From 1b69d04db6f19462e161e6cebee956b142dc6f2e Mon Sep 17 00:00:00 2001
From: Andreas Herten <a.herten@fz-juelich.de>
Date: Fri, 8 Nov 2019 15:48:36 +0100
Subject: [PATCH] Update Notebook to 2019
---
...dsOnPerformanceOptimization-solution.ipynb | 2417 ++++++++++++-
.../HandsOnPerformanceOptimization-task.ipynb | 3213 +++++++++++------
.../HandsOnPerformanceOptimization.ipynb | 1841 +++-------
.../HandsOnPerformanceOptimization.html | 528 ++-
.../HandsOnPerformanceOptimization.ipynb | 471 ++-
.../HandsOnPerformanceOptimization.pdf | Bin 54013 -> 85793 bytes
.../HandsOnPerformanceOptimization.html | 2492 ++++++++++---
.../HandsOnPerformanceOptimization.ipynb | 2417 ++++++++++++-
.../HandsOnPerformanceOptimization.pdf | Bin 74614 -> 129532 bytes
9 files changed, 10086 insertions(+), 3293 deletions(-)
diff --git a/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-solution.ipynb b/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-solution.ipynb
index cb7fcfd..5208f9b 100644
--- a/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-solution.ipynb
+++ b/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-solution.ipynb
@@ -1 +1,2416 @@
-{"cells": [{"cell_type": "markdown", "metadata": {}, "source": ["# Hands-On Performance Optimization\n", "_Supercomputing 2018 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 12th 2018_\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.\n", "\n", "## Jupyter notebook execution\n", "\n", "When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the _edit_ menu above.\n", "\n", "You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n", "\n", "If you want you also can get a [terminal](/terminals/1) in your browser.\n", "\n", "## Terminal fallback\n", "\n", "The tasks are place in directories named `Task[1-3]`.\n", "\n", "Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description."]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Setup\n", "\n", "This hands-on session requires of GCC 6.4.0. By loading the `sc18/handson2` module before invoking this Notebook, we took care of also loading GCC 6.4.0 into the environment."]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Tasks<a name=\"top\"></a>\n", "\n", "This session comes with multiple tasks, each one to be found in the respective sub-directory `Task[1-3]`. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.\n", "\n", "Please choose from the task below.\n", "\n", "\n", "* [Task 1](#task1): Compile Flags \n", "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback ([Solution 1](#solution0))\n", "\n", "* [Task 2](#task2): Software Prefetching \n", "Improve performance of the CPU Jacobi solver with software prefetching ([Solution 2](#solution1))\n", "\n", "* [Task 3](#task3): OpenMP \n", "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance ([Solution 3](#solution2))\n", " \n", "* [Suvery](#survey) Please remember to take the survey !\n", " \n", "### Make Targets <a name=\"make\"></a>\n", "\n", "For all tasks we have defined the following make targets. \n", "\n", "* __poisson2d__: \n", " build `poisson2d` binary (default)\n", "* __run__: \n", " run `poisson2d` with default parameters\n"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Task 1: Compile Flags <a name=\"task1\"></a>\n", "\n", "\n", "### Overview\n", "\n", "The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver \n", "\n", "Your task is to:\n", "\n", "* Optimize performance with `-Ofast` flag\n", "* Optimize performance with profile directed feedback \n", "\n", "First, change the working directory to `Task1`."]}, {"cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task1\n"]}], "source": ["%cd Task1"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part A: `-Ofast` vs. `-O3`\n", "\n", "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. Right now, the Makefile specifies `-O3` as the optimization flag. Compile the code using `make` and run it with `make run` in the next two cells."]}, {"cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"]}], "source": ["!make"]}, {"cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n", "Job <5033> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "1.13user 0.00system 0:01.15elapsed 97%CPU (0avgtext+0avgdata 10944maxresident)k\n", "2560inputs+0outputs (1major+264minor)pagefaults 0swaps\n"]}], "source": ["!make run"]}, {"cell_type": "markdown", "metadata": {}, "source": ["You can use the GNU _perf_ tool to profile the application using the `perf` command (see below) and see the top time-consuming functions."]}, {"cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "[ perf record: Woken up 1 times to write data ]\n", "[ perf record: Captured and wrote 0.172 MB perf.O3.data (4125 samples) ]\n", "# To display the perf.data header info, please use --header/--header-only options.\n", "#\n", "#\n", "# Total Lost Samples: 0\n", "#\n", "# Samples: 4K of event 'cycles:u'\n", "# Event count (approx.): 3867635297\n", "#\n", "# Overhead Command Shared Object Symbol \n", "# ........ ......... ................. ........................................\n", "#\n", " 72.02% poisson2d poisson2d [.] 00000040.plt_call.fmax@@GLIBC_2.17\n", " 10.16% poisson2d poisson2d [.] poisson2d_reference\n", " 9.99% poisson2d poisson2d [.] main\n", " 4.69% poisson2d libc-2.17.so [.] __memcpy_power7\n", " 2.23% poisson2d libm-2.17.so [.] __fmaxf\n", " 0.75% poisson2d libm-2.17.so [.] __exp_finite\n", " 0.07% poisson2d poisson2d [.] 00000040.plt_call.memcpy@@GLIBC_2.17\n", " 0.02% poisson2d poisson2d [.] check_results\n", " 0.02% poisson2d libm-2.17.so [.] __GI___exp\n", " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n", " 0.01% poisson2d [kernel.kallsyms] [k] arch_local_irq_restore\n", " 0.00% poisson2d ld-2.17.so [.] _dl_new_object\n", " 0.00% poisson2d ld-2.17.so [.] _start\n", "\n", "\n", "#\n", "# (Tip: Show user configuration overrides: perf config --user --list)\n", "#\n"]}], "source": ["# perf record creates a perf.data file \n", "!perf record -o perf.O3.data -e cycles ./poisson2d\n", "# perf report opens the perf.data file \n", "!perf report -i perf.O3.data | cat"]}, {"cell_type": "markdown", "metadata": {}, "source": ["**TASK**: Now change the optimization flag in the [Makefile](/edit/Task1/Makefile) to `-Ofast` and repeat the steps in the following cell. In case you follow along non-interactive, call `make` and `make run` in your shell. (If you are in the Jupyter Notebook, you can actually click the link of the [Makefile](/edit/Task1/Makefile). In other cases, use `vim` which is installed on Ascent.)"]}, {"cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"]}], "source": ["!make"]}, {"cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n", "Job <5034> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.51user 0.00system 0:00.52elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+264minor)pagefaults 0swaps\n"]}], "source": ["!make run"]}, {"cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "[ perf record: Woken up 1 times to write data ]\n", "[ perf record: Captured and wrote 0.086 MB perf.Ofast.data (1889 samples) ]\n", "# To display the perf.data header info, please use --header/--header-only options.\n", "#\n", "#\n", "# Total Lost Samples: 0\n", "#\n", "# Samples: 1K of event 'cycles:u'\n", "# Event count (approx.): 1765737747\n", "#\n", "# Overhead Command Shared Object Symbol \n", "# ........ ......... ............. .......................\n", "#\n", " 44.65% poisson2d poisson2d [.] main\n", " 43.84% poisson2d poisson2d [.] poisson2d_reference\n", " 10.28% poisson2d libc-2.17.so [.] __memcpy_power7\n", " 1.12% poisson2d libm-2.17.so [.] __exp_finite\n", " 0.05% poisson2d poisson2d [.] check_results\n", " 0.03% poisson2d ld-2.17.so [.] _dl_relocate_object\n", " 0.02% poisson2d libc-2.17.so [.] __readdir64\n", " 0.01% poisson2d ld-2.17.so [.] _dl_new_object\n", " 0.00% poisson2d ld-2.17.so [.] _start\n", "\n", "\n", "#\n", "# (Tip: System-wide collection from all CPUs: perf record -a)\n", "#\n"]}], "source": ["# perf record creates a perf.data file \n", "!perf record -o perf.Ofast.data -e cycles ./poisson2d\n", "# perf report opens the perf.data file \n", "!perf report -i perf.Ofast.data | cat"]}, {"cell_type": "markdown", "metadata": {}, "source": ["If `perf` is unavailable to you on other machines, you can also study the disassembly with `objdump`: `objdump -lSd ./poisson2d > poisson2d.dis` (feel free to experiment with this in the Notebook as well, just prefix the command with a `!` to execute it.)"]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### Interpretation\n", "\n", "Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with `-Ofast` which enables `\u2013ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part B: Profile-directed Feedback\n", "\n", "For the first level of optimization we saw `Ofast` cut the execution time of the `O3` binary by almost half.\n", "\n", "We can optimize the performance further by using profile directed feedback optimization.\n", "\n", "To compile using profile directed feedback with the GCC compiler we need to do the following steps\n", "\n", "1. We need to first build a training binary using `-fprofile-generate`; this instructs the compiler to record hot path information \n", "2. Run the training binary with a smaller input size; you should see a `.gcda` file generated which stores hot path information for further optimization by the compiler \n", "3. build the final binary using `-fprofile-use` which uses the profile information in the `.gcda` file \n", "4. Compare the performance of the final binary with the original `Ofast` binary \n", "\n", "**TASK**: First, search for `TODO1` in the [Makefile](/edit/Task1/Makefile). It defines an additional compilation flag for `gcc`. Insert `-fprofile-generate=FOLDER` there with FOLDER pointing to `$$SC18_DIR_SCRATCH`, your personal write-directory (the double dollar signs are intentional as they are used to escape in the GNU Make syntax).\n", "\n", "After editing, run the following two cells to train your program."]}, {"cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-generate=$SC18_DIR_SCRATCH\" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-generate=$SC18_DIR_SCRATCH\" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_train -lm\n"]}], "source": ["!make poisson2d_train"]}, {"cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_train 200 64 64\n", "Job <5035> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 200 iterations on 64 x 64 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.248743\n", " 100, 0.124046\n", "Calculate current execution.\n", " 0, 0.248743\n", " 100, 0.124046\n", "0.00user 0.00system 0:00.10elapsed 5%CPU (0avgtext+0avgdata 5248maxresident)k\n", "512inputs+0outputs (0major+115minor)pagefaults 0swaps\n", "mv $SC18_DIR_SCRATCH/*.gcda .\n"]}], "source": ["!make run_train"]}, {"cell_type": "markdown", "metadata": {}, "source": ["Now, a `.gcda` file exists in the directory which can be used for an profile-accelerated subsequent run.\n", "\n", "**TASK**: Edit the [Makefile](/edit/Task1/Makefile) again, this time modifying `TODO2` to be equivalent to `-fprofile-use`. A directory is not needed as we copied the gcda file into the current directory.\n", "\n", "Run the following cells in order to build using the newly added flag and then run with the profile-accelerated version."]}, {"cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-use\" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-use\" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_profile -lm\n"]}], "source": ["!make poisson2d_profile"]}, {"cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_profile\n", "Job <5036> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.47user 0.00system 0:00.48elapsed 98%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"]}], "source": ["!make run_profile"]}, {"cell_type": "markdown", "metadata": {}, "source": ["What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### References\n", "\n", "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n", "2. https://perf.wiki.kernel.org/index.php/Tutorial"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Task 2:<a name=\"task2\"></a> Software Pretechting\n", "\n", "\n", "### Overview\n", "\n", "Study the difference of program execution time of different optimization levels with and without software prefetching.\n", "\n", "First, change directory to that of Task 2"]}, {"cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task2\n"]}], "source": ["%cd ../Task2"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part A: Running"]}, {"cell_type": "markdown", "metadata": {}, "source": ["Look at the [Makefile](/edit/Task2/Makefile) and work on the TODOs. Please implement compile flags as mentioned in the Makefile target name.\n", "\n", "Afterwards, compile each target with the following cells and submit them to the batch system. Follow along accordingly in the non-interactive version of this Notebook."]}, {"cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_pref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_pref\n", "Job <5037> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "1.12user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10880maxresident)k\n", "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"]}], "source": ["!make run_o3_pref"]}, {"cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_pref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_pref\n", "Job <5038> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.77user 0.00system 0:00.77elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+264minor)pagefaults 0swaps\n"]}], "source": ["!make run_ofast_pref"]}, {"cell_type": "code", "execution_count": 15, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_nopref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_nopref\n", "Job <5039> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "1.13user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10944maxresident)k\n", "256inputs+0outputs (0major+266minor)pagefaults 0swaps\n"]}], "source": ["!make run_o3_nopref"]}, {"cell_type": "code", "execution_count": 16, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_nopref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_nopref\n", "Job <5040> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.82user 0.00system 0:00.82elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"]}], "source": ["!make run_ofast_nopref"]}, {"cell_type": "markdown", "metadata": {}, "source": ["Do you notice the impact difference with optimization levels? It's always important to carefully study the interplay of flags."]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part B: Analysis of Instructions\n", "\n", "Compilation with the software prefetching flag causes the compiler to generate the `__dcbt` and `__dcbtst` instructions that prefetch memory values to L3.\n", "\n", "Verify it using `objdump -lSd` on each file (`poisson2d_o3_pref`, `poisson2d_ofast_pref`, `poisson2d_o3_nopref`, `poisson2d_ofast_nopref`). You might want to grep for `dcb`."]}, {"cell_type": "code", "execution_count": 19, "metadata": {"exercise": "solution"}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["poisson2d_o3_pref:\n", "poisson2d_ofast_pref:\n", " 10000da0:\tec f1 00 7c \tdcbtst 0,r30\n", " 10000da4:\t2c fa 00 7c \tdcbt 0,r31\n", " 10000da8:\t2c 62 00 7c \tdcbt 0,r12\n", " 10000dac:\t2c b2 00 7c \tdcbt 0,r22\n", " 10000dcc:\t2c e2 00 7c \tdcbt 0,r28\n", " 10000dd0:\t2c ea 00 7c \tdcbt 0,r29\n", " 100010b4:\t2c 62 00 7c \tdcbt 0,r12\n", " 100010b8:\t2c 5a 00 7c \tdcbt 0,r11\n", " 100010c4:\tec 19 00 7c \tdcbtst 0,r3\n", " 100010cc:\t2c 22 00 7c \tdcbt 0,r4\n", " 100010d0:\t2c ea 00 7c \tdcbt 0,r29\n", " 100010d4:\t2c f2 00 7c \tdcbt 0,r30\n", " 100010dc:\t2c fa 00 7c \tdcbt 0,r31\n", "poisson2d_o3_nopref:\n", "poisson2d_ofast_nopref:\n"]}], "source": ["for f in [\"poisson2d_o3_pref\", \"poisson2d_ofast_pref\", \"poisson2d_o3_nopref\", \"poisson2d_ofast_nopref\"]:\n", " print(\"{}:\".format(f))\n", " objdump -lSd $f |\u00a0grep dcb"]}, {"cell_type": "markdown", "metadata": {}, "source": ["If you feel up to the task, you can study the number of L3 cache misses using the corresponding performance counter, `PM_L3_MISS`. Either use your knowledge from Hands-On 1, or use the following call to `perf`, in which we already converted the named counter to a raw counter address."]}, {"cell_type": "code", "execution_count": 35, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Job <5048> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "\n", " Performance counter stats for './poisson2d_ofast_nopref':\n", "\n", " 2829292169 cycles:u \n", " 136018637 r168a4:u \n", "\n", " 0.826136863 seconds time elapsed\n", "\n", "Job <5049> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "\n", " Performance counter stats for './poisson2d_ofast_pref':\n", "\n", " 2654990243 cycles:u \n", " 128824827 r168a4:u \n", "\n", " 0.775593651 seconds time elapsed\n", "\n"]}], "source": ["for f in [\"poisson2d_ofast_nopref\", \"poisson2d_ofast_pref\"]:\n", " !eval $$SC18_SUBMIT_CMD perf stat -e cycles,r168a4 ./$f\n"]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### References\n", "\n", "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n", "2. https://www.gnu.org/software/gcc/projects/prefetch.html"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Task 3: OpenMP\n", "<a name=\"task3\"></a>\n", "\n", "\n", "### Overview\n", "\n", "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores.\n", "\n", "First, we need to change directory to that of Task3."]}, {"cell_type": "code", "execution_count": 1, "metadata": {"ExecuteTime": {"end_time": "2018-11-07T13:47:57.724441Z", "start_time": "2018-11-07T13:47:57.718745Z"}}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task3\n"]}], "source": ["%cd ../Task3"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part A: Implement OpenMP Pragmas; Compilation\n", "\n", "**Task**: Please add the correct OpenMP pragmas to the source code and compilations flags to enable OpenMP.\n", "\n", "* **pragmas**: Look at the TODOs in [`poisson2d.c`](/edit/Task3/poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for`\n", "* **Compilation**: Please add compilation flags enabling OpenMP in GCC to the [Makefile](/edit/Task3/Makefile). The flag in question is `-fopenmp`.\n", "\n", "Edit the files with the links above if you are running the interactive version of the Notebook or navigate to `poisson2d.c` and `Makefile` yourself in case you run the non-interactive version.\n", "\n", "Afterwards, compile and run the application with the following cells. Non-interactive: Follow along accordingly in the shell."]}, {"cell_type": "code", "execution_count": 37, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"]}], "source": ["!make poisson2d"]}, {"cell_type": "code", "execution_count": 40, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n", "Job <5052> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "500x500: Ref: 0.2571 s, This: 0.2946 s, speedup: 0.87\n", "1.48user 0.00system 0:00.56elapsed 263%CPU (0avgtext+0avgdata 9664maxresident)k\n", "0inputs+0outputs (0major+273minor)pagefaults 0swaps\n"]}], "source": ["!make run"]}, {"cell_type": "markdown", "metadata": {}, "source": ["The command to submit a job to the batch system is prepared in an environment variable `$SC18_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to increase the work of the application."]}, {"cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Job <5344> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 1000 iterations on 1000 x 100 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249743\n", " 100, 0.210080\n", " 200, 0.184635\n", " 300, 0.166526\n", " 400, 0.152783\n", " 500, 0.141890\n", " 600, 0.132978\n", " 700, 0.125511\n", " 800, 0.119142\n", " 900, 0.113632\n", "Calculate current execution.\n", " 0, 0.249743\n", " 100, 0.210080\n", " 200, 0.184635\n", " 300, 0.166526\n", " 400, 0.152783\n", " 500, 0.141890\n", " 600, 0.132978\n", " 700, 0.125511\n", " 800, 0.119142\n", " 900, 0.113632\n", "1000x100: Ref: 1.9872 s, This: 0.2385 s, speedup: 8.33\n"]}], "source": ["!eval $SC18_SUBMIT_CMD ./poisson2d 1000 1000"]}, {"cell_type": "markdown", "metadata": {}, "source": ["What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n", "We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."]}, {"cell_type": "code", "execution_count": 23, "metadata": {"exercise": "solution"}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Threads: 1\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3037 s, This: 2.8420 s, speedup: 0.81\n", "Threads: 2\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.2998 s, This: 1.4320 s, speedup: 1.61\n", "Threads: 4\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3135 s, This: 0.7168 s, speedup: 3.23\n", "Threads: 8\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3145 s, This: 0.5278 s, speedup: 4.39\n", "Threads: 10\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3153 s, This: 0.4848 s, speedup: 4.78\n", "Threads: 20\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3190 s, This: 0.2016 s, speedup: 11.50\n", "Threads: 40\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3243 s, This: 0.3057 s, speedup: 7.60\n"]}], "source": ["for omp_num in [1, 2, 4, 8, 10, 20, 40]:\n", " print(\"Threads: {}\".format(omp_num))\n", " !eval OMP_NUM_THREADS=$omp_num $$SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 | grep speedup"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part B: Bindings\n", "\n", "Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!\n", "\n", "There are applications which can be used to determine the configuration of the processor. Among those are:\n", "\n", "* `lscpu`: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.\n", "* `ppc64_cpu --smt`: Specifically for POWER, this tool can give information about the number of simulations threads running per core (*SMT*, Simulataion Multi-Threading).\n", "\n", "Run `ppc64_cpu --smt` to find out about the threading configuration of Ascent!"]}, {"cell_type": "code", "execution_count": 48, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Job <5076> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "SMT=4\n"]}], "source": ["!eval $SC18_SUBMIT_CMD ppc64_cpu --smt"]}, {"cell_type": "markdown", "metadata": {}, "source": ["There are more sources information available\n", "\n", "* `/proc/cpuinfo`: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with `cat`\n", "* `/sys/devices/system/cpu/cpu0/topology/thread_siblings_list`: Holds information about thread siblings for given CPU core (`cpu0` in this case). Use it to find out which thread is mapped to which core."]}, {"cell_type": "code", "execution_count": 49, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["0-3\n", "4-7\n"]}], "source": ["!cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n", "!cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"]}, {"cell_type": "markdown", "metadata": {}, "source": ["There are various environment variables available within OpenMP (and GCC) to specify binding of threads to cores. See, for instance, the [online documentation of GCC libgomp](https://gcc.gnu.org/onlinedocs/libgomp/Environment-Variables.html). Examples are `OMP_PLACES` or `GOMP_CPU_AFFINITY`.\n", "\n", "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n", "\n", "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n", "\n", "What's your maximum speedup?"]}, {"cell_type": "code", "execution_count": 24, "metadata": {"exercise": "solution"}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Affinity: 0,1,2,3\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", " OMP_PLACES = '{0},{1},{2},{3}'\n", "1000x100: Ref: 1.9854 s, This: 0.2326 s, speedup: 8.53\n", "Affinity: 0,5,9,13\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", " OMP_PLACES = '{0},{5},{9},{13}'\n", "1000x100: Ref: 1.9828 s, This: 0.0833 s, speedup: 23.80\n"]}], "source": ["for affinity in [\"0,1,2,3\", \"0,5,9,13\"]:\n", " print(\"Affinity: {}\".format(affinity))\n", " !eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=$affinity OMP_NUM_THREADS=4 $$SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### References\n", "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["# Survey<a name=\"survey\"></a>\n", "\n", "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc18-eval)."]}], "metadata": {"kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"}, "language_info": {"codemirror_mode": {"name": "ipython", "version": 3}, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.6.7"}}, "nbformat": 4, "nbformat_minor": 2}
\ No newline at end of file
+{
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Hands-On Performance Optimization\n",
+ "_Supercomputing 2019 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 18th 2019_\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.\n",
+ "\n",
+ "## Jupyter notebook execution\n",
+ "\n",
+ "When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the _edit_ menu above.\n",
+ "\n",
+ "You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n",
+ "\n",
+ "If you want you also can get a terminal in your browser; just open it via the \u00bbNew Launcher\u00ab button (`+`).\n",
+ "\n",
+ "## Terminal fallback\n",
+ "\n",
+ "The tasks are place in directories named `Task[1-3]`.\n",
+ "\n",
+ "Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Setup\n",
+ "\n",
+ "We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc (GCC) 9.2.0\n",
+ "Copyright (C) 2019 Free Software Foundation, Inc.\n",
+ "This is free software; see the source for copying conditions. There is NO\n",
+ "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!gcc --version"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Tasks<a name=\"top\"></a>\n",
+ "\n",
+ "This session comes with multiple tasks, each one to be found in the respective sub-directory `Task[1-3]`. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.\n",
+ "\n",
+ "Please choose from the task below.\n",
+ "\n",
+ "\n",
+ "* [Task 1](#task1): __Basic compiler optimization flags and compiler annotations__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback. Learn about compiler annotations.\n",
+ "\n",
+ "* [Task 2](#task2): __Optimization via Prefetching controlled by compiler__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with software prefetching. Some compilers such as IBM XL define flags that can be used to modify the aggressiveness of the hardware prefetcher. Learn to modify the DSCR value through XL and study the impact on application performance. \n",
+ "* [Task 3](#task3): __Optimization via OpenMP controlled by compiler and the system__\n",
+ "\n",
+ "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance. \n",
+ " \n",
+ "* [Suvery](#survey) Please remember to take the survey !\n",
+ " \n",
+ "### Make Targets <a name=\"make\"></a>\n",
+ "\n",
+ "For all tasks we have defined the following make targets. \n",
+ "\n",
+ "* __poisson2d__: \n",
+ " build `poisson2d` binary (default)\n",
+ "* __run__: \n",
+ " run `poisson2d` with default parameters\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 1: Basic compiler optimization flags and compiler annotations <a name=\"task1\"></a>\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver \n",
+ "\n",
+ "Your task is to:\n",
+ "\n",
+ "* Optimize performance with `-Ofast` flag\n",
+ "* Verify the cause for performance improvement by viewing perf profiles of O3 and Ofast binaries \n",
+ "* Optimize performance with profile directed feedback \n",
+ "* Generate compiler annotations/remarks to understand the optimizations done by the compiler with and without profile directed feedback \n",
+ "\n",
+ "\n",
+ "First, change the working directory to `Task1`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task1\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd Task1"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: `-Ofast` vs. `-O3`\n",
+ "\n",
+ "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. As in the previous task, we use a `Makefile` for compilation. The `Makefile` targets `poisson2d_O3` and `poisson2d_Ofast` are already prepared. \n",
+ "\n",
+ "**TASK**: Add `-O3` as the optimization flag for the `poisson2d_O3` target by using the corresponding `CFLAGS` definition. There are notes relating to this Task 1 in the header of the `Makefile`. Compile the code using `make` as indicated below and run with the `Make` targets `run`, `run_perf` and `run_perf_recrep`. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 84,
+ "metadata": {
+ "collapsed": true
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_O3"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 73,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24897> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's have a look at the output of the `Makefile` target `run_perf`. It invokes the GNU _perf_ tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to _perf_."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 74,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24898> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16264721613 cycles:u \n",
+ " 28463907825 instructions:u # 1.75 insn per cycle \n",
+ "\n",
+ " 4.738444892 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Next we run the makefile with target `run_perf_recrep` that prints the top routines of the application in terms of hotness by using a combination of `perf record ./app` and `perf report`. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 75,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24899> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "[ perf record: Woken up 3 times to write data ]\n",
+ "[ perf record: Captured and wrote 0.739 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (19102 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24900> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "# To display the perf.data header info, please use --header/--header-only options.\n",
+ "#\n",
+ "#\n",
+ "# Total Lost Samples: 0\n",
+ "#\n",
+ "# Samples: 19K of event 'cycles:u'\n",
+ "# Event count (approx.): 16254596654\n",
+ "#\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
+ "#\n",
+ " 65.50% poisson2d poisson2d [.] 00000038.plt_call.fmax@@GLIBC_2.17\n",
+ " 21.21% poisson2d poisson2d [.] main\n",
+ " 9.18% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 3.28% poisson2d libm-2.17.so [.] __fmaxf\n",
+ " 0.74% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.01% poisson2d libm-2.17.so [.] __GI___exp\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.00% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] _wordcopy_fwd_aligned\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ "\n",
+ "\n",
+ "#\n",
+ "# (Tip: Limit to show entries above 5% only: perf report --percent-limit 5)\n",
+ "#\n"
+ ]
+ }
+ ],
+ "source": [
+ "# run_perf_recrep displays the top hot routines \n",
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Now add the optimization flag `Ofast` to the `CFLAGS` for target `poisson2d_Ofast`. Compile the program with the target `poisson2d_Ofast` and run and analyse it as before with `run`, `run_perf` and `run_perf_recrep`.\n",
+ "\n",
+ "What difference do you see?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 76,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24901> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.41user 0.00system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast \n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Again, run a `perf`-instrumented version:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 77,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24902> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 8258991976 cycles:u \n",
+ " 12013091172 instructions:u # 1.45 insn per cycle \n",
+ "\n",
+ " 2.408703909 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Generate the list of top routines in terms of hotness:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 78,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24903> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "[ perf record: Woken up 2 times to write data ]\n",
+ "[ perf record: Captured and wrote 0.382 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (9728 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24904> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "# To display the perf.data header info, please use --header/--header-only options.\n",
+ "#\n",
+ "#\n",
+ "# Total Lost Samples: 0\n",
+ "#\n",
+ "# Samples: 9K of event 'cycles:u'\n",
+ "# Event count (approx.): 8268811890\n",
+ "#\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
+ "#\n",
+ " 81.12% poisson2d poisson2d [.] main\n",
+ " 17.97% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 0.79% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.02% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.01% poisson2d libc-2.17.so [.] vfprintf@@GLIBC_2.17\n",
+ " 0.01% poisson2d libc-2.17.so [.] _dl_addr\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.01% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] open_path\n",
+ " 0.00% poisson2d ld-2.17.so [.] init_tls\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ "\n",
+ "\n",
+ "#\n",
+ "# (Tip: For tracepoint events, try: perf report -s trace_fields)\n",
+ "#\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If `perf` is unavailable to you on other machines, you can also study the disassembly with `objdump`: `objdump -lSd ./poisson2d > poisson2d.dis` (feel free to experiment with this in the Notebook as well, just prefix the command with a `!` to execute it.)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### Interpretation\n",
+ "\n",
+ "Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with `-Ofast` which enables `\u2013ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Profile-directed Feedback\n",
+ "\n",
+ "For the first level of optimization we see that `Ofast` cut the execution time of the `O3` binary by almost half.\n",
+ "\n",
+ "We can optimize the performance further by using profile-directed feedback optimization.\n",
+ "\n",
+ "To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:\n",
+ "\n",
+ "1. Instrument binary;\n",
+ "2. Run binary with training, gather profile information;\n",
+ "3. Use profile information to generate optimized binary.\n",
+ "\n",
+ "\n",
+ "Step 1 is achieved by compiling the binary with the correct flag \u2013\u00a0`-fprofile-generate`. In our case, we need to specify an output location, which should be `$(SC19_DIR_SCRATCH)`.\n",
+ "\n",
+ "Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension `.gcda`) is written to the directory specified during compilation.\n",
+ "\n",
+ "For Step 3, the binary is once again compiled, but this time using the `gcda` profile just generated. The according flag is `-fprofile-use`, which we set to `$(SC19_DIR_SCRATCH)` as well.\n",
+ "\n",
+ "In our `Makefile` at hand, we prepared the steps already for you in the form of two targets.\n",
+ "\n",
+ "* `poisson2d_train`: Will compile the binary with profile-directed feedback\n",
+ "* `poisson2d_ref`: Will take a generated profile and compile a new, optimized binary\n",
+ "\n",
+ "By using dependencies, between these two targets a profile run is launched.\n",
+ "\n",
+ "**TASK**: Edit the [Makefile](`Makefile`) and add the `-fprofile-*` flags to the `CFLAGS` of `poisson2d_train` and\n",
+ "`poisson2d_ref` as outline in the file.\n",
+ "\n",
+ "After that, you may launch them with the following cells (`gen_profile` is a meta-target and uses `poisson2d_train` and `poisson2d_ref`). If you need to clean the generated profile, you may use `make clean_profile`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 79,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24905> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_ref -lm \n",
+ "cp poisson2d_ref poisson2d\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make gen_profile"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 80,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24906> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.28user 0.01system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Great! It is! In our tests, this shaved off another 5%."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's also measure instructions and cycles"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 81,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24907> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 7925983538 cycles:u \n",
+ " 12253080719 instructions:u # 1.55 insn per cycle \n",
+ "\n",
+ " 2.313471365 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Compiler annotations/Remarks\n",
+ "\n",
+ "Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done. \n",
+ "\n",
+ "To generate compiler annotations using GCC, one uses `-fopt-info-all`. If you only want to see the missed options, use the option `-fopt-info-missed` instead of `-fopt-info-all`. See also the [documentation of GCC regarding the flag](https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info).\n",
+ "\n",
+ "**TASK**: Have a looK at the `CFLAGS` of the `Makefile` target `poisson2d_Ofast_info`. Add the flag `-fopt-info-all` to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use \u2013\u00a0for example \u2013\u00a0`-fopt-info-all=(SC19_DIR_SCRATCH)/filename`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 82,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all poisson2d.c poisson2d_reference.o -o poisson2d_Ofast_info -lm\n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 207->207 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.\n",
+ "\n",
+ "**TASK**: \n",
+ "Adapt the `CFLAGS` of `poisson2d_ref_info` to include `-fopt-info-all` **and** the profile input of `-fprofile-use=\u2026` here. *(Be advised: Long output!)*"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 83,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "Increasing alignment of decl: __gcov0.main\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_D_00100_1_main/48 -> __gcov_exit/55, function body not available\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_I_00100_0_main/47 -> __gcov_init/54, function body not available\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 295->295 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:122:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:88:5: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:72:5: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_337, ny_124, nx_286);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_316, error_118);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_127);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_311);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_122);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_129 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_132 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_140 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 53\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 50\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 47\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 3 times (header execution count 9800)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 33\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 30\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 42\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 40\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 60\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 23\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 3 times (header execution count 100)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 12\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 16\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_init (&*.LPBX0);\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_exit ();\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24908> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "libgcov profiling error:/gpfs/wolf/trn003/scratch/aherten//#autofs#nccsopen-svm1_home#aherten#SC19-Tutorial#3-Optimizing_POWER#Handson#Task1#poisson2d.gcda:overwriting an existing profile data with a different timestamp\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c poisson2d_reference.o -o poisson2d_ref_info -lm\n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 207->207 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 47\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 44\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 40\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 27\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 24\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 37\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 35\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 51\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 18\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 7 times (header execution count 99)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 9\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 14\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_ref_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Comparing the annotations generated of a plain `-Ofast` optimization level and the one generated at `-Ofast` and profile directed feedback, we observe that many more optimizations are possible due to profile information.\n",
+ "\n",
+ "For instance you will see annotations such as\n",
+ "```\n",
+ "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "```\n",
+ "\n",
+ "The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "\n",
+ "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
+ "2. https://perf.wiki.kernel.org/index.php/Tutorial"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 2:<a name=\"task2\"></a> Impact of Prefetching on Performance\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "* Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
+ "* Verify the impact by measuring cache counters with and without prefetching.\n",
+ "* Learn how to modify contents of DSCR (*Data Stream Control Register*) using IBM XL compiler and study the impact with different values to DSCR. \n",
+ "\n",
+ "But first, lets change directory to that of Task 2"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 85,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task2\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd ../Task2"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: Software Prefetching"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Look at the Makefile and work on the TODOs. \n",
+ "\n",
+ "- First generate a `-Ofast`-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!\n",
+ "- Modify the `Makefile` to add the option for software prefetching (`-fprefetch-loop-arrays`). Compare performance of `-Ofast` with and without software prefetching"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 97,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "rm -f poisson2d poisson2d*.o\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make clean"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 88,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "make: `poisson2d' is up to date.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24911> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.39user 0.01system 0:02.40elapsed 100%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24912> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 8271503902 cycles:u \n",
+ " 481152478 r168a4:u \n",
+ "\n",
+ " 2.412224884 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 98,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ "cp poisson2d_pref poisson2d\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24919> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1.92user 0.00system 0:01.93elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24920> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 6586609284 cycles:u \n",
+ " 459879452 r168a4:u \n",
+ "\n",
+ " 1.925399505 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_pref CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Repeat the experiment with the `-O3` flag. Have a look at the `Makefile` and the outlined TODO. There's a position to easily adapt `-Ofast`\u2192`-O3`!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 100,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec poisson2d.c -o poisson2d -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24923> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24924> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16445764669 cycles:u \n",
+ " 645094089 r168a4:u \n",
+ "\n",
+ " 4.792567763 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 101,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24925> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.74user 0.00system 0:04.74elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24926> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16239159454 cycles:u \n",
+ " 631061431 r168a4:u \n",
+ "\n",
+ " 4.730144897 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_pref CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Observing the results, we see that SW Prefetching seems to help at `-Ofast` but not at `-O3`. We can use the steps described in the the next section to verify that the compiler has not inserted any SW prefetch operations at`-O3` at all. That is because in the `-O3` binary the time is dominated by `__fmax` call which causes the compiler to come to the conclusion that whatever benefit we obtain by adding SW prefetch will be overshadowed by the penalty of `fmax()`\n",
+ "GCC may add further loop optimizations such as unrolling upon invocation of `\u2013fprefetch-loop-arrays`.\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Analysis of Instructions\n",
+ "\n",
+ "Compilation of the `-Ofast` binary with the software prefetching flag causes the compiler to generate the `dcb*` instructions that prefetch memory values to L3."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: \n",
+ "Run `$(SC19_SUBMIT_CMD) objdump -lSd` on each binary file (`-O3`, `-Ofast` with prefetch/no prefetch).\n",
+ "Look for instructions beginning with `dcb`\n",
+ "At what optimization levels does the compiler generate software prefetching instructions?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 114,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=gcc -B poisson2d_pref\n",
+ "!objdump -lSd ./poisson2d_pref > poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 116,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ " 10000b28:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000b30:\t2c ba 00 7c \tdcbt 0,r23\n",
+ " 10000b38:\t2c b2 00 7c \tdcbt 0,r22\n",
+ " 10000b50:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000b58:\tec b9 00 7c \tdcbtst 0,r23\n",
+ " 10000b80:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000e64:\t2c 92 00 7c \tdcbt 0,r18\n",
+ " 10000e68:\t2c 9a 00 7c \tdcbt 0,r19\n",
+ " 10000e6c:\t2c a2 00 7c \tdcbt 0,r20\n",
+ " 10000e70:\t2c aa 00 7c \tdcbt 0,r21\n",
+ " 10000e7c:\t2c b2 00 7c \tdcbt 0,r22\n",
+ " 10000e80:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000e94:\tec b9 00 7c \tdcbtst 0,r23\n"
+ ]
+ }
+ ],
+ "source": [
+ "!grep dcb poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Changing Values of DSCR via compiler flags\n",
+ "\n",
+ "This task requires using the IBM XL compiler. It should be already in your environment.\n",
+ "\n",
+ "\n",
+ "We saw the impact of software prefetching in the previous subsection. \n",
+ "In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance. \n",
+ "In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching. \n",
+ "\n",
+ "IBM XL compiler has an option `-qprefetch=dscr=<val>` that can be used for this purpose.\n",
+ "Compiling with `-qprefetch=dscr=1` turns off the prefetcher. One can give various values such as `-qprefetch=dscr=4`, `-qprefetch=dscr=7` etc. to control aggressiveness of prefetching.\n",
+ "\n",
+ "For this exercise we use `make CC=xlc_r` to illustrate the performance impact.\n",
+ " \n",
+ "\n",
+ "**Task** Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the `Makefile`: Add `qprefetch=dscr=1` to the `CFLAGS` and rebuild the application and note the performance. Which one is faster? \n",
+ "\n",
+ "In general, applications benefit with the default settings of hardware DSCR register (`-qprefetch=dscr=0`). However, certain applications also benefit with prefetching turned off. \n",
+ "\n",
+ "It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL at default DSCR value"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 117,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS poisson2d.c -o poisson2d -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24927> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "2.26user 0.00system 0:02.27elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+477minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=xlc_r -B poisson2d\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL with DSCR value turned off"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS -qprefetch=dscr=1 poisson2d.c -o poisson2d_dscr -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24929> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.58user 0.00system 0:04.59elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "0inputs+0outputs (0major+476minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_dscr CC=xlc_r -B\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Does Hardware prefetcher help this application? How much impact do you see when you turn off the hardware prefetcher? "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "The DSCR register controls the operation of the HW Prefetcher on POWER9. It can be modified in the command line by `ppc64_cpu --dscr=<value>`. However this needs admin privileges. IBM XL offers a compiler flag to set the value through the compiler. `-qprefetch=dscr=1` turns off the prefetcher. Observing the results we see that the performance without the HW prefetcher is twice as bad as that with default prefetching. So we can conclude that Prefetching helps the Jacobi application. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "\n",
+ "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
+ "2. https://www.gnu.org/software/gcc/projects/prefetch.html\n",
+ "3. https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 3: OpenMP\n",
+ "<a name=\"task3\"></a>\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores on the resulting application performance. We do this study for both GCC and XL compilers inorder to learn about the appropriate options that need to be used.\n",
+ "First, we need to change directory to that of Task3. For Task 3 we modify poisson2d.c to invoke an exact copy of the main jacobi loop which is `poisson2d_reference`. We parallelize only the main loop but not `poisson2d_reference`. The speedup is the performance gain seen in the main loop as compared to the reference loop."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task3\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd ../Task3"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: Implement OpenMP Pragmas; Compilation\n",
+ "\n",
+ "**Task**: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.\n",
+ "\n",
+ "* **Directives**: Look at the TODOs in [`poisson2d.c`](poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for` (and once it's `#pragma omp parallel for reduction(max:error)` \u2013\u00a0can you guess where?)\n",
+ "* **Compilation**: Please add compilation flags enabling OpenMP in GCC and XL to the `Makefile`. For GCC, we need to add `-fopenmp` and the application needs to be linked with `-lgomp`. For XL, we need to add `-qsmp=omp` to the list of compilation flags. \n",
+ "\n",
+ "Afterwards, compile and run the application with the following commands."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 39,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -c -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d.c poisson2d_reference.o -o poisson2d -lm \n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "The command to submit a job to the batch system is prepared in an environment variable `$SC19_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to invoke the application using the batch system. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 40,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24951> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1000x1000: Ref: 4.7430 s, This: 3.9363 s, speedup: 1.20\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval $SC19_SUBMIT_CMD ./poisson2d 1000 1000 1000"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Inorder to run the parallel application, we need to set the number of threads using `OMP_NUM_THREADS`\n",
+ "What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n",
+ "We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 41,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 4.7288 s, This: 4.9791 s, speedup: 0.95\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=1 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 42,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 4.7125 s, This: 2.4914 s, speedup: 1.89\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=2 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 35,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.1065 s, This: 1.3836 s, speedup: 1.52\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 21,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3868 s, This: 0.5272 s, speedup: 4.53\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=8 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 22,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3912 s, This: 0.4612 s, speedup: 5.18\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=10 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 23,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3864 s, This: 0.4037 s, speedup: 5.91\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=20 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 24,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3773 s, This: 0.3045 s, speedup: 7.81\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=40 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 25,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3819 s, This: 0.3081 s, speedup: 7.73\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=80 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Bindings\n",
+ "\n",
+ "Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!\n",
+ "\n",
+ "There are applications which can be used to determine the configuration of the processor. Among those are:\n",
+ "\n",
+ "* `lscpu`: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.\n",
+ "* `ppc64_cpu --smt`: Specifically for POWER, this tool can give information about the number of simulations threads running per core (*SMT*, Simulataion Multi-Threading).\n",
+ "\n",
+ "Run `ppc64_cpu --smt` to find out about the threading configuration of Ascent!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 55,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24465> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "SMT=4\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval $SC19_SUBMIT_CMD ppc64_cpu --smt"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "There are more sources information available\n",
+ "\n",
+ "* `/proc/cpuinfo`: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with `cat`\n",
+ "* `/sys/devices/system/cpu/cpu0/topology/thread_siblings_list`: Holds information about thread siblings for given CPU core (`cpu0` in this case). Use it to find out which thread is mapped to which core."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 36,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24949> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "0-3\n",
+ "Job <24950> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "4-7\n"
+ ]
+ }
+ ],
+ "source": [
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n",
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the [OMP_PLACES environment Variable](https://www.openmp.org/spec-html/5.0/openmpse53.html). We also have a GNU specific variable which can also be used to control affinity - `GOMP_CPU_AFFINITY`. Setting `GOMP_CPU_AFFINITY` is specific to GCC binaries but it internally serves the same function as setting `OMP_PLACES`. \n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Using `OMP_PLACES` for binding, and using some magical Python-Bash interplay:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 43,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: {0},{1},{2},{3}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{1},{2},{3}'\n",
+ "1000x1000: Ref: 4.7315 s, This: 3.9090 s, speedup: 1.21\n",
+ "Affinity: {0},{5},{9},{13}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{5},{9},{13}'\n",
+ "1000x1000: Ref: 4.6485 s, This: 1.2829 s, speedup: 3.62\n"
+ ]
+ }
+ ],
+ "source": [
+ "for affinity in [\"{0},{1},{2},{3}\", \"{0},{5},{9},{13}\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true OMP_PLACES=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "In this case, we carry out the same experiment using `GOMP_CPU_AFFINITY` which essentially sets the same environment variable `OMP_PLACES`. Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 44,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: 0,1,2,3\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{1},{2},{3}'\n",
+ "1000x1000: Ref: 2.3964 s, This: 2.1361 s, speedup: 1.12\n",
+ "Affinity: 0,5,9,13\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{5},{9},{13}'\n",
+ "1000x1000: Ref: 2.3925 s, This: 0.7030 s, speedup: 3.40\n"
+ ]
+ }
+ ],
+ "source": [
+ "for affinity in [\"0,1,2,3\", \"0,5,9,13\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Great!\n",
+ "\n",
+ "If you still have time: The same experiments can be repeated with the IBM XL compiler. \n",
+ "The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is `-qsmp=omp`\n",
+ "\n",
+ "**Task**: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 44,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -c -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d_reference.c -o poisson2d_reference.o -lm \n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "xlc_r -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24956> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "1000x1000: Ref: 5.6783 s, This: 2.6528 s, speedup: 2.14\n",
+ "21.56user 6.18system 0:08.37elapsed 331%CPU (0avgtext+0avgdata 23040maxresident)k\n",
+ "3200inputs+0outputs (2major+1098minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=xlc_r -B run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Run the parallel application with varying numbre of threads (`OMP_NUM_THREADS`) and note the performance improvement. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Just as in the GCC binary we see a similar speedup with higher number of threads until a certain point beyond which the benefit tapers off. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 28,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2561 s, This: 2.6432 s, speedup: 0.85\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=1 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 29,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3071 s, This: 1.5343 s, speedup: 1.50\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=2 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 30,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2617 s, This: 0.6936 s, speedup: 3.26\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 31,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2728 s, This: 0.3402 s, speedup: 6.68\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=8 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 45,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.1678 s, This: 0.2869 s, speedup: 7.56\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=10 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 33,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2813 s, This: 0.1452 s, speedup: 15.71\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=20 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 34,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3284 s, This: 0.0981 s, speedup: 23.75\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=40 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 35,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2918 s, This: 0.1439 s, speedup: 15.92\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=80 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Now we repeat the exercise of using the right binding of threads for the XL binary. `OMP_PLACES` pertains to the XL binary as well as it is an OpenMP variable. `GOMP_CPU_AFFINITY` is specific to GCC binary so that cannot be used to set the binding.\n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "We are mixing Python with Bash (`!`) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two `$$`)\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 36,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: {0},{1},{2},{3}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES='{0},{1},{2},{3}' custom\n",
+ "1000x1000: Ref: 5.9792 s, This: 2.4122 s, speedup: 2.48\n",
+ "Affinity: {0},{5},{9},{13}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES='{0},{5},{9},{13}' custom\n",
+ "1000x1000: Ref: 2.3101 s, This: 0.6884 s, speedup: 3.36\n"
+ ]
+ }
+ ],
+ "source": [
+ "for affinity in [\"{0},{1},{2},{3}\", \"{0},{5},{9},{13}\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true OMP_PLACES=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Likewise we see a higher speedup when we bind the threads to different cores rather than to a single core. This handson illustrates that apart from compiler level tuning, system level tuning is also equally important to obtain performance improvements \n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html\n",
+ "2. https://www.openmp.org/spec-html/5.0/openmpse53.html"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Survey<a name=\"survey\"></a>\n",
+ "\n",
+ "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc19-eval)."
+ ]
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.7.0"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
\ No newline at end of file
diff --git a/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-task.ipynb b/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-task.ipynb
index b9f231f..3b02952 100644
--- a/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-task.ipynb
+++ b/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization-task.ipynb
@@ -1,1206 +1,2085 @@
{
- "cells": [
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "# Hands-On Performance Optimization\n",
- "_Supercomputing 2018 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 12th 2018_\n",
- "\n",
- "---"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.\n",
- "\n",
- "## Jupyter notebook execution\n",
- "\n",
- "When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the _edit_ menu above.\n",
- "\n",
- "You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n",
- "\n",
- "If you want you also can get a [terminal](/terminals/1) in your browser.\n",
- "\n",
- "## Terminal fallback\n",
- "\n",
- "The tasks are place in directories named `Task[1-3]`.\n",
- "\n",
- "Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description."
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "## Setup\n",
- "\n",
- "This hands-on session requires of GCC 6.4.0. By loading the `sc18/handson2` module before invoking this Notebook, we took care of also loading GCC 6.4.0 into the environment."
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "## Tasks<a name=\"top\"></a>\n",
- "\n",
- "This session comes with multiple tasks, each one to be found in the respective sub-directory `Task[1-3]`. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.\n",
- "\n",
- "Please choose from the task below.\n",
- "\n",
- "\n",
- "* [Task 1](#task1): Compile Flags \n",
- "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback ([Solution 1](#solution0))\n",
- "\n",
- "* [Task 2](#task2): Software Prefetching \n",
- "Improve performance of the CPU Jacobi solver with software prefetching ([Solution 2](#solution1))\n",
- "\n",
- "* [Task 3](#task3): OpenMP \n",
- "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance ([Solution 3](#solution2))\n",
- " \n",
- "* [Suvery](#survey) Please remember to take the survey !\n",
- " \n",
- "### Make Targets <a name=\"make\"></a>\n",
- "\n",
- "For all tasks we have defined the following make targets. \n",
- "\n",
- "* __poisson2d__: \n",
- " build `poisson2d` binary (default)\n",
- "* __run__: \n",
- " run `poisson2d` with default parameters\n"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "[Back to Top](#top)\n",
- "\n",
- "---"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "## Task 1: Compile Flags <a name=\"task1\"></a>\n",
- "\n",
- "\n",
- "### Overview\n",
- "\n",
- "The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver \n",
- "\n",
- "Your task is to:\n",
- "\n",
- "* Optimize performance with `-Ofast` flag\n",
- "* Optimize performance with profile directed feedback \n",
- "\n",
- "First, change the working directory to `Task1`."
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 1,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task1\n"
- ]
- }
- ],
- "source": [
- "%cd Task1"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "### Part A: `-Ofast` vs. `-O3`\n",
- "\n",
- "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. Right now, the Makefile specifies `-O3` as the optimization flag. Compile the code using `make` and run it with `make run` in the next two cells."
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 2,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
- ]
- }
- ],
- "source": [
- "!make"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 3,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <5033> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "1.13user 0.00system 0:01.15elapsed 97%CPU (0avgtext+0avgdata 10944maxresident)k\n",
- "2560inputs+0outputs (1major+264minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "You can use the GNU _perf_ tool to profile the application using the `perf` command (see below) and see the top time-consuming functions."
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 4,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "[ perf record: Woken up 1 times to write data ]\n",
- "[ perf record: Captured and wrote 0.172 MB perf.O3.data (4125 samples) ]\n",
- "# To display the perf.data header info, please use --header/--header-only options.\n",
- "#\n",
- "#\n",
- "# Total Lost Samples: 0\n",
- "#\n",
- "# Samples: 4K of event 'cycles:u'\n",
- "# Event count (approx.): 3867635297\n",
- "#\n",
- "# Overhead Command Shared Object Symbol \n",
- "# ........ ......... ................. ........................................\n",
- "#\n",
- " 72.02% poisson2d poisson2d [.] 00000040.plt_call.fmax@@GLIBC_2.17\n",
- " 10.16% poisson2d poisson2d [.] poisson2d_reference\n",
- " 9.99% poisson2d poisson2d [.] main\n",
- " 4.69% poisson2d libc-2.17.so [.] __memcpy_power7\n",
- " 2.23% poisson2d libm-2.17.so [.] __fmaxf\n",
- " 0.75% poisson2d libm-2.17.so [.] __exp_finite\n",
- " 0.07% poisson2d poisson2d [.] 00000040.plt_call.memcpy@@GLIBC_2.17\n",
- " 0.02% poisson2d poisson2d [.] check_results\n",
- " 0.02% poisson2d libm-2.17.so [.] __GI___exp\n",
- " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
- " 0.01% poisson2d [kernel.kallsyms] [k] arch_local_irq_restore\n",
- " 0.00% poisson2d ld-2.17.so [.] _dl_new_object\n",
- " 0.00% poisson2d ld-2.17.so [.] _start\n",
- "\n",
- "\n",
- "#\n",
- "# (Tip: Show user configuration overrides: perf config --user --list)\n",
- "#\n"
- ]
- }
- ],
- "source": [
- "# perf record creates a perf.data file \n",
- "!perf record -o perf.O3.data -e cycles ./poisson2d\n",
- "# perf report opens the perf.data file \n",
- "!perf report -i perf.O3.data | cat"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "**TASK**: Now change the optimization flag in the [Makefile](/edit/Task1/Makefile) to `-Ofast` and repeat the steps in the following cell. In case you follow along non-interactive, call `make` and `make run` in your shell. (If you are in the Jupyter Notebook, you can actually click the link of the [Makefile](/edit/Task1/Makefile). In other cases, use `vim` which is installed on Ascent.)"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 5,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
- ]
- }
- ],
- "source": [
- "!make"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 6,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <5034> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "0.51user 0.00system 0:00.52elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n",
- "256inputs+0outputs (0major+264minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 7,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "[ perf record: Woken up 1 times to write data ]\n",
- "[ perf record: Captured and wrote 0.086 MB perf.Ofast.data (1889 samples) ]\n",
- "# To display the perf.data header info, please use --header/--header-only options.\n",
- "#\n",
- "#\n",
- "# Total Lost Samples: 0\n",
- "#\n",
- "# Samples: 1K of event 'cycles:u'\n",
- "# Event count (approx.): 1765737747\n",
- "#\n",
- "# Overhead Command Shared Object Symbol \n",
- "# ........ ......... ............. .......................\n",
- "#\n",
- " 44.65% poisson2d poisson2d [.] main\n",
- " 43.84% poisson2d poisson2d [.] poisson2d_reference\n",
- " 10.28% poisson2d libc-2.17.so [.] __memcpy_power7\n",
- " 1.12% poisson2d libm-2.17.so [.] __exp_finite\n",
- " 0.05% poisson2d poisson2d [.] check_results\n",
- " 0.03% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
- " 0.02% poisson2d libc-2.17.so [.] __readdir64\n",
- " 0.01% poisson2d ld-2.17.so [.] _dl_new_object\n",
- " 0.00% poisson2d ld-2.17.so [.] _start\n",
- "\n",
- "\n",
- "#\n",
- "# (Tip: System-wide collection from all CPUs: perf record -a)\n",
- "#\n"
- ]
- }
- ],
- "source": [
- "# perf record creates a perf.data file \n",
- "!perf record -o perf.Ofast.data -e cycles ./poisson2d\n",
- "# perf report opens the perf.data file \n",
- "!perf report -i perf.Ofast.data | cat"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "If `perf` is unavailable to you on other machines, you can also study the disassembly with `objdump`: `objdump -lSd ./poisson2d > poisson2d.dis` (feel free to experiment with this in the Notebook as well, just prefix the command with a `!` to execute it.)"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "#### Interpretation\n",
- "\n",
- "Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with `-Ofast` which enables `\u2013ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "### Part B: Profile-directed Feedback\n",
- "\n",
- "For the first level of optimization we saw `Ofast` cut the execution time of the `O3` binary by almost half.\n",
- "\n",
- "We can optimize the performance further by using profile directed feedback optimization.\n",
- "\n",
- "To compile using profile directed feedback with the GCC compiler we need to do the following steps\n",
- "\n",
- "1. We need to first build a training binary using `-fprofile-generate`; this instructs the compiler to record hot path information \n",
- "2. Run the training binary with a smaller input size; you should see a `.gcda` file generated which stores hot path information for further optimization by the compiler \n",
- "3. build the final binary using `-fprofile-use` which uses the profile information in the `.gcda` file \n",
- "4. Compare the performance of the final binary with the original `Ofast` binary \n",
- "\n",
- "**TASK**: First, search for `TODO1` in the [Makefile](/edit/Task1/Makefile). It defines an additional compilation flag for `gcc`. Insert `-fprofile-generate=FOLDER` there with FOLDER pointing to `$$SC18_DIR_SCRATCH`, your personal write-directory (the double dollar signs are intentional as they are used to escape in the GNU Make syntax).\n",
- "\n",
- "After editing, run the following two cells to train your program."
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 8,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-generate=$SC18_DIR_SCRATCH\" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-generate=$SC18_DIR_SCRATCH\" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_train -lm\n"
- ]
- }
- ],
- "source": [
- "!make poisson2d_train"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 9,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_train 200 64 64\n",
- "Job <5035> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 200 iterations on 64 x 64 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.248743\n",
- " 100, 0.124046\n",
- "Calculate current execution.\n",
- " 0, 0.248743\n",
- " 100, 0.124046\n",
- "0.00user 0.00system 0:00.10elapsed 5%CPU (0avgtext+0avgdata 5248maxresident)k\n",
- "512inputs+0outputs (0major+115minor)pagefaults 0swaps\n",
- "mv $SC18_DIR_SCRATCH/*.gcda .\n"
- ]
- }
- ],
- "source": [
- "!make run_train"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "Now, a `.gcda` file exists in the directory which can be used for an profile-accelerated subsequent run.\n",
- "\n",
- "**TASK**: Edit the [Makefile](/edit/Task1/Makefile) again, this time modifying `TODO2` to be equivalent to `-fprofile-use`. A directory is not needed as we copied the gcda file into the current directory.\n",
- "\n",
- "Run the following cells in order to build using the newly added flag and then run with the profile-accelerated version."
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 10,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-use\" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-use\" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_profile -lm\n"
- ]
- }
- ],
- "source": [
- "!make poisson2d_profile"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 11,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_profile\n",
- "Job <5036> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "0.47user 0.00system 0:00.48elapsed 98%CPU (0avgtext+0avgdata 10816maxresident)k\n",
- "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run_profile"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "#### References\n",
- "\n",
- "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
- "2. https://perf.wiki.kernel.org/index.php/Tutorial"
- ]
- },
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Hands-On Performance Optimization\n",
+ "_Supercomputing 2019 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 18th 2019_\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.\n",
+ "\n",
+ "## Jupyter notebook execution\n",
+ "\n",
+ "When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the _edit_ menu above.\n",
+ "\n",
+ "You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n",
+ "\n",
+ "If you want you also can get a terminal in your browser; just open it via the \u00bbNew Launcher\u00ab button (`+`).\n",
+ "\n",
+ "## Terminal fallback\n",
+ "\n",
+ "The tasks are place in directories named `Task[1-3]`.\n",
+ "\n",
+ "Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Setup\n",
+ "\n",
+ "We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "[Back to Top](#top)\n",
- "\n",
- "---"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc (GCC) 9.2.0\n",
+ "Copyright (C) 2019 Free Software Foundation, Inc.\n",
+ "This is free software; see the source for copying conditions. There is NO\n",
+ "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!gcc --version"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Tasks<a name=\"top\"></a>\n",
+ "\n",
+ "This session comes with multiple tasks, each one to be found in the respective sub-directory `Task[1-3]`. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.\n",
+ "\n",
+ "Please choose from the task below.\n",
+ "\n",
+ "\n",
+ "* [Task 1](#task1): __Basic compiler optimization flags and compiler annotations__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback. Learn about compiler annotations.\n",
+ "\n",
+ "* [Task 2](#task2): __Optimization via Prefetching controlled by compiler__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with software prefetching. Some compilers such as IBM XL define flags that can be used to modify the aggressiveness of the hardware prefetcher. Learn to modify the DSCR value through XL and study the impact on application performance. \n",
+ "* [Task 3](#task3): __Optimization via OpenMP controlled by compiler and the system__\n",
+ "\n",
+ "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance. \n",
+ " \n",
+ "* [Suvery](#survey) Please remember to take the survey !\n",
+ " \n",
+ "### Make Targets <a name=\"make\"></a>\n",
+ "\n",
+ "For all tasks we have defined the following make targets. \n",
+ "\n",
+ "* __poisson2d__: \n",
+ " build `poisson2d` binary (default)\n",
+ "* __run__: \n",
+ " run `poisson2d` with default parameters\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 1: Basic compiler optimization flags and compiler annotations <a name=\"task1\"></a>\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver \n",
+ "\n",
+ "Your task is to:\n",
+ "\n",
+ "* Optimize performance with `-Ofast` flag\n",
+ "* Verify the cause for performance improvement by viewing perf profiles of O3 and Ofast binaries \n",
+ "* Optimize performance with profile directed feedback \n",
+ "* Generate compiler annotations/remarks to understand the optimizations done by the compiler with and without profile directed feedback \n",
+ "\n",
+ "\n",
+ "First, change the working directory to `Task1`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "## Task 2:<a name=\"task2\"></a> Software Pretechting\n",
- "\n",
- "\n",
- "### Overview\n",
- "\n",
- "Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
- "\n",
- "First, change directory to that of Task 2"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task1\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd Task1"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: `-Ofast` vs. `-O3`\n",
+ "\n",
+ "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. As in the previous task, we use a `Makefile` for compilation. The `Makefile` targets `poisson2d_O3` and `poisson2d_Ofast` are already prepared. \n",
+ "\n",
+ "**TASK**: Add `-O3` as the optimization flag for the `poisson2d_O3` target by using the corresponding `CFLAGS` definition. There are notes relating to this Task 1 in the header of the `Makefile`. Compile the code using `make` as indicated below and run with the `Make` targets `run`, `run_perf` and `run_perf_recrep`. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 84,
+ "metadata": {
+ "collapsed": true
+ },
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 12,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task2\n"
- ]
- }
- ],
- "source": [
- "%cd ../Task2"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_O3"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 73,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "### Part A: Running"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24897> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's have a look at the output of the `Makefile` target `run_perf`. It invokes the GNU _perf_ tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to _perf_."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 74,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "Look at the [Makefile](/edit/Task2/Makefile) and work on the TODOs. Please implement compile flags as mentioned in the Makefile target name.\n",
- "\n",
- "Afterwards, compile each target with the following cells and submit them to the batch system. Follow along accordingly in the non-interactive version of this Notebook."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24898> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16264721613 cycles:u \n",
+ " 28463907825 instructions:u # 1.75 insn per cycle \n",
+ "\n",
+ " 4.738444892 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Next we run the makefile with target `run_perf_recrep` that prints the top routines of the application in terms of hotness by using a combination of `perf record ./app` and `perf report`. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 75,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 13,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_pref -lm\n",
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_pref\n",
- "Job <5037> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "1.12user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10880maxresident)k\n",
- "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run_o3_pref"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24899> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "[ perf record: Woken up 3 times to write data ]\n",
+ "[ perf record: Captured and wrote 0.739 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (19102 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24900> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "# To display the perf.data header info, please use --header/--header-only options.\n",
+ "#\n",
+ "#\n",
+ "# Total Lost Samples: 0\n",
+ "#\n",
+ "# Samples: 19K of event 'cycles:u'\n",
+ "# Event count (approx.): 16254596654\n",
+ "#\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
+ "#\n",
+ " 65.50% poisson2d poisson2d [.] 00000038.plt_call.fmax@@GLIBC_2.17\n",
+ " 21.21% poisson2d poisson2d [.] main\n",
+ " 9.18% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 3.28% poisson2d libm-2.17.so [.] __fmaxf\n",
+ " 0.74% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.01% poisson2d libm-2.17.so [.] __GI___exp\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.00% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] _wordcopy_fwd_aligned\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ "\n",
+ "\n",
+ "#\n",
+ "# (Tip: Limit to show entries above 5% only: perf report --percent-limit 5)\n",
+ "#\n"
+ ]
+ }
+ ],
+ "source": [
+ "# run_perf_recrep displays the top hot routines \n",
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Now add the optimization flag `Ofast` to the `CFLAGS` for target `poisson2d_Ofast`. Compile the program with the target `poisson2d_Ofast` and run and analyse it as before with `run`, `run_perf` and `run_perf_recrep`.\n",
+ "\n",
+ "What difference do you see?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 76,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 14,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_pref -lm\n",
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_pref\n",
- "Job <5038> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "0.77user 0.00system 0:00.77elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n",
- "256inputs+0outputs (0major+264minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run_ofast_pref"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24901> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.41user 0.00system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast \n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Again, run a `perf`-instrumented version:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 77,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 15,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_nopref -lm\n",
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_nopref\n",
- "Job <5039> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "1.13user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10944maxresident)k\n",
- "256inputs+0outputs (0major+266minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run_o3_nopref"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24902> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 8258991976 cycles:u \n",
+ " 12013091172 instructions:u # 1.45 insn per cycle \n",
+ "\n",
+ " 2.408703909 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Generate the list of top routines in terms of hotness:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 78,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 16,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_nopref -lm\n",
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_nopref\n",
- "Job <5040> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "0.82user 0.00system 0:00.82elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n",
- "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run_ofast_nopref"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24903> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "[ perf record: Woken up 2 times to write data ]\n",
+ "[ perf record: Captured and wrote 0.382 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (9728 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24904> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "# To display the perf.data header info, please use --header/--header-only options.\n",
+ "#\n",
+ "#\n",
+ "# Total Lost Samples: 0\n",
+ "#\n",
+ "# Samples: 9K of event 'cycles:u'\n",
+ "# Event count (approx.): 8268811890\n",
+ "#\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
+ "#\n",
+ " 81.12% poisson2d poisson2d [.] main\n",
+ " 17.97% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 0.79% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.02% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.01% poisson2d libc-2.17.so [.] vfprintf@@GLIBC_2.17\n",
+ " 0.01% poisson2d libc-2.17.so [.] _dl_addr\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.01% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] open_path\n",
+ " 0.00% poisson2d ld-2.17.so [.] init_tls\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ "\n",
+ "\n",
+ "#\n",
+ "# (Tip: For tracepoint events, try: perf report -s trace_fields)\n",
+ "#\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If `perf` is unavailable to you on other machines, you can also study the disassembly with `objdump`: `objdump -lSd ./poisson2d > poisson2d.dis` (feel free to experiment with this in the Notebook as well, just prefix the command with a `!` to execute it.)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### Interpretation\n",
+ "\n",
+ "Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with `-Ofast` which enables `\u2013ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Profile-directed Feedback\n",
+ "\n",
+ "For the first level of optimization we see that `Ofast` cut the execution time of the `O3` binary by almost half.\n",
+ "\n",
+ "We can optimize the performance further by using profile-directed feedback optimization.\n",
+ "\n",
+ "To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:\n",
+ "\n",
+ "1. Instrument binary;\n",
+ "2. Run binary with training, gather profile information;\n",
+ "3. Use profile information to generate optimized binary.\n",
+ "\n",
+ "\n",
+ "Step 1 is achieved by compiling the binary with the correct flag \u2013\u00a0`-fprofile-generate`. In our case, we need to specify an output location, which should be `$(SC19_DIR_SCRATCH)`.\n",
+ "\n",
+ "Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension `.gcda`) is written to the directory specified during compilation.\n",
+ "\n",
+ "For Step 3, the binary is once again compiled, but this time using the `gcda` profile just generated. The according flag is `-fprofile-use`, which we set to `$(SC19_DIR_SCRATCH)` as well.\n",
+ "\n",
+ "In our `Makefile` at hand, we prepared the steps already for you in the form of two targets.\n",
+ "\n",
+ "* `poisson2d_train`: Will compile the binary with profile-directed feedback\n",
+ "* `poisson2d_ref`: Will take a generated profile and compile a new, optimized binary\n",
+ "\n",
+ "By using dependencies, between these two targets a profile run is launched.\n",
+ "\n",
+ "**TASK**: Edit the [Makefile](`Makefile`) and add the `-fprofile-*` flags to the `CFLAGS` of `poisson2d_train` and\n",
+ "`poisson2d_ref` as outline in the file.\n",
+ "\n",
+ "After that, you may launch them with the following cells (`gen_profile` is a meta-target and uses `poisson2d_train` and `poisson2d_ref`). If you need to clean the generated profile, you may use `make clean_profile`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 79,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "Do you notice the impact difference with optimization levels? It's always important to carefully study the interplay of flags."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24905> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_ref -lm \n",
+ "cp poisson2d_ref poisson2d\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make gen_profile"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 80,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "### Part B: Analysis of Instructions\n",
- "\n",
- "Compilation with the software prefetching flag causes the compiler to generate the `__dcbt` and `__dcbtst` instructions that prefetch memory values to L3.\n",
- "\n",
- "Verify it using `objdump -lSd` on each file (`poisson2d_o3_pref`, `poisson2d_ofast_pref`, `poisson2d_o3_nopref`, `poisson2d_ofast_nopref`). You might want to grep for `dcb`."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24906> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.28user 0.01system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's also measure instructions and cycles"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 81,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "exercise": "task"
- },
- "outputs": [],
- "source": [
- "#!objdump -l\u2026"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24907> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 7925983538 cycles:u \n",
+ " 12253080719 instructions:u # 1.55 insn per cycle \n",
+ "\n",
+ " 2.313471365 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Compiler annotations/Remarks\n",
+ "\n",
+ "Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done. \n",
+ "\n",
+ "To generate compiler annotations using GCC, one uses `-fopt-info-all`. If you only want to see the missed options, use the option `-fopt-info-missed` instead of `-fopt-info-all`. See also the [documentation of GCC regarding the flag](https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info).\n",
+ "\n",
+ "**TASK**: Have a looK at the `CFLAGS` of the `Makefile` target `poisson2d_Ofast_info`. Add the flag `-fopt-info-all` to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use \u2013\u00a0for example \u2013\u00a0`-fopt-info-all=(SC19_DIR_SCRATCH)/filename`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 82,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "If you feel up to the task, you can study the number of L3 cache misses using the corresponding performance counter, `PM_L3_MISS`. Either use your knowledge from Hands-On 1, or use the following call to `perf`, in which we already converted the named counter to a raw counter address."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all poisson2d.c poisson2d_reference.o -o poisson2d_Ofast_info -lm\n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 207->207 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.\n",
+ "\n",
+ "**TASK**: \n",
+ "Adapt the `CFLAGS` of `poisson2d_ref_info` to include `-fopt-info-all` **and** the profile input of `-fprofile-use=\u2026` here. *(Be advised: Long output!)*"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 83,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 35,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Job <5048> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "\n",
- " Performance counter stats for './poisson2d_ofast_nopref':\n",
- "\n",
- " 2829292169 cycles:u \n",
- " 136018637 r168a4:u \n",
- "\n",
- " 0.826136863 seconds time elapsed\n",
- "\n",
- "Job <5049> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "\n",
- " Performance counter stats for './poisson2d_ofast_pref':\n",
- "\n",
- " 2654990243 cycles:u \n",
- " 128824827 r168a4:u \n",
- "\n",
- " 0.775593651 seconds time elapsed\n",
- "\n"
- ]
- }
- ],
- "source": [
- "for f in [\"poisson2d_ofast_nopref\", \"poisson2d_ofast_pref\"]:\n",
- " !eval $$SC18_SUBMIT_CMD perf stat -e cycles,r168a4 ./$f\n"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "Increasing alignment of decl: __gcov0.main\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_D_00100_1_main/48 -> __gcov_exit/55, function body not available\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_I_00100_0_main/47 -> __gcov_init/54, function body not available\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 295->295 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:122:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:88:5: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:72:5: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_337, ny_124, nx_286);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_316, error_118);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_127);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_311);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_122);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_129 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_132 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_140 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 53\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 50\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 47\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 3 times (header execution count 9800)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 33\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 30\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 42\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 40\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 60\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 23\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 3 times (header execution count 100)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 12\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 16\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_init (&*.LPBX0);\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_exit ();\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24908> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "libgcov profiling error:/gpfs/wolf/trn003/scratch/aherten//#autofs#nccsopen-svm1_home#aherten#SC19-Tutorial#3-Optimizing_POWER#Handson#Task1#poisson2d.gcda:overwriting an existing profile data with a different timestamp\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c poisson2d_reference.o -o poisson2d_ref_info -lm\n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 207->207 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 47\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 44\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 40\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 27\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 24\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 37\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 35\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 51\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 18\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 7 times (header execution count 99)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 9\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 14\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_ref_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Comparing the annotations generated of a plain `-Ofast` optimization level and the one generated at `-Ofast` and profile directed feedback, we observe that many more optimizations are possible due to profile information.\n",
+ "\n",
+ "For instance you will see annotations such as\n",
+ "```\n",
+ "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "```\n",
+ "\n",
+ "The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "\n",
+ "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
+ "2. https://perf.wiki.kernel.org/index.php/Tutorial"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 2:<a name=\"task2\"></a> Impact of Prefetching on Performance\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "* Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
+ "* Verify the impact by measuring cache counters with and without prefetching.\n",
+ "* Learn how to modify contents of DSCR (*Data Stream Control Register*) using IBM XL compiler and study the impact with different values to DSCR. \n",
+ "\n",
+ "But first, lets change directory to that of Task 2"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 85,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "#### References\n",
- "\n",
- "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
- "2. https://www.gnu.org/software/gcc/projects/prefetch.html"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task2\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd ../Task2"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: Software Prefetching"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Look at the Makefile and work on the TODOs. \n",
+ "\n",
+ "- First generate a `-Ofast`-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!\n",
+ "- Modify the `Makefile` to add the option for software prefetching (`-fprefetch-loop-arrays`). Compare performance of `-Ofast` with and without software prefetching"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 97,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "[Back to Top](#top)\n",
- "\n",
- "---"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "rm -f poisson2d poisson2d*.o\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make clean"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 88,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "## Task 3: OpenMP\n",
- "<a name=\"task3\"></a>\n",
- "\n",
- "\n",
- "### Overview\n",
- "\n",
- "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores.\n",
- "\n",
- "First, we need to change directory to that of Task3."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "make: `poisson2d' is up to date.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24911> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.39user 0.01system 0:02.40elapsed 100%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24912> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 8271503902 cycles:u \n",
+ " 481152478 r168a4:u \n",
+ "\n",
+ " 2.412224884 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 98,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 1,
- "metadata": {
- "ExecuteTime": {
- "end_time": "2018-11-07T13:47:57.724441Z",
- "start_time": "2018-11-07T13:47:57.718745Z"
- }
- },
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task3\n"
- ]
- }
- ],
- "source": [
- "%cd ../Task3"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ "cp poisson2d_pref poisson2d\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24919> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1.92user 0.00system 0:01.93elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24920> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 6586609284 cycles:u \n",
+ " 459879452 r168a4:u \n",
+ "\n",
+ " 1.925399505 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_pref CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Repeat the experiment with the `-O3` flag. Have a look at the `Makefile` and the outlined TODO. There's a position to easily adapt `-Ofast`\u2192`-O3`!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 100,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "### Part A: Implement OpenMP Pragmas; Compilation\n",
- "\n",
- "**Task**: Please add the correct OpenMP pragmas to the source code and compilations flags to enable OpenMP.\n",
- "\n",
- "* **pragmas**: Look at the TODOs in [`poisson2d.c`](/edit/Task3/poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for`\n",
- "* **Compilation**: Please add compilation flags enabling OpenMP in GCC to the [Makefile](/edit/Task3/Makefile). The flag in question is `-fopenmp`.\n",
- "\n",
- "Edit the files with the links above if you are running the interactive version of the Notebook or navigate to `poisson2d.c` and `Makefile` yourself in case you run the non-interactive version.\n",
- "\n",
- "Afterwards, compile and run the application with the following cells. Non-interactive: Follow along accordingly in the shell."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec poisson2d.c -o poisson2d -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24923> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24924> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16445764669 cycles:u \n",
+ " 645094089 r168a4:u \n",
+ "\n",
+ " 4.792567763 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 101,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 37,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec poisson2d_reference.c -o poisson2d_reference.o -lm\n",
- "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
- ]
- }
- ],
- "source": [
- "!make poisson2d"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24925> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.74user 0.00system 0:04.74elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24926> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16239159454 cycles:u \n",
+ " 631061431 r168a4:u \n",
+ "\n",
+ " 4.730144897 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_pref CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Analysis of Instructions\n",
+ "\n",
+ "Compilation of the `-Ofast` binary with the software prefetching flag causes the compiler to generate the `dcb*` instructions that prefetch memory values to L3."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: \n",
+ "Run `$(SC19_SUBMIT_CMD) objdump -lSd` on each binary file (`-O3`, `-Ofast` with prefetch/no prefetch).\n",
+ "Look for instructions beginning with `dcb`\n",
+ "At what optimization levels does the compiler generate software prefetching instructions?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 114,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 40,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <5052> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "Calculate current execution.\n",
- " 0, 0.249980\n",
- " 100, 0.246028\n",
- " 200, 0.242198\n",
- " 300, 0.238487\n",
- " 400, 0.234887\n",
- "500x500: Ref: 0.2571 s, This: 0.2946 s, speedup: 0.87\n",
- "1.48user 0.00system 0:00.56elapsed 263%CPU (0avgtext+0avgdata 9664maxresident)k\n",
- "0inputs+0outputs (0major+273minor)pagefaults 0swaps\n"
- ]
- }
- ],
- "source": [
- "!make run"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=gcc -B poisson2d_pref\n",
+ "!objdump -lSd ./poisson2d_pref > poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 116,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "The command to submit a job to the batch system is prepared in an environment variable `$SC18_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to increase the work of the application."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ " 10000b28:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000b30:\t2c ba 00 7c \tdcbt 0,r23\n",
+ " 10000b38:\t2c b2 00 7c \tdcbt 0,r22\n",
+ " 10000b50:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000b58:\tec b9 00 7c \tdcbtst 0,r23\n",
+ " 10000b80:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000e64:\t2c 92 00 7c \tdcbt 0,r18\n",
+ " 10000e68:\t2c 9a 00 7c \tdcbt 0,r19\n",
+ " 10000e6c:\t2c a2 00 7c \tdcbt 0,r20\n",
+ " 10000e70:\t2c aa 00 7c \tdcbt 0,r21\n",
+ " 10000e7c:\t2c b2 00 7c \tdcbt 0,r22\n",
+ " 10000e80:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000e94:\tec b9 00 7c \tdcbtst 0,r23\n"
+ ]
+ }
+ ],
+ "source": [
+ "!grep dcb poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Changing Values of DSCR via compiler flags\n",
+ "\n",
+ "This task requires using the IBM XL compiler. It should be already in your environment.\n",
+ "\n",
+ "\n",
+ "We saw the impact of software prefetching in the previous subsection. \n",
+ "In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance. \n",
+ "In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching. \n",
+ "\n",
+ "IBM XL compiler has an option `-qprefetch=dscr=<val>` that can be used for this purpose.\n",
+ "Compiling with `-qprefetch=dscr=1` turns off the prefetcher. One can give various values such as `-qprefetch=dscr=4`, `-qprefetch=dscr=7` etc. to control aggressiveness of prefetching.\n",
+ "\n",
+ "For this exercise we use `make CC=xlc_r` to illustrate the performance impact.\n",
+ " \n",
+ "\n",
+ "**Task** Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the `Makefile`: Add `qprefetch=dscr=1` to the `CFLAGS` and rebuild the application and note the performance. Which one is faster? \n",
+ "\n",
+ "In general, applications benefit with the default settings of hardware DSCR register (`-qprefetch=dscr=0`). However, certain applications also benefit with prefetching turned off. \n",
+ "\n",
+ "It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL at default DSCR value"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 117,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 3,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Job <5344> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 1000 iterations on 1000 x 100 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249743\n",
- " 100, 0.210080\n",
- " 200, 0.184635\n",
- " 300, 0.166526\n",
- " 400, 0.152783\n",
- " 500, 0.141890\n",
- " 600, 0.132978\n",
- " 700, 0.125511\n",
- " 800, 0.119142\n",
- " 900, 0.113632\n",
- "Calculate current execution.\n",
- " 0, 0.249743\n",
- " 100, 0.210080\n",
- " 200, 0.184635\n",
- " 300, 0.166526\n",
- " 400, 0.152783\n",
- " 500, 0.141890\n",
- " 600, 0.132978\n",
- " 700, 0.125511\n",
- " 800, 0.119142\n",
- " 900, 0.113632\n",
- "1000x100: Ref: 1.9872 s, This: 0.2385 s, speedup: 8.33\n"
- ]
- }
- ],
- "source": [
- "!eval $SC18_SUBMIT_CMD ./poisson2d 1000 1000"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS poisson2d.c -o poisson2d -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24927> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "2.26user 0.00system 0:02.27elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+477minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=xlc_r -B poisson2d\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL with DSCR value turned off"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n",
- "We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS -qprefetch=dscr=1 poisson2d.c -o poisson2d_dscr -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24929> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.58user 0.00system 0:04.59elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "0inputs+0outputs (0major+476minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_dscr CC=xlc_r -B\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Does Hardware prefetcher help this application? How much impact do you see when you turn off the hardware prefetcher? "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "\n",
+ "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
+ "2. https://www.gnu.org/software/gcc/projects/prefetch.html\n",
+ "3. https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 3: OpenMP\n",
+ "<a name=\"task3\"></a>\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores on the resulting application performance. We do this study for both GCC and XL compilers inorder to learn about the appropriate options that need to be used.\n",
+ "First, we need to change directory to that of Task3. For Task 3 we modify poisson2d.c to invoke an exact copy of the main jacobi loop which is `poisson2d_reference`. We parallelize only the main loop but not `poisson2d_reference`. The speedup is the performance gain seen in the main loop as compared to the reference loop."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 22,
- "metadata": {
- "exercise": "task"
- },
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Job <5379> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249995\n",
- " 100, 0.248997\n",
- " 200, 0.248007\n",
- " 300, 0.247025\n",
- " 400, 0.246050\n",
- " 500, 0.245084\n",
- " 600, 0.244124\n",
- " 700, 0.243173\n",
- " 800, 0.242228\n",
- " 900, 0.241291\n",
- "Calculate current execution.\n",
- " 0, 0.249995\n",
- " 100, 0.248997\n",
- " 200, 0.248007\n",
- " 300, 0.247025\n",
- " 400, 0.246050\n",
- " 500, 0.245084\n",
- " 600, 0.244124\n",
- " 700, 0.243173\n",
- " 800, 0.242228\n",
- " 900, 0.241291\n",
- "1000x1000: Ref: 2.3303 s, This: 2.8446 s, speedup: 0.82\n"
- ]
- }
- ],
- "source": [
- "!eval OMP_NUM_THREADS=1 $SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task3\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd ../Task3"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: Implement OpenMP Pragmas; Compilation\n",
+ "\n",
+ "**Task**: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.\n",
+ "\n",
+ "* **Directives**: Look at the TODOs in [`poisson2d.c`](poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for` (and once it's `#pragma omp parallel for reduction(max:error)` \u2013\u00a0can you guess where?)\n",
+ "* **Compilation**: Please add compilation flags enabling OpenMP in GCC and XL to the `Makefile`. For GCC, we need to add `-fopenmp` and the application needs to be linked with `-lgomp`. For XL, we need to add `-qsmp=omp` to the list of compilation flags. \n",
+ "\n",
+ "Afterwards, compile and run the application with the following commands."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 39,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "### Part B: Bindings\n",
- "\n",
- "Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!\n",
- "\n",
- "There are applications which can be used to determine the configuration of the processor. Among those are:\n",
- "\n",
- "* `lscpu`: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.\n",
- "* `ppc64_cpu --smt`: Specifically for POWER, this tool can give information about the number of simulations threads running per core (*SMT*, Simulataion Multi-Threading).\n",
- "\n",
- "Run `ppc64_cpu --smt` to find out about the threading configuration of Ascent!"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -c -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d.c poisson2d_reference.o -o poisson2d -lm \n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "The command to submit a job to the batch system is prepared in an environment variable `$SC19_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to invoke the application using the batch system. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 40,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 48,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Job <5076> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "SMT=4\n"
- ]
- }
- ],
- "source": [
- "!eval $SC18_SUBMIT_CMD ppc64_cpu --smt"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24951> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1000x1000: Ref: 4.7430 s, This: 3.9363 s, speedup: 1.20\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval $SC19_SUBMIT_CMD ./poisson2d 1000 1000 1000"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Inorder to run the parallel application, we need to set the number of threads using `OMP_NUM_THREADS`\n",
+ "What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n",
+ "We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 32,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "There are more sources information available\n",
- "\n",
- "* `/proc/cpuinfo`: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with `cat`\n",
- "* `/sys/devices/system/cpu/cpu0/topology/thread_siblings_list`: Holds information about thread siblings for given CPU core (`cpu0` in this case). Use it to find out which thread is mapped to which core."
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24945> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "\n",
+ "libgomp: Invalid value for environment variable OMP_NUM_THREADS\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1000x1000: Ref: 2.1046 s, This: 2.4171 s, speedup: 0.87\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval OMP_NUM_THREADS=N $SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Bindings\n",
+ "\n",
+ "Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!\n",
+ "\n",
+ "There are applications which can be used to determine the configuration of the processor. Among those are:\n",
+ "\n",
+ "* `lscpu`: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.\n",
+ "* `ppc64_cpu --smt`: Specifically for POWER, this tool can give information about the number of simulations threads running per core (*SMT*, Simulataion Multi-Threading).\n",
+ "\n",
+ "Run `ppc64_cpu --smt` to find out about the threading configuration of Ascent!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 55,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": 49,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "0-3\n",
- "4-7\n"
- ]
- }
- ],
- "source": [
- "!cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n",
- "!cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24465> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "SMT=4\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval $SC19_SUBMIT_CMD ppc64_cpu --smt"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "There are more sources information available\n",
+ "\n",
+ "* `/proc/cpuinfo`: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with `cat`\n",
+ "* `/sys/devices/system/cpu/cpu0/topology/thread_siblings_list`: Holds information about thread siblings for given CPU core (`cpu0` in this case). Use it to find out which thread is mapped to which core."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 36,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "There are various environment variables available within OpenMP (and GCC) to specify binding of threads to cores. See, for instance, the [online documentation of GCC libgomp](https://gcc.gnu.org/onlinedocs/libgomp/Environment-Variables.html). Examples are `OMP_PLACES` or `GOMP_CPU_AFFINITY`.\n",
- "\n",
- "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
- "\n",
- "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
- "\n",
- "What's your maximum speedup?"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24949> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "0-3\n",
+ "Job <24950> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "4-7\n"
+ ]
+ }
+ ],
+ "source": [
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n",
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the [OMP_PLACES environment Variable](https://www.openmp.org/spec-html/5.0/openmpse53.html). We also have a GNU specific variable which can also be used to control affinity - `GOMP_CPU_AFFINITY`. Setting `GOMP_CPU_AFFINITY` is specific to GCC binaries but it internally serves the same function as setting `OMP_PLACES`. \n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 38,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [
{
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "exercise": "task"
- },
- "outputs": [],
- "source": [
- "!eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=\"X,Y,Z,A\" OMP_NUM_THREADS=4 $$SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/usr/bin/sh: OMP_PLACES}={X},{Y},{Z},{A}: command not found\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval OMP_DISPLAY_ENV=true OMP_PLACES=\"{X},{Y},{Z},{A}\" OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "#### References\n",
- "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=\"X,Y,Z,A\" OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Great!\n",
+ "\n",
+ "If you still have time: The same experiments can be repeated with the IBM XL compiler. \n",
+ "The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is `-qsmp=omp`\n",
+ "\n",
+ "**Task**: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 44,
+ "metadata": {},
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "[Back to Top](#top)\n",
- "\n",
- "---"
- ]
- },
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -c -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d_reference.c -o poisson2d_reference.o -lm \n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "xlc_r -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24956> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "1000x1000: Ref: 5.6783 s, This: 2.6528 s, speedup: 2.14\n",
+ "21.56user 6.18system 0:08.37elapsed 331%CPU (0avgtext+0avgdata 23040maxresident)k\n",
+ "3200inputs+0outputs (2major+1098minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=xlc_r -B run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Run the parallel application with varying numbre of threads (`OMP_NUM_THREADS`) and note the performance improvement. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 45,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [
{
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "# Survey<a name=\"survey\"></a>\n",
- "\n",
- "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc18-eval)."
- ]
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <23926> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1000x1000: Ref: 2.3932 s, This: 2.7175 s, speedup: 0.88\n"
+ ]
}
- ],
- "metadata": {
- "kernelspec": {
- "display_name": "Python 3",
- "language": "python",
- "name": "python3"
- },
- "language_info": {
- "codemirror_mode": {
- "name": "ipython",
- "version": 3
- },
- "file_extension": ".py",
- "mimetype": "text/x-python",
- "name": "python",
- "nbconvert_exporter": "python",
- "pygments_lexer": "ipython3",
- "version": "3.6.7"
+ ],
+ "source": [
+ "!eval OMP_NUM_THREADS=N $SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Now we repeat the exercise of using the right binding of threads for the XL binary. `OMP_PLACES` pertains to the XL binary as well as it is an OpenMP variable. `GOMP_CPU_AFFINITY` is specific to GCC binary so that cannot be used to set the binding.\n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "We are mixing Python with Bash (`!`) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two `$$`)\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: {X},{Y},{Z},{A}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1587-117 The string for the OpenMP environment variable 'OMP_PLACES' contains unexpected or invalid text. OpenMP environment variable ignored. \n",
+ " OMP_PLACES='cores(44)'\n",
+ "1000x1000: Ref: 2.0988 s, This: 0.6556 s, speedup: 3.20\n",
+ "Affinity: {P},{Q},{R},{S}\n",
+ "<<Waiting for dispatch ...>>\n"
+ ]
}
+ ],
+ "source": [
+ "for affinity in [\"{X},{Y},{Z},{A}\", \"{P},{Q},{R},{S}\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true OMP_PLACES=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Likewise we see a higher speedup when we bind the threads to different cores rather than to a single core. This handson illustrates that apart from compiler level tuning, system level tuning is also equally important to obtain performance improvements \n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html\n",
+ "2. https://www.openmp.org/spec-html/5.0/openmpse53.html"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Survey<a name=\"survey\"></a>\n",
+ "\n",
+ "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc19-eval)."
+ ]
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3",
+ "language": "python",
+ "name": "python3"
},
- "nbformat": 4,
- "nbformat_minor": 2
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.7.0"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
}
\ No newline at end of file
diff --git a/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization.ipynb b/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization.ipynb
index 673f2cf..8453a7b 100644
--- a/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization.ipynb
+++ b/3-Optimizing_POWER/Handson/.master/HandsOnPerformanceOptimization.ipynb
@@ -22,7 +22,7 @@
"\n",
"You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n",
"\n",
- "If you want you also can get a [terminal](/terminals/1) in your browser.\n",
+ "If you want you also can get a terminal in your browser; just open it via the »New Launcher« button (`+`).\n",
"\n",
"## Terminal fallback\n",
"\n",
@@ -37,7 +37,28 @@
"source": [
"## Setup\n",
"\n",
- "This hands-on session requires use of GCC 9.2.0. By loading the `sc19/handson2` module before invoking this Notebook, we took care of also loading GCC 9.2.0."
+ "We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc (GCC) 9.2.0\n",
+ "Copyright (C) 2019 Free Software Foundation, Inc.\n",
+ "This is free software; see the source for copying conditions. There is NO\n",
+ "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!gcc --version"
]
},
{
@@ -107,14 +128,14 @@
},
{
"cell_type": "code",
- "execution_count": 70,
+ "execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "/autofs/nccsopen-svm1_home/archanaravindar/SC19-Tutorial/Task1/Solutions\n"
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task1\n"
]
}
],
@@ -128,12 +149,14 @@
"source": [
"### Part A: `-Ofast` vs. `-O3`\n",
"\n",
- "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. Use `Makefile` for this task. At present, Makefile specifies targets poisson2d_O3 and poisson2d_Ofast. Add `-O3` as the optimization flag for poisson2d_O3 target. Carry out the steps as outlined in `Task1`. Compile the code using `make -f Makefile.gcc` and run with targets 'run', 'runstats' and 'perfgenerate'. "
+ "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. As in the previous task, we use a `Makefile` for compilation. The `Makefile` targets `poisson2d_O3` and `poisson2d_Ofast` are already prepared. \n",
+ "\n",
+ "**TASK**: Add `-O3` as the optimization flag for the `poisson2d_O3` target by using the corresponding `CFLAGS` definition. There are notes relating to this Task 1 in the header of the `Makefile`. Compile the code using `make` as indicated below and run with the `Make` targets `run`, `run_perf` and `run_perf_recrep`. "
]
},
{
"cell_type": "code",
- "execution_count": 52,
+ "execution_count": 84,
"metadata": {
"collapsed": true
},
@@ -142,8 +165,8 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec poisson2d.c poisson2d_reference.o -o poisson2d_O3 -lm\n",
- "cp poisson2d_O3 poisson2d\n"
+ "gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
]
}
],
@@ -153,7 +176,7 @@
},
{
"cell_type": "code",
- "execution_count": 53,
+ "execution_count": 73,
"metadata": {},
"outputs": [
{
@@ -161,7 +184,7 @@
"output_type": "stream",
"text": [
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24738> is submitted to default queue <batch>.\n",
+ "Job <24897> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -176,8 +199,8 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "4.70user 0.00system 0:04.71elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
- "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n"
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
]
}
],
@@ -189,12 +212,12 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "The makefile has been modified to include options 'runstats' that invokes the GNU _perf_ tool to print out details of the number of PPC instructions executed and the number of cycles taken by POWER9 to execute the program. You may wish to add a similar clause to measure other raw events using _perf_."
+ "Let's have a look at the output of the `Makefile` target `run_perf`. It invokes the GNU _perf_ tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to _perf_."
]
},
{
"cell_type": "code",
- "execution_count": 55,
+ "execution_count": 74,
"metadata": {},
"outputs": [
{
@@ -202,7 +225,7 @@
"output_type": "stream",
"text": [
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
- "Job <24739> is submitted to default queue <batch>.\n",
+ "Job <24898> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -220,37 +243,36 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 16199088860 cycles:u \n",
- " 28463939531 instructions:u # 1.76 insn per cycle \n",
+ " 16264721613 cycles:u \n",
+ " 28463907825 instructions:u # 1.75 insn per cycle \n",
"\n",
- " 4.717812694 seconds time elapsed\n",
+ " 4.738444892 seconds time elapsed\n",
"\n"
]
}
],
"source": [
- "!make runstats"
+ "!make run_perf"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "The makefile has been modified to include options 'runstats' that invokes the GNU _perf_ tool to print out details of the number of PPC instructions executed and the number of cycles taken by POWER9 to execute the program. You may wish to add a similar clause to measure other raw events using _perf_.\n",
- "Next we run the makefile with target `perfgenerate` that prints the top routines of the application in terms of hotness. "
+ "Next we run the makefile with target `run_perf_recrep` that prints the top routines of the application in terms of hotness by using a combination of `perf record ./app` and `perf report`. "
]
},
{
"cell_type": "code",
- "execution_count": 56,
+ "execution_count": 75,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/archanaravindar//cycles.data ./poisson2d\n",
- "Job <24740> is submitted to default queue <batch>.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24899> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -266,9 +288,9 @@
" 800, 0.242228\n",
" 900, 0.241291\n",
"[ perf record: Woken up 3 times to write data ]\n",
- "[ perf record: Captured and wrote 0.735 MB /gpfs/wolf/trn003/scratch/archanaravindar//cycles.data (18993 samples) ]\n",
- "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/archanaravindar//cycles.data --stdio\n",
- "Job <24741> is submitted to default queue <batch>.\n",
+ "[ perf record: Captured and wrote 0.739 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (19102 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24900> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"# To display the perf.data header info, please use --header/--header-only options.\n",
@@ -276,66 +298,61 @@
"#\n",
"# Total Lost Samples: 0\n",
"#\n",
- "# Samples: 18K of event 'cycles:u'\n",
- "# Event count (approx.): 16162306943\n",
+ "# Samples: 19K of event 'cycles:u'\n",
+ "# Event count (approx.): 16254596654\n",
"#\n",
- "# Overhead Command Shared Object Symbol \n",
- "# ........ ......... ................ ........................................\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
"#\n",
- " 48.17% poisson2d poisson2d [.] main\n",
- " 26.08% poisson2d poisson2d [.] 00000038.plt_call.fmax@@GLIBC_2.17\n",
- " 15.84% poisson2d libm-2.17.so [.] __fmaxf\n",
- " 9.13% poisson2d libc-2.17.so [.] __memcpy_power7\n",
- " 0.72% poisson2d libm-2.17.so [.] __exp_finite\n",
- " 0.01% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
- " 0.01% poisson2d libm-2.17.so [.] __GI___exp\n",
- " 0.01% poisson2d ld-2.17.so [.] do_lookup_x\n",
- " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
- " 0.01% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
- " 0.01% poisson2d ld-2.17.so [.] strcmp\n",
- " 0.00% poisson2d [unknown] [k] 0x000020000002415c\n",
- " 0.00% poisson2d ld-2.17.so [.] dl_main\n",
- " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ " 65.50% poisson2d poisson2d [.] 00000038.plt_call.fmax@@GLIBC_2.17\n",
+ " 21.21% poisson2d poisson2d [.] main\n",
+ " 9.18% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 3.28% poisson2d libm-2.17.so [.] __fmaxf\n",
+ " 0.74% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.01% poisson2d libm-2.17.so [.] __GI___exp\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.00% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] _wordcopy_fwd_aligned\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
"\n",
"\n",
"#\n",
- "# (Tip: Show individual samples with: perf script)\n",
+ "# (Tip: Limit to show entries above 5% only: perf report --percent-limit 5)\n",
"#\n"
]
}
],
"source": [
- "# perfgenerate displays the top hot routines \n",
- "!make perfgenerate"
+ "# run_perf_recrep displays the top hot routines \n",
+ "!make run_perf_recrep"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "**TASK**: Now Add the optimization flag `Ofast` in Makefile for target poisson2d_Ofast. Run the makefile with the targets clean, run, runstats and perfgenerate.\n",
- "\n",
- "Compare performance of O3 and Ofast binaries using the following makefile options- run, runstats. \n",
- "\n",
- "Verify by generating the profiles of both O3 and Ofast binaries to understand the cause for performance improvement. \n",
+ "**TASK**: Now add the optimization flag `Ofast` to the `CFLAGS` for target `poisson2d_Ofast`. Compile the program with the target `poisson2d_Ofast` and run and analyse it as before with `run`, `run_perf` and `run_perf_recrep`.\n",
"\n",
- "What difference do you see? \n",
- "\n"
+ "What difference do you see?"
]
},
{
"cell_type": "code",
- "execution_count": 58,
+ "execution_count": 76,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec poisson2d.c poisson2d_reference.o -o poisson2d_Ofast -lm\n",
- "cp poisson2d_Ofast poisson2d\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24743> is submitted to default queue <batch>.\n",
+ "Job <24901> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -350,7 +367,7 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "2.40user 0.00system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "2.41user 0.00system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
"256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
]
}
@@ -362,16 +379,14 @@
},
{
"cell_type": "markdown",
- "metadata": {
- "exercise": "Solutions"
- },
+ "metadata": {},
"source": [
- "Measure cycles, instructions."
+ "Again, run a `perf`-instrumented version:"
]
},
{
"cell_type": "code",
- "execution_count": 59,
+ "execution_count": 77,
"metadata": {},
"outputs": [
{
@@ -379,7 +394,7 @@
"output_type": "stream",
"text": [
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
- "Job <24744> is submitted to default queue <batch>.\n",
+ "Job <24902> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -397,38 +412,36 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 8261506450 cycles:u \n",
- " 12013095395 instructions:u # 1.45 insn per cycle \n",
+ " 8258991976 cycles:u \n",
+ " 12013091172 instructions:u # 1.45 insn per cycle \n",
"\n",
- " 2.413121525 seconds time elapsed\n",
+ " 2.408703909 seconds time elapsed\n",
"\n"
]
}
],
"source": [
- "!make runstats"
+ "!make run_perf"
]
},
{
"cell_type": "markdown",
- "metadata": {
- "exercise": "Solutions"
- },
+ "metadata": {},
"source": [
- "Generate the list of Top routines in terms of hotness."
+ "Generate the list of top routines in terms of hotness:"
]
},
{
"cell_type": "code",
- "execution_count": 60,
+ "execution_count": 78,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/archanaravindar//cycles.data ./poisson2d\n",
- "Job <24745> is submitted to default queue <batch>.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24903> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -444,9 +457,9 @@
" 800, 0.242228\n",
" 900, 0.241291\n",
"[ perf record: Woken up 2 times to write data ]\n",
- "[ perf record: Captured and wrote 0.382 MB /gpfs/wolf/trn003/scratch/archanaravindar//cycles.data (9722 samples) ]\n",
- "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/archanaravindar//cycles.data --stdio\n",
- "Job <24746> is submitted to default queue <batch>.\n",
+ "[ perf record: Captured and wrote 0.382 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (9728 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24904> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"# To display the perf.data header info, please use --header/--header-only options.\n",
@@ -455,36 +468,36 @@
"# Total Lost Samples: 0\n",
"#\n",
"# Samples: 9K of event 'cycles:u'\n",
- "# Event count (approx.): 8264083365\n",
+ "# Event count (approx.): 8268811890\n",
"#\n",
"# Overhead Command Shared Object Symbol \n",
"# ........ ......... ............. ........................................\n",
"#\n",
- " 81.24% poisson2d poisson2d [.] main\n",
- " 17.83% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 81.12% poisson2d poisson2d [.] main\n",
+ " 17.97% poisson2d libc-2.17.so [.] __memcpy_power7\n",
" 0.79% poisson2d libm-2.17.so [.] __exp_finite\n",
- " 0.06% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
- " 0.02% poisson2d ld-2.17.so [.] check_match.10253\n",
- " 0.02% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
- " 0.01% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.02% poisson2d ld-2.17.so [.] do_lookup_x\n",
" 0.01% poisson2d libc-2.17.so [.] vfprintf@@GLIBC_2.17\n",
- " 0.01% poisson2d libc-2.17.so [.] __memset_power8\n",
" 0.01% poisson2d libc-2.17.so [.] _dl_addr\n",
- " 0.01% poisson2d ld-2.17.so [.] do_lookup_x\n",
- " 0.00% poisson2d ld-2.17.so [.] open_verify\n",
- " 0.00% poisson2d ld-2.17.so [.] strlen\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.01% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] open_path\n",
+ " 0.00% poisson2d ld-2.17.so [.] init_tls\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
" 0.00% poisson2d ld-2.17.so [.] _start\n",
"\n",
"\n",
"#\n",
- "# (Tip: Profiling branch (mis)predictions with: perf record -b / perf report)\n",
+ "# (Tip: For tracepoint events, try: perf report -s trace_fields)\n",
"#\n"
]
}
],
"source": [
- "# perfgenerate creates a perf.data file \n",
- "!make perfgenerate"
+ "!make run_perf_recrep"
]
},
{
@@ -500,7 +513,7 @@
"source": [
"#### Interpretation\n",
"\n",
- "Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with `-Ofast` which enables `–ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
+ "Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with `-Ofast` which enables `–ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
]
},
{
@@ -511,66 +524,71 @@
"\n",
"For the first level of optimization we see that `Ofast` cut the execution time of the `O3` binary by almost half.\n",
"\n",
- "We can optimize the performance further by using profile directed feedback optimization.\n",
+ "We can optimize the performance further by using profile-directed feedback optimization.\n",
+ "\n",
+ "To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:\n",
+ "\n",
+ "1. Instrument binary;\n",
+ "2. Run binary with training, gather profile information;\n",
+ "3. Use profile information to generate optimized binary.\n",
"\n",
- "To compile using profile directed feedback with the GCC compiler we need to build the appplication in three stages- Instrument binary; Run binary with training (smaller input) and gather profile information; Use profile information to generate optimized binary. For this purpose we have defined makefile target runpdf that depends on two binaries- poisson2d_train and poisson2d_ref.\n",
"\n",
- "make runpdf will inturn make poisson2d_train and poisson2d_ref\n",
+ "Step 1 is achieved by compiling the binary with the correct flag – `-fprofile-generate`. In our case, we need to specify an output location, which should be `$(SC19_DIR_SCRATCH)`.\n",
"\n",
- "poisson2d_train: will inturn be built by \n",
- " - cleans up old profile data if any (rm \\$(SC19\\_DIR\\_SCRATCH)/*gcda)\n",
- " - builds a training binary using -fprofile-generate=$(SC19_DIR_SCRATCH) along with the usual optimization flags \n",
- " - This instructs the compiler to record hot path information.\n",
- " - Run the training binary with a smaller input size; \n",
- " - you should see a `.gcda` file generated which stores hot path information for further optimization by the compiler in the path specified in profile-generate \n",
- " option\n",
- " \n",
- " - poisson2d_ref: will be built using the profile collected in \\$(SC19\\_DIR\\_SCRATCH)/*gcda files \n",
- " - This is facilitated by the option -fprofile-use=\\$(SC19\\_DIR\\_SCRATCH) option to be added along with optimization flags\n",
- " - Rebuilding the application with this flag builds the final binary that can be run\n",
+ "Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension `.gcda`) is written to the directory specified during compilation.\n",
"\n",
- "**TASK**: Run the following steps to generate the final optimized binary using the following steps. Compare the performance of the Ofast binary with and without profile directed feedback. Run the program poisson2d with increasing values of xiter, yiter, ziter and compare the performance. "
+ "For Step 3, the binary is once again compiled, but this time using the `gcda` profile just generated. The according flag is `-fprofile-use`, which we set to `$(SC19_DIR_SCRATCH)` as well.\n",
+ "\n",
+ "In our `Makefile` at hand, we prepared the steps already for you in the form of two targets.\n",
+ "\n",
+ "* `poisson2d_train`: Will compile the binary with profile-directed feedback\n",
+ "* `poisson2d_ref`: Will take a generated profile and compile a new, optimized binary\n",
+ "\n",
+ "By using dependencies, between these two targets a profile run is launched.\n",
+ "\n",
+ "**TASK**: Edit the [Makefile](`Makefile`) and add the `-fprofile-*` flags to the `CFLAGS` of `poisson2d_train` and\n",
+ "`poisson2d_ref` as outline in the file.\n",
+ "\n",
+ "After that, you may launch them with the following cells (`gen_profile` is a meta-target and uses `poisson2d_train` and `poisson2d_ref`). If you need to clean the generated profile, you may use `make clean_profile`."
]
},
{
"cell_type": "code",
- "execution_count": 61,
+ "execution_count": 79,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "rm -rf /gpfs/wolf/trn003/scratch/archanaravindar//*gcda\n",
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fprofile-generate=/gpfs/wolf/trn003/scratch/archanaravindar/ poisson2d.c -o poisson2d_train -lm \n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
- "Job <24747> is submitted to default queue <batch>.\n",
+ "Job <24905> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
"Calculate current execution.\n",
" 0, 0.249490\n",
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fprofile-use=/gpfs/wolf/trn003/scratch/archanaravindar/ poisson2d.c -o poisson2d_ref -lm \n",
- "cp poisson2d_ref poisson2d\t\n"
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_ref -lm \n",
+ "cp poisson2d_ref poisson2d\n"
]
}
],
"source": [
- "!make -B runpdf"
+ "!make gen_profile"
]
},
{
"cell_type": "markdown",
- "metadata": {
- "exercise": "Solutions"
- },
+ "metadata": {},
"source": [
- "Measure Execution time."
+ "If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!"
]
},
{
"cell_type": "code",
- "execution_count": 62,
+ "execution_count": 80,
"metadata": {},
"outputs": [
{
@@ -578,7 +596,7 @@
"output_type": "stream",
"text": [
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24748> is submitted to default queue <batch>.\n",
+ "Job <24906> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -593,27 +611,34 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "2.30user 0.00system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
- "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ "2.28user 0.01system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n"
]
}
],
"source": [
- "!make run "
+ "!make run"
]
},
{
"cell_type": "markdown",
"metadata": {
- "exercise": "Solutions"
+ "exercise": "solution"
},
"source": [
- "Measure cycles, instructions."
+ "Great! It is! In our tests, this shaved off another 5%."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's also measure instructions and cycles"
]
},
{
"cell_type": "code",
- "execution_count": 64,
+ "execution_count": 81,
"metadata": {},
"outputs": [
{
@@ -621,7 +646,7 @@
"output_type": "stream",
"text": [
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
- "Job <24749> is submitted to default queue <batch>.\n",
+ "Job <24907> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -639,16 +664,16 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 7907873140 cycles:u \n",
- " 12253084984 instructions:u # 1.55 insn per cycle \n",
+ " 7925983538 cycles:u \n",
+ " 12253080719 instructions:u # 1.55 insn per cycle \n",
"\n",
- " 2.308084004 seconds time elapsed\n",
+ " 2.313471365 seconds time elapsed\n",
"\n"
]
}
],
"source": [
- "!make runstats"
+ "!make run_perf"
]
},
{
@@ -658,168 +683,29 @@
"What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
]
},
- {
- "cell_type": "markdown",
- "metadata": {
- "exercise": "Solutions"
- },
- "source": [
- "For the problem size `NX=NY=NITER=1000` you will see that profile directed feedback improves the performance further by 4-5%. "
- ]
- },
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Part C: Compiler annotations/Remarks\n",
"\n",
- "Usually all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done. \n",
- "\n",
- "To generate compiler annotations using GCC, start with Makefile.gcc. Add the flag -fopt-info-all=`$(SC19_DIR_SCRATCH)/filename` to CFLAGS. All annotations are stored in opt-record in the $(SC19_DIR_SCRATCH) which can be viewed as a text file. Make target view displays the contents of the file on the output screen.\n",
+ "Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done. \n",
"\n",
- "Specifically, if you want to view only the missed options, use the option -fopt-info-missed instead of -fopt-info-all.\n",
+ "To generate compiler annotations using GCC, one uses `-fopt-info-all`. If you only want to see the missed options, use the option `-fopt-info-missed` instead of `-fopt-info-all`. See also the [documentation of GCC regarding the flag](https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info).\n",
"\n",
- "**TASK**: Build the application using Makefile.gcc.record. This makefile generates the compiler annotations in $(SC19_DIR_SCRATCH)/opt-record. Using make file target view or vi, read the compiler annotations file to get an idea of the optimizations done. "
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 76,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopt-info-all=/gpfs/wolf/trn003/scratch/archanaravindar//opt-record poisson2d.c poisson2d_reference.o -o poisson2d_Ofast_record -lm \n",
- "cp poisson2d_Ofast_record poisson2d\n"
- ]
- }
- ],
- "source": [
- "!make -B poisson2d_Ofast_record"
+ "**TASK**: Have a looK at the `CFLAGS` of the `Makefile` target `poisson2d_Ofast_info`. Add the flag `-fopt-info-all` to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use – for example – `-fopt-info-all=(SC19_DIR_SCRATCH)/filename`."
]
},
{
"cell_type": "code",
- "execution_count": 73,
+ "execution_count": 82,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "poisson2d_reference.c:85:17: missed: not inlinable: check_results/17 -> fprintf/20, function body not available\n",
- "poisson2d_reference.c:70:31: missed: not inlinable: poisson2d_reference/16 -> printf/18, function body not available\n",
- "Unit growth for small function inlining: 145->145 (0%)\n",
- "\n",
- "Inlined 0 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_54 and *_57\n",
- "consider run-time aliasing test between *_62 and *_67\n",
- "consider run-time aliasing test between *_75 and *_78\n",
- "consider run-time aliasing test between *_82 and *_87\n",
- "poisson2d_reference.c:52:13: optimized: Loop 6 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:41:9: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d_reference.c:64:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:64:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:59:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:59:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:50:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_386, _378, _390);\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:45:90: missed: not vectorized: complicated access pattern.\n",
- "poisson2d_reference.c:43:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d_reference.c:33:6: note: vectorized 1 loops in function.\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_386, _378, _390);\n",
- "poisson2d_reference.c:43:13: optimized: loop turned into non-loop; it never loops\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_386, _378, _390);\n",
- "poisson2d_reference.c:70:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_134, error_144);\n",
- "poisson2d_reference.c:81:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:81:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:83:35: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:83:35: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:76:5: note: vectorized 0 loops in function.\n",
- "poisson2d_reference.c:85:17: missed: statement clobbers memory: fprintf (stderr.0_13, \"ERROR: A[%d][%d] = %f does not match %f (reference)\\n\", iy_62, ix_61, _63, _64);\n",
- "poisson2d.c:57:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:51:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:47:20: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:157:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:156:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:155:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:154:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:145:9: missed: not inlinable: main/32 -> check_results/38, function body not available\n",
- "poisson2d.c:138:31: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:98:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:95:5: missed: not inlinable: main/32 -> poisson2d_reference/37, function body not available\n",
- "poisson2d.c:93:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:91:5: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:73:29: missed: not inlinable: main/32 -> exp/34, function body not available\n",
- "poisson2d.c:63:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:62:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:61:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:60:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "Unit growth for small function inlining: 234->234 (0%)\n",
- "\n",
- "Inlined 4 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_84 and *_87\n",
- "consider run-time aliasing test between *_92 and *_97\n",
- "consider run-time aliasing test between *_104 and *_107\n",
- "consider run-time aliasing test between *_111 and *_115\n",
- "poisson2d.c:120:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:85:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:103:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:103:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:132:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:132:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:127:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:127:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:118:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_557, _553, _563);\n",
- "poisson2d.c:107:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:107:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:110:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:83:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_542, 0, _545);\n",
- "poisson2d.c:67:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:69:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:38:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_542, 0, _545);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_557, _553, _563);\n",
- "poisson2d.c:110:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _195 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _197 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _201 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _199 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_155 = malloc (_7);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_157 = malloc (_7);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_159 = malloc (_7);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_161 = malloc (_7);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_542, 0, _545);\n",
- "poisson2d.c:91:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_239, ny_221, nx_124);\n",
- "poisson2d.c:93:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate reference solution and time with serial CPU execution.\"[0]);\n",
- "poisson2d.c:95:5: missed: statement clobbers memory: poisson2d_reference (iter_max_239, 1.00000000000000008180305391403130954586231382563710212708e-5, Aref_225, Anew_227, nx_124, ny_221, rhs_236);\n",
- "poisson2d.c:98:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_557, _553, _563);\n",
- "poisson2d.c:138:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_245, error_242);\n",
- "poisson2d.c:145:9: missed: statement clobbers memory: _118 = check_results (1, prephitmp_426, 1, _131, 1.00000000000000008180305391403130954586231382563710212708e-5, A_125, Aref_225, nx_124);\n",
- "poisson2d.c:154:5: missed: statement clobbers memory: free (rhs_236);\n",
- "poisson2d.c:155:5: missed: statement clobbers memory: free (Anew_227);\n",
- "poisson2d.c:156:5: missed: statement clobbers memory: free (Aref_225);\n",
- "poisson2d.c:157:5: missed: statement clobbers memory: free (A_125);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_146 = malloc (2000000);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_145 = malloc (2000000);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_144 = malloc (2000000);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_130 = malloc (2000000);\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all poisson2d.c poisson2d_reference.o -o poisson2d_Ofast_info -lm\n",
"poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
@@ -849,843 +735,78 @@
"poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
"poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
"poisson2d.c:108:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
- "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
- "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
- "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
- "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
- "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
- "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
- "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
- "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
- "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
- "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
- "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
- "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
- "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
- "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n",
- "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
- "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
- "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
- "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
- "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
- "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
- "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
- "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
- "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
- "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
- "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
- "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
- "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
- "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
- "Unit growth for small function inlining: 207->207 (0%)\n",
- "\n",
- "Inlined 4 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_84 and *_87\n",
- "consider run-time aliasing test between *_92 and *_97\n",
- "consider run-time aliasing test between *_104 and *_107\n",
- "consider run-time aliasing test between *_111 and *_115\n",
- "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
- "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
- "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
- "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
- "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
- "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
- "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
- "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
- "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
- "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
- "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
- "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
- "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
- "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
- "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n"
- ]
- }
- ],
- "source": [
- "!cat \"$SC19_DIR_SCRATCH\"/opt-record"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "**TASK**: \n",
- "Modify Makefile.gcc.use to include the option -fopt-info-all=$(SC19_DIR_SCRATCH)/pgo-opt-record in CFLAGS.\n",
- "Build the profile directed feedback binary using the steps below and generate the final optimized binary and also the annotations for the profile directed feedback pass. Compare opt-record and pgo-opt-record and note the differences. which file has greater annotations. How different are the annotations when compared to each other? Can you spot any differences? "
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 74,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "rm -rf /gpfs/wolf/trn003/scratch/archanaravindar//*gcda\n",
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fprofile-generate=/gpfs/wolf/trn003/scratch/archanaravindar/ poisson2d.c -o poisson2d_train -lm \n",
- "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
- "Job <24750> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
- "Calculate current execution.\n",
- " 0, 0.249490\n",
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fprofile-use=/gpfs/wolf/trn003/scratch/archanaravindar/ -fopt-info-all=/gpfs/wolf/trn003/scratch/archanaravindar//pgo-opt-record poisson2d.c -o poisson2d_ref_record -lm \n",
- "cp poisson2d_ref_record poisson2d\n"
- ]
- }
- ],
- "source": [
- "!make runpdf.record"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 75,
- "metadata": {},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "poisson2d_reference.c:85:17: missed: not inlinable: check_results/17 -> fprintf/20, function body not available\n",
- "poisson2d_reference.c:70:31: missed: not inlinable: poisson2d_reference/16 -> printf/18, function body not available\n",
- "Unit growth for small function inlining: 145->145 (0%)\n",
- "\n",
- "Inlined 0 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_54 and *_57\n",
- "consider run-time aliasing test between *_62 and *_67\n",
- "consider run-time aliasing test between *_75 and *_78\n",
- "consider run-time aliasing test between *_82 and *_87\n",
- "poisson2d_reference.c:52:13: optimized: Loop 6 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:41:9: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d_reference.c:64:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:64:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:59:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:59:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:50:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:45:90: missed: not vectorized: complicated access pattern.\n",
- "poisson2d_reference.c:43:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d_reference.c:33:6: note: vectorized 1 loops in function.\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:43:13: optimized: loop turned into non-loop; it never loops\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:70:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_134, error_144);\n",
- "poisson2d_reference.c:64:9: note: considering unrolling loop 5 at BB 25\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:64:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:59:9: note: considering unrolling loop 4 at BB 23\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:59:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:50:9: note: considering unrolling loop 3 at BB 21\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:50:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 9 at BB 14\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:47:25: optimized: loop unrolled 3 times (header execution count 432180)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 7 at BB 11\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:41:9: note: considering unrolling loop 10 at BB 7\n",
- "poisson2d_reference.c:33:6: note: considering unrolling loop 2 at BB 6\n",
- "poisson2d_reference.c:37:25: note: considering unrolling loop 1 at BB 30\n",
- "poisson2d_reference.c:81:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:81:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:83:35: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:83:35: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:76:5: note: vectorized 0 loops in function.\n",
- "poisson2d_reference.c:85:17: missed: statement clobbers memory: fprintf (stderr.0_13, \"ERROR: A[%d][%d] = %f does not match %f (reference)\\n\", iy_62, ix_61, _63, _64);\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 3 at BB 5\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:81:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 1 at BB 8\n",
- "poisson2d.c:57:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:51:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:47:20: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:157:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:156:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:155:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:154:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:145:9: missed: not inlinable: main/32 -> check_results/38, function body not available\n",
- "poisson2d.c:138:31: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:98:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:95:5: missed: not inlinable: main/32 -> poisson2d_reference/37, function body not available\n",
- "poisson2d.c:93:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:91:5: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:73:29: missed: not inlinable: main/32 -> exp/34, function body not available\n",
- "poisson2d.c:63:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:62:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:61:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:60:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "Unit growth for small function inlining: 234->234 (0%)\n",
- "\n",
- "Inlined 4 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_84 and *_87\n",
- "consider run-time aliasing test between *_92 and *_97\n",
- "consider run-time aliasing test between *_104 and *_107\n",
- "consider run-time aliasing test between *_111 and *_115\n",
- "poisson2d.c:120:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:85:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:103:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:103:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:132:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:132:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:127:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:127:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:118:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:107:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:107:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:110:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:83:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:67:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:69:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:38:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:110:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _195 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _197 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _201 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _199 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_155 = malloc (_7);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_157 = malloc (_7);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_159 = malloc (_7);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_161 = malloc (_7);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:91:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_239, ny_221, nx_124);\n",
- "poisson2d.c:93:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate reference solution and time with serial CPU execution.\"[0]);\n",
- "poisson2d.c:95:5: missed: statement clobbers memory: poisson2d_reference (iter_max_239, 1.00000000000000008180305391403130954586231382563710212708e-5, Aref_225, Anew_227, nx_124, ny_221, rhs_236);\n",
- "poisson2d.c:98:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:138:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_245, error_242);\n",
- "poisson2d.c:145:9: missed: statement clobbers memory: _118 = check_results (1, ix_end_162, 1, _131, 1.00000000000000008180305391403130954586231382563710212708e-5, A_125, Aref_225, nx_124);\n",
- "poisson2d.c:154:5: missed: statement clobbers memory: free (rhs_236);\n",
- "poisson2d.c:155:5: missed: statement clobbers memory: free (Anew_227);\n",
- "poisson2d.c:156:5: missed: statement clobbers memory: free (Aref_225);\n",
- "poisson2d.c:157:5: missed: statement clobbers memory: free (A_125);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_146 = malloc (2000000);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_145 = malloc (2000000);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_144 = malloc (2000000);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_130 = malloc (2000000);\n",
- "poisson2d.c:132:9: note: considering unrolling loop 7 at BB 47\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:132:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:127:9: note: considering unrolling loop 6 at BB 44\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:127:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:118:9: note: considering unrolling loop 5 at BB 40\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:118:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 13 at BB 27\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 9 at BB 24\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:107:9: note: considering unrolling loop 14 at BB 37\n",
- "poisson2d.c:38:5: note: considering unrolling loop 4 at BB 35\n",
- "poisson2d.c:103:25: note: considering unrolling loop 3 at BB 51\n",
- "poisson2d.c:83:5: note: considering unrolling loop 2 at BB 18\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:83:5: optimized: loop unrolled 7 times (header execution count 99)\n",
- "poisson2d.c:69:9: note: considering unrolling loop 11 at BB 9\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:69:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d.c:67:5: note: considering unrolling loop 1 at BB 14\n",
- "poisson2d_reference.c:85:17: missed: not inlinable: check_results/17 -> fprintf/20, function body not available\n",
- "poisson2d_reference.c:70:31: missed: not inlinable: poisson2d_reference/16 -> printf/18, function body not available\n",
- "Unit growth for small function inlining: 145->145 (0%)\n",
- "\n",
- "Inlined 0 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_54 and *_57\n",
- "consider run-time aliasing test between *_62 and *_67\n",
- "consider run-time aliasing test between *_75 and *_78\n",
- "consider run-time aliasing test between *_82 and *_87\n",
- "poisson2d_reference.c:52:13: optimized: Loop 6 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:41:9: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d_reference.c:64:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:64:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:59:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:59:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:50:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:45:90: missed: not vectorized: complicated access pattern.\n",
- "poisson2d_reference.c:43:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d_reference.c:33:6: note: vectorized 1 loops in function.\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:43:13: optimized: loop turned into non-loop; it never loops\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:70:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_134, error_144);\n",
- "poisson2d_reference.c:64:9: note: considering unrolling loop 5 at BB 25\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:64:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:59:9: note: considering unrolling loop 4 at BB 23\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:59:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:50:9: note: considering unrolling loop 3 at BB 21\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:50:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 9 at BB 14\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:47:25: optimized: loop unrolled 3 times (header execution count 432180)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 7 at BB 11\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:41:9: note: considering unrolling loop 10 at BB 7\n",
- "poisson2d_reference.c:33:6: note: considering unrolling loop 2 at BB 6\n",
- "poisson2d_reference.c:37:25: note: considering unrolling loop 1 at BB 30\n",
- "poisson2d_reference.c:81:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:81:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:83:35: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:83:35: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:76:5: note: vectorized 0 loops in function.\n",
- "poisson2d_reference.c:85:17: missed: statement clobbers memory: fprintf (stderr.0_13, \"ERROR: A[%d][%d] = %f does not match %f (reference)\\n\", iy_62, ix_61, _63, _64);\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 3 at BB 5\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:81:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 1 at BB 8\n",
- "poisson2d.c:57:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:51:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:47:20: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:157:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:156:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:155:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:154:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:145:9: missed: not inlinable: main/32 -> check_results/38, function body not available\n",
- "poisson2d.c:138:31: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:98:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:95:5: missed: not inlinable: main/32 -> poisson2d_reference/37, function body not available\n",
- "poisson2d.c:93:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:91:5: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:73:29: missed: not inlinable: main/32 -> exp/34, function body not available\n",
- "poisson2d.c:63:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:62:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:61:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:60:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "Unit growth for small function inlining: 234->234 (0%)\n",
- "\n",
- "Inlined 4 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_84 and *_87\n",
- "consider run-time aliasing test between *_92 and *_97\n",
- "consider run-time aliasing test between *_104 and *_107\n",
- "consider run-time aliasing test between *_111 and *_115\n",
- "poisson2d.c:120:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:85:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:103:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:103:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:132:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:132:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:127:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:127:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:118:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:107:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:107:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:110:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:83:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:67:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:69:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:38:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:110:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _195 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _197 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _201 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _199 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_155 = malloc (_7);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_157 = malloc (_7);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_159 = malloc (_7);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_161 = malloc (_7);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:91:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_239, ny_221, nx_124);\n",
- "poisson2d.c:93:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate reference solution and time with serial CPU execution.\"[0]);\n",
- "poisson2d.c:95:5: missed: statement clobbers memory: poisson2d_reference (iter_max_239, 1.00000000000000008180305391403130954586231382563710212708e-5, Aref_225, Anew_227, nx_124, ny_221, rhs_236);\n",
- "poisson2d.c:98:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:138:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_245, error_242);\n",
- "poisson2d.c:145:9: missed: statement clobbers memory: _118 = check_results (1, ix_end_162, 1, _131, 1.00000000000000008180305391403130954586231382563710212708e-5, A_125, Aref_225, nx_124);\n",
- "poisson2d.c:154:5: missed: statement clobbers memory: free (rhs_236);\n",
- "poisson2d.c:155:5: missed: statement clobbers memory: free (Anew_227);\n",
- "poisson2d.c:156:5: missed: statement clobbers memory: free (Aref_225);\n",
- "poisson2d.c:157:5: missed: statement clobbers memory: free (A_125);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_146 = malloc (2000000);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_145 = malloc (2000000);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_144 = malloc (2000000);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_130 = malloc (2000000);\n",
- "poisson2d.c:132:9: note: considering unrolling loop 7 at BB 47\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:132:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:127:9: note: considering unrolling loop 6 at BB 44\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:127:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:118:9: note: considering unrolling loop 5 at BB 40\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:118:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 13 at BB 27\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 9 at BB 24\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:107:9: note: considering unrolling loop 14 at BB 37\n",
- "poisson2d.c:38:5: note: considering unrolling loop 4 at BB 35\n",
- "poisson2d.c:103:25: note: considering unrolling loop 3 at BB 51\n",
- "poisson2d.c:83:5: note: considering unrolling loop 2 at BB 18\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:83:5: optimized: loop unrolled 7 times (header execution count 99)\n",
- "poisson2d.c:69:9: note: considering unrolling loop 11 at BB 9\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:69:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d.c:67:5: note: considering unrolling loop 1 at BB 14\n",
- "poisson2d_reference.c:85:17: missed: not inlinable: check_results/17 -> fprintf/20, function body not available\n",
- "poisson2d_reference.c:70:31: missed: not inlinable: poisson2d_reference/16 -> printf/18, function body not available\n",
- "Unit growth for small function inlining: 145->145 (0%)\n",
- "\n",
- "Inlined 0 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_54 and *_57\n",
- "consider run-time aliasing test between *_62 and *_67\n",
- "consider run-time aliasing test between *_75 and *_78\n",
- "consider run-time aliasing test between *_82 and *_87\n",
- "poisson2d_reference.c:52:13: optimized: Loop 6 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:41:9: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d_reference.c:64:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:64:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:59:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:59:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:50:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:45:90: missed: not vectorized: complicated access pattern.\n",
- "poisson2d_reference.c:43:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d_reference.c:33:6: note: vectorized 1 loops in function.\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:43:13: optimized: loop turned into non-loop; it never loops\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:70:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_134, error_144);\n",
- "poisson2d_reference.c:64:9: note: considering unrolling loop 5 at BB 25\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:64:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:59:9: note: considering unrolling loop 4 at BB 23\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:59:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:50:9: note: considering unrolling loop 3 at BB 21\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:50:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 9 at BB 14\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:47:25: optimized: loop unrolled 3 times (header execution count 432180)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 7 at BB 11\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:41:9: note: considering unrolling loop 10 at BB 7\n",
- "poisson2d_reference.c:33:6: note: considering unrolling loop 2 at BB 6\n",
- "poisson2d_reference.c:37:25: note: considering unrolling loop 1 at BB 30\n",
- "poisson2d_reference.c:81:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:81:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:83:35: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:83:35: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:76:5: note: vectorized 0 loops in function.\n",
- "poisson2d_reference.c:85:17: missed: statement clobbers memory: fprintf (stderr.0_13, \"ERROR: A[%d][%d] = %f does not match %f (reference)\\n\", iy_62, ix_61, _63, _64);\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 3 at BB 5\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:81:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 1 at BB 8\n",
- "poisson2d.c:57:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:51:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:47:20: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:157:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:156:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:155:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:154:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:145:9: missed: not inlinable: main/32 -> check_results/38, function body not available\n",
- "poisson2d.c:138:31: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:98:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:95:5: missed: not inlinable: main/32 -> poisson2d_reference/37, function body not available\n",
- "poisson2d.c:93:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:91:5: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:73:29: missed: not inlinable: main/32 -> exp/34, function body not available\n",
- "poisson2d.c:63:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:62:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:61:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:60:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "Unit growth for small function inlining: 234->234 (0%)\n",
- "\n",
- "Inlined 4 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_84 and *_87\n",
- "consider run-time aliasing test between *_92 and *_97\n",
- "consider run-time aliasing test between *_104 and *_107\n",
- "consider run-time aliasing test between *_111 and *_115\n",
- "poisson2d.c:120:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:85:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:103:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:103:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:132:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:132:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:127:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:127:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:118:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:107:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:107:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:110:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:83:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:67:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:69:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:38:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:110:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _195 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _197 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _201 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _199 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_155 = malloc (_7);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_157 = malloc (_7);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_159 = malloc (_7);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_161 = malloc (_7);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:91:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_239, ny_221, nx_124);\n",
- "poisson2d.c:93:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate reference solution and time with serial CPU execution.\"[0]);\n",
- "poisson2d.c:95:5: missed: statement clobbers memory: poisson2d_reference (iter_max_239, 1.00000000000000008180305391403130954586231382563710212708e-5, Aref_225, Anew_227, nx_124, ny_221, rhs_236);\n",
- "poisson2d.c:98:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:138:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_245, error_242);\n",
- "poisson2d.c:145:9: missed: statement clobbers memory: _118 = check_results (1, ix_end_162, 1, _131, 1.00000000000000008180305391403130954586231382563710212708e-5, A_125, Aref_225, nx_124);\n",
- "poisson2d.c:154:5: missed: statement clobbers memory: free (rhs_236);\n",
- "poisson2d.c:155:5: missed: statement clobbers memory: free (Anew_227);\n",
- "poisson2d.c:156:5: missed: statement clobbers memory: free (Aref_225);\n",
- "poisson2d.c:157:5: missed: statement clobbers memory: free (A_125);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_146 = malloc (2000000);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_145 = malloc (2000000);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_144 = malloc (2000000);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_130 = malloc (2000000);\n",
- "poisson2d.c:132:9: note: considering unrolling loop 7 at BB 47\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:132:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:127:9: note: considering unrolling loop 6 at BB 44\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:127:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:118:9: note: considering unrolling loop 5 at BB 40\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:118:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 13 at BB 27\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 9 at BB 24\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:107:9: note: considering unrolling loop 14 at BB 37\n",
- "poisson2d.c:38:5: note: considering unrolling loop 4 at BB 35\n",
- "poisson2d.c:103:25: note: considering unrolling loop 3 at BB 51\n",
- "poisson2d.c:83:5: note: considering unrolling loop 2 at BB 18\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:83:5: optimized: loop unrolled 7 times (header execution count 99)\n",
- "poisson2d.c:69:9: note: considering unrolling loop 11 at BB 9\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:69:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d.c:67:5: note: considering unrolling loop 1 at BB 14\n",
- "poisson2d_reference.c:85:17: missed: not inlinable: check_results/17 -> fprintf/20, function body not available\n",
- "poisson2d_reference.c:70:31: missed: not inlinable: poisson2d_reference/16 -> printf/18, function body not available\n",
- "Unit growth for small function inlining: 145->145 (0%)\n",
- "\n",
- "Inlined 0 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_54 and *_57\n",
- "consider run-time aliasing test between *_62 and *_67\n",
- "consider run-time aliasing test between *_75 and *_78\n",
- "consider run-time aliasing test between *_82 and *_87\n",
- "poisson2d_reference.c:52:13: optimized: Loop 6 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:41:9: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d_reference.c:64:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:64:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:59:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:59:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d_reference.c:50:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:41:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:45:90: missed: not vectorized: complicated access pattern.\n",
- "poisson2d_reference.c:43:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d_reference.c:33:6: note: vectorized 1 loops in function.\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:43:13: optimized: loop turned into non-loop; it never loops\n",
- "poisson2d_reference.c:33:6: missed: statement clobbers memory: __builtin_memcpy (_384, _376, _388);\n",
- "poisson2d_reference.c:70:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_134, error_144);\n",
- "poisson2d_reference.c:64:9: note: considering unrolling loop 5 at BB 25\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:64:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:59:9: note: considering unrolling loop 4 at BB 23\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:59:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:50:9: note: considering unrolling loop 3 at BB 21\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:50:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 9 at BB 14\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:47:25: optimized: loop unrolled 3 times (header execution count 432180)\n",
- "poisson2d_reference.c:47:25: note: considering unrolling loop 7 at BB 11\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:41:9: note: considering unrolling loop 10 at BB 7\n",
- "poisson2d_reference.c:33:6: note: considering unrolling loop 2 at BB 6\n",
- "poisson2d_reference.c:37:25: note: considering unrolling loop 1 at BB 30\n",
- "poisson2d_reference.c:81:9: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:81:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:83:35: missed: couldn't vectorize loop\n",
- "poisson2d_reference.c:83:35: missed: not vectorized: control flow in loop.\n",
- "poisson2d_reference.c:76:5: note: vectorized 0 loops in function.\n",
- "poisson2d_reference.c:85:17: missed: statement clobbers memory: fprintf (stderr.0_13, \"ERROR: A[%d][%d] = %f does not match %f (reference)\\n\", iy_62, ix_61, _63, _64);\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 3 at BB 5\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d_reference.c:81:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d_reference.c:81:9: note: considering unrolling loop 1 at BB 8\n",
- "poisson2d.c:57:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:51:14: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:47:20: optimized: Inlining atoi/24 into main/32 (always_inline).\n",
- "poisson2d.c:157:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:156:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:155:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:154:5: missed: not inlinable: main/32 -> free/39, function body not available\n",
- "poisson2d.c:145:9: missed: not inlinable: main/32 -> check_results/38, function body not available\n",
- "poisson2d.c:138:31: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:98:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:95:5: missed: not inlinable: main/32 -> poisson2d_reference/37, function body not available\n",
- "poisson2d.c:93:5: missed: not inlinable: main/32 -> __builtin_puts/36, function body not available\n",
- "poisson2d.c:91:5: missed: not inlinable: main/32 -> printf/35, function body not available\n",
- "poisson2d.c:73:29: missed: not inlinable: main/32 -> exp/34, function body not available\n",
- "poisson2d.c:63:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:62:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:61:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "poisson2d.c:60:41: missed: not inlinable: main/32 -> malloc/33, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "/usr/include/stdlib.h:280:16: missed: not inlinable: main/32 -> strtol/40, function body not available\n",
- "Unit growth for small function inlining: 234->234 (0%)\n",
- "\n",
- "Inlined 4 calls, eliminated 0 functions\n",
- "\n",
- "consider run-time aliasing test between *_84 and *_87\n",
- "consider run-time aliasing test between *_92 and *_97\n",
- "consider run-time aliasing test between *_104 and *_107\n",
- "consider run-time aliasing test between *_111 and *_115\n",
- "poisson2d.c:120:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:85:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
- "poisson2d.c:103:25: missed: couldn't vectorize loop\n",
- "poisson2d.c:103:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
- "poisson2d.c:132:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:132:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:127:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:127:9: missed: Loop costings may not be worthwhile.\n",
- "poisson2d.c:118:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:107:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:107:9: missed: not vectorized: control flow in loop.\n",
- "poisson2d.c:110:13: optimized: loop vectorized using 16 byte vectors\n",
- "poisson2d.c:83:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:67:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:27: missed: not vectorized: complicated access pattern.\n",
- "poisson2d.c:69:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:73:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
- "poisson2d.c:38:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:110:13: optimized: loop turned into non-loop; it never loops\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _195 = strtol (_1, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _197 = strtol (_2, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _201 = strtol (_3, 0B, 10);\n",
- "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _199 = strtol (_4, 0B, 10);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_155 = malloc (_7);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_157 = malloc (_7);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_159 = malloc (_7);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_161 = malloc (_7);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memset (_538, 0, _541);\n",
- "poisson2d.c:91:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_239, ny_221, nx_124);\n",
- "poisson2d.c:93:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate reference solution and time with serial CPU execution.\"[0]);\n",
- "poisson2d.c:95:5: missed: statement clobbers memory: poisson2d_reference (iter_max_239, 1.00000000000000008180305391403130954586231382563710212708e-5, Aref_225, Anew_227, nx_124, ny_221, rhs_236);\n",
- "poisson2d.c:98:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:38:5: missed: statement clobbers memory: __builtin_memcpy (_553, _549, _558);\n",
- "poisson2d.c:138:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_245, error_242);\n",
- "poisson2d.c:145:9: missed: statement clobbers memory: _118 = check_results (1, ix_end_162, 1, _131, 1.00000000000000008180305391403130954586231382563710212708e-5, A_125, Aref_225, nx_124);\n",
- "poisson2d.c:154:5: missed: statement clobbers memory: free (rhs_236);\n",
- "poisson2d.c:155:5: missed: statement clobbers memory: free (Anew_227);\n",
- "poisson2d.c:156:5: missed: statement clobbers memory: free (Aref_225);\n",
- "poisson2d.c:157:5: missed: statement clobbers memory: free (A_125);\n",
- "poisson2d.c:60:41: missed: statement clobbers memory: A_146 = malloc (2000000);\n",
- "poisson2d.c:61:41: missed: statement clobbers memory: Aref_145 = malloc (2000000);\n",
- "poisson2d.c:62:41: missed: statement clobbers memory: Anew_144 = malloc (2000000);\n",
- "poisson2d.c:63:41: missed: statement clobbers memory: rhs_130 = malloc (2000000);\n",
- "poisson2d.c:132:9: note: considering unrolling loop 7 at BB 47\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:132:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:127:9: note: considering unrolling loop 6 at BB 44\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:127:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:118:9: note: considering unrolling loop 5 at BB 40\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:118:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 13 at BB 27\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
- "poisson2d.c:114:25: note: considering unrolling loop 9 at BB 24\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:107:9: note: considering unrolling loop 14 at BB 37\n",
- "poisson2d.c:38:5: note: considering unrolling loop 4 at BB 35\n",
- "poisson2d.c:103:25: note: considering unrolling loop 3 at BB 51\n",
- "poisson2d.c:83:5: note: considering unrolling loop 2 at BB 18\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:83:5: optimized: loop unrolled 7 times (header execution count 99)\n",
- "poisson2d.c:69:9: note: considering unrolling loop 11 at BB 9\n",
- "considering unrolling loop with constant number of iterations\n",
- "considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:69:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d.c:67:5: note: considering unrolling loop 1 at BB 14\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.\n",
+ "\n",
+ "**TASK**: \n",
+ "Adapt the `CFLAGS` of `poisson2d_ref_info` to include `-fopt-info-all` **and** the profile input of `-fprofile-use=…` here. *(Be advised: Long output!)*"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 83,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
"poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "Increasing alignment of decl: __gcov0.main\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_D_00100_1_main/48 -> __gcov_exit/55, function body not available\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_I_00100_0_main/47 -> __gcov_init/54, function body not available\n",
"poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
"poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
"poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
@@ -1700,7 +821,7 @@
"/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
"/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
"/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
- "Unit growth for small function inlining: 207->207 (0%)\n",
+ "Unit growth for small function inlining: 295->295 (0%)\n",
"\n",
"Inlined 4 calls, eliminated 0 functions\n",
"\n",
@@ -1710,6 +831,8 @@
"consider run-time aliasing test between *_111 and *_115\n",
"poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
"poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
"poisson2d.c:108:25: missed: couldn't vectorize loop\n",
"poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
"poisson2d.c:136:9: missed: couldn't vectorize loop\n",
@@ -1717,19 +840,19 @@
"poisson2d.c:131:9: missed: couldn't vectorize loop\n",
"poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
"poisson2d.c:122:9: missed: couldn't vectorize loop\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:122:9: missed: not vectorized: control flow in loop.\n",
"poisson2d.c:112:9: missed: couldn't vectorize loop\n",
"poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
"poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
"poisson2d.c:88:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:88:5: missed: not vectorized: control flow in loop.\n",
"poisson2d.c:72:5: missed: couldn't vectorize loop\n",
- "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:72:5: missed: not vectorized: control flow in loop.\n",
"poisson2d.c:74:9: missed: couldn't vectorize loop\n",
"poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
"poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
"poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
"/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
"/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
@@ -1738,48 +861,60 @@
"poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
"poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
"poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
- "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_337, ny_124, nx_286);\n",
"poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
- "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
- "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
- "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
- "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
- "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
- "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
- "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
- "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n",
- "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 47\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_316, error_118);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_127);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_311);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_122);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_129 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_132 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_140 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 53\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
"poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 44\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 50\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
"poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
- "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 40\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 47\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:122:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
- "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 27\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 3 times (header execution count 9800)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 33\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
"poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
- "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 24\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 30\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 37\n",
- "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 35\n",
- "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 51\n",
- "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 18\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 42\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 40\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 60\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 23\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
- "poisson2d.c:88:5: optimized: loop unrolled 7 times (header execution count 99)\n",
- "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 9\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 3 times (header execution count 100)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 12\n",
"considering unrolling loop with constant number of iterations\n",
"considering unrolling loop with runtime-computable number of iterations\n",
"poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
- "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 14\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 16\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_init (&*.LPBX0);\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_exit ();\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24908> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "libgcov profiling error:/gpfs/wolf/trn003/scratch/aherten//#autofs#nccsopen-svm1_home#aherten#SC19-Tutorial#3-Optimizing_POWER#Handson#Task1#poisson2d.gcda:overwriting an existing profile data with a different timestamp\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c poisson2d_reference.o -o poisson2d_ref_info -lm\n",
"poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
"poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
@@ -1882,20 +1017,21 @@
}
],
"source": [
- "!cat \"$SC19_DIR_SCRATCH\"/pgo-opt-record"
+ "!make poisson2d_ref_info"
]
},
{
"cell_type": "markdown",
- "metadata": {
- "exercise": "solution"
- },
+ "metadata": {},
"source": [
- "Comparing the annotations generated at plain Ofast optimization level and the one generated at Ofast + profile directed feedback, we observe that many more optimizations are possible due to profile information. For instance you will see annotations such as \n",
+ "Comparing the annotations generated of a plain `-Ofast` optimization level and the one generated at `-Ofast` and profile directed feedback, we observe that many more optimizations are possible due to profile information.\n",
"\n",
- "`poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)`\n",
+ "For instance you will see annotations such as\n",
+ "```\n",
+ "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "```\n",
"\n",
- "The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations. \n"
+ "The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations."
]
},
{
@@ -1926,25 +1062,23 @@
"\n",
"### Overview\n",
"\n",
- "Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
- "\n",
- "Verify the impact by measuring cache counters with and without prefetching.\n",
- "\n",
- "Learn how to modify contents of DSCR (data stream control register) using IBM XL compiler and study the impact with different values to DSCR. \n",
+ "* Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
+ "* Verify the impact by measuring cache counters with and without prefetching.\n",
+ "* Learn how to modify contents of DSCR (*Data Stream Control Register*) using IBM XL compiler and study the impact with different values to DSCR. \n",
"\n",
- "First, change directory to that of Task 2"
+ "But first, lets change directory to that of Task 2"
]
},
{
"cell_type": "code",
- "execution_count": 37,
+ "execution_count": 85,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "/autofs/nccsopen-svm1_home/archanaravindar/SC19-Tutorial/Task2\n"
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task2\n"
]
}
],
@@ -1956,7 +1090,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "### Part A: Building with SW Prefetch flag and Running Application."
+ "### Part A: Software Prefetching"
]
},
{
@@ -1965,22 +1099,39 @@
"source": [
"**TASK**: Look at the Makefile and work on the TODOs. \n",
"\n",
- "- First generate Ofast binary and note down the performance in terms of cycles, seconds and L3 Misses. \n",
- "- Modify Makefile to add software prefetching option (`-fprefetch-loop-arrays`). Compare performance of Ofast binary with and without SW prefetching and note down the performance."
+ "- First generate a `-Ofast`-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!\n",
+ "- Modify the `Makefile` to add the option for software prefetching (`-fprefetch-loop-arrays`). Compare performance of `-Ofast` with and without software prefetching"
]
},
{
"cell_type": "code",
- "execution_count": 38,
+ "execution_count": 97,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "rm -f poisson2d poisson2d*.o\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make clean"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 88,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec poisson2d.c -o poisson2d -lm\n",
+ "make: `poisson2d' is up to date.\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24714> is submitted to default queue <batch>.\n",
+ "Job <24911> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -1995,10 +1146,10 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "2.40user 0.00system 0:02.40elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
- "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n",
+ "2.39user 0.01system 0:02.40elapsed 100%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
- "Job <24715> is submitted to default queue <batch>.\n",
+ "Job <24912> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2016,15 +1167,33 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 8264018720 cycles:u \n",
- " 483312952 r168a4:u \n",
+ " 8271503902 cycles:u \n",
+ " 481152478 r168a4:u \n",
"\n",
- " 2.412880428 seconds time elapsed\n",
- "\n",
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ " 2.412224884 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 98,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
"cp poisson2d_pref poisson2d\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24716> is submitted to default queue <batch>.\n",
+ "Job <24919> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2039,10 +1208,10 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "1.93user 0.00system 0:01.93elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
- "256inputs+0outputs (0major+481minor)pagefaults 0swaps\n",
+ "1.92user 0.00system 0:01.93elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
- "Job <24717> is submitted to default queue <batch>.\n",
+ "Job <24920> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2060,45 +1229,39 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 6573061567 cycles:u \n",
- " 459043186 r168a4:u \n",
+ " 6586609284 cycles:u \n",
+ " 459879452 r168a4:u \n",
"\n",
- " 1.920153470 seconds time elapsed\n",
+ " 1.925399505 seconds time elapsed\n",
"\n"
]
}
],
"source": [
- "!make CC=gcc poisson2d\n",
+ "!make poisson2d_pref CC=gcc\n",
"!make run\n",
- "!make l3missstats\n",
- "!make CC=gcc poisson2d_pref\n",
- "!make run\n",
- "!make l3missstats\n"
+ "!make l3missstats"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "**TASK**: Repeat the experiment with O3 flag. \n",
- "\n",
- "- First generate O3 binary and note down the performance in terms of cycles, seconds and L3 Misses. \n",
- "- Modify Makefile to add software prefetching option (`-fprefetch-loop-arrays`). Compare performance of Ofast binary with and without SW prefetching and note down the performance."
+ "**TASK**: Repeat the experiment with the `-O3` flag. Have a look at the `Makefile` and the outlined TODO. There's a position to easily adapt `-Ofast`→`-O3`!"
]
},
{
"cell_type": "code",
- "execution_count": 39,
+ "execution_count": 100,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec poisson2d.c -o poisson2d -lm\n",
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec poisson2d.c -o poisson2d -lm\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24718> is submitted to default queue <batch>.\n",
+ "Job <24923> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2116,7 +1279,7 @@
"4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
"256inputs+0outputs (0major+479minor)pagefaults 0swaps\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
- "Job <24719> is submitted to default queue <batch>.\n",
+ "Job <24924> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2134,15 +1297,32 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 16233775058 cycles:u \n",
- " 632495173 r168a4:u \n",
+ " 16445764669 cycles:u \n",
+ " 645094089 r168a4:u \n",
"\n",
- " 4.729401619 seconds time elapsed\n",
- "\n",
- "gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
- "cp poisson2d_pref poisson2d\n",
+ " 4.792567763 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 101,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24720> is submitted to default queue <batch>.\n",
+ "Job <24925> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2157,10 +1337,10 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "4.72user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
- "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n",
+ "4.74user 0.00system 0:04.74elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
- "Job <24721> is submitted to default queue <batch>.\n",
+ "Job <24926> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2178,19 +1358,16 @@
"\n",
" Performance counter stats for './poisson2d':\n",
"\n",
- " 16244358719 cycles:u \n",
- " 619811301 r168a4:u \n",
+ " 16239159454 cycles:u \n",
+ " 631061431 r168a4:u \n",
"\n",
- " 4.732843397 seconds time elapsed\n",
+ " 4.730144897 seconds time elapsed\n",
"\n"
]
}
],
"source": [
- "!make CC=gcc -B poisson2d\n",
- "!make run\n",
- "!make l3missstats\n",
- "!make CC=gcc -B poisson2d_pref\n",
+ "!make poisson2d_pref CC=gcc -B\n",
"!make run\n",
"!make l3missstats"
]
@@ -2199,7 +1376,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most"
+ "Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?"
]
},
{
@@ -2208,7 +1385,7 @@
"exercise": "solution"
},
"source": [
- "Observing the results, we see that SW Prefetching seems to help at Ofast but not at O3. We can use the steps described in the the next section to verify that the compiler has not inserted any SW prefetch operations at O3 at all. That is because in the O3 binary the time is dominated by `__fmax` call which causes the compiler to come to the conclusion that whatever benefit we obtain by adding SW prefetch will be overshadowed by the penalty of fmax\n",
+ "Observing the results, we see that SW Prefetching seems to help at `-Ofast` but not at `-O3`. We can use the steps described in the the next section to verify that the compiler has not inserted any SW prefetch operations at`-O3` at all. That is because in the `-O3` binary the time is dominated by `__fmax` call which causes the compiler to come to the conclusion that whatever benefit we obtain by adding SW prefetch will be overshadowed by the penalty of `fmax()`\n",
"GCC may add further loop optimizations such as unrolling upon invocation of `–fprefetch-loop-arrays`.\n"
]
},
@@ -2218,7 +1395,7 @@
"source": [
"### Part B: Analysis of Instructions\n",
"\n",
- "Compilation of the Ofast binary with the software prefetching flag causes the compiler to generate the `dcb*' instructions that prefetch memory values to L3."
+ "Compilation of the `-Ofast` binary with the software prefetching flag causes the compiler to generate the `dcb*` instructions that prefetch memory values to L3."
]
},
{
@@ -2226,33 +1403,32 @@
"metadata": {},
"source": [
"**TASK**: \n",
- "Run ` $(SC19_SUBMIT_CMD) objdump -lSd` on each binary file (O3, Ofast with prefetch/no prefetch)\n",
+ "Run `$(SC19_SUBMIT_CMD) objdump -lSd` on each binary file (`-O3`, `-Ofast` with prefetch/no prefetch).\n",
"Look for instructions beginning with `dcb`\n",
"At what optimization levels does the compiler generate software prefetching instructions?"
]
},
{
"cell_type": "code",
- "execution_count": 42,
+ "execution_count": 114,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
- "cp poisson2d_pref poisson2d\n"
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n"
]
}
],
"source": [
"!make CC=gcc -B poisson2d_pref\n",
- "!objdump -lSd ./poisson2d > poisson2d.dis"
+ "!objdump -lSd ./poisson2d_pref > poisson2d.dis"
]
},
{
"cell_type": "code",
- "execution_count": 43,
+ "execution_count": 116,
"metadata": {},
"outputs": [
{
@@ -2285,22 +1461,24 @@
"source": [
"### Part C: Changing Values of DSCR via compiler flags\n",
"\n",
- "This task requires loading IBM XL compiler from the environment. \n",
+ "This task requires using the IBM XL compiler. It should be already in your environment.\n",
+ "\n",
+ "\n",
"We saw the impact of software prefetching in the previous subsection. \n",
"In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance. \n",
"In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching. \n",
"\n",
- "IBM XL compiler has an option -qprefetch=dscr=`<val>` that can be used for this purpose.\n",
- "Compiling with -qprefetch=dscr=1 turns off the prefetcher.\n",
- "One can give various values such as -qprefetch=dscr=4, -qprefetch=dscr=7 etc to control aggressiveness of prefetching.\n",
- "For this exercise we use make CC=xlc_r to illustrate the performance impact.\n",
+ "IBM XL compiler has an option `-qprefetch=dscr=<val>` that can be used for this purpose.\n",
+ "Compiling with `-qprefetch=dscr=1` turns off the prefetcher. One can give various values such as `-qprefetch=dscr=4`, `-qprefetch=dscr=7` etc. to control aggressiveness of prefetching.\n",
+ "\n",
+ "For this exercise we use `make CC=xlc_r` to illustrate the performance impact.\n",
" \n",
"\n",
- "**Task** Generate XL binary by compiling using the following commands. Add -qprefetch=dscr=1 and rebuild the application and note the performance; Compare the performance of the default binary with the binary compiled with -qprefetch=dscr=1. Which one is faster? \n",
+ "**Task** Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the `Makefile`: Add `qprefetch=dscr=1` to the `CFLAGS` and rebuild the application and note the performance. Which one is faster? \n",
"\n",
- "In general, applications benefit with the default settings of hardware DSCR register. (-qprefetch=dscr=0). However, certain applications also benefit with prefetching turned off. \n",
+ "In general, applications benefit with the default settings of hardware DSCR register (`-qprefetch=dscr=0`). However, certain applications also benefit with prefetching turned off. \n",
"\n",
- "It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B. \n"
+ "It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B. "
]
},
{
@@ -2312,17 +1490,17 @@
},
{
"cell_type": "code",
- "execution_count": 47,
+ "execution_count": 117,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "xlc_r -std=c99 -qarch=pwr9 -qtune=pwr9 -Ofast -DUSE_DOUBLE -DINLINE_LIBS poisson2d.c -o poisson2d -lm\n",
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS poisson2d.c -o poisson2d -lm\n",
" 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24725> is submitted to default queue <batch>.\n",
+ "Job <24927> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2337,7 +1515,7 @@
" 700, 345.423208\n",
" 800, 393.963155\n",
" 900, 442.314962\n",
- "2.27user 0.00system 0:02.27elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "2.26user 0.00system 0:02.27elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
"256inputs+0outputs (0major+477minor)pagefaults 0swaps\n"
]
}
@@ -2356,39 +1534,38 @@
},
{
"cell_type": "code",
- "execution_count": 48,
+ "execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "xlc_r -std=c99 -qarch=pwr9 -qtune=pwr9 -Ofast -DUSE_DOUBLE -DINLINE_LIBS -qprefetch=dscr=1 poisson2d.c -o poisson2d_dscr -lm\n",
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS -qprefetch=dscr=1 poisson2d.c -o poisson2d_dscr -lm\n",
" 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
- "cp poisson2d_dscr poisson2d\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24726> is submitted to default queue <batch>.\n",
+ "Job <24929> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
"Calculate current execution.\n",
" 0, 0.249995\n",
- " 100, 50.149062\n",
- " 200, 99.849327\n",
- " 300, 149.352369\n",
- " 400, 198.659746\n",
- " 500, 247.773000\n",
- " 600, 296.693652\n",
- " 700, 345.423208\n",
- " 800, 393.963155\n",
- " 900, 442.314962\n",
- "4.63user 0.01system 0:04.64elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
- "256inputs+0outputs (0major+476minor)pagefaults 0swaps\n"
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.58user 0.00system 0:04.59elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "0inputs+0outputs (0major+476minor)pagefaults 0swaps\n"
]
}
],
"source": [
- "!make CC=xlc_r -B poisson2d_dscr\n",
+ "!make poisson2d_dscr CC=xlc_r -B\n",
"!make run"
]
},
@@ -2444,14 +1621,14 @@
},
{
"cell_type": "code",
- "execution_count": 1,
+ "execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "/autofs/nccsopen-svm1_home/archanaravindar/SC19-Tutorial/Task3/Solutions\n"
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task3\n"
]
}
],
@@ -2465,80 +1642,49 @@
"source": [
"### Part A: Implement OpenMP Pragmas; Compilation\n",
"\n",
- "**Task**: Please add the correct OpenMP pragmas to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.\n",
+ "**Task**: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.\n",
"\n",
- "* **pragmas**: Look at the TODOs in [`poisson2d.c`](/edit/Task3/poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for`\n",
- "* **Compilation**: Please add compilation flags enabling OpenMP in GCC and XL to Makefile. For GCC, we need to add `-fopenmp` and the application needs to be linked with -lgomp. \n",
- "For XL, we need to add -qsmp=omp to the list of compilation flags. \n",
+ "* **Directives**: Look at the TODOs in [`poisson2d.c`](poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for` (and once it's `#pragma omp parallel for reduction(max:error)` – can you guess where?)\n",
+ "* **Compilation**: Please add compilation flags enabling OpenMP in GCC and XL to the `Makefile`. For GCC, we need to add `-fopenmp` and the application needs to be linked with `-lgomp`. For XL, we need to add `-qsmp=omp` to the list of compilation flags. \n",
"\n",
"Afterwards, compile and run the application with the following commands."
]
},
{
"cell_type": "code",
- "execution_count": 17,
+ "execution_count": 39,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm -lgomp \n",
- "gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm -lgomp\n",
- "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24666> is submitted to default queue <batch>.\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
- "Calculate reference solution and time with serial CPU execution.\n",
- " 0, 0.249995\n",
- " 100, 0.248997\n",
- " 200, 0.248007\n",
- " 300, 0.247025\n",
- " 400, 0.246050\n",
- " 500, 0.245084\n",
- " 600, 0.244124\n",
- " 700, 0.243173\n",
- " 800, 0.242228\n",
- " 900, 0.241291\n",
- "Calculate current execution.\n",
- " 0, 0.249995\n",
- " 100, 0.248997\n",
- " 200, 0.248007\n",
- " 300, 0.020992\n",
- " 400, 0.021114\n",
- " 500, 0.245084\n",
- " 600, 0.244124\n",
- " 700, 0.021475\n",
- " 800, 0.242228\n",
- " 900, 0.021712\n",
- "1000x1000: Ref: 2.3959 s, This: 2.1408 s, speedup: 1.12\n",
- "11.08user 0.00system 0:04.57elapsed 242%CPU (0avgtext+0avgdata 24896maxresident)k\n",
- "256inputs+0outputs (0major+739minor)pagefaults 0swaps\n"
+ "gcc -c -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d.c poisson2d_reference.o -o poisson2d -lm \n"
]
}
],
"source": [
- "!make CC=gcc -B run "
+ "!make poisson2d CC=gcc"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "The command to submit a job to the batch system is prepared in an environment variable $SC19_SUBMIT_CMD; use it together with eval. In the following cell, it is shown how to invoke the application using the batch system. "
+ "The command to submit a job to the batch system is prepared in an environment variable `$SC19_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to invoke the application using the batch system. "
]
},
{
"cell_type": "code",
- "execution_count": 3,
+ "execution_count": 40,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "Job <24651> is submitted to default queue <batch>.\n",
+ "Job <24951> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -2554,17 +1700,17 @@
" 800, 0.242228\n",
" 900, 0.241291\n",
"Calculate current execution.\n",
- " 0, 0.249975\n",
+ " 0, 0.249995\n",
" 100, 0.248997\n",
- " 200, 0.020870\n",
- " 300, 0.020992\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
" 400, 0.246050\n",
" 500, 0.245084\n",
" 600, 0.244124\n",
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "1000x1000: Ref: 2.3963 s, This: 2.1526 s, speedup: 1.11\n"
+ "1000x1000: Ref: 4.7430 s, This: 3.9363 s, speedup: 1.20\n"
]
}
],
@@ -2583,7 +1729,7 @@
},
{
"cell_type": "code",
- "execution_count": 5,
+ "execution_count": 32,
"metadata": {
"exercise": "task"
},
@@ -2592,21 +1738,23 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "Job <24653> is submitted to default queue <batch>.\n",
+ "Job <24945> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
+ "\n",
+ "libgomp: Invalid value for environment variable OMP_NUM_THREADS\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
"Calculate reference solution and time with serial CPU execution.\n",
" 0, 0.249995\n",
- " 100, 0.248997\n",
- " 200, 0.248007\n",
- " 300, 0.247025\n",
- " 400, 0.246050\n",
- " 500, 0.245084\n",
- " 600, 0.244124\n",
- " 700, 0.243173\n",
- " 800, 0.242228\n",
- " 900, 0.241291\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
"Calculate current execution.\n",
" 0, 0.249995\n",
" 100, 0.248997\n",
@@ -2618,7 +1766,7 @@
" 700, 0.243173\n",
" 800, 0.242228\n",
" 900, 0.241291\n",
- "1000x1000: Ref: 2.3823 s, This: 2.7327 s, speedup: 0.87\n"
+ "1000x1000: Ref: 2.1046 s, This: 2.4171 s, speedup: 0.87\n"
]
}
],
@@ -2628,7 +1776,7 @@
},
{
"cell_type": "code",
- "execution_count": 18,
+ "execution_count": 41,
"metadata": {
"exercise": "solution"
},
@@ -2639,7 +1787,7 @@
"text": [
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
- "1000x1000: Ref: 2.3780 s, This: 2.7215 s, speedup: 0.87\n"
+ "1000x1000: Ref: 4.7288 s, This: 4.9791 s, speedup: 0.95\n"
]
}
],
@@ -2649,7 +1797,7 @@
},
{
"cell_type": "code",
- "execution_count": 19,
+ "execution_count": 42,
"metadata": {
"exercise": "solution"
},
@@ -2660,7 +1808,7 @@
"text": [
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
- "1000x1000: Ref: 2.3788 s, This: 1.3734 s, speedup: 1.73\n"
+ "1000x1000: Ref: 4.7125 s, This: 2.4914 s, speedup: 1.89\n"
]
}
],
@@ -2670,7 +1818,7 @@
},
{
"cell_type": "code",
- "execution_count": 20,
+ "execution_count": 35,
"metadata": {
"exercise": "solution"
},
@@ -2681,7 +1829,7 @@
"text": [
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
- "1000x1000: Ref: 2.3779 s, This: 0.8743 s, speedup: 2.72\n"
+ "1000x1000: Ref: 2.1065 s, This: 1.3836 s, speedup: 1.52\n"
]
}
],
@@ -2842,18 +1990,18 @@
},
{
"cell_type": "code",
- "execution_count": 54,
+ "execution_count": 36,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "Job <24463> is submitted to default queue <batch>.\n",
+ "Job <24949> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"0-3\n",
- "Job <24464> is submitted to default queue <batch>.\n",
+ "Job <24950> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"4-7\n"
@@ -2869,7 +2017,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the [OMP_PLACES environment Variable](https://www.openmp.org/spec-html/5.0/openmpse53.html). Examples are `OMP_PLACES` which is an OpenMP variable. We also have a GNU specific variable which can also be used to control affinity- `GOMP_CPU_AFFINITY`. Setting GOMP_CPU_AFFINITY is specific to GCC binaries but it internally serves the same function as setting OMP_PLACES. \n",
+ "There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the [OMP_PLACES environment Variable](https://www.openmp.org/spec-html/5.0/openmpse53.html). We also have a GNU specific variable which can also be used to control affinity - `GOMP_CPU_AFFINITY`. Setting `GOMP_CPU_AFFINITY` is specific to GCC binaries but it internally serves the same function as setting `OMP_PLACES`. \n",
"\n",
"**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
"\n",
@@ -2880,7 +2028,7 @@
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 38,
"metadata": {
"exercise": "task"
},
@@ -2889,13 +2037,12 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n"
+ "/usr/bin/sh: OMP_PLACES}={X},{Y},{Z},{A}: command not found\n"
]
}
],
"source": [
- "!eval OMP_DISPLAY_ENV=true OMP_PLACES}=\"{X},{Y},{Z},{A}\" OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
+ "!eval OMP_DISPLAY_ENV=true OMP_PLACES=\"{X},{Y},{Z},{A}\" OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
]
},
{
@@ -2924,7 +2071,7 @@
"exercise": "solution"
},
"source": [
- "Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores. The binding is set using `OMP_PLACES` in this case."
+ "Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores."
]
},
{
@@ -2933,12 +2080,12 @@
"exercise": "solution"
},
"source": [
- "Using `OMP_PLACES` for binding"
+ "Using `OMP_PLACES` for binding, and using some magical Python-Bash interplay:"
]
},
{
"cell_type": "code",
- "execution_count": 26,
+ "execution_count": 43,
"metadata": {
"exercise": "solution"
},
@@ -2951,12 +2098,12 @@
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
" OMP_PLACES = '{0},{1},{2},{3}'\n",
- "1000x1000: Ref: 2.3966 s, This: 2.1410 s, speedup: 1.12\n",
+ "1000x1000: Ref: 4.7315 s, This: 3.9090 s, speedup: 1.21\n",
"Affinity: {0},{5},{9},{13}\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
" OMP_PLACES = '{0},{5},{9},{13}'\n",
- "1000x1000: Ref: 2.3922 s, This: 0.7029 s, speedup: 3.40\n"
+ "1000x1000: Ref: 4.6485 s, This: 1.2829 s, speedup: 3.62\n"
]
}
],
@@ -3009,27 +2156,29 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "The same experiments can be repeated with the IBM XL compiler. \n",
+ "Great!\n",
+ "\n",
+ "If you still have time: The same experiments can be repeated with the IBM XL compiler. \n",
"The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is `-qsmp=omp`\n",
"\n",
- "**Task**: In the Makefile add the openMP flag and generate XL binaries with the openmp option and run the application with various number of threads and note the performance speedup."
+ "**Task**: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup."
]
},
{
"cell_type": "code",
- "execution_count": 27,
+ "execution_count": 44,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "xlc_r -c -std=c99 -qarch=pwr9 -qtune=pwr9 -O3 -qhot -DUSE_DOUBLE -DINLINE_LIBS -qsmp=omp -qsmp=omp poisson2d_reference.c -o poisson2d_reference.o -lm \n",
+ "xlc_r -c -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d_reference.c -o poisson2d_reference.o -lm \n",
" 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
- "xlc_r -std=c99 -qarch=pwr9 -qtune=pwr9 -O3 -qhot -DUSE_DOUBLE -DINLINE_LIBS -qsmp=omp -qsmp=omp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ "xlc_r -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
" 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
"bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
- "Job <24677> is submitted to default queue <batch>.\n",
+ "Job <24956> is submitted to default queue <batch>.\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
"Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
@@ -3045,19 +2194,19 @@
" 800, 393.963155\n",
" 900, 442.314962\n",
"Calculate current execution.\n",
- " 0, 0.249975\n",
+ " 0, 0.249995\n",
" 100, 50.149062\n",
" 200, 99.849327\n",
" 300, 149.352369\n",
" 400, 198.659746\n",
" 500, 247.773000\n",
" 600, 296.693652\n",
- " 700, 29.493030\n",
- " 800, 33.799893\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
" 900, 442.314962\n",
- "1000x1000: Ref: 6.0360 s, This: 2.4149 s, speedup: 2.50\n",
- "21.04user 6.65system 0:08.48elapsed 326%CPU (0avgtext+0avgdata 23040maxresident)k\n",
- "256inputs+0outputs (0major+1099minor)pagefaults 0swaps\n"
+ "1000x1000: Ref: 5.6783 s, This: 2.6528 s, speedup: 2.14\n",
+ "21.56user 6.18system 0:08.37elapsed 331%CPU (0avgtext+0avgdata 23040maxresident)k\n",
+ "3200inputs+0outputs (2major+1098minor)pagefaults 0swaps\n"
]
}
],
@@ -3212,7 +2361,7 @@
},
{
"cell_type": "code",
- "execution_count": 32,
+ "execution_count": 45,
"metadata": {
"exercise": "solution"
},
@@ -3223,7 +2372,7 @@
"text": [
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
- "1000x1000: Ref: 2.3086 s, This: 0.2739 s, speedup: 8.43\n"
+ "1000x1000: Ref: 2.1678 s, This: 0.2869 s, speedup: 7.56\n"
]
}
],
@@ -3304,30 +2453,30 @@
"\n",
"Adapt the following command with your configuration – or follow along accordingly in the non-interactive version of the Notebook.\n",
"\n",
+ "We are mixing Python with Bash (`!`) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two `$$`)\n",
+ "\n",
"What's your maximum speedup?"
]
},
{
"cell_type": "code",
- "execution_count": 38,
+ "execution_count": null,
"metadata": {
- "exercise": "solution"
+ "exercise": "task"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "Affinity: {0},{1},{2},{3}\n",
- "<<Waiting for dispatch ...>>\n",
- "<<Starting on login1>>\n",
- " OMP_PLACES='{0},{1},{2},{3}' custom\n",
- "1000x1000: Ref: 6.0379 s, This: 2.4154 s, speedup: 2.50\n",
- "Affinity: {0},{5},{9},{13}\n",
+ "Affinity: {X},{Y},{Z},{A}\n",
"<<Waiting for dispatch ...>>\n",
"<<Starting on login1>>\n",
- " OMP_PLACES='{0},{5},{9},{13}' custom\n",
- "1000x1000: Ref: 2.3051 s, This: 0.6816 s, speedup: 3.38\n"
+ "1587-117 The string for the OpenMP environment variable 'OMP_PLACES' contains unexpected or invalid text. OpenMP environment variable ignored. \n",
+ " OMP_PLACES='cores(44)'\n",
+ "1000x1000: Ref: 2.0988 s, This: 0.6556 s, speedup: 3.20\n",
+ "Affinity: {P},{Q},{R},{S}\n",
+ "<<Waiting for dispatch ...>>\n"
]
}
],
@@ -3398,7 +2547,7 @@
"source": [
"# Survey<a name=\"survey\"></a>\n",
"\n",
- "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc18-eval)."
+ "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc19-eval)."
]
}
],
diff --git a/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.html b/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.html
index 42729e6..53ef1e0 100644
--- a/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.html
+++ b/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.html
@@ -13017,45 +13017,6 @@ ul.typeahead-list > li > a.pull-right {
.highlight .vm { color: #19177C } /* Name.Variable.Magic */
.highlight .il { color: #666666 } /* Literal.Number.Integer.Long */
</style>
-<style type="text/css">
-
-/* Temporary definitions which will become obsolete with Notebook release 5.0 */
-.ansi-black-fg { color: #3E424D; }
-.ansi-black-bg { background-color: #3E424D; }
-.ansi-black-intense-fg { color: #282C36; }
-.ansi-black-intense-bg { background-color: #282C36; }
-.ansi-red-fg { color: #E75C58; }
-.ansi-red-bg { background-color: #E75C58; }
-.ansi-red-intense-fg { color: #B22B31; }
-.ansi-red-intense-bg { background-color: #B22B31; }
-.ansi-green-fg { color: #00A250; }
-.ansi-green-bg { background-color: #00A250; }
-.ansi-green-intense-fg { color: #007427; }
-.ansi-green-intense-bg { background-color: #007427; }
-.ansi-yellow-fg { color: #DDB62B; }
-.ansi-yellow-bg { background-color: #DDB62B; }
-.ansi-yellow-intense-fg { color: #B27D12; }
-.ansi-yellow-intense-bg { background-color: #B27D12; }
-.ansi-blue-fg { color: #208FFB; }
-.ansi-blue-bg { background-color: #208FFB; }
-.ansi-blue-intense-fg { color: #0065CA; }
-.ansi-blue-intense-bg { background-color: #0065CA; }
-.ansi-magenta-fg { color: #D160C4; }
-.ansi-magenta-bg { background-color: #D160C4; }
-.ansi-magenta-intense-fg { color: #A03196; }
-.ansi-magenta-intense-bg { background-color: #A03196; }
-.ansi-cyan-fg { color: #60C6C8; }
-.ansi-cyan-bg { background-color: #60C6C8; }
-.ansi-cyan-intense-fg { color: #258F8F; }
-.ansi-cyan-intense-bg { background-color: #258F8F; }
-.ansi-white-fg { color: #C5C1B4; }
-.ansi-white-bg { background-color: #C5C1B4; }
-.ansi-white-intense-fg { color: #A1A6B2; }
-.ansi-white-intense-bg { background-color: #A1A6B2; }
-
-.ansi-bold { font-weight: bold; }
-
- </style>
<style type="text/css">
@@ -13089,7 +13050,7 @@ div#notebook {
<!-- Loading mathjax macro -->
<!-- Load mathjax -->
- <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.1/MathJax.js?config=TeX-AMS_HTML"></script>
+ <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS_HTML"></script>
<!-- MathJax configuration -->
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
@@ -13116,7 +13077,7 @@ div#notebook {
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h1 id="Hands-On-Performance-Optimization">Hands-On Performance Optimization<a class="anchor-link" href="#Hands-On-Performance-Optimization">¶</a></h1><p><em>Supercomputing 2018 Tutorial "Application Porting and Optimization on GPU-Accelerated POWER Architectures", November 12th 2018</em></p>
+<h1 id="Hands-On-Performance-Optimization">Hands-On Performance Optimization<a class="anchor-link" href="#Hands-On-Performance-Optimization">¶</a></h1><p><em>Supercomputing 2019 Tutorial "Application Porting and Optimization on GPU-Accelerated POWER Architectures", November 18th 2019</em></p>
<hr>
</div>
@@ -13128,7 +13089,7 @@ div#notebook {
<p>As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.</p>
<h2 id="Jupyter-notebook-execution">Jupyter notebook execution<a class="anchor-link" href="#Jupyter-notebook-execution">¶</a></h2><p>When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the <em>edit</em> menu above.</p>
<p>You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.</p>
-<p>If you want you also can get a <a href="/terminals/1">terminal</a> in your browser.</p>
+<p>If you want you also can get a terminal in your browser; just open it via the »New Launcher« button (<code>+</code>).</p>
<h2 id="Terminal-fallback">Terminal fallback<a class="anchor-link" href="#Terminal-fallback">¶</a></h2><p>The tasks are place in directories named <code>Task[1-3]</code>.</p>
<p>Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description.</p>
@@ -13138,10 +13099,23 @@ div#notebook {
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Setup">Setup<a class="anchor-link" href="#Setup">¶</a></h2><p>This hands-on session requires of GCC 6.4.0. By loading the <code>sc18/handson2</code> module before invoking this Notebook, we took care of also loading GCC 6.4.0 into the environment.</p>
+<h2 id="Setup">Setup<a class="anchor-link" href="#Setup">¶</a></h2><p>We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!</p>
</div>
</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>gcc --version
+</pre></div>
+
+ </div>
+</div>
+</div>
+
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
@@ -13149,17 +13123,19 @@ div#notebook {
<h2 id="Tasks">Tasks<a name="top" /><a class="anchor-link" href="#Tasks">¶</a></h2><p>This session comes with multiple tasks, each one to be found in the respective sub-directory <code>Task[1-3]</code>. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.</p>
<p>Please choose from the task below.</p>
<ul>
-<li><p><a href="#task1">Task 1</a>: Compile Flags<br>
-Improve performance of the CPU Jacobi solver with compiler flags such as <code>Ofast</code> and profile-directed feedback (<a href="#solution0">Solution 1</a>)</p>
-</li>
-<li><p><a href="#task2">Task 2</a>: Software Prefetching<br>
-Improve performance of the CPU Jacobi solver with software prefetching (<a href="#solution1">Solution 2</a>)</p>
-</li>
-<li><p><a href="#task3">Task 3</a>: OpenMP<br>
-Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance (<a href="#solution2">Solution 3</a>)</p>
-</li>
-<li><p><a href="#survey">Suvery</a> Please remember to take the survey !</p>
-</li>
+<li><a href="#task1">Task 1</a>: <strong>Basic compiler optimization flags and compiler annotations</strong></li>
+</ul>
+<p>Improve performance of the CPU Jacobi solver with compiler flags such as <code>Ofast</code> and profile-directed feedback. Learn about compiler annotations.</p>
+<ul>
+<li><a href="#task2">Task 2</a>: <strong>Optimization via Prefetching controlled by compiler</strong></li>
+</ul>
+<p>Improve performance of the CPU Jacobi solver with software prefetching. Some compilers such as IBM XL define flags that can be used to modify the aggressiveness of the hardware prefetcher. Learn to modify the DSCR value through XL and study the impact on application performance.</p>
+<ul>
+<li><a href="#task3">Task 3</a>: <strong>Optimization via OpenMP controlled by compiler and the system</strong></li>
+</ul>
+<p>Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance.</p>
+<ul>
+<li><a href="#survey">Suvery</a> Please remember to take the survey !</li>
</ul>
<h3 id="Make-Targets-">Make Targets <a name="make" /><a class="anchor-link" href="#Make-Targets-">¶</a></h3><p>For all tasks we have defined the following make targets.</p>
<ul>
@@ -13184,11 +13160,13 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Task-1:-Compile-Flags-">Task 1: Compile Flags <a name="task1" /><a class="anchor-link" href="#Task-1:-Compile-Flags-">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver</p>
+<h2 id="Task-1:-Basic-compiler-optimization-flags-and-compiler-annotations-">Task 1: Basic compiler optimization flags and compiler annotations <a name="task1" /><a class="anchor-link" href="#Task-1:-Basic-compiler-optimization-flags-and-compiler-annotations-">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver</p>
<p>Your task is to:</p>
<ul>
<li>Optimize performance with <code>-Ofast</code> flag</li>
+<li>Verify the cause for performance improvement by viewing perf profiles of O3 and Ofast binaries </li>
<li>Optimize performance with profile directed feedback </li>
+<li>Generate compiler annotations/remarks to understand the optimizations done by the compiler with and without profile directed feedback </li>
</ul>
<p>First, change the working directory to <code>Task1</code>.</p>
@@ -13211,7 +13189,8 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-A:--Ofast-vs.--O3">Part A: <code>-Ofast</code> vs. <code>-O3</code><a class="anchor-link" href="#Part-A:--Ofast-vs.--O3">¶</a></h3><p>We are to compare the performance of the binary being compiled with <code>-Ofast</code> optimization and with <code>-O3</code> optimization. Right now, the Makefile specifies <code>-O3</code> as the optimization flag. Compile the code using <code>make</code> and run it with <code>make run</code> in the next two cells.</p>
+<h3 id="Part-A:--Ofast-vs.--O3">Part A: <code>-Ofast</code> vs. <code>-O3</code><a class="anchor-link" href="#Part-A:--Ofast-vs.--O3">¶</a></h3><p>We are to compare the performance of the binary being compiled with <code>-Ofast</code> optimization and with <code>-O3</code> optimization. As in the previous task, we use a <code>Makefile</code> for compilation. The <code>Makefile</code> targets <code>poisson2d_O3</code> and <code>poisson2d_Ofast</code> are already prepared.</p>
+<p><strong>TASK</strong>: Add <code>-O3</code> as the optimization flag for the <code>poisson2d_O3</code> target by using the corresponding <code>CFLAGS</code> definition. There are notes relating to this Task 1 in the header of the <code>Makefile</code>. Compile the code using <code>make</code> as indicated below and run with the <code>Make</code> targets <code>run</code>, <code>run_perf</code> and <code>run_perf_recrep</code>.</p>
</div>
</div>
@@ -13221,7 +13200,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_O3
</pre></div>
</div>
@@ -13245,7 +13224,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>You can use the GNU <em>perf</em> tool to profile the application using the <code>perf</code> command (see below) and see the top time-consuming functions.</p>
+<p>Let's have a look at the output of the <code>Makefile</code> target <code>run_perf</code>. It invokes the GNU <em>perf</em> tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to <em>perf</em>.</p>
</div>
</div>
@@ -13255,10 +13234,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="c1"># perf record creates a perf.data file </span>
-<span class="o">!</span>perf record -o perf.O3.data -e cycles ./poisson2d
-<span class="c1"># perf report opens the perf.data file </span>
-<span class="o">!</span>perf report -i perf.O3.data <span class="p">|</span> cat
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf
</pre></div>
</div>
@@ -13269,7 +13245,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p><strong>TASK</strong>: Now change the optimization flag in the <a href="/edit/Task1/Makefile">Makefile</a> to <code>-Ofast</code> and repeat the steps in the following cell. In case you follow along non-interactive, call <code>make</code> and <code>make run</code> in your shell. (If you are in the Jupyter Notebook, you can actually click the link of the <a href="/edit/Task1/Makefile">Makefile</a>. In other cases, use <code>vim</code> which is installed on Ascent.)</p>
+<p>Next we run the makefile with target <code>run_perf_recrep</code> that prints the top routines of the application in terms of hotness by using a combination of <code>perf record ./app</code> and <code>perf report</code>.</p>
</div>
</div>
@@ -13279,36 +13255,73 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make
+<div class=" highlight hl-ipython3"><pre><span></span><span class="c1"># run_perf_recrep displays the top hot routines </span>
+<span class="o">!</span>make run_perf_recrep
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p><strong>TASK</strong>: Now add the optimization flag <code>Ofast</code> to the <code>CFLAGS</code> for target <code>poisson2d_Ofast</code>. Compile the program with the target <code>poisson2d_Ofast</code> and run and analyse it as before with <code>run</code>, <code>run_perf</code> and <code>run_perf_recrep</code>.</p>
+<p>What difference do you see?</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_Ofast
+<span class="o">!</span>make run
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Again, run a <code>perf</code>-instrumented version:</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="c1"># perf record creates a perf.data file </span>
-<span class="o">!</span>perf record -o perf.Ofast.data -e cycles ./poisson2d
-<span class="c1"># perf report opens the perf.data file </span>
-<span class="o">!</span>perf report -i perf.Ofast.data <span class="p">|</span> cat
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Generate the list of top routines in terms of hotness:</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf_recrep
</pre></div>
</div>
@@ -13327,7 +13340,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h4 id="Interpretation">Interpretation<a class="anchor-link" href="#Interpretation">¶</a></h4><p>Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with <code>-Ofast</code> which enables <code>–ffast-math</code> option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the <code>-Ofast</code> binary natively implements the <code>fmax</code> function using instructions available in the hardware. The <code>-O3</code> binary makes a library call to compute <code>fmax</code> to follow a stricter <em>IEEE</em> requirement for accuracy.</p>
+<h4 id="Interpretation">Interpretation<a class="anchor-link" href="#Interpretation">¶</a></h4><p>Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with <code>-Ofast</code> which enables <code>–ffast-math</code> option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the <code>-Ofast</code> binary natively implements the <code>fmax</code> function using instructions available in the hardware. The <code>-O3</code> binary makes a library call to compute <code>fmax</code> to follow a stricter <em>IEEE</em> requirement for accuracy.</p>
</div>
</div>
@@ -13335,17 +13348,26 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-B:-Profile-directed-Feedback">Part B: Profile-directed Feedback<a class="anchor-link" href="#Part-B:-Profile-directed-Feedback">¶</a></h3><p>For the first level of optimization we saw <code>Ofast</code> cut the execution time of the <code>O3</code> binary by almost half.</p>
-<p>We can optimize the performance further by using profile directed feedback optimization.</p>
-<p>To compile using profile directed feedback with the GCC compiler we need to do the following steps</p>
+<h3 id="Part-B:-Profile-directed-Feedback">Part B: Profile-directed Feedback<a class="anchor-link" href="#Part-B:-Profile-directed-Feedback">¶</a></h3><p>For the first level of optimization we see that <code>Ofast</code> cut the execution time of the <code>O3</code> binary by almost half.</p>
+<p>We can optimize the performance further by using profile-directed feedback optimization.</p>
+<p>To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:</p>
<ol>
-<li>We need to first build a training binary using <code>-fprofile-generate</code>; this instructs the compiler to record hot path information </li>
-<li>Run the training binary with a smaller input size; you should see a <code>.gcda</code> file generated which stores hot path information for further optimization by the compiler </li>
-<li>build the final binary using <code>-fprofile-use</code> which uses the profile information in the <code>.gcda</code> file </li>
-<li>Compare the performance of the final binary with the original <code>Ofast</code> binary </li>
+<li>Instrument binary;</li>
+<li>Run binary with training, gather profile information;</li>
+<li>Use profile information to generate optimized binary.</li>
</ol>
-<p><strong>TASK</strong>: First, search for <code>TODO1</code> in the <a href="/edit/Task1/Makefile">Makefile</a>. It defines an additional compilation flag for <code>gcc</code>. Insert <code>-fprofile-generate=FOLDER</code> there with FOLDER pointing to <code>$$SC18_DIR_SCRATCH</code>, your personal write-directory (the double dollar signs are intentional as they are used to escape in the GNU Make syntax).</p>
-<p>After editing, run the following two cells to train your program.</p>
+<p>Step 1 is achieved by compiling the binary with the correct flag – <code>-fprofile-generate</code>. In our case, we need to specify an output location, which should be <code>$(SC19_DIR_SCRATCH)</code>.</p>
+<p>Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension <code>.gcda</code>) is written to the directory specified during compilation.</p>
+<p>For Step 3, the binary is once again compiled, but this time using the <code>gcda</code> profile just generated. The according flag is <code>-fprofile-use</code>, which we set to <code>$(SC19_DIR_SCRATCH)</code> as well.</p>
+<p>In our <code>Makefile</code> at hand, we prepared the steps already for you in the form of two targets.</p>
+<ul>
+<li><code>poisson2d_train</code>: Will compile the binary with profile-directed feedback</li>
+<li><code>poisson2d_ref</code>: Will take a generated profile and compile a new, optimized binary</li>
+</ul>
+<p>By using dependencies, between these two targets a profile run is launched.</p>
+<p><strong>TASK</strong>: Edit the <a href="`Makefile`">Makefile</a> and add the <code>-fprofile-*</code> flags to the <code>CFLAGS</code> of <code>poisson2d_train</code> and
+<code>poisson2d_ref</code> as outline in the file.</p>
+<p>After that, you may launch them with the following cells (<code>gen_profile</code> is a meta-target and uses <code>poisson2d_train</code> and <code>poisson2d_ref</code>). If you need to clean the generated profile, you may use <code>make clean_profile</code>.</p>
</div>
</div>
@@ -13355,20 +13377,28 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_train
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make gen_profile
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_train
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
</pre></div>
</div>
@@ -13379,9 +13409,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>Now, a <code>.gcda</code> file exists in the directory which can be used for an profile-accelerated subsequent run.</p>
-<p><strong>TASK</strong>: Edit the <a href="/edit/Task1/Makefile">Makefile</a> again, this time modifying <code>TODO2</code> to be equivalent to <code>-fprofile-use</code>. A directory is not needed as we copied the gcda file into the current directory.</p>
-<p>Run the following cells in order to build using the newly added flag and then run with the profile-accelerated version.</p>
+<p>Let's also measure instructions and cycles</p>
</div>
</div>
@@ -13391,20 +13419,38 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_profile
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-C:-Compiler-annotations/Remarks">Part C: Compiler annotations/Remarks<a class="anchor-link" href="#Part-C:-Compiler-annotations/Remarks">¶</a></h3><p>Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done.</p>
+<p>To generate compiler annotations using GCC, one uses <code>-fopt-info-all</code>. If you only want to see the missed options, use the option <code>-fopt-info-missed</code> instead of <code>-fopt-info-all</code>. See also the <a href="https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info">documentation of GCC regarding the flag</a>.</p>
+<p><strong>TASK</strong>: Have a looK at the <code>CFLAGS</code> of the <code>Makefile</code> target <code>poisson2d_Ofast_info</code>. Add the flag <code>-fopt-info-all</code> to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use – for example – <code>-fopt-info-all=(SC19_DIR_SCRATCH)/filename</code>.</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_profile
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_Ofast_info
</pre></div>
</div>
@@ -13415,7 +13461,34 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)</p>
+<p>Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.</p>
+<p><strong>TASK</strong>:
+Adapt the <code>CFLAGS</code> of <code>poisson2d_ref_info</code> to include <code>-fopt-info-all</code> <strong>and</strong> the profile input of <code>-fprofile-use=…</code> here. <em>(Be advised: Long output!)</em></p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_ref_info
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Comparing the annotations generated of a plain <code>-Ofast</code> optimization level and the one generated at <code>-Ofast</code> and profile directed feedback, we observe that many more optimizations are possible due to profile information.</p>
+<p>For instance you will see annotations such as</p>
+
+<pre><code>poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)</code></pre>
+<p>The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations.</p>
</div>
</div>
@@ -13443,8 +13516,12 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Task-2:-Software-Pretechting">Task 2:<a name="task2" /> Software Pretechting<a class="anchor-link" href="#Task-2:-Software-Pretechting">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>Study the difference of program execution time of different optimization levels with and without software prefetching.</p>
-<p>First, change directory to that of Task 2</p>
+<h2 id="Task-2:-Impact-of-Prefetching-on-Performance">Task 2:<a name="task2" /> Impact of Prefetching on Performance<a class="anchor-link" href="#Task-2:-Impact-of-Prefetching-on-Performance">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><ul>
+<li>Study the difference of program execution time of different optimization levels with and without software prefetching.</li>
+<li>Verify the impact by measuring cache counters with and without prefetching.</li>
+<li>Learn how to modify contents of DSCR (<em>Data Stream Control Register</em>) using IBM XL compiler and study the impact with different values to DSCR. </li>
+</ul>
+<p>But first, lets change directory to that of Task 2</p>
</div>
</div>
@@ -13465,15 +13542,18 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-A:-Running">Part A: Running<a class="anchor-link" href="#Part-A:-Running">¶</a></h3>
+<h3 id="Part-A:-Software-Prefetching">Part A: Software Prefetching<a class="anchor-link" href="#Part-A:-Software-Prefetching">¶</a></h3>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>Look at the <a href="/edit/Task2/Makefile">Makefile</a> and work on the TODOs. Please implement compile flags as mentioned in the Makefile target name.</p>
-<p>Afterwards, compile each target with the following cells and submit them to the batch system. Follow along accordingly in the non-interactive version of this Notebook.</p>
+<p><strong>TASK</strong>: Look at the Makefile and work on the TODOs.</p>
+<ul>
+<li>First generate a <code>-Ofast</code>-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!</li>
+<li>Modify the <code>Makefile</code> to add the option for software prefetching (<code>-fprefetch-loop-arrays</code>). Compare performance of <code>-Ofast</code> with and without software prefetching</li>
+</ul>
</div>
</div>
@@ -13483,7 +13563,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_o3_pref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make clean
</pre></div>
</div>
@@ -13496,7 +13576,9 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_ofast_pref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d <span class="nv">CC</span><span class="o">=</span>gcc
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
@@ -13509,31 +13591,66 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_o3_nopref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_pref <span class="nv">CC</span><span class="o">=</span>gcc
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p><strong>TASK</strong>: Repeat the experiment with the <code>-O3</code> flag. Have a look at the <code>Makefile</code> and the outlined TODO. There's a position to easily adapt <code>-Ofast</code>→<code>-O3</code>!</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_ofast_nopref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d <span class="nv">CC</span><span class="o">=</span>gcc -B
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_pref <span class="nv">CC</span><span class="o">=</span>gcc -B
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>Do you notice the impact difference with optimization levels? It's always important to carefully study the interplay of flags.</p>
+<h3 id="Part-B:-Analysis-of-Instructions">Part B: Analysis of Instructions<a class="anchor-link" href="#Part-B:-Analysis-of-Instructions">¶</a></h3><p>Compilation of the <code>-Ofast</code> binary with the software prefetching flag causes the compiler to generate the <code>dcb*</code> instructions that prefetch memory values to L3.</p>
</div>
</div>
@@ -13541,8 +13658,10 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-B:-Analysis-of-Instructions">Part B: Analysis of Instructions<a class="anchor-link" href="#Part-B:-Analysis-of-Instructions">¶</a></h3><p>Compilation with the software prefetching flag causes the compiler to generate the <code>__dcbt</code> and <code>__dcbtst</code> instructions that prefetch memory values to L3.</p>
-<p>Verify it using <code>objdump -lSd</code> on each file (<code>poisson2d_o3_pref</code>, <code>poisson2d_ofast_pref</code>, <code>poisson2d_o3_nopref</code>, <code>poisson2d_ofast_nopref</code>). You might want to grep for <code>dcb</code>.</p>
+<p><strong>TASK</strong>:
+Run <code>$(SC19_SUBMIT_CMD) objdump -lSd</code> on each binary file (<code>-O3</code>, <code>-Ofast</code> with prefetch/no prefetch).
+Look for instructions beginning with <code>dcb</code>
+At what optimization levels does the compiler generate software prefetching instructions?</p>
</div>
</div>
@@ -13552,7 +13671,60 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="ch">#!objdump -l…</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make <span class="nv">CC</span><span class="o">=</span>gcc -B poisson2d_pref
+<span class="o">!</span>objdump -lSd ./poisson2d_pref > poisson2d.dis
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>grep dcb poisson2d.dis
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-C:-Changing-Values-of-DSCR-via-compiler-flags">Part C: Changing Values of DSCR via compiler flags<a class="anchor-link" href="#Part-C:-Changing-Values-of-DSCR-via-compiler-flags">¶</a></h3><p>This task requires using the IBM XL compiler. It should be already in your environment.</p>
+<p>We saw the impact of software prefetching in the previous subsection.
+In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance.
+In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching.</p>
+<p>IBM XL compiler has an option <code>-qprefetch=dscr=<val></code> that can be used for this purpose.
+Compiling with <code>-qprefetch=dscr=1</code> turns off the prefetcher. One can give various values such as <code>-qprefetch=dscr=4</code>, <code>-qprefetch=dscr=7</code> etc. to control aggressiveness of prefetching.</p>
+<p>For this exercise we use <code>make CC=xlc_r</code> to illustrate the performance impact.</p>
+<p><strong>Task</strong> Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the <code>Makefile</code>: Add <code>qprefetch=dscr=1</code> to the <code>CFLAGS</code> and rebuild the application and note the performance. Which one is faster?</p>
+<p>In general, applications benefit with the default settings of hardware DSCR register (<code>-qprefetch=dscr=0</code>). However, certain applications also benefit with prefetching turned off.</p>
+<p>It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Measure performance of the application compiled with XL at default DSCR value</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make <span class="nv">CC</span><span class="o">=</span>xlc_r -B poisson2d
+<span class="o">!</span>make run
</pre></div>
</div>
@@ -13563,7 +13735,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>If you feel up to the task, you can study the number of L3 cache misses using the corresponding performance counter, <code>PM_L3_MISS</code>. Either use your knowledge from Hands-On 1, or use the following call to <code>perf</code>, in which we already converted the named counter to a raw counter address.</p>
+<p>Measure performance of the application compiled with XL with DSCR value turned off</p>
</div>
</div>
@@ -13573,14 +13745,22 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"poisson2d_ofast_nopref"</span><span class="p">,</span> <span class="s2">"poisson2d_ofast_pref"</span><span class="p">]:</span>
- <span class="o">!</span><span class="nb">eval</span> <span class="nv">$$</span>SC18_SUBMIT_CMD perf stat -e cycles,r168a4 ./<span class="nv">$f</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_dscr <span class="nv">CC</span><span class="o">=</span>xlc_r -B
+<span class="o">!</span>make run
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Does Hardware prefetcher help this application? How much impact do you see when you turn off the hardware prefetcher?</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
@@ -13588,6 +13768,7 @@ build <code>poisson2d</code> binary (default)</li>
<h4 id="References">References<a class="anchor-link" href="#References">¶</a></h4><ol>
<li><a href="https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html">https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html</a></li>
<li><a href="https://www.gnu.org/software/gcc/projects/prefetch.html">https://www.gnu.org/software/gcc/projects/prefetch.html</a></li>
+<li><a href="https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0">https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0</a></li>
</ol>
</div>
@@ -13606,8 +13787,8 @@ build <code>poisson2d</code> binary (default)</li>
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<h2 id="Task-3:-OpenMP">Task 3: OpenMP<a class="anchor-link" href="#Task-3:-OpenMP">¶</a></h2><p><a name="task3"></a></p>
-<h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores.</p>
-<p>First, we need to change directory to that of Task3.</p>
+<h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores on the resulting application performance. We do this study for both GCC and XL compilers inorder to learn about the appropriate options that need to be used.
+First, we need to change directory to that of Task3. For Task 3 we modify poisson2d.c to invoke an exact copy of the main jacobi loop which is <code>poisson2d_reference</code>. We parallelize only the main loop but not <code>poisson2d_reference</code>. The speedup is the performance gain seen in the main loop as compared to the reference loop.</p>
</div>
</div>
@@ -13628,13 +13809,12 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-A:-Implement-OpenMP-Pragmas;-Compilation">Part A: Implement OpenMP Pragmas; Compilation<a class="anchor-link" href="#Part-A:-Implement-OpenMP-Pragmas;-Compilation">¶</a></h3><p><strong>Task</strong>: Please add the correct OpenMP pragmas to the source code and compilations flags to enable OpenMP.</p>
+<h3 id="Part-A:-Implement-OpenMP-Pragmas;-Compilation">Part A: Implement OpenMP Pragmas; Compilation<a class="anchor-link" href="#Part-A:-Implement-OpenMP-Pragmas;-Compilation">¶</a></h3><p><strong>Task</strong>: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.</p>
<ul>
-<li><strong>pragmas</strong>: Look at the TODOs in <a href="/edit/Task3/poisson2d.c"><code>poisson2d.c</code></a> to add OpenMP parallelism. The pragmas in question are <code>#pragma omp parallel for</code></li>
-<li><strong>Compilation</strong>: Please add compilation flags enabling OpenMP in GCC to the <a href="/edit/Task3/Makefile">Makefile</a>. The flag in question is <code>-fopenmp</code>.</li>
+<li><strong>Directives</strong>: Look at the TODOs in <a href="poisson2d.c"><code>poisson2d.c</code></a> to add OpenMP parallelism. The pragmas in question are <code>#pragma omp parallel for</code> (and once it's <code>#pragma omp parallel for reduction(max:error)</code> – can you guess where?)</li>
+<li><strong>Compilation</strong>: Please add compilation flags enabling OpenMP in GCC and XL to the <code>Makefile</code>. For GCC, we need to add <code>-fopenmp</code> and the application needs to be linked with <code>-lgomp</code>. For XL, we need to add <code>-qsmp=omp</code> to the list of compilation flags. </li>
</ul>
-<p>Edit the files with the links above if you are running the interactive version of the Notebook or navigate to <code>poisson2d.c</code> and <code>Makefile</code> yourself in case you run the non-interactive version.</p>
-<p>Afterwards, compile and run the application with the following cells. Non-interactive: Follow along accordingly in the shell.</p>
+<p>Afterwards, compile and run the application with the following commands.</p>
</div>
</div>
@@ -13644,20 +13824,28 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d <span class="nv">CC</span><span class="o">=</span>gcc
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>The command to submit a job to the batch system is prepared in an environment variable <code>$SC19_SUBMIT_CMD</code>; use it together with <code>eval</code>. In the following cell, it is shown how to invoke the application using the batch system.</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC19_SUBMIT_CMD</span> ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span>
</pre></div>
</div>
@@ -13668,7 +13856,9 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>The command to submit a job to the batch system is prepared in an environment variable <code>$SC18_SUBMIT_CMD</code>; use it together with <code>eval</code>. In the following cell, it is shown how to increase the work of the application.</p>
+<p>Inorder to run the parallel application, we need to set the number of threads using <code>OMP_NUM_THREADS</code>
+What is the best performance you can reach by setting the number of threads via <code>OMP_NUM_THREADS=N</code> with <code>N</code> being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example.<br>
+We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler of Ascent, from overlaying binding options. Also, we use <code>-c ALL_CPUS</code> to make all CPUs on the compute nodes available to you.</p>
</div>
</div>
@@ -13678,7 +13868,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC18_SUBMIT_CMD</span> ./poisson2d <span class="m">1000</span> <span class="m">1000</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span>N <span class="nv">$SC19_SUBMIT_CMD</span> -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span>
</pre></div>
</div>
@@ -13689,8 +13879,13 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>What is the best performance you can reach by setting the number of threads via <code>OMP_NUM_THREADS=N</code> with <code>N</code> being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example.<br>
-We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler of Ascent, from overlaying binding options. Also, we use <code>-c ALL_CPUS</code> to make all CPUs on the compute nodes available to you.</p>
+<h3 id="Part-B:-Bindings">Part B: Bindings<a class="anchor-link" href="#Part-B:-Bindings">¶</a></h3><p>Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!</p>
+<p>There are applications which can be used to determine the configuration of the processor. Among those are:</p>
+<ul>
+<li><code>lscpu</code>: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.</li>
+<li><code>ppc64_cpu --smt</code>: Specifically for POWER, this tool can give information about the number of simulations threads running per core (<em>SMT</em>, Simulataion Multi-Threading).</li>
+</ul>
+<p>Run <code>ppc64_cpu --smt</code> to find out about the threading configuration of Ascent!</p>
</div>
</div>
@@ -13700,7 +13895,7 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">1</span> <span class="nv">$SC18_SUBMIT_CMD</span> -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC19_SUBMIT_CMD</span> ppc64_cpu --smt
</pre></div>
</div>
@@ -13711,13 +13906,11 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-B:-Bindings">Part B: Bindings<a class="anchor-link" href="#Part-B:-Bindings">¶</a></h3><p>Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!</p>
-<p>There are applications which can be used to determine the configuration of the processor. Among those are:</p>
+<p>There are more sources information available</p>
<ul>
-<li><code>lscpu</code>: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.</li>
-<li><code>ppc64_cpu --smt</code>: Specifically for POWER, this tool can give information about the number of simulations threads running per core (<em>SMT</em>, Simulataion Multi-Threading).</li>
+<li><code>/proc/cpuinfo</code>: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with <code>cat</code></li>
+<li><code>/sys/devices/system/cpu/cpu0/topology/thread_siblings_list</code>: Holds information about thread siblings for given CPU core (<code>cpu0</code> in this case). Use it to find out which thread is mapped to which core.</li>
</ul>
-<p>Run <code>ppc64_cpu --smt</code> to find out about the threading configuration of Ascent!</p>
</div>
</div>
@@ -13727,7 +13920,8 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC18_SUBMIT_CMD</span> ppc64_cpu --smt
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">$$</span>SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list
+<span class="o">!</span><span class="nv">$$</span>SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list
</pre></div>
</div>
@@ -13738,11 +13932,47 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>There are more sources information available</p>
-<ul>
-<li><code>/proc/cpuinfo</code>: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with <code>cat</code></li>
-<li><code>/sys/devices/system/cpu/cpu0/topology/thread_siblings_list</code>: Holds information about thread siblings for given CPU core (<code>cpu0</code> in this case). Use it to find out which thread is mapped to which core.</li>
-</ul>
+<p>There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the <a href="https://www.openmp.org/spec-html/5.0/openmpse53.html">OMP_PLACES environment Variable</a>. We also have a GNU specific variable which can also be used to control affinity - <code>GOMP_CPU_AFFINITY</code>. Setting <code>GOMP_CPU_AFFINITY</code> is specific to GCC binaries but it internally serves the same function as setting <code>OMP_PLACES</code>.</p>
+<p><strong>Task</strong>: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.</p>
+<p>Adapt the following command with your configuration – or follow along accordingly in the non-interactive version of the Notebook.</p>
+<p>What's your maximum speedup?</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">OMP_PLACES</span><span class="o">=</span><span class="s2">"{X},{Y},{Z},{A}"</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">100</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">GOMP_CPU_AFFINITY</span><span class="o">=</span><span class="s2">"X,Y,Z,A"</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">100</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Great!</p>
+<p>If you still have time: The same experiments can be repeated with the IBM XL compiler.
+The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is <code>-qsmp=omp</code></p>
+<p><strong>Task</strong>: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup.</p>
</div>
</div>
@@ -13752,8 +13982,7 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list
-<span class="o">!</span>cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make <span class="nv">CC</span><span class="o">=</span>xlc_r -B run
</pre></div>
</div>
@@ -13764,9 +13993,31 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>There are various environment variables available within OpenMP (and GCC) to specify binding of threads to cores. See, for instance, the <a href="https://gcc.gnu.org/onlinedocs/libgomp/Environment-Variables.html">online documentation of GCC libgomp</a>. Examples are <code>OMP_PLACES</code> or <code>GOMP_CPU_AFFINITY</code>.</p>
+<p>Run the parallel application with varying numbre of threads (<code>OMP_NUM_THREADS</code>) and note the performance improvement.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [ ]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span>N <span class="nv">$SC19_SUBMIT_CMD</span> -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span>
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Now we repeat the exercise of using the right binding of threads for the XL binary. <code>OMP_PLACES</code> pertains to the XL binary as well as it is an OpenMP variable. <code>GOMP_CPU_AFFINITY</code> is specific to GCC binary so that cannot be used to set the binding.</p>
<p><strong>Task</strong>: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.</p>
<p>Adapt the following command with your configuration – or follow along accordingly in the non-interactive version of the Notebook.</p>
+<p>We are mixing Python with Bash (<code>!</code>) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two <code>$$</code>)</p>
<p>What's your maximum speedup?</p>
</div>
@@ -13777,19 +14028,30 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="prompt input_prompt">In [ ]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">GOMP_CPU_AFFINITY</span><span class="o">=</span><span class="s2">"X,Y,Z,A"</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">100</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">affinity</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"</span><span class="si">{X}</span><span class="s2">,</span><span class="si">{Y}</span><span class="s2">,</span><span class="si">{Z}</span><span class="s2">,</span><span class="si">{A}</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"</span><span class="si">{P}</span><span class="s2">,</span><span class="si">{Q}</span><span class="s2">,</span><span class="si">{R}</span><span class="s2">,</span><span class="si">{S}</span><span class="s2">"</span><span class="p">]:</span>
+ <span class="nb">print</span><span class="p">(</span><span class="s2">"Affinity: </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">affinity</span><span class="p">))</span>
+ <span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">OMP_PLACES</span><span class="o">=</span><span class="nv">$affinity</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
</pre></div>
</div>
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Likewise we see a higher speedup when we bind the threads to different cores rather than to a single core. This handson illustrates that apart from compiler level tuning, system level tuning is also equally important to obtain performance improvements</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<h4 id="References">References<a class="anchor-link" href="#References">¶</a></h4><ol>
<li><a href="https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html">https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html</a></li>
+<li><a href="https://www.openmp.org/spec-html/5.0/openmpse53.html">https://www.openmp.org/spec-html/5.0/openmpse53.html</a></li>
</ol>
</div>
@@ -13807,7 +14069,7 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h1 id="Survey">Survey<a name="survey" /><a class="anchor-link" href="#Survey">¶</a></h1><p>Please rememeber to take some time and fill out the <a href="http://bit.ly/sc18-eval">survey</a>.</p>
+<h1 id="Survey">Survey<a name="survey" /><a class="anchor-link" href="#Survey">¶</a></h1><p>Please rememeber to take some time and fill out the <a href="http://bit.ly/sc19-eval">survey</a>.</p>
</div>
</div>
diff --git a/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.ipynb b/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.ipynb
index a28d887..64fdf69 100644
--- a/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.ipynb
+++ b/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.ipynb
@@ -5,7 +5,7 @@
"metadata": {},
"source": [
"# Hands-On Performance Optimization\n",
- "_Supercomputing 2018 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 12th 2018_\n",
+ "_Supercomputing 2019 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 18th 2019_\n",
"\n",
"---"
]
@@ -22,7 +22,7 @@
"\n",
"You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n",
"\n",
- "If you want you also can get a [terminal](/terminals/1) in your browser.\n",
+ "If you want you also can get a terminal in your browser; just open it via the »New Launcher« button (`+`).\n",
"\n",
"## Terminal fallback\n",
"\n",
@@ -37,7 +37,16 @@
"source": [
"## Setup\n",
"\n",
- "This hands-on session requires of GCC 6.4.0. By loading the `sc18/handson2` module before invoking this Notebook, we took care of also loading GCC 6.4.0 into the environment."
+ "We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!gcc --version"
]
},
{
@@ -51,14 +60,16 @@
"Please choose from the task below.\n",
"\n",
"\n",
- "* [Task 1](#task1): Compile Flags \n",
- "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback ([Solution 1](#solution0))\n",
+ "* [Task 1](#task1): __Basic compiler optimization flags and compiler annotations__\n",
"\n",
- "* [Task 2](#task2): Software Prefetching \n",
- "Improve performance of the CPU Jacobi solver with software prefetching ([Solution 2](#solution1))\n",
+ "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback. Learn about compiler annotations.\n",
"\n",
- "* [Task 3](#task3): OpenMP \n",
- "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance ([Solution 3](#solution2))\n",
+ "* [Task 2](#task2): __Optimization via Prefetching controlled by compiler__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with software prefetching. Some compilers such as IBM XL define flags that can be used to modify the aggressiveness of the hardware prefetcher. Learn to modify the DSCR value through XL and study the impact on application performance. \n",
+ "* [Task 3](#task3): __Optimization via OpenMP controlled by compiler and the system__\n",
+ "\n",
+ "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance. \n",
" \n",
"* [Suvery](#survey) Please remember to take the survey !\n",
" \n",
@@ -85,7 +96,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "## Task 1: Compile Flags <a name=\"task1\"></a>\n",
+ "## Task 1: Basic compiler optimization flags and compiler annotations <a name=\"task1\"></a>\n",
"\n",
"\n",
"### Overview\n",
@@ -95,7 +106,10 @@
"Your task is to:\n",
"\n",
"* Optimize performance with `-Ofast` flag\n",
+ "* Verify the cause for performance improvement by viewing perf profiles of O3 and Ofast binaries \n",
"* Optimize performance with profile directed feedback \n",
+ "* Generate compiler annotations/remarks to understand the optimizations done by the compiler with and without profile directed feedback \n",
+ "\n",
"\n",
"First, change the working directory to `Task1`."
]
@@ -115,7 +129,9 @@
"source": [
"### Part A: `-Ofast` vs. `-O3`\n",
"\n",
- "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. Right now, the Makefile specifies `-O3` as the optimization flag. Compile the code using `make` and run it with `make run` in the next two cells."
+ "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. As in the previous task, we use a `Makefile` for compilation. The `Makefile` targets `poisson2d_O3` and `poisson2d_Ofast` are already prepared. \n",
+ "\n",
+ "**TASK**: Add `-O3` as the optimization flag for the `poisson2d_O3` target by using the corresponding `CFLAGS` definition. There are notes relating to this Task 1 in the header of the `Makefile`. Compile the code using `make` as indicated below and run with the `Make` targets `run`, `run_perf` and `run_perf_recrep`. "
]
},
{
@@ -124,7 +140,7 @@
"metadata": {},
"outputs": [],
"source": [
- "!make"
+ "!make poisson2d_O3"
]
},
{
@@ -140,7 +156,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "You can use the GNU _perf_ tool to profile the application using the `perf` command (see below) and see the top time-consuming functions."
+ "Let's have a look at the output of the `Makefile` target `run_perf`. It invokes the GNU _perf_ tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to _perf_."
]
},
{
@@ -149,17 +165,14 @@
"metadata": {},
"outputs": [],
"source": [
- "# perf record creates a perf.data file \n",
- "!perf record -o perf.O3.data -e cycles ./poisson2d\n",
- "# perf report opens the perf.data file \n",
- "!perf report -i perf.O3.data | cat"
+ "!make run_perf"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "**TASK**: Now change the optimization flag in the [Makefile](/edit/Task1/Makefile) to `-Ofast` and repeat the steps in the following cell. In case you follow along non-interactive, call `make` and `make run` in your shell. (If you are in the Jupyter Notebook, you can actually click the link of the [Makefile](/edit/Task1/Makefile). In other cases, use `vim` which is installed on Ascent.)"
+ "Next we run the makefile with target `run_perf_recrep` that prints the top routines of the application in terms of hotness by using a combination of `perf record ./app` and `perf report`. "
]
},
{
@@ -168,7 +181,17 @@
"metadata": {},
"outputs": [],
"source": [
- "!make"
+ "# run_perf_recrep displays the top hot routines \n",
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Now add the optimization flag `Ofast` to the `CFLAGS` for target `poisson2d_Ofast`. Compile the program with the target `poisson2d_Ofast` and run and analyse it as before with `run`, `run_perf` and `run_perf_recrep`.\n",
+ "\n",
+ "What difference do you see?"
]
},
{
@@ -177,19 +200,40 @@
"metadata": {},
"outputs": [],
"source": [
+ "!make poisson2d_Ofast \n",
"!make run"
]
},
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Again, run a `perf`-instrumented version:"
+ ]
+ },
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
- "# perf record creates a perf.data file \n",
- "!perf record -o perf.Ofast.data -e cycles ./poisson2d\n",
- "# perf report opens the perf.data file \n",
- "!perf report -i perf.Ofast.data | cat"
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Generate the list of top routines in terms of hotness:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!make run_perf_recrep"
]
},
{
@@ -205,7 +249,7 @@
"source": [
"#### Interpretation\n",
"\n",
- "Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with `-Ofast` which enables `–ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
+ "Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with `-Ofast` which enables `–ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
]
},
{
@@ -214,20 +258,34 @@
"source": [
"### Part B: Profile-directed Feedback\n",
"\n",
- "For the first level of optimization we saw `Ofast` cut the execution time of the `O3` binary by almost half.\n",
+ "For the first level of optimization we see that `Ofast` cut the execution time of the `O3` binary by almost half.\n",
+ "\n",
+ "We can optimize the performance further by using profile-directed feedback optimization.\n",
+ "\n",
+ "To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:\n",
+ "\n",
+ "1. Instrument binary;\n",
+ "2. Run binary with training, gather profile information;\n",
+ "3. Use profile information to generate optimized binary.\n",
+ "\n",
+ "\n",
+ "Step 1 is achieved by compiling the binary with the correct flag – `-fprofile-generate`. In our case, we need to specify an output location, which should be `$(SC19_DIR_SCRATCH)`.\n",
+ "\n",
+ "Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension `.gcda`) is written to the directory specified during compilation.\n",
"\n",
- "We can optimize the performance further by using profile directed feedback optimization.\n",
+ "For Step 3, the binary is once again compiled, but this time using the `gcda` profile just generated. The according flag is `-fprofile-use`, which we set to `$(SC19_DIR_SCRATCH)` as well.\n",
"\n",
- "To compile using profile directed feedback with the GCC compiler we need to do the following steps\n",
+ "In our `Makefile` at hand, we prepared the steps already for you in the form of two targets.\n",
"\n",
- "1. We need to first build a training binary using `-fprofile-generate`; this instructs the compiler to record hot path information \n",
- "2. Run the training binary with a smaller input size; you should see a `.gcda` file generated which stores hot path information for further optimization by the compiler \n",
- "3. build the final binary using `-fprofile-use` which uses the profile information in the `.gcda` file \n",
- "4. Compare the performance of the final binary with the original `Ofast` binary \n",
+ "* `poisson2d_train`: Will compile the binary with profile-directed feedback\n",
+ "* `poisson2d_ref`: Will take a generated profile and compile a new, optimized binary\n",
"\n",
- "**TASK**: First, search for `TODO1` in the [Makefile](/edit/Task1/Makefile). It defines an additional compilation flag for `gcc`. Insert `-fprofile-generate=FOLDER` there with FOLDER pointing to `$$SC18_DIR_SCRATCH`, your personal write-directory (the double dollar signs are intentional as they are used to escape in the GNU Make syntax).\n",
+ "By using dependencies, between these two targets a profile run is launched.\n",
"\n",
- "After editing, run the following two cells to train your program."
+ "**TASK**: Edit the [Makefile](`Makefile`) and add the `-fprofile-*` flags to the `CFLAGS` of `poisson2d_train` and\n",
+ "`poisson2d_ref` as outline in the file.\n",
+ "\n",
+ "After that, you may launch them with the following cells (`gen_profile` is a meta-target and uses `poisson2d_train` and `poisson2d_ref`). If you need to clean the generated profile, you may use `make clean_profile`."
]
},
{
@@ -236,7 +294,14 @@
"metadata": {},
"outputs": [],
"source": [
- "!make poisson2d_train"
+ "!make gen_profile"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!"
]
},
{
@@ -245,18 +310,43 @@
"metadata": {},
"outputs": [],
"source": [
- "!make run_train"
+ "!make run"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "Now, a `.gcda` file exists in the directory which can be used for an profile-accelerated subsequent run.\n",
+ "Let's also measure instructions and cycles"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Compiler annotations/Remarks\n",
"\n",
- "**TASK**: Edit the [Makefile](/edit/Task1/Makefile) again, this time modifying `TODO2` to be equivalent to `-fprofile-use`. A directory is not needed as we copied the gcda file into the current directory.\n",
+ "Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done. \n",
"\n",
- "Run the following cells in order to build using the newly added flag and then run with the profile-accelerated version."
+ "To generate compiler annotations using GCC, one uses `-fopt-info-all`. If you only want to see the missed options, use the option `-fopt-info-missed` instead of `-fopt-info-all`. See also the [documentation of GCC regarding the flag](https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info).\n",
+ "\n",
+ "**TASK**: Have a looK at the `CFLAGS` of the `Makefile` target `poisson2d_Ofast_info`. Add the flag `-fopt-info-all` to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use – for example – `-fopt-info-all=(SC19_DIR_SCRATCH)/filename`."
]
},
{
@@ -265,7 +355,17 @@
"metadata": {},
"outputs": [],
"source": [
- "!make poisson2d_profile"
+ "!make poisson2d_Ofast_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.\n",
+ "\n",
+ "**TASK**: \n",
+ "Adapt the `CFLAGS` of `poisson2d_ref_info` to include `-fopt-info-all` **and** the profile input of `-fprofile-use=…` here. *(Be advised: Long output!)*"
]
},
{
@@ -274,14 +374,21 @@
"metadata": {},
"outputs": [],
"source": [
- "!make run_profile"
+ "!make poisson2d_ref_info"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
+ "Comparing the annotations generated of a plain `-Ofast` optimization level and the one generated at `-Ofast` and profile directed feedback, we observe that many more optimizations are possible due to profile information.\n",
+ "\n",
+ "For instance you will see annotations such as\n",
+ "```\n",
+ "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "```\n",
+ "\n",
+ "The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations."
]
},
{
@@ -307,14 +414,16 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "## Task 2:<a name=\"task2\"></a> Software Pretechting\n",
+ "## Task 2:<a name=\"task2\"></a> Impact of Prefetching on Performance\n",
"\n",
"\n",
"### Overview\n",
"\n",
- "Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
+ "* Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
+ "* Verify the impact by measuring cache counters with and without prefetching.\n",
+ "* Learn how to modify contents of DSCR (*Data Stream Control Register*) using IBM XL compiler and study the impact with different values to DSCR. \n",
"\n",
- "First, change directory to that of Task 2"
+ "But first, lets change directory to that of Task 2"
]
},
{
@@ -330,16 +439,26 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "### Part A: Running"
+ "### Part A: Software Prefetching"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "Look at the [Makefile](/edit/Task2/Makefile) and work on the TODOs. Please implement compile flags as mentioned in the Makefile target name.\n",
+ "**TASK**: Look at the Makefile and work on the TODOs. \n",
"\n",
- "Afterwards, compile each target with the following cells and submit them to the batch system. Follow along accordingly in the non-interactive version of this Notebook."
+ "- First generate a `-Ofast`-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!\n",
+ "- Modify the `Makefile` to add the option for software prefetching (`-fprefetch-loop-arrays`). Compare performance of `-Ofast` with and without software prefetching"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!make clean"
]
},
{
@@ -348,7 +467,9 @@
"metadata": {},
"outputs": [],
"source": [
- "!make run_o3_pref"
+ "!make poisson2d CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
]
},
{
@@ -357,7 +478,16 @@
"metadata": {},
"outputs": [],
"source": [
- "!make run_ofast_pref"
+ "!make poisson2d_pref CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Repeat the experiment with the `-O3` flag. Have a look at the `Makefile` and the outlined TODO. There's a position to easily adapt `-Ofast`→`-O3`!"
]
},
{
@@ -366,7 +496,9 @@
"metadata": {},
"outputs": [],
"source": [
- "!make run_o3_nopref"
+ "!make poisson2d CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
]
},
{
@@ -375,14 +507,16 @@
"metadata": {},
"outputs": [],
"source": [
- "!make run_ofast_nopref"
+ "!make poisson2d_pref CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "Do you notice the impact difference with optimization levels? It's always important to carefully study the interplay of flags."
+ "Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?"
]
},
{
@@ -391,27 +525,86 @@
"source": [
"### Part B: Analysis of Instructions\n",
"\n",
- "Compilation with the software prefetching flag causes the compiler to generate the `__dcbt` and `__dcbtst` instructions that prefetch memory values to L3.\n",
+ "Compilation of the `-Ofast` binary with the software prefetching flag causes the compiler to generate the `dcb*` instructions that prefetch memory values to L3."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: \n",
+ "Run `$(SC19_SUBMIT_CMD) objdump -lSd` on each binary file (`-O3`, `-Ofast` with prefetch/no prefetch).\n",
+ "Look for instructions beginning with `dcb`\n",
+ "At what optimization levels does the compiler generate software prefetching instructions?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!make CC=gcc -B poisson2d_pref\n",
+ "!objdump -lSd ./poisson2d_pref > poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!grep dcb poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Changing Values of DSCR via compiler flags\n",
+ "\n",
+ "This task requires using the IBM XL compiler. It should be already in your environment.\n",
+ "\n",
+ "\n",
+ "We saw the impact of software prefetching in the previous subsection. \n",
+ "In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance. \n",
+ "In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching. \n",
+ "\n",
+ "IBM XL compiler has an option `-qprefetch=dscr=<val>` that can be used for this purpose.\n",
+ "Compiling with `-qprefetch=dscr=1` turns off the prefetcher. One can give various values such as `-qprefetch=dscr=4`, `-qprefetch=dscr=7` etc. to control aggressiveness of prefetching.\n",
+ "\n",
+ "For this exercise we use `make CC=xlc_r` to illustrate the performance impact.\n",
+ " \n",
+ "\n",
+ "**Task** Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the `Makefile`: Add `qprefetch=dscr=1` to the `CFLAGS` and rebuild the application and note the performance. Which one is faster? \n",
"\n",
- "Verify it using `objdump -lSd` on each file (`poisson2d_o3_pref`, `poisson2d_ofast_pref`, `poisson2d_o3_nopref`, `poisson2d_ofast_nopref`). You might want to grep for `dcb`."
+ "In general, applications benefit with the default settings of hardware DSCR register (`-qprefetch=dscr=0`). However, certain applications also benefit with prefetching turned off. \n",
+ "\n",
+ "It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL at default DSCR value"
]
},
{
"cell_type": "code",
"execution_count": null,
- "metadata": {
- "sc18": "task"
- },
+ "metadata": {},
"outputs": [],
"source": [
- "#!objdump -l…"
+ "!make CC=xlc_r -B poisson2d\n",
+ "!make run"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "If you feel up to the task, you can study the number of L3 cache misses using the corresponding performance counter, `PM_L3_MISS`. Either use your knowledge from Hands-On 1, or use the following call to `perf`, in which we already converted the named counter to a raw counter address."
+ "Measure performance of the application compiled with XL with DSCR value turned off"
]
},
{
@@ -420,8 +613,15 @@
"metadata": {},
"outputs": [],
"source": [
- "for f in [\"poisson2d_ofast_nopref\", \"poisson2d_ofast_pref\"]:\n",
- " !eval $$SC18_SUBMIT_CMD perf stat -e cycles,r168a4 ./$f\n"
+ "!make poisson2d_dscr CC=xlc_r -B\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Does Hardware prefetcher help this application? How much impact do you see when you turn off the hardware prefetcher? "
]
},
{
@@ -431,7 +631,8 @@
"#### References\n",
"\n",
"1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
- "2. https://www.gnu.org/software/gcc/projects/prefetch.html"
+ "2. https://www.gnu.org/software/gcc/projects/prefetch.html\n",
+ "3. https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0\n"
]
},
{
@@ -453,20 +654,14 @@
"\n",
"### Overview\n",
"\n",
- "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores.\n",
- "\n",
- "First, we need to change directory to that of Task3."
+ "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores on the resulting application performance. We do this study for both GCC and XL compilers inorder to learn about the appropriate options that need to be used.\n",
+ "First, we need to change directory to that of Task3. For Task 3 we modify poisson2d.c to invoke an exact copy of the main jacobi loop which is `poisson2d_reference`. We parallelize only the main loop but not `poisson2d_reference`. The speedup is the performance gain seen in the main loop as compared to the reference loop."
]
},
{
"cell_type": "code",
"execution_count": null,
- "metadata": {
- "ExecuteTime": {
- "end_time": "2018-11-07T13:47:57.724441Z",
- "start_time": "2018-11-07T13:47:57.718745Z"
- }
- },
+ "metadata": {},
"outputs": [],
"source": [
"%cd ../Task3"
@@ -478,14 +673,12 @@
"source": [
"### Part A: Implement OpenMP Pragmas; Compilation\n",
"\n",
- "**Task**: Please add the correct OpenMP pragmas to the source code and compilations flags to enable OpenMP.\n",
- "\n",
- "* **pragmas**: Look at the TODOs in [`poisson2d.c`](/edit/Task3/poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for`\n",
- "* **Compilation**: Please add compilation flags enabling OpenMP in GCC to the [Makefile](/edit/Task3/Makefile). The flag in question is `-fopenmp`.\n",
+ "**Task**: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.\n",
"\n",
- "Edit the files with the links above if you are running the interactive version of the Notebook or navigate to `poisson2d.c` and `Makefile` yourself in case you run the non-interactive version.\n",
+ "* **Directives**: Look at the TODOs in [`poisson2d.c`](poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for` (and once it's `#pragma omp parallel for reduction(max:error)` – can you guess where?)\n",
+ "* **Compilation**: Please add compilation flags enabling OpenMP in GCC and XL to the `Makefile`. For GCC, we need to add `-fopenmp` and the application needs to be linked with `-lgomp`. For XL, we need to add `-qsmp=omp` to the list of compilation flags. \n",
"\n",
- "Afterwards, compile and run the application with the following cells. Non-interactive: Follow along accordingly in the shell."
+ "Afterwards, compile and run the application with the following commands."
]
},
{
@@ -494,23 +687,14 @@
"metadata": {},
"outputs": [],
"source": [
- "!make poisson2d"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": null,
- "metadata": {},
- "outputs": [],
- "source": [
- "!make run"
+ "!make poisson2d CC=gcc"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "The command to submit a job to the batch system is prepared in an environment variable `$SC18_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to increase the work of the application."
+ "The command to submit a job to the batch system is prepared in an environment variable `$SC19_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to invoke the application using the batch system. "
]
},
{
@@ -519,13 +703,14 @@
"metadata": {},
"outputs": [],
"source": [
- "!eval $SC18_SUBMIT_CMD ./poisson2d 1000 1000"
+ "!eval $SC19_SUBMIT_CMD ./poisson2d 1000 1000 1000"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
+ "Inorder to run the parallel application, we need to set the number of threads using `OMP_NUM_THREADS`\n",
"What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n",
"We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."
]
@@ -534,11 +719,11 @@
"cell_type": "code",
"execution_count": null,
"metadata": {
- "sc18": "task"
+ "exercise": "task"
},
"outputs": [],
"source": [
- "!eval OMP_NUM_THREADS=1 $SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000"
+ "!eval OMP_NUM_THREADS=N $SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000"
]
},
{
@@ -563,7 +748,7 @@
"metadata": {},
"outputs": [],
"source": [
- "!eval $SC18_SUBMIT_CMD ppc64_cpu --smt"
+ "!eval $SC19_SUBMIT_CMD ppc64_cpu --smt"
]
},
{
@@ -582,15 +767,15 @@
"metadata": {},
"outputs": [],
"source": [
- "!cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n",
- "!cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n",
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "There are various environment variables available within OpenMP (and GCC) to specify binding of threads to cores. See, for instance, the [online documentation of GCC libgomp](https://gcc.gnu.org/onlinedocs/libgomp/Environment-Variables.html). Examples are `OMP_PLACES` or `GOMP_CPU_AFFINITY`.\n",
+ "There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the [OMP_PLACES environment Variable](https://www.openmp.org/spec-html/5.0/openmpse53.html). We also have a GNU specific variable which can also be used to control affinity - `GOMP_CPU_AFFINITY`. Setting `GOMP_CPU_AFFINITY` is specific to GCC binaries but it internally serves the same function as setting `OMP_PLACES`. \n",
"\n",
"**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
"\n",
@@ -603,11 +788,96 @@
"cell_type": "code",
"execution_count": null,
"metadata": {
- "sc18": "task"
+ "exercise": "task"
+ },
+ "outputs": [],
+ "source": [
+ "!eval OMP_DISPLAY_ENV=true OMP_PLACES=\"{X},{Y},{Z},{A}\" OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {
+ "exercise": "task"
},
"outputs": [],
"source": [
- "!eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=\"X,Y,Z,A\" OMP_NUM_THREADS=4 $$SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
+ "!eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=\"X,Y,Z,A\" OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Great!\n",
+ "\n",
+ "If you still have time: The same experiments can be repeated with the IBM XL compiler. \n",
+ "The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is `-qsmp=omp`\n",
+ "\n",
+ "**Task**: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "!make CC=xlc_r -B run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Run the parallel application with varying numbre of threads (`OMP_NUM_THREADS`) and note the performance improvement. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [],
+ "source": [
+ "!eval OMP_NUM_THREADS=N $SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Now we repeat the exercise of using the right binding of threads for the XL binary. `OMP_PLACES` pertains to the XL binary as well as it is an OpenMP variable. `GOMP_CPU_AFFINITY` is specific to GCC binary so that cannot be used to set the binding.\n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration – or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "We are mixing Python with Bash (`!`) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two `$$`)\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {
+ "exercise": "task"
+ },
+ "outputs": [],
+ "source": [
+ "for affinity in [\"{X},{Y},{Z},{A}\", \"{P},{Q},{R},{S}\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true OMP_PLACES=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Likewise we see a higher speedup when we bind the threads to different cores rather than to a single core. This handson illustrates that apart from compiler level tuning, system level tuning is also equally important to obtain performance improvements \n"
]
},
{
@@ -615,7 +885,8 @@
"metadata": {},
"source": [
"#### References\n",
- "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html"
+ "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html\n",
+ "2. https://www.openmp.org/spec-html/5.0/openmpse53.html"
]
},
{
@@ -633,7 +904,7 @@
"source": [
"# Survey<a name=\"survey\"></a>\n",
"\n",
- "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc18-eval)."
+ "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc19-eval)."
]
}
],
@@ -653,7 +924,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.6.7"
+ "version": "3.7.0"
}
},
"nbformat": 4,
diff --git a/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.pdf b/3-Optimizing_POWER/Handson/HandsOnPerformanceOptimization.pdf
index 00d2d949244f7bf9ddf9aca6f4e1db96b1adf874..c441f9df6b2b229cc5a4f31e855c472c5a6e884b 100644
GIT binary patch
literal 85793
zcmY!laB<T$)HCH$ee&V$4=zJX1p|frq%1BQ8-2IToRZWceYc#%l2n(}<ouLWeV^34
z^pXq(BV%JrJ3Fr8lA_eaT&{{+Ya>thX>S+2`&|1M`!Zj#_`3%jxS0zX-oH7ccj8f2
z$<yAw8_a+Ho$xhj<LpakSPt%GUa4vKs`Ax_s-MLlyI24J6SUx8`RU*Cr~C6y^V|PC
zI$gA;>-6o~o1b5u`ZVdU)aQMB?*9AwD(1(#IrZgw_vSS3E|#3*-8V0IheKZRtB7^U
z3$GaZ-+z2{&F)Kcw&@-H^*io(<FhU2H(iX8%&*>3TB_CeFh93$-=E)apFe${uVSrp
zZ+*4bg_Q5NxBc3&r-J=;-2Mq)KJI;6))=&3+<iWyzuHN~m^g)p51t;~Z+1~%zi|rJ
z*XOfso@!;ST(o4{kKYPz5j8Q+XEfSY3mw_;E;@-b==P3R9}i6Nbu~)2U1{REDzN$Y
zmcZMqe*6l(yy%bp9})5Q*Y3WZr`o>ColmDCFWpT2xc941F+rP!VqOZD@BVbsytwMG
z+`6+{9_H+}y7lGE?WGG+E9buASo8SU>%V)e4#n#)v|K;;%D+c$#;%h(`$|ot)V}OJ
z|LJJ>YvGA+yRS`|{Do8bmGs=X_fMV<W!|#bLubB3-|dWe%f-K@KfA}Co29T_*vnS`
z%Kj4f`fV4uvnL&x{CU^;S8uFO+P<k>n*OEvwb9iVU%sEPoY18f?5*Od<f9-hYqPm>
z*Hiv^#rcbGAC~!59O>*m{r{Dmbq|<L>tzpoIxaNnegA&_|1ot7YrK!IV2ucVlVyHU
zU*9lUDErfviE2`vE-RgtpB_8d*tYRA&$|y9x3#A{j>`zzu(-c@7SAjJffN?QtMz+c
zN6xcTFD^Qf$t3zK=Dg-EzIU%u{+PyCYBn^joqzMG)0#VRYhT{_B2;;E>uTFP=cL=}
zZVG)jG`6+{O%90KmicMBs?f3-ZKjMa6DC9-IJO|cHDdl-KC5o#2v?EbS(E-Ojoe-r
z_Q8`;V$0U1hB?`(3X+DtpYrN-pM_6-ZpS2Wl9lUqE{od4z{H;~7-!v((R{k8f06uy
zijWFhxn_y`PQP})D>r@mJ<X=t!RA&kb4r(;O5>D&@2>3<<DDJke^#I;=Z(YM=ed_|
z{cLJYJ>ay;a7)-$%PSg%-lsN+x69dIz3g~tZsEl8EvuJmnOJuo&UzrT!*RI=@9I<A
z@&u<yZ((J2H)!n9c_4Bk;_bv5StiR0z9$Q&T5>P1n6g~mt>94D_B8cJi+bO`Iv+K0
z8~aAK%>M?OnyCv~IG<cRH{Unha=k;lRTBS<zBT%rSlR04*;nlBjrwu)N5cPh?W=}S
z>E0au4%3Am^KdvVy6;~lQ@${2y&1EXK}zH1mfOKsrpEr+U32-5@@yZYt$_~@)w!IQ
zxTE8cw@R0eG}oJz4KnK@;}-f}?|9NbV~UdSM~CFlDDRu6*RlQ0w2KTb-M%u9b<0A>
z#me&@ncXbaKel%IyaN}TTN#TtEIG|~%zV$%)*>0s@8?AXV(X{R*(g(<(6afI<qc-V
zHxGVjZs1&Dy3r@oZKG8HH}?YBd93qW3zZmn*6j>;Hd^;aH^t=jtDW+ebCZm1YF-<#
z$m}uXXQ&N#mesS|5bn^t_f~)YcCQVt(<B9&;_rr)1ln9PF;#KvcGTQk-nDqAO-RqT
z?2nR+W({}3sy$-&yy31B+n3<!R4y;uV!^%ZTiTt**DC8a9OEy(dUenK8g28**DaVF
z1%i}j2ly|RoU`!smNbr_4PPZ+Zht*x3#$kFkLd6HPHlM^mWJAlBBrl6O4nUpQ2Fu}
zm*eE8hr%L@4Sbom{;RzAlZSVi@aefT<VCIueDITCXYotW^vq|x@aLV%?&m98d_GQa
zYyH4kYwxAnW^nUR@ezqlzi(!2u3X&qjjj3mncErKDq{W{|IIum|8m;ah)myvY0d6R
z+b>6zyghmRyVj{_-3`uB?mJDmHp?FAo-p6R;D1$AaN2cujdm^<Wye>h@3t4F9lvb)
z`b>K2v<;6rZZnt3?40|JNAy*^?@13Hv$AI%MN{4RUMQ8YcHLXa?)oI|zFST9E|XWS
zqTDH|Ix?+H-7${8z8?BnFkeltMXsz#_V2j~Leu+lyxmJyfA;itpLK2e7O@vkRiE;&
z7Zf}swf${D?}_leZcpBBYQO$`-O|-A8p%55s-0ghu(njMlaJy3RQ5z(+3Yr#er%`R
zcb{#~u5YqD(HASGx@=3_mrySA;z{?5f3K4}aa+SW=gyWa#h9DyXZF6>$h$P}Nf3Ww
zbaGb@=cU-*`w<4ty=_{%U9(-%H_s5gbLrv?izc^3=F)2?zSethJN5dHWavSyzMz_Y
z3lHW_*jtdmIpz54im3e)-7o)Lvh}v7lS`j8dsEihIc)A}t<9X*y^l@FxXUCd`h3#W
zO&d$Tw`asFrj=z+stWjF_iARV+j7Hwr!P17zB_Ynr|q*x7aCsH8i@EvG_P#Cc(A0r
z);D3w%jdUp&Kr2-%zJYC%`cxx^G@-f-7b5*^V_ys#rGBertgz~S;ym@>B3=hS?BGo
zT@`s|J`?Y4aaW!H#IL5t^DARW_?;C#_vIJ<if+%!|1@it>ikzn1)e$tA5mU&Y=&G(
z+7Iy;kshBPzt%baRQzP>M$gx+7yQnDl5}>s9&IDUq@CbfeRGYR=?kN6+qD|%jnwO=
zzjB|S8h*1WQK+iZE_dZc!OJUkof(2ooiovUf34-=Jx_sa=PdItxEQXuqqHXDdtr`C
z%}T{g{Y8hmM9T6Xr-m*poxGsawfX%0cCFecu5TrO2G13p)UqeKcBx`sYs63Qy^+VC
zUa$9I`~RzV(qf&|yWNvJ{En`_SLr-?_AyTB83xOy#zwAxZ1?-|r>d>xAJ#SLhW0((
zUR8I+USzHPxfZU(yFr$DxnG{<p6A|vHTQSs!>LT`?^<~3wpd@kztGp=wdAZTFS9KY
zJa71KdGD}ipTmPC%(M5fSw8<G7?s&}_P@cJsoKl-FfX}yzC?O*e=yU@_2)M2S(q~G
zRHD}FH@5|@HQjXiw0xPo?}b^ZQK~w65vjSVk0(Z6e)QwqlE<sIYH!uhkg`+Srg}c(
z^-9hk^^4Z(3&wBtwqxVx=aJj9-B-?jlc4iPo0_S6A9n?BzvMqVvF_KgGlw_+Y*jwr
zxqZ$%b>3ffeL{JcBz}FJ<(_xv@s<0#U7p7tcJ@6RDz{$f%_`>b%7{jJ^OW`<yT6^@
zI@5kj@7vi9-Rwy{aVOIvs}nquvrT`9daXDSc52s5?X@~TpXUE(y}173cCOUC6liNH
zH7^C!zA~~PuAyaWhTPD)8+G!WvZ=tn&mw=BK32pp(O_3$>%4uS-dX0Kim=|9#2p`g
z-3Zmof9rQCQPAmw|E0X?vFpCBE}FK!JEG!`zmDIr|38a<ZmRjct>#CD*ulsLjyn0<
zfBc&;&nM#EI(vCb^RD8zmf@3%3m@J&_2q9#fY;IOpK2!Q)vb!2qv^+9VURRuS&YoX
zA8zwAudu1C`Fx2ZcTw(5m!qwJ1uH%F9GRQ2xPl|^&&%oC<)^*Br<N|*rKvitS;(hq
zDt}4gZI(-hFRwTxu)fnde`3M3p9<j-FAlrk*IQ^K9Oq`3_A&NS{Ja~+>NBpd^O-XJ
zW>#Ni`s+Dv>{1q8N}pD+3+4T}@?+YTG|RJI#+Rb4TxZR-n|uECx8<ip-Wl%;efejS
z+vSdZ$L8k$u29(Zg}qHEt@^N^@Iu)=iHCkZRj>@6WU^#-?8P&iYCi;Tn|;~w_Ls9S
zj!n6AC1vITw&SX&HlA}4+w@?{;^o)!3SIUsUVMYQH={LsVn9mKVx#+)9BqwW_J#VK
zSaDkG*xpli{4PD4-CD2OzTH|Hd5hautNV7@u{VoM-P<*;%~5r8%Q)=g>1CL%6_~9V
zIcHZJbNO2L3o_HJU%ZfZyd}{oV%*mAf12u{>hhNN+s!U%crMl`zrs8(Dw;!*eeLC(
zx3dl$+k8>m=(14x0f%)3cZ9yn-Pm<2{q^~y8#DggYLzWqqx|Fjk&P*C`bQM|7C#Dl
zG;`H6hc5pufj7OLtYLW_KiODC>*lL>U7LkmJ^l8*5>%djgoVX;ZtPma>;Dg*So+;X
zFEOqw=Fql@-a8D74*u3FI(Q=O$sy5qu6hY2qB|2p7cr{{Xw*JdJo#H{-PdJF+;>)V
zx%O{!OjTAtYO;LK>UA=zr+>C-%$#!4Q0(P}BPJ(WZ>8xjyR+tg<`LC}mQ|C?@0@xX
zv4HET*7jFnzAv~_r!IP86x0}~xL4@gDK;MA%T1i~Qp?UuuKB69M$NFfR=kPnY(nBA
zA&axCHf~CqcunF%&lkJ32~xpx-X`>)?X)bKdgd_ivqt0jhWtUfeXIBQt6e^DK2}?+
zVwb*_-7k^86X!G9A4Ho=^)eZ?O=@!Y4V8*&Wtw<)+FC8!nIh%Ne;yqza^80Dwpsg2
z6_Jw33Ma20ykPP0yVis+c4wLQe`K*SXEMF?o;%c1QF}v%UFr7!iLV}MsH;lPzbMcX
zUh!-G>EPKB`-~-JJlnTiJ=5^~v&m&i@%^(EFCQ%~z2}n`uwnlCrk2>ZT3!qhxx&*n
zz4JSC=;wDKu}kI4x9{t`UgEdu_M{?}i|c-Fj=XxaQuW-TpC_c>nlBK(eN_I*lI{;X
z4oW<1T(e8!>}=6$jW0%z*Udj?TX?uu-s!qUZ{*wZgzNJjIkI$U7d~_O<G!qZM?~G9
zI+>q;ub<2_Q%Xq=+dcR6D({%|nSoZaN0r=9y)Emsxwd5K`J;_XlugU0Mcz_O;EpYJ
ziQe>RQ_dFs!Wh$FqbZ)Vw@u@}$oRPL<&_8Rb`z#3PUC-?vf!nM$cM{EdKWSM5PTf=
zjHht>k{`)`HD{Q$&tIK>R=b9;@Nm)0?V;aTv|TM71fCqJ=~|k}(tKi1^OT3|dlW4K
zkG3@=^X~ZgR%*-E(r3HMUC%i?uFVZ9&Es1wZa+(-*gM9*YSv?)&%Jrm=T}9!Ge&v}
z@MHwUtqeP_(9!v>C+XHK!})Tb3gXXdoA`Noq&iv`yl7r`GJO9LQL(r^H)novEGk=i
zJ;Uc}&C8M}VyrtB*E%sDGz{kXp~+h!5->4Fy5#bEp&0^2s!wce9xN+d8n??(=ik{w
zvfcadO})mXn{=Jo=)_a$hx3*N9KKw1vP6LS*pJ-_Nms63Z(aG|>?e1Y&OA#E<wtdi
zKQ|r8|N3$J)8$WvUxligNcK#$?9h}ipOnWlZy(E95vi+%FUqaws=IHM<2f_itaffC
z+a1f#UYEsJgte|TRt~v!@bI~1@57uf#TY0b_FA^OCB@+B(-ap?17>5+oew&%B-aW*
zW}dR*=N4A?)@{d|{HI8oi0)6_a@nM3-E)Us$Cqcfc3%A>E#Xkbv2CUA-D>9N3a3m3
zOwPPnII%Wv^~ug-lfG?=+o<nlwam2o-22EmlNQf;b>f{^xca}$yE{%@iv7u;Jo~9L
zW1#d?=X)Ev9nO8pczR1otSR%g^tvQ=-Z$C{|K_c`c<$KS?+;$hdlY9IZFsWoY|4qN
zsm-2TLSF9et5tVA^N%+=qGWt!*Pdgm)l;51uX%Y_Zf2>-x3Z>5JEpYx`g@!blzrs!
zaz;$HufkF}E%Q0Mqi=a!Gb??z_sOykJ03+CojjAU?)*oAPZn%v<c{PSc`d)S>(R#6
z*sfVWHC?xtFUoryU28AlI?wuW@sV#6{FhxzsAyS!>D`95d5`XYS-_^dwl&)6&NElZ
zw)wWxQ~f*qBgE$1RNmL;w3pvDIc_22jcrxS`c;cPfBdPPy|vad&PQFmcauo+h5tvF
zs|Rn2ncCs8pTBEs&53B)+$EQO<-8QUW_z{RJk&YArthKk3B!$NZ|c5WUG}S~S1#-J
zil2!Mn@YKZrtCRax5w#t@bisg?>{n^3U<A|6<NB!)MU}BuW4o*BC~v^Z{N4f?e*_-
zZ8MW1Ud{}iy26a7LZ__eRol`w^UZ=A9olnbmZ;5)PO2-NQ?PMkU*Y@P(Wf=sn9bCG
zFLUwMWEBqUd1AEXMw?a6oR^>2*IYd=8SG`4e`m**!qusdxmK!9oOUKYbLR!U=a01#
zk7ll|J|35Fy1R2z^u@dv*{io19^9bt$1OMc{MKzJee1X)9^^4t9o@O}Fvr_pahoDK
zY}+>-le(D^7%2AsSHNLCr416rVY`oPvD>XGU=<bb@~ZaiQx>Kdx18Rr`@bh|b??jb
zea}AsKHc;8ri7SQ>%Av^F88x_e||lGW7mAu7wnC@;?CH#UGcI`n_I;iq@HTN@$u65
zZqrJpN1HxPJGoK+>kpA2<F|JvZxhk=crE^O*=_eCg%dj!1?KHrrf#2hO`t70<IpKp
z_GyvH&Qh27&7GaC-d_}BchJ=fet1ITa=`Aqt^ebG=UlBx6S~~^@mbjB)s78Qb+hk?
zEqG<Ewz=mR>u1M5tlRjbyfSXBJ^g35InUna8yUKJ?)Q!B>h`7A>A8hl|K54qr?koQ
zVf?K>8<~%5nZ|l<oioKFqJF}&)i;?Y>?-~H{@9A&8M7Q;%$4StS!XueU`A<Q<^Fv>
zzrUQF-)rAphuQKnHYTR!Wo&3^fZX!h8hx_Qcqi|^=bC?*)@8&q9ByZv#IRHKMf3#c
z-G8h7iaqZAH<;q$HN`7+`rW)U68m?1al01&^^pHpK7pO_|H<a?5Bn`@{ucgt`}o84
z{q4;n-Xei=c6lHE&G@(J<A=h(=RG;Z&s{vU{}a>wr2@z1{r~;Ze@5z0owd_nB;?Ip
zyziX5ZtT3$Fgx3wOZm^0)GX`T5IHxqML^8%UGcVVq3+B3_@>1h1Y8W$<64*~J$L>R
zU)>4WzLm)$HF{6u<qP!e{2!m0n)WmP>Dql4PTiVr=yyS4+w~_Nm%3CXVqf~0Jh2T~
zw7urkybphaW<0LDYrP}9n`f2YjBV%H1v<~TxK+75(Tc5j7n^!^YtF2h1xpwI6utcE
z4Y#b=b@isRKUW-D88~l68n@q%f-8sg#aiBKd`;r@obs!MbHkR8uB*PJPg=7wgmuQf
zC9lF(e7wFXGcCQm^+Ude{M-qVc`s8+_$(3&)Q|1_qjIurp5{bpu2!vnRlXUHDO%#p
zo%5z{k@|S$i*{>qKxoFpwTnM)ySOkw;IP+%&(&7`QO;L4U;I(R^j3YddidgA_SRgb
zEW6XIj!g}nd_vl$Ja*x3>E!7O&XYf$`cbm?r`4owoi<-HwnS~2+N_|mH(~Yx4P%+x
z8PTpnNBv9Jt4*@C%w;hQIcuYO>Q>_NwqyF|R!v=~$#N-f3B#*PDas%D-pB~9o0uV_
zXyF=oc2~r*$6k|py5{T_bJRFpakVA(`ce5C3lhYZZr7_gKJ&Sm+UG<y{{xB3R=wew
zJa-9uw%Y8t_;l?y*9hB40pF#Gv)-o9Zs6VKR~Ft;GcQt$`Ec(RSBJOjd`}x233zh6
zu%2;>)A8W4_3f+~nO2jJ{eAmbp+Rwhcel}j!$&oXe|)!B<NyD^y70z7hFh;$e`Y^F
zc-LZAMoC&&dg_|!HELQrJrV-!qmRDX+qs50pzw!mnfceroZ0o~Jem>|8VtT@XIrn^
zHIKbq?oq|OxX6z&`a7=VZ<aLiV)?B#d-u)!)Z*=*gmd1VH%|RrYioDOOJI8D1L<{7
zR&ShO?yeWG==Y@f49269y2^W}#(&|}RpmC{w0_M!<4f^-+oxDq{ch=4cz<etzFg2-
zg9V9&m+Fo?JEfdHbI0w1neno$!iUn!@*mC0?a<iLD=m>*a=r1+I=|Ji_hwz!%rk!J
z5bx3Q@`rju{hJ>NB5Q4Im118W&A$9`&GeQAMplFCO{Q;|xuo*exJgeed-U9Axp`RW
zB&WyOF0JCO+h^O>M+!b&{^u&I)=$s7s<GFX78o99_;;mgYoVrGkVVm^isv_1^C;#i
ztFs?&`#Jx$K+BWB*p`oehfLej?oF-lRx4oAR=>M*-IT+#jJ7}3eCc;oLiSb4zCw>%
zCaJ}mPkym7tq|qk{k1mlLWf_AsrbYijV}*oKJX5dPQLUuBi!lxns1RB9ZfQ<994(E
zXO;wRnD?W9O4Sm-f>{PruFq0W4A<LLbLH2c)ms*?o8m9hk>~c%CAZXP;;z?M!fZmK
zJ6<SRFnxS>z|Cpeqxk-uUoZ2&?)-bn-{S5=5B@#x;=1=g?A$*qEM;@P;dh5rAOAZO
z3a1#(U3TL_;H?l#^+%2CP13Y-x7N<{XWPoMu<X~q<;ov8x__VBIb}l2=Ja=_0ZZR#
z>91$kNf6%H-NPlhaMt`uMN#J;PvpF8JF8FTrBq15)u61Wie3?w4)eOpE`){s?$S4D
zZam|2{dSMr($zCoM?1UUlC)AjEVw*rR$7Pftiu;iNFL%SmA1EM?tYl&v1n^|!M5hR
zGj7SO-LyO|hU2-{vYmUjp1f~;?&r##!XIpZ%{{3)L;2|8jiQmyt}tlMyCd@7K)FHD
zJb&|6!A*M7aoe`u{h}Ma`a#*WPbar*+~29Anf7d(`nzNLpIv`9Xs?L&R`NW1skG`$
zOo9BeS#LR4)>K})(h^*LoMX>%Va<mJpS*r<8)*Mg)PKw2e{H$?OZP`61RPx4|9Hi$
zRRa1)W3*ZxvYkGYzM45zHcWPg(*LZF^Ilqg>U6lj(=3~F@)ebrl`&ZjD}%PK^mlk&
zXSmTdXU~y~-|jx&(tCc=53j9{4f6#|AMaeF`I57FCcDDUln#Xy#b1mU9|ujC`?HMA
zLHXDWx3x_T-%`@2EEfK>sIDtH@#CR0HNt|Y4(<JvV%s(^;8nmGMf-IJX9Y{Y|730d
z=$`MEA72_*8|ziF*BRS2Yp!OWVRS-hwdp>AH0e;T!rBX8r~Lje7*NT<)+dy+JEZOe
zdy&{#=D4*p*}u%+6Vax=>%*R~yA}Ih8VYO3tz)ZO_4MSo*R~5gyNWJ8dU|2W{D0R>
zHpKsue12R0<!y=YM|4uQ?Jv>lIK0`p>Q~6$Bi|iQ-JBXLp;dlX%CC~gKVI|r42?z2
z0!bTo@gMNGsySWow!_?=Atm0e!Tg6sk9u&dPIll*b>BMYpIYk02RFCY?ew|mU%z_O
z7jJvzu)h=QM2s)aKl|(ApRhZ3-Gnr4%e!u0vzz!Xp-F33d18~i=k5C%IdbpJ_T|j{
z{qnu&w!D|=Eo<#BonPWwePFlm9OqMdS8gUASD3KrOy9(UPOhKvyAQ>5Gekc3c1!1Y
zBOH6Z`@`k}haJL)^IU4pneP4M{TQhzwj_VC(SoDXi;CBCm0tg=QKdU?^Sl{3=HCiZ
zu6V5xuz1K`vEWy@7IVQJtqWGao1$eE{$Ju*J!wau<Nms(XVRh`2z&oJw&`J}f3xuf
z<KG+CY}zY)NbKhM)=53<CvH3*=*jA`uJm1>huYfvJ8akUF1zG%I%kdUtz)N+w2pa3
zF0)x%St{9lpyF-zYO~O<@8+Cp(`t(}i<tgw>-0}k-jvxd6AanFEvEA#>gAj6`>Y?1
zK3-qk|BO*;o4=?$*Oa1z*G&?p{nWTTW2!)H{<gg_%{enO_k6l_Z5302y29kZX8DX0
zjoXShlx~b$cVXgQS9>M3lhaG*dQ1I|3l6k=aiiK;mwo=}pP|M7<g{m{KYe*=)v143
z4k@n%qRT_&N<;4S9J{*LWpALy4INgOvk3*=KE^hmqdn^<Dh2y&J*1%=W?_HX<J0_&
z{nC4%-wB$Y#&k{O>Wig|^-pVEo0%o4mi6w~srap}n=`mS?GB$7mm*!wG4Dgfi?dq6
zj)xvKKj`FNJbk6V*#r%li^qPa>z{6q{pYfj&E{!%?G@_-$2ZKlAQO1Y>-sUb9G0De
zFK10nk8;>>Rpf=DNAT7l#ij1mcQ3a5Yx^E>Z<%U;OU8+FiA7(w8J=11UCk<R#$kc9
ztJB7A?%SW@nU;kr_b8lc-o<w7@<WdwQ?mZ2-MSn2T<c%>wT+uDU+9bRy`OvMYD^`c
zbN*h7&C~d}BA;}MzSgU~asAo00{^RT7Cn5IDq1@;)qm-#sk`~r=49GD=eqDqY}r}G
zv(@~|oYu}x-4a_`8~y*t)o`};8-l8gUM>uhh`qZ~D*Rh#v3SFm*n1}XuXf#CwKlmx
zGQxUUKELZZD>36l*RsuOdxc(qn(}5wnx^sQ6=ik3${qffe-u9Wcw_UF`CHyKPm`07
zOnd3rx9vnnkb2L`BLBps{lW{Etp1z+EZo6D|8e%S>)CD5{iX#Wmz<77NIriRV?05u
z>YYh;HJ`@X*q-3ny&WnCXZMyB9o0(YZxQe|V+_i-oRX6^mqlsc1(WbJ+v(5S&Pg1f
z`m<kMO^4~TC(ohTM-T1_&AiQD&dT{`#~hAJ&)zT8@$T5x!hX~A<%7bn6OJA~@_SW}
za+qzZQrSO6za}NI8UDZPCjDzsKf>KqI(zbiO7@KGE!we*U(b)9GG#^S;sQ-2_9Id6
z)YcXS>iKPCOMDZkGk@i}?xWV1Lz}*}KiKnh!#|0&rV?=x|CDcSKKL&$)$ZD{Kacr)
z)`oaaFFCN~zp%}Y6Jdv5O%j+Guq7{P>4)-q$K=mymBM)|dJ`h;+)wP7e^xsq{1%hO
z?(<J;VgnaVy0A-fwyWEvIV)Is9yQvCf5>|mtny2Q>3R3t<#i={jka&`%Iy&S)Vhx8
z&h)QO6m#dsPx?Hg%tY=~8Pm4Q%r9LxXt|$WwEgCmsio`rkKMfR?7L3&gUX)hrm1%q
zKD!ohOLfJD%TxKz9)D2%B<anTP0|;-P6}EsynW)+!^#Q9Q({-v&M4V&T2Z@XzpB}H
z=^5-_y25YoF4?~N$-5U0b&Hnve-E%(`)2AqUf1&>631`<|MvR&v$+@68ya6qd(^BG
z_cUJbj6}e_ns&z2R$HEcor2YO`;@}pYJ7RkGu8e}%I}sN-;>>gL&}z<tXsZg(aO}^
zyEPGPvhM?qee9Q&IH8lDA!^uqZem@$$;)Hu7O`(nzsyRwsmd3yx#dOZ^7)H?s<NzN
z+ZsRV?`!>frumIC-eBpw5H)ISY-nPP+;`a<eR5s$PTsrEwf`{tFRf(g+0HQ`OCeHq
zMeGFEoiXQ>PAL8T-|ewTBXHNA>(0BS@2-1p?Xk#0ZsM1h_3EXQe%hCES^d}dxBplD
z>+|-@)7QUSvc$t^<>l?SJtY%+tZJu3>&@L4`@eG9rI&v?=l}op?fT{TeTK~?0#-9p
zCxreio_gu|LA$5@XXe-B>IX^1&Z~V@p?BWSzUSi4*=9f2wFSOn@8S3<pZ{&TTh&%k
zzhm3B?6+}yZ*t<Ul<jq|+lszVSDy55Zd;yO|E#UNaQo9*neKXDHQ)G;XNyA9*FJ94
zk9!oHo+vlnQu36{#yySxzk3+(Y2-&4#aF%)dE+`){_&ztr(F#(fBs2`KX<fuc;v9W
z?zJ9U`P@Y^CKl~-HS1;GuAU;PC7XL}lhaRWDcx54U8nQipKW}{Ch_TFs0Y_#-MqMm
zhf;Ui#IADYlH@dh{X2AKRJ>iOMe~QXZ4(`vl%~C(x`L&PBR}kEv-R|2v$%8CO_Eu;
z>fvnXy#;S$xB9J#Qa^mYNYY7rzSqa4GfmEX5q|c5i^kW*MjuXx`<H5d4p=f}Z}4Z?
zO_?kg=N9!wuMwJFzLcN)>@2<B3CRm}&hfAJEf4o7zk7A7RI=#ytC1&@zZJMybj$fI
zo@aKDcQ@az&aKgmK{r}Yz1WnWWxVA6l`T277IR;Ptqa+zy>#Wq*IjyB-)S9UD)l;h
z>#EQ9fTIg%?O?w@@6n{xqAb(7MSh2@Bo{7yJICC04PTC1eY-^R=50UzT(ozTd&hSD
z@7$WTU3IQI%zn=Nyl2MZ4y_=|W|^g8L9eUJq-%PWo<=X2cI}Yljl$wDQuo<aj~Ygv
zzir;{xZ!40YII^HPkPaDfrHH-zjA8af2y$6jjO_W(vG77#xwJ3`;OI%%>C_g<)+oA
zzVbf>2j@AuBp-Fyl7GR&blKjJns^KT<aJT}ml9@7*cq`zOMLf{yNl+9=@)8euievN
zeKe;)ed?iO%?Hy3jQ8w~+SE5O!f222+cmn{f%}h@m&i=HdBwq9MPqZo#y;2QYaM(t
zmEQhx%VaQR42+z6E+O#f;ef4=B1HKg9h)KV!XveAvid)+v>PA8UHaB_7IKSq?SG(n
z{o7qB&$l;3x87ybkkDT`Ma=Bh<P$07*DnS{|Gcv*(U?#DID7t|m(6@%?*2@^W4_Ju
zCRg!=FIKJAiH%bFJ8nH&^n8YOPV$kiyAy=gW$S27b~+KQ&|=7*(A=&5B4=u-)57(p
zFTP(4)!-^Mm%Fi}%|X#(C$Ib+MT_mi*YD0`+08wdaRX=TD~qc9!v{>f!WSmF#CpB-
zVPu`ZEW25~$}(c#o9WIzY9VLtJ$_j$`t{IHLBaCG%BOQx3s=8?Es=NdmPXCv-VV>T
z>nzvp>XGYC^1Sjsl2u(NlVjeqHx`F&mY%uXAoTrA^YMh%QszT8U9-1k2+00t$U6IE
z=319%f2n$bJgtA{dDk*k?7r(^o#~?V{)|lQ6I(^?)e&p7DtpB)?N;3M{;W=I_x8y<
zEFR_Fx^z!6Cr`!fZr?$x-isnKIR}5QIUN_Q-!{jX<>^D6(AW7_g8P{ZmY-&hJzy&3
z{mjGgvVr^(&F9)OIg4*a1~Knzh*+%2q}Z<Qu=DuNhIwYEH0;^Fe6>Gha%i*V;!Kx^
zQD1j;Z19Zucy@Q_*?UVOELuC()yRIErMPqJ)U91I(;x4ZaA(=WdsP2Pd&PBbzKoW#
z6J8fS*sYEi{BmHS1E;TM#D9ltK5U7f3tto#o>i6;o9-fAcVumt;mugXg?&-`Z{^w+
z+fEQ$lw;%P&)ffT$0Q!Dg0BXu?0u61`BfG%#CWPU@pbRNTIE!)v&~^cdRR`!x{dYw
zdzMU@zs*24EIdES<ml$&f1PUV@n2n{?c`4vSjo)R<iB35Qo2+-ulmUov5?PCvzK&f
zU$=4k<Fjt=3a>X?`+~pAJfAdw%iR@4e8ycf>JRPa<~nnXGqa}hSM>jH^2d%On-xy?
z>a*sOG&3yi=rUNNtKh%?$@%@3`vVP@be>)_d1A;S!#%y$#}lr33+}0xTAO+D<MrP)
zEz$-F2ac(I@$(Pe#=fGM)k3&alx6D4O$+DF-7lWGc@m$}t%o<F1fD%uC3ey0!k=D8
z1J2260asFvUig?eS0rpF>(>1z&rTP~N?7o8*0DL0Zg?+Euot@Q(Y0WUu4T05^!El~
z6Fm)sLfX?Kb;9nqJF%WT9bNgt|61uPx%dz7*mQK>PqUt-u&3{U=&uV$FNbG+Nj>&@
z-}%QfvzPbY6x%xY+*I58PoGbymFG<KyC$1|H>&i(+=(w9tZqBCm!aW^za6_osp@mB
zYki-)CHLI;v8v2ICC@YM_zSr$=5v#GTs|-Tuhzm;qhR0D)X6#Qo2mp>c+TjJazA=i
zYtb?}<D`$AYuBciK3;Ys<h>2cu|;+c$F2%YzPr4_^oT_^%W7V&s)dW=wlH@07rZ%>
zbGh&CJMC)>N7c86sd}9d)jIZP{kEHqT=RXO$?rEXug`U~oYUmtsKjA->*-t__q_D1
z&|<qS7aJpgP6<DqF;`0^a)JXN*WX}H>FF|_k`G<Pr_Jn5{&>`6-Hz8=-ldCr+`XP(
z>@C0Jl3HY7u5+Hu`k2d1j(u~~Ur6pVwT{`-{qjZKyxn@g^uH!LiatAR<-L_%(xQ4o
z#zAGB2eBNvZ}l{F4_vNXCMNcJayy?+`|dRr7kobcJnXc{u;l#uym_AlYkxklpW^f~
zw3j_OHOOXDP3DRfY}0a{Ke>H^e`CAMtgw~W-}x?+j%iE(m3n(_+y3hn%U90dc(2lX
z&Cf~3Hy=Mcx$@MU*vnh;niecN{k!W?^lblYN*U&sQuXhjuy`#N>J^IH@9wxJNa^dW
z8<)9^`0H0Uoe8#IsBn7PJB<?`-gn4W+&HZ5>#4}=abBGv;{5f`UxK?>J~KvWujp9&
zeD#KHSB(_})mJ^|^^}nce}4Ok=8rFl_VED;XUuv7r&|{toFT~N9PC-U)%j!Cr4MOI
zQMa6S-1MqSW!zeHc1Fbf(BN}xTPrW!_?IU6bSGCx>9fl#KHta|m2cVqUN>LJclWK;
zS3>u>WNj6Y+P<}H7xVjtX-OuTZj%=*n#U?Kf%8+4er~~*=gHCs9?!JXEk85&l=3#E
z1GAm4NnDBM%xcoLUFE6w`^N0oPnJv+%6=j7M)8W>anD_92RNOr6eZ3nv1hEGeMd%1
z?$M<NiN()6HuW2;PGaSjX_VR({pH29nfH`zTi^QLateGevv8%Y*&&Zw>GI<>GsG_*
zoOmS9-t+av0v27B=q|Yoq1$?GITC)wSFSzO6m#8uJ8g!M3e%JQ2V~Y2yPS`8S^PFL
zP0>u}XsGwE86FOAU;H|xzDGCFKjKEpl!7~K>#yvdw&cOud8&q1OX33MXUt#T`aIM^
z__X?5uM2EqVIm@rmYvj@_dIL){NH(D(afoeY5O&0dN#S)tL9b-w0l3=CFJ=w`C_Bn
z#aU<ad$lSTaQA<a+Ffzo@L{ySWuW-?F3Yc;o(|c6Vm|7975U8bThi$5VuML94=?pn
z-yVBA%&)QMw3GOO)ywxqhX=p^dm+~Ij!b&2LUg8}$o<?|w;$d9Vc5uRGHIuYyxPL(
zgKml2?`noL@3r2<_@3|Weomzsmb_DsSC-A4yW-4Lt{Neuy=sf*-rX=$OSQ*0_|dd^
zQ{4EO?=r^UDX7$1`(I)1D$Dml)yBfd&E>9JZ`-DJ`rw&eMJnDK<=U1>?=N!Q_{Jhe
z%&X(Z4|cw`BbWCci~8rzS9N*4A<vw7qObTke>~W|Y=+ax7mIhrfAH3n4}81Jd7sR$
zi}&liLq&wHJyV~*rO0rrP?L8St0CXC%;1Cr&8r_r%@4ZC9QE00<9RQ^rBU6r{JL3t
zI;{S*ol{@lT6Q?9+)Mq=DYN-U4qs;FP`@4`b!UF=jC_O5Hy{2wnK3o&s$2BjHKKhB
z<L|G?sd#q4|F_l$*X7e=qtqAi?r&%3QexlG`QDX%Z=bZ*<w_AVLDxgES44KqI(@a*
zk*!FnFi$&|`O6)Z%W7M<Ocsyjo3xExY45YSLOoT_wnsUzzRWr^`-0N3tv}jsEzbAO
zuDH3a)lKb5e>jiRiS)0ZnU0>|e|c+$?2-$|7F~I@dv{Orz1ZJ158nmny!zY|+3lZQ
zqOZCkA>1bLesD6&Tcw=xTQ6DbZv-f2#<&@qZMNjupJMWJUv2n)ttY8wwF=?C7V~a<
zZ&{s|e)sn3yn9s>CsmmmTl3dvr<Kk7)p_R4CRH=z-_}Ps{ykrM^Xv6#&hk#BGP&lQ
z?SKCI|7SPTVDZH4jF=b_Gm&CsYH5hv8QB_jvQK5Z;NIWbpV((9HCtbJ(ah-GkiY47
z+K+&nrfa)PCg}gGWtHT~I~lc{-MDpKlF^9-9euH9-<~fs>->M^8RP%cZy)!+egFQp
zzx{1rPr=lX+rQQA>OJ@8+_`uDf92Fm6`H0VFBgT(=lg%?c6V8Wf5P-jT5~4;lhVA(
zZZ|d4?7V~2O8Y`KgSp27@62-hbh&cfy3dzxz4P(E^y|yscmE>YRMw__*nLm`<Fxda
zL&;0-|NOc0dDpk+7te&=VtgiRRNnT@Z}ku3wevDnKTh-4$kZvEe17`A>%y;<BBhS@
ztNlx;XP&R}{@2Vs({JAY_xs=ZHEOzR5>0o05W94x#zEtVRqqXzfVmNEB`+qNtp8={
z<HfFNGGklRmdve3DsEj7TXisGi`Yd)-HQi~-|l9dy7C3X%Ye@LI;&3jTf1JKwJ!d$
z=EJ7^?s)}^l(l`%ya-OpIO*e%wm?WNFmwMc#-n1s8N$93wI@7?*~;^0pSbn!Yb7&F
zKd#-(G5^_mi=a!7Ba`lG-rksGXi&SUH^utor3Z^5bUG4pVzneCbI)8_x^beWl(&?J
z^o*s?ng#7v>?)Wy$1-v*M^Awj4|~$|rOs>TD(Qbs3K0nNO*_rQ|8MQrP*y2{O~w<o
zy7S(DnH@WC?xa{t?Ta2uzr5;DR4PB)BOw^^`)chO1@Zp)+3FGA`=%{C^GoU4k{JEj
z3m*n+cTZ2b|MBKa+ulu_X$MYbuX!kt{vhvK>b8F$_jxX4%JOJ%<m7CZ`@23~toi@%
z-UavbPwbk`!z3rlqjR09KTQ3s(*s$)k_9bASt@Na_>cPJo;dtX`d#iT_03o7)=4<k
za9-=@&bHRCQsK^!FSI%nmveh!)QbT7bKm1<bTWns&swYWOd!^0#^QyYbJNrE15F-F
zob{S>R;22A#<iRonprBRjJwxgo%?D>X5i$Cn4iiS%3q#L*k+;HuKeY-X_~gZLm8{F
zg67iw|AOwEo0KH}eNvO3UEzb}vh$`I%c@4lz7;%~>cPx^NR3<8pf^p=nSYVRigcEz
ztuIyX%;##_&9mp}%e6HvX?M@Lf3Q9!Ta@*n@#)+bxf7EYB~I|&ZMyb@&b0c8LVdsP
z*`GJLZ#^&HOYQi~RLl3y+nnMYr_@gKs*aw#&+w3w_m4}7eN83L#V+eqZRaW03VT%3
znJjMN?Ek!B?!|=o`3sNCvkbr7XJmiJXnFV~Vf*IYPkbJqS#~w@+`Zz)OKoCFGv<B%
zGRbDjPScjKSM3v4uABIIZc^k<t_!-gvQE9TR|~TrI?BGuT>8bH2|<Ai`uo0EpS{?~
zw1Kty@TOJiS8jTm&2*Y5dvS&Kr4^Rd85UFjtUY6OHUDnht}9-x(zlBQ{FF{?uW&nN
zclqbnpB+<7CTA}?6}u+;$PwN;9Z%)jnl+h6H#*JfYUK-LeD|a?t<Pqn=@XfPGZVS%
z4%$81u-ck$1>de6tIA#G>y|t>-8bvBl&yY&Q(d3|8=IpchggthzTqrZrUTs<Hwstk
zu6ZanLt>|%^z+YIe{V%b=6Ia@wAygvLUxn1KDpS>bCkr*AGp00e=_mQ9Kj`TZ6Eve
zJyq^22|WIe=i+1qW>?WiUCw%MWRD%}h*dt6y8nO9=Bkv%LaW!u$aVXQ-j@8#choXs
z;o`IlzjIz@#LqEFuy0r~S1VzyVIq5S>v5JfYpPNX`?>N}*ak0L)F^x{YNi8YxcmGm
zevUs?4E**_&T?w>lHMh-UznG>UG#9;bVggt$@+COa}#c<=qrRrE$Dc$dsD$Y^Q>0B
zh}?|zsb5MD-qJW=`EBE0zk_@s-#N_E_2zuBTG#9>zAM%;<b>?PL_Kel_6@CURUcAU
zocZii7!oYOneVv2Z0pJ_SKq>(NjlS$`8GQDN@ljZcBi?n5_h`(*pqRu*v$I87fJ{7
z{@)jgQcqhcc=k$8eU4SB?2qT~Z|~n8@5w2p-j$&%qJCMR)yQCnb601EamAcVqL+*N
z&(2=-XpYE}$~3KyFA6f(y*<qO_fm4@$qn5OtDa^vUyG7^_E#%ry6LGq_qfvj<a^#X
zU2yQ1dqpyzk;bZaEpG3VwR(%AH~H^L%ihUcKf!+?uk@ex{cn|&TOS7$J}c<|Xn4+X
zhMUB$zb}Q-KKWep|9Q21>bEeiS+-{{lrIzPT`hLL|M$z)HQTlvcKPbE(`WbE)2Upi
zx4gSE&pc3Gy17`+Q0VJft-c*|W<Hnm(g~FlIMzGULH}E8^prn~%DZ3xh&rpAB%1q1
z`I%Y34u7UyTnk#aPYdPQ8Nd2{Sb#(2w7jc%D{p?A_PV&N_=SMrTf4C5x%Ul|pGs|T
zHGFpeV9d*_(lfb#UMtmoEx*Kk^07sg4*ANg_Y<z>-JWS&ZsokyRc31KUq90mcB${1
zZ+)Ipb<>1%=QE3mn@cRGaL@cE{V*brx9-}cL+fsx(XvYXEj>N|?B$i2>D3`qu6X*@
zrhUB?;4OW5yVCi`Ia=YBVnrrv7JgrMC6Dvht51hDuDF(;4O+T&Pmpq;_N&>mh5IY!
zT0h!6=X=!o1ve`N_so2-V*jUXSC$w(rDYmH^K4$6%QoM|ez9A65#O8a-5U$rgR*mU
zd$$G8&77ObcxK<L^F7Q*-LwCO9~Jw3wpu>T@KvqDT<%&oKR36F+nUeV9(>5w@jPbF
zgOcCZW9lT9w@i9>^L78kKi>Kd$Cj?O*Xh<?a{Azwl%J<gA3b{X#F3YAi+^p67S#^E
zo}0Gx3SS8Oms2mVZaQP)+o6<Iu6Wty(X7m@Q$~{eHaAYtoU$f5`0v6t7Uy8S&)05s
z%}EH@{>AIw+P=#2>^)YT_bWZ)l=dz^5tHogW@c{U{zz2C>SJrz>Bq;u8O3lsj9neh
z_nwnScfsR6&;6RO{Z}ME-ITkL=e|a!;49ZVwhyb9Tv*rWwNtV3;;FlPmOYGom0T+N
zxS8XUO|fWlOIel5o(Y}SFZ6C1>Hl<NjJmv{X;IPUZJ#{u2j#0bDduWMo%Uy2t9d;>
z?@px5r(?%cciScOai01v;oI}}O7YWjsS-h^0Mq}AE~rhJ{r>c-8Q-Q&deSZSV8#}+
zr&F#km8<%>wSPOqp7Oj|&r+UUFj=d(_wOpnwK<kmn>l!ow}tGUu<sGmHFbfC1I$%S
z#jm2{g4wtn@6{|n7R9)#y7I}D`>p5oz1QU2Td;Y@<8GY?k1xo?76iMc@h7fXC_2mY
zn&>+nvCe-R7byp}+8vkE->mxTOl;lmCp+9D<IhcZeEW8GPt4QZ<y+0IrMXM;C+wJZ
zR<cLo*@s7GmWFfw68aT*e*edNy-W`8Yu|mdDl`1>devF^o54S3&AG&;zlZbuvnk3y
zr$3wW_!9q~Fk{)wpGm&*45zmKo8EDGVfnEK^F^O~{9F1+W_q2&oXZpE)%dRz)~`$P
z`EyL^R$f9Wlj_tB|1{<X|EgJDBC?2q>#>4daj=b-B%_{;#%o=F>yO31f;dCxSFFxa
ziFv(x_lduLdV)oZi~I{L1=j66quX}bu(;^5?Tf9eT`qJPi9O%TvG9=R*Q4?Ozlm1{
ze_Mm4k!eCqBh%Qx7-hu!Z1nEF;+?$jK8w^fUVK%@$fn1X!o0-WAa=qbx&G~56Zg~$
zc8aKaUHe$ud~{uXf4^(zqp!-U@BeKMwx0Pvc=v+O|BJr=eLq(OJVc=Q|Mk|pe;=>!
zZvX!`_xtnpclGzT#R_P8Z2fKZywh%q$jPO5ufLo>KPYNy{PMr;##;543jeQPzyIyO
zPo<tKrUu0MUa@+5-TdN-*k?)KqvP&w6F0bKzb?Ya%|(3K%--4QdzW-O9W#0VTx034
zbMNiWFU*WT_bmM8%4wIbi`Upa-EZ7#_x!uO%ydrU)VkJVjVH7OjHb#i(aQRkuvLBO
zRt~rGkF2I#?|u}dmaX(NGVa^5JyQw~y|NIV_V)1V{O_7yU)ng#K7Q5w!rH6hA&>kM
zR_K44;c~|BlSTiWyt9jYXC7y+JAIZXj#Km0?&}*~>F+-ld$HMP!v*%qZ70`tYyaHg
zueX1-&2s%`ue8=a-cnh7Q%2?Hjqm>s?+EDm*VFdq%|GYk{(l;BrYJu3Ogu5CP}(Ep
z7>l)_=+Ei4Q}<3iAIWc3bIIlD`US7oB|24|TIRmb_IY%r_M$DlpZBaPalPDrRj>HZ
z#K5fwAF_WBI&=N#8Wz?&hl5k}PSoCI*tmFBM%Sh}tE^ib9(bz6I+VY0SSrIBbN^st
zv>T^lXx_xL25(NO35MhdGkjcmWHBrA)Psle3fn&U&Xf?dc-S;^ozA4Snq2R@R!cw3
z$$Hkpbj|)p+>^APmG@^9hnLlbr}H0+a&LI^V11h80glVNa~-$(&JH;9W97{G?}9k@
z={F?q*yAkWr`?&I+2TLlELg?2*!<K*?==Sdgyp7RemuWCY6+L+<5y-gS|?AxR6OPU
zvS)5TuWgZe`uO9f_YY5&{PB7<)oX`A$79`0^K85BhSQJxrmS9b!LDf0yu8f@DW*ni
z<9%`rIbYuu?s`<Rzxb1HpvqsCebO)PO{$6i7PvH~Z08S?Hn|h`xfivsly|)Nc-p@V
zgQ_VvstdnpY;te-;$gXCTh;Vz?c7##e}>l+Ka`i{P5r#u=OL#a`yunh>~iU7W{y)g
zj?DVPX~E*4+@xI<^+sx5NXg><6QXse7tY)&a7x=v?;C%`vyKR%<Kpi%ln*5`L~;D+
ztC=V1t6ck#Wny4F=d<ouCYk2Yc}c8ag~~IuJ~7=tyFjmkSJch@OQ@1mxqHPAIk|80
zuZj<{pDH@ZKHZo3iGobX(m9GbbGJX(dq$M0w%0-4VydW?^k2*43#|&7_+91B@z}|C
zb~PA?a866U$#&!+XGT=1cFBF`2lB6WZ|Z--+R7EaEpW<%SHBgz^^BwT8C7qnyv2J(
zY2$~brEA|UT4|d2P<jvR1QX}Sv)mT0c`<ouM69LSdw1)nQMW&-a>hMbnl@2FW7SpR
zb@4YBmvt5{*}Eq)UR8BV=lO=eJZB?gm-${dKDjkS?!?76xj&}7-u<a+`P)pL%Ll$)
zP=2@7e#x7qx@*^DhkOYwzv^gadG+|#FoRbGRv#=E7#^4F)v%f<wz{myt@F^c9;2U6
zzDikk3Cz?wXX=?YHDjmFm)d!|r-!P{x6Up5bS>=c^-EvFY!*mdu#$c9c+&M%mTi{b
zZ`Pilzsqkz-I|)*1Am@A+Vnj+_MG=a-@Wr6Z2TVlxbpnl$#FiX9xQl$ROEiq4iSse
zn}7HZ?fi76KAAyFw@=8;>-d72@MF``C-`k{m{WH3<iVFYtgY%sZc>3Yk}k`mcA8JW
zez9{&m7!lEi@C>HmUUKt(oWy}c)K9$?TkMxGFR6|TZgsq<jl)reX;b&R587k>WgOT
zD(q-Ks2SIo9`$<j?3X8v-=D0%Cw%VYwXDl?L(V1tKA{%>w)&!9^9At}jW13f=Hm^y
zWl)gNS1VQVQRdx~D7Rwul~eYWiO$%ztAA58bGx_37D+wD>-|pKv=;iFa9f>wa4G-f
z%5CdzHSXrQV5)G}OJHkaxvgrziof;_n~vwSId8oFs%CBx=h8EqOaFQNyJl5C=hwo;
zJ&E4Qd%xB!eK+aCyQJgqS6r?YaJ<<qdtdKzv(U+3+=td`{q?o&TfQYTMX%|w@YT;A
z6{c^u?cKD+ySi@%`@B2VbC1rEo0u3mS?f{x)~7y>0jKO=uT%aReYbSZFG1G+mg9Oq
z)`d*l6&H43Q^{FRca7KPUE!18X~yyIsnDIT5uLPT{vx}cPtUm;!wsa<WKT!`J^bPI
z%z1596)S(e+P+Et(njH9zq=BxT(eaxv|Vp)4A)kiCDwn{H`2M|;=CJ$k*7PP-*UKe
z?47EqD$?@1sIIm6+O?Tg7oTWvT3GBZ{racMxm$&j_9vN3J=rdw-&1$!t%+ItJ!wXP
zzn1e%CrdS6%{ukRW>(+k>|WL7MOrIruN+ZJGCzB+>vynT_}q0SUwxc+Co~#-%hPF*
zt`p%tXyaR9rT6pc=1CW3Wt1+yEb2J(UQonB%`IGq4{2U{oxELU(}#1Bk9@nYZD=^u
zp)z%a#oX^R^rE?n#WyhKM(TPmKH#<^^PR{?j;hSeO~&zug}<@AiG6TWZ^8c?iwk@l
zUT{vU?pyY*G4`AC+mvZ1UO&2ZxFX^HtaA&^O{ExYO6M**r+nQ~$8*ukk7qYZE1T?2
z^ZgZAzfN>R$n__sTT%_%SQvT}UkFe5`XJ(X&6%LnF~6;+Bt^$3d25z_TUmG|T|M;S
z4#jt&LD%MLGYWaG6mCi0Fh9go^S4;-lm`|?2fmp-HI><(=QPn}rQFfv?t98B@5o1=
zy}2lkdrjfy-+nJA92RI2FVl<g|MNTSU*L{Cx2>xSW;6%-2QF5>>*YOb`?IvAYxULz
ztYNCp^q#V|Wm@)oZqHvkeb);~`G1$%WiaL4$IEBs3b^Lv`j)7;|JhPDBenc=Wa6{7
zn;&nLt)3aY)``>Ok@0MU+KIQ$WZtZ}eA6!VeZ#qTw^I$@WpCgYIhw<1G2#3AS35Mh
z!W?hrDT|vgi3%{SvNGOygX#YT#f}$e7uCC0J@z!J-?#es(?tfw$3L=8-J!}A7L?8-
z!!BBqzjC9fgj)Z`z_W#VtJ)`-b{V@(*s~<T`7_T&t4j}$Ni3Kp5GnGNdrOIB);gC0
zm&3s}*V?|^ypz3f9{<zN?Nij&?q8+0IY()^!JYc(?JM?vxVQIg+@9Fm-Zx9iBkiKs
zJ09bxv5gN*`OCaSy(H~rOG};v|BYksv(N52K3n|nnicbe;->VOFA%-z&*OJh(x3Uy
zE{j_ql=hd+F8?B+B$G8WZz8+K@4l}N8I=X%UZ<0Km{%OUcChzub%lEQ#o&D_^5j`I
zIxnpZ+><-+$jX$~X=Sf9-btIbxBhuGMS&}ISEBs7?dP{{{+hJ5?LpJe-QSegr{sJ1
zIqa@o7ptfh#<H*X{)|$a=55lw>~l&ja@#g8;S9=O?eZlnP0MJOVL)EU#(8B)AJ6!&
zew#X-VV(NtsO6uEzMfbVKf%rS?2Bm^QYUKb$Lg<(oPI>;%kvWEm1cVi_+oPR$lm?5
zee3q~XV(f{ni2kPnzi7~yPq!Ap1wGDZk*%&0#2Ep(~lC}*XOhIp5&JP<GRM-Vyt-2
z#zQmtvJ~7ZzsPKu=d3wxTKZGX-`gVZZC(HR+FM8eCs&SkD~oCUHt^)$t1)T8|E%2)
zx5S<JH0w{nYx%=syDPpOTX8}-C*U>HMs+8qP}@nH&X~wV8Ap8E8@Xo6U$+D4(SBT;
z?53t|>f+}Vvh2Q*@r?cG7AK2^*^hpGZk=Dz=>9t7dEUE{zidunvsZrIcInNNBa@O=
zEPD5^&r8t#Txor5a9{g_#az=$fB))o_A9-%kM~V$?v<Oh>tdxDgiY*iBHzruRu!>6
z$ZBF_+*>QN_^Klr>%OwBn<bRE{$;tpKZAQ+dWX1QvNu!raz3Ne=VkWraJw_VQ@CJ0
z|LY#%*89D(LJ|gUb9e2CFFjiD=V)bj`F1_`O7rh74>&oGgeYg*^Ew?X4L|s&Svl_F
z(_d*fmd#I2i~bwO-55VD(l&Ia#pkg9Yl{mt4L^MqTl6<={o2L9?g_13uJp&Ncj4{@
z{LAJj?>~C0&#|WY^;waM=GULX-aOXM-h20wRgb-pa(cy1E{W5-PMiFyIJK|!{Cd~+
zj}du$7FOs!y&|~za_r8D7Ba8;rgjB7wf8jCX4-6>zwK9<!Qb!pfBE|^#(u)mJs@sD
zjDeAn9nyIfN25>XsqBP|K_7XxkD=8{qComXZq8jE`4FwEF*fZVkM7TRV`+}wWai1d
zc=wW1K9eM0ceh^o_IsJx%>OHOU;K&Bd%r(7XobXQ|6J#PFSp+P_wV}Av;Y6>`unr=
z_ebL#$0Z6^zg3;&-m9TG$@})}pYz!@*PfdH`QuT^sq0H@{=NR1e>*<zyYN<_Q>!z#
z`ThKSuK3Bcl{f0n=5PPn+_LqwsOPbc7l*cND*U|acV*y-8EJKXL94$1x&7R<B-S}a
zk!xl{TS&{*yAgp;Ctb)(`_5|@pK_gDaLU}9Tgz|qda1l#UU@JrEP0vM=C}5r+&5``
zW?pF;zq970!&R$(P5C``&)-+B+h-D3>s4vRf8kl>Q{K;NPepHiZvB#O5*+dMxc~M>
zo|Hqaal5AanEl#xNqUXLYQ_EOQS4!+#}>Hs>08NVNZ4_y2>poTw49u=JCn=p)710d
z^7Z}}&uBfQ*TvZQdXrSow$KkY%<~@24W79)Q|-dkt3ht8u~XXf*OdG2lg{hEm>|~G
zbA9W~PcLE?ir-^nQqn7V*zKBj-dNWmx8ilpYwm<6dps`Ii}keKywBx!_>}*R&nM1b
zn)m*<(ods*de0{R{q;Fu*%q;bFCt^^Gp3zN@l~-u{n~Dp=HpF|k7quLwfHUNQK`Qs
z_`G)MjPJDytn*%m@H;<fICto(dn1o$nZTF8cIVyg>t6Y6RNLh&kjBOELinsgg;L-p
zo_)+sY$8j{ji%ge>^uHY=YQS#LlYJT%$Gd;z1?f2uE<a0!1kE<SHX7`1j5{fJ$%?t
zFu1seAH3|vaXmebv4(YNz)E(9HFxH%KflfT#N*wvR~(<Ev}~5C`hJ5yynJfgk6$}3
z1+1~<Id*SdgWBAdqMV?2M!H%vtNBt^+^su!Fu=I|n8UR6$r}_uty=K3qIMde6T9*G
zQ@dPc7pDX#O1)t}^6+J{pNlx#L67T^zf9iye7DdzY&_%ew`1RKFohH@4_Ktt^!kfX
zjDuKGkg@gbQsz1tuP3Z~W*!fnaIN)*x8aMBYaRR?*Q6Co6Qr$dHEtwcQ*L6t{QUgl
zrl$@p$0o<}A8+D4ez0x*i7gQ&p$%JA0uBr6n6}y^{nL5yV@*S|xL$$7$9#USiWy%!
z8XN;=IJY?$Z=4lU_F}uA(fOn5`r#W&QorxzDv4%)*jD>{@!?G7i9Z+@9k^yGdGLgq
zp}owp3R@wLWQQFVW(ynQBW5^HXuh&}d!ruXG6zeBsJ_WsuX?gsEACbQ`CTil=DPiF
zc`%n3pXjFG8*jIteexhi|Nq{<$-i4eaz7o2(Xou(5ERYvq{?Bdk7z_z8gobQ@gtEo
zMQ@Cg*rRzolN*`~TX=rXd^LsNwB6YJVAS(R{$?B2v52(3_Ucd;z4FyIZt@qNU0=1%
zPSB{<pLuveU!c6khYW4*<b>i0XR=pYhppOX9rj7!XyaF-BnF)%@1-dgCF0Fz8K;SF
z&lL-oin+aE=At?NUt?D$bX*PYS^C^xRpsN4B++Gt%Y&XyS#HH%@%HJ$qc7By)<+vP
z1m1~X&KCOdnsmj}Jy$wQSl69c=oP<fL+P&hQ4839-{F*ectQ1cMi}o)iL_ORS<ZaC
zu$)mh_0v~Zo%&s?l22y`w9oOmDk!?5ZEDt3>92jN#S#m1j*FdKcBXdKvSsD=;ddsl
z+_6M$z9Q$fJI1qW=iOO(e&aVWwiFe$j;UNbnk(0?ym9C1+aD+PT$NgTXQj(^!>#|O
zf4Aand1%w0dPeDD?L>v#e0CGNvbQ3B6SRGmg`ZE_emb_<<(w$%&D7s{rNLRV1Kr<T
z+A!}1SHMc4+GkY@#3K`Po|aAJuQ6w8w>MXJv}W)y<dE2RqajC%J;G4SZ-bNe+ZBP$
zGd`|5UmtO^?Y93Z8L5g83A>L{yj8tLyQh{51#QbuyEvm_%{#Y)>!0~geV$yJYa+TN
z(|-NBAII*yRnA~$=Q+0IkeG$Xip#r?JT2RPB=K~@qnf1;n_{Q1KIKvW_-LQ<`6aKi
ze{v-=J^B+f<B{;Uoj%%YUsbkh#dK-yRems+%XwG*<hS>lCY}n&4Vrh%FJn&QCFOOc
zoH952X3XGQTleRSGuxY`7dEEG@S80!d-;3ag&0*0FE_!^?`8(N@6&Fm{<NK79kBJ7
z(hKgxYuL|t_WC<5t1v#WY^vFdcR}|&r%j04*vDqPVA<3;Q!gEI`(H1zMgN}F=Dz%Q
z%xcYZ?Z5L}KPKS&O10zs{QMP<ozgUaTsy5<uFZLOh8xpV`KZpGkDE`Qf1SKKZEeKm
zqW{@<e;TsC`_3qkv;Ga2&4MGZTfU0ExjyTyn$<$Py`Rm`{JWO+d&4H~-JzeE^!~Sq
zo(_-G)!nsKdG6z9*S7ktKV39?d6K7?_fh%hJ1RDxJ00r2&h_e4tMl`t=BBSWmb_WX
z<l@ql#-_=e{BAEV-8ePosQn7o*yq<ih3IElMO~QLw&sSkRq5>1QjvF(x<;p5pT!>1
zy*B@h#gxliPL@2ey7WslCR4q~SafZ}`vobA#U6Iv_1AwHgl@j}`{dUQpZEnmk;3P9
zMlxQWkew-hOrdg(t-|_hrvv?`wdb8pk*c5bLEOo&U#q3_;F2ecHFr%s(czI?mf@u(
zsU<DIm0N2v?eI*~EBSI}1}Tj}3orDWeov{-4liEOr<Hg8RmMi^s{K~G?rJ?@Tl_Wf
z#u24T{_BkmzY`zx-1xdP(fexpU!x7y+LQR2ms|A9HW{$;`5cZm=RTnLR_o&tO-2o-
z?~8Z%9ocx)MXTzWN3!gZ8FK<fP4dhoCLjM|q^>k`eI!SJiv3aXunV?d3u4bDJpR*_
z)#m9R;8>nFJI=^|>w2%YrC;YHF0Jcd*R)p0E^Twcl-YCMTqzRpt)720@a3tw72EfT
z&Z}j2;b%OmGx^_=){kq?X6bWFP2}MS*%s!hK553~ck2o&Lau*`Si16Br-JkcYx7@)
z4<mYK22J*Q790C#>!oAqyC*zU)Z_NPe^}*<-KOWScfOdAoVE9T$7F^Z3=1B#$hLC@
zY;rZ)&a>Fdb)nPQU!nzWGuHUWdU`*wc;6)S;mon6)e>o|!(`2GEc*UE@$$(j8xPqq
z&VJmZ9z3y=OTOjpDaD4fE6(<^9DXTjI>*ePH6bpUZ{L=4S~}n6`b(7xFV3iBYu}V3
zE+)<W#?Pefu##>1j3w#8Vg{C1yagYJB<}snu5>~=svtH~x%zL+9JQpvEQ6aSMy|SM
zv!6{9x;ryEr(}Ct(#c+_Skad|`53a-PPjOA<AS%Vn3y_4Yr>pXx2<tzoBn3bLbcy1
z)w5?vNa|YNJ+fx5^sLmZ^z>S(@Fx?UgH9}T^wpYIv>{ab<g=hWkvY1z&&jURV@b=Y
zn=fhYbZy<NUXFm2Jxi}fHLQGp_hsnn(!j;{H#X1tyUJkGW53u9+f%c@s%4-1no+ae
z^Fq4FjCHzvGIq@Nt~Shnu84eht=3=k@p7JS$lNV|@4vP7J09_B=aypm*lR1letojX
z+Td$%yXN(yhaRo)E6nP9o||2BtKrs1^O8+^#Wz13Jv_m&TjT9=e`mhS`@dT}+3>~q
zxuW|tyGn&m+U@d7h4kJm{j9rOe)Hz^#5YGiANc5=+WM`~eDl0jk@J++Y(6F|7Q6T1
zo!siDf~K!7hu&;D>hk1>+!236`(ydnJ}o;kwe(V-(B(Z_?@ul0lXg0|-RRhq#SBk#
zRsSzp{qM|$Rf4ylEjVv+H}}znsgh>(QG#*Xd86#t&iR*U_51LpzjmqD&quIKto(fP
z#5)DEJ<k`H{7^h_Fl$zk=7;JFn)4Xz7FYjDP8IOv%)d2_>C4+Sg@?DP-uc+RENIsw
zUFYJg_VvG-7i~H9cg6amu6x^+Hz<a9FHK$UfB3rM9rjBy=@!S%MhZ>(qmgvE=HB(^
z7iXANZ53v@9U<twvaoIIM78P}^S?|<I+pq;sd;1eu>z&AZ(3_M#3=4st?o6SzhTW!
z!;ZDh4@{ry<nCEoTo=FL^fc`?|G!-PU-)Y~+wQ`bwHtPCl;3D$Yp1upd}8CH^ZyoJ
z`Z8^O^4YhSe?PtP|3}z8-Jgg5-STi%4Z5=Xdy|iF{k{y=D5-^S<p1=&`R6w4`Mr%V
zYSK&-Rv&3sNP5zIecv<r{eNE8XMf?igxM!BH707xz}y_=XqBfjr~8z5zMZ%J1AoNC
z73LQ{C^`r!Y*#7fE|MtU{(X||o48NE1ziKz2mQP{FQIwC@ymKsT>@(+Px|%Yw4-dw
zuRwPGm-B7*{IB_Oc>Ce^|5NzAx-=GB)$CsK-Mwbh#}98G%jXA4O|_E`y7lF@r{+VY
zrEGudtbR!Z^NWAXt@WzC!`}Du{%f1A=rt3A_6EoQ$z`e)SjTyX`+kD|pC@uJb?2Nu
z;c<Rj0(<7R6){pzAF}-GP3O-4BB%J6UC;O$r-9wn?|Sc&=Y~yDe13JVf5QgH`F*Lo
z=Bac{`(~^WqxbDFtAeELm6!G$tT91u-)nw7-@aev+K!o~Q4xpNY~-Bxb6(wRlV8TQ
z?uYgSOWRqiWd6!6l;l*N+QEKf9oLt}kBVIxIkQ?Cs$D&VxB9!k6%2F{xVUWjJhv;8
zUJA~TyTHG7S$*r055B8>9%es#y0iU-Y@qFTpO|ZWo?Kh&k1kbzxAVk-rQfb!nY3W)
zv|}IDd|SWlUcK9!=~PxXOV0u^p@umYVz>I=e4Vr*=)HDgfMdjat=;pcw|?p2`nGQp
zr|Q%K!C3(!l1sm?xxh3#OyE_6?FOe?YFW(}cxpQGo;iOh|E&0H9{<nG`TQ%N|9<&m
ziz~-Fkq2ufc}+Z9xVm)ub6;%G)HD_=o_gb&i(Y@)WcNieEq@O^R=braK3~-}QoAW!
zMQa*oh2FFeHkm?A7k~AxJmt^qTFm`gDf-bCnHLjR$z1$!OzKvZbMqe4l&8I|mER_E
z{xK8g{I$Hz{M)yq<qxc8Ub&Iytac$sd2dpj+Q))dYd_ABmDG89!)0NuaE<ay=3Q%f
z3%4cQn^mvkzj;!{^c~NnET%~m1gNxiK76{;%do5R#NKd;TVFy=S$sEcKl@kjLW9Db
zJU8Z_Y<1TRolZICyUz?^t;om<n-*}TXWoS!7B8N(d^P=6yvb<d6EhFjPfs#Gh^+}y
zTzy3&@yLyd=EeN68_pb=w}}1vljkcI3EKGtwQxJJ#7JIY*~f40VHv@?N`WV)SmBLE
zSmz1P9-Z@#UN)Oa-<Y@m{m-1|LiS(X)^JSB;=I$7)V;Gp@^DD#?+&})VT-q0&s-<J
zH2kgOtiUtXyt39`*PYpxqsVKpncMrY!(9>o73Oiv?y+!8oml^;Y=TqklP(sC`Txoj
z`09UuUi9($d<~v&?D~tQ%Veoqb{yjgEJ&JiZc^Ct8$2wkQ-UvNIP9!npCKzG);|Bz
zn`L!INpA~VCtXvDuy?#x#%#*yaJ@_5^mWxO3Mn1NCJR<HKS}P+H!^o}J7ORfRK~kH
zGDjp~x#vEKt#<wntO|SDD$k~|?Us1qQj#Xt`Q7iFoNCXdFbNK0y-7@)*7UAmo8PL^
z<-y{R;*{xPv~l7AvmbkQ@=dU_+<MmP23zXIeTmIp>toA3^4-fLqI8wsczG?in3`c`
zD*dSV>+zZE`ZYZ!nXOO1>umMw#QgQI5+*3Q{8$()Vj6ylL-}!RaEe0WWUVa5UF;Jq
z6HdK<f3ok?oSLZGORKp3i%(5!R-3XtW&5Scdj8q}ql9PX>M~C1oqC|`!(_Lg?_<R*
zj<9Fysj}C%=-QUm`UD+*DwO>s=H23hXZx1*trINFJ875w&F9JEo!;;7ZDJ4F&~<s`
z`=|4k@&*W7%{%%1VwYla!J)-ja>fTTxX;$T&nTIvUD6kIP_|fVC(k_d6#*d|JPU3t
zJ*<{=-l*u^r21o0j?>rm+C6!g>g~GA;gK8D<>_;y<mbilEBY>b%s0oq%WvkRV?L2?
zocoq73ZA{*Rj=4$)@Ac43paS3E%J-boM^cH%+@8TQa5Bwzin>d**UeQY^QFHvuMI!
z=`3ASYni1Bygxj5`6yMJwMSyN$%8wRayyhw_J*H+H;M1um01rh%9rY8Jn~%>*!+uc
zX3(wZl~U)XG9H+fSu3zp_l{bD)L!`|0$14LAME>`cqC%=N2WdV+zaM0o37ZeKG~=2
zmbb>V**o)#+{_E^?-2dIQn`3P&+g{mxiX(SAKVXBFmisDm38g5;q(OwOrgOSBUUt@
z?tZ3Sytc=paqHyci~dy$^KK~Si=MH0+J@A(+TE-5I{9us%BjAgR2tfO?@D;{h5gf)
zKAZac#l5p-|F&h`vAX;GQM#XJo22C-neOQLlbY_*2Tnz;zJBG^YW`I|AC~>nNie_h
z<m#OELx)P|nsz21v|F_KFo)SgOAas2X??0y`BrC^t`gd5eat}bJCjbG?3UXaGViX0
zoM4|eq5s^6>6;Y17%Pl_8e|=Oa87LH$!Oj#BRehrmwK1iH_Cmu5oP<hf$6oxmsQ_w
zQ(yIRd}g^iLFb!d+INQy9yPx_o;82Ssko{%YkT=Ewp$DeP5!^FuS#$137m4_AkXy}
z(>ph+XDKiLbaPtX@wu~a9_<rs-RipFjKYhHT=Rp@N39G}eaP?qd{HaA{S-~^b+4>=
zBd*VOOF#T`@7FaOr{y$W@|kV7apStFYzKGj=`g&rN&MJTmt~ndbr-dA=DL1=xq5GW
ze6o;q#1+QPh5u&ors?m}JrUvcEpLx({>67wC*IiFR20tiJ7_O|l7QW>bnd#xAEt->
zak~(&_tW=5UBKD=Sx=9LZJXj3(~&pttDdPeSJd^ZZ*DxRG<#^|y*M(sonNo$%+ycr
zbFLmPym0o1%WI7wtIrNKMtu)<$v^)#Q*mATkv)e*f9@4LHcL^fQ1ADT^I=8_&%+Mg
zWIy4y^3OU|l?IvVMZC=S#ja}FmL0BIZ`IOJK1FPHXv+z|#;)a(pPH=-Z|=VPM1QIG
zw9WTse_kiIdxz7b%I~k26y1EeT_9-tBJQ`LUv)APpMO<i&R3jhyvp-gYG1)Si+2K!
z4$p5yzkR~Dv`9}d(tc6MVUFi&T07O$b(T5T9*Xr0QA?dOf8)D3$B&0?OX^LwO55?y
z>%`BSO%c;(Y;21$YjN6cxnav;Pp3)Y;mxW6QKw~tAABzR%)CrtHIu*5%R|-T2l<~V
zwrHnI2YvWnmi?7yvH!G;J6gQw#k6;<4D<W@{KT9G*G|+-(S6!7DQm)mV_)Z9;&-s#
zo2+H@ms9%O4JoHrdo`0gSM>+}@|!5VifzV<zdxm_ULC%<X+tA#-F}D2MHQ24zSmxz
zILonT-;X=ue!o*HmQSeM9e2yRCp<SnTGI8^&lDT$p1Mp6Ylk?O&He3jWN!9Nm^kNC
z+KU-QR@q-ozHu+q3f^Jabm>!l%DVYCcBkG92z()BwlB=W?E=5oaf2WAu{I??bSpnA
z3;#?DN<HA~Id!FG-`jgL?B<@@`{s6p#f!rgdVVuLKf4h-vyipEr}}rLoTh{L59c{I
zcV77;#(#z<PfdCDTsz?dx8BJd-O0YU^T+WfUFH>fX8f1#d!}vf*}c!dT&ZSj_fhKz
zk=l=GcCshmo!fV9ZRJi2mt#wkj&1+@<5kGiJ==_W<`?lT$eYWh;*gtNYqKi$)cN9z
zTfSwjxO1Z~Ja*kw^Dl4rtX-WVzF?DC|MQ2xH(p-&Am_j&#Uq!0A3A!Vb-}d*WraP9
zk~hsXQmsCg#i;dW&iO<4XROj(7hk_TtJbo=bWMVrXNA+v7ZaB4z7Y3o-zMF)-=5vs
zW%$BB=-89kwd{)i4!M1-uiyEk++|#H>QL&1=ev4OCV!Z%E9K07v*L;G(ZzY&eYyR&
ziMt5RNql^G=3Q3a*&A6eYkDse-Ku;lP2<tFL)q3VE?&McpS$_=Ik(iK3UaCk*0VoR
zd?IaDG;?9YhfPawc)W^qYY)zvAEJM)&(88wZ9d=4+f#+R;xevhd248DzIeN(ywE?T
zG$-cyokzzX&5Gk%eVj+<eV=RP^an@p&A1f2`3KLk?X&k5yt0yvzP^i7EOh#=t;-ii
zoR|K2dls*1x?guky2<a?ykA%K-MQTv_r~90x0=Dcr7O+#&;6cyIkBEcPLSbg(xx3*
z-{br?FO&ARD!dtVartDAS0-~lTQkk~J$o=%m$x-|cOlbGrX5~S)?NtX4;As{{JQjZ
zoO3gi@kB;BYuh8rCF+5-U#7g}Thn*_;*QmZ0gSfs>I)j5J_@hjF;Ok(`Rm|&#f$on
z#qX_P=zrhXD=x$}ovqj4zPS7ErM20t7kBMfh?QS=LQ6EF?d(<YG{)0hdka6$++`)i
z>?Bobx@Fbh*RISj^WM69??14$uf_1-3fA<UT9u`rKVM7FbDSd|d9bv`{noGFOVUIw
z-n&d#{kYbYZ?_5SnU!64=Cq#D{T}z$DCq7EFSieAxteP)>-NO#nR0$kxv5cRkJCQk
zCsAB`H%iUxPhDqS5Pqyk?vilzz4y;~S)>AmH?rn_ExA<toM)|M(;YPtQ_Zg1pEaLc
zdFC9thT-wUwrF=T!#O9qpB7)Lvg?`paA$0#v_6w}Y2o3(3#|8#zjnW;b1rAj@7#G$
z`$G1eO8&I<%=8fFxj}~xU;bWs?4kbJ_Pz!C_w2QA{Ox%A_hX&e65lr;*^t#!v1jS`
zT}O7l`?Sj1_VTK;lU>$?*Sb~AdAFlK^xw}jf8~W+cKIIHzW7Kyqm6TSVfh80gg=@g
zKlPjCqOM)&EvRTLijw>_!NqBQ7>iuD<kOpe=j&?DX}iz<|B-2xx`XG_JG!g>U(Z{T
zncH&hp`4!$hxmVs18i@e-4pn0JAq}wzCT^MKPE?iP~ZBe$biXY#o_(@x1L&GbZ4)q
z<mI#9E|%)L<>80*hwt;(-(A0K`BNvg`b$yLCAvqA<8I&FF7dBOz>2N=Mab{W*6o59
zIh9zhM{&%}YZ09EZ2QU3C#(gh?{V0qzHS%1v&p<^<12T8u%)Rgmzr%1yp|X&f9{yO
zb29tMI`-d2ay}yeCAAd-Z)$((Jn4J0(eBPBV<pEgSr4LeZ$7`}@R4t|D*v|Ic3Z3c
zmx#+GR9^^E|63}MZgb{st@hFv{CqyjFXt`a^M2h=;U884Y2Mp)_wgStJtX;`NA|*_
z-@L5vcWwVA!>i!>-$Clu`HaKjnGC167k_nAyZbhKzh+x$-r=e8W&GyXKO8!CX8)II
zIYx<2xvG!0nir;i%KLOmCcc%w!m-?0eS^8>s?txdrG@L>^-b8q@o1*Rx?dZ<9F<t1
zc*mRR$gJf(2j=(G%y&>wX<JcQSK<3bI_N|87dDIh^Rq(xe$Uz<{a@}%ZeQ;9Qwk@8
z>|$Q*@sHa)Z{9_lFYEXJ=2ttk>pGTRusLzPU}JMT<e4!<FL+B>{>p|ne#T9VMYmJ(
z9Qksq_e_72_<hoSe$B3`(5a{UKk201c^l*@BCq5c6e3?|zoNON{*{2qpXu$N^Vu~+
zdOif-X0YG8t-bzV?ZS`!@%;6F%YM9kUciyW6(qYamhnRSq@<$D4{txcKR+mnb^hs(
zhb3A4t>pfHek<R;zjikN6)%nTZ5OMW&d)r4f;pk^hiKaNr;`O%_j_wdDlg(@PI;Ub
zz}V0{pTn>|wneP9=Q4+{x8#y3uXwH;v)p$eNm2bs!KpPlQq$53Ctt{$apWM&H|{!V
z&A;UdXUdb-hhIr!WG-cLk=bj0qRQb!e0<7+OHnL~-Dka>EE968`{ZrLIqZSe3ogV@
zm}|Wu;HJivp3Z04jT8U;^*MjS<?+j8V`1UN%0}N^)3SIo*teebNw5r>S+8;;_Owxf
zc8|&9LyXhd7Wl_!6fU{nSf{-9-6rPUTfd*wxD&TI({|oU-uV-kveoU@UK{Shetm1W
zzP9ZGoxid<bpkpDH;xFtF=<kr&BgHYwET^x|3TdUny20856f75^<h2x<5tGczZoBW
z)|0E#<~_w}I`eQn|9`7LK@I0OGCA;Nuu4mXrYuoWHl1J9ckexW;<6V4;%3vsd?vfb
zEb6#8o$b)lm-($elW&{6+kR=oHG3|3=iBz?YrdZUvE{F@-Oh#2&o=jL_lkIsm}+Bw
z=VXP!<%TDIJ@0?=?2s_j+<tBbXPMG<BQ^6iU)T#Aw>CX$D|Osjv+jqEQQfiGmv%qp
zeHVK2&Y`x<{pr#RPN-%0mNok7&Q8m@e80QRc1O7;Yg(q%g$+d^H(DG0T-LWY{%Tp*
z^GC(I)mpDEZFWbMSoHeVxrVuYZx1ING`szQMYu`Cu;<R@7dZ<yB>cY+{_X9DlbXNo
z<~-nk*LR9zuE?gj7mkLT$BMR?rp5HsO=gs0tFY%fa??l4;NWxq$?yC-HT@^Ocg%d0
zaWXLS(2@D!k%9*|oxOHwU9be_?&&e}ABBF9y4E2%bIILh2EBnvM^4Q-@z#a^Ug-M=
zlFz%%wHCMXFZ>a=+oDXS^veeJ=c{;`F33o3edA$kSkIEUBC%?3|9Ppz*IUzC-fjwu
z6ymM>GIbI2_VatZTsZ=_Et%T3lX+Kl(IbxQl8S;Jof;LaiGSJ~efsnkYFGSb+w(|@
zXLiA#3zu2^56YMRP&xfj|Av(PX4{#Y8hbV8ocViJRDFS=qg<&^W4h$=WWSlw&mXt@
z)=uuPeY!7ro^tD|<=aeIH=2F?uFR}{Dr$qkY2g5#>!+Uln!*?-F~f}2W9HUXk*)C-
z%QE#(uDN(~$D`Soy!>@NQZ#n*rlwqKY&xnb{Y+R_HBBp+P5LzZ8}4qOM;kktCq3o-
zJ$bWg+F6mG)3ok!6zu$@m+8lGH8-hcuT9X$xfafkYnI*BY)dkVc{91}@rfC~bHuIG
zlEaNemUf?Aqwm?><@Rgd%1yFNTbMptD?81uT|B?4Z^L~3_A3vaf<sQKpJFQB`(crX
z$b62CD!udTzO9XmdfHk#qa@)wo5I`~Jvm-Er~mS(ZVRxJ@L9jpPENRLLA{ptgs+t~
z>9SjdkIigyx#YBZmdlKbkAs$pP4S=PvW?4AT0eA)+E%s3_|C-a33*-jms~5Wy{K^M
zp4bM<s~6YkhemTR(UFz9T-CN`&zdyVhjEL`gjLvu!spLkabEE0nu|*1fiCiq8hbw}
zEqHR`=B_6<`!*Vu>TF3-POAuBZDaNQ=4Ed++sip$(v!{gJy!~8dTJg@53OT*@k!6o
z-^Y9R(GAN2`}>qvw#I%`TE!QXJ2Udvs@f&WtY20#n@D}ED}VRr>vf}($JEPz+zcr)
zuel#n#D43}!|I}2JJdTX-~68O@?}k-zaiVXc>SO^%^auBX>6M<!mg_J`go&aw)DCc
zB7aT`*?ZkN{c~mDD^~5_2M<^)zg@Q`*`3Sd;I}IKH8rmfKl^w5v7DhGOSy?4OS*~R
z!SCxBzebq}uK9W6wCT0RXS-8nIaU>W&t5t)xNG4PyUJURHUde0JEA0a<V&7f*46pW
zH`HHDy1+Q`!YelJe@lLD>UEwl$HDa1mA$)9-}{;urptEW>5na;uiM^dU$wfhUVh@&
z2lq<duATj~T-M@>^QL*Pa-Ad3AO6)^SeIxYaEx>HV;<?9-dts-(Q3^GiGH0=j;y@p
zH^*$Zqg1*7I>YT&M{evlv`tSq$!dK0VD04ez%GlI-_ITVWmoxd&(7?wqb?n737W?O
z!g7**F5Y4DzQ1X5x9+<#C5GKS)@Hnq-mu)R+VWB-;nmaV`<7c2({^89cTj)fDb2Tc
zEKeTH|DC#nrEtd2Y1ieBIh1oQH~TX8ae!@S@)c8OJMPY<99C=IeKHqnTYUKF0<*G&
zC5n<ideRqN_w%(|U?&<iHz#xJWvjJ2^RqN3yk=T0X<T*ryYl{?#E8559zPMaz1z2d
z^KR>k_2-YR;XK+fv1j2jj<QuA5-iJ9c5p8|*;ex8oRY=0Wd(}CcTei<Oc1kiTaXjM
z@N9L9<jvhtQ)75#uS!f5yQ|L8`#kLKVXplrRZ<O{)mAd)FMFhOWQ!c%dhJT((-yN1
z82UareW+T&JtgT4(~LP@SF3!gvpy%pI4MsPZnOM2Ep6t~n9DnEuXJIOz0rH_rIS=f
zpK79OTcdi=^4q~i&ku(`zEM6iU9SGLMta2ZU7y=^PHu^MBDDF_rylc3Zl85zPxx7#
zyB(jkB`sw(>(b?$&Z+jLN`6RMXkK?FW9mc+w~Hrc?+$z!pR@4Jorx(C2e$TH*{e0B
zN9EzAB`3A-#`<~v>iMiJa_89g=eaxbZ|W@lQ2k}MsDak@M>GCJnF^$)JzcT%${}l;
z=VGg}O<ev5@v<|=zNt#B*nEd+YuR#-C;mJF@1y=$aD})nyu)c!tsZ^lL(QINUCDKi
zHhPvVpLcU{<I#yb{4a1+Z9Wri%)+4EZd}H3|F%roDZA89hHvy&%_-WtRdw|uxy=Ww
zOnq-Jkvx~1w)p0bODjDrH9xI>t-B)G<Y-}Eui;wVY2S0Le=W&ma^9)twoYKey>pjt
zDptnzx-Po5W0vcoPZLfad38!>m(sfLj*iFlK2)ZyUMXcbJLR=1%eCz~*`;6KOg;Xr
zW6`!P*><@;i?b)Bm96`%JnQV|OKc(=i(e-zr}Eioozg7fn7TFhjPR^!Z!(^Rp3IMS
z_h$81RsS29+$~wDy?BAo-4~V%ry5S}-E>AM`s}LvUoyDQnmF~(x-8GWt|H0Ui-Fxt
zb&Uddx39HoPu8EbNUyhU_rp0~=j?uz;ru;{MLsR++OGwbt)f?^t;tq1)t|Yja=Bjh
z?CM<ksdK)2Y9+OabpM>^qxJ1@_@u7aH}_|*<XOgivt@U5qzZdWlr59T@u`h=IWKQi
zwMLy#tywUM-}TMve(xQ}Jbun;tll-p`p%i#jvc9IcmiePS1AQu;%vWpK<-tRtYBJY
z47ZKB&cxLUZQPURC%!4Jt+=s#gIhM^$HzI-Vl^(r?N}kES@1hBL9HV7zH_7Lp&zM|
zEq^r%{Cr;d8htQW8$HF*OXaPLBSW_C`wPvj6Xo_yDro+;{6lA-*5UgLrxyyzoeq93
zQ~LYm(rHp<!vCk4E?g=ecGQ(?-?zmMu0Ph@b7l;idMZ^i=KJCU3<clSBpa^IdbEh=
z(#-RIA{M-PE32IJw|-V)+59x;PU#U2JEsFna}u|Q1e%uKT$=DPDCuAIDiJn^cdF5h
zY#yx^%vCoxi)?S7D|BS_y^Eg&KMK#zw|n&F`7zdQw~ulEdtqBs{U$?f%DDxVvJZ~E
zUCR=3DpBWJf6OXl568aSpMnjgGzCvSW}lFHQ(I?N*?PGv&65=_X7E4wX17BB)dk<r
z&tubB7yhyop1<!L7hkN^hPMyj*3ZA%yv0DVc6RUgIHrm>mpJ6#<bQh8*}b^JnLEFm
zXT#IGhWELPt(DdG?lh^tvo>_k@+I#QKR27FiL_o}zLUu=Hm!SG@khzZ<5TBo-0}Hl
z|7V)QbDcHYoL2sRf4p#Mu=%@pLDQcXJpFAXy~B>nv`ga7W0@H|9`#S|?8$I{SL(Py
z#H>_!Uyb92H=MdxtlzPRJ-)y2^ExIKj{7agvhLLFiha*67W{vorisSh!u+|5B);Xv
z=3UQBo^<03%dW?|j<UT4c~|bwO=6a1w&*ZE_ismaQ>pojA9>|W^Zx%hq;O?>yr_vC
zm+4O5nY!0bL}p*U`#2})=G4sfKVR&da`siGMeIT`yAxizC7gfH-2V~Mb?W-7sX|q|
z3wavvhS&){Ir#F6DepXGqg&r?JgdEKZ`BD~K3%TD#O3q-;9nCBXUN_b?|5sv)KYm?
z*J7{4{#L8)&rcpzk6hhUP{t^~C*nm)_3uq02}}QpvOn`Ewwhga<@fxrCwh{*zFkUb
zwOqZ-Yi-QR3u2E_WZDkRoi}MKcbDYu6NO39)$RE|51fx*wf~%rw|jfv^dhU7*QQSD
zI&4+4n``gW=zR(r;(0ZyWo51zo-sR37VYH<>b~}>wei`5eXGlIuBF?3f8A%Zv3}Rd
zi=poOD!5<Y%;t$JU*4E(D;*J>xJNHMYtvqi#coDc#tnZ1*D!YTGV<oUJ2c(f!Ed|X
z?V2fkPRWZl$hU^>Z0*n#o|k78eAB${%*vG}osq#ae{gRVwY7a(RA?vFf41sUO!<WM
zFJ9J67Ga+<$$D}&vvjGaM~Ud%lGTFWW54D3sI>Kk&U^g(cIf<>2eS9}&+T%2Bzx$<
ze({v9MC-enJ$YAb4{_|OoAEJoWBaxlAC8o>J)ab*7d+KFB>kK_Q%g+BMKj|`8n+@(
zPwQgXm$&Ilv*g9_npIsl0|K@N=yQEs)U_*2ZO(0`hF{amr!;k(V&(X<=-~d>&F#5=
zJnBo;J5M_NQ0#F{^<DM)V^H`7*L4nhAGDNDPK{5HJCS%nwS7~-b^e0adl`(^hvsyd
zKS{{)bKV(Tzq51&>rAmzn%NIqR@E<B@NIp4Z^B<Do%lI^v)v32{@WDv|Mnc)2fD@v
zR>4oU#OMA<K6!Al19L@w&xd8DVp475b}akf8fEc(Fx0-?yl3Siw~Lp)D0WDk-aC5|
z-_NJ||Je_!7KdYYZw(EJTlr~>a*A{8rTFY-0|AB)<@T;})@~OgR_k_L$hzB78lBBu
z?fdv;i<I+~_PXCM=YC0KY>h9PZa@7-;k1-h?`HnZjhg9v{#2AG+paPLp&Pubq9X64
zy6XOp{4u@v!Zca=)_)?r2EX6h+W(ysXW#dHn_w6FY2%Q?u``~2om4bmG$829(}j&Y
zH*Z#{FR?Y5sTD2OCYJndMp*gY$_+=?W#oS_`dT}a$!D$BwjIYTzddQbR-WPT^~}aa
z9LpbXvArpFAye@nM`)v{RdRR4j6U5>4x;~${hW3wZ|NV=G_L0#CEi)s^{efyzSK2k
zlFgM<s%t$NPpZoX++%xcsbQ{owlaE>|E<Z-zCJtAQsS^SQ%qmAP-^y8r(Q)3e$fdv
zr?_K;*2-PzJm$OLh0DRVEr)~7n$GMpyZi0)rl$#KBC2YHDpm&n&1O9xzBt-8_2GB+
z4(TNs*y4kzo8ycP%#ep0Vkh3uYBmr#_P(|&rmSbtv1@A;4y`KXcpaAYn<3L?mxHg<
zmF|6izsS5i%EbBSQ|<F7=bxvi#$}lr{D0E3CLy8s($?mCJ$_8P9anD=FACGP)M;7$
zrseCOHyUrYzi9Y5y*6;3UHtyO^vj&GpT0`o`lxs3SC!_W8CNe%`2A_afggub<7|I1
z?g}z*Q|&xH!+z?@x&r=fzgC>(jeE9NKD;@2SL*8Qu!nx8QPu5yUk)(o`)-W&xITGV
z+dVE$zZYVARA#uHnv^89l6}{|<?Vt~?@ImDzLWF(VhVrLjO8AdLgyza`Q4h<G3TQ7
zr8NI2zbT5jd@k%o?<Q<hwB%NLCE4mx*zm<|`SB~pN>$NYW@NZWHJp!nXK!EN_e`=!
zX722LoR1GPJlh{6HIFko`MK7d`olhx&tnO9BLiZ_c}$E^7C^^Nj?Qj&5NN&sSM>Ld
zpswjx?}{`{$Z%@C*P%OQpQLQ?!gZSr`G0=DYU?borSf5?ae7+6ab4Jx)|YAz_upKo
zmnoE+F6I6FVN$k`yH)8iZsm(5UynTb_j;d{t$p>nW7AiuKdj#;65l_2_wwh5FNCZ0
zyxV@*zxe;hef(21R=B?SU8*8zC|+4<>wiUjne36Ox10~<)1H1#S^p^QGLQSN*n4|#
zwdJJyJ$k0BeRuXJqdxm9lV`d-Sh{jD_r+*ilM=brvvN}Xo~&*D>3Ouu-hQR~vY+L0
z-b-X}|JuFDyJE)KhmH#ll`PaK5^-1PG+w6b(KJEbe1Yw`BzGY#Eym0VOlv0`Pv=)S
z_n1p+i?hDAKsdwE#>E=!tsONCg{5kD6y4@;Jakj6B`$>Rmck^{o#M-6wiHiUP_>7B
z;jSd7Qcio}kom$gNBc~KSGivZxp3t}l%sLub*@>nIJJcI&Z=2A?dF=r`6r|x-1uSL
zq|gHUlV1z$d)F`8DV~w_PU>8?#fpdP+-!NvzxM4nv6*|k|9Ot&k<{9M+m`Q(u<qY{
z+i<(aX`k}8f7c3Rj;WshTgRyP?Wr!dST`gp)=g1!+vM9>j|~Lc-v1TZ^=6ZZd`Xs~
z#HK>twr!EK+V++`@w^sr+O1IU)3@x`@>Lt#SnPV!{#j;RiVZg1H{bP_b<aD8%@*JH
zahJ&EADn1$ap`x@xQ=z@#=(Eeovl@0zFRx@-t9)0UH0$p$N&FRQ~O(@{o3=J$;R8`
z_g34~UE3)rJGb!XXS<0SGiJXk*niH%<eyE~Sv5hWzY$6C&kMx)Z|B)u?Eji0Z^gBD
ze@FY{cLM%=+cSSHYW$<KX^Bqq4w;)fH!gmC#fod~{(wy<r>z&6UG?)+OU?KBvFQcY
zbMM|aTXX$${OPYc!IhGk;wKxrn6@#gBqrZFq~_R>G(*5O&oOk#Q-e#~e?(Ta$$LE8
zkSTD5SNO_PVGm`sPgBIqc{_|ZE1c`;{3_s@+7Z+k@mKQ5E1!UX6^^zCj~%I!Y5h9m
z0_RsICv_fSE5(HiCU&oKJaN=OQM@zD)M}%JOVC6e;qR<pGntcEudFk)XSO}fV%z&u
z@<pUt$ZEsIOBF93Yspe|cibu^Y`o%Hlg{fzw`pGuM7deJzb7qJNOrB<8y3BW-*^&J
z=*Pm(+Hv!L?wWW1_RBm)ozp_)7V94FGTgiTz3Kd8nP0p1*RyW_k%KKY7!i{iOf6A{
zdSj==ci(alIeP!I=CeZY$TNS!1soSQUAQY2yKudsWNw15QxAXLZ?$5+yiA9O<}-JG
zpJQ!aKI!eX-Md~LeG-0Mr2Rs7ap?5e9j4#caXb>y3!UqsdTP}R&yV|eJj*pYdGFIV
z+x={YC%?Tv{`mcS`+MKcnyyd%wTplH@AP|n<?Odz>NJ|O@%Q74Yzt59{`#=+_2JU+
z`O=2-)t!9WlCQ3PFZU=l(ro#M+?d}tr!HLms=LB6t$jz)j_q&LceL#i&)gJoq(JCX
z*D{{%X8KOkuTP(F=JLbHj-O=}VxB8b|G(#N+Bf&y(fNC%x1YYc!Pd7o#^scf<peL~
zm8u?z%b7CYsOWz5YZkaXQ6Sj(l$4?I1-9or8QsMSx*yLqO}IO8Lae_d+ZMLSlm_;<
z0!JHd7EM@YaeRicz~;*n*)Ixqu9`V_m1CP}vzRl#v4HJ_2L~7EvB}AB%AMh2;qPPN
zXA?E~@W3PF?3)nZL$)(`Ile#eC|I)a%%SF+O%D!EeBiivmzZ1|^TY=SCs**V=x${{
z$`i15!d<83=|11MKQG)RR{LzhnaK;zEIKbUrBcY6#kBa1z@7QcM!k9kA#F8$r*w85
zSg3BpdG+cUhO4up7Tj7iU1rs5&lT%Z*{qz({M}Won*99TRjx^W{j$cCvrlK$>xU;M
zKRB^?v(0Xq@+Z^($j!Uo690Ger#GSd|3y7L^(>op?{+C=ZA)j{O&h(hT9~j+DqOcB
z<k%_CI>XZy=9A*jaBV5QVcGq-yEscG<uOYg--Ac@N)KD~BwO*_DpU(At?>J_E%SKr
z$M51*jp60kk|t4$@k~ujP?`l(e6w#m2(-QbC0gyHdGOM@GIPe{f<KNtzR|CClt)JG
zZi*1vRP$l~`Yq=tIJ219yq-DtyKHazwp(k<eQj6Ve7{fkdt$)yT`PYT<mT5(UNM~W
zTJ7YPRj-N!y6!(+`eM?}S8HF?iA`USHhcf;!>?a|m(S17y`1*)@%(~)wIY8XKdhZ?
z8QLm*@cnl^4c$G_@84fnj43<2GW2uhqANK*tJgc%hx4CaW!sj2`R4k4T(x)STs@q)
zwO{>r^!3^n?+mu7YTd4T^qgaFhCRF`n);&qnare0D?4t}`+et?IDNl9Enn|xJ?~V#
zt9@&_esaaEXE&8DX-#S_k@D16^j*@%GttR^;v4QSJZl%KR<bRcC%kChL@sHi;tX$<
z6qoJ*w<WCs&svJ!E!nm4$-@&CpJbk#6h72EG5SNol6e&$CR7MS3GZw?a#&bJZb6LA
z0v&Pw^Ajor4mB^<w_GKA_iW(Y^$H=@ih;T+tCr4NA-3qV<*H&E3q>1=i%sljr8ied
z@GCz&$ocT#h3wNUJ)9E@p0qw$dqKS0-q*HUxmT=PxwpKVU0*aHZKh+pk$#I&cUHTR
zzKZan-^Ce*K8?b<7DYzaeHxv+#TV^%d*?E9;;WMDM&5nzsuXu+uYBNYdUbC}=e~+5
zk-r`VK5GqlHd8?Vy=|HK>1F3$hghF*m*QXEs__3$rm2ti+v3Gjx9^h|_qkJDSn3<S
zEbq|stT(svPIb-yf5zA(23sm4>XZmmW0YHqW2c<WyX7F#`rYQJPH?E;gP-0iEmIr<
zGH07;on*H7=Ag@Z+i1gy+v3lxPbp1v_+?XEd)`v+{>P-fG272x?3v%b?$(mUH_dWw
zmpGS8txCCfX<erK<wI9|OGTdk&x-%H_~WCEZ|0w>HF_LxU%zJFy#4>{YTn#>_v}~x
zt-1G3Km7GI)s%OcU-Y+qHD6AOJo;MnxzJ+S-(7dh(#)hKjqRqJwm<%mxbEGnce=H^
z@BB8tl9_R9?#`;%oKJJlu6%$0vfWm9W6876e-;KG35}i0R~xx>VNQ9b$?Q4ncg|dU
z?t0CXTfa_MZJDw+cXj>Kf}3`G?epjR+$outIZ@NW^<WfZ<TV$)$?*<$!6K|L<hm{h
z?ON!Uy?w&9cQ4Wvr?3j1dLz5#OvsA+%%yx$f$TRqtzCo5syON$Ta>%Sx)(R8==8-j
z6o|N3Y^vhwTl4E=V9OJa7WUr^ClB+g@K2b`KE<+fN9z;EmUag(zN?LEWLL?~v<zXN
zShA|Qq|4*p!pjS-Is$TEzPtD`Ou(v{W9~E?+w_+)6F9F1E_w3t-N%3x%yL|tzEx;k
zcvPxpF@GM5tw?~FtH$R!YZ^9bEZY6_E?enw=L<KV-etR2A~JEo)nJbctK>PWLfA{(
z1UaipCuHq%Un!<(T-15WrzOkZEAamOjx#ptO0R1CK2=<h*}>Cwy5_e`{?$y0bC1>d
z589m2GhUu~`T3kTyZ;4Et^d!s_@TKxwghHMObu&dW{O&jUF=?LAaLydU+qsjPH$Y%
z=~uRpV{Pujg1y$;qWRvPeYHh<O<>UP3;+L_+`rH)9Q7kr?X&Mbi(^mo_7=Zg@hE=Z
zdVZT##jmB#U;p<kJ4}7w7Co=6S+}Z0xc--h^9IJeTm8`e>c+1A*?;FhexHB;`}^AP
zohFr$U*pf7{+K*_Qr3^2EB(I@cyYhK*8aQnN#65wd!uxemOlEv@>A)T@~>NRehJ&3
z&$h2UysuRJYl#kj;?n}@HvM-udbK*cl%{xJjJ|7LBfDx=?NdD&-AOMsuOwO9x*l2g
zQ-5A|j`iHv=TDt~`7+DPf0B+a)6_<eC8kMIDFVV#N@)%&oSar9O<TZgWPCwvrUEnT
zDaRHcrpO~Lae=``_Z@;<ujqP+#4)FOD(q!j8c@O9xhtUW23Jd*2gj?BrYCQLg0Hk*
zTj0Au=G9FxmQtanU5(c*@60u0ImHyIxFUO{<~uLZIcCS#3Vc~3@TKSum*aUZ$MX+r
zSZtZuIsGIo{0p3%`!9I&J?H$M(E6ZTko8B>67e5Nh4m`2&Ii4&J6%$*+8D5Xhcb6*
zxB2O=>tA*_@BX<xsEu)gd97S)WrO_Z==*!-vuXKMN?fg66?MFy|KJ7w&+9(?=I-?S
zcMMyyFe4^e7@Jt2B#TSF-OUC9Y!7ztS30ie{n0wlu_Y{A;MU!s+o$j4$n19VRk`K-
z|8G{@%S2`_f1{Z_g>SZ;;JVp_MODgj6si?~f<e~(2Us^kL+Ut$GL?0`C7K^<a
zT~@W?*ZNh;Y47`QzI^^+pS{eKklz<euUVI_zdpVAT~SzY^Nr7&)t~9=ZTe){b47Gh
zpiZCeu{W2eR7rfDyms5xU6*d#d|#V=Y3-hkS5<Y-`uu$H+qzW6{aJ)((xsgmek*3R
z?Gsv}U8$^Hwrt`Ci}`Lz6W9D-TWz;3R{CS~OUpZVeIND-@8}H7I-=Mw?P{&jWW@1%
z2mjN6geS_AiWeU+{Io&;!-j2*MlUXFxTH*Q*1IP5(=E8;qL|ZMML&L9<vyK@N-3Oy
z>c?MRT*TY5%ChKJMEA7Mb_)cYt_tqU*Gk&==&9cG)AOpITvd9s@avb;KFTxGX8e!V
zet9)}@2lsr&wes5s9a}_Ery6&24RHSQMwce9wvHFUVqfKtS2ccI@ht~s-M6!v$WF*
z`DXLx222)s>-_)kq&c8*BK@7GTfGnMv6-@IzQP|9uPYXg*H2%2up-Y*Va~@(ryJZa
zd`p`3^nX%DiNLea#rJ13D*pLt^LF+7=i65W1r=`7S^n%+_tTdVA0(pQXnd=<&T;Y0
zr-1v%IopHExGtG6aO{6`?QzPx4HcPJCFkF{)1G<p%O|Gp;JDv2e=Yke%05MSgTm=Q
zWpnt{L%J8|vCMLdGw_%xE4idqHnL&U-|gqNPbqtMBtJRE;(V?}vf=qTi)=GpwLG%d
zhq~OFAQa>?OL=?TYs(_{Ta$_|HGc70w5Q`$uEko>C0oM}?MZU$@(58}XcK*+`<!HO
zYs<MzZp++^^D4`>&don}hxNKz`R@MHqUOmtoBs-bxc<WGo_Eb2p|x91f4VOpZnOZ)
z<pqXDNJB)yrAZ}~1*!VZo-S_rc_ks01*rzk`fiyyC8<UFZaIl1sV=F>`6;RTKB;->
zB^e4vM#d;*>(sF75aDpC`4dvqx@TPC`8aW-@nf&6VZ1MHm1eJ9w<}5a5%-E~;w&91
zpFBL81mqQXnNF&-v~x%VJIM5EG=;dXcB|PI-N$|{OFsH`-u1GrW!WEdUYmdUb=)s&
z`_;|1W#7-uu`Yi1_ubz8@4x@u_dbSGZE~VeXUda5(}kw5i?B*eJY*Eyr&s=Ii|6St
zGk@N+^3E{wlTfue>Ux2xm$S!Z!6B9~t)6KMS8=?I7Tomv;)?7kHNU<$e_C$f6ZFYt
z#rqneHn%@vf4Jl21ZAZ-Tx@y;6KAjlCmuFr>pu3c_3reRzt=N28~bGbVwL<AGy7uN
z{6)RJbFHqYY>V!h$tG>BYWA&M*!l10H!`}vH|g1)4oNs`v}xBJo^PqTvlpa_&*VB1
zsBC?2%IRe<d3U`jU6Uz0drADkXVG89v}b1;KApYxe8x?c=8LDqX6Qt&`LXrI=C4+(
z?V8V?KYJlMVz%S?bN+rlNsoPYnw2dyjGdo-b@ODkjcY>Jr}LW4uB`Iu(X|iG-js4|
zjd1Q7!<+m)YYZ2bPJSe&?=@>n!NEI~-9pbZb6d^I=AJy1_9E9Rd0Xi3$;Z2Wrpd&b
zscj3q!J@-9yUF)(T2lRyX9<N7Hq#n(4^B(!j1ZdEwKUOfO@LOA`Zf(6Etf?+pA+tT
z7^#LIisO0B{X53W`&Z{0KRM0or}ZC2{Sh&o_#?tFa3d$5R_m0%`OZ=2OZSI({EMl*
zx34E6rA%aLVfwDj4mG{v_LA5y2TNC$x@8^P)WZ7uoAqkxvg_||2k%HuuAF%~^x&S$
zHP`AptZMaAr6xP+6#hFoamM0^j1Bu&T|e_jY5((>4R4*e=RUeQ{hqJscawip9fkc@
zFIQWlvUk?S%A|Vd^GBJ#oZS0-54#Q5(hqA6^XrDM`x(k!o%m$;$GvxUKPgk>I&?KI
zJT-IW>Cg~A|MSsd6TFU|-^21)@)Cpa`nkKWlx{Rw*{Ww!$aXo;o!PnUDvK6>?9BSI
zXD?2kU!Y;~>GX}=0yo)?@VRMTv8vs=AMt+Dg`YaZDr+>@#fTJ@+;+Y?qhNpPRK
zhrHCL(0T4LxAx_VHg|Qb`n2WV%PW^3Ctu_+`4e{Ldj0EtF~LXAD4q!EnDq3<*=<a-
z)|@nyI?c86j-j=pU|7SVdG;4)PTx5@^TvVD8>aj=r|<8)$zEK$YVY2FTh4LfS004k
z_x&gHaPjh9SN{cx+tq8>H@$m5D_ce;{8f?I^vj<=PwqLr^2(KKdB&4vpXlvf*_do`
zvTE;~<$jaj1Wool(db-qmYx0Hf?s||lxNNkbZU6as>=WQSJ{t{x3w`J(&T?K#>j=N
z=1+g;7ys?itBWR5pPcTwzhGOauiw4=Gl&0c&c5YLj^}nxf5=&M_)gM03+DNZ=agA_
zRP8LQo<$r!_vw<=t%r94vMxR`wUm;comClSxufCn!ujos_pr3Zw>?OGxghF^VczN;
zaTY!@_YZ!(Z1eC;ZR_b}o8lkZ?DUblxW762^a3uCmagLxckF%E+&y))^#E(t4zpEj
z+l$#|$Zu~8<5+Yp-YDRZrN_Hz6$iCjqddaOmGik4ws=jPsF<|n;mQ8c`El_QD&G(M
z3fMFEs%%e#@Do;-mPfVw%tYRAo1bZU?Crv+{(^INc`TEb8Z5}uy|eY;u8juwfBic4
z>{{&l?H9Zb{%d&CG-291mW5Ut$F8cp+gtVe+dCPyl?smUK1~S?&$6p}f9PbN);-%v
z>f1thSfnlW&hvU|lXmg3Tln1M?qvziitdX!JEn2FvQ&Oh<h)nU5wx8n)8G7gj_An_
z{qJRl0&ZXS8#m0Gdj0n6Z6cX(zI8pCzd@T_y6(fCGs~r3H1=GOHtn)8-BEd?UQAqg
zdHin?-kJNWd^gQ4dnhli|4}{2_>$Pt{InQmi%Qwn|8qF{7d+7X_v5g}B#uv4&VK$|
zzcXxif@rTyzxiKLXG3rKMqjp;rgi~#Z#&0__6bXzC4T>x?`e#;t};lTT>pV}|C1xX
zn_kwwd{A8A&1+@Cw<W`BR_CHX9WO^s12Nww@dUA+#)t^t^H~9lb30rTA2l!bIIxgq
z+o2V+neQL$=*;L|F-fF7e#WcG9i8$org9Z0?v+}!<-P*vq>SU8qF$xhOdbwJueAd%
ze%$(s=Xug+t*TzltpWaP{qA2+P1ZTCeXcyauj<@~<1d?iW7(Uod^{}XtbKXgtmp}A
z1NpR$*x20UsZEmWd(mpYUtiQtMC;1-<h%tBO?FLb|Icq3${EZVQ2+B~ZrZgCH>OOV
zFv%@hp<#RVs(TCA3j)f|{d3kZ$gt!;|MS7J{muGcSHJt>kgRVRccMARbj7kd(Y+N<
zf0YQ=PE~ksyXO9!<F2WO3%O-XzX#msxziBnv&XylL&e*kGd;2S&n~#?#%%Rx*6ws$
zw{!JmwZnTu-~RRw4!_^!x?{$h?=IUX_H%PC_0=hH_%d<ghnWUGwF;|GT4evtapru>
z#OcR2;gPuNCy$%ePaSVo&-?M}g!ucbS9WM+W$5iU@lUSZo04AZebz3D^Tfku<=#2(
z`_``r+`9bm9Y$8!Cr%rK+*UO|+w1zZ#BcUOxA4^-tL9(-v8^rN#=KyCRm`+@z2fkl
zS}WhiJk?`MVUc`UzH|H3-7&MfUv_=E?xl4~G^M)wTGV$pF3-!;_2$-;J10ng_6zi`
z+#Gu`;LfD^D=$V)PF|K4zb|q5$qmQ2&b?jWG4t?Kt%LxR#PZ#%R^NGdOKVoi8(TGV
zbJ<hY{&!zm{!Z*r2rM_@=VV>AzWPnwg1=9;f0)+st;u#`hw=K0PnVhhZEoXBs667t
z<S6xKhw&|gpWpqCsx|HTReq&UyvT#C%lobP`c9*|X`i{*_+MhV@5RcxXUVq59<MZ_
z8mEYK9a`usp}BtT<2`fIsx8x(Uu50Ga`eRCANJX0GOSHZs(riO^-s(%+&wGv#qp^N
zqm!37xb}#?ue(vausUH{)@1(~!Tif^xyk#rpPu9+xc{M2+lPn$6s_;(l~29>c;&h`
zH{=8#e!m`GwQ9AE-1_ZWNiX&@F6KC{g{6&ZOw?e$5y}u*c0$5~1c8v0D3zk-#71Uk
zW(JAh%&V~(jx+#AYNN~0!W^aNvNSY1Bsg4bpQ6h2i!(BXlRw5)y}ueYed=<f(CoJc
zO-~mxx^pS@^t*9P5PHY!sh|-$k8Sc~hl&m^Ar6nCK=-zGn^&u^-CDJwrq^%IxdUg4
zu35awxSI7mX3rYa$v)R#%|Ex_^7We9bGM#bSFhjuz5JY*LStixgTjpYV$$=!ylx3d
zEz$~Jv1qz3zkld+(d^Kw|A{u@hxT%+u(&J8Xg>JB&?i*H&Ggfvv%p2jV)HC}L8VLk
zAGzi$bu9Swysq*~sm4^DFK;hAp5$70RPf|&4ygtX0m)BnPD<Sq#831nJW#)P?U{k~
zRZq*TTPqCTMac<@e(kn!+wQ&o<Z1Ju(~B1^I{fF8t(I9+)cR9;2|jI1yh|4qulOn$
zQp7U#(2SZW-mNVQSI&J^Gu3Q*QqfA2uVSliEVw4U!p1gujqx5~$E8oRd8Tw;6c5-J
zy8ha#P_|OOP@m^3zD_ZW_f=g|KgI2+@s^1}S(~nIarA9%51!(<cG;(1x4W-ScqA9b
zu`E8{edcQ-&%|5u4s9O)_kQ@%tC{)m)RG<i!FIn44n<@wDU!6Ey4Y-$NtE-Ee!=Ai
zznrFueeB&np&-Rls5i%9iU-#wPES`mH!cC$P2MlOr{{YJuCy;~^iVDG6i}1k)ZqIO
zx5N7dFArx)2a`+B3;93$mwIbi?>ND?LUKywo(HQgam&V4+C6WGVx7`%|FAVP`l}W%
zWB%e#jPuqATzv6$*({x+XA`;K&K7uD#=H7>VTWBck7!Dd@5G3eMo$Ezg14Co`=7Ue
zdC&LVTb8fx20MhO9^cMf?Ja%ry6vHP2dm?!ea>7`!uGU2l<BfOcS20NK-v|pbyIkr
zTv&S4)hfL(Kihlp;>H!LH#0N5sXlX~ZBeiFHo>jG*G`K%^Txn>*4}ajFOQ#dyu|N3
zWAU4jIO%Iu;#bk1sd^&k!c1no{vH~ACXIXJCXI_L4uy8x7U;fg*}LLT&uP9n=P${B
zDy}H}qV&(y?#Is{jh4;sikCdq4kxh2tG1n<_3zrdga5A{(7$lQBeQe1aH(_g_MD70
zB^+h^Q~OSCjgg93TOnuu>tN=iHNU^zh~H4Hy}vK!lVi@5XAQ>|D@J|0wZ?pN)06Gf
z^3MD892K+iuHtLuSBn?o{LreD>V9n*QxpHEt|#i>%+kNlIjFXdKlb<E3OAXDvnFaM
z)y@29r#O9W$6lka$Mx6yeVJ5nPAc&HJHuQqwbpD#ZvPG8kyQa~6FFzRonO54)5@3k
z&m~{<v3;?1Qb4C2gWTi7q=tL9790#;Y4&W&#PaV;c2D^c^i?a^H1n^Y&i`rrKQ}I_
zG2gY9X^ncum3zgE6$RgzHhh_&vumMD&m;rZ+RDd|rfRouUm{TR^FQm9d+cwvbci09
zke3&^y}+ulYGISBuIp^ZX5-n7dpnj~bWhOsJ*TqpPw=}HF9N1s4eR;*a7)DxjTMEH
zjhQugx3&Gf{^q&e=6UDqKghG$C3~*p&RxYe?O74u?49y^o;RGm)nR`7^;@$^yBjw>
z*zDf2e!}0${8o8o2Ric(UQk&xSLD|zi<J#}b|0cl<Tsd1;(yjGuTx#M{Qseye2;{u
ztdE$n_2Pz|@0QHV5HOGWB~{nH)2>uZbn06@FT2&d*yG!aAJjy6togG$X{nV-)Jp5;
zYm`p?I`}g1jgrM(W0xJb@4jLE<85}iqT$U6CT+W;OQ!1S-BbB6aqD&MCuIlxMFKva
zE3XmI=ZrpZ`o~Ux=Y<mIs{XK?>X?z{YUiPM%j-|u&9GI+inni{yU_FQim$7KBkXt7
z<V=zH*0W{q5q7_~rwS(@-^Q4#r^c#dexh$S%Y>bZPn4c*j54YEf6#8y?w&Hn*_{_8
zjf%w|n%pWC`%>e0Zp-4z-=`OI>gsOWE#EBsqqCl=d-jx1Ywae*O6Po^{O0J@TlszU
z%JaGRKY!8{H~0Bt(UO#xc89oHlPpgs|L=C5r|+QuZXc^o?7mAUvVD?%{)+JSmgk(l
zu%G??{oGB=*|veP(cApnFUE@I%F4FiX%V^8v|i3hP}}<Rhj$_sGi+7YKb%s~yC`B$
zeR*kAYGgpDua?G<pD7!6zy4I=Uq0cFC{Od1qo4053H0uI?VS*CXj1w1o`zlL=3RR{
zFGxK&mGRo&Bc;z<H*)@2|NOe^x)V{ecQroQ{#jLtS>0jjo;6Ebx|3Mn&Ivp*bLY`I
ztxMGnV)f_Wale{!G%%tu^yj|`MUQg#weq*9yFNP{_TWI(FGuT#hhG^t@bQ05)|Xpx
zRZit$^dIeua_x6(1oCq>$Y-iNwE3|-?fS%rg|}-oHD>mia<=W+aPsx@JGn=v{5)S7
zmz0%0N1OZpmcF>D-bSZRrN|weW_Ik`TID3RJc)CUzn`18|H!1;`VICwTaL<vs5H&s
zx__HXBXruc$E>e!FI+HX%@3)H{G`bb{BDOI*{b}!kN5H4={@hQPCBnl@MTi_TK?#|
zMa-rf*Zn7c|GW3=o;6D*cW1YrX<X)KXsB(_JMj;XN80AiYH8NFW$kCL1>|q$-{z&F
z_dT(?dGWct_4zBNpLzK0+}t|nkMgs#>Xu`xvWXgO#%K>D2qYLLBuwy1$PmaVU|<k*
zV(Q0cIMRR^sqFz16LXY7nAQmPh~S&s>=aeh6_kp2lKSQdi&Qh7mOkC1+}9<_ezYmc
zQ=r9zvn_A2hqsyI%OICo91^a9O{r`fdyK!mn=N@-<g3y3#FBL}vF_%d_w8ohXMXhe
zz2^QW8_wUnnf>>5eD+^$rBk<N33a;o)pb8xJIgxxiqm`5TLq;@c6>~>yV%AW_3n%y
z8<*aRK4A;52d+%nsv1X~?6Yky`_)`Mw@X{doPSE6M8;mvM^1r@KG;pWu3g)>QQBqQ
zucbG4Mv9%g&AV`kQsDeL=Ggk$&(D?9ShAR#p0TpIx@0dBKGy&B!widH>6r^Iv$VTt
zAFr<Pk;tu_yXa5KnxcJGwJXf@v_ck#xU*<I?+>l&57oPLY+`6{(GD{wNdb{Xe76@|
z$qUfq{S<u6h4tm4DF>}`eNwJ;Y&^A8ah68voG|fKD(`nK*q~4v60pEXP07Gxc@H0F
z(7DKGyF*?s?d=HDyRwGAtAi`3PjYdH8!M}(^P<Jd%XjT|KhDyB^;Oed^9kvT?i{X?
z>nhs*M*8vBmy4Dri=CToSv~Rnn~fJgUOdGcHP^^MBWf*Yx72U@cR^3m6-+y%e0mqT
z1;4vektJk0bCy%BN3D<Va+{TsSJbcgoa&zCeRjE6z`ZH_TW3T${S5NfwD)pdxspL^
z0@orZ4c@I2xt1PU>7g~NYw4qvZ<D-Ie!jVQB}2O?)%Wd!I?>h+z39m3qtT@u&%5e3
zf1AcL{fXz=C4ZNCFZsP>{Y$N1*^wK1dwu7e<=-ZMqwRLx#q6r}rF;EuXI7uCUViiY
zw)nL8i|JwW)!2_YzW%T*-|7D8zU*BSZ2i29U9WNJsU5cxaAxS3@%Ye0uS!?0TgP<I
zI4uq3;#hp!erHE)?Z(_+DqkC0ZvEQTrTlbdwV8O)&t+L5=i=<eOj_gpAAhVkWTo?s
zA@WY|di(6Wo2QNIq$j)zUl$rRv0bZj4^z&QWHr^6bzBB>e@F-L?tk;>&FKk}cIJ*#
zBUzjm@c&_8KAN9&Ao^d<C*C{O^O?8b`pfUHbxZy3ap9foR;~(<m0hQG(4Q+<&gSRL
z<C7mwtSNiolXLi?U;yjqCHmUS);{_EnEg}htd%pKeQ^3A^Ji&7aNEkQdu-+ArwOI3
zmG<4R=IX}H!dwUY0=fDXy!iBz=Ou~vRX*5kWoOdR9&9piqw}Ig`p*_Q?0bDZZP%JT
zYxb0^QUbID?zg7y2F6>YYip0wW8dy!eezr@v(oo`x7`-SE`R-B%8$4}O3@A8?f
z>@71ZpWHdN#cg5z;ujn@4$5p0<yK_rT%<B(&(Q=ApXZN)m7nk0d_(l$i^t-d7TA9|
zzjuMZJKNrS2h!8>?ym}7!JoHorNyS-Pn><8>l}{TypiW}#RKDV3AV>l4%1HGEYEd$
zz$W)V?HAvjBLey-o*XwbV`0nV;pQ-CH8q_n)mOOV?619B^L&)fHt8NbvL>r^%ZwS$
z9tQD04xaR>UcN{zP9WphA9jlmQM+u?V&{r?v-8evm|F4SGk?N?o0Fd|Q>$Y=Td{9j
zT9)|AzuVY1-(X}c{;Kk8&nJ%hbq}5VM61`RCeH1A5IFT3$G*}J8&7U*oMXO4cB^qb
zw;{`vNozH7yA_M8rOfLu?}?}v<2Pie?rS#=dNx1Qa>X*culHt6uGxI6{CkpEb&Uaw
zb<$IBehJpaefzZ@-%V_gS?<w$eA6r~sZ{n;{GqdMUjKO`FI6o6{?-PAw`(OA`6x{&
zv16PWF+=N*n!VT^o72qQ%^P~8jhsux_kRt#cx+X5hJEyRhL@`=Kf6?L2N}QnI_ce>
zs4OiLsV|myu05G@c4OC9cA;;fX4AG5eCu@2cf8-~o_OnG@3MGDqZ3yopL1N{ywGI)
zc!r8ZVz0~__5U5Q*M7g}2rV>dO7n1%U*oE(qr7a@`NhfCf3$tNe0;UY?tnn0cB8-|
ztK%`bw>PDQPSW3fWdDX7(|n(^CTW{GH7bq!qhC}^*|Pn>jL-?kMa50G)tuk9*iing
zOLO6TUheMm*%fY{Ti@D6oixr3dA#fv??Y+fF#kVgLSG(9ELHD$TfxM)$F|WqOYGXQ
zP94wr{7M@GUZ==$UgG=gk;oRCd}F2eg~e+-W!}B`6|$)5a733Y%j{jxS4_11A@ef3
zQifsmQwdMTgSwrzam5VBw%@mAn9ZNTZ)Ii^^NGu@a_Z*^69b(#KTuomXR#oR^@!S$
zq8{Z?S%Devk0<55Jihrb7t<`$KhJ)&Oh{;5mF0X?tna6)rq)^Os?I~tPsK!V1n;*l
ze*f{2Rjgq6+x-l_38k+C!mQSx^0f(CcjB~`=AE;-R;7IkoOf=n&)?=|q9*pCbm>*=
z(><Hcd^ywew(G{DAGN=`PHS7pm;~)(tMq*`!zN2<SK%X;T^}F!pUCFrzZP5jL{YJO
zqw(=GUWd*deBKwcuf?lkg4g>yclI@Hf3FZBYkpWRXvv)elJA-~|2XTczIpY9kEeWI
zD^Jmw!5gNO%(A|!_4|Y08=mh?^}eZeW3So`(|vBgtnR(xI{whk?%(v(NqtY&1bzNm
z{?fy16UVhlH8pA_7BlxQm{KS9L2{eauB@;hX6G38d;Fg=Gd8|t^LMkv_y4Rt;*U<2
zm6BTVJ9j#NxQp-eg17PSZkPAAMZR66lp%arbcSmF@!WS?YQ<a6II$O&85D-Rc%O3Y
zniG43@cgEBMbp)CZ9NC~bS=2nxc208(|rzSb3St~$$z4odV+^(;~(~dVii-N(^fO3
zEFPqM6LY(Grt;6mqu03ahCWH#u4^8kyZiqp)yopkryjC@dfbH9k#V6kGc#YCtYz7O
zD=u7DZn<2xd(g5qEs%Lf6SqN9;`iDAijw-y@wRTdaogj?$<n14lCNciEm`Mi+8C2u
z7a6<v(dGlEHarlxD|LST--f2TxAA+Pw20i2(|3E;5LRGt*v9xrbilheDO|0$m)y(k
zbdL1=v+YgKn<Ytq6k9BtAKR<2y0-o_k0`pc%wkL9-6!2W7x#4OtvYvS;>2kSuYazv
zlhYCmF-iWBts!w<j(uj=E$R1%7B2Aa_9_d1*Ld&7;h7)L+;yB3dr(F7_{6f?dl&v%
zmFgUnZFtH2`vXs>#Kfj0lP!6CTUhcdX6^h~cO`#f?bhwP7IU@VXDztlxp4BA+1pv)
zpXf3z$>HkqeOQ=vU2%?<%%qSd6aGi8Q54y$)@vwXmRj(u#lo3MdhH~s)%!LYJiVQF
zdGSjR+f#jyl*^<xX^9@XsUwxOE6#gT*;AG28&fS=TX$Eis83csxhWtvYt361_M6Rf
zrx_*kziScMx~%t^+U9dtoL`pO*8k0T_UpEy?U6iwo8GnU>*fjXYW-bzdd?))H!DQ%
z{oA;Qam^AR?s?y;fA9SH-s4;h`^DzV)A_=W`Bcw;W}Eq8|NZwTcipnTk?y(kkNwg+
zKJNwmd9UZ5_5Z)$wEW-tCA()-Jn3~_{6c95*XM(wf0l98O>*XVmY{L%p~jo0_J;{F
zJYH|OCO!%GjF3N*cs=Rohb1Sq_geJkt;~|Lu1`GtQ{mo!zhC#?YwSDq@7;~kcWI~J
z&f<R4tRI^F`p4n!?mg|AZ#v%u%y5||7tK<(<Un&6-<sdWHhliGEF`QS{(GC8Amj7v
zqoqaBjdc!ol7gv<$Ne~R<R<Nly>~)(t<ApUXXfSiT$%Q<y7?kkJ!__n)3%>RFQsog
zD2YAts4NhayHHvsa!Pkb<57*a-tej3{9B*Cu{%{Bnlwp!!yS!{_uUlqESYlex=v}F
zEOvf6_sOCKZuR}k3t|^nuUhg&KXem|-Li*==kKdOnH;{nGDE2BxdDIf0dLN=%g$^(
z!l!S3SU~B`nOknh{e<?&D7}`cN|81H#+g?8=aJL(;^mpk9`{F1{$BTB+9g9%-Cg>T
z1(km}+JtJKx3r)661^ZObRN%Ron7nhuRNT))c4I)Zo3;Z-(QS*|K(cl&;1j_{@;w{
z;aqNI9pBXBarPyXZhq{>`~Q`Dr{4<Ne>7;)$;tn{mf5fD|7l`;X9~CQfwtSzrtGqG
z6im1k6P&E|dRks*+Kjw5QTKzdSG;%>{GsyJpAelGh1j3(@A+?B(RNsmchbYDGq-n(
zu3g40D{H^**kA1(H*!)m(zOH1MEUJXEKl8?H7nD@?_tPl713Q&eU&3^Dt=k$%(}f}
zSA}$a)zN=j&b<&fbp3TI;^w=zrBU{+THlm&Y#Bd)H7H@<e!OAR;~NoEB+n+xD|rWb
z7j543iF?PPLx1j?nr<?Ux4Gl@;meL+uS7dTs%LSEe0>m9zcjAoiB--{Jux>1efzB^
zMYgwdTdX)48Kw{&@FVk~=FzF@E0~U0ZZ37ps}7UXdS&>%Jumz67mZZy=@Vy8nKe&S
z`)b|^&a}fHJzD0nH9Gq^Tg{Kl_PTvi&ew8Pu(sVdhmCdI-qu_0+%s{$_N=_^l5Lw*
z;LEtY_?;?Gen@U<x8<(wPGHENFRv}bbep;5{ywpl_x>D<>}y`Svf_wdveNq7EA_94
zpSW_VSHg4FQkRCiI@(j7KCyRNsv~Ld@oh)meW7Q{pAT-DQn*^AI65%S_v5=eF?r>h
zT3LYyW+ZLcSJK}+(ZGFDd_#{Kt6+bFZ~B3<`o)hu6AYMLw3}*d4bDU_6kvL?K&Dds
zdslRWUBvmTm*P|7@6@+tugl5Zb!eJ-+N)JR7xd`f@^CH4{nA*wY|}LT>-DA_-;Z8i
z_VCe8j_D3Y$7>kR?XO|>k~5C{^zg;ugWM(QpNw|os+HKsGuGbni?8UjlR0)%+9S?h
z?c`71K#RpXT<UQV8vV^FC-V<4DYTeb;hmCn@B4#eb}y=W+*d9U$ZILsd!)ymv;Y0&
z2h+?sqn}RJ`0uoeFPH7U#qs(EkGnj3XV2t2C?@kH_m$|EJs&zY_+Q+r#mCn)TXJ{f
z`dMZ?-{tLWg11eaa+ZCE<Evi=zrNj_R$Q*TYZs?Qfz21$FWK7`JnmY1AWM9QS)=IA
zqSOK*$Be!|#&?b1#GSTX_xIJK?OoNz`iXMU{vT)V+LU>E-!A8;>Z>O4uRHg&&otiM
z@{RBlQMq{=+j=Tyo6P>=_{+t4j-s=^_i2}vl3iylGxW>7xUOYv_#gG8Yvql~o9urp
zi|ag{_dbeniDfvxq-vG2(v^4m(<Vlip4wht?U-V6<4~%^g)FhT&5RG2{~mF9-+t=q
zr874|=C3`u>5A;-3t?iry*;nKR%QIS`1M+Q?E_(z+}6n_Pp~eYK6`tZsY%$J48GMI
zw@=QsxV3F_Q*QppIhMIU&oGGgA1<7uV#)eaSY&qV$uGMz*G4D)y|-dol+@MTk!NS6
z`P_K@z3Y*?fBf!@S?O+PBl`;5=4UQ@*Iz#QW#V<ExbXX~{AXgWg=jwZJQA3>#^92U
z&^j(>9U<+Aex@7$6}Y<Bt@ZJD^_V~JOtN%L!_$I*qH8)=-OsyC%sG;ja_7iuy^98`
zPX`Lk&-YoM<o&s)`g7$Brv08?Mdi<8lXka?OCEGNB&+^8N6MdB^h82KN_k@4J>^B~
z?|hxaA!?N6d01n;Li*C>0)?i9M+#o(W^R))X_pb%x8`?av`NJS)g_I*lXJX%eCM87
z^<sX{q+MxC?w_Ci^2d$Xog9+IpMQuu2pnJI;(N>Mt7%Z!>)Sg00^T38CKnvlS+(Vi
z?faOkDSx)Fej##e-H9_k#wkTduE*|>zf%5N%9K6xCdd4|c>T){?wYFD7Dl^NJ!;sx
zabeu%n>{X5CLfXB{HnW8!zWPmVPfFMSyy#m>C4aCd3)1z=H3O<M4rDnv)gAvhD)!L
z{-v3%nn?%w!vuHjc-QgqXy2N4kyorMRrU)=7K=PRsWI>He^uW@^Hl!Mdm6H>VMeXi
zQQp`sDyp0>gGCE}wrq0J&+Knfe4N|inccO;M26!{N|CN}&Gxtusi3fwlnpao4yk@V
zvvtpjUX9K(&u^G=-2UKPFJDtHK5>cczSaZZr+wr(=-y|lzvR4p`29x`?Q?|NO6GAY
z96533$)rg}+j2vfEPSyy+Oqoe^&UQ^Z1XFvi&pgO+?N)!Rukj2`?(=6-*MuU&0FJE
zZQblxw)f07g;$}|=T4nFH}%Y^)4F@o)35LO>~c{rxh`*;Vg2&h{D-IO@>f0m&lL6T
zF)Oxl9i+uIc$U#%Od=Uf2-))|^K^fkdHk+f42F5^7qA$PeARC-i7R(ZEzJ#(m%puz
z<g5@*y*FP#O=N}Z0Y|IF?^#$_ELooj1PW-I>jvp?L|4UU&W_8xtbF#a%;ev%YFqAU
zw#H2<n&SL~jnQ=v>&>q*<==0IexLiJtb9M`(WzfA{=HGV?@Q(^^_k|zpa1{=xBsy&
zzqF;$<Bt`S4{vYJzbRr`efe4V5y_1+{q<+;w$nGWHMRS2Q|z1Jw?Z|E4GhVda^AfO
ztu<w*JU+Z+tGi?0<i0@i;GcUBCOThS@!!zqExY|ci?)v?kD1Qj?@i}D7`!vQ|G!~H
z)Pvsqrsc_t-4bkaY|J>BZwTLrH#qY6@1G^Bg^v_=z7`fT{v7m~OZ;5)hJ=q&g<*+)
zSw&v+3LZAGc0Xg8H=R3vj`|$|X?-E_CtX3U8)tGJ;{Dt*`#I<K4(=W0!L7zeXCB&d
zWXg>%Dhn+<C1&n$);Q^V=A@|F^A6R>otvL=DhCG@x3WHxnK<G1?8!6cNz9%-Lt#pa
zn#wbd&p|@1+1vBGMfT~OJ0qOu)$4!ejIYSKx!NMn)YQ0wKYiLXZ?3()|0Em#Ig@Sd
zJFV5q>wbN3-1(XL^|O64`&5hGE&G~%Xj=~V?CI)fh1EKXTR1x_H?DlS^Yg#T!_h}9
zB{#m5_`YDn@eM0u9OoTMOD^89^M*~1-8YkM0?{qD&ALbD9Suv+PguR->W%IjQ8}{P
z7;ksv9x_Xg-oTNg{f%oIZ}*XH$?KD<H%RD+bsrAe5aQFaG+8o2Y+A49MyGq*_q_iR
zb9d))rJTS8%*q?R?v>not-P&w$C>$u#XsJ5_mlrAcTazY(hIRa$y3x{TwS2=9^SXl
zF74|K?_~2{TgjYy57|uyPtNT)Qm8I3_kTl~jp4sbEDu}ubUqYbZxiHe;}bJ?%e%lo
z7nhxsR-ezaRzdlMo|nhXHYewMi8UWg<#+FHD8Ke!;Qsz~(^do+9}TQZ-^_UA2kVaO
z8Cyhxk6QIQc`v^0EPs5Dz+!L17OkFRPHk6ahsa&Hl9u_c^u=w>Xx8B8fyP<FYfV*W
z*Dg67_HGqp{$WmwDO@%$BKIhS%PpQ@98_tU%ow|O+PlB(U*=BPJ7?w6i{2Z096FhH
zu-0|7YwURwqyL^i{O1~RJI5VYd))F&ytX9F_SvO5^;BAvmaNa##Z&ij%`V7RP2M(<
z*VSyP$K<FqmcK!PTRWU;OxVsk8c&nq+g<q};PA${U#9-y|LRt)UU)Lx^?KR#`N<wz
z!n``J%%0!7q-^#1ZT!C;ytvi;<>#`)&9zq+E?W_CY=vRQ%L#vD8rM}M_V+(muRr}=
zzdh;E<kI%*I~aH}E1zHfeqy__o}!$?be5xB@6OHfnY+GK^O7;!+#a_p0s8Y+U0ZcM
zsNBB9@T$ol7l!4B0t>EeFR|L=tW#NMCB$G7S-=?efmx4zbAMAZ%l13jmtU@ZA^T{h
zsQ&^%23sAEbRE9kM{_q_zMiPg7tXnFi~VHZ_60|FFKzN<uPnGCZSdOHZa4ckfqGth
zv3WCQ>3gwWe`t3n#c;RerKMX}eDzu!$e*^VaA#oRlu2CX!If9m1it!e#CAqzQQ_h^
z$>|<?wmZJD$#A+KUjAS*L$CAe4w*`y8+=;oYac#Sy#CKjV}WdM&(2kLE4HuQy?M4&
z!G5K#vrfsz=9d)RRqX!9$n)Ekw!gQ37w<V^joR+1DLr|r2_`*WjAm<(tv{Nx_MYcL
z3l%3J75CHstazCuSSt<Qu82N4CDXJg%75`x>kB`6Pp??o?RR*k_^bz)<qrvd5f9&Q
z@kj3K-^QEWFW(qVoENf7a`y40RU4ajPi?p>&r?xZaHLeCdiMEcHQqO_$Zb8B5whyw
zovYlXu2;UzX*5=k-j{9X6>n#-qrTx2kHn)@FXl$6wJCPn9<7d++I#nS(k`<_?ShTV
z7D*bPU-w_P=E-FH(8Bxw%b!(i{HtR+qUg_4DV%?JPIt@p9gaF)r#()u)6IJK<!v^*
z{K4;?dyjriGB8v!jh!+tJodt^m#;q*_WrVtn##Yu{`})>pMM&>=h*0fO{eR;=Sh*b
zp=MXYtJjzXeEM^)clYDwCo<Pcm)u)ac;!&x2Sp39mYP#P{@Cy}ytaGzd-Y9TmYPR5
z?Uk7poaX1+Rm`+QG_QGOfRp}fH6^3KACuG%YOmibvb$+knfx||1on@A_1ZJ#zxF;@
z%)jX3;*t~o42O4AzWH&ra?K|5OCLDDy<$F;tF3jkRy9(0Bfr_I*Pl8|zD^U}Z?Njz
zUbjNsw|ZsiUY>gc(=Rj$h?{SEUNisKgCP0$P5)W<)$^us^v}6iyk_chCXO86n+;-t
z8c+5`?`MntXx+5B?~zXLMwhvs#%x>XP1!BE`*_m2txeYxukDrAo^HG=$Zh_5R(;Q<
zlhRkFvp!!ac>g(%2ydUb!y3)UOLd%A+D$i`CK-CCyh0*->B*>Uw(3NQdpQ@4eqIUK
zw7}z}$TX`Nzl?jLjv8dea`v8BSZkjzGh2Iw`OH-^U*0c}FrTS==V$xGc~4ulEmm-^
zJGkjxe9JGkqtUi^-!Zy#-&M}2zZ_IP<==wloo^KQ)xYM<>6$mA@z6ZGT~A&o794j_
zH)PdKzvQrwLGShd?xVh6-<|bZ>a4Q%U!m`wd^Y`keDm$2O%J4p8Z^ELC>7bluvU4Y
z@p`)#{qE;uD%)$yA0IR@d2}+<UoE`r?$QJ6%p}<#HDCCA-ZR~GviJJ^RnmS9j*OR>
zLqB9EX-Vx5<Cb7vT-YG4_pmk8D_^<t>9eWd{k|>9WnJAarx>OaWmx}s-al{ibN{~e
zyB|2N|JZlZ-b)#u-U<A(oS&=5(IeP@p6gFA=j!7rTPJ^2<(2%Uk=!D*M(S3VbV$!U
z#i?b@r<09BZ#Q!KD!%gmmbPSh>au57M9o;!j0`SmT#Np$sQvZe&iCdOm6qEdv_{vI
zwbv_Ml;3Wi7sfYh$|}a687%wk4ANO;>Q=wtZW3>IJ8iE0c+1a)D_>n&k>aJfHAM48
zS#)Z$?MvMf@#jxOmbiR>US4~feKMbg1)JNC<4scL!KXR11@{JBV4VIS*6;jD<|&&G
z2E;g~PdQn{?cuq0)2@{(>+Csx`aNnklDGaV*S#x&_2HG9Aquvk`dyvdjs(04dKYDL
zXR8IPsHs~r>)i$AQmIS&n%78ph)Ib^K3dQ+!=qBkH2tTjMAb4uy<Z%k{_|GOYq$Et
z*zs#-{_+DW<L_A9x$5?O*PfGdrxKqnYqsOs{;>LO+IPiOhYo96w)?MJQ?PjYT&V|1
zea6Rfloa~!Hbk$M*!*=}(CN=tW?EfPJDt_L@Pda4*IAd~X)=-3pDXI)D=p{l57`$y
zTi=iOuB`m|DFI7g_NaHSE1zb5WrFCcU2hJRL^{=kKV19%-ri4nvKnoHb2{=PB|US^
z_q?{e#piA=v(VpER_evVb33Q4%G}GdzyDC){B^efg*&GjwF$|d2ueSD%j>E1YOUFK
z&YpAI^)E$Urn2pi#iif3c1@YmQ&PapXQQxk;Y{c9g2e|{BzUy0Ub1fCqGQR7e+#yr
z&CcIY=^V@M^XyjNg1%q>y05Qve)i(e+3E`skv6fjR)s&ETpcyJzwRby`b~?JxTc58
zE9W~|%in9-&9V9JyRDO+YAuskpyAlqsCo4wTkgy)JvyaETN&KAUgh}gY}NkTn7FW_
zx%a{0^RGLKOoKPfEUb_?mbjK_LelOJe9P~a_GT3rmp=7!b3cCk)#FD=I)zi*?9RB)
z&7Ctr!mD=<lfBuM_T?)>UzLXJ>(UId;Njc#eZkW8>*McxTUx9a57@eU>EC&ikGlKG
zSJpnNzVq~6{4Jl-Yj4(0iAu}LTz4b<qHBDU{)2aBn>Y8>*lyI!%+s0`D&~;bdNL-H
z^>@tK_<0L$=55*&<GRevD_zHXR_AG<(k>>Wg@*5rg#UgJ9A*%B<IpkzUe$X%6E$L0
zR-Vg${o+Gu!-N|TrS~70_w={Blk&lhTjKJrc$+3H)L>&>$SUK1e|d5L#@adOZr+s-
z%yaB*U)|cY+VxA`-~2~Ox*w-(`{L*_HCFY*EFaUdU$Gj8oZIGwRs22TcrX0<kAnMq
zm8N9QG_t+$<UykArHGI9%O_|2nILEW>J*pPiEw@u*|+QRGgmQ~{?^uyJRzVVyYIEO
zz0L9O^(QRLd4o4hE!-t>%<|5=ymfc|p1ZC(SlVBEL7Ay@-fPv~eKNZ?^x8I8^RlT;
zcUrhVMMX|hYkIEt6nAN-t6a$oPpvDz@OG|P{PDWI@7HDOU!U>hi*}lGv1n%En`>DN
z&NsvEs94y45#Mv-Ovb7eT(b`+FTZ)QPVT$eETfW-|D>uUxbI#~e);a*<?_Hj2M&(L
z&24k{*C!?T{he=fV$Qu=QTwfPp3k$Dd@##umEJ6_dfm6?E58`}W>xb{jhQe#(8v3+
z_jZFP1{Zwdq<6$cm!EMrHJ;g(v^ON&XT6hpVe_?!W$L2mCjI&(<vy4B%b}y++qOND
zNY=~ekji;&U)*bT=8eY5ODR$zU)VxE&e!?1)wBM&T;5;1egD1xAFB45zf8O{NcUxk
zB)d!N=@*OqCFZdHTykoG;R;46!!3fXn@uaWUduSO{KAj=Z|4fT0*!m()-SpEWT9AO
zO3=%Exjp49(R{AX@eftaCNG@yDZ;nZOp5E~=d08GqHp;uoK_UfE2|{)j=y@l{Tp@u
zQ;ScRw|aXY{GaUKXS{6IwgtydHO_wjFIs8pma^I()%w<JJK3l2=UqGf=NpA+r!58r
zX0ux4T!k$s+Z*rV>}^?mPWF}UBj%gIMTfrUaGyGNUv%qCq2*S$yk0I2^IqS(CgghB
za<MeE{Z$sP553Q?{qjukgSy31p*C41^B9F|M-)`6{-{24OD(rv62;!Qeb4__(N{jc
ziZv`bA+K`wU+SHo>2=4{?mhc``vwP--t2$8vl@~DZMLstIs4CxL%w-;qR<ov-H4(I
z5mS1c;$s;E>JK042))vjD}Q)nyY|8lRy+&$TmAgmYklQ+e!KFgd8Yekm!DeyInw%z
zV7z-w^q=I4nPr8a)juS96cm2;6u9Pm#A8|KA;*+weJ`%BXi>R;yZP*d`-NS0_e;0z
zZMV9%OTCHj;Q@<v4ae;cYqn1kmfpHx%ZU`NrK)0US5De8J6275NB^!lQLDqUO{SeV
z`1$f7lbOF381?Pz^3b~*y}Iw_cX`H)GXZLPL93@~Yi?bTQSoeT$^V9>|1Tu|U05w8
zWbpd#*$+zM`D%qfttP)!$XmNtdZoXe!!FJV9_90wF5m8dzj(S!-4h4x=cZaOKVN7$
z{FCWiT~K}7;+L+vE{-SH9I;hUOkHj=$4QtaJn{728-G_m`1LVI@8F{F+JbWirw;G5
zc)LS<d3?90-&vOQCl{Yu{#S3=>a)?!NBLmD7OU8`{swKUp7ewyYZqV8&7F3w@7w{N
zusfm&H~Ege+h#p$?bR=g7R>rZCP(Kk*r8%0cCKafi{G2Rhds?*-m1NOvK`a*M|00{
z%kTN8zf(GN!J$9$UKYtV=@a;+C$=R>tUkuOt$xkE+PN+Lb5q?Nd3zYAPF0)STJl0C
zidE{X&a7QC77K5&lXc}f$#H$zt+@%Y!T+y$oyyyK>;3;rx0`Fqn08OLSiLL4g5%zq
zPD|y*Tdd|zODy42t<Bn5>?zSH*DWE{YNEk*B4y6JiBhZYy=z=-*A)Do!GdYs_k$<)
zPf~oY61VQ5`rZ#M`ApFqm2>C&uKB25?`XmE?NU(k%ozS{qNyF1R`S$(yKi24@sdyU
zw_ALs6P5?_y?e0c+!f~V_jeTRD;`Qo@dun;&zO>Q>b&Bm?3eF;rgp@%U9~vE!tO7!
ze&Wl}O}Sf=w)?56s()K+r@yOW_1|js8Hd#bqwel`d)qPX*ttDhukbK#KX&n*iS?O<
zY3}U7?y<j^ji+miKR(w!zc%2gV!GIL<&-xTkB+eU-#UHLTJ`9IB}xnR-tJf0U*<mh
zGLPP2tvg9l4M&zwSvz&&)ytY(dpnF9^fHUruHVGo!l7#SzbQ=aPO0d-hd$ryRw&&v
zsl6$ya9n-+naruTF2$xkTX1F7qCUG%kH6QI|6YA<>*~$DQgc}(Qh)eIK49tDug^V2
zvX@J=<6hhTnI9&tY<R?Q@^PKJYh}FcWY^nY<8r6Io00RWQ)v0VWd<jdKYL6%$*Mg+
zW=Gc5#{BQMqIBdEHa(UTT|8BqiA%6msId9iitCf3W}ka!F!Qe81I_$*1>r2Y?{|e9
zQmkis;w<FH;o$$U_OY|Y`&SC?1)u#6bLM}zdHmt!BXg&2E!0!krf&WF<gvEZ^4m%@
zg>4*`1gC3yw(ecAaoNs2{|jQySD9Glm0kHfJ#}qVfrsUSJAPf8!h9y@Y?ak#JDz(+
zv2Bg^+7su6#a;6)C;t(vStZ>j!=~siA>X~@#@1u@W*g2=a@=xwYmA}9ZQIYbPsBtD
z-g$bvFW73a{H)cvXVZ<vR!`j+(Qq@zXy@$a2UC>S@=jP_#ab;n*DX0fa!X^%{m$hp
zLQc7cd^F&nW^DHBhv(^!!ppxL*9^A|ejjM0TGF||;_jN|N2VENtE!)tv0_xY`Im8p
zrSHPDbnam1+TG#XW+n@#vg8J^=q)g5P};oo+M%Uw=PevLz8=)bOE_L+Hsx>O`Zhzc
zRR`A{`VqQ5AkruJ>xX+UtqTP9+ve2mtX>?>vFpK^-)6zZQ+~!A7rLNSbMk{KyUzRe
zCeszuix=-olH6Gw|EKf6rQMoM*Ja}(-nH7ste>|1aDsl#BVD_6^=bchgodnPF5J{o
z`dXu~>2j2h^<?3gms6Zl4WwQi^w0nDGpqT*9_#t>4J@4QYi+g)osH>RWG-?ena$3E
zqj`6B=x^_>si};c3?&<4^K_?9p6MC9?Znh=*G`)!vhcjLp0t(6H-1k^8cWByuCu+p
zTwKwsDl#jMoI0DP*Rs_rn}18N>{?b=-;%`5{+GND2DU%!ma}>yo0li|be3H0XFK+3
z6PNjmFa0}<Il3c4MsSfP_p_kDiF4;y=gOMt|5YzLR-1Qwv0?U$BNnwzHFH!SiZdO3
zv(R%#)De#<DqQ=ve$%#?$ox;mVi9Z2xxOFb$6M9z2mk!}e6`=_o5ugk?EmE6XrKL7
zZFcLI6|1?7OlD>=R|xV>5jA29&0G;NYv#uGt}xkK<#qSdeZw=;o7ZmYlkxd{^o+No
z@m70P!}*(^zIG70oc>&Ea|&lkS;V(j8bu#xo(WdBzI|Kz;I~_Kxp()!IsbQu;Mr#G
zbK&#a7S|t7pPFN^G5+1@GpA3U)-KI>BXwa$uHe(jkM5r6a?1_pd1rmOaQV~Rgw@-^
zZ7+l_(bwd-@2j_CN~M@xxx~M<k}U`Pdw%{?jd?FrH{<00#;^L%T+S=ASKj;;6WHZi
zvi(iWXPJ`Tjlr}1PMr>kVs-eoP=#~#0m;Qjy{_GOtv9=2vb4kWTWgk17W(3O=TwQ8
z`fhERcg<nBoioi|>dcl1IxDfot0BdCwaS_^HxhEHE{jB2&XN6UAAOQ_2cz5E>_yS-
z|Jt6<KC??})v2V_MVB^AP2J<1y5q``jBE|{>G$rM>^ZdWYURR>zgC!D$=$T;PD;Zc
zy-FtGr#p|R-jb{_)4A~Vp-c2CW^3;9rN;tTkMe~x#!gh+E*6lQbu7ZodJeDe+O-8%
zHA$at8oS>wohTEqMJC`$)CaGq%t*t~6(#20FS9By#tCk<O3Q0Jl~^!s_L-+<w_fLD
zzgqMv`>L0I?56XFLjqM@wOk7gquQqV-4olHy>(K^E{Wa8Q_3P7ol{Rcm_IGmD$07<
z^h#@G|F35vbFO%_d+hmlQijDXzG=GthnqS(dS3kAB(3+S<Be|fWPy!oNApxT*w>%!
zcFY!=^pa<hV&&7-b{2n)X6&7JL-*!R<u0M4SN8w^=vz_p|Mr)&>w@q3t$Ocp?AX=B
z7Y=7+^2DxfdQiP%?&9qg9r5iJ^7j%JclOzo*99*0TFEl6_-=lzrD#ZCo>8P{_q~M^
zi{dJDq~_*#zkQ#3>pt6qW1jyHUTplVD|c_pcb@tEvQOo`O{U29{49LDZ`rk{Wz4P>
zkGx-RXP^47`0>8J&v%m_zBusnU-M#TrvGPaXDn?>x9Ll~QG7>O!qD+i>h+3*bvtiq
zGbFpYbv3&iA57kx?AQM_%3|iTx~i---yS@8C-rXY;h4-#S2yw=<llJC{g1!!f9@a4
zJCv~v*CXxuz%v(svbQ5UAz^~S27S|-9q;8IH(Rsw8BAbNVVJ1Lz5tsANGnuHoe!`u
zHbmKMy|#+IMELrycN~=_%Z@a#3(in9Ue<D<n*VIu1?w4n@Ait{oVGe9W$M~T_h#qp
zJbBFQ*s||(2RH>L1u${8OmgDpys5nHd}O`~^TT@uMen2EdqyR%pSjky{{Q~>@|*_J
z(mo0TJTV*crf&ZH?4owqI&RJ)?s*o8Z>;p?Jr_jpc=|{wPvE=Yi=zcR$2Kzcy*p;f
zvGbr=Muq;vd2wcQ{W9Jw#C+v?D4#e(VW)Zh8?%M_>{YxE|4Uo$x^8f~Q*f`{v5veM
z44aSJ9f`=1_-1`g^ToIAyDh8Qc2w(r{leAvrF)8>lzHNzDXfYMFBRC!-}h+u6jt+*
zOSU{$QXutw!<n<n3My7g2hJbVy>U75rbYYI$1-89E}zr?%DA<#SnsI*xr?>Rx~0N8
z=fkZx3Y{Og`U^gA@=t7W>ydfB;ox=sLyA@^HU}F2OTQC+@l@>L)0YaI-@b5_cYOI;
z6!B(HUGw2Z6CQZ9i`L1-9Tbe?OFnL8;Zjj)`NPev?VOaoUC<+AJ(WA@d(8K%CDkNc
zU)1q&+WX$vEmc?BHfm4P`;l|?*ZE_v$6Tx`+FPtEs$6&dWOeJQUj1>^gbxba9eV07
z-cUHhDBZ++)GcxGM#CEsIWlH)vpakbsU-(*Sb4)F$I6U(cIVkcc1NEjecteMhVru(
z-2-At^Ao!_9K9iwBe_i`x^eB1u*B>QTW?t9*luGiK4z9Qdt>g6vNNp3NADPvpJD!d
zWRF4q8S&4D|7>K?5#T=5v{67ug!}NJjT|~clM@0Xc+`$2ZM4u~^ghCu@SLTY<ul7?
z)_ShW>mS$i-eBt%&fSo~sCeR)Te-IAH$gTo?bhwN({9Bavp+2DGTCM0BHJSpN~g<K
zzkT<v&-U2&AA6PupZB$0UHxx&c)@yYd6%1CH*;?{{yOjW@8@>+YWKaCzWKf9Z^3Vc
z-@BKHS8kcV<NlMn)JKAks{e-9X#Cia*ig{$zTr9ZaXuaSA9K7Go+**2o%*lv!Y@a4
zyRKxrw0)_?vKsAG?e+Y9QbuJf&R+PF*xbI9f3<YkljQz~wr%U%ddr@PBs4|)-090`
zyt#J2r`H<46IoL>YFy}f%=&WG8Vf5wO@EK{iSEz!FP)Y9c<xnW<B@zT4~;_MjjDGd
z)?R*+wdD=t4@JJl7Zy_8*&<Opo4BI*on|^XU9Vw%vn<elhO(QG;Fd=70&71_<0G>Y
zJ~lhE{cd}G;qfO^<(Vg3?~BjpiuYf`JN@(@(ci`rJLYDvueH<lS^a$4queiNTyIZj
zn|eI#`m|}`&zGCOyRgC6wEeTxi^I!p)i>TdlJ+$Hl&Isj&wme@?+jRybM5g5Zdvwk
zGCIi~>io66w;gvqs(Dzh^Xtos8CQBDA8apPU441t<hSJy&iJ%VXa6U1RJTJc=23y-
z_Ks-HbHORc(~L6;*i)=OY?7%BvkO(w?(z<9oMFJmnJ#v4XGLIAUeX*L&54>$#ewtL
zE8?2CPQ0{;tXbqUKWfscLkGPDcZ>9U1};$zWN0xly6~{lW>)ucaTRfq=TF5AJ{IV`
zv1bcUdokr^Nnu<dV|MzzGu4X^GerA-o3NMpyiHk^+dK2k1v`J-JCi@bV18vx+0RGU
zuk8J}J+OY+=IDf}FLs4JJ6skfXs2~EMonpgw2G4Q!vpN!>(c_|f9$RJdF`dgi3R-&
zqpqwtYjxs)ozkaYkC$ai-_$B`-&1h;Od`)K?OW?kWu+vV#yC$f^S^yuH}UO`$3};L
zmQ`G1*n9hJe3{N(=B?!?nqxG1M4XGB1gp$?yen4f_yKJeDYe7Qo!U1uqi>#@-{bQ6
zc|}U#?`i8ZZ+!l)9&qN|g^s<UtDDP>8Uy40oilm4_ww>ps!hHt7<L75?GjFPm>hm)
zPL}P=Sl9RYe4ke(t-f3O`i`dc=IuvP<g$h4R2^)+#(g$9vB^R_MfFsJP2|}H3s>?P
zPi%;5HGALW{?Kqo=JiUA9rld8`su=393nl$nPXYGCwB@>Pz}~_2-4boR51VBy_yu0
z-mt|hGmMH{<qD)`afBay>94?@Te<Z5iJlD~1G@t^yym;}#p8r;o9)q^Cmm)k3OW&b
zX2*k6&wF2voveNsV)3`^_s2QhS@YGd#cJ|<Eji@1ID7SJ?^pZhT(=VcD%)>xh5c)_
z71J%JRgB+GrM_9fxu-=@pS3)m>ml1YzZmgAQ5hLQL75oQ!0_<U(BliF&FZa>*K?ok
zZ#t*XTCmcQBi7_mw_vW^zDpN2Y?yT6*`h;d?%23lTc=DtT^7AFW^GtNres&opV{V$
z2XD+YZaAr9z^PFD{e$t0A3hoPSl)L(v~oGznQ`xT^YYmo=h}E=7_V=b|Hb{#jH`Ch
zGcUX-&2GPZcz06EZQqE)FJ`{@e0ox=`^~-CMa#}DTys&$c&^dG$u)asFx`Gz`Fm>V
zgt(Z{df|_eS=SP${hq7PeV<cQ#KVMrcbN6%6~`w`^yoREuDsyIqkplyZe^N|=h~B`
zH3Lp2Np32QzTnrXz4!9+s&W?TYs+lzx^2xkfAwv;kLYe`{|mlvKU(|d#OVZh{+uxL
zaQb=6%dx_9&b_vMVRt91WpS_7;tOvWJ|B6gWTIztiSz98iwCX?#{X-mm?Lh`YkKHX
z!Nris2O4QgKGkzIk_2U<H#2TLx+YMBNwwnn&h2cgE3!ANca3!v4q=?j({#()VCm!1
zD@(F22ZwwL5j8$juH><7d&QOQE}I<N9o&ssrCr!$S5&<EBk|RB+vE0+?yvd-Cp47(
zUbvwrmW#bBa1%%0nLqb-bk|Mn&hWQ0nK~(|IDheizaB>gSJ)^$DLZiUQQFPA^<`?y
zl=RAf%H3Wc#V7xIft;_E?Ze5DeUF#in)~+l%soq{6(p~Dd+Wi<*X{w$_0gZ$kJRt{
ztiD;`pIY4Wow-}}n}6LoRDMOfY-+%B{jKqP%L*?uEn5_<Q#NOR=CdoT^Y2?;F8IiP
zPh{`&2U`u<o5Mbe=`}NdKTufBviwYPYnuKKpBOo2p&94mT{#UEtakB;Y`k-R(ITUO
zbBlV-{%oyS_BlPU=K5BrX}?25GIpk&Wn;XPAg7(jqsz<9J@uFI+PSwsRP?Hu&OdiG
zbxOvKFq4PxJ(JSq&g}lNfK%=P_mc&`<-X_N<2OIF&}~!e@5#5-w*7i};Qn6~CDUEY
zq!&!htLB(;cFoE83fFrXS$%av?yJ_FtrcdxJo&G?ad6_Y4w;*x%zIr5rWD#Hbyaqr
zXA*9o{b0)b&Adz*mp{IkCot>m-lMwrdZq1WZ!0ut?zq`#ez7!ZIgidmX?c;CA8gqh
zi}<s<lw?9H&+7bP>+)G)y+yUorX*wjI*uixhMruRNfx4!GO@woA#q0|?(*!mm8@mY
zGd#fmzUG_C60=hay0_VXGI{lH%kqGl`ERbb7q?j-+3@^KI>+s2QqNx>KH$re8Lg<F
z?~}bC@x3-bx6xF|Or4vu%w7}Lr35Y7?xw%+WZsP9eQUbj=B)B#WN>{aH<2gkaYV_C
z2~AfH>8#ZYJ}>fo>zp5bc4>DM&!%<n-8wH?BYuO`q27-@ANX!NTCcUPGn=OR{CxQ9
zl2<mnHXexx70v$Q|H1e6!AYXb%i`pYZZE%mzpizOEB`&;DV>vDxF396?WD(WPGq*-
z!)=036^!*<Gc}LDh*>T#$8maD#GEH5Tv)qGa*sP$->s9;xP4FRt=z^7DsEnaAw_o!
zt&6uFVG;JZbNt#igSY!G$S__$s>XCvv#>wY;rU@UGXee*qx>?<#vFl}LYsJUb2piF
z==MJUS7RYq8+7iEbA1L!ft_KK+v*D@CY(DJ%`I4c+g`7`yUK2N_f(NLA5SbUb|@3%
zp4$;tdrY-^)``7B>#R5L-?T%h>C_UhYKx!J2loHwOjS{QyyOl?p=VHY`j+lHcY75l
zuDN*DSAF4e-6)gmM<1m2BndxW`fp!ILgRGJS&_2R4_-OyR_tQDbx>}<{N9UG3QCG~
zgLfC|M!2@`X7ZJiUeIr38c-Dw8Fa2?%Qnr^FXrpT@4F{-T`2D4yR5u(8cR>|rz&v2
zxYFY@QO|o(68EI8;?tY7B)148J_;yxsIizlMN0CKqd(($t!xusSGQ?<G)|vix8|<%
zedhTtn>IwJ`)~YwNLRUC->T}*fp^ntt)8uUp|^!G_@>^9A3HhUz0z8DyyEljZ|1H7
z#kzfuR&}K1-YZ>pN+sx*6_ft0Y`Y6TH#LT~n2HA~S+NQKFFYAv>#-#J*DUwm-TN6&
zseL=kn;>Y@)^%@p%X|BpBEFit)d|~AUwCa)*_-`f_W5N60qfHqRz1=)U^O#V<^1<F
z`{hf$)wPczOl$O%<o9#Od)-shU)O9V*09@!?c3%8){noh28-)Iv0R@VaL78lz`8<w
zot^TD=A}*%?BU(#7hKxfBKzn1>vPsG_6gj3o6n`3rX3&cF}-~9%55v`XNTV}-WK{d
z>(R1zTYq`o%uzeD=(ca%_G`1l?*0hM)U;`Q8}Q-Mzjpid(g|)H6=$bu8AV0<c}!S7
zvDvlN+p=iiTB&<o*L~Z~O;=vo9Z__E?cL5DYMZQ1^rx3BoSOJ`_o?jOnh(!Bc^Gb4
zI7@dnb{pN??l}ADg=arwi@2YsmTgFx8qvAim(5DwY{Se7CZXD!=iVOqq5In0=!wA3
zq=IAZs_n9;*RRjx)8tP)*($W3MSL1dUwHWa|8H+i&Hp6fYj)w?q8+9#(TY0`B+uM@
z;%vYlW%0zTM;>3iXg|Gd!hVPMx=DU+3k=0%*FBH8^T6Sa&&*vmwn8E;hOWOp9g<oV
z6(uL6)XR6fMC-{H^ZD^LCno$o<DBzv-K3pw^iR)Lu2i;kfBBttqXyqUmgx@1-@JM3
zzW*|hiQuUX3<6%i*Dzbzn@zZ}_*{?A!bg)P2Va)B(%AR#!>j+n3-lB!Zd~Wpa}}*f
z^YTj)5`DtCNH=E1$wRGgERw_3IXAPsKHZmZ6=i+GI>t<Y?RFNK?@5B3;*Ya(tCsa$
zinpCU`)rKGt#xZJ-1wBluHGbYD}8C$<OAKY%g#+z4Z6tbDYlVa*?La5%86y)k8gHh
z-@Ui@x%iQzlYe|Y`1bYE5T7+l$BJYxKP)ZH3YDGz#A5!Fix=-Kxy~zQCVkTTj<0sz
zuZU>}lBc$Q@e#CK$jx_6c(=nteV@giE`0&Nd>Gzsc|Pa*GNtu@%S$4)HnADMSMZ+3
zWqm>{<KitX;qUoPd;ZkjY^|BMSL)}jjeJ`-tPoSHSFcg6@qg)9?2_o9>zf?t@!V46
z$lPPRHJdw^T%5CghJ$BI)7!&3zss&4d?z-WhnYP)VqK)O!_WP@9(GDfYVItVr)TYd
zelhcB_8rSUuX?d|cl5p|jxMjo9c~_qewDpiKDF(-X#4*!(k^ml^MAB&dMtYSo_E2S
z+n(0B<*|~7{y9##zwlsv<D;(s5e+8}zt?yCzuPEv<(=5dz|&{fJUv|&URNHz=$+NS
zXZi=e1#BypYm6&8T2a>(CG5O-XJhH2MMYb8b*v4^x?&iTezV>nhI8V?b%C#+-kEXk
zIP;_;lUF*s1X8AMypy<yO{V5xiRkn4H!j!yK0o#6n$Gn9)6-V|_ScG@5b}BPdu@G*
zgUS{ccihgn|8Kz~#y6iEYRcxYpWlA;LPLkXdhQLayDv}ec;8bp+kfFc`@R;l=eKX2
z4>4k7@sin<oV#XoS-8e#{oU)1hy<#1a(jJ@DLV9*UE^8w7X^s}ef28y9-3zv9o~L-
zhwK9<1?^l$cfMft0|&dro_U;o*QEB#Ve$QQLesV!{lxr5yDUOr%aU6*hcgc9>*?zV
z{F*4SYtcm;;n+5@#TzrYik=r9w=iMUYdZZP!$R_1N}Fg`pyscPUq3E({1Oy9RAcey
zwDtD~7g{xrH#w%b<uy&MxpIG_f^NOxo9`#Rzc9YC=HmQOBl_P?s;ph!rv4D~X33jL
z@(*+h6U)3}nzmn9rzgtpbn`>)gBs(jTz(;MlliX5thwR*{<e1E+O*ZN2cz^)bv@l7
zcG*?#RBN5?lV#t{Sp|frR_f}Mx2<5h<&pOD)JcT~K9y7Fn_ea7sNY{a)l+yTgLmm3
z_HP%e-$nkOF?G({b#3B4-+c8C#NOrJ@sIWA@|$gwyXW`(5Uf42V0wM}lK1)>Gz<6`
zs~z9$PJgrE-)%<!ZI$9bl-a_{3b+4OvXJDtmvxXeV=-r+VF9x*r`zx8(-S|+ZaV*3
z@z{#xC(`<6cdN{Lsc`Q=!|KQ)@oksC=PKNMq3|z!$62+ajvYN^&-3q}GyAoe-){E1
z7Y{@4iJejjVeqwNRp!`q|KhI0rS31YpQ|g+nVzUH$!%{}znG1!a?Ab2CL8%OlD#Fa
zh~HvAu(d#kb#9rn`VHxCsa&5L!`xFfOFlj6>=xYFcC%-fs^Y_vEI+wl4^$3GUip3J
zZtBjGWpUHLo;zmHcYmkst0U$qADVduo^#%QwQ~BT%$-h(;Y^`3r_>rfGMUx0N5ka&
zZ1<+x3AU4Gu!MMdtqV_iX7@vAQu@rr8=~z*>ikOsjbBfB^rQNRczR9l#lUNdo*z>c
z!&}X|?uUH5EOaQFDQ@<Qx`)A1H~A!6e=XkE-|l={JGc5Zk51H|zSPP$_1-cHyc1mv
zI6D~^-p*Ve`n{6x&gWBe*XM8bzj@-dI7i$+F;;%NJKsKdH~(GfE~>wMU(ELt-}x-X
z9;+3{cbu<?aaR`HR`k)RO~&N@dZUGxjHK_pKlb_J<7LLKZ%srRj?UA0sTD1(KGkdH
zQf`&U;epnb`;^b?GgvP55oUZVAD_D_?ZlLyrcXuBz7;8Yp;IIOvggUfNt|-#2a;nC
za>sqytbF>+Li;Xb$GAm2OO~vh*;*8zX|dPz>kEI*oJYDZkFM>NKE-CYZdc(ywy)Yk
zwf%O97sTJQSV;acJ2HEH)T~pX+QzlQk7iC#=(^n)FTDFUJB!}JX^j15`fWC^E|&$x
z`QO*r8GBlZ(>!vIMv|D+K?TJ#;-O77&3zZ2U281*#keJKL4Tz*UsZ5*eA%8o-c|(?
z*Te4!L_B}>Wb+Zz#>TrkQ>WF&Fa2v>E9*W%dEU1B|INI)uSB`6?DOMV?4<RGcj>V|
zY~E9!ue_PAW^umc9!vUFF2C46kCb8x15ao^(E7M}>GtK*jaSU)jE{U$`G}|V>5-q(
zAA*@@Z8V-dF}uO0UyngDRp(CP(&Wsn(jckFyZP307v;}(`OIlh`Jv~rpSjabo1+((
zXm2#l{F=G--Zs(1o&Ww_Ii$<IjQ#v+i#-z~tDPzX@8-;2?Z@h6CAh>ha?9Kf*38OC
zsoXd*=8|8>9{0?Dx#)bYRN4NoEM<?+9zMZy%P6m4X6<9e7j+t+1T2}$4%Iz+?D*C5
zXHR`|`nKq8Z-4DkWldkP?A5m28={XLanN1tdGfL3me5Df!x}G#Drim&)S475;&nSp
zzg)my*?PU|>)LMTER`JRh_!!KvqXQI<ydH)$GG+8_2hFU^Do;UF<|xDYWH*N66Q6N
z(lSg^Q+vX=w=7NFmiPPHYV8aUnIDg|zr9uXAn)E`HE;LTmHW!-j1|sY$Yhwj^yKBV
z^qXxBmZxQBZVB<e6}_ybGX8GUwrTGc_C0zRm;Ye(i9_yAepzw~Zd`(!f<>oj-YA%8
zUv#`zDYo!jYg5hbGi#(5G_=Vn&5QCsq^@*JexrQw$%N;P%n9{+#ZxR+TF<O)x)?5Y
z@9uZYB@cHoTGlwOUa)5G4C$cC=MUw)#oPM$o#ZCWtr0GJdiukX_=yD@zdy^p=p%a2
zy>|Yy#i4JD?3sn%GnTQ19hL5ST^x0Or_8LzUy4$lff8En+qmnKPEHQ^ar%4#gLR<t
zJ+^H}ei^K8E?eS$NxSKs=*h^Jjs>fy?7H~jjpIkg4VR8gscJqOty{Eb-Hhcs_*?H}
z6>qT6S9PDZz{4-h=*9DT>xVBp6=VZ8&6%>tsWxw2U!(p0qC>_N{>-OsX8A7-4J=sf
zc_F9e=J&*H8}BP@R^A>dlRVSPF_34{hiy&4+}%<SGj?r}T2XY9i*<r)P|dF2jV+2j
z?WxB;N?SO{S0v6YtPpT^c&Nu+r_0{Dw90LEIP<9ti(Mz=eRa2pRLm5tlj05bEpXlP
z_x<$4y%UTkt$)D3q5nt854J;wCr(#$Rzw_f5SGbua@MWL@Y>;FYkVSc$)PZ(3CrXI
z=5l=fr@8WH*4wMyv6q*wwlS9s<E*y%e@CO?u^i*SyZsMp4gG7XkM=GUJ@a;!ad=+P
zcJ=vEwl()&RsL(6VHLb{;ky%M0_SDSN<~aA=c=m;PY%gG9CKqqZr!PZt!u4Jou;}i
zUv>G)$tgPAQw)B)Y>yV}cWD-5-M!<SzUAxFW=m)D>|Xio!lwrv4J^|+?R8t3GQNL!
zn19Xk;@9KXA1<<=H~UThpN~)YKTN1OYU!*#FJ}q+(w!>DZ!cS>%PJ(LE4a6&@i6BE
z%koE(tggcOiq<XGee<{c<G3)@<mvWM@71-zp$=<P^g`!lSUTI^vCdZWj{8+|`VFtP
z<QB%CkBuY)TmxhjmF8C1=ol8xpUqdsYm|1cKC!C1vsxqY{Pr0Vw#pYK1kUff*;gg;
z(5mhA@AG>;Ff;MCZR@x%BE(<Uv`EM;oUiCxzz3;|9N(fBZb<d3JHhgOpYe)_wARDd
zG%Z_Lc8b@23b&1O@wvI}M&_<Z4-DJhu+}S_*tTx9c|(y<x0_(`0TzpASN7dH|9VCz
zi{<=q9>c3lrvqdDuAgGiWWTV_$lsXr)S25Y&k|O5t#aTmmFRU5)9qiWeMo6#$kG_q
z^b4QPX>6Hy`Ozn*1?(YrGZRuO&b{2+5iT-u+STc&8p0L1A6v&iG>nuy#k9+1cHoSI
zhdAD~ADT96m%b@;Wz7WXqrK1HA6Mnr$&)wZY2CzwuRN-{Qk%s}<5GC0E(+2}Idfb?
z`sM+)n7D+U(WTlNPv5t(c-dTWUd;RG=UdjjHX9mR4+|U-Hro1<wWQHbeY$$8{2^(}
zsjAi1KX&icda(MsaPW?mH&fz&X*J#X<8^OoQosVMj(0^*#9tNtm+)HWvvq~Vs>`yz
z4}}W1#!a2+cy&JG)i6=<=#VX&*Qlk>S;t`abIn8JBV|^LR?guGe|;r#*Y|a{|9b82
zTF-C2ZGE<)%Wry&gLu|vt_s2QBFTcV^k?;!C$~@gdguF(mzVbLTx$~VTVc)V<ErMk
z@@8O0-2GHd&ka#d{hQz1U1R3atZ5mrRNY88;2g)jfUcUYX3eWIS)cBaah_TcP%wRN
zR99%{&YU(IeGTn2EiZWkwrP)F);zqrKk~Hn)m_gg)}%7%ueJRp?0@3I(obwx(>}ec
z=9^vbth>V}-*)NdTRW`(tozcjazW?zkJ9fVQsdN3et$PVDlmJIkx!t`r^fh${SWw@
z53g(cJ#8VEkXB)TuEX<0*_9G16GJw6O;t;tz;V_w=|#`#l9xBDwC)FN3Qyj+Wc}o$
zlm1qoJk>EdHB+(Xe|)1)2AkLM#llTxJj-|Qs6XwgZJ<<^k(BMVQf985c7LPu0`oa7
zL4|9bmInlRPKpTMT6&=9?b(K-zZ5uTy!mX&yYdOw4;JO?7T#M+C+aL(ko@=E-S*Y1
z*k{eU-NhHhroC!|f}LuUS(a(%o)1f$TeYkW`?|dSe5M;~KR;91xBLlHu3Biwl;CxV
zCz#^GeLCH9uV*;_d>U@@>MHxq<sN%#O#^b{%?sXcGhL+-SCPBbGuZck(nJ0yHJy5E
z7jzh1F`f6n>*qINo9HkB_2of9kDRaatb8MN;L4U&p<my`$Nc@h+o|}<?H=Bxf{BGm
z(S<%+eCDl~usK6*m4o*~MVnr6z2;R0uf9h!U*W7Mo4Iw$52I+Yurp5iU%30s1mCmH
znRp~7HBXy+F6RrL{Rhr(pL5kb!l&a7&(izM%+HfH>HnU0=l}N@vz6V+hgYU-)w1oB
z?7L#aV)~#fW@?;&==O$9|EHa9Kh>ZgZ|PRbb?Ev7X%)rpX_udT{o)ZIb$Hg7n{7#}
ztd?0AMNN{}via`w6Rq1WFJCROW5VqA_!V}mG@l)fWbTQvpZB=vZtI=n1zVp+#`RCp
z5Kr-5ET@$5sOq6*<+Kic57r>ISmBcwLiUOOpB8zIY4^nd6(`=g@((m?*gcjDZK`<r
zKz8C%X{j^JJWbcE4Q8vKTlZnskFa05_t(#iFF&JGwm96nwYNB(S#0U~pHpV7pMS%w
zf9BTjQC|{&9oa6r|4p*AUizIKo@Y;ppJ(q*Jw4O#<LjofOY^dO_IpL%wEXZwr&O@y
zLj=b+o!@H5J<Q8I+^1~Z`TF1<Q5&wES55z)HsDyWj`>5rocoPg)!nme%h$#~ezeph
zJ*w$?ddsEc32)U}HhO%Cm)GGe4fr9hx%XW4_xbWxf@)WoZwY<pm%a1<Ub~;hN<W$9
zJ$^h3pUnC&OT=v^N6;h<7X|wc;r{&X@m4X?%rnd%>v`B^nQ+9FImn-0K7VfI|J#D|
zCVU9t^E`T0OVs1RU#7x~mv>j&t<U%C(@DP*`e}m2sr^<P*KFFf<KLUMvg0RT+dNw8
z6TBq)r&yKFt9_E3|J;3R|2@2=^2i}<k#)D${khVM{-#B}S?Sf;u}tM>;nJ5p=iUES
zuAO%`+cx^@6uF?!Wmc>wf1eeb#A5&Kuj9Nec01}8i*E8i`1p#J>4JNKy>I@gKRvc^
z-X5nj|Cd~-ZmrXvbVy}m*OSYRc6IIx>V@r!VisP@5Y0(!u%2+yc|yY93;(?zKm2#6
zX5&#YrA_Dk@5}v|rD$86<=~||dr^ls-{K`VKdKfTdl$M#k<070&8-C{x8}|Hki~eV
zd$MuruO~ep)vw*_s>^$=c<TR`16wvn8$XqN8Oar0^l!gNmW^hdYN{uD=0vUeiv6=i
zU)^{zW5tdcjd{^b)7AAi-&NMCuX}cFvG7jG%3pVUqOFT>+TKuW(7K`a<W=IFAI~;<
zZ%)^bHmm-&G}SymZdb|WU(9oJni&eXxM!acleQ8SYT6#;Dk6UN^R!<v8`(a%s6X4?
zAiXa9dtC00xNmdUe)U=Aq@Nz5)atY}V6$WJ_Op@u?u0E3(hi*xwD`)Zncb)P&aRxF
zw)(=0sZ~+hr>?v<j{K4N|9}7Py;CPeb6-9hHMKAO+-d#1-GU|8!`6poh;H#pnR#gg
z!-}gRFW0o~{4KUihi8WJ+|+E_<eBFUq9(?zznp#JmptFy9Tg|T7GHc-BJyj?|NqDD
zR#ctzefewdNv>vrUB%@`pDTUaz3<_~r>#3@&g^kpeU)K>Z$JQJCU@AHlu48B6`i!c
ze@L!p?$hZLYAtqM|Ily1Gdt^N?qi$k&c^5frsmH^yh_X3&$dm7yX&}q-ZHuA)lw!=
zr`sZ<*1oG*A9u_wc9~i5^2@0yuc~W5)HCQePHDupu)&;|1N}`<HZ+1(PTWX{|Nry<
zzIx;TfB%2)Wn_|(kbKa<+#x9~v4K&I!AgLe8=GZFTQ*5uMPX)WfpVi;ZzN|-Na*eT
z%x7dpTpn6=8JE3Vvg~P7TIRb(ZceRfkA$?iHsv!P-e9bAcu~R4@3r?G58d%9)C_9h
zB6a<ft~U4OSH7#a?|Xi2;rs7%BTJ5Lui9TOA@OeK^Ld}=r2n-4|6GzWHPw}qh3R<d
z?dy5b`|Hjf^PT-f@Y8egGq)efO9($(eBo~NdH<*LALR)4Ys?e<5X2pqcG5m~YP$A;
z)494^w}maX-tf)*=ljY-%Qd_vn9cjMbw(0@?Y7G&>NlO-W?aJjZQJ<(C2lW=o57Pc
zTzr?;Evpj`_`h#%_xwB8@1H$y?<Bd#C*sAHgoEALPbPA9JvgASH0NOxtNxj*e`B?x
z->MXUzbUWP-_6H6?WfC+lXvE(%$}1X_sPOYN~bzZPVw_s{!g6T6HYRx%=>4ZYIyNq
zS4QLlCplrszjG#^G1jn%5>pRvIdx+BA^S7;)o<PR7ZZN?eEkH?=0B<Hb8f%VES~DF
zkt*H%{kOZz@6*rDPCmKl@848Muhb`-etw>6*yvZ}`rr?Pk$P3#jKrWXHFt|Y$Cy`F
zoV)#ETiV>?*Gm0PRV>wUSU2J4Pwm>K!;9w1$=+~Y$MKzW@d<A)-=!&6OcqT{U-^8+
z%_}lldQV+I2XUPb=3Xhj()CKxE7dH{ty7}hbeD$(tzI!*GyPO+l<(T5F)PDYe4Qfw
z)Gv2|S<q|^{w(9KT3@BMw%+op47$DI?G?XQrCH)zr{8k_8L)4P{8N9{0LKsmQK#0R
z#1IQnN7o=Ptv1k6UM!5tGjcqAT;lSk%sv+p79#X@Khrs<6v3p6iG{~69o}^DQ}XJ{
zu(^lATjehnzbyFGk}*xg|4V!6hOhj|cJkJ>In#MJ6>dDaacBDL+~0?Hh&~QF#(Pt>
zx4x?8C(CWltlvk!rGC3ElV-6(_FuxBnLf*={>*!{t+T#^S*>?wh{?-$PiiN3tJOc>
zaqi<Qk?)#&CoNYwH>vEUSLXdIXM^u6*zI_?QlVhuW0U`vKKLK!v`YB=eZ$R3=U<71
zr%qokwy<f9gkEIV@sGWS4rhhN%`j2w{2acy;!eiofTd50Hr?2=X3ZB7&kF1C_PRx{
zqj$cWXQ4MC_vds0vCF!bTN?{wA}X}Q&;N6|zP0$SP8_H8(Hj&0G1!URO3kz}EeL&F
z5XJ2or8aN<TfuVe9Vzp_ENbFmNnEu0_Pztg-)s-B<;(ZVTKiH*`h@e+%F5oK8tZp;
zX-}@JG0V~IzAk&@ar}Jc#7i;v%l~I4>{#+#C27v#S^iJurj?l$+ol|!f7DCwnm(gp
zXlLWSn6{G1?-!c0zvnq%+`2|qr6%Fz`!<2X*sMrDxu7e1FNufpUSU0cYSOAnY@UkO
zx5_`{%x?Mlwo<;t^VE)g69n{JWG6nKCf{oCb>9Y`sfSa&9tR0cdaS)^l~JU?p01VQ
ztt)40Mix$DjyWR8BD{{Xcu~<^C6z3_CF}cNYrj7AS7EtqW)rj2`X3SMrB8~NUOKkw
zm(;Wwj7g>s))xM7<34#SHtLzKo!GSQt-JN~vmK_-@}DC+E#~`<UR}YyrJ;c_6%o?m
zr^|X)DZNx-4J@_}KkKA-R>QHP@QK)O8yV5xZM)9886S{-llg42{%rHs4Zc5oH%`C5
zT3_jmgO0+ijlVWkZ#DPMVKg^UxWe&oWoyz})%iS$NfP@%v@b8ORBvHbE)f5kGL?Pt
z)n#k%N$D<oyyM0{=a^*aPnK0NcePm$+V5v>E1heSKKJ*osgCjUovyozpPOI(|Jcu~
z`bxW1XWd_x&9XpY_3>Qs?=Hoj9OwV$TAf*%eXHaCglw_h7Zqo|3i#H~yJ*9sFU}ly
zzI?yoety-KX&=`0|JcsUmU-e!3hPB@?TgH3{w7<^%PxEC_I>J<ubav;rS^p=TP$kj
zPI+7rq4=)**`+I9dP?lJ(|79J3g4A*alL``yx*^`PwamGE;8(_n%EQdQt#h8m$`{W
z1nOQFn>S(ZyqUd4d<B-8T(8efu<f;88qu?M1IzBmybgAWLJ{{Rcdib9tS)u1RM(|H
z;uE8SK)G8%q~}Y^!kHSX+I8EzOFquhzCY&`i=fBsdrz+91jIG33cYnJ<a%kX)$`uD
z=8sqEC*01T&&Zy@|2M$BhUv{g-Z!ipq~GX16w6nY<H+a#8s5^%TH^jyF7NeA#~)2s
z=O63P|ESgbV4K*hlPiTE$Qth1!IAb@ljFCYQ6lecn^j$gv5oq?llQYp>L`gFNsc@G
zxO4T{hoSc`E65xQ-e8*DaihPa?h(V=2TxX&vM%o|^Wlnjb29G=R?Ocw<KN`P4_ZsY
zJM!P!X^Neh*)2Iu^MLTXbN@<TM7W2mC)qGrUKK2vyDs;#>uTXWDrvvASLsJ?E&F_R
z;`{Y6`-(CPR;zEauIJP-QP{EM@~hL29Dg$Z@nL7O54_)adDX4C*Ir8AT&XFP8k^%D
zoc;8G`n>BOZWNg;Fpra1?&0)LDQDuQhpFP8C5FoqTqO>jRC=Fu|Jf7cYe!q3F4diM
z?!_*P(g%&U-sjdcy*3e1@OmU-dp9Op*Gh+JcbjqfI*$UbJzDpI_WkWOo)o;ik7v4a
zgsD~Jp3BwW+e}yl7x4%6RllF|qRGOIx!ROZWrAAv{Y&qz8SVeP)BULOBDQsw79HKa
z&g*svCM9e-G2L6~;^jLH@1)<W{#3G&{C4R~;E9`l$BPZyOX575Jl0>jXjrmI?^Wrx
zS8}1PvvrPdHc3wW@J)xe{Nv+pgTTiJTt3ygvHP5`keJ_Db}T3O^wM+ZmcLioSkRW2
z^?<G6-UsG+ExFIm9C{p?#&G+uh47IF6`%5?qqr|0+0nQ!uJ++w?Ms^vcS)wQ=$6Oq
ztaUx0{)z9=KZc~-KMi551&)rjF{<4AFHfzFvG0B!^|ifVCC7ii<!uwHnXN*q9K!i0
zRF~ax-ly2-a3FnJ!AyCv`JZ(|R@d_9xT{($W#|`nd%Cqa^pZK3SO1gEtq+vWSV#YO
zXL&MXX=u97<`_9)<>}_fOUlg5%U}JGb}Gp~=pQx1>x#`sZ@bp>LiRyZ_j+2^E?y;I
ztX^)k@GKuk<H|4bQHfeRLle@N?$#Za+Mq76CGhw8^*WX2|71TNeHwG=Q)qRZ%kA{e
zFVb5k^41+WZ0uXs6t=7F$=-PydNb#89!^MaD=kxh87*IQ_l3jE`1)jTUCWSJY*i*3
zJd}S#aqm>gJDkw{<9Mk3gN=^W#>q8;IgLIOU;a#E4tvBF$h7>;T74ZYeb>FR?yt{#
z=kMQotM=!uv-&5^r|i>C@9y&QmQsy9b?e5Hg01%_>0fIUy<xwDd)lwyK&gM4Nt2|V
z4?9n*VS8i5@Y(A}>$}rBylzESR@UZvtw%rpC|_vRdb)FR%aqh_3L1y5s99;=s>q7S
zx4NOWAWt$udi^`WWL93TAD%0$?{G_5EEJTv=45Rdqn7f@Kv^<TS~yF7@y5_q*WbOp
zl&a6nx;f$Aa{K$Q>f<|CTA3N{x7hlZGotUTAGe>|UB}<qHA`-ku}w=d-}5oPg6qbf
z@-@4cPV9a7h-X@>=%Q<4O?qV(50!T8`Towe;%Cwi*Y<UvKi<-fidf%kRqycoR%IH~
zQSr}xkEZ>Ymzx>4^I)}r@w8PQTv<1i=Wf4N`ElQ_39~Au2?k_(zM4_s8L)Mcj;P^G
z;g~%;!!N0H*L`g0H266Ep}yc_C20?fy_=b*F8N_%uOI84zNFx%WKNM-_}o)QmoHsf
zRXp3$GqJ8T(ejk0#TuC@_jcFiwxo(#m9O3{WE~isnw=OruQa95J8kEdbxSvOAAP-X
zX~pS#d#$(qoMY2I<FxDhxfX@r<|gmnuDrlN?0xAy^XzG#KMH?M-MhDU#kRFQyY*xl
z6}NY6ecb$@X@}NgrM)Z<tsnURICaowtBIyvsHDFeXS}6TiCbN3=1fh=k3y-=S-Fvp
zagnA?w>QK`gvEs3Qm8!R%@Uuf*b$+({q_SDuDGjd&(5c%<s80e!Leulp5HejZ@GnX
z8DHOMcs5ADPpBnXG%fSau@ejy$&W4yiufzJND4YRw%p)8`K8~we|^MzSsD5M1CRA9
zy0X?wI#xBVyT+<gIpr?*>lf|jeDf{Cb?^5S%$cv)VO_HM*~@gnBiX7I%P%L^PvE<}
z?$Sy}mdn3NE6S&?dviT`mfdA1n~9m+6FuKw6j{%7O62E{4U$R$i?w`a#>!tipUlfq
z{CFWVi+#-6RFgk?kCQGx*rvGjQD&My=SKs%bhhl}JgW`3KbY)`{FA=t$VVxqj3s%^
z!O9PQ6wIB*`Ec#GuF7!bsZk#GZ3Z(;tor?){8`*Ip*}@o`4qP$&z`Gqu4Ij=>s&Hf
zB{J0~i6!ZZ{{I8At?Bm|ZhLpkoN>Hf|DX2^wGElhUpDiU<;vdT+O6jIw7W;0d-2EO
z4YJ3YJ#OSL(@~$C&vT)|aGrKcJde|^*?*D>LY}5RQ%zg(?yKbAm%JMee&#*e{?8#<
z{&hsYcB9y)5}EMnQ>AyRHT^8ysG^<zag}0V>2YOCvDqgLS8*A{e%$7km6`U-A=&4F
zXTQV+<=$6&^Mnj{-FWkB+r`yC7vvw|)m)mXV$&}$&%LfyRCU*a?9TN*LGoupC#zIT
zzZYM%x#Xdz+O%|~9nF7RU%hI&W0aZeoZ5Y_cv{wiu*V<nUD&tKzdb?gXF}8uC#4`A
z@ve%gE4Ftokl*YP9=mYumiLUIQ)b2X?+s?YDi|pkT=eqGWPOKJzm28YyPASMx4#lS
zZ4{g68>253+Ub<M(_otBC0RL1)pR?PHL_FHSKTP)tQK1~MQin+ZlkrvUE0RlOGB?N
zy}C5V@#FM>n96fj`6gZ0)-Jp*H@WV%$fd(A0akOQnWXPlKMF~ku>O*_Y=^?^8Eh|F
zAMVqvD=N{5jFVU5<qbdTviVH<UCY9lCX*`ye7<+2b>zfYk4Yte&u-S9Xc|%BdW5G>
z#<p_HQmc!yiBmQH1vcETNxD=&<I{-~-`f?YshsfnC)_-DzFE6*rCzW3C-&Cbl_yyz
zUJ5@`5YS%IzrN0t*(z?H`{833n-48mb?Z)9W_jwt#A*6@!nP|VZLdu<`mp|B-^Fi5
zroZPexpnwI<Lepg{<FXJ=!&!AVLqZ1m9cm0H?7ay@?6u8PH5HLTD<Y;8Py7zW&7he
z#L~}SldbI*IrdU}{#@ly_Ged*ZEAIOJz6^HTFWsF*O;V=jK5Kn_+&kEZa0fZ{p`F^
ze0!Se>238Dk5{jDzTQ`^^*Bo~`R?m*vHf95*G0JXV;g6NClt<Uc(80bWA@z_RWmeg
zma<3QWq9o$F1$cOqf5_!TAOT#$Jx}$oT+m@PG7v*$KtQz+=g_ny*|wOmpm(Mrl0i`
zS|vQYiffnF=Rn;Xob~@+KkVn#=%3y=J7W6<p1l#r78QAL%c^{9ey~TP?`YX0--_TI
z6}!HEQ`bXkH*VZvJU$_NW`*!}vs~$0v&2Mq2mESU^s+>2c2(Q&SFtAzz6Tz?{lY2K
z|7q^avvph3d82pEF8($_V)cpdTP`ef7K>HQd;g{3%Iq+|^GVfbYRtuT4m{fZ{vhMd
zX&rOTEO+n7*;V;}_UY;C<90vU=Mew<X6Avu4ZCta?P(XBDfXdsOP}e{^rg$vl>Z;K
zzVxVg&E%rw;wM6`wPn7CFFrb~e&NL?R;A!HJrmYfZoD#Ap5K$%v}97b*!}XI-L4$k
z|JSK8ZsfdlI8ph=<%HV5?`ngh^6o#kVz=dGuV1>~{@t#*6O3Bfp1t0*``$zU498`u
zGXM879<+IJYUQ_jqxI$g)~^43ss8?i1Q{P)&H}~dg}2q#oz(GJq4Mv~QMnmRVUyb1
z8(B}9uh_O|$p*KeS$8Ew8^0&I$F$|%tK0X#bjz<hXV<R1^lj0xphE)pE<_#DZ&7SG
zJ8!q+@{3>Jo^!q!<Qf*^R^*y?(NSz#f+KI-6TO+D4Usyr4U;_A&eA!l(U<+YsaD11
z)P!{>>O&sz|KGXq`ecEQ-AO0a?bMVmCs@Uv{K_u4TrK?aN0W)2GkgwTei@>-V*Za@
z)uxvQz9|Y?ZnpO?m^(+EILM#q-<)0X@%@FiTk03(OI?3$%b%FEI^APspL&h8w2oa>
z^@63HM;FE^1s{sdOZmb5)ANyTocHuyJK}d%zwi2Xd3OHVBID3>z81q>nYCMbm)Irm
zdo2B~yL$1?EhVbr^SX_Sn1r6*x&8d!bFZtq8e6?@E?Ko9t?p;S0>|QzqUTR<SLhsi
z{m=T6L{WF%i)T;v>U^#)))9ZKT=@Kuv{PV3Z2Bd|m%rD_%cvC{I-jKf{<_!#Q-Px1
zG`DwiD@vzxrYFu5)7g4w`WYs>&h3Kc%jBXr6+0ht^Dpr@Qj>JznV6c)e;$5)<HvGW
z)9xG%x}G`v<J;Evi85X1R(EGlHZ#9gCNVpJFI|6kwtrUR)nD;xhdfHR`_yQDh{zKC
zWGfW8X;WR~-o^e)>;HS6oOv)fJTd3&o()MN!2uJmX6RJOe0B9NDNCwgd-eTp*h}4`
z)6Qv&t*^iR_H%}q%S5MT%Zy)t-XX4P{cZ0qr+1+@Hp>4jk+GN-ogX`WcC@_K)vt09
zZ&tj{DKQrh|2}Qu2jy8yyn{CE`}Xz3;%iq;d9P&a{M?(M|4&@Zyj5&rxK^uR>-vy=
znwjt4{(SfO)rS>KQ)Wz`e<OK<SGK^b1}$TD)3<j|ANBrN^z_n!pZitr1#fJYoWCq*
zItxpS&W*iWA5CMrnl8Vh!{~%zw7y$p?mY)pE*YEYmo-~k_s*Z6^!8Np&(Bf^Ip0=S
z%`r@i{$m+Cd+LEykLH~-Dz3SwtT-)gC3xlH4c2^~V-s&UEqdhi>-8(GT>*kT2W%qN
z?o7BFF@Ju%?nPCz?!*mJYBnby-nqK<e*^RX8w*qYCcW9_%kza>`9{+5K9~4>7XR@2
z>fhNf7v9?PM#;?D+FILDC|3UZy=xKIZ6$u3%h6G^f9#PvnJ;X$yuvd*j=CRAIh^}+
zif(T@b~isjxcyS}=jhd{hsFKt|J>a9RqsR}|Bu{hJM4Dq>8Tl8s&>zL!XKWP7Nj>_
z(Q}1oPLAHnm0E5ee&(+`H0hSjx{#zL((+mFC(iHm{*kkJS=$u8lP5S5!@K3StDZYJ
zCGaZW<_#r=uk$9vyvn$D=)|SOH4e{Tad+lzZ*1N%FPn?&WUsHw?4s>Uw3n{lu{l>~
zMPg^<<iwM?zY6bXz3g6QcKrGe1>5I~54<+Fv$k85AM+%hNx%1S=ftzhp~?nfyW>y0
zeDI3U-M8y^Q?$8M)#A^4%T>(7>N{L-l-<?4KEW(BIoLZky)^9SL8dF0i>GQ&t}C&+
zU~OZmVbsRk8}x+bSwz0&gfp%lCG5`mmwkRV8^4bKd}CMP!GEEm>WgnIS~TgAn`d^-
zVkgPTGP^#^UoW`M@aO63?_XcOSLoa-7?7d7<&2}<n-x3m-ni8{U3ty2nVjP1YU)4#
z5X^q|@3(V){I3wEZN~XY);48FFCAr^n^|ggP;2w*E2=Nv&YEod-YdU;)AQESxb-_L
zH(791b6k~-ys6c>?q}Z&r5*0Z*ZQT;s<_^};v^NAFI#{5#$AEgsUClHb&GFrOy2$b
z=6}Zzx!XG!-2J!9S<IX9wc*&~{j>gmco_aY!KR7*u2OCOl?%19X8C*lQYZPYz3yK6
zNVvveMjG4v_-RtRnqI$lz3}PzD~ahBi?>|7-Y=*7UdG|fy9;~N*2QJLId<#e(_Np$
zt^c)u+O9WEezoA1L$9|#EcO1UV1HO*W3ms&P0wV;eVwcY!8bCm2}&e4ea@3|`LWMn
z$E1+tOKoE7);BRo)o8ylTQ4GGb%49N>AK~oje>h-`2X-pR@^!Nm~f)tg%4{!aO7$|
z=ut8$mR)oD%=CJZZRWzc+CAlqdY*je;7$Ixxq>N9<&M8`ZvXMb9><W_m&NSI`X79B
zsyws!jj`MQ)55F?|CHCv`j>G%tW3%;cy1({_odm(w%FaO-sP$9>yl98*kP{W^8em@
z1GR0uCTE(s52-Dd@(_H_$X9LpN3Yk}dQac=`)uJAb0_|pd+Yjx2U8l>w45;rdX#Fu
zVz!a=9EHmp->%QS;4J*?!|V+PbFEK!S{(J(<T{e#R>Ag6<#|-GYig*_v7NdL=kdt(
z`}lNxvfNVt;bHWHh`<wvYU-{}o+iYT^-$wt=bsfn_N;vSwtu#3fh1e2Z*L`Q=_F^v
z7@I@id1q`p-jextg8rfD9~T&y>3vUFq_SgXw!@!_T{~xFubC*(`8`v2kLR`i3xfIm
z=?O;L4uoy~5`VnnPm8*6S<Y7hl?Qz>+DCpwmfMs)^(vR(K7YvPZL{?C^)~J4k{038
zH|Ag4ukLo}&HjYA4QZF^O~2jW(peJ!+yBARZN~E#X=KfOm;J|*FH`i-EUt*izzE01
z8^h)m-8P)Fwl{6fwPyvnh4bEpUWtgv$-aGmN_@nm>l4EN`Ni*iYq78L(f^MB{Cn@`
z9>p>RVM^T9V<`8ggXcRw{Qmp>`v3p2HQ(32XJBR%U|~>d<<`Sy1JYhpk|#M#Ow5pv
zj_i%%j0q0ER&V&ERV&XX+3@1;MLV`cO^>>r6)k#nt=5jkS9dXRT63x>Cf-+Cai<}N
zG39<rpo?Q*^xA~%wGqizS6#ei6}zR(ZF!dU+|O^88sF!3Y>IvMU2xm)cXiYL|C~RW
zLqlENOF_V9f6mKEuJ^Rnb-#ZKT7AV>eSH|4TkFrPoLKK26YpreX|ixBT+=Z5>*FU*
zd0Qr*baims!Et`4Md@Deb#|S<O`T1;7O@02|GfXVji=R7FZsuRH|dyTx8y8#_bfib
zbWq^r3d8=L|NrDoujJY~|5I*j&{FPeM}6iS<=*1)FLmirvoP>bpKq6YOI!bD-T6u9
z=PSj(dM&@oEW^3^={~L2nwKIwuZ!&5e(Qie@3j33!cP93;<zu7v8c46W^GZkjo0cc
z8BZIfN<&({9o77_)$w4|PRF&JQ$N?6o-$avK5XR#_mv`Ey)oiTrs&6QPg`3vF>2{#
zeg1FZtG0Okd~&jAXV}Uq>@Q0M^L^&7>gJlW=o*V3?}OtR%iC_UXD+|>%r|b!q3X9!
zp5|BUeTmy#(4`lp&L5Edglp+Ja|<8qrO)@>%ktfkFLAW~<cf(`Bu>qh^6p)HELeDj
z_zLlromW&|sb(3Ow$AeMT^zS;S+Mbn@RiqBTwgi!iq0$ESH4+hrtP!5D+4~S_<4mX
zOAvGyiSE*Afu}X~PtA_<T)QYNID5s`D^{o4qx^Fhm<30#%$=hD)LnBWgH{LEGN%;*
zT0LCLL#E8?TCinGU)Q29QyRMjgP3mboEGH$`{RGhz>8PvSw1%9|6W+v{d=x(6}Rsp
zzvYpBJN?96&Ue<hn!m7p6!@s(qx-qPbB*W7+b^#=cgsHGz#GQhv2DAb?3^QWBY#2C
zgSC!|%~ovIyzAsQC7n3C(9Ze%j=s{+U2^Ge25HZ}Z4hsZmHRe1rSHl9!_D*Lf7C3U
zedO%vwS_IOJCZNG4BC9g^IKES_BS6r4xjkn*X8&w!}=G`EVjb0H$Lrcot+?XrC3XG
z`bCw@t|JFKcpjW`S-UdO!bwc5TXdyP;Y%%HR`*8}Cb;cs|J!=+a>wjhN*lV?{)l6_
z{IV-kDfEc(T%DQxt5sVTevbBQG<!W`bLY7$&t0CanETWAvhF>xA4eWl9pb9e7CsyL
z@%F01CL`1AaxSK&KfYbcPPwu>`Qx09Pb-dn+I6UTrpEacty3<`nu209e!t2%5EbTk
zE&D{mG|zeEnKu_l%bPqCS3IV?W>fBS&d9rQxBtz%;J@g&m8(foS5lLRWap--b!XeH
z7Q`!C%)aq%cJnPJJF)A#!!AT#kDB>7MrM!d&VA43q;pz@|Gc;I^QG6P_pAIBFucI|
z(02RkmzT0+bq|;<X8tvKxB1P^3tBg&b`*(8@!A>8NXx4UeYZ48`=sK@X%D6v3rI3$
zN%$H2^?&o}PuZQyFjq_<g601Pr)8lha@!3!dFK36F08t)A12$8u<(b2R~^SJ_v1HJ
zqT9W1D91{c@Lb!n<-sf?k$vk;zs=xj;s4LaV_p7Z%hA(f=eT$LW!$Cm@5h^y#naF0
z)ro%GUGZ}M6iJr!otaL0C974c!{e$LmEvt@>T7wJPs>U%Y^%tvT4rcHvvqylY<;hD
z=epNUof;}~eZ$`3g9XkU=}s#CX`S2!tAiS+dcN2m`6y*gM&?o7)@{CNQ48KGZHfG`
z@mgYh@v3r_-^xCm9X#ictThd)aE<A*zIJu-<=!28uUuPmaP5a0HLW$zb3J+=3)yTe
zv)yeX^IKxc6VXDEKHe^O&07s$3L@ulv43vTOL3gkxcx=WUA8^{E*C7x+~3xnqWXvH
z&3qXKyJ<234_jJtmn}*V3FtY!D(80CgKw$5E6Vn=`)$*e5sI3AFuUM~-kJ-y#QKif
zEx(!(YY``(ot?F@$!0&d>dMN8<_{(vI~}?8{YLq>ht;)mvO^zUVm_*||5v~{yGYN+
zmxFC256U~9)i!=8JKOQNveC<v95Y<RZ|6&$pY*iEqsy@4Z|+8m8*^^8Bvh8i&%fyz
zrdabzzjD)!b+vU)d)~5q-7%xpbAPYG<X>B+U-o^rC++qV5$^4WA5|UUQqMcH)yTep
z(;)iNnq!)rfz~Cv1r~1azb~gg`<GgPfz+}4|9)CetNc=UJgr~p+Jvd1*Hs?#uA0*t
z_$Vnxe$o9z9+L%sZ*LR*b+SWRa8h83U?k7q6B-kkYQMkU|6SreYxjd<?G_t>B1H$!
zRY{+}a&ksY@7i|khdTSIG`&A>ZU_IaXI!?hB;3tx_SD2v9#dBtX;*&E(pkT1_rpH(
z$*$U*RlPgz$nzggIP>rK(z|w?MirtdOgA`;6l<9OPnF?Y%l|37o$crQaC7-}_wV<d
z@LX<h^EiFdzp8|{#>+V$-v6!g{7--0t!13NJZt%^^3EB}+!VMdXmy~9)$(^ee%+If
zO*=A8_Qz9afyH?m54MOM)YTUGbjrKwvQWj-&>1dnN{bdGD_7rHzsFlTxBS@N9aC6Y
zd6nON_%!EWwY=bZH9eakpQl^CXB7OJz@u*RTja&L8+Yz}Z8&>&X2rUe<GVOy+W+j}
z`NhxYr*?_m{CArBiH;AgCtDI{zD>FCUEgVc-?mx5*yQeCvG}F2%We9Vn+$nuXQmkV
zt~THQbmNiRpYF<bKYevGSo+$nz5lMc2GsQL|NrpOZSnmEJLc{EGx_QA(}!7P&wQ4R
z(Ks``X4l(jqdj5=-9CCIH}BK9VySC;b;VS-kXNs@UUx@XE4`ceQT0W_cE^^4P^P53
zGM0|xYI<L4ymzI(51Ap**x|NEQ-6)s#<Q08=G#}e>@wfwwe+fr#&r`vhVK?43VsJx
zEmMn$@8RpLzf=77maL}XA+3|A4y`?Ve5o(*@%3*V4R&X5NZEW}Y`x2oo_7qbKNc_f
z{`b1~?ym=%KdW%(Y}oPX!;382l-TrCp&f4)PqKbg6tg%aDc7WG_M66n0#Wv)W?rGw
z7x&Jyb5>AVx1ll3tf`P$ws-5dciCmrUr9X>sED0zwrKZJBh78>9nyDS2k+@M7C&%Q
z>BSGGt~{$VHfE~ITCBN-IX;|I&(5!Z)NLW=cj2k&4%Qg4{|+iCAC7RQ2uOY?(CKLj
zJic#rYWiUdPR%9y;`9F=P}&{6_rqM7+n+W(JN8w&Ni1*i1n*~UUQvGy`)Al+;`|q}
zFy~;JSl!aIx;t7`C(ntRV!=8w{-ua&m=@pdL+ARx%saf}>N|rg{#EBDAAjkvpyh*Q
zc~p^EadBDKnH`zOyL`4a)My6HI%SlVa`owst4Frms`lQR^`T&`nav)z+czSDFK3i9
zeAwPF|LE6=U+0Bp+GeEIJ?)8E^<?FlHmj6tH<t32Z%N>ay>no3w1r=Sd87SVoei#*
z>9hR=9$(?r`F^Blv#(*t#oHx^Cp~<?{<2mi`TqX9Ow1JwF~_POaJ^gd<nr~xJ9B34
zn|$$@=c9^H{wKByZ-sxm+|RP+-}u&6@K&OO`~ukza<%_jcJj8Sz4g_Y9Qsc8X8v(c
zjqL|ci(Th?ZKf@;)<;16<mD;H^Y5wuJzP`nlIyyGXVTHVrwuN9h@OpkaV#h3V0*`j
zXOrgK{rPi6%K3fI^i$3I^`q?aFLuqWllF>UeDm(^&s|+he{V^RI`ZVT(qrcMzt3`C
zu%B6Mt+3^D?{Tp_mGvuM&vnXE&)L@A@$<tBy9u-8dk$szMzz^ZRAoEV86^BsZ?B1t
z^+s05XFEh|pX}Xo`**Iup>-8CSLN(ZEK=j-T&&2mPyA-S<wc>SMxE~LZc(qbJ?z`L
zd&+JoHfOlyZU51i*KO~!f6}A(Tb+*0ne+IgaZj^&p3OG1Qny~c#L(TV3lAD^RIK{_
z?Lowr^YMv%QzkaPb8(wC;pu;+ya%5iKK!+b%d9|O@69!heAe~dl9i!rZMRK)!ENyQ
zP$&0>Z%x^cE4P;3tc=LLTYm5MDW1;{xGcj@+WG!%>&QKJLSf3}Rcr1k%(lqC*K?t1
z!6}bvlFpLCla;Ong)ZD>AZPe|Qtj4Nz0yn9Gi+s&_~%;t>F)O8+fR1~?#R5ScQr5l
z*xh7pyGHKKMaG+NPPw$K<?tr<>+5Q7&)K2Ba?V-}4TGs^R@ud;CY??-efrc?=i?;r
zYrpfpI`+)0O3%Li^78Rpb{Y1&&bI$GXY8E6{5{JN9Z$La4_-XH*6eQO^QQev`Gc1G
z8)9D-+fGjwnDa1=V`ss<f)KT5OP_qY!^Py%`Lk2Pd-3AOCBKxC50;6|I+=9IXIjUk
z?dnorgJLUVEDyQv(2#G;-C?<0W53g__6c13wly`CH^<n8tf&yaSbWQC_w<!H%kIv9
z_pP1np1f8ke_+G;_^)M|U;D(ql=ocvoD$-ArT)#^hey9nIO@E9vS{?oddns;my7)O
zE$y=Jsr?t3w^IMFaM!A@uannYmOb{0UGB?YX?FXH^%hyGO1A_IH(x%OqIS!97O$9g
zo<+*7RYyEKVirxVytXo9<HZBsC5-2-Zr+=4R@hE)`#gKue#R4XFaLYvpj>pK&RRO?
z&z+^E<y#&E+~2(|dL6@4E=PyGnW^Ur_t|#$w!i-N!{^wSnVJf3pX{`I`~B@nzJB4w
zJG<;%!;&@>T)i7TX9s_C<D;K9L-QJ(4xN3QSaJEcj`@~s8`<`$6B{)wjAYzT_83}t
zAHMeB&J(980a8b!Lql9*K1BC_6qLBk_QiBsW#jJWU&I^EF>9Um_PD_5U3tLp`-%Y9
z$%PN5PUO^kppsSeqTQ$L#oBpl`V%iIRV3IyGCI!LU!J-Bi{*Q7zQX*SUibgVxTMU{
z_>`%}asKP7t6~={<;wr(K9$s)|8eOprApOxj~16snb-fsbI!KdHwKTwWvfj}Go5!u
z>ukRKjiHda+POqtn&Wk#wB)ZT;eY)$L>X=I|J`jTcA$5oNHoiItA+8;YvTe~?>2S0
z%n94Geckh-)fvrg{xvtQrYz)eKjX#iELn2v-<L^CobwNhZT}ULJaK={)Y_j`d#1m=
z)K%SDa9U5P&UISwj3YN^tw|~VAu9AT?aV3u+xn;V&+D0Qwsb9Da51Iq-KXDwkHjf|
z-L_-iE0$Z|Y;#nKCuu+KUwL#P$D|`yCFaEOeeCA%b75b6)m3Yzor<nfPwKuk{+AzC
zM6R$+U2*Vt<Rv}6tKTf{mVe7Qx&C0&@)?$uzh6|hYoFc4+i|2Tcplpw?ejdp<ejUo
zKCD_VELL)?^iw3qvvs!_&+OG#V@<j6>$An>4}uemlC~e_w7Pdr{GVvX-#hG^cQ6RY
z?v#1ID&p~(l;$noI-l+bw|+l0L-lLwGP7x?SBU(NOY#?db0S_rqIhLt`o{m8MfUuy
zXOIZ~{lCG$?$5rg#;EQGQ@)-sKTxOqWXp$@EImhFp5k_SV%N>1a>!br`ETy=-<CB;
zB<5ID-!i$`rx3Fyz0mIblI-MFhYv6Qx@xO?O&#Nr<YNXA?Qwsd9_=mptki#Xe$4*b
zmtUSM+;I+IshahR^W2R`dt^-(`o_D53A|syQPi>})a?M*hjpu(l{518rHFp>sAS(e
zS!V4_t!m%spK(UBOStO4Zv3KT7hS+KSJNtCpW5!&=^IvcoW0*9zHaijWmWP~3n#_L
z=V}M>%=*(|A^iO0<H}jqU;e2qXxvvX-?nJy&PkbZ+%~p**VX)g`}1@7x;gQ7i`BOa
zMI{KFufFs<;cJQB{C9rlmxY2aUA(3_yS(LXk>809e<P2{MPGW=Z%J`=1zcb9GO7Li
zvu~vxKgA5nuKV@;S+wy^&PjWled^~G7yTD5$`5~9(5|JNV8^p8@xWF2cN>d+h2-z6
z>MtxzsPVD4*i(>fo_MijxqNz6PKW=UJO2U}B+AM39r~5t9Dg(Z);YId>$%O&Y_Ok}
zbD(+u^y^ignMHZG@?7GoR<w%ed-6Yzu}*Sr*OSe*Ej_T)K!3A=d(@PuWxR^fol~An
zvU%{|?CO*0qF*Cd+|+7`&n=tizOBi`WxCMp33Db-n%@>>^7hoWo9BLqiRnk`%@sMK
z{wwctfX^aBTfbQeeQWcW>MWj=iC5?~-FCfEDHwm|L4C|@Yiq0ehxIcaim()>tC)+*
z=f;RX*~(z3AGA68`W&0PcPd#O5812qY+ZaTKt?#Tr_bMc$;0iIi)2skzI1s>;L?qu
zTX#NXy~**-cBcNV*a<V|D*s*ZdgGB*HlO?a7i?et-0^)Mf7H1M!Jq2W)o%*PZnZUC
z_ADd1(s$iXlS}dk53aL08TR|~w&bUwJAbA|#i!`Y+vv*1O!*Tzl|k<1Llylg2mAL`
zsj$eFM;0k;a!C(RXIYrE$v4Fy^_5iWZ=>ukU!5NfPM6l)OTOTGjP>*0XxCpp!B^}e
zlhaCfo_zN_yJFhe+e%mGTGr<7&0D)IZ_B%lMn#t4za5GcWTbnwWM(zac=F`!ZYE)q
zAA5zL7p~jLwe-r7(3jqGX1dB9f6X`dm&>oSZ*DKKSL&`~Z|<41$=#0s)Hi{TD!uhr
z-m;!qmSSzr^8a;7s%~FS%Id`%&&8ig$&K!5e$^BZQ5qbz^UdX5=N{`+GXK-O`DXX8
z)GZu;^e&d)4$FVnJ?VNy2g~_to9P}8Gdy$8OuQL<{bz>Bioi25o7#gKm0xa}^lfKt
z&<3XViN?N*e|GRU%{G@;a8290(IqT8^4f*w4o<O(>HbUCJA4*ZoL{_GHM84V>{8E^
zwTU~|hgMA6p~?N{`+j}fkN>KtJM2lXiZJUsvF_Njd2gCup0i65-p;;s(V`i-yu~Rl
z2G!@jT=nhx6Y{Mi-}%<o8;3RX(uKLpuen%GKPKwl;#s?G?dK1whyKW44ybf<b6w!m
zX=<pGc;sWm-H?b`SCiZl-$w`eom%k!dHK5=pKEh=<}}ucU0*GxvBBWu9Qz$!Sv~Ai
zc~9&QoVa<zh04Ol?LQZ6S+wls!@FxPNC<PCT~j+tv9wG+aQeI{+k~h7{Fywr=S*jn
zZN{nnkM?;d_%FZbaZ7_Yr0x8NPwW4<^m|<@7g_a2eU@I@w%dIvdZ)8Y7EcJ4f5;_y
zX63Kn3*~lRz17tDS~FTEux|6+X&bT<=d!e1elJwjc7Au}Z9yN6|Id#~Z|d6g<qDIT
zt*))6=P8+sQ(j6*pVpSP5|Q^)Qm(u%=VCUoL}J4a#-w!(mk-~5aOm9G346BYY+e3*
z)%kyw&+eRD9e#e@thWukdS5zsA9=P%DmkafN-6%IS^O4TJ!a{px4M38y0s^0_UF2j
z*7<pzoX<V0Pu}pqCj2M+hL7D3>s9Q^FE$+6!caDEhxt=`R_9$?X8&r~?LOa#ss5<q
z`DyHzb_Y*7$QE94GHY|#vCi8ZhjU%;T{-bYg!g^oocrqza4gM{`?p~htL>SBCm&Ld
z>E4PHzjS1QdAd~>XT9=8lZ9&zP8WP!a=qx^XSvShwQQ?(*BzJBsoe1LU}5&^(#=z6
zx)g5Q;P5H&N9*>eiyMzU>+@e?{yav(MDoGTR6G0GTx<JocK=8$ojSF2sr9DvCmX82
z&p%$d$@}=!t{<|Dj)`l$7T*)yb-MCu-;Evv&0@2gWlUGx{@3IjJT~?9o@J{0`tFBs
zni*-CHfLt0+`%Vm$-&q1GIL9|Y6Xc)+x*?f&(k_F{{IJOe;3~kJIZWsq_D1?y7Qo`
z+|(;yr|c_<nm6M@xOIN1nW@G6%>OSA$!dO?Uoe@+wP((k`s8!jI>srt8?PiC`4bQs
zeko+eK6TT6+mNay|F*ojcJ=J~^U-@RRWo}ton#73Dw2}tc(b8+d;UVpgGxKHS}W{q
zuHN_&TOOELF@J^kyGz2;)_?yPdGFtZ`H|{nDH7*<cDYYZ{2o{N<HNn}3XTQm?bdBi
zyOr0yJA94%8lMRHS7m?wwU#lO6#rTK^W!PY9Sa#9ugV(-P4e`&UH>cE-Mg?-H_0aC
zrs<A|vbEFKP82j}iTUCE-}t)Dh2pnjj+WDPo_v~DbZcW-SZufSP1Yv?kF5T_etFw1
z<-*I{U$;Bo&MP>wcH_3=499N;*6QCoK6%1ArHO}PT3@G>hVHm}#{NN<%kOWu`hyld
z|IxB`+GWn2K6yUptvSy1tq!^T?#UeMIl`geJ#WvE&F8ps@qx=eG0pC*f~9LVv27RG
z_<$w8`_;VE7fgBS{G7%YT3VF^nxaD9eXuXL-edm3T2DET`=D>gjT^VGzTPNmV&i_r
z#i`4M?YLmo&nlVUraF7wr*&01A3cBmnK4tI`+3bNCX!5-e+2L7=CkCg^4AY**S?{u
zufI=Q>uqU9y8EBZhbPqwpG$r<33QnM#V^OcP0D>^Ji}kng|5Fkmb=`&Id7BwY`bT-
zuH1@su84Z5pKNNEn9r_n(Do_n?mdZ*g%8el^xrx4?c0gH{VJz2B;5VFPblsEq2Hz3
z{X?)ID&(3rW73AFhi~hwb6@Mz5_?lirn_T8-_!6LnO%COo2-ko3s>ea&y?72`Rc6+
zZ++L?>Px5oG(@oq-|1Ri7^XLSeyZ7Yuit*Y=S`+8-?+Qz)z6g*>i>mOC4FB;>@xmz
z_KJjvw9HEG8#CkIZL?Qi6`|d2aHM6k&(i<@*1nD3a(_oeZ-UYGu#l*B<Ki1lcc1++
zVqaCC)LIqiRp_p2Z}I9qx93cWyh5qKxP_XVBW^7^XxLHnT|FRQXmWI{)I>wg<nTQq
zk=>^S{(O299o{4yChh<9i`l$aPx;OZ3obl9Niaq8<FA=E{7MVzn*=wAoYkE$VdLqJ
z&vR~`J}7zY(2VnPF?)aARC(0LcGE80ux0Yj9(mWmE7{x5dnQ~LF@B!eW_dJHP3miT
z)atTIzTG?<zHH~qSZVw*e&)?(U2AN_r%yZN6g=%_ikA7zoppN5|M|DBaf!n=6=O=w
z<`yGU<S7@>WY32G|Ns4dUHd<J{@?lU85rEsIM-n_9ck+fsS`b>#%3s6YFZ;XBSgZt
ztrJ_tCCaFIYlhz1Eb{_q&nqcOkJ=o2x!-l&(ooRo*!2E{((Jpnv(wxKT*6!ftk!s~
zG!!xo6i&Y4S&}5Sy>Pxj`k9sAjNb3Py+dW@liA0=2j0EEcl&$G?{hx?|89Sm>(s4T
zK^h{>VO2U&yNkZ=XziZXn)jKv^6=TKwk50gr>3p5agpoFVU1jn5~%pOHf8J7Hz~by
zbJyB!-TKb+fP1?^aeGAm;`X2^NwfDp{rT77K~DSR^5)q&?on*}w#m90a<%TgJR$M!
zv-`7WJe7YX{-Lqn!6PIf<j-v3J9pI1^YM1~bKH`VSXy`aZ0@Cs)1PIx{$}}FbHDfC
z!i^ITI%eG$sQ>zh^OdXcm3a#{rkHrxTA8@K6&HB1yL7Tu$>b{6hUr#TMhZ$ndo7hr
ztaOUbuzc-0u)w1<C&OjKtUYclIlLcwc|L@@xBp;wS+SUj`ATuPUi`v&!csy@<0Q5i
zDySGaynVGQ%$<#81@q!XhnD%xkr3Tm>6CHgS?BB%vrcqI&)4solks!T>p4f~X~&4y
z8QeZxyz_X=iX9qXmtEl4TJJC=|H8Esoa_6%C)&@Ga^hX?7LdHM@JdA%lj+2&$yMD|
zvwv}a<@>5<+C0ni?4moG{8@aa{j>aa7l;LVuULA;D9dtd)2BrySJ+Oqe_HxyMeh`Q
zQHR_`Oe+<#xJ7+(7dx#q&}!;h_-KWS*3GV<h&8?omwi!uzviTh#x||IZy(<mAMD!t
zY3r|BRr5-9eJ`edVSdJC-8uL8vd!_gSZ@9CkH5aJ{_XR-?>+tt{uTZ;e}imR^@(rY
zjGx*5<UG&(er4Z*fBOzcwf|<G%YKaSO%YGYp3}ZhPwj}GlQl&vb;GCKj&%hW&gHlX
zS$Kc?718?hwBw}+G4WZMyYx(1RX;4WzWmtOCFjLY_Pw^jv+bJuFB)99?y<e5X^VW9
zQ1?9-)m3w1k0&zlPCc{y#!R{Oi9Kie1K96Zac9nUsS!1rV$sLQCY53RbF$B6Gfpvi
zhVaRc?Ot&IJ5V6?v!ZvCm82^7JhqUawM_}%zWtv6*zoj*yXQQmXFd96Y7(|hzC-(Z
z8h7PyY2MvG*3Yg<?{IwcJ8I7SHuse~t`t5HsQ(zfa_wHP2Wfe(_x_9Y8C91rloG6F
zUj2lxZ}tqChg&zkGzj>R_OftsLb{!EU(FGgr<0gs9++Bi9$tH9=7W_r7EhVp#cMEx
ze*C#SJhWWBzI!=`ZLx{N*E`3ixP41cc*Vb6uKR}UI_>4_qfNpN<OT43p83x9eP#FW
z58F?Pw9NhB!KmM&yrJu+w&Hx}M~?AZY(i7hpBedXn#?(^%+ijPtx?#bL0DYPzR5#%
z&$}x5{-StcA!$tq^={isnY`6z@;|$6`&G8BSyba6*~<I9vg+W*soN_hK0o2>P%4vJ
z;;4P-FYA7R_g50`9sd91ovhN64Zgqou9bgGX#c>#fAL(6W$2{&o?V|GXl3_D&it$x
zw$uFZgb8!nguXC^r9NPscka)Y?b&wGx><fpYiixUb_%X2j^~K4soVU6h5t&k<c!rq
zk9*i+P2<x~tP}RE(+WtendLWqqF&907Q1~J_vY=iId^Pb1<$vm`yApFrNx+eza8i*
zD{@R*opb-y4IV!ZwrPhBm?tQd`}Z~MJ9SBAi*nq0?{h+8ZR@)~Kib%FVCiRJmS%^W
zA?vPA>$ubP!zI>d)~i(6I|s}vy;;`@ouAtpnbRD<-(F7bKs5USwfHT~r#bE<80^qo
z*F0PF&4k|?)=aZ&YM7X7Yi2Q7Tg{qf#d0U@Z@b~{be>61Ztp#6;8MpMv*yh5hkZha
zK12jO+UnH*do|aLvj?Oky!p1d9?@e{4O_B?D^CARthwgJ$;Z!q$}LFS_rvbo;~6)e
zgc_6|`n)86a*Fn(GSADYos3T=&6=-oFQ>Ax+bE!0<A3*4tGdYrduHWEZnnzjzM1z!
z#fFD}pS7G@ec6^}+Z<y~7a2Pyo_*Wx|1H|E`FI%zTTQ9R+%?&Pw_h(^#P%;FyTC)V
zv%)t`TgQeY*tg5AU{xi<HvTnFOoZOs7HViaPM<nMb4_=on9cm-hrfLI<*rm>80wl4
z@BD14>2cS~%R;N4eW>{KWMYLv*k2Q|Wd~-<`ky}faY4|d61Ck&FT_<|ShjUtX}B$a
z?dFf!jch@wjVu9Ig+jg_E3ojHlCUBrwJtht!5yxdCFjpiW2p0I@;I?jTJ~9gBU2`S
zx6reNzbt-lxR5?g<Deq1vG(2cH5=GgIh;K(<<-j_@AD>It)H}azK1|skifA8dChyL
z{%On!+m>%<l*_ZM#!%Pd{FR#C`DZR}c#yGV_Ic&Co>SBg3n!@BESTMDcVFb%vu8@N
z`!?8`+Uof|y(|)5zF^}f-wl)XmzUMBIVUW6$!B45%xWIfRQnZqhYQ}Vh~c{Dm2iJ|
zyL<bZOKjOJ5@*uh>R*$6bnvI%g{?n#JrY?f_tr;r@$|K)zC27{vr2<Uyf?F|d78h|
z%1$M_L&p|BSvBd+oTi$Eo1dlxO4m#KR$Z6fnQ;3=>(1phuJ5*J?%ub|_v^;<?@qj3
zec9&zu5V%wTTb@BZ<u@3v?!-lWqoJp0saFKCv{t9JAU+#(%ygg-Q}lk$6kEB9nA8l
z_&oa-wp$yRqO&8CKi}8AQoVV4mHA;dRW}|53-=Y@jlP*3{L%j2zNbbakLh*0v|OH9
zZeDRqQ#8k#3%crBJ&Sie)vKJb$!g)r>fLVAI~*UpI#hP_ci6c{A5Vy0w)ykzm#JIl
z!p51=Ng=n3uWe)FJ{k3WMyuBPbu$+*_GhQ>vzkzscTce~T4-DS&m@yR{_k#WI)%=4
zS)b)X1JAwN!_>H$xBJZ>kMPrnPpr$1_Y`E`KE0Ful-A~MUb9^~u~EyqUW<zb%EgC=
z#|0`sc<}Mm*Nl5Ajn=YeoUs-~EU`OExz9%HGu<_I*m<*FRjwuXdRbXr72i~@3)3?h
zq;!?Go>&;q$UbfM@pgrGT|Mlc#R=lS*H6{ed%R~Fn+tc<j^hRWb*Hx;pFdr|A}e!~
zSp)0Q-sI<ep~ub$6nyP|wP~_tLH!bq#>-&^oOh=7e|WiZ&0L%JqK!P;C&jXP&QOdw
z;IZafKF_CTdygwDp1*>zZ32h$rYqkhN<Yo*pSW%9^erkc)b{=3y(6=q**u3`B%E92
zc6W;SH0##8y8q_BT-vr-O#MxdiTRI3ch>$9maG12o;Tw??<A)ip||F25)lkPtUrNK
zs>(mq+Ume`@1C;9MpKTmpK&{Oc2S$|&F)QEM>oyYY;(%xZ$8i0rtKj3p81%9{Jr)H
zd4+vT9D<VjJ$p+9-p}T2(7bWv-xu2-4`<!+Sz?^FZ=RfdH&f+l#rF%8YhJr{2W*(E
zE3xnPR>!8R)7IFintu!`I+OLY-7$#iwR4S+*hiBaJk7gaKDu=15BDCI9e&bcjfX1z
zdLxxN8J|byURfsIQK2d4eEObF_>cB0H>DCC@?TkO+qX+0KR?sFOeSVVkl5<brGh$K
zQf9vo_&V$5TP?XRTP0T|q~|^JhCseui(imZ+PtTF{}R)yC;dzhpSZ<wQocu@!LP^6
z@~3&9$YocjDV<#Au}PD&ee=bY8cLx@ByCGAoYyVR4h{}9=~vd-_<PF3G`H7Z7e+OD
zUv+yGpdr6d(fQc3;1#XUx4c!?xvJp0VNQ5WlcIGTkKrrcP_0O#4FR`5&v=%V%w3);
z`nj6BO#E5L1&!oe7iROHmRnZ6qW5{w?RU{L&cD04NnG^no?o@||6B7u?6a-@Zpk*U
zVry?n^xk9EF+0SqFG?KnkK0z4xPe3DbnJ2C`n{d!{xR~I_s_Kr5w-p+D)aBO$X)Iu
z_kSchCjNaNup?_`_>P2)RbTf<|5fHV``;_q+<5KYRcT^7qGm=vTqm^cQJ{g0`-2Ld
zM;9xE?{Q?=FeNE6vPiJ*V=BA5_REc*%ihZztljJW#jCFK_rLr<M|9IK>)#ESYyPj|
zhxfzH+qk!;|M)EKY@Z=ITl|U7{MAZFJfCo1ub%8=XCWxpCZM`Pt?cpkkMVch9$w#g
zSXaD4{NC=Hyx*A0w8H${W*m}vCJ<oWzo7pVKc{@ymT<Ypw{HtGuRd~0VN257RnzZ1
z^?7NN_+8|M|GM-bx2&IB(XCyFf=e}&eIv|jRk|J~n^w=<r00B8a^rFl;U!Z1_O8c&
zB>U@LUp}`|zO%qlTBcercEL@dE(S|i&RVzs9m~0FgX0!W`{{p0tbv_d=|)W3KBY3*
zFLtx~jtg3H9B)WZ(3xd;+i6L~RF=a&1<tqhShw#BSUW>)=N9!-SKF!-WADCuJkdBj
zx3^!YPcR^8{RUx$ZxVIWz5S;Z`}4lB*t58|eWU&U3mYnC9DnuGQ>vtB?}Vr(9#Jih
zl%0tyMcJG)S6;YQrL?|l|A$WXvKqzraUQ#~>krixY0vy$%`@Y`!h<ee$qDTxCns&-
z`01^Bv9S1pnvP!Fj)c4Iom+c(Lvttjcb-!Sxw7@`t0(WJ{+LQ%5>8G&>iXGv`T20x
z4Hq?5?_zs;mE-pf_3w7s>x90&v)c834p;xHn;+GeofWIgHgIHq$To?C#ko#RtMEnP
za`~Or&+2Ug!UPveFTOu<jkrmq;>FuPj%H53wb>=M@Rw_C-lcGDshU$c2c8{ppI^W5
z)v?;T1Kuri>`ZUs<5bUok=XkAZGUwuuV)<dmsD^5#S7Rr8VVfUx!TSldq0=ZBL4Yz
zZyx&}RJ7dZ$Rr!Loicw}_bVqzZ`vb2G5F?X8&1LYJx5d?Ca{!sCLHgbW}BM*^tkrx
z@amkH%?4f!v!1Sdu(ICa$@)`SX1_yqZ~RqRu%ks_YJ74|ZEbR@b|=$xtw#U$Gi4g*
z?3=p#*F5odX==E7<#nDYliXE@e4UuO7sjE-|82Ify)E0)9>P7*(j)xj^#ymb1jW8=
z6Mb3rw=X2|FWau-mwyhG+Q`{g-}&&9eXsW7qsiI)sSQe>#9p$jPRc3EU(fqJY%TYE
z<0Lk@{o==#a94lPFrU@$GBZ$V-uYt-x7fDcy!n&WDP4W<x9;khOP<}c<SsGqJb3HO
zv4qKI%&pAgo2Gcj%(3CTa;opfvIBpc>T3(`Z4sL7`}1?R>eZXoD;~XmHbHdfp|WPh
z*RAvKf7`V)xc$+e)=3`@?CJeuX2baKy4gG4+p`t6v6-dE`ED{yt(&{Y#k5$dOHQnA
zsj{oa*;7uQ`Zd8{eodR-TNchWXWP0teD7=;zkT90U~JsH^wh^S5)ohjn;E&)1pll&
zGBfpFG;^)lJMJGVcCEQELwkw;x5%BPvwf@D{HCwG6}V^df-etcC6WRsyBdD7leppW
zBTbcO%cR|Bny21q?>wZp)6z4^r*B_Cc*^2Lrb0ViL)ZILraVcx{X{h0{i%&l@}tF%
zlO8S3J(l?T=8WTuL*?c#*kh#moqOSA+bo+|`kbxeTIWr+9(0*uzGH<%s`<*Qsit$7
z+qe6=A6&TZ=tqr*3I0WitDa_k2yMFclS@va;PayCizW9Zv2y#Qy8p{_TxFU0tLM4h
ztzr-9P~E(9UHrF~ajoLZ&0M)M+wOIBo`hV}{h3Y0_qA)b=k^&m|5E#;EqA2-exuhB
z=S@@3Wla10YQwZmC5jsbV{a~QYE)mkjy+>p_tQTXcF%5pi#S?YIdkfq{Cg|z%#N5N
zGI{EhDVkqy{6EiPrOleE5SkO2m6@7%@|}(2wtBP1b9Y-m31&XJ#StDqS>&zI#el<#
zb4$#uD*KGuO3ePxJ9OlMkC4#<?(TJmJ^p2#H+|zJP(7*RmGK|RPgmxt%qX3@zV`5;
zjz#5$!7f?_mYEm+g*QFj*yrDH(c^;uBe|1(4;K81czpC6Zy!%=*ST=+0RD4AWo=KT
z!rpHx`;xKtY+BQq_aE%sYK?zgGI#fjt-N>8r0I2!n8fWO$u_g?7IoZ<v^%5|G@2I9
zF_?U$dfq1OCu|4j7i|1=YNhn5)yC&0?c4N!3+ww|CM~A&D|7#R_;TveT(-pe%MY8_
z{#8G2(mh%A=T?W$MgQfO(_iM6y(qQrFe>-x)px39SvS+_%!F^bJ=58Gr3Cj`JwIUe
zF(ENGcXpSxrz~gQjR(>a&%Pd8d~|_`g(++C`afUFFBaWC=wQKmc$Vs=WX?(jiNqyR
zEbHcMi4^qWUEJ#%Jz+}X<e65&s-JGOPK@#^k+IA;Y|H$t{qrSe(GAM~yynO~*!7}U
z)qVSQlWk$L_79`C*}AV-nXhN|(cJ3!xjDiH*B{pkRwu_>S+c%N3jL@vUGaR<4!dc3
zf0AFk3)*u<F-%PI)1tr2BKAB8Ulo0B-{Y;PYIkk>wuIq(z4_!LtJb^!U;NH!*ELO_
z<hN&@O_DIy*L-zm`I<BLXIm~kk`(2gAJg0O@5wB^Khj<|SDw9;?Yd`Qx-8!$<(I!*
zO4xgpt8V?zP2mwSEstFBe|uA({t3_f=ci9yFe^z=>&fvF&p5w_qVqhSeUhjYF;u8e
zQH@v4oq6PylG7K#J+obT!mngs7QB@DR$wEqSrXr?7n6mWWw*>wy!<?J#@%FZ_cvOO
zsx_4pH*WH1S*dV&X{F`Q=atr<pWa(5o1JJqk9G63!_yKsupXTD(4hOXLAl*_(XGDn
zC;K<w(6U+Pe>KbaWMtc8r!xl}e+egbUA(k!-_16cz~Z*GG0j^%wqD-kR>8l1!OqIV
z3uip5(Rux5Ys5wE`|IZ<9@x5Ee(r?%n-@=8e^yiLytw7x;2U2>-v7F*(9^Qq#C)}i
z`SJ%ns(ca2ll;BAv=TgLpWMgg_GH`igAPJofvHOq#9n)<topAk)Lq@?x_0q(SN8L@
z^ZhSL_Pk6=^efGr{3;>EHz31f&i?67zQ4SBIo<!QutLPQW!|5Ar%9Y-T@kuXZIgf5
z_DO#eY*z+8-{RtLGh_at1B^e+V*}>Qja{+I$LXU}zv$$qbYY>2N!Jsq4*fsz;Fw&O
zpSSBSlh6{*RRSsO&vr`HDg71Q!8Q5NC5Bw4Q>Q+eUXZDj>Arvd%KIf7PbWWX*Ps3V
zT6mgjI5+3aL{90@e_tOi<A0rf+b_qW|91kr%_aHoqA_u`_Y?D3e#hm0WqG#Iv{EDB
zNc_Gx;c*cyXO|gFe$3q+le^yY(vugL!`|FEnf@ZMY}tueeI7pdLYvGL?Y|Pk{#^cR
zzE@^p_^IwG6|a+@&F}v<XZ5Y=tt{u}Us>d*{>I~^k5|ajCr75Ja__XBs27-;{?qo5
z<Fnkk?Li`6r#)qF(>s18snSz%&3YZJw3fPS+D^H@m2UqFvQPLWZz9Fp_D06J?%zWH
z+}x#mzD?Sbl<eqNw((-wf&YwK6Ms&{HkgWZG!UK}oG^xJ4K{@B`;l|H|6F~%_*O;+
z6DAI9Yx2#IR{fJYRBL2zi854M8eSbC94<DWVP=ExN~2@$%kItGx2BHi(ds5$t<ahh
z#<g85O0F^}GCDDNDzGpvUU1^fhoE`fjGh%N8=4$huY|pgDqUN1&2CZI^#yCUUfsHO
zTj~2ZasO|47G&?go3H-a_t_om;`8tSoO?g7+P&rH&qSq(8AWGqJ$=3H$hof0-m96V
zhQ}-u74u)4XwF*~`ltJkNf=i}!^bHd>a`WOrq6$I+|xB<;g#>puDm}fR`PNFulk!u
zFB|ww`P=mT)eMHEm!`K`zs$ZAZp+KQbbaCu50f~vw@=+)?4RSZ^u?qptAnRRCC|EO
zV{W38^Uo-DTIBp?LDDaGCMJGvvc1mn>uta<S=XYrb15f07K`2z`87Ad%4|yC<(Vzl
zOsd-aywx?Ya`|el^<HYOVSdq)ZRtIS)n9ha(K&o3psLyGti$Y*Up!Giu4|imwK~pp
zdY%zHrRvs-uuIyHujelaf3d~QN=-L%rK!ft2?DdsG^TF3m2oD@q<Wj7yXdORtJ9Wq
zotH?<J|<^$=Gz|eO-a6Y+9u4ndV100Ywqn*pL<GH`S~x`=<PD;Rs0gUXGXz<8{Icb
z-jwEunMp@?=pMayENSD`8@q2*<#2D~z1^p|K}Cnx`#4VoU)ramF7wKlB?Q;BSlgdz
zu@za`z3*$~>E|oni`+M_{r|-L>iH+-F*boUJm(Lr_dfG4sDA2yeMVNs@5k4+1vajE
zy{&tGj?l-Ye>W|g^y^Soz{cZeL?>u8JU?>0OmxcXlV57wHg-EP`ky-5TIq9BZq<q<
z70>y0?%KUZ^iRk3(xTe-GgGtrvUcjWUH&+YHOVwt&v*K2tC?mC_ix?($Ds9IW2e!R
z=6V+CBaPKD+A$wmv|l{&RqEa@H}Pwx<>qZ07A=x+Xnpee$L)+?*TTQ)dTyR=J)8fk
z+;fK8(HBjMoc3>HTCgJM%bjBhmWh%7PM(VApMU?pDNo1J$hXd^f1fB;SI&R%Ts=~5
z_VhT0j#aXmT-#>cnefWO=0~XcoDf%5y`qDD=QW#dzU$on<ICGkhi5(&Yd98ez}GI*
z=6UP8<^9)ewQqa(mRZKATb})!@lut)<jwnc*9&)^7My6`zffX}%bLV)BW}mUD5c5X
zV(HUUR>`D(oig|5$}Kgs?W>#@J^WV@#@ut<PRr6$?#-iL|F0YWiRjo>qkiJyvhZoV
zKM$L$+&xoJ!~epgXJyD_HQmXx5^vx3Ik><`&p5d6kcuYnEIVZnpY&~0PX$g<>Atu~
zGxfk@>#GO1+-ux-POz)l`-p9hbGdRxP5q4Q&rHi%wmp;#zYuCZ=jpi$7n>)S74)j(
z&rW^vR>$}aoAviuBJ~Q&;jwbEuPS)d16uyK_gKu#)_$R~`;Bh9#-oQ@r)%AF4lK;L
z74zlK-*837_{#s!(+s8^alBp|F0f@+<U@1A2^x=Hx8$T1{d71g{-9R%>&XPmdFm{S
zU!~1jX=lQ@Ju~R_RFzLNC3{jt0vpw8%s<O7@7uDioKeXyJTA9z()@?Ie2s=*)aIxw
z^W}R*$?UbNd-80qzh2D0ih|E+3NJ*hV*6Q@!tZ=}n*3w(CV#PMvz3*EpEtZMR;j#Y
zH1*WMh?TofNVd&3mJH3>o^UYj&aJawzor+ZY4KPsl)v{z;Eax5zf+z22k+&pWfpAM
z<apitmetz!6%IFSv>q8hsJwYGWu3qR?}|jD$x}=Ym;`+gx@U1fcImN%In9q0X0Uwx
znfFL(|JL|tg$zy)PDH%m;NUoQC61+kLU6$UU@aC0<~Rot!8V3??bjE#r>DI5I;*Fw
zHb1|*`u$~%#j%d}ci%5weW(0R%jET<$F3JvI`D4&p)bv@d`(^YGfzXr)pJv>ET6!1
z-b(RX-m*?5ySL7MTNA41a(=V3kXE;vZOfC<I^n7T)BftGA3k(i{dH@&WTYAwVe8cP
zz-P}%X6*})9{K8Blr&#c^qb42MyL6B?_3SOqmr91Dp=*6ZYwmnc8{^Y`GeF=sj^m8
z#++NJ2Kq|(O`_U$ubi9wX07oj`_1($C0WF!Ww*vp3g?@elegg2_D!d+ojNX}Y4y5I
z^w5i!0ou(jANoCdjvO==XjHSC!CJfUeKY$6hKIjXxv#mdkJ<U|)!U2Q;(Gma;x}J7
z?Yo2fRqUeocQ0skC`|94ayR;yEEj{vZ=L7t^EHa<-G8*CbDd~?C~*AY4*mD{mMf{{
zdiKp_Z(8`z?%!DlA0^R_4>1<^j_&%A5Pm_je67Np?7T<o|F^SzU@YhR*}<LtdZySW
zw){-p?cuZLh)FN~ec@)(+toV_xBlX>Hd5h#o{|wc@mLIp$n~RP+=nGSxbx2G8V9&M
zkz2QA6N|<4uFoG_E8i|J3*PO&QF-~(^vLkH20s?;+gds0Pqh2NTT!dhxOZDQzT>pN
zSIpLXuczQJt9kMr?>nMSfd@O*>(;#HH(K7Ot95Bpw9fA?FNQ*P<BiAtVv_jZYzyA~
z=GnAERrQYq%N}-o-4p!p?4v*x(JZZb(^W5AQN3{GkITgWXPW-)xN7Xray9S>a}ImY
z{wDMCjaBSdlhTT(7ClNoon^LJl{qD1_vhayCieNWYlv{ZKe<g)=#V2%&;AzS?HeyH
zmfl|Z+<T|@%ZVm)9=3h1S6KF;wWeWB?(+|Oj+Ao?2{|loS+MtV+p*52X}L3JDaovi
z_2K@(V>$hC+S&<W2Nyl~m)=<u>!ED;@!QLMg)qjBr);x#E<T+0ma(bEI;LPR_pjY9
z@^^kba(mqSO?uxRF}-QAVY2#J44%x9zm?Zcyry}|s`R|!&kq+qp5AYE{(4yBn^SE<
z)|FE>-`3EWyv2ZT>cSlcyCp0pnP1oRP>I`=@$b~bsoQx(zMK{+f3r8Q&U@aBJ%uyt
zs#9lk<-FIjJ!JB8yV040B1WC25}GQLpA_EOw{6z1by5>-3)X$G;XGdZ@oj9r@4B*=
zF^=<hC9Ld@IIVurbJOvRO}cxze71LOUcLEMPEpgguQ&2JPu~{rWlLsFeWc~{H~Gdo
zj*zWWN>>(lm5WbrN^>>JR_lD7;dzcJ^u+pQ)vZ?&I<udLRnD@~PWrv<*7=?Vx##ay
z@y<LV!4uLU9+Ycby@D~rZ;sY$m4{9D*85i~@^XfLEj3@#VmXyT{WRaEGFPr-gT{)q
z&+lT2M6GKkG+7!%e44Eh@Zoud^yPUvJvl6=#k}{Zo>)9bT0?$*VQA&sTW1%(w|UvH
zOuC>|`^AmLE8Gn(%6`?j<NM>oPJ^g2*TOtseFc-{ziywOH*NQ#b&D9jIF&eUxwP?c
z+WU;RYj2k)#G7s?k5AyAAl`0za+gx?Jr_x%-lZ>7d>jkg!iAkJ0&Bk?nf2o<=j%hE
ztFz`rhJ==e`cB@XeSF#3YtNUTs}x{YKd^H9yStx7_~RZ5eXO1@mTla(?9i#>2BCj;
zt3EQdiJvh0*skTh3$HC~h>%do)7&0be%e-A)ZlMqZ|1Ru5-TsVgy@}<U%h?a*5?b$
zm(1JqN~UiY-(mN_9Tx2;?=PNMU3D-*|EoyH2H6i!wSTToUi$ZY-}6Pei3PKF^llFO
zULo_Gd++|UO=>dxTH20i6ty3@eeL9ypa3JY=*<?wTbbP+O=xp|q?UiXe$U+E!)yoV
zzg+UB?}*tnbLI}$RH2HQ;T^AM-S}JjiP3gC^M&voMc3bnWbIUGKP_i*Y}@Sv3MSqc
zo~m<gOl6Nwny9{=d$#qD@;j9azHx}Y_{k-4e)6T4hy3KCpWM9a&Jy|R#|JBkxbVpK
z=E6GZl1mI>zvg7m`Mpzko7cJGkgXwM-b=$zTjuRg*m|wN^wRU6+dBVG_$!_Ce^(Wj
z2D%y2wpKh1bYt|j6bWJ}DG>*X96vfXX)uVfFwekdHqtr`lAGtoW++=`*M?OGNSDgp
z7bx?--63{n?z>xebI%8@k9e>uJ=*qp&erprL!wsg70YC1;porta1v7BkE=YOrt)yA
z0F!5<nnI&fd_k6D^tofNZiK9jd!KgBUH0|a%EuLp|Ap1h{b+XK+P$-1|LwXvv$m#w
z|K0n~KmV<fw`ZI*ql1H`X})|#-u-8>`b+G0XNxPR*G=78)*cn4sNKVFd))SbZn~?A
zyONCRgASHHu_A4uw@QpZPx>Vu>|ez4qs2o)ST)Z4;EnI{e-hlMD|Z~IDF0)%i$_Ov
z?UVV_VuWoa%z3)P6uViPs*juwulxJA|Nq8Y%aoQm>{{%h6@L1P>1(4+o8)R|1zyoe
zRb84@6&rR<ng95D>6d3Ezxa73S}qQqdOD+XPshEPj_*$`*et1=KlLKNVbLy!D6N$m
z%Vs87h3}fbF3fYP)-IQmA=)AP&nzhRPSuJ!t#LZ@-?}Mb9;#ZIg0Z^#le98RS5CRN
zEk9px@6?H_U#6$F2JQ6nS|96uS!?AKz0lR7{po3|R(l&sq<W?q+PhDU+-TrARq}t{
z+iQ1k?#!Nl_v4poZ}vtTU9mEaJ)`kbX=Rks>VVWcPiCLh&{h@YsJCfYqCE42Rf_GW
z=oI5kvNv1bxD`!^d14UMaAJa&OJ7Dyhk~84sHg9P`9gxmpUfTCKA6ubrl>39=i|e(
zxHUtfnrU{-mnVyoJ#Tu3b_X9=mQpMd)5qm-$yKqO$ws4W8{a93*(*17UWxo#y|raq
z>4CcsANaq@z0hn`yJj&j|CP$@t;=?~Nms7fb7rNo!X~D5jUm5$&RG}NTim_sIx~iu
zb-i#pn_rdnihmyGZ}Q3HF=!txxL7q~%kO~CM-?3GnY_>Ye|hqJvN@wpf8(ZXr*|(>
zH^yI&55J$YE%KyPt|YTC$6K~<)(St4o6rC9b1fVH)tt<Xpod)zh899HR^M(N4EA?6
z^l4;SQq<JCSTyC~$;h%6iTNA9`vmS?=e>|)-v%~^pQ%9=^V0vR`)}wt?fzD;{dRA=
z<TVewzL%molEP_wuZXC%W?%d-oN2=qRN;`qV{-i6fof&$35vV-&hJW?e(%i}|Dbxg
z?EL~q?%%NAE2;O`bNQiA%f_Yv(VvgLm{|YJpJL9`|6}IVW$}}z-kAFFd)!RrV{BQg
zSEQ`yE=msEprbY8pG?iqofCGSm@6`Wf~%Qb#+|Ya6La*6Szi}V+LruFYhF>u@;~l5
z{Ck=XxkeqG)!cG2soW&cZ~Nl2?foluWj&pxRryck^S7tL;cdH5+`AlmS@yE{0sil`
zU6t)OJ?>OrnLI=LPTi%KzE-cwe)bEW3s^mMQ<Sb-%-wM7f)WD@w|heBtTDY?j|IhU
z{}LF{YsLL@Ph4P}?~Ev`J$Kg}Hg;NCchT~Tk-+@J_s>h;-@4u*bA_sZ<NZ618#8&9
z#iR$WZp_{I=vLff`!{kDMUx*aJQ?n2cA%rMe5;ne{JmVQ#X+&pthycB<vXVA`|~~c
zL3*R~3BCGhOg$R|biVq}`6y=4eA3G<JUSxI^TpS<jo;=a%Q|L#(Qkf!@t!c>cKP|X
zH97ZJ*%lN=P7by<dh+Vi&J+A>%l|MgI)6Ffc1CW_)<Ee5#w^2Iv$p>D9K3XWtu6bs
zo=GdeD83W?7&v!*RN&f>*Hx<y9}KwfA;W(CNV}J+y!t9>fhl5>4|-*txux6A`MGhM
zp?G%rw*ZF6e0Sb)R_LcqxnFSfxkPQPaMs%7*=K%CD;M1y@6WL3d)lwU{$nlMf8Gs9
zR^*p3QN7x7^Vz#kWv#cELyjL1@u=jj_pVa<9gyf-XlS`q|6Q2i($HC(q*kpvv|+no
z|7O{|&n>q-<D>Jh_*cBy>boo~+go@mYu`P+rq%C5`EUMz_&|NJ_@qAe<LT!F82ytY
z_p#_*^wucubzOh(vs=pLcpV3xtOLKi72XO}_-Ox_QOvnB`9VX)o++jicixDc(6+Cy
z^4QJ7djbB2H$LVEZ13N?`=j07_|;k0%%V=-Km7aR)pDtb{+TH`nJ+C~FFcpH(}4Hv
zQJ<xs+w!s`vO~nvXMD)ozcGKq-t&)D(hvN75&r+`ukbU*$NGW|B;N=euisoHckAq{
zX;RE-`#xDXc>H<Ww{!V}pdBph8bXfUnbz+hw1fMBLhqet0?v+NK7ZcUebwHq@KH*j
zreItB`9vd0t;Gt(|LpG?eqh$0BJ9(#{@LEYK5>GJQ%*>h?mO(^F~w(U<E5$6^VY{-
zIQWpWd;)Xc``3<23E@#Wx_g9d_Pnmz@?`k}hL<z+k1Su+<bSp1jpzraTS@X$&j~U7
z6kEe+vh`|VJJYkv1vOtEz4|ffui_7%nvXM2Jey^kbI$IQYI3B;e2?%$Ca;|=D*AV<
z>^%4GrNKqX4Mu5pY+tUJNjR;o+IF-o+M?x0)c3jywai~HUObUJRJ~`ZmBUQkWooMH
zUahyjB>r*E&a2vW;;gq9JFR0|BNXrZ`t)ms?`5Lf`5(`Hdsxu^rnS+@GC|wLYyNRu
zy_2&#|3K>DY3m$Lloy%h|4H1jO2vE8yER;yCnj9Wuf9Ix%cP&ZY?^V|8q=>%*foi1
z_8fy}n;coMUVHZT>*Kx44tmvQKIVS;pz)f=58vu1I+a!?e?GnZIj_i&YtvSyfD+FW
zruR<BPTjh2S?8I{GLvMKzDSkdIBPY>?v?I^-nxwsUbZfhV6=Gkrqld#U9Erbp2&+g
zoWu^BW*qas6mxLfVZ)ODC!O;fTgn}aeu({cPujlSp!t?hmhP{+R)-Q^f7Q|X&L=S^
z|Egb&$L++FSu>xBeKNSyvg)jA!PVfFHJT<ymkJ-5MEHC(y#4U#d9ikGW3B}e!r?k^
zzI?Mf`T5_$f(aSrvwYREn|Dr|v~I=i?@Id;x25FeoU#5SwDPT7kiU?6o0GF`NWr1v
zYrX1+)%7@cy#27R^-9b8yJwX7S?|q0f8po1&!JO#FMBX@UO%S4Y~QK3CSCiVe7OB^
z*3>HPmA8AGChlSRmn6G!|I?EH8PAWrSRQgc&|jrtzP7})CHpz1Yd*jDNo(f`$@mN<
zpNhF(zw&Nd#-<v&=T?D%{P{frcIKj%j|wKOHjS3Q?R9159Xn&uFB_Y6=1IvuI-Rrl
z$h5h}p(S!U|0I^g**d8m|Gx2iL5oK3{br^9<KGt>EZsNz_CqHBs#|X)cB&Xo7t~F2
z^<M0q<h{K1+mDA^%FB1SEGjNre){14Z8!hEU9@|L^7lIRf8ug%>N(h&W=M-%Noe*x
zNDv4~3CZ}H5b@v`3xnW1riqwl8zBwKhEx`$>iZ{U1()RN`zaV2nd`e{78RE$7#f&b
z>f^T8#K;Wgq?WVMx%szGZTxpHexpNSAb0DE+WM2W91o^HPTubM=GI3~DMxvU&a##h
zixf|uJo)$Q_nE4yK_)Y!{O(-qa43pTOG}?ObFQNKB;#{S5}jDpKA!hjW)d7{m^xFi
zP4oDgjXZbdbfY#cki9(Pz~1y7qDm~wm^nN;Iejt~ORV6~F1YGr(O!1Cg}>sPPf@{T
zAFeHtyM7&zahY(<D1Bx~0^_@&=I!qL`T8tvqzlR`%w{~{S?6f<e8!eZ#@#=|s?%Od
z&K1j%nSLh8(XWb)bBS@AW!Co$ai6M#GoAN`&#OH1{lWW{wi#`TMhTYF*0Fs#XrHjD
z)ykqhj%i(h@G=d)B)<n;r4?VeA4-@kZfW`>(ayT*z|5c+<Bl~VFIal6NAO2zF22}T
zIgMS+;e6q_4Jl6*?M;^1Sa}}4GwGb=WhcjzF$Y98Oh2HpR;2BaL{za!M#S-8{wq1W
zJS>e{i%p^p-zF;YGqX86%NWZzRU4V+a4>V8&|4to?4i)@<^6`qnCptDRmcLt7Z(Hg
z#WJ@nW;@u!XwG94)}Wjb`65(+OXiwJg8ySr7K6_-^jYpYhz4xnoHK!?km+R8M8?x%
zOj^$jtTwRCZZw(c@<>u_q5$`!2wsbWkB&rpG&KahIB<mFT>~GxM1yZaLrcTPxorLk
zdR(pxxE^S8lpdNS;4qgZ=HS_AzX~=zn@sP6bxU}RX0%0faLDNxXe~;8pvWZA;>uz=
z@ye`a6IpF|)K6$uEO<I&g(HKr)+KHpmwODY3=<Ab%3}JdG)tx3pW{Ka2WQ!#pxJD4
z94rUyI1b8hC~&gYWLc8Am+3;IDC>eocTXurr7cO$vzBh`%eauxWS1FQ+@a|yIfMPl
zq_72(;_D=3o^7rvteoCwyhATty!mm>KBwP?I)OIMwMTuQNWb@3_voC!^9uKF&37^%
zkKH{zz4^1N-M%wnhY$X|`0+vRx8qp_RkQD<v>$!CV8@)kxP5<wBli6*o6h^~`R5wb
z9ee)oShPI+^3Uf79rw*h=~n#sZ1KU94@wFwM2;S|TNcY#cl!Rb=Eb%Ps}m|G_J%vJ
zP`JoqmMF~S9FZrabj3AktF~m>y1h|T)xVlv3xAoo_tHVb*2T-$><n({`uII0#+<b!
z=+n-x=OU9ntY25T;ofig)A7saZ+`t!=WuS3`ObH|eQh-bHnF{CW`=5vE`q8xk50%p
zds-Y7T`#=jz~mOjGuIcz#Ll^t|Ne4T>Gm2!i90>VPoDi_cT(KExyEkxxy_z-aY3IS
zX#AdZ_S5d1i}t&teLw#+`nhG-+`pAp^Hb`-tpB__wBBm}IrD^~CwhFFlJbj`IxikO
z#Zvh7qs1zpaLI2g>pr>M2=iEZA=vTfa^@XLpKLYk^~0hc6i)Y_e!50a{bWUQ!jT&-
z$+IWja(78$*PnOt)IIyUrMYY?LSkk<T6yYU=DCkwUtee8GuXek``^Yrr;pWsYGjNl
z3MkPrUc4wi=Ss^Og^M0eOF}!3PU6kIt*3Hl=YOuB8sa7!mMmQxsVa4Q_ZBt7OSVV-
zZ-3Zbw(x()-38P7FNYfaHjC(UeUTo2DeI-zER}$=xgIKJr*32{cp#j8DEPr`v31uL
ztX0^%TIyZiy7yw{@`e_%Yd0O8!}PIaPVAqu0KN*-y>r)1J8XOJw(ur%sXd<gn}2lV
zGXMGJa{kk!w(n~+-d#}Zw!Z0X*1wz0aA8&TE~Pg2C$|fm-rd}B%Jxi;_ig9+E39fa
zE;yb3^3BQh)R!2=mbW{d*sd1`w%Fb;4$PH!=fk~l%l7hEL9gm{5B_g_&h_u{Jnd;V
zFMlSTj65%XW?gmMHI09@R*`(UDQ;K8_8bWfTpP+KeM;%m{{PR}SIwBRiL>a|&%Zfl
zS>F_$eFT_J+a2BhYfgK3?k}Ic{#LVJ@cM?ncUm_`VPf9vAg5&c+_#esslNT9vD4Y;
zj+UKJPs-(`+m~E-y0R>D_f_!;%{#tTo~nK_;ccx@W#dg9-~7(Z1)eW&|L$0JNp_RP
zJeJG%ma9CP75jarZlum}sascjZfReXHkIDI);HMr)i$$0C#4BfY7g8B+!mv|vaJ7{
zyz=c5)6?JnmDPXe<vT0zU{inPznQ7^DUYkY)>ie3%vHKpx$DF;?Ny<jSEX5awU$e(
z+)wMi{nBRc#@8igzi0H>n`@kGpK?z8rnuGx^AP6zAlDVU_GbqK#ZGbRos<!@sAT$H
zUvYCulS>OSvP9(1)y&qj*OxX<dRk-?s}uf3t?|Y_uhqW~|B;*WYmx7%q~4`dR-JtP
z#(UcjqqF<WZtUFEu*&U#31`BEuDAOW{%+0|dU4w5;M%MSpZQOUZh9$|dE4vP1J=W0
zPNmNEMzIOnr`6tAoL18n<ByH;>Fv&bTJ)ny_SyFG2b&T!a}Q_w{N7l8Z&^3{F^}C6
zH_d{LJ~qiNIsI}*?{%F3M};&0<5Lc;OMiJ*C`LWB-nT`qB!`z{7aL1Bds7~3(~6Y`
z?S(iV{|d^U=e&Ng)r5kX@$2^`vAvlVo4)$v@|S;>IhKB!kfoTkqF5ld;YR!I`e?uD
zI{VX>e}0-(Xy3OhtKg*7jwzeg-#p>xQsed9TwvbcdFS43KJoU&rw>iFGnY+Wx#+7@
z&w6Xg?Z2!h>)&{(E3{s8<BH5)M%m{#A6doUS@yZ`+JjFU+gbt?7sg!pzHq@Su8N!h
zfv*o@{Gu<v=;4mIp|g9bWcrCy@$-{wU!Kj~_2>SjhnKebI&3r0GF|aH#@hK9qgScM
zhD?ctyEtcuU0xx+^p3CBo=g9tRs{Xswyy5#s(<YF%9*~$e>=bJ%-M5Ufr}3AJGS@E
zd9P=y&fUqKaeLP_xt0T`e;=LxCVKtb)SG6FNw-&@)pso@6?`?(L+r5LrqsIq5{KJU
z9#(C<XX*YdUQ&6=<!9P9`t3>~J2NNBn{%|f3rz~NU%Y5n)G_fpS3h0fJUzQo+o&}v
z`IF4lQz^D9*Rcfdau!hB#u^y9ZEe?|=$+GASAK}7I@@~ku6WoK=2KhN`JC4Cys}Lt
zYRh@^{<-V3Kg(-YO-wV6)ePO98C;$fY@U*lRV>TjrnatcU$(7E{ukww(pMGo&7b5-
z$NrDlwa)Q+Ej!Dq*vYIC-BFE_pKYUClDuA~Wc#es-6fW9QZd(e*-OWN?lr}UjXvT}
zewiLt@(%x*^gi^LS@rpSe&O?PsBef&xc&O#?WU7|Eu!D-J-<S{l<#k6oFRAN2aVX&
zYIl7VamIOXS}(2${eS9cyv2X-^$89itL)WSQhS}>&v>jA_k40r!PYb9-i3eJRUR2~
z{maX~wzbjb2m7w=4GaxAnz+j={iiRFS6#=Z(j}r3GTd)5swJgV)EK38WR<5py})L=
zNvclG{hHE*j}s3I<hz~YnLAlI{LJT?7Xp%<3&Nd$ZE;?Ea8b*(#SaQOmEQz>3_Q_v
zG9y``%71fZZb?Oz%lRt@**&Z`-}rs|<Wu{5{~xbk9~b}RMs&;T_!RT+n@b!I>+jbM
zzxnp%AFF<WJu{US{@S+P{-nLqV~fS#{cm`F_fuy$SJE2$=+nQvy-SO4wK~YB-mfs+
z+Z_40-Qv>qi`Htk@_V}DUu{#|X`8%ZzszynHfb&S$F=b+#yeU<_8UK3_;vT>oT|?&
zB{u(m`s{8A(-nz#N-LQUMaC<yeC_Qozdhv7ku<mF;D~Iid-Ly~F*?S6-0=Li1;2Br
zIZpdrGEaYV{EuJfKdjw9{pl?Z<-5NY3m*4i5fzKev$>G5ZPELJZ2s`Hj4g7RPOUu>
zjt8zb3Uk;nH3nBbVNl;v;W+(<=e6%lY2_Q)wV6U>&d9gFN&B~*{rK|d`*V*>KcDgM
z5qI$2KffkE+g|$l^`$@8a%SGYc4qyyGyQiZpDjN>uhGuJ`@8Yk`z!a)`Z@KM+5fcv
z4oeJf^-htyv^;@3K`Em_dam$-1CF{3tp|mf+7G2==+rtMPnNjET6pBQg|H;^1mzdP
z2C^#zM0|Z1s-8^fu<SjNBYUaZe_|_#@}`P6hgDsl<#Y&D$g49Xf61Mpk*y%oda>ZF
zwCCFi=e#p2g0G4xUF>k(a8xfWuC~xM<Itgkt%Y~|Bx?UZ`*QeXo$qn>pE;jo7`FO0
z{E}zXD7n-c(0wD+dBx_I3kO;XORhPTDuhI7^1N?Z`Jd0{c4)-$6}OLN+(@{>D=T~Z
zz&hqRR#qNKKRr$_T>M#PnqU2nAH^SM@07??IQa91sJUIM?K#&!myRYEYkNevO3N7e
zoS(8>**NbB%Ylb4O!yhEu^->FV6kM2;mK>a6%%}fU%L0aSY<i4m2+{!Ug3k&V&<g<
z1iNUvgnAw8@I0*SnlbZ&SCP_04fFE>dWtzsZ~Z)McUx6|es$OF?4I};A0;<L?eku@
zVb#<AgsV5sC9K-MfGhiJ>VuCp9;f4OiCaB9#dSn7c+s_w>a5rFLa!ODJ}Q>Bbyf1T
zbNq@AG$bGUwdieNN)aq+<aaT(`X*X`?$sjh3tyGW0!7MFx4U+&d(j~N+|~M$lTz0U
z>6;fEEv%%R*-t1bl*;&B=9RUZ^yTAy;UzBvCLX`=)nt*=gk=*|3xyPdmLw<H1=uV*
z>B!SL&25&0LU_qHKQU(KjT1~8CDJ&*1i2|muk@Z*()-;=&U;aDLDA)aIPQFns|kCQ
z7G7-2X>Dm+)?$7mphGx@^-=$l>#d7iZ}1q!xZNx-T6nnIJUsk;c3J(tmL(c`ukBLm
zH557T+KIffTT=1Tc~)QJ-zP0gg6m6<8{Y1FoYj~7;)%?iSu*W<RhJk2{I%_$bmr|_
zIVsVfuL)?k&5gAyD*qENSb3(uCZksR?}-WgrtC*W(qBi`_g7AQb!7G3dFtjRZkJr`
zcC=O8Jt<uG_wS2WPgtha%sv14d1*^pVwzm|jDsJw-6sTXvt;UBDVkQ3Z7jHEebP)}
z{`I%Y><jLHY`*Z4J@(e>-3QAe(_HTIDhSmH{y+M*|IZ@xKi!H?cw=Md?OJ4hB`S7x
z#+LauZi^Nu`4+Xf@#`3_c_+WM&Z03>v@I)W*E^O!`~Sav^5bk}rq%zR9p-1duDp;x
zxOem29p=$H-1n_l>95q-vbJ9Bo%NLP=Dj!d+<R9ZytkOkMD55Vk(Jv+8$YS8?<$d=
zZQWY!-~RNX^HNPSednEQb0d!{h%EeZ^plAAk{=S{|L-%%<}80{#=|UL(y&{@*Ew*O
z(sa&c8orJ*yn?+>a%G*^p7TW9J6PrC2i3J4JLRq}ce8Z#G0ADacUI+*|BI!kqLSNJ
z-*S4Q*wV3D)wAWomSYiXkEL@xbUM=0o^kodT!EUeuV3?re|_?GrdHVHH4$g~L)SQ!
zhpq9tCirG$Qm^~5$AYyf?e@Eqa>Em4kKGns&6(@zv-RWU*`m9S_-VN(&sx~st-Gmr
z(fRP-+x690@69#Ixz!~hv3}hiyTr@&rFA7AAC|t{w?6DVyX<Wpg}0BnJf4T@e&Y;x
zVqd%UM*H$}w(-i}7K#*i{nKZWSuJhx$ojxbz6M#THHydcuJ?vry?@-ch3mdk&n>5(
zHO$L>KGjT2`o$EzP(qUHl4@L%$f|<J<;PF9e_q{u{r##h&u+$<+(?_4@^-(fIZtZd
z3uV7qI}>kLvwd9hca!tP(j7VB-q*ISsZU<(UKjgv{nz{Ul7Fn^pO>g424riSn0qR^
z$k{&=l|O&)S@>s7xAcykb?k<_?ghQqYOB_8yW!Y#ZTp1Xrv&6qDE#CzD%lf$E41s;
z=V#j_lGqJ*o%%BOX#oGz3Dzer2CBW)Ip*BpwY6E|ZRG27K{iL0U$OSMIIH-LR!h#t
zj+>4h-~MK&Yrb*Te8U&CrTu8^(g5AKO?>nJ^ItZe+{cxgmr`6(l$w|eV&x}gaT%If
zAhs)m2Sg%*Qq%N<^7Bg+Ktm-#`kr}d`3gn~1`0v?o-WZg28o7-CI-nCrm04WX3423
zmMI3NNoHy07Aa;X$wny_c7#>L>IY|5rE(dXS?Y&J8!8wp7{%fnZZS4Op58OP>>Y46
zQR4W=`;%jY<}SGtv~%tqA<x#u6QU>Hc^_1y>E*u5VouhXElKHTA|hs9ymiUsFb{Xy
zt?cB)8@DBzQ-AKiUzua~dG7O{`uApj|L=9_OG#VEg3T3duMNe|KW2zK<hZBl`$w&1
zAFS*S@YQr|H_Q$4S(G}h;?Z9D_YVZ5?j21G4m-8~%z@?`?aIsc#P6Hi$UCFc*jIIZ
ztu;6Q>{CK^Gke{AuIkqsvClR<9rFL)^Wu*dX^(A!Kd0Y6cYN}U<gA~MK1-}SoVtVQ
zaYPq)f$qBFPBDG1Guov$NQgg7ifQ65)SULfE2edIA@8$f#=IkYcSxTu5T5o}>`qT;
zL}T<uq3sX-)*O?%(`UVbPk0A+b|L?~W6V4Fr9TRoJ=C=6I{u+$Wwp((=~pfb_|$QH
z+sFB>mWOW}Tk!m*mzN$C<Q6=ceg7is<;8mMi<cc=d~f>VfA4B9e)%%9Jo}3M-e{lo
zYI|<`CKtXpyL{X8z4Ws3(0kW?;*0lKFZQdxZ}NBMr*}qwb3gA(75n}0Zt}F-FYe}w
z)xWzd{Wbsnl1SIy^7<?P&)4WzWj}p*)O7EJe~kRIL$K^%FD^+eDyb++P2;jKH8e5f
LQdM>JcjE#8G+WX3
literal 54013
zcmY!laB<T$)HCH$ee&V$4=zIk1p|frq%1BQ8-2IToRZWceYc#%l2n(}<ouLWeV^34
z^pXq(LvssrJ3Fr8lA_eaT&{{+y%D!3-8SQ?eINcJ&{|ki$@Q+B`An1Pew$~nv!3Z^
z#=pg9z2!95qm!-`?%$`tcxzJm<dR7}m5h!878VUcHQW&s4#<m%$ox}t+{+!O+cbUu
z<_1y2^qCxDuNP=M-n6PoL%1+5^<dIh)f)BNq4k&Vt@?ZGP5Gv^SAET2%>5V6cWcW<
zh2L+lUO&oy<;Wwg(>tyn5fbcH;a)W7Xp^YlRrk*;%DlJk@2Z&A(zEvfqkqK9kUfzL
zJZ%z`{9NCOs5&0mWVFufUCD;k9jcy9X|CqFD+EF-xK}^C>2$?<>tr9FgX?y9X}Qg+
z6kD!S-u3<%pH8xC#OzpezFoC5BG{G1^wvpuA1XAc)|HOg_1tFfyIpSAE4riim{!-%
zX-!xr`(dYP-s%7Q_WrfrwJ!VWioG%-o%iq6)O{9z7yn*4ecx9XsdR~HRg2j}`4<~U
zeog48Xxz8sgOR+M|J^^j9hX;~X<}|!!(eVuutB3Kc0$@4o7|?|4=!`6rXSyEv+k12
z6Bi-w#&0p-!vjxs3ph12CoV21n6FZEpT8`qjZZJ{=9;~8-eva3aKCSO(7x#<Z|(#S
zL(6ykrPX^USKpN>tXx|4_d=AqRbkKumEbGoq5aMGr!H|)vQql5^ExGPXZXvqh6Pue
z>$c1|<@zJ>O<(h8T_3wyj=H@!k1bI5K3x86fn4Hx4`=BvnRm@E1=o}(G%#=7qIll8
zYIbbYq(@UTMc?)Q^^`m6T2cO;?|ateZ33k&0yzhm)lVoaUjI94D|6FBI};X`n{tH-
z3PxL=IfOlwNL>Hzd$`4Q-G6dtJuYq7d3kNlqthw2>3e-wu42epsH*erWz{!njjWUT
z7KaziDV`GhOru-z^qhU?rCGz4X1<&|W!Y6v(_OY_h5M&(v2x|$mo2&c;Md_5;au$>
z>|}VPr+cT>+_=HopK#+P`_-=kPu856>!z>IF7wj#mHkSkIksn`IDaRoKiOrvMBrgw
z4%=N*uJi)}tByZRTwG|%%WKqLy<YUzgp%B_S4k&(m{|n-FJ^Neneo?k-3zJ6)uJ})
z`jbD*U(F?ucvjEi&}FVXj+6cyw@OWXBXa0;?3)XZ<6J5Z9(1%4h<Y$1bIH`>kGa^R
zms&O5xghLv*ElDN@!`IED`qW7T3_9E;<8D%tLlM}>_t&8A6jZ`)bi7sbaUmjHB-D7
zc)M)jt(wza^ev%4pqTH`1C0{RcF~(MPc|OksA}<jMpt`p-u5kT;&NRJk9*zhxV2PO
z>FtHxJsSig77P3mR>|6|p7!2nx04ft<r91DoBgM!x=obI_}uY((mT5Z3kSZiI+wiX
zpE=?;-A%k*(eqoxAR;rXb<>Yo|CCxeikO08ET_#~QIuGy?0NjciI-e)?8mK3ChTq%
z+H2EXUG?<MqU9B)a~Fv%G~AlAz~_ck-^Y2oST2A5`>S2TPFJur`i%FY6YQ=|myDIa
zMR|TWp1GxVTi>$6NT-9W`+Bs)tfXyDSA}|h5c_9qQ#?oARmaQFe%;HR>9^)J@fQ7l
z{zkV#RNU3;&DqYqxzCe-Ynz;T8Q*L?ebc>vJzH9XH{Yn(#PA`g$$Nt)f974mOM71U
zEL^L5W=)KCxj^<F|1-wc#kCyIzA>%)d1%p>*Tu_Hc@|8Ztg_VBf3gK5i%;R_<#$#E
zA4|JdmM-A3>p1J2%Xz)R_xCEksLbnrr7X5L>k{{lX9~R6C8o=+Z1}|8^Zwh9mo_ds
zGIRaCl+M)5lhJbfaNo@98`pRC4S(A`^KWvTifDc+?zXMtX8e+QiwZ*SJ(k)c-F!v&
z()p0I^z{ycm6OVzFwZVwy&AE-*)lkw+9GhOv+)JD&t}|W2g7a6ELqFCDqOz%dux@{
ze2pvL*7(D6i}8UJw*!8hH_EdlcK>Hiyd6+5uSNR%&4)UVb>4*fryXK3{E)llg08bG
z`<#y(et-L#E%Rh!hp@Eij(=ALuS+R+K29vsD>9z&c~gVBILqCAX-~b=doq6(^J{$C
zai-Sm&6%tc54PrMr;asjIGuaSc2=$nt3i9e**X2F2kh~$O;gvuU-jbm+XXc<zw5@@
zp3dLAFXi0l>cpmbjg6mso-^Jzo@2#x)lNxshS;j_AG#x^SDx9C?{u{1m<;#dO;?Xh
z;0~MHoTENrS=TlGTXm6~1rK+2ZhqbO?11Tqzg$xS@(ZqR{hoLG!9vxu*VDFi=C;q$
zE6m?!w2w7liWE}=YpeaN@PFwCj80F87fvcua(Pkc`eWX#oGte*buZ)-*Ll5?xx(q<
zw$y!ZSRDH79<NBx2)D5NoNfPp*=gg=QLTFyg?a>Mx$IGmyLa#91#kbEP2%2LEH1uw
zea*S=PG<b>T4rwf2!X<#%g?KdpIgq$99!@xw)<DFVV#ZQ_srQoe`l{fwaSqFcJ#SM
z8~qQO4+Ogkr--wfaH;-f=Qvsa>#e<+{?4CVsd*{TS}-*)1ypYunh{rl8k-_lplc(%
z^KZKe?E4<Rp`A%2AorVRQ7z}K1$;sZTPJI1bg8g+#oTPs3{uow{c`>LH{D92UemY^
zciP6hJ9F;b<JppI_nCweekeNnJe23#RQln!QbnOJQ`5vFlUTY#9at7w@I9;YJdrT5
zo^OAER$9DxezU!pFpr6wNt$w7zt-!nT}C$g>s}U^cTa4Hz44~{=cQap&M6vRN!uT1
znWeL|uYdJ*na0%|KUTi=r@yQ-?J4F<|IBeN&7+|D+w8B)W=>RDBcEtJ(c;dEv{3Wc
zTTY!};(R^z@9n$0Tf?`{*ZF96_El=)%WZn{p~){}FW!yT+T69{tG<Ns+V0dFX?el7
zc4dFOc$U@OMeSXZ`RXlim*t%|VETT;H*@N4>jP6CZD{+PvGUoscRCL}CZ&}8++$GA
z{#pCMl9)E_rJqt~*+yKhwR|e&q`K(h&2>MGtmTC5I+S`k6vSCBHaxu9{33bptFo0F
zmuxz=b1T30)zt5<rR8xxna2WsG7rC+ajD+dC-cyeDQY!7nYWHlS(fW_^Q!Y+Gxn;P
zM<?$~HgXEFTm62C!^S7#Kk^QIc(ZT!%@wciz71r*yUaIg+A5uYtlUP5YvgBitdTc)
z_;<nmD*G3ulapti2wgs_G|<vb-Lb{Du>Ahbt*L=ekIAPwzZJW?FL(7E6%JnSE3?-{
zEMFQZ-7g}?ej)gtYJ`H{jm^8J7c}&%>|%MqF5{fq$Z5cx#l7g4xZ1(?>+RZS@3uVX
zSln`PW5TU&9tRJDqhiJrP8`qKzSJvvdX-t_wXD4+JK2t1U1xN2rrnBWo80YZ7yrt8
zIp5t(qSB#SzwL2FPM<Qzh8nZ!L2d7=W_C+9dic(^zPdG7j!ANgg4ayJ!dUC43I%)1
z-a5ZNaIkZazr5^JZQg%V8^136w|bY(>$$9)6*2GgR|u+dZ8duFGGl>=a$o<HYAaD*
z=RV)qExEB5YhK-Qn!Ln6ee$8a+!=zWa|53Rmrhpe;EWgapMRt`t4#e#)~%ZFC#S9s
z-KD#`Sg4t4AIF@Sq%KDH&1;v1C~r`HmaM#E>q5a<kwV{?+`<%uR0~TsipET3l(yi$
z@!$qS+&a^HJMFllIQeF>NDHQ>e=T6IWxDY^x?zTo%LZ-^Mq?u#!Q{XVZO7W*$e(_Y
z{9)gA)y8RpHQAG!4R<WoR$$?LQDd+za%uD`_G;duUCK7o_UI^itbUsE`^nx$qYn`)
z?yfA^%BAaLc|d2+YQ_tVkCr`qJ@wtbjJICtSFTKnP2A>j;PbXq2XD;PHTy6}T)Q_&
zJN5TgFL4PU--9c()pcuqx{YF|o#I@-)+0D1XW`}5aq9zJ3t#<>DA&nrkoEtnKC9Qo
zV^iV8ztziTKVN%1<h9TPEt!ncuU9YMJSyoNI435ctWC!J?_KrX2EuQHPcjR7Z#+KX
z=f`{2ZnKJ9ub6yKzvJ;;`O6|{UW;=(C;Hxe$nhX(*M_Bei*9dcXjXjsC@?HuT%dJ-
zROijwsJpV)4#(cL`EoU--r?;+kwpck4lpR(a=%u@&B1Gw8sG4#<EI+yTW#Av3W>Ye
z+<%oRawVxGzxh1TyGbeQc(eJ{96bq6`~Ek!DGhH_FSPXR*cT`JO+4V}um8*PZ`%81
zG9H{EA<x(H?^Cl+G3T4S6c2rExl>*OhJq#&raqm_dm(@|v1+y8^?Q$wGXyl4x;Rxa
zED>W^SXid|aJ%;NjTKiGF=m}Ud1Rrx@binNGVF^O0wa!82j*xwJZlM>@}YudZ++&S
ziw`v;cgRm(!WypCDtYR-zK_Y_*@bqV-8X-=-ahy}mq|x&YHv+MS#e!nyv$+IW6C0*
zpUA`r)_yGeF?ZIhEWXR33*}!*%5G;ler_3)tBqZ*U&h5n`%0zebt{x}@Z0aW$mSIg
z`hLqyMk$HUp{&zN%U>(H+i#biDq7{}TlaVY51Z)8^A}FBd*6yW__s6p^9xb#BYvsz
z-`DOnaGmq{CD-(;I}dNkeck?>xipzec=h-0%(=_2usUzf_~Gj2ou9Vx-4&kLlI5rU
zdE63{DtMMGwz#JCiRIm+^V)X=ZSBHe$(KiMFAnxyc52gK`3x<Uo;0PM0>14r=i8P!
zFSNJkY>+c=kJ<ZO&&bi`ONW=f$-xP)Z~7RBRPEzzi_zQkfXiJe`<=c)21iWra%D}<
z%^~g4ON`wj4;_;gSpWae6PHaNr*k^|J5hh`VuAHcCx&O8#}}?!clN;RW1n6ezvg!0
z@6MNL!D2JnPoGdPUVrDw3YL=j`R=7c40r2<5B$BgY}<v)NvrR?^kzN0Ra)vSkK*Er
z7PCH1>uHJVSFcLDi)5RgU$FN6y5Hxrwyr%>QJQ78yk|%MqQag@Dw6f`mH#i-?_GPh
zj|-zRG%_ZpGBmQVFhi~k&qn3uKiMYu?_T^y$Ip&zemx$qzBY;N_;D^={cLR2^F{KH
z6c(tyTUQj7shs&Y=*#i)Hs`FRTdzImy`;2(H&Db<=G<%JCF1+_eke7_@AF{!!`5As
ze2-a7*KUvOthZSo&t84By1QcH1|7~HkFq%9YveZQoDvH^+J7&#_L=*nYWsPAOTNB+
zboT1wk4?t&KAc=KKW^8>gG}Pl2a}YHw5`~Vz3=NTaS8RBrdH(i_56WOj=gL1rr+q=
z6ZYIUI>0)A>#I_C{Vq9+NusN^uRp!%QvXk`cKgHE{+0Tf^=UIQx0uA#7J7X8xMa#l
zg_%B^su_0jRX1$Af8p|nzj0RAmVT)5jsMiX$7pto+3&2|Q(~`QStJqhV{MR_r1#}7
zGp20!y&b&!=Brz$cErE_zjaFLu@n`*DW5VfMVu<**LeDA#+3ibt!t__-wGE`YJ4W1
zlE!#nau-LW%p|$>=ZxO%{q<w6?D^x59hXT?KcK@`Q~8Wrvhw{b>zpHT`@+6@1>g1%
zl0Lgl-~8$R(0DV`uQBY6A=+8fzt-&#W&2sN;YVIZt<pgsNzP!g#tDBSHtBTp*xz66
zFV)d~eAld*^F5EHn$5g#UUvFng^IY2NAm8}|9^u@|ISHCIb$^QM&c?x53{$D{x*AA
z&6j<5)QMfc{q7!~8cD8W$&7_gn>3bKY|L4;YNvMglxU?rG2x=tpHnBzN>CHxlzL}U
zTa!_C#dY$E;80%QX>oV2)J1;Z6<Qg8@0IwC*@ynDnss~sM3Dk#>)Q=I>AcEJ>5~_~
zJs<c;hv%@N;^V1r*UekZ!R2SPea%F{+_g)6OJ48gShx4|QqkC1h7&(^zVMLQFhMEz
zLifTKF>=!P-)-Yp5dQUR-OI1OmusAEo;WbEp<~S(6M?&YEZ-ase_nOQUiQN3Q@6@Y
zy`@SwP4woFvgs19I~F<N<mBEMH9_Gf|7VL&F!_Zw?_msHdpml$^_>kT_QWoKZt{X(
z#QoD=UGw5!k0s64{(E;VbsyWMzz0+3)?SO{7rMf_ch(Dr^Wm}UUeAl2|91QLH$41q
z8h*YX4$s;7=ih9lKl7e^X8u!m?m@!xT-%-qA=Tx1btbRB-~QL}f|E(&?52f`!E<<d
zm!DMXoB8D8%1shyf7vd)*YQT4<8|SO$0{fNLN8@pE50E2?9_kd2ah!^1XX6(OFViU
zd18`+ZRoYdk>?NAuy3kk`Te8tMtbtIBC)>(Vy+htr891NUNMuIfpf{K0>hre2K+Z~
z+`jrMZ0giqjFq$a93r1(7HOFoe44GkR4FGfWS?I4<X)!>j|{FXiJ!^ZkY?q*>~;Ck
z=B0bKd5fej-{qH|r}g{kx;NK1|J3P9_%FEBxaiWeXVa6m-#n)@sUh)n<9*30!|q`A
z0vW~R_PVbZfBn4i!1s9k+P^#F*zTO?ieS>>QS$zKD>^GyQKW0H>O-mKe@e>nZ?1k{
zu&Rjpq{w~YNsT`*t@q_?vug@fe!=M=P;f_OueAn?^5HYVX4Ur@e^oi2jW3J!U#n}t
z#I0<zbNR!Twg<uYa?Dz~3XL{0>6&&6GyZq$Y_5wx!>P<1Xpxg>#<S(i-gD)9W7o?#
zb39I&yi5JS+G9FCL9^8ue&$Bs4W4PVIf3h-dU5E1Y=<wUi3fJ{SV$V!ocyPB;MQ$j
z2B+vx55$8mtluppwQ_l&roO-~rXTT_I9&TncSZ^K>YhD$`dknvPv4#3*JYvi1Kl&{
zmpb2j7oF#x)_W+t_Ojia!gIDUG7^ey%>obOx=on63pVnbY}4(%?4)1n#%|plc-+SO
z$b(3wUo|(UG@L#XJf)+gv{!KPY~{HPr7d}<ggAdon-&-@lv^w#BOK8g_4JqGl<Y(k
z-HpGtuDoRQa_a2d)G1dkT1Oh1SZy_2d%;+H*1o6f&gB+%2WS6U_hM>D=*nv5V&lfD
z6}%7H47cy9-B7a7WnD;ic=n5(W=o}eqBCbK(t35YwRUr7!lAuw{0nAwG09o%-zsDx
zFWhyx_HVQPVY!LMbCUL4Hp!d4>4nBQKE<t!X@%dSRyT1-WSbc7Ty;XUK+pH~X%=fu
zJ5%Xd&y&thshzZ3MDvQT`C6kgsf3L}tqaTfmaT}2a<SOl%68RphHA=;TRMscx57nv
zUu-E04%nXeY0lmx6OZ}H>ux4go2T~nNNw7BC}`KF{AA{;q`-nh6Izw#h5eE)tzUAe
z?|a&Qb1M#>-F<v)1|kljSLdnTJ6E&qVAYhyOdi|Zg0>aT(*o~kcKJsrJ5AYPrL-!<
zz{9P1&6AC<`j@jeImHRwDc`bYQG4)?O2g#>M`Dw`s~p5mn#@SdIOJh}SZ0c#qmjo;
z&cIVY6l!_eRn>H7vUhAc^U<dv<Af(0(+%<6ym9*~Kc|%`GW?$youS4VFe%1tZLHb(
z^0lYtzO|paH)YlalLeFH6r?7n*G)^k$G>Xp&L6X<fBCa8_k{1CMJ!A1?XLRm^!kz8
z)A`SC@bNf^G%fhTx^cQiwP(-$ty8`TufMdNt;6B6Ov!`ds%PxB<z;Eh-=+C#J<yr(
z?#GKA-ybqgVosRwdg3zctezllhBnhicH5_~JEV(dIz~P&sXQ{zd4-eQ?czswZ!BMB
z*85ayity%)DvwQJS0|aA`naIr)u;c5gu{K0IjH;0c<b3AcAm$9>tp$4C#jRU$BVAt
zGM1^>S!vP1(a0)vjU%IX(Uu2IYZ!emL>{ilTx@S9yyh&Y;V$0^?-H9hHYG-8$mwxA
z^G|D?cX5;BvjoNR-L31kKYxB@!|5$$Oa8`WFLDWGwd4M+!et{NJ4fnj&r<(`Eq+PI
z_nC$6G1+*C|Kh_(H>~b_{3x)NC9L-G#|80;U$~Pt>o~JNt+e~3_F~V?sZ;;1dr^FU
z)h4x#kIK`lE~sUm-#NMAE@ybhkJGug6Kr1#Pk52eUr?#we)a728H_EfMH1Eq#u?iB
zxqsjcdAvJ-msOx&ZQG+QVrkAxwuW7pVUfXGaCe84)6QE9*7PpqU+~uRtILB0kv@#x
zHqN$>KgySJgd|Bn@JXIvxsSKft}Ik!o4~6h42EVE9XEsGBl8Wl5;rqn|0{L+`VA)G
zH__(SSJq`_zkP8gBcM${fO)y&wo}}u(gLkJw{BEjSZb>B$|N@8sEWcegP9t~WKXt5
zIGc;<+IdTw#cnhH_}t*rd*w?zgj8n#*yFd5k0<F=;U`Ng+1~Ox5$i8}ydnX)1txoV
z`dHpB?mLot;*#Q<_p+JwHg<_Bo-<_boI82_g83xb&qWe)$%P7!Zd?;8yChRAnY8Bo
z%ntcStYK!lrOU)Q{4*5VoUE5I&;4L2CDE&XA#cICgOLXoi@5!)Oi`}(UC7#d;qkvG
z2@<!rJY99<U~wMH4ksRiFPbrmj|*nqd9LUEF-h<7%_leBEV&(|!K&xg@%330=P5hi
z1GjH<di408eWUdBnaYF{F?;L(&Xd2%BL4EnzMPN>o8PB8mrLF-SY<Ss`GC#KB+=GR
zcaLPIXO}!HrrX}t6K#viox?O^iM6SQT=ycE*XCW~d{Gwd3#$6_`drI=!kC0=j9AWv
z&O85XSIfsMq6TN?aU2o6QD9Lx`Sh;Kx%Sx+3U4&5W+nVD`uOkt`ujCA*e88_;d98v
zFM5je(bUb{)90xs+`7GNP4y|a7`fd``+|<e8vOWly8pS&-=ANfw$HbpS6@+D`fvOB
z^grTLKAvS=ZL8!N`79vk$+<e~)$>$7y}Px^JhYXyw6`$mM?vmf;aRSm7jH7|To`*%
zl6}kWqSUTniB0>bwu<kWwvJu8q4887Qy2fXbQ?LP?81UKH=kx%2h5hs_X~V;yZrx>
z17G&;%r|1a?|gUl@7J3bd;Gn+seGm1;*Hl<i}PJ(Fk8jXboa+4Ip=G0Ssf~>^Sz7P
zJ9!>0Shn}^J?HPCtIhmb;|<(|Ym-mdzq5~9BK^PL{;&9l#vFevZE*`?+Tw=hW+-j(
zwGok^fvLLp>l5U=f*iKoYF_?ZV2MY}mdwX0Vi(#~G_F1IQt~NGl?%HPfB$ClrJ!lC
zmjxa*IKMmd=FIluwR^1lD<u92>M#4yzBgy>eM7Fj=l8!oS@rU-)%UdN6U7&s96K^w
zKOxo2!0Ygp+0Wzu++TgtB<09|qn~E~4_BMlpYZ2YGL+<~-Th+c#^q+whtdxI+kfkR
zrEMy6vPS1khv)srwdA=}J04UhY<af*(2e~QT`!;LdiH;dss5_wP0y1KX(WA&@R3=#
zIq}hp8auVV3yhO@t(GxKn|>tt!ov`6ziRo64kww6i~L2_o}K@*j8Yw66>E#_NZOF-
zBPp@?$3>&lKQ*5w@TIPN`m4)O_n^wL4f^}UeQ!%hy}5M0Y^iwXeDSX>NA{hFihTb-
z*!S<-RVTvQbRSRqe{55E?}@Y1Ti5Vj+?=yTDYI<eCmrFXD=Ke)HoxnuyzJJ8Orgap
zz6z@(xR{$C#Hz0<Gg{i<XxsT`ozpe<SHC=zrfkYkun2##cV}Ham(Z+5Q@$LC)p)z`
zzn*@SvnJ1*$k|7Z-TQ4@ntiTzf!<P$gCb}617-=V{i1JC(fxJh2FVU#vF57}9p>pP
zhQw)BD{WYwu_pCs)QtS>nd05Cfx7-jSp)yIPW+l*_w`%XJRY7V2~M4G?XzO$his0<
zhTr*osY!n2S7GbM=MvwwmUa8;inGcXxg1`}`<Q_>@y7AhmY)>31bdl_OBTjeGF!L(
zuwV*h$&g7{IccqG%ZX40%~`jPt~EIClh;+}+rCw)WKzbe^DM67Q_JqBp4oLP=f3;>
z3o=d}Ume=xY&iCwpUd}m!TDtZlNRYb4U?T-*D9hR?L6xUckRlp&z;oyFCSlMcfZ7S
zru3%lZ=b67PQ5YB+WC`>xu&Sbb&2h3_MG}|RJOWG==a{%+lyjlbFQiv9a^X!6yaE`
z|7zI;Gh=trDxo7B?2C8LS1#T6?YV~R#s4xZgLyBc9pRqPDJ0@@G9pE0<!8mm8nTT&
z%BDHZdTRb=PQiTt;tFTKt^M^rC0u&dO$WvqWwWmf9A^1F-Dyf%V&bIG84PbX@9wYp
zde&v%uc)H4dqQ?D56jtde@arWoypPthw}bjjqgtT+P!soXe7hL`!#cU`qD1Ueqf-<
zD`0r9MvObKTZ_4M-H||Nhl#J!0?GqLBrTido-a3?5@7Zue(T$1J7=7o7TGo}v^}KO
z^UqVCu++fLn6%!gS+ThbmN!%fyEZ6iTwQlMJwhzeWv%-c%W0N>_NGb)>R;HlL-)ek
z721LCe05gE)VR;Sn#N`;^IKNrPSf;l7Kukf3KxmAhTS)vl49`pO2h)G?E7r-``?}2
z^I5=q@`c+~Eh`Fn)j2j_h_2RAX}^AH_cyz@{gWM>c{A<jIrZGO^9?XzTx`%*>HL-D
z-kn)97K*E>tbe-b?x$O2Q~S4OxNgh(ee7?<0fPhFn-|<?nt0O0H~2r7*4tXeZOen$
zUq4xBz@HUxOE28lSZUi*R}rflPON4}<{yeYx*%}^_vbBz``7%uRU>fU^;%}m);)9W
z8eXaPnQjYP$f;E)pnG(0fUN&TvyTn_-<SQH{3hzHSebj{u?5SEZ~FRqEHaB`?DFH6
z>#LvcdUwnE9R2u@S(<WlwfOe#*EhMfUtqT6r-CJqS647hj*w|rUcE!Rbeq;Yt-miD
z{ob-{lgr$6U3C76x9isF{J%d<AgSbynA2uU2JiW2niN8cb#A9Knnr$@c>UX)doeTA
zd19_6iq$FU{+kjcn>OpmuN4n1t8VZMcNlKa4Qv<fmH%cq$6bP7bcPE<_O?yizP&uN
z@mprnx<g`aKiPa4B%amlu;#ys+IY`lv(ZoX^)qAgyDqJpnYdwN-0SpL+HoF!iQBhs
zzx48+h5E!EMuC5Q)74@_;@mEny={p3xV2(NU)hp)ht*p;=a_5hUKQD6F3c8p@ao-3
zs~orFANSI~eVyA>WWCO2V?S+9YroZ^%_;%~Yhzw>&y~!*9@cDgspRnOy8CI@w<Pt>
zQT(xD(F@<+^?RmN6_>s}VYIUG#KC!+_OA$WzuxH;7Q7?%z#fLPukO`tDgM0S*^S$B
z0rDx@c?t>f2I1|D&L$jg*JQ(A<&+)|{q^VN#i>$${gVzdXo~3{P~l3jN=h~?Tyn0x
zb8!iaTHh5xHST*`^jlRs1^nC&Kk}H%FXsM|BSK~+`@*1;(-#UQO2~eU2#Hy_dEq0i
zT05<ES5_tzHdG$#pP0XNX4=DNOF0(RHELe@v%+pxyz&F(Es7fXZFZ(PMt9<OiySZT
z3h6pCg`<3@!qR{5-=6PHeDZYKd#S@r(I<92oqE^xzYo*-Gv_)kyRZJfNc;DrJ2!6L
zSbKh#S-#gkX|d$nHuHBqG)P)!5h$}*#qj&x1<(I~Ib451|9t>vU1@AgOkHVYVvJH(
zZjF3heEXQ-|9ktJ{4PlrxY<0M$8d*VGJ2!%_OP`sS=O@ZiyIFpcS^8iFqpXie7(E8
zx}BZZbK0Xm`906?-P`wWd-u)bl5<->w4YOAt3MpMZSLMZv*zwgGnSg9;`8E77VBs3
z%ac#O$@*!xz(vq5Ud2-=W=W}ayn!T}{`|-9yNz`7X74!|xAo9{^VG}Ney=&ZdehAR
zWnarJ8suaTXe9dFU_Q6%t)70+=Bj-UY<T!8cOMMfeep)ds?chW*Ot9Z#)l>)w9lQI
zx7jytUEzo822WOI7E7hP+;{P;;M_CwmS&sBJbWKDeV1Q!uk_iZlqjB=VpD`xiWub2
z*tah(&HU|^rxh-f@*<=3mS^a;cyv7BVL96=a<U=C*fZqF1|O?MI|bj{aP4Uk_fa`e
z`l`fu3eUORuiIAT7k&$ldb9QN^qu>E)}HRPEjj3TGtubLtNX=XhaIl}<l^v<WUb7u
zdz^4zVxxbI9-rqLq2t-5pMMwazk5(jt;w};=bT-!Djq(``xm?XvNC-$TP$O$=&h~I
zE0vYK4$ZJ;+O3s8YgNkW*In`Nv-ZxiGV0_AuRDHH<ZO$^%o@L<sb+8IhG>c^Sd_}&
zYz>&S!K0_eCTZuIcPjG7m>qp5O?UCAWWWAhAZ_0B_~$Hp6nvWgpA(h}sZMwJS@n3$
z>j~4|-l)2kk#8NdYt<{weJdJnm#t}D=)Xx&P|A2aoA0NqQ+Xp4CvJ|fdo$~+)ZuBf
z8aKNx<k#X<7xcLQrY?C6$DE&gmw&oh)!|^WXtD2p+xO>9+SfibP<z`suQBm{u^|77
zBELr+{VVEzzd6C#$G_3#SW|$?LoT0|V-9MEq-Q=@b?Cu0E!T?)OG_J_f6Mko^lJ1Z
zlr8d|&9A;p^N$%vV`k^-S!Mh8l)iq(E%n3qX<c=)(vryFozC}*-lRX6A~ieOKu_9(
zhuKVw=X~4J(B8I-Z&nrUJiu@{cFp!$o_iunyVf17*p@jnQ|82~+M7)3g7dePZirHv
zvVD5@=hxf+{Q7hD>C-8hCpqS<GuQS0FS^J0n`VRA9?@r<8y}TA%b$7`@Z#yM!p)nu
zuiSRyH>Xnd=8Ev7{&_Q&GpF^h{JA~FG1-8_?%S`>P?4*Qs^UfyaudHu7A-!$QenD)
zT&2C~p;^1slb79e<)}!P^pN1U$SL}*bbyn+cYkK1mDgHT3AI(9Ca;U$TQVmi_fW;m
zyuUNTyVFHEA8xX;_0jQ~?w!tcR(Q`WEmtL1`IpxY#|qc=`n~bn?pt4F^6;#4lHCDW
z>kkt+C54vuZ7ZGK(e=Ro6l3enj!WgsBPMVz)RDW))M|HxaZiV!z0l#`^H_IGvAz5+
zeWygPPo~nOhU_OZ9+}TSep)?%?SRAN6rZ2-w!c0v=dSGc!&!Xh4N*bK?vL9h+p7jj
zX$yIM`!AsB`(SC;)xSo6vxHbJn72)|30uOWB-PSxT|2eo-TbGE-(S0ICm#JqO4CDk
ziR!mEf_3(bR6RA0%$UaZ<j>A63iFlJ?g@3goObfzfhE$6u2n9-cE6XK5~|bDb+h93
zQ_0@-`v1Rv`g8W`*PB22B_!V-PKfr+He$LQCGp{m;*R@it$BV~s~&|s5V3rF^zVz^
z1|6I|s}`SLqo!XxCvlU&C9f`ym*O5_S-kJ-jJaMXdmQOBI?=ds2FIzGtg7!-X>v7o
zar-lG$A13vrDSeuO6)sMmm(?l*Oyo6d<xQEy~Z$j`TMfmeP{n);J)0z(w%cF_ao~Z
zPOi9yurvEwHXp1yfAe9%40Roe!s}1>O%RU!SGFzo@4C&~EM?QGZ5OLGvO6TL<Ck*L
zjL({_Z*}EBUs2f0iATTRufAq~j>A96(%K;|+2iPh+7lJq<`#PX&NJLS=kJ}D66;hH
zmn`gK<YK(hIcJ>`!^}hpw(tpoml!5!s`32UdC({EL*DY!pSvgC=*w;6=5Uj>blRx(
zk&of@@)&u!btZp~KQUpOsd_3w=yJ!Oo!jo_#PlZa{9w5`a<_N(fwWyGX6vro+{*kp
z`!&}EkMiA0=W}?O52UI@epr;Dbs)Y;`393?)dAaE9D)+(q8qb5tl-+KF2i_q_MR9e
zpDmsHj_t|bt$(ZLVqQaXs(_5&V%v}h-W})8@=afn%J3$Lo6UE5(F7jdQl=UE&%It|
zTNpXtdh-=AR=!R9b0$yAHs5__V%oK&orwiy@$){oF4A7ivw`U!i-(58(L@P*AtAnU
z-=$UkRe3fhGD3e}2PdZaWJvU8rk>V!oO%4Z(vvTzPJD?o7dq7Pb;)PtvY5No4_8&^
zZ`<vZzt#W8-sLmCxz1g;J@d2q8(!~<b2aRF*RB7A96IEAqUKQ2xz-QmfiJG41bo=G
zVE!z&TeeeclQL_L@s!j!R=3ye)V+G~+A)>#4W_R>bpIMYy~JM{UKp}r#f=AB)K{MV
zvaDd|+}(#Rl*@QG+-a_kW8yl${A%^2h~!PI5*6<6ug$BsZnOR~b@{2b9uBqYhehj}
zwuh!Ft#8`0=IEsc@0&XVMUEVnzjt%ZyxvPZ49v^^d4vd6XTM|L<J-UabA(Xhs~F`t
zTNc%uheHL_G#@(kY9{nHTkmdgTV9*N&$RXAN8c$Q%U|;{=e@~eTfbOgq3?&PtF?Dt
zvt?gYtNS_Yvy0t2jZVqZLx06y&Q|9u3ypX)QP%l}^pd`E>E;xHF6Cbw$rA4@z8zWY
zJjF27%78=k`#rICzg2-oCCoA}_A*S1e0^M;bD0&d!)-hL4Udj*HkR4&tLwKH#~qga
zjxC$(OV;uWe-coX2-i4v-P+$iBb=}9LJdpAext~f6BrII|8eSiDocm#y}J`xf2`+l
z`~PpJOa$wfPidPz$lmN%)3|eV71PEBS)ZLtWu_FyzMRYDxWbc1;)$H1(eJFxuD56E
z4;b+7scDw|6JRupeQiy_=9j)-Pxe03{&Q4m!OYL|gCx9}e(NsO37oI<vh(S(mdQ+6
zUrHD<O*L5F$rbtie)@k+RBiLcqdwQmr~NtgC3xSq`_VEkGqh5)8{Qso%@H^4>y4V!
z;rhM4&`?6*V|I_|+D}gQKb9U9VU~`yIC9BRy+r)wQ%+uyhk_2;tBVZ97QS(Lu{F;q
z-H^{gKWxL%+BZ>VYz-ez)bB8oeJQ7%QF7er_DV;-jT(Fv#xJj}2z)7c%s5u0SND(i
zg-0FP!YWR3d$vixyOI$VA~%28%+_gI1{1dKklh`qK0A%s-H!36CP!7Gj$G%Tw%z-r
zK1mDJep86}ERnw>o4f4Arn3Bp{u4SXbojnm@0+J0bI{%6;M*6f0q;*&2c`LaTp1kb
z@<`%)nf~MR^*`l>{HD8LX~~%r(~>hbGBQM-fm|DvTYTG0=wID_r?(}~Ud-fBW7zj>
z+2z8y)0YV^&s-C;KD^v!ip%XBjY*=OyZ629-yf^v&?qA6zRj(WL$Bdz(#9LLVh^62
z=bXRe!{f;T2kgr?MfmQU@7ovuk+VGV&9NGz-LpP495tHZGx5ps$u~clr<YBZ^G!W6
z*L7!<>QSNimdjII*4+N7mEAE-;AMZ&z5L4p><`aOPAgpe`|fq$T}nk^!fT`CKVI-$
z%`$J|WWCN=qF;QtvgV3ypP3aQwt7p(ice;d=_hUc`|IZa<tYC2#?p7Ur{}Y3la=}B
z{3}Z;|9$+u*?jZq%bUe_#rumJrZbilhql)rw6H0=CAC~B#{I11Jm1Q>hyGmqQa0Vx
zJXxvWdip|r)Ae7>P9A3yVd~x`Xg%wf!GqaO{l$yf_i26Jv*&bIwot~zXFrWKPOT93
zTq(Tj)$C7x-`ge$hHuvtTk>X!+Fak(f|(tQ=S{n^;81(z)YtwYTI*lj3{?%Cn%=$S
z<f$f2KmGJm=VG4u+wFPR^*VdgwgjO~7A03}?w-8DFY$b$6i@mq?{|OFcI0dopYw6T
zcdNhuYK+zR%6OYA@5=Y(@9jIOpxW(p^YPECV&10A;r{NT?vbuNyA%6Ew_mblOjI_C
zNoUInQ@Yh+c&_e@i^NxpeyI(nr^NX6EQRalb6T%EwPxLEC!O8@ey{9zZ`$ctf9X!?
zUG;Tc1**bERTJLkv5UF+u*`e)<Fqf6wK9jzR^tyVd&IdvZ@Cs4x@(`-^#}LGcD>q{
zvvPyO@n+V&i5Hq#8%v&jcUgCNS>WxB$9HCg7k}Z3IA-fSb=J{MYGT_XKP}~Zy?pcc
z>F@34#h$DC@!`{xqd&L5KPJPmqNDQ8Wx15;#zJ<&hjz>iv=ih$cxS~yo6yO^iEF;s
zJSm%ecvBI_s!g-+ls`y*bfR+VGRv(y7p)4}uJbbT=}FtG>QBx`^XNq^R(a3Qc`)6p
zugkl?vg!L4=R(hlja_24|5JW?*IfN4)SY`baOX>wCKe0!q-j@f9nW0$GXCVMQ^B8#
z^Zc9Q`;y)*=q_7$yJ)7~ag+Dl&)n6xe3pOaS^AAN!aH(FGPfX$k6W7Rl=sipaZcVY
z@#h%Z#53H!H-0US+kP}-O{c<+)?K1(0uNtH99Fc@&%2a5d-aNAnVl1Bg7g!7?*8yv
zxhm%3l+(9Mcu%i9EL2{Q^^5gac-706(=(+j|4J}7E>1i0RAs&QUW@XTkxMSlkn<L6
z*>~{?)7BT(>iPHM=I#4mVYkYY<G`V7D(l{B{Y+UDF)x_$*K1e(_Gv6?5nFrfG*6{w
zt$A#AvWV4gMXjfQfkHn=$DMYs4e$OcExP+ULGfbRuPZY$K3f$8Y>DLG^F&zmAmfDA
z!upo{`Dct%)_wTp^ONz)!KM76i_TZ%Pf@rdu;+L~_pSJZ_PhUr^5_0<+Vm+tK#09T
zL`bY`!53x)E)Ss@)_1hpAKcCrn_$Id*B{dIc4EP1cS9By-u#FYPcEADB|op-ly{=N
z=6{U+$+_2(=kfk%kQQ*#GSlCr-NP5I`7~gLeCLG4(TAI^EG#RWZ2EFh=1Rd|8wD<z
zbM3zr*_eOYRB%q5SBUVGrJH4>nP-cMuhU+4d+RB-4v!BGPO7fDPhTCLu*qxHg}uwl
zLcU&2v2v4J*y?C;E-Z9s-XtgOZl*~^hm#!=YDKKprd>-}{l)u}r071MjN^T-U70K2
zK0D1dBlZ953&)~PUz>1Rrm@J_M?6s6J@~xogX4R0+}2N2nzMU0kKpp>MNW%6W-oVX
zc(6l-%d2@x>a8WOL&KKLs#zSJx0d(XqdOaewhKu`PU#jgGgxuDct%^y?a3CAQ@Lyw
z8s(q5rP2T4^5?h%ube{|=1T}9yo`Jx<gj1h(EG114qp}cv**qC6R%ybzH2dM`*Yyg
zN9KofR!wd;2*^6DzLM9QE9l_mAZaOeM#Tlv3%XNVn;t|?@Al6Ows)TDx>(mLIy~3u
zRzzFu^(g{^C$n$vIkaHnp^H+YdoO;goqM-zLbDvhu6L!eky*Q4RLW*AEl*pWQDnv!
z#&aTa>Q=d@xnW!C11Fx#$olpn%ge*YDg52WS!MmJqFC;!u3N!$_;TjnEVs)-_pRhB
z`%V?k{j3$6l5vZ>p<o4%=2x|rgfIV=$H&!GxIKO5wvxd+b!M<<u7=Kntd&zlwRo@8
zv2&c<lOuiD$^26Hy08H5)c3qkgA<ak++20`Y}~s~+>;rSxIcNirfmzFx{mjnNgU%z
z@xHM3H=6c*lIE{0>}U7BW9Q6JzIUMJNv&U!>deSLo6Yh!y<8*AcVwr<rRaDMj_Gv`
z&z*a!Qx?hZnsP{5F22&K|7vvbf8*?|{D{ZFg|@v9Cw;qE^gMQB@wHxE<Cj_1T#<3}
zjApF=*8cQT`;yD|zb`oTuGAnR&$w1gXM3fr#I)OM<m)3+i#PguJv=X~X8QC_au7%E
z=7R<=R-L(57x~pJI(A#j#7nmnIB%<dsd;g@&#*6I#oa^VR&J}07jD%PKW>mLl9;nf
zG3)s43KxBjx$>uE?skhXpPabhyMS2!=F-cDx>k!@Cwb~0di-$DRn9#foB~;VTGJAz
z%REvLea#gx@tb(mBDoNYuGZvM*?^bgEwa4(?5?dXds8BDVWEEUn{|BLWphi8%6TNl
z2pw3}E`CPXu6s{LKkMddndUdgPk%9Eer&m{^1-Kv^Q-Sn37t9p`uF1xwjM~1y>`>a
zJ)nMO`qveT-kjg0Kc4+mWd7;QFE*n$Yu8`e_cC$y{?D_&Ny_e(=6un6D9yoPv4Yf7
zUPpa}pBL2>zM4MrdM25DM6=&bMgC0m_LE%7)4R@E_yjGNsJ_yA%$-%6c~?&0Td%77
zN7^HfvnymR=yA=S_bX4CZR?jUldeAFSh1>Y!o7k`k3J~-73~o)@-tZZ*QpIu1|3fO
zb+!52br$yB;_qZlo7JO!DEBd+kx|MJ{8alq;p?wSmOp;KzqB%Pm+VEi0_|HIyH@E$
zo7fu`&CHBiu>MMJ&I!KtFWy`^*7(UkG-}S?<)2#Feh1XlR#g?cxTaqIutrcl{_bUN
zxjR1P%gTP;342w1|KMKMunV4Bu6w_#^-wu~IpnVED;tT~DXw=PEbUsRTy*=5N~=b!
z*pZ2hI?{)#-juO#VJ^5h-=}uMZ1yE@Hl5HcW<ELX=^|IxSu>9B%@bQ7qV?qFpItGb
z{P$HWV>V=1%$oPgdl%ON;hCA&FMZoxb@BMRX6I+KSG2fKdw+l4oP9F(+saBe7ulZT
zP$<vQ6NyvSF1RsWeSO^C%I|64UcXpcb|OV#w{=>U;;#>yE#f!*Gt=DXG|OI}U#2EG
zaq9HSpr;%Ux=ZzXSzl)`b28qrllnR->+4IQDOR)FjACUTsxp^dc%BqsC0QnAJ*lbe
zt|`kl$vGyBmy!xE#2cTI40$B`<D6tHKg-16?RWW(X|6P2iCD4lmW^*#>>IwB3+Ak_
z<zV&h^0;rp@=-x@%gtAY3%%DJ-@ZtHQ^0Q-$^IULa}}>Ma(xoxa^Amq_eJZ?;+?5M
ziY-s>)&^equXb>|ok*iql(Ncw<qct5*2?g%X3S*`@!fa$dQRuGCrcCO>ui}~k=oMc
zai*)+NBa0R@x2<|nPDa`ts+gPe7G#>wRfs|8RxUzn>?SV{>)r%^`YqJd@1h>ETZdt
z|23&pFa4-<=<&8+dG#EE40(lzYxSGo#IZLEm|b3;HrH%DbNXDf?2>l<dvltL|9<`S
zba%b{3!@Wq4{y6`E!xepX{PCGGutD!6WF8r*!LWFJ}kWNynuqo3)>~UY}1%|zt(r<
z_Zyv)GV&<jFeU!~^PB6>|NH&s$!lf}jz>(~e{+K^@9b4%kX2*)xS;G?3RhKUvt4lh
zi7ER&GM9x!D6`ggHQ6utWBGB$1iATld{ye=8{FO&syS+|=5;M%yJ$Oqj(^=ho-=HI
zr=A|T_D$7wT7HkuZRLFL@6TIKT6_u=d7P2(U_!BoDpSkDTn&}(tB09wIfU-9C&%P!
z?5=XrDft`6?|k#>VUN`B3M);Pu3u>TmEXoDarJ$HeLlYpC$F>V`NqDB%~R#buJf5}
zL0Q|^OFG}Y$ZXLzBP*w>WXj#wS1)f>y}IJtc9ShLkGMJSKBx4Zhh4^KVn)KR%kBJT
zJNdb-CuCIY{pZoi#K`wZvGZFR*SVk{)&9my-nUoZWt-1!lQ)fN!}SACZU#3?E=#`j
z+Ss=4M=<As=tP09yKTQdW*4mJ;lK7l;C`^1o$$qjH@(ag=Iz&g7ktynNptar^WWaL
zpRYO4?NWE)ec4_=^O^sWKb%lwx*+)eI=iE#tjM((_T>!zsr}32;<hB#7Zyrvllx(_
zG4c0yzg)gUbBZHO@Bi;v=6<gE!#t0*eiEBJ>OF;L)Y<=k{6AXt+)2!?jfo*qT^rP$
zdbL3({SF%l?ENfS?87&=;>``e9SRO2qMRBgG6D)ALFQ{uTvF5Q*niwpYT7D==x;ab
zHt%`fXewY}&0(^Avg4eCo-Fq!FdOnjdI+3-`+I@qNr6Q-n~SrbEi<Xi(fY2rt6}qs
z&kN?usRVGcGAmxVUiaGVhxwfBCsBWz^j}ql%2ve*9iQ<;TE}ZTllsRlo7iV*VMnG;
z*NdN|6kYlJ`t--?$DaHtsk#2{_~C~yvPACgc6st((YZ@zrOr+qyv*kUCuLps@rlvz
z-cx%&QP;#Pb*K4;#f$g-FkQgG%=z9S#hJCmX3B)d313RR5}C~7t$u0L9Icl1ye(zb
zb!V%n;WHnzr?Y?FnC4)8sJQcxTGofhSA4V|^X-tg3gdk`<4m7;a9Mb1V)@BARll;V
zR%>lq8>(NOJ?p&H<B5jzmd%aLd8M0|ShZ-I?u44ma8b5`>V3t5YQO&l3Tdt8`YAJ2
zbjiwHztXO|<ubkf%3E{W*!tZ6uRROaWlp&!!Q$9`-S*$J2g3S`H*sUeqoIKTA|iuJ
zlS(QJQuRYB3sMc8@od2~H#aaw?y@ee;*1dvzy01t#dw>S?&Xb>!#uBCTN*X}TGrZ#
zLgxUNRg0Q-Y-n*+(BZUhk#(A<(o(D5@Md?{sbv{gqqgd8&04!HXl?G=)6w^DtxvCL
zuijtHcJ!A2{rBJMzKhS1PdQgP?f>Tai#sGNCkb@AOxpVTO}O=q^96DHZ=SSOIB@s0
z{)vZozdAO)J9N5#hdF27o{rfDvmUH*D+n_7Gn(|jBmZ~!wf)m(`zvs44v#;2UU1`j
z4*&j*KRoxe$_Jm;<@_wz(qSN;+PP2UuSS(g`L#Lx%M!&dr=OXz(?-KW>&p4#;i|KD
z7d&*D=|1JnN$HzhJ1?DgSDf8c&pDl)g_Zr3=&Hy~yhS^Go@!68{LOkI_4IYioj0u4
zP4bO?`a&mKHT;D0UFH<OL^;N~c}DE^{yUvC=j(mu<C@B9^UwL2<dP4cf`77VPT8E|
zJ=LgurcuPZm+j^!m2ICMJk&jNQ$l6Ng84z6_0xZ9rrIrh$2Iki_cE?YKQ(Qb5A9+3
z_M`f}LaUdUj8w~g_DAM3rUXw2TKa|OD|gsQ1NY^k+xk;Oe(qY$vd_!@Pn6h`-ta2c
z$~MkhHnq!(z6WchYE>@rdhl3rdgV{eX68GK4dQ*YEUx5xv=}U%f9SdYJLA-aMe^SR
z)-&uZ>R8usYW}a2^Df@9wXjgg%E-TPVb$_Ke?z<~8@{ix&;4M(nX67!x=eg#v(E?d
zqw)KdIWE2sT-15wilA|Pq?3;9Q-h+tJGnPD7fgTB$)PkMKtP$JOTu-JOOK*XM|j7^
zBdc3f9fj8m@)-72%uV3r{K)oE&2ZiX!O0ybk3=82eMBTl%dmHb-yY{P3injg1UGkD
z9*KEWn<V#<(Qtl*Lybp`%MF3k9Z!#NC8;KQuTgr#QJrMEv8_Pino!>nt3#bXS|a>&
z6wHLq9Z^d%-q>287}L47rHe!98i$;9(~r0iT`38}x+B*Fmn!dauQ7deu;YL3vy5%&
zPo7L<KYQQ!f5@MCI~r<cpHb=$W)ZdPk}LhKI(z!Ny*0HZH)m9eJ=R&IqqqN&Y1z9Q
z#_LbT*=Edi+V3b|k^3@ReP%(Mym?j3)7R_wuS<w5+kV9B(}&(C=5qUX{yh8s*!dX_
zz8Z52PHTl~O-wtTG4uWrLE*{jXJ?h9Y&!iUcyhjPUCP0~!IuSc9!;C|&mnDLy6*qQ
z@~sM+xBt$x{grxPF6a49^A!i>YZf1Nvfr34y8FuQgY$~ibD7T^KEXZnS&cL&-z<^+
zJ-Je56PASk%VJo;*p+vaxAa!N(F(Sr%?JJ0pKQA<bfxV`$~VC}pItw{t0-~5?u)+a
z6nX01hbA2{-3Pfdm%Ncpoc(A{kc#8mC;QHbRBNuUnO9+|HZPo)VSkpv$K1n5_U?Qy
zVP$Qgq7o1&926fP9ULznBd)UbL7n&(X$kWs5i$0$9XlFdo~bMJ%v;i567XI8S9IIU
zay$R|zr|-xc`mnG`=*{>kh|K8t-D-mE7B&}2{QfDGmCE9__9XPESWVYPX5@l8KJyW
zmYcMMr)?BlVaOYJ$8$2r-s@>?YbHnvOUQ^wL`a9+4++10Lm)(4VeON3#%4u_)%FIh
zpK#@a`WLOwB1RL>c&}b<v|g=zv0H5Cxr`N4C9J>r26%=}p7G*<*W?$1+q*66GSBav
zHv5&+HFdkd&8lXl6ED@hK7O=w#{09A&Cgy4tP|e+r_96uME}Eh)jHn$hkN&QKaIa~
zXT{rnUusi$tt+P;cl-H8!1^m!1k<KhnzqH!Z)fe<_j|^n+ZT7uPHV5Me|AXx>)GDD
z^Y7<7i7dES5L#>aWX_*`$Cpp8@8n6;uMKz|w)esot9;fPj?co=TlD#7N^dItSv})`
z-xn_7IU-YcX_}>azIf%ICZf7Q_UZk9`<|EF4efNe_g`Bu%k`hn;hiRnE#sB=wiz#*
z8NDQN!LfZDD%|hP*QF$gZ}zkC_`81gGxeilbJP}<Hspz3{l2)X&N#$q!=#7l(Vy3U
z$P9Ym9_o`>np<isR>GxW@;KXs@9CliSNnf(Uebtv_V3=y8m{;1_4VJZ<@hJE{oHbY
zN4I1~+n(G>CxuwQT|QPEBR^T)?F{E!KX$7LDK95axpYuN$V^?uBHN6ovQ>Ci>s=*z
zu7yHTTSV^0Ey!)<dpn^x$3*3s^Zb_I6MKx;eTx>fI3BOG!N7I)j@#k77xnEg9h;=3
z=Ay8PPdM}NzN`1zV#O`9jkFZ_y9}<{)T;U~%f0+5`|`!CCui>cU;8k4hm&H_N7s4V
zqWIPcuG72tr(#=N%#F~ZMQeY_c;wZ+vWe$1d*@wQTxX)%aHa9wwgqpBr@i=AzOMX5
zJNpikWq~neI{TlNc*}e2v)FX<;IWg5ttOH}C0Vg+`g-qAnY(NGmOByLLch-S&hst4
zSGHxg>RmRuqm8Afyh=l#TzufN?U3Pd_0vy1t{i+)wu8^Uysm29(#2mCN}T<V)Xjf&
z!6>xPThG;gPu;48+;)re^#AslZ;oQUt-L*N={$yBJB#Me(^RwGEq@(oyeU1$BVxM1
z^I03qIHIgo4z74pCsQaP+9_}`ns;4Wi<5=sPNzCH(OXu$r4viM`&%MpmkMf64B<Hu
zR)6`1V1CMUw$HiE^L}a-r&t~S@Gt2}f1SDbgBLFQ;{^)!YxCPO-p%Jpy40V!F<QDe
z$4&A{!kKOb_xNdPDY=)nXU|Pkeskew%Ce;!t=#YO&GCAkot9BHBk1P4cefVBEqFIK
z=BV(Vo3pO&@p#3fms)6gdtdU@!qC^;f4a*&UL}23m2D_4*!b`Ayz~8LbK6gg82x`?
zrN8YKm))za_ne<)x^4GgBs6)&OZAX~%RRHr4f;Ro&3*V@aARuwleLx772FTh1O6O(
zov+N+yZgViRKLe<ZQ->s3IT7QOq;Oes+vl;!ta0VC*Elstv@h7eV<*9eeNf*-Tlw=
zcemV|;FO(KS(lmW{r#QXf;QO?zqU^+IkeG6Ej-un_M4*?>B)1nHyNI*HvK(u#rKPm
zCQ@r1?{DN)@n7;g%f2b~!K!L?W5Zt4Ge=enDKa`3w`U2Y%b&KKI`=ZGP>N8s9?uV#
zcNR-RPBUCix>->8T69WL-D%TbW}DpjCHhaiXxUJ((<aF}XW6}78(C&P-k|cY^H|iu
z()Y}JcJwb=|F2|Krv02HdFw0;C)QWqo8%d`VO7|_!&O|97sSlC>{uOhaAwqlg#Q=+
z^s2v)IJorJy3=zjqJN(6abEo2&cbEx+n0MPR;Ir;(vFaSwe?o5hm`QCzcPRC=osB~
zS-SnqJ6D4fYt}A*m2v&Abh{wu3tleWe=p4EJ&m7jIJLf6eZ5Xa%2lT7&!@Zg+Z<H;
zd~FYhy6A7&5Tni<_GLzMFUUl`jC>aVfa~_ko3pRg?K$!D(c<>n<@+~pI<QjCRne;@
zt}>9zncvOsbFA~@fGFqk4^@B8^QfN*|DSs(ruAWUo&Ix`E7KNjo|d7ssNy1HN?H59
zss`%=SJY=$eae>Zws!n<_2Jv>B@ZN@zTa}|qMn+>AB7DbHw7XCj~tt_T+E5nM0{FZ
z^~>M+tG*v9d^PRs&h2Z;8uBkK6X^OU*DLj=#!*R5L%fFL<oCkOzSFi$oNf8w$dB&D
zWyh+PAIp1me8Zl%5582mx`afu9f?lvbH4eE@7o5UgNM0VJ=9t}zM1`yx^go>sqxSv
zX3_Rn=QgfzzH@Z)(swhnJSI6f`FLft-40OO;JDuA=KZ*TCHv<qsu(Z4@;&<G109n_
zM-!_aou7v-4oqC;we|k#mUr71t*LF~op$1nUf2KjxA`SeOusS@^gUcE)ezBfWJ<-#
zMlO}xCU;-X)J)Ko44Lp?u8i1psYxeRc~k`*Vy$gmCzJlYM&rQF+R{T`c@9+m6q=*^
zI_F<b{p-pfLVWHeC$gWG*6ox}|N1QS>9wiBU!)c--}#H-OQ5~Qmk(1!?Pu>lcwx5C
zw^zjrWv(wW_uqX=?UvOCeQ(hPm)g?;@8~PPH1C|XMl0sdO8Z^4c5}9$>lOE2?8wVl
zbt>tA>Z2D2X8c`r%Ki-7z1QiJK3%K2<)*{Pxqqg7`{CoPmI{WO9L)ITs;WGvn@`&1
zJTuod{rPoW?HIPt4-OYt>&NA6&lCEPA5_Co)jz-K(b~KNQ#Wzv&M4dXY(aZwLP_-p
z0lztS<+mIz6ke<(*ebZb$*8AFGHmY`scU;UE@T(1|9P4_KGvp#<43bo-yTt^!@nbz
zCIl}2uxX~=t+}`6-tf!2w;}L&_g9I#wX^hk4LQa57q0f;>h@grBF$vh@3|`tw@L9O
znyg>_-91cg>hY<rUvt(*@dY=^zx>{~_p{aY(u;d6a}F!Cqy#p_N;9$Uo_DWb>N|It
zZ1Hrz)iM?BF9j`BxW5&v9kjANs(Uy6M6TN0tJhz@I>1r7ft`ChkLO0AsS&$8dz-|2
z%@1_0SzK8=Wxa*6nzG;QdAovT1*W@QT=!7(^rdHq(mF0_Z_>EpbM2*c$=m|TEYnm^
z%Y|z$p4{WOW@mFI$A^Z8;aUN6wnfff;_2l-%|(?x^UXooKP7clTr*D`dlPh*)q?N%
zrCF28mP=i7a9x|H`f$lc_5TZ2+c8X^5^#f6E{eS;#$cN78}{3!mp9zpsqAd}DKhu@
zLWYvLKmSi$rnuYZyX2Px8?PLD@_Z@hmA5Q2r<I=laMUSqi=Sf4i4aAXe_Vx8pUlo)
zQ|t(@S(bnG>%*;@wHKEC)k$9AlyAJ`=>gNuqYi0D{xNnw*fHn)`7~yp@Q&Mj7IUt~
zmP|3vzBOq#hlbt!|1;(vYWGU?I(_GIpLmYo!_M7jx(gNx^xk(?%odU={k7-JFB`MH
zzxML*=3Q6i@@TDWO26x|^4tkurll!doTqwze3{a6qO|baj1}7Xoe@u3=Lj~sEh@ca
zXE9@!`ahSHtz1uTJlE%C{C(Cj#F$0<`?c7p<GhP5p4mLhk|+Dx?>%WUY!7D{E<b%C
z`M}r0nQA=0ySOz<Pk%k{boBP5hqnsWvWt1>erqs@ZRvB`R-gX7p7G~OjkD*wo2I0!
z-N*9n;hsfe?+^0@%x<)PFE8)!+rM|2ENemf%>U)v{0`^z>Tl59bu#~7(t@3iEpB&B
zy`PvHEtR}^Yuc8}Z>y_S_o{RAGfFuZM*o|($@~OsA(y|ls!PhsE}@`E#Y=^A_`Bu*
zY(FZ0PM^`xqI$k*%HvF3apj=S<oh>PyDH_%h8<h^vhYp1{E3`XyZwK5KaSlZmD?4j
z?HRhnWyzxt=BtFCE&ZV1%zy5j)1#l$*WQ|9(X%{K{QK3|TZsx4Ct_P-9-FgVJEJEv
zl_?_cc-#xU`x3>wd498goc`dIaPMpW14_*CHWe*Ty>}{iw|Px^@^O2_%4Yw`YWH?q
z=w{ANJDzOz|3KzWk<En~$Gq#sUp?S!^wq0Tez<M&7Cuwne+hX)>Eh{AQ$(k~(Oq#`
z&uQBn*65J<fY57AbN{bizRqw{_v!bGtS_?InB`tjJ?Qqg{&2-<gU9E@4Zf-DJaF)z
ze6oUY$*r7Dv-sJ}JJx))3;exc!M7!fyo*0R%d8K*Fkgn}ah=h2i50c8bXWa1c)W~f
zhe3^s{p9(_*gxOP;Ie1Vb=Uf07}e4=(OlnI=bY5VcRW#_CTxGOx8nNEKN>NaYb%RQ
zINv&D*Vnb!9r=6l@WKfDdo6tLE9bZbzFhyMM#R9NFZ1lm^En3#o#zzHw7h;Xah{1x
z(pN>xO1n4jwkWOT+&}ZfZ&{OH_a2?Eh?>=A9k=yb^lq2c2NWa@wVbi=4_Fd>dF9E#
z!<YU`*={?0aq_oCXDrfBaETo@68+cZy+|S0+pk&S@?zoW=jY?Tar%1(D{x-fb)~{A
zKG#D(K>Vo0rQ*jEnfY_VjQ%Z2QBPHUx^km^iS(y}*-gbRdSY+4B*@o0`%K$4rQc!i
zn~jTR{du-(-?v@P1x>rQf6}nO)}_}g`1$dMz&`~)51u-~l4><?>%VTJ3GY}>6-*Um
zXOFwDf9r1lxi7N|rFUqt^`&fkA{%>Ui*3I0xw^}hk7ECoYVHvGb%lL9OUcnY;qks1
zzAFN+wuoyKxXsJnT(#+{Qp6tfwaPzbUh7|Tb>uns@wU;SpzZSfwhQeVeK+z%{_DFp
z&1w5C%N57A-`qPvKJrZTls7qTt3&?3oNOQQw)NYSTQ{D)_!H5rlx^#$#Oy3EyVUhD
z=jI1`^9+jF;?jBQtVG`|VH5xBYP&n}>#O()uRmV$i(R^Krqkx5r4!~a2~LaJ+?zD3
zRsTzM@A2u`<&Fa93@-24kbWd7_r9=t)8@R^zC|}ftsYnH6<;&Wq|12y%*&Tcc7HA|
z<U91Y_{d5|rssZ}ci)+#x4ChqQ`7JL?OgSpPt@0KdVc@e(${vb>mwKbu3ydbkTJdR
z<}us2-!^(}|GD-bZ0$O??%lgn+6`wmXx#H$*7#CQaMjUXk$GR;67FgJIVflNs4zcf
z;>7785*m(v&2qioTs7TuRojm|UCuRY_WH%CcbvAp%Bb6T@<M`yl!$5If=ikulT37K
zd-(F-Uk=IQ7FG;n>YTOXN<`h+1clSP3ufN(S<7go8W1oc_tHZV>F%X7vW0kC)C^}H
zZBv?ial-V=6N0<%J&n4#X49TKUbUB#Q-n6A81o7(T$^+C$?}&;Dhn8U!wzq5F4Xz2
z_*||mX!7JcN*~4B*Z1DrC&^!&{PXjhkaLm23(mINaE3g4dhJ!akYWBxPsZ}Q$E$qS
z`US7o^f>W#;)3by++5e2&GN0ojJIp1msgsHU7qN={*%4={&bJLr>+(?>HI&h?Y^=3
z*Skx{?l5M%IbO?-y7bHJeY~1u^3R4fF3vJXJbrlUuUqj}U*ic6=X|lPbACKDGRl7U
z=bK&Vo{lN{yanH)ev9pIc&ffw>A$dCY+w|p16x?cjI5vG{GBc8#TfyQogV2w>dzNF
z*Ocn6GdoYI|MeE;UB6@}o86x??<ecId#z5TMowG5EjVz)d;OKA&mGpYtW~@y@?E(i
z;3nrh&Z4Vo^<}1!lY?$VT%0?-_xP{)g!S7K&u!6|CNID19`Cb<DjRlG|2AE1q+yyO
zt!W~*_n3x_%#%2W%3s^$Cphh1xpm>5T|X`_TK{?Dl=esb#f$V^(tlrk7oGNEE+fC1
zX1e>Abt$haJTDk-(y>+yuzKNqC@Qz$)Tv1!W%0N7P1trsS>YzP=}pn2`|FO%<z($;
zoUY+=^UjQnr)=v26&0n;H}okl)L6e}$(nU_`I+nif8E1Q{+Re$?EL*bTUG{1bSwX|
zSmqz{xmnU&a&F>+(>EsGezx<r`F-YXhuTh*ewu2%C;ZTFk^j%`&V0eU+HObCyzrQd
zN7Tfs-G8Y6eDLs8&Q0GV{|a9`>A%!`K{!xs)>{n|HGhG5mvV1bam&6~?!D7;j_u~e
z@_;^lmS!W<kiY-T%4!#UH=4oUo%OcbHd^HFEBym3p)-&9*v#3y#ralFPJdi&=Q@U>
z4<4b7#qG{!dgh|H_RB=@Pdfk5Wjddq9Or6dp{uJD%&#+;FSu9!LvC-6i~IRM8h4q0
zUs-#qYX_65O7e}Vg6<Q%xi@8s6@H61dbP*2DD;xwgf(o>e=WFo;@Om4J;EZhl=d;O
z^N4>cb=?)Cu|Qzw7cTotr=-hP*-qiUTei0K<b!}WG1~XHzl=E`n$iAkPT!_mf4UBx
z(Rn2-ayaK)KvwOC8nH#iW=`d4sfoTv4>E7Hy})DoYSV*Lu{XJz=N9z+c`j)8&h<b0
zb|JsR?7!chmw#hl{^{SDwJdTAe{6G`^5v<%uEyr0n_j;5{d;16e0lG`x!MnPX1wbA
z!c-TutGwC2QQM>Jz%&c~r)>4D+7^c{t<#-8d9CctBR5P;{f;`O8rl|qz7sL4-@ET~
z8$;H)Ty~emk}sa_KJjSv+V-%hkDrgsp31s4NaM`O^()U9z4|(HX5}rNJM6Lz%1>p#
z&OJNz_XCa@Caot$MfSh5DBhQS_|MgMw`G@pslSo?es|T@y`OuwR9~3s_^di~wf%-T
zV^6Q*&(~!4>9n4k)mOVz)w1*#_o0_}w=TRb_pSJ}{I9?u``8(_26HXd>SWxG`Ul+p
zw<Fnl{ram{xn-kmuVkI)Hr)SLGizVU^PBh1pO-E9d7~@;c(%Ry3%T-gSBZz*Oo!wS
z-O{m8c08vdwe{nV3CS(XTikUwreB<EKld_|?4Qp}r#`6}&A#d%uB`5NP9-&u^S;b>
z&9CQwFW9huM(NQFk;U`PUz<I&Tk+xNZ-ph#Zu>CjsoC6^87b$Sw)yu<>lc5;jdr^p
z+Hyp6;_gSGt8NtB@jLQjaw%)kH<xGgx@_B{cAq+Tz_hFGh1rIgfh_Fonpw@SrLwZ5
zUO8upSk%jOnyV(P)sg>t*)LST;m1a{;<oiIb6SGiZuifAl-RM><5tJ1ipP7aK5GSW
zb_M-+vzfN<)UPdSB5tz<K2La?w^}#jV4(ffn9av>+rtkSsZYE8O=eRQkLdrnwz-Ej
z!uCA4_IjpD(ntMIov&SH=={If&(tb^E8y7Pqg%S;Q{;|UvoUiB2WQJ($=EOO=lI|C
z|B63l{GO&7_Gi`pJn3o6zp@_u^~UT?S>NHj>sm70b-HUxt9M;%{&QS?d0O<(M>F-b
z(oSk!`@hD0t!{<0z`x67yuWQdqis!p%`a@6S}h}Kl{c+zp=_I_$$#DB`&=(A{eLf4
z{`l0x1u>@ozZa$+l<mlPCT7|_>)nET)@ds`KPUgt48C{h?S~_`jRl(177ErcD8DfK
z@_X*`pxra(pZ9-y{PUA(Pvw2L?{9CPanUyZPebfS#`~8VCbcScmHu8SUwUuR`{#c%
zw>nmc{X70(Kf_P|6KyM&P4QdNK5NY+8<l-$-mQN5kM~@FMO)s(Tkm=8It@2+=iaDR
zvHhpXQo;~*BHADz<&3`0+Whw#_AFQQ)!(XrR(@+YNBb^^Txy)Gg{G{$&_@gNWzj9C
zALZGw_*k6&kQuSz%amBj`5aBQmi{8^{PffQ9Ce@UVKH^<@)G|PyVp`1mS&~j^LyBF
z+}zSEv(b3>vE=Etd6qw}uRrUayjE+0$!3H0DwQ+$3QtXV5V16B>Y^j57d~2>u=QG$
zJ)5^hJnC_@XoT9!3C;rd-Fa9iEPHkT*Ur1)zoh0q*zi^DVafy1vls5(Epu*TJm)om
zx3cK|!noCJD>}44F4sD;Go!?*EF?|DgXy5XnoUWw?Qdz%&F3o?mAY7OUvP8NUvK#o
zkNrOPUph>d+;Ejet@@kmd*7FbHlKN2zOP21_1KbC9>E@8*EN0j7N5p4dqKC=m8kDK
zMQb14NsKpNxvKBV+zW5bO!BlopZs><(%pwEMWvs6C1x*;53k7mGjYb}jdMGnKCGxI
zDx0|e!0l<ddA2HV6?89b{c(8_``<Kv3-7k~3e(Mg$8SA!Kzj40>b?!HE?wAhV~fXx
zilZ~CmadT4>-J3Usrr&x?-djGFVubRQ)9Qfhc|>>MT)6(qJYO*74hG*+8;mrbp6M9
zbNh3joxXVpM9Ehq-ivi`iinxuB`Wen{yFPDfhqFAa^aIMaj<J%akkrkZ&u6HNXFEK
zQyZ(3`nlDm=WbZPV%x7fnqnfiBzmP?&)c*&?}(`QS|bo$usp~|X<cvhxhoT28!g&m
zSW(t<Y3=1;#}s$BPBV4R?GFVe3(3g8u$$N$euVv`{gZe%*_^D2%K|KJo<FO1@$0Lb
zRz+6xYLq^vuRa%g<IbeOIf*`J`H$)uSX})x!D6+^dbTe=d0Ta@9P}4DZQHba#TgZu
z^YNaBg=G<KJ1q`~f09o6d+pOE))n)vK6&(MKJ(F5%d5L;nc6dFIJ5lYIa}(O66+T0
zmgw>|>7@P7+t2PkV%NA8F=7AiC9X-g`j34sKfC+LO;O8G%c;xGYx$?9&2=xn(N@g5
zbZVPa<25HA?*P$+g^3=wvT_4$Ts##NmF;}r9y)#F$%ZE!|66T3`8E{lYA$;cB_q-*
z6<nfgzE<yIis$J7_ETvE^L}b*Tib6IQE@HhK3WkKwT~t6*_Pm(RHudU-2Sc`j!o$5
zG0IJiiJQaH`)vBdKU<YlJ?GBinDpYo>_=~EYSSDeJ!3sf%g#Mze<S~=CbrFT?gZ)S
zlV?nJx@cBw@veB|TcJ-I*3?S<@nx3G>bQRNxr&H)o7I+LrIHtF!rRmrYON~g-INyW
zeYv~!_WM6`wC>6|UN4^)!S+Db@%pBoESuG$zr7}O=&y^Ox?HDt;tH9+;uSkjIyUvr
z7moE=dDLt6rz4ZL6it}Z%$!@Dw@`MD{_gk7Ki$f5p6hV=muB5Hrn6Ex-HTK@COde!
z1ywzooP8<G?^2d$ctP||(F(J;nSv{(GOst@AQ`>o<H~Iz%{v~{q`A)gXykDrrSQkP
zjL;48{|(=t))FneID76UrHEV4CMt4oZ-}T3S>LGj_uU&g&#>*43oK6EnzPNbzsP?k
zV+Ds`biw|G$uWw?r>j2Bkqlp3zw`ImtGoB>{CgJr@W0A}1;R=XPtK9JQsWfmGK1}n
zPtxk?XFlEkcmCqysQ6d)J57&gm`EA9_rE)TMoY?;O)~Dc{Ts=12G}N>jTGPq{1QFc
zY>IY5)7nVRh~V&B`vfOfZ|%~(9{ql+pU~E=dOgV#{Z?PQ`)J*cMsB5Orp66wIgBbg
zoSuF{n*<ef65S`La%gDnh+bQ^aa-`(yV18Zv(tBdZ;AUq^MA~(sO;(W|Li_LG2}0J
zZ+ZWB?fu_-@0a)4<b28KIsVx2wv$tM`?{H%le04lmc+<ODtq7i@LcrdCy93to!&do
zDCau=dwJoEaEnvgpFaA@oc*<V`To;?^929<OPIYn%)aCOAI>}FPyW7;p542zMX^qz
zUgX%}A2!<WRqy%Uo8EtR?)M))DM{DnSpL(V9K7jr%Em&kkICF(Is)#g-SrhpajJnQ
zk|rN9zQ8XjR9WP;NX64Ke2=e^p!Q>xq?JB;g6r?d?8$z+?CPVLtIgjEte>Xhc}eN%
zqX%C!cIrg!aTL?_4Dv|pTzWDy>F4+T9g8PVPD=5LESi+Gu41Q?jAqypk432&6IHaM
zCLOZ<Aimt=l29ra>#s<i;Ev?WN|UEdlGtbzspH)>ZTaznXBY1;J?ZIH=oMvDHLqjd
zg0)SPMZH{0X6d+Ux`{Q1v)1+pN{X+1@<uD=#YT(Bu6sS6eSXjP?ELH_sp%cmF?EuT
zYP!0YQr|?)peGY=p3u9crRr<Oq!_E{-Szk6uKgCiLeewEl&5RBeS5m-)8zU`yZ7|@
zcuq|3U*juR`MR%mW{~1U@8|!fG}})T{o?UQIrfP35t;b9f9gDsfA$^vcc$8B!-VZ3
z(RNxK(T_I^-ad2NN^6qGgXMw}(>i`82^-k=98pUW`KDmTxp-sy2F)1~v)g=+%N>(D
znwGeEqvZ^HGx6E&wryvRGi^M+QS^paj_5ZNv6i*Rt|eq|EWN>&qr8ptb`NWE^oHCU
zWjX9(O|41R8)I+e<?xH0xL6?aO|}2w+2>pPl-xsqp857t*0!|d=L_$eIqQX4xa^Lb
z3$T3rI%9+3^5atPym!67wd2mm#>c;Fr~lYzu9$l0>hI0(_J3`-rGKVhb@IvE!q)?m
zvOYM@XOC9zRJ*BG>$CcWzr^h)ztyhWncTY0T`D(Iy6$hz*59%g)(?XB_b<?oVNc)O
z?-2iK!>#D&z6-iov@bvAR=H(xV1u22aD!ZN@7hD}R&6wIunJh3d3xQZi}N3?y2KQ$
z=fQA_$@9>s-Y@fBJj(eXBG|Y*Eq&GFVh+~2!-vvm7i@@)*178bGE>uUU4V1EvBPYE
zl;&M>3=D;`4-Y=}*|<zd+T!p5wHM8|vwps95=*}pCpf#YS@xFFPP=!<uTGiizw1g~
z!JQ9Hbtwz)EZN$8`HbKD=^~Mxu1C~e+{#Y;`fX+X``07u#V-3k#7c0ve{6eD={0-p
z^(V(g(_eCJ-T6&PDvV2K?YVcW-W@W!Wq$u`bKQqTtst%5o;Q83Uw<$8{?)U3k?X!V
zS#bk<3C7G#e0d+^miV{{8!+uY7{Vjed3$!@geg;%FRs6$*dg94-fMbdXNOs`tcLr1
zw+o31Cw6faTrtRSXfU6px=cqrBl*E<iyeu(QrZ`PKAmx`oWJ{c&Kc&+rID#CS<)}B
zEvn|Pd&@rGHBHyos*9)X^qD`(OqZBLt!kB5Fog1~Fb|s%<)vd%?Y4INVbOw@pU+pu
zF;1U(o;fb<y*OKQ!PRRWIc1D*FJ9e~kuRNL#Wdsj@9%5H8dD9u#r4y~H*#mDc$Z##
z{jXhq4@2#|hVPTZUp=%rC72f$qVwwfy9+!_Oy}B#71`9g53KCj_hQ<XRXo`{S{3f^
zym#A_K~E@OuQqK~fpy2Z`$GEZ&o9L`-O`@dEY4W?GT_d)uMY!PyB0RBd1|)yc2xE%
zcAm<ow=TR(?Jm!YSL|Zm)ODpQH8w8xm%>NqIS-Zde;w4aX!~;I5Ua?Qw*n1&XRZwT
zTcx+l`PkxPS|WEJG%H_J_2}<jJfUN8z?#UUb6HI8GcTLxKVUPix2thrSSLAu^8xqn
z(}4}aHw@og3Vm~A?HiM4OAk-kdMD&nf%WXUU&W4Zp0J>C<-XH*1K*Who3fMRo9rC!
z)0H<qsq)uxAOCeA@?IX7U5=jfo@)2=5p&zL0ys}eJ}?W7dgcAb<oA}ed76iJ96ci<
z9vLPj+}Kme!Y;|y`fX*D>D`OBLhD!r^5@+7cx3W!r)eEc2hKScUKif{@|Nl~lk6t-
z<*~*FLTN&u9)`Tw{6)2Fj|P*9Sg!2DrYsJ*y6MjyJ#VyZs5pAqc;(Jt7v7ZamW$mq
z`zd1>kJygsa(8a;{haeEbPMm-f4UaDk7mbz6lZOo@x^xboB5&ha_nzaKm2)ax})X>
z=Jtc3uA8%OdSChJJo_1|+M7GuB8z^0-hTVnJi)&wY}jR9yI-s1Zd)GoCPdcWa{}8o
zN#?Va3a56Qj|gy?eO+?$0l5kM4T*+@9trz%EFW!=XXfz{pOSSWczbM^>HL^7k*T|w
zZ#e8+C?(o>t0(6+@4ADGFD}=x^)-3z`raGm|2*@Mx%|22;)i?gYdBV9e$hWIG|%Dv
zewODy_nh-v7}v~SGUt7rZ>Veg(_`ZAPH0C>wMdU|$hPM_dicirt0e}TUfeA^%`a3S
zXI=L~YK~7_{n^zEZ>^8gzIA#RL-8_wGar^mMNblMf6EI#+H5f)C;Xa*wRUE+XYj(}
zCN4IU>s!B<#_ji+ZxwWz^PJH3u$YvehjbOx>{Y_V!*%@Qo3mJcZT|1XpfWM=ZsFhB
zat5C3MTu>*&sC-G+UamQLB#X`*MtRP4N7cJ#8zEx6X^TM&}@X3k1!lIP=DayMK
z&54=y{7!vC;_1EdAHq35En(GEeYIysQ^?v`vDMYGYr6tdm%bL>DLbo|_duS?iPW{n
zqNeb&uV7s}N5zt-bV2LnC9!E!4fhL$7#~Q}GQR!$VTRN_e*M7sjb?VziXUYEb-(jj
zA-%6q=oycvZQYBTAv=W5rFdG*zQeJe$KF+w$FWUj0{f%`agT%5QZ6r+)-cRIQ1k3t
znaqhTJ9oNr-rOQofAdOC4y&HZ|6UHq|DMj>2YWcAIcn`1M1rQDQ7-h|=xcOb>GbVf
zcCFj|0#6<u`*kYsf^*G|j~w1yH&4FV9WuM&P5Y4x``nkczn!afiZyHTDW8gXnY$Tf
zitHCBneTc$o&DY9a{{VLZ-TBC-}`?4Q{RD`c89a?E&g=+`pIi6uPxlSZIbE)_2>ti
z>T8&pUDh`pTyuH1PXFb0mUGe1qTKtBd7giL?->sZOLi*rgohPjoX<BUak$j8pAY%B
zmS6OlX6*Vayyy83R{4u%39KkNS$Lt1!#v}aL0Oadfrm-S`)Vcc>aJ+HY+}}Fd0T>6
zLUn?K*vs!ymmJibUVAusrl!yA+}-@FznP)gZjH;T#fuAzL}mIVE_BA_S)DLU*wpvp
z=;1TfXMY8KuwyoVymnbuvaNz2N0r0WTXVARx_;*>5n%QU$gJz_6w7e>V)=XP#D<%i
z9qy5*Oa%8Gu9rSo_fV(g+pmKkH5Q8t{>pms>08i)Wfx1<TuEM5U3e{=y@98er$IJd
zZIR%yWK}m?jveBU+pHz5tpf8~9$WSV6-TT*$dYGNpq-Q`%f-o<&SB*^ci~Nk4bk<<
z%?IOe9_U{4SAE3|hM0xtnS`cnm1N@N+934vRjNgNd*xo`E3SVxKjXKOtYKgLG~7VR
zdd-`i54Xk1r5|%Ydn`aym9JTL52M}AOZA)o?f5G2W7dar|IXX&td;wIzPDR!mBTvS
zt7|t(ICMs?U1<4m$8pJu?xw^a_4nm>aOa=X)wSHtc5H*%ES`Uda-1I(I-Q>&$9+k8
z$&1<@S5~Z<P`-KVhf1ks2VWff#dcejyO`ttqVr}?nqns8-bqrrR4`$+?immB7v5W<
za-%jWt=%x;!(YAa2ln1qemY08*Cq9g@jdSo^Cb^!y)S%K_Q2obq(aM=gVr9+buM1#
z_O{EvpS_kjhQa2C$&>w3wG1=z{J6Bv1Zdy<y1h2jjA7r@`%`Ype>6PYcljy9Qp1ZU
zo-EQhD*1lxzModU6`V$DS3gc<v9I8ke7;2WMDMX3kMG>^JU3z1VeeA0bKf3Ygxn4L
z;JH@!`pOkI-AlN#`ln5+Tvr&{=C<1Z{HMNyH}CO!H15%EyB@q|+T5-$Dzkl?Cr@Zg
zJFFB^%YRU3#v#8Zk=cq`ciiSO{avMZbEfRuVusV|Hy)^Nz9E@$q-#~z+k1sxb=<yZ
zA9OFf&d$d7b<Uc-EF5_&o}c>pBru}deUbU0_LKV$o&Lb~^on%j%?m3wOK9!<_~2T5
z!Jl;Lwg`K{%yZ%!Z2qb>OQjSTJ^5d4Bw{kDS<f%=cD<(aJ()YHYYn$EJX-tjLxbzZ
zh0-pr8e3L&?63+*($R3ZGPOuy?+3m^wgplz-{0lld*x=ETX^y9#&2HyFVyW~`!{f2
zbzn?s3pvc>Y-JQV_sqlDf2)7g#WxA;^U7YkQ&Pa(zeQK1al)1l8m>`!v&!agw3yxI
z_08CB=R@X_9R~9ZEE?SVuN}L^wWg~%rXY0r-uSr>V%-*ey>R|{s)S!-s%qe?wd;?>
z-}tY^^)joYo<&})arQ#-bhFJ(W>uBBPSzHVF_y)D+CGR#oOm}eO8i1C(}y`vcRX&`
zHYe_~MASC1&MSvzc1M_m+1C{vK7KMl#&dq?))NQil79Yr+s_@|T$=kS`RM!0G0e|S
zwG<UkHe6Nt-sk;(O$)aR{Bx?@H#$2;H~rYlblJ3c`rjRFvTS@kSL$WNH!`R7PTTh0
z@aW9xLVKA`izaNHGF!LjXjtky_XEf8oD=#}FXFenDxcL{uCMa!dU09l|19sF6YYO8
zExnte`DpUL^NZY$_-Z9SU2v+ParI=E;;NO~&J^^lJj~0^yY_{}l_Z;tH-=#a#fpW+
zpI1sA@(@@$P0Ly6g~Fb-JL*p4WN+Qd@Qk&<;M$&qI<vFCmbM*l*e|rCFfsm!X`<~4
zJ>AX1@&^k;!d97R3kq;X_O`nuY&ciDwQRwhXPdr0Vk@60{qH<`f#>%7zAVO)zb4L3
zcvD~zni4y6UWmr!oQnn<e`v<;UATpT#d!Nfwe}Zg*-o#zGm4HaY+f;?f7#dP@^155
zwls_VZhUp`*y6Wqd7W!zuOBRB`lYkmFjRth(v`xKv&$|oT>RSFYSnLs!oM$`xw|eY
ztNL+E^u)@RS5xGU%uTo&`s4iSnR)j7R#9g^`?Wbw`MPQ8;ld?5n$*_6S$j&O*Y&N|
z&WMTPTFaT8gvuHnn&#Fk3;2oiG4?->2)-M*YlF|U>Fa!%&aCk@SNNIu{6XqPh8Jf|
zv~y>>|M{DuXk}5U9{cRiTX*FWA*++ut919j)PLQberxhv&1)tHR2r5X61jTHplY>B
zs&>VkgysKp<z$p|7(zA9#P4;^p8oNqg5FMJy%|}(ekKR6OYHb)c%n^YW1qDA<9eH>
z$$r<qGl|Nn-QB14hwF&ld#l3G3W<{RGM65iq;-dmIv3oEoM$d1w|rl0S*s|=V~4Cl
zFTV(*y=H2e#vk1K_Wxs<qiVx$lRDk{RsZFm>o#0Dawy|KOVi!8neD|<ro6wBS>~u%
zu(Pd6*XA_a)~h9$;m~DxeMZ8Pv;TA6*z%uiJeG0!%*1WyTc1Am4=i?M{CGg=z2d9=
z<&y<%iY6W`W%X87Z|^TCuH2z{bj7BZb9(M2I;7V#nH~IH`245PZ>!i@id+l0td3se
zJGfVVZ-L0Fwm-A%;~3Qss6T9}c;Q_4HuqcKGM(v*y*5pFV(9v!U}|zt7We9i{+6Vq
z1D`UL6xMd+`>Z`%B3kxz{Vhjv5%Kf|g6iAUsx$r{vNf^n@#9^o(8^I$ZN21Ti`_Tp
z6MJN@Sv#)~ICJHwQm^_G%hWu<0?Si(pPIHAwS3^(ap%Hy_X+p6Z4WPLR1DerRCOzh
zw3)|TD}j$!R*7VKWE?uMT7UbuIbstJzq_5S@#6Nh&2|~S84eBw|2Hc=Sou1qIP~T&
zZ}$E9PB8~6-xt~ZVYzxYOXORYhGgHuuw~1|j%JsMD2wsEkoxk>Y2B+wb!FQ(CqBtq
z{&>2%wSnOKSFbHT)nwmsQar`D^Px<T<U6qi&E3DNckR~XFn{=<D$zeHJIi}Ywz&rj
zm+!fMuF{zmvfo?Zzq0wT`yJc6Tc_XKTFhQ5xMzLhBJ;*7E8B$*fkG=&OB}K_%mm68
zrc8eC)@e}3KVc{Lg6Zv(?|lt${`=`uCwFIg5P!_|v%h(M{W|btV#J>=rlWu7hb@Sg
z{Ksp3ZmHGxE!y3Ur>$5em(AIfW4>E>*@d#|=-Z8A%X=%!7OdKH>lKqvHUIAg4re}1
z?V7SBw_0NTliv&#`(Hh)J*nvV=ds70`w#cs@0CCK)$6;*BJtY){cG~R_${ncE=jrJ
z(X?9nK;mIny@Q&|Sb|UTbH01=`jzc{tw@~|@x`X9uAEgrzi01rGQD=>>wmW?uk%a4
ze|WS!wL?5ne&Jj8b{?|}%lEen2S`-===89^loCD8_H^fqV+tW}?f?AJTDQ?}x@z{e
z*E8l@%z3%bIq~uP%CGVl$|g?8{g=F@R#biYoz}NgUhOD9RT!r~he`RNi2RN26{pY5
zx1T@v{JC0V0XrsU`yT@TM1Sd5bkEg1b>DgKot<5^Q@;B0#qJII#Z@EGmv3F$`dH=l
zmjfN*52NL6T2~!>zggbuNwh)U5|I>Et-I%v+;Sh>k6q*W{lX%q84NW6#|vhun`g<+
z5wAbU{&`hNZ${mVRh1vNY`v1Taf+S&mA_Iy8WhrNHa|J`W&W&Zvn@1J)$iZ6+__D2
z*JZPbWrxib@>w1{*I;K3_~tw}?X{VR;r{Riq6c^$ILmC32yXGrW=)seA+3IW%lA?l
zA+yF=f4p3iC371?9ADN2%Oy;54BMxYbzkn6@SN=H%8GVhSpOJ*`Plg1>GEdTU&lgD
zuT^YFl#kl@I8ZR+w(QSIukP~vTCppN^-;%$8%J^v83?@h`jD~yf?-oq*{hvr8W#mK
zOnqY=zE7la`qFvF?{H_mkZ|`C%}}qhp6$n7eQ=^qaoSBWsnDgbbX8`qP+1?ppexIm
zo9n{G7S7Xar9=)M$}jj`yvE_MeSdTCGrr>iagwrgRnOm_naXFH*2t4&yw9v@eqX%a
z)6GvS8r(n4+aPOlQ|8>ng<qSO%1q_dYuO&q7rwxNt#bJuzs&hAcmL>yMEC`;t#VvF
zPvLO=&9f6%iAK!#>vw0o|KoGR`NJn~-4zT}j5gkt$C>z)fBnjLXHRF<9DA9&y@lOj
z^WCRXJ9P^`CoJo$3J)+1xE6AcN&S%Bt-$zaE6lb>zL2Zw%Uixtz{9tP<^9EQDVd`S
zCNTcoW#ju!t&A^rM%;{rlg_oTEO89(>~7B6zN@EJZOeqECBGf{W+b|4&;D}t7gwk+
z%i+5PJN7!BNa^U?r6|$Vv-zXc9Nx@tOn=)B6fW3uDI)4xVz!3vf9o|1PFtVS((0Xi
zC(nPu!^l=U2Af1Rm5TlQZUyvJ88Ul)$#VA8IlcUIGFMPYb);fV=RH-96lZQ@ZX>Qs
zGbbs<26`@3i{BhSTb{=_(|P4Wo`?>v-G?=8(jF<#;eYu_;M_`?b9}YU`_^2Tcj!;W
z!>%b)V%QEZn3Cka@sW>fS!2xk6Lv@Vc`PO_im#imc6I*7?w?OKU7v90%2%m34;a+V
zi;B~7@9boF;a|_i`u1*WSpUS2KOXBGJI7b+esS{M36V#Si0Yl6D0VwJv-@IO)?(%>
zPfXt{ySYu2wyBWL`_#Dh9;*nGM?XuE75fJ7_eN8m?@;04*<l;J(|4z-pBl4|)`<eG
z*;iFdw$*#rHE&}y&~G$e?tfroLBIJdz5K$^@L98WrkuO_vf++Ij#S~+Qx9~Qve+`d
zWbG|u%U(FY!Sle1%`bC&0@t2vjEHzxvZKfT-}ZL-mEYcdvvVzdsO|NL=b^_6bLpGy
z-rZZb^vyl<UU0_jHOIB*rU%ycif-_}aj0iW&e2`!Gb`S$E74h0sxK62nsY5u=RwEQ
zMQ<&HbuXOa-u~|V{bsKpoQ?U*sw*~DcSkG8ynbBs<o@GhJbkUg@1(cf)>(Qk==YAo
zf~KWslH|?hW~u+vzY?sKgKbRRgq$&TGjkK<4P0|0DkGv(clmn>`{l0nl6r5crx>+1
zbamM6thYW}x$la<;EL7sZg5ag<ZN+h(3$YoHqyh1BazcP@)RHILWk_==-sy#tzGNA
zTu;{h^@|wUy>*FvpU?btDs_AQ>QCXr%5%b>YM;l}y?=i<Ve>M-)QvZDw9K-8e+w=2
ze}6+tda}Yq6_xI#((@yi`%Gxzo6ahq!YZG{E<f)e7uPZt<-<)+LJq!NR)1#3p*@G@
z{G0tT|Fk9l=7`|DJK|MQ?>KFbet)F?TJX<;e{u&O`6O<Plqr+^o%?P6jr;ch-~4ec
z-ndA8eqGf)ua!Q!;>(w5ww~Yr=hNv!E?Ykw7fzhwq3Cy3=-fg1B3Gf6I>A$x)`acw
zRT0v<>~ctCql-$6r0nMGyU$Dy?aVfNFR)H9wetz5P@<3DB9}QTcQpKjQZ-%txq_Dz
zEqroL{*X#$*ro{0X=$4CMD@Zsgj={~YIyd<3q4isOj+1m=`_=0o@SWFhxCOJiat}-
zQ#4&qIxP*+IOjF#iAwvG$M1ueM0zds(L4Hl<`0(cu8CSxx>R;XEE7rUH4I$+{>BO)
z?|`;1T1S3FX!ss9pS;x5QZfHu<uV_Ebj=p8z<EYX)B{y~#I#c;v6x=#c0Kvzk*9Ex
z$fGq^>NDr7Pw5U4esWT1Vp32Z$JF|t>VIZl=QNt>V*7Jl$){`fU%RJrO;mY$v;OFV
z|Kg%GDt|gn4)q=q$Ufq^|IehON%xKWZR<}ZT`Rh#bxrD;)wN|C)+TOT!{THVYG*qC
zY1-z9^37?R-X*(ksL<*9s*%l?t@m2<+U7OSBR)r--z>I4Y?Ii=xD?&AwFzreToWZ%
zZ=A9<;#}1=wQFf%CfaRVTetRZox8Pp>txZ>w+;(!4q3C@Y0K<GRTA7QlKfLoFW<2I
z$DvKj&dpl&Sfwbvfcua<@BbbF!TOnHZ}g6SVPCfWEXRDqlJAFhuRkNRPyOAz@9$zu
z?jQa>x2r<wQS3drd+D3EeLDO@_WSz%_ly4ToLQG<w*Peb*UifBOX?h6)_&g1BWbCB
zV!Qg$K;w(*9)+`FyG3e59$cIA%wX%CnU0&cmB!pS^=@BZ+q(9Jn;&2N6S~25Pr2_I
zCBe@U2K<6&il2)u(rn^?@aVybwHNKTEo5iB>%8FR>pWW}jkAyXOLNjKCK`mFw3)Zl
z;Oj)@>cx!_v+Ew!D?MB$-)dTOu~|;u<};r_#AM3}@)Gq*0tZ)f-!qjvdc9)4hlps3
zRM;v_*Og|+Qva-&zsy?uc5O)L%+v3R^=4(LeBH>l_Q;P#eI}<Y_M|3n6Z~Z%VDq+6
zVq0J2oU<uYlIQPTA+t;4!zI1z2X-GUeIo1nym`95+k*G@&ra^MJn!D{%71S5ezxQ*
z*I%ErSL<k=^5{oTMjBh>dilQ_mOm8<$me|<FV2}AB$gX7XM^&qr{QAn`&29aN-gxC
zZF;L!E2dpKuT}fn>1nSFH(fdV@aP-kj<cDQ^D^0w?PM0bdwuToi`v(X^*`@(nE9gm
z-?B3~TUC@lI=qwXP>!?N{I=MHyH%O5OW?qaMad624({&Saea5+fzn4&Z%zoc9c{6-
zie|Wdi?2sjc+UfAp?|GYX6baOJ<E;WTT*)IK`7%L37h)Y91nLNyMKMkUH0Z9)yMtf
z#Xk7WedpcN*K$_VI_!;VTHo@P0Jo=i9hsQl@~HK_f8g&k`-e{H>|G8y3qsud(k$|C
zJ}+xv__lJLSk8yenHik@Tg5L(@mFXR?AFS*-sh_AnetRFCCTDh!G=p|w>+dXWM?dJ
zSUf%I(b**;-9Nv%8`?55{4i5dyS?vvD(hi^tp#%qcg{TfKJk}j56fkqnanTVGG4J+
za41dI;pU5AyIIF_+ss~fEIsLUjc;a*x6j3COC6RLuur%bms)In=gNk)bMuzJZnR?m
zTNU-Vt}*AyE&UyRRn^70Gq*^cS-{#lb!MXK*J~0N9Jx*?oOAnd?AztEmz>^<uVyot
zvbHYcNq%;SbC$`<Ztq9lJvUSA@4C%j%<Qm`$9VaiXioN*e1E*HcI7fYS3IxrH0RNY
z9jq63h13@IO$?eDCQ<I7^yI?o(;V&=A(!%Q2{T38nw+yeXIG`IQ}KbPe!o)WerAre
zlZ*5(tnzTUyldaig2XQ`jvkx+L2iyljq=orm6mrZPyLO0I`L@xwyN;M#T%S<8uiWm
zc1i77T*5r185u|FM2>R$9e6!&b7}jjlWP8352qLJ@bF%Gl~->=b90%=0(Q^Y%NN~A
z&(>GY)jhe)BwHq+aDtb*MNiPC%SC6O7&bOB^}42p2w1#T7xwRHy#BAWMy~z`pWaWA
zDO28=JYPR2!Z$0<Kzj0-m1&oBo^C0wJe_yja?h^YX@73s6Usin?PZL>f(1R_F7?^7
z-S!NhfBEL~7n>%36YHDzdy1v}m4b$4W;ZO1|BEf<lc*08|Ga%^#QeRkSyQ7-0$8Rr
zNhj-1IPvJyky9#N8~t<==W)gTKDsjUR!ZgFkj2e8TwZHcrwKj#*lDy$AuM>y-HdZ;
zzccv!Z+OZt(OxJtYfWre)Alu|v+@OR8t43c`e6BjJI{)*EvRljsp8)~Z_bWmnGfdu
zdRP19MDCYIsv>TC|7D(X+3Fb;Bb2^KB`qfS%%z7D&M^KGc-yXJk@A-Rq)k)mvKzr~
zv-Gb7miX6iI2!g^KYweW@$u}fjgJoec;hMc`q=eP(`Q&(<tbk0k&5yZzpy0mR_e-a
z*QQr|owEPNe0H`CO*gZS`FCGUGL7(F^NdwEGUem4Am>${Nyoki&e+S8@#B<_Wwg*6
z&$7MJRt0jG;%0w8JVCzng}?CoqQ0Vh_Wo;%eP=HwFR#iuC+4U3o%OSlwR(T{6u!qQ
zTKbJlGQq{$mcA$#ZdKV9VryWR-B9}{Yy+3DNo9@t{ELR#GkANNZx&Vab#?ubxVLie
zjeezj_42liFIbt@&ifmlw$@?N0r|#@kNB>-2L3imSjHec@kqh7HMjO=*M18+^Zxj>
z+NQeppSAxJO>%Q`DxUZp{u-d@GI{Fqb=<Ym9UQmvKc2b&_`&=?kwRWQlb+7V`T2S6
zGnL{+Gfh0%4ZdVM?~Gu+dLhfUXQ8}R-@^F(^UF%}cx)>&rtkg6e~?Ki!~H;{xAohn
zW%1>aQ`d@h<hwtfpBkYpuG4XbajNmsOKOVM2EU^k9coNieVzGT51C8%8(%)#bHK+)
z=C<D~<KjO%=WkMsDC&OBbMwbBeeR=1ImriWx>Zkp@t00++qmyofbfq~XEG0YZoI9x
z`Jn-;6I<1rtMctUTP&ITq-Ni4s}9eOJaCQY$a42d)6Ylm$|_5-xxU7g-(X{I^*TeF
zF7eOP=01K_wR4Z0l-Am<Vu!Au@h!@_8_9ge?TXVpo&2>g@~mtxlnI?Xq%Q3?!+`tq
zl$lo^=p?_}l)R*WwfnJ7L9?V-$DQ&mtjW!D|2W4S=M2w(n*CdP@~Q8KxGF+76(4Ur
za>wEZlm54Ne?7h~on#YuWe!_(XJz1PIpH@9fvz>uAE%%BaOth;@2wjbe|qe?WAmct
z0<DGB210fg1_sOPYE^fxGuq64Pswg?$-m-*8YLpKr+#Q&nk*S7UX$>L@n7h>-xqSX
zh~MVjX0!U-md2ciW{Z!0Udz2Edrjt!MTgm{1a14(uAWKq)4pDREb{#!-FaHYU#ERd
zSQ@;uTP@4r(G1a$IX6{oTLt6GX5=khy+F*o!|n{%TZJ^&Gy0K@uk&;7y=;;9{rReP
zUjClyO?L{oc6>f7b)vOm)~yU<Z})dc%=?*?&$$F&IU-hce%hHg7CSwrE)r+__^5B;
zcIAb)!k*^r6g{J+8a6XD^(xa+W>MXx#u+i9@yCBYHEYXSc2CFjY2J((ezS^ni!>CT
z#oTIH#>KX9f6bDQXRm#0UO(UPjrLi+TYev|S@q7T(#wBmxtsab+P7g{+qd0)mD5?K
zk^9T&L*RqubI-l&Z|2*y=6dM0*9rw6O<0aSfAmS{pmpE916x&ZYHiYRU=1<MvsJR1
zc`SGPu}kr<u32phdMm6w=XFl?i-=bm+c&LVwtktkX0xHnEUS+b?62*<y?v(Du~O5i
z@@3|~e`$Qlx)2!lcKwR&fsE42Rc#K>N<F(#j48B5HT}}dg}bh%XKoFXSTLbUY4fg-
zvo?kkJwCOZ;XY{CH!E7_^v$(?sZ)fjJv|Nh`F?8M&fURNn8ud8g|Sp&M~aELz{4Gf
zl$(p2g!Y}AyzbeDDB-_9`6N9YQ>TdLPFj+Fs{HtVg~YT2K}z@Z*NI+n&acgX?vQ`~
zLg(v9o_>?3(N`pb;-dPNpU}@}__xeHMsKCEcW+wv$tjALUzj?)@V1>g@d1l<uDZ>O
z*gC$$=ek9Y=hbF@p2nll<z0JrV!=Yz@Avqn{mriQI51zC!Dc#Ri|4I}540K0e*K<m
z5Mnd$kp8!0Y0F=QPYYNUbB<&C#yu+bLSAd4vR}^W6coCiEj*+1$;2;{o8&CQen;J&
zufM-lj?Kks{l4`#;vZ=iWjQd;WXnvPqhniq`90^BKl^|B+C6>eabv=6jpEv0ua@xG
z@VuF9&f{{((Y<^5@^6QaO+3sjJ9qu|ZEdHXUcQxlu-*M8`^PJPRd=1tf3Qn`SCx$~
zSN$w?)^CSDUCAxHUEVF%>fX7W{gK@f<AWX#o?N(CcS+#Fi?DZsPahw-HLdgqb4+_=
z?ET(v*FsHq-ENS0buQSYt?iPvVdsi{hhB58syiIp9Gb5?(dhi7nQ}9O`Bp!iIrX=g
z*R~bU4tgJ{s&`wxK(cC;^=yBhvM;f}+$`r`IVNqaF#EyDBHN=^o3b+2ys)Ww%(HTJ
z?{tw@TwY%sB-Iu-w=0KhFFe%UR(0?l>+GaUmnMr;I*7c`=1(eC?pyG4@!ox(qSB3d
zoU$|B?cACVU7Trs;9I==qVmHRctwtdKY1_mHgGA^R`b=*dD<@Adz16mxnWJ2$#nTI
zb$2e8vz!w={AgQj)7JwnYyIv`*r`2vCSP{v(&pdmIQ$Mu>&}l?mSBxKzg~`e&$U}2
z!B@i?wJWmMYAFjny8P|fsimzeZ~e8hk>6^#_IY#soP@>u6#Cz<C^elDIx8lnlQm!U
z+Fb|Hg4Y+Wtn!bl>SrkBm}Oe$y0f_?Z0h!%%zpNf4<=2~{3sEXU^YYK)^u}iQA^9C
zGjFY(b$@cF)>OxJ=FcXW@A#EF(<g^(vt_#J(!P^ZmN3;FUAj)+WTIN?L%GHGlE2K}
zdx2m5)~~knaVzV8YyCPG_wZfm_ck$}Wm{D(R?C)_xGjmvsASKr&wej@&*j6(2Ydfy
zbxYgqD4lkhx9q$?rrzXfD|D<kob<Ko-(z!KxBS_om8IcvHNUpCtkD&4-D9-ny=|lP
zo{Cs6k%_EQYo)jxpI-_S+pje}BS<AFGtNjwFfjkUbIu(X%kVs|1=ibtJA{AH(9F|Y
zb(HmHaM-)`Tdgy0rKwlUue<x>obM0iiZycG^B>fnw=v$>Rj;x&jP0e%n^!GI_48#d
zv>WctTKDbQvR%JECI2`#{XN_EkN(@2x!G<@zq;zI<X`!k$~TX$3Z#B*Owc*!|F&K!
zn<?1EYv%IFGgEg;$3_*cS;f^FK0nvikN4h>cdHo{YUFAIRaSVrw0*Hq>$|diR*^z~
zK!xenCbbvqe*apvuI<dZ^QTT<eAS?(`L#mXu<YHYU)(30r<q#bUNCFL^{`c}NjK)~
zD@>0H+gBCe<K>ss=M&W)#PsHz!pZWLw_aCCy)hFHc$z1BEg~ss7T0w{)g_KeQ~s`U
zTN);RDI%`V^p?%JUEeQgUJA_p+%$W|wQ4RF%L$Wq=cqSMTM+zb&6<*pcj7PHxOU^#
z?G-D}_0G9+-mNdb*-g0fy29ef+;fWF+w;@(GRj1cO>*7sWvn=T!vATH6^&072!Fk8
zv(WJ)-<QXc^-up>TPOyvU%2>2NOkCmkh$9(Ca@}fea0;PefiS2akU(_K2}U+_B%SF
zVqF^*gugw_@$V2{U36Xlg@ee;6`Ou8Y+7+OT!gF0amwzR<sq*Bc<k1Dx65{Tr<AD}
zZGIc|Y196Hx7|dp{dsoM_{xEa3!QZxcF3^x6h!eI_Du;BTcbMnmSO!}{!nwP^}KyS
z>gVO&X0ac7dqv!(=}^p!%^MPu>$do<nLcZi;`8SWEzg!&*O~n@YT9qWIQf-=9>dqZ
zD3QSOk0)+)_OF{O&2%Vw+41r#&tjb2bSAsK3ik23v*OAB&hT>=wtW%HQjJ~PCw68v
z+gg8BkA|y@X1rZwveKf=RX@t*QR7CYJuY(>IyFD^JQpeG-6EBhUl6Fj`<Io`qh}Yy
z&ODWG*!%Oh@wA+q_Utoe9#M9V&gb$wG;`U)#^QWFCA}KwJNHVnHY&ItvA@0X|5A^N
zs81e${7Uo6HnMc+K3W*i|KQ8QYB&Dhdw5fO3zjcqnUHE>Dln(oO6kW3(R(o>dC#jq
zIDLtmtQgXDk3ppF>(<YI)2%Ox>~M<-;#-#VxIFFnJ71sR)aA!a7QI}u_m^J|@4fo&
zT<_lNFAuJKcuD-xo~<i%cIlp(?-^TiujcCKHjzuAQO~(7ZqL1P`{4c0uP?5Vzfgad
zWzkBd^mr8=&!lDVYeJ_^UFgZec_Q{^=)z|U_!r1uFexj@ns9decdOfX|3)qEG<zg1
z<eD`{XQt1+usP8ap1o6-<$ZTJ^=qo{!aJW5A2XFt_#67zVD&bc1fj_@BA==*xgVLD
zU+^m@MR-<V@#cp=9%!yA-mr6>+@FBFFJgY%RvbR~oZp>i|4welSE~2Kb2IerON7r_
zcK7<~cUQ^}YRIg#Ec<ce;VwA~E_I`*>EUt(lP})ell4NqkN@j@p%2G?EewxURo-Ua
zV!cb|?7ifZTZ782jvRV+#<rT{p7>3pwK748(b31EU7vTKbab97HL3IK??;P2?9Np9
z=q26Rf9Ru=ycXMcR{00{$-6$<Wg3bHFI_0R>j$6m?giQ({#DkqMn1iK_09B@l_6F(
zXJ7Tc>+;B0wdZ1&-1oW4^$)80yL^(mIaaK?eedEd;VAo+TGK?myPlrWPN~V;7M8n3
zWTHi;ntSJsleOO`um7RSHJQicfBUblrX1a{B!gWAsfD-ql-b9+nJJV>FXCJHL5P2+
zTD<u3NUeC~o8Cvy#7j+OX*u{}WnYWep7ruJtx}%VQ`ZXBlxwxV3TJv&_<HNagiWv3
zEbBcJcxln`$%Tsdc&5~qIuyuVxSR3H-NWD~|IWn+#hGV09Qmrf;u%vuN3F~s+55Af
zYkt15amF*PR3YQ<n$MQ9?p9?MSAO_=pM7m_T9SOoFQv^Q=T~uV)2TM}{d7<JD9dIi
z!xy?y$qRjEccuTT%KO@|+Uh|r%c)N3_2*x0+jW4wG{DvT=1hY$w?j!4O%+d1eB*9n
zyz{m13;%wzqLsO8{$+eswea~MHviN+e^C*~jVBXkocVlOn)%?(WWhbD*R1xrU%PeT
z%}b_N3R@)N_0=j$(@k=ZrLAD-Uwief?elkE_=GL*%y|)1TIiXVp{lXrsnu0Jqr0kQ
z&m7rKcm<oj_?*BJd2QWyjj2|1&$T?SHaxQH`;RMrL9c==xfJC?(p){+j3=*oEN2z_
z_QM6c1zHEQd1qL8&pK0jfwwnR_Nt;pb0*_Aen;N@n-?9u`$lJf*@r(LLS;_$EZU*l
zIlIi>d7;bK{CR(WSgn#j|7^K<`m4>pyTYH&{<f~*<;k-98!!KT?v@{{vFw@o`LtC`
z2YVQ9i9g_a$MxCnU%%w#s`pj(Rqew6r&y+)arhSbpL>bD=LIf|TXqc%knWr(eoeQr
zp)vCKb8iG^M99<Y`wtgA@8URBTC(Eb_ery+aZij+)AR6bYU13n&}o6rrc)dA5A$)g
zhfb`O6mtyVEcRC2S|E8T%Vg`C7x$eqBX9kh8?A49{yQHt-?rNO2a2EXyj^|n=iK{0
z=YH<%F*Ht2F`9WP`K(pB`gzIKckU=DE6p#Q@l;SiMEcqCOZR8%+n)01Ke^=p43DOj
zDzkXksPDS3RIhLFpa1X)q2n+6drWPO=J#EdpTE5Q__13C&x|B@Tk5D~Zrj^@_>Q<z
zxyq_-?5xLcoGmMuVVo^yYnT3fe#Y~Z{O7N>o0=HeS(WUb8-K*iI=tGv+WPy6imJlh
zcPCF2Ec?fJur9H&Y+_?s9HaD`YgKyRKWwQ!nf&$Ebldrx{%_uX`1!kus|xlTyt;nl
z_{@`=B0v2+!<4o$M{@hAGlxr*H+SAV^uws`n$~6($-^@Y(t0eDk46|L&n)cWJ#73~
zlGS+P##9!!6g9)iGd0f$p9x8m`mFL<#CYz^qc@EGjrq<PpOHz^Gwz>xUMEG&&|AX&
zSZ8Hp<;2Ky7RLI9+7jZ&rbeDzlXOR-Jk9j8(dM>~DSKvYJ!6$-yLsZrG_lzMH_pts
zYP9pslr{cl36BycWqj&exjJ+H%+-<WcW(aa_j&ai?&`pO4sl0%9|pOdbQkT}m1lVK
z(c_D&-S=l%9{coU#sS0hwQtsRt<{U(e9b>{zg?KbYN_9UEw&e~zq#j>|Ks8x_QwwY
zC>Js8dar(F&IQB5Lr4C48K>^~Bz${@d(OuNqIaghdYroA*AbER9(%ir{~et<^Plx~
z&fT6_f|9`kTluvPehbyO6<o1IV!9^xIkEk$&v!I(ZQgd{*ohN2Zk#xA?8dfj$BvmD
z^b&4AA9r+RcwyZhpP<Ft7a2=jViDiQw|v(8G;OPe!S#+j39AF*E{X0nXU#dBknU0!
z#U;6T&Zq6+(>G)u6p*>YDf(!4j?&Dmce#5$NX_<HqbT0sy~gVjhit%>{Tmjst({{3
z_i`^k+gyQ@P9=da#Ah45n02-FbjBjro2kK?0@Jk?SbvhMTr~A??x$w!`vKQ;PMkR9
z+Z>{|%xqog<(P}A2ioJ_3wQ60nZAcD>d;|_sX8ntIz3;VTDpN}kvC^^_aTc(O5ZiV
z=KEL%N9`%+Hz~}$xb@>qN9*kz8y*}IdU9!}L#5FQt6K|~x>@MBul{8`|8QvEl1Hbx
z`uDz$Yc($T^=0C{y)t@|GgdV4+RYAMRxF`dul{uhf6jw78-A|x;X2mM)^XBQS*zl#
zz!K317Ls|#3YS%^4!<b6Pv68rR7~uKT<L=XZ};{DRn~uIzJb$&r1g(Dm3jGT{rtk8
zo?2F9=O8vcV(t|MlSA)bUGn<##XSA#yID*PwsS&vh22^_b??H%*GrZKvBU)|_$_QP
zafZO=%qN16yzbPi2w!A5`|h#w3oVQ3+9xKjJCgJ<DSXQT^}<!}?)AP7pRaV8Mf-wR
zSk#t9drmLa-nm`dYLQ}O!Gx?x+a*p+PnJAg#(E}x{=5m(=g3cIpEzfg-u9>!?TeSH
z^A$65%RV`IQ9$e-(^Vz0%~y3V=)JmrXX#eU+x=RxEq(V@TV+GvOup{?_*;TUhlq#T
z>HfVH&)%JV>~8Vz+-+NR{lmv)qWt!HuIV_s==n4DZA-6b&C0sXvf{+rGk0QZKZP8R
zKJc}6M`|C(>Alx?R7})(wC>EAQzuzX#ji1$GQ~G-kDO<{WA~c|kBJ8loqn-)tL4T0
zyY_FA*DZ-It!WWy_T$>I?dKxz>hr5}pFi|{*LHcgi=WQ){35>0KywM6u%~bChF-Z?
z$a|`x`hBm0yx^hWrt%|gZYw`EeX=QA#^6$rc=pO8&uRCo-=FC+eK&8l%#pPlme{-s
z@jUr1DC5B_VW|+qjhYKL?~iW2X7O?J?eZ@Y%8RG|b2i*}@$&6tt<;BV%xAhkihoP^
zHt(&IiZt)epDRDAH2zt4#;4V$<GItRP1Y<riapZVEfq%<j{KVLYMzoO$)N5h%e5#e
zu+@In>&NCVT{p+o_+=NryVWiK@Ld13*FS=TG}eVGSTCPHgX48#?WF#CQM-S>_oo+d
zWL@$uda2GXwDZNTAGNx?Ti@Pw%=cSvQ`OR=Xs5z_cFsG_Kj9j`|LxDvwDO<v{L#mz
z2VA#zJ>DvLw)Jel(XVk;kDL9^t&LFQIkf($u1MY7>3p+aG|tat_`mG+(bQS}(edxw
zKF8eu6P;qRv8L>z6Qhk^-XD#w$LlSRt*w!i+4-{k*#nC$JO4f{Kg49S=tIHhQlTHs
zetGM^&v|S4_u|^mv#)iBJ~3_G-2I%X@q5IBON%eZRs4Q^`+4}z?yw#IvgFquQ#t8&
zQIaF8@lNv9#w}9(uU3C7-}Y!hTv6Jr(zX}cU(3yOTkgGX(h_{XF*Np<m#j|8iZfd0
zZ}{3BXIQ!`C)`o3;3Q|Jp_0ml*EcI#c4%<gE=gaW=Gc_$QybMAz~;&tntG!>dTQ#@
zsnrYSEw=n0sckE#pMCUddEwTTQ!e{mKYMp}*rZ?)ZcT?{9=xZ{9Ddbg(9w``f=3~l
z&BV*}h=X5Tm7gD5=*|zd0os2~Tr<5~bKH5tjW3%m&!x{+o6UYS*v!Ov?hobY-Tw3c
z9o^yRIOUD?k99leWfw+$xp=E)?O~C-YO?=#iL{^BcG&*%)S+Led)3c5>6hQ#+dqR#
zr7zUoamT)<Bi9(e@|i`tUcbA%U3vb~<O_d-{_B2}v|06mzjUUakE6WI@lWz~*ItMz
zWo0n!*}m5%T*2np!ULbu=5MwB+2OpZ#HC$)<usKCN=FZ7c!(QihExjP^w;hA#(pmG
z-g*n!`M(}rulp#pXY10e#K54$BI|;=3o-=)WkiY;1&fS}Hh(U2id&_%B-i^e+k=j#
z8N6Dzy_Vd!VBC5B_kO7<Qd`5WEkA6PWhi(*<iVXvi@7(rj?`Ps%i37zAe(LDTrZ|x
z*I8!KA2hQk%6W~Zgt5xkSh2I!9l8&{tWH`ZR=d^g{>y?z?@I5c$^PFxr9R}M{gR}@
zk4?sz2O{QZd~bhp&iO)w{vF9()8FU^y!@K_bMcJreZCDhoPS+tdHtzx&feRtje$8`
zawU6}-!i`{SR6g+#3wa2&c)B1zx>}MU;0DBG+%3}mu1&O(PNVLEiD%CNea%`8z^%v
znZxebzqQ5v7o%O3Z7xrLBcnU-!YRJj1@flTdT(}p)9?^HkzwZQI(KQ8x?J+*j+x(|
zl^)QI>EiJ_94Y_5#QROl_T((i;29yaw!Qba>}kI9Wy;%wZ=+ZL&G@FXe1dx?cjQ;&
zjYls&s<gOv@4}IzPJ0g8#;D}yHSv99tWJ%8VK8IYB>OA<A(M6;mAK=vKWc>o>*Di=
z%jI9Bo^FUTb+UDro*xwWh;6dP)WW3;cORZ}h$ZWO^DTLoriH;Ghr=!xU7p|lWP@Sv
z37!?4#*hAdcu>rLZQ@IVcD<>KmisejWQaX_tWjH}Qrf>~RcXb+pKj*6*{(DmGd?+~
z_CVM3EfI^iTHgQcb^f^R4~f5zW-m#4l_b-`H7Tmj!P)T1^R25_9G_?M)B5VnK<$ox
zBV7$+CH1D$>BfF%4|eUj)RQK<qVcm#PwbQL?U@f63>HjlKjt{&#Pk<uxoX)Aew6ce
zyNjMUy*)-P?&!+uBrRsIP2QWeHfg`GiQ1a(9~$^FJtC^GPg|lR>8m%-uCwW1Y+0Vq
zmFro&RzLdJg$WmryQ^s$%~xH$BsP;hxT~z;t+(=~)#vXwZ;p3g;@_o{P<Zvgx>FBk
z{i#j39T!*<F~uwT3EONo#a9;-wO#EZ?B!&;vz7CX=;pCo`t@a&Es$Q9m2)(?sdy&S
zod-w1Y<)Q|uyAuC-wKn={3UZ6^&jZIX#IZ5KP>LeiWkkZFU*X5do(nM&As6IYU_2E
z9_g*^;c}5P6;!s=dZ+Wd>jbONXMff{F6H_|g`%OsPuA6!ws&3F6+S1OqyO=`_?x?r
z&%G}B`hHtjWA{1#>kA(5pZe;(2H%UTCe5od>nx^E6`vu(seYdGgV~#tIbOHr7Y9x+
z{-giC<ZAXqjqg7%*(E7Xc=~^T5Xa7=*PN|oI-dNvFa20S?9TzwL$95)58s}|&-lIU
zvvs?xk#v1f#qTwR0&xc>e2?7s|6AvDb6&;V;%{B6-%a6jjXfZ0=ficlm1~DgiO$+f
z|4;PY?S3tGP4ll7lhs3!@@2ojm@?`vs0z!B2+XPUJG>!7gLjeR@&!xQ6#PH8Rn#uT
z#79>Ata##4&*}b)1NS+f4B!5=MvA|9=hyXn_IPx#ynnQIZQWYy{-@WUUVnDIm1((n
z+C+`7k87-DXL+wa#V5mgT`iuCZ|C>@Z{(Lf+P&-cm-8W~WEEHSZ&HkmR9rOqPovH}
z=i7{5t9Ay;P3sK^U&VH+U3H4zo=vl@e3rky=C!@;OR!-R_Y>)B=TiUvDqWZU-~B-T
zt*2ACW#2Y0{eL0&dqmRTqnE$GuP>Bd%ev>f-ug^Kj>pWtS~7YIPqv-PXh|y+@(iCJ
z*}-jOvh8edN$e_yM=8k<Ga_mx#h9pThkmj@eOG`vY*WZ3{aM*<S!Ipiwx7H*r)ySs
zeEJ5rw0~xj>3tI-#m!W#jYIw<FSS`?@_0gn9{>K;*JeJ7uIsuVCMB?}U)uNlpJbz%
zmXGeFdOSMu`PW{Nz<;OXwA^1WJ(?XQz3Z#zwU3-ZzU@f?@Bb()R$1L~*x={Zt^ZYn
z<0_fv95Ow5*2#3sRMC=y^O~C44y|hP`kV6q;LQ!<uA6GaH{Rngwy?YVGxxfDtKN&d
zwX2#R%Ss>idoItmuJ@3K`s(>2TkbsCrdm{!8Ze1Jh4J~b`WL2xyX@Z|Qtz^S#dq(*
zqMr){ro0N3J1ml6`}_CtR^JDcci8qghdclAJNEI_*Tx_-xknkVvoi$ye#xG<ap2VY
z{Vw{+O0hE$?{?J`ewe=e<$;2ahjwnyk13d?{bFs+im9>J6HE5jrRxN5adqgMz$00C
z^P}ZT&e_u$&HuNqj-Q|BY-zWulGo^H(}Su~<ASrZZvDNk-ppSZ<107YS!aJGQ*MCG
zm)f@i6W+h9-F~n(<Ncaqw`#jkt=AWd!*zdnEPWF%qrUm(gunkx*I&3b<?Ia4ii!G4
z$@^#hTYZ7~c+~B0#e1p+R`hdRzv*~MsJbTXV0qOqorq75US<C+Umv%#YUl0C-tv4g
zza9wnUBC6gdP(l#hbm6j*1c)D!JBQYyz+$AzKMS<+tVX_ZA%n)f2uZW?3riS8_``O
zzVTP0mR3OJtJPu;UFR%O_?mO|Yj=JutJUpQr<U=oDqb&kJWy!erP}YUE1xLsUAg0n
zjf(yLl+!VZcTIjKI8=%LY5d|RdvV9?UM5@7>Wf({H`n@Y+<vdMEHVA&wsUREW|<xA
zQi(g1`uzT`7H|F5_)RzK6&;^1JS@6v1-Er+w%&(i$(12NTC;?#u5rD5Fn52d_3>zf
zlTQzRjc%Czb&>Mbt>IbDtAs7$jxAA7Kjvh?629`(B@4UiH%9UmoR@Z1?%(maLgv5s
z`*f!LXZUsr-M?{R)~?&@KT1BBwmEVFS7fW^1m?_88GB9Jy;}PP8M?SvcYPOoD%<yx
z?cKMp-+#N-ZByQo{?0$^a)^m|xY(3;8K>@X$K(bdt(&+;VHp>H{Qa-lfBohP|Ik)_
z?mx%k_p^;>U$dle1pctvXOuhZ@ke2%D^q1HE?fu-X1m__+P!#&`HhA5q?hd#Us@|E
z>l6~AQe)WqH}UXC*F#(%ZqA!<VBzok8{*$uC#^PYUwbLpFJ7#+(6ajfv4_UbS1E;a
ztJKX_^k4gHc^a>ss+3Xn<9)n&``<5ly!Fzug+FtIzDeHjQtLgj-14PrvYW|+%WVgH
z$`vLraXT7cZ>r20wQ95V7dF4|i$DD<l9gAQ`BO{fxgziEvvv#k{3K=^brNE6VSUfR
zu~<ovF|$>(mDM$XWB1N&Gj8peabpQb<N26hwlYW6<6KXlX0qSsu<N&Z8E=iO@6TNV
z=9S+n9;pSdOwrh6FwLg*^#!S?&*myW^ZzaQSjqd@<eATYbRYB#U%al2f5l$+6>mD;
zHhx!sbKJbZr1I8=nKC>l9zEO`dOxeF?O481VgzrSqQsfIELK0X+~!ED+e!6LzxFJK
zzwg@r&v%2KKA0k<lAt%YapGLvmgf4bC)=htPb^gw;ClG#MMU3!mG!oALXF&}PXd_N
zcP$B*6EQ#1q|f{Da;i&>=ehFl^0uX0R%)FLRmiG5WDuHX_vONVo1~KB35QGO|M6FO
zR?5Dh>h)o#8^85so_+r>uVZ%NXQk}67lzl$dkVjOfAQ_wnSY)}eIM?ha}hTA`AzQ@
z$GP32-}3rSPuqES&*!gGwXfgX|K^oL=Bi7NF8^YZ`SQ1g?`_GRPwVHl@852}UhBcb
zOWD8OxBtCa|I%eo??!Rn#ci8QGFCO8uVB}Vlv<nNVZF`a27|Wn-mt><Cx6=6w#|@u
z`JMA+tVL5#sPc)NRsT~HPnnCqe<ZO`VVBuH)*Z*%W?f5MaPNB3+-o79juvwLn5kj1
zP*p!vL%Az{Rm1fiwd_{=MD1EveOR=jY004_eG4mt%x(PMg*;FFomHdQxNXhz<ayje
z4`0Sl%3HmAvz=dSX-q;wNAm_ro717&i(Z#abLZ`yaG6&)Me>SBNJ%OC+|${MH=MEi
z@baZ&|GmXW9h2w#7H=2J6)9c7$#=5fTj<kW8>6I#Xr}aNmorCLi;}-bTsUs`<Kcsy
zJCFVQta|h4xeEOkzk2=qzH_(r-Lc*zyIkV-#|;N;j~sfdbN}y|Lt8Gm7JO)E@U6cZ
z6?E*}1=o1-zC#!H#^%cG->DG7TzY`FG-LHGi<3^;*~fZocU%m)rRpoW>U-cZpRF5R
z-13AZ@0sLFUleecWL+`W)pBR>#<_WU2e--1TJT3<uJgXYZT(Z2XNP@Ji+fSqSGU$F
z|DUsFwa3G4n^J7fIa@87C7;+U{V4G8<=oXfa$a4MKl7dU{MM8Y>LSjbvWXX)Y&?%m
zs(cn5x$o)0Qkm3((ESVl%B=m}xwSIo&4vrTQd^W>J-sKJTazh%@z$Hey(zBpR*&8O
zO{_okkad%-se!to^%=AIXV;#;l5+ia^0rM+-#flcC=~j+)`Gc=@dne1dA^|+@5wzB
zoicsWc}LI4r6;Vnsa#vh^6y9Ak|P@XouhXj{r$m8tNZlF%*@NoA1<9<Dt@Z4u(Hrn
zJ7PwqYgFH@+)T!4i?@pHy4arW#r<Xeq_#&v-lzS16F<Id`qi{IU+zw}rCXNF|NS*N
zeYx0%MJ>o#mTPWli99UY8o?eR5`KMNYGT=<jzc@xg3>mqOZ6Han-mf7ripdxjj5ZO
zO3D&n#s*k$D_vS*uGz&Z;;oa=Dfn2*^i`J6AD=ayYO`k4q@Phe|NVYypuyYpo0apd
z%b(o|zV~@vaoDjLvyF{rzPVbt?(!V5^HQ&#ypfR*uqd0+eM<YvdEeYSQo4K3dE}p1
zVt!V|QLpn^uJ(zi>4E!~NA5Xku*oI6D)rxs%suyBr0(%sB;L3^Q76TytaZIr&65v;
z%5IiyY_2P=S5~&)=$A2k^f&SFi_$rJt6Tm~dH8MfU7H%_!zTM)zq!j~^EREoTrSVv
z+@R-PO#1(@goeGlPu)J7zqa;4)Nj{Ss^7ZB{fZ+q><_2jd-Rk0zT7|cRO@NQ%vb;A
z)rD-2t4(E2d0!%Zy5@vWf?BcPGrectI$_3nbJn-svT5pEc50Sk(Xpn)nQ}d%mnXkG
z8ewe4bN)-(6d(C#p~w0v=L*iK^Qli`FrFZJgeUD@(!I1cgTpf(p4okd<C*QV+BA{R
zjK;Gh5BDryUUQ^pxw>zP%nZ#QW#g$ckNKph85qxuJd<;*voP(M(dQYUL0QGk9g$~t
z7|9!-o-ws&v7~bHnU~5tH;JZk8M_<HZw}c!xybvc)KBwgC!g6zu|(WSwY}l=ZQ3E%
zkF4gcr|X|?4ga*8_x-l*(q;26|5ZP2;O*Rh(Y7Nr*FVz#`@H#IXI*~1q}2X(=(W~s
z;cHcA^P1Uz-2CJ5-rwKuEr?mY=kD>1=Uc;b<7`u`Z<$^BCcay=?8mIz+2Zl_@4rUp
zYizj3HQ}eLwq@JN3*3^Hb(*r_%jU<;oXOz1p~m^wo|sQt*ZPId3F8zEVA|!TH{C+(
zwaL_f|D<~N*DqbRZr!F$ej5&{N7*=~eyRC%R@k!5Lwu@rGq>|0rsO@XOq<-qr^yAe
zm)%y}8x($Ep>~y4Icpv-H)mgz`}XkV*`*6_A6&<GNxMXK&eYA{j#Yly!~R`v&ivAU
zs|{vf7O%hfzi49pR@YBQHNAe+oGoACFWP=u#C=lM)}Q-MYIb@qc-VAm%hgoVpS?_5
zSnjGfbM3s_#kAt%$HoSU#{Ks;g&8W8%~I}Z550Zw=s`!BK<#PyA6Iweo;r8xRJ&7N
zbVX%@o9tqtgv7fEZ{&}z4L#!EP-}Zdgri!-K6h^s>$?IKPH~Or&)42#WnZ}O+k>fV
zW&AQ;uWXt6ZuOjG%YFIWRa$<>Y9BGm7l$REU;MW6PKDgL<rmkVU3YETwfirozdyWS
zdx=-&-Oo=p9lduhnDt+7n$zP$%E!yM#t9ibV^_`3SjnT^A<(3#X=6F-%0I)4WzIEO
zTq?y2T_-+u5>1(!^6b&K^xz!xr_0h0rM=6zecr42)s>*|^*j6ec0b=HlWUPGaH9W`
z*wa-XE`042ocepVm)7Q@t|nWSJ9oFa?z~m9{iTCvWmwt9W4;@d?rza|^x7z<P5H@>
z`)^Mf%l)3dKJa(d&#(%4ks<}>=_|WKGq23s*eQK@vf`9(ogSTgKW}q}TsSa2Zt~no
zZYNdi|AgFJ(PYsWlknKq@##&5N~O4jV(EpBOEj{@|L>C7Z@d1^>*&XC1z)^=yHUB|
z`n_4QyIG_@r5{Z+I{7`aK6UrUERQaqPWAZzhDXb4L!=ypmTUIvEjyO4x#(EEZ|`}n
zD~vf0*9d*N#J2a%2F7y9CzGbVYB=}zV`!4&4aOsVXVlL5pIn{M_(ngYeRHelPA{Vb
zzK3dl1x|eI3H$ZsV)y+y8gMCo1JhoW>x)H?7I01dQn+{1J^7@Ti9u%BEfW{CKDjZa
z<?7+^ybs>Jp51-T26}m$xA|G{_-_5VGv@Y%iCr%kK73r+yjCe~BkR@xeW}!AFLI;g
zbj=qWI}ks)>AUu7QPqHzL5>}63R2&SUiV)6bu0L+`uV_;#O(*KcWym!dvcz<pN#IO
zz1QCO=m`8^PrcguRq4sn9N#-ZThimt7Cy*}iL-QTIjQ&gxt?NN%BtL)nj$WjLc6U2
z3J+qp*`=S>n^?s*dB21B-m{JF#`3!LB^NGaL}Wf_Hwsem3q8d7zxc5<|CP5{jpui~
z7CtXHt)x80)b+tx$@^U$xlFa|cCA?x)v}0prnK4n8)ahK7|I_mI2GLTf;W5bU#2UK
zEBCGBHgj3lD1Ur@-}jXZ-OV@p^ZUe_)o=XFVJLR2;nn0XJh#tUPMA0E@W;rH3r^n%
z5secRkP+IM_ry7T%Q5pj+lqveOXsUz^1k}MdTmjr^16^~>`q$@SnJcu)za1r|A{bF
zS$;IjQ(|I{aI&K4mO1fhIn50(nV)YAbiDHXYJkS0oNXJVUcGKs@S5}ebLzGQZEOub
z>%)cY#r}Sn*LqQ2wW_B7*x?Pw8a&zVMRU1{KRqvRt8$mczTox-t0{*qCDt44%evzD
z#C<~1H2de1Cp^)e6y&PH@?MB5@$<4bZ`QWnF0bXi^R@Z%Ts8ms=3yo6s;&=vXPf1o
zY-h^by(~ic!%fA57vtn>thvuls4rrFx@c;MbLfgokJ+ncC*NQ6W9_TkU)@wqc0_c3
znpn=M-NSTU^z~%#f)D-A*E?#8EPtZ5OlA2IN15aAgT>pgT%9`Q>b8n`5?=Qd@`bcF
z%~_K0U9z9$pU^D(eVx-DEirle!&0xWk9*&bk5^_YPuHIxXW*D|yira(P)6dJl)?PC
zALSZsGE)n3pJ^VybnVi$<^|99$rU$yy)mr%W5IaO`RA=ayfXEfUN_dT8gxpR<+FS{
zX!reHt$xn|+f%CDVU`><dk-&aKHfLaqU7tV&Z*u<nKnhgeDcO9>_P0YGk>>)*1tQ_
zEB##Te2#z3p9RK0Ki1cutt&e6XMIlt_xjkqYS|N}om^URGV10F$H^1_9646euzVF`
zk;|XCXEPU^acY}ozCyuKVgF`Mty5Ojy+6MT?=IChky=>lug}Xj`KkwFk!J6*5`#6u
z9_1@7HX4dF9XQ9Yq<C?1#5U2}>$u-}+`E5cZRuMJE&=cNkq<n!m9)4zT=1T)G}Eck
zxx#GsLy0USz5*+=*xi;D#R~5JM;vT-y~=$jWpHG}i39Re78UFfTBm38IHF?Nn+t{(
zr(;Wh&6>ff($o9#y+HKMWfx1<2}Ibj+&up-PP{!~anaW6e+r%AqfDwd{n?uDHHZ0#
z(I)quS$F>gGQDhVywU#j?ED;ANu60cE~L$8<d2`7@<86h(4z5L<f}On9Cs&3ZJM{C
z`I<Z9_jm3m65?e_KEKSJSpP|L%bGLgCpb=btYTlssO9Za6MWIJQhL&%J=fkuOn%NM
zvr)B(YuC#ZwpkO+KboezzxQk3+=nJrH+h+#1zzeo{Mf7JKl73wA|2||l6NM~R?VGN
z5WFm}SxL|~a?5UKQ3+MQSI#1OM<!K0j`b{_@A_GI34da5&`SCCSB0JTp0P9ElK3Ve
zCjCP~vv~g}Kla+JL#>f1+P@a}9^yVE)l<d&-TK?1JuFGh-Vt?@YPyMs9g}bETXbXQ
zgty9DO5dyWzB2O-`0e#XxW}AjWj=f5rO&osUaZ?*nY1H6K(b8w$Ni-$nt?xKI!c#M
z*k#sRFD@r_@7&*Gj}-OX`KQg#;t@y>nKUVTcl~?qigPPg9?0%J)wP6wSF)G5Z?C4c
z%lpO0dG`7?Gj{XrX`a&M>7~4r^GMgzx4u@}x4hmn-M8ZY;YzuJKNCCo{0kNKmA*}H
zj^JOru(j5lHN@fF{6kWFudlnW`+Vcw%*U>WZ!Na3oO?#aZ@*Bqmx6jizHg@C^R2v}
zZkaB;)gSob+tbixS!sv1y_v>no4I3ih7)VRs%u`ej3!?!QJs{RQ?0mUYX{TSsTGS}
z+rQnGxn-{Qn`=L(OgnUIGRwcx@0Je?)_C{6)6bk&W;ORsV!X(s<+-!|&DFm1*@<nc
z6ZgS-UWxPvhnKo2uRFD1PLg_vccG|y<}A^<TJ;w+7wz+Xe0iVH&)#FwO;;YXCM>D@
z#kBX0llH8wc1xz}Y`-V5KEd&l#<tM%7uy<mnN5$Zj5%Qv-LRnWWL=EZ>^;g$4f4+G
z#H7xSm^|e@;~l2Ul~*_T=3LfjnqPdLcXDsYGL6btA~Rq0{E+$Yt#ZRmS@U*2dy2DX
zn%~K1f2<>xpMH4j&i_A2?XPp*ZGUv-+KMnQeuY?5)(!R7XRPr}-kPpx`$nbGH|WN(
zH@uAj50?E6zA~?jbJmuy<I^l1vtBy&oj-E@mPoYisYK};AC0rSN_`hi30r2Mne^(W
zpX9{cSJE3jqkW%E+_}&(MLtuD`%tCD_B~SHIVW$Lx`9DPXxcG--thh0iyec->NVa;
z)G*IKRB17{+T#C}b91x{gjXeh5-+&9IzdV&PyL~0mCvQqbq3cecZz8G2K;+a(7#7X
z;O%{t{hIl|OTO_tTC`5mE!+R+!9EMm<t3%FfA$#&Os-3le>->EnHB$>X3hB`uxzek
zp5!_yEuPan(Y$;1?hX0svGcP0q!ty929x-{{{Hy+uS6HOIk&S~a7TJR<(wM$^p;cK
z{C5Q#s?HnUs;#wtsFv$+;*0&x8Qar+9EuZsYGchGuPtN$bo#MxXS-bZO`{b}yB>5E
za4_=sosVEPKCt`N?nCSCcCzd^zdD+~>xuK9z@3-QKDAA>+7+Mh-fZWZocDamUv@a@
zXYXv;t#>k0+huh~@D;}Sv8&wnuerFiuXWq1n10z0oaH|zJBzb~9Jn@dX5CCj(FvdD
z_*V$ecX4|2%jJE?AJeVEn)6Sn`N(KCdGAdxdt|r$))VcC)xtUc$Die{;nD5V;#hQc
z(d$`9`I@8!D$73kXR2l{ozfDrpih8veMj4YDn<S-qarE0kC)YD7;g#HS3W=YdCC8y
z1vBcNh-{y0lXLvnQML7=>e_J=o*TZox-9bEQpxfcPA9h9eD&4n;w84&vYQPTL-Ljy
z%HPu5<}W^F!tOT-%ll(v=gyM9d~a9OM<!K$`_lQUd*AQ>|KsD|_W_y@YA+Wr3SAeL
zb?vm*l8T*Mo>uPSO<2tQvf3o)QOG9svW;avr=)h9m@W0JHr{qq@cYecp`|7Zn0;p4
zV7skpcl%SZ&FO_wZ&odoDU#aQKP7*6ZM3?&m$dZ{gC13-XNzT+b-k=|-{)r;WUbs*
z-0{|A%hnUCt2cf<6uneupA%E(q$klWGq0?9IqPZQBi%W*^UrLnxVXmVueN@=ByX<z
zv*)h+BSLonIvZtfdUE=$o%{4&1h0SXsGi%sy|iM}>7Pe+_*9PTo+w%SNM7r3$uAbU
z?HYA2@0{m}aQ5%JwZ7)UO$8t2qg;CwS8jju^3uWs>l}FnnRfm)Z@8#=qSvf6C?-nF
z<@U-|3nC8Ox5}tp^KYiCsrCe}{>ZFZJ|`}umsw1yHB=Q)-&X7NBe^Xp)-P#qgWs};
zi<4K_>h@0X-nV+{gqgckRj1cDr(d6PXNi(&fnefADZ!vilZ_sKl(u@Ab}YYh4b!?E
zTNZpeda#hqvhLl&8neSK<yF_Z7d*IN|K`L6z3k<Db35;v&V2I!+<u?On!k2d{qf0n
zKAyIABg3@o(!Sk-Arody+|cPf_0h98+E31kGAASjcxKEEV7glQa(804-}L$OR%<ue
zJlJ=0f7!$gf4i{B^OWT6e;%5neu&p1YWn=MANf-2e5LuB!*f~G3tfy=GZc>)&w2cL
zud>bMq{i9r_ht8U=w^hq&*@wIDq*!;>f4&Reztq%{!Y1({KWrAOa$Y*iE)iGe#STL
zcN#zawsq-7!3Ac=HbmTbcVD8i;Zp0LkKa#R-d1#^$MUk;`n{DQI~M%-@@1m2xc~Vd
z4hC|?e)_*JHQ4n(nZ$2&>eGoF-M78T3Wh34TitC^7j3+}@5j3vefL-NL}_@P^*QZv
zd8uPkmBUO!gHqL}SFis%>}6DIC+H@8y1jXW`M292H>^0kaCwi}jW;H@4!C=_epNFn
zoPO!X{uh328??)Jr?LeenzG-t=Y!Jzc@cN3rO&pAWvp7WcA+0n%ZI;*rhZ$bGAYm^
z?Bw;`Yh>OV6tDaDO?L0UPcM_b#kS5qGT+(z)T8*se_!>&T%DYvoHA~#O6A;Vz`9JP
z?^oC1!#SR+CJUbSu$%0!=9Bb!zV}@}`vd78&l+!sK4Otsp;DQ1@aSdtuLa^-%W~sx
z#d*&DdSfnAAz$xlo0lD*J{t;{7OuH%ka#m7xJ_!|_YM>Omqz<TX1Hl6Z{<0d_U4{y
z_%8XR^K$1hg^X&p9~EH^-?p}Zoi*o&rA+m^50)_{=a#+8S&(=xvcxdQ-kbgEJ-(oM
zDU0u{TNkOsa&VT)`v*tPuIf@fz&q2Xatg=3o%d1?%G~ew?CRxia^N_@eWv@_595z{
zvl>!2^i<2wtUPg<|MSgU#(lG|<m^8gZ*FOtAbk9Lc?55}W6PeNHDAswYWX!|?*ZFg
zMNh2-%Z}#tYipieYT<u>jb`8S&@G#mED2IM_w7{w-LTzUHGG>(N-uwW%s#8n?0x)W
zt$))OT)KNPcf*16J@O`{|KkLu%HL;SEL6X7JA+St+xN3-837fur4JnVe~XikCskhG
z*1~TF&uLpBA&WVG%`L9)D9;RbE8C{|Zp#+iJ==c2SlD#oC)33r8b5mHrWhP)kqKte
zS4?icccLcc!Fi87+0-O~I8(;SUtV8QKJ8&#x%Bie;mb!(_bb#cy|n-5!|N>q)0AAU
zb#2bPKl^fG6oWV0qT(p4s6{(nYnNY3iCAuGs#DqiW$OWVwbjo{=2dU--Mp$>?0w0#
zjiJ+TzK&nSDw*+d(ZQqPaT325ar`We@LV(hxxC~}*9E1s+ak&%>%T6V6Hxr@Rc_x_
zrR+b?`#p|^Ok%5H?FkNHn6+R1M?-bVhg~1;HvVLsa-(L_-zJNHub<2k=wLf{{mAjD
z8=O_na>e_)Tjb3SYUmQY7<f>A=g$8Z?roYgL6xa9M8)R(B5!@$FZ%r@tA5G^{{ANz
z_@{MOFRPsXWdG0m8M@a0s#i>XbWxa1XK@{$NzvA;8`N8r?`&|K60}C1&sFSZJO9O~
z8B><ZJhZub@#4=*d*7d%x3}cwn#&poyH?yz{~av*D|pkEztYE^%fIDieS5ky;7@;*
ze)9jNOds}IXCEj&9=kBk`S|ZY(Lqf|v{oy9^A)()dURcuY54L7`ZGQ#y<TF%oH_4G
z>w_Ze^Uq%Y4*Ie1&9ATSPP@{5gA@xwQ;%+S({KE@Nbc+xv*}-FtX?j9kU4isz|1Yt
zp|1m<@2rfv*K=p5<%ZeXYuD84+-c>Bk9FCVf3>@C@6{`@?G9ddj-`B0;=gsjI3?+Y
zM806#f2mUIXBCgOT6h)4Y5fmuGOn1CY45FhXM*^Ji@yCKA70<S{ZdBb+=-Re&x(b-
zR+g(xtKYxewqeSIs<xXCtl1VTR}^PTehYlZ)8L{g@ksd66&d}dyQ*1NHq7Qx*K)b?
zXvMktt|wPWuUUCA$elyJ_pJ`|f5nG7i9E0Mxc2;v{d-g2`c=|q!*kPle)H^!7oQRL
zclN3s)1LD8%@<lbi(~6nrfL7Da(i0+es{Cs{{^j^0Rf>-t5eb(R7DuB2db~-+A6ek
zk(Q>5-cs+sR=-Tw1v{|bo<HYhf4Yv=E&-wIH=RQ_XZYnkU9Gi{=gp}FLZ75B$mHJ)
zD}N%(Y5cEbgIoUQ`v=%(_Hi_LG@iKW@puiJbm>D^&q(Rj3pl1tP~3J=+c{lz^~;|9
zh1o?Lt&O+~HsmsfIE1;Cx%u&l6zcDvJ4r5V(jLpLZTot^JqfqE@j8ehA^C%Qr{Kib
zP0rWy&ggCQy)78EfKg9F=TQ8U(>7&m^2GKvJ<Aq-UN~dRiESNx9J4PMn?Fr)HoShZ
zG3kb><-AXUcGAmFw7QADWqaIWe@yey!Oe|j&6OowrzXl27bH2h^@;A9sn4~gM)mY2
z>4>uH<;ErXE3RGK|HUUZ^O%_T>9##Kv2J~S;fqdBJfl<1!5uICWB<MX?465yDzFTG
z8<KOWu%W4?0rKE?Y)Ey1@Kve%xtHHEZT0dOFPXA==VE1pEv9?6?CtSdxHM)7f3#P0
zqQvA2HXgI4sF-qY+crHcvv}LWwY3)%HO)NZw_FmMV3)w+&iceK;)|YmI>!ROuA-~=
zwC7!YET4C3@8dIf=l)Eqt4`bh{`>EJ_scW97B0w`6&Dc{7WJj*-KMiUcZPMBZcg5*
zRXX?T9^<MnYra^h+B$l7C#gtit7_NPx6b*0^;@(-o}lZ+Gc}e$_w8kRugi5FGc8Jx
znykU<Ykb=`RQpoMs?gVKr{>KL%rDRgUlF1m9Q<^~!YezbzP>bN_LZo~S2N8er)Hh@
zd%A4P9h1~~S2gQRw5CVsdim~j%)O?$*6QiBPeG}Lfn|vkjg}ki-?1Pg^Va{uzrv^X
zJX;c^y|Tv3H#1W<b&1Xv-L^$SD^t8&*Dl_)B1fzJmdbqb0M{#Gt7drhEX?q7)LrIx
zssHlcFZqUjGhT0exiNU-eyyHDpE+#(DoOW!ql_MIzEik1fcYfrGPl<kGx{obRn|1V
zv{RFLF8tc%&7MB<4$t>zQ_eleop@&E7mkjJafffp>{|3sqkabKv*6@~{1-CA&dgCO
zK5_V_>dmLk>)D%X+uIah<<#unSgKLMx3~Ak=gDch{cG}P<euZ&oqS*KNZz`2<}cZ7
zZ2zz@N4@uXDlR@%WZCDcEgSb5-MMt@QXk(Y=1|7ddAH^6eyXmlu&aA>?e6Y?XDXJi
zzkFD>yCkLlk}v+zvUvWJi8jacjQspq${yRyk$!VnJ)z3F{nN#^U6b}tzje3XHG9J0
zc=O0=p}S7$7usJ;{`T?R=Vsae4&fI(4mTQY%Vv-2=elvk{3x%|JfX>c_ZX&5`mbT0
zC3xpTQ`VlZabl|<+iUOhop<kJ%H!^13wF!@y5P}X(ByjfUh@Z|S(_@kJ)YSa<!gH-
zI>uf*I%obv!IEvcMGdUk&QpbblRIQN&H6v3T<(`CQoOz{OKgioeE8z3BY78j<sM%A
zA?{q^=sktiPFc;@{llh?OtwENl7@%(W(!Cy;J#}1cI*Do*CG2`Zag=S+huvNT>kuX
z#^3wgFMJF8zW6v>ldZDNkC(UN_uTeOlZ)KaWb@^2u*3uP`+v9d+f6Xnt5a7?tY<l?
zU*Y^}N@y5I^@6xnX>QWHwit95M%q~YopAhk|0LEP%LT8s+;uu5XE|BORh;~jKkXlr
z|H8%XY)(2SvbSx!)+$&%^`)f9v5HM>>z^E$7<88VX1(^ln#ZrZpX=4y#D%d5PwY7S
zsCMEp>AWLevzJc3bM*PFyOQVgqxs@v<=$^HdHJ<j&c43ACA4Svp~csEPjdX0_&>qG
zDok-lr6^xA>tpFDiAx_m>U^Y8{IKM-!kzfy16DmZ9QN0@)CW9KmYLpSIwMk)@%d8O
z<d`qzGrIKc1>6_R*?-btc70om=b8UUPCe~AJMD2}m&3|t>m2TxK4!CTifrQLZ+ANU
z|LO!?wOZ+$au?nA^)+m`qWF9}mtMvJv1j(t$Bh#NFIm0O6U+J1BDJebDJ*YUsF?W^
z%M;E1?w&Uk9^b4o$~$u2V8MOysTmVZB@5V!pP!m|a;4wxFpnP5kDJf%olZXBJ+nD5
zjlajSUf^Kd&40T$v>sM?$}Z&oC}^61!=(4Rhc{Y$yJyRjP$j*6mw~nV@`7LA&K5t{
zi?y5m|IV6*w2d}ejQX7qLjN=MdzoI6S+~8cNcB!z+&VqY#NCe@i?#)Nv8^eMp2cG$
z(eqAokJq8dp4&?OYP+`Xczgb$&M_Y5Nf+N<j!!&wx=E+{z}@TOALRRP@ow)5Gkh5=
z_+Hqtamhh$i<0>|Ja4ipZ~cGq?%F9ok%UI00*eanrmgcA=k30(XtCF?GPsZXAVa;Y
z+U3h82@Rnu>fZ0G+UnnZr-SvsihkJIYo9f~ra!+s>m%Dnv9}4mZKrsCXlcDIh^gKD
zeXX8xlBwM8x&FQ7oYMktG}gR0aAL~OkYM|ykFOr;`=9+*<n-<FBaUaO-qRkm-m%tL
zaqs0u>xIjA<h;-BxZ9z2|6!r1U*~uC+VtLEdQXj)T^GJ|>ZEMJU+#~a56ezCTRito
z;xwP>@w>~CA4&h5^=ir`>%yw%|NPj_h30HjJCo@>qg&`%XR-3<WlX<YUtGSldB3S*
z;;u-6TPY<gVl(_JJYM<!V`d8zoPw<`G9qIl*2vV%2)Qm=8&(|={9M*fvA{-Ysqieb
zjc+YCuHJU})QY!JvDasd=Id&hXX`dK+|hq1#L*aIpeX)BRYlQ3NN8Oni^_ryuGOwu
zL9?^Mo3C!mtG>3=>h`+ZyX^YkS-<{xvusC1ssG)7k)Letf1dNc=DGFz-~aA?uCwP8
zlj$k6n8ThVyzOmAZSL3XjT;vwU7UO|dt29K?$1W)`FkDgJ1+7UzG!dGnB*ON@X6EM
zJ3aNStak&98tZHR%vF4n$n#hy{#4(Mt<~wrwe4rg&QEUNu!BRV<?z8^u2W017cW{o
zzs@y8N%!>WlcL<$-Q4ZBrtw;@6HoJKm0r|x`;sxY%cM_DHi=aq?H+DhaYG|<j>jRt
zXPollqTI%tiY{|+Ze5g<7<5f9u_(#MX;DG(p>I!QJMBAv@n104f9#U#`m$ALhwJH$
z^2ym2FBNw>wjNB4;o&~iw6Wk#Tn_U$u5E(d2h9>^Z(MuB>`kbSSoh&=$<`YxbOgN*
zP1-p32D76?d+fX;JLi99O6Px`6_9j+#k{lHZihljrib}HsdiU=onP7;r&;W-IQ_`q
z`LUDa6Ylo>b)lDTtH^|WG}rsfRy*l>hVhJpE+yA?bbH9Am|S^R*ZH&S{(|e*83Y#D
z6->B$fh$Zu@Py4t>wtGFL^@A#cUPs{omcmr(P+<VAzSO_uyYI2Rp#8koTgVR@b($+
z^QK8Yhn8PBHDU6OT#2@uol#yb^83;<nVBP4-WpAB(P_8p^!v)D{J_5Ky7l=-r)*A~
zm1UDI<?>GoYg2i6>dwy2uvPNmotqasc!ul{dAme<)-B&{!X5|yKhyiX|8TCefsf(C
z4DI}~)yJPPv?+h*Vds32p;s$1cV26pw~=_>zHL*T#GlD?vu~Nw%oU&_{`aNLd0p8P
zS~t{h&3-(!itjfc-{gl<_Pbs@S9l~dUCy=tlMnAZn;@2|rMo=BRlTP$C}zIc7?5DP
z=|BI3=KL&XWs^<2Zm;&_)X4s_s;T+sGA#v(9LvpJ3&ZAj{GTwZEnY4=(@?wG$nOx#
z7nzN>tZVu-Hwc}v`ohs4`9;S$$A7i;^;Rc4XYYH~Pdt2M8Je6Tbol2wKVq1EZ2{xG
zaL@f!^UgHqRM!0r{33Hd!7S>4Q-$jSg~bJ~hEKZpBrIR{vGh#(E2aqzP8$x%*UJfN
zu-JBK6iuFZ{6Y521-`pe-uEq-psi8ytRVhc-}^{=8Hb<o1uyq?+k}U@^G0%)O3jT^
zIxBAIsPnP(L-~=Pr?;LCvEET)Rynt9ztpMDNq(-^9m3xqEWc;-i|cT*(hPRVu9hF`
z&Kx|XJae)E!{LHM2G0s+rQ1AE&RZ;c(97oIeigUw7gww^{>e<As&hf+@a7q&pR>6C
z@n`9OUd$mTCh1VxxI1fuoWQkggOUKd8n%1+a)p1ZR=ngX5nNv0o1ZXk(;wxTqB_6N
z@%X>|F|SZ!=j6XtDUWk&Tz(pF;*j6><>JRog|@0y-YW~1o$}}Zc`?X7{g2kK&!Kv?
zeRhYW&wgI3F3);2?dt5(pU>9asz~D2(@x!49r5n{BhE9boSRpqT%D~D&~$Bq*({x7
zZW+pKr=nMNYkPMGA9s<TApY>h(rI<iH%=>j{X_DL?0^0pd*y33I|ZG8oR)uCaPr+t
z>1DEPhn{3!2wZSE)6ucx_Uh_icVeBl`$#m}@td$Ho!!nA#$?sH&1>7{B@381Quki<
z+xRhQ3$MV$c(d>OS}O{=c1$pm@m6zoi`@E0<JZx<t1SAA1<f7Q+p8{cs`#~be`#8J
z{E~H(T)p&BMVXF7Z&yoJ&6Q{No~lv0v}WbSQwzIpyggm}`t{-$|IV>a5nu13;Ceu~
z<Ij|^J8VJgg2UdQ(9vLWQp<j|WWmY83%5*vNbmb?`&^^@xay2M)drQWOq(s@b0_gk
z*qe}NxP7O+TyZJW#k||`zKveXvo5SpHu@3Ox{pug$hnPo%J<E6TiBek#GtpX>w5Ur
zu)v@|wRJ@y6}xTTGCzs2bWdlD-P@J2xq&7B!@rJynMF;Kvsf=zyFHT3<0<z4cu3m!
ztgqkunUDY4DOkn+j$CE6;`+&XDgTa!r!KXXWp44}RpXNCb<0R<UbcCSOH81|@7urf
zgjf%@U%zdyq}2L?^Vctag^NEjO<%uMx_05IxvM7g+937oAzn$}m=@l=YvHnS1CJW}
zp_g~dMLKuB4_gv*wADQB+SLXrlcpJ;RVUw<b2>fe&A+!M&z@g6dHGtVnSQNJ!HS%#
zZfhNxFEHk1Z!g}Hyy`cPk3b_&zST^on@f*YMulJ9sB)r3$#ZMW{phPJ8o4fbec3wm
zjByRaitwo+Il5^DE0kUSf>u6kuvm5{e!i3F^u`y4p4u8G#P_e5+--h#UC{sOp<P%0
zm|oTVnlwR3e;!-#{GH36rd_${&%UF4M&!Dwr^>^hZAiQJLRVMhMzoFZ+-vg;MHt@;
zq;Gb6e8S|!l(h8I$2LWuRuuQrUi#Ct^^;40U~%f3cDLxs+c%zlS<aYf9I{$7!0g4V
ztA=M3>hIk-(zNsJ+ilye?F!@NY%Z5p{5ZlAT{2s1e$$l_wUBA+CT67ucrM|2(p$e%
z;bV=7JKKadULWO8AJ*m1t?P_D)M9j6QY*{qnX%cu_l~D8coi8-zn_z#_Ty;mqsNmq
zesCZ8S+PnvQ9t{Hk^QfickI?Oevv(Wr-kpMo&V1S<>*PZN`W(fxU33fd>NoNX?mFG
ziE^j6zCmuYuV%m8IR8yjqSUD^A}?M?M)~~Um9&38&H6&tp5C+O`JA)0SN<scvTI?T
zLADSJ$L_u`MMcx<cYhMzYzw`V^X!u8n}we_cOA0cesO!_6elK?<Jx>rs)HIMVvnu4
zayDsUMC03O@18lYeH^DS%eqvF%VhKYlWE5eUq2bq_;ow?o5bW_P2XMa|9EhM<;KH;
z+Vobp)gjK3f%gJ0-+i~)w`TpD)pc8ro?p5&gO%Sd{^q<S-<*?qO?4T0ufFeOo1=Vz
ziBF;>FEV!SKCy4Ym49zd^Ygzw@#K_;>N7+wFTPainC)#lRj$*!KCaNAxXM1oaqAuh
z|5?FP7wWUIsyKRvCojIssds2)iTKH)(Dm!x_)ULAta&fJtme4Q-~59ccP{=oNB-zW
z-@E%H_15mnJXdt#ZMe5~@TE(aZwRd3QlYt_;o_Atv8|u>&$(V}cfK}b=1Ifby&P64
ze43FwrK(?7<)n%J-*)YR(7skluA8MQZ*{n=`VZ~BzxqSSgL9W7)(K{-ZhD)etbE-%
zmwWvmft}B#<%BP4Z`8U`TDs`huDxeY*YsMnM}8C%sLVRgz{-5>VseJz?fl4tem<$y
z#*2+4;-1b~x+h`N>^bYc--){x<g3;@>91eOZ_}%hr?V#|@^8#DN}naur1-nBG0by)
z1&iCl>yKZ&wMv<tl&raD+thnmZs%`H@mfeq%~9csz9S;WmFWFvVWaBf+_szZPhQ9}
zZEu}xG~<lhw@nuxWO40of4A@b{o@(`PGPnyEfB5BkjjEoegCAa;F4T@KLtYrQ+>D0
zqT&(-3ln2~x6GW9)FOSioWzn;m(=9^lvI76)V%bP3<YCTb2H>7=GCa&;M*QT|LXQT
z-RHY5I_vwyDYtiNaU@QUjfhatuNIMSv1N<xIpSsH`R)4s%gVPsHp$F+^K6@5-$jpE
zFJIREdO1(-WZ(g2MW%&9HO!iY9ZXA86)G3-@+ld3Hc2}!$yF#7P=Aree9*E_@R*d`
z1&x_E94Z|p^qCVsxW9C`p}{Hg%bk-yM5g1&Er(=|<;qXI`#P0mlqDrPLM|LhaMCSF
z;O#3ES+Lq<8avzbCA|lC9(}_0c+JJeYZk|hE=Y(M-f92iG4lY2jP8?1bs~?eWKKwY
zi##KoE$pzQ$8f#tewL~Qt9?>rBEI-9;MsLTMT}7=T&8I2p+MKhzCsuE2a79|n7Ajm
zdNgsUaoy=>QDA5gSbIV6!b6G2GLKCiIuy=3WV95Dy>Oo(Ug(fh;FHp@POEX=j40I&
z-9E~D7?z4LD7Kw6+UA(EK&ZzhIQve&XKF+C#q&;;4$GD2+>p@HX1)K=<J3h(&ILRD
z8kQ<NX)sMQld%&t>?`3r!g@}4lgI}T1+7+IAuF{ORvWg63npAv2R^$U7H`?ZR>S#Z
zfwP00=?~^MiSCbv>!K5yUMiI^A5N;6^k>!q<tF9;-i=H=n+0DU=Urf0!hJ*QKEn;c
zC)`f@jPDedHE=Pl=@gCeICk76*F=cZaRXnT*phv1HW|VPlbJNs4UZ+yd2mT+q2ko+
znBxHllW#b$JKEE5uz$h<Z;wf99VGs!uI!lneqZwR`EoM1Pk&yNCjNQ$taE*v|JdHv
zKh>$d-lC~DezW`y!RpS*7asgKy?yuq_qW=@$T^!2%ZBa#WOw><<WZL&1~dC^928A&
zkF5Q>zJ6n|`MUK>j1qmWOt5)!DA-12#+04{TetoD9Umo5yk64KZ>X)p{C1gR!jCpK
zt_+8Fk&!+6F||j}@$ubU`a5Ft;@@Ac?akr)efD97;!p1$k+UZ!cfLt*KXLJrh=RnS
z*awY#!Y<-B#FUgAIV7`0ZL?1uVk*lx*|_iN-)q}GN6*@RvORCo)i*|K?{Sw++E;!#
z=UH2=r=5vQMe-q*mqI;Uyh=BtM2|WB5!F((a$>ZKlss%cC2~<f(><|$T`hmKL!6$p
zaXIQN;r+N&MMf({aAC=boo_Ba>EzyW=ZnPVM;E%R{_~p%hP4}taje$xONz2REOGLb
zQe4u?{jFY~ET@Xh=$2Xc=z{u3kG5%-t8$&SC->?_9<P5oEA>fTw0{C`gw)?(e>(Q*
zE*7oU*tANk$ddE;Y8_Qyk*uioE+TFbQ#3OETE575Yu?H6HD*H9j!ef%iA|f>UYmqC
z{yc2i6!!hWf1cm%O;7cUkHp5>hv-Y}mhC@$yGBmzsjl{?$Z5Mh4cFb!*}YTuo&1f}
ze+;#||J-}}_1)|(XD-d2x%$xVUv=M}Y;tK8E9H17D5+I1;83FHlJRv}X)43if2)2@
zPhj%M+hjGTHtA}_k=^eMYd48h8g7{w@2GlpI@_F(Zc&rK^8ydO*IJ%?w|UpDTMrk@
zD!N^N^Zd)bpTTSH3T#()i!$epG4kcUZc}6{xPqzd%w&!^O6xB&nCuZRyK<u7{sm>_
zn!`N-h3mUFEuQ<ubCT>#xAu$_DbYt0KT5q(QNA+siif+JoIr{3t(kV5yCiR>{rh;r
z?6beywe$NQyKOV9u-dds^~Lg>AVss6T)Am~4oW*m&vf0c{V;Vy`Tl>m^96E*4#oeh
zG28v_zVfzJSJ&<^=9zhaUE9iUx@$!)U4Ec@)K!D+y~n+<#l_wkvP!pkvsMVLIgnQA
zKkJ>Xh|uc7uJRI>eV3%$Tu-c%c1fDtVU!t>(lV*$LtEEmf%b#dH{P}0;M9Fzs%9M3
zdSerhV^Tnoy8q!x-)y_D9^_k`8t`=H(q9WsH?DrwRkqzuq;}VdS2J(_Ed2L@^L7J6
z<l(y_M}HrvEqPoySM=wKaIat~3vcffr(0hxhS)2&WHR3puH1Pel;v`X*>2C{@2e+0
z*|lWzCEH8WzIm>C<L&xrhW!_nmA;c)yeGZ#Ui4INhtKI~c9$w4U++i1Jfieg-{bIA
z=G(k!Rpxu&%W}*8-%dX&GpX$8f@zzUE`52v^w*11$IsNh@D<-GYNE>+7|&pJ_29kk
zgzqI@T()6~vsS6=#fe{kkrut^vE#+x`b;V%tE9`+BRjty^lfr}uEs3qZ+2FQS5c-z
z{Ghkaz6mli=PungRlh0rWE*ScowgWWr*%<js?kn$m#^(R`X?f%=}*nP=aF@{CmY92
ze>}A-dQbmTwKd-b0_U$<!_}2pFeSXyQa*Ufx0e<-O*$o7qV{p<b!Gl#WPacqzWZbB
z-gT9~SbjJ~@>eW=^PDk1&?GFH$u9S%z;+w4*}e_(ckR!}h-Myq*?ijnT!MF1_585^
zF#?*hUKusDv)Hum&bpNM>3H2V|Ar9xdPOr&(`LQ^eKD=?Pj-ksS-a)Zi_euOu2=kZ
zp5?#l?qwT`;-_0Q-e0bHW02h&^yIr@N`>~mTkpc&?p%EJP9gWYtquETOz(epIpf7f
zpYlD|?`|u}7m0|trXjFZWJ1XOAKZ1d*5#pw`p4HTazAtBbok+P4UO<~5zE^59Lp?K
z-IICU_h`&T=F@3eIp*z4+ztl`-<zCe9W*)hMy%)kiE>51n`LG=zk2&@!YXTb>vWyt
zVcs*Up8ejmQB#y@qn6M5CAn9mm_pQ~uLRj{eY7sBDsxv>PMEWq_TIS#S2y3RS$WpC
zRNTaL=Jv>&inWh;f8Cg87Z^TY_2YUI`3b9@mIZBn;nlnUir3Z+MwxF#uDQ8&@7tH{
zypHuAck8;FowHM3)_uJAUP||ekm~7netqxh*IaiN&xzYrez1MbKJIr0&XKz^KQwHW
z7w*kiZ9cO<cIW4Ff%<(yV*lOFCZx=1-cn}GuCF4#@qqow(nDYMw%0%Z^L}xfoL6^s
zU2n9=kuL#<ZWjjm$W+Jfyj<xk{cY7v*NdwKUv1m7%S$Kar;tkj(bHS}(vRv^cF$9e
z^!vW=oATlv<;xC)KK<$S-f&0$<%>nq66d!63aZbX7PR&LE4K-6;~)Mo7rM2bciU`1
zvyLw*(T$VO?kQ2f_+?UC+>aX#a`zu-F&J_%ceA)}Qc_Voxbypl<9R;~Ykpq8FkADk
z&(fD~rH)*la^sV-Zij<X&C5Br46m{8^2?vC)1Eag?zlujyz-ZqQty)Ur7lY>@zigx
zmwmylvGMgrySx8?&R=>t%p>QA-dvj(It!$J`kddp)n<Wt@(tx*`A<4G7ckt-@L9-s
z<<+hwzZZ7y^%h&Vuvo}*@`azm+$%&MpPQt8R_4m}{kG9(yvjYAk3IFyn11K2^96OU
z@0(3#Y=2WZ>F`1CyBqQkK5MJ-zAF_V#Gm-zSt`3A_H#z6;nCY#X<`#K#4mZO^=58~
zIkaz)!F#JWr|r5e8z+6_?*6E+bGB;#(XYPo_kP}%-F)`P`LB<gZGRrvb9nFmFTZ=l
zTBa^)_c<FcS8_gT#@2Q9NAFK%xi{(Vyr-{K?Cn4G-+Q0^cWc!@$-O_Us<hXA=Ih`0
zZ1dd{btzkS{_|N=@@P_ikJmr@A35T>(QY5Uu9~;__?y+o|CG3H*nV@#zSS>|^ZIUW
zd3*Wv?kz40J$9?#7irrZ{&UBIxvT$MGJd_|ul@Y!+&OZ8D{8*<Uw@u|{?bjIxhY3K
zzN;*p|LY`QjH%vc_K>=N)A{f2+H#wb_1a;6z1@2kznz^@ef5TY?}TadoPK><*TZV<
zBs#74`={G8xh;0DxcgW7O=;xk6UVD&{tjt+$CaJgvS_|<K;q(cf6pGydG_K~QaQ_S
zy_mg+mr2e(H{-X(yrtjYY_AO5Jt-uSKScNZosx@LdVSs1H`;%FJ@=HyH7Cp`!qj`U
z^|#&H++k1puF7+T-Jg1Hqvop0j{1@{r|dSJvQ6qSpZrwALNPn+$&Q-&rfcs%Tes-b
z)VCXJzfF?e5dLOD?fduxe@^t><orEz?k3~1zS@tf)o1T**#0Km?7aHAHOE|EMx4sK
zF~e6vEp7XozSKANZ)|68te&>H$#%p2h?81jn=PwT<Zrlt>->8z@7d;=t7o28-;nqu
zanYlS#o5(%GHD`zebS7VF8RZ=RPWxNzyIFy->v-KeKTclP+GQEqm{PTN3o?oHXfRa
zR|S^$@RW)2d}fVU<TE|Es5dya&&Ize{YJr#2M7Hh92Dlh7JIMk$E@AaeD_$sOydi)
zWzW9*X+xj=<A>YXr%zvhuT)*ld`<O@I^RnHtDZhQQD(Dec1zmBCmL}-EFR^myU&^9
zW4}%)B(!VY<kIUW=f+&%{I7Q6yTF0_hco{ksI`8uO>z0(dnt=go;{Pk@M)*Ni>`kj
zQ;E0#ta<8;3EE*tPfeDX`2K0=Hmmr`FMp<++rN+h^Z3V;B`3HKmY=`-wr1M2*L=O(
z%2q_Q>X^=#>AGVmvdK)cBO;e&4yTywm)rh&IoD@ydws_B-}&d?%}*@9nR`y{XwRA-
zlVbcY-uZFwuNnK&{Fg6Vwb_M!RayO?dUwjpAA48t-7Nk#?c&cSng2)RSZcaeq-(aQ
zx-U9k<dT1I;$`#oeev?T&-%@SRsMyEIBH2O;^1O+-QTuU{zu`$-6EarnwxqKt2!Rw
znf0rG3R_-abilp*8Jl^N?`<egIJ<xM9>>iyORCv;KR*{vnX1diTV3GI^wrxZH;P55
zjGJli->-L%>Z^CZ?9UFU*rDU`&!DkIcAd#pg+>!y3Bw|uzaHVLvP_+EOLsnSJKPXl
z(e%flm7Sxyzs7nkztkStdrmca39egM@2NkxpEp;=UbTm9lIfn=+ePZr4!q-%ElN$b
z-&!t~HCLyo`c2{cr)7%sPrtpZk=yPvIp{jmqF5<@rOOTuA99a;Q<&{1TPBno9(ywF
zw&02-Tc0jqfA*hsq3-W`%#Nj@ks(^wG9oB7O+P3<zeE8f6Qu8%mzJ+!q+p;Br0?kx
zZDVO_VQyq$X=!d`k(!cfVVRt2Y@V2CZfb67lwxX>lwwC%MXY{sW>qSep^=e(c(kE{
zk%Cbyo}Q}_>fI5l7u@}r0z}*%R&Q+J)me~qT<=@gjew2eIt{#z8_(Wp?wnm9ey-<S
z32UFv9E-9iLI&#PpPoMXm;c2@WqNUk_qRm;XAfD<J&<v^f8S+APuFp;L_wJX&hS?~
zH)pYX7cl<}U}$~icT-q>O%r$5mH*k!hHIQTr(_*Scb3d)jBd6s`+R_JO7@GZd7>9s
zOD{0a4dA>Rz|VVu!#04`H-LHTgZ8u!P7F~Cd7~D{)&%f6&S;i=(Q5ai-BF?;P@*YO
zqEVpdpoPx?6CWpuWeH+Q0*$E=4H`KIxYiuh%Q<8wlkwx=zLx=$>%$87_D89H`FL1s
zrT_gf_g{ZfOV{W}uC}&l6<P9-X;T63_FpeWf3k-qXnw(r(&CcDqLPZD)HE(rV^bp&
LE>%@me>W}wHi$0n
diff --git a/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.html b/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.html
index 8847eff..639b1d1 100644
--- a/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.html
+++ b/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.html
@@ -13017,45 +13017,6 @@ ul.typeahead-list > li > a.pull-right {
.highlight .vm { color: #19177C } /* Name.Variable.Magic */
.highlight .il { color: #666666 } /* Literal.Number.Integer.Long */
</style>
-<style type="text/css">
-
-/* Temporary definitions which will become obsolete with Notebook release 5.0 */
-.ansi-black-fg { color: #3E424D; }
-.ansi-black-bg { background-color: #3E424D; }
-.ansi-black-intense-fg { color: #282C36; }
-.ansi-black-intense-bg { background-color: #282C36; }
-.ansi-red-fg { color: #E75C58; }
-.ansi-red-bg { background-color: #E75C58; }
-.ansi-red-intense-fg { color: #B22B31; }
-.ansi-red-intense-bg { background-color: #B22B31; }
-.ansi-green-fg { color: #00A250; }
-.ansi-green-bg { background-color: #00A250; }
-.ansi-green-intense-fg { color: #007427; }
-.ansi-green-intense-bg { background-color: #007427; }
-.ansi-yellow-fg { color: #DDB62B; }
-.ansi-yellow-bg { background-color: #DDB62B; }
-.ansi-yellow-intense-fg { color: #B27D12; }
-.ansi-yellow-intense-bg { background-color: #B27D12; }
-.ansi-blue-fg { color: #208FFB; }
-.ansi-blue-bg { background-color: #208FFB; }
-.ansi-blue-intense-fg { color: #0065CA; }
-.ansi-blue-intense-bg { background-color: #0065CA; }
-.ansi-magenta-fg { color: #D160C4; }
-.ansi-magenta-bg { background-color: #D160C4; }
-.ansi-magenta-intense-fg { color: #A03196; }
-.ansi-magenta-intense-bg { background-color: #A03196; }
-.ansi-cyan-fg { color: #60C6C8; }
-.ansi-cyan-bg { background-color: #60C6C8; }
-.ansi-cyan-intense-fg { color: #258F8F; }
-.ansi-cyan-intense-bg { background-color: #258F8F; }
-.ansi-white-fg { color: #C5C1B4; }
-.ansi-white-bg { background-color: #C5C1B4; }
-.ansi-white-intense-fg { color: #A1A6B2; }
-.ansi-white-intense-bg { background-color: #A1A6B2; }
-
-.ansi-bold { font-weight: bold; }
-
- </style>
<style type="text/css">
@@ -13089,7 +13050,7 @@ div#notebook {
<!-- Loading mathjax macro -->
<!-- Load mathjax -->
- <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.1/MathJax.js?config=TeX-AMS_HTML"></script>
+ <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS_HTML"></script>
<!-- MathJax configuration -->
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
@@ -13116,7 +13077,7 @@ div#notebook {
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h1 id="Hands-On-Performance-Optimization">Hands-On Performance Optimization<a class="anchor-link" href="#Hands-On-Performance-Optimization">¶</a></h1><p><em>Supercomputing 2018 Tutorial "Application Porting and Optimization on GPU-Accelerated POWER Architectures", November 12th 2018</em></p>
+<h1 id="Hands-On-Performance-Optimization">Hands-On Performance Optimization<a class="anchor-link" href="#Hands-On-Performance-Optimization">¶</a></h1><p><em>Supercomputing 2019 Tutorial "Application Porting and Optimization on GPU-Accelerated POWER Architectures", November 18th 2019</em></p>
<hr>
</div>
@@ -13128,7 +13089,7 @@ div#notebook {
<p>As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.</p>
<h2 id="Jupyter-notebook-execution">Jupyter notebook execution<a class="anchor-link" href="#Jupyter-notebook-execution">¶</a></h2><p>When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the <em>edit</em> menu above.</p>
<p>You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.</p>
-<p>If you want you also can get a <a href="/terminals/1">terminal</a> in your browser.</p>
+<p>If you want you also can get a terminal in your browser; just open it via the »New Launcher« button (<code>+</code>).</p>
<h2 id="Terminal-fallback">Terminal fallback<a class="anchor-link" href="#Terminal-fallback">¶</a></h2><p>The tasks are place in directories named <code>Task[1-3]</code>.</p>
<p>Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description.</p>
@@ -13138,10 +13099,45 @@ div#notebook {
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Setup">Setup<a class="anchor-link" href="#Setup">¶</a></h2><p>This hands-on session requires of GCC 6.4.0. By loading the <code>sc18/handson2</code> module before invoking this Notebook, we took care of also loading GCC 6.4.0 into the environment.</p>
+<h2 id="Setup">Setup<a class="anchor-link" href="#Setup">¶</a></h2><p>We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [1]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>gcc --version
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>gcc (GCC) 9.2.0
+Copyright (C) 2019 Free Software Foundation, Inc.
+This is free software; see the source for copying conditions. There is NO
+warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+
+</pre>
+</div>
+</div>
</div>
</div>
+
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
@@ -13149,17 +13145,19 @@ div#notebook {
<h2 id="Tasks">Tasks<a name="top" /><a class="anchor-link" href="#Tasks">¶</a></h2><p>This session comes with multiple tasks, each one to be found in the respective sub-directory <code>Task[1-3]</code>. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.</p>
<p>Please choose from the task below.</p>
<ul>
-<li><p><a href="#task1">Task 1</a>: Compile Flags<br>
-Improve performance of the CPU Jacobi solver with compiler flags such as <code>Ofast</code> and profile-directed feedback (<a href="#solution0">Solution 1</a>)</p>
-</li>
-<li><p><a href="#task2">Task 2</a>: Software Prefetching<br>
-Improve performance of the CPU Jacobi solver with software prefetching (<a href="#solution1">Solution 2</a>)</p>
-</li>
-<li><p><a href="#task3">Task 3</a>: OpenMP<br>
-Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance (<a href="#solution2">Solution 3</a>)</p>
-</li>
-<li><p><a href="#survey">Suvery</a> Please remember to take the survey !</p>
-</li>
+<li><a href="#task1">Task 1</a>: <strong>Basic compiler optimization flags and compiler annotations</strong></li>
+</ul>
+<p>Improve performance of the CPU Jacobi solver with compiler flags such as <code>Ofast</code> and profile-directed feedback. Learn about compiler annotations.</p>
+<ul>
+<li><a href="#task2">Task 2</a>: <strong>Optimization via Prefetching controlled by compiler</strong></li>
+</ul>
+<p>Improve performance of the CPU Jacobi solver with software prefetching. Some compilers such as IBM XL define flags that can be used to modify the aggressiveness of the hardware prefetcher. Learn to modify the DSCR value through XL and study the impact on application performance.</p>
+<ul>
+<li><a href="#task3">Task 3</a>: <strong>Optimization via OpenMP controlled by compiler and the system</strong></li>
+</ul>
+<p>Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance.</p>
+<ul>
+<li><a href="#survey">Suvery</a> Please remember to take the survey !</li>
</ul>
<h3 id="Make-Targets-">Make Targets <a name="make" /><a class="anchor-link" href="#Make-Targets-">¶</a></h3><p>For all tasks we have defined the following make targets.</p>
<ul>
@@ -13184,11 +13182,13 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Task-1:-Compile-Flags-">Task 1: Compile Flags <a name="task1" /><a class="anchor-link" href="#Task-1:-Compile-Flags-">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver</p>
+<h2 id="Task-1:-Basic-compiler-optimization-flags-and-compiler-annotations-">Task 1: Basic compiler optimization flags and compiler annotations <a name="task1" /><a class="anchor-link" href="#Task-1:-Basic-compiler-optimization-flags-and-compiler-annotations-">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver</p>
<p>Your task is to:</p>
<ul>
<li>Optimize performance with <code>-Ofast</code> flag</li>
+<li>Verify the cause for performance improvement by viewing perf profiles of O3 and Ofast binaries </li>
<li>Optimize performance with profile directed feedback </li>
+<li>Generate compiler annotations/remarks to understand the optimizations done by the compiler with and without profile directed feedback </li>
</ul>
<p>First, change the working directory to <code>Task1</code>.</p>
@@ -13197,7 +13197,7 @@ build <code>poisson2d</code> binary (default)</li>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [1]:</div>
+<div class="prompt input_prompt">In [4]:</div>
<div class="inner_cell">
<div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="o">%</span><span class="k">cd</span> Task1
@@ -13217,7 +13217,7 @@ build <code>poisson2d</code> binary (default)</li>
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task1
+<pre>/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task1
</pre>
</div>
</div>
@@ -13229,17 +13229,18 @@ build <code>poisson2d</code> binary (default)</li>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-A:--Ofast-vs.--O3">Part A: <code>-Ofast</code> vs. <code>-O3</code><a class="anchor-link" href="#Part-A:--Ofast-vs.--O3">¶</a></h3><p>We are to compare the performance of the binary being compiled with <code>-Ofast</code> optimization and with <code>-O3</code> optimization. Right now, the Makefile specifies <code>-O3</code> as the optimization flag. Compile the code using <code>make</code> and run it with <code>make run</code> in the next two cells.</p>
+<h3 id="Part-A:--Ofast-vs.--O3">Part A: <code>-Ofast</code> vs. <code>-O3</code><a class="anchor-link" href="#Part-A:--Ofast-vs.--O3">¶</a></h3><p>We are to compare the performance of the binary being compiled with <code>-Ofast</code> optimization and with <code>-O3</code> optimization. As in the previous task, we use a <code>Makefile</code> for compilation. The <code>Makefile</code> targets <code>poisson2d_O3</code> and <code>poisson2d_Ofast</code> are already prepared.</p>
+<p><strong>TASK</strong>: Add <code>-O3</code> as the optimization flag for the <code>poisson2d_O3</code> target by using the corresponding <code>CFLAGS</code> definition. There are notes relating to this Task 1 in the header of the <code>Makefile</code>. Compile the code using <code>make</code> as indicated below and run with the <code>Make</code> targets <code>run</code>, <code>run_perf</code> and <code>run_perf_recrep</code>.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [2]:</div>
+<div class="prompt input_prompt">In [84]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_O3
</pre></div>
</div>
@@ -13256,8 +13257,8 @@ build <code>poisson2d</code> binary (default)</li>
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm
-/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm
+<pre>gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d_reference.c -o poisson2d_reference.o -lm
+gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d.c poisson2d_reference.o -o poisson2d -lm
</pre>
</div>
</div>
@@ -13268,7 +13269,7 @@ build <code>poisson2d</code> binary (default)</li>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [3]:</div>
+<div class="prompt input_prompt">In [73]:</div>
<div class="inner_cell">
<div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
@@ -13288,25 +13289,24 @@ build <code>poisson2d</code> binary (default)</li>
<div class="output_subarea output_stream output_stdout output_text">
-<pre>bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
-Job <5033> is submitted to default queue <batch>.
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24897> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-1.13user 0.00system 0:01.15elapsed 97%CPU (0avgtext+0avgdata 10944maxresident)k
-2560inputs+0outputs (1major+264minor)pagefaults 0swaps
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k
+256inputs+0outputs (0major+480minor)pagefaults 0swaps
</pre>
</div>
</div>
@@ -13318,20 +13318,17 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>You can use the GNU <em>perf</em> tool to profile the application using the <code>perf</code> command (see below) and see the top time-consuming functions.</p>
+<p>Let's have a look at the output of the <code>Makefile</code> target <code>run_perf</code>. It invokes the GNU <em>perf</em> tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to <em>perf</em>.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [4]:</div>
+<div class="prompt input_prompt">In [74]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="c1"># perf record creates a perf.data file </span>
-<span class="o">!</span>perf record -o perf.O3.data -e cycles ./poisson2d
-<span class="c1"># perf report opens the perf.data file </span>
-<span class="o">!</span>perf report -i perf.O3.data <span class="p">|</span> cat
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf
</pre></div>
</div>
@@ -13348,49 +13345,121 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d
+Job <24898> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+
+ Performance counter stats for './poisson2d':
+
+ 16264721613 cycles:u
+ 28463907825 instructions:u # 1.75 insn per cycle
+
+ 4.738444892 seconds time elapsed
+
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Next we run the makefile with target <code>run_perf_recrep</code> that prints the top routines of the application in terms of hotness by using a combination of <code>perf record ./app</code> and <code>perf report</code>.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [75]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="c1"># run_perf_recrep displays the top hot routines </span>
+<span class="o">!</span>make run_perf_recrep
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d
+Job <24899> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-[ perf record: Woken up 1 times to write data ]
-[ perf record: Captured and wrote 0.172 MB perf.O3.data (4125 samples) ]
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+[ perf record: Woken up 3 times to write data ]
+[ perf record: Captured and wrote 0.739 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (19102 samples) ]
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio
+Job <24900> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
-# Samples: 4K of event 'cycles:u'
-# Event count (approx.): 3867635297
+# Samples: 19K of event 'cycles:u'
+# Event count (approx.): 16254596654
#
-# Overhead Command Shared Object Symbol
-# ........ ......... ................. ........................................
+# Overhead Command Shared Object Symbol
+# ........ ......... ............. ........................................
#
- 72.02% poisson2d poisson2d [.] 00000040.plt_call.fmax@@GLIBC_2.17
- 10.16% poisson2d poisson2d [.] poisson2d_reference
- 9.99% poisson2d poisson2d [.] main
- 4.69% poisson2d libc-2.17.so [.] __memcpy_power7
- 2.23% poisson2d libm-2.17.so [.] __fmaxf
- 0.75% poisson2d libm-2.17.so [.] __exp_finite
- 0.07% poisson2d poisson2d [.] 00000040.plt_call.memcpy@@GLIBC_2.17
- 0.02% poisson2d poisson2d [.] check_results
- 0.02% poisson2d libm-2.17.so [.] __GI___exp
- 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object
- 0.01% poisson2d [kernel.kallsyms] [k] arch_local_irq_restore
- 0.00% poisson2d ld-2.17.so [.] _dl_new_object
- 0.00% poisson2d ld-2.17.so [.] _start
+ 65.50% poisson2d poisson2d [.] 00000038.plt_call.fmax@@GLIBC_2.17
+ 21.21% poisson2d poisson2d [.] main
+ 9.18% poisson2d libc-2.17.so [.] __memcpy_power7
+ 3.28% poisson2d libm-2.17.so [.] __fmaxf
+ 0.74% poisson2d libm-2.17.so [.] __exp_finite
+ 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17
+ 0.01% poisson2d libm-2.17.so [.] __GI___exp
+ 0.01% poisson2d ld-2.17.so [.] check_match.10253
+ 0.01% poisson2d ld-2.17.so [.] do_lookup_x
+ 0.00% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x
+ 0.00% poisson2d ld-2.17.so [.] _dl_relocate_object
+ 0.00% poisson2d ld-2.17.so [.] strcmp
+ 0.00% poisson2d ld-2.17.so [.] _wordcopy_fwd_aligned
+ 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start
+ 0.00% poisson2d ld-2.17.so [.] _start
#
-# (Tip: Show user configuration overrides: perf config --user --list)
+# (Tip: Limit to show entries above 5% only: perf report --percent-limit 5)
#
</pre>
</div>
@@ -13403,17 +13472,19 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p><strong>TASK</strong>: Now change the optimization flag in the <a href="/edit/Task1/Makefile">Makefile</a> to <code>-Ofast</code> and repeat the steps in the following cell. In case you follow along non-interactive, call <code>make</code> and <code>make run</code> in your shell. (If you are in the Jupyter Notebook, you can actually click the link of the <a href="/edit/Task1/Makefile">Makefile</a>. In other cases, use <code>vim</code> which is installed on Ascent.)</p>
+<p><strong>TASK</strong>: Now add the optimization flag <code>Ofast</code> to the <code>CFLAGS</code> for target <code>poisson2d_Ofast</code>. Compile the program with the target <code>poisson2d_Ofast</code> and run and analyse it as before with <code>run</code>, <code>run_perf</code> and <code>run_perf_recrep</code>.</p>
+<p>What difference do you see?</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [5]:</div>
+<div class="prompt input_prompt">In [76]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_Ofast
+<span class="o">!</span>make run
</pre></div>
</div>
@@ -13430,8 +13501,25 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm
-/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm
+<pre>gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast poisson2d.c poisson2d_reference.o -o poisson2d -lm
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24901> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+2.41user 0.00system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k
+256inputs+0outputs (0major+480minor)pagefaults 0swaps
</pre>
</div>
</div>
@@ -13439,13 +13527,21 @@ Calculate current execution.
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Again, run a <code>perf</code>-instrumented version:</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [6]:</div>
+<div class="prompt input_prompt">In [77]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf
</pre></div>
</div>
@@ -13462,25 +13558,30 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
-Job <5034> is submitted to default queue <batch>.
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d
+Job <24902> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-0.51user 0.00system 0:00.52elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k
-256inputs+0outputs (0major+264minor)pagefaults 0swaps
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+
+ Performance counter stats for './poisson2d':
+
+ 8258991976 cycles:u
+ 12013091172 instructions:u # 1.45 insn per cycle
+
+ 2.408703909 seconds time elapsed
+
</pre>
</div>
</div>
@@ -13488,16 +13589,21 @@ Calculate current execution.
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Generate the list of top routines in terms of hotness:</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [7]:</div>
+<div class="prompt input_prompt">In [78]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="c1"># perf record creates a perf.data file </span>
-<span class="o">!</span>perf record -o perf.Ofast.data -e cycles ./poisson2d
-<span class="c1"># perf report opens the perf.data file </span>
-<span class="o">!</span>perf report -i perf.Ofast.data <span class="p">|</span> cat
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf_recrep
</pre></div>
</div>
@@ -13514,45 +13620,58 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d
+Job <24903> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-[ perf record: Woken up 1 times to write data ]
-[ perf record: Captured and wrote 0.086 MB perf.Ofast.data (1889 samples) ]
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+[ perf record: Woken up 2 times to write data ]
+[ perf record: Captured and wrote 0.382 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (9728 samples) ]
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio
+Job <24904> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
-# Samples: 1K of event 'cycles:u'
-# Event count (approx.): 1765737747
+# Samples: 9K of event 'cycles:u'
+# Event count (approx.): 8268811890
#
-# Overhead Command Shared Object Symbol
-# ........ ......... ............. .......................
+# Overhead Command Shared Object Symbol
+# ........ ......... ............. ........................................
#
- 44.65% poisson2d poisson2d [.] main
- 43.84% poisson2d poisson2d [.] poisson2d_reference
- 10.28% poisson2d libc-2.17.so [.] __memcpy_power7
- 1.12% poisson2d libm-2.17.so [.] __exp_finite
- 0.05% poisson2d poisson2d [.] check_results
- 0.03% poisson2d ld-2.17.so [.] _dl_relocate_object
- 0.02% poisson2d libc-2.17.so [.] __readdir64
- 0.01% poisson2d ld-2.17.so [.] _dl_new_object
+ 81.12% poisson2d poisson2d [.] main
+ 17.97% poisson2d libc-2.17.so [.] __memcpy_power7
+ 0.79% poisson2d libm-2.17.so [.] __exp_finite
+ 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17
+ 0.02% poisson2d ld-2.17.so [.] do_lookup_x
+ 0.01% poisson2d libc-2.17.so [.] vfprintf@@GLIBC_2.17
+ 0.01% poisson2d libc-2.17.so [.] _dl_addr
+ 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object
+ 0.01% poisson2d ld-2.17.so [.] check_match.10253
+ 0.01% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x
+ 0.01% poisson2d ld-2.17.so [.] strcmp
+ 0.00% poisson2d ld-2.17.so [.] open_path
+ 0.00% poisson2d ld-2.17.so [.] init_tls
+ 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start
0.00% poisson2d ld-2.17.so [.] _start
#
-# (Tip: System-wide collection from all CPUs: perf record -a)
+# (Tip: For tracepoint events, try: perf report -s trace_fields)
#
</pre>
</div>
@@ -13573,7 +13692,7 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h4 id="Interpretation">Interpretation<a class="anchor-link" href="#Interpretation">¶</a></h4><p>Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with <code>-Ofast</code> which enables <code>–ffast-math</code> option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the <code>-Ofast</code> binary natively implements the <code>fmax</code> function using instructions available in the hardware. The <code>-O3</code> binary makes a library call to compute <code>fmax</code> to follow a stricter <em>IEEE</em> requirement for accuracy.</p>
+<h4 id="Interpretation">Interpretation<a class="anchor-link" href="#Interpretation">¶</a></h4><p>Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with <code>-Ofast</code> which enables <code>–ffast-math</code> option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the <code>-Ofast</code> binary natively implements the <code>fmax</code> function using instructions available in the hardware. The <code>-O3</code> binary makes a library call to compute <code>fmax</code> to follow a stricter <em>IEEE</em> requirement for accuracy.</p>
</div>
</div>
@@ -13581,27 +13700,85 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-B:-Profile-directed-Feedback">Part B: Profile-directed Feedback<a class="anchor-link" href="#Part-B:-Profile-directed-Feedback">¶</a></h3><p>For the first level of optimization we saw <code>Ofast</code> cut the execution time of the <code>O3</code> binary by almost half.</p>
-<p>We can optimize the performance further by using profile directed feedback optimization.</p>
-<p>To compile using profile directed feedback with the GCC compiler we need to do the following steps</p>
+<h3 id="Part-B:-Profile-directed-Feedback">Part B: Profile-directed Feedback<a class="anchor-link" href="#Part-B:-Profile-directed-Feedback">¶</a></h3><p>For the first level of optimization we see that <code>Ofast</code> cut the execution time of the <code>O3</code> binary by almost half.</p>
+<p>We can optimize the performance further by using profile-directed feedback optimization.</p>
+<p>To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:</p>
<ol>
-<li>We need to first build a training binary using <code>-fprofile-generate</code>; this instructs the compiler to record hot path information </li>
-<li>Run the training binary with a smaller input size; you should see a <code>.gcda</code> file generated which stores hot path information for further optimization by the compiler </li>
-<li>build the final binary using <code>-fprofile-use</code> which uses the profile information in the <code>.gcda</code> file </li>
-<li>Compare the performance of the final binary with the original <code>Ofast</code> binary </li>
+<li>Instrument binary;</li>
+<li>Run binary with training, gather profile information;</li>
+<li>Use profile information to generate optimized binary.</li>
</ol>
-<p><strong>TASK</strong>: First, search for <code>TODO1</code> in the <a href="/edit/Task1/Makefile">Makefile</a>. It defines an additional compilation flag for <code>gcc</code>. Insert <code>-fprofile-generate=FOLDER</code> there with FOLDER pointing to <code>$$SC18_DIR_SCRATCH</code>, your personal write-directory (the double dollar signs are intentional as they are used to escape in the GNU Make syntax).</p>
-<p>After editing, run the following two cells to train your program.</p>
+<p>Step 1 is achieved by compiling the binary with the correct flag – <code>-fprofile-generate</code>. In our case, we need to specify an output location, which should be <code>$(SC19_DIR_SCRATCH)</code>.</p>
+<p>Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension <code>.gcda</code>) is written to the directory specified during compilation.</p>
+<p>For Step 3, the binary is once again compiled, but this time using the <code>gcda</code> profile just generated. The according flag is <code>-fprofile-use</code>, which we set to <code>$(SC19_DIR_SCRATCH)</code> as well.</p>
+<p>In our <code>Makefile</code> at hand, we prepared the steps already for you in the form of two targets.</p>
+<ul>
+<li><code>poisson2d_train</code>: Will compile the binary with profile-directed feedback</li>
+<li><code>poisson2d_ref</code>: Will take a generated profile and compile a new, optimized binary</li>
+</ul>
+<p>By using dependencies, between these two targets a profile run is launched.</p>
+<p><strong>TASK</strong>: Edit the <a href="`Makefile`">Makefile</a> and add the <code>-fprofile-*</code> flags to the <code>CFLAGS</code> of <code>poisson2d_train</code> and
+<code>poisson2d_ref</code> as outline in the file.</p>
+<p>After that, you may launch them with the following cells (<code>gen_profile</code> is a meta-target and uses <code>poisson2d_train</code> and <code>poisson2d_ref</code>). If you need to clean the generated profile, you may use <code>make clean_profile</code>.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [79]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make gen_profile
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100
+Job <24905> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh
+Calculate current execution.
+ 0, 0.249490
+echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated
+gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_ref -lm
+cp poisson2d_ref poisson2d
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [8]:</div>
+<div class="prompt input_prompt">In [80]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_train
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
</pre></div>
</div>
@@ -13618,8 +13795,24 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec "-fprofile-generate=$SC18_DIR_SCRATCH" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm
-/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec "-fprofile-generate=$SC18_DIR_SCRATCH" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_train -lm
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24906> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+2.28user 0.01system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k
+256inputs+0outputs (0major+479minor)pagefaults 0swaps
</pre>
</div>
</div>
@@ -13627,13 +13820,29 @@ Calculate current execution.
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Great! It is! In our tests, this shaved off another 5%.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Let's also measure instructions and cycles</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [9]:</div>
+<div class="prompt input_prompt">In [81]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_train
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_perf
</pre></div>
</div>
@@ -13650,20 +13859,30 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_train 200 64 64
-Job <5035> is submitted to default queue <batch>.
+<pre>bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d
+Job <24907> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 200 iterations on 64 x 64 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.248743
- 100, 0.124046
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.248743
- 100, 0.124046
-0.00user 0.00system 0:00.10elapsed 5%CPU (0avgtext+0avgdata 5248maxresident)k
-512inputs+0outputs (0major+115minor)pagefaults 0swaps
-mv $SC18_DIR_SCRATCH/*.gcda .
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+
+ Performance counter stats for './poisson2d':
+
+ 7925983538 cycles:u
+ 12253080719 instructions:u # 1.55 insn per cycle
+
+ 2.313471365 seconds time elapsed
+
</pre>
</div>
</div>
@@ -13675,19 +13894,27 @@ mv $SC18_DIR_SCRATCH/*.gcda .
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>Now, a <code>.gcda</code> file exists in the directory which can be used for an profile-accelerated subsequent run.</p>
-<p><strong>TASK</strong>: Edit the <a href="/edit/Task1/Makefile">Makefile</a> again, this time modifying <code>TODO2</code> to be equivalent to <code>-fprofile-use</code>. A directory is not needed as we copied the gcda file into the current directory.</p>
-<p>Run the following cells in order to build using the newly added flag and then run with the profile-accelerated version.</p>
+<p>What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-C:-Compiler-annotations/Remarks">Part C: Compiler annotations/Remarks<a class="anchor-link" href="#Part-C:-Compiler-annotations/Remarks">¶</a></h3><p>Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done.</p>
+<p>To generate compiler annotations using GCC, one uses <code>-fopt-info-all</code>. If you only want to see the missed options, use the option <code>-fopt-info-missed</code> instead of <code>-fopt-info-all</code>. See also the <a href="https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info">documentation of GCC regarding the flag</a>.</p>
+<p><strong>TASK</strong>: Have a looK at the <code>CFLAGS</code> of the <code>Makefile</code> target <code>poisson2d_Ofast_info</code>. Add the flag <code>-fopt-info-all</code> to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use – for example – <code>-fopt-info-all=(SC19_DIR_SCRATCH)/filename</code>.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [10]:</div>
+<div class="prompt input_prompt">In [82]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_profile
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_Ofast_info
</pre></div>
</div>
@@ -13704,8 +13931,74 @@ mv $SC18_DIR_SCRATCH/*.gcda .
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec "-fprofile-use" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm
-/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec "-fprofile-use" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_profile -lm
+<pre>gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all poisson2d.c poisson2d_reference.o -o poisson2d_Ofast_info -lm
+poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available
+poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available
+poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available
+poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available
+poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+Unit growth for small function inlining: 207->207 (0%)
+
+Inlined 4 calls, eliminated 0 functions
+
+consider run-time aliasing test between *_84 and *_87
+consider run-time aliasing test between *_92 and *_97
+consider run-time aliasing test between *_104 and *_107
+consider run-time aliasing test between *_111 and *_115
+poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.
+poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.
+poisson2d.c:108:25: missed: couldn't vectorize loop
+poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized
+poisson2d.c:136:9: missed: couldn't vectorize loop
+poisson2d.c:136:9: missed: Loop costings may not be worthwhile.
+poisson2d.c:131:9: missed: couldn't vectorize loop
+poisson2d.c:131:9: missed: Loop costings may not be worthwhile.
+poisson2d.c:122:9: missed: couldn't vectorize loop
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);
+poisson2d.c:112:9: missed: couldn't vectorize loop
+poisson2d.c:112:9: missed: not vectorized: control flow in loop.
+poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors
+poisson2d.c:88:5: missed: couldn't vectorize loop
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);
+poisson2d.c:72:5: missed: couldn't vectorize loop
+poisson2d.c:78:27: missed: not vectorized: complicated access pattern.
+poisson2d.c:74:9: missed: couldn't vectorize loop
+poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);
+poisson2d.c:43:5: note: vectorized 1 loops in function.
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);
+poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);
+poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);
+poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);
+poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);
+poisson2d.c:96:5: missed: statement clobbers memory: printf ("Jacobi relaxation calculation: max %d iterations on %d x %d mesh\n", iter_max_130, ny_139, nx_195);
+poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&"Calculate current execution."[0]);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);
+poisson2d.c:142:31: missed: statement clobbers memory: printf ("%5d, %0.6f\n", iter_237, error_219);
+poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);
+poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);
+poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);
+poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);
+poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);
+poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);
</pre>
</div>
</div>
@@ -13713,13 +14006,23 @@ mv $SC18_DIR_SCRATCH/*.gcda .
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.</p>
+<p><strong>TASK</strong>:
+Adapt the <code>CFLAGS</code> of <code>poisson2d_ref_info</code> to include <code>-fopt-info-all</code> <strong>and</strong> the profile input of <code>-fprofile-use=…</code> here. <em>(Be advised: Long output!)</em></p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [11]:</div>
+<div class="prompt input_prompt">In [83]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_profile
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_ref_info
</pre></div>
</div>
@@ -13736,25 +14039,220 @@ mv $SC18_DIR_SCRATCH/*.gcda .
<div class="output_subarea output_stream output_stdout output_text">
-<pre>bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_profile
-Job <5036> is submitted to default queue <batch>.
+<pre>gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm
+poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).
+Increasing alignment of decl: __gcov0.main
+poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_D_00100_1_main/48 -> __gcov_exit/55, function body not available
+poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_I_00100_0_main/47 -> __gcov_init/54, function body not available
+poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available
+poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available
+poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available
+poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available
+poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+Unit growth for small function inlining: 295->295 (0%)
+
+Inlined 4 calls, eliminated 0 functions
+
+consider run-time aliasing test between *_84 and *_87
+consider run-time aliasing test between *_92 and *_97
+consider run-time aliasing test between *_104 and *_107
+consider run-time aliasing test between *_111 and *_115
+poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.
+poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);
+poisson2d.c:108:25: missed: couldn't vectorize loop
+poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized
+poisson2d.c:136:9: missed: couldn't vectorize loop
+poisson2d.c:136:9: missed: Loop costings may not be worthwhile.
+poisson2d.c:131:9: missed: couldn't vectorize loop
+poisson2d.c:131:9: missed: Loop costings may not be worthwhile.
+poisson2d.c:122:9: missed: couldn't vectorize loop
+poisson2d.c:122:9: missed: not vectorized: control flow in loop.
+poisson2d.c:112:9: missed: couldn't vectorize loop
+poisson2d.c:112:9: missed: not vectorized: control flow in loop.
+poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors
+poisson2d.c:88:5: missed: couldn't vectorize loop
+poisson2d.c:88:5: missed: not vectorized: control flow in loop.
+poisson2d.c:72:5: missed: couldn't vectorize loop
+poisson2d.c:72:5: missed: not vectorized: control flow in loop.
+poisson2d.c:74:9: missed: couldn't vectorize loop
+poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);
+poisson2d.c:43:5: note: vectorized 1 loops in function.
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);
+poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);
+poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);
+poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);
+poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);
+poisson2d.c:96:5: missed: statement clobbers memory: printf ("Jacobi relaxation calculation: max %d iterations on %d x %d mesh\n", iter_max_337, ny_124, nx_286);
+poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&"Calculate current execution."[0]);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);
+poisson2d.c:142:31: missed: statement clobbers memory: printf ("%5d, %0.6f\n", iter_316, error_118);
+poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_127);
+poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_311);
+poisson2d.c:161:5: missed: statement clobbers memory: free (A_122);
+poisson2d.c:65:41: missed: statement clobbers memory: A_129 = malloc (8000000);
+poisson2d.c:67:41: missed: statement clobbers memory: Anew_132 = malloc (8000000);
+poisson2d.c:68:41: missed: statement clobbers memory: rhs_140 = malloc (8000000);
+poisson2d.c:136:9: note: considering unrolling loop 7 at BB 53
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)
+poisson2d.c:131:9: note: considering unrolling loop 6 at BB 50
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)
+poisson2d.c:122:9: note: considering unrolling loop 5 at BB 47
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:122:9: optimized: loop unrolled 3 times (header execution count 9800)
+poisson2d.c:118:25: note: considering unrolling loop 13 at BB 33
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)
+poisson2d.c:118:25: note: considering unrolling loop 9 at BB 30
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:112:9: note: considering unrolling loop 14 at BB 42
+poisson2d.c:43:5: note: considering unrolling loop 4 at BB 40
+poisson2d.c:108:25: note: considering unrolling loop 3 at BB 60
+poisson2d.c:88:5: note: considering unrolling loop 2 at BB 23
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:88:5: optimized: loop unrolled 3 times (header execution count 100)
+poisson2d.c:74:9: note: considering unrolling loop 11 at BB 12
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)
+poisson2d.c:72:5: note: considering unrolling loop 1 at BB 16
+poisson2d.c:164:1: missed: statement clobbers memory: __gcov_init (&*.LPBX0);
+poisson2d.c:164:1: missed: statement clobbers memory: __gcov_exit ();
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100
+Job <24908> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+libgcov profiling error:/gpfs/wolf/trn003/scratch/aherten//#autofs#nccsopen-svm1_home#aherten#SC19-Tutorial#3-Optimizing_POWER#Handson#Task1#poisson2d.gcda:overwriting an existing profile data with a different timestamp
+Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-0.47user 0.00system 0:00.48elapsed 98%CPU (0avgtext+0avgdata 10816maxresident)k
-256inputs+0outputs (0major+265minor)pagefaults 0swaps
+ 0, 0.249490
+echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated
+gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c poisson2d_reference.o -o poisson2d_ref_info -lm
+poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).
+poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available
+poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available
+poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available
+poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available
+poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available
+poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available
+Unit growth for small function inlining: 207->207 (0%)
+
+Inlined 4 calls, eliminated 0 functions
+
+consider run-time aliasing test between *_84 and *_87
+consider run-time aliasing test between *_92 and *_97
+consider run-time aliasing test between *_104 and *_107
+consider run-time aliasing test between *_111 and *_115
+poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.
+poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.
+poisson2d.c:108:25: missed: couldn't vectorize loop
+poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized
+poisson2d.c:136:9: missed: couldn't vectorize loop
+poisson2d.c:136:9: missed: Loop costings may not be worthwhile.
+poisson2d.c:131:9: missed: couldn't vectorize loop
+poisson2d.c:131:9: missed: Loop costings may not be worthwhile.
+poisson2d.c:122:9: missed: couldn't vectorize loop
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);
+poisson2d.c:112:9: missed: couldn't vectorize loop
+poisson2d.c:112:9: missed: not vectorized: control flow in loop.
+poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors
+poisson2d.c:88:5: missed: couldn't vectorize loop
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);
+poisson2d.c:72:5: missed: couldn't vectorize loop
+poisson2d.c:78:27: missed: not vectorized: complicated access pattern.
+poisson2d.c:74:9: missed: couldn't vectorize loop
+poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);
+poisson2d.c:43:5: note: vectorized 1 loops in function.
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);
+poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);
+/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);
+poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);
+poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);
+poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);
+poisson2d.c:96:5: missed: statement clobbers memory: printf ("Jacobi relaxation calculation: max %d iterations on %d x %d mesh\n", iter_max_130, ny_139, nx_195);
+poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&"Calculate current execution."[0]);
+poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);
+poisson2d.c:142:31: missed: statement clobbers memory: printf ("%5d, %0.6f\n", iter_237, error_219);
+poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);
+poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);
+poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);
+poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);
+poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);
+poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);
+poisson2d.c:136:9: note: considering unrolling loop 7 at BB 47
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)
+poisson2d.c:131:9: note: considering unrolling loop 6 at BB 44
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)
+poisson2d.c:122:9: note: considering unrolling loop 5 at BB 40
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:122:9: optimized: loop unrolled 7 times (header execution count 9701)
+poisson2d.c:118:25: note: considering unrolling loop 13 at BB 27
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)
+poisson2d.c:118:25: note: considering unrolling loop 9 at BB 24
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:112:9: note: considering unrolling loop 14 at BB 37
+poisson2d.c:43:5: note: considering unrolling loop 4 at BB 35
+poisson2d.c:108:25: note: considering unrolling loop 3 at BB 51
+poisson2d.c:88:5: note: considering unrolling loop 2 at BB 18
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:88:5: optimized: loop unrolled 7 times (header execution count 99)
+poisson2d.c:74:9: note: considering unrolling loop 11 at BB 9
+considering unrolling loop with constant number of iterations
+considering unrolling loop with runtime-computable number of iterations
+poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)
+poisson2d.c:72:5: note: considering unrolling loop 1 at BB 14
</pre>
</div>
</div>
@@ -13766,7 +14264,11 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)</p>
+<p>Comparing the annotations generated of a plain <code>-Ofast</code> optimization level and the one generated at <code>-Ofast</code> and profile directed feedback, we observe that many more optimizations are possible due to profile information.</p>
+<p>For instance you will see annotations such as</p>
+
+<pre><code>poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)</code></pre>
+<p>The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations.</p>
</div>
</div>
@@ -13794,15 +14296,19 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Task-2:-Software-Pretechting">Task 2:<a name="task2" /> Software Pretechting<a class="anchor-link" href="#Task-2:-Software-Pretechting">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>Study the difference of program execution time of different optimization levels with and without software prefetching.</p>
-<p>First, change directory to that of Task 2</p>
+<h2 id="Task-2:-Impact-of-Prefetching-on-Performance">Task 2:<a name="task2" /> Impact of Prefetching on Performance<a class="anchor-link" href="#Task-2:-Impact-of-Prefetching-on-Performance">¶</a></h2><h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><ul>
+<li>Study the difference of program execution time of different optimization levels with and without software prefetching.</li>
+<li>Verify the impact by measuring cache counters with and without prefetching.</li>
+<li>Learn how to modify contents of DSCR (<em>Data Stream Control Register</em>) using IBM XL compiler and study the impact with different values to DSCR. </li>
+</ul>
+<p>But first, lets change directory to that of Task 2</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [12]:</div>
+<div class="prompt input_prompt">In [85]:</div>
<div class="inner_cell">
<div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="o">%</span><span class="k">cd</span> ../Task2
@@ -13822,7 +14328,7 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task2
+<pre>/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task2
</pre>
</div>
</div>
@@ -13834,25 +14340,28 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-A:-Running">Part A: Running<a class="anchor-link" href="#Part-A:-Running">¶</a></h3>
+<h3 id="Part-A:-Software-Prefetching">Part A: Software Prefetching<a class="anchor-link" href="#Part-A:-Software-Prefetching">¶</a></h3>
</div>
</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>Look at the <a href="/edit/Task2/Makefile">Makefile</a> and work on the TODOs. Please implement compile flags as mentioned in the Makefile target name.</p>
-<p>Afterwards, compile each target with the following cells and submit them to the batch system. Follow along accordingly in the non-interactive version of this Notebook.</p>
+<p><strong>TASK</strong>: Look at the Makefile and work on the TODOs.</p>
+<ul>
+<li>First generate a <code>-Ofast</code>-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!</li>
+<li>Modify the <code>Makefile</code> to add the option for software prefetching (<code>-fprefetch-loop-arrays</code>). Compare performance of <code>-Ofast</code> with and without software prefetching</li>
+</ul>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [13]:</div>
+<div class="prompt input_prompt">In [97]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_o3_pref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make clean
</pre></div>
</div>
@@ -13869,27 +14378,7 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm
-/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_pref -lm
-bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_pref
-Job <5037> is submitted to default queue <batch>.
-<<Waiting for dispatch ...>>
-<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-1.12user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10880maxresident)k
-256inputs+0outputs (0major+265minor)pagefaults 0swaps
+<pre>rm -f poisson2d poisson2d*.o
</pre>
</div>
</div>
@@ -13900,10 +14389,12 @@ Calculate current execution.
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [14]:</div>
+<div class="prompt input_prompt">In [88]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_ofast_pref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d <span class="nv">CC</span><span class="o">=</span>gcc
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
@@ -13920,26 +14411,49 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_pref -lm
-bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_pref
-Job <5038> is submitted to default queue <batch>.
+<pre>make: `poisson2d' is up to date.
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24911> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+2.39user 0.01system 0:02.40elapsed 100%CPU (0avgtext+0avgdata 24256maxresident)k
+0inputs+0outputs (0major+480minor)pagefaults 0swaps
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d
+Job <24912> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-0.77user 0.00system 0:00.77elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k
-256inputs+0outputs (0major+264minor)pagefaults 0swaps
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+
+ Performance counter stats for './poisson2d':
+
+ 8271503902 cycles:u
+ 481152478 r168a4:u
+
+ 2.412224884 seconds time elapsed
+
</pre>
</div>
</div>
@@ -13950,10 +14464,12 @@ Calculate current execution.
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [15]:</div>
+<div class="prompt input_prompt">In [98]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_o3_nopref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_pref <span class="nv">CC</span><span class="o">=</span>gcc
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
@@ -13970,26 +14486,50 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_nopref -lm
-bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_nopref
-Job <5039> is submitted to default queue <batch>.
+<pre>gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm
+cp poisson2d_pref poisson2d
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24919> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+1.92user 0.00system 0:01.93elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k
+256inputs+0outputs (0major+480minor)pagefaults 0swaps
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d
+Job <24920> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-1.13user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10944maxresident)k
-256inputs+0outputs (0major+266minor)pagefaults 0swaps
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+
+ Performance counter stats for './poisson2d':
+
+ 6586609284 cycles:u
+ 459879452 r168a4:u
+
+ 1.925399505 seconds time elapsed
+
</pre>
</div>
</div>
@@ -13997,13 +14537,23 @@ Calculate current execution.
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p><strong>TASK</strong>: Repeat the experiment with the <code>-O3</code> flag. Have a look at the <code>Makefile</code> and the outlined TODO. There's a position to easily adapt <code>-Ofast</code>→<code>-O3</code>!</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [16]:</div>
+<div class="prompt input_prompt">In [100]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run_ofast_nopref
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d <span class="nv">CC</span><span class="o">=</span>gcc -B
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
@@ -14020,59 +14570,65 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_nopref -lm
-bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_nopref
-Job <5040> is submitted to default queue <batch>.
+<pre>gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec poisson2d.c -o poisson2d -lm
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24923> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-0.82user 0.00system 0:00.82elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k
-256inputs+0outputs (0major+265minor)pagefaults 0swaps
-</pre>
-</div>
-</div>
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k
+256inputs+0outputs (0major+479minor)pagefaults 0swaps
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d
+Job <24924> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
-</div>
-</div>
+ Performance counter stats for './poisson2d':
-</div>
-<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
-</div><div class="inner_cell">
-<div class="text_cell_render border-box-sizing rendered_html">
-<p>Do you notice the impact difference with optimization levels? It's always important to carefully study the interplay of flags.</p>
+ 16445764669 cycles:u
+ 645094089 r168a4:u
+ 4.792567763 seconds time elapsed
+
+</pre>
</div>
</div>
-</div>
-<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
-</div><div class="inner_cell">
-<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-B:-Analysis-of-Instructions">Part B: Analysis of Instructions<a class="anchor-link" href="#Part-B:-Analysis-of-Instructions">¶</a></h3><p>Compilation with the software prefetching flag causes the compiler to generate the <code>__dcbt</code> and <code>__dcbtst</code> instructions that prefetch memory values to L3.</p>
-<p>Verify it using <code>objdump -lSd</code> on each file (<code>poisson2d_o3_pref</code>, <code>poisson2d_ofast_pref</code>, <code>poisson2d_o3_nopref</code>, <code>poisson2d_ofast_nopref</code>). You might want to grep for <code>dcb</code>.</p>
</div>
</div>
+
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [19]:</div>
+<div class="prompt input_prompt">In [101]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"poisson2d_o3_pref"</span><span class="p">,</span> <span class="s2">"poisson2d_ofast_pref"</span><span class="p">,</span> <span class="s2">"poisson2d_o3_nopref"</span><span class="p">,</span> <span class="s2">"poisson2d_ofast_nopref"</span><span class="p">]:</span>
- <span class="nb">print</span><span class="p">(</span><span class="s2">"</span><span class="si">{}</span><span class="s2">:"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">f</span><span class="p">))</span>
- <span class="n">objdump</span> <span class="o">-</span><span class="n">lSd</span> <span class="err">$</span><span class="n">f</span> <span class="o">|</span> <span class="n">grep</span> <span class="n">dcb</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_pref <span class="nv">CC</span><span class="o">=</span>gcc -B
+<span class="o">!</span>make run
+<span class="o">!</span>make l3missstats
</pre></div>
</div>
@@ -14089,27 +14645,809 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>poisson2d_o3_pref:
-poisson2d_ofast_pref:
- 10000da0: ec f1 00 7c dcbtst 0,r30
- 10000da4: 2c fa 00 7c dcbt 0,r31
- 10000da8: 2c 62 00 7c dcbt 0,r12
- 10000dac: 2c b2 00 7c dcbt 0,r22
- 10000dcc: 2c e2 00 7c dcbt 0,r28
- 10000dd0: 2c ea 00 7c dcbt 0,r29
- 100010b4: 2c 62 00 7c dcbt 0,r12
- 100010b8: 2c 5a 00 7c dcbt 0,r11
- 100010c4: ec 19 00 7c dcbtst 0,r3
- 100010cc: 2c 22 00 7c dcbt 0,r4
- 100010d0: 2c ea 00 7c dcbt 0,r29
- 100010d4: 2c f2 00 7c dcbt 0,r30
- 100010dc: 2c fa 00 7c dcbt 0,r31
-poisson2d_o3_nopref:
-poisson2d_ofast_nopref:
-</pre>
-</div>
-</div>
-
+<pre>gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24925> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+4.74user 0.00system 0:04.74elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k
+0inputs+0outputs (0major+480minor)pagefaults 0swaps
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d
+Job <24926> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+
+ Performance counter stats for './poisson2d':
+
+ 16239159454 cycles:u
+ 631061431 r168a4:u
+
+ 4.730144897 seconds time elapsed
+
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Observing the results, we see that SW Prefetching seems to help at <code>-Ofast</code> but not at <code>-O3</code>. We can use the steps described in the the next section to verify that the compiler has not inserted any SW prefetch operations at<code>-O3</code> at all. That is because in the <code>-O3</code> binary the time is dominated by <code>__fmax</code> call which causes the compiler to come to the conclusion that whatever benefit we obtain by adding SW prefetch will be overshadowed by the penalty of <code>fmax()</code>
+GCC may add further loop optimizations such as unrolling upon invocation of <code>–fprefetch-loop-arrays</code>.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-B:-Analysis-of-Instructions">Part B: Analysis of Instructions<a class="anchor-link" href="#Part-B:-Analysis-of-Instructions">¶</a></h3><p>Compilation of the <code>-Ofast</code> binary with the software prefetching flag causes the compiler to generate the <code>dcb*</code> instructions that prefetch memory values to L3.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p><strong>TASK</strong>:
+Run <code>$(SC19_SUBMIT_CMD) objdump -lSd</code> on each binary file (<code>-O3</code>, <code>-Ofast</code> with prefetch/no prefetch).
+Look for instructions beginning with <code>dcb</code>
+At what optimization levels does the compiler generate software prefetching instructions?</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [114]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make <span class="nv">CC</span><span class="o">=</span>gcc -B poisson2d_pref
+<span class="o">!</span>objdump -lSd ./poisson2d_pref > poisson2d.dis
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [116]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>grep dcb poisson2d.dis
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre> 10000b28: 2c d2 00 7c dcbt 0,r26
+ 10000b30: 2c ba 00 7c dcbt 0,r23
+ 10000b38: 2c b2 00 7c dcbt 0,r22
+ 10000b50: 2c d2 00 7c dcbt 0,r26
+ 10000b58: ec b9 00 7c dcbtst 0,r23
+ 10000b80: 2c d2 00 7c dcbt 0,r26
+ 10000e64: 2c 92 00 7c dcbt 0,r18
+ 10000e68: 2c 9a 00 7c dcbt 0,r19
+ 10000e6c: 2c a2 00 7c dcbt 0,r20
+ 10000e70: 2c aa 00 7c dcbt 0,r21
+ 10000e7c: 2c b2 00 7c dcbt 0,r22
+ 10000e80: 2c d2 00 7c dcbt 0,r26
+ 10000e94: ec b9 00 7c dcbtst 0,r23
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-C:-Changing-Values-of-DSCR-via-compiler-flags">Part C: Changing Values of DSCR via compiler flags<a class="anchor-link" href="#Part-C:-Changing-Values-of-DSCR-via-compiler-flags">¶</a></h3><p>This task requires using the IBM XL compiler. It should be already in your environment.</p>
+<p>We saw the impact of software prefetching in the previous subsection.
+In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance.
+In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching.</p>
+<p>IBM XL compiler has an option <code>-qprefetch=dscr=<val></code> that can be used for this purpose.
+Compiling with <code>-qprefetch=dscr=1</code> turns off the prefetcher. One can give various values such as <code>-qprefetch=dscr=4</code>, <code>-qprefetch=dscr=7</code> etc. to control aggressiveness of prefetching.</p>
+<p>For this exercise we use <code>make CC=xlc_r</code> to illustrate the performance impact.</p>
+<p><strong>Task</strong> Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the <code>Makefile</code>: Add <code>qprefetch=dscr=1</code> to the <code>CFLAGS</code> and rebuild the application and note the performance. Which one is faster?</p>
+<p>In general, applications benefit with the default settings of hardware DSCR register (<code>-qprefetch=dscr=0</code>). However, certain applications also benefit with prefetching turned off.</p>
+<p>It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Measure performance of the application compiled with XL at default DSCR value</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [117]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make <span class="nv">CC</span><span class="o">=</span>xlc_r -B poisson2d
+<span class="o">!</span>make run
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS poisson2d.c -o poisson2d -lm
+ 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24927> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 50.149062
+ 200, 99.849327
+ 300, 149.352369
+ 400, 198.659746
+ 500, 247.773000
+ 600, 296.693652
+ 700, 345.423208
+ 800, 393.963155
+ 900, 442.314962
+2.26user 0.00system 0:02.27elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k
+256inputs+0outputs (0major+477minor)pagefaults 0swaps
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Measure performance of the application compiled with XL with DSCR value turned off</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [9]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d_dscr <span class="nv">CC</span><span class="o">=</span>xlc_r -B
+<span class="o">!</span>make run
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS -qprefetch=dscr=1 poisson2d.c -o poisson2d_dscr -lm
+ 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24929> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+4.58user 0.00system 0:04.59elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k
+0inputs+0outputs (0major+476minor)pagefaults 0swaps
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Does Hardware prefetcher help this application? How much impact do you see when you turn off the hardware prefetcher?</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>The DSCR register controls the operation of the HW Prefetcher on POWER9. It can be modified in the command line by <code>ppc64_cpu --dscr=<value></code>. However this needs admin privileges. IBM XL offers a compiler flag to set the value through the compiler. <code>-qprefetch=dscr=1</code> turns off the prefetcher. Observing the results we see that the performance without the HW prefetcher is twice as bad as that with default prefetching. So we can conclude that Prefetching helps the Jacobi application.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h4 id="References">References<a class="anchor-link" href="#References">¶</a></h4><ol>
+<li><a href="https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html">https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html</a></li>
+<li><a href="https://www.gnu.org/software/gcc/projects/prefetch.html">https://www.gnu.org/software/gcc/projects/prefetch.html</a></li>
+<li><a href="https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0">https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0</a></li>
+</ol>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p><a href="#top">Back to Top</a></p>
+<hr>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h2 id="Task-3:-OpenMP">Task 3: OpenMP<a class="anchor-link" href="#Task-3:-OpenMP">¶</a></h2><p><a name="task3"></a></p>
+<h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores on the resulting application performance. We do this study for both GCC and XL compilers inorder to learn about the appropriate options that need to be used.
+First, we need to change directory to that of Task3. For Task 3 we modify poisson2d.c to invoke an exact copy of the main jacobi loop which is <code>poisson2d_reference</code>. We parallelize only the main loop but not <code>poisson2d_reference</code>. The speedup is the performance gain seen in the main loop as compared to the reference loop.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [10]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">%</span><span class="k">cd</span> ../Task3
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task3
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-A:-Implement-OpenMP-Pragmas;-Compilation">Part A: Implement OpenMP Pragmas; Compilation<a class="anchor-link" href="#Part-A:-Implement-OpenMP-Pragmas;-Compilation">¶</a></h3><p><strong>Task</strong>: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.</p>
+<ul>
+<li><strong>Directives</strong>: Look at the TODOs in <a href="poisson2d.c"><code>poisson2d.c</code></a> to add OpenMP parallelism. The pragmas in question are <code>#pragma omp parallel for</code> (and once it's <code>#pragma omp parallel for reduction(max:error)</code> – can you guess where?)</li>
+<li><strong>Compilation</strong>: Please add compilation flags enabling OpenMP in GCC and XL to the <code>Makefile</code>. For GCC, we need to add <code>-fopenmp</code> and the application needs to be linked with <code>-lgomp</code>. For XL, we need to add <code>-qsmp=omp</code> to the list of compilation flags. </li>
+</ul>
+<p>Afterwards, compile and run the application with the following commands.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [39]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d <span class="nv">CC</span><span class="o">=</span>gcc
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>gcc -c -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d_reference.c -o poisson2d_reference.o -lm
+gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d.c poisson2d_reference.o -o poisson2d -lm
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>The command to submit a job to the batch system is prepared in an environment variable <code>$SC19_SUBMIT_CMD</code>; use it together with <code>eval</code>. In the following cell, it is shown how to invoke the application using the batch system.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [40]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC19_SUBMIT_CMD</span> ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span>
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>Job <24951> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
+Calculate reference solution and time with serial CPU execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+Calculate current execution.
+ 0, 0.249995
+ 100, 0.248997
+ 200, 0.248007
+ 300, 0.247025
+ 400, 0.246050
+ 500, 0.245084
+ 600, 0.244124
+ 700, 0.243173
+ 800, 0.242228
+ 900, 0.241291
+1000x1000: Ref: 4.7430 s, This: 3.9363 s, speedup: 1.20
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Inorder to run the parallel application, we need to set the number of threads using <code>OMP_NUM_THREADS</code>
+What is the best performance you can reach by setting the number of threads via <code>OMP_NUM_THREADS=N</code> with <code>N</code> being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example.<br>
+We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler of Ascent, from overlaying binding options. Also, we use <code>-c ALL_CPUS</code> to make all CPUs on the compute nodes available to you.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [41]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">1</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 4.7288 s, This: 4.9791 s, speedup: 0.95
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [42]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">2</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 4.7125 s, This: 2.4914 s, speedup: 1.89
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [35]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.1065 s, This: 1.3836 s, speedup: 1.52
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [21]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">8</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3868 s, This: 0.5272 s, speedup: 4.53
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [22]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">10</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3912 s, This: 0.4612 s, speedup: 5.18
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [23]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">20</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3864 s, This: 0.4037 s, speedup: 5.91
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [24]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">40</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3773 s, This: 0.3045 s, speedup: 7.81
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [25]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">80</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3819 s, This: 0.3081 s, speedup: 7.73
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<h3 id="Part-B:-Bindings">Part B: Bindings<a class="anchor-link" href="#Part-B:-Bindings">¶</a></h3><p>Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!</p>
+<p>There are applications which can be used to determine the configuration of the processor. Among those are:</p>
+<ul>
+<li><code>lscpu</code>: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.</li>
+<li><code>ppc64_cpu --smt</code>: Specifically for POWER, this tool can give information about the number of simulations threads running per core (<em>SMT</em>, Simulataion Multi-Threading).</li>
+</ul>
+<p>Run <code>ppc64_cpu --smt</code> to find out about the threading configuration of Ascent!</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [55]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC19_SUBMIT_CMD</span> ppc64_cpu --smt
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre>Job <24465> is submitted to default queue <batch>.
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+SMT=4
+</pre>
+</div>
+</div>
+
</div>
</div>
@@ -14117,18 +15455,22 @@ poisson2d_ofast_nopref:
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>If you feel up to the task, you can study the number of L3 cache misses using the corresponding performance counter, <code>PM_L3_MISS</code>. Either use your knowledge from Hands-On 1, or use the following call to <code>perf</code>, in which we already converted the named counter to a raw counter address.</p>
+<p>There are more sources information available</p>
+<ul>
+<li><code>/proc/cpuinfo</code>: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with <code>cat</code></li>
+<li><code>/sys/devices/system/cpu/cpu0/topology/thread_siblings_list</code>: Holds information about thread siblings for given CPU core (<code>cpu0</code> in this case). Use it to find out which thread is mapped to which core.</li>
+</ul>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [35]:</div>
+<div class="prompt input_prompt">In [36]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"poisson2d_ofast_nopref"</span><span class="p">,</span> <span class="s2">"poisson2d_ofast_pref"</span><span class="p">]:</span>
- <span class="o">!</span><span class="nb">eval</span> <span class="nv">$$</span>SC18_SUBMIT_CMD perf stat -e cycles,r168a4 ./<span class="nv">$f</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">$$</span>SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list
+<span class="o">!</span><span class="nv">$$</span>SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list
</pre></div>
</div>
@@ -14145,54 +15487,14 @@ poisson2d_ofast_nopref:
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Job <5048> is submitted to default queue <batch>.
+<pre>Job <24949> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-
- Performance counter stats for './poisson2d_ofast_nopref':
-
- 2829292169 cycles:u
- 136018637 r168a4:u
-
- 0.826136863 seconds time elapsed
-
-Job <5049> is submitted to default queue <batch>.
+0-3
+Job <24950> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-
- Performance counter stats for './poisson2d_ofast_pref':
-
- 2654990243 cycles:u
- 128824827 r168a4:u
-
- 0.775593651 seconds time elapsed
-
+4-7
</pre>
</div>
</div>
@@ -14204,10 +15506,10 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h4 id="References">References<a class="anchor-link" href="#References">¶</a></h4><ol>
-<li><a href="https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html">https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html</a></li>
-<li><a href="https://www.gnu.org/software/gcc/projects/prefetch.html">https://www.gnu.org/software/gcc/projects/prefetch.html</a></li>
-</ol>
+<p>There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the <a href="https://www.openmp.org/spec-html/5.0/openmpse53.html">OMP_PLACES environment Variable</a>. We also have a GNU specific variable which can also be used to control affinity - <code>GOMP_CPU_AFFINITY</code>. Setting <code>GOMP_CPU_AFFINITY</code> is specific to GCC binaries but it internally serves the same function as setting <code>OMP_PLACES</code>.</p>
+<p><strong>Task</strong>: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.</p>
+<p>Adapt the following command with your configuration – or follow along accordingly in the non-interactive version of the Notebook.</p>
+<p>What's your maximum speedup?</p>
</div>
</div>
@@ -14215,8 +15517,7 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p><a href="#top">Back to Top</a></p>
-<hr>
+<p>Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores.</p>
</div>
</div>
@@ -14224,19 +15525,19 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h2 id="Task-3:-OpenMP">Task 3: OpenMP<a class="anchor-link" href="#Task-3:-OpenMP">¶</a></h2><p><a name="task3"></a></p>
-<h3 id="Overview">Overview<a class="anchor-link" href="#Overview">¶</a></h3><p>We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores.</p>
-<p>First, we need to change directory to that of Task3.</p>
+<p>Using <code>OMP_PLACES</code> for binding, and using some magical Python-Bash interplay:</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [1]:</div>
+<div class="prompt input_prompt">In [43]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">%</span><span class="k">cd</span> ../Task3
+<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">affinity</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"</span><span class="si">{0}</span><span class="s2">,</span><span class="si">{1}</span><span class="s2">,</span><span class="si">{2}</span><span class="s2">,</span><span class="si">{3}</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"</span><span class="si">{0}</span><span class="s2">,</span><span class="si">{5}</span><span class="s2">,</span><span class="si">{9}</span><span class="s2">,</span><span class="si">{13}</span><span class="s2">"</span><span class="p">]:</span>
+ <span class="nb">print</span><span class="p">(</span><span class="s2">"Affinity: </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">affinity</span><span class="p">))</span>
+ <span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">OMP_PLACES</span><span class="o">=</span><span class="nv">$affinity</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
</pre></div>
</div>
@@ -14253,7 +15554,16 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task3
+<pre>Affinity: {0},{1},{2},{3}
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+ OMP_PLACES = '{0},{1},{2},{3}'
+1000x1000: Ref: 4.7315 s, This: 3.9090 s, speedup: 1.21
+Affinity: {0},{5},{9},{13}
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+ OMP_PLACES = '{0},{5},{9},{13}'
+1000x1000: Ref: 4.6485 s, This: 1.2829 s, speedup: 3.62
</pre>
</div>
</div>
@@ -14265,23 +15575,19 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-A:-Implement-OpenMP-Pragmas;-Compilation">Part A: Implement OpenMP Pragmas; Compilation<a class="anchor-link" href="#Part-A:-Implement-OpenMP-Pragmas;-Compilation">¶</a></h3><p><strong>Task</strong>: Please add the correct OpenMP pragmas to the source code and compilations flags to enable OpenMP.</p>
-<ul>
-<li><strong>pragmas</strong>: Look at the TODOs in <a href="/edit/Task3/poisson2d.c"><code>poisson2d.c</code></a> to add OpenMP parallelism. The pragmas in question are <code>#pragma omp parallel for</code></li>
-<li><strong>Compilation</strong>: Please add compilation flags enabling OpenMP in GCC to the <a href="/edit/Task3/Makefile">Makefile</a>. The flag in question is <code>-fopenmp</code>.</li>
-</ul>
-<p>Edit the files with the links above if you are running the interactive version of the Notebook or navigate to <code>poisson2d.c</code> and <code>Makefile</code> yourself in case you run the non-interactive version.</p>
-<p>Afterwards, compile and run the application with the following cells. Non-interactive: Follow along accordingly in the shell.</p>
+<p>In this case, we carry out the same experiment using <code>GOMP_CPU_AFFINITY</code> which essentially sets the same environment variable <code>OMP_PLACES</code>. Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [37]:</div>
+<div class="prompt input_prompt">In [44]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make poisson2d
+<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">affinity</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"0,1,2,3"</span><span class="p">,</span> <span class="s2">"0,5,9,13"</span><span class="p">]:</span>
+ <span class="nb">print</span><span class="p">(</span><span class="s2">"Affinity: </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">affinity</span><span class="p">))</span>
+ <span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">GOMP_CPU_AFFINITY</span><span class="o">=</span><span class="nv">$affinity</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
</pre></div>
</div>
@@ -14298,8 +15604,16 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec poisson2d_reference.c -o poisson2d_reference.o -lm
-/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm
+<pre>Affinity: 0,1,2,3
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+ OMP_PLACES = '{0},{1},{2},{3}'
+1000x1000: Ref: 2.3964 s, This: 2.1361 s, speedup: 1.12
+Affinity: 0,5,9,13
+<<Waiting for dispatch ...>>
+<<Starting on login1>>
+ OMP_PLACES = '{0},{5},{9},{13}'
+1000x1000: Ref: 2.3925 s, This: 0.7030 s, speedup: 3.40
</pre>
</div>
</div>
@@ -14307,13 +15621,24 @@ Calculate current execution.
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Great!</p>
+<p>If you still have time: The same experiments can be repeated with the IBM XL compiler.
+The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is <code>-qsmp=omp</code></p>
+<p><strong>Task</strong>: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup.</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [40]:</div>
+<div class="prompt input_prompt">In [44]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make run
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>make <span class="nv">CC</span><span class="o">=</span>xlc_r -B run
</pre></div>
</div>
@@ -14330,26 +15655,40 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
-Job <5052> is submitted to default queue <batch>.
+<pre>xlc_r -c -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d_reference.c -o poisson2d_reference.o -lm
+ 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.
+xlc_r -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d.c poisson2d_reference.o -o poisson2d -lm
+ 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.
+bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d
+Job <24956> is submitted to default queue <batch>.
<<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh
+Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh
Calculate reference solution and time with serial CPU execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
+ 0, 0.249995
+ 100, 50.149062
+ 200, 99.849327
+ 300, 149.352369
+ 400, 198.659746
+ 500, 247.773000
+ 600, 296.693652
+ 700, 345.423208
+ 800, 393.963155
+ 900, 442.314962
Calculate current execution.
- 0, 0.249980
- 100, 0.246028
- 200, 0.242198
- 300, 0.238487
- 400, 0.234887
-500x500: Ref: 0.2571 s, This: 0.2946 s, speedup: 0.87
-1.48user 0.00system 0:00.56elapsed 263%CPU (0avgtext+0avgdata 9664maxresident)k
-0inputs+0outputs (0major+273minor)pagefaults 0swaps
+ 0, 0.249995
+ 100, 50.149062
+ 200, 99.849327
+ 300, 149.352369
+ 400, 198.659746
+ 500, 247.773000
+ 600, 296.693652
+ 700, 345.423208
+ 800, 393.963155
+ 900, 442.314962
+1000x1000: Ref: 5.6783 s, This: 2.6528 s, speedup: 2.14
+21.56user 6.18system 0:08.37elapsed 331%CPU (0avgtext+0avgdata 23040maxresident)k
+3200inputs+0outputs (2major+1098minor)pagefaults 0swaps
</pre>
</div>
</div>
@@ -14361,17 +15700,25 @@ Calculate current execution.
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>The command to submit a job to the batch system is prepared in an environment variable <code>$SC18_SUBMIT_CMD</code>; use it together with <code>eval</code>. In the following cell, it is shown how to increase the work of the application.</p>
+<p>Run the parallel application with varying numbre of threads (<code>OMP_NUM_THREADS</code>) and note the performance improvement.</p>
+
+</div>
+</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Just as in the GCC binary we see a similar speedup with higher number of threads until a certain point beyond which the benefit tapers off.</p>
</div>
</div>
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [3]:</div>
+<div class="prompt input_prompt">In [28]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC18_SUBMIT_CMD</span> ./poisson2d <span class="m">1000</span> <span class="m">1000</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">1</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
</pre></div>
</div>
@@ -14388,33 +15735,9 @@ Calculate current execution.
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Job <5344> is submitted to default queue <batch>.
-<<Waiting for dispatch ...>>
+<pre><<Waiting for dispatch ...>>
<<Starting on login1>>
-Jacobi relaxation calculation: max 1000 iterations on 1000 x 100 mesh
-Calculate reference solution and time with serial CPU execution.
- 0, 0.249743
- 100, 0.210080
- 200, 0.184635
- 300, 0.166526
- 400, 0.152783
- 500, 0.141890
- 600, 0.132978
- 700, 0.125511
- 800, 0.119142
- 900, 0.113632
-Calculate current execution.
- 0, 0.249743
- 100, 0.210080
- 200, 0.184635
- 300, 0.166526
- 400, 0.152783
- 500, 0.141890
- 600, 0.132978
- 700, 0.125511
- 800, 0.119142
- 900, 0.113632
-1000x100: Ref: 1.9872 s, This: 0.2385 s, speedup: 8.33
+1000x1000: Ref: 2.2561 s, This: 2.6432 s, speedup: 0.85
</pre>
</div>
</div>
@@ -14423,23 +15746,45 @@ Calculate current execution.
</div>
</div>
-<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
-</div><div class="inner_cell">
-<div class="text_cell_render border-box-sizing rendered_html">
-<p>What is the best performance you can reach by setting the number of threads via <code>OMP_NUM_THREADS=N</code> with <code>N</code> being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example.<br>
-We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler of Ascent, from overlaying binding options. Also, we use <code>-c ALL_CPUS</code> to make all CPUs on the compute nodes available to you.</p>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [29]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">2</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3071 s, This: 1.5343 s, speedup: 1.50
+</pre>
+</div>
+</div>
</div>
</div>
+
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [23]:</div>
+<div class="prompt input_prompt">In [30]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">omp_num</span> <span class="ow">in</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">4</span><span class="p">,</span> <span class="mi">8</span><span class="p">,</span> <span class="mi">10</span><span class="p">,</span> <span class="mi">20</span><span class="p">,</span> <span class="mi">40</span><span class="p">]:</span>
- <span class="nb">print</span><span class="p">(</span><span class="s2">"Threads: </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">omp_num</span><span class="p">))</span>
- <span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="nv">$omp_num</span> <span class="nv">$$</span>SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
</pre></div>
</div>
@@ -14456,34 +15801,42 @@ We added <code>--bind none</code> to prevent <code>jsrun</code>, the scheduler o
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Threads: 1
-<<Waiting for dispatch ...>>
-<<Starting on login1>>
-1000x1000: Ref: 2.3037 s, This: 2.8420 s, speedup: 0.81
-Threads: 2
-<<Waiting for dispatch ...>>
-<<Starting on login1>>
-1000x1000: Ref: 2.2998 s, This: 1.4320 s, speedup: 1.61
-Threads: 4
-<<Waiting for dispatch ...>>
-<<Starting on login1>>
-1000x1000: Ref: 2.3135 s, This: 0.7168 s, speedup: 3.23
-Threads: 8
-<<Waiting for dispatch ...>>
-<<Starting on login1>>
-1000x1000: Ref: 2.3145 s, This: 0.5278 s, speedup: 4.39
-Threads: 10
-<<Waiting for dispatch ...>>
-<<Starting on login1>>
-1000x1000: Ref: 2.3153 s, This: 0.4848 s, speedup: 4.78
-Threads: 20
-<<Waiting for dispatch ...>>
+<pre><<Waiting for dispatch ...>>
<<Starting on login1>>
-1000x1000: Ref: 2.3190 s, This: 0.2016 s, speedup: 11.50
-Threads: 40
-<<Waiting for dispatch ...>>
+1000x1000: Ref: 2.2617 s, This: 0.6936 s, speedup: 3.26
+</pre>
+</div>
+</div>
+
+</div>
+</div>
+
+</div>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [31]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">8</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
<<Starting on login1>>
-1000x1000: Ref: 2.3243 s, This: 0.3057 s, speedup: 7.60
+1000x1000: Ref: 2.2728 s, This: 0.3402 s, speedup: 6.68
</pre>
</div>
</div>
@@ -14492,26 +15845,45 @@ Threads: 40
</div>
</div>
-<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
-</div><div class="inner_cell">
-<div class="text_cell_render border-box-sizing rendered_html">
-<h3 id="Part-B:-Bindings">Part B: Bindings<a class="anchor-link" href="#Part-B:-Bindings">¶</a></h3><p>Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!</p>
-<p>There are applications which can be used to determine the configuration of the processor. Among those are:</p>
-<ul>
-<li><code>lscpu</code>: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.</li>
-<li><code>ppc64_cpu --smt</code>: Specifically for POWER, this tool can give information about the number of simulations threads running per core (<em>SMT</em>, Simulataion Multi-Threading).</li>
-</ul>
-<p>Run <code>ppc64_cpu --smt</code> to find out about the threading configuration of Ascent!</p>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [45]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">10</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.1678 s, This: 0.2869 s, speedup: 7.56
+</pre>
+</div>
+</div>
</div>
</div>
+
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [48]:</div>
+<div class="prompt input_prompt">In [33]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nb">eval</span> <span class="nv">$SC18_SUBMIT_CMD</span> ppc64_cpu --smt
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">20</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
</pre></div>
</div>
@@ -14528,10 +15900,9 @@ Threads: 40
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Job <5076> is submitted to default queue <batch>.
-<<Waiting for dispatch ...>>
+<pre><<Waiting for dispatch ...>>
<<Starting on login1>>
-SMT=4
+1000x1000: Ref: 2.2813 s, This: 0.1452 s, speedup: 15.71
</pre>
</div>
</div>
@@ -14540,25 +15911,45 @@ SMT=4
</div>
</div>
-<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
-</div><div class="inner_cell">
-<div class="text_cell_render border-box-sizing rendered_html">
-<p>There are more sources information available</p>
-<ul>
-<li><code>/proc/cpuinfo</code>: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with <code>cat</code></li>
-<li><code>/sys/devices/system/cpu/cpu0/topology/thread_siblings_list</code>: Holds information about thread siblings for given CPU core (<code>cpu0</code> in this case). Use it to find out which thread is mapped to which core.</li>
-</ul>
+<div class="cell border-box-sizing code_cell rendered">
+<div class="input">
+<div class="prompt input_prompt">In [34]:</div>
+<div class="inner_cell">
+ <div class="input_area">
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">40</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
+</pre></div>
+
+ </div>
+</div>
+</div>
+<div class="output_wrapper">
+<div class="output">
+
+
+<div class="output_area">
+
+ <div class="prompt"></div>
+
+
+<div class="output_subarea output_stream output_stdout output_text">
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.3284 s, This: 0.0981 s, speedup: 23.75
+</pre>
</div>
</div>
+
+</div>
+</div>
+
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [49]:</div>
+<div class="prompt input_prompt">In [35]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span>cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list
-<span class="o">!</span>cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list
+<div class=" highlight hl-ipython3"><pre><span></span><span class="o">!</span><span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">80</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep speedup
</pre></div>
</div>
@@ -14575,8 +15966,9 @@ SMT=4
<div class="output_subarea output_stream output_stdout output_text">
-<pre>0-3
-4-7
+<pre><<Waiting for dispatch ...>>
+<<Starting on login1>>
+1000x1000: Ref: 2.2918 s, This: 0.1439 s, speedup: 15.92
</pre>
</div>
</div>
@@ -14588,9 +15980,10 @@ SMT=4
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<p>There are various environment variables available within OpenMP (and GCC) to specify binding of threads to cores. See, for instance, the <a href="https://gcc.gnu.org/onlinedocs/libgomp/Environment-Variables.html">online documentation of GCC libgomp</a>. Examples are <code>OMP_PLACES</code> or <code>GOMP_CPU_AFFINITY</code>.</p>
+<p>Now we repeat the exercise of using the right binding of threads for the XL binary. <code>OMP_PLACES</code> pertains to the XL binary as well as it is an OpenMP variable. <code>GOMP_CPU_AFFINITY</code> is specific to GCC binary so that cannot be used to set the binding.</p>
<p><strong>Task</strong>: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.</p>
<p>Adapt the following command with your configuration – or follow along accordingly in the non-interactive version of the Notebook.</p>
+<p>We are mixing Python with Bash (<code>!</code>) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two <code>$$</code>)</p>
<p>What's your maximum speedup?</p>
</div>
@@ -14598,12 +15991,12 @@ SMT=4
</div>
<div class="cell border-box-sizing code_cell rendered">
<div class="input">
-<div class="prompt input_prompt">In [24]:</div>
+<div class="prompt input_prompt">In [36]:</div>
<div class="inner_cell">
<div class="input_area">
-<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">affinity</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"0,1,2,3"</span><span class="p">,</span> <span class="s2">"0,5,9,13"</span><span class="p">]:</span>
+<div class=" highlight hl-ipython3"><pre><span></span><span class="k">for</span> <span class="n">affinity</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"</span><span class="si">{0}</span><span class="s2">,</span><span class="si">{1}</span><span class="s2">,</span><span class="si">{2}</span><span class="s2">,</span><span class="si">{3}</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"</span><span class="si">{0}</span><span class="s2">,</span><span class="si">{5}</span><span class="s2">,</span><span class="si">{9}</span><span class="s2">,</span><span class="si">{13}</span><span class="s2">"</span><span class="p">]:</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">"Affinity: </span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">affinity</span><span class="p">))</span>
- <span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">GOMP_CPU_AFFINITY</span><span class="o">=</span><span class="nv">$affinity</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">100</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
+ <span class="o">!</span><span class="nb">eval</span> <span class="nv">OMP_DISPLAY_ENV</span><span class="o">=</span><span class="nb">true</span> <span class="nv">OMP_PLACES</span><span class="o">=</span><span class="nv">$affinity</span> <span class="nv">OMP_NUM_THREADS</span><span class="o">=</span><span class="m">4</span> <span class="nv">$$</span>SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d <span class="m">1000</span> <span class="m">1000</span> <span class="m">1000</span> <span class="p">|</span> grep <span class="s2">"OMP_PLACES\|speedup"</span>
</pre></div>
</div>
@@ -14620,16 +16013,16 @@ SMT=4
<div class="output_subarea output_stream output_stdout output_text">
-<pre>Affinity: 0,1,2,3
+<pre>Affinity: {0},{1},{2},{3}
<<Waiting for dispatch ...>>
<<Starting on login1>>
- OMP_PLACES = '{0},{1},{2},{3}'
-1000x100: Ref: 1.9854 s, This: 0.2326 s, speedup: 8.53
-Affinity: 0,5,9,13
+ OMP_PLACES='{0},{1},{2},{3}' custom
+1000x1000: Ref: 5.9792 s, This: 2.4122 s, speedup: 2.48
+Affinity: {0},{5},{9},{13}
<<Waiting for dispatch ...>>
<<Starting on login1>>
- OMP_PLACES = '{0},{5},{9},{13}'
-1000x100: Ref: 1.9828 s, This: 0.0833 s, speedup: 23.80
+ OMP_PLACES='{0},{5},{9},{13}' custom
+1000x1000: Ref: 2.3101 s, This: 0.6884 s, speedup: 3.36
</pre>
</div>
</div>
@@ -14637,12 +16030,21 @@ Affinity: 0,5,9,13
</div>
</div>
+</div>
+<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
+</div><div class="inner_cell">
+<div class="text_cell_render border-box-sizing rendered_html">
+<p>Likewise we see a higher speedup when we bind the threads to different cores rather than to a single core. This handson illustrates that apart from compiler level tuning, system level tuning is also equally important to obtain performance improvements</p>
+
+</div>
+</div>
</div>
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<h4 id="References">References<a class="anchor-link" href="#References">¶</a></h4><ol>
<li><a href="https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html">https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html</a></li>
+<li><a href="https://www.openmp.org/spec-html/5.0/openmpse53.html">https://www.openmp.org/spec-html/5.0/openmpse53.html</a></li>
</ol>
</div>
@@ -14660,7 +16062,7 @@ Affinity: 0,5,9,13
<div class="cell border-box-sizing text_cell rendered"><div class="prompt input_prompt">
</div><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
-<h1 id="Survey">Survey<a name="survey" /><a class="anchor-link" href="#Survey">¶</a></h1><p>Please rememeber to take some time and fill out the <a href="http://bit.ly/sc18-eval">survey</a>.</p>
+<h1 id="Survey">Survey<a name="survey" /><a class="anchor-link" href="#Survey">¶</a></h1><p>Please rememeber to take some time and fill out the <a href="http://bit.ly/sc19-eval">survey</a>.</p>
</div>
</div>
diff --git a/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.ipynb b/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.ipynb
index e6a4bae..5208f9b 100644
--- a/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.ipynb
+++ b/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.ipynb
@@ -1 +1,2416 @@
-{"cells": [{"cell_type": "markdown", "metadata": {}, "source": ["# Hands-On Performance Optimization\n", "_Supercomputing 2018 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 12th 2018_\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.\n", "\n", "## Jupyter notebook execution\n", "\n", "When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the _edit_ menu above.\n", "\n", "You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n", "\n", "If you want you also can get a [terminal](/terminals/1) in your browser.\n", "\n", "## Terminal fallback\n", "\n", "The tasks are place in directories named `Task[1-3]`.\n", "\n", "Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description."]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Setup\n", "\n", "This hands-on session requires of GCC 6.4.0. By loading the `sc18/handson2` module before invoking this Notebook, we took care of also loading GCC 6.4.0 into the environment."]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Tasks<a name=\"top\"></a>\n", "\n", "This session comes with multiple tasks, each one to be found in the respective sub-directory `Task[1-3]`. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.\n", "\n", "Please choose from the task below.\n", "\n", "\n", "* [Task 1](#task1): Compile Flags \n", "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback ([Solution 1](#solution0))\n", "\n", "* [Task 2](#task2): Software Prefetching \n", "Improve performance of the CPU Jacobi solver with software prefetching ([Solution 2](#solution1))\n", "\n", "* [Task 3](#task3): OpenMP \n", "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance ([Solution 3](#solution2))\n", " \n", "* [Suvery](#survey) Please remember to take the survey !\n", " \n", "### Make Targets <a name=\"make\"></a>\n", "\n", "For all tasks we have defined the following make targets. \n", "\n", "* __poisson2d__: \n", " build `poisson2d` binary (default)\n", "* __run__: \n", " run `poisson2d` with default parameters\n"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Task 1: Compile Flags <a name=\"task1\"></a>\n", "\n", "\n", "### Overview\n", "\n", "The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver \n", "\n", "Your task is to:\n", "\n", "* Optimize performance with `-Ofast` flag\n", "* Optimize performance with profile directed feedback \n", "\n", "First, change the working directory to `Task1`."]}, {"cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task1\n"]}], "source": ["%cd Task1"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part A: `-Ofast` vs. `-O3`\n", "\n", "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. Right now, the Makefile specifies `-O3` as the optimization flag. Compile the code using `make` and run it with `make run` in the next two cells."]}, {"cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -O3 -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"]}], "source": ["!make"]}, {"cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n", "Job <5033> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "1.13user 0.00system 0:01.15elapsed 97%CPU (0avgtext+0avgdata 10944maxresident)k\n", "2560inputs+0outputs (1major+264minor)pagefaults 0swaps\n"]}], "source": ["!make run"]}, {"cell_type": "markdown", "metadata": {}, "source": ["You can use the GNU _perf_ tool to profile the application using the `perf` command (see below) and see the top time-consuming functions."]}, {"cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "[ perf record: Woken up 1 times to write data ]\n", "[ perf record: Captured and wrote 0.172 MB perf.O3.data (4125 samples) ]\n", "# To display the perf.data header info, please use --header/--header-only options.\n", "#\n", "#\n", "# Total Lost Samples: 0\n", "#\n", "# Samples: 4K of event 'cycles:u'\n", "# Event count (approx.): 3867635297\n", "#\n", "# Overhead Command Shared Object Symbol \n", "# ........ ......... ................. ........................................\n", "#\n", " 72.02% poisson2d poisson2d [.] 00000040.plt_call.fmax@@GLIBC_2.17\n", " 10.16% poisson2d poisson2d [.] poisson2d_reference\n", " 9.99% poisson2d poisson2d [.] main\n", " 4.69% poisson2d libc-2.17.so [.] __memcpy_power7\n", " 2.23% poisson2d libm-2.17.so [.] __fmaxf\n", " 0.75% poisson2d libm-2.17.so [.] __exp_finite\n", " 0.07% poisson2d poisson2d [.] 00000040.plt_call.memcpy@@GLIBC_2.17\n", " 0.02% poisson2d poisson2d [.] check_results\n", " 0.02% poisson2d libm-2.17.so [.] __GI___exp\n", " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n", " 0.01% poisson2d [kernel.kallsyms] [k] arch_local_irq_restore\n", " 0.00% poisson2d ld-2.17.so [.] _dl_new_object\n", " 0.00% poisson2d ld-2.17.so [.] _start\n", "\n", "\n", "#\n", "# (Tip: Show user configuration overrides: perf config --user --list)\n", "#\n"]}], "source": ["# perf record creates a perf.data file \n", "!perf record -o perf.O3.data -e cycles ./poisson2d\n", "# perf report opens the perf.data file \n", "!perf report -i perf.O3.data | cat"]}, {"cell_type": "markdown", "metadata": {}, "source": ["**TASK**: Now change the optimization flag in the [Makefile](/edit/Task1/Makefile) to `-Ofast` and repeat the steps in the following cell. In case you follow along non-interactive, call `make` and `make run` in your shell. (If you are in the Jupyter Notebook, you can actually click the link of the [Makefile](/edit/Task1/Makefile). In other cases, use `vim` which is installed on Ascent.)"]}, {"cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"]}], "source": ["!make"]}, {"cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n", "Job <5034> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.51user 0.00system 0:00.52elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+264minor)pagefaults 0swaps\n"]}], "source": ["!make run"]}, {"cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "[ perf record: Woken up 1 times to write data ]\n", "[ perf record: Captured and wrote 0.086 MB perf.Ofast.data (1889 samples) ]\n", "# To display the perf.data header info, please use --header/--header-only options.\n", "#\n", "#\n", "# Total Lost Samples: 0\n", "#\n", "# Samples: 1K of event 'cycles:u'\n", "# Event count (approx.): 1765737747\n", "#\n", "# Overhead Command Shared Object Symbol \n", "# ........ ......... ............. .......................\n", "#\n", " 44.65% poisson2d poisson2d [.] main\n", " 43.84% poisson2d poisson2d [.] poisson2d_reference\n", " 10.28% poisson2d libc-2.17.so [.] __memcpy_power7\n", " 1.12% poisson2d libm-2.17.so [.] __exp_finite\n", " 0.05% poisson2d poisson2d [.] check_results\n", " 0.03% poisson2d ld-2.17.so [.] _dl_relocate_object\n", " 0.02% poisson2d libc-2.17.so [.] __readdir64\n", " 0.01% poisson2d ld-2.17.so [.] _dl_new_object\n", " 0.00% poisson2d ld-2.17.so [.] _start\n", "\n", "\n", "#\n", "# (Tip: System-wide collection from all CPUs: perf record -a)\n", "#\n"]}], "source": ["# perf record creates a perf.data file \n", "!perf record -o perf.Ofast.data -e cycles ./poisson2d\n", "# perf report opens the perf.data file \n", "!perf report -i perf.Ofast.data | cat"]}, {"cell_type": "markdown", "metadata": {}, "source": ["If `perf` is unavailable to you on other machines, you can also study the disassembly with `objdump`: `objdump -lSd ./poisson2d > poisson2d.dis` (feel free to experiment with this in the Notebook as well, just prefix the command with a `!` to execute it.)"]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### Interpretation\n", "\n", "Depending on the application requirement, if a high precision of results is not mandatory, the users can compile an application with `-Ofast` which enables `\u2013ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part B: Profile-directed Feedback\n", "\n", "For the first level of optimization we saw `Ofast` cut the execution time of the `O3` binary by almost half.\n", "\n", "We can optimize the performance further by using profile directed feedback optimization.\n", "\n", "To compile using profile directed feedback with the GCC compiler we need to do the following steps\n", "\n", "1. We need to first build a training binary using `-fprofile-generate`; this instructs the compiler to record hot path information \n", "2. Run the training binary with a smaller input size; you should see a `.gcda` file generated which stores hot path information for further optimization by the compiler \n", "3. build the final binary using `-fprofile-use` which uses the profile information in the `.gcda` file \n", "4. Compare the performance of the final binary with the original `Ofast` binary \n", "\n", "**TASK**: First, search for `TODO1` in the [Makefile](/edit/Task1/Makefile). It defines an additional compilation flag for `gcc`. Insert `-fprofile-generate=FOLDER` there with FOLDER pointing to `$$SC18_DIR_SCRATCH`, your personal write-directory (the double dollar signs are intentional as they are used to escape in the GNU Make syntax).\n", "\n", "After editing, run the following two cells to train your program."]}, {"cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-generate=$SC18_DIR_SCRATCH\" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-generate=$SC18_DIR_SCRATCH\" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_train -lm\n"]}], "source": ["!make poisson2d_train"]}, {"cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_train 200 64 64\n", "Job <5035> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 200 iterations on 64 x 64 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.248743\n", " 100, 0.124046\n", "Calculate current execution.\n", " 0, 0.248743\n", " 100, 0.124046\n", "0.00user 0.00system 0:00.10elapsed 5%CPU (0avgtext+0avgdata 5248maxresident)k\n", "512inputs+0outputs (0major+115minor)pagefaults 0swaps\n", "mv $SC18_DIR_SCRATCH/*.gcda .\n"]}], "source": ["!make run_train"]}, {"cell_type": "markdown", "metadata": {}, "source": ["Now, a `.gcda` file exists in the directory which can be used for an profile-accelerated subsequent run.\n", "\n", "**TASK**: Edit the [Makefile](/edit/Task1/Makefile) again, this time modifying `TODO2` to be equivalent to `-fprofile-use`. A directory is not needed as we copied the gcda file into the current directory.\n", "\n", "Run the following cells in order to build using the newly added flag and then run with the profile-accelerated version."]}, {"cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-use\" -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec \"-fprofile-use\" -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_profile -lm\n"]}], "source": ["!make poisson2d_profile"]}, {"cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_profile\n", "Job <5036> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.47user 0.00system 0:00.48elapsed 98%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"]}], "source": ["!make run_profile"]}, {"cell_type": "markdown", "metadata": {}, "source": ["What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### References\n", "\n", "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n", "2. https://perf.wiki.kernel.org/index.php/Tutorial"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Task 2:<a name=\"task2\"></a> Software Pretechting\n", "\n", "\n", "### Overview\n", "\n", "Study the difference of program execution time of different optimization levels with and without software prefetching.\n", "\n", "First, change directory to that of Task 2"]}, {"cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task2\n"]}], "source": ["%cd ../Task2"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part A: Running"]}, {"cell_type": "markdown", "metadata": {}, "source": ["Look at the [Makefile](/edit/Task2/Makefile) and work on the TODOs. Please implement compile flags as mentioned in the Makefile target name.\n", "\n", "Afterwards, compile each target with the following cells and submit them to the batch system. Follow along accordingly in the non-interactive version of this Notebook."]}, {"cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_pref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_pref\n", "Job <5037> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "1.12user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10880maxresident)k\n", "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"]}], "source": ["!make run_o3_pref"]}, {"cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprefetch-loop-arrays -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_pref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_pref\n", "Job <5038> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.77user 0.00system 0:00.77elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+264minor)pagefaults 0swaps\n"]}], "source": ["!make run_ofast_pref"]}, {"cell_type": "code", "execution_count": 15, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_o3_nopref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_o3_nopref\n", "Job <5039> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "1.13user 0.00system 0:01.13elapsed 99%CPU (0avgtext+0avgdata 10944maxresident)k\n", "256inputs+0outputs (0major+266minor)pagefaults 0swaps\n"]}], "source": ["!make run_o3_nopref"]}, {"cell_type": "code", "execution_count": 16, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d_ofast_nopref -lm\n", "bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d_ofast_nopref\n", "Job <5040> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "0.82user 0.00system 0:00.82elapsed 99%CPU (0avgtext+0avgdata 10816maxresident)k\n", "256inputs+0outputs (0major+265minor)pagefaults 0swaps\n"]}], "source": ["!make run_ofast_nopref"]}, {"cell_type": "markdown", "metadata": {}, "source": ["Do you notice the impact difference with optimization levels? It's always important to carefully study the interplay of flags."]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part B: Analysis of Instructions\n", "\n", "Compilation with the software prefetching flag causes the compiler to generate the `__dcbt` and `__dcbtst` instructions that prefetch memory values to L3.\n", "\n", "Verify it using `objdump -lSd` on each file (`poisson2d_o3_pref`, `poisson2d_ofast_pref`, `poisson2d_o3_nopref`, `poisson2d_ofast_nopref`). You might want to grep for `dcb`."]}, {"cell_type": "code", "execution_count": 19, "metadata": {"sc18": "solution"}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["poisson2d_o3_pref:\n", "poisson2d_ofast_pref:\n", " 10000da0:\tec f1 00 7c \tdcbtst 0,r30\n", " 10000da4:\t2c fa 00 7c \tdcbt 0,r31\n", " 10000da8:\t2c 62 00 7c \tdcbt 0,r12\n", " 10000dac:\t2c b2 00 7c \tdcbt 0,r22\n", " 10000dcc:\t2c e2 00 7c \tdcbt 0,r28\n", " 10000dd0:\t2c ea 00 7c \tdcbt 0,r29\n", " 100010b4:\t2c 62 00 7c \tdcbt 0,r12\n", " 100010b8:\t2c 5a 00 7c \tdcbt 0,r11\n", " 100010c4:\tec 19 00 7c \tdcbtst 0,r3\n", " 100010cc:\t2c 22 00 7c \tdcbt 0,r4\n", " 100010d0:\t2c ea 00 7c \tdcbt 0,r29\n", " 100010d4:\t2c f2 00 7c \tdcbt 0,r30\n", " 100010dc:\t2c fa 00 7c \tdcbt 0,r31\n", "poisson2d_o3_nopref:\n", "poisson2d_ofast_nopref:\n"]}], "source": ["for f in [\"poisson2d_o3_pref\", \"poisson2d_ofast_pref\", \"poisson2d_o3_nopref\", \"poisson2d_ofast_nopref\"]:\n", " print(\"{}:\".format(f))\n", " objdump -lSd $f |\u00a0grep dcb"]}, {"cell_type": "markdown", "metadata": {}, "source": ["If you feel up to the task, you can study the number of L3 cache misses using the corresponding performance counter, `PM_L3_MISS`. Either use your knowledge from Hands-On 1, or use the following call to `perf`, in which we already converted the named counter to a raw counter address."]}, {"cell_type": "code", "execution_count": 35, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Job <5048> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "\n", " Performance counter stats for './poisson2d_ofast_nopref':\n", "\n", " 2829292169 cycles:u \n", " 136018637 r168a4:u \n", "\n", " 0.826136863 seconds time elapsed\n", "\n", "Job <5049> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "\n", " Performance counter stats for './poisson2d_ofast_pref':\n", "\n", " 2654990243 cycles:u \n", " 128824827 r168a4:u \n", "\n", " 0.775593651 seconds time elapsed\n", "\n"]}], "source": ["for f in [\"poisson2d_ofast_nopref\", \"poisson2d_ofast_pref\"]:\n", " !eval $$SC18_SUBMIT_CMD perf stat -e cycles,r168a4 ./$f\n"]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### References\n", "\n", "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n", "2. https://www.gnu.org/software/gcc/projects/prefetch.html"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["## Task 3: OpenMP\n", "<a name=\"task3\"></a>\n", "\n", "\n", "### Overview\n", "\n", "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores.\n", "\n", "First, we need to change directory to that of Task3."]}, {"cell_type": "code", "execution_count": 1, "metadata": {"ExecuteTime": {"end_time": "2018-11-07T13:47:57.724441Z", "start_time": "2018-11-07T13:47:57.718745Z"}}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/autofs/nccsopen-svm1_home/aherten/SC18-Tutorial/3-Optimizing_POWER/Handson/Task3\n"]}], "source": ["%cd ../Task3"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part A: Implement OpenMP Pragmas; Compilation\n", "\n", "**Task**: Please add the correct OpenMP pragmas to the source code and compilations flags to enable OpenMP.\n", "\n", "* **pragmas**: Look at the TODOs in [`poisson2d.c`](/edit/Task3/poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for`\n", "* **Compilation**: Please add compilation flags enabling OpenMP in GCC to the [Makefile](/edit/Task3/Makefile). The flag in question is `-fopenmp`.\n", "\n", "Edit the files with the links above if you are running the interactive version of the Notebook or navigate to `poisson2d.c` and `Makefile` yourself in case you run the non-interactive version.\n", "\n", "Afterwards, compile and run the application with the following cells. Non-interactive: Follow along accordingly in the shell."]}, {"cell_type": "code", "execution_count": 37, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["/sw/ascent/gcc/6.4.0/bin/gcc -c -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec poisson2d_reference.c -o poisson2d_reference.o -lm\n", "/sw/ascent/gcc/6.4.0/bin/gcc -std=c99 -mcpu=power9 -Ofast -DUSE_DOUBLE -mvsx -maltivec -fopenmp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"]}], "source": ["!make poisson2d"]}, {"cell_type": "code", "execution_count": 40, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["bsub -W 60 -nnodes 1 -Is -P GEN111 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n", "Job <5052> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 500 iterations on 500 x 500 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "Calculate current execution.\n", " 0, 0.249980\n", " 100, 0.246028\n", " 200, 0.242198\n", " 300, 0.238487\n", " 400, 0.234887\n", "500x500: Ref: 0.2571 s, This: 0.2946 s, speedup: 0.87\n", "1.48user 0.00system 0:00.56elapsed 263%CPU (0avgtext+0avgdata 9664maxresident)k\n", "0inputs+0outputs (0major+273minor)pagefaults 0swaps\n"]}], "source": ["!make run"]}, {"cell_type": "markdown", "metadata": {}, "source": ["The command to submit a job to the batch system is prepared in an environment variable `$SC18_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to increase the work of the application."]}, {"cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Job <5344> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "Jacobi relaxation calculation: max 1000 iterations on 1000 x 100 mesh\n", "Calculate reference solution and time with serial CPU execution.\n", " 0, 0.249743\n", " 100, 0.210080\n", " 200, 0.184635\n", " 300, 0.166526\n", " 400, 0.152783\n", " 500, 0.141890\n", " 600, 0.132978\n", " 700, 0.125511\n", " 800, 0.119142\n", " 900, 0.113632\n", "Calculate current execution.\n", " 0, 0.249743\n", " 100, 0.210080\n", " 200, 0.184635\n", " 300, 0.166526\n", " 400, 0.152783\n", " 500, 0.141890\n", " 600, 0.132978\n", " 700, 0.125511\n", " 800, 0.119142\n", " 900, 0.113632\n", "1000x100: Ref: 1.9872 s, This: 0.2385 s, speedup: 8.33\n"]}], "source": ["!eval $SC18_SUBMIT_CMD ./poisson2d 1000 1000"]}, {"cell_type": "markdown", "metadata": {}, "source": ["What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n", "We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."]}, {"cell_type": "code", "execution_count": 23, "metadata": {"sc18": "solution"}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Threads: 1\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3037 s, This: 2.8420 s, speedup: 0.81\n", "Threads: 2\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.2998 s, This: 1.4320 s, speedup: 1.61\n", "Threads: 4\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3135 s, This: 0.7168 s, speedup: 3.23\n", "Threads: 8\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3145 s, This: 0.5278 s, speedup: 4.39\n", "Threads: 10\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3153 s, This: 0.4848 s, speedup: 4.78\n", "Threads: 20\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3190 s, This: 0.2016 s, speedup: 11.50\n", "Threads: 40\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "1000x1000: Ref: 2.3243 s, This: 0.3057 s, speedup: 7.60\n"]}], "source": ["for omp_num in [1, 2, 4, 8, 10, 20, 40]:\n", " print(\"Threads: {}\".format(omp_num))\n", " !eval OMP_NUM_THREADS=$omp_num $$SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 | grep speedup"]}, {"cell_type": "markdown", "metadata": {}, "source": ["### Part B: Bindings\n", "\n", "Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!\n", "\n", "There are applications which can be used to determine the configuration of the processor. Among those are:\n", "\n", "* `lscpu`: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.\n", "* `ppc64_cpu --smt`: Specifically for POWER, this tool can give information about the number of simulations threads running per core (*SMT*, Simulataion Multi-Threading).\n", "\n", "Run `ppc64_cpu --smt` to find out about the threading configuration of Ascent!"]}, {"cell_type": "code", "execution_count": 48, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Job <5076> is submitted to default queue <batch>.\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", "SMT=4\n"]}], "source": ["!eval $SC18_SUBMIT_CMD ppc64_cpu --smt"]}, {"cell_type": "markdown", "metadata": {}, "source": ["There are more sources information available\n", "\n", "* `/proc/cpuinfo`: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with `cat`\n", "* `/sys/devices/system/cpu/cpu0/topology/thread_siblings_list`: Holds information about thread siblings for given CPU core (`cpu0` in this case). Use it to find out which thread is mapped to which core."]}, {"cell_type": "code", "execution_count": 49, "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["0-3\n", "4-7\n"]}], "source": ["!cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n", "!cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"]}, {"cell_type": "markdown", "metadata": {}, "source": ["There are various environment variables available within OpenMP (and GCC) to specify binding of threads to cores. See, for instance, the [online documentation of GCC libgomp](https://gcc.gnu.org/onlinedocs/libgomp/Environment-Variables.html). Examples are `OMP_PLACES` or `GOMP_CPU_AFFINITY`.\n", "\n", "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n", "\n", "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n", "\n", "What's your maximum speedup?"]}, {"cell_type": "code", "execution_count": 24, "metadata": {"sc18": "solution"}, "outputs": [{"name": "stdout", "output_type": "stream", "text": ["Affinity: 0,1,2,3\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", " OMP_PLACES = '{0},{1},{2},{3}'\n", "1000x100: Ref: 1.9854 s, This: 0.2326 s, speedup: 8.53\n", "Affinity: 0,5,9,13\n", "<<Waiting for dispatch ...>>\n", "<<Starting on login1>>\n", " OMP_PLACES = '{0},{5},{9},{13}'\n", "1000x100: Ref: 1.9828 s, This: 0.0833 s, speedup: 23.80\n"]}], "source": ["for affinity in [\"0,1,2,3\", \"0,5,9,13\"]:\n", " print(\"Affinity: {}\".format(affinity))\n", " !eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=$affinity OMP_NUM_THREADS=4 $$SC18_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 100 | grep \"OMP_PLACES\\|speedup\""]}, {"cell_type": "markdown", "metadata": {}, "source": ["#### References\n", "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html"]}, {"cell_type": "markdown", "metadata": {}, "source": ["[Back to Top](#top)\n", "\n", "---"]}, {"cell_type": "markdown", "metadata": {}, "source": ["# Survey<a name=\"survey\"></a>\n", "\n", "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc18-eval)."]}], "metadata": {"kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"}, "language_info": {"codemirror_mode": {"name": "ipython", "version": 3}, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.6.7"}}, "nbformat": 4, "nbformat_minor": 2}
\ No newline at end of file
+{
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Hands-On Performance Optimization\n",
+ "_Supercomputing 2019 Tutorial \"Application Porting and Optimization on GPU-Accelerated POWER Architectures\", November 18th 2019_\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "As for the first task of this tutorial, also this task is primarily designed to be executed as an interactive Jupyter Notebook. However, everything can also be done using an SSH connection to Ascent (or any other POWER9 computer) in your terminal.\n",
+ "\n",
+ "## Jupyter notebook execution\n",
+ "\n",
+ "When using Jupyter, this Notebook will guide you through the steps. Note that if you execute a cell multiple times while optimizng the code the output will be replaced. You can however duplicate the cell you want to execute and keep its output. Check the _edit_ menu above.\n",
+ "\n",
+ "You will always find links to a file browser of the corresponding task subdirectory as well as direct links to the source files you will need to edit as well as the profiling output you need to open locally.\n",
+ "\n",
+ "If you want you also can get a terminal in your browser; just open it via the \u00bbNew Launcher\u00ab button (`+`).\n",
+ "\n",
+ "## Terminal fallback\n",
+ "\n",
+ "The tasks are place in directories named `Task[1-3]`.\n",
+ "\n",
+ "Makefile targets are created to cover everything, from compile, to run and profile. Please take a look at the cells containing the make calls as a guide also for the non-interactive version of this description."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Setup\n",
+ "\n",
+ "We are using some very fresh compiler features and use GCC 9.2.0 because of that. It should already be in your environment. Let's check!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc (GCC) 9.2.0\n",
+ "Copyright (C) 2019 Free Software Foundation, Inc.\n",
+ "This is free software; see the source for copying conditions. There is NO\n",
+ "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!gcc --version"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Tasks<a name=\"top\"></a>\n",
+ "\n",
+ "This session comes with multiple tasks, each one to be found in the respective sub-directory `Task[1-3]`. In each of these directories you will also find Makefiles that are set up so that you can compile and submit all necessary tasks.\n",
+ "\n",
+ "Please choose from the task below.\n",
+ "\n",
+ "\n",
+ "* [Task 1](#task1): __Basic compiler optimization flags and compiler annotations__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with compiler flags such as `Ofast` and profile-directed feedback. Learn about compiler annotations.\n",
+ "\n",
+ "* [Task 2](#task2): __Optimization via Prefetching controlled by compiler__\n",
+ "\n",
+ "Improve performance of the CPU Jacobi solver with software prefetching. Some compilers such as IBM XL define flags that can be used to modify the aggressiveness of the hardware prefetcher. Learn to modify the DSCR value through XL and study the impact on application performance. \n",
+ "* [Task 3](#task3): __Optimization via OpenMP controlled by compiler and the system__\n",
+ "\n",
+ "Parallelize the CPU Jacobi solver and determine the right binding to be used for optimal performance. \n",
+ " \n",
+ "* [Suvery](#survey) Please remember to take the survey !\n",
+ " \n",
+ "### Make Targets <a name=\"make\"></a>\n",
+ "\n",
+ "For all tasks we have defined the following make targets. \n",
+ "\n",
+ "* __poisson2d__: \n",
+ " build `poisson2d` binary (default)\n",
+ "* __run__: \n",
+ " run `poisson2d` with default parameters\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 1: Basic compiler optimization flags and compiler annotations <a name=\"task1\"></a>\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "The goal of this task is to understand different options available to optimize the performance of the CPU Jacobi solver \n",
+ "\n",
+ "Your task is to:\n",
+ "\n",
+ "* Optimize performance with `-Ofast` flag\n",
+ "* Verify the cause for performance improvement by viewing perf profiles of O3 and Ofast binaries \n",
+ "* Optimize performance with profile directed feedback \n",
+ "* Generate compiler annotations/remarks to understand the optimizations done by the compiler with and without profile directed feedback \n",
+ "\n",
+ "\n",
+ "First, change the working directory to `Task1`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task1\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd Task1"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: `-Ofast` vs. `-O3`\n",
+ "\n",
+ "We are to compare the performance of the binary being compiled with `-Ofast` optimization and with `-O3` optimization. As in the previous task, we use a `Makefile` for compilation. The `Makefile` targets `poisson2d_O3` and `poisson2d_Ofast` are already prepared. \n",
+ "\n",
+ "**TASK**: Add `-O3` as the optimization flag for the `poisson2d_O3` target by using the corresponding `CFLAGS` definition. There are notes relating to this Task 1 in the header of the `Makefile`. Compile the code using `make` as indicated below and run with the `Make` targets `run`, `run_perf` and `run_perf_recrep`. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 84,
+ "metadata": {
+ "collapsed": true
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -c -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -O3 poisson2d.c poisson2d_reference.o -o poisson2d -lm\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_O3"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 73,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24897> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's have a look at the output of the `Makefile` target `run_perf`. It invokes the GNU _perf_ tool to print out details of the number of instructions executed and the number of cycles taken by POWER9 to execute the program. Feel free to add further counter to this call to _perf_."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 74,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24898> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16264721613 cycles:u \n",
+ " 28463907825 instructions:u # 1.75 insn per cycle \n",
+ "\n",
+ " 4.738444892 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Next we run the makefile with target `run_perf_recrep` that prints the top routines of the application in terms of hotness by using a combination of `perf record ./app` and `perf report`. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 75,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24899> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "[ perf record: Woken up 3 times to write data ]\n",
+ "[ perf record: Captured and wrote 0.739 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (19102 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24900> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "# To display the perf.data header info, please use --header/--header-only options.\n",
+ "#\n",
+ "#\n",
+ "# Total Lost Samples: 0\n",
+ "#\n",
+ "# Samples: 19K of event 'cycles:u'\n",
+ "# Event count (approx.): 16254596654\n",
+ "#\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
+ "#\n",
+ " 65.50% poisson2d poisson2d [.] 00000038.plt_call.fmax@@GLIBC_2.17\n",
+ " 21.21% poisson2d poisson2d [.] main\n",
+ " 9.18% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 3.28% poisson2d libm-2.17.so [.] __fmaxf\n",
+ " 0.74% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.01% poisson2d libm-2.17.so [.] __GI___exp\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.00% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] _wordcopy_fwd_aligned\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ "\n",
+ "\n",
+ "#\n",
+ "# (Tip: Limit to show entries above 5% only: perf report --percent-limit 5)\n",
+ "#\n"
+ ]
+ }
+ ],
+ "source": [
+ "# run_perf_recrep displays the top hot routines \n",
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Now add the optimization flag `Ofast` to the `CFLAGS` for target `poisson2d_Ofast`. Compile the program with the target `poisson2d_Ofast` and run and analyse it as before with `run`, `run_perf` and `run_perf_recrep`.\n",
+ "\n",
+ "What difference do you see?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 76,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24901> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.41user 0.00system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast \n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Again, run a `perf`-instrumented version:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 77,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24902> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 8258991976 cycles:u \n",
+ " 12013091172 instructions:u # 1.45 insn per cycle \n",
+ "\n",
+ " 2.408703909 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Generate the list of top routines in terms of hotness:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 78,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf record -e cycles --output=/gpfs/wolf/trn003/scratch/aherten//cycles.data ./poisson2d\n",
+ "Job <24903> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "[ perf record: Woken up 2 times to write data ]\n",
+ "[ perf record: Captured and wrote 0.382 MB /gpfs/wolf/trn003/scratch/aherten//cycles.data (9728 samples) ]\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf report -i /gpfs/wolf/trn003/scratch/aherten//cycles.data --stdio\n",
+ "Job <24904> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "# To display the perf.data header info, please use --header/--header-only options.\n",
+ "#\n",
+ "#\n",
+ "# Total Lost Samples: 0\n",
+ "#\n",
+ "# Samples: 9K of event 'cycles:u'\n",
+ "# Event count (approx.): 8268811890\n",
+ "#\n",
+ "# Overhead Command Shared Object Symbol \n",
+ "# ........ ......... ............. ........................................\n",
+ "#\n",
+ " 81.12% poisson2d poisson2d [.] main\n",
+ " 17.97% poisson2d libc-2.17.so [.] __memcpy_power7\n",
+ " 0.79% poisson2d libm-2.17.so [.] __exp_finite\n",
+ " 0.04% poisson2d poisson2d [.] 00000038.plt_call.memcpy@@GLIBC_2.17\n",
+ " 0.02% poisson2d ld-2.17.so [.] do_lookup_x\n",
+ " 0.01% poisson2d libc-2.17.so [.] vfprintf@@GLIBC_2.17\n",
+ " 0.01% poisson2d libc-2.17.so [.] _dl_addr\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_relocate_object\n",
+ " 0.01% poisson2d ld-2.17.so [.] check_match.10253\n",
+ " 0.01% poisson2d ld-2.17.so [.] _dl_lookup_symbol_x\n",
+ " 0.01% poisson2d ld-2.17.so [.] strcmp\n",
+ " 0.00% poisson2d ld-2.17.so [.] open_path\n",
+ " 0.00% poisson2d ld-2.17.so [.] init_tls\n",
+ " 0.00% poisson2d ld-2.17.so [.] _dl_sysdep_start\n",
+ " 0.00% poisson2d ld-2.17.so [.] _start\n",
+ "\n",
+ "\n",
+ "#\n",
+ "# (Tip: For tracepoint events, try: perf report -s trace_fields)\n",
+ "#\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf_recrep"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If `perf` is unavailable to you on other machines, you can also study the disassembly with `objdump`: `objdump -lSd ./poisson2d > poisson2d.dis` (feel free to experiment with this in the Notebook as well, just prefix the command with a `!` to execute it.)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### Interpretation\n",
+ "\n",
+ "Depending on the application requirement, if a high precision of results is not mandatory, one can compile an application with `-Ofast` which enables `\u2013ffast-math` option that implements the same math function in a relaxed manner very similar to how general mathematical expressions are implemented and avoids the overhead of calling a function from the math library. Comparing the files, you will see that the `-Ofast` binary natively implements the `fmax` function using instructions available in the hardware. The `-O3` binary makes a library call to compute `fmax` to follow a stricter _IEEE_ requirement for accuracy."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Profile-directed Feedback\n",
+ "\n",
+ "For the first level of optimization we see that `Ofast` cut the execution time of the `O3` binary by almost half.\n",
+ "\n",
+ "We can optimize the performance further by using profile-directed feedback optimization.\n",
+ "\n",
+ "To compile using profile-directed feedback with the GCC compiler we need to build the appplication in three stages:\n",
+ "\n",
+ "1. Instrument binary;\n",
+ "2. Run binary with training, gather profile information;\n",
+ "3. Use profile information to generate optimized binary.\n",
+ "\n",
+ "\n",
+ "Step 1 is achieved by compiling the binary with the correct flag \u2013\u00a0`-fprofile-generate`. In our case, we need to specify an output location, which should be `$(SC19_DIR_SCRATCH)`.\n",
+ "\n",
+ "Step 2 consists of a usual, albeit shorter run of the instrumented binary. The can be very short, though the parameters need to be representative of the actual run. After the binary ran, an output file (with file extension `.gcda`) is written to the directory specified during compilation.\n",
+ "\n",
+ "For Step 3, the binary is once again compiled, but this time using the `gcda` profile just generated. The according flag is `-fprofile-use`, which we set to `$(SC19_DIR_SCRATCH)` as well.\n",
+ "\n",
+ "In our `Makefile` at hand, we prepared the steps already for you in the form of two targets.\n",
+ "\n",
+ "* `poisson2d_train`: Will compile the binary with profile-directed feedback\n",
+ "* `poisson2d_ref`: Will take a generated profile and compile a new, optimized binary\n",
+ "\n",
+ "By using dependencies, between these two targets a profile run is launched.\n",
+ "\n",
+ "**TASK**: Edit the [Makefile](`Makefile`) and add the `-fprofile-*` flags to the `CFLAGS` of `poisson2d_train` and\n",
+ "`poisson2d_ref` as outline in the file.\n",
+ "\n",
+ "After that, you may launch them with the following cells (`gen_profile` is a meta-target and uses `poisson2d_train` and `poisson2d_ref`). If you need to clean the generated profile, you may use `make clean_profile`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 79,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24905> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_ref -lm \n",
+ "cp poisson2d_ref poisson2d\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make gen_profile"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "If the previous cell executed correctly, you now have your optimized executable. Let's see if it even fast than before!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 80,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24906> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.28user 0.01system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Great! It is! In our tests, this shaved off another 5%."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's also measure instructions and cycles"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 81,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,instructions ./poisson2d\n",
+ "Job <24907> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 7925983538 cycles:u \n",
+ " 12253080719 instructions:u # 1.55 insn per cycle \n",
+ "\n",
+ " 2.313471365 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make run_perf"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "What is your speed-up? Feel free to run with larger problem sizes (mesh; iterations)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Compiler annotations/Remarks\n",
+ "\n",
+ "Usually, all compilers provide an option to emit annotations or remarks by the compiler. These remarks summarize the optimizations done in detail, the location in source where these optimizations were done. There exist options that also indicate optimizations that were missed and the reason why they could not be done. \n",
+ "\n",
+ "To generate compiler annotations using GCC, one uses `-fopt-info-all`. If you only want to see the missed options, use the option `-fopt-info-missed` instead of `-fopt-info-all`. See also the [documentation of GCC regarding the flag](https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info).\n",
+ "\n",
+ "**TASK**: Have a looK at the `CFLAGS` of the `Makefile` target `poisson2d_Ofast_info`. Add the flag `-fopt-info-all` to the list of flags. This will print optimisation information to stdout. If you rather want to print to this information to a file, use \u2013\u00a0for example \u2013\u00a0`-fopt-info-all=(SC19_DIR_SCRATCH)/filename`."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 82,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all poisson2d.c poisson2d_reference.o -o poisson2d_Ofast_info -lm\n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 207->207 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_528, 0, _531);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_543, _539, _549);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_Ofast_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Let's compare this with the output during compilation when using profile-directed feedback from Task 1 B.\n",
+ "\n",
+ "**TASK**: \n",
+ "Adapt the `CFLAGS` of `poisson2d_ref_info` to include `-fopt-info-all` **and** the profile input of `-fprofile-use=\u2026` here. *(Be advised: Long output!)*"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 83,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ -Ofast -fprofile-generate=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c -o poisson2d_train -lm \n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "Increasing alignment of decl: __gcov0.main\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_D_00100_1_main/48 -> __gcov_exit/55, function body not available\n",
+ "poisson2d.c:164:1: missed: not inlinable: _GLOBAL__sub_I_00100_0_main/47 -> __gcov_init/54, function body not available\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 295->295 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:122:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:88:5: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:72:5: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_632, 0, _239);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_337, ny_124, nx_286);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_64, _135, _313);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_316, error_118);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_127);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_311);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_122);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_129 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_132 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_140 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 53\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 50\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 47\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 3 times (header execution count 9800)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 33\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 30\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 42\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 40\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 60\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 23\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 3 times (header execution count 100)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 12\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 16\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_init (&*.LPBX0);\n",
+ "poisson2d.c:164:1: missed: statement clobbers memory: __gcov_exit ();\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS ./poisson2d_train 100 100 100\n",
+ "Job <24908> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "libgcov profiling error:/gpfs/wolf/trn003/scratch/aherten//#autofs#nccsopen-svm1_home#aherten#SC19-Tutorial#3-Optimizing_POWER#Handson#Task1#poisson2d.gcda:overwriting an existing profile data with a different timestamp\n",
+ "Jacobi relaxation calculation: max 100 iterations on 100 x 100 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249490\n",
+ "echo `date` > /gpfs/wolf/trn003/scratch/aherten//.profile_generated\n",
+ "gcc -std=c99 -mcpu=power9 -DUSE_DOUBLE -mvsx -maltivec -Ofast -fopt-info-all -fprofile-use=/gpfs/wolf/trn003/scratch/aherten/ poisson2d.c poisson2d_reference.o -o poisson2d_ref_info -lm\n",
+ "poisson2d.c:62:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:61:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:56:14: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:52:20: optimized: Inlining atoi/24 into main/33 (always_inline).\n",
+ "poisson2d.c:161:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:159:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:158:5: missed: not inlinable: main/33 -> free/38, function body not available\n",
+ "poisson2d.c:142:31: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:103:5: missed: not inlinable: main/33 -> __builtin_puts/37, function body not available\n",
+ "poisson2d.c:96:5: missed: not inlinable: main/33 -> printf/36, function body not available\n",
+ "poisson2d.c:78:29: missed: not inlinable: main/33 -> exp/35, function body not available\n",
+ "poisson2d.c:68:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:67:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "poisson2d.c:65:41: missed: not inlinable: main/33 -> malloc/34, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "/usr/include/stdlib.h:280:16: missed: not inlinable: main/33 -> strtol/39, function body not available\n",
+ "Unit growth for small function inlining: 207->207 (0%)\n",
+ "\n",
+ "Inlined 4 calls, eliminated 0 functions\n",
+ "\n",
+ "consider run-time aliasing test between *_84 and *_87\n",
+ "consider run-time aliasing test between *_92 and *_97\n",
+ "consider run-time aliasing test between *_104 and *_107\n",
+ "consider run-time aliasing test between *_111 and *_115\n",
+ "poisson2d.c:124:13: optimized: Loop 8 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:90:9: optimized: Loop 10 distributed: split to 0 loops and 1 library calls.\n",
+ "poisson2d.c:108:25: missed: couldn't vectorize loop\n",
+ "poisson2d.c:108:25: missed: not vectorized: loop nest containing two or more consecutive inner loops cannot be vectorized\n",
+ "poisson2d.c:136:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:136:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:131:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:131:9: missed: Loop costings may not be worthwhile.\n",
+ "poisson2d.c:122:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:112:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:112:9: missed: not vectorized: control flow in loop.\n",
+ "poisson2d.c:114:13: optimized: loop vectorized using 16 byte vectors\n",
+ "poisson2d.c:88:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:72:5: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:27: missed: not vectorized: complicated access pattern.\n",
+ "poisson2d.c:74:9: missed: couldn't vectorize loop\n",
+ "poisson2d.c:78:29: missed: not vectorized: relevant stmt not supported: _27 = exp (_21);\n",
+ "poisson2d.c:43:5: note: vectorized 1 loops in function.\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:114:13: optimized: loop turned into non-loop; it never loops\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _187 = strtol (_1, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _189 = strtol (_2, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _193 = strtol (_3, 0B, 10);\n",
+ "/usr/include/stdlib.h:280:16: missed: statement clobbers memory: _191 = strtol (_4, 0B, 10);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_153 = malloc (_7);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_155 = malloc (_7);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_157 = malloc (_7);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memset (_524, 0, _527);\n",
+ "poisson2d.c:96:5: missed: statement clobbers memory: printf (\"Jacobi relaxation calculation: max %d iterations on %d x %d mesh\\n\", iter_max_130, ny_139, nx_195);\n",
+ "poisson2d.c:103:5: missed: statement clobbers memory: __builtin_puts (&\"Calculate current execution.\"[0]);\n",
+ "poisson2d.c:43:5: missed: statement clobbers memory: __builtin_memcpy (_539, _535, _544);\n",
+ "poisson2d.c:142:31: missed: statement clobbers memory: printf (\"%5d, %0.6f\\n\", iter_237, error_219);\n",
+ "poisson2d.c:158:5: missed: statement clobbers memory: free (rhs_202);\n",
+ "poisson2d.c:159:5: missed: statement clobbers memory: free (Anew_124);\n",
+ "poisson2d.c:161:5: missed: statement clobbers memory: free (A_123);\n",
+ "poisson2d.c:65:41: missed: statement clobbers memory: A_144 = malloc (8000000);\n",
+ "poisson2d.c:67:41: missed: statement clobbers memory: Anew_143 = malloc (8000000);\n",
+ "poisson2d.c:68:41: missed: statement clobbers memory: rhs_142 = malloc (8000000);\n",
+ "poisson2d.c:136:9: note: considering unrolling loop 7 at BB 47\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:136:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:131:9: note: considering unrolling loop 6 at BB 44\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:131:9: optimized: loop unrolled 7 times (header execution count 9800)\n",
+ "poisson2d.c:122:9: note: considering unrolling loop 5 at BB 40\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:122:9: optimized: loop unrolled 7 times (header execution count 9701)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 13 at BB 27\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:118:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "poisson2d.c:118:25: note: considering unrolling loop 9 at BB 24\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:112:9: note: considering unrolling loop 14 at BB 37\n",
+ "poisson2d.c:43:5: note: considering unrolling loop 4 at BB 35\n",
+ "poisson2d.c:108:25: note: considering unrolling loop 3 at BB 51\n",
+ "poisson2d.c:88:5: note: considering unrolling loop 2 at BB 18\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:88:5: optimized: loop unrolled 7 times (header execution count 99)\n",
+ "poisson2d.c:74:9: note: considering unrolling loop 11 at BB 9\n",
+ "considering unrolling loop with constant number of iterations\n",
+ "considering unrolling loop with runtime-computable number of iterations\n",
+ "poisson2d.c:74:9: optimized: loop unrolled 3 times (header execution count 9604)\n",
+ "poisson2d.c:72:5: note: considering unrolling loop 1 at BB 14\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_ref_info"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Comparing the annotations generated of a plain `-Ofast` optimization level and the one generated at `-Ofast` and profile directed feedback, we observe that many more optimizations are possible due to profile information.\n",
+ "\n",
+ "For instance you will see annotations such as\n",
+ "```\n",
+ "poisson2d.c:114:25: optimized: loop unrolled 3 times (header execution count 436550)\n",
+ "```\n",
+ "\n",
+ "The execution count indicates the dynamic execution count of the node at runtime. This information determines which paths are hotter and subsequently facilitate additional optimizations."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "\n",
+ "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
+ "2. https://perf.wiki.kernel.org/index.php/Tutorial"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 2:<a name=\"task2\"></a> Impact of Prefetching on Performance\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "* Study the difference of program execution time of different optimization levels with and without software prefetching.\n",
+ "* Verify the impact by measuring cache counters with and without prefetching.\n",
+ "* Learn how to modify contents of DSCR (*Data Stream Control Register*) using IBM XL compiler and study the impact with different values to DSCR. \n",
+ "\n",
+ "But first, lets change directory to that of Task 2"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 85,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task2\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd ../Task2"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: Software Prefetching"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Look at the Makefile and work on the TODOs. \n",
+ "\n",
+ "- First generate a `-Ofast`-optimised binary and note down the performance in terms of cycles, seconds, and L3 misses. This is our baseline!\n",
+ "- Modify the `Makefile` to add the option for software prefetching (`-fprefetch-loop-arrays`). Compare performance of `-Ofast` with and without software prefetching"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 97,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "rm -f poisson2d poisson2d*.o\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make clean"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 88,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "make: `poisson2d' is up to date.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24911> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "2.39user 0.01system 0:02.40elapsed 100%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24912> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 8271503902 cycles:u \n",
+ " 481152478 r168a4:u \n",
+ "\n",
+ " 2.412224884 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 98,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ "cp poisson2d_pref poisson2d\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24919> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1.92user 0.00system 0:01.93elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24920> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 6586609284 cycles:u \n",
+ " 459879452 r168a4:u \n",
+ "\n",
+ " 1.925399505 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_pref CC=gcc\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: Repeat the experiment with the `-O3` flag. Have a look at the `Makefile` and the outlined TODO. There's a position to easily adapt `-Ofast`\u2192`-O3`!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 100,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec poisson2d.c -o poisson2d -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24923> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.73user 0.00system 0:04.73elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+479minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24924> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16445764669 cycles:u \n",
+ " 645094089 r168a4:u \n",
+ "\n",
+ " 4.792567763 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 101,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24925> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.74user 0.00system 0:04.74elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "0inputs+0outputs (0major+480minor)pagefaults 0swaps\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS perf stat -e cycles,r168a4 ./poisson2d\n",
+ "Job <24926> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "\n",
+ " Performance counter stats for './poisson2d':\n",
+ "\n",
+ " 16239159454 cycles:u \n",
+ " 631061431 r168a4:u \n",
+ "\n",
+ " 4.730144897 seconds time elapsed\n",
+ "\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_pref CC=gcc -B\n",
+ "!make run\n",
+ "!make l3missstats"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Do you notice the impact difference with optimization levels? At what optimization level does software prefetching help the most?"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Observing the results, we see that SW Prefetching seems to help at `-Ofast` but not at `-O3`. We can use the steps described in the the next section to verify that the compiler has not inserted any SW prefetch operations at`-O3` at all. That is because in the `-O3` binary the time is dominated by `__fmax` call which causes the compiler to come to the conclusion that whatever benefit we obtain by adding SW prefetch will be overshadowed by the penalty of `fmax()`\n",
+ "GCC may add further loop optimizations such as unrolling upon invocation of `\u2013fprefetch-loop-arrays`.\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Analysis of Instructions\n",
+ "\n",
+ "Compilation of the `-Ofast` binary with the software prefetching flag causes the compiler to generate the `dcb*` instructions that prefetch memory values to L3."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "**TASK**: \n",
+ "Run `$(SC19_SUBMIT_CMD) objdump -lSd` on each binary file (`-O3`, `-Ofast` with prefetch/no prefetch).\n",
+ "Look for instructions beginning with `dcb`\n",
+ "At what optimization levels does the compiler generate software prefetching instructions?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 114,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -std=c99 -DUSE_DOUBLE -Ofast -mcpu=power9 -mvsx -maltivec -fprefetch-loop-arrays poisson2d.c -o poisson2d_pref -lm\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=gcc -B poisson2d_pref\n",
+ "!objdump -lSd ./poisson2d_pref > poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 116,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ " 10000b28:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000b30:\t2c ba 00 7c \tdcbt 0,r23\n",
+ " 10000b38:\t2c b2 00 7c \tdcbt 0,r22\n",
+ " 10000b50:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000b58:\tec b9 00 7c \tdcbtst 0,r23\n",
+ " 10000b80:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000e64:\t2c 92 00 7c \tdcbt 0,r18\n",
+ " 10000e68:\t2c 9a 00 7c \tdcbt 0,r19\n",
+ " 10000e6c:\t2c a2 00 7c \tdcbt 0,r20\n",
+ " 10000e70:\t2c aa 00 7c \tdcbt 0,r21\n",
+ " 10000e7c:\t2c b2 00 7c \tdcbt 0,r22\n",
+ " 10000e80:\t2c d2 00 7c \tdcbt 0,r26\n",
+ " 10000e94:\tec b9 00 7c \tdcbtst 0,r23\n"
+ ]
+ }
+ ],
+ "source": [
+ "!grep dcb poisson2d.dis"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part C: Changing Values of DSCR via compiler flags\n",
+ "\n",
+ "This task requires using the IBM XL compiler. It should be already in your environment.\n",
+ "\n",
+ "\n",
+ "We saw the impact of software prefetching in the previous subsection. \n",
+ "In certain cases, tuning the hardware prefetcher through compiler options can also help improve performance. \n",
+ "In this exercise we shall see some compiler options that can be used to modify the DSCR value which controls aggressiveness of prefetching. It can be also used to turn off hardware prefetching. \n",
+ "\n",
+ "IBM XL compiler has an option `-qprefetch=dscr=<val>` that can be used for this purpose.\n",
+ "Compiling with `-qprefetch=dscr=1` turns off the prefetcher. One can give various values such as `-qprefetch=dscr=4`, `-qprefetch=dscr=7` etc. to control aggressiveness of prefetching.\n",
+ "\n",
+ "For this exercise we use `make CC=xlc_r` to illustrate the performance impact.\n",
+ " \n",
+ "\n",
+ "**Task** Generate a XL-compiled binary by compiling using the following cells. After you've generated a baseline, start editing the `Makefile`: Add `qprefetch=dscr=1` to the `CFLAGS` and rebuild the application and note the performance. Which one is faster? \n",
+ "\n",
+ "In general, applications benefit with the default settings of hardware DSCR register (`-qprefetch=dscr=0`). However, certain applications also benefit with prefetching turned off. \n",
+ "\n",
+ "It is to be noted that DSCR values are highly sensitive to the application. One value that works well for Application A may not help Application B. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL at default DSCR value"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 117,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS poisson2d.c -o poisson2d -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24927> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "2.26user 0.00system 0:02.27elapsed 99%CPU (0avgtext+0avgdata 24256maxresident)k\n",
+ "256inputs+0outputs (0major+477minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=xlc_r -B poisson2d\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Measure performance of the application compiled with XL with DSCR value turned off"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -std=c99 -DUSE_DOUBLE -Ofast -qarch=pwr9 -qtune=pwr9 -DINLINE_LIBS -qprefetch=dscr=1 poisson2d.c -o poisson2d_dscr -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24929> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "4.58user 0.00system 0:04.59elapsed 99%CPU (0avgtext+0avgdata 24192maxresident)k\n",
+ "0inputs+0outputs (0major+476minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d_dscr CC=xlc_r -B\n",
+ "!make run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Does Hardware prefetcher help this application? How much impact do you see when you turn off the hardware prefetcher? "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "The DSCR register controls the operation of the HW Prefetcher on POWER9. It can be modified in the command line by `ppc64_cpu --dscr=<value>`. However this needs admin privileges. IBM XL offers a compiler flag to set the value through the compiler. `-qprefetch=dscr=1` turns off the prefetcher. Observing the results we see that the performance without the HW prefetcher is twice as bad as that with default prefetching. So we can conclude that Prefetching helps the Jacobi application. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "\n",
+ "1. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html\n",
+ "2. https://www.gnu.org/software/gcc/projects/prefetch.html\n",
+ "3. https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Task 3: OpenMP\n",
+ "<a name=\"task3\"></a>\n",
+ "\n",
+ "\n",
+ "### Overview\n",
+ "\n",
+ "We add OpenMP shared-memory parallelism to the application. Also, we study the effect of binding the multi-thread processes to certain cores on the resulting application performance. We do this study for both GCC and XL compilers inorder to learn about the appropriate options that need to be used.\n",
+ "First, we need to change directory to that of Task3. For Task 3 we modify poisson2d.c to invoke an exact copy of the main jacobi loop which is `poisson2d_reference`. We parallelize only the main loop but not `poisson2d_reference`. The speedup is the performance gain seen in the main loop as compared to the reference loop."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "/autofs/nccsopen-svm1_home/aherten/SC19-Tutorial/3-Optimizing_POWER/Handson/Task3\n"
+ ]
+ }
+ ],
+ "source": [
+ "%cd ../Task3"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part A: Implement OpenMP Pragmas; Compilation\n",
+ "\n",
+ "**Task**: Please add the correct OpenMP directives to poisson2d.c and compilations flags in the Makefile to enable OpenMP with GCC and XL compilers.\n",
+ "\n",
+ "* **Directives**: Look at the TODOs in [`poisson2d.c`](poisson2d.c) to add OpenMP parallelism. The pragmas in question are `#pragma omp parallel for` (and once it's `#pragma omp parallel for reduction(max:error)` \u2013\u00a0can you guess where?)\n",
+ "* **Compilation**: Please add compilation flags enabling OpenMP in GCC and XL to the `Makefile`. For GCC, we need to add `-fopenmp` and the application needs to be linked with `-lgomp`. For XL, we need to add `-qsmp=omp` to the list of compilation flags. \n",
+ "\n",
+ "Afterwards, compile and run the application with the following commands."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 39,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "gcc -c -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d_reference.c -o poisson2d_reference.o -lm\n",
+ "gcc -std=c99 -DUSE_DOUBLE -O3 -mcpu=power9 -mvsx -maltivec -fopenmp -lgomp poisson2d.c poisson2d_reference.o -o poisson2d -lm \n"
+ ]
+ }
+ ],
+ "source": [
+ "!make poisson2d CC=gcc"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "The command to submit a job to the batch system is prepared in an environment variable `$SC19_SUBMIT_CMD`; use it together with `eval`. In the following cell, it is shown how to invoke the application using the batch system. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 40,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24951> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 0.248997\n",
+ " 200, 0.248007\n",
+ " 300, 0.247025\n",
+ " 400, 0.246050\n",
+ " 500, 0.245084\n",
+ " 600, 0.244124\n",
+ " 700, 0.243173\n",
+ " 800, 0.242228\n",
+ " 900, 0.241291\n",
+ "1000x1000: Ref: 4.7430 s, This: 3.9363 s, speedup: 1.20\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval $SC19_SUBMIT_CMD ./poisson2d 1000 1000 1000"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Inorder to run the parallel application, we need to set the number of threads using `OMP_NUM_THREADS`\n",
+ "What is the best performance you can reach by setting the number of threads via `OMP_NUM_THREADS=N` with `N` being the number of threads? Feel free to play around with the command in the following cell, using 1 thread as an example. \n",
+ "We added `--bind none` to prevent `jsrun`, the scheduler of Ascent, from overlaying binding options. Also, we use `-c ALL_CPUS` to make all CPUs on the compute nodes available to you."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 41,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 4.7288 s, This: 4.9791 s, speedup: 0.95\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=1 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 42,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 4.7125 s, This: 2.4914 s, speedup: 1.89\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=2 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 35,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.1065 s, This: 1.3836 s, speedup: 1.52\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 21,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3868 s, This: 0.5272 s, speedup: 4.53\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=8 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 22,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3912 s, This: 0.4612 s, speedup: 5.18\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=10 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 23,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3864 s, This: 0.4037 s, speedup: 5.91\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=20 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 24,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3773 s, This: 0.3045 s, speedup: 7.81\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=40 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 25,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3819 s, This: 0.3081 s, speedup: 7.73\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=80 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Part B: Bindings\n",
+ "\n",
+ "Different CPU architectures and models come with different configuration of cores. The configuration plays an important role in the run time of the application. We need to optimize for it!\n",
+ "\n",
+ "There are applications which can be used to determine the configuration of the processor. Among those are:\n",
+ "\n",
+ "* `lscpu`: Can be used to determine the number of sockets, number of cores, and numb of threads. It gives a very good overview and is available on most Linux systems.\n",
+ "* `ppc64_cpu --smt`: Specifically for POWER, this tool can give information about the number of simulations threads running per core (*SMT*, Simulataion Multi-Threading).\n",
+ "\n",
+ "Run `ppc64_cpu --smt` to find out about the threading configuration of Ascent!"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 55,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24465> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "SMT=4\n"
+ ]
+ }
+ ],
+ "source": [
+ "!eval $SC19_SUBMIT_CMD ppc64_cpu --smt"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "There are more sources information available\n",
+ "\n",
+ "* `/proc/cpuinfo`: Holds information about virtual cores, including model and clock speed. Available on most Linux system. Usually used together with `cat`\n",
+ "* `/sys/devices/system/cpu/cpu0/topology/thread_siblings_list`: Holds information about thread siblings for given CPU core (`cpu0` in this case). Use it to find out which thread is mapped to which core."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 36,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Job <24949> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "0-3\n",
+ "Job <24950> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "4-7\n"
+ ]
+ }
+ ],
+ "source": [
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list\n",
+ "!$$SC19_SUBMIT_CMD cat /sys/devices/system/cpu/cpu5/topology/thread_siblings_list"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "There are various environment variables available within OpenMP (some specific to GCC) that hold across compilers to specify binding of threads to cores. See, for instance, the [OMP_PLACES environment Variable](https://www.openmp.org/spec-html/5.0/openmpse53.html). We also have a GNU specific variable which can also be used to control affinity - `GOMP_CPU_AFFINITY`. Setting `GOMP_CPU_AFFINITY` is specific to GCC binaries but it internally serves the same function as setting `OMP_PLACES`. \n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Using `OMP_PLACES` for binding, and using some magical Python-Bash interplay:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 43,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: {0},{1},{2},{3}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{1},{2},{3}'\n",
+ "1000x1000: Ref: 4.7315 s, This: 3.9090 s, speedup: 1.21\n",
+ "Affinity: {0},{5},{9},{13}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{5},{9},{13}'\n",
+ "1000x1000: Ref: 4.6485 s, This: 1.2829 s, speedup: 3.62\n"
+ ]
+ }
+ ],
+ "source": [
+ "for affinity in [\"{0},{1},{2},{3}\", \"{0},{5},{9},{13}\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true OMP_PLACES=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "In this case, we carry out the same experiment using `GOMP_CPU_AFFINITY` which essentially sets the same environment variable `OMP_PLACES`. Running with two different configurations 1) Binding all threads to the same core 2) Binding all threads to different cores, we see a higher speedup in case of binding all threads to different cores."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 44,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: 0,1,2,3\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{1},{2},{3}'\n",
+ "1000x1000: Ref: 2.3964 s, This: 2.1361 s, speedup: 1.12\n",
+ "Affinity: 0,5,9,13\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES = '{0},{5},{9},{13}'\n",
+ "1000x1000: Ref: 2.3925 s, This: 0.7030 s, speedup: 3.40\n"
+ ]
+ }
+ ],
+ "source": [
+ "for affinity in [\"0,1,2,3\", \"0,5,9,13\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true GOMP_CPU_AFFINITY=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Great!\n",
+ "\n",
+ "If you still have time: The same experiments can be repeated with the IBM XL compiler. \n",
+ "The corresponding compiler flag to enable OpenMP parallelism that needs to be used for XL is `-qsmp=omp`\n",
+ "\n",
+ "**Task**: In the Makefile add the OpenMP flag and generate XL binaries with OpenMP and run the application with various number of threads and note the performance speedup."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 44,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "xlc_r -c -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d_reference.c -o poisson2d_reference.o -lm \n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "xlc_r -std=c99 -DUSE_DOUBLE -O3 -qhot -qtune=pwr9 -DINLINE_LIBS -qsmp=omp poisson2d.c poisson2d_reference.o -o poisson2d -lm\n",
+ " 1500-036: (I) The NOSTRICT option (default at OPT(3)) has the potential to alter the semantics of a program. Please refer to documentation on the STRICT/NOSTRICT option for more information.\n",
+ "bsub -W 60 -nnodes 1 -Is -P TRN003 jsrun -n 1 -c 1 -g ALL_GPUS time ./poisson2d\n",
+ "Job <24956> is submitted to default queue <batch>.\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "Jacobi relaxation calculation: max 1000 iterations on 1000 x 1000 mesh\n",
+ "Calculate reference solution and time with serial CPU execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "Calculate current execution.\n",
+ " 0, 0.249995\n",
+ " 100, 50.149062\n",
+ " 200, 99.849327\n",
+ " 300, 149.352369\n",
+ " 400, 198.659746\n",
+ " 500, 247.773000\n",
+ " 600, 296.693652\n",
+ " 700, 345.423208\n",
+ " 800, 393.963155\n",
+ " 900, 442.314962\n",
+ "1000x1000: Ref: 5.6783 s, This: 2.6528 s, speedup: 2.14\n",
+ "21.56user 6.18system 0:08.37elapsed 331%CPU (0avgtext+0avgdata 23040maxresident)k\n",
+ "3200inputs+0outputs (2major+1098minor)pagefaults 0swaps\n"
+ ]
+ }
+ ],
+ "source": [
+ "!make CC=xlc_r -B run"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Run the parallel application with varying numbre of threads (`OMP_NUM_THREADS`) and note the performance improvement. "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "exercise": "solution"
+ },
+ "source": [
+ "Just as in the GCC binary we see a similar speedup with higher number of threads until a certain point beyond which the benefit tapers off. "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 28,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2561 s, This: 2.6432 s, speedup: 0.85\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=1 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 29,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3071 s, This: 1.5343 s, speedup: 1.50\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=2 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 30,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2617 s, This: 0.6936 s, speedup: 3.26\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 31,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2728 s, This: 0.3402 s, speedup: 6.68\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=8 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 45,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.1678 s, This: 0.2869 s, speedup: 7.56\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=10 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 33,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2813 s, This: 0.1452 s, speedup: 15.71\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=20 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 34,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.3284 s, This: 0.0981 s, speedup: 23.75\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=40 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 35,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ "1000x1000: Ref: 2.2918 s, This: 0.1439 s, speedup: 15.92\n"
+ ]
+ }
+ ],
+ "source": [
+ "!OMP_NUM_THREADS=80 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep speedup "
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Now we repeat the exercise of using the right binding of threads for the XL binary. `OMP_PLACES` pertains to the XL binary as well as it is an OpenMP variable. `GOMP_CPU_AFFINITY` is specific to GCC binary so that cannot be used to set the binding.\n",
+ "\n",
+ "**Task**: Run the application enabled with OpenMP from Part A with different binding configurations. Make sure to at least run a) binding all threads to a single core and b) binding threads to different cores.\n",
+ "\n",
+ "Adapt the following command with your configuration \u2013 or follow along accordingly in the non-interactive version of the Notebook.\n",
+ "\n",
+ "We are mixing Python with Bash (`!`) here, so don't get confused (because of this, if we want to use Bash environment variables, we need to use two `$$`)\n",
+ "\n",
+ "What's your maximum speedup?"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 36,
+ "metadata": {
+ "exercise": "solution"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Affinity: {0},{1},{2},{3}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES='{0},{1},{2},{3}' custom\n",
+ "1000x1000: Ref: 5.9792 s, This: 2.4122 s, speedup: 2.48\n",
+ "Affinity: {0},{5},{9},{13}\n",
+ "<<Waiting for dispatch ...>>\n",
+ "<<Starting on login1>>\n",
+ " OMP_PLACES='{0},{5},{9},{13}' custom\n",
+ "1000x1000: Ref: 2.3101 s, This: 0.6884 s, speedup: 3.36\n"
+ ]
+ }
+ ],
+ "source": [
+ "for affinity in [\"{0},{1},{2},{3}\", \"{0},{5},{9},{13}\"]:\n",
+ " print(\"Affinity: {}\".format(affinity))\n",
+ " !eval OMP_DISPLAY_ENV=true OMP_PLACES=$affinity OMP_NUM_THREADS=4 $$SC19_SUBMIT_CMD -c ALL_CPUS --bind none ./poisson2d 1000 1000 1000 | grep \"OMP_PLACES\\|speedup\""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Likewise we see a higher speedup when we bind the threads to different cores rather than to a single core. This handson illustrates that apart from compiler level tuning, system level tuning is also equally important to obtain performance improvements \n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "#### References\n",
+ "1. https://gcc.gnu.org/onlinedocs/libgomp/GOMP_005fCPU_005fAFFINITY.html\n",
+ "2. https://www.openmp.org/spec-html/5.0/openmpse53.html"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "[Back to Top](#top)\n",
+ "\n",
+ "---"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Survey<a name=\"survey\"></a>\n",
+ "\n",
+ "Please rememeber to take some time and fill out the [survey](http://bit.ly/sc19-eval)."
+ ]
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.7.0"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
\ No newline at end of file
diff --git a/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.pdf b/3-Optimizing_POWER/Handson/Solution-Notebook/HandsOnPerformanceOptimization.pdf
index 4357a6fc0c527eb59248f5962db5e7b0e70630f3..283819d6780fad07fd929bfced380e2ad08bb0c1 100644
GIT binary patch
literal 129532
zcmY!laB<T$)HCH$ee&V$4=y7^1p|frq%1BQ8-2IToRZWceYc#%l2n(}<ouLWeV^34
z^pXq(BV!A5J3Fr8lA_eaT&{{+bE9_mX`2h3{jU9seVMP=d|n<Ai3sU}XLnDENyqfM
z`Mo(X`^WdQRTGcCo0j(O(6y<vVpU#Ou8NrR^!!o%b$_2vXL?(|eb4`me|~KH^I>v)
zP2|DlJNDe2E)}-F^{MJ_(`UJryZ=7Fy7R-sIrZgw_vSS3E|#3*-8VmYhlATJvuU>@
zG;+^;eqUkwd7V%5=b)VT+s*iVn(r?)THE(n%3}S*Pjx@;WPHl?FaG)dw*G1R`w<cm
zd*ZiGVQBrhb)B7_{=I{{w#w`H)*PQHuJN|K@Q-YV@HCb~Pg#X{<Kq5kcNJA~>ZD|>
z&;K<=`t{O{lL|f^KjIcq6XSg5NL#kdkqvpKoa~(9c}2DXx?N{p<ywnAS)ysVLtk4m
z{pj`gp{5`IANDW&n0NQiv)1B_q9+PH>F3!zK2Bn@pWYJHq+NeGEBxk73GScV;UT#e
zuh#W0nzJ*GD>Lj_sib4d8l(I7XRH4CDs-V@hj85m+X+FImJ>IgRXf|V@!Q1xKeBQc
z?WtXER=WJ3i`>`V@8{m<JXK?S8EAPezv#Ht>+tUx-%@Y=R`D%PxwCM~U#H(otZxd(
z>aX#X_5UtymVKrCS#{L!Q|D9I%`39*y?h^-q%c*^N5VtOMa1Enb&HU0??>m~cM7-I
z^(?fXZkn*;&(%dcP2_VvzGj$aKW)OFia)hKzkkv`9(}7?_rR=20@9KD_Z;zYywl9?
zKEX7t)4ARKl<%Bok>r;rcuwS83yjg8@;7Qp(1ykRb7zUp;t(ilF?|0ywEqv!ES-Z3
zcWZ`xdA`m6<>v?YdVTT_hc+?YxOr?|v;Lt|zp_HxpQxVN@>-xeuI=VF|0ah=8v~2^
z+tn6t(QE$mSfn(V>#!KZDUTB`HW-#{NLq8gtkqAI?@nY$XMo4g##_1fd250g7~jrw
zbojp2#-LwBt<pHZyYlwWo#z>j)Ht@6$}|d}&)BeMJA?2dMbp~8z)j3=wsWojoy+m~
z;!^uxRpFES@{et@KJxhL5l6p^?UomU5@u{Yt9o>qiIXwQ;n@-!E>}u#nHINkgHOZ>
z%eLobnPrw<XD4l1r@BtvKezlsfq2Xm(YHBgxmKu~MikG9%{A~6IkGC~`mHRZwY&v~
zLX;)kmuWqCaAL~al@hlti#|L$u`vDF8q=eTEo?(r*mFv=%TDRN{BiH?O{1XN2iX(0
z`8Q8;X7jRWw2Aw<`r)eIq7nheuQInTKH+_}iDCZ*|FC0$-20!%C;cyYYIE#!4%b9M
ziI_?A8J#`8+;;5yt2O8K?sE;oJlPLUB-VEA+SOHmJ>K?Tg!=Lo)l)C*;e56t*+tjq
zN^g#($dQo7E{l-b<J;`qBkM!FN;4igU05;odC`xqwF~!KslVH}>t@7uo~Zi|E=mSu
z&V17*nJl|gz2kR4+5!cgmq7<5+^21c@Kb*HZz|8F=l4xBawb|&ZBZ|<UfHaWll~((
zAvnN!olln9I<|mp(F58x#dh3}V;DGgJe$ZHxqa6*vBf*j+nFzJ33l0gcsrxB!anKG
zEEm47TYguOeH&ZG2Gg%6Gq;Q1ICWZ5pea69w>Id`^DXN<io}|O{@z(xHtov=gInv5
zrZe4OsCptLSNd_6-VfvVtkEhwpFg^_Y`*aPUXNwQyj`MCe{^n}9d+$`{L{2Uv)mLk
zoK__pzbZ{<OFxx5&)8D)#vh-j+oe@CtxhofxV^#Oy=UKowVkdU3F}|By*O5SH#4qn
zOVomwC2PZ6iwu05*Zoz#TiE~3XLo?@WxEm|<sH@vG6p7%Ydzf98(y~ezM0j|Q`FUF
zv+C=-u=myLLip7pTHHz(u9hz?b2InO;5R5awdd@vMY|qpUb3J3!}w;Af2ggp$3*Ao
zpv>#W7fX-KlUZ$%Q(ZTAk=3i2f!wz{_j3H?WIBEO@SLKus+LWs?QMiiqAyp4GFi*{
zMIJ8-Te-!eFF;=8c;K$g^9~DZqg>qd5*};F%v}HJ1VeMHqQgt~SDpg4*B*BqpT`{*
zH)VrE`Z1-r6)QA2Q|9Q+;a~n&gyZmQ5zAwpimIM6H`V%Hggp=Tmw9Qjd&$-V9(tPf
zRUI7-Por<jDt+22uT-=*=i%3#sXrA2PrW&%p%QLW^Cx4)uhwSAb-!OFyQR&(zg68Z
zqkrnPp9|l}Bt4h&Uhh@8@6_$r+l`ff>OOv-*BFz2Z=u&|F54Yf<qGm#{e4&Dxp{sO
zcr6ty;_)K)W&WWVnmx;8q8EoQ+U%&qbZtwxTJMA)7stgZQ{Kc+KKpU*zcm31W&C1l
z_RdxGov^keS##3y*L!YT^t)gF{bc9Y1B#5tePtGU-_@3qvOFyvl(#p9i}Q`>ixpDS
zw@&L_)@~c}tIc3}y4qQG*S^JZT^C;Lnw<a3ByqB}*z-%W#f7i-_n1i(E)#7&*>>^Z
z%U6G=9q=&zw)Nq>#78%5UasGK*|@mM!`>%fdY$^E+wWTAe*Uq~wVzh|e)_HUfSj~d
z_qLRNype6OdEMK-EQgn!74_*~Im65!M7)+;xu18*H?=i8Lq9Eb;}Mrx(y_z+s>J2y
zZ9RJ$1zUXN<Ig+A+|STh?Y2SikhU#%dQpSj{X-L{t?#<JXBzLT$fkqJaVr^Wr>NCs
ze0uzTQ|Y~?mCC;ue{D$l-aCWOR`OVXkeTMAPidhN?-sKxx>9NXb*Y>h+mws@z4C3(
z`F6a2<#sE0#Wfe%()`b!Ta~jzmv=6;p1<GUvnpQnp>e2xua-(_#r9uOxA$Mj60`|B
z{xsfxKF`1R=6t=Irj%;?*;<M2|8@4t35|k9&s5JCrmFv%w)*zZ+Ie#yDAro<`u(Q-
z%Y`cp?_+zG7ViCXZAb54=cb<H#)3`}*;>1&_7<gFeY|Jau3gg`COx0pzibUNlbgBT
zZNCS-6Bp@~-Yr@t{Z~Bp&h)J<727QC$|Sw7WfA^<@aI`ihn3;xHeOt*CZ1N+Av`+@
z`(4gnKAUHnzi~g`vx^hldlD2l)?OD#eXeVee0Roe=fGX8>Gkmr=J`6y<fp%Qe8WvC
zZ}+0G>Ptq2Rhuqk%xFvZ`D~aG@b1~H9Lx3F?7#B!ncV->(X?sSF|~%7hs?sht-8dw
zt-Pi)X1VgD#DwMZ%6Cn&yuAEsQ`&cp=bA+e*G(<leQ_yE@#~su9qVgXtGHcq5ApBo
zblSFJ&2bF|!+ODx^;ar_%jaHmJ{}}x<z2Pccy`SBi?Vxqeh8*Emb>I-u%3^(7|$Y7
zo4)cRZ_wwn)26Wh3rIP0_T+<bjUVYYOV@0@#;o@5SO0#-i@(<eaHZy@KwDR-c`2aw
zn6Uv-4KpJPQ{;x(-Kdj&Cfj-UJ=c82x@%D_qg1hh0Q-rvC+{@LEL;Bn<59aC)A#IC
z4VtoIY1J+^zXXZ0|Cg5vX?88K(|h%Hx^JG+Uv=~Km;NVJ|F8W0uJrfo?^TnF`me2e
z`&D`GevgUs9(}9&|2>pr^(A$Vg^#_f6kVi>9;&|j625Qw`<Gh_f4)CmANPOajGMZX
zyEV9tRm;>v5}z3@?sr+(@F;2NUM2puAM`(*_FlI2%-pmU(ULna`cyqWT_bLsE@ir1
z{HgKt>R)mnbPG&>tZBR#^U-&I|CXq@>nF9>O})SB5P$fI5)JvnO|6rs{ydr5`SrAG
ztM>c_wv!qIUM`-sqbpx*`=yOyzK+*Ij;#>CHCx>EnBV$J!=;ybCW`Cq60$Yrn0joF
zwpmFt@6vl~JI}PmBx^00{6zFlbXA|W{|2Lt8#*IDuKG}9Xw{wFwQy?)JJ<CfvDR$=
z&$IeY)V?^IKHFr~PJy5$xrd`Nq@(&}w0C!NnwXzv<6C;+Ml17XjlLP%jjm4&JhkRP
z`1Bn01(!}-KPY=ecJiH4KeOC#ZQu5JniOAcw>~Sy6!|`3hqqy9erX-kLy?tbleVat
ztng}>c<z5*iQn#@OJXDres0Mwv{6V^`PljS(G|7nryeOk$w)nMx#Y6dM3aTH=P&#=
zt7=!~&lhQ&0n9O6Rx7l3zEYYcdsD|dic{>9@0<O5mBh;1)gm3Tcg7~&+4>>X`Tf5c
z4nJcnZ3QeWn_U~jZB|U&d?9hV<JRV1^Xyj(&rNbO*yk7WUOTUj|K3*1Gi+=sURqqg
z7FMSoy}ZUodH>Sb3Bi?3OX|v7Pt_mZ{wwTU?2J2KTX<Zi+dt4K-1;rhVGql0Z+SL(
z)11(f&5z_cFKM<(Oy6Jnf}i1`<d(ZPw&l5qNme`({vIt8m|qca$GTU-*P-W1LEGyZ
zcWGXmcME0(Trkv|UTPO;cutD<+BFuF#D$F4ou)?})q6YHd^vOA-kljeb1vv6PrbH#
zdv8azSK#u@f&+nfIa~rJf7X2J#$nsx^Wkr^b8nbQSJ!hv!^5&J{>&3zN^a_W^i?!Y
z`$NF3ix<8Jr(b_Q=dY#0WHU{d8Nwddg8vjQ(5b#8zF>ul!l%7Xv-0>Hm!02o@=jD-
z^fsQ2f)9cUU*9bZTejk(rir)G4$U4%1L35M^T}@+=QmBaV~(59KaXACW2tt9r+t(7
zvRg8H5}pX`nQ&P7LzI~4iK#CXBs)Y77EZQ_*AANPeE3?=Vo9;{Tbeg%SyyV8z0#AP
zwVEOI?tO+Ss)nZuQY2S7nwYGUFmzA<?51q@`po6YzY;8SG^<bTG2A7!|CD|2Yhz!j
zhp9n|d$*^E{}ERUDh)ldcWQ~QLCo1Gw(sj$Of8v@si=BrJ)1Q7H`CHZeMP0omAsoG
z4#wW>^WHdh@8SNNdn&J4pV*|k@$I3LLcdy{?`m3pe3#JUnTG_w^8OV#ca6PQjW_c2
z>IK%z74=wL)>UlF`|>3CP`~7i74xLKzt_olORVE9*{t?*P4jZj`qT~4;io>Y_+!56
z=OdOY9`$^yRxDthF6LeQm(zK=#@FRZ?Z+4-{58M+*YlHI_Id7UmR0V1v-8-D?oXR|
z?$?*QeoIs8=R~vK_OcFAS@TM5Mnlb1cAd9>nj2a_%d_e%d0+OPi8<$%`p;Zm%dU*{
zpINh6`1Q*!DV=q{nXy;Vx={At0u|XKkC~K@Fh^P^ndFDQUwG+tPy6$dS|R?v_sdVN
zE=nwqo9hu*-Q_Q!P$;x}`b+t#{{@SFKfix`VtjqfnrCnF*31pt9l9?1(dM-pdp)xP
zm7f+Kci9-HvNdVtRj#ymHk->_4v6pi=aJnVx^3s)nz_F;Ri`$-<NsXqhK2FnjdMJO
zZ|vOXpJF%ty6NNk(5=hQN|nth*FH4=&f9G^9rtgYtXlVa?OpByuPO!s5tW)_muxO=
zKltpK*xrRTi&pX)_gOwXw!Qe;wXM&WeG`?HV=}$B{8gW2!TRSJiD#5cZObo}aMioD
ze)a5LJ^5O2Lim(R4AY)nt@_H=74dcD*StBhk6$k3;b+~eclJ-y->HkGU3e~Td6B(O
zMJz0?X{`_2%-xHxnO(k~$G)3$<2z1am5`06<{CP>Gpe4PyVU%zR^jv@-zdGq&W}ur
zUTvLS@UcAXUBut7>6u{}p}Qr6nuGRV)vI59rC__$TeVY)Cx7hybR=bU{I=NR3ETfK
z)4BM3U5F2x(e3lIw{5chx8-m4@5EZc3?GmFBA4?Y9;g{;9-eJvIrD2$;=N5DU%xfJ
zId^+}<{zf3A+xHipLwaLSls+!+vRagb!)Wk%aSe+vE+HvbasDy!*qK7u0qy=zJMju
z?p|}z`lQBf7nrf2`j*+PNGZ4HmICiTzP$J8$g_Vs{0i1C$Gd|6ANypvM=IAcBjzXb
z<=Tfwn--O(>`e;K>D}~ub%y#$`N}zAi|5@s@lq~)`M=bM7OgMW)g73<dGW7>X@Trt
z55GEg=C)nz&d*a?KW+IrZS^n4)4#R0*!P%Tn-P24xU<y%<BA2jo+m3$Pnom2o2^?w
zWveIqx~mU<D*ZXx-4>F1ENq7Po+_vGfQ6@wtzK)~InI6M$X2ryU&W=zz6iKyoqlq!
zi)rbZfL#yoT)yRIbnj}U`Q*r_g~vC(ndEvPXu)H(E2S)F?7D45rj(VIoPE4%N6W0I
zoOO(jS7h$r`h8II)SXbf<8>=O>@=^KnRUSU+yTyi^Pjgr_1c`JeL(i!6vZENcIs9;
zRX&{Z>{ZTRvoBq)-W&%Oe>Hxqz4YqIZnNjVwDwr}tP~AhdvL+q)mJ+m)+{?E@bbKB
zLJZ&6i~l25^?aJV>gkqcb?=(8dZxa<Qu{B}%Rax<{#Pjb-=^ZYtlb^%iEnQSrD&GM
z|IKUNwmd(X_ug{eE4C}rudQA6OTA9z+V-+ALBkt&Tf_v9&+oaJ`fAmocUzv@3$D&J
zlyrL6^wjP9yGdOOjLrChiqGV0UZ`|@x#6<dDzCEWne|!he#?KQu`HjG*RR$0#7a&`
zW2d*eJCkT7pUYZ{jkA9>?%i_ys@|4+&0m$XU#oBHbbc3`GVQ7M#>q{exp#A1z89iB
z;rKd5@ndSa{#)3^;*?iTd-49<Z`&3FQ=R%Br}LZUE7oIaG8q%qWHL50KyEU<mD-u7
zV{UYR{S5gTW|4au53;c3FrO=1c&1>%H4pt;34i}LMy}D4yF17Ec;2Q%)y$c@x=Zh0
zd1%jG8}M_!9m|RTS06sT-#&jnpZ#Ac{zA`tw+~;AuB*M0<*6q-Pp^K?luvP*e<#()
zeNxwax+O^M$^FKAH~+|l{d!<r^FF7Iv1)p~?)lGxsm48ri`W1BD}C^hF>^1Uo&J5%
zi^4WP?xj~$sL6brT6lb-xlMXw!OauT=W$-se8ih5$d>u|yui=uW65EU&Pz7&Yv`Z)
z;<n?A_G7o5_h+4%di=yMry99R+y6hm{C+rHJ*d=bL0RafHQCFjB`Cj^dRp{zPv@7*
z>!0WFEd4OmQo1(jimU1K#hnfQ0YCLyPx$Z8y85_(!;A+;l1}<s4xc%@CHd#dNEjCu
zhFx!N%zg1_N{=76#AfBk^CfS6J>Z^uby2RBjH7N`<Ae*(m3kYs#J#=z-+21;+&|MP
z<0r$>Yt^n^KFRppynX@yvxfsjuV!C6w<){7qb6=bPsT^v8PDtwxqX-#cv&zvFN{C`
ziH5+$1rL2Eu-JJYiQHJXVJ2@k|1-nv-E+4(+iD2fbWdvPm_PYCceAb<SLdHA`V2Sk
zs3}iRDT?RX(9QeMG5(ZX%3JoIdiQF!xH+9j&vx=sHx#Vvmd?KX+sS!iV^fxex7ezH
zigo9i8)gNq(|E}9P(^~fm;0cp)UVW4GAFEjH)cMc`So3=v`K04{F%pcqBf<6v7G(e
zIyL)VMQo$0N>b>hmfW>pgeL!Ex$Vxq;l+;9<*SliD{pVe{;4T_EIt3_ojaRU4s5)r
z{%Pvu->-GQb6f5;i_K$7$hO+s&?m1{XWx)~%B!()@+?kY-g7tZi0xPIoV29prp4D8
zzMPwu{affKqxih;xwGdpwlz0Y=S`lx!)fImleaVOl!PoR=Cqc2ds?%LU!~f0rlz?@
z|8tJ~Lo1{Of}FLkKU6-seh$|iktu8948BivHIQ8!+4ICrA?owid!2%d^Nw|Gn4KMR
zv9gdOi1+!Dg$cQ=6EA14Ui8SMHBoc>A<Lq>igR8lK0mQx$A;dLH@|qwW;gCRWwSUv
zRy*zNodb7Mb`+R%eKvVERqj&6eXk#}#!dTT{a#&L{EJIl>c+COVc-5<IJ3iVdD#(>
z^J_)5N?jhrK4xCR^<HMz9`Q<*C!S`F`Ez^szg~1BK_%ZVxb}|FZ<Pqq_*qr!til4B
zCwhNQnsd+M&<(}Jzc<UOuI+d==}n!%o>>*&Ty9OdD-cz+>g>Y(YBEp0oj#~{QJ~q$
zn~CAtGwDk&3?Dmhm*^G|nI#$-p6OEOnzT?^V3zN4)kXKuIbQ7R{qxy6wJzQEoYXs~
zt&<a-Jlp4-jpAmUE_rOn6U)q|3BgO>{{AO9aZdBx>-=F~7N30(W?dBUT_CV3?DeZ#
z*Y9K|mjB3>H?Z-3CaCqyA&NnI*3svuey_i;#{d79V{p2~lMB3FZGN&b+h(^+c^dTW
zR0C(>jPT!+Z|QSh`nKcb;eO9>{>Eq@qub6CUR~(ABxCfsU4CX}w7u)RZ>4IUeqXiT
z{Va^$vb)(VUF?FjZI9x)Uq1pA--&A-cgy9p+%@BL=ry&O0v376$MWYLUcN-%`$Ov6
zkGGs7kA>`-#_h;;nQ_+l-sKE`KeXO5j>%uZp^%iYl0j>mTYlzABZV7pXWvRKI>>$X
ze6>f|mB8H(?k$Yx6yaGEHvf0#+%h$Xn>D+SX-!$J6Yncx_*!SC2J?5%V!LN$|F?b7
zN<J3K{m(q-+Q(mf_a(Mm^!fX83(q+{&%^GQKULhhJ*z)&>C=xFrZ0Oc_|^K-CpZ4f
zTlTybd$&C1&Lz?2xD}_f+<313u=}0k(cGxN_WJt2#v$pAjA7s3cSVO<SwuhJySYz+
z^;bfH!J5>2yWG}2PxM>gdi%$nE6**o6APmAYzy)Zoi$T!nYHCt=5>>Y6}A5yTzrDp
zC9XA%RkunI+R8iQ(FK3s$z>PvHit<c@wUHNdX{JAF99*@t7lKTUJ4b>eJaBBY4dZz
zm!9QuL2Qq9U-_4<p0S8OPE^Fd_3s7u6TjCwBwxK&Sm`}&;TE$a4Lk3o9C*EveZH~c
zYVK*~pG0<;MeJ$c;o5gqIoW>g{?(dhkMlQtKFfLfeyDK7?3HQ1-C_g{FRy-($9Un;
zD(-|AY$c}0Z?SHcbIxu1sCRVhTKh9LdJleXJ8@S0Y3}}TO`Y(f)a{A83m+NX%~z1N
z(%V?_{zUg33&qa?uE+9A-n$rj>2lQaBw8-B`FtZc^IeGJjW;Z_U2I(chM02vl2|m$
zX}v;K9gE<zc?y=9a<|oNTTWaw-m`-{vrofLMasF<`Kq7EcY#Gm!q>hob@`@v@yEWc
zuhq*I&1@G~`!@NE&Bi@;O&&*_GWNyl#=TTqru|&U^I~4)FBY8}tDP6>|GqY9`i$u6
zj0ar%j^0<dO_yG~E-~eOd1H9oQlaZDhuU;Id(AZtIX;-W#k*QUM{$yM$<b@8W%qL3
zUAFXWp?~(mt3KN0FZaEXt~NfZ=cVnXDY;E`e$eR|7pn}XZ8M18m_0xJw$aU{hL3XN
zm#`e$eoE7OdF{jZnqkJy`D<?eQ}xdK!W0<gA^j@Idw*r*)vUP9X-_XN^P5}Qte$^S
z=Zkdymt9+T$m{KSz4hOYlUJ8#uM+vO?PM#vSZ(3QTea`x3`<|{(4Nk{d8xdPU0sr^
zd8Fd*85J>Ew;rthk-&GYZ1?~5v+ll@`knIT%HBJNZ+&z!e!X7!{>{yomshW8K6$u9
zw#2N8DLi8K)Yl=i?vyI+DE(U+tFP{rFUvDe=|gFe=UP|YMn+NP-~6@GVy~WhE-ljh
zIpN!@9jp14PYc?<;@8#fDj^#!*L;raDTv9=d9~@z%(rTPv?Zt88yxfsm}a=~r;oCW
zVez^dr+(d?`O(hTC}X?Ivi-VA9JPAu4%a8O?3Ug&ajm$^vu$=&KVBU<Ja^re@0S>-
z@2=gmdwYib<EQ%hOIjXQKFC~}7o2czdW`ne{UM5e&eyjr`{(4V@T&CT((-$9TxBI?
z;whq&w_G|>e*B?If$t$}ReP3oyQFtmJxTfgHM%0*XQ%c52fNa3;*P!e9sT3uh3V_6
z-LAZk_@48nApLIf;((g<4R)KDvZBxXE&R)*w)NVQ!~CCbmd!HWIJ>@H@rS(L!c#lG
z?rr%M=Txh|@ZbMw0YUoaXRGG@c3L;Xu0{2j)vs&PQ|tdd&M%YH@xs#jvLL4QWoTw@
zgwpzow9bEGCh+dF_80bP?tJ$T9CB@lX(-{m!>YMwTIp5MlmGP(DW-dLJ+gb0U3BcA
z*p@{nD$kocXZ@@W?Dnb8`Z(d=(@(!jfBvle`NdMl(dGG>PkVF&B~qtPo}_eZht2x>
zc}q*gHEVUm*Z8Y*KZ}`{qmf_wYi;T5r)Mt5pWm+<$GmX5JBOvS%Lhvz&X+#~6r+!o
z9QZvapu)p+R;$)4OT}lU?E-ep_U}1Z)l%ew-nMc`{;>`BwRd@ONpJ1)mNU1NTOxGl
z<SGhVNj^-C@LS*#vSz-k^rM)W@9b<>f4aS7>nqc{*N?DfG8Jul{%K13+NW!~eHzm(
zG#2lWd%td7u1v*|eG~6=t~7nRmT?i|mpx*!3EH0@n!aXsY%pAG7cggjZ$>oBW8?j&
zzdu^4(8@A9Cnjuh-o2^2IhTklxL?^GF>#tq@un8#jrEy*PTzj4^Ip)Xt2Jk;%Hf5~
zGLmxhee*l^T5vhM;OM>bt%ONZd8X3rYbpCY6SUdW($>})EEezIrtzz}g~h9G!|uLb
zNv-q8R!rApE1YiUH9`4!aNm}-{bFHv-#p^KzQ9HzR-<(3=G32dwcEINxw*A6R9%{I
zXGzcp=Pt3MVO6@>HtYKR)pKp13D`8+i^tvCRb0$+Yvq}f7x#UfQ}E8{=*L$snQP6I
zz2r33ef~8s>YUJ-ecY7_ljPQ2cV-oEG_yB9dFk)k@9`&%WKQotdFapWSu>|P`5dWL
z_Vb#?l9sr6qbhqq;IzxL&ulg3+wcBdZCi@ok^>(CW%m@V3XS8sIzw94{l?FVg$s>E
zJsDp)T1W=BWhMH)37XBg@@ep*<)yw{by*X_?#zCADt}eXPtMd+yN<-Q*V(;DQh#5-
za&C!vNmYEsx~jFSEmZYVO<$PRuVXg3sdkY0S{IA=)&0K{R=!mG&e?fhbESj=$CBAk
z(;KHu>y(IAxhk~D_h}H5P?L?CSa`e4&%$GOUb?M%@J9c9w}_UCnC8#PA%_|x|NSU6
z`f+~xrJA2dyI-U(o35pwG5vs_an6LjYr|elJ@t4)kfoqv-)YW{^}0;6C&cu3D_m=8
zzop~(<0*I7ygQnwri#9r7b(8f>Xj9TB%jjEwdS9#8C$uJUGp-3uKG*9`-I#T%jvf{
zl35&U&aRHzCB33>L*L49Gr#INKX=YL79m)(Hls|XI$}mv7n`YG4r}9<GZ#OdZMd7y
z?RO@!@?j~jsPBi(X%7@HtJ+;!ops`l#P|0+oUPAV*D@L|wSDt!)j!$38>fz|mfv1`
z<=U=~6)THZ27MRZIz90mzuxD<*yEyQ>ic3coc;z?MYS8O?7MsRal`bV2fpU5V7s@_
zaeB%uy}07HPttCRs2pvY&-yt<D*RSU&#T5`v9CEUdY;|zWnH(}I-Z@e_obp<uhl$1
zRo8u?_VumH#0BNIxrx8Nyie>-{*C8smqc=P-5%FH?wYF<x6AdUo>ugf=_QZPbzRMl
z(|K^oS!AF4<+QCU{IeaB^;1Kb?^jDsyq}S__gYbLfXTGe+D<GI7hV_DCYvmJIZyX&
zke4{)m#Scf<=^%`i{Z#q_qgw%dvxj9qq;|R%Tnb0R$SWN^D^vp)68;X-G`gbGP(Y^
z71md+v|{4P8?Swt`=*)7emJ=A^z6X54gqf!v)n%F9rr%1V|$*vV&~dVa=k`#PMlkl
zEq{E^?v+ndY_&WBS2uCoSo0<&b2)ce(XInOXBKY>uzVOeRdU6bWjks^j$T<O8~Qi@
z+@3?nZ9jL(e>js})-X%zd-)eu+cOFap6e9NsV;qc%Y4(?|2OBS_V5<24v@R0R(;}U
zre@nt?)|Z@Y>V&gJi=!F`_rwr@7Y)@;_rxjI4n4^a`lZF^Ahg))IRSOnrk+#BgN<C
z?io1;yl)$X&U!bmz24;a`>b#6$t`i~`b8US?s)GOw2OYW&&cuInl*RdovoJncX4)o
z{NA72PREx1zw+!?=h=#~LRkjO1)tX?NZT8Rn4Ube;kV2|#(66by*rX2>uz!C`5&)O
zcV-{1E^}_Yt`p$5{Mr4ab*<OG$+wrE%M?t^QZ$!o+h8pbd_CLa`uZc~UY`X@*A*Q5
z#+H3y_IIVrv0ndXm4BWseQu}nlCbM1<n2ShweI}ic0iUpZ>GVlwbG(b`nzs)6g$r|
z*O_K#(zg2VmxN~p7Y>|S)@bdL(OQ&q@>F``!{0h}pS>SH+3@S>vkPk#=kuKYd(0s0
z*A5BW#x0J;rMq+gwiNC8_v^g0+r#D$cOC6AR1ds<`ecu|;uc+T{hxoA*E0&ZDZaz3
z%}j_|7h`B?VS!ScNuAD<Hve#M{SEmKZ8G~B541D+Hr|W+5%%T5)HQReGXDP$I9Jft
zFRFUF@3Mu%1f3;I+YVRmfAMs_tG?F%%iaxt;&t2iPoFP8jeq_lq3cif#>)A1_hq+i
zxs$n7ZThtK$MfZ<e)7{4o?5iArftgp_TM~DLjK*n`6pwkPD()Ci@!7W=X)J|sk)%v
z_RqGA^<htq);_Iyuvzi$eaixQZr07m?bEn?8;@<y`S)eVzbWPEe|`mATT>wXsXUHf
zQm&-Z=UDPb^P2M}hmsFAnh51Bxhu4`+~<v7N;IR>gI|AVclq3KvH5mliS>3VZtGib
zgOew<c$VD$VzkTijD}UP%lz_9Ka5Ve3o3edTzFC{erHO>q`$$5ha7&oEb;LF@YH;v
zu0!L;Rj)kWZvA}2<g=N~t?iEr@+YeKSSpD3u<rMK-_5KsMOk*Lvc_{^U(Q6fb&vYA
z|J2xq7hgDZ@m}@~#(i~5cBbrH-(?ZJ{cY>@LuY>*bRIu6V^4ot(bWU{=SW+8jZxTk
z=;S=^4+hVfuBQ66CUvFkV^t0@UKN|*5}KkE?K%D5$GgQ9T0c5hN9?;?dE({9okbOK
z;fl4vekcC=z3Dq&(0t+o>q%AK!$;mv4fZJ)NtluRV$I}B4~5q+PQIOX)^2aLpJk>(
z`;+?HzJZTrU#NeObotrIykG9Lm`^`P`~Jt(zA_QVa$4;!U46NGv&T36^LG@*_r71~
zchc!<d1CAImHU`CTM19IX3pk`JN07cp-)E1B|F!CT9f&Njc;q%^hueMC#?*c$Tjn<
z@fpukagnAmIi2StG&{6eF6j&E2i#dvoh)b+eBzS}W6;JR&5~`RH<umq<`KHPp{GSV
zbY7i)j`Cg>Vf{k?D?h!Hjy~m|f2rxUR%TAR@sx|&F3PR-d;H~C8yfj%7VkOzb$8Fn
zzXhu~jV}u@&nr}T$Xw@M==??LOt7Iv$Hb{2Tx>Tr{rq*x*0kPe+On2Sr-G@Sd9&fv
zeanNaBb+;?`Yrb|P^<~)Ej~Z{-0Y}>+TTJ>yy2KHGh=(!i>%Z&4Yh`0Q`E&O<NVL_
ztoB`*=c3eI`sI0-zvQ+KuYV5ByxAoSKmN%PcyOeAa?YZ|>v=2nDs^4brd}v|m}hx@
z(;TBIQOTzm9iBTR1gqc9VmoKN^MvvvC%cc<M}PY&7zkfcd#SSe`HZ~jGtA0D1@}!m
z@knaQ+7q2Km47=ahD=@5z2??W&7L(LuO=_Kcw?%;g$K5O+m11_o$YuWw|Q;dZ@*Q>
znymYq@^-yHF|T0l;Z@cX4;8i<Ne34Rx8z1$WW8`w@#d<AKFPJKgzrBIIC0VMxU|14
zxBHD>40W^Xr<8@H7CLPE_lA{i&62#>oNX=Je_JQ>luL>Hkh6BO_%}V~f^$-eR{p{1
z8@zYfKGO2|_xQMM$E<icv8hX^xo&7W+nDrq<|6yLxjO%T@9eQZEObpI<h1y-Lo&-n
z`A(e6SjF|EXY<@!>p9t%KQ3FNw`cbb&La8wXX=DiysQqLozxj`m%b>swqyP`tH(!<
zy!o2`MPj>q^|p&nVT<f0H}{y>gmbI(>-VTGS$(~*RWzJu{e!F<7SCUu`Tm>n=iG!-
zh6)lZwR!wY6JB@AYKpJ?vGvZLgy3H$Q(W%4aP*s&*`1atXRLa+FvPGex;!E9vDLW=
z2FB8R?X?Y3cOJd_SLtK;s)+&0&is~v%X}L0HnOi;wLO!!e6^+L1qZEXjxE`L8oz%l
z^L~B7vu#;<aMQ*%HJ0O(+HNhokX6#Y_H@GpW~0)_%WHGrB{W?9n}5X3mc5$2Tq`wG
z+B+=b^p;Z29m^Bz;)1{K$a`fqt?J7B7PH&MZao{n#ndjneT)0mmOp2wOQpN+n<(kF
z)c0A=${o{qF8e$#n~--o?0)<fP0!%JH!?ftt$HYbG*l~Fdbj?GbbsY}+vkhUN<O<V
zy=U2{1fi+kv$ZZJukYJ8BX3DpOyIf48+n#))nu9RIF!j?Pmk>T9*)fD8TW5!#Wd<#
z?W~dMc4t#rB6nl^RI_c>6C<x*Ikr{kWJ2~TtNVsq?tDGB)Y8HJqx>1otviel-?H2+
z7n|Gm@uu+7J?&9@-o5wTV88QN(Pr15Q%hF<y0?42TuYr@g-!Z|wPjV2Jx302+;6wi
zJlFG5d?nBI{>-hn-UU0VcQu{$+whRJCt-KV)$Mn?R<vC>UEX$H`B2mPv;UTN`0V1G
z_&D2OR^3^_qq8F>|28T4VX0*KTGB-(>hy1!iF-S@9xv-Hd6>84`f*X0=<1%C?9-M<
z3vfo4eC9o?b@$x)l!}P&GSU2(m_F2H+&u57T5@A;dz<s-S#tIAU*_bdm$Gt2@8`Bz
zy1?g-zXY@U`va4i7AkzOIJLA`!eZ-<h1%QV-ju9=dTe!8r~16zxhH2D-7dD!`6Rw~
z%c3b$5~DvXj8QXAxO>fghVAXG%ibQFsMjl(|64`XU|*`A<hnWP|Nb3WmG)w)>Fb5x
zg*|7@;jNz)$g_3J-rA30VP6ujeXHKQ>~7EVUj>W_8Bf<{E&qO9x1>z=;yvfNd7A{M
znc1wkH0S#MT@9OceygqB>7961^kjDQ=R#{&`QJI&c6n8ouhqExZBi>!Z%?0fL;Lfx
z)4gl|UVIqyreNK@hA-wW%|3d6PI6C+DA~H{q{mGCGkw!-g|@9Weg6Mq^2Oi%a<5%&
zr?t#96i7DXu)Od?;F?78+P5jE-`oFXpV0GmDVCOi2~qu8V{?=hvTq};`=6Ncy!)*E
zhy9rP#Hxq`iQNo48D0wch%b?N-?QFp;-CLYi(FnMW|yf~UsY|6U7s8I$oq+Vko8~F
zm!3!EZ7VzWXI1?z{`JwO_SYt#(=v06FIRTY_m4Vzu<d2d-@luK)SjGvS#flCU2pl+
z`MgJ7uM%0Oug-0DZrd!kCuemkS*}}MKW2St*6Y*q&*S^v7;J7Z)n{F@)M<i*a>^l-
z)6!xctuJp1^ZxSM{PJe(>r-{EfBr?a#I=?MYMuM+xXy^*NOV&B6RE&b1G6(T9_^M1
zf99rkr|ntu>DyY$TaGPUpQ(Rz&6l_v7da$nyi5J&=eS$N@QabYK-s1V9|A&_nM<n{
zE@27Wd6nCEYVeF+Gi$B2nntF^3)&)9T3()+mgr>i)zHZ(+ucK@2-N*P>~d|<{dody
z?Ho64+PputDlUAgc(p=oWk<5pq1UobiuvAk^1?>^E6Wm{+U6Z!_0YNHT9sm7<I`^(
zih<o}(I?_E)7DRW$x_7^HRog!Z_O7=xfhWKlpRm)>fdhiTB}e^glmUm^M-}2{ihZ(
z9O-@9=d<=<MM_42E`#g34NrdxKJt)n$q-Gbtx10NW@W9;MB5^tiN#C4-a998UDHv9
zOMCkJ789lbhp3bud)=(C1olQ<!<q9e*R<po_LvGLN*xNj)G48RP(u1|$Dz3vQVGWx
zre670#3age+9mDm!%sr2+p><YVJzHy|KMWXP5%}sD9yhAQ$SPyjEWw2R?Mp7v!Y)s
zHXTzf%w4;D%9=Yr<7OLqzH(AmpXeX|#wB`j(=x6HQ*2z^kGD(mbhO80_1#!*^Qfic
zjQsYpmLFp0d5mVQV!FR>kzn?zj#U>gg-T7>ntE%GJa7Jr!ieWz?|9gqzUs47GjEd4
zr^}T=o8zYa5cm1=^=9MDy5E%sKfYg&ar$}k{6X!rsv;HkySr>B{Rx~rUFXo_4MDpt
z1LWRLpB!wks#naB=S0JAR{^WsWg36Jm~!Qwn)+qsdS%~&nOBP@u)m%6CNZ%@mZf$>
z^Mb|6?|r^axZo>Tmi<^uS-#h^b<&~P^VcT6&pmy8a>2E_ZyiZIPjr_x91QYzb&8vQ
zVPovtrJEd9p15dv=E%<_Zm*xc`X=3b<IwRyNtx@juP@uSW6DmM30Hac@+8-7=sTXR
zxsJ1?X5Wbxkpok<$*``qYl`5TxbK{vr?={cjGqT;FL+Orc$RQ<S*DWp!--FvCmTzB
z<@Pi<zpOJkj-mEpVR++wvz1cQ_M6^QvY2Z0Y1;nF5_zGOj{Q$s_2R!5NiQ_BY`JiC
zX=R0t`u(?cMc0FKWPeSH3^|g0z1<<)C1z)SRQ7bGS65@tEQl_f=<B%5H>>b*z-`9U
zx(rQsq9qyo(~jLzxbs?TZ<<R(wd~W4-xLFnuDkZ;<|m!CAK%)2zIn~=_O0~j-#OX;
zx993sKHaC@z4Rsft0u8DjtjvemYsVyeVfHyQx|<)_w>G*D<w~PXY#ktkYAQ0k^gY}
zYK?>`TGIm_F7sdiR3zcHw!&Wb&;>cyU9CPg^%k-mJ9?!&PUGgY1jXIu!3R|yZ_Jam
zEl)3>p1$zEZ|lA*t1FwWcm9jMr><^wr}M<eiC3lw=GnzfU#{|L&FTHCch$U33cr7O
zX~?s6chxLkO1a*MexS}b*QzPEzANHN3b%9lBMrWJE1XW>kNvLv=xK7DWyQVO%71qy
z9$zhXLcQ$q<7?ILw*Ec+Ft2!(`;19PdUq&pF?{&tugs+@S`V8q*|^wVeY?#xto81d
zW}e-XO9K8sn`m`b^4UHug;&*2`v13P2yY58%?z)+zcN(Gf7`lHzkl%#J55CIMex5}
znpt^!Ptl~CvbM~xByIfEpa0#Xv-G$od*A+lANcRDpZyC<F>OvvF>PdKhBEnkHEL&{
z_HMzme?@;WepT+aR?skbAZ}r7F!N!8iR~$wAOD3XZ!LJ4<~>VN|NFzZL>05?QGRRR
zT;A{ZYsn}3TCNHIS6@DE?_XbM{r6k7%~93LeE<CKe}1gFIw{6#@7ntP8maHEZ)Tpp
z-*Rcl33+C(pP#<_-T&GiG$H+^jk169cbm}Xt>(X`94}j{IpO$ck@cCSl}{?&CC)KU
z7LTl(Wd3pGyLd5n|LW<P6=8e+J$}3X_?GP4l)oRZH_M*d(C6K<r84c*!<lyCFV~-E
z*%|P%WLl<n|BvSV;j?aazG+?Q(Hh=eG}FyWUTH3`r%L!y-c;RnmY)|*EK#|Bsp$8~
z`}<vY^_r~BSfrt~Ud8dqN0D1f`o#x*Gr!OOqG=<tc5zj{i+5morCLeqn=p~8?wVVw
zD^A3?Ju_0VyDc>R$U@)y&oA)FJky+>Ulh55f7*)!_Z_z!Gg)es&gJ=V)gRNByJOBS
z7jlojVYyv`o4+93V`l9i>uAp<%-tH?>8;5EYutV*i13%h-*NjnW6yluNm~0GV|cwv
z7-SyLTWKd2xB2>e2c~z?8zz6)7Qnz?tTBc6)IW}29``@)Vfwb!QTfJl0|)&t$pwe~
z5B45ZG<dp-QR<+~zNN?P8Lo3Yvu^$!d`Pi&&oBMrkI5o=E2kIebch(Mwjarh@Mr#_
ze=c)jP>@*2u6#y!iz${(PKCnDWIp_Jzx-5reURDh^uYhh{hy2dEM3I5Fg>67&ho0_
z-}seou^Y}lkeqWV<oB-suViw9MO=cCp7gx*`tx<g{hwCb82L@&zFfNP{?Ny3iuj+~
z`UmbZUQIXXHhx%Qw&94tv0Kj^4^G!MJil;L0kiz4BNgXjLu}`+6<mCO?a3e&hf0T7
z{<Tx4@Gx%OkoW!Ym-X|suQdm1UD=ss({Cw$M(fBkW~<*UF-h`n?>8;44-x<OZMXH*
z!l@?L&$;Of^q#rop?3DmU!S$&ve909A7*bd{lWU_<)#LQQd?<Jy(c%iCB7aPWl~ED
zf36>^D={<C`9ANuFFJiqTDR0IqWYa;Qu!Lgo;}!n#VNCq!FHkpm-(%wX9TWFXK<NC
z$<GU2-BdO23P)4@vj~|BUaL;+J~`QIcF%#ml8RDn^~=4QR|uTEVAnb+DEXS#8@tO3
zI}<l9yZX2`s^vpu`++TiE6%1?URt~2*9_UhZTIv}rk}ogvnjW#E{MHq$}Bl6?bIXt
zpUt`vaQ)sFx%G+dAxmvr3%@xWyQyaEzASy!%T)g9UB7h}C*5$JZlYE3)N*UrIpvR+
za!xpf`>b2?=#ikp(Y;x_*ZL~YbKg02)6B~po8;$uCU_oPqA)e^%<*n{wQugFLK?;&
zcoa&*KTLntc|m&Zr!6b?teDNSpl6o%>Ky?J+Gz!UQViH-4j%foz~0?>+oy$rwyrDu
z(k@CaSQPL$E$e)IY(~QRICgU}-D^QfYya}kQ#_IoqZ0V<X>Zi@%-8k4Q3rmXV@!W5
zS^H8;!t~VYW2cWy^t=A-W?Q#JVCK~nd(}53FE(czGnUkBKjYGLuXowpxJij37PqBT
z#g4sNz1a8LgdNA7Hea}!Z~aBMOJ`@+f*+dd+`Ucp#mQ%v>+UkRo%Z>D@RO4L`BM!H
zYA^j;X>;AcZGYHC)lFj4zGvttr_8QU`H;>p$E}@VWEeARx>I2Lj$WfummVfjhS()X
zBy;P2tlH#MqdBjEv%5LX>u9@m-pVUGrluV*Dcrb`RVk<cUh4|&#Ls$5xV?_IEGWLg
z9O$Sk)|0cgG@;)1!jX^9wrQuDmHgU}al!e3OU_%CYwSh0CJ7{Ri_E$2#AJ2!X!4)n
zLrcm%nPQ@f9A2lC92V+N%6K%TyGmaASElZl{TGhCShDVL(6@E#-tV)$x-Rz9k|&vd
z?-WWl_63R6W}0Q@C*Qh$ZCi=|8r%PirffWQ?2eS%lO8AAvbM)DM!wJXGHwyeKfAE|
zmrBqV-4FMIdFHil4Z9pUE7I+dxJmre+m{}d%B<`x?!KuI+HgiH;PHvAJ+J%L9m}Zw
z+Lz~1np~`CnzTb<#=(4l!FFq}Q!afz*WRtXw{Vu*?qw@NJoY-5HjAzP`1N}2)|m_4
z%5?7jDs8#o5mEj9Fr(M;$kK+B(|!a9Mc#Bx(2smnA+)>i?@#0RtBcImU;e%)-{9`o
zj?6RVMkz|KG_p5re-_oS<;-W-_20OfpE@1+<)$9C$NZz?<rG^Tfkzu2Fmjx9bdE`i
zvrTdly2kiOtCV-?zJ<@1|L*?nH+}uJorML*GN1puoVus?Se4L=mVa%#?(jXSdhg^r
zk=^4|VT}KTBfL-6PAJa0BkJP*BC)XZ{D)Yc6+S^dp8dHq(|hhM7w0+JZsOUs>qt%8
z`XIj7KUiz#t}TsV$-iI}#=g*W`O1ZNS#^S)wMDIEyL3!M3K>*|a{Qexb+=EmOgy}-
zZrQPQ(;m#;wQpJz_t8VbY-!3T?eFaRwLCj8SLQ?D)8yPJpP!-mKfHIQ?C_s)u>73*
zqpRx|Hm%4OdgdcpoVFlscIS*Uv&H<X{BuLYpRWGLwSs%gj^>ZoRo>biez?V9(r%ZK
z3%{EEqRe>KI~-6kx)5ISGfOVSgk3a+`^eAN-pX}fwr}coNt?C%thefXuEKc_KJQep
z4!_lQV99gKx7R8gq#3>?aTs+6#imBBevxb_8+y0;Se4<HuZvI2Yms|B{l%99SMEq>
zZFlL=KNe*0hHDRt{N4!m$?v+JPUpVxqU~H~2J=V3UpnfK&o@k+<R$#i=c|=7Z{``{
z8(ZRz?Ni+RI?nX_2F>#Pzeb$LQx45;(<!Z*RWsv7rh=T6$9<!3UgsMZDvNFl`e^#w
z>Plz-RrNUsm#K%^Skx*oRDBX=yui1fgW<!z^y>{Jw--$*uUY;_>R#%T8O}jAXJ5q%
z&eb@2$D{D=vP<`t9W~g+WVvk3`}#c>m}iRenW`$MsI4+MTbuK{uW@gV?$?#=w@jx`
zu-GNESz>ZA_wz@K_DDW@^ik_f<=L{SS3jF<<^Nd7@v->))c>=@HTW#GFADwoadgUa
z5&qS8=4wV&Ys#lh=l`|uZ*OG%fA43?dovHuZO-*yP`P+#<5%6dJB5#aZ%gD<{MjY5
zIC{(Uo)vp~zD<%j9x5+%eD(KV{U0wrxxIAv<#(sIDgM*BvcU3ww9JgGz5ZI~5B+_S
zxVyVoXz7lLwHNlyz5Cbml<Gs?POlBTew=bYw$*Ca-nX%5DVxRgdjYGy$b!n%AKxb5
zdl@`O;PRs9*OemMyrsMk{5xymHP7vPZ?Rl<ebwek!TZYsRtDP6DEoTR^qFR4Myb_4
zed{endAjFKE<~5f2mFxB5)Md<_;&lW;7%KjHRjtlpI@XRp6i-!y8g!9HD{a8dVIgZ
z$$yE{{L*LB?`2C^_Rg~9e0tf%b|!0x_m$p9;%X+}k2J)u{kLTHqDB1W&##-L_<7pp
zU-8%2e>Q#Z{aY(7^)_m+JDRh$Np*Ve`8Q?>zdxz1{hE~-YZl@xed5opbv#=oC7q@_
z|J=It+}mwS`)2$L^C@~Ku&pTExl_+LM0}^(5uRn;=li8R`F9`O@u{=?OySi&-T8hU
zZ-Y+c3ml%j>}=JgV=f#U>^S_No}Kh)VJ(Mu&!;NU{pz!)|9R{GpKZVO;**#yRa2s-
z7K}^{3{j>QqOJ2)b_$;RtNo0B#dMv0jE;?r?-|QdoM%3CSbNdJK<37OdxO;{l)CTE
znRjg+)8V)~Uo&T&PqN)twc-7z_1*hc{ORAva^nB0)4%QY_t)>O`Tb5WO4n1hwsP;&
zf9f^GFF#fNeZ4uzPBS&n%W`gg{O9+g_v7a&uwBe<h@KPbwg0(X=)Wi#{!*1GW%u1~
zXJ;N|&$t>9UOMY|!@>CTMZz&G%O9luxqUtL%~#eLCQ2&*x9+rEA-=JzUeDdgG0#el
zHT`f@)rS`)b2Ks!gxNHmUMnVa?%-p#$~%^;m1Nfhq^+Nudhqnu%;R^suiu&V$9Ve(
zE*+7i8Bg!)*H`~lK6z2++$I_3Nm8f39g~>zrbt^vh^z3<X8CS+&TaDCou5kegeH0(
z^4(wiJ!txtwxe#dGc`?iM0{bN$z^WiGuJdj;mA?x`|6b{O!KrCx=gCoTcX@Q<By+|
z@Y$>@_h<ZH%Idk{f@w#pZ_L_FD~dg@&yt$9a7oDZb-QZ=g_>U-GR=M+tG)l4l7W<<
zyQ}UWm8DMaY%*7i^G;U}=w^)Gne@k-<r>4)7gy%;d#pF)i@x8*c6ECr|2qeM?)lc&
z?-r}`h3#ZI=X7vVnVIB`W@hnap1US=t=cc`<<2$3cUxe??}~*J=02VM*)QbEj13Zt
zXK$Mu;J@Eu;at(i4?T--3cSx*zGnXEHFrBCgd!Pd=<C!xJ}dP3*kZ%qJFcc4(|TsU
z?&L4qRS7>5Dhs`zzCY72HFAsD+bv<DC6kvQE7uHN{cEQ5d3HXw-4WLV8D||_<hl7k
zqOL()0OyMB+ZeVRJQ2UfHF>^_>Md3$`_4dChKorGx7(dh8K>9ODcGBB-yL0<pPa{f
zf8y^Sd3GFg)znoU6vwXi*tNml<DrzD@4cCp4WH!J+|BCD=KXp#X0iCwSn-?Drrz<k
zZxz-_el&ZVna^Wp{%~sE;t$&%O}rP`#;Tm~uJC5Fi{UewY)heYt0!E$uM#x5jBmz_
z-yQ1?nZERP^EbS7;kx6JiF1y3G^S6T`tiw8Y4f8#kG3c=DA@<8{B6JP`C)3W%)u9$
zOI;bIe^*PW@f>P%h_dx_m?!w`+KLn0%O7cPn;54qnA%YAh}rkz!}ED--&<bUcqHPk
zDd*uu$^7a^c9$P1mlari$j)shbAIFecQ@kaefj+2j{8$*!|y*f2LvtH(sAwA@k36@
z<%)GqQIC5rKMKk@nC>w3NXq-JPSY#D)?Q6pfBa14i{*W-h7oT|IVQ<zc`k1%cQ<Wd
zWS%R#n33zmBRz%L+TomQr^GR}@ZMtL*?8Y%(}RHXIm*s#$G+_rI;QDm-*h7TTSStB
z+h&6qzMt3HH#%3_$)DbkZasTN<9&;cy`MJTtbLy}EC2pwwTCM2W{ch4qY`ec`DM?w
zvsY}ZcRpdyHJAFVzqs|Mdfwe54L2>kwbtm>XYK!IX!vo4uTA<P|64Oz54?@Znm+YG
z#9hbOy3^<HrypokT+(?vmEBlz)6KnKr#()%wpz8~|H};CD?NwrJ-@TIozXUDO;Y}s
z3A^SlIG6Z9C#6j=MD29jEYtl9_blJU=e<R#`wA2D@dD9l8EqH-aNkXc@{wv<q_vt;
z*8kedqgmmL%D!*4kQWy{tg^{ub``_JTIc-Ct?q^WB}-){U(<Cro-1nAb2TZjOtht<
zOseZ*g?DFit3gZbYCgu)-TW&*o?5#*Bhgqa+JU7)m*G0kBIy<CSqZX-OYdJR%v;7N
zuvxQ7_R^^^^Pd(*C#{!hh-aVhc2{P}?@4J<XAS*s!ys($;FowMbV+Ei@j2$hNx6R<
zpW3D7c<O5Hn0$@xX1n!|JvZ(7me0{}nyNJ4d)2b!DQp3pY0Iw$`fJTj%{lB{wfKg@
zw~#g-<3nNbE1!lKh&@y=QnU8?W_L8&x@WPI&s0bE{SGa@wTIRjy*!h{Y;OCmUxN9X
z%7=7)EwldJFC;$KA55!>so6GT{~xo=KjmLfJ`|cJ62k4|b1dhvx#V`^6D$(NW|qOf
zRiB>8Holu2@Vl^FXvZ6$Gi5?5T`6WCo#igdS^3w`oqI$f;{6N5JgI3XI>n53y_S{V
zHc6$fM(W%n)n^rFZ$FZK?PsNPa7OOK+|*LVAFB%71h=l$`r^1pbg_>3zO0?R?nXQ;
zFTNR_J7m;fukWbV)*JJyaMhM%sfL4{7jN%%TB>s+*Izqei#Pi@yQd!E%$pzW;a2#5
z@3>;L_J`Bmv)maPeLn8%IsEiwrE@0tg}I7bfA?M45$EyrpqJ6tLMPpC3gVU8e?>Iz
z9ck^~WA>ruhOtL)INJ%ExigR5xjaAeLEq(xMcpFyXDYl*Gc~+gUq;OrkrI?!AIQcS
zwNLbtt>_nCR#~3pt#1}T-deliT!*RqZ#~ru$8LGeK5<5c_w(OfN_}saZB0&S&+6wl
z;(uI}ek0`8n+TP^pOp@r%C@{4!m{Z19hPjZ=ezd=Ej+q}NzUTtf@S4>yh5*p-j}jg
z9&TB;t<NV&HuBopE$eyMdcSOC68RACsVqK!iRsKb!7bJqTKr-aGuEz^DVnM@)%BvE
zR`;5zX}7nPPPw*BAYolQ%ZIhX$7dER&5<iH2`h;H9U=1OV%f6zViCs2SJb$!DwoKJ
zzIhYo+osfWbwgR;*>?{&><H+Lt9;w3=N0Nz%`r`z%iuz}&%$jo@3wR$UC8Z=&p&=#
zCfRu3wwyJ4yzV~Leh}uf_uk%3R|-2Ngn75L%;LJ%ytGU4klyavYp?F)NZ#95Gx<VC
z^>c0OhaYC$*|qS0xy1plxgT$w7ye<__iO2$$LkYbWjATgWczQ=?q}<6Cuy*Dsm#C4
zw{7NGdh;5ax;-zs(xKA#sbuSF#+7ebQv+mA)f|0cl(pt>!E0NKoShZ3Phao2ecq>F
zZ-D*d1$SavBlFhnm~}xr<WAJ8oUNfX_d=?UPt9uWtMZGTCo^x-pAC{;t^V6;J_$q#
zdfngiBzg1lYe|hyFWbHMJbcb8W=j0>`KGzL^=q=XEjsn@kZatBOS^y1GHNQV;(XP9
zn&WXv@+@;3E#;oc`?mCKSopj~GVySGrljK%z5Y`y-@@I>%#~LQ?Eh=<YIEl;|CHbp
zALDOc=}!&*7;d)peBnO3#Z^;6|N5tg{h3j<c)Hl?w|AYm*QcqkUg-Yl>YlmB&%M6c
zD%El?^2OUTP7zuBJ)Xt4&0bz{S~tDz-IZxx*AFkg8d%|C9N+Q1YD0illIHu*t#4K?
zzN#-%Sn@6W@|25nqB7q08x|-0wAd^l&+TsgBI>2usVmQBp8dPz)~lKaQ(tn=`{?bo
z`_#n`_qMiOTfa-q@%+tw;opPS{MEa%-sz;Uxx>mP)r8gM>wFmN9o}E6>(PmOv2E59
zvB*1GiN$~F<#l@G<u3aD`yp(0poDY3$7e+y!?M^1n)Uzg)-xRKKD-%AQ_z&CK}!?V
zRVI6-PUmIs%)P(<1OJmAm%WV&-HcYOyRr+*UB%1hoXHY<`(NKp#ItkK(<7egUDf_Q
zi<JXw_Aj|y@Aoa`m3_5W!2j8oU*|8MUq7erm+Jgafk|JM%fJ8nd9tZH=hH8bf2~hX
zesok~`s;jyW6rhfj<Lkwf1R)Mb!GCxf6IQcSRZ?CsqfF|)GEew_KWDM+O_-Vg-bR)
zDkzivoNu!Hy+*N;<FT!LhO6(#E-Jrs_Ic^?BNky*x7W0Z@A>ol_<ik%4_mg(N}5@e
z&VQm$vpoIQ<&R!1uMeG`{C!{D*N(u-%A#|&!t-{V{&P3}Y4!7-i4)m+!#U+2TGiHy
z@2C~YxUz4bgVU`u+)vV*nBQA)SFA6YX4-rDkYn4eZ*ONTC~<z8c%oN(DsKWG`ysaZ
z6^@4AY_m$FUYyZNu-4Y=N>xaCJR?nP(G^c~6K=gP?;0;GwbWL0(dc{dN93jWM};yo
zrGLL3r|``#{>yE5b-&=Bm-Fjy^ZffBzNxXq%iYt>MDeNeYz3W!NVfFC+d9#Abhdod
z2$**DTxijQcTybf+If-8Q+Yq!^0W*+<G(+wr?y_o@Nubo+v)a`<~LeIU!Aq$5B*d2
zv@s$_B~VD;rNlQRKP7faY{vAQ)LVOPI16U?J#jgHfH#&=eOYwk%#^3PX4e<)|M@x~
zx8nUL@o?tLF{}4K^my^x#e~6RTX>Q9+0;3|W6tj=4N8~!_WNL}{kvZ~_D;xs*u2KM
zTzCS1hWXrAtqYqBqxp6&<+SS%^pC5_*LqR7FR!mhLdvfC@wd>Lne%EhpY3u!VLk2k
zgtCVT!6*Kz^RecAsGjux!gN6`sb8rUF7qN!*k7ETaVq!5izTrcR~r9h$rRdivY%M=
zEMWbu+H);pRg+SGHy#Xqaxs9zrQ^{Yx6mJ2f%!S7H?JveaY()5&f&7U<k%f&r5m1y
zje9L*yB<D%JHxg^nNwnVSzv*U>j4XG%~fslN`L7qpE!`(@v6-A^R%Wft_l3YGHq*P
zI96!nd<&Sm@Al0b<pTD1KN{(rKU+5OXux^-$u=c{`^+M_qSr_D7piRh|6;Q^Q)G5-
z=~syzYJp0h3a)mB7+0k({$$bgDTFWoFO%!$rUgHFO?W0T>Hb=q_VCJahEnx9_6t7+
z9bQyV`mtnh?(R<l3g;>f&#&6Gfpv<!Qg6CqXCA-Dw_hv)iW@KdmYY!DvBjAAlWtO;
zXE8tHuAbvN_a;|XvrL%Ky4mt&)hBnOooQRYX_$3}S;>aGH(tCf!`HBRucy@e^|wmj
zEZ$Tmw|$-Mo?6wGOErERvLWjjE>E7iXV%+-o!8b}n%ENcv+P}uuj#`zH3svSvgw)$
zv2V8IY3WTYpZPQ~OsUX?^ZAeMY%Pf=w#A>eQY+E9BD3iLL#VUjj35d1l-1eiy<IJ4
zq#8dx@>lC%bezE7gauXGwtV_?$Vf!ng8#hY`-}-4CoTwlJZ4+4*FUn?<*DcQk`p{}
z2P^N3Zu0ka&N;pHVD!glVO_^>9}2w}wB%5cUiibvpb4kaSDI&5ds!)eXF0SwWWV0N
zlmO#eF^3E;p>=Ma8k$o3SBA_MWENdt&3J=(vE<!`!0Tpil2cnZ6#edAw_Gqw>geJ%
zyxCjU?axcPwR|RXnf9_(>vit3<)0Qz4tS=r>UJIXs`Z9vBlm8}%3hKuI{8?<kGR3-
z!kZtXZ~u3?QXQa{n=);O`CZW+?vBO&?+<NnZ0wgW30x3%&FzZbDzgR^vmYwgCS3b$
z@H4XhsNnLLck|??O`n;nJ6CX)?@7CJO@5W}$(45}RbRTkZsJ?t7wRWZzj!?7O0}7%
z#PZD#qt9`%%Ul+2ZM%55a*DU<(dtj<iWlDgZNDI~#K)(9?>mm{x9dMPKM57QR<Ss}
zboDD^(R(JFKE?l-{i#;((kU^4%8~=0-`r7rY2=@x6ysRiEVErAaOYt^zpnmJ+iPcA
zcn_!dooU>sb|H7WZE%|d$KQMdcg}O)MdUo{zg*KhDlIo@TFaDh-Iv*$v^1&~1wA=^
zZ(79U;GiYnK1}|!uJwkSYK-^a)vgb%j=X=tfAId}P0vHC!lv;Dzj(SnIJv0F-opIr
z#CD^%uGgII7*w-&FT1cJy7<zX>cxjuzwTV;{@$)+HS;d!H%D93i>~@QEZq3Q=+#+M
z4zmx6^{-~6T$?^8BU*61a*6l%rhN(gyiazXT2Ury%yQN6-^=MM4oe)Vu?TkSTluPG
zU5VSn1NZmLWV)&5koLy0|H;;yb9L{e9ut#aXt6x$P=#Wv@`^L5XP^FHThhE~nfs=I
z5H78iYd`<|a8%FBBC~);{-y(u{4INKrJbE`cj)DP-zhxRTxrur8PnJQr**sw*ep2R
z^jpEcJ8PqltV)mja_@Uf;fe(e|NO0<8WtZ}Hh;E7-Q&XK8BZ_xvupXtf0(l5N!cv}
z-V>FFXZ>bfQTBMzm1*z29j3-K&SB4dd+)*5%&^T|R;ypJt;=q8TX@P&C9mYU!Q7R+
z|EJ%3qEc9`X7t)obL)&x+O3XH3{}7T*M58D(Y@tgppo61mX~(|A8DN5xIQvq+PmEH
zPvQHYU3${>`OU1OEVtSmzkF``wvD6KR6qLVmpp~0tGP=)bEzzl?Y||<cu(f$jfK5B
z2i?s|Kb|nC68^V_f9?q-H<6ofWwTC7w{3mZR=!@~bXj_Q_Op$jm2GF1JxIOc*|f#z
zUYPVfE3wzB5BE;K7`5T7o4!iwotC{pXCLI6+I(-E-{<?F+fP5_S3%1Be?0tC`lJr1
z?OV*}cEUVD>{{Sl*Gnrhxc2gT&G>sl!Toq&rJdY!R)a0f-5zYO+&ztA`WBSwclK;3
z_u7*6wEvyVy>%MRioR2>9h_Jk=X&y%vS;LekH(bWZQD{f3~%|i1!RAHt|}Uw{&&wa
zzhwuqU9)oE#U9Ijw(h%V#ux9LR|RY#Uov~XmoDj>;jgm)nQ-nrS?!Y3Q{5Bhn17vO
zqQBbQY_r(m`<rH3y<54qrBD9cB`MW>#(b4K2jZ5VuY8^`@#g=17eys<PM_H~;hWf{
zZ~FvSCG**3^nY%@Gqpo4WXH?h(^a26U7K0^^UJm)dtUJG)!X>s^N-rjR{q=E!u4J;
zKmRnb{tDRfBAat<{^}UMRq8QM1oC!0I*|R;Ej&0yP5Gb(!}HJ;mrABCJh#F5>_h_@
z_6Mt7%kF9$rad;f`nq|hO^v(JCBON4Wp_5@?Na!-w_y7`H`|i5$;+!1_`UY2Z}Hr&
z^rch8kx%$>)8AKi{8<t9PKIatPyau`Z+D8*=HDrEt}eSn|5t2PX6LuK-F8jq^C~T|
zFZt`uJg&UVI9A8BDv#yF#d%5ZT}uVl?|HXoJ;%am@!E;iZJI{PBHpHTzCD4~4{y(6
z{$Q8nls`S_w8RsYkZFG&>+feiQ}<~amewC}J7x?mP>y(c8+|tKwgJz+_nLp0&ZeBa
z#vky-Ai?^LnZowOcB{NR>o<S(g<S&|$t*1uUSD9e*z}M3Wfl9bB_7LvUf$CCp*}Zx
z$$#;)56eHN-mm$WCcV2js{HJ~7y0&ElAc#xe{+0(ztfkw_oEc8G+X+jE#&`xRA{mJ
zRs8yULCuHjRrPuwCBB_H`0=u{{%`jhA&sndk==~nj|B$^tvvW*o4LWBZv9s|GIr1Q
zN_AvZJ2Z9r_O8)%SAHQWGj$!`kDOiH=WP#9yYcoh$N#$Z8wzqtLXL1h?)=D}^hIjQ
zr!U$6|7O){bz9y2vU7pm$*{loa(?acnx8Pa=3H~#W;@S%mzB#73&t<--*Rv#+sP$|
zvKvn%?whv9#i_SsS<im;30;#a_k6mNYMJHIrJyCrvZ~3A(Y=7x#mH)}zIVvw0tQcm
zWp6|FJTk0iHDGZ2zKQX(e?wt7lkk&U4Uv)teXZ^3McLYH>Mq)WZ4($+ZLi;#_Ym0Z
zxNY_fTY>gO=B%nSyF4|Ib$n-@An0&5ZRWdsj0wJn^B0}GG=-l*;E|x}meuU<>u1L%
z<cLgvV5i2P-RJv&Zx@HiQPto5jvM89yv|QzEoU~=?A^YD^(Y^YLV5b;W&L?o%o%$b
z64%^h_$AQ5G(S0H<$>Z!vKBm?7drMVQEhcO_fRU+^aIa&%V#rK)(NaO+xct5J01fr
zF%id7h1v7D9m|eN-eY?v+mLB)uzlvv#+_*f{5u;bbgpDnZ3>^EvCyhW^U!J5>rOLm
z_vm}CNSqQaB_sc8p2M?lj}O^<Tp7cw9)wJOu<7p3X)!@-yPLjEiSU0lVe%1<BU6uN
z83amtDRXyeuaGkLH~lb8>#>0ee`j##vM4u~lt*(ls=e=t%(y7<x-m8~TXmwqt_^|)
zI-<FUPsFT9-8uh&lCV!!s6>{)&j{(~hHK6>%@(!_nbWqtmGAMctSftL7q0ZPVt*SF
zyjT2)qr#nMoc+2ppZ;3N{PpGIpPyB2pS|c`+<tudq{Bg;J<5;IKFv~{I(zeN?V9Z_
zavWW2o@`>B8>_O`N%;Hv6=tjFs!Le4x!jPhy>@BYz1oF#rapOI>8Dmb6RLQ?bIV}o
zn(2qN&wY+&vQ*@d4DeBCdZ(yb{ZPteW3A^w*<!W6;@IV@1wWgdu8CeEEVV6d*OXk}
z8r|<d*ZCZI{qL4X-jdXyTf8&(8{d3yazf?eF&pE+CzE|j=d>;Q{8>rzb40Kv+l=p;
z6+0Jvobx;B^KCtIQ_gjQ8{d^rey}NLMO*0PIqOv7qqq2Om>y}n<58F9aYOx=S5_)-
z3D;`3TeK_rkG|+fwGaEvGOrxI`)Ix1ue8N$-h4>4(bLv&D0E#?*O}zaX1JAok)v0`
zd9|C;>2+HUYO)Gg<)sGv=2)G1`iZ_B4`=-1XSYJO>Kx%XVDO_Xc=gdUeLnB|?w&JM
z(wkWBTdb+|eCkq5p?f|;bBaCBHmy^y58rp7YU`|&b7z)jr1d=KVvgu|W@laV?#;6&
zxm7dciZm9kJS}tc+4ArQ%R9fH;r)L3%C8f3w{{fA?VjGfD_LX7>%^_ozLxgy7kec=
zA!vtKou10lyI*7P+hin{U0cQ?@mlfYM)~>wn3R^gUcfBo%!r!rFtkM57#MvpPuZMj
z?|1EI>_*z1Ql88@MH&}$FGP3W>5{YAxOchSt^f8ROI$joPT#bcne%wf#_W_e6E!|f
zT%^$@_fPW6f=BkY6~g~lKm7Rq;q&%~m!})F6qcRi<3Il6m*48bdCwju{(qjnR887G
zN;7*((NE|5-#=Cco|1X|UEu$t&&fI6e;Gf0p0scF@!z{o`b?Y^8|9TC#CkD#{)^o+
zqxYQpJK@%x`>R}zx6ARq@wCzFtd|bT&$-^;=_PekdBaqdh)c7N$nSTRvgtc6#C{@s
z$N$$4x929zzP#BshE1~j^*7Iu8t*l-dd(ZpNL-5kb>r<3;jqG$9nlf3r<HT+x<vw3
zA1PkwZF8n``K>A|`>(gtRdVL;zZGWuy)m45yF;y0+?&c3%dKjhW<1v6GrPzoHRtxz
z=NFD!dw;Hwe3$xkj-l4hT?I?-iksA!J*l00n9DNkwQi37FZ14szAn3jX3M%~`&dYG
z=PA_QsN`Un=GFZ0%ex~re3>?llh4jLxG<7mSM}N6qQlc#J}IefQQ-biH|@UH{=46|
z?BZwId-8W*x$TSYKd~E58o%4^;LUU+*Lm)QF5kONA1q(4@Hp^#+S9mspSG^nZFxFJ
zxNT4Qwtb5hc*do^?fkGlRMwuUsOI&?o2S+H+CS{vm9hJbj&$#8olS)gt3~Fh6<oE+
zWPi4RS?Ef^n%na0Z1fk-TbC#~q0ymUIJvKaTV-nN)}P9)#>adMBz9_TFy3aD_{>~(
z?&XQLdqUiw$mLm8CCe`SU$$HPbw>H+i{CE>@+RnC6+EME#kKz8#iQ=)IuEZtaysRn
z^}8xTJDo4r+t0dJY_gbFY#AwER;&>$_juL<9l;5iiLD>{5}qzOP<p^ivn6?6WYnr!
z^$GX-w*J4paa!4zmr)CrH}_}p$39S9BCN%dl{F=O%d+VKhvVXwcxInk#=TCkyLpq%
zZ?CkuA2S<$ghk!{{$kPmyxBHOY4z&Vdn=cgl(9{eU#4*UneeN)sGGC+w<WT5#<E8x
zX3zKF7FDf&($nxJ@wNi%+1$q|4gV+fwXCs=-fhs5@pjhK9o4rC4$j${elBN9>cx&z
zk@qH9-~C=RduHc^w@LeQ`NAhvaIcTZJ|ceHWp{Q*^StfV3tqR>$fruVd%Mq_pzKip
z$gSGy#m|skfj6JtsL>CXT7S4mTUjsQo<gbDF~;kgHofRlwbsno^T2AKYh~@WbLI^?
zcbVeMQ=a=piFBTueYj0q?PvSzc3&sgy7$xmtdM3pqi-2hv}x`4ovJQXoD+X8tz7n=
z<14$zqzh`F+6?M{8S!>(o>jUoiOF_Wf<@JbIi>->UQDui;Illeqaf&H>{7XwTN4Y9
zh5eY*v*L)#@zQvWzdH=tg^oV`5twG*C4F&?V8pqPX=ir^Z+gPjx90AlP`|g^jT<<g
zOz66Q{g%X@OjDiDs~$Dy=L+Y3cyn*_)2UMq?)})YXR>Xj-tH+!86TSz&XIlBeDZ3<
z5&b^a$*(;aCM2Jk{ak8-fBdtl=X#|MIxIC4Ki^gCXY0)<kuHAxS#ivkq->L3uRn`@
zPWbQo{cO9_`OamNP3FqQewdnZ=k@_rWtTJt=din}Y_C7~NOpJ}-LT8p*(G<|!m^sS
z;Dr@^-;?H__sbEOx9*O$Oy<+YI>j?P^`8s#Jb&L&<>Q~mQ>L}#!m&$77Hx7^JJT!o
zs{0oc`8O|`^;&kW(3Z+{S4wtgUntlSo+GqszgowYxmOeClw@x`lfQWv<Gd^LO&1G^
z1uYN=eq5e?OZ6o8%#w_$*J7WGKP%jHwd;obhEJzoZRFLlv%A*Xu<rVk6})QaRVN(}
z2}nF$R&2lWeCg?p^QP_bUz^kTB(KBmzT`1EiK|vgucx;=_FB2H_8xO<aZ7*m;rRNw
z$LjYldlhQ$a&2#td_rwV;1m_*)1Gz?S7#k4p3`$BZ_e30a(++wc5iMzV{(Y+wPRbJ
zb+?IgO5G2sncJQ^f4Z{X)^m||z{4r?Gkw3MB$((L{9<9hml4}{#4-P^d5Y*q+2VzN
zUo_ep>BOW)f2+S5r_y}(+!F<Fi4)o#U-)M6cI@)bck#_LKBe%v|K$JXbDXL@o*`N8
z!pm<stbhJ#=H^+P&&=O+SFo=Wow(`wl<6k#Qx_S%d|LNjMe~!<e#h_MCmd9n`^>ub
z{4f5UcdyRQ^RhW_YJA3&K}4)0#d>Oh{N?JL$?ul4EI84>TVS1f>YGE%O6OwlbMEoi
zm{)JLqkqw)>8fA8sIp|Zo}YSk8^iSOH!dCDUM+s4b+fWEbm6=|A$uy<ns5AVwqotk
zKTGwRSBHvKF0l*F+TmgIX~$d}`JiV?`kp2NyDk)dT@hR%^v7{u+MQSJOiGL;hl`uN
zCkCv~&9BI7c^-6n{~zA-&zd@1AMIu8v3OtRn*3-&aKzs?zINFKde0-yzcKr=?y7~@
z;vb$@wP%V=@%?;l(~dutnmT69ljHAyX;{w8b!EYVsgnG6C0-^jXbhBBWvM;dTz}sA
zdatF|k^QRIQ;y6#-}&14euqXqw=YBSzod=Jd9qH6ADfe+bVzae*7-k+XD0DY2+nq|
zG_T^)zPY`J>yXqn`RfX6>TW6gcyHNU@NFuaqHIF4z~^?uj??+u%#1qn^I0P^?{=@f
zwSTScZ4YtVmia3s{Gzr<8VDad>34YU=Q^H+Q}iz6H&omJZJk&({oxBNH9v7%Ck)Mu
zQCB+b?7M9s@b<Iluf|f1pht`>noKILyBOXziD;jG7xd)+?1D548<oo1lhV<GWvs6k
zc<ifua^~@cb^k(2RsY1tO1}K>y7hN#?aeUx{WnkL&AH~g)%M46{jE24TnoMZ@433-
ztz+_A1$srC+L_l^*Wcq*<o|c;m%qjRpT+-Ue!fogzh0nMK0iL{=C50`)~)gMT*c^h
zW$i&FThj%xHJ+F59$WVLR=BR{-NRm8Y7aE0U%4f}zTk+^q5YRXPc`~Cb3xRD2cJ$|
z*E=0#dzB+#>teAJA~9T5EaeQ7E^gf0^5Xip-zN@Th>Wj(wMF*O?3%YxNdnis&NK@?
zKD1djG3ROXWM+lNIT80*wSMgVc24EyclDDNhrWqyc7Ok};nsyO2hDP-)+Eb!@UXnf
znfCG-6Vq~?nn`C1p3miIsc32bBzfZZ^Ubg4#wzfb^?84obL6n{akT<=Uxlv6lVT+_
zbJfJ|2p?Gzaqm)N+gqciJD)h5%=ZeF#5cXsH9oP9W$$gCLv`N{E@@gPY5hfjHNd~N
z)0@lFw!Jzh^o_Aev4<gNjG~4^ShI?nTqGk~LI#I;#>SVnJe&B}d<ZiNhr;*X&-5bP
zpG{raC*UZorV!FD!)m5iq&?-;L!+q%M+NIw={;}1-aaG6P3=>4A?w{s23ZT2&uTnf
z^*Bqo!TjqEPbvS6uP@3JDfBR`(+O+dIBlIwO2&>wVUN!l^KJ<$nUks^;_G@S<k8A$
z5`rr~xutZ@=WBN|%h>VyM(oVnw%0@*UL^G#I1rSWBQaBG>&$P$Q&O5d*LF2_P2dlH
zT4HR=J&(^MZO*RC3pU;fawwknMP^&FU#aJVmlpgsxhKy|n}6)X%=UBF@8lo4J!4(b
z(#1yGcorX&XXp3NcWqy))Z={3eEO!TbMpN1)_=G=>DPe}73JV3GP^zET94h^pLi?f
zTTH?W0ilPNUo2M6eQ`J1I{Mbkr#>^U&S~bK<=Fk`$i4Z(rC06zR(|_*!Earg@z<Rx
z9hM?fcM497*4cM^<07VR5#4z^Cr-ZJvTpOqr%EprWw$w;(&4jStkY_%Df~S%UC7??
z-m0nnn}0mmxBT9dng_2IowhVoZhK(4UM<a}mgoA-XQ!8n6wJBi=)G9F>FP;4_MAh0
zcCTa1@`@@izk8aP&?g?VT(S4nhTyn4;&RM4C!UrMU+1}eY4^mP`}MA;|E#dvBp~x@
zEyr5_=GhAUV!3|<rpodQUR?h@SZ%3i<;=$Cm;KjgcK`Szdt&)4mhBF6rYQ*S(1}%0
z6L?~<@>AHqqyu|>o-KUh9B{w=Nzv7sSqg2YJ1hRJZ~Y<i=eG29|6S2$|IHodC^PUi
zOD&f;U82}*-t1AVv+K+Y<{KW|a?INu{^TD!kW)Fu^Jw|94T-te<-H>=vU$F9NL=gm
zdO>jR%gTK%O)Fzd?>>6Vsdh&9gP48fgVXMBJjG7@)&1~~``y&&oCSOSNw56;@w|Oq
z!b)lL^6ouhBKKl9KDW)1KmGNf&{>C={@c&y1fTkA{#bwiqLuseUWU8mvWkhXYnw0V
znQ5MTIZXf7jNYw*(-wbwaJ{bP2lw(yk0!2E^h0^gh}vXeXk?6f#z~BKK6u0?{0IM*
z;4hnaL~MC(@a^V3k$Xe2Z}M%v-~Xp~PFc~eygcvqvNG))H<@VuSIR#-m&`T$KP&d|
zoBgqui+;~HJ-+|!_S(;1+cVG9F5jB-Exr18-tl+VS51FE{LY;0Ww^}$+l#lB+uOH2
zfAsPEzvrpJnf<cKm&`8zJMg~w^||xkc5j?tP?zFU^6l5b`G>zgemwWltAk$~=H1#R
zV6*qKQT@{L@{OX!#oLbEw+_xedbs1*)}zi%IS={Tf4-Q#LHNZ%Plw_YDYB>be7*F$
zMr8YfEv)x8{w+VwA@Jpp|BUsErx{4=&fzY1&}PtKT+p8^FlX(#G@G5wzmzTR^m;#d
zux5FU$|PGaVI9w%Tb1<~e{A#KKk=xeb>*(9S8D}cB<49Ry{h`+DC>m>*BUq$8HzX_
zJJ)_=%L4w-PPGav3-qrW`Ux6x*m-BaJN|E8rs3x=8_u13*dAlpQ>moT*<+NZ_+!eN
z1Byy-<xBk^3x{WxwYGL;Wb5ll_Vg)DN<3H_Vd1x6apob@9sF+(W-w$GmR(@_<5Ur^
zyyB`@3<EDqjDkkUiH$B6>+TCJk(_PUV~}`<gWZ2Glhom>+FOE4O_nNGR;}#&(-%<s
zWtr8a6HoQ8U0nF;g_%{?!Pmv5PqzO!TJq(NuIynR;lsK|ulN04=WlUl%3tQHW|xdY
zll1zB@<oLji6?R$C6yHB9o9N?Xv3`Q5i4#8ttnvL;oGmE>+148X{G7X+U--q=4IsH
zX8JPY#|E31Lb+TknOZs2h2KB!nUsF-K`7hWD9Lbv)26Nkm5*d19!fqkc%0?Ww`8^k
z$E`kNk0$;W?Orj*Pme_1XU#dt+?~ZD>NS5M-^!`IPZgxqCTO%Ac22WpU%nzh@Wt`1
zOgV~+Rg;~TiET~ca<h}0xZrAawm;9oL)(>A<<?0rdocN%&(z)rVmE^d)u(yhEz!7Y
zx>w-X=df_|c3zRYagWni2zRyfMRznldn@BGyW^FmcX8DGXA<iters~_J}~Wv$qSdI
z!YyJtUH)CmE}gg}W#TEayK4O*ckOu#T#hv?y&>kTy0c=bWI|Ix=IaSlH|cCT-70^1
z;!5kf8+JJ}Tfc7oFp)7x>z>_Qzo3O{ORgk&ER=XO>F8AFzl>Ze9A~-``L-=cQnP<<
z%5kb|>H0a|^FmHc=J0HBz39y=GVj7A{b)ZM2i|nQrH(=~EW=jCzAe%6-?{dLuK%O$
zpG8VimhKFFwEbZFx1>KZ;=hg`C{BB~KP}+G%{Gb9(^q06mD1;@Z;Dtbn73}n<$D}^
zQs?GQ+|HvmW%B;q1)H}E1+DmesNq9fgTnmd26l6=_UqbdS*&?9#dz+eZYE#B$6b-W
zXH!jg_f9FAa4Ku+)oTZ{Qqxzj*`~Al>&CQgabey4k?S`_wR^4J7?rwp>ZV&uHs76a
zm2ZB-)fG0f_huVqr<!fQc{b7f`-IJ#qEoj|-E_Nj`R^VccmB?mU3xwfU!GZ>uq&dY
zzU!&m%*@oCr;1)D?fk0l^Dgq>n<Xpjx6ilV=d5QbJ&jYY>ZNwV?FBDq9$XQ{BiE4n
zXl}ph!As|xmaaFfy1O@QSGBM8t@NI=om0iv*c_NG^JVe<(ss6w(<83mIMZsyarXP(
z--o$YF}6f(Z{EM=z|NiFL8j%qa^LrIX7cc>WbFR?>y!Ml3#@*arJ}h3F{>C1%`D7N
zn|UYuo|y6MeXjk7Id<Y!X^vk43YJgA7UVnfJuSPQ%lGrYw@;UdroQQ35v!}h$;Vdc
zWiS4*qcf@We@Sinv-;ZUZ~ojbIe&jw{QO<>>(9s9@f!Ym_tMDR{>w9ymW9=`{{1>C
zVJ#(GmHa2gXUpg3e{OUe=KbqgR?}`~ZfD0L{Y<cM(Y-dew5l-Mhq_-59RGaVW%ZH7
zchwKKds{8oTEFbQ^W4HkU!UK*W;pc@*W=h#wUxZLIM1-%f4Fzemml9R%a@-PnX&eq
z&yvH}8u{`?=6-nT9iJw%f&E(1x5+cvTN1>*Jzgm$C#IZet>C_V?(f}zf8X>!tn@sT
zyK?Q}^i!<{Gd!#Ejm!9F814IJb$!2BPT$m8MFx|nx<)@Xr6|^T&wD1v%9nU6=oagp
zUEMR83*$6-TeTJM^_-meuWV=HZI>kiw=RA986}{#xqDBjS3)Wy(=~qcMDLkiPxMWA
z5-K0hyk@jqSJZfeRz|b-niI3xo(i*GF$z1rspiaDO*8F%Q;zpC1sFY*-Mmv{{j}S4
zb=kf*chBF=9zJ7M%)~7Q>jh&EU#i*g#Wker>?Pi5q2C{7NFOK-=WJ~Ky-VZOj=ay4
zHIhS(cZc!5xosZJlJeS0Fv6(+`+3&riX;0f{kMc(zjSHC-g&ay?{_<WPoAh#F`@9e
zQmSCWn%!)UogYh`S~u`Ks_QDrRt_lJp3AZ1XDRcZWXDy%rMYw!P9#jdsMOry+dFwn
z?*qRl74C%#uWpFj($6_ny=SYp%9milY}LFc-gBP%`zo(Fev0W)w;k6Ljr{x)4gXiI
zYp=dEHIw4dl3MJUdV=9lS>&s$-`5Ek7{8KY+!)@K_pKx2uiEO}(o6Ljs&g$r*=NW~
z@vir-_!zMveNM&Y?9kUqN2V<(?@i5{!I|a|*rB>A*}mib;trEDI<Gu)Wh`cxF1f(>
ze$J-@SMB^v>gxWu9$CRL>jF#eh3i7{5pSEOSMJe#e~@X>*)2;hh<BQ9e4je`?sbor
zr;Mf5dY7BR9v1CXo!-nXa^C0hxs>N;>U=aBL?<2m#}e#v_iKu1<2G^2=mkpyud!c0
zb;N0gzR=p~*LORK8G8pTkXBiKsK!&P<JPo4P7l>4ORwKzeQ;h(_s;~TZmBr?ZV&5d
z#u>u$w@%%1Uf-Nqy1}{j=GLPpwg=AHz5Q3flML+(+CSGm?&K4S=(A9JKJhfOuk^Ah
zJI(JLzuo6CyDZ1{_SLB(zwd0j)#w#_d)c8m{Ay|ZTi?qkuPT1|S@U+Hpdr)8ZF!=H
zPv6-URJhXOw6k%qSdm%XBi*~j+S^`OYUe$g{r;rN;cwrk1~&Q2ZJ(miC1<9~ID1xL
zd+M5WiKvRMDF<)tGdU<%?R-kc!QZU#<f`I0<5T+z0>XQ=1o+<{xNxZ2BFsVH#jmGP
zownWa+crGeG-cz$oN`v}D+((+kMEtBvDAva)@i9({*>UGoc$H6v@0sR=Kf@0?%B;>
zd3o9Y()MlJdk%Em@z}D<<j=GeiKlxT>;HYb_xM*)VC@Cn9d_p?vmJiKa-nbg$(bxC
zXC_Rr+@o%JwY??g!4IYF&%bOv|6Jq5WERoGEBE%zpO)US%H`IlIbV13Z)Q;W)tR)~
z<<qq%d?hlerwz-Wt9+EJo0fCn;Q8ne!k;P*TyCCk{zU!sE04l62ha1XhetjauY9gO
z)8Sd7udn{jg)8&)t+y@uQLDFb`}~qai+4$B|5bdQd^>J(#h$sv?UQ}tx?ZRJe)rJw
zWl4D4-KTf7T)7jb+@8ifV>YvKjex|R>x^gSG8$`qFQ_|x^jJL$pM2sL?v?HnjF0l!
zOL){w`6YDgu(*Fkya{K1i536n8LzVJwRD@k?}jt)C=V<Z^%Y%K!_-jo{jchQ->Ys-
zwGUeMP{k_G+w|Q<^FS%%il@^*^{0ONuxM^`>qG7zDs$hiF8{umDdWdS1}EcMhLYn_
zQ}`R~comiu2f1nlemf^>TNtEjuyODC3*UcM+^iCu|3oG|SLJuXeZd7&Z#T?hKV`%>
z_l>N31^bNojA?cUY)n2ge6Htt_BkwZz4X&AtF^C6uU)OasJFO%yDMk3#LG$Ap@H7p
z)dhZvseG}nb#1(==sA7o?SCd;az1X$3%nKOWu6oIOJT=0#!&IEy=y0|)izoB?U_~f
zmwpAQL+h`f%m2IY@a)vRA;BEema6nUkG^t`f7dZD_p-i}=4`8&V2|vbmu_aJc~#yC
z+yCPJ$IUaV@0g^`IrctZ>i1ON|KCFDLPV`@?3S$h5ju72Pm!7za!Fpx=WJf(RL0Ez
z`}6X8#wEX-Ut+1-jESn-OpFas>o)8D$Iw#^%bY(d1|4|V#ZV3z!?<<|G=^dRkjL^#
zz_LFVPVU~8@P=!WYS)~fX%_lHXX|}$bNs(<7WS+DV$Gl1e>P^<{eLO-{a*RpyfXLy
zKh}!A-}`d*-5=l06`v@yWVT*dEHL*=di{yT0;lF%fBbva<1_ENKZb2#QtN(S2;#pm
zDQc_yzdmmN2eW=PZ@)6(*Q@0B1z$?SS5;iM|GMSz)<fmN+!Eg(to$l2_UlD=q5FdB
zb-By;|KI$(Souu*W2<Km1J<3|b8%v&lDm;X8$(MWqsmXgT}{7NrSJb<;LCjY_62Qr
zj&s*o%N88J6LtFs+dkLwsa`oUlDV}4%v>I<g#uGj*Pm!Tx9Eg|bM@s*`m^6TpY~z@
zV!)r<Husl%mW#+v``dY8mnXID@OWlbbg_NG<C4AWY!30<n&NZkZ|bC+DO~#t@9kiE
zyYS6h5mO_-jH;L;_u^+WyI<he-D52<D^knxT&a(In{Ud>y}e5(+r{{nUi>ugR#e{S
z-zyj|`dxp!b!v<!v*{fBMdzJsdnbFY>8(s-n-;v><(x&S>*vo(nx7+@O)O^IW{5kV
zS)qP$=B&nBXH<TOioGr~H`JT?*X`<)pzDFJ?j1Vox4iRBY8(67uZC?h7dO4u_%F(<
z*uea}_jJ8x_wDKb1?J}l{}bSExA;-^-2Ku~57lFSifa|vk2Weia$KaDDiB<|z>b6E
zwy3h)AxA;!lXG56*gUsgC}48kDA`H-fp~@c#=aV!s-}*f&;=`JzB-&}vvBUJwk+=}
zeIfaW)1HS%#3;X;bu8$DRQ>KL=Wd?aGnd7p*zs)FGWN^&w|=^sHK8(9$JID_jy8`%
zY1{3b8QTx4#YnB5{93|js`5`>HSOEmFS}GPTejhH$ZpBveX(0_U6{OX*5joYoXb;|
zo2`gEAm`us@cZ<cj~1@k8W1-_+j!3;rN!m`nMF;UzZQHCx)|gVb+>B6*V&~Pg4h1w
zV!wOTTw~RCrEty&(Y~Ud#{0Ii^0#{hIc-lW`yIEtkmGUCVINb}7n>_;`}^{0Y=cXV
zajVoRo6DcB*)wtaYSUB(7Dpjng9{R~J{r!<fA^GO{#nL-MMr)=?dYE?zu>>j=Jz_e
zu1=dH*cH89xck)0Tt0sH?p~X+Ej<18vCu8M=PX)qsbr>0(7L>o>?N^b{$CE{FtRNO
zbKj=7;FD3PN7=kYsRd0db_qZ3PIw|#^vwHJf9k<qManXh&T~ZtST261FM0N)hy1it
z9A^)EyOmE9v}nlNAK~`9H04R0(JqC>G6(KyEtZH~Yu2CVx09ze+%RfFe_yq=f%LOi
zpPl``X+D^7_~f4ptDPlZNt|a%?P^HjTwTP^lD%r99qXD1CyQefcP@Ly-uPwewYtct
zZn?WQ++Sw==rDR2G|f57K|yP}Tc(R{Hmhk&MSAL`m`yu(uuk~)KfwHQmp^BUgoa?&
za!Hl-=dV>Ihu)obeeS7wY3>PM6M2eC&IGuxOAWU2Xy;Thk~IkS{@b#zU~TtSjcnE%
zPSYK^R@^>ylwr50b5oT6>&=t0tG%7>%nv_f{QK65O%s>79@Gl0%(lK)lxr;Xa8cYB
z{pTO1wwiypY&D<4ebnW5dPnoR&vFS?p)viFS{EJ-IewJm!n_$5r8bz(?ux6CPyRgd
zyveNg%+NO71qo5xH?iFPbeCzqzHLmB;FnZa!(g?Kx*xI^m)&FBw1Ml#oe5`CqeME_
z@kpPMxE1sB&4$7i7vAmdlE10=M1NPFr{>BVVb*c16QdlL#7Cw7a#>*>pyH_h>3~7}
zkAlQ9>mxtctF4Z|c;qZ^WqroF`O|gN#R5(5Z#i;=UHN&|*Gv4i?;STy)v2>S=Q!QY
zL2CKy9Txu%m`-~B!Scf%qyOvyb#D$mKd_83$L0Q^?+1c*l`r=Fd@b_#yO>+illDu0
zUH|v%?f+K}Jg&o1wwn=Ewwsz;qLl4XzWJbj%lGgf{71AW|6*Y)*0^B%Wm`s`D}P~H
z@h#goE9>_!(s*^)c(TPyldno{NfkN4LSK)3_oz(!Vd&G`Re$+`!oSrIKmPvk`tOIA
zo8|vVthsvlZ*WbG`AWkX26yKFv(aGP9F!L1HGj$`f%6|5)R!8}Tl*-xs_pZ~a}Axk
zZ#1J<-?{Ys@tH~QCb?Q|xl>vxBE9Hs@XJtB+x9!R&(vjH{o$E@c3X<M?XmO|$%<1Z
z`tRI6bN;8CCz@ZI*WKvSuRL~TaYevH%YUW+e@j+8@i1a!JHO30jqOrJnOxHLwj8-f
zi9e$1mlnScn-yhTYh``@TwwmyUq(Bwi?!btU!AqT;g%Rj@4}}KcXK{juG&-bwB%j&
zI`5!HVeiO2yAGs^2{QI@KVgzTc26Y3YCf~?+DX=xS6@&5p5AQwsq^u=3Mb|1M;$(y
z#I&Sov&8*Ud=j(f`)`5b@U7+z&;JW=db_GDI&}8u&AzLper;X#tSt0)Xzu2%w?a47
zhJ<dm(w4dta`lD6!6n(79QTK&K2F*0b~ot5i`ctmrMFk?vi09p9<rl-t%i)K$|iet
zlNUP*|9<<dG56oDu&3YVzWjCe>bj!;&8Kqf4^K%^T%&a7@YPL^wtnciv-JGzs|&ww
zI1>AO?V`4j|Ia79EIq36Ge6H+tI2i2O#_|JQV|x??{8+Wzi2*V?>gbPi}%$oc<i>-
zV(*Csf42YKbl`xLYWLM&>ymbN<-WhY@U_eDj(mwr)*+K$pDqYCo$PyV+Woof)}8#(
zduaLjG=-Ru8K3gQ3#1M%{_tbrbN5%J4=uCm1@0L}2ednj&KH}nt#d**U`D0Y^WGNe
zXs+uiON!!+jWXX{37)q!H=T2Z`xm`))6zY?1eDdj%{cCTXKxGtufkPt=Pkbe(}z`4
zu|L?o!1Be`>C2m<=DxPv)FiWy&2rXRSqq8Y!in4#Qo1oGHWo3i<aFb|9Nkc~I8Pui
zo%P<csgpFym#^CLXG+AP&o*5WCV>aSp0M533~GyvNOKE2{!E~-{eYy?hh^WC`mY>w
za$UQwD}wXiA&n0!-$$k${dTTvuUKZ#+nzXa?K50?nsXlSyVLlS&#qkS(bOZX`iz;D
zAwdihri(sxe_g)utVlBF$C*<fE^qz0?p)A@h;s+0Y-bf{XPw*bT>WLj9(zCbY4t81
zwes(m9eygr_Uu}BN14;y;D#;Jr5Zwmb5cvQUx#Z3zdSrID<!b?V6qYKC7b%)&*hnK
zwJv&fvQO^r{@SIo-`)#`)%O?7yS9sC%f5*r4O8tlroI<$Xa3o0VERRXv3aL>QncHf
z6j_7ftC4y0ol2(J<-JYyoF8tJn3OWHkxwJ=;GGcN_7{hhv<}QYcEMpCH;4Yp&akBg
z(OseD@kgA6Q+-eB{+PNX;wkeg<L6WKTK$e4Ejy$bbf)~pP7e>+YE2W(nctl*Es-_u
zcGGY8GAHbY_Q{<F=1S~)E4~U%ulOIbHY(G5o@i%HZIRX^y+z{F7Cm>0F}b*BkJY3N
zs%yg&?S(J)?Kfl)SbiguWAXN-*NgRcg*|X9*pVZ8U2;i+Z>Rs~nn~ND>{iTas+k~m
z*>r8sd9Tcx$&b?Wly8(gtqQw;M0I-ROs51xk5>+{Q7q>=l%57PY%Tk{=mU#4$Jxx8
z@hsPl@K;S*b^qM!hPTmc8O=iVg1+2}@tky7we+@)rqe|^>u5)l)D7I)7qZrCq;9%A
zbGeoEqkcL5P}}Yg4i8=Rrd7_WeKq4(*OrM&em03a<39wL^GrD}&w4}taL2*f58o$t
z&F->JHfVckQX_xC=j*52AC7-$6=z(Wc;}GR2^nL#$8owPB@gafySIl3-A&fmHT6Po
zPx&u}=$kWqC(n(``B+hua(my66X6j*qkqUu>RV(L603Xrut{9t>pSk-au+9P{@?nd
z+An=gm%+uWOwD_~1Z*&fx%0Rrqr3K-sq1O3zZ)00%c?w(zk6EbWpu<z5%rVb3~k#|
z`JP!Xy7TVgthBDBYy9_Ro%()+NA)U8<bzXMi>g8||NL3;W6k`nuP!839WI;czI#vX
zvo&Y_>c#(kx#7g#>8CpO1?o-<y0N@PyVa64eb4FLHVM72xf{j4ac?>C)YZFN{EKMK
zmn%nJ6y|-oXMRxj<lNb{KcC8ndRf$Bsc<Zbsc?)8K|@rC%a-3pp6+|HjqfgKzt0r+
z_v{x|xHkASyb=1McVf|<tG26m@A&!O)un5SM)cHmuS}R&$|qE;%Iy}BZ`ZH<d4G|3
z=)cR}CQs|P?)g*x^I_%BPgnJ$)<&l8wb`$GYu&Y1Pg(kF;^wX2Z>#xL==^^B|9k%Y
zuu(K9G`;wa<Iru(o1e0Me@1IQJ+;2q>u7;pWOjn9p8vfBR-4AH8;<3ARW6@W7kDaX
zL)U$&YO{}?MGadI$X{c-@%+~JqNn-0O>gyDhRM~&JLm6Z`}NtA`F?h6tI(uB3)l87
zdl&HZxom^=o6K923uWIcm@?TmZdbh>5v9BI&8{h5Uv3k=k$r4!+*0;EW&%4jbXHvu
zQqTS(ob4dT5G~;<rsbc>`)l2unJjL#w=HF-Z&>+Vt?jqV_k|nVr<$4Fy8daMjp6i9
z&wA%q^;^~be8pD$WWz5Rt{bb99J;+yf9P&GXCIokW3g3rYl3a{*>fx(EWW$^e))=R
zr__AaxKI_TiiX1%{<<-BrpU%VFDZ1*+Vm?;x25|g<E~j!Zy5ab|ENS2+&_3u!S(Le
z2EBEfyM$JheeGiwb?EpLJb7Y6U5m+Up><dHn~J<;Z>p4Y>6uyavhnzf6&ab15j77Y
zkA<i@aVe`hWrUq$E{oEb8zr*hs?_lW!`~jg;pdn{7G7VL_Nwr<`E<cEX<KejDE0^q
z+c>LzQ_HQ{+q4h0txL$8*}}YZo#*o-d@et}%-LP&6tLQlwP<zW#y`?s@Am&Qsr+$#
z`z)(RLf2(PABE~Je{m}MoTTX-|9eUwj>)LLEQ#2BQhB59imlr{d*XAqDR(md-fv<%
zP3haGwH>c-wa)r-dnTXRTCKM!A3q-0a^d$UeS;<TGiQc1%{FGWtx|v1HgWUg3V|cL
zlSDQPt6bG)`LOg$^5+NB*JpZW^BrWIEL?oCuy<M_&l8WTw-=gndKCCXZc96!zw%vR
z?X*q3$_LzCC-Ri6-#er5=hND(Kfw#0oMXTG;6`)$`Xko8n+jj%8S<7)-*rt}Yx>S~
zu`?o>IujM<Z!fK7sV(g-e~=oV>i62pT=My{w3xJ+n?+LsH)o~aOmC}s_Hz4@SBIou
zx7bFn)4FP=DwExi$MLi%eWjGC{8O)BuiVo|O+yWIGb$HXZuoV<W`P>l=a_(=&6CtO
zsT(E~KTm&p@>ZzWrRmpdkKWSV+SR{}XOrS9&a5DLy(hg6D=%F!i57nKRw_iN>9urH
zRMFynkDm2Ch(Ex3g~|WMzU50VU0m1tW1?Thf(;9o|B7<xa<Q^0^i+_)$n`ELATMTV
z{0UvhccPbPP06l%sKBVPrOv|t@5?ahHy$gVUu5;!V6k_j&atu(L$!?|6PFx{{=xEY
z&%NxFwzNrm+Va`fKNNn?5w}*c<IvC9H%ca_ygPBNL~6|yt5c^US#LYc{dHSw_ra?E
zw@rF8E+}3TJ5?+(GwzV({DQiby@zB@UdU<p-aYy6qL<O)E%Ri`W;De2y)yU``F9<!
z<f?vOS%GQ4Y}5_X|H?YI2CewU9DLN_q23g=M=1sdp1pnZ8}9zTnsMLoc9Hw}`n~OQ
z8}Cic;M{5WVBV=CKI<4)9x!Fkd~wpw#o}fNpX}t`X30sG=Y2a;W}eahn><+|K-KaW
z&o-k~6>De2v00odFXn%=eih^1cekumPl^5I{V2H1?V4JL;xhA_>!t`i|C;w9BYCQX
zP*mq?&c6>YOh5bKh{>%flVt5C8G9dA>w559mhld^u-OS`gP&8v?$6lsY}#wNyl>@C
z4BB><-~K!CMX>b0iKYH^=F%na-j;Xe7%`hxcWes0Xpntv=HnY{R#!cF(3@{~y8J;|
zk4T_g)Z?h*J-frFd0%W<XwRu&m@xH+h67vC9QL=r86-RG#1$6X@H3=rIV_@3Ua{%-
zk{GS#Jv09(%AGN}aBBLr4;gd0W9lzkMeV*irLQbs?)$pREq<v5_b%t!RF&1*KU@3c
zdW6yE$M>W-p5HsPsysLGkNJ`_OY47jOqi6Xr_6V1dh+iR>Y+A%Te|+%X-S!{QjYi4
z|MqmB$m95ErN!^I+-o#kxZ<*`Y2w#&drsb5ZpxqaJ(lrSq}^-tuxF=lL>%Oi+g#Ir
zT+*a##`d}pkI2hB*;{McZJlnjyL)JCy1@Rl%8}bYTlVC&LgNz#_s{2ie!k7N{ijSD
ztBk}0hG$NX?Pqu&wrqaBZ0d6J+fx@$o?;z*{@%TFldkSsTdcZa=e1&+pABtKst&(A
z8_l$Ir}eJz^~wr6E<Q}wajE{i=Khg`??bhCjkX<h|FYv-6+_tXFB4z0E_dswQ`V`P
z`fkG6I`OG*Z$;hQrn^>abF%1xxaZT<7=J8wvDcf$dGIs)OYd6i`VxmVv-jUS_pSKn
zs|1DIbpf?eN0!BBs2-QURR4N;S=#G$RzW9n4ppk{l4G!4<H9hzuwj0o=)pC|17;W2
zlpkMv{(hO9__I&fwaRrqMbBG!Yh%W3zW#l4e()98YI&^CJ!;~3-Q<MO%Wf9U=|5RE
z9oL92xl(s(_2s<duCo>u6$|HGSYT-W>dE(IjZ429W@&z5TYTEKr%72Y`f~pH6VG-_
z{QPxB?Z}}ifxEm-qgv0tnIriAE8k+D<?n7c3NF{Z_D}Zeix;wC?Wbc*tHl?ujlZ&7
zUg-2|+n$KEQI1wxe_4FyRnNKp&vS}>Xy1(fGdDk2w*C;=c=GDArHk1#B$bR_R+`H-
z$_2Oem-IDW-LY2Tc~VjEmqexS7RRlo_h+YYFOfFStn!YRsCxCrvejOs`0kVQR||w5
zAG#76I7KXE+WvysdI|QMZ>+z$`t~ok>;9QFa*OYtWICGdw0>5NQk~APqT?P$ZpyM>
z_r~wp9RDVr^>0G%i)%&yYm|T7H~4T?x%kt%m2BIt<lp$yU$(5`|NA#SpTD2`Qk`B^
zpFV3}`Lh4+b6@^>J==42`O;P{!*!afiTmqb{X1R%kHum?+Z@csi-i#}jTb{>L)4Mb
z$di4K%>?#7*Z#qrcH$AA$uk27A%*QM`&ca_bgTV}J^uYS(3y19b<ysrWeo{T$!@yc
zNo)JJPDrXZJn8aBeP;8I_zl<Bzxh1#?~l#t^VqXZt?x%Xh>|Ls)Wj})qw?=#SB|Hx
z1&Z+->Z~L$=>I%<#JsH2*lG4s5xe?dpZ{b}F*VxsOhCDYt!kd}C3lIN-ib>-exD+!
zRo!s$lHSr~Jzu_ZC>wF7OMCSA&9Z)YlT(HNl4u@VZAY=|p4~rmKIhA3KF!{<K9K7Z
z%aoM3Pr<XcKHF|4z&<Ncv0usdUhdLdp^PIj-S+}lT>QDV?GW>+&$DtHqv!FjKPkt0
zQ9Izx$GNj7Gnam2Tr29*VUs@H<vmaMMgBzri(=0`+x_jRnnPwr_O}JXcTVYUNL5{A
zGAZKW713ipUWYCm<y4y7c`?-E0{7=8ZUgDx6O4bYKE6XwqN`IzTvjTB`{6W!y<b--
zxV<R0eb8g_f<1(H;>wyl&%N`{`4qgY{ljflH8J|aoA2>z&Gq#s55AxA_#kKJ!T#=B
zo_ee|mrhO#+%~DBRbA5BE3>j;gYy0`_wyYJPO?{~aX7_#XYh8P+N>X#8M{8kzj{vO
zA{+6#<^4v{+pL{<V;b*BIu)jKi=C8rS2`TW?Wgp3Nlbh29v25`r-Lh`XGE$h?y`LL
za_w=w3fmWkD|xOzDA&9EewWXx18eUJ>QxpVzQprSqQ&r0Pq2JLfx;Z;jtx(y{9G|j
z(P{Y|_AN3T&!dtKEDzb))15o%QsA42i~hNu#mm3Q&6^atC4;B%saEW!ccSa-#EuAd
zD?N@1H(FmPW31HA{7j?$eU(;lR%DRb$(Q$EE%NW0YU;iGN702<idjm^>-)}#9-Vv7
z$Y<Uc-KkcNp5{h|ean{pGCz_rQQT|lUeRfGuP5vECSMLykkt6px_@KM)-C^L9QeLN
zUh7&N_wI@RTk0lVV)L-Es(&?;=fmPneL?3#15&N7MPIc(7^?ontTe)B>1roNO|yi8
zUkrNJ&IslOKaF4!Iar%}H|dQ}X5Px@u|ZNLGfrm*^gUkHcUtwha@>o6(vz}jeGj?r
z?RjW%kB#r&x~ajU>rLa^uZcgk+hM|W_SV7OyleFs+paxyI~aTX$~2>gA;sssPA-_W
zZAuJt$@@~4t3ukWOAnbocrM*AL0{ebUgNpN+5E@%-E|8+x&FoTz?_+?OHXmF`0C4Q
zpnZLj!aL2xCmZ&8EeiCWJ$u(P^CP0)1DIz&$Xw95F|>4UsMNMc3r}C&V6x(y$VyX{
zrQ(n7GA(Eg6+dtxzx`Rv?u2Q{E6!T%(=N=E7A;lhkG*GBdFSTCXs=&f8SU5Kb6z;O
zW&VcO_YQ2zTKe4X>&v}*x#zRgs;|ZT_wGs4Q@2sQAg#Ao``(Q^tB;+1_48Ss<*60R
zE?D>IwVQpP^=<xDJHwkkatR@R=AWLGoqKU{vdDiM&z<#?jI4Ki{jQI{CViAG=e%1-
zb!>n`Z>YCy)ys6}l@FNzpPv1@*o!aqtMv5E4(pDXE)+=!P7=0Mp3Elkc;&IyzcM<C
zQ>;^-UfXRuw_%1|?zWnysH_ar$$k@hj;=j%S7WWt%9Iz=YJzmNyDl9JNwPe1b5`u`
zvb%hHeI%6QzPolkyAfu2wP^i^??T#VE;N6ABgy}*XDw&`^C_PvA8WL~b64Dl|L3*p
z^dkPU<qr$Wy^g-PIk!&y?#lYI<>%jhxjFZr``wrSe^}4|AHM(d&-eeW<!dh(^88hH
z`nQOAp0sRr!E2fC_sS1m-^cv-+hKE^e-o^*l(9se2VrPtfKkTY0hh5~n0<@m8yw}C
zjxZK^8*Eo-zqaD`a+yj0?U!_gs(5F8?Ritk#3FH{$t366z7G@U{F`-V$;0_GC$;{s
zs`^>{>tp4wFP8q1Qzia>shJ%UWH{wz$`-8;b#Zm|m#o;<#<m^5Xl}PM<L&wnwwHHk
zKewM(KQk!coXP~JDPncL6Yt4cMl5JJzVATwN|EJDLQSMQ*G+Jl60o%D-?NVvlec)?
zYH5^_)mhE$@~o9($%Zvkw<TIdF77(%)8V4XT^(?!_20T6<D0KP{S?WmP?EkNwnbZF
z#?}Mhd5{0*`NDtZ!-@6%O1!@qL%HKV><s;CSF0!d%Tc`fpoJy39pn9q)oaYpo~hA0
zahl<I$0p{>Y@V-Wj+7jE__U@-ap79?NfLj4rWmhod3B^I*l6{;yd`FW59d5FcX{zH
zQSsM<oXf$<E%7HW@g&`FW39eq?X}nUi&pl1)9(@?y&rFE7CN-z^NEI%rz?^p3?i0g
z#k+g%Tyj3gA^3R7^@N0`MDFa6<l>VB7Z|yH0(Z3}9&CM4t75<NNZskC!>6?3Lz<rC
z@MsnKZjfb-a11fmn^QKERcG}RtDSQe?l7J*EBN5|N9Ud-R4+CEb=LCu{vEeD%=Vns
z4VrYh=hg0CSB)2IDLv}TUp3*}wh6OeuAf&|R-t-SvEkLK>)tm%@3`uBv}nHZi<ifC
z2d`UwL|dHw;hxj;?D*4`c%DkRt!}M#ZvH%V-p2yL8Yvn5T2c0YzkNRU;C}oH&E=Xa
zCzU*1shFa+LMA!5QC4cs!AIwOkI63WTy<lU>hZevzdBRfTedZrC3<IMo4*zFO*rs{
zRpk5Mm1e#6-`6yzJzH~glUQHUw(l*0tq<PS7@q!=Wp^-TW#(~izb%Ucle?eKTCn3l
z^T*HtVG&`&4e5=I70rjr@9li>+3&r?<g2pj&+nJ?Z46UQf3J}qwd76Dyo+=8UE_Pb
zU$;!`TIF}%g_BEuPdx0`Jok+L`>TRo){gT3s#L>Uen<Uc6MR<M=bBf&CE-e(r>|4f
zn&9bE6w>naLqv5GR!BL&X*y>-Y1IywbBCVE#&hsnyg6|F;)TLHN$I{nW-oc9S7xNR
zRq?#nm9*UxPK0@%>}r)iBAuXcLMVIoo+%gaPCs5Xp~Trg_S@~RUIJ?;@$B5tx1jnf
zo1NT>e^bBhNI189&YEPt9sjH{BVMgjvDjl&wU9ma*0!cH0f{$R>n83GUA$|~E;Hu6
z^_jf0nT{1pI&^hKzTBXnzvOz=ov(!~(`M-WNKo3z{pyO)n(93Hhe;EXZ0<B{|M#tL
zuUSp3#FTDc_ON-<*VgT^UZlEevq74YiHbtii@==1R|^s|J#YKY-l>*T?Jc`jI<#Lj
zWa8zZk5|?$Y@L^~!;U3x)q%C`%%ZDSczK7;j=y+5d!x;133sj?PeWNZq<-mJDzAA_
zLFV@C%p+%$^_se53nN*!E!w%0Rl>7c<4aZkl$KB-$4!41Ex%FO@@igFncbQ%cPjds
zBvMtmZuF%%?)|o-E6MD2(Cx=RmK~{l({r~j_QvVgYW*=e`>#cqZPzpEzn!-(H{tg_
zM~M@W+zBeT-xe}x<?lZEBV4yCaB=cuAM5@_dp_Ql(~WVw^M}v;&T;d+i|==_EtgBa
z68(bNZMt&L2MwjI^R~p_UMG_za%-iN`=j!nPl}gseLP~(cF|hhf?Yz?&S9p!@uzm5
z-3H3{w|kZ^+v(i%?4!l@`w{EhB}_LLlx<MfKJ52GYp%L>R&L3i&1Zkk3z*iob5VwD
z-L%cR>mNk_TQ9@0_1S@&@<OGIQJvM!4-Z^Bwrr)cIq#(Brt9vU&zzn0>d#(Y?(Nr|
zHz?eG_Ch^(^Sx_6=VeunRAqcx;bnb#Tgs1%HLs$Mb)7og^L5VJo3gRLwz&yS4_k4y
zvS?yf@uLXKyMHnsP8X`4v6kl=dzwv$<jh;$mzQjD%1M>k#QBe3=Ff|e%E}UsKl2w|
zzU5_q;pKnExzqSdel_~3TAh`D@r8MA9eYXCy<We)URL?~T61IU+2_18$arYBd)0Qf
z<x*bJ$v5W;pKf1k8?!%+bLr2T_=L@y?rpa942g_g(Y@o=*W@*^X1}wa9-O}Y{oa^&
zLZ5dW%X4)U{cEc~UG-6XtO5JBsc&T^-p22W$-8dznmeS!`@@1_AJ3;cx8-*;o~Zl#
zQ2tin#~3WN5OMp`El^J-dK+b(uWSyw$?+S1M*p0L$(#~;Oi!4Wct6ndP>j45syXR@
ze9PgP50hO_Z{oLDV(=~_QsnvHv(4K2|8>d?f6k9pdh$Q5@?W*#>bu+i2*;eUXWsF&
zKVE(I|GMb9x~P3Mz1}gwCS{eoQry_HuRXc^spQ}1gA(bVt?cD&ZvElgc6|T5y6C#f
zU20nreJ7M%k?ZN~jEi5|Jf-xbwbcoe4{J=)PTtbLzv<e#bx8r$*2YH<b|sY0x%hr(
z+~wA*)pveBOzEjTw_fox!`Jrm%8Bip<~oK|)}Py-&b8yob#c4toW`kf8?!b%yRcSm
zBCDs<mE1Eqn=-;@AG#Oo9(DhvUbm%C^|=@KN~V|Y$m9Ihl2RM_YV{i?<_kIh19TMS
z8EWcv*mqUxXxcrDx$tkU)w0slGr!fzR`1%`mv`Dq?&oE>;#AdT3$7oi`}e6ddYftM
z1CMhHnXK0d7K&+n5YkBXU~oIGeusPQKanSM9h8ipTr|3pd0zTn`<bJzzt$|*nlC5M
zf6e`F(IS_Vla0Ht-03<{>HJagqnYuG6N@sWWAzvB%1^dF5g8~nLFa8s@}1|)Pk65p
zj%#^(?AMg173+W8{;=qN@``=;#V=KRwcdBxowjKGy59+FCI4)ln7-^4BiD7~zxQ?<
zh4D5m&$8R`^Gx>_jR4K6(B4Fg?BD0Razp(T?oXOOe`ov!rH&Jk>bth`$bwqD4RYT%
zzPl}ac(Pa}U;mQHPP?l7K041hlqRF7eMflO%Kfrk*JrI47EX8+qP;&|@x~h2J+~KK
zz1n<a^?^S@{>$XtrPL2q9(*5pdC9D@NjfaGZ%b7wqtZ?6Cm3uv%j>&m(bS%nbBmwb
z$*2l)sw@i+{p(dHcl4BI_C(PGpM6<ZcwCjrj0-+=n=yXU=ZdWK%w;=O{q6*xnmP5u
zmIa<kLXU$CtX9jsIqDpK^uZOUQs2O1s-eG@?1^UDvgMVKgjB?qhPOM*PUajBnO#^D
z{K(VWrStw4KH2=)cQ&k;pY_+Z)Bms`ll6@5eRJ+fYOL6LdsnRQ`Xx3YlTXX+?zLRS
zJ@d5U+H<?Y(;gf<>$^kIEbpLith4+3<Ph%{Cp^-*XDQrE6PaUR8T$L1RN}7giO+4d
z(j>Wq-EPM&JE>_T_`E0Nnskp)d+Yu4k~??%SaT_d|GIj`WKZ|SmAXeN6DxPNPn_kt
z{OPsOfS2kEwM+a&EWD$f;v1$`vx>avx$M<${dKvc)sK1g@n802Z;F~8^yT-7&<5*5
zT@9fdK{t59Q?51??fJ5!q_!=p;~HP|&aZjuA%ZzkH*>zWY$>@=wQMriBWB-cmko}t
zTl>O7`bg=jgqKC9vMe^6Z$G!uuUfhIrSaXbXFboiTr>RmCF<~d)oUM|<d>X}{CF;M
z<JlLt{4&42Q`=f)E&b#07Kh-?i%hh)uUqRVxa96smG3<bLJz+<FTbSpt61yr954Ij
zm(H*JvU~ZZFZ*YEZ9l*A3%k$Kyxz2EzJ%*3S07&qjQ*eV*hjYcTu@qj&E-O`<(muJ
z=GsWtZr$>G`u8mXcKY{&4{U2HuC87ZD6TGdweFbD$$*3XE!D=W8RysjIVKVCOS#%`
zmgx=7J0i@13*wf9<%mS7g(iO9nIxk2Yh!O=(c}v<VVSkOoa_2K!uH?N&^s(*So@^x
zG2cP|-G80x3|88N*(;>XPq}Y#Vj|<$m)!oce8LY+Y@BKWSYGDIE^*~{m8p*sD4Qso
zE;nnL70;Aa^S<oq<NkOfQQA$%cs@t$@7)@E?egE%2K~tWtfp(l8Mi2r`)WDs(Vd#d
z8B$k0^`E`%Y3A>xXIH(y`0!)G?rA#v=DfQ2yKNEUZRfoL8B-5l-QAV3vERrnJJiCp
zLMO$_a-zK8!-v-{*IayPbS}cDb^0^a_Zi~l7oW*47j@t9D?L>5OxFXGufDrAGH<*L
zvf{Y+D30ZQ$@TLew;gjeyRFWbFkLA2h0yB9Q;rp!TK9Cd)PC*jGrsT0R_~4#Vw<_I
zXn#@mP0?D5T<7`QJTkijUQen^4$)`35Xi^bfBV(;3#-&$<t%lIj@!Sst$ek4#nRR-
zm*=KPhwnC>`uR&4N86k!bN3u{-uwLNog<<=?=Hl$zmkdCQnJssci~)ZnNKtG0;Zkb
zH79p__DWYK>s?o``Yi62Fn)C7o`A`l1s0oZD_)AOa(=r`!}K1j<kOmXp%B-uBE3`A
zCq?(Je-SL*DU!$@d)UM)+>*B}#Gz!L#O7`LcNxuP={_AWBV5OK<0sKew?#GQm#n??
zP4we2S&qGXIg_{(x8A$o`RLBo&&R(OlvZtY$hTMRU0JtW;e1jXgJ|oQPKMyV71mo{
z*Rt*~x8>sQ);ngrJLa5UyZq+v_50qqyp?4w`g(dF*SeS2>w~5Bc^jDD%UQMRRWEKo
zd(OZ~vHvKG+WLs9iRpfYJKWBDD>z87oe;AKI#{YZ&wVyugPeStV370OsnvIEcZR?8
zdKQ0C;mr9v6H4-wH)dI0nbo+@@p5dh`Srz|**(=at~Hg*%`D~J@#e*1(Y^T`D}L{<
zm0CD6pZDI=POBp~{(fY?zjErE@ALmLpEH;HgIP;k5_PPCk%c+RUZk_pC-aog^6q=C
z`Gx7#m3<6t{EU+rcB;;Zp3r=I)4yQ8lfUJI?qmgLe&(1dZ`^cuZB#^z-6oG;ALcjP
zt@z|Gzew}{)rX(&x7+`<`14W!--4^tLa(*UKmYVsa9)kwAB(@AeKnKs`|8*ySJ+2Q
zJ)Qq+^B&vF$7g=B)J*;UY3;SAw-=m_(DlDJ)qGV={IQ!J_Rf06<xyAXB`#_D%DaBr
z1Le=B-?>jtSTM0M&e-wQ-2Dr6nQ{vAo*dlK!o9t4S)REJ_kD&)gZX7jk6u43@rsao
z!Cg7;jp>~Cf8*`{h5eAMs*74`E}ipG`=juV2W3&(V(%*A?HbehCT%F}zPGhgzBTL5
ziDg^P_;?BT&tcxiqy8+&R@LS-hegOk@wsB>g8KMGWY?aZeQo25mb-g|=PnT3B5^8b
z`Fz8R-#*D5W4zC_;mR#@A)yIIC+<tB?UpWoG_xz*^Uuq#Yx~wTE#iHuEoGC%Jx?fZ
zMy=~=pS^nikN>3272hcGOrUvvww2+YnzK21f11lZniuiEa&fTJ`QkV`VCq42nUBxq
zQw~;cO2{|Qp4`)Vx$#=nzetf40u5Q&_t)0F?l{rP$ay63K$P7>&KnEAUu5X;NRr{w
zE%kY<9^UHTbKtB0kqNC$I(#vFNms<C2x|O#AZfk1|6!7<+v8&QiJX&L^;u?rP2)P>
za#%rny~ngAvro+1I`>tK=pqI0;NXYa5$b^|2JTx<+?K1J5vF{lyx_2h+UiFc^R>>1
zZQOUa>DI-XjI=+CzJB0+7sK0}njU%Vx<>0q?Ufvdwp3<RTO5DK8^XHZwM%F2VzbQR
zYkn7squZ=<4NDhU2v;fyEe_jcyl_qKtP5v@t+WNE%zp03t;A7z<g4Nr%ZF^mK_+e2
zG*3MJ8~wb-&sWVPzT)%>cB#*Sib+iE$_J)*W<*}Ic$igne%@OdnH`_KUAIT)39!5n
zV4ibi{!15@Ti)t3wkOPvS);U{J$hl_;fLKyZTeN>cAO1&*bjPme39*GEq}&c`|P6h
zdAB~d=N^wdj`_tkx^aAaRQFtUo7(==xzhWLdM+?F-hB~feet6C^`5!=deseCE1KJ0
zP0v~#Tln=OQ#Du6`V8HQ$+OPQQa!%)X3e$x4#w%*ZXA5j5}q;hTx(KV8XwoYSJjn#
zN3^blO_!hZlsW73#is{;*&fm2^Ev9hz?n@Yf1X)}z3beT&`msbzXXn~Htm15K=J9>
zouUT1j-L6c7p)?UwFRQ)SWILrnDFZPqD`k1+p;(&xLm&&*7N4r!IH-u)!7~9=FcPf
z|EM|`E8Hy>FnQJI{o>Ja_d80AkCjiEx=neOIahOky_e3iu0*lRH+!@%-F9lcUjL``
z<EQ=m1Fwkj7oHBhnmzq`^v{l?o2J;XWu*0d|25Mp=jM}2E4JE#H#Mb;G#BLjvALMi
zQf7H^q0EC1(TaDj^3PjoyUs7z;fR#2e(v8rZ=N1`)W3J{1J-sGy^!Pncl6|^Hg=tv
z^Db`93$CYLOYa_?_%iHkv3p<QWy4sRn|E!Cw6x<kWu-UVaAjsV*gH}ELe3S(!&SFl
zKRH+TA$wcHv7PTdjYa-m{U?1%M(y60>tEN*-9OD(y-4A{jrIB@|MMRNzD==l-@9aq
zvsPn3fr$S}$?mMk?LzUqjw}HQj;}LVT5R8~mS;W4%i>T_vD)&(&qoaoOv((#ch5&z
zd|wo_(aqjR;L{VA+oyjn&g|e`lF=IHyUcZ&P2a;S_m(@~xM;BX;g!4h?(dgX-v7Mi
z%2e*nEY1B~rSg2Qz8Z6{*|Pe%l#153>lUoRU$lO||J9PXZAZ(4_q$pW?|IuE+VDwO
z!06m<>G#rEC5~6Ub?=CrHp|;okdSibF}nlrw5`+f+;f-{8JMoj|K$?=RMSE#(@9qA
zdiaL)#fF*9(VkZM>p#tT`$Vf%{?G4FF5P9K@tzNlJ0w;$9J+0@MgRH3e<J+3yN(}@
z*G*ke@Mxh!@#Dl|8~=bCKgFB&SYO)bz4EW^XCc|C?H0e-T4(I8kZhIMcKf;JmU4}n
zxxT^so@H_sUd<L+6kWe^N^)JFQ@J~T<Du)17Mc48S4?i-FCuOz@ayhvBZ)&Pr~jwN
z><QakxQZp$FkbW4jXxe2pDotEbzpKzr^aE!RO^Bei{e-1Q{#?>9^cgW^VO$q>6P3I
z_Zz3k+q{dr*8c43r^Aa~6;80;E)K6+rCG7_*N$0M>kS{B+wODuUhh^n%{1SwN4sOh
zTpmoBx$b4<h3P*c)^%T=aqgqn+6aZC*=OR4u1Rq3*zziX^>}z_%qnmH`M-BDhTFHE
zanRJi*&vj^Ol<p7uUj)z8jIHY&Cxj9cC}5#&3uaBqB?)K6HDeSed%@OZSMK-$@io7
z301hwEPLj6_R}J^M>pDw8%iFe7dPziOO3Pr<dMIBmBHFg*RE6sz1_BU%C=KVlfrJi
zYjgj6_TE<W4Q=M9isIL-Zne_k;8<R?dP#r!ndwVSOrL(vZi?KPCYf|&M%q^XO%*=h
zPF-7+-uSRHE9u?BB^#rc>^XjG&U*=$X%DJS8wQyc%l6J_DfqN4+<Q^!JKIc~7s5|g
zt#>UsIVpIu=1H5cM#ptqO?mGhad+I_u_o_Z*=pIR&$A!8efpUy|M=c9Hp!Qp3ZGn$
zsLojPKBuQg{$3k<%cCsYce4U2zOGET6($~N`R%CeyN_4aY4I|61}|muN?y@-D)?z0
z|0jk58O;~x%zm0mZP(k!<-U~f@tFwelvm%QvRAIS{(W=H*1Mh7Zzju3+EY-x@Y$96
z_|tMs)A^&*ncF3MW_3&HzML0rRO?wK7RB6{=ac8_l`QYKX5TkM52O3mUrsIj`tDxB
z*T6frws+swp4g~<`-IIyW!ts9X^~gI=4~!sSh0V8+}251;_ORa%su^NpG(fSEqOLt
zzg9*)tv9=Wtovh3bDe;8n(u<f+Pl+^`)Zu%De`r1o2MTsRC;~Z%J2^1lkUQgr!~c1
zS)2Vh@wv+B&(2$8=bKrF->P>~sk{4osZVF)Wre0R-JZ3crDBgK%`4P<xqOc4Mq||{
zdxSPjEM?o`cj<~TQ%Gu3`=Ufa-Ppca`AQ~9S8ZP;<=t)O{+RGl^SO=Z-}Iv?)6WW9
z=9%rVUMrc%eQ>`|d{gvmZkb&(bM{B=ZtmfzxgGztcGmtIwi6At_iL^R{nC(o*!q6t
z{lLGWuO~iwx1)RY<GB~S-ucH`9MzuJ_TlxSrRq6PFCVd)wn6pg0o%O_9t#-P*IJu9
z=}KQqx)mZ-*q{<&EgrOa-oEC<i{6EWyK9X)Wb<Z;Pv5>;Hp6Q7S&j0wtZyf)xQqV`
zE%&*#MRwPs{H?tb{)YB{o~{1>%<DDp``&Y$&*R=d{}L>3*}s3&s~goX*Qv;5TwQot
z=y;ck%bu@lNnzTjRTpbD|8SeT<8;5?`P%8rcApV?Q7rM^%IEaiu(`LA%2hniUR>)k
zJtQq_U-bjQ_~clVeBa*RE#hk(cboFt{>Xp+?t8b@Ud`KEIA?Lso?5iI?rK-+?m71-
z`8fxtIDTK3-5vcg+;`%>4a>7FYbq~HIQ%SW#{sUSySc|AwpZS6k~*@&VOsO8&y3TT
zruJv(ZM!QG#rQbp`&2c?+eJl_-bQu3$=H1J%%e$9cD6f;JTG&v_vbu(tyA>x))hCb
zZ>+d3e&iX4>WVCnTb0We&R@)Z+HQSZ*n>Tsk2hz#b?vcpG>TllT*7mTs+Y;_m-@*U
zI_I4*$^5_bqHXMv#e1&oT)BJ8+|T!KoRGQ9lW0<;=vf}fI{#neY^B~g@Bi*uE;ZZU
zT~O!!+t=6bJ$f~#I&y>fdiS|!pJnaVS+(HrtGTYfgID}KzuX|(BYb&u<nb);^BQN@
z-Z|hKI_XAs$kO*LeYNXr>I<cUg-R-S{rEh~@SjHB+1)}FY&-naGSgnXs!Y9g|NjU3
zd0}_cu(ZBSh?)(zFgCNZLt4rgbu#Y>bf)H7L3~3aKcf+2k@th>3C^o;+k$6mTDDG{
zGAnEQbRIU(WVdZDp`xq1w!S=gIezlIke|z!XPl}xmHj`bu4a~9?Z?Ss#bJ^D{?9-C
zW1LqrZ=T=&`mj@(OP%U!HXJ-WS<|0w-l6n}@Kf29>2Dr4dp<oD##GM4q_m~=>YP@I
z@N-@-7k98%eU6#OwbxQ)pO^I2#qIYUUI#6-m}nB@a*svT`Y_iCw$!(WzPIYPt=gf;
zxj_ET;>ekG)9?DM3EMGs=EBQ&)|Gwl)4RdGWQJ?b;`9$pRm!$@73o~l!?#@8W4dkg
z^+%VIFHYF@B6Whq1^s%PzfJ0y>F=MYw2JVun|5nmYtl07Vu;wd;84+O&tB(uGM6sK
zy(qe~xV7oRL}AZ;<@xt-N|b2N&bgA)JNbd_1+f!3Cl;IfO^`Huw%}>_#$vVi+`F3j
zWM{K^RYtY4L>AU;73E&P;cuU9@MX4@N#bj^t-ZBL>i6E+(|M~U-yAYm=UeBt(bsB{
zC7;`ePyYUsXO((AU0%5H56{tu`~TWR{Q0dNI>X|qrMoJt+KWinyP^^W&D*7>cy2n%
zx|pxntn%<%-q?e0Kh2CO`6z7{ER)f9b<WBR`&!+o-Bk-6-xj8H8LM9V;<o8)^vdZq
ziQn$!{Z%+9X4a=S{nP^IH1%?B<+V<87MDd=^B3&Bd`}`x_VTs3?~C@X<hZj-vnb8s
zPJ8+)TM?$Gt|v1q59wUGRQY03l9bq)-dC&4W#d9NZ-4l;efy)^GIFo^Oj$m^Z3(ik
z^$ea>%)$4>rb8k2-054|c^8+4Z-{4lZkM)I+2}h@_|3}|ALR_>G*_=xbKJnD%YJQP
z(k_noReI}pX(g_3%8-Bkdg<55!?9P3I=f?Ru4gf@Yk2EDFfw`Y;tjLzMg<v9=bhr(
zDZDxyySDADOQ^I=SP<L5s~x*qJC9>WT=JTC+*yk*&spHG_A2Az#2+V082SJ3sVX)f
zS5lF#-gK5dl5^!((Xg*4mallZ)_$|SL)5BCCNnp_HOZZy9ox8KNw)$2juW5e&D-*Q
z;q|S3YN6NnX|Idbeyp%-O6d~2%b&L?ue8<hp6^y~&m&v6Gk5N{U1fT=ljhC5S9sxC
z>g~9X3x51{`mT|=u(R(mqm7Df^rjWT&ZhG>i#lf;I+(65TjX1n-<#I4tgc$hB1yvZ
zDVM(6HG|9R+9K^kiazM&Zh7T2%U7B=uK8u<l8%kJdz#lzn09Srm1vRfvww4~x2_EN
zw2+&1^P;PNt#u@}oe%nQZDVSsMM%`~U0dTeTuR&%ukH7`H2Qj?-?1!-l}5Li(>1RB
zm74U(ar+vfiTr<$J=6N<cj>K8_1d3S!p&!|^gpl@dHC&$d~w85mAPUo)r@ohFWfO*
z`ugi>@w@nsT-Lf^^j5twboPz91V-+dz~X)7_d6vvpRU_7yZE#AkIKEe!I`<9jMckV
zIb4W+x7t?q8fVA-(3lM^V!JO+eS5?9lVQ{G;=ltA1_7?+Yb8!T_nVvM*IPXM`dUt}
zXYSX97Ay^&eqdQXd-BTHyY4N&<?*Q7S}#cH!=JX&7X^2ph;@GwJ0JA*>t41gitN#s
z1N%=u+`!xPY1u3u{~2Fm`%a~Lo^sZ6-X-m$|A=c6TR_VhV+BFwF4d%`%`X!qcd?u8
z7g5V<Fv#95R5ZcJNUThP^YvE8YW^j<T#q_e_smglG<(nKq8=&s!c!sZMTho}jR#DG
zG}IZ+Xk7SM8o_#Z#hV~*X-=2in#&u0o$vmcEWc~v0p?7Oo(hefpXKEjZcU#2cAmq<
z6U&$B+I_CbDd+m|+2CK%WiP*-_CNL53|IYC5Kxt#@a;<5k+1`*hqC?KV>sSQ_qd&!
zck2VohDq<QxnADAE&O$nDW}q7gBvG*ciI+x?eB8Ga3_o9+ld;_<a2AkZAskb_TbRL
zhrc)7el?L{@ulL_e=R%jx8CJhue9{(j<7uXOvq3imjD=oWh?(Nvw0=0RWEYn}l
zo^zq)TIJhK<`!S9uO*xPP5gNLx0u@YDzlUS=1)9kcy{#%7q@m+HI7w|tiIwG%;fCY
zs<#-P(LB1nsbBir^V%zc=dE7o9&SI-ckSHC<7;0?FVLQyHdU`zA}M_P0gVprxi4kU
zX}pr%u+%KA?^;a78|4P?*rL$q1rIk(U+A6t_l?i3qJ154`*<g>QJ;M%XK&r%lj|hz
zH+glbxTxNho%%;1B&h6L9s|qMt?yn>NqT$c?T;tbAI~Nx-8Ve{Uf{D&`I*msmG=zK
z*9(6>v;L5cpw^D$d0?@B#|_S}HEG!8a!bx-=N+vT(*n%d{_$<+5Hjg^m=T^ZN4!a*
ze4*ew(HqZNGd(XjRmlB5abHE_bNo-8-%WRXm&rak-?}0GZeL;?N7SvY%x!%ePp?%l
z<i2RXxj!=M*=C-}^(JXIPao&`sb}cj`1oS7d~@WnKfW8i{_}3|I=DEjW2s^0H@<&-
zddDo6UJ=zic8dG<diy`j2@6VOu++Fj%}N;?m|7s$xJRQ-_8D&%1dn%R6~{L;@-aSQ
z*l9U4dP1{_)XyaOIo@&hYqZ27Q-Y&&6g-~%+&x?8+Oa6RPakeO%7^@%Z|4&HZ#`f9
zez|{tDt|n5ufLFWTI==U=WA>1qxQ~B;h%rKzD^@`-_%dPN=#PTL@W+5Siblx-@H@m
ze?#xBFnY5r#6(@#{+D&Pa&F~g+5MXPBVIl+dhKNVX6?RUo96pFST<L>&R~webAWvf
zyY>P8V~4|wSJtFdecSbC@3j5#%NK`lEQ)72{5ARUmC(LJQ+GJ8%zvw0P_l4pgVh=#
zqtA?1451f;d0aTd_`;V7_*`{*Y$v7o_Upf#Y4R4#vEMctsLk{L|Lc!*g#YccyJOZK
z=Q{jUNBi5#i)$)RXRNrkYfs>8)4u8cNtJV0-idJSe-@+G7#DarcP97kZJW;7H7<6z
zopL;Dag6x&l<yb9B##{1X~6bp#qNz!mK`#A$;tWdd~?)ancdp=@ZQ-qNn4HXuYT3F
z!N`C+k56xwpZoIza(?0I*{2x(S&JsmuK2@MJJo#sN@?}DPU+(4T|L=#8EF?UHqRE-
z`=YEiA!hF3>I<>$mD{)^3U12R@ff|ym}{2m{IG4ugTFo2XT+@%UccY9Yq9Oa-u9~^
ze;xJ&l-o%$FLInIx_aS7=T$GhMO8{T&EHm7e6L(i=m^*5e$lg?i>x@?XO&)CEOLx5
z?zX{t8#(2ZYqPdw#vk)FiT=x7^dW9z>3V_X+pIbp14a1zj{QC>`q54L_@e^8JGXy{
z+5g(+a4yH;Tt&|ot7DM|IlcNGC!gi`b|~LLTtavvm%Dw*<8OP}HI^*&yH|h4ta-f=
z^M`%M*H4#P{63)OjCEH2!--)>b#J}neUSX`lZMW<(!00B_eTDjb7{6w?Nq}<n{{_h
z6$@H4cdl17&*Yi+yB_GCtGMA4CgIKbeD|ipw|-aMSc-TaT+cdiCh121y!?#5dBKky
zravlPy!UkK?gl=mMd2$n&0<w1Pbi%GlI3Q9#>%Y;x({w|e=RAp>1dPJ{AADdVXIW1
z7xY+Z2+v(AasDCGy(f3SM;4o;iR`$ztCX>zX0!PDYS%OGM4vccJnZA~al_lnzj?*!
zkEiU-@bQ25CI8;9XU9IjJN@13-_(2FTaF18FLjL)4=dOvUUNY8#z9`T&xhjl#TScI
z{QKltS~2fcfow^qL!0^MT%mx%vnT#FD|s{pm)-8t7F%^P=M?+0tg;-Q%|3U0wtQam
z$Yrg;A&&Q1dzclO{~l2E`TH{@dw2HS@}=c_!XHYXNj%yk>au<f?{B_i(=!EC7HApo
z6n}lx&;3Tz#<m)sCzmbaW=zuLJkhFh$wObm?6<DQx8{P%z>SC3go|+Ro>{+M%S>Wj
zrsfZoz!P6i{qOUWU4A=@;kn1cX~r?{*eX?1HZ)zH&1I&Qe{1`_M}KBD9gkeEvEwV(
z$Hn;@r)7tgKA2FwE{6a9?JwN>|4Rj@pZ4&2^EO{FAZM-SJj16U`6~?Wt=gHfG=Gw#
z1oNq*8+T5d=sR=M?#qk$JIuPY<|;0@xn-)trS8}3ubGN{{#$k<$M4^V`Ik73Z0b3k
z8a(6hdiK|)^?#mz=s)bAk`>LCce6S;d1>d)FZ|8{+CGho#WXhv+os;Wl={$cN{meG
zSLNmv4Zj{72-ugISk1mJP5xNcWbSU;Fmtw5dwr9SOlFm;jf_q7-l!wIawD7U6{fiP
zlU7PzwNRbKv4Y2E_2V`!&yJwBEtg)KwK2SL$@-fj`F_Q}s48={cfWMjUD>%TP2bfv
z{#Mjh;RQd9j$Zn1F8usr=8?J`AuHm2H8p$!f-gF1C51Ge?$SFuPuk>Y<gfC{dz_fR
zxU@Rl|G4N(Mc45Ktf~)=T-~%f(#y>H+Lu%3)&3eDO>JX|`k2cSHNEG}kq;|-*PdR$
zyYqXFa?~6t4bu-wZ%=(nkvCy<W}bK^+AKeNfB2{C*DPLVo}a)boiP2)zTEg(`41h%
ze!n}%wx>|2@8I>VpO*Q5^I@1_`FS>@%R%$CuHoNW#l(7>KS!KgBdi=P_;iW#g&$XH
zS0C)^T^`8sZsL`7?m@<Ur~mGBF}Sg+)=he&_F<LH&e|S6{1K;AZY<J}Uuei9^YE5A
z=P$=c_l~r^VqPUJbExZT=DYZ>lO%03B<H96JiKk&ohez`7j}xZ>H92hk6Zn>?~|Y7
zEzdubjx7nkoj2uA)Xys^c3ao>&TF!H``olE{onN!{MGy=Eq?`XmjrI-I1p}ipmlB2
zE@{>JhNvLH5Ux3Ajx@JOe`;s&HB?WSGyh`EpF`2>KAhT-y7a;n??~5o%r%z}O_-={
zxjjemXSR^nW2Px~9~l{0-M=ZPxw)+7JfM|XV4Ud^GJl11)|uF^`(7Jgw+Op6$&b10
zRnz>#dCj8puJT=3vym$$KH}r!#k@-&oqn0vbXGh2!tJH|Hm}m$+iE3s=A+5_h1r{T
ziu~*E;w#;LZqegYMu}!`ermDi-4OAeBz5Hd`@=aNC-g+`U)=X=+eU?BJ`a`|R%t8N
z-TP!*+j8w%+tx6L)GsE>uWV$S=CR;<c1*xEEB>B}`Gz+SthyvCx;pRJ(NFgSX06%(
zzx3m$`}V>A1wo_Dg~bm_?j(7vnHgnN`NrtX4VBZ-;pP(ktskR(jrM)XDw~&Lv@KBQ
z>_UV3A3F`+-r|4yvR~WNtAE9%dGmKDM{6*o@u|GFpE)P2X`b=Y*sIoI`fopYU$|rV
zI5x2Frs3qQZ1Ifay_Q?w-6-*h<~ztZQ~5HZ(YCMW7B;V&^GtBgN(-*v{LQg%4A%Cq
zKI*w9`a*b1!^fO=_bOyLzkWWc#kNG@Tr#h$rP5oIPiol~#~;nVo**?p`KZaE!w2ML
zOZ$$`;jpSbtyIA^>w9SJdfD=e`E2v1y{*{srR+|S_PO2joA#}otNMWHii|~M&GLJa
z*A3j*C3Ewf1<x${s@BvKdATsMKkWrWuE?y^!0-A}OX~jJOJs7eeWCOA$lHC~tJg49
z-?}|5#Nl1i!c)&KC|zlm5`V;EwZmZkzKaq!-TRK}p0DHxxiF#n>c-XW8v<keUiLiT
zWYPVYckt%Z$J^Ez6;)nUIO8L8&Q-O0&W?;TdM_t>E}K)n&Zkwcc+aNJC!Uf1QELKU
zJ%1AM+vjIZin*M8=%!7bxj(m>ZZz3E*LeT6BX({jlI-8qrZ=2P7mWW{r;>1^Xbxj}
zfSg6}?JbFs4T%>0hr>QHPMl+*T9Gr|>J`UD#rFb}&n%Ue+g*1#eoat8&ew_4R(rWV
zu(`st^wZj#>-Rs9vDn7AG(o56tXIWSueGxir)sFL*z0mTXoZiriTRndF3~T~Hoi{%
zEp{eCJ2Cm0m&Ix3+hVz)cT?7Bxc|1EsygG~mdDXIR()Z)&hSHvQ<+6*m65a8#?$X~
zre9JivAA<r%qQ*zTU*$=t((PzBx7qU*T~ALE?60A|L~7kDF5Gk{n9(V)*PR-Eqm#S
zb@>nD_@%m%SNxbXC3K0so5z6_r>>b7m#;V`mr-Bj$d$m(kiy{C7S$4DJN;*>Qfsx3
z)wXXwmwBa(<8Lf_VR8TC;uFT!(<c_rxEsZt`9VW<kAQjd%H3OQ)${m_@9DUy={e~v
zxg=mdc_Y6UPyVwF4|{91S&e%w{|e0wjhiB1+FK>Dq{D{2-7D<LQ9Y)7b*aOZ6ZaL*
zS&|<tYmmDlIrK3@{<l>N%op44T-Ot5_LoN~Wl{kXQ><L7lE#&%lD`(48;f}5M{7>|
zn{h#O#v@Uy)q6^hR#(5M*Kk|&Y4g3Fca5bQ4>!&BzVhatO`C#Z*z})iiFbKyuCF|F
zHbpZ0YhX~2&dkK}<oq=U!wTl#o+K~&{MB0q!4Pq-&Fj{ryFWR%Q&hue`*NMBR?l|r
zyObI>;Rlm<<s<f`f360;&%4iB$b5TO`l3qx=8wS!?qMQ-jUt+>-|oJ+d{ugf-P$9z
zvjn!K94=YGm#cpFlECxBmu26U?fsxN`Sh+W`{L&$JMH%J`z}@ay5XJU<_$aA<$fQ1
z%hk&p{^*yl^XCtjKHf3=V<)3K<?Mn*|Kkg{6k0t}zVBbSrSOt`*W%tKYwNoz4HL{>
zK2-Y5{rYhwyJAjIofh-{owuC#M0hR;JsFU-Ue;^Pet+wz-Dg|21ov03l3VfaPhQ+^
z{jU#<CTR-G`t%lhYj%1k$)5c6(*Hl}t%clmnEeJr1L7vijSNhXI}UH9cIPRZZ#W<R
zg8fJ|-~EOMN)CDsyI9^eiNu}Wr{((TzpI4|U;Js?liOOflT@mgKb`csuJ9S(f0-wV
zo&QfZH~;uwRPpckhtH26o(`}7pcJv1zkmJ5&w+YMGYTqf{#QncZPnvm^88`Ofg_@O
zd-BB|2|s$3eP?~<{<?$9t%bwi6xXEO5taY#l%2M)v?E~4;rzENzkFR8dvT_caQvH@
z29NTE6W!K{wO-L_i9PoDtl{*|(kd1WWyV{=(oz}fOipe~3-^Rt9%}yaVYA)k!pqEG
zRLt+p-2d<EkK1lrd`uo~{^|K%a7C==91WG(nT}7+$SAuzKFELUzI2AtvFOK!MJZwH
zgV~;M`<$s;zIoa`u6L`yZ@&?)e=Ex%e%~x+qn#`>llvzz*`!R<^<rDgqNK&Ve9cLn
z<%>Fu7cP)pG08=0FJq=`h~4Cz{{5%Y|1CA<UQqU+g;%v}gSvN|{DFmg)wNQNz0TSC
zRps2)1aGHD+f!ynKKWtonA%*Gd;Uf7(ml1xse-&2T~^u?r!Kg@>-hW#CZo4?2G2qY
zA9?<n`{b$c{w3<xFFdspd3OJF%H#4ERGna~zI#&d>SOs)le{Kt+h3I0bB0IJBXMeP
z#U@#owHd7?i}Z9S$ZgrQd{tnT^4f{^ni*f){uqjUsgB)u&SHwU0AtKThg5a-<MSu-
zG}xbJd@SGBC!+2dI<xWEgpL$Gp_95NPJ2u|;N@QS!{)d0d$yLGr)~l{Ns&4Z5!N?4
z)oP{oZ<(@)Ii%w1krO>aSEtpbJ6kekEKIJP$m3G{CTnMyS8B7drHFc5Lq+oROR8-j
zIKyhSC;v%hu03x#wJG+wkCn4rU8RkVQ<eOG7Fl11hzsrZb#rz8{qB9GUS6QIW9g5!
zFHX|4E7Si<w?;KJB^L+X$|y>-S*O&f_ePFMIV9k*qQk!V+~(~be;!7wL`_%!7#yEI
zM?+t1|I*_*2i!Q+{&gyRJ}+s(>(qFs;kH$nd*PZ0)?HIOwz*hkh^1w$5xRAl?d);q
z5a(-QrKdP}>yC7JUcR+jD8+D(`AUx}ZHHI9aUxj{o;Rm4bPJ!lE6HAwIHgj$@8wn3
zn=h8ilx@k{WF@NCsL~hLYPP^8G4T1O=HlY2+1L3^bC&$!$=$KUykf0tD3?pRo`m<s
zEzhb%e!QP{%&y_-hcZ3ItTntJIqdw>{F|g)EyIKg&Y8{N(O#J|W9c7_jL&+SqJjdf
zUu~Ht*|$ddpXq#XVBXKKrs<#KE}gX7r6Z@uKDlburD?O23#VWHyl2wS{8yC`t5<DP
z5AsP}6y6p)<5llg)8zrLUrAoc%Ck)^?hrRDdda+Whvd=riEpcQR`+UrU(dF7d4c5Q
zpsyh}V)w@VabunwJL8S^7B=^t>m5C%)_C;1<$7FcSf=n)dqqv(q4RR0kNf;jZl5a5
zCC2sk$26rXVf!C}EKem`{2uPuy~8w1DWi(XcO}=^bNLNBzFIl#$>zUs*|cHJIfrY9
z<e5Lk78VIkSZK<#&9W!oiuIq?wNH8}foB;F*Ks#6t!%ybyx~jenx|Dv<+qnxX1~&M
z$gZ6-i+%6(YzD*V9k(5}dVPK;5nz-4^6&m<oBZv6f7$GBKl^v&{3R#-k8O<#o3SkK
z$kjVOAw4nSg;#p|GdJ$yH$S^$S@r=H&0jgc{g)iMr7L^D;cax#Qt!InY!0EEm5vcQ
zH=mWws#>+1zgRTNJ9UjmEQhb=-22mH=A3(5zVH2IuiIvi7r4FE))L6L{?(RU$YsJb
zE2S&;9yc~zUwD3Pgxf0R2irn8e}@#?s}^n#l4o77B^j$dDMWs^posh95U%|}iZd_f
z&sP0iyXVrfgYR}e)8a~gVtlu+Q1|Y%z$N-Oy*{qm_{?ta9L;;|f?G6Y4fo7exTx~B
zJ=x^rvt6$FvFhq;RVsf3>t?<Ve6#4fzjyU^$ytZ9xi3aes=C#AS5^8}^nslhE^%#t
zt9Y?)YIWkY@KwjQU!Q0DZr!GaGr8Lu<gOfH3Q{?>>HXE$UN>!WE01aKSZ#a#`)#JP
z5nn!^QG24)${W!8C`nz+=gEQT%+(1;lUA)Pb-Vmh@kSv3!PvF!529?BY_GLFGS|@S
z))lYNZwHI#&vO=<*rXQyF5-gI(SS*YS?kV7P4-@H`DCKz_1U6xrQZigKbySNBe|u#
z$YARGvym%~p8DLeSM1H!+}x~_=WgAaerq3_+nc>&x575Rl{!B)Z(YpgS5c|H%IkyV
zO5#-H_ni8vpms1g)GFO*hSVlLr=3|pGljMlyk4|d^mVZBgeh5RYWa(A<**)Ca~9dE
ztor8K?#pQmCJQednZHSBMe%{PKg-R&eYm@X?JHYeiILJNCuxu0tC~{#>ZV1WsGH`x
z=hM%+TdB_7llOM&Y<GHfTx@CcgxXW{Gh{AANi3P@wSIPCyVcc~@3t)se)C+TS?t_*
zW#cF6rCy(AX1&kdIq|q?=`+*7S*(dws~0hD@QX-#S-!~9((caHWqK#$?rk$$6_Tsd
z-JyB#KwrwWU8{W(qe6}{wy!)_R(84f(CgGq?-ZV&P~11AzqehyR=RJ}u1t2b&A!tL
z5(A$y9h{dHviEAl2@UO#x4!#Jb*{)S*FJG{V?h0jWyuR0<%*kc^7eOBU65bnb^rXz
zn53Rqv!IY)iwic!9Lzj^K>dzGi-p5Yhhs0NN9p-*RG2mE>f@LEXU=XaoxWnqy`VR@
z`#sLi%&c3ozP9en>hk&9HQzp)wfmutPo#dz4&%NG`+#HjB;EUjt*RzIylp?DJ8ebo
znaNo;9hVpWTAvc}_;+mAoSV0PKA5QJtv=_{|IIgN?hkkFippF$_uI8Lk>-VIIho69
zmnc4!n()eP=4~(ClsjL}ow+N&T<^)zGk4{8Wj>R8u`~7AJIDV&&e(3K+I;5j;kB>+
zR6LW@s7rn(_u|)?JG0N&M!bkM`fgiT@NA!UpNzT#r}o3QGk4opd_J?bF67y>3fn)2
z&%fAyBYW#P&iA#(E+MYM`~QED-+W;HO>9jZ;$|bwO;K->n{^g+#Ztrj+OEpR6+&OF
z9UeG0Zga@jEz+*)c$?yHyXAeohGwUfXY%olYaO>WY<#V9t;1ePvrf5sS-qO6=bz<j
z{6FIT+V?M?AGb`to-<frb8%W#<?+-0UR^iOT|WKse7fe@IdOgux$9Rm|C{zQpZSNC
z*UIm^)w#veqg|E%oNF;z^!;>9_kPopGUfGq|7<gIpZ+&y@0ltdzN2@nwUk=zdbM6>
zMm$#O-WDrXmJ!b`R%F3+)V=-lm+DXYagkRhNUT&WeWttHZhm0%`!y%;T@Vd<*_YW9
z(&WwO9JbOkwdcTj{mCL)EPptUJ@#i$`}O#`aP$8<@0Xt@i*KH;wfwYKqHKEi<MpEZ
znh%JD6`ksod=_?N?=ge9!i?Hq)<g?>ynImV^WxhV-n*txMZ-7kvE9lh+rl$f|JJ&x
zNng1vC(p8EUm<0&^hU#l1yM)EvadgI(oH*%7{q(^vtsYX-x2)zJGU6j;_7C|h`GDd
zOLZ;BrtbM>-?NV1Z9e#F%Q@zz<skxl<2={LeUB2_uw%pQEk|27wp`!BoMC)4Vzo$h
zgGl|VxGwwH#XC*d@?*~Kn5^>Vr|i?CKfXNrX}G+7#%eFMQ!ke?AFSS4^`y^r_B8k8
z=LMOskFDNmd_;F%dT_RbZR-EH@b|~o7hf~mSzSJb+v)7=D2~k+R#bU&uhh2C3Eprg
z|IN`?#(oj2tu2)V0=<u$ot<fSf6L;jVm<zAA}SwlefQx-s8h1ew8<GNZh=27PAUf^
zw%_T<SlZ^bqw4(TDVOA#KL0<F&Zf<6RV_SUa@oXatG{JsJo7!@*73|2Wm$N&MtfoO
zM+vJPCtieQXe^Ds>TG#;)xWBsa&OsVul&}_xxe=NJax}IueXkGL^tl{y4!Ly`c^_#
zewOlg&fYmZzp4thHyu8iEB9pI{5z~W&AzSAKKb^!)aI89-hC^d7=2wU{`dsxCBm0}
zN1wdgb-eVQ{-oWaQeN-Ez6KO+U$uYUox4Hp6L&|e&v`ff`ekVbYvDD2{;BNW>QuBN
z^q}pJ*s_LDraAU=_XpnjtiY3#lYH$^()(Li*vh$6TTKgW_eUQ*e)4zqQ}g;IwPJhb
zKPb1`Q@`Tb<Uf3;)0wVzzObuf*KOMObMwbAn@_*L8ts=}xbaV3NdMFF#lNf{l<&S$
z_$O{@iNXHp{p+8WzrNdiKf7*w=`P;==34@P>|MzFPAubf+vRfIZx+?}Prlpm^JewF
z?z<lkmuJ07<Jf&!_NYWz_VSjwXV<>TJNlTJ;j?Smb|>eXcUQ@sdKdQf`sUryuMEn{
ztDawc^X^Oh$#?4>$i0c%xO??w-y3_E@}_;CX0{``LN%rFs_2(IR{8Yz@4vp0wf|uD
z(V<d%zbEfs-`)>zrazzmxV*+s&wi>|*L}Ck;l2gh`=dYpVYshax82G4C&;GvVPCI*
z2H8|sUiCGjuKd^fC*|P<{B`Sp+`alz?a#eSY@e&WrSIta2sr+H{qruD-H*Sv_21`K
z9sc{r?}F^X_ciW;$wl)t1d8<dyPX9RUhDUrYX0=HyO?LrloHSPLC>{j-n4zmaDL_g
ziG{Kyb=$wJvZ{Vkx6b2sC1b^F1@7hdCrzpM?S9gAyr;ST+%(ajTS|PTR_;CQC;fcd
zjPsTWzrUEX_I_OyKGo#=4U;5|<2U*hpLq56wM21z+o|w<C2zK?<HrC=y;l|r^7?Bp
zcD<|Mi!Ym1KBF;JTV>@A-kSBvks7I`?Y!s1Rd@cVIk{|Ns&V3Llczg3+!OUJbx`$e
z3~JwE^1kST$rO*zcOsO3U%8yL<Ysmrt8dLBrd?9JZF`zr0|RF##mDzs_5YPI+kZy8
z&}MOaO;^L&Zj+r`uFX6tXx`?upseul3!%1_ynCm&+G_1OD!wVWS}J=*WaZuI(<5BY
z>!$OT&ushD>EdZG?y+~r*AwZ}Uw4V*GTy5{UH4?^=0B6TE?E8!-PgA=%f5^Mz1O`*
zPj;;g2*`|n`teolx0SCtZcZwFd_!M)!u!LIS|X&Ty=T`>|MKrmm~wO=;~c-J+7q3+
z`!lz#@XNMIv9O7-JlTI(zDD_B7Uu`1J?~V5_|FI~`6}nXwXCK4GT-!x*4E!L^%cwW
z&h0C_kbZwh@J)YV%ioc2gm1<_vGDh-f2nmvaDxN$Zb5C)OKDz*=6qViHS39*?y<L<
z{!1Fpypv}3{f5=mBa@?U3u>RWPhWA-cV*S7la{fGIek1wcJrjQU;O!FUe(6BxV$}k
z?&zLg8C9Ff@wL>`<ALR_V>w4}ryfu3IH7fHiKE)-FZayFel~1*VGv-xyh(NE=697l
zU;Cao<-RW9`>Sp2AJTfGceyCZy<FhiWZoKiKqBVV)i>Tb=ifyh@QG|zPkX+XLvhX@
zfhXVP|1rPJj8eu{LlSlLma(y!CCVUx^vOQu-EZfu|G*zHafNKg4*><v3DHeESuMBR
zep`M@@%?{(LD#_Oli9lsXDG}lydA1KQN%UqPS*GNb8nse;{VoW$^WA7AK%}dU-#zU
z5Aof>K`xS4-yN@9zU4yH+iw-8kKOIRY=3{rpAYTze}CM5ciuju>9u-;_=B$|2c>=;
z2=q!{dd0A|{*vjv^5uu`HF+vucKv7Jao@r!Ekro<;G^>{HjVB5hwo|3c_+eT^G0zA
zH*d?qCHeJ>?!QuwW&c$%ovmulvf5eGmRX%FU*jiV$uPg}_ZJuC2gOZ=hXjA=25$?u
z`0gVi^C3oDFiMf>;H(4vTAa(fW^wAAFfm}-lkI0)ctmJUzs85kwz$NH!WU$d%9X67
zV?A^iZpg39a_f40=HrCtsSNuz=o_-!+^^4bU%ha{vUggaj>><FIvjn|P{BepEn=eg
z9<9w>zB?>jW=t=5COGr_rUQ@oH$Oe0+{SxccggL@9U{%^nsWs;j5{i~9I6gh<$8Kz
z&+_*hgzPUIYR-RDzG};=*ET1j@8?Q?R8ZDZo*5j;erT3?FEeLQRYvCtTZ!r78`(8Y
zB<?LWzFZWr<>?kB^R*{hwk`Vo<?6ll>kT`3P2P#IDlYrEX=A~<>MNakcXw*sY<)jB
zrN3P7bDo&+p4Z&r#w9z$wz?^utW$gMA>$AsWwqByOGnyUGlb)%{){7;4=?fkeZ0PV
z_Wu8FX;<Ix>4-~XJ9YP-=0u&qhYK{1MD$j3C2lzW*X5~8i^bV%8WGc<H;7uj%H%te
zlU}L0+vVrG?JgTv&fhbmY-dy(pHmyJX&JLA<A#^|PCEAC7kE?*!;QD(c;CE`|I)X8
z-hsu=D#mVa-si2^U=hMQW1~k<na8T^!o&8SjvwYZ2Qmj`-%$^FyvF8z3BO(Xs%^WM
ztu+u(TT-5xD%Q~*C0XX^@|;KMX1LBN?)5k0Bz!ORz6vXm`6$B~e=M_`!&E!qoc%7R
zH7gt!a8}nzb(=A0R!K(kFvcF7Iqz}fbkp^>T;vkF^RgbipXYYuxYTS@!H08K)lNC8
z$+_o(hrG?bn+D;ZqkUKy-`{O^)6$(4VZTUs@svw04R*XNM>-pHOrDmUSbldB$2R-a
zy$f$FNS5wys$h+^zI;ibIeMqj)y!#OOij-s8+~8bOfuT}wl8wt4nC{Ny0a(dxc^a}
z$M;xo-AtQ^@7KQd^WWB~QVhNry>n~!Y!ex$iX+=)(=5WT_=_F2ytH@Ow9j1OHQ(JN
zve~Aaz0~BouWmi@tg+Fvz-^MnR~=OvGR%2R6pt)F+BZ#bv41F^EzA2*_kglp&W22v
z7kal$SR&#S@m^Z0XyVBrlOrmhZC37xIQCKC($RCGD$$q!EI#z;VrS^_=Ac)0fm;7=
zX>^6YSoqze=Xt8R|ApE2!oT%BpK?{+cxkI+MaKr&xyjQOg{ey&6r1{2uFLC&S;mXK
zi|^TfyWkUN6sf)3z)9%yx$|o`m9_^RebUVKy(5U%c+Tt{PiD>8FvHAwlf?bP(y(Jc
zZyh(X7rOt+a<!ZE?1#IwKb}hOZiq;#vbVkb<>ou_%t(nQ6~&XPA<e1*QKz|=9jN|3
zL-WL{10HIzKVBT#f6%6-;NtRQ%m2N-_Da&Ne#)gixePHnVFy}M*}WIrefVLyMtO(n
z8tGK+o%_93IX2y1r|-O5fgxvZf|lXFWQN_=y$WA1PTAO!@>`>J`H9{T<})GpcRG|@
zJ?@m6c-ZX6q8&c0(FYeD&}1<=sb02a{Wk9HVZXx`o&Ttz*R2@8*?-BL-xpSO-@A52
z<iY34#KYN5iWelBW~VP+Sh&l1`H6cx?Z-Uzd)wVEbT_>CxS_RqF{jD9<~@@25t=nd
zM~&yt<FB}M?ZihG&Do{%=jzYg7JvU*`oVfXm!5N{KFTT?HPr>Uw4MCVt<<#et$!(d
zo5I|me;j)Hde-0HpVlK+U)Cd6=hxHsOjAAY-t{*-_P%dy_)~wse&wB>XSL?%pQZ5Z
z&{R*(YCBTq?=!#m<l;BCXKZ=5nPHBEd2EvP6obhI>mKwc7WY4I+{1fV{?L_k%IE#p
z&bf2NC#O$x2IIADA|krY>iI%!$r~pyhVa|^UwU7X^)&e0*W?Mi^lYXD7qs?Af0!nd
zqH6iX$}G+L+p(!{vW1^kB!zuEcG=^q_0!6GN2e~__C&;K-jtnZEobkvn8bM1h`n*`
z@>uKK_NG|bs~(Z5Jw>yee?QW{62iMI<jR`MEUlW4H}6sF==*$Uok&RRj3-=@-Hfh0
zF~J$;H;kp4Kg`&pmO1GjtH80sFs7qt3tWDA-k<R@Fvhxex!1;E)iY@Zi&bnE<~%5R
zQEJDsV^xm#{@gPcJXC$U)(aK-c^q|2x+i?1)4<iEXy%QpB{2!6hju^feY*Em@x-lW
z?DaPn89r-vs+d(7{4wK3?&U)#^jH_Nq-*T5E-rL5HTG&at8~&$dll<Hu6;5LSH-Tp
z<+0`eyD_I=<*MszC5|k9!rrm|#v*}7(Q}Q}EgULt8kL-Q73mhu*1B$E*h<;jq6dF2
z#uVj;ANT*_w0qhtH&4%t`OEjdaVYtg?3m3ZVqp4l@0Mllhh7;rPPR5)w8#5HQohg2
zRgpDpmv5hqvnX{vc5SO!g3Hy^x?5KRH1yIA<OJ`Va=S3fg?I9n4JD7x*tKrU-6)}L
zBzbt_{_e2p(z>4}mMr{aWRiYBS!h9wkJ|R7kM(9>?$w;T?Pk!W9m1YcBEKE08KyI>
z+^Kk|)p^Z6p(k92v;;F+w^f}ul;w3|?(TbhTQm%eB^xA<ibnH&eh~EAZ{9-diLdT1
zq)rkNi9EVi_XE93snCCPJFWNMz>nd<VYLRPh!``_O0PSusqeqOKXVS2rgt>=~F
z1iuS<zHDFic5Khq4UPJC(O);f>$<hhrY_cNg+;%neQ#)OowsA5{>u5CdWvWFyo$N{
zamf=~Rq@>mm&9L)N<TCsVTGvn&QmkXf9(9IcHeQ9eB{B>D))Q8ejiy@JU4g#*896A
zq^4ZUy}PP%>yi^(Gm=FuwKeZ1`cKZyRT8aH|GG6x^Xikn*^fh(P0`NPEbms>ruBpE
z(cH7{ZdpCC_7X+9OYcPRK2DyPJZFp93#N(lZWeEr6j#V}>pdd2=j_s!;OCc4T|GAI
z;K^L2z~)6e-u5mr&b4@%BK&{j#=M+Mx8^m!@Qb*rY9E!k=l#7B@uRxsZPzyVuiE^Y
z|M=By(`R`L%H@mwZ92ZsYmZLG;|(=iYIs7v%GN)8>vcUPqH`7B(z-KB!IqAX-y8n4
zTCtqRDtUXr8~=SwJCf>~=ILhaxalv|_9^rRhiinDJ7bvGlw)%9mz7Ud+Tm4mb&`+G
zk?QSBO1#(nHsj#SE4!+Hh*_u5Mor%Ir#O4mFN>|BPnMgvOcUJjQG;93YemiWRr<X<
z%cN3gG@A$*hD%DtpSf_-Bc%RwvfaG!eH*78vi!|-{l>%ddS_l0YyR5her>xE|G9$~
z3O3~~%zbHkj{EG>6u;JOVt;OVtoAllEz{g2oa(%=c+Z#0H)6Ms`M$|W_EXv;cx9c_
z@1<3nd&4xu=dCsGzg*1!^!M4S&lWG8_WmqdzGhOl)*}~>$-T3m+9@+_o7T_p=~~sV
z4G|w^1|7O~rmFo-Xw~wLmv4(ab6P~U=1K%?`<wQ9&6Y#Ec$Z!<l}mWg@#6D*p9=-g
zzpGx+s3|cHslGKW(O75A^ILKiJ2%Fx4|4N(b9k;tS;#5Idd8E@-W~@x9Gy0w^K|yK
z31`9ss<yJ&T#nz$;-hH!-PhY7?uJCT_krYpq23<PB5%IG#hTv2m$tLz`*q!e>Z_Oi
z&5gZrjz`P4BV6pih^->m@~`GE)2=8jDgIeqD!fG4t@ZsxE{}5ut5_p+>aKD*a8LL(
zJB%StVs8x7wLS4$S!7*#?r-Hh#?j#Y?PKdL&KLi0rf(6BnI(|_Q?`8X%;dEbG|u_W
zS~KhIUs1oQvoBu|F6~^Fk~QT%)24VM3B}y0w+`6sZTYKs(4>lO-fIc*ze`G$9|~{l
zx_no;NMuv^%(}TX!E4_?yzA_nr?qmPtF-As-tGNUo(TPT@qFg=Ukf%$KG=0R>R0*Q
z$#1JdKV4kzT~hl|)LfeN2Je>e{f~}u87ADk^32Zrt#9we72ChWO#d~<x;|9m(zG^?
z*`c%5D_8gkWSw@^yF4v!>H1>9)zJcPbnW6gpUi(Qsi8XOOZkU4A4?lIDZSXysQbR}
z|GtLX%k~y;kMO$7`t<F#^Oxkmdg{MXt~Oq6AAcZ?xAW{Xxr;R->z00~YLbpW;IuY4
zV8P+6-|sR$UuAQdP_@`5zW)2yljiM{=V!_m_&xZ>^ve6tVV5;ugTJw*D(_$OR9u(i
zm1mZ-$L-Ag;NJ>2M7me3dD3shXl?POtvXq0!#?XfQ**6$|M=0^*Z8aBLfdp%UgntR
z?>-3s5RlM(bT9AS!FQ*Hj-Q#PF4-_QKJ7H)f*lj>^o@Vr?{IGZ@NRbDhKce^FW(c~
zxMR+ixoNMGzaN_TGiU|FL2mhNNuOsWFvf3~l6Cryx^0`fMn?PL6V;!UUN}bVzkB1_
z?7bQSds#NJCmy(eZ(DA)io>Ft5BjS0ovJrJy|G{U+jsjv;(rgWeu<^OWk}R%fku`l
zs3%!Pd*`d{<lXl?{EGaBZC|D_9Bya0$*?o`QS^l7)4A_et7q(dbDv+ct14`NwNOIx
znJv4$bthhN?^06z_9=JbX|DfI9*O>mpZ@*+H2eK}_J3ujDJ&{ZtNh!te*P1!MZKqA
zKb(L5wbu61tMlUA4!(AHueWY>`Ri5p;^(c|bRqcQ<n}OD+pCT9*1np5*UV=B!_TMp
z$Ne`mG@BR~*Cg<MswVeB^R%8tkuRQJQ#zIYj$c1J!=^9PV&?ff-d?=^9xn@m7}PTh
zls3r;GqD=5*BLcNeF^)mB*wd2qh|}VH}?}>ha^X%?~d<}O%@W_k}cw@B;5Sjto7${
z+YgU_s9d_d&Zq0ewCOW%uM&)Kh||8rxy|eK<gNviS$*axF5~Pz&>Y~qD&aYk)=I4n
zij&(8U4Pd6<I-<IqsZ@`Qm$qX`L497OjT|SotAU>{9MJBb{$69U+GgSuUx!t^kd=a
z2V9~yC!ha+R<z+u<m=hfoU?a(Fx@|UJoa>BxP;fUZ7hfV?QgRxMl3!f-7?LI_3XEa
zrg>p4d}|kZonD(!a58U(<+tgK`#!#Mt#ncO8hd-6_TD|oS|@k#pGbbPj9dN5glQ+Q
zZBshGM8iN-w1`>A-)-kbV;`kXrze3^j`)<O`8Dhi6L`9%AdJy~rK4`PZp6pu<yC%q
zM=xqK{_uOhA}9ZQ!t~W5ala28UCp)SmQlxiot@{79cs9~^I``}-~w^RJ$Bdc-8el_
zzvj|8E?K2DOuB0xhF)M=zx6g_T=xGvpT9}wADMf{rN8#X4Po2;`yxy{%vh&IWloA(
zZ5!($?D}Qdf|7;mHxJ&fTDWIj3ln#mz44TGn^x!d?3;hA+LaZ;eUS08vWAH7b+32l
z*WKD5`dFsd==&Y5ntu-aOeJ4M$=U=xvRkL2yp#L5|0=_rXZIw2PL5(^e9V3{{4k@{
zJ@aJ;gw$1Q)Ardtk;_~bTOYe$-dFwW<4U2smUe@K5e;rD&@{2d|2_k4xI*?CT<
ze<-;!e|q%Xxi#sUm6@-5FD~lct90$>rhAKv&6^g#d9_xqy1&P4;pX<M^X6X)vTn<J
zb#dYBD8EI~RTJ*M?tU95>nI+k>i156{b{Wfjt}CxWuksFpT4iZoB8LTcB)O?#|P?R
z_b#M-IW)bZzuaW?<$IwOKFUjl1CJI~+`91iq9fOz%qx!-Toakkr!hX99&Z>eEcWkF
z%qRDGjWN=*tU653G+k*ZSYXYW+RO34jGbx4p2v;(t@jl71UV<obXecN`;Z&!$!9DM
z$*Y&MOkFK{&3DqOV*(elm4AOx=e)AD=KfNt*8aRDS?ZN}i}iXLINe`q2AyFv+_(G*
z&;9#yhp#3?EINNaGkuA;;tkHnhifIG)wlc&6tIqvb(C$kP-SSB`5|xdpq}S>M$5&I
zVl$E(mMWdEY+5Y0AbNk;8lxw^(S;d(8*4j@Gz;w1r<6PWnmJ*w_t(T$m#jrUCC;)d
z&$+1lkvr&H?K3vVyvpZovHohJISem)s-DdVdY1Dq@Aah((HXUzWy|&mz5KPq-@HkB
ztMJ`|#^MVPwEkLbdS@zfyig_U?ao}5$2&zXi*oPIQak!7(YWm3W{YoKYquziw{8+P
zb-l^mYHj>QC@$MYFqzMKR$DOlqQ^#0d*m(Er_|niq+7eb_}&Y<_oi$CnH^cPw(b;k
z&TN`s+@-lXM$_%=&D*saUhJp1HgA)gTz|iBxlxdYuKt0~zbv(rTRrubFW(y{v-a*m
z>jQSCk8P9Y=6sra>w~3iLhX~y1~ZG__fIOn-?w>N!{*P^AJnME9<Vsp`EE<~rBbed
zCH(g{x(12($S-($JL!PLl$g4Bo|a?UtNhkGv2Xtq7JY?lW=qt+y?>H_KHc%Y)!;tM
zm(o+tTH&SVi_YFj=iIM5p{_xwUf+}1==%YgBNv_2;%7yB%*ku<`eW)AE#@|5nzAy_
zkzYpJl)X+D%Bh6P9t?UqO>RkcX+o>_@64-=+hdk7PdMiFUqiO1quJttb+3Kn3bCj*
zM)_K=Y=N`xm-`<~du;RMK=aaDR}b<8&2-@w%1hCT|CgXLZ%(sH`b58?d#i-gmiOkq
zdVK7LgO_$w@M4)~HFi3pyPS;sLMqNLye^aQsI6yxX7j}Dr(Pepe`d$oPjTMY*H5^2
zJi6e6n2)2MMTxp|P0*yg@=q(1Hy@1OS?R_9bBdhht%CQ<)+{N|eDpT=(%hpCnj02b
zWU@|PZF2kBioJ7o=5bD!7A$xW^7P5G-`>lrFLDIFxTz#}iuqfW!r$b`3{AD88r5M-
zrOe`9s@=R4SY;Qt<=MI^+w+n)q_2Os(Nh1-!X2}FkLd4zT@i8VF6V+X$sOzkVP#5r
z{&y4Y)NY1xS*`r+bz6H@;5Ej(uUDs?o5NH$mHAKO>ZD7lVG3=U3SLhcu7@8@F^W0X
zkg}%ABh*J}=Ub^mU)Aa!iF`YK=0E?cIJM&Yd8?{E`szy7xfPX*H5Oj&Ugvmn;wtg0
zGDcg==e)dkd6TB&w*^x(*9m_N3O!oF`{qT`<gUmKQ}s)_GoNyYzUo&vQNC_L#Yyvh
zsWES|BMSW5^!%Lnrm1^9HmzP(ADhW@_lZ#PK4&v_wPhuWK|8Khth%J%p?ON)$i(@q
zi<-dkPK5<i^rubCTq&Zt!zoGG)p&)C@tUUwCE~V?SMC`-|C6!D>b>5A$}pLq{okKx
z-P$aaxIV=A(?tG{W_b^H7cE+3e0L#_+IQFQ7aQlu6n#=(+P$)G&02;_Ii;0h%Y++r
zCainpDd6>Cmgv&7#=ALu>oixcnk7HC;@F~$E4I_KHqC94SDtHdZJL#Avqp|h@17sN
z-F3@^qP6^emc{SX+vBpzvFhP1qsQO&FyDK5J4$Ct*usgvnvG$LHG{cdsb?N4EaVXV
zx*@AH>!}hK%NvQMlV4q5xpk|)l*{Vvs%y4e&ARR#yS6px8eeawKl803-MQHeE3bzL
zy1q_3JNta*#dOD?YKh{pS_w)^&wTaPEH=OT<m$F?%cG)N#j4Zog64(25*EF_Mr8j?
zue&EIm%BckKYP-=xtC^tPBoELTG`qk-?hrh%Xi}W7calRdnfdAt#Zk(XD<qmS_pjR
znDFvi(ww><tw$~XE>J#d5kAlNiExACywB~zbMEq6Y)Fc!-%<8ex9{?ygr!FeQh8YJ
zznimZx{{Q+@6DK)(hK}wDti~nF5CQQ>l~AH5{w^$A9I~B-@fvv@a1W8ld6lmOjzG%
z+db3umzlkA?c=aTW?b>vOG~9bA8g-m8Fb!!$BXHw8O@h@Z<3w5ul99R(LwpOPd?5T
zX^@>IYq8ttyrc99mBuXPke&Cwd^P4ik*hIhvhe21)h}ik&9xUuU6WmQVfvqqrxzyd
zx!$Dv$aYr$TeW#+tfwkc1r(=#@0}@-7B=VW1Br7^1|Rb_DR@6f+L<-~%8N;J78_0C
zeUi<8W{%H3wZetI`oX&AH#mmctvb2m(zGeb9a)Yt2cH^dpEG9e_2-<BovIrcZ|l*q
z_{$njzNagXtypW2e>`lC<+mq=I~_h|ihr1ytC^MmO)G7-*p{ru!@g5*Zfh!*w37c-
zl+|+6G0rLM`i~vD)6OROeA{XtJvDsuEenAc7jJRfe(Mu0Yq+shYR?sm?8x2RKbFiq
z%dVTY;=;MLGPZInIkK`Z=ic~x@M`A!KTrGRpZ8bW|4z1jZ7TQYv~qpr6aD@_`@f&q
zxbuU*_E!haoY=Yh7IbBwEti{AFE@3=zcuVklHOOpPJLT`GX6!`RJASlUTfv=*>ZA|
zsXEhIp5K3@)?eRx-TCjvBhKGd<(`J@wr<jO*?dWVjnIsB@+EF-lVm?PR{p!W`Puw_
zxhT`!d7hKDd6?K8%IJ^nU)y`R?sQ(v;m9mYX~~K4^^cvz<=uH+_t#zAy!=MVjLjRH
z3SYc9Fu~y4r)N`)wVuCUSeTkvAA8+*LioP*;vTi0ZrOf?cC0e>(L31B*d5(0v%PA=
zI{V$mv;S=gWlijjoj2#W{HlU09jgyd-P{)c?~DBOx$axBHQ`N(+Tv_xiL&A9ZIpNZ
zV>5w$-@{+<yLj{6Z<y%ZnAdnO>Pgrah2tORRb~ADzfh*AO<$8&#`^}7X_@G4k>|73
zpDU;O*ZUR;eVji>=E(n~s=vj*KBoTqbk)C(@7k-EPo?thJq_7~mrXzYUR^Ok&}kpn
z!rGd8u|sj6s%17ia<HEFXY-kp$y(Pn{q*`jk@1^SjE(!GW)=$Y7n=0PR)k&Z+H_sX
z{c}nCGlhP&>~(ATwlTd=lmDStl^U|=_bFwI>o5PDs{2-UH2nXc<EQl}f7#q_aOC9C
zodG$={n_?9dka2Y8>h9{rZv*rc7eo+O`^VwbN4Ad;&D4-m>uBlBoo|z>Uu&BTa#w!
z!8ems*W6=^-gY1=d(wXIZDx$RH?96&W2|Bdd1!k;M8&m1jaiMmblRr0%T9ek{3j$G
z9<r=>nf0eI@pkHGxr9x}Lb$#E@a*2tyTY7tdX6I3vL!8QbyJ*{Zi@|j&$!9;Sa|m3
z^b=hxINTl9ty=icX=~T!J8An?9t^qq?6U2WH8vN&^GnOU{<6_e{`Hp|rMJRLcYUei
z;YyyB{j^$U>am%@y#7JUHoq^+%-pGG-z^u{xo=^!rrnIFm4>@+8|?VDQhmw3g0r=Z
z#|4CRwoI^8d9UZXXYQxdelfT1x?IeZGMJifF+0B4|8T9>?eqszvzMeDpLF#O@5Q9(
zJsP{F&GLS{txUgd8bgxBo7vB^SknaW&)zVvdEz5I9d4=ct=y$QAJ5bFe06K4?%O>N
zZk4@Sq3L(w+ud(hzFGSwoh%gcf4SuMjZ2k#=3m}5@0yk53RBO{o<po>?s7dBU2Ahb
ztz=60^Qdc@CN5!n!`#cPGK8nc{QGp(=f!4@NoH5P&OK-pzxh@2PiH|gXLh##p31W!
zE7n?vR$s7F^WRgnbA!z5W7=mGqMj_-cKcP2b>4L2KH>9i=66>Jde2<E<eZK6qK4$#
z=Wp@{#0Xeur@AD4iIA^8=GrWEiHX(CwCz#a?H!teX#r-(`X9APSL)R~aoh66wEyAr
zE8h23=(=*NOf&C0y>?ASbH4VDT;@~D`O0Q-scw0hDKzcZO0KkNQr4?x^(IEh&hEEP
zwLF=w63)F|B)8Zt$ZPH^%i|ZSjE&mu)U;Y-rxpkJb*s<zQRETW%yhJ6VV3gYvs0Q5
z-1O68UA^bS@@-0=bGAOY#>{>G;GN>BS57_L%{Q<6vt(m*wCurU&c7B|t(`rsLwEZ*
z55d&`zvIqNNN!9|TsCvdp7mR2PkueSHFnZ<zo)O%`u{E6b@G)r^Lg3x>;JYF+ir-;
zx;#ZNZ%Oo2(br0$V!o=T99(ah+qKi$T(UZ!XG$3H3Cz*Vewr7~xzfV=#uX)je}`sW
zUAgOms?N`_fD3%vr%d|$GF52v7rTF~Zy&{TSN(eXy!YV#{ZF5Z<rNhdEZP$J`niH^
zxQ)F}fP|ceRoJ9U=NS7r_gcI)zbDLOyn2C~;t6S$Z)sh}{O>QixGU!5{9lEqg`cKB
z4F9IFxozslLuzKP9U8OPwk<F{GEKq#g|+qT=eAQ1orznsXvLSkJ2qZC-mvyhi>#Bz
z<CR6NYYM)03)bp#Z@jW5ZtAQhy0UXu2AxSg$LMg5In4MO<NJp(GR@}e-klKKk@r&g
znfvaVkAL>W`dDVCOp3XD`0_O4`+Kk0P3(J~X_6bAxqiyKOKKO)+Vf{`n_B96KkWCd
z-1NLsjy=h>t{IYi*B%~T(_<UK{+082`cu{I3ZD}tv!p9sdYfLJaL{S<wNC%@cd_W(
zRkzZQ`}gSFeKX~mo6M)DDXON@zuv0a-qy5U{K;m8rOY(mLWNCFK2MzTqk?6k-tG2O
zqXVbByw7aPR{qs~VF$No&;5B)3Bhyke-?@D+yA^uz_j4Lrqzd!uh#{e*gx<(zoaZF
zPnP?*)#u1(?~Z<Mp3gsXxA?88JF8#B&fMMp@822Q9lNB9`!(<O?_}@JPkDE69eeTm
z8PzK;IM4X)A$#Ud`>O?ZiEnhizCCjH<lA$%U)3MG{ZQcEvg6BNzV5kQxx1QeuR;0!
zqqlWaf)AE|+{S48I?7=Cr!xI{&fgUCzdP;zRK~woJOBHI`}4Q6hnGKo%V_@Q_-*D-
zy?bn@+-lw)zQ;D?h2oytC4A4`I+q5_56Jp<@V?)jy$_mq{-1oN_SxmKU%f}yo{6ms
zy!7|ZgMIJUPQSLPcE)Y44sRj5(|7h({LB00{Onz8S?W8vs_<v;7Vft%=3n>j*d69|
zxB7RoKarkUt+5r;JBm+p7O$H<bGLYnjrM<)Vq@Eq{YM!(a+&I_66em|`+fHP1OI*<
zwx8F|S&yxvBWh={k+})VNqTQ%PUk(@C3yF__AmA`6Ljnv8^swXG2B$05j&yvcJBR9
zv6KJxgN{Cm&RQ)W^e(|*$K|~4Ako*gT_WrK|68t{eB}R?li7dN{ont$`}^|u%l>-K
z<qC_6pMCi&oWEZ~b5ZZ|_tWd^CSS_ZtbJpbX1VNseEgbiU%o7#(olZz^NLOL+U`o`
zv;He<+t0`~(V3}sW%v&n1?k0~=N@Z&=9P2r!$h0y*71qW;nz|!U+`@z5tmb*cI#uZ
z+2$>ti|T9t-+p=izJJgPfe*{g<m#<*{r>-Wed<Adee&<0lf^H3PkqZCXtB%Vn=zNY
zim1k(OACEADunpC$f{>u5?XgTLtEjH%i^2Lu2Y4*>VEO6nth!-C3AXqmV<zi&XSGZ
z>m-7l+#U&Up17;g^qd>_g{((C*QT@W&0>uG*tmcH<L}n}7fZRi`eouX0$UlEWX9O8
zTs|p`#d5a5naw$d?H6J;Pt(jV+pzFbRK<nOS0>pR|5i2BSh#3*hip{n{*67)l%r!e
zif>8t573#c{zA^yJNeh*BW7yTt~@Vtl{1=W|McgRyWgK|?NhtRsJ>@sz_M$*r>yVV
z?>PP9ly!Z^mUqv#Dy7!c&En$uwDM(5{F`XiEuUAP)1IGx`E+*TmyN+L0%F|PCI&xI
zN?hd7Kjl-hx$*^zL|vZ}hZ_nzBcg-n)b8zB9_jX`O0#F`@h=8TncV+Xtdf3j8JE;;
zzJkMPg{$C`6v>OS0uv=%{I|?vcw=n8RK!hRV7a%@Du;|2EFLdKjD;(GIieYqV`m;v
zPMP~j=Vw9cE@y$+9GudeTjx8zVfk_9RE<shjkZW9r?<5N(TN8dszo-kXT9>hb6YKm
z-(<Jaw*!^oN8~L2|Ld^T+9jRTf6wN9?~TP*%9Xxp+}1x5r|?Z;QM<yo#$b^nb_HEQ
zDqDmdXa82LR{8VO!1^D{jMxKaiQjJ@o-x;@PclP3Hg&P&rm7=qp3_5ab8cTO>ebrb
zSgr6pwl(*OT*tT8ca2~7-LT#8kMTftNU8l7xu3`GKCFIHuV2^C_?`Q}bOyU)FAX{B
z9>|Cu;-8}&&aKOPj`e%;504$zZHM)F?lT^~FV&~4VcYfai|^9b+l|W&Y~J48Z_1N@
z`_lQu9mlrt+_Ud`s9bsI!`r`~CT4{5$Va~KdZ_&RzWibJ(DK6{I`!Z5J{10MZCJy#
zBVJ<Pfe)H~{0BcMKi80`dm^(misel1Vz~n!8b4gO`r{tk9_;H;eCYD)f<xQdzn`1_
z`a{UtbCyDHI?pI1*S7Yqlqvam(E8@b&}t#B*PIoH_b(OnQ1<N5V=Db|O#Dr0)sfod
z?FZBSUC#f|**RH8_Nc_4DX05b<9EjVsFP>7{zKTmiqGQJq!T*+T^AQNtSg+8cJ5F2
zswJ_TN-r-z-E2BD<H+STb-ryv_wMawo7+`ZHsQ5wOM{Y4m}BKq(G3R}CMo>yeQ1C9
z-;-C<5500tyl8C7Tg5-EMEmFbhNcoj!|e8h7h^+X*S;zB{l5OitogpmQ;WF`{8EKa
z&HZS(&a;*MxZi&jQI>s5*1W4GO62U&T3566TeR>w5p`zniI1wUYAV<+Vhve-;Pxd$
z=2`U{nh$(bFWvWkZS;q-vx#0w(vNcIriNTUdoz5~$@m>7C2HK>#P{?tnLcLNF!zmB
z?F5zw+Q*mfzT<plwG4~t*O`ZR?pR!^aW{5FP`KIjAl4l_GtcZ^`RVVmjE)1XFOvJp
z*YyN5UJ<H(w`o%8L(^SS<sNUZ#W;6!-~QvDD(>WAF!A)c*5<%$Th3LcDc@FF8(Tgo
z-5hmrcT<56<6QS&9IC}1b>hq~w4HbQo~o2`ZQk5DYBu3c387Lv(>t&He6892^l$WP
zEt}%iCw=~h`kvc!Zf)w<o0B!8p3K@<RIzD!{HEz`1?p#mGj{P^N;!D-&`kGO`Q}Y-
ztVid6xj&2BYt0AQ=@VD9I4s_>E}*W{_3O_l&ZlZ}p8J)umtIL$+i)jzTCwBCxc8Ix
zX-Z$_I9b!r7Ju6M*0uX$rb~NYf6NxC=V+L8d+Fh$JA5M-*he!>I^Q#A@)UQjeVgTf
z9m}|K-Q%kJtC{wDS!dp#qg!d>o)q-x(A^_PJ}$eVuC+k#_uc}%y(g!crS20dE_JNb
z+j6q}`NbQ>wW?lCoXZ|qzuwlG`7rCrs&wP-C!3Bqm|yjru*mLFg?H+O`@bg6>8dgj
zu@AkyS3BX*-h$b?ABi2!`Pa&I@@mEBSEdfRvz^1eZ+B>}$bHsl;_D!-w2m`e)T8`)
z1=IXr>((iK%y^V?T}txl$8wIf6C-n7-rQ(h)So+x)9$#>Qd6#Valh0KPOmw=FKqt0
zFHaQp)3ld;-W+>IJ!Y15v&rVI&e9hXpD&H~YgxFGf9~|xQfs>Ai1F>4xh?+Vk35y_
zGcC1F-`#gM=pN_uHzH@7Q|_!e)7RAY+0wK)JnG=jLyXK-lNP%CO|y?KP2;jn=X<Z5
zYm&t68_*Wy<<+!?W%f3&hF80uZId=H-{&Qf{i`F{rabD$me(d}k@uUPajagnHuU<M
zEWVkuDxa?_wVu9QEce@<C)?ite$yL$!O8RYi+evj=EZ5=IJ@`b#22UTS#<ek3tUqD
zA7v6ylsY^6$xYpf+qdmrbI^3^OvwrD9h|jZwK}Hfcc)dkKRfy$DCbMrPOY3PO1oMa
zjUMFW@q25piS}5vd%DUNvn4Mt_p4km*tGAeT<db*=Mz`&)!8mDtbeOui|uEd4Xf4%
zxh-hmo)~-Y*{5)|ox!YcTX?e!PupoVh&Udc#kJ+?*6CbqEEeSxxQ=m4Kc2;vX<nx;
zDsyj>d*wk%#)E5Ceb_dit7C!SpE5li&1RufNA;)Fd(0DPWwy{TSMG3lduZX#l}m-)
zT9X$!DVMF!cCfxsow>j<gOfdT!Cr$6tC(`<y<f`YB6v8^&%iCCcCFpJ8FNnrx%~Dh
z%SvzUimp9=wJ!OE*qRBNb9$W&@@C&$`_*^1V_~qGj_E0V$(zf$+#JGJvYtP!_l)1o
zg4y`<mc2Dw{iE!^d~jIn)Y8B5tEAd;h1;vYFn?Rdu>HoneZuTBs?4sgn3#W~O(ci2
z=8JLa>zvy`N^2*-3A^mYJ!{vBo&WDv_twPto&5E!?CF=2y`3Vh+qm;)iO=7oyyb9X
z&m@7vI^E%F@;|3`NZrq_JpEm9zggk@^K8t!OkDO}xMh>!oZ1xNtvzY?X+MJ(wtZVI
zO2Rfom;HG(_rYVc%R4m6FWpbr{QK*rQ(X*&2cz%p`B$Bk*S)VeTJ7)txjUG8)6BNF
zJlk878yDAK^W7qswV|7dBm3R%hg(D?&ZeJP_N4O9maEO_vre(KDZ9S@pRr-BgXLy(
zp?fvwUK?gQJCxtb&3vEIoh>+Z(w{hGmwv?!8}9kGUz0OtV62K-!7qGkgR}`x&D`A!
z{WTN}yh^S}+K7Jai;PM*zd3vD3waCnC$V>xpKG|*_4rjE?mD{E_jYHIRm=A?L5}SD
zH4$xR?YdcIP0PByAKTq9v{7(mZol{~P5qr?#Qu^u&pJ;mSJzt}y#3C$Z{g)@_?Zs>
z`?fUNGA;6ZOUC5|_y6Qv-!u6}?Dp3GTbZZZzwSKWk<DlQx3TTE(tIhsJ>Ng}oa~&}
zV!Bf@;O=VUx`os3880=D-n(S?JbzighFMiYU+#Tg`E-W%TwmYap3k5EmkK^^^xt%u
zrOiBVo_kd}*Z1_BSABH%>ie`P&n9Ep#(k1Jyr5OFCH1QhTygvUzW#@NY_y38md39U
zQ4_U>1_mgjJiTFi`;^Uj-hCFSVm$A;q>eGUn_(yOrPxP$9uGHO&n!B+^yF)K&8|sl
zHhyBequvIHeEs;iKV_eEy3SvV$qjXDch(x)$Ddh#{^paVM_!zL>Udf`eY#NB&O84t
zqF;CSyvVrlCeI}L=Y8hh-CK9;o6E#<iLX!f*VD}&1+u#u)1KIV-)^tJf7Yf7@iKQd
zqxb7glx08bZT2rQnaMb5v9QI9%-(B_A5@x{moS(AJhtcWxA(X6+XUAualhxearvY}
zpT#WYOjiN*)!x~x)mARtMOE_uBH6b4U1;85$m+R7PeshVNH?W%$!z8qYHdt&uX<Gd
z5O8~a${=|AM#)728+$(;;;Ci5f8oO^=RawZvs*N>t~ctmc^zJU{-6QtwjI+9Ux=Mp
zAUQc|%VX&!n<e}WJ}CWsy3imdfOo1(fP;!-^|PAG|F5K<Q_C@NTfO=b-@$XwmFGvC
zoy*#h_snJL+4Bwz?xzkdmJcfV9(PFfpxI8FS1YA^AAG6F@LRIl>cYklmW?GWtDnrv
z@)66`K3wGCGI`Un4+Y	#0*L);<<;;*M$Axnyb0&0wB?4E`skc$(R&E_ubzbLvC}
z|DwS2r;MI;-TtwCqvh{oKAltcTV%Uuzmk~b5yku7=Hu!&!7`_kov*8$>@)c`Gkl|B
z;+Z#h&q#*6H<=nN6L<CR>;KW;_1B)?Bd8wCve#^}S4G&SOy(qGxiG_(QZ_*$y;Ucy
zLMOOIMgN;Kzr!GPibO)~x|!?lzgoTaTx+DTMz8XZJM%QMy?4rL@h)yixO8P!ygOs+
zQoj>FpDJCCzFDfxUbfLP`~N(vPurA4&rLq)`&;>+plsyVX?g53@>o|*n7&&<cS6B_
zZu!Z=OD*G~ZO#U|o!w-%!|=LS==1kVuS<8Gl3Tf3tm5qQYx0(DQ$x)nCY$oP@@c+1
zwX|u+x$7mnP3|9WJ$^lZ@1rkL`(@XEp4u1K_S`tnsMy<cq2-SYSA==&>J=myU%Ect
zvT?0E&#|=Y?oAUDcvSbPRj+^fTj8CCk^glu_k*{U*Gqqn<T31)y7kh9^Jq=h{-Do0
z&cp{)1u3m?5ZJTnIh#@K*R5`P>l=M*6XRrTS=U@+{Qhiq(EqzS5k&_7rJf6YjQ#T{
z*YG|Mo4MV8v;DKz3#pl$(|@77<r3?H=~*Xj*FDiIyy|ex`pC6!3jA%i8slvu8{W8u
zOtG*k+UuH-=e$>W&27no%U5^(T4wmIP2#xlV$%SLQd8M5xBJqU3MZ~o$+UdFlQo>{
zRL*MK@F<2CKlfY6zsOe+tN*=8_|NC-fu#kZ42EvTeY2W)AI@!etFY}pD<5^Hu2b*5
z@4xiUpa;wkZH`Wvc4xM^dWqJqx3B%}8N!keSYs(REQ#tf8ylnSsgJ#MH>=q|gyBK;
ze5L8Ki+==XTL~<@%+*&ay|uPMqF*l{xufdh_WARa_eMB8^geg;e&=kl$c0x=+>yV1
zTFC91Mr^diuajO0qFcPI>kee89uNI7eXc>@?0Ky}g=7uBht!Cq{`*zA@5|f^cK72J
z%zbw#FaP?T{s>-P*LZohhz+|=oH}mc6#imadf^*`XXi4rB;GdX_LasTJbmeVZuSPR
z?K#EX-fuM3liyuvl$-fnh12i(<!>qRhZ9;@v`e`C-kjA<EVSExgfaHNG{244YN?;v
zZ-m}0FW&LowQiF}h5B0$OHG%zOqb>z@f3X7!o|Phf-|dR=GkgxSr5(3*!<oz9_m--
zrEwNd@=sn}w!)G{bdiPTA{)608Kr*ev&<V83A=8QTP|F9OVa+)(Y5AVVjc$lPJj6{
z<!uZ{jY7nurBYspKl9GKrxW?(FXM{1<*eA^gt$Ep1}Iw`VkgFDHyel?z5iFF@{NRR
z@7h?$7T@Rz>u#^m&8haDyqf!p#3H*dxBcHwwbhw$K>zuj+4jfJckMCQmHw|iJC)OR
z^_D=sU1bGAH$<~?_e?Nenrbh+Zj<1(&zl!-$X~?xO#a;c`R(hEUw^zZw}tcPCzh`T
z!g-aJ?+?$=TydgyUvxsvGbR1^vHZS?tEQ=T9-m=9?d7`i@@*AiZ$;&n?JZw^ATxOL
zQC022<uYsbIx;UyZ=AmDhTf9bC&L`mTUpe7c;ke6lv7XE$*g2|`+NCt$EkHw{|M%7
z`kb+a-SJHD5}D3)4?*=@Uzc+)I5WRGMfpupTwCU1U6eQRnxf@*mn+6x6CW{TIxjzd
z#aQXqnJAwvs=5c#&wigTXQBF8yhmn^^nT9AhZ&abzr^-WXxrMks|x-xsdrD|zzljL
z0|NyEh5V!}E*l$tx6GW9)FOSioWzn;m(=9^lvI76)V%bP3<VQo13RSCU}GoW&1-TH
zXnkMVwRcOVRAXzbFHdLFHnpww*Vr8-bUSrYj&k4McY2<uzSp9Y$Is1~X?w=!bM8jd
zvkJ@Px2^tW(h;1T<$e8KnzoR;mAP6+QlR<O15f_1-u9+BbD#F_?VD5$|7zLA-<u~N
zUmhPDJxil5s&x9Vmml_aPFeA?<-+@Ff2Fj0V!wZvE-1_Pn;W6S^w4F_r_v4OhLP9q
zPLw-V{(hU<?HQYA$b8yczj^=HzpYV5%^b}-pNj;mR_~g%t~PR_?%A15t}o{K=$O4b
z$z=3(|7RDaw-u}EysX>YcXIPPi>4|c-MCzZ&2x_7S_h#Po1HAV1=A`f1q5n(G>c@k
z+1Sl)w&>T+xFEInj6)3Ll$Lpqm_!td_#VCT+2Pz|C)~POlw<D-nJq3V*+u$3tQl`L
zUB9L<JCz=7c_mc0Y{fev8`XIx!mEy72)S_OLzJU&>vgVKvpBVc^eiXa9*ku@D^jPm
zWA)4e`;%Hb=0B<0G5=Wj#XH<4rscfPO=YeWgfE`gw!7+Vywtp1A0Jyvvlqt9-~V=H
z-nCtY->jdBr_OnK?81GiZ?X1iKkw`R7kw&^EzS*zigQ!cB5?A_tiuKZZSU*2WTU-4
zE?V1~;PF-}X4gd4s?roy(<_?-=Q!Wnx9#7;b}6ABCbNE~Up(R(du`q9h(9mcH*XcT
zE?f8Yn@!A~jc*eC=XPJ}**bS-$D#c%XB&2PZ%?bquR8r`{q(czZrt6wFSl)9VP;Xh
z|5^X^ci(qwg|6A^9&<0&{j|}eU3cQnPkx(SE^{MG)KDz{*^TEX<8PkbCCm4%%<}h#
zwO{92Z=7AbkyripX^Ve1?(OPS4V>3rpt;G|Is1Ha@|yfRC*4-3{43hF>Y0B3n(Y5K
zQ<=6OzF1>kw{xD|+3S;CmQ*fKn!uDQnIqY8*lkN&Z?k}_pdwd%^C^eVi3`$ya0am2
z3-~8aR#>5}wBj@S1Q(wtCr$IT1X7ndPLULRso3Qqu#{!uFXfIaoeLBKnkyQ5JHBXe
zy%bz1{7ACJPeZAs$=Oj!Jg{*>w~~{d%vG<Fq%sZ_rc9+*+^;5cxp1wBO{(Lnn#EBi
z^f~y#WtSCUNp8i?O}!jfi@KV!RMnORtYyo*bhyXlWulf8m(+V#=f?J~OIa(;qUF*Q
z7_|;RUKYN`=5wm;|819J_ROEI`lcY_aB8wu`u}S+GRKZ;|EXjAbmvDpwzNQ0E85J!
z7$yHtx!d&!l>a}A)LNc4Qn~w5o>kC9t7Y!Z$YUQFlP>LQG~_gl`|)k|7fXx7Y$E&o
z%CA~po!YgzX0rKLg-7<a(eM5$_>~5K{deo$uS?yF?XUZ)OkMG+NTK!rr=XI+IeW$K
zzJK({^^f*>e*Jv=__+JGy)T}4^7HQe{Q0i`Yku7D)Yhs$^sc}9d8^X%*?<39{qVjw
zCu(Dai_?+E8-H1SDgU@-)vd+<4cG4dI`#eQXFBuvZrZJRlOOy2rT#ui7a`s0igQ}_
zY+CzgtG3sD@v}GGyHjeWC2f?RpZa9>tNq`Ndy|g`|J?p!`P(AfJbxkG>5ZEl#Fk8Q
z$YKq;vC=`aY*JXo^d^DJlLU-Yr|>4FEzmued4cb_!^$Jp%uZ$APTxw{E-m0ap}=gj
zL@ZF}gd$gFTXByiyP>tLrGV_j6|-)xlv)yLoN@7s0sBQ}PC1iTiyj`F^uTew<A(&#
z4+{bvdHJWgTs`A=Rpn0ey`}>GX)NxVQzR=cSPR%h2-z%Iv97WGqk}?4fyzDRS4j^X
z4Vi5;Lhc=CZW3d@+9!F#_Do-tbjjRSKldiT=5?Ddy!3b@9&%w*e8FrntKcs!OZ}c~
z4oJHoEGGNNV@2sHo`AVQ+#%9GHm}&Kyzp9^<O`0Qva%Cja9GL8PK>AwxxVp}iiOdN
zt<B2rP0H<d#cwL_%!vN@x}rug?{DtYi#zi^Z#^~j?R74zLR;6cqQ0t>%eAXBTY6Lu
zPYLYgnL7DTUD(4smHqQgGtxH}%e6@^zh-i6S>+#z<4@PVKmJgntxRI~<(4(l|7X08
zz3sdF(0h6QgHeLm5~wjTg_WrZ$^ikfQx0Z5h89-4byT;dc<Qz|%rX(Uu_g8LrZ*Qi
zPF=HbWA9^`Pv5d%OINK>Z1kUb@z4JXqs>>#_U68FJbUlgwma>PvagrDI=po6x72|4
zf>pCsF0Wd3+EJr!`twEJX{%!|?iG2yAZ_;k_4)O2b^mU?y|wJ@rQi1-mi*QE^}GMO
z@3RomK9>FWcP-fRp=|H|*tUXuX2-+MKJN_LY`AtG&tKoWx~snl)cK^ZuND2Z%W~!J
z{;PKL-mH)PE_rUDWXLQXv5&iWN>ig-y~C$0)SngX@$BUfohvoQcHJr_SGV8atM;})
z=fCJ*k$JYg4>M<_cYCb%;JcKmvbV)#qLh&fm#xbt{|m}n+NM2`?5OhVsB+V?5uWKR
zt)!?N)ncTi+hU|XrC9xHuPVR$v5C(W{O1YC_1tM<SN?jWx&27<g;^)kJ#1_o9x<Le
z%6zKm;|GpTc10VDUy~N!x_DRS>Q64Gt-?ZAofbvC>daEKn!niUc(d@KW<lMG3wsQ^
z+dmoz$;}av<Jr5-=fn&FPWL#Edt9{}e_Xz@+2T^hMvF@o+bcc<F{VzFEYbMvq_KHL
znZ{=)BgH-T?P7;TEG{-Fs~4XYvDi|fxGVdK-xSG}e&xEGpS)NnzG(L}uDFokyWZ1Z
zbi2=7sxOxs$B{aZ^W&#`$>p1KUhVPQ+FY~bV?h+hr}Oi}4sW@%&n&9)?eD^aNA`+`
zU*D4YO08z?v2|~kt@-hjKW%|YHMV4CLR2y{Mk@#7vmZH#9KHWp^V!B!O|{S0n>ftE
z1hUFkzCC1jEM-fh6=!nIhuir^&o8udaqUlecCYwcPkC70_1hQtlD?^zeZ9y3=ti5`
zFYof+t>WzUS<Yb|7q@(M@Y?@%=c~<;J6Hd@{_4%qH*fEs&i?uR`suvut$F`GJoCT4
zUdyg#Pd=BnRQ-`#-=p8Rs^nk)eLMBT^IVIvq>U$zM5OA}N<Q+vvv$_I2k)1s?@xZ5
zEbaH`-QqjT)qiK-um8E3zxl-Gr$$q?71q6zs(-oq)^V*lHQP=cKOdxH_Beg!&kKJl
z-tTx@u*QD1g|7a*V^1sBb*C;en#k(eSuV8XL*EkNnL;T-mX~DS&vWQsV4Bf7i>u|`
zw3c^StWoNd*8J?LmD+O9Zo;XJlR^@fG9GMJJ=i>}`GKS3%qDiu_oo(hv08>G9-N=#
z`y=>`Up=2oy^2N9c?GLQVpdF<R!v5}?)?f@N(Y-=)1+4EPP!hjcH)AzEbm2rWefa-
z&N(_SZ_Dy-%4&VbX7!bgW$rYVxm=Y}D)xCEELZ!EJb8HcKywOv#e@^L4m7uj9hPpA
z&YZA-F;b&gMZ<VUy4We1&g~58DGOa6F45h3VB*ug$2_Cw>{eFvKRW9~@J8!TCLa<O
z-kLaJc_Mq2L%BP@N?O|#nG&fbPZnIBSfTvcgy~28{I-K8hxfTh)b<w2Xl7nmnrM1<
zs@aeCkvGc!Zqcs&&m5s<xENc~B5FC3sih%m6&sj;%Rt2K{!h`mHzuJIEViv|nV_*n
z=<+V##do(HH!<W{sS?=G{_nS{-TYq$Dii!<e%~uzzw`3fvdy`D;>Le3ZojG*cjn)X
zE44P;ukY)6W-8ZrR^_yB?HavBU!v!p$}@_!pL%!uCy}#XPJdp1b^iVN`|WJE`c{|!
zHkMnLpIZO>)1MiY!N+ai+wCvE_~g;swC{EMj$Gb$b9>sx84|{_+hosA|9AJ!npfXU
z>!jw_zJGl~es=TD%E<SZCg0uq`rac=GjZeDI@4e7=y|egTi^3_Ym(y4+ASi_w4J}C
zBiDYuNND!2{a<&c-+sOCSM;~XuW$bS@GB=Q<+RIdm$)sSQ+~Vr<cV^W{GArS_jJNj
zk-&*3)+)}|xZ})t%c1z*m0Pk)wu%bsxLk{KxfV84Gy7<DLOi4GCjL!D1p<6=EzNVC
zEh36GT8LaZ`sjhHMa0~u=1?0;77sq9(0Nr+GH;kw*q<t~uLuY~v1Y<WSruhLH`Y)s
zgHKAwrKakC_^kAY>s91~=dPcGPx|jX;olP97}z*p{9*c3^*YPFr9XK2g&#h5{SleB
z`tbQc$M$%Ehtq?N+vELg+T;C>%hXxUwER<~_0WK;YkBoC-Exa3C6^yQi1=*5BFM*Q
z-|sB&@|xiJ*VhV{9Y0ufevizGmh1NI)z3~^b-d^bWszA^v{0L`ygyJN@kxik6NfJf
z>_*&00xS8iSgqh!l98FpKZ{jKhRNLMMcUGTsW&Uq=BZsYxMH&R*he$Jr~M0KVt?03
zOv)@-{?Xd+RI;%Bvg+$)kJo+q%l@G&Y6Z4LX+}(WXJUqO3rOsgyP#^d_5ELwT{)3m
zxhLhe3h;DJc9{0|p9$}qIIC|e;C?{&p1-%OoW-t9IIvvutfXwE%=Fz~&uv}YRsU<>
zz0WKB)@Iwp|6091QugPaT^d(cWxfv7_;Wqh)TQue6z~1jeW8DzhyVQg;rHq7*SGIm
z9W%>j?XUCCUVhl;>uFl!^y+tR!qV3A*MHv{6>TqG9;YidG3ev%ke^1sW>-a+{hcWP
ze5?Gv#QVE8S6NK?y<yXSzQgL}Z``K3xC)+H`eN-n>3uw*XZIDQ_lceSB9gUn?mVTW
z>wm=0n{Jc!|LXqq`Ii#YrRtv3#2QaIuw0Td<TdKx)D<*V2vJlFG4#H`Hd8XA+h;-}
z(^CZ}wMLyJr}HZ`XX-1gRC*<*!gaoJlgfm+<{*td4KA@7?m4VZ`YJ4?T8>3!D>Yv^
zWnWMW;QN}x-DJh-c=y0-nRkBDOsS1)1g>noB2vDT%WwAMFpjD)jw;K%)&=RU3(^bh
zn&ve&v#9gTQMXvQ(EWvK`{$PG4Nf19b}-c#UgECVxWj(Zx`hc#vlU+o?~PayeP?3p
zuA{S4k7n<>zwqv#XwAb6CuZCAAF?^{{pZ^C_xzbfm)Z2ZvI$-L_;-84i}uR!AAecX
zO#ki0mM+YRNf*Y(<|u{LB>(JY2Z5vS>$+^0O?hxXgImd^S3xCr#mR+xXO{Z3?%2KN
z*@XP~yXuxD2V3u-tA70E%&o|o(-%q4ti3WR++@=>;ofDA_e|Bx7G|sKTu=5re!TU}
zx!+NdWo0X_y5AMve0lz>-#@<YvX@yh;rGSTYb*DdzW%(j!1-DL_uH2XJ1V)SSALan
zd*yx6OYFE<^0(_#swBQPuie(W=fdflZ!4uQ&E2u-s^#i4o~7Zh+-y51T82%UQKHxN
zc+*11t8+?p*QFdiw?#-y{IZvh(AoVvo*yX7{F%Qtedlc7qT`%*Q`GcLM3ye#n9>}i
zQ5(R>#q%?uP4UIq3HuF2EZS^RIBhiUJ6dM2-#XZSgKggehMfU=Ejo6q_!joq2qX!<
znp2m;QQ8*$W12+Xlf&YUFRhmqdbd9lJ2$zWZ{Gc_^VkySwM%}q>GiwtnN@RT+Vr3Q
z81~mKlg1V;#En`Sp=Q@hf!WOl0t^qz>yO%&^&}-l=Q_58s_$5Pd&b*)xi;1d)fAWU
z|NU-kdGjz6SA5ZQ?&&R+p&^^L3)D%eW^K9fb(5*Zs@Z}Z>3hByJrLZocf;mC@jShJ
zhyGl6@jcCfBmUd(>yLhad|MC@RJd(TPVw8LpGtHpd34J}s`q5KWR!(o`F`<Hz)3&H
zS!@gsuW$L9aaLB$f34Ny=GngubZ5oIB@~{@whz9)W#6NQOHACG&g36|S>UwQ;+5Ni
z%qOhPQ+$=1x6E-9TafdA?PuMkbFDvKmJmPpS#!?D8R>o(=WS7%x@7BhEu~x!PEED5
z6QkFEmD#DB>uI@C!FIBS{-P>rziidaw^I$)&zPtr6zZXoul2<BdCy8Gr{{CHEORr?
zt1R0(H~-un*6V8JyX&WG=N<Do`_2DA{NmRY`XAp<%~%>c?SCEP$HfbFVTlVvq#?85
z(xj5gf>eEHPZziRypoX0f>Z-%JOgP)Mn)(t$f;q~5yIhO^CuWgX7$NE_VM72nT1QQ
zt&`fY`t_={>vk>b_@NXLEia%jNkib1Lr0WIvy&0$!-vd0nhJeNT?%0rR|(F4J8jnn
zyK80F_l50_x~o+eUCp+>{*BA+RrkMb-d+A|xAFbG&;Om<`Fzg(o#AboA~Sr{qzz2-
z++vd6Y?#UU*(Y+>n~BM8XD+8dNqePp@MJ{cf*(Ti2RnGwHW^Li6SUlE;n8?X>506H
zhMe2NYmEyZ%YA9t*OIvDka$z=>+TEMkFH;ej|g%JDOh0mD7A;<V9E@s*X<i4M84&}
zS*Y`5{}x{_^QEsIU;28}Z1LHDv((hmu80Kp-Ay`UaAU{yq@+4)n;ZR4)v|X;?%MKn
zna%4W<>jaSo~LE(VBs}CJw?4&YUjdbIeMY_7gF_}`gmKetz(>?eKmSbOxw{Zvy*E>
z4`=eaX-`kzr1N~qde`|YuSb4;8IxDG>)xT8JXLv7L5rq!o2K60G4q&7?Y7YKy%Glt
z!}iWIzL8`3c+ukAosv?MRh6q&swDTGtI<j6<y$8$Xk9(6?8uoNn!)L+ephdN%Dodb
zxz+pn?UG|YtDh~GJ#Q4bsbuDiD{M)JH)Omq&EYX){LO48IlJ#@GUpAM9KC5BnHxoP
zM2(XoBlg7<&AXz}!?i20v2w%iO*}VSCzv}~Ij=jdeu`UUv8e9Cg$sXOuz%eBp}(`d
zqkpRNl${&GRn^}A69~G!{@2r_OQo;2)XEuZ9}k>xYJNttLYKw)_m{UCRZm}*V{~c#
zdGC(sUAH6Sw(c#u-uLIMcDMPC7}YoKTh;&me(`wuq~jY-9+7!2FWVV;Y9`z7=wBx|
z<DYMv&t)=iW%4B3-Is2?xb^MT{*4(w3bR6-Hho+BWt#Ht)wj;SOYD1R^YrfHz*R@?
zTl@(qf4C`r?{;Rkn4b3!^e0IlwG(iwj9u#Lsu%8PXnf8_U%$0)egR+GrX`GR@ph%_
zzUJ)P)_2nUBirRNcTw+cVJ1`F?KXJ(q{UG=-nI5ZLw?db$M<duW}CG(UV2!}aXae4
zG=86^ovJHNh%0<uv~|71naJ?EDWdWMOZZ#Y^O_1jH(M8fViD7#H5OZwW&VD;AaO<h
z>u$-vZ>Q%z*FJUQB+E>uT9L?aGSPC*VJVWm(w|<wIT)=dDArKaU9jZnoQ-qWMjf1f
zLs|a4PWkIBd7qCf%gY=Btt&&99PqyDy{qqoV{+R9@qqi$hu<|XdRv#<+Wfpb=A7x#
zGh1TTe)P3cUAj-V{VdOi=lA+soGngP?%k6foHD6fY3qqb=a8f7)qEG`x<8tD=<KQm
zjp8m-8G^0Ne_!#PTl2t8{U_&~_X}72mArlV+1`yA0?mhx&;H-HPVm<w-j6*qKU|T2
z$hu4MfOYvMCdYL%HnrKGITYT&?ZLUX<2?U7i;%5bcVq}ZE?B3r$z@iy%+;%=Q<)aO
z>U_-lLg>K)h32~FiUo1^oIUG$Vzt+be-~~y{~X)nV6pl6=b0tXYVJ)wJF9cu**`*J
zQx`Byv2lO!urBFjl$}{djI;H95A$uG&#`@D&sfpUu<E}hqv)KbC1x|l_Y|#YQCY=X
zIhj**ikBSc^hLb-b<6%edUb|jC-XGc$8UA=(_H5$oL5kicz=27!3uM`!qbX%?6Lpl
z-rqSl>F<59-7A0n$<<z?{keGU`*my9-v5>-`<Y{xI>*|5;tJ6RbQT^Hua3SsFQDm7
zl~%W$UF$8ig<QFNHqQ(F8e+RCr`iARUA~tGx1V?T6mkBQd}0D`Va;7}jhb7Fj(V0I
zYpvgw)@MJl&bBn{R@Swvy_3TOZ%lTnb!&CtZg1AAW4R>sWTW)*-T#Vq8s#ZA%bXYg
z%j($adVYqlSx2RRq>RQNnO56P6FG0)`1=3GjSt?dsxO8AtUmTT)qP#ZGVM?EE*4B?
zlC5Gj$&i`Vuqc4X%k88AOK-zR13@-t9-Sn$u)qvkrNE9a4_25z2;j(0e9-&u_l0_k
zs`iOo3l8h|eLAU<WL9$GsOE;ACc9a?J=#)cg)iY+w9k?!MPYZP^c9~P)v531%-uX8
zGRN!W^~>Ft%C_E$IQ?+a<=lzvw(2d1&)l1J&6e@khq~^C^KxU398YyyRW(J(zwesU
z@*|V~$$nU_UGy<{W6LhlTHde87c7nc**~e#>``j?^{hMk&5?bECofOza%^76nEz>t
zY{IU>1yO2$?Ia|=EMr()TX6Wkj>yle<+m2@YL?NOr8t@C>x$cfoA1X>OH0xdeDd92
zJ^k9}MMkqPn97{VT>ic+zPT}HPq$OWo@ZZszU017H)JiX*{LsdPi@uDTi3+ecs@@v
zeQW0z|4V`Gp^)VM>Dzj1Y77FUmO8PxsNa;{C-gh0dfIgfb-mds8YwXfhflIC`EygG
za+%KA`j~0Qgtp)R{cF9b=(<T<ZRd53+@tR6-r2QuUD9+Hw<jMCR$txozHj}4sJ)9V
zmrH({b#i8q>T6%tqB60Z#7$DXqPx77mYu%xieI_)Rq);W)Aw(>b2z79`cAKxy)`e*
z#6(zb-hN-SUcYQ-*!o1TuIr-HM6^#XTlUoZj&`Ps|7`2>CwoPCU*GKKdexQv_S7oV
z(w&kMt+Io@tT^-e`Wqg}px8sZWldHV?1-t<=~8-=7hXB@c5LLa34L?hor<%*&Gf%q
ze#^g3s${~H@*W1x%2)Zf-_|Akc~btNG3BjfwUXfM%@Nb*hJE&Do7E^Rz|P5%x#96S
zjl(DITTWfg@#&I%fbhKY%B&jaP3P`C!@|E^&2W$Fw~i^F+~?GkEIG)`^tI~I1IZ&Z
zR41$V>ROdGHJ-1VVYy+lfAMAyZ=dJEEW*J7TMySv{8#k;=q;~nYY#4c-2J_Kj(-^Q
z{gbY$KFi-&OtR(@QxNU__27aAAII7|N2-6G=smh;u4DQKClT3K$Nz_J{HY*dvZCP1
zuY!UX)_LK&%&|MPD$?KRUJ`U(GF#rvs)E~ORrt>%C!-I9M7}*d)1dX5$VY8Er5KLS
z?fIp<@22MUFW<AJGTEi&{{E#`g5#cjtk|0>xc47(wbbE7*xEuwO&=Jc%p_$eBs@qE
z2uX=Dc*=H|NxY4XL8hEV7n|WoV{)W6v<xlGQMxEgLu&(q!^P?qRr<5eZ0p?kORn~P
z=Gx?m`l+k0-AZNc=3sQ^($dLNX7S*B->M>TMQc9Wq}dKXLL8eIJ>_0`{S4Sp7m%In
zDzQJzJg+>l{7z(e?3$>z_uBhs-c7N-TwnJ8wsrZry~Xd|&oh2L=l)*z1{O9Yrp5%v
zq^x7xj(kZywM1l@gR6eL;Iu{ObT6&?^6#*P-i&H)4~{+ui=YD!8D&(Th_QXzDD-fG
zN<kWLy-~!H`lH?VoCF+y>i_w4Lpne!^2zOm>MGrTyObxD3z#qnDi}TSZE+D()}J8h
zc%Z&AdivQnLHTFGF1z&>e((Bl;zay0qg{8aeDmf_j1rWT6n!=QeX!xfkR~nd`G%68
zOs906TPAd9<&pxenT9{NNM&g{2k*X8sg=uj^hv<VSGge*9oL!%?5tYmkzQdw$Ls86
z2@T=J=a<*4j$0eLn&(x{s*ZDEuQikB&AK(=TcMWd@e6KVD^CVpW-}EPa|>kKI_s3N
zRB5)7QjhD-hdp-YO2^v`T{7yKB$fVrKlI2<b$Oen`LVhwe?GcDJSL<(_3(?8Ia>pl
zUX(gi-)AxNqWFp}9r=5DTNJe?Xs8GVDxEx$ro#G@<)nMbiM)mSvHKKVm;5`zs`T`P
zsN;$RjRhJH+#lvGG&iugpvWa)xM=_9|C97|-#<{YT9D@O>|;~5mwrjb)5<x_S)u`Q
z-&(|%rM(K3V!G$}^sh~%+M<iny`hp%&LoT9verBG(q^&TVZ~oLTv{hGW-INBNj;&u
zWtmQ{)x2rtr}nnrsTa8-d+>o?h}?C)H@dkl`PT)i<=$6Do?1TXlH`>CrG`GUg$~@{
zQ@j(kRzmmP`DNZ+T&3q0&AT>@&D}OII-8y0#`l>Uj~w!~-Y$IAG|byHt!##@@4eju
zs!M9hc5>&HGpXOnn)J0I@#%^`THATfg<YBbnlB{kOd9vbRR;G~918WeDUf~HV!Ptc
zp3{7D&R>%ISX@zfCFx&j-M`NQ8ZDdK6^o|$G<_DWX_e5+|C7C&{eQac{(xlVrAPDp
zuCShs-E?rXYV(Hd4Num@oL+j!wCnlC!}9Z#^>)0U^Vhhn{_1u?Z~LCkD<;`CQj1de
zpRLkX_NcQ?aj&Z3+4kY)vF8Cl3LdF$WZV~^^8U<?IsP9U<0k%j@^9Iicb{+k=soy)
zzw~V>?;UARx?aTR{%x3WI_y}F-@k{ivme{4%-I}vW$*JfW|ET+l{U1h-^|vrRcJn8
zapu$cok2gJe93>lF=Ln9mk7@lF7izMA6YK?r(etdaCMdM&FYiYzc0mj*|_Yz72+9H
zFK~MQ^XB?>8uwe@X)trL^xoR7&&Yo+j%ixnBv)bYg9e<FFHSet@9&elH_hk6(*NQC
zzZf=LRtjTi-o4{#?4g%GUb=FLZ57F5<T;na_|0jOn|!m+tT{J3KKbuVUgRtq7A<(L
z{o*6z;|^12&w1&dIYIK5{O0&yXR7x7Ia#mqqe*qA-<FL+5wpB3bDkaiQ+w7doB3YK
z`pvgJ-l?5x(w7t7=k(3j=FN_)jK_DgGz)L2a(bctxP@ugj{~m*YaBdS?B~z_U-0wt
z?%L^kWdidT)!bav<J+dYb@>;Uy*mzGivDl;?AMvdh#lqriI-Tf*8X6d!z_M)De3Bc
zk=0jDFPgHe`y5wNzPyEbPb}N>UF@9YyJzohsXx0+diKdao(-RjeWKnydgQ|-A6dP}
zDNp@R2&3rxe+Sq<PP)*#r)U1}5=oK6k-`28dOGrgo-{5w<nze<%r_y^d7Br%wefw`
zw)Xy<W4{hueSG+c<oSTxntT3#d6#2eSz{gWEFtJX5%ZfFo(wZ*JL%QPt#)0U#rJK}
zozN}9H+)*oO_;Y?V)?~$+zWnd7cG-LyH>MTY0H)y@8&V7f1LcELCjq9X>_g1Zo3Qb
z)NgiYU%fx~ho7DJzY`}#ZLQ9$YrpK6#M~mvb+l;O@qgm&Ht~&ecWR|0tZJ7|44--A
zQ;USRw?60e#lg({?@VtsZtYvK=3cc!`^CAbxw5kDcUna5w5^wO6x6o<|KgrX#S~M`
z^^c|$tX<?0vw6AE)|&hP*HA8v13y(Z?*9BLqC9-UA5osxD^DlBH|%h`_gU2-<M5^L
z+jtvqExx-}-A2`K*&(Jirbl?6FWvn3N4(oL_cdoCrcd(tWc?#)Vx#baptx&6XIwWm
zm7UXga^{YZz39tYhwlCP-&sp<J<`xPtW{I*q4|;T{vq~5!AhSKwF?sN)-IU)A@OT;
zBVYWp<8gWGuF0u9y!}V}qFwvl8iD*3+vPJA9@_jku6BK*aq>3Spa2`$WFela#KqUo
z@3ijn{5-#C?~y0#BEs4BRod>gYEPN+)YGP^)#S*Rb?S?L=S#VL_$}_XZbQoZ_iz2z
zC0t||{iv|(%=c|fUQ<*nE9#uAMW#$hz0dSsH;3s??(eo8bx$6BG_JTW`ncNfQvg@Q
zZpO*KtUtb%(TjZadie?VfA_4EwCzqE-R87x=f@Q@W{A!><q^j`>E@d?A8-22zMHuD
z>h+s%nAZw*p02$S_kP9J+pk}}Rw>@`w|M#ekU!G#Ep-pD)!0OhDq}PU5(E+q6A~tP
zC1eO>6fiIdIx+QQGaP9QjMV0UiHSK%8>BUYJtFw#HakTXbp@p&o}|7x!Xnj-r=?H#
zDED>o%DmB-#^Ts9CE<lx*OnCFDSEC;IvWB-R3p*^PFp0u-#2rkPoRFWtlRTbS4&Hb
z-|x15P#wJIz3mSVzTb1^?!C7E-_~pUl@_g8t2xo*$(D+1i*|B9@A>pCat&{|dylmJ
zGvVBb1y9e(BrbT+^v0rbF-x4vhOnj}oiCe@8NWRe{&vl7rNYk+FFP8q%~sJ9QRRA{
zT)Fp>xWsP9C#<&7Gb4aL-)<FD;?3FU>D}y>4D_m*lp<Od%sKA;4venpNVT2>$*~
z<E;}pt$YhSwiNAYKi+b2+I7uS{>QYQufE=%^>~q!saL7Xf}(#BOV39<b;^~AysFWj
ztvrR%QSsFVUaz-oPcJFhr}~6U(DRa9bNQHyTd&0G7^y{_LB}#n7kYhK$IZnZ<j%^f
zm8`_nTypBg1cj$_QoW~X?cE}H(JTM#uLT_|6rRp7^hym0m|&!0WmU9lcXf%+hlsU$
zH)@$H)lS*2dA?Ai{PaG<xcPc9R>!LEow><9{fwNnto+(L3$AK1H!r$6Yr~3-{~znQ
z*0BXOuQ+%`Bg^EeU+;3Yz{x8$ue4t|ex>Bp<XMijPG^_wS+Vm<O%~%*=bu5{D~?_f
z%F=x5zEZ<f)PXg~F@$63LbITSAr_*ZuE85aY_`Vwbn48voh6m)x^B%%*(>W830zFo
z&R$)yn(t%G$NXz{lO9g8Svhsf{*sL)^S_wJy4LG*Z`-)BbMw!G-<WR~+}5x1eZO_x
z^-SyB{L1$xzwdoJ{YLxOnOB#e9I(icz4yiUMfJ@>U-3=HJu{b`h&p)mNd$Wm2ZLa7
zj75;<dJo6j7VmXJqLwxo?D{*o`$j3}HsAS6>?Q<6$A=d#u_-<O%t`NlNbjkeua7q^
z`|vu`M$SK?{ivKm_wJ8h)6eQwS02k}4*Gj)tC#kt0M-8RhTfVpOIj|3GA<LYXI}Mi
zdhDE9Nu$Hh)m9jFCvbdeluuaj!<XmBiTS?!8mpfdKM33X|DmVr?<K!0l&+hGeqFWt
zaH#4J&%{&B>h_;2jBF;I=bO`+YhlB*Dq-H2)T3Lb{;B)p@NVtQu1vn_!_Ob+v(7g1
zS^M<u#~&t>6|(PUG^MWXn>#mvqpT;Wf+eu1K=QN5?t(MS_pg3j$Y?WX(Z?QHA<<ex
zan?_<`^_$=K23ePv^w>4^m#Se)W+(svOEtr#!jozjgr~g_?bU?uFR=(Hw#u3N%QYC
zd0h7D!b{t;Iu;6N&L}@OTV~4jSK8J60N=BQwGkYO9wMp>pXjnVCs~Qlm$JM*_vktf
zH~IY&SpP1pf5%#DW%S}Jo1Jm-@36T}wFSk#M<?D@mrt@vm;5%XH+9A_=J`7h8%SJd
z*c5YaU$!!n`3DB`i{%g0+A18??6+-HG$=@paA?q6v-0GXf}X><*WRx!O%}=&%i&Z@
z3theV=n)yGrk{uS{nBo!i<*9Haxi{of1vUDrKg6kU+syo$#}uC=1BAWdIsim^K{Hj
zKWb+j`@GdOcz3|t?Y1+H@^Q>qt$y+ORQDgnyy8WxZY1k+zV>11U$w<8B8Xjnwv6<N
zO6S{w)}OZuI6G+G?7Fl~!f5;Q!q4VsyT7J?6ky))yuv~`^<U)F=B!u2@7G$MG+#aa
zP27>Be*U%u_cb*y85bv9`52zM$C_utW092~Bf^4R7r$}%xzNk_x0U?uwPzl^-7O}t
zY_F;F6i=2(=hPL1xr1BpKRMnsJH0ZYP+(f2bJ~h`kG98Y-HIsXf1CbVPNCK`e;!A^
zW7e^~ahK+Xd;7M!HQTk{o;7EaW!T1l4u|%gIyOn{oZZJ4x-VYWiY$}8oAFCjrP*h0
zVV}cX23-r|KF=n>YbOrYed@oFy7zxyfGE3iMw`l)gHa)ejFMLUvau@b&DXH6-Q%^9
zTU513Ls)fr@tLdJbhE=$brZ#Zi%(notas(iV?Jhrs`<|fxn=t&eG`+|%=Kx<qeaK2
zy}wyz@@(<U41;%@5AOc4r!9R-$nAHFr`lY0GtFCdS74uE;@2bj>W=z-fmMfY^r=5M
za9h}7Nm^p=oD-A2{8SL)-s<;kqQYB)evz~brfqX~vDn_4b>d;)I)7Cyi=L$sE*n->
z_etGsK0k5awBw8$qWUI^2;7-;<FP4wLe1~J%o~0-d|Qxw`0&y74<?6O@=HppnsDcR
zdey?573z@Vq;uRN=qh95&pMrF)9P%j*C;HAJe=R>&%qhCVwG7)&chk;lNMd7StPDq
zv;GWg{)}IDXIA%%UwzrNBkMnF2Gi>3_Dz>6L-QU@EOLzt^L&_>cX?$oi|2!Lb+wyK
z7a4ABSg*bIUS!&o)Me>uIpIgdj=#Sh9us%);ev^uOiyGh9e)xie6d$p{X(yNy-Qqn
z#fEpUl=(R$TjpA(g$b4M*_A)}q#AOlIi%|Ovu~n%YS|CI+9COEg3?pAa|QBqj^@i*
z&5UjB3r~xZU+8i$XM>QWa>Z5cJ?!_}ZQtugpAtLx&h+5AkEWNde@qLv;QRjZ<9?mV
zB`WDtXRY7w?H)AAb;GnLkBt`|KKX`q>6gt7=Qmuv5W4x$_CmHFuD_O_dGl-0?7iDL
ztH0lO{G~PT&4mjNck_4G>`=<GYR~>(bbnvA$(by1QODU5>kjBtTIUrnf4f^FMa-&W
zb93inr+RhcZDLjjX8jPc;aM5^)gXoESt!d^z8qh>^&eTY(q_ji)GFs|I;RPAJhg9_
zt-Nw##I=)G4lt^3?l<U4JNLL>cU#<xbxP*Db9Xjo-TXUUcjhsR^&<Z??U!Y8aB<v~
zkSH*Cb7CX&GPR%u*VJZxV^m*b+AVcJG^Rn2d-we>(}jym@-(I#wR3awU#HzFx7ls8
zQYr5$&L>izPrrI4I-fPHjkW#7wTim8>>^*X|36YyUvlAFg`g4pmZoOOXEP7pZz@Vx
z57xM&_BGF2=9Jf??WyT0>Y|VN70*b@{8!Y}&^ojK;H1Z9M;G$GS*ChI;_T`di(Z|P
zmCTf^vh9EVc%hJ^t8v@@wx*A73MZ^wvG=2%4A-tLD^kjz%Dw2dog;bvj;zW}QJ<h9
znau5XrB|(wGE><qrd(UhKTkzOOwGe;=HUx29v|A4&y;@e{#Wc~c=c-QRRup{8QSNH
zSb6TRjQ-HKCc^pnvIRFp<d&Do9$I-w(bJXd=lr8mP8YK;Ni-)0vn}`UFjm=cQETPK
zSJK?U``((_&g#3e_>G2L(sISsL5g$EE?T(k^;VV4aK9qS=xM<LJG$F{nTJi%;l8@`
z*e$=33AQ&fHHE)D@bd`FzLk+@^xDUEO5F9&*Zb4f?U%je{N~$%%QvcCys}&vz2p7%
zGn$bH`l22_mOm|h!0XhO2bue~&!7Fj@5z}`0b7gOmj$V3wtUaLH<|7K@4xfbtp3|v
zJ8jD8|G}rue&PIhrQq(hKVSYw7eD{+wX3_+{hozMmtNCq1NoSv`KunNTec+dwI#{+
z9n!0i;5&A3!I5<Wo^?}Oz8;h*)1AieKWopW?#nNxty;G6^Iqw5>>vM?U;1C?{xS4h
z`SEqd=FxfAV-n<RRz^i1myZj1V(gRXpVEFvWz&}%N*k0}<+c<h+@1fpu;S8@gEu(8
z=gTucPPo*6{@}#JWvri0c4%;0mN_MSp77=EC%1PQPe0nGeXB`cvaRR7bWhk1ZI8zS
zo6b)6ynlmDXrqApspjqv4eKs14#+#mqvK|nvu9OI)e7D8?}7VP3r&t`eCpQu*Nmy~
zq+rHty+u58wp8qk@tDSJ`m5Te^|jR9VAZAdD<`RbFyxj0_vNp*+-|G0&XYG-w^wE`
z$9ZL#ruM0pRqT{(6G%+IW@c43>A_Qh$mbW7U+>uHZ}9rHxcIJl*1pEF^=D@7`Odt3
zO5e)Ni?t`4PQP|HnDW9}-6mz(eWt0aKBY*cUo8FWC!438o3uXW!_jj!{m-hFZ_hjZ
zzjw>8^Y4;8EUw=CDU{-p<tLs|`MT@xKl#i(SEl{aoi@p5-k)%zzn-;cmd|{+EM^9S
z*_~}muAJrUVz~TlmQ--$_7ZQC!zIRREV$y_UB%}&y}$fydD26sm#6B#S9ZCY%9LeJ
z;8}lWPwbiu)7Td;eiWL%i$8cYP1)Hpx^dI$iVsUq2fn*@#mBjbYk8FF$_s0Aq)t9M
zc=>qZ#XHBY9=-kJvi7&-rEdE>G%tl8J5`py{_-!)kd5+*uQ_L|Z(nG=!<KD|?D1oZ
z&Sc2_7K)q{HEH(LsqqIy#h$%iw{qgzpN}4vH7!4WHGI9tl3SO37OY^O{!9DoBIWCe
zr%N{)Fjf3q>AP%)ZOj2j-*cOoZZ;kA=l0iJXX7ZUaBk*W<D%P}KLrH#@3b$8p1sUn
zGkRywnI%`g_(ZNObM`Zk?sHdqW5OYmBz5WMm#E-de%~@q1<sB7xRJf<>u0^U7mBJc
z^F+?RFX{DNBFklI_0HO-lO{Y)zaaNb=ZkI=<DS~OH3b|w`~r3Q8$G^0Ge1`(qvLy2
zxlB^H@~&U)!hH_QrsW;+xTK@X^d>EGp^o~0L9Nu2KO8q5Fa15uAbzI)w52`KOK05f
z`jpjG{`^_dZtuXrZl+@*&0m*R%k;EZPWr=?Vxrkm!IEReyzRHNtUpt;q(YS7lh=)j
z_qbXG6xg4h*u5kC4%3%|RqLkx)B64BpK(laYW8K(ZF>xZ!_Ts(=3I2sYRX>B|HgRg
z)|#FFRy*#{pJ&D^e%^Hl%OuNZ+y(!hN(X$JadHyByF6dq0_(}s59S&#{QZ;b#pSYJ
zhl;;EG&pz9;nOce&vW@5ha^)2Ej~7Fx8a+lx}SIQv4<&-bSHLJ-fx!vz2LQ>+|o%-
zI|L7X6*87nwb|$29DPqEXR80AKin&=wn$DpV)>8R<yG2?yJt!`wmneD4_?3Ed8>O%
zZP&Wc!UBOiXKwILTakIV;QzzJ#;MYtH)|TCm#pJooPR&pvi$m@)w3Izk2~LY+@Hp6
zweAMjnzF`h;fRUa;cV)HEzeHaU$#B`ZtI=Lx7O>`t=?h&S?JluuOjzXYR}&GX0>%l
z-6E}@4_?K*i}+kR<*=P&+{1^G*(W--dv5lWU*d0aMAGQXrY$~GW-Yuir}<ma@+Aqq
zY`@pLuA9*{cY5uG`R323n_SWEbG^b^;InGEm6Frje@T)vLzmb5-Y)Lq)~p-T#<(`^
zqNN1Gz6<_7U+PxG=$`HI{d#wnj`v+d-rUrUyQhTgjS~{E-*xZTW~Q8}+cy|b>xnU_
z+F6+$>YRME^Vx>34Rh~wF3z>qzVS`${LIVK&hrF(w2(ROIVpI8o#V#H89w`GU)v>m
z?R{_7%B6vOXJ1{Dd8}n`-Fl6>FJHEDFRx6yGUus8`j44uul`k-&y*|EeN_52_RE1Y
zslHQ<=DEmBJ;tScR_Wm?gCj~Co4#r^{N-P?;l-^}UqV{GygX^M?I3FeAG>P$*`V4l
zIU*<BbexX)?s+A|yLp$4>c^kQzG!VTIlseve!J#}*-KQ))9<Nni>{a`=%aF{(CGZK
zk0t?)JZv6i;?JJjD*ZYYtKS)*xOkdL&nK>;uqrOWaA9|z_Is1kPCFK#aC~@rm;4Ub
zX7)HAsT&3-w;j21<>Z>Ke<w9oX1o3RyJu!!|J|wG6AaD!>ls-J&v;!~vNj}MId$v4
zv!6INvaC1b3_r8T>-3$^=R$pt)$i`|Tllu?%o!7BHTRmkt-l>>x33Q@o-nQV!`p9P
zjd<;1oi7P*cU_{+7a=YE%5?Tw4;M?%J%+3Iz34b35-KPsCb~NE?xKI69?smYTWxuP
zBg?b!+_^ocSvuXWXnawc8#zNs>`#K|!tU37GgNO({5rvDufo(P;!4SVi)<v17XR44
z#Pi1FCHFVaIVC2r*?;Mdh27JZs3_L1Qqr3r5cWd#ZJ<HQ9$v#t--*-Om>cK%tV`{9
zf34fg`BYaAkJ)689Vzp6P7~iLs5<ed%x<<s>ADy1o8P}znzZHcc7Y%FjOHo+IJNOv
z>zAL8Uq#D2o=|*f!Z~Bc7KxcNV}gQ?ZCl%ABC@w!J3asBtc#8bZ`*bSh<u&+wEA$;
z@gqtH&E>DX)zz3(wC$Sr)-tmb>!#lnbJ_c9)8<W^H}^zV?ri=3=FP0~c^bD;&zwJN
z_U!$Z*Sa+~-}`R;6K}rr(RZ%Yyp-aSqSVA(5DT>L%*Y67ISuX|Xc%Kg4-6)R?D>;<
zy1&gle%CAp!#ws2AS2NYN1C4^b<D`r(%b-fmD}1#&I;kwd-Da<L{_*SaI{+do`r?Q
zlJ$u|pn$fyZjjFPY4^f!EepT(#iQ8#_>+BA@(ato7K$&Gc`C5e(M2%dvCQ`R-M?j5
zYt{GNy_@e;m|9z6Up8OAcGfe`XJ^m+`M>}F`-;<t-TIDH*xY*}D;?JvBWziH`C0f8
z$&EAp^=It1(>Jp<wfk^W?3>}YLN$pE49S^t-n|K}HD#wfKD=bByJO$vzCiNepL-7`
zI$vDz-_YhQyZt|lwvQ!`na<zuP3JuryfeK2zhOnxgWmk6<;jcP5^Qp8%s82E2;Yb|
zIP&=KpCzk>j}&&k78Wx89Q2t>{9N>ggpX2%VTpcOMPBm?9yYOdKVz9UojZPx`W*pj
zeIfBDT|uoIXL26m{oFG9Ip_8c?j7a9t;R=Z9@=qa%8f573oSe)X6|s-IO%)lq^R2S
z4%Nt=o1bwi2L~0mvObcTIN|r~$us6j%$_|%VM>ac$}^77K|-zB+w;3c_UW8EBb?{e
z>wo5qugJN%+9J=?)VP8_ecCi{uD!kgBpd%ZlWpuft<}owetmG<`I-6kvwbrAREyp%
z`<i`dTMqZ^>FQ^N)jEq?I6Eshu6()k^S{c&(MK#LH@=kkzF@=g4J%_D=N(E*F5a;7
zhE0y$H<N7w(Ji*kx<}_74NK5ZSiRxujqV#!IkMXrZ+GM#GE0u$z>%Z<jcXfk_mOSM
z>yxTCO6Z7n9}e0O;?uG;Su#RwTCe9ur+eG?y#En%cjs}XoWKRl${W4zmE6l#-qyS0
z(EP*VA8)(+$^VqQr$0mKh1j3uDe5n-F3@)m@7rgW_I1YTWb<BI(VTh@*-Zvd&h0o-
zxLi)||AsOf!+)Px9=7c1d?>u$F38u$Cua7RcY%K{E;}i$KA&r?g7OJHFOQpTPR{od
zYYL|K-wi%s{k*>8eSElg$cmYdH1=-3$&mDi=}xvuRM*O+xo!)WUVOW-KRK@B;!+7G
zQMY78=U2X3{25uBO@3Q_c{^oo)5^~)B)4{6n>FdI-KA9R@~aKoAF{}JvC4hXiJPF^
zA9zA?<(@ek8`i~ne{XF5<##IX+?7i&R3qFLxG>&fvOnT15?8iv`uFzinrqzi74BrY
zDVs~FMs4s_yF0}z)mV2b-?Cd5z2cjFcWj+xc>6?~((Irmp1Q_Obt_jyxhUF6H9uP*
z>D||I*QR2}!wUVqr!P;fZ+MlJ@hSA;*1J!iOA4)>Dx~!FS-SE{+gHVJo9!%ieG{l%
z`0_-8eCZBNt!ai?JW*Cp7{V8<j=S;r@z0a@Km9$uaK?|6yAw_I7@VW-{rqD4>21g9
z4)zPFjvJeH8><ID&ra<6(&MPUSkUxJ_wrR+Uv0f|mp|6Wl-FMHz{&y<nXkHY`Q8h7
z&zs-J!r&_w(<1g~!Rdo<mOqd<sQdj^NtJe0dqvh!;S3fBe(p<>-G}oEXTSMUdZY7j
zXp{ci^C_1VH9qF8Ot_>dC-c?nOjYywcdfQZ;t%(?t1o-j9dxjChrD^@nY^}=Ox>)#
zmolz2&&-O+3z1m5<j~m=IpgdpReO6FjoUNstO!4*I?0=V?OsOrLl-}2{!npve8Eb|
zT~2sg^C{2l9hEmq7w~drxGN{GTgCq>_w~Cs%h>L$??~Ew;l`QXk`sZQ=k2c_Z}z|Y
z@VEbua3QtTC-YP#7e{kS@GcH&@Lm1UJb7E#`y~M~LW-P|7N^$FWn=7VvYAnKW^Kx;
zEmCf4mtQ<3oAKvZ>Xl2b=?|`S`+j)I|B&NXcXqr?W&f{#58fOtDVyoxuXVTQY_j3i
zh=X^%4!q-M+hb$#$f{@W+4SIjsySKxQRyaHR}<d7YTc!jRekQj&0yX8rt*u|%gERm
zI8?In6o!8B(-l5E;pn_awYt6c-aR&qmA-hm!y))$&&=o7{)^dtIyryUj`#nqKii7b
z+cQ3zpx(Bpb9>@B)kD#D6vURME=j#EW?KI1+xF)E2fvT(dsJmOLt^4AJ+Jet^)lXm
zDgCj-?XT?GQ|-6+KQDa!vt~y5lpFflVn?4ZN$L8gDgA1-ZJ4x1W!>{*wjUjf_+DED
z<zKYOdT8-OK!)4NF14mkj{QLC{11PxzF}js`}k(QAmfFp?5uYy8SijyKNzC1aC)h*
z;7pC0lan7zjnzN8Zo#s<?YbN{9RD<&PQ2w`tNi0dbH*19*-yd;4D#fx>q_NTzv(rp
zY1;d(q2P9xS7E$onD?9J-dDS7lw$X$9@XzzRXjg1#(Q`7-I+l_`Bx;jC>%N3`=;{#
za=RT@j_-a^-}pYBS@Mu_`WKn(rz;tlqJzyAxQcM?e1GkJ%i15Z2Tk1zryYw}bZ&`c
zbCmz7yFGW44X;NX%r<y^uXpNc$ym+B=dU+SUlMe3^Oem_+!s3Dm$P-T9p_fKHsxc`
zw1pw_y`@k0XuY$x>DhWIWvyv*tpQK|wv3rISsIZ6OH#PJWqtlex~)x;*t)L8?a76I
z^UeFtie8!R6Uz6iJfLUx8L@YN4tt#cbVyX@3TytMP4D7cez6^mw!Qm~(VhFQaz_2(
zpz@x73z~O6Qs`Itnlq<s-i*dW^X#@gd7W5r+(F%tRX6>T!#)PR*Z;eZ`hI<P+H0w^
z%G!U0zI*c7^!KvOw+}ZxkREE#_#&WGWDCPu<%P!U?OyaBKNnZqUQ_;fp@GSxlbQZ%
z;azu^9$054$^NMM!sqjz>8_K#*YB^A_G@rtyu=*(Av;M+YJV8F1oPs;264T|t)X7|
z%9UTAUH$I&bxAJk>UKHBFr6sF`p5VFoi;!B_glaF!Q=XmeJ9mk%J}q7;GgCETs@8+
z!T$4He}XwzA5Ymj`Kv0g<S&in7NIp#x4NW5dgdukEo(lVY!rICk<(Z4mG`&QCCgKn
zJ-Z@m#+qhia7p7@^mj$=uLpO&H?OF)-2UKbbWK@%z0yVb?fQ9Pe6yykV*HuGvd_*S
zomHl8^^5K%@piY<=Gu?9{9L&5)uk0FUYc7&G+&fOrzYFJ)GZNz{zPPn%lGHywYS+P
z^I2K2x&1iaBxN3anloE)Z_owC=?`N4&aZ5qviV>@jAQzglSSMfo@+PlTDkHc1IJIl
zN6kj^)_>)?cO|esymB){!8TOCt8?3tfLB59qHOMLwO|!BbxUTwyP#YubxB|I8VL_E
zDG|v>3tDD)R4SRK{|uF=+9s&?i{sOO-paTJt3Qk#zh>qyKd>_Xj>V;`ZqIk^IVpE4
z@!7IwJGSi)f4@olp1A7JVNJ_+|8;8$7Ehlm^&qLw_*jmTLjT={=+zRNzpe{9{rSpF
zs|#wUvw9a^@G#*z>k>RoCbs&sOMQH$<=p)t`+{fd`|;itmOnowVCl;q_3m}$)6B0-
z5M8zF&7qP=r<(AGYv147`zudYqb+bwM}DNFXRi64*Ou4#+|6Z{`kTwjy;yi|_q0`+
zdwKTvAIh7*a_<M>&Z$OiLb4}<($C)VdMdqIYxbS9=iGMv3z3(poc71!((de7FE2N%
z9gOU96Ruo1v(S3S#e@)pC5J+Tt_K7@-q=vL<JPmS<~Q~%T-U5toa-LoUi*LP>nn-H
zCI6o7%~+!|PtW)2>Q5*4u045t-<y`rIWk82hd&ta@n1NXKmXv}mYaFsqfQoy2J-|=
zS#aRMlvf$e+kB$jrdiF5Vo+u+HCuM)(A0kqHU#WB=l0>@^HP_cvovp<v9RHJydjLy
zV`FSZ`{j34ld}qpOP_kVxgS6N>hYr_ox&+@c4yq@=FXWQ;nh2b$=>Wr`|_2cuS!Gq
zb!moJ@bK;WK4Izl_3`(;EiKlI2W;KF^zXdMN8QiK)aE{_zVq~6{4Jl-Yj4(0iAu}L
zTz4b>qHBDc{)2aBn>Y8>=x)@^%+s0`D&~;bdNL-H^>^&q_<2ih=55*&<GRevD_zHX
zR_AG<(k>>Wg@*5*g#UgR9A*%B<IpkzUe$X%6E$L0R-Vg${o+Gu!-N|TrT3qh_xQKF
zlk&lhTjKJrc$+3H)L>&>$SUK1e|d5L#@adOZr+s-%yaB*U)|cY+VzXv-~2~Ox*w-(
z`{L*_HCFY*EFaUdU$Gj8oZIGwRs21ocrX0<&w~GZm8N9QG_t+$<UykArHGI9%O_|2
znILEW>J*pPiEw^3+1KmxGgmQ~vTNx_pAgWH-S>L6z0L9O^(QRLd4o4hE!-t>%<|5=
zymfc|o~y1pSlVBEL7Ay@-fPv~eKNat^xAe;^RlT;cUrhVMMX|hYkIEt6nAN-t6a$o
zPpvDz@OG|P{Bg#;@7HDOU!U>hi*}lGv1n%En`>DN&NsvEsO&KKBEILunT%B{xMm+t
zUVigpo!ob`Sw<x<|4CIzaNoU}{PNwq%jJQ64jdefo7?8@uTM(w`#ay}#GHG#qV}uj
zJioVB@_}z@NUVwIKJB;WE58)_W>xb{jhQe#(8v3+_jZFP1{Zwdq<6$cm)~(WHJ;g(
zv^ON&XT8(%g63-x%hW~BP5Sjo%6%^LmqSOtw{3f5k*t@`A(ivm{<m6b%gw+^OHZ1t
zc)`2kk$vQs?8*Pm+1~k8Tl0VVzt-<F?0ofwmPTD#X~fsTHT|ONd=m@qPo7g9lLMGd
zk}{OJ(z8Egt-UzKf8odfZ{|KWS)3-ZH_mhMNoU<nCzf2gZ&UG}W1CD@+rC5IIcm<T
zPdCnbm182ZNqm+5{Eb^?I_W-LCS~PfaYz1JUi}S!xhd|`-}6nM*7X1Qyg3Ozrn!#2
zQ<=^0|K0AQmHGD9hwpJ^QNsLs_0m!6KiqWO){>F<AjgE$wo9!@tv;<pSc=nquKk_r
zE9{$=KWTY?S8U4M{nlB=s(z(grd~R}W?Gzd<jQqt{dCXx)P421cJ%)JUoUzWKkzT`
zQsuGs$=l(vrqj{u%Mb4}y{F!ld2Hol%CG#tb^8k8tGkn5OtANu_4o9S&*%U2`0hFT
zefvg1mKd|Y(k6^Y7gglNa?biYRlttz`vDb=#;6TX1U70&w(i@-pzyD)RdDqR*4uV%
z$^9Wt4@)JS>PugJmM&lM{XU=j6Wi=M^Y>HYpKmUEp}eoJqWEL~!_-?#&(1&4u6Xe1
zS#`$<enI7#YFz9SXId{R59IXRx1D{q;{L}bwF_TIRP&eaDa~h*J=9nj!`NTj7R;xr
zZkFYkIq_tOm$yz-uxh6HZlACM`BKZRVQa5t=uB*WzPvTl=&NI@Os(j|m|fe$WVg#(
za9x<O&?ja|n09zjmgB__XQE&JWAy(2Q2E#4?<Oh<*LKf-sH%6*=h3H9wcCz&qN>e<
z=2tb82rEo{XXovoKY!nIz0N-;8pF<Ihg=e0z}fzpb<W=<|9IRlcSm(JO^WQSc62)B
zmub<e#<A|;wCat&gAaZcz8TZxy8hRLIf+x+iwkcR>ifO#^7K2)lK%AKQ_KJAEn9sy
zy7?#{4A^27yVl>JZPk;WkYw%R3%a?}uJxTez!P>yG~p)Sv3J|7XRW>Z#nFOUzsTh1
z+$B3yY{bsBY<}^3)Az8axyxI%cTcwC+Wu(nxo-JA-}HA%hb}nuN8ZaK*(QAgzx2em
z1c}whc(>KB*;hNarGIX+yCZK8<J75YlUqw(=tQwfebt$@YsO;XEq1c5TqilMFS|82
zAvXB`HLp{7TW`Jpf9ZB}O&Qbf$rj6Zg;;RhJJD&Wy?Be&+-Zp=e5$ooJBvLfI_0`0
zq*_fh*iNL(nKw~t^}TnEi|v|%-!oV+t^0oP#QsT&&sF2rJzT!$!;*ZaXpYLc^L^KR
zRIhim;Q4kbD0yZK|2EOoj!P?fYQ5bzFTHrlC;Hn>KJy9bL40o?tT}h3`TDy%3icHb
zeNE&R%!+3|ab(JT=Ox!K-Ti!8U?<b+!VXTpxjJ#mmsh9U&NwPJ*T>86O?0he$;+_c
z-~9~QeU-NCs=U3u`ApB;%IuYrOnJSF?`D?Ga6a3|x2$j1ANDl8aJ^%5`R#u%=yE=%
zqvw9&M&Xf@yz{n9pIq+Mbx_mQDfU*qYu(#Ev*nU8jiEb^nlN<wYes7;uU;N3QYDzq
z5PRuaRD3EwhoD!@Ki0K2JAO;uIX&|obAaoX%wL<V9s2z9W?t6Xx^(xcGma}mU1e*Z
z9Do1!-M8?y+2QHZrdFH=r#{Tzbih-hK3+`2SV~k&U@vdI@<Y{N#-ogrj{oiJ`n0dg
ztb5z*y|=XP7T$a^S=F!BH*tddvxzE`xx?&s7F=D`eE;3%tr0c{QjXhdxoMlTh$wNX
zK4R+)T(7>>Z0_Aequp~41mC;!a2?03d!;*Dod2<$Xjhpd&@k`N?_cc&_qICrJ$N>^
zQ8@1T=KjOWJFT^|AICW4_?LZ~+{=62F6U*CT1A7$@^e9xxT*t_eTysqPuwx@YevbP
zH!Gg6KM}R{!G<D-opVJ}*UnJ8nPnZw<9B1mVV=nFsEPB{^}6mAnf=uL5n|3`$>h{$
zWG7a*F}t@uCt<#7b4Gji&Lo3v)z7L==xRK;GkIE{Lw2Fx?9w@B_0x62w39b7Y`&RN
zWX^t2(>+>R!LgL<o1s<ju?0pM%qRB=`v<O^(!JtQqMS~8&X>;MX~OFMul)no6)n5B
zIK}IQkYnMlNdHdV)N5XT(=1DvJT|d21r*P6I(trRS=+C&`8meN)lPBTTEG$Gkjdbh
z=DoJnn|FR;lfbLyz&i)}p5|!$ejLY}tP|3FxAo)dxP_Z$EPM5E@8$9b3U$>t{uF(8
zTQ5*@aOU^iWzRG|@9a}u;QC|ILvOx_d;F}~0s7C~N{<>9J^%M(^8ezbh}8Agdp6wR
zuHF%+o8Nvg?#Gennsa_Se+yTyh+uz|!uk4I;3JmhTW6H1sqeU?*?KC$<U;fOdoMm;
zWjk0|ZnvL<Q>ZVhvPyN<P8rvHjSa_nY6>{m%CE2Do04_<6jMsF5yS2~QCjN8lb7X8
z)XrTSo_B~t;!?Ruw&bjR6)(?l2+S3oEiEl7vMuD}r4JobXP=GX$S%Dmm$Td|io0vp
zi$iJimrQS3+;>3Sw&aBMojW$COl^NXtL4*C_MNZi_1lzvo4`g3C6{2aGfNjKS=)cV
zWt9{6%l}Q!?>pPvldoOqEd15-!@~QJK1<h4=g9?IJ0@y)iqvGk4KGk)|KnNU!u?~8
z>_`1RZr^>&KYi8@nfq{a`v13eKW=Z}H^1#`#{DudTr?%a=qlR>B`FQ<6rNR=0ydf$
zC+mx@wc7Ll(7*Gu)?GZu7M&_<IpbOP%;`;OS@m8?cIl_DH>xZ<e=aobq|l2u8{S+E
zeDY}X%;o;2+qRoGz1{lv)~@=S^M4mA&teykSz*KM{=e_M*3E=R4R@x`oIZJa_^S&y
zOcxm5Ryw7Aw0eSQ@2zzbcj~u2_B(z1Kv?ejss*b(;)4YCwZ{}#ewL|u$MGlHn4@vN
z#OJ@>JMO9eF_`?H`E~r6&Ux;9pEiBnxk#k@Mc&Pw&n#a^B`-IdJ7xOHEnE$6oIQoY
z8jan$r>@y}J=Ty>-Mm3>Yoxcj>Wj%croNowR~Bw@r+v*WA>-Ujk!FTVW*KHoWjNUu
z<{2?_<G~wW+BCNmSy;cS-!_T6fVs!&y32OHzr5$oXO@PBOg$R*bV;K2>5BGK1uHu*
zUJLNk+q*lnVruQ`&q>K&4YOC=PAT1alHq6UHx{*1#WTIP82`wLSn&FA=e7{`GO>4l
zy$iXzWY;n6QgY7IU2y73&&HlI3+Y)=(GN?09C=cl-oN*?lI4O7%LOO4KA5`Y(x#+U
z0Wb5WUAp>dQLR!|>6trBQx8ATF`s!lcgyvg*RHr;y}oK{+^&@U?JE~~b%k_2O4`Ds
zJ9m$6(f2IXl_iE{{U_gSXlgq(tugP^>yRf`F0oz>4U+$QHq&C|M81g?e<xdV^z7qV
zAM<2$WP#+x@2Tc7KLl?^v#Bd4o$a#q5a262+wGVwHt8kLBE`zDtL-fQa?RK~?}qZt
zoyuK8N3ZPv|IxSN&4Jrr&aS(B&u`WHgk#69Cca2GBa<g~ZPSD59dj3Nujq(xw~)V=
zu(-3&ro1k2q1Q^5d7tm*$J)w<1m+n<dUoGiII$?MLPu(De)rq=xmFk09vt)hfAC`C
zXI;5_TfXzm@0Wcl?`<+gw(Do%<9*AnJuPE)t$5`9dOQ2nf5ngY^?kmZ{P4wrpZ}T{
zJ2U-1TRUTEQ@TxG;*H`v!V-p#k5aE!B&^$cOPe9t&8@51-S}Yg-sE#VU!yE$KC7$B
zTJ!C}gLhKzwjPej+;nv#??L{J=iLAJ3;*X%JklYDZMYt3#|NId0F)D5vl9{~7;Mlt
zt=aKj{&BN4JD<S>78Qnxdh83ZS%9=amDKqF3lj^J?bU0e*h_@3&yrK@KQ_hUfuqWq
z36jnY0#)zNF8iKkwc~r-lei~Jt7AmJ-%B~|YqrhMZ14AHEF3CR9<VrhE^KYNBY6G&
z+rAWqH{ANWtIKrdZ>GPyI(Ps7```OnW=s`T6X5uzAHC^F;m=iqr^DCqausmOn3r9N
z;@9uo@4ce3P|>{MH_MkK3%17*jPB)+-?ZFu&^FmJy`WrwwIBPIbcS^`tsndid?tKR
z{C_XEMxO7B)RF(@F&DBo{5<p_&e>p*wAz83jr|5<(mj9MO1XCZeVX^}o#Q$=?>n|k
zhieZlKHe{9UG`|tf(I`G<n`Yd?EaX1+)pC2ZhqQxgIjqkyJyD-q|BJ#Ki7M~?z@*R
zmT9})Udz+<Wc{xjyxd${iobp+`}lne_opq<?>C<G6M7~nmG)dzN`;f}cg>c&Rr&J`
zcAr=fU~l~|e?|NLcJCSOyXUH0xT}0c@6PSRc@rP+`dY@_%yjtewBI`_S=uUZ^h%|s
zi$8mq_*_;_&ZhRvjDpVVFBSx>H=Orp-Uh>ev9k{yn_+LB^>EeH*Y3LUM)?O%m2$ha
zySLu1WNW>(v$|#Xk00&5b*n#~T5y3uyuRoE!i|hGWX(8!i%9qN9&k%qyy4;vi5wX-
zx!DcAN7NE0Z`8aIk|Sj%JG=4hv1bXzhSg^nXSbg{E|$Q((e*}Bj^;L==)ScF!je{R
zG`$g(BfCwm_=s78^hVzsVL8&zS}c?GW@tZa+nFpkL;YFr&xCz5<e&AkCNxF}i1oB4
zER2v4>+{^ep~KaEENMf8j-2;_Z^`Mb%q+z$#jN!lmCrxg_sv**vo|baNrOO2+2Y+(
zIjXt;wu&CI4)b24_n7-b)y0ayo(yxtNhh}4OWM17cWu9+?YZwIFLrKB4?SQ1jq6S6
z(`Js@`^*-a9pg{0{}G&j{@(p>obQWom95BacpH1E+2&ICo$nJR#M!&!zh?jK`1xXW
z|6}`o_v)+u3;+G!%(c+(!t0ma%+GiYcrt#coKTNE!^O6#v@EWypZyr$9-EAh3;zH3
z^7^1%bG`JYOph}HcQ1H4@Bf>lkh8&1m09^t&XaBKTt0e#doCVX_2UeiX7?Xoi+S^x
zo#|PcEV}83iP+B4e=>Vd>YfX{dVl_pP;O4vJQcHcsa0EE$!u-A%Wr>3Q9#ytalzIY
z<;lT}uN@2?`)ABv*AQ#+ihJ`>52dEC4{CXbHJAL7P}Zp`2*{7FXWkXNFHJ#s$BEyM
zCNF&5VY<+0<@%a<E9UKjxeJc2o|##AKW59+dD{wZPiojSW7Vuln^x`F()KoM+s0o3
zdA7ALru?qBHu=Vqhy@XH3AWeOOBM^edM&r-+k0VO{rAS8wvvL&p}S=koIWpFb9#}a
z`;jG8Pi*GPM#o6=&(C<>aM?&APVPrCi@APKLFeJtJ5G767UA<mCm&tV_;y0>aqrEg
zGb=;(Z80zl<y_<>p)x0oBmHqJrzPjcOGj9i37mD?EB|2UM=yoN$x914gJQi3xwzT&
z9N!4pXhqJO<-sU$<$%NP=Y<(&a?|HeiJvli`uQUVZ*I8n5i@aJXv#&`ovlCD^kxdb
z?=7><F}&rpjd@<(n-4vA(vuz<ecYQFube3NdB>MeM_*srYgis&zijjD`lT=0++H0m
z3loa-dYPkkX1=t7(&5Da?BDB?0`q^&t++W&)8oVf#Q@c;3n_953HgGRj|(3^;muKf
zC2;<}MXo{fua0e+#?lcHtl`2-u3p}Du{%=w-Hed>b-eQ~I-fVYJ^ih>)q-WsHV39U
z)ElbU*LS&Wp0nNcxX}l$MmF!l`bS=2uf(=}c4xj=BFC?vbGv44n{9nH)6q8}TW<vK
z)GA<qEdTJ~YmdqEtoLZmK78Xx(E{Ct3$z2duIzL@b8?z+<kT<wmI?HGZ95gZ``Xk^
zhLv7D%L3gFo)nKMn0vv5>sWWOLt2x_iAf<mzf6wiNeNxhzPN1e2M!x9^LIAvtly6>
zT9}%%hfRa)h0+4u^hBXR9;2m2mL5KdX15>v&YR8Gzu=;l38$~axf8s8ZjT>$Sv=)T
zS!lZcOV&z8Gn*_kj<T`?z5OkBSZ)N*nam<`(rc&B$sWF#SuZ^EXKY~iYYKk1MfUW*
z1)s7~VnsJKuxKq&*|cocnjP;>Tk38OoNvhCyf^Uv^1};Em%K_`FLR7d>Y%s8m!N|0
zO9j?F&pEO+acW~@Vq#<C)Woe@r%v6<vn%uK^z1*&FWljH@gs;~`is^Jp1S)auDyCD
zD=Q-uBy%SuV#379r!&&n)|T8#)1ACBq)5v&cHX@5*J3IX(!o{+vsm01Q{Kr{u?f%P
z^p$>CYZ%j=VZnE{vi?Jf@WGiA8V^6n`nI{wLt^WxY;Iwhon>bOsxrcR3U+gNOV|p}
z-LEsVzUA&_tLjTSr-XEa9(yX=96a8AasSzA+v78>UmgtoW-P=VWW>MmPCiTI7rzw>
z^@}tw>^wEm&6cgU??^|D5l@%iyW5fZ3DX6h#As~mV)M%6ao(20dqc5sS$Re1Tdrj1
zDADCnE7Q*2yj9z?u_pX*;jWFkdF>O=_A;KzuuDn%c(eFg#dVJ5AFB9w@NbuRz|nk7
z!Q|$_n8Fnty?ngo4#k;Q{+Aw^ZmS^o{FiSU?*^3+6VF%nb0pb=<JG$&SlmO`yXOdI
z3$QwJ-mCn!_TZ`eQg3#x(9U2wwXiuZ;4<Un+wtzcU23|Kn%;p%&3lhGs#I-`jxDy3
zE)dva9=IWH$@QjB%lrBR<x_0a4F3E5wUZILkR~rOZIgFGy`h)T!SlxZjNdGYf1+?}
zF28T<l4D;MI$kK-@y8%UMy%-eL!ZLQCG*{-g+t|LZ_e5NHT-OQ-?rfXWAmkc-z@ok
z@!~Am+}ktm9C;St-1BwT0ngWMfz0)>f2@zxzx&ZVlfPbced+yOr)E3u&3}CR^%Lvv
zmBrI<t&7{eqfjpRo~GFC^LH(BFYWm(yZp6&-TB#%4o6n_UTJKYx~J7U@&Da|FL?(&
zFKaA3$!sS)-L<h|*=O;<8j+439oANz?~_+-5Sg+uqgiyn*zuBoDqXLmv=es5yRbep
z^)};pP<2_O(7C|V!=qq}{^d(MMf-zS2Y-4YlUucMR<2|6|5++#%NqA|^GiIe+vmWy
z<9*rxYkvcn|6Fk}E3PkyJ}PGa_W$bnE=!_rR|s#pu$I4ZQK#uIp(nQ{D(s)wx@z|k
zean2M1g}Z&InspA2uz$`e$+u<P$n;;eTGupBjW=n9=ch4+^w^hA!@7rF7+0*%k7WS
z&bRS$tL00aP)J(#U$><Bh9YbC&tvSaQ8nKg7u+~2rOZ*CnkUS?zfDQ`72n!Uc|N&2
zt94mcw)!lxu~MAErQfw?ovzO66rH@byYqTpG|c|`VR0X;e^Zvyms=jU5Bn>>^RMGN
z`u*g-y_W7fVtK0ls@pcKJ-6{*9M2t(3D=%(+SGQXN~*5nX<}WQlI>L0-N_D&mqdJu
zo=iJ&)Z`{_@$Gq$`+9vZo?={ZBKGhk2QQs&KEaC$OBI3_pS@t{y7TPu{Y(3$wz;Qo
zbe$*l{p=L;2+?PUcC=Tv<t&%ICTA^tcvA86Ra>G$_3j4btXR`y_~Tgh^1SqupZe8f
z<<4y{e;r@nxFnMQp6`^2lU=wTY+UVhj^9sM+Wzr2p;QHBv&clf<1ccS^S|TwR-ZLD
z<-E&}?k#KG9i{J<@n}TdmwGE_cwfVfOE9G1ZlQGW)FT{^d+!K%N6)B<&){m#Oj2&l
z5!rD&Wx>~^zpp#lqgJZ#?r{j~@ZpQ<+GY|Zb>vyw=6V~Mc029lf6MHz{I-;ncUwF)
z!$_(>W&-z|4z&YYujhr&y?N5>Q~B<vmu@Pop3utY(yjM*Ql#&bD8}pa?u7@2Fgm&>
zwd&6KVEQ1wwl!$dgrbXg3<?XA_Es;MeCBGa;>0ZnkGg6vbk~ouxb68qu_HnF(W2VD
zn*VF~lT|ir9Xo%sch>h3+bxZ@b=F0j4gZ~3CbcZ>u+)Z0e5Gu!LazCXC$3puvS87U
z9?q=H&?%Sf<M!1ow%lg9rRna?D^uc}Cd+#|s$EzmvC_tOmdDX2>S9l)q=gw|*d01@
zP;38=05wBZn-jhBIOc_htbWbSt5X>`YkFwl=Gpt$?YPqux1XDrbo}RRd-+`-PHg<U
zd)D`?nZ}EDU*Pgd+P&aWrShGdp)viRo|V5#=TT0ZEqgRn$amHqX`iKmOFnF4m)mkY
zX2I9AzuKJMdW%>}MAZJgoo4@gg2(NOx$eEY_px=keLKsWplH*@b$@s3d;6NAZ}oSp
z6Skke@YtxbclGnx=Z^&hEKho9^+?a)m$~85kNZ-yUugQR-jft5TBD~Vzn?q4>%N-)
zy4J9?hTRcd-!>Lle!P8kvbg>et97XXf2^wuq$|YNS?Qi=RdtHs4)63|Jjt|$_s{Xy
z=d2&>6S(&_okuZED=yk&c6Ikkqt*74!|&HFbN#FJXxg`}zqnrds2y2zTQ_d|v}m!r
zKY}v#Z2sj2e7N-OxBas1li4^b&Q3dVIXc46W5V(Yb0b^5ZyN1e`|5tzb>D7tGp$hT
z$WM>I-YG6Dtq5)4KmV#p+v!!=w5t+79-N)}f^kbxOPG+Yc;c4a*09qHu72MAB<<Yc
zl=O328=1;yGHj7sQ!o6HMWtf%+}j<7v#;kSoN(Oa`e2K=H=o_C@bIf5A!0`+eO9&m
z=%>T^X8pQd|6gy_j(=|SN@u}M&%)K6*O&?#j~k^==-&5&UGL1I&fANX>vb~K>)Ynd
zK00TLgYsFcJ15L`AMCp^(Izu8%HjlnB3EfD>(!9ZlE{F7l9Ut2COa<QXIKAG(fOMx
z@1|d?y_Q|F*Ujg8-upyi;ce*yiz0p~$MC$}ys1z2XV~5Q9vV)691PxW70alNs88Ci
zrED{^%h$`xGc~~4s_XFT|H~cbx_{WPUOJ}f)W@DDZ(3Zm4#>L9+8MBTCil&Pwsn3&
zNqpC)zm4C%eamFo7%}~|%UNVTCkS$iKh*M#YMXW`)^_&fqcIkjmT87Ze@<doYf!kA
z-RdTLzB_i+sflU<6F+vQZDd!Hp2Mwv;@J1&n;p1!@9ljqe&nd|kKV@I*NaxJkf@we
zDSP?R)~#Bpy7Qk{%zrX*;+-GYdBw~mPk7(a<*t)b>V3ZPR87^a4w>7m?5{iKE_jf>
z?4X8HyXKx_OyRd`uVn{My#7x#c3Owd!86t;<}5u_wn=i<)@>(_{GF_zzdycAaen%G
zyZW~gyQLx`m?!TO-6ynf`IiNeiv$*Q`K1O1JdYGPF!#{cn$4WcFRs}>!@-fIIk)-j
z^RnxQ-lfgvVPjtvxz5wq;imnrhnz28p0q4kXQmx~d@=K7_8q%EuX?d|_w4;o99>R}
zJ6t>y{c6|I{-D3FyAJ>VChH()B>R8%rpJPZ?{#lKf7j7Er#MdV$Unac_ZKzRH$D1P
zAKh@m`Mtj5|IJ3J>+Zx>1|IfZ{qS^IcwKq;f_G8>p6MUhmawguuQ9I3wIXhlm#pjJ
zoBy^hI%G6;)r7SHD?$T9(qHa3j4_-zZ&}#OlshNR9hX;6TOB&Hgz<z<>P}}DQ;QEx
zFSX9SztOqo@As)c*GB67=RX_t`+bO+^2%lI_d{b08r%!!Bu+oL?XBZa<{Q5mf4-66
zpZC6Fp`2iRU;KX6s&?bYyOi&w>b3s*^TBe9)%9CtyP7^K2OL^{`}z@!tTHy^>djf*
zy@{=pZZxEPKYBcNe|yqCSN?{&H#6KHt+YF=ZcuZ(_b#8_kIfg)KKP*c%EBVx-Wi#7
z_om+Wc%T2_X5g7g*COW4&-+<s+%zp@<}#@>KaRd^m2|Qfa!xmCN?+uCQOShM*JYi}
zJBi+7hgu6XWf;{D$eww?WT@ph-$&AzDbW4^(+u|Zx|@akbyo7Jvkg109O&ITp??>j
z@ug3qhxg6j>mt9v?(oEh^Wh)!l@rQ;9Q>?qV=#M`>Nlo7A@#KOBbo)qZ{LLMv!B6x
zm;0P)pzec;r*UcZUmHK&{c_J%=(*OmQ<7^wF3^eS%QlO9vM}aDsiOJP`4{^omY>K?
z+j-eV^+JpB=`bILx-IsBWo#j`sn$Q`)_TYs|1op*W7`e0@712abX@02PPpivge|!>
zjPJ7I50vVku=fd@<MSuw2LB6Tw!MF}m22x8ecLv3+~7{SZ<oye{We?0rn9>bh#PKL
z)V=$z_(9Hu2V%y04$?gZOXj`K?=iozZx2tm)rouS+a(UqS6UsZ=Qqh?+1v_d=5=YO
z=G^dmzr8VKY3q;q56vc@P)oXeTV{Wy_3an!b`b_g@10$|#&kxfKKm;pt~SAx`^sC}
zv-&SwKj-UUq3>KTBKuKGFX+RUEdq7zK}Vw&bZVvr$Y;D}zkVQ4q~r#x?1T0V{FCRi
zn?z{^ubZmc8{Kl^>KU~wx`Ld`12z|3WLFnBw}Airz1UgK=`Xg-d@sY3{B!c{1z|P8
zT=o(Ch4VJ7omHaQIql?ABi4XvZkzqv<~j*aU~hd^@yTO@`KDAMp~fJ`lO>;e=CH>*
zec~x*DLdVK==~3dz5D$5C(6B87*_9}U;XKi<>7obi@!E8*$Vw~#=3^AUB#0(FZIk^
zxzZ@@;<KX%`MxxXwk>{r^U;b9C;O^%e>7KE|9Y_cWqiDKseF`Z1GAue<azJ*%d=kO
z_eqpWu8F$(@<)GRMG=z+>l71pOLh0fr_Ux?to3Y5Px)-q^6JHt&66W)^Z)IibF`%D
zbA?o*^z+=Fm`VK04@*w}bh=-5qx_MA)mHY^>5UIPPtEU+O}T7TZuw2l?WAPf@e>87
zZ!CYYX~Jq3=_C2SE@^T;G!$F!S+Lij+3i|$+v2(v=N~V;@XPD<yeQjlvAx3DHnVr?
ze(UC&)MEINv+m&SjP|MDE5q)8_B#IccCN7b2f5Ge`t`jH>~eXvKUDK?x<A$5z9;7Q
ziQUaIt%brn^p7mRo44S+;I^cTL2WVy*Ox2pJ7n|r&htaYZ<3E)Zp}3jX>fg~^HL{P
zR$5l}tj<hz$<32)9W=IxjI%qTmng-V@%u+svZ%A#`L%l5`B{r6#HJtool#=Z|F`GE
z3^}>CUhW?jt>=x*;`-+}hp|X_iGpS5(KFL4-P<ok2lf4tIIni6Z{LokiM5&kk54r}
ze|Jfp<D&N$(s_P8nma|6`Mz52<<R2Ud$Jzh%AaJ$EaVa>{UI}@ypr+%(JMg@uDv=q
z!F^AZ|A|*GW*;rjicoo|)*;^J>wZ{5rKNHa*M#NQjb_b>3UIS;2rDva+5WJ>{O*&~
z-=}vkk!>%mJ@WP6)^mJu-E$vH-Z*>5`o{OKefxzzU1YI*lX?4ZzLBZJ(nOI}Cyoec
zt+4TnPdc&M{JZYF96f_WmdA?SEcUt;e0n4<{_s<`lOJDr*L>Zb#!Gt~YkPlmIemY;
ze5LmB^LJZ}zhu1X<8fQYwkgI+e#!Jr1-n}lmaPiuT-ozIYLD50t)(_+mw#z#4DPqQ
z!;@D0)Lm%(k&xxrmR_rfn%@58$FHEz=I3V%S*ZWLz?OEl@ThC(<kCJZR*ek1_(hRh
zdOH|b?wpzBo2SOSWmj8rPyfe7*K4K9_Wxxmdwlk=$Jbdwc?m0X4?Di7GWerlDOz@D
z-=k#4($31B{JquFyr+HLb?4~M<tvY^y0&{m^e>Bg-o<|BSQ|EZ{JEo4FiS{9&48;i
zXj4pL?Bj=Ak1|poZ>*SG9vHB-Wi9{8iM68nzHygNzdv$rv03Hrv$GzRzWS}97QgV4
z>fUKzlDn6@%qq>)Tzrwi*K_68)Awpu6-f#G-v6Vk*80|;+Y47rOJBV8)a$-{w|Ot4
zOcRW*?wqy8=e2@C_DtK!vy${up6yJ37ruSr)}8XJly>YkfAw>h#s7&Dy3J~J+=>J@
z1wNa8>Sn-1{i371O1XjGTL0L*?m07CVSZblQcCNw$BzZxcHdA85xTLl-u}k)!!n9~
z8U87<=_R79<=^MEo5cJsG>cz!T9bR%1Gdl-n>)G3L@)evyuiF^a=Ylcnq<lU+(LSv
z?rbrgv{L2AiuZR*mTdBUwmALKodbOTPg&Fk#+IFqT;3cU81JHN$mN)Nq3U11P9+mp
z{XfpSGkw9`+ndfB#+_N!@HX?|mro6wTaC<qG3<zSy1S<S+ky=Hh{O~xn}f@A(=6k@
zH%z|c?v!`+^(`5Ff3F~e7t8Cd9%OPV^y+WEd2%P~t~oNg%=J}I8<u~b$M%fVdcK#o
z@&mTT3r;d@es@^M`97msb>3u)Q#S6+i!@Xo=CZ1bikKY`Eb$Z#c(&nbkpl1XA0^)z
zKQ>F4f4=f8e8c>E{|^~$EmUS}Y?_<q6352nRnnujU24ju0+)sQvqBQ`B7>Oc2yzDZ
z6jUyGn?K!|cY@J`^$)&p*#0BrMb#m@1BZ(ZD<aM~DF3<AVP~qmr*omw)h!IW&p*8;
z)MQfEpXK`Eu)Of>b=P9{-Z0hQeQNXCgQ6RbfBwwIDD#)&!}@(J@4GD@S)b0Yi#m5_
zrCse)vC{NICk36KubK6Db^j{UX)m++^O+0JC?;3e3E!=9GSUm&sdl83_r|Nm9HEh`
zgAPquS<<ByI@QI~*wJuvc**k0i_bF-9CA>9zVm4Mo}0%^7T1Yi6^Rj(cqabd<lAx8
z0ES#SyFXWre(w7CgDvLA@8ZR8AKst;hfi+OdJFd#2QDw)(h#zIiN)(FS5zIG79U+S
zUBAFW%qYFze({2gJGm*%1-}g+UzcUO_2}51YpY&py%$jp(465ddVIG3W&ZCz^N&rM
zo|ZFjXB}(vHRbpiE!I;8UCJ&?nEOAdy1m=;(cyYV+O5Ch$F^(iZ)f^i)BWk9F%z$p
zUGfvND~B1b8^z4upT0|8AVXn`+n<FKDn2NSh#KEAny`0v)AbIv4d<EOot^x@*=@)3
zxff3daUa-z=C7j43ELN&?O%U%N_uq7l`r@Sd!JE?{tsKnZDnikv2=aZk@7pkl>G3-
z<m&8m*Nq-?6zQ#(NLt16ti155{Um3Oe@$A4Om1punC6PkV$#*p`g=D-(L`wGR-08?
z{G7{!M2wd9IG(mrzF_6oe449)FJSBPhQ^1HTdq6B1-i^CkJn<AW7X+-UDN8Gx=j1i
z%Ou|_M=S2bJKl58TAjT=<oolFj(#lWZTscL^`BhYdGOiqR`ykgUUoiXo)x<DVPMFn
zAie2E&H-YZn!fJPxnHw=l~}-``@EbY6&q$p2_G>2#%(pT-j17Hx#M_x#-&d$Yhugz
z{Eh9K%r|Ryec$rAyn5>S@b&7)0+J_R-1a0?V&}`LYrKy#)Q2+MeR@JaBz0fl6rYJn
zVTBdTwP&4BeV{ehNO|rmYu;5lT6#vSGm`K5`q_r^?fDe(m+45^t;L>mM8Y#eXWsh$
z&i30@y|c#st+%awD>jAAp5iZ_Vf49DaCza&f~xdm^*c{&pH_OO{Kv{kwwAgk@xB$-
zoIR~>j-4mtGNP`hYI-j4a_Tpme|O0xclIEo1wP&h#}>Uw+_UgWrIxQ%=v}5$6_$3P
z4;UL`%d+%>)Shg3x#>>BnH8QIZy!(Gz~ldzd;Pn!arahCGv<G;`KKZ?_o947%`UT6
z^A&2%<@bwke9_K2w$$ph>e{2n?mpbE+TQ!KcTeyA4G+C8M=d<LeZFDW4T*_)JqZ)V
z|8UhYZGZi`{Mvo<{>}*jjk|MMqs&iF@X+U6K55FS5Y1+n3^5^R_n4Kwr!NIo{(5Aa
zwOMtq?3rnAj(dlB&C&4Xef0Z(QHo1&pp=}X{DxYMyQXI|>=GH>j+B-9lyXj8do}f$
zQGvjhmdAly>n3JD<<bfidc0;+PtUr&l9G{fxr~c*<(8|7#@NkIc#?HI!#CHWTO%WW
zd-nB(r&j$tw(Qy==4XKujZ_-^I~#bVN}2R+GQ-!M@#<KbG+F3U%9*an&GR0gv~-y4
zC@PjVeWj$){Oh8@4+L*tnUzrYTXbvIRmN|cLiy3GQ*K}T7QeZ4t5K79TySbg_(l8K
zl7g1ui+s=OFZr^S_xZl0eLs25PkMIz)QTrhb|i>8YwdFLx~i*{D*OBSmxsG|I|U!P
z-NU;~FtIEtJ1}HP$TWrN8#AP|+`S(-+DsMKo2zC1;#(xs6~-OqGq<+<F!GiPI^yL2
z#kkK*=soM4hevW;{kXYhIbU$>KXBf3{>7`aeLG^lsNQF^eV(#K_ix{w@7H6@R&^yg
ztyGz6q&rcvE5wH7>f=u`Oa0tKO`9*(pM2W9^uX-(GKZ~M7q0&Jb<%>?(;1)KqAWBz
zlX`dQ)Cz`*hKk5(dUD*j_^kAY)9aPNQ+ei`KYLI=L^e?T<D)ffZ0q#T-`e@kDUTy2
zwa)6MvgIMp$jd9*Iiz;RRg|r>eDpck!K8Uz=aLMe``z_tHFDeTW@rd4{B@51L&`qp
zC4nDj?kU;eobs-%>hkd_M{o8suLO$DeVFy5>{jmk*Jt!^FBjc;VYQ5t+wM(u+?UhN
zE_n9p*w%B2%iix4tL(7*=lk<~_03<WUz+C!27f+Md{F77QCiBKTMe>XerFxlzqsa$
zPDM%EE{<E3IxN-G>gMJykiNCJed;F5*9Yc2wc*-%*Yu*#b?yb*WIyQ3xkt>l6EUxR
zAG7_|N#}XTOxV|X^DlE(dfm&BJn_YTJ1ya?g&+K9)i!^Bci-Mkxn~9Y7PWWsHXGjW
z?Y|h1wALa_aHfFE$tw%5p6pR#R8bG=aIh0lm%E?2yJUqLmq9*T#e|wG3&i*2w8uTW
zzt3!0ZGQVCr$e!Z8m!@(Qx)cY=RV@<|Nfm_{Pig!amRLOr8;h$k{{}n>FOGIcjnjJ
zZ^GAh9d$MCeW;f@`(@yY-)bs9@=d?~n!WYok$-1c!o@=OSaZ7m_SteHcBYJ&r{||6
z?@JQ%`hR_jvfX{XXxS!1-4(yy?%|#EefAj@PQHhK9p`PZ>8f8Wy2<xga;TT+f_s6z
zZ~j<6bz3}dkJFj@<rk`3>$E4`(b(AW<f@-to%@1%p*W+MbwSxravJArPq@@PA>r=w
z|K5)u?z<DS@ury4rtkgt-|Rm(;oZ%v3On0;ryW;o4Y(Ltc+}|Fv(PzqOm5qEZ9VkD
zBf99#>dF9hzf%)lO}cbkCcfDE)6T84HS&-BJDI*M<;11SMo+Ijxl?~4v?6Ja(Ja-s
zi@2v&@>HHt2|udh>UzZOYx&JL`*ku`e$5NK`T6p8dD-)x=dV6ZiMTz(>;1tn#*jl{
zis9_ZN4HOln`K%2^3qIWecipKuP-f)p0-{l^(AY=<k>e<rbXP>oG!3u>M7+N9`m+e
zdD?NcLES3w{;#_w`*uCg`rCUcf30q~(Gx4l#Udx4dhP7uow23#+NRWNUQ4|eE)mIU
zbvqgpQ&kXlHzj*#)ba}#r2^ahrXM^c`S+i^f3>#icD7}mre3<|=S+{iQ)T>M-Bv%<
z>nAdop0FuTVhmWhV$#d6o4=ncpY`5cyK2|gr(Kngj%2u4%?(?g_IjUv#pg$=*;}I)
zUk%D|+yDRg<%+74zHfd#J5#h*U{_K3(e%ULEcZX0n96D?DcKXXIFNZk=k(yZm7k}r
zPY@EmUv#oBKY{<)u|I#$-IKYS{X$)$&DW?tIXFlAV)5s$#Du)IxFcUSM9E)SZT;fv
z<r|NS{P#vB+ZhBezBupo)zcSNoxPBfvGvv$p4j#L|AZ&CvUX!zf`N357w$D4CMa7k
zL2Ep2B*g#!`G4Pk<NtsEfA3{vl97;n(7@awDJ`*qQH{aMm752PWyVOGOG#btVP<H7
za#nk9Bxg*>)qne$&&Z0nJhbXEE_=6R+0&-9%y*64oLaLU32AX{%4a^j!C2?;qJo>>
zYwtTAy5m)-8PvW->iVatdv$zPPY+xFZ+~?Azxn@Xx~A8Bz4xD&`LONxJH>U+x7XC)
zzn{-CWvZyE0EbNM{Tq8ymmi<gJKOA}@)Lc%8QZVg8>pReU$}ewy!of>58qUh3$)RG
zuvBc%nMw86w9bb$OurqKmAlrhJn3!zr~98<&Ie3U$hG;IZFp4fSI+W@|5GOCroUi&
zlRJO0i`bNgP0Q5+J7)R)_4%u};9se=xc$!c`)ALuZ!wCPx#2?Q!KUfgPbdqE9Bg#-
zx_O9|J8stM-@8M$-ST|)u2>;NPFz+>=X2+W$vdr2npvK-c~Y2S67g-1jnlK&@=t`t
z6ehEuu=!hlGHKCY(TkfLT5Z&fezB_0Ob;m7qT{!YW9mfz*7_Oy{kQC&ucLP8e7s^X
z+t1T+xA&b5ex}tQaLSBL{zHGqx9MkRt50_Q{rhy&lv5{CKR?$_W}5q?>%dRO6u&Qj
z3=S`O@pISnXFK!0eVDWTV(wY1{@mAdrhN2@Y=}|({5kv=OS`MJt=)$97{PbKZWE_Z
zndNnIMW&1L`C$Ek%_}Xh#GLA!<*&OSZi!fselY0NFm2GeVOx8m{MIfB3tS&`ZVLC-
zSyA3QgVqPW)-*pg_m)HM60<<LD`{EWU#+%Iz2#OJbbF=j6}?xyS>{{&Z~6ZW*f&M~
zsXuFgV~ByMQ)^ISh=r)5Ymk>#Ti4P@D_IzoXXJSLxWwg6nSCxIEJWz*ex`FyDS}BC
z6AQ0jI=t!PC*jqVVRH}Z9_oL&v&5n{$i#b!`me)QH-5Dn#`ny%+jg4m=8hXrZrs`a
zb=%*>J6(mEkK5i%I%aRZucqlO%htb1zm0ys<};SL!dGu_&PVO?sT%Xb-$(3^Fit+^
zqcy7}|I@#du9No{$2_md?fN|>?xe9$y65harCah}JzJSCBp>tLM_|W|!ddlMKhhtw
z$Zq&qeIw^&dTICSO{asq102J6rt2Ji{KM^`;8rbtAE}8)ey)zR$=l+&BB;nRGAAl5
ztg1_8&)n6{_7}gdjVbq+5kIl*PcTRK%V{qU9k9^T*)w%@di|o;w|3@<t#64<dgD>g
zA<y}ClgT`(9m@(W7PqSC3ZK9Jtz)-n%qIV;Ku5MFgNs|=)*m<iW`BGwU%pq?+Lt=g
zC!Cj7R`&kXSih@FdvaZoUXE_}b=gCY@6S+9ycA<!@INzQ$CBqFNplX*@_#NjvCOR4
zHs$#Iqh5N~^cf99I~(uCw3SSL|8P6od%gq3t!rdeYBEl~Zxkqu&5HDs3%auRns_Mh
z71!gZCas#p=BaqSl;I&~cFWJVmGWhtr*`a{AfV?WJMsB6`BsCkjT?NX9!~Xo93(X9
zvG%4_Mv($J2d{`FW~F(B&3U3AUMO;~<@KRCD`K{Dbk5>_`TFs%Q@e`n&t$sKdeCre
zwcRyO*`F~hw`^RsuT3?rA%eFeTV~%yr%h$sr&Z4GZ+)8d_MP@=sS8h^O@7{{YHsr`
z!Mi0nGgL%e?%J`dn`9TS^04xBymIGU=;jID#$Ez(QajJ?<2Zg~-@?4j7kNIgRnPkQ
z<@7UO#T?-}?>A4kKTU0s>|k{<4vQ~a|1GUk_rWr@7O8{wTLRW>eKJ|mhv%{QzYlNn
z-nVQ>V0pJ_-kMJdTW)Q+dVBHF5*hK^^)s?_ob8VLt*BnA@Z)&-feq^>KU-5AzuoA?
z>&q|NUOa03EPrRuhF_^E+g(22F6m_uN%^`%_R+iyJ<f+a_xrZ*Eb*28C|z_m$=pgM
zSoYMvjgBH=bLu1-XaAG?{UdVfHKBW-1^;h5sK7gO-Wr9f4ArWJo%KG)i_fmR`(odu
zO>x)sd=F30VsYPEFfm7lU;9Yy#h6)9CtaT`5I36c`rAuf=a*>OVea_9Rz)xO?mfNg
z=8>j9OY$=9uU|RAq`m6ttwX|_E*ozSJm;vV?wWX4a?|$0=_1`5MW-c{?_oTl?!)wY
zxAONdS${&5cPwmuA?R(#!1CxEhn}{kt-7A@(v+j`3n%Ku6@Q(5d9DIe=H=))vv^Ns
z2W<6<p8aZ9p3n0=8$YMnd{xc){kHf4!<&cluR8o2Y<D#7W%$;xSM|=bwMUpAzFqjQ
zYr}?ycPs8G=i1tG#Bcd^S>mH=+)}|i7oBV^OPTg)w@nv6m}A45c>lQCn#9@ep@z$}
zKP*lBbX}p@^+;32nb#FIg<CE6ysDOJajxLnmc}did*eBOnFX?QVz#b(U~;j~^WkZa
z7jA`GPtJ;8t}hbVGvOWA$F+C$k2YBr8y+@dtvIyz=6*ie_b*<#^tdZ5x9ZBfur+sA
zMrl*{5ug8N?^Rb{%iCx6>F-tTaDD!~&n{)+dd;nDEYnM)YC~-_>O1DYc6gxv%DTcW
zboXg1TV-og*F|f(-)3a3mHBbyvz1(QAKRB>+>b>(PsF$A2AS=7rPQf+Ok@p*vqj{S
z*f+Q5)Lh%OW5b>;u9qZZ#h3Hh>?l9k8k#V-jmgDx&Z6bBvsb&iyChsM`1UT8)2~tY
z(b}ih_iVH+Yk$2cuxtulJo)JEJ>PRYoCOqr89Z(IuGgv8k!-TT|FHv8=B97CZ{>79
zThFidu~Eu;Fsr$F#|?`oUDAqdYTY$ERfKHM^1f*N7&l4((Bwq<vog(dWJ_$sizA-5
zs&stOl9hE<f4wL!ef7J*&>bg>Y`l%d*z(U`DwuA!lSSB+&qw8bddY#79_NoUGs4f5
zZH;;Lrsi3m5O?~^Wo&s2FL>=Ac-}}mr(-iOm2Jb_eq%?TzUA+>F5GJ1+avu#`l|T-
z-CDNhYi0zxT_}2f)jFKDM(Mr6JAQ?0$M`ct8N@}d91q>}!8-Ql@voaJ)`iw{Oz-;e
zz9frdr}+ZsrR^nO8Fr>0w|yl2h=Zjjk2Cy7;>XJSPFrt%f4~>g6wLL(A~7V+I$X7O
z^MWrKcF`PqhxQb=zMnh2CoHP;n91p<hK5Cz#Xjlj#l_3d?N@a9F7{_h>q?ii!uy{D
z|IqoE9C~y1l*{I;xD1Pn)P--oX%U#=^?#}8vPsdbX=)eVny+E3ZgP~~`v1qNN#&34
zi^fNtQ`L>%`aN|^@-&bAj?y<DxZBuVPEpTTn(cA#b?2i=H)k5~l%++amv`;^D|o$^
zZ`#kV=G(4Ja$cIbz@53<q-p)u3Eq#)c(!%b$9!d$lis3ld`$dc*a?xD^YYCSRPzK>
z3r^kpb!gG2pewfv*Zgbxa_;u(yVKX-{nJ(B{Ia@e+lm!imIPi}x~A8cFZ{Re&)-%X
zn0~i^KM+)}<#pKJ*{7&DqvFJ*^9^^|65=@bU#OkA^q_~Id;0TZOE*}^{nrawUSL@`
zX~L#6axAVhZn>ztO8fJ^_GQ1#awV5J=kZnBq%{j18SObs+p9g9)lHj}qcYUnPq+BU
zrl~0BBs;x*Y_iSk>#fbsyWWH@NDw>o`_1#)f8{j{LyxOH6JNf~p7r#`o0$hsuCRXL
z(AzI^``m$`H*Mzsg#8(3fA25fZnDWRX3oJ)8=S7(a=LJoPd$cd`uF?0S=PsK?SFEi
zR55OMDEI5Ff#Q5w@%s<>EclUhFYu1g{mr&#waqK^la2+6dOGq}Fa3PYO8;KGc#!aU
zBPK0fPg!9-&Q)PkmMW<gH+7qTkD8ToQD4qLQ0<@5AJwFo6lTu!veE^UMCLyeU;6q*
zkBFXqv+ub_tAZobqN1Wo^NiIs&#c$eakpf3pVEBkxB0sM4QHI5=aq&hsb{Uq(Oq+D
zbDvJ$NuBA}wuY2#tgziCl7Dx5{`0c=!OsPYm%jKrS^fOp>YDR>EF#Aq)t>vEezwT2
zaMQju+qVl|-8Qw*e2KEc6S0kDH3Bgr;i(}@rYqEl|KW_coU(kH+R^1&k56hm)OPPY
zcVfMPuCc0e+@c(hwYu6Tv{$<ou=8=8?pp07t>RzN`9NEjMey{~GU+)k2Up)Z^XKN7
zp4=PK?2X?qm(REGzP-XL@tKv`vPiB&o=pX74(asXmXKsf_pz~RdgR%nq3op5A@Eya
z=D))0ouaSzHa~tWSn-F|Kk%)wvWNb~tGgOh`ZxVfkp27O*~HDyS3jLC*!S6(rLaBk
zn$2IGri!&J?@O%S@Lx*2Wol*Gkr1_?@BO-qQ~y@^TuzVbP&dj|HF{|~>rtq~OeJ}F
zLuZz(kXFsjua$pQu3=2fvny(N;J$imPuu<6m>j7)hD$<h&c4x9lsoqLOv2kOiK1zY
z^NyK!|L>V@5$DX(dnLD^ibY0#?PVjyJy-1v-)Aie;(n<vz&6?K`NhsT^&tYE`gtBt
zxzusx&z}^t^$n->Czec7S>5B=<IwZ#s{D@R4|irKR7)BPfBy0QXg%j;7PH)rx@iaB
z<vQO^_<m$%&BTu>i6(W7Wd|w*IDem=@-*dBFQb(HvEtSb-HsmRljC{%R_!^n=LpZ$
zy>`p(cXOIm>|6Nbf_%msVcFNYsu!H5^*w&IsYv;{i-En~w3Mehai%J(-c{%vPcpW2
z3uRoUeearw?pYuCj2=##mxjz$YXbLP_f>KepB;POu*y{aOYaWFr6#>C&jSyAKCyno
zA{X%z-H%f}Upep0nw0YX0kc!-G(DS>M?U$mlyBT05F0xoT5WCaikyqp=Yn{zWXar<
zzh(dBrht#CybpK0hfCK}C&BeXU(E!+B#XUVrM;#1o9&9Opv|v0mS-)nQqoTH(wkfN
zN_E8<&24$U;RRYR&GsfmF4JDq`Py}{R$zvX_%S2bU5AwyKRM&xwsr9*7tw0{<nxE7
z1f@=`4_sF2R+yT$bjzw;TXs#+)%YjWrF}m7d9T~WU8Y}dC0}|!!>MY9fLHrx=Y+-8
z{&P<8d<v~v$t;*6e3@ad!5#CX{(XGvr(ZiSS@>|(jEreN_gJg<c~5AQI<&B}dh*f3
ziw;C2zp2YDP~F7Fy}onC!HLPs`?iLt?^0ePbkv^jL$tr|FaBUT>7NG$j!a^id9prh
z!{^P<3e?Uo{pj{@!UlcQoeDv}W=ZnBn(*%A)%9r($4`IGxM5M{ArTTf`?lWObvhP@
zJ{`?nwA^&@ajPJ<f2KDKZ^f;9cHnYI^$+_5*~?%5FOc=T=-r;!5aGhTcKh|X<hqB>
z54$8bZCF~Sw{6Z&mj1(Eu5%xJbZ6(T<owBpENY*AE?)BLz@A$X+cvCNvBU4utP2uL
zJG^=L*T#EaTBzKaTm8XF+g`Lf@BXfon|=Lxb)~s4es%h##^@%kss4NF(RHskvyvW8
zU2X7LOQ%m*;LaDLgtgMK{Kl=zzbdSjPWbyWi%Epp_2Sc$mo_vDuH2bpw6MqcUs1_a
z&gJ%tgg^K!oSykW*NU_M`6f$Gr_!Wgf5-4tyH%{#56{=z{^?|V`q8LB__f%qh2p0p
zLijQlHXo@Ih!Hsa@rK_X&it<4BkCVMF7vp-B0bysfklz7aQ>rf%W~asUs{wT?iC+o
zGS@RT_<T|PUUkW5cIztGW7oV|xkq{L)%D-jBx-*zI438REc$bw*sYQii&h_bJ6CQ}
z)#Xz!D|7m1zJJ!l8d3Az)<WUB(8kMZ>EhFUx69QOKRvbe_4K>qFS6~ed2a}YP4}Jm
z-9YIw)4slI8<+i9bLC3Tll>LxQ8j&~L3ufSF^g85UzceSR`WzfDt4Pg%BeH0%N*{m
zXmtK{XSVaSkWc3xvFm-`z>)ZL|7uT%oTe`YH#okTMBLx;y<y5U>u<ZyIi5eXF#hFx
z{~hlXmiR1e{JHDRy7C{Bqb6iZ9<R@JFyP-^wCeBtbFb(AzjnR$W&QmL2{JyqoCS)@
z3vcVKJE`NdLgn9|?|Thc*QoUOF>p`H56E@(Ozd4^vfEIL>HYD(9Xz-9d`|i=o$>W&
z^zEpnZ(VzrwkYmdxwR#p!<k|B-tSH47D!vqX<xLoYt4?HCtYV2IqT>gY?9h@BGyQo
zVROVT2Gz+?rje5ZWv*Xi{q5N?MKN~b{~ZV9{}tD+S637$zdFgk#>r*b!IE8*Uh^yY
z`L0{`I8#a3aAw=`%PV67>_6W2V!50+>x4r{Pu0Ga`E6S!G|L~G$9Db0qx(yFxB4%#
zf7SIR_rBKAu=5jxWc`1XnMKt6Ip^Rl-0i&AWm(JaJ10Jff0}$Gdhay7(!%{!@Bf9q
zP2YYm?rGYpb21#sC6|9?NPE^Et2u6NBk<j=IP;~Ko{f0w6Bd<IJIl}2ADFr-Iv{KM
zCeP5svwuE4bZCAy<H@_#<sTwiuKz7xV)(@D?uD}_t0SL%e-@#4%>B{1R`ZrcA9kHv
za`+N|l)a79la~2M<L<53cgR+FVs*CXj{VNm?VRa}^Tc$v-kE-e$*yy|p!qVn_)Xc)
zhur*2B97E#op>guCi7p0U*Gt#+|{%@M}w|s&i?rJ==(&Ou5+uqvnQLGUn`TC9l-Z|
z-HvSk%*Ly~;u8;fly3K_vHVbxCHl!$Dst1Ny2!nY{g>AN*F1Ui;N|NHIcN85ND>JS
zn0Pfqr%L9ltA9yZQU%+q?|0)~>K>hTPFrk!{qeWm8DcIIot7;#e%-xex~lc}{7A*`
zT5l{GYOMI=&ad6R?)2HU{8L|5@#~abDK*<QyL)x@>3|=BzL!!pBjW#8Jt=(sY8Trp
z(`kR~Hi-Wh?w;+`eIa}5p^ih>weC;a^8H)Qch}O25Js;vr~Tg;dMq{NC=Hl8vw7CH
zw~rq!t=L(V`QT5y(EHUl4)&Z6D)V4sauR!UGpsPW@zrMjFc--uGuDbPUSpQOK!}xZ
z-uahP4jsCu@3*<^>Bc{-y$_artF=8Rv03+D)4H=>4~&)^jPbE~t!#AVDeqj4R~c`b
z%-J4$<R}IfD*i1k6^&iN!ImJm=32~_yfx>~ub=ipX!cQq2wvfNPYV9sy0!n{!ToQq
z7%lfGyRFCeYpGz)#^Urv>&=_eSMT5ZdFz*g+^90a*>mU4ovP5eu0K0Jd(CS(o|<yA
zX%prbF4^YEzWOS^U@<?7{U6yJ&V4#Xw|5=8n;#(DekuBM^lH_^)6dm>`uX##-ibc`
zAGy<Z*zMHQQ!}<y?Vj_5KRhulNN>8L=L*l99KDq*wcI`&&0lwD(k+{HAxTT5<ul(;
zp5N*HBWLrnwkdokPjDoLchB3RdhYO)z^i_nH<TE@&YKYPD&yXv6PFU#I6QyF-I=$&
zxp~LDY%Z>oy}mB9i?%P*Ub=cmajwpa#Lmdci6>=$dEU=@d3u@I@#{YXY@a_qAZ>1M
zZ@4Hw=1Dx8e(&Lv6VED#DjS4VJDhg;;1!{}Z`<=l(dJTBn?L_8S2Yi>Z*jd*b~o?(
z1hdfOVDH%U(y*TknXX(e?$w@LSEF{p+Qw4DsExNb=n2cSh<w`_XIwo>*qz^B@cDUo
z^XvHUH?|cXYz!4$zUaoHMUx)6d1luvc9NVdv+INWdf|13KTrSW|MK#^K<8G$fDGj=
zXB6$;%-C`F#;wlj%4?R*<P<+&SO5A*VD__v?9T6F{`oL%GtN)6wkbP$=_upe%u=g^
zTANp2QGGFYR<P}R&-XREzAr6}Tfe(<lLc2b$5qM5n_8Xge)i2!+Tm|}&0p%QitD{g
zPEvvSvh}xb+!dIe>hVWcxA^wX;`ZM=|2uxj-QL0A?!R5mV&1H;6OKLJKkNUCi{alB
zZJOHeD%Iv+xlk)>mcQ38b&~Je>&HtU1=kqNNMnnSnI^TX>Gf;Z3!k39l9+z6c*{le
zK3VPeG7e|vUD%_xE-vfMv0D$H?)oHd{jdGgcD-rxs|B|lk}iM1>-|r_{;<TxWFL;3
zp2>{+T3HLcZ)9E*lt^y+oG0b-qtRf;q>$uGZDQ-zH!(=nXumOAFCt@gfV;ZsdgZ5$
zf_rB8|MW>#tZX<YoG5tV!=4Wsxmpi;luU|c*PK2RUN5rETsT*|r+iV*lkXb5$sa#g
zFvY3ZI2h;pA4}|U42gYNEPky2!AGacGmqaGyX`+6%$o2odCjbU71zVcr2K;C#<F=|
zn!Rj`-M#8vp8CEn2{nNo<{B>lZ@o8A+s12hrm6do+F~gW!S{@OzfJz+^*UScX}o@)
zF}$LA;veg+>kl5(WQ^pTnYiT0>AV2*R5J_5Wy!bWZ!c(9JM+*yG102LV{$=Pc#ufP
z$(|29XFShseb#kql}b->w3Dr*jo{1~0#A#79C)yJ`@xNiCba(i^KQAWs>GE;hKq!M
z8h&IAzI|KXy!!zc57%s|&s?um+LL!ww8%>s<gVwud`mIDW&H!k#GIIS3tc=5jITHR
z_*hbGcs)YdQ0TpJR7Lk1`Gv~&<jx;V$!T1h_HuvU#~+;jYHx1567)DAyEClg<L7r3
zZ%$2pXCOAOedeup^Zb}v#`8u6>&!OTFRu6RX}MW{a8JV7CI7SD?9UW_x$oQj1Kzpm
zcFuuUjP73lQ7m&w>xZf6hE0n$G`S_OwRyTN*&<r{Y{c5L6K_4Txx0GB#tk>FZQHN8
ze}l?;#r1#g?kl=oSo8VFf5HFqD{kjL!8QkhbOt5v2@fMn<g+HT6A~ue`0)Gh_v`=v
z$JTsb|DJ)FO@M_#sg+v~n+-_YW=WpqFg8G)<%r^p5x#zHzJSp}uULuWNsG!K-P;wk
zIy72bq?}Q7>aHuS9ydG{I1}TUSNJ$=W7w46q_JqhinU=EqO`9?WM9o#yYJerFSfOz
zx38BUi~DuYT+!pjw>o+62lsZqj{Nt$-cL#4&0Ee!hYz!^yOsN!zwdk}C#SVE>spGw
zt3%3){#&0iB+chYAGA{96O>VYu%v8`&xzhG<`!HZ4BQ{Gy8mAH{o_{g!u=^-ZUQX|
zMGF73o?A$;S@b>nKRLHRCi&%tavARq)+Wa<9f9X}{{NFVowaD{`lOvP3$L(cCNJ|X
zES&jn8CQQGJ1a|8>gAZ2d$Uq1=jujVxNbXiT;D}>DJSQopBhg$i3X?01*gngrLoW4
z<J>&Cj7Mzgm8s4*9pKw2OC>N3YY`iL5I^CqA5IzU;&`R<_KPK&{e_p8u=ETC{3R
zxOVCUX@;*zz2(rF^YYi$*D2m1UTe?Tp739IMI&Xh`soy@RwuC)SKIa*m##81O6c;E
zHMlxoe(_AHEZHS9vu2uCp42*DI{EJO-MZgmc2{(nMQMKyUzu9edQDv;#CYlRee<+R
zXZULztoQ7Ectj(~#IQ}mdyn59M;`?>p~)R5kBB5`J#^aR>7#s4X_>Hb$M(*dM|6_R
zKJtCkGn_ud{fv4{C+iWXB*Bd>5nejVVFKEm<%gO-sDue;bLc1OZtRQjuMmDODBYoZ
z#4O4Bq4N)+A}7Zm!i_Ehf<7$*3tc1@s)V>$EL2tEloeD~VVP#Et32)I$Nx8lFCM99
z`Pfulb|P<O?5?|!4^0%c{~Wn@u5R*Q@kKU88j_xN;^&3bJO96a@czSH?RV?%yu5zv
z{yzKHFCP3@mOSP0+VJrGzqT_PvmN1)vHNgG;k@H~-g&ka=O3(}qh_5N{CCCHke?G&
zotfHX7H<7Acj3zQ7uUD$|8}6VrMTzw-b8DaGNINdvsxzSO`a#Yuja>b<wI_B8kRA}
zFlSd~JUZsiDP^&A!s(nvvz2ZrwEQl9y6D;!53?94sne%|CEwf#)#K_)Xl|CiX?y1U
zlX-62R%*@(HJZxrYS6o;L{Rf^{pSe>xes<96N$V~BxQ2*xGYb~G(B^r@TUjf&rJMe
za&WoY^2uozX0FUwH|u`bQ8kUA&E?A#0y^HW+tKH~$G&Z@c%0zPh`{&~4eqDXJVzr>
z1VkJY^L#J!IsW*G57k*V6BbSqkBR*G=9_yj>(P!P-nCh`D}$o;U*EJ}`o=}8FztS~
zkV2mYHpYRwZf*Zrz<$HDsk!mxKFbw84|0pD64=&O&Yfv>@aRFAU&n57=PiBVbd`O%
zzE|Fk`rfw!eM-WIs`A2bFTHpp>cFHaVki9N?<A=?Xz$fc_?lxPv?o!)^X{JDhYM3y
zeMqdzzVPZL2S1Z2kE-VH^y+1<26v4b{De7tn(AjLp4NGCNnuJ``0~FEZ|{Ha4P{k`
z(6O5lB+w%^p-}APwarV<TwLd4)E2sZ%Z9B_HFD~+-xO+c%>O5=k@hcO`ciFi{SW(p
z3+__-_vFt??)W+JR?HuFSG<Uu;?8-lSkFl>MSAXLZTrOn53TjZy10zLq-JIMgzQmM
z_MSia%z?S_;$4%9KPA6j@^q@|R-gAV9lJDJiVuG2zWK;$Tc~2XsOO98Hy@>h-Me&Z
zn@FVD%+xBwGb#+C_1!JKV!p?eZ9X0~nBg(a<U`t_?UHZ&*p|Owo*6&AtGQ>TFUOtT
z^9{G%^h{H%dFH-9B5SXiy<*(a`@0);6B8b-Km6f{U2gfNlQZt;%$$?-sBFfZzF7~S
zoc)>hs#=fn*`>2uW~z!s?PcPJQf}?oV)61yd1J#wgLm)z?;Bp(>~FPghnR8T&6|b3
zL2;8rd3pq<Z{R(+OZHcTMoGf&Z%Xl+WyhRsnDh^NALQ$2;GDp}?`L)YHNBZE@v{_v
zbWY%AYg*T*dqGuY!>y1ly2tb_9jvlHKL73~|3h*`ZR&2L;u&vzoYPjh%nUtxdh5nT
zyZ$}PDC3UXB(-?^#>bBf4OrY{D`KwY_9(PUDu^CXo51xj`iFk4$`Zv3{-Q!J${nuO
zEP3c)d}#Tb{}+y*Kk~QaKt%d3j)(VuzvQkvZk8Y3&eC#)Wu8HrG>@f5;KIPu!m>f-
zZp+=hlDv~o^Z%<6JF?<7yG5^O!EGMbJ)b5$;Ni_b+t68Iaj3}1Y|-&|TDwdxhH2k9
zJ<U)-J@H0u+`P+`n-3@Hzd5n{M2O$#4wm1WD>T$*PM51YmvrBcUH6;s(c-0hCbw@c
zJTd*W%KwC$6-<>G-Tr<%M7ZPU_h`0x9cGr>$5<wtyOsBM#qqF>7mr^k-L&Q2qnno#
z#e7XGb7wx@b>{MyyS0;F-uii2`dZ}Kzu!_P*r&+X)mQ#}x_;4|bM+q=zdWwr#$mPM
zy44PW2#@-0f92$&+7l-Km{K=?oy8GFTjkL86I;SxzFIBS6>WV;{$b(L7XjNHSvH?y
zak=~IgLuDZ%o{WJO{MQcB;@-xMD4M(-=enh>_P?o^*cf~neU2P8K|yt&8)BeV&RGS
zIR~_6&$`hjR1)#$@VwX8-+OP6yLMr9_cqh{6HMM(7uNDO?Tlr9sk<-QR!Gov6Z4Zt
z{gds!RQq3A%P%)oZB|0_$<E`f>Yo!oR~b(|@Kk2;`)=(UQLRl^T^2vB;BRT1Y|+rW
zDZ!Do>dI#wErvw(8xBFU9ric&q?ukjetoUp)vNc_KK#=2ttjta7^EZHA${|8@Sfh_
z=?5;Vy!gS@m4EY$jhU*k5o>PmttmoEu{%CI2`}WCqoSR4Q>-B7m+i8T2RI*pWVrMH
zrGx~Za$kM;@iVO(P4txZYgqoD==p0w+Q(xtt4}-a;=DF<PTy?XvjJ!J2Ta}(QLs2)
z&-!^-r<Jy~{|uG*^Lx&&|N827aoT=Pr~i*+J9h0d$oe_!=cJSTu?ZSHwLd<*Kf%g(
zS^8RTscXpO*Smz5+gtrU*ejjBJtxKMgW|2bhqJ{N?MbVuyOg=B<lDEm-mj-eerwxm
zo_|MVOTp}x7$fr;d0&%-1^zYqd`N!UC3fw^0sSQ#TXPdXxcnAlo!#_7Kc%ZF^hAHe
zhJWlu=LNT2eYNIDzty>3_Tvl+Egw|N+wA6VRaJLgb$?HCs912velbm+r!%)+n#Hy4
z>VfTdU%2_^M$B*%77JUQnX_hX?iMTgA2SbJfAZzbpA%~g<c)&&7gXs@ty{VNpjoYH
zS>~Y}-5E~P-aMXY+T)vG)-3O-yQxKtz1vpd)IRAea-Ev&-GK^T=^MG(_*vsmyhs+g
z_V>2$h579Fr|$W_{*m9s`DareKRf$rW{;V#$Z_rJ2j4jJ=3h{May6oPf9iYA3w(cn
zH`M&+Khdq;o2#Kxo$s`BOIH5NM!DAT84r9X2e()5mi&^YHKF^#&V!%J<sa8Aw6j*)
zt{B02^3lCi4e=P?py|6Fv#t|4zDe-x(w)1xxdL9!`+GLtJAZD(qkRb{S4HlVJ{7fi
z=H1<&yS-I@n?99t5zSD3#GF>MTjoteRhfLtHHH6oj_^J=EnOUZ;%$rWwGUea=1pb~
z=57kmSm}5EimH(XSMj1NE@jjGgl0{?(WH?5W|iHixQo3uVn33?Z1<<}##VR=_d0uW
zw6-g~)>WU;<U1j^D9g}!=FFm}WsUc^Dk~T~Pp-Ll?c&>9_Y>QL{@8?ht_=PhQ{S}l
z!>_Z?pDjCgV&c-CqH^BoAG3Fe%$+7C(*5oC$zH~wKd&-$j-KA|d;bPwi+Q$}0^&}x
zyh-ieWqK<5ppT*I>UEosh04@&ro~!Sa$buq%G$HSX#F0JxMzRGE6b%Hof0kd`%+hW
zxb4H+UmXrpX0KXtQ(^9ggo9ru96GRyqfN^7h4AD<p+cz>S82!vichLFIpsTBC7*F4
zm&U)@Kfdg){(L)ici@b~gL+T%*t530Hv6U$J!S6nsfU+LF;$hBXt%TOZ(Q-Ce@nQQ
zZQ5`lVA|6KCw2YTO}G{oa>eL|v)lgXo6kxIx+Oi{w#_)bZ+q>Px}4qSpYwUPte<_p
z!bR6xZu^TH6R$bDTl1WW{A9f${r<+7SKhMIrByAC`-&GOe&CFqI%EE%<HAxO#8t&q
zHmy;yvhrK4eNB9s;f0>*kJ9$|Oqnxn#dXguqJ^Rzx@q0q0ugP}#ogRms<jad9!MLX
zs5C$E`0$iT=7O@zS8l4Bc`51E-xGQDH;(^y7X5at=)=F&%VVYHu4Vn3v$({Dc}ntC
z`T5@_1ZMAel6e2#nOQ6AZ!)R5O!7CMH)C(Y=6Eju{_sDX`d5B@GrX2*mi(19{L{y?
z&hoVZLOZ1reK*Xwm6;qUyyoE*rtYccxkkC68$3DU6(?7&QqNE9me}DW_9>+Q<x*~e
zqo-fm{wuLzTX1pCGj(aHhlj7-j-2plx!&1~Xl}mmd$V`H)DrL$;5f27?G@AW%bRal
z>_~jdRuM9PCexk1_?MUeU6_^l!NPFT>Oa9LM+4Q?EvxRH+@Gi*U}L`L*w^Wo@BEP~
zvs`<=sCoCXiG^#Flo(iDTQ*J%s^yuzHZAMso+nCE0>n;6ho)4-T$uCxO`yb8wkM@+
zRdcIT{);!9W7az9?QwxqyW)W1w-o`Nk2XG-IFq$6K|gELOLqTlFV{96nf$>+&}O4t
z;nKG)>bp#$f6pnO<a;N2{lxA2SOn#?yDDbZ9n{{NwRHk#-|@Tkvn$KaE{|L(CKh-!
z?6Z$z(DBV3Dt?#swn={Exh>}#JL|zaG3U2iN)+B*jjOtFoXx5FN*nwBCnfuZ*Rb)w
zyZmoeKWo8aA62!3Q~fr)t~tZKYD08@V8`W-@7KP^<nc-_eYTnR7`MLI#)q3PI|r)G
ztW#KQ^72E`jaPRjtkVBBdEG|6`19{lcbPnjQ@pdSt0R8m!zXNoeOA^tCCWtIEnaWt
zcH;No)+*BK@yYvGn_(g$yQ`+&e&clyTVKEB){R^K@JRa2eRS$)a@Gw27SBggVsGyA
zeCy@!a%f)^dem#Bo`tQ=l!g29Jg#s0cu638SI}v>go51`E|(vd+>czWx}S|XFX`2l
zyUFr0D~+<u6=byiDkTfn|4X~@U*ghgzUA(7H>@yTH~D-+#Je`p18(1ss{Ho}DxY|^
z>z9PL#m3JQS#<NCcmL<wQunGka?bw7HAU&=q3cQ)6ggZ{75n*iW!iVIm8rZ>+sxWd
zuMzqmm!vQF=0&`MMCsDP^v?VFFYAB*XE0dx?muH<&5yZ>O;(d1Z29?O`+*wmCng^x
zd8=~jzn1Dq*-G#4YW;qn<x|}8_MJ6HBxYD#KkB+md4ldm!yLx*k6tHb9XhnA^i-+)
z-g>qp$;SjD`qOG$tJmGwbmZ#m%hKiR=lm3EsW#wp*T3v{(9(O(_9nG0FJFDsdLt>>
z*Z*kSs)7jCc~`#{h@915aDBr%P3eoro@9Itd3k%*D*j0`mt1}`*I#A&&Cjjs7k55t
zKKA<3?ysELT{D~3zIpU@(mj*UYb00w_!>Ix)PaVR=Ql8y&9LcPezaNMpGD}z@%}`E
zS@Yd@SLhTKf8O!y!@Z;3;p^i{=HHuqozvH#;km73?S?9;aEG$xyIyu`N@m=iqHFC`
zZmE)TOx@LUQc-EI`t5I`VvE;xUOr+r|Lp760^Gg}-s?rY`01LwbLSWP82Lxx0@wa}
zCMAYG4VbNUI8l*DW#Rv=@8u0Y^KRLGev@a0&Wrhz{rhz8UGa^$a(CtVopRETCc8bi
z4{=?g?yeMhH@`XlXZ)>mYM<7xR=G04{+j23sr#oNuP%u!iu;w<qHcTjfb@1%_oDUk
zri%}U_idkN_ab1*{Lm@dntr<1Ssv#`G<`0x`%phSi~CgfZ=EYSVu$p_cAq$Y+d*iJ
zZ>O)vIbTf=cimTIuD4^_|E~6)w$odU?UCS(o2*AwUKu}9;XHO@mfC}M=eY9Fr$!d>
zZF8sJJU#Qz`rzVp_vh#T=$lSyaCj$E>;3QJvagMD-2z|P47ZoQU1@xFZn&VzpXOy}
zWVZ&Y8Vde;@uQ^W`(G>nEy|VQZo5=A>1^Ajw|<cU`@w(W+^)O1Hy0JD%9q+MKigF-
zy_ebM?j63$vYIPJ=XyTa@2siHE=h^X@_D;)Th}d{<j5BJCg<3nU(B-eZ-4B1wWq15
zmw(eVy|`(*dY9|N7B-|?Ssdj~eX!a3e9MBT`xfxKP03mu!r8cjC12cCa?_UHOMfG`
zs<N@|ar84^{mms)`BBr4<+ojS={P^B%S;aQZ9Vwvd3NQrv$vHl&b6#f-j}$}FyH9C
zVX%>8`1ck&iDR$6dYP#R2cA5cxtmGo${Tid{Y{m>vZe${hp*KxT`+COycPB}zk*M0
zFMj{1jzj;aEKiWpA*r7+oZjk>Jf-emeam{~S&Ftf%YNyUP+94VC$6|`j<V;Ry0=zS
zIzqm~DSOr>?IrWCmd{K0VEw_rs=NRC><rEyaf<R=Dsx|UPr6>&!Lr@mX1d2ik6n7t
zUffyvdhgY&D;mqzML4fKFs)=};qO0onh^`no>;kTV$BiugKMMxC)liZjMrMbTI>Gx
z<_=D^iuwMJ*E@X<Rh(aZcH!+ueeCASuGty)Zcoiydalc9KeN1-U)_&`q5|({ip^?1
zRGbx&?ElTNG+lmUXJ+Hd6&7i~_f-bz%&^X%6+U^7!c@ywoAv)zy_qQT{Bx(Pb+)qX
zsYhOioK^4LzP|fcLgGRGmn)QvHyb&uO1Nssv+T&nHLF&x(Aj;o!^z%c>6}SU`}N=7
z-+oOh;<C8f24l-uW5<P^UC;j<=v;SIKIn~eUGJwmM_u{6xovyfRi<wB^_Nd+KOzy}
z5q9@9&&tieJ8tzX`7(FKo$LKqy*}<aF|#rFpXm2kred|PX_wSI8<cW(wZ~swZWS_X
z<6@6Y+lwZKQ`0ObZM4~av7^bq{!ExxV%*pJ{Ly_OD<0qSZF>AH_^w3x;^p%yXY2TO
zr+p3bW^B8uw{6)))2$pu9zKa_cJp)AOjFt1HSKWQv!IlUpwIo+ey1lMSS#)8%g6fj
z=1y0k)n2c33QxV<qUxf#&v)~aLsAj`%>}oZef+-H+1v6S{Lxh(w@-7e&Up{r6r)SV
zlP0NL2`t^YZR5s8S@G?E-niJzaqnfUFh1At&wIYYE3F&$d%640{PC$e=cRAH-E{u{
z&v#mbudkaMZ+yMYBj!Z+#VxGjSAN}i!qc+<U+F>(UA@Uyy-lWmd|LW5RV?PmtmCbF
zf|zIPzkFJ0Kg<8qedn}~dk%%okZ@~LTlU+S`^es_Etd^ecyHe6f2ZJW@WKb{EnaM&
z+TQanCp}?`u#Zu+Oyj2KW(Na)Eturn>){vp`P}KB@3=U^J;XPpXG>f-SQ1^y;J@ED
z>U8WYi7)STqUS!!x0rRL#qtHW%T0TO*gv0AH@+<XU}Py<zv<1%XX@%_);`ls%ThXc
z^pMi8oEu9m*S@GoE<V0|QuJrNM4|8Hw|7mL@7r?a$Q#!`8-A@?^=p%?rFGHFz3h6M
zeqPd67y0x<_Fv<|o}jh2L^qvIJll8UOaN!8-OVDVD{lWvYA!Cj`r0-|D1DQ>){Hf8
zVw7i3O_gzbs=B%D&D}}6UPz_3n3-kVO3s(yIlTYpzu)r`<|Y)qs@QasN6NT#a<_%s
z_3t5Hwn)wUJmI=ke!1D(VvFnlJ}&(&^uqj~m|TyB<h$?Z_H38Ty6~-erE15oeXI6u
zS*elipJi9P<4wk4$(L`lqto<0M7`bFc#yGTqK2X6+<wC_m+A}K>*gg0#TYx=<iFi2
zmi>NRIL}+*3m=(__WZ8CckkU>|DQ9zCPjbxG&`{0!{*7wX#e?kyVy;d+^-exUHYKR
z+hc#0UDh(E-y04|*9UQYQ(q}uEOt!$*~Z`B6j%I9-qgMM#7gn0^{+EDSFUiLvf60*
zG~J1D?^T;BZul~ao4u@GwlPe6{ewxZG5@@G*&Oq_@%n~anxNgB9dYKy$-g)F7%lcK
zH{W=dx6jJ_+E<AO64?iCoPPKGr*HU1kGwT$ua|zknzMT1`v+aFzrRQI2d#SiBW6S2
zWv-n%c_I80EN-j|a_&1O?sk|p(06{%g@fPh+-*6({g9rxXveg*lftfU(ml#^<FNUn
zUv^LCPhZn;`{LpEA}Vq$dI~KU7aiWFb@#OD>ziI_H)ngxcbuw<J$X<pB&x!#)BUr_
zP1h&))zgKe|3(~jGm2-dWS^`Q`@{R&+k~@+>t_o;i)6VKD^c-tnxp07Z+x+GTRh|s
zoi<);_-ArM^h0^;>F?j}ESxXpXvvxUQBp8&k??<4)>KY=rgI&qsv|Y$zx+Hq-`rBx
zIQU_hq4v#)mev%V{fGRVOJ>fyIWzqDG*9yfoFAi=rY~RpGx*J&$#L(vk4Q<pj|n=Q
z_S}&5rd#sNCGW5FtXa`jk`|uFY%iMfBYsiNUd86AZ+a@sP1%mJxn}$=7YnhFKhdLe
z(c|#qq(c&`PQ5>{?c4fW`ElzU4Q58GYpqV-Tr~60-D6Efp84krF6BIIn{)0EudUnR
z?1+|=h7}TVjka}%);DsTkWZ9Y<M5d~(<V)|Y=-%DS(yZ$o?m}XZq1(We`N7RhSTiq
zY=uF`3za#VDsF#rYzSB`qWnCqrBd}Ko6y(I@ym;9T1(&Lze<01KPdmf{!K+ZQ*Zw4
z>(h00ZOd7FFd)&OsH|`OKl$oIdxf#heIT9aiD!1j6n%DO!_WW!f4{E%AASGt`(g$L
zw+Eb?u$hjuS%uV@6?0P)l!Hopqu67D!|(4CeCHeDqT-gjZk}}d+Ul+AuGt2<td%y8
zc4PA7P+rqgE1Y(tx_Lnpb6{Y1*hbU5#bpuEt4ptJzxplv{M+gmzkmA)nUw6kuWnd%
zUGo3OKl7j0KNS4-jivv%^qWnImk+I-eSCdc=~On)bpM$5PZ#st5Kh{+rAhh}Z?Y@{
zJ4fHc!V-PIq~pBXuUme&V8On7HQ%?`*%!~=clx(hxm}RO|E7HXcPVp*Yt}sn{!3qe
zV{ob_spW7+)A`4ioo&4pNB=(e->max%Db;oFJFo-`4HUVKieerRL{|%$cm2n{p+2K
za!*ZpDSYPsX4P&r&zNr&KfFVW>g`X?Q}FLRt<w2g<)qX~uK8al_e^E;{8shz9p{H>
z9(z@#_iXc+QnX%R=Mi<6RZ~vhpQ&>5!E5#DE0!q!6Z(_i5#DfJZs+vnE|O1PiYf*E
zsPdea|EHiiWu}RTV&vQB4@FhJPx6_gG4aISJu3UST64}lImxy3yh>)r`IP*c{+gnO
zuWCP2J{Jp5-?{dtLE+a^N4>At>?oO}<fHg;TAPQ>UXkiQOiw27`o7yuN@$0A!-4vY
zBn`u!2?B>Z1&??ni5Rw3%$D%p<J_ZsOjx+1^N324k>S)CZa(U10`CQlJ0p+eB=H$e
zubBU%0W`fLCcxe4dL$`{bK{f<Hyw4*1dDi*=*C$Es@DYajzlHNZk+dH!VQNU6*Ixv
zow`TD4$Zyc_(r8naCfKekvPyy%pQ*XL(@Mfe-r%Y!XW6uB50&EgG2F2OH+W*LKg|a
zbHcX-Z9aFI3(u~8`2VJ`;gR}^=T*6HQ#VDYm!Em(UTy!|eMw7+y8Xho;Al<euRZ%G
z*T);U7q}ew`t5D)((|kAj{iPa^uY2=iN~Ak+v8*Q|EklMy>2n>>+;{zEqco)S<6a0
z@U19~dt2ja5nqw>U|wT7`#<+5YE}j>-Z~~W+eyB$jIjQ3Yl5Y-P`gZR_U~x<PUeiR
zdp_HFWGzdMUl2ar#%s-^#$fxy+_<N663^5vJoV*2Rtg5Y-;uS<Pq}pBcELjB?3%PE
zZmTOw=d70%zps(Sm8R@y(fX@EHO0AR&cXh^#eW`cVf@JR>T;-4TX2@->;3E7bdwbN
zwbrCc&z`bIKlooy%j6ZGI+v`UboY7uX|a9#PAy<9Xy)KukfpGn)6dIZ=R#nWYu|3Y
zxCb*bN{kP29#eOF9;~*)gmZhbtM4K=N$Hh*shbZme0f;^?3?7IHxh^SI9742*Sa;o
zl55G;sT<gSeHQ+G%^)L>OKQ_!*^LHe|4OfmSC-#AEbFl2!1`6qxr<(}x}o;@N{h$A
zwyG5k8#E84tIu{>fA@CDlD|(GUo1(MN$$On(Q?l1-sy9{=U!NOxv%?~TgR-FHBx7`
zya|dEh@W%lL^#_%^}AA`FYm6j(%ezI{oxtak5eDqK7Bp+u1TTuzgLIuDrjB0t!@>h
za9jNOOYOS<S0-6M@0ZD1!y3oV$G+|2^{o7DvK%RqnO!=KAN@OnS}*RaYwun>a~XH4
zY-VKo!wrhOA&zrz2E1Vo3`lu!tyJNYO!Xm!e=An)`=qeDf%Ri-t;@kx3M={FidL_C
z=~3o-yHG8@EGsQ?!U4fstwr6THX3^_7CgSlYn^BPMSQAjs`?VfBfEc`KAw^)7d=H&
zFI0S`N$1IvoeYAjJ!-d}i_kuo#{cWrrhQQ}&$-Goxy34NUKIRADr#vTPxr|V&CcYy
zp9Mdvf&>q{sdOFYQa;TinVBicAT<5tlnEi?Gb2v_lCwD>cV?Zh>9UY!b!m^%sdClJ
z)@)cY<={Nq#1_#X-Q{~eXa6z2Ffq7BHH#zV@uA5#UuZ0t@Y|*0g7UHd{8J8WumAJ#
z__bvT@0rxBt*!_-osTn>y}8itdis6azqQ&i1v6VGW^hWi{aUP;AsNN>ThUPF-;aZ>
zY{z&{xfYk@-r0Tl@uRu={pB*Yg(+{D8knU{^%Qhp{5g?R%V)OFAu~OZefqmz)YfL1
z?ufozurW3#s#yLeqlnkar54k&3w5rSXg24%8%!0_QYihnyZQ`EMZ^>3Xgj_=yxSez
zy(-?<-JHd+OvbeP&fe?iSx#`CQt;(``~JG_&D4A5dvaR#_UuVlTVedLp@~uH#`n9^
zPt^UM*P#&4F#YrFEo}A%^Sd^k3q2w8K1ow_bw*yA_GyO){|-Mqo|*Cek!4Am7~iaC
zo02XHo!39#B>ksb{I-7V{v9=J1tq+H*SZ|teZY6a<2~hFn<j4x-ne=HpIpnCvsY+N
zy!<dRtJN%awRhSB@%{SB&c&Zhj+S_4CaOM`$ZJ{njGtfq_JQ@5``0e|8#Gr;VdFg8
z5bZYOA4_w7xAQ)zQLbKfr6%I+wgU&NCZzj)N-(oYarOETmKLk9Ti2&2?ZeYIdSAl0
zIwmh=`*D4ll!S&~+V_?j>selx*GO?p$qf%$^<t9LoR0dLl5qzr*Ij<F^74)3tIu|3
z1^KPXV-U3tfAvwU>1His>B^fGCq%Bfr^y-h+%2;csCjXISM)Q!#Jv|DhzKONE}Hae
z^{-i%&$O~_ySwR~);Eh9Z~mgJ(%+KDBahwBu*nzr=5`}2aPgF0ri?rJCNk-5%hf&`
zHpi@}m0BzR`M1}<eK}_t8De#=-?Le5cJp9&wd?-1Tder+|2tM)aOb^=)b+<VTC<E&
z<PDm+k4w$%pYh=2Zf@U0%VQ+|IFvXXJEqFE>7XXV*UNn#olBRW`zPAh^uFfrWphz}
zlM^eaaC)v!+ng!GwUj62M{CYndx3o+YDa#!8ZGUbyEQx0gyZ?KfGtdG8}?o>uwa?g
zzppN@bN6bud%nUv6<v#&KDj*2YBsX!3e9S;vP(K2^lNYGb@_QW|7**f5D{Op>~gB+
zu8_HI52lAnhr3*#HudT~$+LOx+}9PVtcw0zkGLV<cjv#e;^PxeiZNWRHZS;E%RUsJ
z^SdGw@cO+j_lNKI^s{r;-M!Co;rO?D%jX@xTAYskx)WjjBf0lQQ~B$%^Air#-rKtA
z*&3a>e*8jHAI3#1vCMk%fn{k#tVG<cAD7)P+?%s+&ZQH-53v+)IDROxo#R}Ec%O4V
z<C8@13!n5-`=7MRiOBrXmElW&Fvqc}^n@dmTJfp(8Xj|9lTwaHXrFa0PP39XNM9Ck
z_NjT)!aDKgi~miUoWT^j>dT5dF3P(<KEBA{Cei=PUXHzU+qFl|?FAD#6O{WJci-J%
z%Y06RbGF&LIhzx1-pb~emt24Eoz33L#jG7m+;&`j(R1*?%_~W_a)kcP_O?~4>iWFq
z((T!yEsaYx<h>Re-|yx%W_Hl5xL2!thFc+h$5bJ?)>*wFGeqUR4%%k3lzrWnyKkSj
zWQz4839lQcr5Y{NT&5@5E5BcSKH*%f^|{T~I*S&aP23o_ud8?I^|Lk~VvZg#TeXJi
z`K|*`t5ux(gxwbJJMr@Q#sUSA?Y8TZo*fQUn&Ny{|L>O%$G`M0wiZi?%u1gmm}h1s
zZ@zP8g55tMzXu`(8(*ycDZASHK<lH`xz~?=Pp|8fnQ-{b_jLW;yB<5K2;3{-jNXyh
z_hYX5<%Py2b)R0a&ftr7k-k%=f0VUqW%TQvmu1!8-+Oyw-&D;547-cIwih=P{GD_4
z`TCe&J3c?${`~GWgT8{pbF0fM&fZb8)|(`)koD2z`DbpeQ<J}bvODt3Xu{$m{*7mO
zr1I~-Ied)iNZ>}hHQTQzZfj`NU(s|phdcRO&+?y#e_v!_PduOC(e*6Jb^Y4cF0Q8x
zx>p{ac;Uv<FWY4gt_*RsWpmbCowm)v-=gDJ^l{!p(}k)l&f3Jc3tA;;6drWkJIPx4
zSBlx{dFJWgQ#O>^+&9`|`6g~bpJ7OdK-tpNO;xv3k`vh9<Z_x9pD|kTo%7W<)mY{U
zOQ*U|yS&ttwO4HV8^cF4mr2dtZ|1nx)1sTZ;kE1fd-aV~?{CjuafkOomc;4pyg_b1
zGq=R?oO;e}AGYxzySI^1(WZk7+qN~8{rddaJN#_)wz6%<pDhYgU3@w9>5DJF_H2)c
zjoP)tEmTZKt@q=s;(AV%hYj-b&eosB>Wzw=4TVp>PWgELz`{9VdE3)&IQ{6dT44I4
zNkL0R(bQy?qt}s%vog$7Lu97NU1eI3Hzk^%BR28&irfCn-l;@iQgQkx@}>Q}$7erv
ziHC{G>u<7jZ*+XT@EXsXtOl8!H(xlu?z>igk*n(8pZO1Of0$csfA_7$Lp%NHfv>+l
z^sR4*aBYf`;`I`}7b>0c#DME!hHBAYN#4d<eVLo_F=tqwAGsFwha>&tY2Bwht+8#z
zdTV#qyfV5|a`R?J(cdFiO(y8q97ru<cHMgF?`)CXLQQ)dJFn<ppFVF#!NrpIHUEVy
z7kBn>^NGFJoc=2Kz$e?2dRZU8#NRPEWcB~nPhFYLu+5GMoT@gfd($p&@ktSyEq*$K
z=j6|yk4nt+4>MfZ`rGvP%k*Wp@0(@4Qp%dqwa2Np(NADb<T~92+TPx)3KX4fE*`tP
zdwb5!ySFZ0yjAtg=k>eZ2jWLMZKl1G`+RQE;<aac-On#+_IqXcWaDNDcRgznzOauL
zlU5(!VrXo|yx^ex#LQbNwKi)zxVZT>uRc>Zx=_hIbI#1UlRT%3hfOi;+wQ!xMD)Fl
z`O&`5t>Pb>*>vkZ2!2t2eA&b*OlDn>cF50Jb1K;6Tuxu)+%nC_>zZ0ir?b`~Zeyb_
zzkC?ZmOoE@o;J1fzJS?1X5lYK-ub*1Q!Mg}ObwOz(c#z^l2TB<C}g4T3E{k-kFT1~
zTmLHPRP_oG_uH{WKG$p}`s7Kk{Jn}*NN-<SUiL?&dsEm~EULS4&8+%&Uq@{m!@6J6
z|CZgks$6k@GwXZ39TR?CzuWxwE$gSP3@88o+;&ASy*JA}qVD-tS^bWvJvSv9)&*^h
zlUP6NrVU%%(xPaCX{Fa)&dhzY!?z-KT7bHRs!9$=V^p;Q`^BX$F8z<gPS^H`w)?D~
zv+(`upK1?P^UA*`C_k{x=5<+I9TF;3ny~rSwrh#9t`@Al4^O$4c6qG~SvbK<TeP%v
z>eSK$K9xzmdsh2DblT%*S$;g{fJZ}2?zMgQP3KkpZ)OR-qZhkVWJ&t6RFSULhqRYm
zGMHNO^y*yw{Xd_bDL&x4?zwo>n&uF<b+`813-&L!+511M(X(Rm<M*P4-McDpRNL;c
zS&^+Hda{c3H^cr13So}!svE3Kg;)AtP3>-NI=u2XkKvXJo{~Z@&vOJn|9e8NadC9Q
z{A>pcuj97AXP=Gm{U^}vq-!>*iI*kAET{SL1FKCtWj?!YZeBjCckB86jb{@t=%pwu
zTzF?w(dT_LFaAFg78jWRj<c1yP;F9yhL=Ke^RD&xH@m9nN!7e5XIihc*IJWF!$fLb
zbL~Rg$_?&6{5SUAcUj1pp~6|eAu4mJE}vhi=G|(Sqax8?cHCe1?(Q)@dGBARKlZ&k
zd-T%7vsP!O{W!|9cw_X{<1YRMOIIJ7>L<rGZDEFA<+*L?CbyqWbdL3szb*Lu!rAWX
zmQD5Fr(IZZUF40wT<9W4+dlcS^~=?N*@#Vbm438|!HQ%0VdetXnUg~&#OkdJd$yWU
z`mEcdU-jaZlMHX&sF)fNy5OYWkrT~-POeFD*3#C~G+wRrx@%qOzgs_rZd-Eed-d%-
ztND{RRIGME+O@!ag|`k^r0GTpz7^bSuJ~YWh}J^8y!xEU+$His^2-f>TzI{PZ$*;y
z3_Vr*fO}^p-bK&tbx*O7m$i4!<1k{0b)M&(bZK+5^ozDPv-Xrc3*Ip?MLRz|gfVtH
z|JUW|yO|H3eNnSaIOy5=0}Bt8aocG|uF-h>LsVzWe6i=}^6qk93O?kQvqbdwmV=Hx
zA5Q;L5ta6PzxmdNCfgGW`2NlQcP2UdGn?K+?%VPSKVzoJKjjKwi5I>y_p|x`g3I+g
zpKg`h|GGu-MwjQaH`{+qd(nBiZsio~DGGv5PE1=75ay}>)BTEC)dI8U%NP48a*8D0
zdJr0LF6`n~)m@8umal&F-9yxz`(j-7wcqvp_YS=-U3s+n|B27m|G3MzU;pR-(r3Es
zfamsg2dWl-&J@<EZAj;9n(k^KCRlz<q3DaxnYxqi!3%3|yZ=0Xi&4mJzkRNhtn3o)
z5W|;}FC~vAE!<_a$jD<E7x$fuJH&z}96ROBVctA*e)qMt_l@%3r9EEDe*OI2T~!H(
zmnBVD>l5nIDSO)aVUKme4_OOl@pl?KuEv-zYSa$pX`h@MQfGYo(wn2ljO;=f-?+-1
z<2c#7;JfP2<I9Y)uFRJCzdc=i|2oT`XH>PAc-6}TFW#FGusy-4c0!j^UrI)xi)(V#
z*@Yz&zUzPby?@@W+_fPYb}@QJ{08?n$*ksjHh23alj7dy<4&(9XczHLT^LrfCWbBc
z;j6Bw3LYJkj%k}uCz&yRx7hh9_E)Uy(as39XF9)Km;VU<78WOMF*$zzcgfoP`ZLFB
z)XLd+AMw4jw(5N1<;NHQo>?olywhGmXrl0OCG{s?YGk&we>)%VpuWMt+vIBQ#YL`p
z+vV+d{W<wndj2lI%?>uHEqyaKAA9`#dg<beCVKrwn-6iURkLDb(fK-|Qic1<%2QJu
zHfotKdv)UBr%hfLOe>}CA5%2S>DYUO<FT%cQ*(Pr!Dko$X*RC4j6we<Cp0_fJXPS(
zJ6OZ=ZST)7oMM}1f7hGO;CA>&_?6F#Pn$o^l1N&(S9~|KZ1)Yt`j`5g@iJZKIDRSH
zYW_YhaNqB5@qSN>59ejoE$*+0HS6qpWMnE)*lu%&`{4OUyPqEGGAnqJ_xs}H->i|6
zGuvVc{hVg{7#82{;Xk-6Qt<M$j&Kp59Y%~gdH?UMowe?)&gsN6%vCWD4!TX2{lm(B
z^y-S|+Vh*EZ`-}#S+=y(D>ub&k0N7q+Oa?ux31^yyEMv~`hVX3md7uUAuRORB*`}|
zBll@DH~;zO{(V|?CBbhjUUP}WH=YbvsBJCZuzB%k@juu3stV=1-~EWWv+3sb(vwZj
z+2`lj{y6E&puA5!v`w^)Jv(%xC&Sf$`)2rfdiY)FaPC$<wR-cZB4dU%KXV0V=3Pin
z>+~=GtFr6u<<9m44~tLm@?Wf!Ua&l7TSoc}r`4S6vvRf9o-vzpV&kr?SEt0@9!lSu
zIQ4k{<IOjhcEujJ<5YRq?&p{FcZH8nUs&)~c8A#P@Je>?TjnML4sEAWqxTfXJyK>f
ztdZ6|;vS;&`f87{9{+uI#`|8vyOp18ns_~_N%cl_`^75lgT>GN0>f%oaWU}zoA-2f
zrc3X<hB<-z)sq!VUq9K{#Kcy)>pR2F%CK$iX~j+#+j8a4#O`@@>+t(LsrPqs=RRkX
zdAK^DLP-DC%isF@w>KNTTlOL$pw4AV|1n$MCyU>vblejN_MHCoY3NId9dcFwr=QM0
z_NhLRUB6J=_4(z!my%xyEcH6};_#;p6&%-JPTdjqO53Z%?CrartA~^>mjx?^y+8Ck
z=4J6kM$OfohdpO<2?f<$QG9dq+v%KT%eY+3Swq#9a<=?0-nR1L8Ox96pZ7WKWsLgw
zUzlfJ`HEGmt~{u6`>}<I``6w_rf>EM)hqnDx_^t^CxySBZ<nv-nZ;Qz^7d27Y?f(`
z=U-o$tejnbD@#ZwUC8gDl4auxlV!QreqH_4`!2ygzT&P|#hYA@53H|d+-4T{sw$|q
zzA;_MD=zLYo6rI4wR(2**Xii5tG=6JQ_Y-q??$b&#^I{^bt?jkRUR08u-R#QUESIw
zIk2!~S3>%_^9vt8`Zig8y<^?OB@;>nX2i{jy0W&BJ<?=Dkg%)-e_u|_Y%wS0dw;&)
zzZ>y>uJB=*yH^G3)tx5HDYiB06}uK6HZ|{~aQ?l&4-f5?Yg3ZE5jefXWNS#klvz>F
zysI~Ji8i~`m}+OQF?eF0!uVhK*bQS7kvqS>ZQLnadX__S@u3$rr#XsyB=(d!rUzCZ
zsH;@`u!8U8j0lYzlXn_zkBqq-@%?UjW}TPpgjd-&bD}gh>RdS?UY8m8y3_FE$pu%k
zrKKGgAAO~H$=Cf!Wz;i=jMzv16Pso&Ilp*&7jsk1@%0}y_OD<{{ph(f==Gn*8w
z6GEr+*7(OKoX)UV{_14nv(n9<N*5QOus5+Xz94*2HS)Ba{LCxiaSuwZSTp`Sx#X}_
z!7fr(J}ombY|o1Zm4Y*zPpx|PJYH7j&ATJ{snQF%P!aiqS!cRAlgmD>y}R&J6vO3}
zTea_s|2QX@<JK2-CB5BZ*4IqCCDs8fmlg_rda!AJJeQaEr7x{X#m^W2JEd!JlGV76
z-Qe3nx5)*}&t#-sC(QXKaWnsh(6i0r2exy`*M6O07P5JwtB+Gn(Hb|MoFEbHSJ|h%
zr2T%x%k*aaxh%hPaYV-U2k8obL_XF1;re!%HR0Q;uieLg|E+rYxc~f}-;dUueq_Hd
z_ebs&R?DZeHtcX|u=`yrxUx4T`0>%Mpi<AN3y<{7nV<ONj9yY{<fA;Nh~y^bGxHla
z?>Hi-yY1k!dz1Bz=l@^kd5w|1M05KWh6Sp&Rk;zb`Ci-%@ZPfG>p!h@i+e|Ay_?md
zzoR!m#Y*DLF}GDge}WsUyV@`H$DJq(F*M-udw=oktba>cIU7H0Q`LHQZqwKMatfxV
zpQgP(o4>AnmP5bT^UI<BH<xed&s^@g<NT#3d)`;oRa@?;HGA!~_Q$<H+B1s2#P7_l
zOmM9zEv<~~RkSoW<i27idDG{Y+wAXgnP1}03#eHr*_2jqI$-SDI<sY`1oOg|5??QW
zTpAs=KK^&!hd*C}_x0H6UOw>ZtK~Jj=&h@4BGv``_O0J0$Jfv-SNHwSp~$4dWA=9w
zD-)~wxh>RIy?<xZcIu4D<Hf=k{sopUUGVe5+ST?QiT5|oS5#M+vfy6v^L@uT=GsU-
z{&2+qbXJ%2sjZQnl4m|XdTg5hLf^;wUhJ-mXOrcd9~szWKYsl9>N8g{#vg0nwz7qN
z-|q8lyJE~w=~DY6*JRha_pc0GZJvK?%da>740h|@FXaATaNU@Tx5DJ<)6?$v4>B^?
zRj>V~Yx|4Oc>B&ot2`mLF9{#N|6At#-k0-s`nm3pvyWeU&%WW&`dY~;e!hF8YTiGX
z*wYrK5F;=n?%1>ocWtZPP2TbKe|CLRpSeq<wJ4${ywJ#6@8+*KiTv&TzV_Dx7EdS?
zF_~m)J7JeU|2+4DaW52K&J8~~?efyAZxn6H^nd6^-q(Ho$!X_iw>Y`!ImadAPA`>X
zUC?!_Iqz7$$Bnsrcu$|5;$L>q_hV(#Bnwe#_s8o$Sg0wMso$EhN$Y2UQ}m(EyB}^V
zPBqiAynd@>WioG1->!>Oy;ohj<TN|enKgRq;U%ea8@s+Nep&6W?7Ay#!HJbe3~o;f
z3tW_*S5TSKms)x4t$5+avKTh=OMjx|k|XSIfAbHvzF=rH)oWF+&H7w{nk9Z(pO)25
zy7F3SX2km!Z#@;8Lb9}N+~&PYoLhHIEBfd)L-7mdOTxog-5y@sGUr^<vQOTzGuJsR
zSbS;8(RDMu8lwxNz69=C(v*8LN$)_u$7$}rc|IGrq?~yd5R{>z)4(!iqNrJt_QrET
zXQvsjo=~AFGWqD%yIyrW&h8Tx5mAto>frKNy7tHaELpXdP^0+6SMM*pnzCcfZpBaD
z{+8z+o7}BB*zEN8VF=IV#V3?)rHpFh^S925G|apiylK^{-I-s0b!~O$^)%u7>Kg3X
zFK2sl-?c)K9D&c4B6r-Gj3WLW&#r&5_{G_a$!V2$cdtGu&0kdN>=(uF$hFk=5gR*8
ztk4YW>vzp69v?R1>T`MSu_fl*|6=zov;XNFFF#P65?RhHxUxspEpP6`>sdzWM?H#f
z+cGJ1bsZ3PPj?elGt`v4vFnSOiQlihHSvqMj_1|veU<Zp*EPLn;h#+lHq5%3ZuC<!
za%XE&PJhAY!xJlP6Dy^hjp}yDscXu#)v1+gzPz|*M(CMPcE+Xd0XZICVJ{~%Ybshk
z++t$m^K4`N_nXqP^J{|7A1L2Zno=mc`h3EsudI=xNvqrE-1arN_9VaVyOyomu9WKM
z&suqEGFslIINEhz_1fGN%J}6(RR5jauPfM!O(eV(m9x|rFN*YCb7S`;UFHwwUK%ex
zc1<ro#WH(?VM<!z?lcpbo;QM)Y@HLiqds%-9^;x)!+G=qtIiqGg%{VY_|7iB<B(l}
zh)nI<-_hIG^*S$JBeal<%RkbNIkn)=eGb+g6&L=$p0oI;_@y}(w-V17G$}lrRe$hJ
z^79)pd4+m^7AW|-H`;xZUc$FGvVGq%yJh=yT`jHV#?4=+eJ$PE!X*6FWur@9U9Fti
zO{~5=)?)jo^(&t9*4C$3CU#Ab4hF<CZES95$5mWXl$w~!RdEZnDevds{lDx5V*Y=B
zf1i<!M}dJscN6z|Y~~v&7%1c?WpUZq=#x5eY-((ZvKg>7k~2aieA_y)Ra~Nsnzv`@
zt<5rzaTGG0Ib%m+z~iRfi_Ew<xE8&!-gF|Z{Qk0=f-Dn51*h<>4od2A@x5|HqEu+^
zjH9`C)Q?COuiAU2`n{Q+r`peFA8SLtm%o2oea`mzpZ|aTw>NFtwoHWU=!31NpXo$A
zwM@MD=&9oE+NS#jrDfbVpQX>7sm?#OT`}5GOd~SnM2&o8t}6egV}9Gh=HH6?e&lho
z^UO-;wc9Tr*6`YR_ui*JbqWRBluztd?A^9lhdKUs-!h5TLv<G}JiPnt{_GiV#UH)?
z-#EX%ASEH?&TPGP>w=%Pv>a<?&*SU}+5hsDRpp|mKl^T-b=`X)JvhL_$0cCFE%h7z
zdk?toz2W*+P{`Sh`D(Ir)75?d80~JGt#&qB9sa9&)ncVThn25HA97ygq<n1Wif}d-
zw-w1whnlZEF}+hid%u6n{`po~H_qF~$Vxs~KW}^1QLnFeCOq)?y6AxKzk`DRveMJH
zS{=F{FDoD^X_b3v$%z%ZdWsLD)pFIcmDF;p_h}?edRlz0czVT}lDIDiWy<?S-|0=<
zIPuk4#fMq;|2^g($Q9IE-{(Eiex8(9?*g}=$16FnxSVR8<+9gvuj^jlzpYj6RnupA
zo?Vm{oHxbVbUNq^Lb0Ikl}oP}Wm#@*it?-slFCwk>iaWb-xak}{hyYb1v7>yY@N!r
zXj`yih=!=2Yw*SprKu^dnmX5(1qJ_}P<|~%NaVJN`R^ZJcP1Q-`WaE1YkPlJ+p-ry
zzZi>K=bky2e))!eE>q6G<?D;@@BjAsRr!|xC;ol>>wkl9vu)1rqYXcq>$ZK~P@OeD
zzy5yW+{1qx{g{*4$}HI|?mb;rxb%*G^Qx&@sS`f!KDe*o#JL<dAq&q>Z4#|NPanKg
zAtpW}bC;Q_<k8QHyDvSu<eKy1C;MJy=h<eH`>z>Xxb9(I-?T-(OQ`#v%h6SD?jBEM
z;1xZy{6^=z6^T8*`~lYYi?}nTU28-yO|_WD$o4AJ`sZYx%VnI?WLVNCKdyVhU3Z{B
z>Sx8)O<J6)oN{a-F-w~izJ2>W|8e2zjd$0bl$K8Ze%(A|i)@eX<uuNn>{2|tf3BCV
zP3vlW^E+zJ^e*?+JEm|x5Gel`y>jc?pa*HW&T;=m`V4=UEzT0GW?cP*&v(WYk#{>+
ze&h)Fkan_gaYDMCb6?FBmefg1#|~r{2)RYfG(O<`v)~o$o&63BtNwiUU%%>|-%l}r
z(W+-z4X<|gYWBW4eqf7yo~_tM>6q|y5&J9lG{i3C5i_`3egCu2_s98DPjc`dTFkPC
zqwi2w+WeM1{N3#}7rI2I&YgMM;AFGUhBc=@Y8r5`H8S%&a(e16u(0L&-1m<!=4Cf0
zE3v#t*jO#KHu3y3Mf;7<CmxwLZTeeX-R%$U{7=7+DV^P3=>M)k<B?;tX-4Xd`46Pu
zOcgq^`}f>Cn}b_&wiW*0xJys&jey((y<0Q&mwS9N_AHc}lbXAHswIEYl;4+lKYFgz
zh|^C9wUT@A@Q8o??q8ct7d;F4biF)ezC@$d_tp<bpPP64eQ>ay(Ye#h@LI-&(2rZ&
z)<jKwvRp>=$kJCOFEbCZd%t|#{rvsh<BxUT^c+u=f5ZGlbW;N7g#`XJ8S2l|x19K<
z+vjV*aqxhU#1A%}C+|)^ybvBasiW-J%TmvohZb%4x^bS3m|)M#A1ntG0?ejmZau2B
z-AK<-eD12MNOtoLzVD10S0DJCmMnJr!S(n4EaDZf6>dD~EGv*~)aGFfKU!L_wP|jV
zd}_7B;_c-Q4cp6;9jdF7lXo{nul;Y3_Pxh(QqAt|5^bl}AC%DkoVjChOMs2sl#00*
z9)DZK^vqy`*s;mYxuPE>nuW9{g|p`hJX<?|&Vz%=>Oaja4C8C$U*FR-t9o|6xN_c?
zx07;QFX?6O(lKn13Hr?a<ao1cS)o;y*wgxrF~9SL`mS#{n|=LBp6c!189nWf70cBR
z-8kPDdi_fG>xeutjWs*}7JBZ}PXBa+?_k6FzJ-&wo=u9L8xnDTeikR!Iu*h7=Q0{y
z+Y@GaUSw!qCExItS=(~zk?MS%Q$5#H)lxWKFP_e>``}}N?Vf)VIbwZM6>r7QFHC(~
zcyUYR*Lyo{<{71m$aRF;_qMP6koNnsQK8a`DHgpav&^=H%52J9nR_$#xbuCTKT!>h
zA(0G>AyZkV?lqKQoU9_D7CH0&G;yu%j%;^7Z&q!%Kbb+O=!Kj655E9~w}OF(Y(iwy
z<;5z_i>X?yNlaT_*)M+0Np*$fj3`s)?{}Ry{qo=RU6|9!m+O$kmD~;6C)rOpdup4m
z`||$7Q}~y$E{~kG{^R7GQerZ_C6_Cgn08(gsbE^u!d|lZ!t>cpv*ySwIW1haWXaKE
ztM*7IXY-qdY~d>V^mNL%^9dR~Mt2$8+a!($FZjfKHCN)G>}2l)-kfi$zkj&#;OZ@f
zUIynIXP8}g9k#K!%eqU=KYYrXtBYebldgQaDtYhEnX6A(6BikV=3n^KsiEq4#Qmnl
zl#s|dQ|HcCFEN>~lk+rx@yYkQ78>3#OA0)vDL>IV_tdJ@d7-s$_NpfxpB2)-ecfJH
zi^NlptrOA<du5^(Pi7~Jd|);RD>8nV7GN{!*y;BhewSJ*Uab1tm%{k}&S%E8jbUpJ
zT$8$X<4ZaB*L!c4-qYO>*eS@&aZd2n-k!Za1|J{(=2w<$n|a{Z!(;8yKG9LKivPu!
zv`exDL{A0YEt+l<6DfNC#a8FVyf-uowl3WLxJE1ObH$ZYFW>F^T_de@bV5VV+Ka1X
zZ@!LdYEfAm{Y7bN`gNTP4ZpV<$IJRyoB!W%K$q8cLlWn#1MQWH&dYWf>MzY)zsjK6
zKX2cGn{93-HA_}MObofcb^VeL&emQ>o0p14y0S^fF4Hr0KKhlLSHFApT3zk+Vg(mm
zp0b(LZ#poSZ*GsCjBS_Ron5WZ@&y{-NiMkaXge4Gv2Cxdtn6&ty;xqHzQWuqHaY6a
zh4l^0r_Vn26)1N-*1UA*hS`6wuX<)EY%~3*J7?97;|2Y7r?(!TKV85gD|6Ge`K(8K
z&lR_8FG}ay@yfOI=1G|ywm~9?GqrcH<ehT=@ulLL-@N}_4sH=Ali9pyD8?M{SaYph
z=F_9S#}pT@U%}Wmfx~&zm0uF2A7uL{ZreJ2i^>bNegAl4WcM?N=dg=}GppS0erl*w
z&b2%8kM(75-gI5R8<LrMCtP<%|5Ug6_ACF6!F_2R*3GN8-duW8X<bvif~?7xd#B7w
zCg@L=c)RSW#^mqwn|q>Nc%%35r(Ww=x_c5Y>n%C9`Laym4NCWZ_dDO)WB&18e2r)0
zl4Ej{rLxuUnF}%mZ<zV_W%Y-nS1lzy($7@b>gutvewgNb&(Zzov~ID53F^@XHQTcn
zvt;W;R(j<<N_{dT^b>#c(u!-df6P#M^l}3$Tlr;IRn|)Jiq3+$W+pPNALmGIb`xSc
zxA@cw-_znBjcod+?TuLXfPdv?(?bpSt`z3PhSt5?H!1Ip#S(>`CT^>}R3k)8a;_bm
z-6nIdBx1ewADb^KG1H7TD&DK%kX_<(#^!YFpTp<AseC@aPAPM-%KZs4iEoax+fRFa
z()QY~GcJ>SC#D1o@g*+~3aIGnxLlo8xGTo{`jRCJH00W&lfG*nKHGKaweuFH>5F<V
zEEKSdac=AJT^?Y3KJ&I;#41Owbj$NQDxAaSNaS3R)(YK}lCW^ubHlS&kBGcEr}OCh
z)AyPu#1=Xp+p^GHewwY%_W<qlOSjzHZZQAu=BN5vuPVR(w)rn_GUav6w|B)nHXn0j
zpYBRq8$P2zKW);*`SUl`9A2a#s4;DKU;5uF;kkdAW%A1A*~Um3KNgkwcUt5w_mTTQ
z5*-uoz7N<@wKIK2!p5qv`=kFVbDaI}m1%6OyLVNZnML%>=m+bBwmrOPz|{WOLg&%O
z3gde$SvEpR3UVwGymm}w_ZFAj|GMma%)#8|>ML4xD{nXE*Bo7&ep&Bsz+Cfx6;He$
zZobBCn*QUnxU*e`;B2w9^D~w!9r1j^eZ6|JlN^td4bMTZ0N+=~@*nTt(RXZpa(i_0
zgV>7qY34V$D*`4KaUBwQX3*ZSt5T#kq%7jchCOQ+P0r8u=2#)3&2;h9*Hv*<(~XyW
zZn-;oq5aNzo{}q{YUikG@pxZc_$4IqXHTHUtp1B{lD&5GJWW3`PsnFNmdz)T(ob{!
zb+0d<TWQ}}P^VpDH(l>S3TNy7Icr+%lIxFLZke}IKk)P)^{mea=duds=sCv=TJzV*
zpFJSlA=BdbKvJSV?c7EI%{bkI1;TeGlzBMn>W64A^VECWS+sRwoJaWe-xXKZT)q8B
zsnw6k>y>EPpQCbz-xs|+Sd^#9_}0B_+b4s6&&8$0_{-xrD+D(C<p~6JO<Wkm%BW**
zv-D1ahNfUtwBM7kPvUkT5A^9jxyzkd{?`A7f1cv!`hGTHiz~A|GI?^Q>qQEQ>D8}P
zvC7lC)pJy}TWsBbgN4f$?%NdlN$;9u%con{?#lj&ZJwW&9OdNGv!lZCL=UT0y%|?(
zSm@$AS&8<)pWI9LoqA~B-Q(Zp2S2=MYyJPomYaw2zpAZJzvCd(nDFAfiz=VB-3#}3
z^M1<nPnqiS!tHgr+Dlg7w1O*Tb)K)5Zu1EYpHdb$fAWv3r>fouNpG<Haq)Tlid}{C
z<v#>1Jl@FgdwTduV_Ub}n%j@#5}hZ7^Vi;-q^z+aE{DTmW8NzM3%k}cwXRT3FW>rU
zzsP+}(Tyto!Rs9A55DKP(Wdj>{Yl8R5`L#62lWj-EhHLanQjy&r}E#7`uTCyF4K3j
zujzG(G^Fj+uDB&H@N@O1S%>RFnak`sHO%ZHp57J+4!<67(P`0vr5+FN>TQ18^|}AS
zBW3NMLK77dOyBOx-r3-O{qk(CFqwZoSL+_U@xNcj{%B#SlZ)A<&<|HNmV33d?#*?b
z6}R8m^~Q#Y?;^G>NU-f!KR44L>7e5JE{%=FcORAnxZF{in~=KY>^xc1#JE*o6N}x?
z9TEP1aED3adby`A9|bj(gN|_Dh?uf#yMng1e*B9U3$E7|>Nj7xW1ZgkPR%f4`bB}1
zNk7?+FY}&I)7d>$y*cWp++~RZhYzc{PtA?X*=(8bIB|(pHUHH5yCQpx+;1crr0o0f
z`F5On`OAPc^8;kWChYFK-xn)TmwS2Nh1!ov-VVXXuDVDEAG_}99lBX<;fjZe;kPEY
zPCDZ=<;Nj&*Y)|8ks1E22QsryzgW;GoFO02ug5V*C$ul!{qU>*+b0@6ntkg0;qyA*
zZ;3tFo4NnLd+1WuV={`lUEf#b-p;*rsdN|5)pLvH{#50Ee;}ZHmQK&K?}rbt9Wq~?
zuwd$)RJlb*?)j+5O`beg!!7+acS)<6lp4eHi$cLad<zT9rupVwEDbq!xI|4)%gjpe
zR2tK^dCiqJVNd^Yp3<FKv`@h5M)I-b%1HqQ3m^F;Yx^|)bmr;~2wS`=Jo2z*LB;J7
z9+|A!vi*X5-xS1`MXh6!S6wl8-2+A@bNejTSud5RZ9WjP<@sB|C9ftgJ74i`@!s~W
z6GI<OJG0}$+f@fvO?dX~)vaZl?Vh(iUif3~N{i=vo64h~a~UtqSK9Z;vF3;D3Kts_
z@0TrU)@wO)^?Zeeq%KU|t0hqp^s1^Mt#{)b`Qxtkw*?nl=<9Ev^!d!}DZhofgcchM
z3mdn{{r+FL(0Fp7RnXBX%cg`p5qW=7H05`?Udg$<DeWG@*S&W9nzbZjaz~pKW9E|0
z7f%*Xy|`ra&+j2>Qt1;WFvbQ{$hv(DuU?m!-gINKOUV3(=O--xrgCu2=9;(CVs2vl
zx@S5CHEMe_f7>s#rlYWismHCUPUyQw84LT<<5K=*aRDwTj7s)-HCPo)+915<!iHal
zuaEUzy_KVQ;t${Z^eg6z^=ga1T$%b@)`!{Rd`{cPv(pnJf;T_ju3*)?L$FOz;uz;<
zk3G+|cQ@q={NWVWUN>jg!d+IMbzW-!_K*LuuDmwm^2J$g{Qm2;ZlwzFnaG`7|9ro>
z-kn)Xj<4@MBT!pvRlDq7+VZ&VlQoLlr|fXr68>P;3T2+m`^PG?F1(nSn7xUAojAYf
zl`C6T?U;P!;(}kjc8m+><=aFtc{K(H8@xZ|UoWn^U6!XM*kF&RlI^q!+)aXRD;*!4
z@|Z2U?ux;!8CP06Uz$m3ZVq7TKW-58x%1oN_M9E*2mTbqRV`pDdlKJynR$-*{*7O5
z^u21EcIvTujaPXJ_tjEgSG9j`$M-}Q|9!r5zkZVb8)tP@<=7srzNeR--sHJ{{8DQ@
z&)wLr=~hQhDLKb&xi8W!{?Dsa`$oA<`0le8uFVm7@b7(^P~_LIE9$Srxm~wvdAVlp
z%@~>7XR5Aie@>bDv)Xw1ln9^fm%iB(F80S<X03NVxqHjZs<j>FU)CINEMlph-;wv9
z)5ZVy-rgR@MbGrOOaF&mDExUdll{k~q>xw@r=5Rh=$$*Ox8soT$rv%+el@ly{cE1A
zUZOiWBDP|K-JJ5R9S=WQ<w`m2Ix|;k+d;Q8j=Hr$NgJBQj8lHua2HqQy!>$Y=!zNo
z+9qKlT?V`<yHw1r-`nJ;@4K^mYclVf1Dg}Saow<z*`{>EiYcx5-Lt;u+myDyR{p^$
z7R@^Mm&jVP<~gY>dIxx97TGH<lhA$pcyg+TfNA=TuB{&3A-i?5TfTqrnl;;=sVn2$
z?M>(6x<&VX`g=-}ZBF&Ciytd*T4(>5_2$qfn~9h0TkraRe0QJskVb`9@#gO0O`p$4
zJZdmi`?+Pw3YKYGX1-aS;F41>$)mC`B{a<_ZR0$bMX&6YHZDo8cv4kW8**XK_n$6v
z56(~%O9`K9vWm|&guRXB_^(Q@ee>gNYrkEoV{A&%-=%UsqjTb%{-~5}rQa#$+Be0`
zCP<Zf`F=Ry{9`4LOiTUcpRSRStgE_QMAW}T7@n-)Q*2b<`E{B4pMIXa7Ykop5n19L
z>MFXp$gwP8-s*?ym*N|>%{6s+FQhG4qZ7}3cB=Y@AAfoN>Phuip84}3_43_aU1yeP
zDcUKU)J;CMUhmA8#8}_goxRTs<=-^8|9rUb(2Vf=<u{}c?hB8$ZTy*2s>dbpak~ER
zsp4XZ-I`aP{5Yf(emQ%RiRG^=Q+Jnrd|5SRs$$V~?xsTlI{s3x68Fsz{a^C`tiJn;
z&`(LK_rA^e`F!!-=UU&>l^Q=QZ(VVj(|WS8kZ{P!n-`aQI-S#7;vH}^vaY{iLS^*w
zL=o4$sXrSRo;F<S5f{ZEt;N+Xv2gy^Qv$d5o!GYj3jd6M?LBVIirTj)OfUZt938E>
zeXYgz6Eo(|4Nci1`=S1-+|M3tW2q*Hu~a-`wHRA`3^s)A`;l|H|6F~%_*O;+6DAHl
zY^Eb^AR=|7*2vf#<?yuBu<8)uaH;tdQWlvmJ!Nq#?bY0ur;ksZes!ArbuEn_8cd-p
zl-8Y6p5Wl&B;@0?G=S0YWWxfR?50^P98Q&)PgAq+Zp#Yg+xo8WZCPj8^;@?dZ%d7?
zUS9qCxZj5As&nt|SeCyle}3=%?)&e5*WKT#<fC-ZMd{+9A9HUT1#Q>!zqQGA`Bl}K
zVJ>Y)Uz_Y~jF<kY{?nDu@&V(q%S!XO9!VP7#~(W7s^O(KZ<2T2-A=v#Nk6u`@8US@
za<red>bs4EPJNWb*?(UDauOct{w$Wz6FRFZZfo27@3=?II=$$fn&B%=SNdLiepp#K
zeC_3zYuQToojPJ8FS*mm<ic@-j}Om0TQkRL?#+l5X`8m}JoBs|&D&z~8e{IwS!pZ%
zPN!&xM$WQvn|^$WX1Rf?+tPz4o=BIdRz2QSpys>O%qVT{!U=-=*p|A@ObvNvq<Sl6
zX`|uFDQDWg<YxLT^2<(|bj8wZZQ#k4R>Pjal_?==lkLou4uwpTUuXQp{RwAwkKnBK
z?jFI$+Qd%HlbMO0v#+VD<a=|NNBa28J9prTmTzQMui>Badc%j6LeV|C2d5<qMu>>@
zu08r_qi%%Jw648fxktYx-JBt%eSphwwS+Kh*klPwl^&G{Nj6uui^m-c9c7ce{%pFT
zS{alyy+%1!X{*QnL%KhBZ_3@=8mZF5z3!QirNyz<9)H2Ad!`=WvLjb-TeEJR&;gyB
zy<hY%KDv8eCTj7w2<y3febXZ!yjW>%W|-!&mRVyR+u2Ip5XX=;`*%HDzOUPTmmA~8
zu=lbXc?0&&yHzKZ`)}SB3mw+qjDi1mS)C5_T>A9queSfUc*U<=KDbfl4bL8>h&S`=
zXY1ea{w8`gjKO64xp$v66PBE*&d3NifAM|W*9wt>)o*Nm$GY;DZh7qSy`k^!b~(YN
z+HbeWuAP`AlP@y0(WrD;t$b)?Oz4#No3~z&`~BT?YjxT}@l>gA{o*DEzMPxC>ex!*
z*f$6MN4Lpy#$BJ*EL8R}@%9OwAB)!9wwqPQRg$768y1+h)^`1c+6$hC1*D@xmN-n@
zbSKPrKWB>Ajqg`90_OPSUte@QS0Tare7b!ddsn+!hS<7^^NO~VZExJR-TmfdWA^2&
zTOR(7IZ<jK^G<(`THS-I0_yj}Hik|sn*EVM!#?9&RcqAVkGIxN>|-yG^F32_Uhiz8
z|GW1~cG<k$Yx42%+2yaw7Vb#d@}ly~i|yx~nCw1(+dt<)qCor3{G9?XK3TVz+0UqG
z&KFx0d1P8sO`ZL{_^XqcuRRt!IP*{5?#b`mxeK<dt#|cYxFB_5$C6GfllQMv?dMC|
zynD;!vT5D6g-IR1MDo9Lb}XF!z-!Wr+Y|p59a{FKbES2-e(diI3uR^&YYU4bwgL@4
zbA{;{YuU3E=rvsE$cjonUZ7jJ#~Tw5kyE6(fs=$cY>UEFpBNA-8F(tfjdMQF=%
zwl82Q|HTq`J}N^+^2qu*>%TW{I=pt5+^M@SUoCCcwzzTP;PM?4%9oq_9h{{3$Zmb>
z%O_u6y}vl;(44H*LI)%=ru|*_L}J~VnqbexVmIx=>?;=STewv4$DZbM+#=;^&RKVs
zZMk|Z*WK-_$;F8K7cRUA`SR*^@6uQ1<?eU8+AUP}{g(1_Fetev^s)7WQl(nEF7pen
zxM|7f&)MbXFMcSx=dDR~TiuI4?Ez9un-%w+@{f5|&Od>1yMx4>3JwdVru$}Bd+uM(
z;9&i0&cMX8JUwASrw7XqwNFzHMn_~AWN__mn<uw^`*ZUg;T;q5|Lp$r^PhEHQb)ef
zm)8;80gk%=9%mmmyj|oYrmSFA^)^&@=M;h1xeXQD|4R7QT<UKNX|hT7nZSRLW6p~b
zwu0+i5-SsU>dH6em#n$@LCWEk(#<m){WvQbr$0Q-bX+JQe_!RtG<Npz!bRVg{aXIu
z(C(yt9C6AI9aZM?xK4V$=t8nbSvspofA7|Bdzfcj7hW8<WYw3P$*t>FueZIm=!Mbl
z?k#HFs}5`xv(wwMvq0_+r;??Sa`N$6JLXPT-#^2`-temUMn;Zn6~^jS_xWG{O=A4k
z?@@8>O6R1&O)DjM7jJ%-G=pb*h1G*dfx8~dEQGYTHOuBpt=)LVTkU!VYpmV79e?*d
zVm^Mn_wa+F+3R2GUrEUib<@7P>Gnpw(2o;J7jf4`sAuRmL<tEkysxM^sZGz7WmbK?
zrHJ##zrQCiTy#GnzD+IX#)hxIrf=7a-DOs2sG6lF#BX77T&L;jp?~463LKSeNk66b
z+<g{2Q|t7pRgG1pX|5}Cryb2(qZyFl7qUC%@7H~oA6#$UxM`*aO9eyjj@8-?YZq*o
z7R=VfJ^k0iRrNyspHu(jd`kKG;?$Y&@A>!nZl$moPj=w(TeCGQS8LXXr=>S<A6%Z1
zVVifAD?~^|O77=eFK>lD;m`YR{_}TSWlY+8G576@g57?sS<E>PN@7l^%dC0W5n?B;
zry#$e^<7idh8wE7c^Axgi?Tend%9CM<h}iNN#D{fGuImP2c$>z&z>jZcPa06(9{sQ
z6>AvJt&wS;s$}P!?G<-||M$t7^*wUy9-S)JzM@ro=ANwY{8<N1zG~K$(z=~zaw*Y>
zd-dD=BCD$C#<Vwc)=qBH|MKVW9UoB{!LDOM_DfoKZIsnh@wc#v+*!6*a>;FDzBh}_
ztla!BY+z|S6~h?ynrRE)-p2V`s*739s%?0edediJIp>PHKCh2UuTM44_v@{Z^-#(9
zC!<!R_}=HZ?_r<%J0)A@b3FQfkfq4z_2DUU4jJcoEA;1t7ZhH<nDZs#&^MN)?@PBY
ze_Y}JdEw%AsfOjhCq33EtqhsGxY{|{;PNr0U|+_mTlQ}%VEZIECu4{8v`PDNe_wm?
zr?=O+YS+Gn`##QCwED<eU9&0A*6x_Elk!t|`}@1xPk7fqbb0T0SnHtoTZ@-dcgo(A
zN$m|~zaaPLaB=-^t1ITOPq8as-xqMeUA(^SnbJ@BMPaF5r);mbx?LT`w=K-HY2o&B
zB|X9B`6);9ZcUxGrP-}DWAm0r4yC0llhhO+=+=D5xUBG3y7}1h1;GZ(_zg6kO737_
z<#ciNUl1N}sy~B~r)g5=50)C%BPNqO{9G^n_z|MfY7$fO^q%_U@WU6B1Y~v`J^NJi
zXVOEpwxeu$sry?(Og>*cYq0BQfia)l#3e7EY?l3Gc}Mx5W%9ejIS*Fm+uu=`9=|^I
zrgBrso{70r>`%>~^WXTVd`I@06WCg4NSm<nwEs-d=IS3Lh^6F29Qf9d=*Za2%peoZ
z`~{2Qrbr6`NN@g`p)@+zhE)d$m&@H3D0B7AY<;%w?%Lg}!#ApnTHIcJD{i^y{cTBs
zTjm9sI4C@v$m{9B^yHv6|BV(E77va_mjf&+3O|}Qi$*=o%Ca`yw$=81+B$ygt>)Kf
z`{m#Iw=FikN;=^6-8tW{zdiTn@$cVn|Jlud|NQrlj|+~tbR2*Dx5&Nh8~5yQOZV?6
zR##nIH+gH-Zm$qU-7fyUM{Ex1EM>K5cIG*lyn&_diScQnTqe1HFZt>o`ZuwBlVr44
z^VpM_l5Ah|V=-?Wm!r+$zaL+p5ZVy2VNP{aiSvhyI|d37jJgj#$_ePj{yWQW|Mz&7
zbk|C!kKd9u`lyCpU32opsj4Dn)%AfM-l|$fzG2sd`;V`Geew-YQTvJ&b1rF|Ub@61
z%;((+h4QosmQ&g`d!0P4AsMMLtwYPpV^zx;;rHsRrzW*d(g_W?TJ&_i@d=yG5YH?P
zSId)kMIUi(^w88(Tb+6|iFM_qj+1MvzdxB6tueE!CNN56n�budb@B6zTTt^ir+2
zG7b)no#m+MW$eAf^3<8GpPr}A{3u_&{mrS9(SIL(>z!TvE$3W7W@rlMQiblLS`)W8
z%`BXq&$o1iNT|b|j~r9@Qk}9FY`st%V77(#7U$chqJ=gu3_>_O7kYWHt^6sVZ)Ys(
z<@;cskfQJx^Elo2^EuP(PRe`v`0yNdEs&^Wn;G%pP19pfOV3d5-~(+b!6GrepByfw
z+7%1gXq9Z?^Uz$qaTDkDsFG@v8Oyfbf4Z$${UhItHd(t{_uANA?b#}o9J^T5J}&Or
zg(RIVJ+ViHUS8h${@lNZKI!U-CEvvNHO_r)|Fl%NcH+Cwc~fH9r*Yq8&Az!Lwtbd9
zJL4b2h?>GZyB9F*=S|4unR54mvHN<izqj`7`js~MYSZ<MQyONBm)9)b&%UR==6B}$
z`-{E{oPE}&rmDp+(RBFW!N=?6ZT3zS-K42=Olar9MH9L9ytu+++tcfNdzX5r-1Dgc
zEOX<T58N@*lsRAgf3kYS0g(e{%kMsQIxLcVV7^;P(XJk@%W+9Uf}2-A{4bno!WCrU
zP{Ctz{N2H7TiykZv)A@_D1_g8^}%1DUM_dPh|B#u_EXR5J@i<<NxO)-b)nXusVg<B
z9^Kbi&o1|H+0$e9A5A^6b>sK=xkuf8XRq!^36oBBTb$rI#r!Mlk5}Aw#WO8cZ8Uq=
z)L+@~H9`4a%rl;A&((6BOG0Iy3b_C5KauxM>Wton6XCK)deUa^?JU^wCfBYc;%20F
zaQI0swUy=T=It}P()oSP%Q;2<u}t=MUnC#NE>bRhFL++*T)}V6cJr4jPQSEov~*f4
zb|}eEIv_MZ_y5Jl2O4$mb3_VkuSt2A-n;0YXP9yJhuz#gd!&rF?bsW;<*u>MlDeCQ
z6=EFv3*?{lzr7Q^;==Vu%MZSPxA2C@p=EcLBzDis+qhtsT(kX~SMntfA8hQr?!f1;
z!7zNQ(f)?L_aoevSoxNw?G>{Xm^kO}@5~R~2Xs@y?fn_sA~dG`Qa@I}KI2}-!T43W
zTI-LL?E2<V{k^b!g2d0$2V8eNUvl@Udpe)|{oP6CyEm*kxlmHBr0mz6AMMPO|1n&9
zow*`+i^;aA&e9yk9J7eorZ4t+o_u}(Op~i}g;r%prBg-7>($f3wL@;bdsSd?#aib0
zL6gFTK_`!PeiCD8y}4rhgRIGEpA*lNEXcj$d0XZ#i^B}&^6I8q0mBn{J0Eek?Agh<
z?aD=8wUW=(pJwJCW4u*t{db0X@}J#*)~z>u%>L!orK~e4rKMHY&TknH-L>ZoigUj&
zHvf!eNJwx{lm9EPT~lT(@rn&|z7qQ4yUC)$ZO*y#3asB{zwZ2H*uQq$$s1XjnM!7x
zmCL&yys16g{JqxX!^;y(mk1x+__8?Sz~zWAg(UHpE-fpMtqETI^YEpW`p*^Ft|q({
zolwSVv-Dx@A&Z}93JV-8)E4$u$Q8*gIhe07&#<b;_=>Pd(WB@aPZ#IvZ3%yVU1PRs
zuh^gOZw`MrRn5D`wI?V_Npi965muFc-{#K(i&M^BaOF|#T{VeEIqI$X7rrgut7Lwd
zR7m{$wLj{ASB2Y4!zN$0+pQn(ubI<awtLsoV-20FZOR0M>bBQ@t1L6z-4T0CNOJRR
zdp6Gpb`4D?HLGJ!eA;y5$n$Tj<K{AUnlrpV{JC~F|KpQul(>wZ|GO;8Y+vMLd0@wf
zs(;qAKXq%ew)(H;dc)?kWWlO0E!$%Dudj<^->v@U*zeeT=1ma?mKj7xTc}!8n3op3
zJnzVONie>{JJfReD%KOLAEfR$YNj;rAj22g2v(1z?!R)ZXO^e!cy;tjq3Unt4?916
zTINx9HqWYDzwS(cnAU+yqB*lt?PT`2#~i6}zaAp-nj=C(Se)4^Nn5%}*J{_%s&I>z
z9Z}yME37hKy*P2i@=W!jrB)6z&Ky%?UG-+Q^(E<#b9P_Wj_YX34YUhp4&zwA{Ohyb
zA8qHVzID!U{~d6o|2vQWg}D*mIo4V`nV3hvmJWD1VW+Xcr@6CE%2&^c4V<MKR2tHH
zrNZO!_I>FdR%un69eb|5I=$3%$F?mOv<~xBmIU~1*;-lsC!@b`L6CZ8jis#3B`MDR
zsoisqpKob9H*e-^&FPC3+z&G!4gcZuc<r6QD5<PjA8NiTUR3m1Ue+4B<?ch}du*{E
zrhlF{PrpmB;ZWbZ3gfqzORt}p^|V*Ff^#?5xgU|UvQL<miQKDyb>X(cL)!^5`wsu>
zym438XT8xwsm*!Q#V$uz+k8H~mz(Wb)K+HMD?4~YjIZ8ll20=~pcSmS`%$IOEfFcX
zviUa5VnQ)zcI|!fb5cQ|o1%8eaZzsTJ+^#*7S$`r9JwW{w)o<ugjK3XPQUuLuW8+E
z&AGC&U-{$@o5nV;;(YTYA!EaAj=t)1yDy}Co2hM+CjI+DP1dIO_s=QwGv1qj`ohj{
zpG&9oUG`w)G(V=deBZG*1|0j7F6@3dOT>zM-PyLt#x+y^UX{wJKRt8b#J-Oc&v%?#
zJkNt+Z-9l0SAA>jq8ZMoS3WtiymsM7gJ-2zuA4{9mJ)F}>Gk&@lga0xpKns0blh!R
zACh00#T}Mj^3%uVf;<0(jI5Fg+M9covi;tre01ILL3l#dUygY)_mA5hNN6(2<L9!K
zz2Cn{^K0yxGWCjUm(!d-xOP@fNl?^{kckzEwt2ewD1ZCf({DvM^lr%bS%1Hk_IW$^
zyW~EbAA3K<zi(Ez!qzfFT0)7Z*@rR!mz|LCAVDA`B_!i(Ld1h(EDVD4m?mN|+ZZvT
z7*bh~s_&na6<m_5@26mBVWIDqSyWu2U}$D)sgK)UQ$u5vE2`hdWampCpY*q`-j%ug
zsA!?$ru+Q)0tYs&6ILyIlWoc0{(ya!)JYeWAO%h9d-409f8kmAFvCqT_jL8f2<0<h
ze9FT^!{@Ggw{GUNYffCUEt@<gRJU9|6p<)Ye@Ie)jo3Zc#JL=boYJm6K^B@*rg&NC
zPI2N+Z@qMn<GJIBN0maS8b$sE-;)dw&~nkP5z%6s5agu#-gSeMx41)8hU={fuD?|l
z3Et9DT_k=>Q#Yw8<yglO#noGtEEg(wsC`kM%kofFN|W=W>9K`NgpWBGsWP*%T%6K#
zsG_%N(xDcQle}yz6?YvnP)!Y&SorBeTbP8!3aye2Vsms&5-h#86bOV%s;_x&VHhK=
zBf|7EH^gxDB{j|>Hh-aaQyvO`+jK=X<+RsiHm8jCCin8g>Jz6dbk7iF?PfZmEgbOS
zitg&5Ftd{NrdxZN&N4Dv%ASx7ZM&meB`FX*i<RHXYLSDfWJjXt@y>VGf<h$DvYE7<
zoZBoV+F>Q&#ONdI&Zuq8zi<lkdWX<MC3;@l9M{iS%Xejo!)EV?s>jwIxKZq*^_WpB
zuwnA~rb-2-WHq+{20aUp_r;1F{s%N?oI8HP%#+2_mYws@2Jrw92JsMAL7!=n%s~gb
zn^+ER+~_RLpDPh#agO1_Lh}blwG$S1W(zTEDPQc*Ui9;zs_P>&hG`}%1t;b>s!p<F
z&|PG<Br}jfC6pzJ_sC`!gI4xai%PFptve#hASbKKvPjONVcMeaIh|sRRv-5;JqvK@
zX;ImurpBAvl*Z&+;3u?6<HWR%yw>*4uEQJ+X*?!fDFKX`lNK^|z;qrhs?B=>Ia
zV(mpsud8%q=kfJ#lHu@h^W69#dSV>+#GpmbFW!{7tJ%(=WO4DXN~y-rInT~YOr5pQ
zRmIEK-)D{KP3D`2U8>}wBOXovHQB*(9_zFhx}B=$4`vl5t2o|R;4j?Td7kNa>zjKQ
zc>dZtZtsfY+mTfibY#)u-Og!e0}q-CM2O56JeI+KS@DPFscA~hg0b<<-0RKk7jCaO
z&){0Qpvv)4+m0)<RGGhKv;{rxbal>e@7%g<x3ILOLFDR}OeOwqmS;I8EimF`X51f=
zxS*Xe;#P{da+GhGZrjZ_LTzuE|FR`4Kl@O>Z{wtUj)oGZOsUtVaVUN(VYBR6n(Jm-
z!?$tjgcd=L$0t54%4Jio`qbH+R(xXFZcipP>1Kw7lY?a1f}?mZB=gUj>3mk~sKLv5
zC!00cXGxT4mNO+BHNVd@?~MNQ?#r(~E^%m>`snI|mo?iCeO}%n`P5csh0cW|shtnk
z>!b=6bviE4xp?wa&=L`i+TU*<JWaB*)D1g)@bu&5&DW34Uw!c8)gOy0g5vs4oj&>E
zcK(01XY1|b#G8+AzFA{xvA1&fqr015zUe=Bv8FIgr)`O*d-HpC^=5VF(0ev++b<t1
z|9t)P!H<obY&X<|`h+>VyxZrxWsBK2$tMEeesgjhX}RLAvqqvU>tWfMT;->8ZpavK
zoMe}JVXo1dtr5?+Kl~J~7BTx3<C2~84P%5Ysxxfws2^_T`?=a6)A`nNsS^45lNP=I
z5Px^|fu9fT_tx0`TWm2uPE6z5`t#?in;$Qfv6;KhO}f5#UH%T1h4qFu>erwDy3~4Q
zy4~NYVr5@9trv_`QQx?gudc52(EWqEufD4KnPISFhR-kY&9nd26dKMrueW^jqa3@)
zPahxbsECg_x4!%6!|vBtOYEY*-Cw&-?|-tDwe~+7oB8qU`Ic|~{8P=TP*GS}WL@*C
zY(LpcUkv4I`LtepMQeXzocL>jUqQLbSI!sOKOX*{{BPE?ywAVC>ijt>aeve9?j>ta
zZp}Sc>CnDw>O;oTzf;rC?{?hX`;mFZx5TswvJ(HLeS)ulepzHW|K;X?DaIFHX#AdZ
z`|69|ibk5rpY~2$l^^oQ=IA}WDeu27UcZN@Hb8vu`Tl>`uc^x0)x1c4Yq)#YIq%6&
zOARfbd`fOey|G|TgxI+!swdebuf5(kb=uEecArJh+?JUAD{y%|_fMwcE}=&S{<*(=
zrb|BlZB=urYLo7h`@4_-`c=I=`gmmb=lYX{@A_5dulc6m6zK5fZOPUBe=H*WLacP|
z_2(yMY<aCdH<;bbB5}@fF?BBKBzwu<mD#ubV;{xD&xzO9ySM0L#a1mb<9Bw;CO^7b
z?jG8letGrFEK8dSdzqJ3FFxr1c(u}fqt85T?5mFRROo$NsNY>9b!yQ&!`$Aa=0!$(
znXA18x1Y&+uD9)YiCOoHoK4$hH{5z6l%31%c-`7#`<IyhnFZUv+lDR^;&bYkteuwB
zuUR`SsNeGM?kmUkHh+%JdMqdWFX#WZ<2usbljSCV*O~iu#@_UmFAQtXU#jf?`8#@-
zz2Et1&Y@1m`)f{wuU_@q&G~T9`sVxh{+tWh_)X9-^h}<fzI(m&v|SO`E6*HJh`2hn
z#{BOF^*bfS%i>Z)7#^6Zybis#)NWC9^whv@{f^s0GX;9GVz=tnGh{_;mjwOYFwtN~
z)!d?m`Q}2));312|7BPD<uHqi`Cq3yLDRPU-SKkzlWj`5_qT>iHi!1V7dUz)cXRA=
z#k3n2Qj+$@X(rCOvC4aT(*3-Wm0zYA`0cEF5wK<JruWw#GvDj$ewsBiKK1E_3Ced$
zN+V_d6cqmzT&E}SqHFu-pYn@-Y@Tt);PQp+C61Ci<NCyp=1h%0GvTIxhx$+3pyn5^
zlL}4e_Ua$~`Qy{4w-uk<o2TV1{8|*a=e+*%td`ix21+;2mNw6Lsh0NoNX=x4%l?nP
zRul$0nDQ>l>wYnJ-TLFF?WgtXS^EAd*%h(wQ@2S`o!Pp(HZ$)x+*tT(!quy(y}oVw
z_r$xuUw-7=oPI;d_?kz>O_jzp?T<euTGSN!ZM>)dTVod6!5dB+gbc2(Tpjkxy`SZo
z^s+B}=a+GxDLOi{B1$A(SHNR#kJriTJtf+U)%%Z3{(03@Lifmw$zLp<A2Zk-*DGGG
zWBl&f(px&(SCk7);{4Q;Hl-Quk4seh@&1h4ocKQdX%DK)qT_`ZajM22cU|>xk<4u=
z@omiF`_}PpR@PJdD)@Vsf93Vo?#n-azGB{Wpyg=g##wu2pVN~{IUjW*tZU8gS>Iz{
zRP-$PP-q$3^>e4iOx^JA*`C)ToBc1IUH(qZuB7Sl-HI~CpGWl{pPsF>+1jr5lh>a4
zefhc5&fa)XbMNd<Qx1;59&=(68@DO-&w9pE@M%-Tob%%0k8du?*iadL*k$Q1F5yIR
zpB&3y>oz*&x)#JQ7qn_-JysSOv++20&hmNtj|dzW4}bi$KJ=e?-fYbh?!{i6UloL`
znonFTDPxm0>s{=l{&2VQiED=aT03V?Nx92Vx?HC<O8USS=?ClN>-sY`*wk!HE&jMj
z%tVUm%FpoPIM#-U=kj?!SMPfE&gJOk*++jrN?!l=Y-+O;>q4p4trEM|wN>R@O5GiP
zw5V?Lj01B^B4>a1(BD{p;{(UGS7y-<>nyLlDpQQstKZH!_4O|^r@Yw#JH;iYT5n4I
z_byqvzIV>usdHAYJ!*b-`s<Ho$0vuse#+!MrTg7Q*=sAiOEWg~t^K?9-p2i*zq{k!
z>TfGEGrhL1yEJFRwzq#DulsxW-=gWewFC3gckQ~j%hO<M;Plxm19I1SxvpD!YDUQ>
z*2(Km3kBx!FE3dU?Q^WuMtU0ed9~R6+OgLcM1TKQy=7Wfl(zY_b6cd&lqQ<aYEyX|
z@wMvszq;@>*Ie|&PaoQze6>8$)J&;A|M;r@-QW0scgOj0ANn13ZK3wHqizLUZ=Lze
zuK#-`?4r*o{G~QFdG+kjAF<K(Yfo*BDL*c_>)1o-#~Ig?pX%xUm7cgs=KkG(X~+Ks
zzF&6S>)o0jqa7V9Zbg>2HMlK#6A<gO`&<9-?zow{ZDH3|UVeVr?DZA9FHg$e=6@I8
z>vQQt_3^#xonPMwtiHMM%AKe+Hr`9NP4#|d<Qmlb@yOeZc_p_`wZ>b$ck@3f+nhCZ
zr<UlG(}5?yT{v`pwe^nKYTtV++x|ptT_P1NvSiAgnQ2)oPP{XDc$Gb3>6K64SFTkK
ziH(~tbun1;uH(V?J_q0CE7rNpW?zuQ{@d@sPqp8-S^HLrJ>ZSH^~P&;%**RO!r#n$
zS0yZZs_vt=GRS+?o7r0RzTfpYde`h~&(h-xU1l5BH(}M^BcC7TSlP|`KQsPC#SA-@
zy>;I=X&hVq^U;HwKT1E&{$UWiSNQ#b^hVpd{`H~f^3;w#da%7|>!$fX{J)-xyYXJO
zKm6#ykIK5A)wc#(W^ei19#TL3)BC-X7|WOS7yn!A_3P8+?KVBnKW|R@@Oe{3;pRuj
z6Hb5H>-?{6{@ZK)`QJ-qez_G!O8XyM{qgA^rT2RtUD#!^DJ^5E{LO~e)l!x--_JRG
z>`ecjV_XM%_8fKd?%yMyrS9%HUEG|<vES*Y{wc-DKU5^nUii$xG*7MHwoP_vhqsjC
zoF~y1W$&0zr>8CMUHd~S`*YmelP3+|M=G+rKTe&P{lD_l;gc^f*V|P8%j&zkg{ysi
zSNH7GADhnl+v|lNy?XLZ&Ah@|!`O9V^89l1&)yf?dysqIg4yTv`25Z>EPu(h`_bju
zuS=zzi|+IOI)1ch(~P*o2YyNyy?Xm2eBol5KeOzO@4Njh>p%GTY0H*HB|dAOe^Q)a
zY0(<8pCRSKzpvkqet8so^V!GW-SNxQ5A;t~oNT4^h);LX$Ax`1YY!%ic<sLUfF*3U
zpWJypJ#IyI|Kdd#o8KKj&9rHc%=u@hA4%{0_x#DV{rRgm3hLhdmE67j^65hfbFQ9a
zmePH&?{&W8f40|&{K@XqUR;pWwvnG~B%#EVXiz4?pl#FHH9f{t#AQZwTTHXy!9R@8
z&ij5d{@=g8{{P$K=F%U(GSAvSODgMGx%I63*U#o_f3Du1S{HA;|E7_BlJ=SNi=AxZ
z7MA>&^z8kW_`83~ZvS~bhfBF_b7IPy3CkO88af{^<{i>F%y3AZq4lsV69>Q7j+!W*
z?H?*)1pF)3@w{z0$aVTt3g4AC4SfxDCyjPYh}bY=*T(B!YDR}97%2HXdAWSjB90rg
zo*XwUXE?S-Uo-l&fO32Gy@=U7l}ZN>&CZ*@)W;-QFvqYv{CY&3;4A}!nA*JU;%@P>
z|6l)}zi;)+9~pI(-vk(VeH)gYXVfUU)Edx#Bh*==EWttJ{HhN>`ix{buFB~tTPI4r
zZ_M1y^;)7gS|Yao$I+Tc22Va0cUka8OlJ~TUfte!{6B->yqS;mKU7FZMJgyBiJ7YI
z(7aPG<<BP}!;@20_B>po+S7eV!D6$SMuRfPTN_4cVg6*-i8rftuU%gBe52je`jjK9
zpFR|r>Bd>C(JyuA>CNVTqbSk+wwElEHdQX;&=p=gMcBJ5>yyt_flF#P-nCx2Usv(<
z*U8ttvz{hq9%Da{we!K6#L(I253c$=_u!R0N6~9mS8SZzX5XFb%`dv2^@GjXCt6q7
zetI@;^`2^aChX-$-n3lV&3f{ZvXXNj`?cw9U`i3(H7oz!+6M>y|21t-6<r`5x-LnC
zFVvL(!p97a`A?r;Ouf~yZ)UIZt_w$2-?{TKVaAu8yAL1gH556|`Labo_+rW3-Vlz=
zZEHG$8+-&mHrSh`zhBa(lD5%L;Z*N)#XFl)wcIx?oA6jCus?9pA-6YOn(KF-S)42M
zKl$+2=2ua>oR_j651z*67?-@$Y2`vKQQ0(+qNeo@|0ecryRyY=wwqs4X7{eYVJhnC
z_V(3*#(xg4^*H6f;+3C>UAF6_U2`?A&OLGBROhF}NBVn}US5j7Tl{C+^CPpKTga_#
zi@#i4v2nI~*t+ho_fPz~cHD2f?Y*o=S2t|dN_*${Pjq5zUCfm=-8cJXYFFIUSYNp9
z-P_tTr?>j;`}MZ2$~k?B^RwpcQ`h&}JbXD@T)+M==d>ET`Df*?3C^-M-hBHUPhb6O
zj*jEGeFx08&YgK@mhTh`lgjyr7oR<~{kP+e<NQI(<x4WxKVwU)j!!GO%UiMK{u5XK
z(3t<5<a6wQ-HHF!vToIhRc|=AE<d=f^g+rd?%gRyzE+Nl)vOO*+k3Y&sjTVO-gC;g
ziWmNWcGKU#Zl#ZyyukZqJ4H?36@QScF~4_L?B2!;<`L1wj9hze>vw+VdwTWZzP<dW
z|MIS^cl~Gm@!XNvWJB?w@@34wRh52D|C76nY4hhDWpjjfy^dbiw%6uK`}O@cb)U0S
zpM04hcj*rI>@SgWN#AD){%H4nBR*fyYD)O`32$RJtbO0K_;}E*rE}H>{t4FUjPA@h
zHCJQW-WZOfD~fqHn?+o+Tz6=r^nn#W%A-D<o+%x(;kT>xn)AIIVoopK?SEKh<MO9&
zGa_e)q~-olcZr#~l`%0bcSp45%8&W$Q|c_UoB!%RNauZA^`)&ozJ2N+VVjf<FJ1n*
z7)|z@R{t*Tvrej<*GiixuSHXyN)#)p-xoe;dHu~Nh8uPIEc<_6pDbVZZ|ydl+65_n
zxp@=brsqyr*nTdh?_w#7yIS#&=b?MfoRs-|bh71Jtu+U4<mLCQU8^p$U8&N~;?tyb
zhO|sG?hi6AfAlG>Kb-%awXk~XoKtKomz)U~y3OeSR{lv}oqg@+PoYnjpAX~jnE349
z`FGp?b6>pDG>1#n?ZXzaCk59fpXfBriD9>_2rzfLV$1QeuS8aAjdMyhdqy&|5^u~2
zDVw5a>)3c>olmT|_58@CHyc?3wcSo#51ZX-Tz4p?qU6itkNVrS|FC;+yZP+fiQ2hu
z16OTx-&kstb9(Da|Jf3^MI)QLZ{O3jyleEYcWYVA;ffvC{y+TQTz^ijxUS1)g{`(}
zxmlaQ=RZ?pero;v`sevgk+s|lgYG!@tTQ+L-6i<0YvMP7#iqjh*Ym8@Y@SxcFKJb@
z`nT7@E3t-G?8IK42zjg9zEk_S9mCrbB5!otc5AoYWS3kky<_iEj`F1&sy#05^SBq?
zCDyXbc1cF;qc_?ud4?VLojblomgH+biq$-1$Ck-^?zdN9-nZ>bBizsbXOurA?ZK6r
zmjc}}o|=~e+5v84YK%787!j13rXQ4_U!q{BV4x7B@0pjDuK*hD4AS>>iMB~GFf%qt
zF|{-_OEyeOHcPZHvP?3vFiSEwG&VI%GcvLxtRhxFII}91%gEG3KRnt{!C1j47SC9w
znIX!4bJHu~IsS<<?H})NcMIHXW}1EVUiIauwNh*Mx@W()UA^sA^x0XaSz@|{>Vo1i
z9W4_)1iq_;usp1v_owRfN%Qm7>-_E&pE>jIFSqgPtltIteaG4=7~Vff+|zyhgIrpY
z-n?VKEBNFeu-0@;H}p??dgg{&ZP#~OvGqxIT=@@_BTwek8}S@JsdT?uvq|`anB6qj
z-mjMzSp<sRz2R**M<@H*G4H}TO4)y}d0U)2Ahap(<vs5tu4$8wzWXTGxH+X$GS2I9
zPUIya{nUbWIfrcYQw!(iWbx}iEojt9T{8XCjsUa$Q>Sz7aM@;Z$Xi8k>e{&HOO^B{
zFD?HV#2J2K+x%}qlHsSmxmB-hTAk`XuX<%yc1k4ko{+=55wX+ezTeB1H|1&F=OAJ8
zlk@g`6Kb8kQS90@Tip+rt#*Wd(>m%Mdq?fJz~iNI?<UlCehHE<Pu%h*`{CQH2HOpy
zWe3*X;FV4+f77{kBk#7O$8K=nPBhNx)7`+i?Z~$qg4T)YZ<-8mh*Uq;@Hu36L;QE5
z&bQ+SCEA$_6!;Fg$TWSN(coOD!gu(g1=Hk0e!s&XE!fl_hQ#!&w2)1IBr~VKyFga&
zc<2tX@?@!X$8~4)-8N({Keg|8>p%W)k7FO4e{vwcnW=vBe=(MN{@D@|)5Ifm6b)+9
z5<2bc*!jGcmi@S=A8PaFw?~Hfy-UtPXSw(Fdj~%Iz2eHF-uIJ(PJi3y@3C=y?fi+4
ze*LynU26AvuIgE>ntyAiOnd%&(Uz5M?>`5niQW%)s-F4YJ1DN{efGk+kFBGtKAtY`
z`%<*!{p(A!57k9iS)cmu`||Pi2e~3G`!4)1Jv2iLGp`hvBo>ua6s4wd85&ubS(<aH
Ks=E5SaRC5PVtTs(
literal 74614
zcmY!laB<T$)HCH$ee&V$4=zIk1p|frq%1BQ8-2IToRZWceYc#%l2n(}<ouLWeV^34
z^pXq(LvsrYJ3Fr8lA_eaT&{{+y%D!3-8SQ?eINcJ&|26~$@Q+B`An1PezSeA&-GEC
z%^tPv`mCugM<-n?-M>$P@z$jB$t9C|Dj6LGEG!y?YPc%|4$I3uVfuH>@haQ;u!E=L
zD-RT_RcJL|H9H|RO~i4E*R-jb8*bHb-d~w#T=ORHOI`N2wXd}PZj^t$?O%BB)|ZP8
znoO3Bsg135NS!k^Q)`XUIqxtoNf*iVNzeW6%0CO-p1pT}my1;Uoc#wHJ1QTB?1@|$
zs`Oxzs!?&*BZWtq2TwU!$o|oKR4B7J^Wt&7kP}9G8n1rHQGGS{wwIb3V}GRN^f@No
zQ&eq=yYiRI={&BCn7OSzz4W{B>x3hRuZlh|wCGKXU(GiCU5m%M-}_c-?>VY_Z^mBx
zbIt~r`HozY+Wxfu{=Gk<vEftmzQpy;I+AZ!SNB={UHp6H^nG7l4h8Y}Ze8sj+dJ7f
zl0TuNrg7hn51S;){qO$KP4J9da**-SlxArQi->2A(khzv{<9yL{U~8Nb+gbWIdxLw
zoSvCeGR}W35LH#;S7fn}D3@X{FVy*VI78jlV%0;&>1VH<zRLV|V)jE>fj<GJ%?o*j
zPNpCKyRa^Pdse);d(6ET3-`TWBRtnaGh(vlFZo#C#rH#(I4N2w{nvS&61gk<>Rk@U
zmB)TvOwyYCuzjU{-?{AuKM(g*<(^??w#^jUXUzVWqq<V)RzOODyib3^US?*Q6;pUC
zr(L@J=F%*Ye4pio`BHwLH1rPNzFoc}eCE3570N9M%yJgaY?8IMA-kjndCC_mDy(?d
z!OT2yp(*<&q2nCYd-v}?RC)K|`-?s*yt8xfu561{du{nT*K4X}0=MqNGs|D!%wMoV
zdH+M6=1mJGruG%FMLzpddoOc?mM7QyB`TW!t1iCv6sz8&@>J8Q;zfWt@4EZ`U)>%H
z+-q+xd|q_Y=ezap2ZB7-hgbY&Y?||I=H(YpuXZf1>i*@wiX+{BbJ#(<8=O1ycu%I@
z@H-pvz4w@;L5o~L#f_^wX0@=*JiPZh*ISQWM$)w#Q<e!fbsT@O)iUYKKc(<5y{o>k
z&6_M*_~UyhXUB%L=S&YXS<73VEWdHf$D>T@;Zwb`7lrytdmcPkAj6?saK`4+t;dDU
z&FpKdo9<W$xZE|)iDG=X@7~L<1xf3x+fGbYN!>IhVU=lM{;Q9c8jg-#-jgy*-NROy
zUQkuK#Wpu}Q-qC#%#pdxEEZg`S_!VE?4>dVIh^Ngmx&|>m-BCV6PN2+c--q|$E~HN
zN^cEz^lT8ku}I*Tu!`1Z^|bfC+Z~-4ET7nGTl&2A?Nt)J@J#T#=-t`_1r0K5YWnY-
zdoH*yW%uE2A0@tPCTzGAB9;2l^sp<J;1iW4JIZvdgP$IH<kr=<aN=jtJ$!xNUnrDw
zt5y}VegAXnsH^|StSVn!r(~_G4l_5H$~?9$=Un#u_uF#Ankc1L+h&BjO#D8pWlFkR
z&h|$SwqJhn@TS$Q4T~C@xNA;&#aT$(oURHj{2=!4u6=RL5zEP~F>>o)?o7WmubH>#
z_whHnt|vd)D1H+=a&Oz`jekX@o_)zbI5V5I^C$DAX?|z3o}Xko;LW-2z%rdBrS2Z5
z7a6yg<tC;*wYuuG=A+%tGjFbZRh)5Kd&e{G{z>c4&zx4&;N~`I$~UFXNFI(EkDrx)
zUEnV{JL>jaM}e1&;st)YMg3~Nw=EE#sk^E_;>)Q>{Rgv}RqiCkL<KQE6_>dGw)1jj
zM}&ptd=Zx!H;pV!oIl(*iQ<*|?!MvgZqNLi9H%0hpNhL}>$n-eWS&iVh}{FJEz*lu
zSudHt(<ghqgJ9{TvM0>5OIWW)Y;U%l99(S?Jk`<og6nBBZn1;m%x0FBWnC3Z<mOEa
zdGTXa?cLjKkB(pPW1OtXXswuV|5f3u|NJev&WbKF@9w0xZ}$06aDjOyr|^*j+fO=e
z=P;EmICl8m?d$6cCm5bgk~ZD(@2cQ+sl^?S6N~hUxF>WgwW~j8xx4VuQSbDgmq&{E
zPZ)=#*r`Tso+&fw;DW8&3>DsR%`WPHrnrb{hNoKVj*Q0lwQJXFR@a8R-p^-$dwNg)
z%T=y*v)`%~&%4hp{Dp^SPPz^Mk9iNTB))hj=y|XtVDItyM|Pc%5Z}i+m0PM<=gZ|V
zVQ0O{O1@2V9cE=k*<bl~+N+8CyziXYaufD-jqk%3v}CofUbnaWHXHxi%zdVleRHhW
zycGJLru{gY!FQ!1V?@A@#i>v0*>t*jep)InQ+H~S3vYYB@}%2q-Mtd83Jz{+v}#Zn
z$&G$?+K=JM3*)}3w!G%Y#{2nB=k2mEUL9(Y5!Gt4)a^+0qo;4@^6mXH#j;|f%W2`g
zaa-m+IKA!L*4ySAJ~XllhFAZJDSB9Q%isd5`n|0y?n^#CZd^2f`^hW&Z?0M+UA$rM
zQ=t>5ndURdEtGJNoaeka;6i=Fgg?vU|1rN@y1R=jH7^BP4W{O$fGSQybE0ZcGc!Zv
z8gy;M?Y!G=0=4hM4ft6#7DN^1$6V5Pls=Nn#Zs0e>L|$0=(<klkjTmjQ?7h@ZC$<C
zqjl2LLj_6e-K;lnF19T8ct5*+FGJl1b&fyWdrIE^HT#}+{l;O&u4akq&r91al^Q=R
z^E#58Kd<)Nd(MZG^*`J1-TqmSRY^dwWI=)YWwZ5OYmIlS{k<Eaz_*)e@B6RWyHgVk
z+!9sZ^sPQQ<<Wuq;I-zJit53KnE&tckds<#ky>EJ{V=D?!Z>|@bp9*OzCT&)w+m%{
zNJr|gu9`b#rC7rLs9F2pp5nT9=U=K=y;|nh6K8j=?3%mgY}Nej+MB(4mBZ@fSwem$
ziELYWcZ$iksGNf0a{_(J)f;E0UfPnpeK}*(ml?jEGjCcSnA#G5uyV<S;No&ALy^ZB
z1wVDJm$!fLzM!J>ZF*2zw@R(a!rk`MrF@cH4s{0?+grv8uj@F(P`IM^boYwoA0D}u
zgzMg}U8Jn(Tab6QeQD0+&^Wu8uwbvkfMBf+tCn4fj}6XRnC+z-<GXU}YR#=?rm0Jg
z*X8#8xpZjr&c%ft9)A|?WbRh%oi}m!yhGj3-=!&^{=QvZxMcPf&%hH(|HYL&EX^wy
z*qBrXa8GCdy7qMQq}Ou6-9Fnk#XVmdx<zb9{B!=**G`C4&U{((f_JLut=ijeQQiSk
zX@R-7YUYaEdYkaz!-j@=r}X$Nl4k77s4{HsdicbW;r`_p*4$nVkp|HMf%}=>)c*M!
z5oF&Z!6(_JAt|Lc*OR}eA~~z6$<-`mw{FU%8`q4_Zl3nK^_?llI<Kkk&OHf~`<l6V
zZTY9&k^7~NBurqOYAdMZ=JU&dNoT&#V~cP1%gvq(^Guy}g@04-F(Vc;4!0JGirwqh
z6bI}$xBJ<~jSmuh*G-xmn;O(uzcO*}`~RD?y0Y>enC#Z^n}>7={W>N2rNp!$oTJz<
zjyGkE)Gw*oTKTh9socD=b!XJ0b#qI(qa*h#i{0LH;@C5_PJyR+O7rYHGq0cVdvSHi
zzjw2=C!bD<-ZI^UjfMYIm%X}(ZEsa{-_AD1uPoAR2T%QST%kVkzNCZ;<8ck4hnp8C
zEay3L!0mv@akdXdYd@U(q_IFqhwG}Z0RInFfx08Y3U>ExTpo8+SMw)xC?Dl%y|64>
z!S2V0jUV3a;g8iZ3~;-8aF?aV;|{lV2Q3WdD;UHs)wsHNC!5`Q4u4a9-b9PUM>)Hn
z?QJyNaQf-LnB7KeO|CyySn$u8wSiB2S^B<7TUPHeoD{ZuWtf}O*$EBjZyPlyY)`v!
z;JHq`)|4o(!`Hn`6POmUyM;z&pVT|qof7Dgy~jxHYk*<1pXc28=FX(MzYE_<+-b0w
z_gLMFw_Rh|Va4CyT_!$XdpKmW(1laF8C$Pjy?pm*M1w@_xnkG&$IEvA*5JN;)$W>2
zz($n<2_xIHds&0DtzAra?tkDmXOTq}U*N0km?GuNoBp~p+`hC^vQ#<eK7&t#hUm7{
zzLN3J16ub>bidTnl<#}~aNjEV-I+xe3yK1`LM_A+7$yYlNS|#vg-xzo-=We&*3+@w
zli$80`&z>EKE0NC4jVaQ>_0K&+|;WJ_ggk6>HlYid$p|LhiY|eip;-Qv7Iuy@%QPg
z8NCjgW&cf2Th_>Yo*CqjYi085Gr!r_sm#^&drP*t+z)Hnuxk6hwI<R_y=%pHq$M%0
zt*cYgcPf-H+FF|RU5F{#ckh`=xzW0_BWEu2WtFM9_2$}-*Q=sVM{b%_s(t$AU1=HL
z=Ns0Zo1EmfeA}Xh9sBOxtEu{3_>FymoX_f;4{ro(oNW|5_K<ODr(<Tq_p-ddRjGHy
znB^o7KaqT!SF~w&`7}}4nN716D!%r)xKV3f5=X$1BMEjjK@V0`X)OEqFjCJ<xZz05
zeoKkEh0%QbB+Y7b*t&lmdY8bi$9$;O<Sl2?ZzCn+_&H4WTqcnlUI=}7EPHr^v)V-w
z2l-`g;ot8yPEyRbEHzW`KXK$|!6b%O+5NBQcP}_-_}Tu}bNv~UQ@4gEvi(_frPoeC
zo9kUcU~Qh6(b=gEefv5TMdmF_Dr*d17cuFW#5+HkHh0f6TQ=xf7bQIs6*wXxEq^$3
zYUG2vX76~9Ox)e1Ah>%^MxFgLJ(gK*f3Lfps<IXA@_SW!Z{GZgOifOe|7U2b{}Gv=
zDtPDU-08QKcPR#)s(+LD_MTnU(oGt({xkkmO!c_&u+;L&_pK);Pq=^oZ_<SO=D%Kf
zPAPkE@A|7ETR%08ZR^jpI800xz8h=*=Zk#Fio7h$D$K};s4C3Z+zh!2dmHiE|Bjo$
z|9krng@%0fTlxLlLv<w&U#EZ|sY^^<t7N)*^4}LERm)bNZM<T?Z}+{M*3vwjp}&3z
zFn)SdCAWX(+?leEYl=FWuksf<t*YBNS?t_p;q-4xMM@&<jnC)kT+N=IAo6=-@umv%
znWnEJFD=<KWpn7Pb4*VZZLH6G?VYvD?yK6nE2@+GzNzOai~C=)XbYVf_I9$@n#$br
zJ_&*I{{7$Y+uK{!R_$2-WTUvW?a|a?Q(NoGQs&lZm7QnwDzhr5FM5<3D49C-#4NWd
zB_H=a^N#y$^z;OiQ2Az#wP%hCo9nyz)Gd4*e>72Gn_b_J`=77R4vD&8m14X1@ZpOJ
zUl<!c|5lGpOL$_vUq0tVRq7|@0w<C784h(V`%Z~nQml^Dy4&M(FH(2=56^}qF`kwi
zbI(js5SqmJ`Q|yX1#a7dPV6~5pY3*ZA7dJitD^LaKmV4Uy6q&WWFf;LdeS{_&f43X
zf(>+JuAd6%Vwzx?f9rGHwmqiof(aHgHi>ZGo59++clw61`!k*&{Fm+Y;Ay+tY_^*V
zBdlW9o%#Iy@Zs<5`F{Opz0z3CMP}trept3^uMk63c=I>I<Z3q^=A2Wjl_pQpchFb4
z(WF1wY<k+ql}~jeZrl2&RBdc`^LhJjpRdu(soKc}H<y<2zS^|xQWJyZ!o7O~d5;^d
z+^$s;>91h1Ga^-0QnY&Nl1DbiN?XMqRu)Y?;nR>Y>q+M9a%t<d<sJ)yYI&Yd_lzr-
z5fIbfow!3Y;KQ>dmea-?j{YuK(4qK1DS!KoNkxU99j?!8u`u0pZql6#=kLvtYKs53
zB{<<+gNw$Zj||NJP8KA&?6v4CIR5<D3FUhY{RxL(md;^ZcJaW%70HX3)w~{M9Lc_x
zo-|RsRoUlW!^!T3mCVoXiyzstHuTSv{cr02%-^|QN<I7P&FH+I*{|I`G2OX!u<KyB
zWt{1@tj8bPee2GsTkMSStW%ZGPMUav<rin4t$%9W+1WV@A6d??lPSKzUl10yHb*Y`
zqL;6P%x6x`N&JiIe3;jt@CrP?Jam&$E>jnqc{VH8FU`U&VKV%Ad}0b&w_eI$j@a5>
z_A1CeNOaYyMz)GYcg$M$Fg$wjRZ}B{xz)nOqC>P|?V3kli_Was)tuqd=*cxPySVWB
zlP&L7&6`x9^i1ONoWt7L<||j(crWah$cT;K^q|Kyzd<w1<g<PIN%w{6GlK#}V!1DD
zd%o(y!C#i=k5<{4N*-cbl&4lIE!b9bUS_gjkokVWkGV%o6txsWybFV`Ub=P2A-wwZ
zHlN*Q?Q(Y78x$M_`{XQizAW8y-jBQX-NwSkU$RczJN9o|@pxk6!u8(-d*Xi8?Ao%n
zQY&l4k}c0xy?3y4ewg96W$ohBX`k1d9N!RdwOL8d?8KL)2e@rdM(B&&?d^PZQ_((2
z%x1B$__KLmtj=1Si`Eu5U6A-!wm^C6J)J$;7pAUQ*#BBn;?SMcclI(bZn%qFZfD!2
zewE*(Vj=U~yP_}b70-44IrQ_>ZIRFR`;RTo4?g33r{4Ttnqv9=u6~(G%h$5qx3+4X
z_3-$KUFTGmT-|0pr+2~b`_DEnnUtHGzRdUBy6nWQ$)%iu&i{UGxjuQ@)VsV9atrOA
zlm&&X>R&bcm!HoCR{ta4yfysSy?<XLZWCGXqvX`7r42Cw%O<g@M0xL6_i$e1A<@*j
z+}Szr*X`$gU6yCRi+|r1(;5w_6OWP~+qQ_;UpwXSIN-)_R)NC@KYGV%8`wRJd#z&O
zRAT(&uf?SpTai`QxITxLpMAR}c~(RR(|<cxy^n@mGwj>y{uc$GUYVh5c|AE<cin{=
zuDXpH98<ZZ+M6$?TL>Cmy8Ap&piQG|(L?*Ny5H~g?`nP9iKUb#wxeN!TuPsn$PT`3
z@!|L1t~+;1j`?J}PP#99YRl3ojYUh#4}4)ZkUSdLA$7YbxApknU(u089S0W{e0kP4
z;iT8S?^*Jn9{g_LF{nAAu;@cSd!}7Qy^Ta&54)k)p$qad{tq{L8+bLP6xdjn*4?(S
zzqfz;bG9qxeC9EaVm@UZyUF74M*ZBjdy^{9aot#G;W=lq<mw+T!Yt<={*d+$Jv>20
zu&aQn$Mo>x*)=+wpEcgLdnwpqQr@y=&G9wAO7f2!*H{)SDd;+5--R!XWnP-EqQpvy
z^6v(?JIF=f^}o3DBA<D^om=WMH_g*VCTDMcy}f-md-eP6?jjRUNM)?(eY4{sqXOrA
z<?2r3RX?VL@A0VAKEr-+OLAt@q$jPVSNdBdUs+!&QtSHqrdDKK>io{aS6Sf;qU=->
zO-@zLn`XW^DDI<HlzxxWD_aGf`Z)%&D_&M9O1K{U-1^~lk%?$)z#m0Vi>k1Uck)h;
zPwY#+p*_#eyT<DAOPO;CdryS?J^#e!-ooA0i;u15Kc$%c)JIM=byB0{OrM#Llq;k6
z&(nEVa&C1~dIq22%KT5iYIfG=bt$jDbkl?Tl-r_T>uOh|9Ie)p@HlwheAR<+{u{#o
zbv(BwP2ClI!?x>@#;?F~%i<DVd$Fh4kLNF*xyR_jqGwYZgCc#79$l2=b>dKtXNJer
z86R#<n!0D!Iwk2FrTH^=<f&zqZrivmlE?m1<GmxC7ynLd(fH_OaM=CItinAL1)kr%
zsNU@tc2RRm^rOh*4MA79=bXPK@~U-${+b;R*usptPXu+w{7LP-WSuTi{pQetMLS%t
zKPv0X4L;uN@12o%t14X3-i=X>dAb|pC#Cg)FUrlHADElwomR1d`+m0_TW5R6^b$#p
z%-(O-c1>wg?bkBC7o5tv`dr$7x6!7Iv^0kIA<`;E6(?sJy<VD}b@cSE``iM&7JUot
z_RDcKwLJ_;o4H4*wJgSE?i6R!I|X|+%D$PDKC#qXy|c1HSv>8uz)at1p3ZM9TXkXr
zHR3jgbb9~udVj1>Z$s@f^Lbpib50yQzE@qj!SCJLi2KJo74Clzmoz;YsM}@NC-6`s
z^Qf7g(X!BE9aUwon~c-Xm)Yq13G{Y;oRiPDuaKQ1NlhZ{V!@JJ2ZkH*Q_Q`dPptbD
zsbx2<XkG2&htpV}$Zk$~u#45=&9Mc`XL>v@|B<?ALv&HrhIuuoZ24p4Uwk<t@-txK
zysA^S7I&sh(qn#W;&W!_#aoS0q1mx#vvZAnOA=poFWjB7^U*`^iQio(s7P%Q+%v1Q
zgl)~jJ6*T_i3J31_VK?`byI}nM{W4>Jpm`gHXgJ4u^{2Y^LIjP)@*&K;N;oTIrpnA
z|CG&=i{EnFJespD=bfNV_kFI=4Pw%M`kfx)zyEx_E0OR-e}B~dsj2hN2|SSK@^62+
zesAI1fTJA`R(NlnG-I{6!uF#}4Sob(=4$_Xb%Lp%T-=hE1}=7o4sX=>a%#zv)we^l
zqP^Ct<zG_Ss4a0oEPs~gw!4O+7Z1q$^_98F^ZDPCe{&=5_J&jXuV2rb_T-t)Wl@dX
z!sVNuxW?VeEB`Oo5H<Hnl}P67U9+FsD#>U5uFosZeksLWJ-6Wa2ldoG{*ztI?>lF1
z_7b(5VVQX8=s(Mu&bxf%XIeT=KUJn=@MD|sUH^p(CObW@GK{m-i}zJ=%25s$5n5)p
zS)nZd{JOjP&&tc68!LW$xIL<S+m_f(TXt_TzQ4P<`Qp3jla)6&33f|!@a_(u{qE-N
z>0vK@*OoNc)}Nlhxmr46D)-V`<ua${ZQz`5w1Z{l(eJGB5t*}-mtDFUc7Ju}*|Y0?
zR;0yUWLkc9w}|x;KbP1nr`1MI(b0Ql`o8Sho3ilpt&QhSK55lBA!{T&vv%64jT<Jn
zsy&q49XE4j;u%lz$jyn%vv%l9D;zMZb2{%Pb$Hj(rLjtPR8LgezBWv}ed^KbS6kMb
z^c+szw&UvBnNqtM_X?f*#hF%f(1<y|-qMg^|Fg`)3~Q@JTjC{sdREL=F*GX<YN~lu
z{(-}6MqRUHlGJwz7i*6xDzzT>v?huFF1*tGq<q4k8%m~Kto)XSu^|uazt)@B|1&?+
zKFjyk@45HyB)yj2Vg2OVn(Gn*8go8w%G+$_8K-mp_{`>;Pg@^df3=iN&Brj>zre@I
z;%9+rJ9ipm__50Q2OTxKiWa__9-%jd?bdIdb2rT+_bypDzuPG+U0QXwLFl1pJnPmw
zEcf$&5hr286z(Tcdq|^cb@Ru$_aFWLc>4bPIUQ3lYcXTuCY&q`P;0S>-2U5ULjUUa
zA94*0;SS;P-241+ZN@FRTlY+J%e#M1;1!FR8Mpajv`S?;@0a7{ZO*5rDgVAD!pe8k
zK{+eUqQ>HL!HcWEr$!Y1P!F5gV!ux-`}euXO}~D|PrhNlX5XI3y+71qQ`el>5pg%?
zfwzW*_N1iJ&2_>xw_{p8RhrDh^Yiyb*VcYsxX&X}aH;X)H)}SD_@pE)`}1w-^=H~W
zZy1#(ck-Thum1DSS+8%8o8alq8PC23b3`9=T>hKa%k#qPO>eKOOy3)}Snh@EMP}2R
zQN8a<lePXP#(ey8zpb<)mg$nas-vTu*(!g(*t(1<#s@dgoEXiVb7%|Ogu~P3nZ3I%
zZQ5h^%Y$v<vi_(0*1WNDOw-?WjmP=+iLW9@k2C3VPka8;a>4Fx@}7CtcP(ZdU;A%&
zV;`58>L-7u+LRKWEz!(Oxq1wLXCIF2EjuKB|H<k(qM1^UMILUMIZ0G$TXD_uQ{V0h
zr2P%mV2WLtZaQmQ;mVb(io$O;PB#f!oh~GF&eLG=yR~i2*KYf9Eoa**x6N!{W$yl>
z;+KyL%ywP(efcdbWA$A{v#UFPN&3DOnRjMK>8f4ktCW6T?cYCP!Ti!pSp)06CnEK_
zBYFOCUuwD1-L3RIOkXF0r$Y8Mr{AoNSABM#5f@WQ_X-m6_PV$+Xv*=*Z3{%sPkD4l
z^X7p`-U`!p%YBgkFF$?$f7#RZo6l_4EB}9c-(|hZGFjWBimf)rTlA(^8SmO+b>Z;s
z))#u`e|nWZUU&WV<T<skqTkOop0`}(-}BSTL9;~7dG4_(o#&o@Z`Hd7ul)7r+Mmw&
za7;Hnw{2a}@6&I#U(H%u9e(ZXveK6f!7C3&zS$$tw&d?2?!?<6)&5(i@!v>_s4qFw
zYX8gVU%B-Tmwn5wTDSb1|JlM^?EdbKyZ2tY>xe7)AMC#o{9(??Ee!Kty#CBu*Bi<9
z#_RseCDl8lKZ&I3`0815xgLs?QuV%A|JPz_@y&O8Dx_Dedwn~-torHosY^Y@QX}W+
ziri%_iL-wtw_W(ebak7GtJk)kjpS67kIBC}<>dPJ^;e(1xU%m1*V^d0-%m*t&w9S3
zIC_UeP$AEc+GjcZzDHi}-s-n-hY<T$>2KHM6PGD672Z$3_wY;K|E}+z&)3cOTVFi6
zQR8#*<j`mPcvMYR<TOZ$c}|m2+qrvnX>8f=-Ro*E*}gS=7H?O6zGch3$4MuvYwzu@
zd-u+4nIqeug7gB(=Xd5QOi|J4daCL2G;I2NhIxtMz0J(WGv}QVp37Tb%MrKilf{ox
z)^)4JW+~KO51OX8?xB#0ctd+y?OwB5DSZ*K3Q>9S!o~CT*iPFYdTX$1-knyXB+r(&
zOIz-l{Ek;CkiIkHoZ!i?brPkrb~Cqr+Y%fq@a#%k?yqfS?a`$fJ8bsr-LG(BRlmM-
z68o;0hYI>>;=43b3%$Z8<eU2$&eP<Yrm|1w?H=xXDaTxdtW}xa>!yewTo>4XD&&=1
zRlRkCErZ?nw|y)zYibXjyg4~yby2pI?Ey(i=CsLKD+9JaU^xEZ4$tGfz2)<^=W$u|
zUp08~YhO`(&ih>()(B`R_f9lV;7?}aGD_XMdiPS+V-s{*3%F}`J~2qml(S@g-%+-7
zSK0648q3Yh|7{+cu|3p3#LcNz@oLX~HtDsB>M7HlX2o>sT#&lI;^d*g2a*|Ava6S!
z)G6k*sgvBk!&-we@^(IltZFe^yOUpoO4%a$1v7W$-!5Okp}OdTfY)Qs@T=k~|8^a|
z$GEib(T=Avc9VDA75Wq6*Dfm+tfH_|Rbb)5RL^HK-%eK5;c_@6|G8}MgYKvT^ZAl9
zwVD6lZn4?ce#_%i$I&~gwankwth=rpdtOZK%)@yx6>s)mD!3Ut`x)aVZNVuAJ}mj;
ztzyPARr78r|9%0z_vLIYHWzlNGsLw>Z&2iQ?O$7YV-kORY9`zB*4uWk9`thD<CFNF
z6XouyckA&Mw=Y+3eEY4f=+dy+Sw+OKf8*yj7dic7dJjtEM5gClEy`JUw|JY$1*`BV
zqt;84<{3HNF3@M}mW_VbSbJ+$k#}8b^Atm!o#zy6lkzt?9#34mOe5`HFv}r!$GOIe
zBFaxyi`Xvi2%T}lRYW2uxo|pL)p8Ry#&>LCUCh>tWpBTca=&?P%fk0vahdBMJnG*&
zS4ZWYbF)di)dl`{bK`@V-M{5+n>O$Ex6UteCNmON_bs+4v=IEGSj8)!kgk<vqfs|q
zjrnhtz|+bF869!UOp{VFl1^Xd>R%hl@3KXB`8ENas9@%9z8tHlpmkcSShuKdW>&NC
z_|xJpd8=~WL-n7X6>dhy#TJH}`7f+x&Fe|dnQDFYvdX39x*R7=rrN4Kc=oD~v$Lds
zxvCaVRJHJ(==?oX)=R3m{Byswbw<(qdnOVqN^dj#>fcqAf1I<#W~KUK;UsaNN9&_9
zr}^4%Q;qw-y#5a#&)1oCSZY)gVn*+bj7&_C`-N8{UgzJo5c*$N-<rRX{lxWIieJOl
z>t0xvx$W+|Wz*SzC2rQpn7M81)x%en{obv=ub7_Uz3x&<<6%a2^$QcE&hezz8OvmS
zDpgXGsY?>t`Qf;FreS5RrDfG?f0lGk|07R|INzsNiKt9IeKh~vrU~0G-`Bs-JpD!S
zPDbv(0yB7n!<XfTO}#u*x?p+srlO^u4^vVvU%Pzb<(b!P_fJdDd|EV5`*id9GGF2B
zNPpKSI@^9pT6g)T&l5POy2ss_?b)u4r(@2mdZ(q|o3T|Xv9qHyecsef`<$b>e!7~j
z2z#V_{+a1sF+Z)(bJncxxij;*n~s;$hl$6T<ZGvd9bJ~~zV<WovBh&|?|bZh-l$P2
z`R1lmpR>H>4zvF-@{SIj%(Gr!`QaJ<w|w{K&Y9Dr#Whnr!u8n6SJ~4SPUjBYw&0k*
zL$Pnx<t?jItIS24tj}=R>~uM@gS~^ZM(kMQMD2(JdtT~z3O!%ECR*g{FG)N5YgbL5
z=dDZYUK6%qChNKVp;?C0Ursk%?YKvK#_L~hf1<bOZK~*$(cO?2CG+H!+f$D)@2b38
zYfk)A)^$=&_tuu)DcdR{-cv32^JR{v<ByKmc?tXUJG9^SFfKIZwy9d3P_$e)C%}34
ztwING{lvf&=Su#0wQO4Z4i-N-^HTHkBBj;oaWb16cm#G&`zge%<ut9r*`j`y+M&RX
z){vl}$-<(28I$kaTp{%!=!DVI&eZV3-b@$%yj!4jfQi|fiSboL3Wuvxv)ck!=lHFi
z@=|Bnzr|%$mtOmDV@XmP%jcBkVkexp9;jL%mR6OOc+xEUx&!C#=owbFc5Y2qt{m}P
z@7+2{$S0{n??tCj%h{~lX`H_If6uV>U2S$&;c8pscb}CrIvKLs7S!yXFWWU$x{|r9
zV_o(lZa?Mb#6uE1#=m2lPF7BT`X%ns>uu8_JN>S{*Zt7o7T>QrnY+6A&eC6d?p(Pl
z6k{H}GU#eixM}<DCvuln*8QBm?tj7~tL%XA_aTPHW=<ctn2a~KTD9o1GcUO@<LAkB
zXN1_af>}80UU#j!<|f6#Fm)>1Gqt($e0(yVg8jW--wiLm{5ZdSLP4Cz)3XBB^IuQ=
zx^tP}!da7k30nIyKA(DT;k?tQde?c*%26mh*u78Msyx)a`1PvG7frVWg`M+Yv6Vli
zxLq)1b%~<;>4n^~jk>eUE^Y~se#y$J#$^68Q2xv9(rN21e)o+|^qA#-w5jh@)si1F
z)Be_4-U?kGD`hVc<dG!y`9#EPTcstnb+g?Ty3cC7@J}<fuQ^tZu{K|h^L+Zpb<f^^
zUXkq<tkE@N`)RA!@yhorw&kBr%;mYuEY9p7_D;N4a{3Fl*NQD?Y92bp?QWi<vX$Tc
zKy%bdLH&QpN*5m6w^;-md)};n|FgBdm|yv~M<&bHD0eIQUwcC5t~EMqGtKov$d27|
zU#fju#eW^1EPInp_xhR7u^%d!e)B{)+%{o<xVn43d$hf*!tA-bB#V=u=dvyRr+Hx>
zyDG!(yk=o@BTLgw-4QdqT9fq7-<#7}ZKfD=I&}f_Eu9jDtP0+TljoOTnZEJFan6K@
zcB7J+{mZZGUs%sCcvyY;hy7B{P2W;w_^-X*%~C$MeAkK{1$NSjOFcOaE}B>y=!$py
zB;RkjyEV4emwiS<=)4%k&$jc59_I9yg%(d;wwOKk$buiY_Z<B7qMcFUhP=wj-J1eS
zt+})dOhYdI_PZ{gf8K77rKS9P#f7>)Z-Ro&OYa^%EhejbsMw=np?PT7whEDh9qU}$
zmnwA3+_3b)ae4hd*FQhc?*4pU{P^?rdj0iPKX%l29k-PUF||C>{m{xPCqXe^y31e0
z;<i=x+1@6pU8kOC?AfvG4ogvBX~okXrv;y5u6dZ>JQd^9=o-cUg7J+=o{JdE0gX;>
zm3tR`=F4-vZj#WdSvxBvuk2F6Ikkfunr;7n<$67-&HV1ycc;Sc#reN~dnsNwy84o4
z@3o?@SISH-NKR!h&AVTobo5rx@5!rdmS^vL+u^gnaY?7+{@nPhn-YGXX_n_!d%oOy
z(=r<#bwO{Xy!H8VE)#YuI#yb7+dTYZrt$DEYuV<%p&$3&DKHh8C{gtP-<u6q2i{hE
zU;p+k@9~^(He#=xm*{_azgRSV1J?xp^nVGViAt-?c}`wtNV|J9{fmK`lMQ1*d%#EE
zKy}7->)uw}wawb*zR8oZ%RRuk&*_<Q9n143xyG35<!P5^7VyYgY+e?-$76Dfo9nKZ
zaXG8ju6oNe&)Q@E&RrHi|2%v1>TA2Oz(Gbmw~)ficitK7`Y&M8d~eg)yb!NF_B|rA
z^Edrv58ArMY?Y3g#VzYaVcwzI7S%Ji?w7h~b1Xix`*TvtQRTVc>YwUNv%6=yTQ;m+
zZuygY`3--*Prr2YduiwQe@Rx~!&UG5J~xmo`d@x2lBd<qXm0bvRkc>vV}Cj*u{XQS
zGrzk>e5Sv|%n894PH&2OVdlcma(3_Hk8hOi{Qc`ho8tamyt3=W<693ml~}|){Cu^Y
zf4;wcT<imm{cf3Evc00q6wkz*yZ9ww_Pfb1|8YG3-DUPOrkL#xcbrnWTQ#G;NMyF%
zP6=y{h1^@G#QL55wl!Eb@_UWm?UKKfdRJ8a`!o5>2I-4j)7jpecAlQq8|M+7^5*I3
zKu+<sYrZR$1Qwo{S;;zgk3{UZxMf#u4y$gBTU1o(eRg5otW!2`H$>O88UNb7Uh%}l
z2Yq4UjDOVD9WTrEIJZ}pkNfHxcO6SL-&OfMR~d`Sr<PZrPT6U}V{cZ~7J7B&yHb<d
zKOea|@4EZEa!wN4eewX?KM_mI;=dN3J{CXKX>7YK|ADh_j$HxQ_2+dvp7V4V%iRvW
zb#?1cv%mYdx2qVPTey3*uKDw`v6ojheX)70(7?=Q*6UjOp?@y(dRe{vu(i8#=Vv4-
zMtj;M$URsT=DV%oZLI(P9u^CMbw7J9+dK@tp2gJ5u(I{f;#KoG1VthvuU^008<WT<
zxTs?D)Gw-q8k=`GO5UGi@BaVDZ0l!d%JXg%JU0k`W6%{LlYGJH^#Aw!|8goFJfn!E
zDP%!RQ^?TN(#Q^JHS1Z4%>3JCUw+rMh9xxgxS6bsj-0h9_tJvnv$lC&;P(x@Ae6iL
zqG({(+`qqU<u<K#c^i~)qfO<4^3Qk8?+veQb=BdqKhhQR<7ik^#QF1UUwS7A_6W@p
zcwLh>Q9v~=!?H-*W6kF`oByth@)b&Y?<;fOKga54(Lbpbe`2-t-6p<fEbi<V%e<D%
zTT!i7wD(9tY|@)&JQY<p(z5lU7AD5M%HJK*e&k$*vEszDm$qf)C7$+ub9K^k@7`%a
z&GyDG<|fR%H#5AKt!he?;i@eu?aMSzM>^k@i!U)gD6@2ju)Bs|QbBV&Q~d6SyOL9_
z3u8Vw8gsr~=W@)g{7QZFq2iZ&)6?^h#XM5eP~7|Yo9g!icIPTqE52m*EUC#$YF}$E
z{US@Hs<S)kO%&e|X7-6mzu5OC9lPEpy3OxFzM)u+KGWJ}M%&ncxObklP8Zs_7fb5T
zxTn3U=X`$UPKndpN^iNHlJEa~d2yxnED4s%BjypgOI^>*__yhG-~9`BEbON|SDtjB
zsHa-&+@8EumhB48W<8ueHBZ<b=dZVVQYWEc{;Bi@Q|Tuj6}BIFY^Qhv>KYbr%Hll`
zp(gBk=!(p`8UL1rp8ah+^_ozC>!JnwgjCLWr~cDRx@8(wGr!ikltb|t)6}Yq3SaHY
zw7&k`H;YN5%VvF;QA`%6>I81ySBJRYcom+Xp(W?v>YR7pYF@h3n<H&?hqtG!{u9=0
za=OIx=OvSVsgS5NX({DZGhNEoKkV9mdhI$bC1tk13tE<D=S3)2q-q5(wfeK*z^{r;
zjM9^TEEHX+a!5;d!=CpBO6&N3YX@jHtbS5_{+ZmtZ`1As3I2&KF?X=zYkLv@?fo@n
zw|~$4wf1b536{#5@*v~Hm+Kd$?l1im`{ZYeN9Y03g}Ups^;|-C-3$+QSXPmw?dB=*
zYf|&&%>}VLcGnzD)CvpM3i7!4QptC@;wQm*=PESA58V8koBl{3`e5VR4zub>YMqgy
znRjhht`^PyHTSZm=-S49y>os0<>toO)K~N97_dZc)XF;&rP<ED_mpPFy-(89r0%#k
zuYNG`q2>ClFtM$1xw&~a=9sSWk$fe(c3*!G?~>zDRdGEpzBoMpV?X!O?xZ`cd#ArU
zy=T6soB!iQ^R@iMC3YYAd-BnprAkloyNkvD%4?sCUB<6}E_PLJp;_$D$*Gpj6VucW
z+r6x4y>Vo%&gve{y$WgXr$$=8|N8yO57`n4-g@EQJ!?(hPSBCn6aV%5&D-1S)!*x%
z`{H$$=eseh@A3S(iq~GBiYQXh<=h}Hm$mMW2G26FJ)v_QIE_!v+-Ui>DA?TKP^nKd
zyYa1Mr=3ldUOjjq{B&(rVbFsZt|^O`Bpr9v<^0N}SW;n8#4jT=XX}=mn^vfBKi;?}
z>cgZvipFM_w#+K@z2qS@?Xqd^!iAZaraa4>zPh(|Z}5e!Y)m;#r=oB2GalZtbopI{
zJE|HpBKlr@I&>mw_p2F)%=y-T%3D4A)w|gm+8*U4t*4~?ih1`})GuV*mbNk}VcB$v
zqj^Rb7~g05Xt3FPCf$vn%kuMx`~ye*+n2Mv&Is)O*x~q&;iT$GiAmeLpB$}|=~ej2
ztYUBG;kEIByWf<_b7npI7y9|}|1&i^WRx=ueEWMh?hw`0?uq<u(DcLbNrmpORlbkz
zoBb@zOs?9TadTI#PGiruD=|hEYKkd!hgr@am_L2;>Wvlbiu$kDiS`#>-70h;i{W-t
z-RfD#^CU#}teJJ%(^vf0Gy~J-qg*9Y)#hK;9&J{WtMWRxT-DUS;h%3obZKv!3-5ZS
z<*Vl(O7PzBchbff`7>Wx&VRXbtN+hm3!U#JA5Yw!qxGcq%VHZt<qta3`=&@Y{Xg0B
z!aMZbe(74X@PkiNF8pQV5Bm_+u#oj=mDryf;yWcK+^JMs6i}u3VL|%obE=aCYUS_u
zK1h7Z`*C`1bobwta~dx!dNZM>^vJiQsHv9vkzq@nc)gdK-ce7OugtiCMfbK+eM6ev
znTT5o-xa%ZFIMl(x_0JbSA1c~(OHw8O0!nZ+-SOL`d5*AUvIp7AgK1<>~F%s6O;MH
zG7d6witabASa#C8=t(^5r;T^*d*|D|-MVMn1oh^Pn>ZN59GwDW1Xpj^(~vWLXK7lk
z#Ac3pmrgA)=n9Z|eXGCfo_yU;ZoU^&t1)XI6XKT9n3|!~K6k@&^KY98{M#45uvSB1
zv8eB>vq2NsUU=WydivFx!VCPqf(Ayp*%Mttmpn?@U#HT)Y3Y$j;Wc*~6!}V)yz4tJ
z+t2TRF1+0$;;&%(qt)_dn=Vya79Y#6X&0E{Hu3!#u~&QDH%Ki#m-FwYs)pQ3!(;Dq
zTsR6>?vSaho3ZQPy%XFVW&S^$)@2^PlGAecahvC~wJCAB?(u2w=N`V9#`;uv`H^)3
zKRQ$#w=*dz>-z?mR~?<Tdd}M3mp^!SralpV@;jkK>GsdwC?mCXF)z0!-=F)uTX5zM
z?rS|Bk2Yx~FE}9nVU6_PS^CmZhgf!}MQRtvHSRs*{d(o*9d84p+P4Os{l6soa#QR^
zr-nV}o^pN3Fum{hF>Ka`k}77A)(LTpH!rU?IHA2Dg<Ea&)Kf}EZYpk>0?nx@JolgU
zPB^dd@@7|<_io0|GH;Ae&X)e5VYz3Lww9yON7pLv%m4byte0i8EI#q`N13vj_l@H>
zW-cgcI=L($<nE7!j$izqDsgrEuH5jkw%J$2R{PQY)$iOjj$91*d-|T(61VLkFAkmE
z&vr-p9h0m|$Agm>UQV{Q^F4Dz?Lf!e^`7U$Uo2KoagM+FXWM3LfkgsK6D>@fGi-vT
zAKpIYKZj*$p-fik`pe&Irw9eeUk>}!DiI#3R&iF#KW6Hanxo!k%e1d~g!THzO*}GN
zbF13&qZS{Q1#DTh<Xr#q`F!?$^?^?{E3QU=-*?#Skw+uH`SPpnm#ciFqy6>2J@~nC
z?W%WA?ibe_JT3Tq*NaO>OjTFT`y~?1;}CS?XQ^^5)4ImP8iDN|iQZ}*OFBRPlHQw=
zF7cc7_D*)@;wI)PE|V8BTWsJuJIh>bzU4!)`GRS`DtzPpS*0oyMRpu#U3_TY>C?Ls
zR6n#xrR`bs`Ov=L15CR_%(6cF#I9ZK(Q-5O^C^b!+a7H5VPv>s_|}#4c+_S0H+nN$
zzKiPKjJ>uYE>1@!Tj$}K)@S_+uO9L+O~2@KIb5LWac8MaPKWPH1~FBgiog$B?-eCa
z+<0iif-2^fAx#<jr(&Y|wZo$L|4duXt+l)KXzzqIE@q<TGtaN#+<2*J_SR6TjA@0F
zJ9ah&N4z?{y6;!cCat*2%B4jP#xvt@dX~QCSnx^C{8Xr(!><FXy8~Fe<<9qWF@Jje
zV&b10+m#NQ|J=C#-`yV@-+y_Y9~5=GrFM1DRZW*WTdrF2pL^Y=TlVqxu^q?rlWdnM
z1uswDb7oUvk5kKW@yey{r<Tj_n)k{pM%jIx`jbDE)0UNmq?P{O^DO&XiSc@Wf4=ka
z{{HcDvaizIMB5DY*B>cOcjk`ezxk}|UYWx^lNY)>F9g2WeN)?Fz3nZNFK!X)+`dOs
zc^A)m?sRdrz5R8z-aPk!_UZT6A1YVWI3>H8|IEemGrsS?n%rppeD6ug$=fp*=;`jh
zT<Cx7*Viu-a$+^)UMc=$`InUQH(YeS@$UX3=k~t-_ElnDKEM46K8YvxEn2V6m1Ruh
zZTGI&XJKVkUtLqhq<lH+9&6_@e$EfJ+b6zx?V%*&!*?J%kCFMQSMBN<M{_<s%W=Hz
zbm&3$=_s~G`ySq7-jclX?_Ylt^Rs&*3l@rp=y1O)3S`cFE$H;(LBa#YjdFfhY&Vr2
z+{!(BW0U^pt~*AX4KMY?KS{WhAslug`*uP0<sPxf=oy{1Gc;GOz0Psl_pIsJY~I<^
z&Xp}LxV<1HX)E(-?cHLp4{v165<b^m)_pdnc6O|pP|RGnsW0nzo`fG?u8?Er@@DVc
z1xNhZB|aEmIN}kQGb>?-Ot8qLf*(8E9F0w88Gf5A%lMO7HNMo|dP>v2pb0G}Q`&<p
z0=R@%r!=_MPd~CISL(9*-_;M*|405{5!@eIeLpjqC$^6_+rO4;g2UAc-Ce=6m-(B0
zEqs<#wROsyD7&kN=6Gd!vZ-^1O}%e0@qDP=dAI4izgc&EUX{1BqU&n-3NePZqUcNZ
z6`w<5Z>Gk5e3lb(S#?8;t#QN}(PhWY?Jl}(ea={!^7!(0rHAIy^Y{HK&#w99Cx7q9
zm$iAOcHCM|_+Ql46`Wrj_<!-X=ZV{Y-MxHsH{<r1n;HN3=Y30bEfKZ*Ve^&w>qZ++
z8U1-a0h1iECi(sUKK~z+w}Z$MEcJ^iF})j0Lv!TL&D-$k{M%*{wfEOwkYl^y;W}AJ
zb)utGLhj}TjvJPAH`|%CI-cIL`kGMaM6W&d@j*+HPF}pT?!<;XjvVJbb3WhosAd0V
zSAC-SSMwf~3-K?O?);wg%C}Bgj?tuVqj~JBTzN+a<yXGq@7>aq?7l5#=<hpbC*^9O
zay(70*C=!Trg_h22Gt}k_qy@CNk4LLSlOqY&rG);UThisvOU?L&ffNRVQI;iX{lP;
zVb>mBvJjbSJpJS)fww#i1vNGq54|>KWjjd-w?18SR7?AU!rPvuN%JDTQZ;JMn)#$2
z3Jl)n_qud*K}LAdwvEjkPT_n<SF=4lb*bn*<L;I@r}>#T?kU+<VLnH7eZ|CGTwWX3
zZQr<Yk#5a2>jSaNV&<x>H`=!C@$~7}%MO^^<nr7wc3_#%uuH;2n7Jlxq5#W`ASUTK
z{)}!r#I>8mI;B&1n2ja`TwrGH;A=djr=c=Kg<0VabK)Zh--L_%l$Mn)eZ73&JiEW^
zm%l&9uj0A({61G_jsH`+p6ji3&S3b}H+@rWw&vfb|LX4aDC~~a+!x56wjrVGCNB$T
z*VnV_cbpO1HP8I9zcZ(MVc3+dlh^~UoZ`umxRrCxuIrUTN8votMY|-Y=cMj#sZL+8
z=dz)xng251#Z3yI{#opm57WNAE=ix!r18Yo)d~v~_}2cu{%kMD#TT*J&lnkxHnbJa
zZ(X#%g<bG~!K=gf<@1_NF0L&p`SCBE>77q|6CbC8!At4CmDis!i}5*h1kRtj+5b*j
z%+;elEt{RDh{l_3`&E1Q{w2n%;Wsy^f66<QvWmZWlh<WW9)pI-m0>ex+M3-rQCKN<
zKG^H6=FiHgYv;WSkFPF!diCqolW(pb{o4Imn74gKR8*a|xo-Q~(6ZB)-+gnnD5(9j
zRHJg@^lQ~OGaS^GYBaHWtlhXNc1CD!I*VUv%|`B<VM`uY&gHL|m#zBx$)9UcU9r2r
zbvx<o*uCP!+P!7P|1W_CX53HY)=l~45S2H7+PCG}zyEywQ&Uz|S$j3}Ie#<D<b_w*
zRC2?&y|h=j{ls%yJ<rW${+s^lzML#?UCR=%BYYe0gX)yoNq;0Z$=Ywqao1onP^%B#
zwp&N!{q(mR`0HhE_Wg3(@4Ibl)b|MMO?y*<YF&5swz}<HT<^J4zO3&5p(A%cUc30N
zkF9=wUb*+D|C8U{Wxl`uo`l`smvuq!_rALO&Ebip&XR4MZ5%9}@0toHF*v=6_KyBD
zd;UHK%WqS$EQ&B8YTUxe*Z`&Fcr>cI`1TIL|NG}Ry<HNVm84iPsX=|R-{jrdwUd2{
zx2~V-SM#`0)9YEQlBnIH9sTd`u{uO8aq%iz<MUXBW#QfXb{ise!oJy69y$EMdC!DH
z@o!wc-+!I@{So_1&m~O_)#1hdiheETsuG$D_pH9jK0m3(Y$n&K+8&MNE+<dwhubsx
zP46h}obkj+M&XgIlAiRUm+Q`$TlKyQ3E!yPuNM61x~pLN)R_Wxe^mQkX`M7q5$tf8
zv`VgNvdWrsUY>g<9+9@neCct17sF-6*Z<`84kQG{9&`2XEv$?W`?$&QwTyI=#nshD
z^LAP;t^5D%@A<r%x5<Ai1CPe9dY)!*d1{%wfR5*eEAd@Be9|7Ae|_?}VzcXnl-0Ym
z=0D!E?#zLej2pcFX2;nqZ;VijjEnIPs}ek4I@e0*!kU<7Ke5k8`HvVch!7NxQgP^F
zy*Gum<HLbdKlSGONFMxG8#kNr<K{hHi*ERam2aHwGfi7IG|$HSnXu5W;sTzDb#o$<
z*8C|Ax@*~06RLXpN&Kyrc4n`~=L@6$6&gM5`Xzo<-phT(<*oblWjQ-ebbF=uPSGkq
z?;-9MWmCHPN|b|y(4ESxxaBYZrf@kvn;vRw?0j)W9-C*_(W6>d&(_$LpU>`@88=0*
zYXAPC+Y`#>vc6^vY>#*#;~LO#BW2Z;PxoIhD$-bg{_bc0nSSq2FL`r%$B9#`RwzGi
zJNVQ<h-1-|-Llj5!#!_V^*+3|a7`?`#FU?VqRizQC0G5KT0Qf4Xvy>xZu6Ht#yZCu
z1r9$I^Givcd12e-oz;4eboAa#Gw*4;p?P+<SFzIshWQDrznnd*vukZzPx8W(8N4&k
zr8!3gpA!}8`F8EH<Fw`lF-$Wv9E2a=^L(*gZkuAj-Ir;(yXF<n*L|ECXHxhfa(U^U
zHPbiVx4Ca5&L>czabVT*tM}DenoRU2Su$`4{_6B>ar#x7dB-kcPRkd86>iu6e7hdG
zH)f};jjw*X^R-qPPo?O-GY^m7T9NOzVU<bm?2Yy+Mk*ZM*BI7(TB7s0=UiS<*=DUr
zKPPyIIyG#UNq3&X#p?9d>0D`K?;@c?2MZ6{-O%J`S)^F*C$?KE{ilZ2G+BGzMchYP
zB?7DX4RS;G%~P0u<)3BXg!8Y|>LwkH6^_ZCy<*qF%>@s)6)pL^N8(a<fx$P&(vHuL
zrcN#@xfyaoQ&vhF2;IB??~j3wm|iZc+?%Ze_D@7k3KUyDJ;WmT<|+SPlj}?Kg|pf_
zR|IBp^&MT~{`d4xH|NE-76tN|&$^YkI`hWCBq2eas))7MtrD!YIBuwPeY+_)BS7F#
zeT!pj>XP!fb`uVUh51#JWT(h;oye{gueou8f2MH8|IJ!o(%Qm!KCuKQ@_#6|*88iK
z&}X4+`{3c64<U{&ECDtX|Elr0$gSaj|NH+N&v`11PDetftCy>C@-C41oN6y=7b+nt
zl$*Qh@VzCBt|kxEzo_x@3jSuh*RqAbT0^G)j9*vXDjl^J?KkpIebpY^XrBDP<ldH>
z=Vq+Ym=rPR!t6OsZA<$1JU>(YDJ<bZ!0d+akwyoW1$Z>uMSp)IVQXA<GG5_~;t5H!
zi|@2Hhcc@Ed$>hb!E2kvY^7J6owH}pxzH*T9hBGeipgBq`MA)Nw+!e0%e~wGzk1Tq
zV4g(BUl*P)cjObC`e=dikHhxdyDMdROnJh27yVlBqF_a0qtl1neqOKcrR&!H{rUAt
z)uUB4jA=8q<n#VDz6gEEANWzDbg4^M%d>9|2jcZIefYDuy`A1nD%o-CH{*w$ZGs-<
z%oDud^v_jXx8diL!po-=?n!CobA^j1@76q+cy-!b%@5p+&B5<4{m$(Fy+Gut(TnUa
zIg8J4`x47*U)iC_(YE#Ip-Vqlii7V*l{#=MnFssLwc{`SWXa{6Re5V=pxlh@6E8i?
zT>kAu;jYq(sJpXPHm|E%aDPYEzo`$MHSX{}3H$TsdBK*;0-n1>9j;G#?3$>jyzJNg
zJI1?ekG*78U+Uq&Y89a{kN?ZEHm&)eH<~FN{E)uv<IfZ?7mYs_MX{?aGk1mFuyvl%
zuH^RP>|1$*2Vt$}x!Tx6G6E{Qy`n?*vWvDINH_H44=yu&YL@2Ee@LYy=6r_q=WiB1
zR(+Ch6OOqjoeDf7dGDajx%nZz2|_ie{onc@xD%;dIA>l<Pm)`WwbMhnnv|Bl=65<m
zw&yDBxPIzh%h<NNzc0ABB%dkm%a*;CZ=VWTmU&vJ*_Ud5FFg{=wa4J@?FBD8E;s!6
zY~gn|WN(r2w&%hVmYw90583snBCTyt>fyR2#_m_kmh#U1DYZ29)T^g!GY?A5+9UXD
z+RpDD?tMK!ybX?(Y4tK~yLn?O?~=-gQT|W0&Zj-DExInNb}Ig8>)N)M=W=R`_L%v8
z@;uSDJG14`U%L$Jp!fIh?nyO&ucxQqE~B-8kH`G1cK82Urc+{hG1Ipd=H9*SQ+wu9
zsnmI~UbEtC-+d+d^%rlhne)%IIRD9+FUuZY&fT&tk~i@18_gTP+Z6mS*Sxi9*Zg5-
z`nq#w<ne`itDJOltQ(I-I+q=qnc3B~jh*wNx9|58ToLzGw>{+%y0)?U)s?jqw@pl#
za$@uO!oZpNeQVUpOcU)cpDtS$J1s1f{mBWgy>bedHb)9`Jq`cD+xm#Z+4jSZV~Nwd
zJO9pgzNlUw_1fS~v+P~#C($+jnR1uxu5EmKSl52eg8CJ|I@EI%HrajJ{zj)oQRKj%
z2G!j;lWweMuFCPxQCanDUt!gJAqSCVFYZ0Ra(Sze;`G2Xa<79fe|@AQb9Zh3?i*Wt
zrFXO(I(q5umipfvbMGH!s&=}0yvzP|$@S<0X;(EzWeb)ylkRrbzMZ4@Wpg@v?bUm~
zcIQmb@7S~R&(E*ByI0RPf4{GI<&}9VWg%hDti6n99eXEv*F7`FsaDG4>{*eo7rT3R
z%jb4Smg{ZHnaO|a(q5rMiUJ2B`&GBf*gxC;j{EfVWjTT~PKU26@}70yMC0}C)S%F}
z>+NS4pR!4RcU|z#>E%}IlFe%>SFe8YPE<<cX579%?S5L-F4B#~$_DyUJC?*`{#IRA
z>CCaW%Cw^(@?{e9MCpz*AD8@8I_Yuv+C0DL<2!$+RZ2MsP1>|qU>lG9-)Hr+*2X`;
z>>ZgJ64N`fFhE%a^ENCx`?i5d-Tn0!ZVN?-+|^fHvSE|mm(;tnHntmfJ=W-KPhL9B
zXV>3fC#T&#rYalBzPS0JL&?v5i;GM8>^Sc*@dW%Zn4t1Uz+?M^w=zBYA^`#12M^SU
zE_0V~JZm~be?O1j<STlc9(D+uX|yZb7*rNb|8Yj_tAzX#&v{{$#s;0rx1Z>lSvYN+
zSQ%`_A^uzTyWEEv8%*<z|5)v>d2+gP_cY6r^i;3lC2B%*&#q>_KPgF4+V1wYhT3Q5
zx>u$|PCEYW=hfvm49eAVc@Csiu34b8>p63k@I(%wWm}zi^Oh(cUekV9Ln*RZq+^9L
z<3zcRguCo3*<AP?Lkv~>Ba&<df^}CE_WaKhzPyU(_w)FqAHU8YP7hFGbz^bveVi9J
zvncg+Zc)6^%C%hq^K%1#9NF$F70}!Jk=2cTIkPbj=gR)T+_$CNmNxcZXKk9>{&?=$
zIh$D{@7)M}^v5@`d}*ZY)*j9)LGv;yg_u=D92IYIq~5tYVe^N?h0+~QCOqn!TIXN<
zSS?iL^ns&)w*)qB5HiV&pVNOvh<oaWzLf$lE#J+=TB8IT<!{_&^v>N_((j*ep?|LN
z>5TTa36}3}?%dKn<D{m$&-p0U%(FYPHlE#Mes|}#kM)^ho>RBG)<n--6`Q@wqh{C4
zgUX`DbL;$!=8L}G5i$2|hF0<JSt8W}hi<;n_<c+LDf8aoId|4=W|Cj}rR582=ZpKo
z0bljCPSi8*6%Y8S#PY@Sr;A9DV_mSFb3)74)*cQyqh^l9=P&p#5=rc`@wQ+3BqsRa
z$CJN)=E=>^XSrgh(0lv4^l}~{(a&EuT0dClez0A{XW@c}Vh>{+T0MAJ?RwRPggk%$
z=>2YOF4=!Qlk2)GtMj*wQfrQHJF)qYPVTE$D?k2@<xu^d=vJ!SA$jLVZe3G)qP}Lo
z_QYTt>jUNU-hThRP_21h`MGs}>@uyEeayeU<*oVTs_P{we~<8pA8KA)dPs9>=~oB2
zBMajmTRq6W*%NyGH2c(rdbhPKtY63MXPuCB^q}3H-v_5~1T`LBT*IAFD_?e3CZ?JH
z_#qXpPrLG3rv$8waFi%NX20hb=hr9SL$KsDGh%X@k%19PR=XMzoquPWP~H3WHx_9u
za8=S<xx3~P!^+1?JTKo;zp21qns`B|JbY`OsOtK=@%I(mFL_1(++nv|pS^S8LKEXT
zr_B#P)T@i%z}i}$BWU$uzTTvykAJH+Rz22{P~Q07y8A2JzS}*AwBpuWj@NNJq`F3P
z>oxCVGmbtzd#P;CjUxYP+fJ`2jqCojy!+|;*WbH0e_g;Qa$$yK={o0#C*NLiMBgfu
ziGIF)(ay6+;=0$}+n}u2E_mE=q5kO^kxbjVnY)*-R60}_?|WJ{f11t~mn`u`D`r(c
zSzCGdbIg;OIuq8fJyWbc;X&!#bk6+Et5m$LzsSB0)m#1C`*v|`_WGGh-rHiU>Qyw%
zviA3SMBQ0+E5khGP0XWeiQTW)_-rof{&;Copa0rLr*EIhD^Dm);GFm9d6ad4s*{&u
zuS+XaXGYY#scC@=mrr`Z8|jc_ejv7X@5|Y1CcZ6Bxi?$aTQO(}+g$6oytCI$oTqxO
zRyM!4>+VM}&4K{?>ssOrOw+$;hIcGF5?@neJ>jfaqkxW+Q2Q4DOc|l`2QTblF0Yv$
zx!<S3_}P0~Y0i!8GZ}SsQiPwy`}47Bo!*u()BKNSvioVv=Gf^j3#~8yd>bE<Rlf3L
z?Bu<x3f5>T^|T3xdRDD@t<Uyjip)=!6IpAM^1aleTvYg&{0(on&2~MKd}8mT*y!9R
zt9*na9k<^2v@Icg_bLJHeat?cMPFY|as4KBV^+^|?dx{yZ?8G=kKsaG?3{Yb-#dOa
zPG7m^)WvQ5qL!bT9~tDYC}Lmzr^%enalPM%pX*i>2X<WA^X_%TMRDHyVln@?!<UB!
zrtj1{w{J&D@Bv%Cw;9XoE%X)&7_YLtx^H80?2{&M*2=V99ZDOtX0Cs=YCThp;SJW3
z-ZIIFuRKD+=8E~Ug=X&eo%wA&TX@ejZ?D*WQ-ACX*Yv8J%l&pjZj8it;bfs=rEZ@h
z!;f=+9tlh+4)O12)~$SL@$c}>Vufg~m&u*?p3Zzw#m8{Tp5=Ne^K0kOmBoq$x_1qj
z&bP3~tqILJ7e8M+p85ExQ-O~a&M>>GbS&Q2!l(bs>66+ySG$i+?@wIQn#>TSu6i;j
z(0AcO<408oLz?{>jN%?MbtwLJ=<gG>P>H$9QhlHA07HLox|zb10MYY-uj<~nTyePe
z_UfE4wH?Cj%x+5UGauA7|GLV^6V#xXF{{sBqGdJj9FG3xgT0)+dzUvPsuad_y3c0e
zT9w1np<opf?cy79xFUw_$D;t&1IsS<KUb}hGGf?b_ow0U+n1XAF8u6T!8Y?Fi@Wr$
zAB79@(jGkwWDx(l?A_lZ`79BSo;)i2QPJ>Ep<_802On2<a_?Km8;M#H@~2;DF57Q^
z+I>MosM_-n0<Bddhi}}HzIvRItvK}M<PO(1!Q0JeZTbzCt!@qI^04tX4qd2cD$2$3
zORsR_`CE4{mV2D){dsEr>gdmAw;z1o_v(y!bCyQKjjU-4uTE86cDa-P)TT|My#BY=
zTxU*Ra71B8+L?x(2?_sqW*s#(6aTu?u&?)+TyUc5{<GWmh#qDz7WT1Nv`23Fao@)h
zm+d&`Uih%e&Tg@vuWaG@#SU9e78%Z-&2i?9wc3%@6P3(AUop6zxb;V^P2&LpIp6-&
z&|8-UemQgeUehkuxi|BF<yOXtwrLtuvusM7gZr#5?6X|mAhC`0){=wr{HJ~_YCm~E
z*g^T!=d6Co_V1l1idSki|C(^L^Xe-1fa_I@S>?`XoL^m6@vL)RHM95G%Rm2a>1n%h
zLN@1T{F$ZeRv+W~-`V^!;uNpEobST}i~dddTR-3Ttb=9Pudj=lRk!Ggf3TmZ#jx$g
z#{hxi%c=USj-Tc@?!7a<aOMGp?YGoUa(Q)M5RIw4;x*T2Uh3h4sj3xTH#hn2_kMI^
zP1lll`&O^soAJy1xwKxi^ZtK-UcPzt_2={V_H(}e+5RbSN1|-r^0I@)tSgSlOl;;4
z@4eCCHl<TxFXL28oAd}RRX&DIpOexi?|=Ei{=ovp>iKDZ-tsf-&0V`qcI(FfPwwY%
z_*~iY^VnB)(LVNNl~-z$9`-D?`IYgo=jQ*7ha|QC{%$p%W+mcxd20W(o}Z2mW!IID
z)Eq2fxw2%r&ckO4g$K{<E>FvifAIP3&GQBC9e<gx+-9~Zcw@@0JC8u?titp4?dlJS
z#(ZT7N-@>_r}^Su_^}t_r!waMI_E8Nx+nKR=)|pSfAu=PHZ|}*+E%`^(dLZR+?#KY
zn&wZw>ut38YPWcN#kE(;F1b4{GUWH{^10@9VTR9%)6+Lg*=;R&DigBe(l(!hPg%34
z=7hbm^<VXx$5l#Isi-c!&sufX2b-4#MYadcS<Sz`3ZA^*XKwzpuI1PEzBy2!6Tq=^
z_Kl|AuUFov9(y(;$c%CQr*{dn?RGBbJbF2DGgs+W_g`Ng84Eq~Q;D_q;c2+w5pT->
zal!eClZ7lYA9Jj}(#Lr6-k0Z&U32C=<5?kmmgV|P152HG$wkL~@}9Up<e6-q_(|#T
z65h}1bKD($b}hSTZ~15bE_vIFwFhp<{+X-r-F+it*UyFDPVoCiPdIk^tKQ_r?>CD3
z>^~jVv1PUY7uL9qXP&hQ>c3*h+uJzhRQ1iSLoT0K^5W_kQbSMg3p~L4>o()$>0(c}
z-wwO8^}NW$PtzW=UkzAzLrQV8NKnPjJu`34NlU%E^^@e<U15S5EH89^<cEId$=l)@
z7IyaOrf1PP-pf2z-f4E_k6$3mo?dRv;Vk4jgZ<eTrvLw5`9EfPatcdx!IGHff}y1$
z%8swMk<tCP!OevN{!b5BlctqV%$%j<d)fQKmr11)+LuI}FxwlZ?Y%A~-|U6u|Gzwz
zQ*<qFZQcCQnM>s6nu?i{=Xlx=|J>2MpDoAXp}e$;Ox=rwT{1Psay1`ir477vw&obj
z<Da1x_Tu0PeobqM^Ktygxy#Lj+IW{gEt)B2eE;F$($40uX`9YmvzYeig8b7Z-kaSx
zYb-pkUv_;`9lrjqQ2S=7<6&1Nz1#Ap9=gtRm&biG)BeoHE9-WgVoHqEiQRbY<mxS_
zE_uE?S0``JJUiv~vWwR=9H+P}emB=X@!6w4PtDKQ%f#DM{>y#x`o|%oMUA~|ZTg9_
zY3;Lb%|7(f;0DkA(#n?y_Hk9`Z~nT-pVdv&b4tgiJE60_>z>$|@1|5_7PR3(ziUjo
zp!mG$m2vydc6o<ny!!OV&ZPC`Dv>*!!Cx2ENbg*Db%k_hm!_5V+KcJR=QJN)E7>C=
z_LS?5)@u8Wq31sCp6eC9+Gh5$*K2e<CVVN8dDJ6hTJiDkw#e9|EiWCNa-5p?mKX%J
zNvu1RbT3Ht#)nenr?$V+<fDBaxh0+B@Q!Hz@txh8b#L1A1vBsb`+jQMu39c`M;|3C
zBl~*)6oct*x1BeKPf_eqlMTGR;e`gr@@YR7C~Q!7Jo?RHeo88*zmI!S{+8`wmG>_<
z39j4abV7W!(dyTm9&FbA-@db7(r?8(sk0hO<#%e&;X2f6sbOX?t5Gp5Tk`RF1E*e2
zo%1!ff@1jO=k?wQ;MA1=xsU7l$$G=9Yir{^#nw;xQ~vDPOSSmq=j(fKTjbdB+<Ye}
z$92ndCBOVWug#?vhwBdRjNh<k(T+87iL<`gI-OYTzE|kn^~|lpfmMR6pPoET4cid6
zE&b(|IVmTCD#fNRzy9!z(rc5m%V*p0*c_OcTGiY8An?oaTcM3Azm_K63w#r?NFZFl
zYD=Y1cyVs)vvoyBO@3|A`+U}Kcfn7V>*=RG1cdY^S?ypf%CZ0XRn_dy*4cSwD?5KO
zYz$<-Q(7gjsrb`#w#ONDWp9<qvatuJ{a&FK;B@?gTysTthvdrESBgTpN^3>0+q!J>
zUGl)DTTbb!+N3i#?B<rt_9&P=LvGO$-y>6#*0i74(m5k_(szNId=oD;y|-DFcl+<l
zWvdRA{(HZ#_SWxb{Xb3rR5d)&zatanqgrsU)0DISxZ%6=6%((XwfUMERr>t()Z)!&
zxjc5Xty6URGS%(uRece|9#-2p-JlLJkEe=`zw=jk)+9fAWwEvB>Yg<Va^@aeS^ktw
zIym0{m)n}=nqya^c+}q)>?q6jSucC?Xqx56qx15A2XUw`?uhJFTDR%T4XZEHpYM<R
z^XJpgqd(fk{pIVc|6M)5_gJvhBPs5~v$K5GaXk0^dPDp81Qsr4l^Ht{Wc+h<p7Pw4
zdVYK1Iqu2PkN34*x;X1$`MIYa#>X|qHNF3=`}9Tj-13NJ`8Bt`8P8Z_?aj1IB5{GC
zV&@Z{z0K1t7V%o#XJjr}buLWw&F+m{-_H8(l}+Upez9;bXWWdB2CN_4*TgMOD4(kr
zzVmkMwcWZ^rD+eNm~O9>UAB4V?5~#bg^PTG@Atb+=2@@BF=JCfZS_J^!DTml0;U?@
zVpyO(pZ&L(po<!>!q3I)%v|-m3>cy(c|2X%xZi-eDk0%p3FC2H{)*OzdcPkY|0;5=
z=g8*=>xw$nriZU<lI?A1=G~jQk@?T+9o2JwIeC>`Y}Oa#@T#11J@w9l?dxu?h})9J
zda*Oh#`37n#~;h*>pcjvfB#j*LHpjfCHyl|f2BA3f0B_tAaB0+PQO~mp{r@K)~=St
zz9lhBBjplQFHQf^DpSP1f;rk<Xtlk->n%Hf=w`SJdWKHl!~X70kMAV8oJ(SCkzW+D
z0wzgl@2+`pllz3W^$~`z`}bd9<MI9-yWL-yS=0M{<|NVfhukiUZzNaOhF@U15PR8n
z^$kI-h$I`0%*}WAujUJ2;M^t8+g|!$!qg|7Aq*}ajAdmlCzTsi?sx^w@nF~VVA@{q
zKErZNbIZ06^-GSKJG#H~Bz@T+B*oYMP(4UV(B+Nfk^T#e7uF>%@ABk+I_GA$XJw0!
z*tspcR6}LI)+`cPc&UBc#g=Rxj_^>i_h)ZkYY(-&{kX>Xj-{gDj0uO<KWIPaG2e)-
zM#VS$N&UV0D?7SVJWDucO?&cj(Wk37nx(})STDQv>9Iq6)mKiXjj<o=|G!t6^}FWa
zmBi&W2|__GH*D)CC{LQ8I_tGwbOY;a^Ruj9pS?MIQ`Y~Gnb_L(E_UxCH&pJ)K5cbz
z_X9Jp=tbWj96z4N|0I3&Zta><rn56!nvT4y{Bf@?^`qb?MIDyZ>QC=wC!SJKSfzLS
zpLe##XD3PS`?-?ci+Oz-${IK?JW{VnyKATG{PfV9m2r{V&!1ACdW$91`&B*vFWFU2
zr+2Z%h2^!*>*41T)Z8+^+MQQE#dck#`AoJA$LyAFzx8Zx>hB9p=NJE&p|tQ%vtG+u
zkG!ga%#5BVU#7W>|9_PHOXvNswtA_3hR1)mMQ7)I<K;F9lKH@^{A)x06MOx<2U>e?
znwwlu@tGZ6`}qHxOO8vb>(81r*e$ql*^_B${+%7Vhx6zEmG7`n*@szwn-RCH%?x!>
z_0@>T{M&8<b??_tXtWD(*mA3T`a>y?Ia_=K7Oz~Zz+adUaOG6gly;>_es|*U&$Las
zl$NyntuyoShjPZ7H{YN0`s8xX*#SQiC(k%+Z*}tY$?l^m|B8(~J)~TVHbor%))P3X
z?PbcwUjjO2)>GHcI{Z@SiIK-LdGY*r0Wz|C&zG!Tt?j*O-noc?_XR?1+IMp{o87-5
zsrckaMaZGb*>Ce)uV^gVWaC=9RNL&-F^#41AMBUkk=(S>^h|`r>4wKsIMkfAS5~I&
z@9xdB;Waf;?4PXYyE!!LYHnkXyF#SU#6As!BW(+HtQ5TWaW^=96mT$_$j>OE&d+q@
zJ@c2^iU2QprjB5-T2X~r%NUbYmhda7><@0)QFEb{h4*p(t^gmU>d)cavnT&x`MN-}
z*t2tU$B7#&U6YvHmo>bJ-XYq-k@?}#h1$}sP1R;@xBjG;`SU%DkucQ%T6OAC!rHRn
zX)LRp91Xes?(I9=$gxQ_@458dReRp*1U5(8oLDt=spjjcT6dD-rk`3@ckS)5zOdwU
zh0Ib%&Hb;+B3HjYrFSj+^yKT$60S~o>2t*Jz{)>uk7vEjRrS!&)u{F~;@#$2b~(ag
z)2X`BoomZ;!z{NvXwg`~o%(XCa@JYTG;2O9^)*+QMw@bdJGtrYwhKEgB}!9GD(rm^
zI_o5}ik0zf72d#__)GuzUN?EOU0d0oIVC&_arqXukXs;i<BW4EOV26Q6;)R>Yuo(T
zb5Vk0(k6xD)sJrl-E3Xwm|DO(_2AU0Sy$IxnObyO+P(6;*^ch;<G$`ko7V;|Jr{cU
zfa!sYojVV$sb^20`8>6^Fnz<FE4|0h&2W(A+f(;otxn4sA^)@{u1PJIPV9PMBU5m~
zj=i7r$H@tcGd|^-i}Bbf3TAlmin3K4X%=qrJu<P8cfXjKjAG7@w+GwKWilN6x=MR`
zf{wA<mJheD{G7k)=!EGel1p#=-nFnZH=(kyI;eTg<?a1ByEi7XT8jMbUv@D2`Tg);
zQ9n60%+KDsyRf?S?Vn9;N3+-3{SGKReXBSweb<R;t>yLNs;PH(1lL^-{eF~h_UvV$
zsm^kVevO=AHb>{!_PsUhf1*0;O3IU)Ycy{gzE)G@Q*Q}+#;c={ac;ZT^%wUl!$dmT
z1@!n>`S1Pv_wVXWW9vL~HddLdrHSrpD+(H3+?u$qVfUAn>|&cz?{1B2Q7&pT4=btq
znsg^cw4>K;dhEl)H<wAimQINEPTU-{wALhds<*<u?l+%YA20|q|F}G3!<C)Cf4)pI
zQDIql=d#V_P3PA}&HVPv{%u#-d9muJ4s*L#cGvMu<8m$tl5;w`ow0(g^U6F^6Fv2n
zPu{I*3_YA5u~mMv-Reb_dtXO=xL7{@_d~nb?X`1t!@hF|>YumIFIo}7e$VmaTZZei
z=ltCy|8Cuy+isI*EHV7*8n&dzLrzjRK6$oc#D~NO&uiJ{8~u(k`l)5<pq*)<^1q^D
zwMtew>#dW2^|kKI?$gv2I^<LEtwrw6<ji+IvJ7b!YTiXkD$=Kdu16NeyH@lIa<Bzx
zZT0#le$2&Pz_f?o^?ya(p-HDdm+Sk*>+AXZ#ox1iW_OQCz~{Zk%W8#))$besK7W{V
zzUSZRHSPOXPB)X5P24Pg$--(|U61Awlj^&huH<$vWm*1W$FlO!*E~%p{p?Ct1(#K9
zInwuS#-I67i6<oNPi~xWr&|BZ@`*ccZ%yb){p)Gz5$b#QrdqJ+KH=BTFFW#9ZP?Ia
zqN-ErHUGl%8TsX<cI@m9qQ~^4yp<L%zww>hsH-{tHPhtH56VrBvrca;s90N`%g&gf
z_IHwg3R6$z#hX1Xx0fw06Y?=`&|x(ZD$BUXt9RR~mu>Gglc$dBbC{R(YusK}S$n-?
zoz|hgw>3H6tCrsCU2nQ<yZq_rD_L}FoZ9z&5oBYY|L|n6RioW;nUeYIuWV&|KJ#|<
zf}bCzpSxWZ-E4mUO!Vg?dvd;PIuL#_b>Zxbr<7RO!xOeYOEsBzZ{NfY_HU*ys{8Ez
z{9ZNrTUYPgYTh?zzR66?y`r@{O+N18f;sIcb1mhSm~_{Dxz`acxAwVWr)}@LJ#ou6
z9!%ckmiV-7wn{>EYWUXc`?r0*{c6^(S6toimNp)~XPe7?z;GGciVX2Psy31D>^k=>
zi4%987guoDWzvz-s($?)wR$t_c0cG+{8=AdaCqT`W_EA!X<-xjb{bqc)A~Qv;la-T
zYce?$zf_9YJ><I~_hqMCar=irM_Y|+|5n7UU1n?;boYtQzp7o2K78U*y`j~WJW1u-
z`B^?yZT59Pq)n}suf|ftnG@4PH8M0o>7m|@%+0^;CiL%K{6RTB4c3U)wd$E)Sl2yX
zqW9=Z?-cE>9R?Yln>sy~C<XfUnEd~nw!?d(cW_wl%Y8CAp1;jDpFelz+_`rZ^0v<c
ze<a5pcsRdm@yzd^FId-{v~lxvQGUj&E_Yc)=+^7x$z6LM{@823$5+f&?1$(2^Om0<
zr#;n+Tk<ojQ}mts&Df~2XE$|Dc2&95*!sSi@k7sVd4%~@|IEVKY0LiHwmErD-{<G0
z{>>XT_gPu)d^0m?;_UlRZ6udYt`6Smx4Cft_v;_R8Dbm7T|AlX&gNCs9&DNK{~`R>
z(X|KV`=>8so|$x9!(7(;(rfKge;g;tIP@?0$5M4lF>{;Fqnwbf;@dqf-@g=K^_~&(
zVeR4z)xTF=-LT`OfT>4&#=<>z(_JoZXqlaU*~FV?L!n*NLYc=(XN{Q`M6sMWBC_n|
zS6{o?*9{7utvjjmSS80dO}y%t<-D_jj!Sv2GfO+I3F%|zi}yMn$F#`eOy}V`>-{@+
zRn*oLS#403Wtnv=QQ@qf+Z>mP8=T`5S5NLP4?nISKX0$4^uL=&4Y)ckh_Y@}t|)ey
z&l(^&F|x<#+MQ&co{FCfyH&+oK3!0rvZ?5UphqOv|F>MeA?!<!op!RB?|*sXRi5io
zIr6<#IaL=|`DW%`4wyG%uc0xE_{T*COCOtEw3z$g1ZUpv3+C3G#}+P}{;Vh2t+LEh
zbAM#yWpTmIEm?Xcfs358#iD*b|JW{6^+G@VHE)mhA>F0Q_If_=O**G<+wp>RUu4CL
z<4?s}oS0d*_$_a{bu)x7+FE^r&)Yv&?KbP2%I)j1PJ5Bo^LlAtZ5gMJ+gaIOlcnVx
zGKU@)+UOUkYc7sDX!7RUOOyX8f7_VUHb-&_{qE`8@Y60{*XMSWaQ9cY(0{7WZ<R>j
zU#Js(S<L@)SFOz^jl#2+t1r(;GhKFT<IW44ZC)>roT&3CaN@m%CZb>8Pda*OhDykV
z$Or2UqcrxM?0#Z(-abd<xcv6Qk9RL|oYxUJnfCE{$+lTDxpv%3*)!|^m4ccFkEfJY
zrfw`MyT$Dxa-k@B;XV~rMj2a+i;@l@O+VjjoAGm8TervY=+k2lV+1{?TK`$Ucw4l|
zkBq1DH%2t_N}X}>kl>9;6Mnt*mfto`Cr=@_OG~@eC!2V8|28$e$SB9lbS|;w;)*iI
zPM4S6GSd=dZ%;5-)v-)h<9#+yqrv29@v1z_Q!k!0YARMNiTwAvv1R_zS)NXfix1C|
z+rV6>@XfeUL?t1I`x~E;#_4IHvTxK}mv*-7``;>JDlwVoc2EfWMaJ8cC6qg4Bfg{<
zx+!u_xv|I6JkKk=Skp;0YRjUQ=KD@+pIIvoto_VBtKizAJzvU$l07<gT<<3Al-F&W
z^kTy9IsI$-EpHh#yKFR7-@qHo*ZPscY0fm0eKmLcYdKt9j%@F8U1zU)f7h+ECQ7ri
zr)&OXuMVFOU_6~?&JP_uz3YBIH?cb&zr{C?WwqJy4i>YhKi@c?_a4?(vf+r0I~i+J
zt6{n0prDlEiW3bvW;S^%OtX8|#?RqZ2>So|)t9sNWmow*tS-Idv9}2p@Huo+dh3!X
zov@Q~DhEqguCqS<^lo3eO}s>mk<yY|YQ4)BKMxWVb<ulVZ_lAuHbbGss6Im@|K8kq
z-ulYwsy|5^7U#4ixp3_ZHS<s5YV6CKIX}nPlwbb9IT5iNhkO{)xGsFUIQfCP%4fUE
z0-wyj=b;A_e+AuJlXp?3DfOA}-p230)1EUkE?LCEmD9(#D=1QQMSHM)%lC`S?owI@
zHubqCrn#_3u>82cA<r;EhAYsI?}g=&hRwTDnOmQHo3v+H#P*zdhZn5U)C$>lbn1m*
zQ3kKjOMwnoKb;q0$Ox(2elwOaw~$#ZVo#lmg@d9?w8gSiVOFJbj-12KBxMrK4NANI
zFRZ!dz2d2{MdHJ&5AL3So_=`s<j>;gZNw#mRgzAn-o0NJQ(U8$ttd6q`_aeO*Iz7V
zKkvH4l=Fv~i+`%rjPDnh>lp_W9N?VJBX4uY>!II#o8qZ5@{d|dEGNhI8=v!7;5_~1
z!)u{M>k@AlP0yLP{B-KPj~|Q0R~+m<d5uYOX~}idHo=&zZ|B1eER3DLHoRY#qgZ!+
z&d!|w_f`ohn#ld<yLwh>=KB)vRsM|=GBzIfo?w!<{^X9Y+xBeF4zoS{^l6<%`-A^6
z3;*QGpZ!*R^3j|rjSrNa^gEsSCuHcp{Vctl$ItKPu7Gc%)9v1Voq4|^Z=v6>=fB_o
z@3_3EhX3e6VVR!(SzX%obM&4U8tyVRbvn^+Usfa`@^MkxMCI-?7koa?Hc6hQqqND!
zD(h0)E$>V6tHP^qMa)fm5Hi)5r6%3oWV*hh2~X~S@n2sba5wX}_y4$cl)q+oTK|UE
z#vJZSa#kJ}Q)}AvwJ&eqaV&+&bQ#l2<26DSdzVIj`*h$BU;gD((`6ydEe=15?#Lvs
zea<0&pwalqwAidyS27=eeEvUvC6}F`R(W%nsNe=e`QX)y)D**$c&)EStql3ebLHe=
z-&Y}aceXnznynJ5_!To@f>r)4zh7J|dLNuJ^Gpp&H+gKD=zAqpB=1|;0Vah@;!+mH
z)z2oJef-ei;e*0CAGf<2#>>gd-Rp6Cq82@IkI$+*GPABcv5?p0&%O3>!yY^52D^jW
zCjZv0zc?dVdHd0vtk2?CSPnecEwcGxnvvFJRoP5y#)UoGx_)KO&0KLfzU^PAk=quY
z4^@jJVoe+KRlPZ%cRUN5CbTF^cGgWx4NjBS{9)a@pX}2X?a&j}J=U}Tz@CUCBV`5t
zx9Y|}cKeAfTj^7txaw%iO#}Z)=W|wite$kT)RFPCYw9bW%_gfC`CWc9$+zwGR=1rj
zi+J`4eBBgxck#)@zT*os1ntwzWabFi6~9nWNX#hjTE$WK|JnYU^Mx#!trSBeq*cPf
zrAZ}~1*!TWl?ADW&ic5o?lCt<JxOh86=#fa`0e*DD#qKqbT4n59OilD+R~`$*Rs||
z6gmgEtXkBxV?&Fpf)1y3i>%W;m6lrdhBv#rPA$v08nsn#Yu4IrL2Gl@o{qkMYkhh}
zd-eWmwxhTF@4x?6_g#FBe9F1XY5zCRU)&*KIZ2??Wzy4ar(>t(&Um<|Zu8`7hlbtL
z<0l;6{koayZtHaU!hE4S6@uo8CI=&XA1q0mlcMrp@cy^;YwC5)=Q{|bt>4!*UpXmW
zaGqT9$H^7k_RFS43q4cj5KPoNB~+{VEAUIkyET^bK8JOdou6S?Tp3UhvSNP!dM~r`
zhlg5?`ZaD&Hs2&#v~+&IlNsxOAw7N$F1{(+A)8aApA^kJ6|VRB8~23M)7BRkZ7h#b
zoxSbU#mH@5>n61CVm~qGkPXuxn^eBKc}1;3cCpW7MYOmp{<NPl_IUVo`6uomjr5b#
zv{K(0rEa)$nLmG$d)2AtR&nFhgP$%q*e@0OulG6lRE^VJQLUZReMD702UoDQRC2!g
z`2C(E*Hj$~6OR4-NAe9cmn$swdMWuze9ff9K0oao`O_;tm4<WHOs)H|Rp*5C`Y&9c
zc!al9{_=bJZh63|kWU^{4;**a`}`@GjeUoE;=UOn1uO1P;7IhcZ#_T%PWmaQr*`ia
z#xWK>6^Lb=YX5bz?c%M~1qBXQF5FwNFx3C&@0C+NF}#ngzxAL#P4thK*&970_L&d#
zyY|<)3oN>*>>}K`QaNqkrj`h+Q;AQiip7)IAL^YH7IaZqsNg0bV%SyDDd8L;uwEd!
zGn~`ANi9xUB1z_>)j=VlN4$@GlWY`~)dVMZZtvLMsd*$MNy>0;Mf(iLJ>F-O(u9jU
zb{_eC#O4umlHJCJ9}|9bY*3sgc)C;ch}WTM5w14`za7a+;(h3_MpdS>v_<$M=f-(A
z9CB3Vbow4iOX7a$v_mMGQ&iApji5~#%g4PdqD%~u{&cKW_Hr-j|B-d1S@8euGZ%Bt
zpE#+^KWl&bzm-313mJc!&2*Jp&Y@K!V*Bcwx0(Lk>Yu+}Y%=_;b1c#&GN$fm_M1B!
z)8nS@t-4^;TGwp%;nt<=ent;@?DM|tICVX~KK9_IH+h{?pFEU4nQv25{CW2M-uZ@&
zvjVIhP77TXqI7oJMWg+l%4%x<vrJ!}NSS_ex!V0%e@-<0UcOB6#t~i9zl~>{&PV-o
zx94(9%m044>dWZ{YvFmq`2o%LKit||>yplEm95;~Z2Q#j7Tb*WiQ-0Qf0_x&m}u5X
z-Zsfm@Ld1rDnkIX$lXoSueRJz3E+K_);vFMGVgNL6}+7%-zfi?QS$k{r;FG%+3l-Z
zH%+<wkTp^#>cDLy&zshV%#K<v^=P_vvUaBCx1hM6HXpNmY}QLL)Ll(@bi1vyy7<0f
zNm+u2$AU#_OZM&Cwrro?4n2>ogMalh%nkB9Htwj~B~ZwGdFJ0olkRxPzg+N6|I2os
z%kOLE*?reH(mZcl7QQKV&XPW#3)!WezdoE%sa0b66PvS*C;9SEr<`M4H}=@|oHbl6
zrQx5+vHnbwZa}isq8*df1gqAc<%v`@Rx_~BFxX(WV&BSj+cqk$&~u178JnK-q|LW#
zX`JH9hyE`@pJ}Ej&72+<o*L)-&aHQs(A<jw+J<E>W-pkuO3m<M<5aba%6a0&e=f}{
z)-}J{y2ig|QJQzoE2X7>ul08e8{VI-o<D2h;y-F>Ki^K6H$ndJey=}L``V=|#ZT{F
zu`}>??aSXMq{=?&^!I*xsaW<(bOTGu)!?dU+isaw)_yl^-M+Zg{4C$+e`i|tU(J@T
zw%>ohMZ;n7!&SeMPFVh|?e|yzFC=*=?$?5AYpWJ!mfqv~A@EFHk27A*$UNosr|$-h
zb6<$6S!imN2Iribbm8j!Ga6nA)~ELWtv&y8*J`20J^#a%u5|sG*;btCR=m$eCMVs;
zc)Q18hu&I253xJ>F((h`rOmCF@GIW@tbdoTg|EwNhPzs;-n)zbO<R$YpnCZHwrBAV
zFE2UJziP&%SGQhO>AVmL$T)UAL*|sL<7)Yj!b<}8o%y@>@=uX_{{Q~HDYubR;{BAl
zzfj!x0#D^_mC34HZ<hBy-(jcb-!ntlYA#=i!pTeOnoFAlRdW113a;fyeBx3!;oj|P
zC*q{KB|~%9UWeOUvbPkU-^lPd({9K4U0E_c=IwT+g1-GO35i{1h1=IhEsC#O(yJQk
z)8UXJt9GfacJ*G~-FijWQbHW$L=sn3{_>jVb9>p->&q5jJu!3d|LDWZ3tF9?JnFW|
z-6|8K922|f=f|AAJ2tF-;u8JUa>AWIS1R|3<lLG5>DiwQFNT%Ob8{VUKG(VU_I=Fz
z3;g^AnLdkmyosng{c@V!gqp&X$<4i!4|8Q2sl2$dD?(O!zovDmf9B4OVk%$eO54tU
zw&!hzx%Vz!n=a<pQ>MOJbz<?s&YafdKL2T_C$4Bd`L;l|?%kg+G2ZSk9bdG~>-cMT
zbz$l%ndvcIbrpX@oW*P0?#BI+%un0Owaq>6u9q#NRBZv<b6xK%cl@s}PD?p|W5NbK
z#dD^~Zw0oLc{Byy{A=;ZKubt*(RQgAUe4BnprY14JX%{yrCuq$oF>P)(aK9XOlhUW
zgth;cZB)K@LXY>^Z8qD_q0df~v_1TL^o0E1JiP-KJL~o<Jc|2ukLThYd&wh9<S!>}
zH<P;AV|3!+3~|T4eL81P+**=<-RiL0jfI;}`gkXo_U)3jn0oH|nG0_WmTtOxcdP4O
zhr8B0yVNT-o35^$cttYi)T69zwa2v{uev7wL;UT8D@WgXTQNL;ko0%C?R@!MEB@)4
zDgQ2(#^r1kt+|rDxBblJp1k=ks%n9k{Z~9#CTX6ZAon=d>hOK#q|<ySqd%K{5If+%
z;79BA`))i^<^Ro1<R)$lSBu)|u;A87T}98;z8>owzWwE&a5u2)e}nz`n%Wz6x1Q*h
z$)CSp#<^Fq_1c+Fe=ePx_U^8&1F!Xiulc$!Taqh%*4>)3?Pho3xnmaLDamubXMIx+
ze7|^ehDlWOzGP{Sc^==d*0G#E5c=IOElDbSMrXLH6H`MP-&KWkcGHTrte11EoKX1|
zBk`g0PNB!jX^hK`ZhH9WnzqK%Khv_m<fin>8OTk%$eH-CsPahJ4WGTG$(%;V6FvV3
z_ik-^eUH7OP~J8E@5_+Obrzm?V+xa${(st|I(bb(=-NN+Uq#d$cN#2j{<fpZc<X_K
z{}%t0_P@8W$@^>UbnB1XKFybGcl%#k&|!7^a^=UM^Vd?tHrQRs-uiojiQ1IkmcMpJ
zrtIqU%Aa|+J7Gd(wExwM>wcT_DGObY7K!?EG2iC&ezRn)|7`woksnU1V*U1Px@di6
zlkc;&m4bd+->g@p2;JoKNwr#Nx#`m8Gy4ySZVTRQzV>g$#Lq|F`G5P>r>8as+jcok
z;n@3Wkw}|dPtCL4ZO0aFX?yqJ>yP=8elynpyWO&r>+ttKapyc&=(?usUUYH!xQO}0
z8@}4F4CM_g{ms5Uxo##_-t=Vk;oH|e4jP}jpSg8$tdGGD$ApQS6gMsE=+*SsZ4t`Q
z)A{@D()arz?^_>V(S2Q<ANiKy{t{nBk-xT5rZ;{zx!45g{ScV+{&CuDolIr(;sc!@
z#ogcbeD&+Sd!#?H^7g@(AG<nMY~bnKeoUr)(^=U!iK<QQqFfVwI3~Wy{b0Ic^FkM<
zR#$c{zN>SS1KW3Wt9#!yzB)mrv1P`T3q0Ew`Xn^RRc_k9_s`2ZduNX{r<L!wPdXTx
z!PJ~lA`$tiy|6*qcWT!D>6~}-T_b-pN$X7f87uOi|MvYCTUfqaZje3fZOX7wpi}c>
z5VNSqw#;3Zje`$F8Ldz_Xl<#ZXR0zWbmEsKE!@AjVlB_T`x($s{QFhQYsrSspHwZP
zuigA}^WU}4A5>-fUQD=t>h+%@`*W|(u0FL^d-)4f7ysg~j4u|~6})(;sa<DQ-@MRV
z_084iPL}Ij^XHXK_1#kXAby&b!&3gUi+04jUCI|SjSStfBe=fgca25<TxtDjZp~6m
zU#1*w@IG>}(eStH)cP5`d#;~XdAj!Nmfi>^p*mxGzV?3ZB8Q}u#vD1DFJC6<<*Sso
z8{O_YcW!-j*bbg&2iqT(#qGI~cUR@X{UtvbzR20L9*w@+pq(my%iwL&Sx5fM2VQ)8
zpg6~3w|z$YV>LHdB`)PS)>MhFMr*2Gny#r7TyXtC+~?`y`*u|b2!3F1m8sM=Y5Ts>
z>);}{hpEP~TdcQQZ=8E)PvYV}@z;jCew)NfB@5}uIfYLY5u5CD@oa|4ckAHf98;M?
z8FAt7``7qr^=o&(x*4@qW*L+HrT5I$&r8?6UQ}6pqs@`y<RX^cW-MG~wtMAG--*Am
zdZsrw%<==@W#s}7u{Y0sno6s>qIaF2aNEaf_4?~q8wFn_@QdY1PD)bM+E_AKidA1K
zzd<<C{qrx4_(C@yw>f6ErOT}p^m-S?9uA(i^la-{fu-Rofh%UNxorNz>Y>q<tW%SU
zoFW%bu56AdX1^@(fbsD9kOdaGo6S5YPnoCF>BV>HMzi&gmwa)|WGxpWtIflS|0
zQ`NVArb`;TqV9Sf@=W&s=NMkgpr^TT1Gmi<K8c+PI<s%^Z+pEgadWYITh^1!x6U~;
zzOefIU)k5WZ00-T7mdj)drzMC7G80i(@6LAtcTsLi!$aqaZFt4)bU63(UvE<v(`Eb
ztpDM2f7R>5*}=aS`2LPO7SMVx&Es@qworHDnU25ALI(>i=g&XOCb3RnyKI5Q>fJ9i
z@~>@CEf)-^vHNdm-^xGr(3ELAm&@wkP&y=BK2!XmlcLoAcBgA9Ca=C!&iq=LQ}wl4
zQtHk+Z;=UHpIOfDniw>9;%pYLlOjS>BtO2?<ed2W(Hq0S@OwfVPjFi*GxfT>UQ%0N
zQ0o7u^F)^DiH+yurI^0WZeEeb5%zBF?k#=NE{kWTn-)u4Tl2m0tOd^@(`3Kt3y(Lv
zer)6`@l8}b;MKI(^IN;Ns~+C^Fp6JiLev|^#9bV+tvUbCo%_%9DL8P}d~sIIlTo#t
zZw^<w>fURYU0}vse$U=+-YmIlUu&)h=ZyZp&zakHQz||ox@7YGKSv#kn>l)SWluYi
zpW<b_X{&DL^4s6Pc~$v~$uXI<J=*q1H#L6(_al*cVO||4f<#r9Y;s!i*g{^+{zrbd
z{hWBFq=IjD*(Z)&j?!~mB6NJ;#_(>JTUKj&f-gV5ao%pi%_-&cKZzgTond-QWNp}_
zRUVxlM<3>gsGsqE7|$j@XKw3}&-&3@H47#DH|xDyy?e`HhYu5WbL>2p&#`7^tc4cq
zhC6+GFU0OMd{!p;jr)=Q!K><0*X0{r+4fa@;5a?K$X$$gs>;d7`5S}S=d1bdDKCt=
zWOlaic+S7Z%SD=Lj{|z9|I@p2P?l+S%n!Fix$2p+SyF!v-cdQHcTVf1mfp?ifa$R<
zIhNepR_t4_Y7MLP|8W1<<P`Dg_g%{uaaQEqTHxK(`|E$($7zYj=ISTB@hoa={%e2C
zLG8tsn?kw!%-9PeU)L`B=IHp=(^<;x@!8A&RxPl%lsNV`CC@P6w@Gx!|HNayk_CxB
zJnGc!d-<R3y(m)0cB?PsMe-I7R^|M-wum{Vi|$Hpd7_wqu=?Y=%|8QoT#EYqEJNs4
z>$QJ>IBGk7FK&0<Sht5$_TFcU&PA8vU;flcNRYWaD|r6RrpIj-4~>e~Ek10MX>sJ0
zbJ3@o8+S8ZqlD{>AAYya__Fut{Eu5qc+2->t=(SM8P@1%(84*haNYus<;#L6FK%1<
z-?S>HeX;sm*O`UqCWz{^rE2{Vo$lheY}y<)$7Sy7+s@73`$l-)q-735D@s>>%-MHq
zLfitqF2g0yk14au-B^?I$MdA$DX&w($@MSHo;);TeclnHb1U<p-M_XOI;EQOja4_3
zT}^+U4Xu4!+Wvs0H2-N}-5SwYDdlI!6Bqw@_^Em7M9xzsHrap0Qx)!VPkE@N%*VH9
zU;LKc@^fFBKQ=1}<&im=bJBX(ip=VJZgc)F`+Q{gpVvVJx?fiE=W)L1+PQw;tP8UO
z7O&#e4|vdHb3N^A%4*jQ75P!_pDeG%ujy`*nDcmhYRl3*d$}s7TBg}alAHd>?$vG0
zD=iM_&EH(DXuoO3c8!}idBRuzyR2Tn;THGXlUp~Qz4&t@o9nfzIWBB%3TCgnjtQq7
zjI~L8#<S;~#Gewa8=ky+zq+f+4!^p(U*Y=WrE_<AIUBd8b-z}y_gsEvOPcgi6YlsI
z-=+HXufJ<jn3K4yGVy%J(Odh}{aDiPa>=@GUR`qRYn6V4Zl*|jobmGIFUy`ke<aiL
z`*}w&Gt0TTY2`aDW7C+8TUox>^NaoyI_V#idVc>|@9VW)ahsjK{|}Qq#C-12rrzp3
z-zsBx{)^T%bBoT6y?b|RIK#}uz&(?Fm@oS(g>*}6+Pv;PuqX6Kvu)Cm$M<$9DeJE^
z2xywaW-B#a<cFBGH(%#ze^C?jc=yvgT63;m{F5|!;XwlvjjTluOM_piWJdgwkiB<r
z`N}I|YEEldgiH%pZu~Rrpu@EChelgxMlq*&Em)v%d&yx<Gcj+&>nc(lzDY*iyslP@
z74?@ZE*IN-dh4dh)XJSxe=R$HLM7>Bnv{xD^vzW#{VyN&aA1~N+m_D$IO3o4Iomf&
z)YNynKGNrlm)=`zEcfi#r{_0U&e^=oVK#rI(26ss*IwmUO}ZB}nd#lG{;xBl<}Qm1
zo-pCHvZFq~n8-Tz+<Rqf((;1Oz5A56W|?w#+|&BJy7Lq6oa!$4aZc{j+VYL*U+*sM
z-N}5dr)kagEla-U-rMifbnFvjWM`X2$HWhl<6;7D#|NB{6tdIJvix{BHRal!pKohd
zRS0UvOFekA^_y-T<7t04*MI6ZyB2K`YT#M3!Qkqr^>RX-{?9HfIM#Y3{)qg2tvRfx
z`y$Nly2xG6WH0$@t(Lpb()Kg=oW0zwuTomF-a0mJoF2E*`+Q>@XO#0I&3En}7Hkr>
z5q`4D_ureWO=?RwY+PimFWvuj|ADyt!*enNb?og*_DY{Q?3qyb{abceYCzTrv!D!}
zs@}i~ixYbrK7GlxS7<E@&UUUW{kYsU{&}({|55p8F7X}bzAe7H{mcbxW;vgrbA2yj
zPhR^tX+d&IWSP@~k_+uETW>v>GF5fun|<4A6>~e?9X5$)ZPM<l|J!AI<4P5?USP+j
zorV`r@y0B6ax%+Hkac$ojEnS)jQxA>65oPf{c9$DRKBh|Z+~TG@KOUY_b-J$^H)4$
zH_kJ%I_xlgqw@B%#oP1uvFEh%PJI1DyS!q3%XiIxXLlQ4ln$#cSZcF=$D&SOoo{^~
z{68H$JoV<L*&Tl$UpOhhlzpN4A{~?4ff+vY6l|8>-t<+>>Vp6DqGF5cw8QTf$i#E7
zrDm=8^*8s;FNgQ32J&K8Z+)-Yrm^d4d;{kyqu!Yn7U`MoTW;Qv+xtr>mhs7hiL02N
z^R?&1<Y{fGx7;A7GXHR=p6nbOp|Et7RiTb~>lpJK_P+mMTP4}qH}7ZQF7|ILqo;}%
zvSxW4+n}x7r!ZYC<+9GBH~UkrRAxU}wPdbBB=5Pej(aDb)hv}%*D!IdVdRt0d-A%g
zWM_bbV$n;{x+PQ1--J|ai0^tE#Xb4p!W%oo_T^vR*`R%a@2#b5>eip4turI9sB5&{
zoU`!CuLnPMT%PB&ygPI1(5$Xz_N?j!l37<$4^G{^NtDg{q0G<o$~AYo|MBOk&S~TS
zc6+}4jrw;_{?3f%v~l{F+pF>NbbM4mT6gN@>$87Ntl$4m>aTV9p-6+PvM*WxEGc=<
zHlHbc!kY%&Lb+4C|G2^nT9(E}>#0XuZ|dBbnKh@Y`BYNXqh~uen#fO=dCtptWzKEB
z4maZqr^_cE4UgtuyXDdIPIGPUtfc`nCdUWQOuh2j*!c67$Q}Gv4DP3_Us=!A{(ex<
zAd_pdwnp9E!e=$t+kURTyWQIB<^K)0@0EYes(vn+`F(+L^O^6f!s-+DrcIji?AcoD
znn<p>rn0}iyo+9a5pTJ?JKK4??VIP*?7u8tQn$;XI>D;g=Z{5C*Zc+B{uUlDi;G{q
zTHI<|)rzb0#FOfN2Vbc<d2aLG`SYz`eA*~_zwdfo-UZut@45^QiL<oWwrq_oaBG_5
zX_EE$qvCN6f6o4>r1Oic>#dfvTK#y=I_0TvirMP<>)ib2%<(*RM`)jAUhu2=-yIX{
z4PSR9ZhmH$e=YY=ZNS6N-yJ>9ZlB3^$ERY0@n)O0Gil#1mtXj;pHkM{lG&-PRDN`I
z$i{~|=5}0Ef6eveP3IX~(Q3Y}WmD%iW{dp2kegt<n1hcm=nC6)(<@g@uC!m#DEMa~
zl<##QD$?%N^0}+x7(OQRKI4h!wB%gQvt8c&=wX5AiCYAxemqwF^;zf=A(5s3dMk8l
zr+&@!)#x!(e5QE&Zdmk%rp0wyJJWh^^Q~`4_0!q@)-r`vQtRJd9_#kNH5CWfUN`nU
z@+kg^@bylE$bXCFS-I@DEbOW7&J^E&!lv3WA6vTIf=F4igj=>Hx6_xY2H-*mm#
z{0ObTYo_D>n!D-i&D<MrW!mqq3$+mY6BY6LTgh7XpZ)%RXSaPiY8)GKW>V;y|B?Ms
z(I46r{w#kh^{sl+wyLZz_K%sgzgrlU+|l86w&p3$_!r$*+r7m5|K8hnecJ60cVzv0
z|M*mswZO$Qx>;hTcOCbZp9v6te*8o5vOTT0A9ij}S71Bqr2Nm}{Q~pl_r>2WDKoU6
zKmSz!^OL%#?PulH^Ya@nuHN^9arYzUeM=ctxm`tGe+#yMwa4}T`QMkbnm_3L=|5P{
z_-XzG-auc?xdHs9k*bv*H8byqU;ZmSXJG-)-9uaNOV<b`CyC$M@ZF>8PcX*|hAk7e
zCoDKIBR(SP{=L9Djur8KxBQ>E->S6;-z8{sYL8Vxkd>XvqryC&?HtpO-mT!6Q84Y{
z<qZihHFq1?39?id&C`gP8+YbMcfZ=i0<CQSm-9~4Tr*AZzItxY+(Ux>`9(RGn9|F7
zkLy+6Df+NJes=${s8EN@v_!tI9y9hno}zegqt{j~*Ur-m9+zkENEN<0W1FeJ<=A$u
z4L+9?+ZFcpOK>UpUfut-c=!4*rd9_NU;7?9c|d#C!ri;yw(&5}nW`ZD>B&Cly<xlo
zg5i(+LpzEuzG!)~;*7=wmZo~2iWh9v-^?ea&Hw1~x}z-5adYbL>2@b3)Xm&?xlzqH
zVKt}Ew>RDQW?yPepLzX#%})oeUeC~p%O<>vWqm(gPlwaYQM`1;)_29)zYgs@ye}^(
zRCcBH!rQqScS4>`e%rWo_u*h|vvX4qUGv(%?!&Dg%7)LAt%Xk?{`lj`8>RTh?Yg(`
zRC(NTj9QTWak(r1ud{Lm(|GPX>g9ghpViuEo|gJuCh_Xhg@qe4Cn|jGGW_ZlXjs*I
z#`cuIhv|LiLv_y4=Vtz>36qpw!RKMZ@=8f@LX@Z8H&g!OXP>VBI6uE`&hyqc6BV}D
zeK@pdcSFmD9ST#mG)~x`<E~NEuwQ1gPGzYeU(ky7n!3HFoZ6e2PC02ae?KB8?q_C|
z7$2DXb!V`S#uh^<^R9W7eC&lAKfL;>u<e20(itu>(%a{(RKA|-l9~MBjl|OE<;$B+
z^z{hk`U&M7R#a29u)9#JB)z_ae^UL4{XJGUt|<8~EZ8)EcI=|pS2veFDY5zK^5}fn
z+|?U)sxG!TJY$x8S8PJTs-KF5VVQBfFFs3iMVB<hIko1bmIu!Cw3xSlQqrS08+eKf
z8}y%;ANjTRX)1Ss?dp?9pW3r`aTl*D{msI6*`S@{kL0Y^O(%Bs?CLqx@#^TL`cK=>
z?mo&Nuw|n{eYt1%kuCDQ&)?4~@7%0iw5mwUcYf%+Q)jIDpKaiM&gG@eYs$2?b;h&>
zS_hmCPuOzx)}o4zNe)hKHM4HDPTzPk@uc8C?n)uqgvU`qJ}0+YXmFV>dl8iv6}#x<
zq-hKJrks6X^EoiAtS(K{qwAGe*T*eeYB(32$y|QpREyJov3Xqyy^5leskct;*ki#d
zbyolI&umw(NmizUDi;r$AHDhW*O{hGlXgvd^=8g#{u}l`e(vHawo)|LQ#Vv^U6lK)
z;Lh`;+p13zBY&Ixn9XK%MPPmRc~6aLJf#`WU0z)9Rm<^r3JrNLm2!63v}NMl+wcFh
z4BchhwC=slMxF!KP3ux6uU3R<f19cx7$37;%Rlm&QlRB8{SQTxn^~mo)pyMZ>Yi%$
zv{N<nse&at`>k(xoUAS4%kKL>-Fmgns&U!Z;6H0vXPMp*bM+KZZ=BMz<jWED>r2+o
zS#ovKx(C~ev_ItRF;WiDW{*oxFy5B&I5>ADTj9Z<XS$3Yr%qUK^3jLbi>ngs|0Um>
z9;)?dk-2rM>xM07m7T@%7&rb}5yu?*>+Vh4No(>xI}}dYYMDD(?&&-u<_`i&+aA<8
zAK&4WHtp*pOXGD>|BJrQUR_=v`RDBJL;pP;9MxS8O|~>#@v~)1hXL=6nMcC(XFlEk
zcm87ct^2P0FUsz_m|>FACwF)L%utgm9-}?q>KWqJom0ay+l+is7tyoLrWO_`r_Zg8
z<ctUozqL<ra`o0O-RsfsxB3Zf-Ky7<Jkf9UwOwN7A$Jb2PSkC5h!Eb!AT+H-MV<2|
zhrqNA$|r<aL`3hb4ci@Yd*!uvYu|3!YJB&3OWgmN|6^`N9oPH+r}o*&WVr|TiuZm0
zz3+SV{`WGKH*Rc{=s%v^bUpW`)SlXVeQOgH1B)s{*667I=+_Q@vhmL0qxaek-kZ$(
z&M$1RzMv=k$>X^;)4rzr*H8a`SLmO-LCzU>{(}3glXtvd@%y5=nRqRy^IwC15<Tr7
zE5q)c-ZOj8db!#AAAG!d;>euJ&ws*~E_t@(^rJ^p9~~FdjZ*15C;0z^%buu36OO8O
zr!SN@QvEdLii+o^qJ0&!U6jI(c^?Uy8KV@p)iUM!F5m1Uo3H2GQjAmfoVLXM<WWV*
zi$#$=77=2~o?#x(I+kj_O#1nKe@AlAq~uhuNUu#v`zl(UWK_eJcqOG|OjNOUn{>+d
zgZOf<OOsQ$Sbs(81a~~Xw0P3gNf8^3Ty?zrCZ-=LdUo;tQq7ZIg;`ZxRr6ZrEm-Rm
zB;w`jvr8vb(@o4djJ>u$P*Q&BlQ&u^FE(1ZcHQgo?L7Hl&(6<2lB(We9aAUesHUq&
zDfLa&40;-9c|z}&mT9aRm!h+xci-=myY^ee3W-mZ)1I#3_WkLiPm}8(?cUSp<vaCx
z&pKbp+Sh%xGlLW-dO!a^rTPAJ(Jvl<lw*$^OzPAB|4*Ig@z1`4|4vl<Y?!f~C)!Yr
zBl`7b!P{qUTd8$<JXtO*F|Fli(qseso+E0>BHt9uSQl?>-;g;&Vs_)%BXY;&jy_8$
z-dH)q-b_Ba!M08J2-C*n8>imT$}#=MBi6F^;I*XJ8@@&uY!kfQl6%Z)gYFHp9P4e&
z+(!>>n0rGn$9x+*_md0@w%=XHKZt$0b!<ZNs@iA2zVyzu+m`f3)$ZJNPA1m)2M1Xk
zYmD`3?!5fi_IuL%-DZmUNe_-N-d|pOfBqD!m#_Zb{C>a6;jQ?)<C8R>ygiw%X=wUo
z!TILgV3**WV7rB(UzYdezN!s=EhqK*b*okX8Q%Ss>yG`^n<M>U=KbRr#MiYK-*jJ~
zUm5Xs?RT}HqfMcboUKB+76}pZ9h?s9cU}w2EWdh7+R=8!wJpy>ZofER5E{8qlV61?
zb)ib(QMW38A5XrDu8t#@i#A>1+}XmkKk?z_vlchj=}vsL`0ExC_3%{-*Gn$&<=7N>
zmw};SN8g7BAD3AKujHMRm|*<nU~cIjKF8I@)$2RX9y!RDyD>)o|KqGvXVhb}%q{XN
z0_}GN<XyUTFf(oWckQk<hm?ec7B8|&W&Rr*`}Z$r>_w&cA96g+$rXn`*epGJF`Mg=
z_oh<UTQ=2(ysKHJnWg)e`UhIRjehXV(f)^lsOnTV*RMyu^H)^;`n`0op;G+%KJFRv
zJuN1Y{M#$^gVYvtN;KY0&|>sF^7gES$GLM;UxdFDIMRKr`&iVIn3UZc<w6$t3%@WC
zNU2q|$dWKpI3RsiD0mz97UK`5GItDOHyz%{`qbq0?)Ia}=FghHY<ly_q-pcZu$_C^
z_kV0Yuef<ynyhQvfv0Kz1Q}m8YR%op8Pb%+W+I{OyLPGAEZfD`Za?(e@uu~;?fRzE
zXT%xTZ}`sb?9h^xox-=f<J*g@_)XjS&Ey%+a5MkD)^Xs|ja2XHo4juvHQBV(>UHV=
zv;1)lcK!!`pVa0pV0_wPpsh8X=SfLc8v~>Gp_2m5!LAQN(&~M(t4f!d9&(!SKJNW%
zITrEG?bB?nop;C{5y|nKu=(@LJjc7D=MQ!_>?qNw``y;@!gNuE<2SC^*K*eyn=-WR
zDax(*{%NW8_Vpr1nIlu9nm5_)w@dE$p?L1Y#O;3)Ol91wUOiCadR4}8;NBXOm37SG
zcM~56CYyHUoj7>$h0v1YM*}^M1ZteqF-$jaUVP@|Z1aNVnfnFo6d11eoR5C6SXETx
zz{)okWjC|RK82OBeg2eqDk@Q{c*k7bb8Ov@Ej=zAxN_m`yEO&Y*{5S!e(RlUeQJ~Q
zw~PHh<MH1IEbqy2+2!cD@2L(yA2GKrD}Yr?;(=Lcq?FfNlixjQjhdHt96d8dBsxrT
zazjrg3;RsA)^97VOz+;j7F)<7kU#Iv$0L(>JIxnkZJgWwXubN=3#I<im)CaAyTf<(
zLB<Tv6YZB5p5HY6X6GW02_bt*r*N-ItoUI&k6krU_|PNK{xi#-y*^xY?S0jkmv?pe
zHzaL1vG>h`d-LX|UtYH;cLjgmK}mu8e^%|+U^$ZZ=4Q(OJ+D%ZU);y7U$U3!RD-Mq
z&$_HxvB%>Uos++{MZuu(`L?r@PRH-LeDD977w*q2A5_?Hx)*I>G&gDa>mMhZN;aI8
z$T-J-@mTG#LrJ&x+?&Dkt$~Sw$Gg*+>38aB!R3D?5)#@L_#B?M>+PmhKc9SByyA-N
z0X8ij(-j<7(o$~c74Y&p&7Wpmz!P$*Hvd}XE<fRYtObm7_{INtaUSwcu8*7cfu-ty
z>W%}?dCR1}TGcH0Tyy^Qsw^uT{e>Csbst|Xsr=0p^E*>R?qJ=@WzG{^-*1bqo6x|0
zqx10f2O*z+y^iC(T6i_;X5c;E8ODDLA1VpV*XF;!t$el?<B?{O-J8O0hPq1o%<`Hm
z9AdO!>AuDLU;j$|b!Fl$y@F{wwtiANEuP1?k#%9=t{*~`$D%@%FU<d|&ZN-U^{!L?
z#eU|5C4Jn+cS^2V-#*J4#kP2*V^{;rRsj)Xh4mh5&0314s84q`>{4<s>z=Iqd$G(-
z(PvXO)ja>l%oF|a@1*@v6ZCa`bVHvX7h9Hj_06u^uQNl}XjwbYKl1iMQVvU*vZG$U
z@wJ6H77nr*k5x`4u3%p~N9C>2>K@k61#OIG!8>o-F&w}2p{}Cy*Q~7{Z+bMoKQX@~
z{m`8cXD1t!N4)yze16&C&;m2>Lpz_iRir-A^h)F~deT%gf#;jld?Ur#a(5F}Mly>T
zZQuUbVcFT!+J0viPW_?1Oe|TtK=Eh369=20R1|NT%MFJYpB0x(*p(vNnctT^&2n1A
zy-fRntAz~;{N|=1UmM~cE$(%X2s+`HdVldP)+Y=qZQpH;jBDPP1j+?Sho&9+ed&hU
zW^T)lX?t!?-k?-4zhs&QXX5l#^S;*=Oe<$O_f<0Tw_Rw!#<*>9TjjoP4b^G1xydyB
z|J!H@g(_a2)cN;PE9cm$6x=hqZdqmOSLJx#EL}k%LR-3lM}32r#e7j`g<skg+g|Da
zU18w=^4F5QiaH6iicNvd4okiL8q8dFC<ZUz$i0VyZ@L8co6|4l97X3W+~|46@>s+H
zjm9GzT=rj|z_&>t%H7CI%lf#_9a)ol85W5z50?hX$~I44{m|t|gWi{-&yIbJ!KE(p
za(VZ%uT5{{m)IehZ+ulknxW9?0{5z`&rDwH?ebsHEK$&;_4&0=kAv{4^LN+xGM({r
zWI7czt@DHI&(j*;xYHMIKF{yxBDuHY!gYt~+xwZ!dzU4r${F8n-@MO?A>~Cf!y6W3
z!Da(FZPQnd2lh%dzfXAc#%Zsh-xC9`vkz0*gm;`gu(qT1tA+~89Mu<0U)#i(59a-y
z%G*|#$gIEV+x!Jd%muA>(jEcV(l}K#4tPCTJ^f*g?8C3ULE>M}-@5-Y<%Dd{wD<!Z
zZ#Hat?tbfc#T-d_Q)%zeX;wUCAJl$)Id$M{+v97_52O!Sf4Beq<Xhz(dwn~%6-<@e
zOtv3OVpy8Cwe4BUW4ZJPx?F9K|JT+((A%TBe$AIVQ4*(TWf(p2-pt#1oYU61O2?yr
z!lLhwgBEWJx_9{6LGg?ke95z<q;jY0oKdXlwmUw7{gXpqVRxs;amRVvG>i`~(7&3s
z^`&Olmqkj4zVF`S^u4ZsN^zQw(A2{7Rr-qcX}v4<9AA06(f*-Iv%n?x_X@nf1T`)H
z%hlB+Y~w3p{P4hd!vCx<j0wvuEmvp+uT3$(^KI!(h96~)#n&93sPkAmPUoI-OhtKe
z`y!G2J==eYy)qG8dSXV%ad)9V4|G%JOrEA-F7>#txLDoNJF8FQa-_xW{>K5O(FfJH
zZnIgws7(H%>I&<quy;F;g-S}TwX;5L&Gx*ic_LF~cwf$f2wiL0m!5pH`Iam2o^3q3
z;+I_0O@pa(nKaFUL$>r<v9g8S-DGTS^^9?v{{}_>^NHyT1+{lstxG?y`BT@#v{`@l
zeF+(xE7hCc2`TJYI8XceWcS3ovYq!h<kkN3=pUAz5}ePH<~;da(uzmNnxo|pcFt$#
z-(073DMlfo@|Q20>4_&PC;o3w(a11izdPqp*`JuUJv(=tj>^tsJ#w2_kkLEoP?PIX
zjV+lSJFEh-baY&<Of6E_`+@I}Z2{ZM_jkGXUb%TpE!_KeW1kkkgo54eh7Fuo9T-#E
zOb#nKvl&HBKJ;+*@8}<O(M<yTw6f>!WEC*?Z_yWPyzr$X!?h}ISK0iHD(!7v--_)d
z|0r6;Ncc<099Zn0{WzyJ?5LyOj#ZcMt@r<+r+lI6#q-ZxJqr$ex|H$tTDY*@n|j{X
z65b>BZT#H_&Ti=593A0ky?2q`M&>yR>*iS2IahG;Jp1Ccw);gLW5vCqJB1F>j_Y%|
zwBy@dvOd01)sb2)Z!h`BlS{<?VtVM>q7Cg48xwy&K4h9uyZcYXNp_R#4Lgk%-no;~
zlXcJf_HTWzK4Bh)<@W*~1zeb>P&fZT$=N%(_Ujzm8<`h>l|S$LrXh0i(zn$<AD^YV
z%s)`X8lfwi#+|$|bT+%dhr;cfS^vwoUS4@G{$y|a<GAmyosS>0Kkz%?hW|c~O!KI&
zpG*E1uMjMp>{Y?K!Yt;&E0vRT0+;EU%Ot-lJnZN!ZN;}$ieIX_cj}!Ok2`m2vZfhu
zxvXA#>Vk_EN4|chgxT#|Icpm#1!a1S^`mAuZ?4M}-1y);OJ>X+`AXj#3|Xg6y*+Zg
zfNAPnwX=RqE=gsV6$K?$mB+7@UbA3l&R&r>dS!?I{}iyh`?y^Apig7n6Ez8QAxF{Z
z@KhD?lP1w7J~DM)*Aq0fSrpIYE;+ey*V$PY+LC6?DSe@!^>pz|X4dwJ>I-ETT&-KM
zZ2ij&_vpt3_0LNS<~G=!&i4^N)|66Ob86Y$k{dVL&Yhc8=MZDR>t~?g%DIvo^Hhtp
zHmYqEFI0Ca75(=&YufGe&3$3qdx9Str0&gmSt6ska>+^U>ea@l9|i38(hPjke)46*
zg(I;F6B5MlakLn>9zO8c;#y4o6g{o!Pt!9dGa9E~=Jlw%Vp<Yu!tra<8L!*V0_*K#
zd-(kBX->DSXTQKPx5w{OY24}czfS*p6!}dgeU+)KL8rn_jnJ)HbM9$*-sO^0zcI5u
zlHoX~w4*84=j-n;-FjMM)ZzVZ4fisu#lpTHwzB2b@N80aUGx0d@fP|12TLY@-Rt1m
z;=O%+-wvjY`n&vNMCI6QBj-+B+&*J<!ApZVh0~t*HXUDCue1LUSIb9*wJN6OHmz~u
zv-e2;P(FUYg6W*#Jmz_yPS52%{*o!YA}T15C*jb6ciX-k-kCeg?XMw|pWqzk=4+cn
zS){M~P31CCxH==<$K=wp{pMfivOjkS?#cX?ar^n9Pbbq?>|D@Lk+AW*z}Ne`PjX1^
zJ@H^y)6%ZVhuwGVv^g{-E97R0pWAzb1)J>|XFvSA<8#f*+P!*r1$YBk=O(>wdw8!u
z-i9mmaGmb_H7%1LOfCpw`l7i1<F<<9LDNoOTzb>v(~P8&9j8p)wzr4pIXD?wCLHAv
zoDimFvh-@|&aPeDVec1kdwUsO><|v0ZM>)cqueYR$K`D%6TDdVZI->1;UxcGFy$WS
z``M}?9BHo~O*kg}=}yshjvX?h@4BL#EuAV@WAfg-S3gl-QW8?uoEWlIbZIGzw4KXb
zD}kR^UWH_4L_B(sy?)Erd14cl-`&pEd~ti;ZpDn;2nUA~2a1&*tbCtU9(ud#RNKCM
zr<enk?+b1Ih+MrZCGt&5L$Y&eSn3i{^^CG93SxY(q`p2=TKDSF|F?PS6<@NJe|~Om
zZNT~d-Fur)PSJmo6i+cq*6^+5Dd)a$^61~%yLUxeW)~FfU17dfcdhBIt<oxtt!m}}
zC-s`_>HB@?dx_kSSbyd6w@<&@%{hIE<DR*}$=M#ZlP5;3(CEEnWVOI_ibTimfK8g^
zi(Mr4v!A%bdg1h8&-ZLA7S`4MJkol^da3xj_YZ6NYHJe?W$4s7Ha`0AuYF;?%>QX~
z9|X;<J~s92fv2%eKEZjFeY5X!250WxyZ5id>dVJ$%p+n0lS>(wEuLN*ppf>{%QfrQ
zHrt-?Pqhqt;=g~`_h^F3-#V3h?+fm~SL=ILHMQF0qW8Z4?$=Bzmj~__vNEbMIcVDZ
zV8cVD=?_AJ7p?r#&QkuZ{QKPe$vUE&yf4nW^vGrJpWmkOi>0!k{i+u}_0_)Q`-ex%
zQ#-^T$}fJ~-o|ToVg3Gg;Q)zp9~~bCNh#6eY)^O2Sf&v2*8a~gsdXFAO<9_;?e&cL
z7IU8Nb54AmU-?!3LRsg8+@G~uYI)U{-)((6<<*YzTZM7*bJ!L?Fp<B}zT&iA%#I1M
zC(hMt3)nF++yCMCC;3aiYW7@RtsBmJFKwMwJLRh{XY^*TUs^R1emVBFt&de+e>res
z`h(l@cCD)pzW*$5_2jle-4>CQR;|0|gxqo;+>c#z^81BFLNgd@0*)8VQa8_%og-d<
zsQuIGQs0ca7wc+YZrOS!YvYu?26sMN{g|MTUi<q??@W8sv)c=TPWkQKU0jqKT(Ue@
z>1|uS(>;v?`T=}w3*WR`oxR8^ZoEHq!PEmh51eH-MFcPL%w|oO+##)gz3fZ5oRC@b
zttKtkV9VUbki?sH!FCBj4rPrhRrlq71<%R8zF5)ji|QZcFCQEKyIkHZ`|H@X6KfYX
zB+5tad>kkkab5W9rdNA?ey!N$#`>u9!;U4nhYBV<@cNRm{=&v4p|W2)*BpEl%rN!M
z?(ls=jp0k?9lz6^^)kXe&-8`DuU)hJxqly+m{T0~Q%ov!>8o6onJZM*$1mu5y@`wa
z;>8xu^Q+~A4lc?s{QY^2!(sdW=HO?3#|!E#Whd+E-<X-pXPnmTlVrS4Z_@n6c)iD)
zMJpQIKkeJVY;s%X+{J}o52wmZ<<x809?%!Q$bhYK`CiA&`7U?=<c37}1@L7#%$}#9
zTzBv6#8skE^G|v>Fy8<1o-w{*a_Mg6#m?K)%I*stS|uMBe0TPA(;sUt-_GOYZxGvk
z+O#VA(X)d(vR~&dsM@e+=OR(Rw!d2zFT4|&oVWeL-XAu1{gMSI%#z@|w|c#ig_naM
zGgryaS$lln*zD3+tP!jl%W?U|p=Uwj?054@C4c*7D0;vA)*NGSx+l!+b@rF3t8RI;
z?S5EL-8|uBfNZI=0gq(*Wm60BOK(_z`8Gawh@84{&7RBGBEmj~M|g&pObZW_+`Z%O
zJcq-ZxoR0I4*7U|V65$3$n-UdZOW^wZI>*j`HLSBTej-kCZ``pd%Xotw27vQrirB)
zEIqpG(Iq9{#<cz06j;(Owi!8FZV;GS)*f7O=7_t6{H3SLb1qxXnfuGFwsNBF)Sn*@
ziE3)@lxcI;JldCdG_q@s)6RJlYg^<cN|app|6%l9WuHF%)5)jnm3OXuZ+zn*qhH>W
zXKQx-C}zAc{~r&RRn@7ra>|cC9*^vqEB$TVBJ<^jt7moUM@`h$-FW<>*kazRZt5$}
z=HGPh=~FVV_+WPD3G?ke-5M+t<anR>@Fi@&m%8Fasi%ZQLG`wx*+toVeAs4~On4cx
zZH4zsng7?C+;Wu@;@Hyt<~KZkD3{L@d-2t(l}6hOPR?0>nQ^D#5tCP0TMtI^T;aL!
z;%e1fooi0^jFTG!(=Xqg@yKc}^TrLAUKU8!|H<dK4|;p|ZEe@9W8u?|Nh?hV$T!=>
zKYeOOrmXeU``QNP(f#3;=NJ8w(odM5)@?MoSmgEm6rmlvFMaH|R_l>+b%SgA#s=kS
z-P<1f^f+ki-`cf5fnV#PI?KMucN3Ss)6Huv*wg=cV!f|~xrv<LuKW{uo2OWwej}`Q
zz;jwrckLmUjCp_dYp&9Y!#1dnbe0P7gX(7H#&$^eRnCp5jEGL%<?kizm%G+W>b<3&
zV$|Bu)nT`@-ui6SzRMlEP&<`VfI~&4QD9rngtxYl9!?yIoZgYA_*fS@WJgEuzO`uW
zTJPn0vhJ^6#K`WgOXT}}=C4z!+w)g{3LjRU6aG~DJhtxr`@0F7m-(e`yqS}gedX2F
z>@b_X$=NwRO)irr=uFG2Nt<irC}b5QRx?4grbD)-f=ydvhENX=_oQIf?X&+GCUbq}
zD*ly!Y`^Yvn=?tiI}77qWbRb|B67cb{u=iW&A+Nyx{X_prWM{ueY5>$eNz4ZKbt>`
zpFP?&ujbdwO6{daTVv+TS|(!m|I=xGZsDtk<o#P#yKog|O)KF0H$i)nN9xQ)I!`wr
z%2k>eFjGxvS%<39!&5J2#J)CLFL*9?b7gx$uZFjZ{{)WYA|cgBDi7TXCuw*qRt3#a
zn#!$O{!c*7bMvI5UK>rkKdvd*=`=$j<b;c7(w|8h`kpG>vS)-(xqR~3=+tP>eUvM4
zm5rjerk9x3Qm4YONh<1=OXTY3E1eDEN-ff~JAXtyBBUo|VTj7fqee>wvU?&$OOLyy
zMl~5N30FLJ+$oc5-%K6fQ=FBb&l{&US$HdkbbXqlXxpijxG6$$vdW5$aXMaV;{LNH
zEEY&#`p^5f&BB-|vlRTNb%;zZa$fQ0wEe^Ld;BJxRC#r>eCg!v|5nDY3F%PLJpWId
z?f?E250oGJFA>ZWX^T+y`}K6Xj@a*cm2Z9qi*25^Ic)Q_4c9iBwdV46r>Y7}TKQ?^
zPaX5wM|aIOom?uf-E=fLaE*IpaYS*X|HgT#w#R3kubQ{DVe5pg9bc8Q4YOIZRkKB>
z3AQd>bF}E%X0wfEn-@hIuF+kSyXNg0*)?<5L|l`bHe+d$t>~h=BG(Q_IEYqjMccL7
z9u=QvT6!%+R(+y%bF{#}j34PO9Y0QQNiNr1W^b@J*X>8o!rkKcYEz$mur2()r~Lh*
z-*Wrjha3?WeE;d&2fNu@r^+9DzoY)|-)X<kpZ#LI`Dg6D)$?U5mw#sVdOv%9>WLGT
zj{9si`(}08Id)xs6Sd^QVutO{OxjmIJ;yU^>)OYM!pgpvnv~e`%$4nbw!TgG(cbJN
zvCbLC8!9^z=2>qS@D;3K5NCJK>;1Qp*Oudj9Mjpj^4EfnnPT;8lC2MSw(RtJ_VG;X
z%6^#}(!57*eG&dI%xV5tW6i^Exi3E+%q(v>FzZa?p94RH8#&@)Kdk#0`p4<LaOa{X
zpTl0JOEs<@+tDNc^>Rsi>hAtkTE#bO&$$ZsF0s3|V5&s?m5FYW?b~zCm^sO_H!Zj2
zJABQsd-Kegr#I5|_1g8g_G~$A`XR4Cw$lCL&j(Mt1z-H`uiTV>?z7;6U%}6Btv@L7
z_3N(B{hW^!T!rTgo;{=R`s-r(Z(HUty;{5Q?`x$*-&2ctyM@cvRQ;TKba%4nykOaL
z-8*x3d(A)UH8=fX*Vj!;t9o=wiz^I&^CT7DS`vM$vGARP%lFdfsb5ZgJ#%`Get_Dq
zdHXeuqjgVu*i6{X{-{Ize9mu~vrZ2?nw3}#mc59uaQd)L>D||L$^~<Gh*@uHNi<CC
zKPPrz+cxIplPvl_+FI%rm#T3u>a4uIXquZUZ=C3X-h-dF_X+IF{!{CFv(~_4Px+fK
zsf_Qhyxe!>g~yFKo05;lB|WR~;z$j-C#zy`;atL_AAJ9gZ0S57w0$M_5jL;XCn;^8
zr`6`Mb0o}vm35G9{*2CPA0nz5V_z`xv*v}T<{sakq1w6TPO}fsaT&g_TRzg6%xjs2
zMMPdLomOG2c{FL?-#{OK28MmSohP@wFO58C;G!$@ydWv<bM=fmKIMZZ&d(UCsu@iA
zS8Rx64>0>RrGHuB?1g8ok}@|<+Ul%!T_|`<s%C)Z9mXZw!*AZ<GdIo2UVi&!)q-=5
z`{(JE>~}CLdMh5|zIX4=Z9Y-FX&0IfojS8&QdKt33kB8`fpq1H<lmW_zqBm9_{y|l
zR+Cb2n_=-o7T;M{W=;Ablx%j>fBVJfFBk$e4xiC{9(Kra7xVu~eR<IbDo+@DS<bdF
z%5&VJ7rOqWVUX5kuE%^HEizWCdJ0;cRo^bNRormZ`SG2{Zx(Mq)Wm-N!*}gTU#%Mw
zrs@1*THD8Bvi9|CZgIE%xSTzl?+<n^G7svUKI`z@<#*PH%0<>>tTyMG*DQK!YT=nJ
z)6EROGkq30=%V~>v4&p(Q{4ACYpp~4%qmyO+RZ-f8l|;1ub@#@X3JtmTaQ~dqEGFj
zt0gnDJ<XOzK5po04zW6zIAzk@Y1wN1oFXD3Q6X-ujhE|;KDJ0yyp2Ep{A2t32b26A
zJzx42|2x&GvFQWrX2X+nveeE7`B_ij{$|_U!>f1a9-IGGcjLF5+2>dpxlU#1zx|QC
zVQ$G+%el3Ede)iw4`1x`pZVF5mpvn^x!3q}{^~1?kNsaI{o5w;dh>RVwL#vBBFsx4
z9J|4-vT29SMx)N8HNxIAmN#9mE6kc^78z%L^@YOhrl9Q3r7e{;i+XfAban3KnWRsy
zHECCWvxNWB)BsN3YkJxTqpziI-QMv=a@(Jx55WO>pDnX5>^+z=X}PPvU(92Z0{`0Y
zc2!Tzsy+&FEspzdlDg=Y%GD`HW@>awUKiP1VxdypVCV9CAs1idZ)c<agqfP(LU!Nc
zE)}_ZSw3gu)nBKhbFZBF7?tZFEHU5KGx_e1sDDDn?#Fc>&007i>=LU<=9IF?RcpVh
z#@Rl-{^RojhA@NZvp+I7My+AX?!3AuL8<#nj+W_?St@BY@~t<u8`{>NnX*~+;JlRU
zyB9J)SA4s*(w^sMVEaFlk9yBFzL$N-kbQJR#<pzL@n<;?pA^L#=x-|iw8QPCf#0G}
zLK+7QP1B>h>dgaPlDw}nKi<rC<9I&nY6Hi~$ICxS%5HADdB9?Z?)K*kLMO@}w%#iK
zJMhTw_lwmT_Bt$BakD=48kY*^A3=$$B`fs0ZcS%2W}VTY(zZG6>$}+3o2Mt$*>8Iz
z{Kfjr```SFGn11KE2m4XYv)p&wZ^72_Qh=nmkV{>>Hp-||35iDDIjTz&heDf^YcyB
zX9}NP;%U{eFkbr9QK=PeVXqT;{$42L`CC<OwzfF=#X;xYZ?@KRh)r;?W;z#jD^q{_
z?_H-?<!p7Rv6lL;b2w^ef@2EbiaDB7%{Xs1-@MMn@@TPUhE$b?(7u}$b7tnHv7}9U
zc%$snoSBc$|DD2na8k5Q`iW!ab&(p=5~aDHMeBGjtG^{--1F76ZPtPCG(Ta#uDfM3
zdD}Gwj4mXv|7Vf1;H+f9l^b`=?(B{_&9p5|!OnKlo|=1?u5MO-yfaO=s-ZLYc5&~+
zz&$gzzmYM&e){Rx3jvvHw+ie_%bOJZ`kYjP*&?w|DV5o-<(FSJZkt>nW_Q!zP)p>D
z<!4e9&lx?_vv&Knqbg;NM56LF(J8+jBMi)LoU}huI;Y@k*uHK14OiCauW1NZwJw+O
zIMds#`yuK5z5bARt&_q>&t`6zI%9j*-Q*@&g;gE-%Ks)w#_ighvfr#Mq;8KydDs`3
zMFx5NZim>}*`8flzh32fXxg^}(MOg~KUc3O(c1UOS<>FM%INUxCjK7#1N%?a%J1qe
z>%E(Bdh;tE=@Vjk#!)u+cOBSuY*w4LRn7wOi=Q)s&gwkdQuH@{&R4xVk5(Fm=ST8~
zSxw!Yv4~B%FVgYoOy5hv5+76BUxw}C$|`Jrad^!IzND}te<q3U_;>BuT)!U|r>wvC
z{O7UV>Zf^^H(1|~;n*qQA1s}l_VUEsAI}04j?B!6GTqpEXS1rYwVz(l(ig4{dv+wh
zc<ZrZTc~Ap9_wb$&d{{b$gK`99bCO%@<@dry<S+ivv=XFmFC{QJEhf@O<oq$EyLAe
zIo)jG$|lDb`tq4Io45Xb@H(C6x7X(G?ZSJu`YA7;cY1d9xpxg!*}JDcTARDwDteKu
z*X(^fH6lMUmv8?4ctNwy>(Z%TtvF(AcpE?N+_A^yL*HWaf?UsST{>I=j7tx$b$3y}
zyyNY)h+p1!qmGBIs!eJwo}0@r>n-~<tnBKSt0l^*25m0E>T;isSC#)Z3qHPM-Lg-I
z-=)RNx5n{Wom%x*^sAXx!s9J1=@FaPSYA_F(6#Z2kCko6_Fp<{uW>Pp1P0`UO<y%r
zT}{Yy-h@g;i(`Ssrkm#6eHG$nxpDdwmd_6+tlxCE?X*C8nt_{i0DF_STQRd>pSF#W
zg}TPW*ETP+_|}_Vod148qezHb<VQ8ns%16hf4GIx*mP7M{e2YWX#3^b_cFdO-+1Tl
zn0NGpNANz^Ud5+NpH_7JZDM<o|LgG6NmiS#WNe$^Be}_`R#D(rkbkO*#X-N@ocvYO
z=O5niIqB%e==fWERGB$^SFX1d(%bUj-tWZ6C);KTW;94$PEb_Kom6J=!>gh9U!5vX
z=CR-(Qu#S$tLjoYQ%^nZDUd$x{YiP@kyTr#otr74Sh&}$ai)iyr@iKj#^kN*xBmV4
zDf&gCgUXjzzfJxL`ET4TATl9yg4ofMmyFH-&T}~ZfA;++!SDK7cz1RiU0?35n))Cy
zVg8;(1tC7GC>z_Y@@9Q9axY%g?A>Y_qBsAV9Jig#DZ7s4@ANPF)-hkNy?FI;X3&qz
zHkupcr!CLxzP&H{i-wJtjjh0U#d(}=OzQ1@-zK*?EX;c<7ij0VCO5L({y^!Bt6%SC
z<j00C-p$*w@STloN`}T>;mIAZwD-(g@XGv-X~frSLYkj-X1<(g`BZ@Q{F*!412(yu
z^2aG%`)wS>diFx_-J3N@8y3I6WPa-B60>_V7;m!pPkVhagnQM(jf*}%=rvxl&MYoy
ziMGZ?=F}NId~$s;Yg@SVrCzb_7SHMO^!5*W!Wg(<y-mm2UW=w@{oj6_%AR}HfMe}a
z*`E@;T<+=b7;e|gcHZN4HV=@9pSVAGi~CfrEBDsSGnR0w-n{uMKjWrb7h~!!{4Vsr
zFKFS;(|x;)^BS|T*wj6W#o=nkve$&X*}umM&S^4_vfJ-wz_n#wyscQp+N~>>ty;?z
z{^43wsGG`><!^hZdUpqH{aspNmz5lKo_)W?LAP2*x%+{yvouzl>^vdFea~yn?nbQ#
z*A}h}oxkOa9OEki6Y*c7Px&rIhTeO|XYn_s(K~3_k>rfVn+ZV~@%Pq*Jb98~v^AQ$
z-d}B12;0tkGo0@|c(py%XrtztXLB^XWG8ESvi|Avj*U-I@;$Z4)@|?c7v|Lq<^8sN
z<(<Db=>NCSFLU=Ey8HSaudakows%38^{W>>9y>35;=A?l+CA+(oew4-tp0gb%)FxD
zwa#+sH}e!P#i;29MwTT^o?RkWS-CFy-I=4ouh;GQ@g<itGCD!@<Eab#zO&?gcvPwx
z=prVPbv1y+YWlXI&#PjBW+(}If1ae+8S(X--H~IyQ;MH1W7t)^o4>@hf6}98kD_8G
z*rq=H7qz>2S#DX=_m|gu-`{DmZ#w>J;}6bz-_IBeum0&Cl$|=k?}(eKPF>~e1Cfkh
zu9a*xH@|vyy8PkteKqDgdh2(YmR?P+4qJb3&c*)+rytjg;|&ax=Q#H0&y|1bo74=B
zE}2qkH8XPB?u$!RkFQ!4Q1SKVwZ~T;^uFIG#c=5PMRE70T`nme{mG9!YFBBpf8uD}
zz9H-a%iYrW_^VeZ-253?X&EIOpt@8pU**`j*syxW$`hN^(xpR$zuxlN>cBJmviUoo
z)2G7UcL$#IJTuYrb%0mHT}hRndZE>^{>^uvF|E3@_OO(<k5}+Tt7R=F9zC1ludcZA
z%3134>5Xo)k4Ju&vvRdsl^a(ee6`BIQGq?EX!=`~3qmDb^QTVLoA#Sq>b6vLwKVtY
zVrBKOn<qZLzCg3*NcMz_Ys{WcSbE#sSe&a|^ys9cyS<DRr%(7l?XjZqsRH4zw`~?W
ze&qY|II{lfe`^cH;Pnd^-w3G=JrOc@yTb%lrLWJJrN1v<`nJ!iMNUnYakqSo%i48|
z942u7F50Gkq&sxyYw;3=u9A?Oe*zq@ywdJs*{N_Uc3<!+rFypc*OwmVJECf2Ej06H
znQrCH`2V@eT-kM>pGamsIB{X2m_iI6vzvu(`$IJ&ZSHH6{NB#k|E_(N^xW%g?kgv!
z_m^#DepvR3Tgfp|?^I;O21ENO^=sa~Hz#~9W^gLLENd@aFL^M2MuTUmfOtcdyKdKt
z-4!Wsj<{bx$IJLo_wr-wtm1Ww%F{fRziO(f=7oIve?&VyBf5%v>!fvI?%dBpo3AYw
zQgN7~k+xgISF11fVz=1Dii9-``hn>yG>&{wdA_DY)rohj`HmISV{2t43yU+jpMBy#
zaPQAwN$+iM<};u1c{Fk1T60$Qhd!4t9N1~jE-1cFG4H+A)`$s8kLJI<@jpn#Mz>O>
zZn>3tv@&B-cZJ3kWsBMu^DZ8?%Xf}czOz!3F-6jkm*x4qJ`TH@R&(*z+dt*@2<*L{
z!ZCG`IYVoF?c19FH~mal*9lIa(!BD<kGVHDRtsMaxw+D?H^VA({XXI7=I{5DZU;Rs
ztt!ac@x|35UstO;kNb0akoa8l_^*2uTT7;1+sSmUO#N%whwp!?OkT5pi8pV|Sass2
zIH&id885BlLsL^l)S6hHtb4O6p!fp&1^x^vYYWp8&u&-GefzF%?PV9~kG!2qrhe0W
z)bcMs5nHnJx2Jpa?}EtM$Y71_H8*w~kWHz-YH?<jF8holDrsSroG;76Zc5LwlfK!K
zcI8fvg<VD0E18^itJ&?RNbhx>toy3q!_UKlhx6Y#O{nVp?kYX&wE3~A%QC-zU1j|>
z)_}`Bt54RhY{xqGzC)co*Pc#gzmu}Xe7)4K&d1Jwm$U5ISob0{{A9;l-iLj9?wh|y
z6s^57mv7?+%i{BLP2XL&`DnL?+_*OFW7x+^UdxOXLC!`S>*D`t{mGN^*mKG4(c*$V
z7uvfH)*fX4aeGGI9{;E_Qmah1IA^~z*u=ey>HYlg;S)mbV$arjPV-uN`0|>2FV=Om
zEITE=_C?`+%TMg<zOOi;vaO?G)vmX)Yb_UkovhXCqkQqi%<vOG?&Pe!m7%Ftc*&<v
zXyfGH@6_WzdW)z@iu{v*rOmZzYeeV47l)=E&Ha4qZ;9m1h8wvZRxO9TZJx}kiH&&}
z^i%3&oL1VuOQD_)5<T-?d8|0{``5wX6%)6F-_$x-zC36}ooa&I_Bd%ywbkdcZk>=(
z)vdCT6n}Wu;bk~G%X7xNz02xaSSJ6UU@cW+7$DKHqmK1z%mdzq{muIySAVuim9rMR
z9JACX>6zWpJzEmCr#whfsk#4s`TC1KZ<@X0bEZ9tG(9*i_4);s_j`kU474*G@-Iqt
zr`!s$xqQ8NHG9le#(d=fzdIGRVcV~=+O25T+I8|=i=nZQ*kR!#+V0!p1$Z8;D_>Io
zW9vlk+~lwBtMv~iHf{eAQdYlefrN=#bE46C{d^vl^nTY5(;{F0kc-TA-n^W9Me~KE
znpo!#ug-~VlbpSf$!1&F_V06cU$*vnQc%3e`|7MocP@AZB%Cf?C7ZI#`^}kVo(WTz
zi7h(cESM6x``+S^S0z@$^WGm7czy42u(jt(@246~H32h4RVB~(Z0f6iQL^>0^Y4aL
zEbC1ZUg=~SU3WA$nN}Lop1`w=^@cqQ_s_Yay6@7`e{OAl-oE~kLz>9(T+cgOeoJzx
zuB!k1=J@r%zZGWov6EMxH+va3HUHM`Lrc|f*B<x#K3{sT@4^|g?%T{>$;D#Klo8*c
zy+ixV?;rMQKCkz_{`*qa?@#cv8HUU^)Boz*d3ny}O3h0tE-6Y)%muOXld`xBjZF{-
zof1F%Y>av{M{fjYM99<Y`wtgA@8URBTC(Eb_ery+aZij+%Ua>t)YP_PVbcPgO{X^K
zALip~51m*mDdrf!S?qmuX@TUWER(G|AMZP5M&9~0H(KBJ{C7TPzHPPl4-`M&dAs`D
z&$;)1&i&liV`!Y5Vl=bp@1w8U;k*4K-@THO6Z_Y6*ra@a!wUiVssHZfy!3Rd@>2VI
z+(pPo_2Qa*-HU(4{#DQWSuf!xX75#>vgYN)oyI}+H8bVrNo5|MaXRJAu?Z_$UVh`_
zGZt^p?+?%A=IGfp``rVB<Xa|HH7C!R-^-i0cW(FjX%9Z!IPoC2%q}tdZOoh4w{Pzp
zN`84Xx9ogd`^}%fSbw$rxZV2ucIA(?o1^z;b^g}cef>`Onz)^>EPvZqTUWdjTgUa4
zb75U?%`8d%*=qC8o=JRmTCC{w**m`1jiqP!_Lx6TUTf4n*|3K*`PdBOupU$6Ns>oz
z7}yxk3N%QTjCM<KNs}?2H}inaVV@K=!}CVx4JXgc?6Li9^Z3A?17{M_j5g1>c_t=J
zZgWFnikM;d%<!42J;I-*Hcu%`i<yynrs&wz!sIXm?U~nShW4;;o)vj`P3kq{d1q>b
zxn@glc2b)@cgoijzkFnT9{JugE;W2>SZw&(tla2+V%^lzH2qW0PjK%LjJag+bHkp*
z>$=CHYpiOHelXW@kC-R%(EibysdGaYUQNxu_RI41b~i5NQ04vdPpw}0&NlzyStIuU
zc!tE^c}iX%|IS#MeCt@B<Zl1WWj&$yw)IvW6623J=)BS_tZmipWfiWk!sdL|_3{6H
zcca@4*Gb$bI`(wz4dC0B?{M(lpTn&?0%n$P`ygv^{Oi=2TV1ZVx4XNyw_kVPzU9jG
z?HpmVBx*it`R(c2oSZsI%2xDk;{+v(&88L?|A<6i<9qvydy#x}<EKS^Cu}_vq?xP_
z6$MW^!SS?Z|K42<zML%w4wPvK-%o#h(xuq{ajW@Fmn7bs3^6sXO>P;@O|?H$Lo=5C
zc{x9?=F0mPk7tWEJMF*GQg?lgUW}Wt?kT<5UjKe{F<Bj!pO8Mw@{X4MhR0iWWyMCh
zm2s&Q2VONiJL}jcK1RE5KPE)Hd1&<{^n!>K(<+yP{(e4_rtf9qR@^khz(T)8*|TVU
z(*0jcjE}ZfmpL<E7mAdfEw>?gGDESRj)UIR&(qbLR;?ACwW`IrXv-?w&z3PyJ&e|c
z?06sfnQcyE@N&NL_YVsk??<q%`Fv}K&eX#OKdVn3-;*Gl-r5_U8l(_s;^4DRGVow_
zo5ETK_mc(6eMU#4?{@WF*n6;U)dYe6*E;^T&$@MsMf>Fc?Veq;mb{8klroyUY0}BL
z_H(Ap?A*y98B>$$G&g{Mw^sM*kLU9JPQ`{+GVHco9kO;yw6^uy!{(*Sf+XSs7Th+r
zm^MRTv*#1eM^1O@RFW^UoRxbd{6fiMs@93kdPkD})rfC7kY2dx-ND}1;q#RzvuI!N
z3X9sZXwT_o+B>&u-BMBvESRt=(sqfHOv=)f<v-5G&6b)zd4}|4=7)1usqGG0(W-p(
zIDcVnvFwx1i2_2mSf(n8ZJO$PLF?7^J1e(Z-tN|lozZt+vsEVW&D85of44bwaEp6b
zo$lXL@$BE(<aWz{_jbps>m6P$6XmzZb4|z5J=4>EZ(kCYEww6|MdP%d|NYR)M=7_h
z4}7iJli9}MZ4+PqO+oO8@634<r}Ad$E#X|nvXA5aCY$nv@RM~Dm70G}zZjfdyr<kX
zzfwQy#kLne3^cjKH6PqM)_rbmZ2p!x?b~-sci&SlO4^yy92_ut*SUlZO4;YnHEr9L
z9l&&leL3?U7Y<*pJ94rz3r}&LD7-(BMc`PgNl>?XMBVvjWAFI#=&g4g?=W$DDf@|u
zzSU~8T4%V<PgHcSh|S;K*Dp#PuPggLPcAuS%cXjaW9GYd?e<y9b0?vp((d0Oxf6Hx
z`f{WzD_WadPCGL3{?wNzCope}3Ghst-Qe0%nA$6#|F%8i|E3pRXLOk#qy)}xS`iWQ
zu>aYwAHKUTu91$vbZgG;Z?DvMJf6JO>faKPp4Fl!&b?4oI}{Z$`@z%bM*sTDpH*dA
zR$W<?^Xp_&h2AE$vi#uK((LpDw<l}%&r?wD5chQ6?8?vbPkT!3|M)E;bB~=VDvQ_I
zap2myh}@{;I^~dpyVL#ZOMc(E)aqpaC+d#N!|RnVHZ;opSjqMwe^*S*il1A4uJyAx
z|Mkp1*EZqP;?K{D9{79Ym%oz5zgfv=HcfqS!grrJv-`rcclGxRKYNhax_k9^rAGS=
zC%&#-w>fsX-Okdw!%-WX_N*)@iZgY8aQC@I`O7Wh_x}ER_w#AqBGGkwZ#l0vbleoU
z<#%(G0)NC-`?bfIe`Wodt82Yr`ka+%ZykSm{hjN}x#+vq0;ePFIj-&df_`&Lo-+2j
zTsHCiMgNt0H)l@gi%E*{n!-`|rAkds?VOhr|H_#sB?Z!@LuUu6PH}LIbLIT@KrGd>
z@MqqNry1w!uX*_&KQ*iHXl+cQR_fMErJKJm5lIzwaTb~&7&O=D?*uD-mO~1Hm;RpU
zu<H%t4V*A_@~o*--9q(B^h3DqH<|XD+}|jmVD=}cZ}ZKzkIxu4PT}nhQ2uu!?ETf{
z^>*um1yp~_?aN>HJgP@*?-sN8*9M;39gpu<J^J7@*M;02M%%v?{r$N4!r9pJ*NV%S
zIuox73anbMuyLpV-nqUZ7fZjtdUzuBr-sS?E7jgLjQtul&2!bcg)eZsZ~Sw7{?*-V
zol>(H&ac%@=W01FArf)#&E{*aagQ3bWUoDRU8P!R;a{jA6x3QG73vr9thiXYwy}7|
zS8cyz&-eY<8or0+ea@;G(Gw=jzrht_o$<;=<acXKNBf&CZ_dn-oiKR~SJust?R*xC
z6w>~l+Q#$JywCC7=X>jaEp=Nv_3MibZBaePw~Ja7i}|T-WBMrDvpi_c`(StJ^B1Nc
z-59?}_S~WnwZn1h*F<_WC;hVPeinOVTEVYS!>I0iQM2EdScH9_mbbX?f9xsSMK5Fn
zHqMAR*m)(<q;KkPr=;JGE7ps?<GthkMRdiNs?C1{&qN+R=J>_2HserfW%RXqxlZ>D
z%slyb-J5vLu5`skUC)$C;pUdW&yK(9Z}wZ&bj>oKI%%oS(Fa_Md)~{u2xjNyJabQj
zFWZn~e$xMIJD0y$+oU+}rFU7ISbuhEd&!RJQm2pjxMs7caHd$yUb5(1&{5(3O`1o}
z6hD@F@N8cbkKdup`Tt71bK1)nui~6MKVZ(L@BWrOjc=|j$!%O4tyP=xO-Fr#dZ%&Z
zKjV!@A3mzIxD|E%fUEPGgSK<j^K%-^J_=T+#=kI_uxpa_h5nF9yG}~n3D`etg#+uN
z>yOLjU#14nk1}<!ZI_-O6!?g3lEl=)rH#9r=Nz)gx<B{Ud-tY=!6JvlE)`v#-~D8Q
zVeg4AD>;py{0aD9+<$H2OM`B`sY{mo%VuPVB_(UrZc-_ozGqQo!NHqu=GMQi{5xiR
za#HP)uIXDMmTtAYzjf#9%X&3kbsu?y6iW@+npiz`@2_7d@o95Z=#|HzCO@sOycG2A
zoNlD5V5_6ub2|01-?;;u<~-WsCc2_A`l|fy6Yse%9;jRBugl-pC@@j~(rnQxo`etY
zrN#PBOrD;%%g2Uma=H7|-=e8v=~GgtU98-a^!;5|)1&hncD#|9eqH?FtLZPwXZydd
z<T+OzE8!d+zwFCK#l`;ozCjgw-qXCTgZWfMbJ%ZRZ%YZEzu!H5KcDA(k$J}-ecCto
zX=CWl>Vw<1u6Zy)L)82vuNkk?m8FN%x@s2G*;q+kc8clDK2uiIBXc#!Jv=7vevdor
zav|0ohgPp#9USg@#HU?y!KEd0{4AN{4_IH|m7g+CWB-k?Ys}^gZTDVk*1pZr_h4Cg
zdCa<_bE70hJ7ltyyoy5ZME?+-$gT2fK9`KB+rO4aMXQ#bi2eUcPIRH^?77Q0{2!l-
zzxn(4oNE%V@3)0DZa(K8zWnk2safwezrDU{(z_}%Ph<MT$MXgF)X#B#(0k*T?{QOn
zY2fstKl<-WuVg*s`2JJd_K1^$_rLn30wvvRTf=V1pZmD~`f&%79}V0s*Vw|_x2edn
zyt{ksJ6~st)xRYlzEs_HjOcNc-&FJd?qh9p9);ZEZ(ZH*mhidGy)SC#!gX*b*AAHi
zlcjgQfA|*rIs4f)joMRxZaq{fU-tQmDWmR!qNqX-$CAvjy9+Wjco#V?U(nJMaR1xY
z)VL56AKB+;#S&j~KDVBzaev(t?dVPKjvVsId;4ta=azyGI(II;xxD9bMxJk8X`Zh^
zK~<tyXOR85?Uxqr+_5O-|HFXXN2w1c{QhwIo}<>j^}D_Q`z(E<xagOP^puH$DteRb
zCmk)$xV=11+x*nwLyE7wt~&hO(Q>IXyeznVWy#&Fm)9EVwAu<9YYyMCTpzpp*wZ)v
zIez%Ao>R1N?biKY<f~sXdav1E@$cbb{dbGDHt<J14fSP9tXbmuvsu;TUV!A+3mSdO
zD_<EG7y8XQwr%6Qb>5;5GWj|(yxr#~bw5j)>Gdytr*y&rqiJ5hT#L^-oIQ8(pV_=w
z!GXmavw5zaxN`s4<130rug|chE@SVm*O}ZdvRqc=!q1Q8SFM6;j=O)nZNz-&OX1>|
zE9-mIgWKm=^LxltM(&@kbgKTQcWS2XmI~i&=k0c#R(5`?vVy<7p3A>PG*zTPLTUfD
zU;9%IJxfmrT%ooxa7D|sNqhKi%$*)SJ$_>Ny7_PHt)xG#@|gBU<=X2+uH(zC_P+kK
z_yXhI-G18&_axVzSb4|!z}5hXmnmN-H%*H^bFHFJpWov{6VHQ~TK;{SPS2mq{_s(B
zzk5(R%H+LxQ_|j3$sJ5<m)HN_vVn2WqwTE)FOoXyCl>ChTD^WoZ>z+uU$17hDBB&s
z^gX!g)V|+hm8)2Tr|r<0DgWcCrj<d=o`QFGrNzb0ihK#PyLM{bYlBVU`;Dh*MjcY{
z_GsBM=gf^YPj>WjH_qPs<JI-$Q2~AaS@&8>4D~mJ-Q|c`ysYZ}*N+ceWx^*cPZKzu
zKhIHGgugD{mf;C|)&1N8{ae;A=UkZQKh5jbmph@{aY2T^#gCsXG0T{re>E*?_9o3>
zPW?};DNn*D*Ng6YkWqZvuCJV%;cMbUtK0rjN&Wg-KfZ;=6^rfJ^(gB7-__yk;@-tP
zUC7U_S6jmAo}K&S?WJo63mz#Zr>Dhy*>&~h!;lnNfAzX^4<*Bbxn)o2{hrsOczC&w
z(lw>~j%(^>bZ~{t`?JckLiD(%hi-J?-$&Nrjpw$ln#B3{)t$9R1E;i{F1=rysAcJ$
zudTP2r&C}5hDmUQvFAPc3306T3U`IuP2RaJcJM!vw?)dSZ1v<fYriiHy%D%+>)v$>
zSGjIb`Z&Gd;m>lthuYJf^l!e|EwH@!`oo~uBdxN#uIqnLmRuPTq$Ty}Rv7on2QznP
zN*|9l=v>;kb#}w-Z>yBAZasU|acZHAesYlT<;993Oxhu;nlke(ix!ISVOx1;P5hk#
zlfHjTe;YT(XRynDTz=hD{r#tv^^7}IbJ$dz-Rgu=9KBeN_j~)UJ+YR(;n1J1L$<7z
z?Tf4$cUS*C95-?PJC3z4e+xyKr1rKxtvq^L+UPq&_}zeQ{wm&`D-YehUHvv|{^aK@
z|GX~#e0==;zS@$<QjC&LBKP`ctNRw&l*Kg6n)LVg>shP39#|FZ&EVVo%lZm?^Oo-m
zw&*W(?&#`jF<0AYzsB0kKY}an_M=A@CBLP=t=`+7W16->wCc>u(~HdY66Nn%lst=Z
z{lV3EB+vB8%E<p$w)|V};+&Rmv*lnezxd0FwObMs?4;Xm*|tq8bpDj-x9enwBJYnW
zi3X2lPgtys`*=P6tjC;builjHZ4BPKqGpH7adywNdt4_o9S$$s+`pnZ`PZ_I6Bz0Q
z8*l%0TH(>+=v5a~7svSIu--b=RaR<RR*F6kF310r>w93Fzu;+Vqx@X;yZ@zkv+d(s
zUNg5_I;?c<A>L&{7adX*bt?I;FEBlI_ICT3`QMz6xlTKy7JT-jc=KdEx7fFDf~)xg
zZz|qqK7aeoa`OU<%6l7TzTrRd=;OxF`&mqE$MTKR0(jXJU!1=yVf90+ZH~0Mom7AO
zwP!KneAoYfo*Pv8e~Oesg4x{0iF5T@n(~W&TwBuo(9}_YXW^?88GZZbUY6qicz{*9
zXhoyB%Ozd@4#|Xr)7yT$G*a59_T0L<e~!(ifQb*31h(4bP0%uzuYM6PXSnO;`NUoR
z|I~#(@A@0C^J$^tn}6bbpMU@37n6SSXAj@(FA}e<k7@j7FR9K>`>!ME`Xhh&BF<Sg
z)$+M4%VW8!!;iOe$L8Pr{p-}!Z27pd!UZOwnI9`_8NdCmi)k;jiu-x}++q70(}N#=
zuXwrj&*Izr-s~??x_9hGH(TK0Oe>SC2h;a7Pto8Fzp_+juJW4Z)Q@tfcQzOOOJ<Jt
z>G`tTch_@1#pI&`MzdelN6z?g_ULa5rxV?4`f{1p?OV61QeXxD)`-QXQ}+~3VX9N(
z@-4WSoqB4+q3fp}_~yO;w=KWbIzTW=VB+)%^Dmeg_`hO1ep2?;9H0DN^CK6heJ#4;
z&8TGa-}_K(Y5K9{CpWZx{?1<{V0LI#(ABkXX2~uUWLAEXa+oEODYP<Gb}r-dq9_fW
z%ir%?UVVR&eTzlLo6DK+!WQ=~Vl!!U+-b<^v~G2Elt=;hgPh47FD=~j-pCbK-Eg;;
z*<*6~M*Th&>zkVLEHV2Fe@CoOcyZ#c^X<Y(8)f;q?j-BnIreF;{uy7+dF<@pxNm$2
zU#qJ6MkMx0=|c{go0l&h`Y8Rm$uxlZjzjylReck6wrp~Xx!?TScj??qEsa(3X=g&X
zy;Gj3O?>G4d72_`;f%nBsK_akt5a{kcqW#Uy^zt1|AOTc>D1Q&mK!qN{2w*Wd;a2j
z(U&Lo(~Q_f^0>X-9_uVoUiCBa+73lIGmE=s-Q4Q?J~jWmcW~#lFaKMTG*%wgiJFjp
z(!=O<4fknt+g*K!dDgu$-!gx3ss2VY{(d=;U5S6gT=w4Cy|CB(ty0wNy*uXVymeHM
zIdA{b-$F<2_N2d?9~nF66yJPvtHd{Lt!<jt?0<4o%*2jQyw7zxLA^n@;nmR;(YJ3C
z<rpnJcgl!NozlH0Ik=|R>p_36e@FiZ%N<LDYdG>3ZrUI+f4Z3slm6C_sF<$L-rU}i
z2M?W5zEgBV&zpOL%(|@?c<n<%=Pds(a#d6;I%-??C+_m6SD(GD{Cz;~v|!NR7ynru
zQu}H_qoKuU<DzC{jEkBXnphw&%iUX986$H2QW-z9?OG+RTy2TB$7~C^4W&#A9Sg$)
z0y~oeBHg}~Cfv9#$dcu-jWOju%f&**gRzduJrc)P2)ZXf^6l{!+8a2z!e71p&hL4N
z0VjU_{oc7{|K+>E%l)nf``em&X#@v*DNW>w-Eit;;@3x=+OjfDOwDP2Ws|B~fA4OU
zlH04yxBQ5p&_r`V)|DB_w|+A2`{;dUy=0oQhSwzH69uNS4(qP$ZESmOQzR%C8MMVk
z$+(=qv!}e6f2#J=;N>@N`EBOs>f`$Q@e|+XnOWJtzdbqm#Om}h+f!)=MP@(EldE$S
z<g%BS-tOQ(`)RPuvKtm`t`lb4nXPAMy|v@?-n)Oc-(<<%9I1XkVRPMErR7`Ayyt|i
zjo<sNSF`@Z&gz#EB2o{Y`OlHI2r_c}v@2!OQJE~e&6lp6EK%NiGKKr8nZgy5uCAq8
zo#)l2Ofg!Vv5Nod^jH3o0n95JuPB_Fl(h1oh&%7XwxF;nrlvhry^mG~UdhPfF`Xvm
z-n%?*`Mm{d0h3p1UYR{b_p8`fGt=H#etVtIF1WWeEx35a&MC568#XN$3+i5R^omfH
z=GLAlzqJd(g7~H=Z|#cmUAypF(CZanrx<^=d+M~)t1Q6VM}srfD^7btO@fkfp^2u+
z&ofWLS8gp?yCvk;8IuD&#@)uh^=19yr&^x!I~`~6-sqNH{uj-+pOf!z*k}LE+xUI{
zzGF4z|Cjz<AGs%WqHV%ohslkz+2%=^6kS;7C_8Q1G`?w$)BdME&AI(xW^*^&|LMPr
zI!;dzcFt{|C-=*Kwbkp3*WO+GFSRY_#^38Q@z%z3<kw06``q!_Bb%|~SD@xCcaI6;
zi<Z1F{-`s(@}Q|TXUK+3_h+k^{<KbwiklY5d^LdYmD)0f+r_+Hvl<v$l@u1ZUAzzw
zeEC51f^@Hvt{V~`-keSk5>bnL{H=XTiz~~~%FjnDR$D0<uVGyC-6^ei*9^yVCnj49
zY|&QIQ90Ibb8phhkoNbiuh~wl37c*##&^m(e8O|*iXU&MriR<~Z`f1t*8Oh(^&1Ow
z7s*E*IH(@|{-NSlo}!6!gWd*;{am_4ZMDNemT;}tTCcljRb+72wH%*nz@B<E;Qedv
z#zLFw-MO=KeEuj^&)}V1zQ6MryRk*cHqG<<t@f$fn{VgvnsIkxXk*s}3rqfN>0^s8
z8lEwWlwrx*8W4X*e$w9!s~0oBGmdDzR<N3TyG%p!g1g0QlEs^!mGC_{aJ6=`v(4eY
z9OrqP_=Wu{4*c#pdPe)iD=x3yOyNc+)<13hw(Xnvlj*WfiuoUH%5Q&mp8oY7A9v+9
z-Ymb%w}R98x5PgWyz^aVb^hv3jlM_D4x%%Ky1)N1yPoBqRHe1V;NjAXp*6ZsrlpH7
z*t=h6!>z~(GkGn~ygc`OPF!8x;?2_+@BQm|eB0&T=U;99ZJ%*Gw~y!kJeknqB}@0H
zg$GMLI=HxbZsi@Z$9JbKej2*-XDXNZJeEJxjV4)J?+q!bnBRQq$ek&3D=r1IRx!Uj
z@NiCF+TVq%-DjVBqkrbKj*7IH>U7PTk771F&mY;i{JVPX$#l`ZM#T>LuKO4DTTE%+
zz94a5qUXgz8<l4&PnVmtZ95hH>`iR!9+9v^F8MZ0(Kp?Wt`1Uc>%8m|o-J4@|MZjC
zj`c#EFHAX$9Jd91nZNgrYrpdBZEe3Qe}-1eKQ^7P^jz5S)1nLKeLR^tQB&ZY%XF9N
zZ|e*BPI3rn=Xsv%zPu!fp>UxmAE(VeFZHUXJDUZ$SWBkrWi+`pF4%YaPj=9kucGfZ
z#P8kiK4Jd$<NFpcx4-@_TTv-;@%{9RG2uo|j-qpCb?pC<+j#Zc3?2s2#Va~=l~10~
zRz3NnNkw_P3-5&E5h^c_$XcEB+^fj5=#l_i`^5|mv%_B`a(bsHx6l6&kka--exhmA
zx%1i&KFkoEG1EF-HR*zc$*ari=6N+%g8k2Q9n`%_Pc*TZOpU0DKb^^}&^KS}*=4qf
z&t4v#{=T=2j+VC9SH-yHFu53TTApJ2r?TaZrdEQLy5H2JY)$L)ncoNn=*8|~-_mq2
zZO*w(Z%?Y(2EDi$opeHQonIKAX^~}%X2qHX`CIO1UVmqK|JTFRrm`#Le{yX)(sK*$
zmp|X9yn%W9JAb`rH}kb$du?G7S@zEBoVM+in0m1p*5)zk90vC@w?1Zj(<&ZyRDGGm
zq*uGyyDGmlT+8~z$Dv}R@ih3!lFc^1QeIB>_3wK*`^&njG9^{F#jP@PxOdp>ewlH&
z(d@^JownbNmKECnKFeKkZOW(BeYeg|{{Bi!`!$Qk-HCgZ)9=6Yp1^+QYI2c3V^Vp}
z=WqNxyk6hE<2Nf?G5?YL=Cml^_U5kEe=lD+O~3kDHm9{>2ID8;nUP_ZQYX$_G04eN
zTCut8np+rqTZ`*&pV~J)YzrU8*Zy9s=yAUEQGw9+*>8gkd3P4wQs{UpzTo+ne@|-U
z@6G3HozhdG+IL`4Xpzm5=1CvV_vJqce0C$W#K)ksQg#YQhGlFJdsME$5suaK?l0(m
z?aA~ZNw@uP=97N8S(1+e*E#1;ta-8h)kLrJRo#jwI;;QPOTIJxJNtIo=tGiQm{d>k
z$eaF9*syI!SDOBBwUv8VV-&mFCNS4<K6!KI{Qb~P-w%i%xMR1i;_08GtJApNZCGfs
zDf9WcJxUsB-?UDN*vys}6MAj>_jT$CmW}7%DX&aWye2H3RhqHu+@~|wJ$^*}`&!m4
z%(Z2n^QD#i250@5pYv~9K4tO8|NDx#wHD}XRj@QwHDwf%*k-v!XU)Qedy4#LKi}C|
zVX-omdk^RGMRnfuKD;SS`SnH5Nc>r9!RBy_o442h{Br-f{=8j<j}OhV+~e99lylXy
z@0Gt`gnPGn@57_zFW#z7T=LwYD8Pj)I(6cLjGUHNQgW%6XB<6bvxntFmYU#`-u5f(
zUKL*_zj{CW-F#=U?fF?-e^y3X2An8y^l85)*zr>K>xzz;8=fq2-(>DYRo&aSXz!xG
z5-(=xr`=IJ_pLttn~vN)r%So6))L=-HFczBC&(H9T&OG6)MiujLpHyguTc70?zYE~
zYwzw#PWPK{UHAE!<@H`CL%S1eOA2P4-0Wv8W1n?B)~@{dxu;*fcR#kgD-d*2vUaoL
zq9CC&{uWCPExR6nJZ*9M!+#M~|8-qHe_!y{CC|@B&fxepv9bnpjc&;k`dTY{SMhOl
zgs`p4E~{&-Fk#?zd)@kSesSOFSn=DIwKbpLUEMri@ul7w_Rnlw&X=n*c>caWZ`0;K
z^WV(d7j8wH=X*-Iim#lZ!S`sE4aeFC@-Hm-KTI)ia&y>!X_niei4}`77t2hl_@g~n
zMg33ZjRPw77k2YGPhRLD_v*+;!AVty0a|GSMO=*Xa^HndPnvy5QDV(sMX5=)8!OXi
zPWAUN=iR%Y+_WUrWkJBl3p~C{wR6rgHP?R+IKb_}nK12KzD=`8q`Rc)+?8iIPjV)m
z=+p3MpHn)G|BHste!uf_!pRqM`1026&4@4B`00n_jw5$sezKZ9EM$=tzmU6aS@6Oa
zk%|j$8o1Q|d9clZ(X=pnPuTfK?n0JE4Qctck?#aFLLNEaNt=K5a)pfm>q147>|Z(8
zuLvAG@oL)Hs*+FFY;1q8m1ut^!0Oy2%>K~&(&25%Rd>EgB;2dv56n=JD*wJbpXJ+4
z_Y2RWQl^Ve6kW`^Ofc&dtBKP!-+R^}f^pH`b9GDD*`gCQHRm6#;50g@?7_Z7zUFJJ
zc(``SB<rdWk5?*;bDv(geOYj&iCp!ii;8!3g72;DFey5hry)8)y?4!fVfJO5pW6~<
z3Y)24tOyRVSv2vbr47qvw$&<q!lfC_=RQ^2Zn`3WY5n!hg-PE8-_?~X2e^DX{XtKp
zK_@j%;9|yy+4Wcc>Ky7mJ}ra!>@J0?Y_}p7Jt=I@sSnzxfB)wC_J6OeL=VPlH~Jdg
zd2n>%-Q0#Z?-x41vpSvr%e`vnzXkLDhK6*WyyLChT-9SzqQz8Y_-e{Jy|j$xLuyYP
zH*@oHCh_=9GP@{x-`!}MVa2J<Ud@}C<eKgTX6gI$@)$ZkU+o-eb=u8V>+0sRB&(o8
zwpq_MvYx#qQJeSr(@%rBU$(U$)=Ib$T>Q4qDf!~-?eku<>TYSD-g8GXf|s@4BU>fs
zzTDf#8?OgHcD;0~-lUFK|EPBTv9K1d^9LXP>M}Z(Z7xz)>9l9QVsh@Ptlrw29l0mN
ztDbmXn(NJ_HDSY|ppXzLe%~;rci!b&OD=_S2F%^4Ui5GG^0{28=efSVt@QRxc-z_h
z@6Q+22LUenTMeT^{#+3*sp+Uw)vS9J((^8Iq4WA20qrGA|GM`JeaTreBj-S<YKY*Q
z7H!UTLN2bW*BpQP(pGKDT(0};rXE-(nfD>-e|UqS`^&D<Nvq$USn}8K<w@2k=Y38U
zfr__dufM%5WyR#A==iQhJd%;+yw}HcqZd<ZSE_QhPg{64?Bv%~O8g1eB)+e1zU#Ey
zN$k_4>UR?7J#x4fEsvTeJ8S9+&K-TuYc_@a{P$Zy{?umEr<ShY1Dm4`m+bogvH9-8
z%{5p0#I_yZB3#Ab9Br&pQ1X6G&glhHrz;pga<=iA5_wdXuel&W>~HL~x>r+83TMw+
z7sL5dG+WY4n)mgTx1Uxt-Ay>Fw^l!RJ4ce+(lakQ0&C})E8Jeu5ZJwmZ~dXZhX;2^
zZVzEw9C~8oDn8TCpFdefFqe5s3!A&O?c;pku*m3R#UaTbLUv8s1q&afT>SZW7T*r;
z(n8nl8Gp`h;J8*k@dJ}=&`Xv3(jwgp>(4*jx4pXK_lJm|;z#q>+*Y*p(vZ+qNa_A{
z_s6%12L5L)ck9kQ((k(J`IpIuyJL9XtyewjX|`~e#3Sv$zO`HLx$gHqf9poaR{tG;
z85cYX72uGp<C#5UhwWW$J(0foPaX<AS+#0XR??~|syw@Iyo<QI(AdSj&i1+9#Gs0g
z@=j-N1-LD^8R*t}>w3Xz?S?nYT#_Fu_RlU+z2fBTQL56?$bL92w^j8)?DN>S>wPuZ
zpH1HNEt#oK{O|1I(&DP__?|oKFY!msiP>F}H;1ow!Q?Hm?(FNlJ69;ju2Oxr;yq*K
z<ob8kqTf`*J~8#z7BpE_KKc<VcQWDnm7M=?9%NnE^F;XC3H6Ht`)zMpugZ8mZ)c0Y
zO7Sy4(F2~ZXU@O#W}Es_<zEGpuD1R-v}M+ZOEVrNC{@qhzHa;DyA0l+gX>OcoL}+5
zsAJ^?4aUuzA~-I6Y7LmRG}GqcdT%RMmpk=k|Nos{^gpnE(jO1byy}V@v%dBF-&8v@
zeXr7a<BWvmNx|zk9$F`^D7@78!*>^NH{O%Hwq2@a**sq_xNdZ=TF7F*;Q0EQ`!_37
zUk4khUO5@J@X7al+eIHAxaZr~{(2$zsN&0sAM-BgDxW^;vU0}vO?Qr6*y#AASS9z6
zQ;kO5-XgbGVXt|twz^iD@4G5z8f|)|%=@LGo6noJy)5(Rx=NkbTEYC<Y6X+8u#}Qm
z<lDP{+5OI#=3H{^(wEMgDe&se{xj(dPxZgqxL5Iq^7`8c<)fy)eP`#YU#GM9Vnod~
z9{sRW|JtUCPd2|0xp`Bb)6A8>kIs4{?eeYn_Kmd@)c+;*s$E*&#<*wFJ1LgBXr<2=
zRkAC$T(U?mIloTaC3@wH?XAskzU;Ze^hZ@PFqk`J>B0jpu1B<3JkExssxQ(%=`efl
z4DncQpGeojv#WcBPw@QjWdD9zNLk`UTx{c7^B=Y!&9iMbUav|u_fLKuEo7K*$)4x#
zN-KBSUlHk^%aWHav<}JSb?<)sbi$meTP8ae-Msr$w&DnTazb7~Wd(CQUuo-1c18cS
zdZ%;(raLJb#LL~>=C|SC`}RkSOtUTizv@c=BWrOtV9NOoWmEs@W~w)?G*02Wtt`wp
zMc|p_$qlae0_RD7Jg}zBzQCiwEag&zgsJl#$G280FO@d?2)%rJV!rd=<^LLdfASx&
zy`|mZ*xmDd{{80T&(85!aXIm7=s*6>X5uQZx>cq2muyn!vrVy^k}m$qE%*N%o+=+}
zQTp*#tLo*-!)DV<C!POt;(KSv7gv#R(YfnY4jJ-II=pCGg~}wOxjSE6u1^29SHbL?
zLuOT(TUeVF^CV+tZQjmIjrFT`n@Yued~|h2J^O5XIrn8vfxqRZm$84@^N?rSZ{GBT
z<fB4sH(t7Pl;@ZKzQ>$fUOR4olW<7w(slLZu!QZ&-R=1b!uxomvc6rJtia~@L~!5Y
z?v1QAyl0<8teyNiXZO;+UHR``mze9#xw&E8YmYanDSjI7ibG0vUthUMQBzsdCTX+x
zBzc_%R`q1N^b4N5_OMA!{5dJ#u$|pU!ar?&o&WI#4>OKj%+G4I%+P7)e)%X{-0$@R
z#t{8eH*Z``iz;0q&-jRI()7?rn(rn{3K*9!yPa|`b<xHao(t7Y(du7>zF$4|^N@qP
zaDeu!>#M_l$u8HQdu7Xq$hzH+Pc^K(6T7LO!R(XKtjeEHQn$R@lzMon!rw0M6Q29O
z3rR|`-|lQ(q`KXFOC7U3Pju)##gmqwUhy9AnPFc)i=mG7-c&^%>#m@q>MYIzEMjb;
z$Bf<_buo8cz`esDUpY%O=_LQxkLEUaW?#?RZ0ulbZ(E>#xNqBSCUwV#Lu>Awd2>l6
z_fBHMypobT!Vet&+-v#N#nklqjErlfQl_El)2F6PWg8yeE5H5d%x0;q$J3^5PVe6~
zC(m}%)yN<GZgY0;y<IUcc9uon!-<EPOk{KNf{mZQ$e*CO=eDX=?*)YqW>=f%{oTnf
zqcNv$*T)+=b#^(QJiL-FeA<6t=b^pf{PS;JK75z+z5Tw@4_`h$`f$}I?8m7aTjY#p
zB)Ie$HeVM^?c5{yb>6(a8oNrfPZhpcCOD%aM4wUf;MyJj>%Ve7wm5O?(Zr;q@iq5*
zluMFIud}JkPd1#X6nv9+<9v1XbFb6ZtlZnf`%1>$ndgM_9;X_nju%0UD<9~rXuNae
z(2h$53<-ytlWWrK+tW-|v0jvA`n~JNi7)4OOBH$w$9UbkSfu1?dY6%R$;XH%6DD?^
znYeQ1HubQIB7dIOnKqpLCeHpna_e3ceu!9}RIFdF{&maMv*s^u&sQ|Ix%Ou2Gnd-9
z)*qqypWgTB%6-}N>VH8UtK60BzdyBadi6bCvDxNwU&i71Z&M=NbTkk2yK`>Ja#)uC
zao+<u=|{qq1}iisTv(akRc~;(Zeox;+r|SwBzHUTz7Z~1ea8PHZ&Q9xtNR&_6LH7p
z?W%tA<Dr-D4K_LV)05(^z59B9?-%~{Uv~TyI)1jkJNa9dbJWJ;?={zdl}|Wx?Vo*P
zPRje@su0gdGAS<v#3s*AWHX+8Tu@2+ONG)R!M{HYrH-6G{WM2b?AC4TnEHzMmzKW1
zuD^DUTY=Oq)4cibzs(QWay9;o<hPR#3j+S_`*Z5wwS4hk^1t}3cP#(BhbM-;Q)*Ma
zi^E)%OEI}2pJo2}M!c7iI%RU=#%i}c9#tJ7{n5|AF1p$@Z*SyV4<>t+owdI?=f8|x
zuf@1*+oqg;MXv?BmpQ9hotCx=S)g2df;qA%V6m0P{l^y{%-p=9z_W2N8=sM=>8<5|
zm`{8Pc2E6vb>FN7w|kq+d4hy51>HGq)A1zj!<`S3_n3TsT0Wb-WAn-FLL3{bPBJh1
z$oC*{<sXZ!Dej30+YijuO5R?s8N)jzHB;)`k2(%f6|3Ua`>d}_b#Q8CR?@ksvR6@H
z&B_-|Z3VxX6=(7{9$C^8&?b8M?%t-S2hOw?Uu6nflxcf^=cCN>jK|aE7Y6>RHm;s^
zet&^VujbzKrt?f|7wrBLS#F+Z`sC-XeUayXeQ$bscK?f`w==%oc*DB!)6Na){Qax`
zojG7Jx$bk&gZ&=)c{=NK)%RLjZ(!?ceEII2n5x&rfWWm9YNEzR1Mg~0d}$cq+kZ^l
zeQ|f%)x8{@F9Y*RO^lAli=ND0AhbrCQ&!AWDfr9O(Ae_3f_<ton?utpUCy&fZ|7*q
zkn?_-arg&I=BpDyp?OidQx37&9MrmIS`+D|A-Vf<+0rRHm9u!I@<Vy9PMAG){?dyM
znxB9BFHduuvi!-C>y!PJrT-Yu+<J9oOXp>$d)|Fr-#9sMPn#B>H+7@_o(M*t0+*16
z17THfZeKc8&$o8_s`U#_KC)UJwCHBB5qDYEr-eP0KkxesZeQ|ZkN)@5I+2VfCp+1E
zAJorQ^=3Kmw#bqD=X9knm#!}B$Wz?)_Wx`F-@o39r=o1#_7|#lsy?hySurU{_;Se0
zm3f(oFE6s?I!c8{qy?nSvD&JktG;xGNSl$OS=Bx67cVLptBmrpT&j6zMbBBf?B~8^
zuba;NelqJj>*u;%`+n8+mS%lYH;#^-I5j=*UPQ!-bC*_JeVWyqk#ui<ymii-|7`!|
zJ3iNQVw+hY=gwwB3j<T+nT6QU$^em@lJ6hB$efxwHO(d~`Fl?8sl_Z!i*|3gwr-Kk
zF$DpEiw+SxpC#zckd$$A?AUPVOLm*{IiY|wHVHmSnb@6cR-9ZXHMOX4ve~jzfx@SI
zqtjw1nXWI7|93*#Cu{Y#-LF^w+V#73|NZy1``1T1v9JmXb*6+?)no<L{8~1p_BT(J
zb5~_?$rHWiZ72S}i2bPR^rzD$Mq0tBF{$E0eO+r_UHYGUl6$5ps4UbjP>I|3vj6+^
z-;c{=maI*5nc(8|EbG*E7gzWEySW-$rLKBSjohbdWu;~$bljq{F?64{*FN+80m&O?
zm}FJ%+vl;*JS%>R&DDg1=2t)QvhCfoAY$tJ&=b5|KmR>y{<P-IkshDQBDbips%OVm
z$QS-qJi2F^tB|3K)I@&~FOim67rc8Ik9TPK=uhELU(^}v!ZguCMY#H-^-CAEjq^_^
z6?NJqRWc|lHg$YGvgV1d=l&<Z8Hx`yAGmQ~=7IN3R}Q8WIDQcRWFW)cYTK`}h%Ih<
ztj?3w5+&Vhw4QKr=@s17+2dLk|3H7!LQ|$KHVvLmH`KX5avK~M-!JfbG4l~+gFv=-
z&+2EHd0o79T6kUd11{Ag=NGHKID6#Hk)-d2?ZpqyXfRtz`I&CvvCeM@ZrEkHRyU>b
z;f`w=7iXUNs@mYSa?MqZwjFk5T}z_pon4ZenwEZKlh=mjYp>kCbLkG7U$FC2zS@}|
zS0sJz4%xx9uJX?x%Vx_?F~_2-?6YedXZ%YysyJ!W5fr#;Cy(zU?>?g$T;DC(AE>Hr
z^JTl_5ajzUpKsQh_swq)+=(`hwdIy;Uj5+v1OD8}cg^p<t!FAf7!iA2ag{Ou8@87r
zFOEA+Vn}b<aI?K+<&EmNXBR(*x$&=hv$c%tnfLw^RZZ8P?!COh`Rx`yBin>EbN9P6
z2L8@W<2xhq_NB%?IbCZ8C5>Hb56eDnKVz}uo|Uc3MeEq5!P__;dM%}Gb#DC*V$*VP
z^;C5=U)}$-Im+Q}_>SX|20y1K?$~tmJX4(Lp=Y%Pll?b4)GM!2)U;my==BV)fL!ZD
ztH|7w|5+E<-6(#3@yY@19ZgGD-Y=?2tbeKyVf^Hcbn?smcl&eCyMA1m@mB21v2goV
zlbYvu?5k&7|D@KszQy3iYvvzI`nzQ#miF17(m3jMDfoKi36%+=RelR)o>fKgM|oz+
zcAZOdSlG}j=yB{)*_KDAd;h2DHGfWTixHW7D<fm&N2XVDH=jRKkJuu6?8J+Y&8+fj
z|7^0~J-mF|{9TP%<qH8}E|wRfahyUkz8D1f2&SiQzn{J3SnY((GcET-MlIZ~IW=$H
z^xC#P7ng({*AzW$(QKRY-r)Vr3m?PGq^c4HZazMisTqH8(SpxU)@q;4Tefo5X#*~e
z;|cF2eoH+GmXKlNR@iyf{l3U&wUmEPeN^UD>+%`y*yp=7qMorb@WB70Njtc^3v^_A
zB{s&IC6#2FxGfHQY}D7cD`xTk%M&zO{@$JbI{4STUm5Ff^u3n;!R5yK^~rudpJRqI
z7hK3$wcv&AoYm_#a|Bcv^G4h;D4(^|{=33@uIJqq7WobK$}bPf?RT-gwxp@>$>!aU
zFB{7qW9A8Zy`=k$TY>Zgw@+K<-^@<-Sod>*?Ct$5uTm%U{g7f^{Gfxk!|kEI*+B=1
zACFj<%)AkHb6d}mCr^5Qs`uyJ*<oC6`*yD@W3SW3c&0lKnxp?at`}Q<M0L}@S1%>h
z*#B%YvY!2F+NDoNdnZl);<x(VvZT!I^?oy}Pt10R`ntpP-^$jt_FK|yRXnX^Y<7M!
z_qd^Hb@0gsJ|2rI_djLLx=oH8zhfG{GR@T!J22Jg1J~N3DJJV*$Hra@y2U6XqFdf$
zA#eQUoPAT?<hy60--y|0idQag-rIfh%gweBAIlZCf$_8N2Xm}xSoFWAY)XFcCWTMY
zr$4VcX|+)5NnloJrcAMfOpf{e-#h-NEEfx4opbVto0oRQ-u6s(+XdM<Zq+x<I5c+T
zepx?V{Cg4C@4~(On{G_HSupjDPQf$rLN)Hgsj^muECJOAzC?;mJIs-}{(d>z{`X4z
z6n<NVE9&K?mDOv+=DZUyJI`D)<&Hso&HsOgSL)Ba@SWA-#eS6yXYQ{!ys0oYnYVe`
zZA<flV1p>G7~Xq#68l$Ii&$vRP<*<+Cbl`ZCTl^d#ABI6VTMKhMTRoBHcDk2ER3DH
zORN3xl&sUQO1!qOom^U^|61C9(S_YX3!c{3`u^_xSzfxi!OFU&{loDOKa5{qG!HX=
z`e&EgZ4uUwvl8>HvQ^R*mU^;kaBrJ=tWfB-&Y5Gc8~U7P#0O5lmV8e0p&ip#k*?`Y
zL4N&Vx5Yk+=Dj-;Eb(G)a^()@o`!V~-<dw&wfyJ5>wiA>FLv&E+ptmAS5{@SNydJ~
z6wmAvK?0k6B3?zCJ<fRa*S$DC*xTbojq>^JkKXu5-xvGTc0>MO{M|>Br(>yNO$^AG
zyf-&6G)Jytdn4E@gpY5FOU&MC!YsyZq_SJBF~*6TQOJpNuT+n(;aRrk>cgR%VxGip
zs8#uXulo26k;^<sJP)wAxI1tu3UIU(2s`B&&9*-G`<{o<iIwjoKh@pOC{I6U`FZ~T
zKj|)0c3P?mb-JyO+LZHB?Dxs+++5~A%+Gx8K6tnC!)@6$`@gO;m?jh_$hn?Rz{0rE
z{Qcdg^?#-Q#Bco(+9j0QuMqy~dE&3v&kujS{_}~^ysb-)C<(rOA}wQ<eCx~0&C}Re
z+SseBt4^*>_nB{3{3@hqZt`29yuB&2ww}!kzt51`v3~p4yn6HO*_qoeuer{uakBs3
zHDTsIYh&;1E&E*=!Cn!OUvxj~?S{giZN1mD&2L!V$ge5BCExQ_dL1{%z0J4#Z<swX
z-P}3tGrNw^dYK&ubGV;+yj?J7<=x~(9TU^0SgJ0|$Wk!vkn-VO*cK#q#URVVv}x0l
zC09JUmbwKl)|5<j?Ok{*Xnl~((cs4`uCKVh^5YejSG-wDU)4-IXZiRpTo!bB#r731
zuiU=klO^_*@2lQdIn()34!Vn`1$eJay;79Lx^+^N>)K^GD@?CMothTqelzgfm8vY}
ztrKs#SO&~qse2_XOM2_vh!-wzm%L#MP)fR|c9G%YO~!Z3eY1E?Z|Aq2O`rYVq~FM1
z(M&vF^!^g&m-;H_CcQpmvMlvMOQSX0JNX~;9apn`<%u);@nwRf^EbYw?aQJ)bY*mJ
zwfV}6+5Xx3@af_I=`E9=I-Cv;(#||`;-a#7pO5^_gXfZ0-@Nap>Gsm?{AAuqM^EqD
zbSv}g*%_V5di}AcW$(40t!LT)>E-HG;qQy3cQF4c`q#Ae(y!Bn=N`XX7{>i2S;K0b
z$ij#BYd_AF<Pur8G524qV2nt`vE|lcN$NAzEX<clX)Rdb5PNt5?}^9q@*7t0xNgo{
zRA_!tXQ@Yr)9o94{2kA7UIv-?)lAXLY40m5?|5qaQprT&vqM20OT^Cuard`Gh{|cq
z{xf~Q-qPf|2M_O+<@#A%!Xl?$wrsKOfxC;DKm9bb;)?EY_EuW8L+r$g(>LNj<lii>
z|9G3%YCrR>$~^*WQ@#fKm`S80O8?qf)+<(N)4=}!THU%WHpM^H-+n%Q>9jnDT!ta{
z^E+}0k+RKmRAfEbmAhglO%fJTnDTLT?;4#u(`#FrKd*~_c=4xC`x1#)wY3TclRi&K
z@Y<TDsnhWGtnq~%32fRb^CEYj3R%9{sF-^}&r0sK9DQHBZyY^%DS7jRRG!C*oh=8h
ziLKCoW~FkzYyG{Z^%64O*EUOkmi>N^^Tz$6Csm5g&%X1eFTHHvAe(1>&BgeHbCKPv
z2YLl5I`?*W9E)+$6u)?*d(pOe@8-=vxnHtK+%7G@YTn&lM^BgCT=e(ba=-1hMZL~~
z_cvZmJo+$3Kt<#Gu}8n|PdgPfIcSo?JWk!-6;HAhcPkmoU;VqwW!l8mC!2DfrMxfx
z)|Spc^H|==Nq@^yX3dk9sndC~!TRtM&W{gG=enO>U0ih9sjP6tR`d2!wf$->q4Jha
zb!RoFTNkSw**{INM)%50&ei6d1B9lF&hJhA9xK3Ze{ii=?$wLU+jsHF=AS!v=2UBq
zn*K_q36l@M`6`{BTc@AIrqdki8m#B{GV0gNSY3m!k)oA*>b$NSD_vP+>}<c-SSO~d
zYSXWi-(J7}&cY}>VU5QFMI(JfbIy}#vWH~$@Nqlmide^r=am<J^7ioRQ@(dBz4co0
z)q?M5+`rrZ*z|=lCRt}f<cbYbB((P4D*1Xa)c7Fh&rJb>QzuX9b4^dRuaAjkP@Rw(
zx=t&w=;MSrljbz7G&S;F8^!(Ahi^+|Wn(xKdt;Wg?}s17n_j#L@wGAV3QdvVH=Jph
zT4<j?rT4Z%e<O#-HZJYHH4lmujctu(1I1HT9?UcoIHA%I9xc7)X+gmJWxi``w|Isv
zzT}{1AbM2w>7JvT15)aRZwgC`nxAd)cw~8Utps0j$MVk$4>zlDN`9O%{mWBt#{Lr>
zeXQThnEo++l6j=kX0zcyv(rfrEvLSWMUuZ3_SrtZ`0|O^p5w<C+ijS8x-Dm`+uSuA
z<@coo=L^ka;c^R=iM#jd4|B@7$OWfnFY0f3nbvgOTUhVnGF5%a;Im9Um%YzkNPm8_
zY=2dJzk_{z|FZ+rzWsVOO?-!ivX7y#O`U9hODe~acGcq#{p+KZygOP<JW`%KU1hY6
zt>`p|j=k!JGn$O1jZ0nEgfXmr(ZaV?D27$Y`ob>G3=^~CkIX+v<TPLZ+u^QUtWwu@
zOhG!Y(5_qk#-qmf`}4cm7Jke)^z{9@`29AV{m*qAv@b-ZI5y4lNQ=6EXd9cOin+7K
zg0GrqT^*cig0|c*y6$jYBy^UN=r@B?e`ZL%Q;=D)W?e$n{aK2gJWTTUt2CYks>R0e
z>+Y#L7qRx~%j4E19t$Vm6Z^xrxpz|0if<npE;#UMzu@mv{LH@f>f7Jz<D1?~np8Ym
zUSXlW^%(Q>v(?k5^OfgzYCl`jul>qvcEt*x6MpA(Y>Sq>6ET?ox@^zk!$s`Y;qo%>
zE6f!Q4FyAb{B5TcyvW{VYMw6d@6`M)>!R4&rxTB7%N<TO@z@@${#Ecvh9-A;j(yRp
zJ%v`kl@7jn+bX_Mf<Jn1?bCUyKHR!>s>_P6DEQOl&C##7PGCKAb+MeD4U5a}9jxrG
zEz2#J?pUhz_21;H9i_VtfBCF@di%P&3IF-zS-6e&_+?y5RcSVzWy5ymrR3IArxpn8
zJXQT{yYvMPlUM)D{l64et`m(@T&VYg@spvj&;;SzQXE=&hQB7>n?B+7(Vv$OPM9Jl
z+a($?;i$W_*23ujJuCS5Rvy27F_7`W>tv3vTDw*-MtynzsqB1d8)N_Ut($I4ciZ<O
zdZQiJ?yU=4^RspLUs}jA_lxa8{@jH6IsZlYpVS;st(@NEm>sz0$y1vhP2c+b3!WD)
zbosSWsOi=YqZ4P&oXoyezhwpYjPrdu@tYG`|H|AG-8MzT>S|VG*}SO5ng$io%hzyl
zad#BXuTGZ~S}bU9<C50ao3W=t)_3#8sCj$3lipjsSTdWX{>cHQk3H=-=G<n9J{)7y
zJ@IO@zWgMe^{tamEIBeI!pZ+qSmnw+8}}{R^K!=fo17IoKY0%KEzy{v`>S*EJoiQN
z`&^uV9$oxw&CLxZo4)HmxBK_w%G>k)w{@QCd~#~dTl=D8U3WvI-W?M+@%zh!y_Fg+
zOk_WC_{ZEy1{1b#cTTl!H_G^ZwXb{6v%G_1oD~9Z5*8iYT)V$2o^yJmh(}A<pD%0n
z=)^J~^|6n6`KDyWgqS74Ju7|{C4RrP$$07XzC)tV&b>MRLBJ;E+aF7rH9NLE_EAyZ
zn<l)iLikMTy&4nYNTt*6{PT*81KD?<%)4IXcWc+3b)OGUzI#0X?yIs}$HX6&m3_U{
zTWtBVZD+Rqu8Mb$=lzbaPucD#?=5Ze_Lu7wkEE5(Ps^9Od#dPf&=3nYzklEn=jM}(
z9<k>OS6Ok%i<HUs9lP|?IL*Q2$Lgs^C1*;`3^?TxdR6f1zs6G=Y^|f-2=*RZl<3j5
ztzff#^7Z^1yAtk+z5Tj7+-}dg*rwXqUgpm=E<F)A`XZA@*F11$!nbt2!}7EL&%B@4
z!X<giZeERtmS;&`j9H$uc4~%v)YR#*Q<$fQZS8UrWj}K2L_`hu?vqJENy3htrxyK~
zlk$b(_o+?6MVA?7Z0dTHb&n}`SLN?zD`Ic>oQRh=ShmAaq<#LSz&sc3x9U+GoF)J2
z_MVuTbMGDJMj5de5&nlxu`lRw_kX#F`NF~f=X6TK;vW_X8+~McqV4U|A3k~Op{ZJy
z{HG56*yD1gE&S96!<_x*s*0S4)-<hN)V<T^#FdoH3IBSV=KRT7k)Sha>+vlsx?)`1
ztM)0$Gws#Cxy)AJc&js)i);G{TjiL>%G^)&6HI?y=fD5|K)F+Pz(n^Fzl=NsgW_vv
zj-*vwm}<G*ukL!MSS*7`=k$l+oKI3KSmXNI9vNobIlMc4R)w|5p>w5W1vej7eS0SE
zzKdVhI%B_w|HJ!Pt(;zg%@0!5RK6KjSoE@Urgr)k?!1>E>R#;}<FrU#?n<;8_wvxj
z7PnUmTHKYGlecRf=wAOPtY_K9sli9%y2NYp_MCp~njzfnexcOkQrEInnS?7_+~z%B
z;ivN8!XgLVTdTI#EO0Kmn<m?ssxj;Qp~!yi{Ok8tWm|E__T=4_%{lfkASyLwz0kgO
z;eshr3r(W5+`rt+&-)yFf$^Gg!BdrY@fq5i7A)QVK4i`3$J(Aj7F9Vj&OUsoA<(;Y
z>9caFX_lXlNo7<9wD;YQS~J;B?9heZZ}vF|eNDN+ywZ`g;zjn#jR*GFJ{J%Abnnlr
zpSJtB-W{9%_@U&%trM7itV5;?w<cUzH|NVtrb$Ukd!IhFe*L!b;Nlb0_B4e*49#Gu
z6FoWM$H8zmHgS=ti_fj5xc^X<DcziL%{oJ9+oKJuPcA$a=W3rmQM=czzW3O|$d(NT
zo8|a7KD#F*a{oeyov+u{nRP)T3tVq+nRfrxgr)18x4WwUT|c30M(v$&<EM#tiefsL
zCM-Q7d+65Jv||5X?^<^j-w?`OytF-F4|DDl)1u$omY6DDpZ!`kW7Fb|;SckrIBj%g
zKWvKgySRusDsVU7RH5xcflRlB7A!xjsJ29AiSV;(-AL9mhg$6y)-N>Ax8HxX{IKEi
zKde$y)%JYlKRCsCl9rW2+`GWdoh?NtDh}OWz45JoL>t@Z^alyWnVP~=#G~^KME9z%
zT+<x(Pj#A0dFhGX8x!xfKM`;YOm5MX{CK0OPSYamL)e?PZU6qg^PZetC93#$L0)2&
zR?C4CyzFf=(q8TUoz&c!V<>Dk`73YY0v6UqFBZ9dDr?y1S9U4zU&=MMztth@nS%cw
zo7?uB=SI)fE<Ibf?vp->$DBk?CPrVLR9>Zi`SX4MprXPDYxq{2IpK9)C3KnE9xVk0
zAIS~BE}cvN{%&s7A<leLuP$kyWp}TiW=onW^R1wQ&#Fsp;nzpsyfShxUjMNDtM!}5
zw^=G}%8u<!ni8E^5xw8a{^5ItU&?<!%?vP^ROjLKL)*er=JoyZ|KHv>ecsIHD!e?T
z{iZuhiz!QsV&WYx8C?xKleMkSXE^QOAO0{?iqrX5$*jU{r)~Ftu~D{56`t?Z;oszR
zDa3itMfWP(x2_r5dHuF=s@-M1WuBjsFTGfDY{#=3VbV`Dy0d%Ee`QJAdvT&v#-nP{
zNw$hd>i>q_F|=}h_Jq}ctBSG8b^Gi8jWmrU4J{pgFR)maO0qtm*vzxwms=X|;&)${
z{QvNYTYCRX$Ad?X?ERH{MS0Sj3i~OQ%P;nwklO6g@rr|$HA44yG*8f91C5H+FNJSN
zZH&+;THhkIVU>Qc{3F)V-XjvZi;w(2F=NWFFlJf4+ymQgn3)KaKRP_M@r2mgwX59Q
zcOKoTv!+GVaQ0&vXOZ^@tQFRG*@_iQ*{N%Nwu#ZVh^kn%@j>to-m`|Sd_Mp9E`{8Z
z+)+JKOf}23rR))JQ6}T>jGM_e5$iAfZ!?>4hDEe~OV1KsshYPRl@H}a$AtMkR9kRF
z{-t_*Se{|Z=Gp(UJvUW$p7RV}Guck?)cg&<nH1(Z*l}E{nXx3qKh!|x{i6p=DMzon
zZ2uMh-Zy>8f6KeKO1SJ&-s%3Blo7h;)Bow0^2}dKt?<m|Nm|msnnSROdkyn^qj<Mp
z-aGS?Q=|>%Uzz{@$;X^8%&*_1nfFFF-dT5{pYPWDnm|v#c-P4A^4O_tLQ8G>Tc`bK
zu4BK`XZ^JLSxcIpX79NhGvDu&?=y%!-XH&L0!!EORF?ENn!ehVj&+NBo-A+p@#2Ur
zAN$+Op1H>twSM+I{piAriQ>g4c4~f1bBQ}O^+ZX+l3wvCPoodczPTy;and=~m||w{
zGLtv9Q^MwjO$vLMI$@QU+-4zBmTd*06QXDM9@$_vt;Dfo=fAIq#c!;+yioAh`k51S
zj%?DesNWKIs`?|hs=l_9xzxqgH?#~5uNnz2&Q)bGUljlKMe_RBR!fvVhwV$zmz0&f
zF<qhehoWuW+i7QKU!PZZZee@xU!(u!{<YqZ1^JiRSATmTZPMg?f8T>mO#u(LgoP+9
znZdbJe^b^x&NS06c{RF)!I#!ec^j5l@=E<&Msbp3%uJz2C)#5UPfjXNYQJKt$v&x1
zP4;%{zG<Pe3oOr{*vLEK^y-(Z_x^l&>dS?)4X0YQPOSDkZ`b-nvfU`gFR9YaY2Kb$
z9F2S)>n}}-73E5)WP5z?@;|F-R~YIaRo^hH)l@MmzO+91qKfUEFK0xKCQIK^w>vod
zf9}>J>gsj!?KwQ(<Mw$!YCE@U!~S(<@7OEvT-oLOD_3XUmHeE`$J;9|`A&S$dh~z6
zk_l_JUKW4t9(7u{?b<C*jm*^OFV%~i0ys{mO<4XUc(?z?uYKkT6YE1Od_9Y=ajsL`
z>r>zH>89cQQk!|xyHfo=s44fxRvi7N8P2eM!EXlT*pSC}d6)E<3%t9^ZP%qLDe`KS
z%Gw#}NBnMZ_w4n4?I^cv-dpFxSF28T>t$SIV6I%@xnR*m`EIuTpL@2H`fad2bb7v{
zcJDrYlbuyd#50x(rijLTp2z&qonyw?d3KJ|S{%}Dtclum=56=oJE13UZjODvZceOa
z_htQJPwidHoa_3}vhb@UX@~CQ(vw=G5I1o}*Nwe0Gs?qbcGj=@DQ+@Jzje_To)sRU
z%U3?Wy!Z8EkECniAy@9tVeqT?Xk}BTw(ZAvmrI}b-1}2o7_W2wRqcj5%fCCW65@*P
z`MYTIf@S|xtmiptpNd^=K6#GWY#Yh`2TMCEZXEXPP?8N#fBEmPN?V;k;f&o^ulJmH
z+8mj=`tZ4(-C;N4_x*RMe(84pq%I%Z^c&n6E5g1UT(0qc+RO*v4z1R>#C}xU$!twy
zL<Ill@0#-sUTDMyvdVUB{K>cZM$#!38wvjZnMzWf$7E_F4(q!-j-2B@{k~=WgXA4t
z?PV$YMi;8Ki}5NdY`G|J$+i53|FedygLh){3va7=^Q>I?{ki75y2bPJGEFuflwKjt
zYa*-vpz-(%>E~B`W2zETQ&bI}mb$9A7u;Yq`F>{?x4Pbfv*J2it#140lwR8HG|gbr
z+HjM6HLeSmClA{%IAc9O;O3R1-f5p61ipwC6Oxw{cR5jCp4y;R`t+TMbbz68{WfLY
zWl!~O;}>8581>=#w2PN6bE-&izMRNXv61!TPR-T;4#|KCGo#mM_2siYi@w{tzVgro
zUcu5()}ynOO2YzDodgZ8K20@MvE0~EdSh$$>ua%d@7|p|ce!8byP^km$CzI4s<|$m
zuV;DXq+|8zg~5_O{1>g1D$WP2J-4r6vn}UsZxsdAzkk0iF6LNK?y}>zLB`qmhNy>k
z^k?5Ubv-AkyX=4M>b0v^YaE$(SaILq6XlZxRgRs#^f#k(PhHH4$@7HEzGj}gej!(x
zsnGsn+mRi!IbtqH-F4fjIk)e=fYwd6%2J1h7&(Ce-TaId>+&AxYzWf6@@l#L$BN^R
zt?l(Ld9LRa5)X^GblmIxx7AA*SE*kT;C$ye&+*<rmG6rcI741}_|B3llF*!7c0k>{
zEoAM(t_L3RakI8PV7KT{D_P(;-STC{)(>yv*W|G*YGIhD^hk5D|JmOcE107C*_4Y5
zqwVk7-F=d#-Mjq3^`kGU7f)U7t6_U_b-|nVn=jSQSlwHae_!F_OpR|>y?40P9=CsT
ze$u_|=Q(U%xwidRWaF|~b^ph|(~=$PWl^VtZ9A^jFO6kun6Pf9bKxW{hiAqeOA1-0
z=dP4YJTBw$rAFvaC*RMc{KZ0t8b2DIi)78Uc{gp(oAvUuX60}3QF^<}``6E|LmJV1
z?jDc+{x<u0>%H8sc6XmNDFYv?kNcT7e(`dST6y~Y$+f@b4Rmt<*azgKyyP!kG3UsQ
z=y=XH|G#qG8!KM9*vRnL8*$E5{;ckQu^_L|?Dw|2lT^Lm<k{2|Je;f@e*X431p{8Q
zk_G9%bJRn0es@<X)$X#ly8Qq9%<zxf!>lj<4~So)>?^t~|08!)U_0Mb!QcE3c#p}H
zi1=MNrWXD3_)oK)p?Qi|{QYM%>3+Q6E-ZL9zWs4~cbCMQXE*fKxO*%+pK}D2UH<m{
zxo&Ln&wu9Ex0vUb>TP%&mnS%D$z(rqCDu0Ge#>8S+ZnI5Sy&mI6?!SEtaaA+<p;}m
z@0RxjMDBOH!o$04>9zBEca>c=?k-l`I6)vGHM6<SSua~~{nTvT_1S^*rX9H~8M*1g
z)9W%pZ=$5<SZ=K-zxJvoBmGsB)V_!N8cXuGItd4?w-=chC4FQ~n9|=x>!<qnd3u)U
z-{%uN@^|T~my9+!b`n7oHt26r{8^cD_ulzMiPDdszc-y3&Ax8xsm-0D)t{MvJ$YTS
zxJtKe(VIDKljG7CH07MQGs~TwXNF>rwNGZJ+G~e$Z}nO&3x#R9+)WXBLQ_^|?pxWn
zjw?Sw%(<Mm|Imdqb?R4*_Zz(7QQw;wk#?_o*{(A2)i18Eua4bazSRA=WOvPWxv=n3
z_mZIfb(V#{^}Y76>%N_}SK(MSqlsDZidW}j`i)<#oZggIoLgY9d2{EKKRvdh0c<|;
z7q@G0igG->%F&qS)b$~1evVwqqcv7lHrM*xqY7j9=gM4Je|ynIhMN5c7E5UzTqTpW
zFu`8R;Yn_CynWW3{VpA*!bYvvzh3e``f%a0pZ=GuKO`1P?cHwoX>p6%RrYFcn*+k%
zRw-RSUgcu_mRIfXXS0%snoaj#IrlyZVdniKS)+e@Z&%KwQ1vCxr+y2mU$`zn+%U_^
zcEJ@h@4&3-a<v^5(X2OR(=XW0bkvqo)D?R4d&SkM#|~e(?a^|qoB!M6?+5k6Rz3~)
zjs6%B6MVTm>Eolmj}tlrepD<fyeGE%oAqqToH_1%|Nf+R>}-}S3^?)r{l{NH;mbs?
zWxLL`s(lfE{q#2uzRM*noVxrMBLk0C8wY!KBwEi?;jU1>W*zC-H1Wm?8`+Z3dy^w%
ziWbgjvGxw0F>@~eQJ#&uG5@QU?Em-GU@zNApInu5i^D(q8<l3cnM|<%X)|^Ip{w?0
z7N)ynrri>o<9@<U!A9r8*2?Rd|FlZ#FIDjG)m$@usuOccR*{M8&48=RFX-$p(Fl9K
z+I~Ba#hH1=vR`*Tj^VttXx$<wrCXoZ-g>&=$p_|S`{x<2p1<M<Ul1XcT^KHMu~#75
z)gk!c!uU+h<TvVvwrYg@H=SnpJ~yYh!K(3avdISdSCS^)OiwRt-S~2`m<8k0wnE*B
zvlg5BG!|Yz^fMu1>Mq@kzj8PD5_#14TN;eFd8IyLy8Nf+_2r^U6E)T{=M4|zCZrwz
zY3nX39-Q5vX)N>N??Hx=wg;iI{yHA*hXb-BHtK!7ackM!(AB1j2U;fce$Vzet+;ua
z&;>TBG~VX7@^5dx@a|fDu_>Y8_~PK163Qv3enjn9Vi{S!jC0qoniY-8C%g_nxw?9`
zXZ(D9owM%m#h0$m%$x9YS%V_;=60SY#pbQeCC%1epRQ!}syWB5Iw@(iWN*~uc{3Mk
z20kxSemh~I)72CIS!;hUPsBETXQW`D06Lw}#zvp$GYJhXEi93z@79J^21q{_jaQ6V
zZ+OtqrMKWs>{A2J+JIS0w%#?fl@XR+Znr~5Jb1%_h9jzqe2NzuS*;#;zF<&PWSMku
zngYi|X6dW!KUOWiF)yktSv@a()~+&H|Jy$5#-97uUN?UG==R)?U*^9%w`b=2TI>7Y
zt?yNzKA4nj$k+b)(J$W0$eBM+*X=A^yu|I(-!GywzaG*mJo)CD|I6tHRf6qtGR~7@
zW_78q&wn8H-&*n6y0*@LuV2Xq-$~+myk|#4?~U!B(;uIGp8mN1ra}0emZOCh6^^NM
zAD!Lu<;(W_Q$0kD3U{uE`;b=kD_T2jL*dp15sDF$)ov*0tekS<wboIl#It?}+qU#(
zHu(x!&KBXmH{l9Lm$&PoRYr|Vd?rNi@|`e!^2E8y5!3ft=6jzjzx<E;%i*VLook{Z
zjy5UB9ld0r7<NdsaMgxc5|KP@N0UBpG`$g}!(7cD-JY5xy`lF;S`PQL)~5+OBjmOz
zusz-&KlkXrQ!5|Vh$(meoV2<>R<2qnFYpfQpURpNzgKe(A1M;?`C__g>hyqDs$ZNf
zpP%0{)!TnTO`Hddl##f7YtUoAN9P1uZn$Wx9^|jdEpFz_(PZ}OpR=a9Y^URIZMD|_
zzl_Vn9!VK?xF{9`E&mv!W~wKCz2lu~)k?vb==f=DQi~2QmQ!dyXqQmvw^-z#{rrjl
zh2P1v3mjEh<a6&$;QIT_GdMCB*+Unl=>58#dT&G0mMulgQ`{>Z9@-1c;h5ODi({A2
zLxo35)9pIq9!FZWA6>%IGizIx^JF7;zLq`3C)^yj&#cv7V7KomZ`uo`J-X}Cg7{{M
zE#6kN=+Gq=mgc=@bW&HOUXI+q_}qbfK5Ln<vjHDHE%=-#-(7bw;lAbzwHcu=MCFXr
zRh@J6vi84lE}Xx(QvU60ZJipHltPivnuANf$33i^#TI=cEaSRlPmybDOK0N7HBI?~
z&rh#kAI$u~&?5Bhuh`rGcaObU_gCbfKhpWO{a$K?kI-i!=ZO=ZYD=!3r8BcJkmKN%
z%vI~}SDoZ{eDu#(qi?THXKL-A`i*YIFP!$B%-Yu2%i-bR`;Q~sYoU`6gSctybR+F|
z+!H?43nV16sBYYFLL;l+F1Yz>!v<Tm%L<}L^ba1l{cT$>v3)+Xd|_F11@GpDr}NL~
zB;Ks<{=4;Ie0$>m%WuM!)$`wK&yt^Twc^FC_x<x2lWJD%oaU7>OX*7mi*853q3sv!
z^9o;fe^GhAz+%U&8s00fx7@z1y=LDvog~|JYf^Xqj(oTO=;oQ(MQME{FUveGHr%_C
zw(Z85*avEEuVv1f_HH$84W7jPMAV|>tl_2l8-nln<~x`Fs6QC1EHFoOQWg7lwl|Z-
zrsuA|TFk)Fz5lX9<G}^&rh)bsY_^o&F!!@7(&oI|QRX&zZ9nhiCdak;L3vZd%orBF
zd{Gv7fjjNF#>AT4>E{KHG+#cV&bqKSMNZ`OtkCug{_~zDNu*DD*W}15o8U3=`)b=1
zh8*W@w!be*AHSg1!YsJS^h#hrz=<>0Lj+ES#dgQB`nuEwlstLgexZC1e?Z*MnJrfW
z>rH+dNBw2-nCf_W&(bBEq&SvddbjkzJ@boirGG2e|5*I5Sx3(Q)S14{J*qd)-03pw
zu9<R&IjsGvZB6N^$E-;QcCMbuwAS%<X3SBM&tJ2Wtxt5W4PGQ1KiBbNa>}L!9d+H?
zx1YSLkUQb3`BzhxntzQKBW+5gqdl+YNM5yVQDWYEY{4bdQ^sWs3s1k7T7G~_&0GG^
z%RkGPB*q=MYRbQ5XDj!+e}^tG<u33(xkG2}&n)?<8~-;%zudm~&hkG2FGW}VIaYLy
zf9}O+3l2Q}aC`4v=U@#^KY{1o{CP^<3tR%jmd(+6y_Q+<ikX&H)T$861*-1~<s0>u
z+g^D3uD_z8VAibL8W&2WtZo_2F#K0r-@*Pbf_=_$HhbB9c4g<v6-;VnIMxS8J320$
z&>7BE%C%QWs<SDu@W;2w6@f(}FK^5<SGIpY`;4)D`ZL3iXHN4>EmM13?6Tnb1*a)q
z>x@53_*n`6-kkf@Jw4!!ct`QqD+%BDj3!QWz3~0r+{@Cg<q^-sw)B@TesCyi6GNGo
z%BEj0oi<CgC`?|FHCrn*df`_^&CQ#w6Qqmge%WZ?Yh=rMjBlFQ<cjc<`t>@dQEj_#
zit&2fyryxae-XcrZ1Pg$2~8XFF9fM9xMysiT(m<vvn=O=?wdP*MD9L`jggLV)>4rE
zwCRd`*bnZ;2({!O-pvsjVF}Z7%g^t-SHRFAx8BjIk<W_fRB6xr{D4hN((^@bocEjM
z&vd!^aPsf|W^TTc3h9z6U+EJayADpf%k234qx-3S@{X^SW1<?TeR$yZ%E7U`Z}AyE
z<qOI3oK<eWzc(}gO5iEezJ0rQ?S^iF{ui=w`>TtVDQuY&88lB%W#PW%*Y3ZVUfnbO
zs{E^S@0x-#!a}=5pZXp>CKceU=TRFQyRPnoz~5O9vzB`NK4`RMh4G1`zc)&>DsI>p
zvGCoW{Z3ptWAdElvbaZ=ZnYncWN(Yj?K;+^lqh(+y>}Y}tC>Oco#%o2|G(AVec!)C
zAu8kL?7zpC@@V}QN#Wh%ZM`M+Oi<me<mKxgzMlDUNA6_T=YeNZr${IWM_0e}(XKjv
zGxgBPA2(t=55-hF*{Gc9KH~no_3pBp$CF;Y-KM58$1HM9TG`FiPd9|;EiT{xLEi3q
zuMD=P0XfGN8=IRUw+xa~5)u;=Qc^^|^Kfx=7hL@I@$1?1r?+3>_43s}>#4)!IrHy?
z#2<>Me>nHOY`6K$&pOTi;eP$UI=`+S);fLQzz0tOmwnIKC$498Si(CY;frWeLUM|T
zz}a<KIr=(h^*wdZo$)^Jc~x7}%;b`>fkELp{S`W=e9mhG8W<TET)WCTX&RU3R4%_N
zuDH&VH-A1odUGaMXQd`rT-ClRuDYL39S-`i_WrSHU2>)8$Q+Z_KPIgcu5h$oNVvWr
zf%Ut>{|gNXzP2XEop;v#blur$>ELd-in+6lP0h`1+J^VpKlat@{NA^E&d2I=vd`YD
z*T;yzjfgQ^zhw5}scV;PUbSj-%i@`PcduE$WceEXo$5RFcWz&_x4`l66uwT$oZw&2
z{NBZ0vkfn@s($}(U$gMH?>_HlUb3Cpygu<3|MSfIx#{1&Za7)EvE|9<0>4KOUp;$h
zcer$F_2V}yZEg4N{cBrXrT+OtPQ;bl<_q_imz|%|eE<B15C0-Gm>68<_Ap_IOjD#~
zT5g#+C8<UFZaIl1sV=F>`6;RTKB;->B^e4vMyM;F19s;fHsHDQRaEBk!U=Pk=CUw1
zGBeBla+q7w;B#?Z!wbeJxj)}~MU<V{g&s^Se-gOu@6PXLjenjsFssg3WA5V}A<5}B
zYpwWK!JeaA!c`Yv4m|PA`MCdE4wcG1mWQstUb_1J!3$q9;<kAd+<E=aCdXUp`ftsK
zASQ=@ceYMDA}!;m*7If3Mt!p**UlI0*yilmW_nE8@)M`lMHAO`burBcF9?|mW!p~<
z2))|yYM0vt^Z!-W$#Xu0Zw^>)6M6M&^vt^#&zw3p?^r?VzW2J{x)<-cQ&m-`(KaJe
z^&dmLRSve*Y^LUj<}kisLRq~QiyTZF-Ykq|VQz&5lg`Dyh7KW>$`ALi2lepCdNjGq
zolL&U`@GSo>6N;LX;Rb8349WFr9=+3bSs9LFX;7h58Jcp(&kmar~LDdYT@#{o7?sE
z_18<+^A{A>h3L<A&*-bM=#TcvV4o$+agbrv{n`kh#OBNGWy1SX-dJbww24{7?%t^I
z_|;!o-5{oOx3=}HmVd6X+^}k;fk)H?&U4WMwfBuw{;n@GSQdFVC1FwJeQV46&hKn5
zoH=!A-l;26hxToMq2KkX$oWUj#B#gMU4I!*YUDh@OixJb>O(3EQuY0lvVu!;_5BnK
zjScnPGK-2!6buaw4D@k_p^1S7%C(wjqqB=|yKVYgSI@d{ZAA2;J?zZ)*RH*E*lWtv
z<h6gg!k6+3SWc?&735+(RaW<Ro?L?pLz9r|?VX&?i412-zJIa2dztx==MiN^rG-H?
z%9@1*6E7J}(z?VJ?vo-U>?y4(Bp&-DM@mz0DQEwbE*AgPB|a)DEGt-yO!@>C1xsh1
zFwGRo{Nnw=J(}aXnv?iMrzwpNedpZVgES;1UrfB_l&E65W~0ojrF}YQHf%WXsBTfg
zmWe(u?R1`6bqcX|CxkIw75YB6F>^+URzZoClCWuu)tt$(eLih5?S*V^7eaa-3qD(^
z({9tPtTtW6;t<E3plgYHj)}B(%J_ypw)&-Nz_mSO(y9{&<<u97|CLpn=`g=}0?V(U
zJf%-DUPX%Pt#?@d2J;*WINRELA~oh<+zy`(6|dMO4;3tCE@pJh<cgoPeP_=Wwelh-
z&j4}W?}y3`vP!+suvlX08Ln_{sTA)C?}sdm(LC*;P6Z7Piy!DJCr%e=a6FjEAhzSc
zVa;Ds6)q0lo(m)<75IraOlj7LDNkY2G3#q{a^W&&QCZUWqIG4l@1e^lCM6sSX^2di
zJ#|BjM<3fk&B7xtta~!1D#{eJbhAcenx16PU)!kl<G$1crt1xY9#bE1@Jq+|cg$R{
z+Bawt>sLMr>7*Xz#f>Z}QH)PIKe*)a39MmDnWGoLe1M~|&BeaJW#Vo|*(cl;riN`j
zo{oG@9D)x;?lz^RpY(IJ;pBa=jM20;BJ1R95%D?B+ZLR7AjEh1VTYgSiU3ay_U&0s
z2Ce!ho=z&ba?pr1=fq2e4F*kfq^l;kGpt^4I!OJBlC`6D;)wu1AstCCzeS!3Y61x<
z)n0oXMGiSJPhpYd2;y>Xn_1AjkF$4?0^5F%wxwbfU;FDfY>$6$b9=Vm=Sh2K#!o9L
zGn!YI^!sPdll12@H{0Sj%)e=9eZVv4!++M>hYy&)S^g)t?pR-0#LM*Uk2B3QXLP)G
z3ui8?ICt~njy*q~+uxiiKka$Y%o$%aPbTv{J9Xq3=R{Yv9q%r@uUGh}F@IW#gN=i>
zO7pZ(W{+^*7AaGO?<Ldt@BjETXRe&w#_j*SPxj?+&aMxT{kHmWq3%!3$s$)zW_7(O
za6fS{Nz3WxjlfslAvv8tqK;h(ROpOQ`E+sR72^c=7RRez93L+fv3xOpwmo3G#QiDz
zOcuU&DVrkHxyo<OiO<cpyF2abO!RKvUh_lc#_Bhd_e=TL*!OnWwOv|zZ6oJSw_|Lb
zzS5G}s{ieMF5NjI;vc1AxHck6ePfIER+EiERmyhWb~+Cum5&vBZfL4&eed;A=%wA&
zt)caqLJGT&YhAJO%-LY4xh|>o%6eU!dAItErv*pzH><K}x7(Wk`xnB_u{yau^{&$6
z9W7c$DJtDZuPiNGWUh8u??nC1%G1Z6trhsIkvgSEc+EYJZH7|?-HK%<pWAZ%@oWEc
z+wbdr^2|P79<beuC-(W@h$x0rteW@c%3j{?$aVDHr^3J|!IoX^r(GkIzL|E#H^0e_
z(a=7$^1!YIm*gG&zTU7BxY04`Uf~pWwHZ!Fbx*Ca5PUTI#?LpK(u?^b`2W85T*@xC
zJ@s4dJFV|$@9fe%z55Ys^7X@sHMVn)|9ms+>E)9ZuiSrbNSOTbX=z>c@i@Qt$2Zqh
zy<YzG-R~T;b9(CgI`U85KH1f(>E+<M+hvM;lUcr1!u$l+NtvH)dE53*e{3|bCiQ5*
zj@bJ7YtI}=>hKbYHW6Lkk&tsJA=F9wqr~l|=Q`}QA~j{L6-N7Hryu>vC;x`Sa7V)a
zwC#3tZY3P(lTg`{^}$hTmDoDZtWuW+JogQ?1ucA=j#_81G?!md;QaRP1!Gom)(pn7
zqSuSwB&bJwZek9<Ie{l8wMlwM!a+Z)0xgqGjhAhk^w_P#wH>E3hrg8Pi&<K4QsDT%
zPku+%q44G@<}81+ZM4o!UeG+R{ldZ4lNT;e*mZj+$FU9f?f$9T%-+Ctds&9nd>_4C
z-&;xzw>5pWy}B!1SXg~^_1Wks%_SG^MRz5qU#@olWD@qbnZ>K5;c~XpD$^-%yQ^oJ
zzgJxm@mks~cgID!#9(pli}mLtu5@@amtNY{UgXqKmzq2CZMB+b($+aMMb}o}VEs_n
zE|F7tdFlJLc|Q%GZZ!FO&dBX}%|gBs%P+2x7nnB*bSWO}Io)~xXKLMd(cd39Dv!lK
z`?>r?{l~+Vmgkl?pNpF&bUf3{Icc%G>ylUDEe5uSUv<<(=f7UOI>%_|nYQ^>lb-Ba
za(U+d<;#<_SFY0*?M$w_K5>ber<c^EOY2mkCNAE7^UbZp%CeoyGz)W8*1m8}Z&@bj
z{x~wQ?Rfc;?<(1IWS$hCxFx!`Y}eGeUu@Pde)o0n{JZ^spZ6VGl9+6prRY@Zx*%He
z?cR>E`1px|-(7o_z2YdbWooYd@?)c7?02vG^@55UgF81{OP$}a=|^D6lfNt2f0k@k
zy88UJ$DR(p4Xd2WN=#0!yZ&I4@AK2kK9^qpx$64Q%;LN>@8|K`mL<$mn&f-VezyPe
z$1h9WFKuuSddqWL<h5C4Ex+NntrFsab)vHV-9={dL3heHt!}MRe_6V#MsV-R^{38D
zFs0PKY&U<p>u2QUuiIW$eP8j~dX4Y)FMmU`)qH=n+<5Z)@$KE6%d>Q<YM(D(?`OF9
z%`Y?Uf7{i<u1$Yc(ZA?a!NJ*4Pt`KM9ltCXbFI~><7|_wzSt|*+;aCx`4#adoBj53
z*}VB{y6OKMU-zT$UT2oPN)P|}#%wm9t;lqZuzxQ;3&gMVkG`?o_wC7`-Fa2Hx>xR>
zN&o)u^@8^|Gou|uxAVTv7WTQ!xGJ$}iIT<!8=a=~eJ7r8UZ}t5)ZNgDwOy-<IlW>J
z$}M<zw6ZYH^82DyN4h^mdo2CCC+N!5*W0GfE_q^>RH*8=O84l=w7a|C#9he?SY0?(
z+4*C{)jrY07v^(ZFKVnd3c5A#=gHpB#j?%2K70Ruw*1JnM^`p|u??@S3r~x7+?5;t
zb&2tbw~sr$woiR+RW$qKoukg*A1RhT{l03(dar_Cu{nl?x>Y<^W=klg-geNnW?N<F
zu;p&poJ@grhIV($_PmzOe=_ZHZ$-z4zkKVuYJMpi$}rvgV*KdP!DYJ=efsD7ZaQgj
z*YHf)pPXNRa(-@mbSch0a_iHI-99|ew{>OT7QcQiS9`ti^-Fd$pI_bAQ8$<OS;wmd
z@`r+FWM$0#F7|Nyf}@AdaDO}1f9z)6KUEv~jVE$Fcm1e+p73n5(fqe_Qu6=iUwmFz
zDSX%Ae9p7=i>$A#3NT;bF0%ONh4W=~pRaT$zi<7|b~Ry5b9>~=a=v&EJ(+1aDg1AH
z?mqpNKj(jLz{`bStMo&XzWVje``&tU&nfqLd$s?(uJk<L+m-d*Kc6*V`_ZdUliWYY
zSLy9|qM*ImvrhKFlIuK9E5F|9_lbGlQ>PK8xcAa;37ai1GXn2R=A>`?8C0Js7F61?
zm1Wb~1@aGrx0GgFeqX>D;j_O}Ysoz?{kxtMn`Ok>jzm_?Zdeyp)!I^bJls)i?}Dbc
zp$!%v3O^dwe7xR}vSbcRR#hy^BhRIUTXMxBB1$cmZ!6Y{F`p%J$b@CNlkY-DE!X_T
z3B^e--ttR1zju;|&aV2kyezeXNns!3p0o2=GrawCe&{*8V?C{Q@T8-gM@HWY$MuJI
zn0$G_&G5p_R8-1gt(nSOeoej!S1w$z>{L-L*_Xt)b&mJ+n3W{~@lva^)|RyEtWb#G
zQFid|I__OAne}4A&w78ZdGev--bao%t8d0k%ukGdI>ltihTn?aTbGvCPkP#;d2gCw
zu=}DlODFqC1xM~I_&w>+9^M<5?Y8<=*@%Xz{fgdd_4n25+57f<-ENyU`y>C?$9run
z`|tg`=l|C($+cpNdQye`pQnMp-#q-B{^xGNzDZ~HJDz^5{PUX759Y=DZ*oVUNvXEJ
zJ@5B5{|Q@r6z(NG-~1%kbd&V@vm8}AOw~`X7~fD++V7w8QQhsqww(_*@_bCMogG(e
zckX`UeEuIR&(7+dw>GVCqt3CH<p-X=pHSkuf&Xz}WpGQm%4)6huqo{3H+3dG^z_N#
zI(+l=OM~U@i4W&Gz1lfn)b50}h1HK8|DU~of6i`B+GI~Nspsmu(|_iQZIS(zWFdWj
zVG{p)uWja%Z~0bqzpwlD`1R~bb-}6CM<;xf>D=;VwR^^UFZCOSUC%!j<X72#6im<e
zs{d_QG%J*UTJD6M{?YBRjk-c^A?tIJe8a^bKD$`)=w<2Us^|Sb_kMiJw_9rSYR=E`
zZIj<j`)haJ+g>^NhHP%F+P_tKq2ld7b)U7@9`BuRlw)`7XNTzIH9HdPe?*@9v@-qa
z^Ep$^jo;rCHA@$$e;;Z1?+{mgzhUC6oUUG`6yHypdiVQQUwdC1e_>nm+l{ql2l#G!
zmmR6&vKRb+$nS>sH|^@Pem5nn58vBV{>JwA0lQ5<OP-`3HJD#HZKJnYiZa{LH?!_;
zF#K)*t<(30?(9u-?ryw)vqyAIN^#kFyUqG<w0}3po>Lc&H?o&a<ZFJQcI!$`%eLl+
zYF<nE1H4YLYm^rKdB1t`_T0U{f2M5lTbYvW)oA7I^^t99&+DaACamfYc5~a^)x9du
zZ)#?t*GE&ot(KoG4CN9J2cB)NZ)DdNTjP9>rH1z%+kx3Kdu0!NE#5DBAj(|e+B?RU
z^L=sq{$%c7J0l^&{f4c2rp8qVi;cQ1<(F?<QBYrZ%;KT%1Ytjg%hGq7%$OJPZpvxC
zbBcBChxJW0{EUBu9bU7&YnQw6%i;IpQ*Mh-d!BIWo;;yL$|JAn#Wasqe3J?il*3$w
zPyV`@@Gp<6ug_*JUv7@QT#Zd_`M2+r!`FYx)cf&n=Kgv9n{G;Nam(Rd8^NY;8h$~k
z`wQ!O-|D1r=lkmpowMI#f1)+_@T%Jzu4x}EsjKchW3_MIra-%ekv6KYgJZ&VKI}j5
z_%6HZ-;;FPm9@4T>r<w0&iQe2I^(xpvAXvDC%$RRDSnYIx!G`WT8<(A#-pFqH>Upo
zdUN&V#kYjE&(!(kK9Tk1!hRhMxhYQqh5nswzmU7>M%*Tyq+&<Wf|ii~k)|$3H|Fm6
zy6?1}iT2(5=QgjOV3vIIDUX{O$94YC(`sH{TXxH2^W;jQ`+nc+Y;Rew<+ap2`1jta
zLo4#1&ED|$=&DywXD>57e$hT=(RYnYX7{a*W}ThYCB2@J&#{;_fTM?ti|LYnaQk~b
z@xO65g!=Mgmw);E@JHqU4-XC>nov^w<F&llzpBqs8ygmH4twD@cYDw8uU8)(6k(Xp
zIOCD<?%8L$<tGTJ?bEQ*_75`?I^usi>(i3uHb3P~`@c_noS61aILSKgRi5LjS)EKD
zf2gGW`}FSq@&A#!6Gbbp&CJ|qZd~5lw{z2`=rdcoLgrq(+VeU!W4TAJz~;2DZAaE>
z*+^&Z;dEQizl|k%?J}S6gavyK9lBwvGs#Tsc>419_0}~u-aGaOO}k(lxkzI2jU75J
zhfn|Ayn30VUPJ}gy1DOsyFY&VQ&3a-bfRC|K8xvBm3cnBdG+z=QQ_8rk72rZ{bhFs
z?4G6I`9pv9j8)N?g9?VGmWa_8@UTKeP->cfP=0=if}w(eLXf^^URu5aXvQ)~-_s@9
z#?m0!($vf-&B7ol$=J*!&A`aq*ucyzB{jv&I5p8Y#g4FwSpDG4s#GpRGXwqbXhQ{K
z1*2FzgBHdXD5IOEm%PLI5=4%E+$$XWq;~NgrLS%(*NZf}f;?ADnWU$2JtpYnQkNqK
zZXCJrirFORnD~haGaDNt<qt(gzTSV%`sdoc-z*nDx16+P*DS@<*K+rcn#mmZoa1!!
z=V_sztxHo<9;==?5N^@DIV`L^XWhZmKg8}wb{?9MmZZM!XldBN?K9FIimuMEo0aqC
zn2h#U?O!o(4t@=LI{(Gz4gBU2q0Sf8KkN$qY&%QkTXbpsqLarNS3fN`k^ab*cZg~A
zTFYgv6_;Pl>Pj=dx9QjF#W$OjcZhC}JpA(djXw7cJp2z6bq+1wA^knUGVCGWoo>~L
zF3mgb)`j+Aj~Le+XpZRFy+JB^BWwS|LYqd@3Zd5<#QYy-<{YT9X-?lLDg9B#>|y1e
z=C>7$`bqj_3F2`_9rv_tFXU+dxX*OfzIE)^mUo?4eC_(q6Z~tJ)=!j=w7<0f^NnBD
zaz8t_elq?vWA0DOPr=sthM#Qw9vIj3$-g<`^JdSH%$nZ(0zUg&CtGjS<S*}9`rldF
f&KNTZ6_+Fyl~fd^rg52@7#JFJsj9mAyKw;k$TFA{
--
GitLab