Accelsim(Exec-driven mode)环境搭建和测试(GPGPUSim v4.x)

Accel-Sim (Exec-driven mode)环境搭建和测试，以及iGPGPUSim 4.x的相关运行和测试

Wang121201

1481人浏览 · 2024-07-01 16:55:17

Wang121201 · 2024-07-01 16:55:17 发布

文章目录

相关代码
Accelsim
GPGPUSim 4.x
Accel-sim（Exec-driven mode）
总结

Accelsim

Accelsim components

Accel-Sim Components
1 Accel-Sim Tracer: An NVBit tool for generating SASS traces from CUDA applications. Code for the tool lives in ./util/tracer_nvbit/.
2 Accel-Sim SASS Frontend and Simulation Engine: A simulator frontend that consumes SASS traces and feeds them into a performance model. The intial release of Accel-Sim coincides with the release of GPGPU-Sim 4.0, which acts as the detailed performance model.
3 Accel-Sim Correlator: A tool that matches, plots and correlates statistics from the performance model with real hardware statistics generated by profiling tools. To use the correlator, you must first generate hardware output and simulation statistics. To generate output from the GPU, use the scripts in ./util/hw_stats.
4 Accel-Sim Tuner: An automated tuner that automates configuration file generation from a detailed microbenchmark suite. You need to provide a C header file hw_def that contains minimal information about the hardware model. This file is used to configure and tune the microbenchmarks for the unduerline hardware.

我们后面只关注 GPGPU-Sim PTX emulation 和GPGPU-Sim Perf Model, 即黑色框的相关部分。
上面具体组件在accelsim trace-drven mode下的运行和使用情况可以参考博客 Accel-Sim模拟器编译、安装与运行

simulator
后面的运行环境是基于PTX (vISA)在Exec-driven mode 进行测试和运行分析。不需要真实的GPU硬件环境(mISA trace-driven mode运行的条件，也可以下载官方提供的trace)，其中trace-dirven mode 的simulation rate会比exec-driven mode的高。

GPGPUSim 4.x

后面运行运行测试的平台和需要下面的相关代码：ubuntu20, gcc 9, g++ 9, cuda11; gpgpusim 4.x 源码(accel-sim/gpgpu-sim_distribution)；Rodinia基准程序；Rodinia的workloads修改后的Makefile文件。

##install source codes and related benchmarks
wget http://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda_11.0.2_450.51.05_linux.run
git clone https://github.com/accel-sim/gpgpu-sim_distribution && mv gpgpu-sim_distribution gpgpusim-v4
wget http://filedn.com/luEeJVCCazShDlU4ibloXvu/class/rodinia.tar.bz2 
//wget -O rodinia.tar.bz2 http://www.cs.virginia.edu/~kw5na/lava/Rodinia/Packages/Current/rodinia_3.1.tar.bz2
## 下载Rodinia修改后的的makefile： https://download.csdn.net/download/Wang121201/89493888

之后的文件结构如下：

~/			 : work space
|—gpgpusim-v4：GPGPUSim 4.x源码
|—cuda_11.0.2_450.51.05_linux.run： cudatoolkit
|—rodinia.tar.bz2： Rodinia基准程序
|—makefiles.tar.gz: rodinia修改后的makefile

## install cuda 11 
sh cuda_11.0.2_450.51.05_linux.run
## accept, and don't choose driver option, then install
## add cuda path to ~/.bashrc
echo 'export PATH=$PATH:/usr/local/cuda/bin' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib:/usr/local/cuda/lib64' >> ~/.bashrc
echo 'export CUDA_INSTALL_PATH=/usr/local/cuda' >> ~/.bashrc
source ~/.bashrc
nvcc -V
#nvcc: NVIDIA (R) Cuda compiler driver
#Cuda compilation tools, release 11.0, V11.0.194

## install dependencies
sudo apt-get install  -y wget build-essential xutils-dev bison zlib1g-dev flex \
      libglu1-mesa-dev git g++ libssl-dev libxml2-dev libboost-all-dev git g++ \
      libxml2-dev vim python-setuptools python-dev build-essential python3-pip
# GPGPU-Sim dependencies:
apt-get install build-essential xutils-dev bison zlib1g-dev flex libglu1-mesa-dev
#GPGPU-Sim documentation dependencies:
apt-get install doxygen graphviz
# AerialVision dependencies:
apt-get install python-pmw python-ply python-numpy libpng12-dev python-matplotlib
# CUDA SDK dependencies:
apt-get install libxi-dev libxmu-dev libglut3-dev libglu-dev

### gpgpusim 4.x code 
cd gpgpusim-v4
source setup_environment && make -j64

### rodinia 3.1
tar -xvjf rodinia.tar.bz2
cd rodinia_3.1/ && vim patch_rodinia.sh

patch_rodinia.sh的代码如下：

#!/bin/bash
tar -zxvf ../makefiles.tar.gz

if [ -e Makefile ]; then
        mv Makefile Makefile.old
fi

cp ./makefiles/Makefile ./
MAKEFILES=$(ls ./makefiles)

for makef in ${MAKEFILES[*]}; do
        namef=$(echo "$makef" | sed 's/\./\//g')
        cp ./makefiles/$makef ./cuda/$namef
        echo "$namef"
done

# add CUDA_SAMPLES_PATH
echo "" >> common/make.config
echo "SDK_DIR = /usr/local/cuda/samples/" >> common/make.config

然后执行脚本，编译Rodinia基准程序，可以设置编译的SM，(cuda11好像支持SM60,SM70,SM75,SM80等)，并进行测试

bash patch_rodinia.sh
make -j 64 SM=60
cd bin/linux/cuda
cp ~/gpgpusim-v4/configs/tested-cfgs/SM6_TITANX/* ./
source ~/gpgpusim-v4/setup_environment
ldd backprop
./backprop 1024

Before we run, we need to make sure the application’s executable file is dynamically linked to CUDA runtime library. This can be done during compilation of your program by introducing the nvcc flag “-lcudart” in makefile (quotes should be excluded).
To confirm the same, type the follwoing command:
ldd <your_application_name>
You should see that your application is using libcudart.so file in GPGPUSim directory.

Accel-sim（Exec-driven mode）

这部分在运行时使用accelsim的框架，包含一些运行时的工具以及脚本,和论文中关联的基准程序，以及accelwattch功耗模型

However, we encourage you to use our workload launch manager ‘run_simulations’ script as shown above, which will greatly simplify the simulation process and increase productivity.

git clone https://github.com/accel-sim/accel-sim-framework.git 
cd accel-sim-framework.git
pip3 install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
source ./gpu-simulator/setup_environment.sh
make -j -C ./gpu-simulator/
## This will produce an executable in:
## ./gpu-simulator/bin/release/accel-sim.out

## benchmarks collection
cd -
git clone https://github.com/accel-sim/gpu-app-collection.git
source ./gpu-app-collection/src/setup_environment  
make -j -C ./gpu-app-collection/src rodinia_2.0-ft

./util/job_launching/run_simulations.py -B rodinia_2.0-ft -C QV100-PTX -N myTest-PTX
##PTX mode usage: 
##./util/job_launching/run_simulations.py -B <benchmark> -C <gpu_config> -N <run_identifier>
#Optional:
#[-B benchmark]              (From the gpu-app-collection compiled in Step 1)
#[-C gpu_config]             (List of supported configs: accel-sim-framework/util/job_launching/configs/define-standard-cfgs.yml)
##You can monitor the tests using:
./util/job_launching/monitor_func_test.py -v -N myTest-PTX
##After the jobs finish - you can collect all the stats using:
./util/job_launching/get_stats.py -N myTest-PTX | tee stats.csv

## Running AccelWattch PTX SIM: 
## To run the simple example from bullet 1 with AccelWattch power estimations enabled using the AccelWattch PTX SIM model,
./util/job_launching/run_simulations.py -B rodinia_2.0-ft -C QV100-Accelwattch_PTX_SIM -N test-power

在这里插入图片描述

其他基准程序都可以直接编译，然后相关的数据集需要统一下载一个压缩包all.gpgpu-sim-app-data.tgz，可以查看gpu-app-collection 下的get_data.sh。

gpu-app-collection
The repo itself has no data (since git is bad with big files). git-lfs is one option that could be explored, but since the public, free version of github limits the capacity and bandwidth using git-lfs, the data is simply retrieved via wget form a tarball hosted on our University servers.
./get_data.sh
will grab this data, as well as:
make data

总结

上面主要是简单梳理了一下gpgpusim 4.x 和Accelsim的结构和运行测试。参考主要来自对应的代码仓和论文，大部分都是直接引用了相关描述来减少或者避免自己总结分析后的描述可能会产生的偏差。对于上面结构的分析还是比较浅，也有一些模糊的地方，尽量抽取并全面总结了个人认为相关的部分。以上内容仅供参考，后面如果有了新的了解和认识在来更新吧。

技术共进，成长同行——讯飞AI开发者社区

更多推荐

YOLOv8【主干网络篇·第4节】LSKNet大核卷积遥感检测专用网络！

讯飞AI开发者社区

上下文工程驱动智能体向通用人工智能（AGI）基石

上下文工程驱动智能体向通用人工智能（AGI）基石-摘要

讯飞AI开发者社区

人工智能赋能智能交通新时代

人工智能通过分析智能交通系统（ITS）产生的大数据，能够实现交通流预测和管理。特征工程是关键步骤，包括提取时间特征（如小时、星期几）、空间特征（如路段拓扑关系）和环境特征（如天气、节假日）。人工智能通过分析智能交通系统的大数据，显著提高了交通流预测的准确性和管理效率。随着技术的不断发展，未来的交通管理系统将更加智能化和自动化。时间序列分析是预测的基础，通过分解趋势、季节性和随机成分，可以更准确地预