Benchmark suites for GPGPU-Sim

This repository includes several benchmark suites that can be simulated using GPGPU-Sim. These benchmarks are tested using GPGPU-Sim 3.2.2 on Ubuntu 14.04 and the compatible Linux distros (such as LinuxMint 17).

Installation

1. Run get_cuda.sh to install NVIDIA CUDA and CUDA SDK. This script will execute the patch for CUDA SDK automatically. Use the default target directory (/usr/local/cuda) for CUDA. You can set any target directories for CUDA SDK, but the all scripts in this repository assume CUDA SDK is installed in ~/workspace/cuda_sdk4.

   $ ./get_cuda.sh 
   

2. Install CUDA SDK. Move to the source directory (~/workspace/cuda_sdk4/C) and run cuda_sdk_fail.sh which will move the sources that cannot be compiled to src_failed folder. Then compile CUDA SDK by executing make.

   $ cd ~/workspace/cuda_sdk4/C
   $ ./cuda_sdk_fail.sh
   $ make
   

3. Run get_benchmarks.sh to download the source files of the benchmark suites. You can select the benchmark suites that you want to install.

   $ cd ~/workspace/gpgpu-bench
   $ ./get_benchmarks.sh
   

4. Run patches for the downloaded benchmark suites. The patches will copy the modified compilation scripts. The following code exhibits how to run the patch for GPGPU-Sim bundle benchmarks.

   $ cd patch
   $ ./patch_ispass2009.sh
   

5. Now, you are ready to compile the benchmark suites.

Compilation

0. CUDA SDK: already compiled while installing CUDA SDK.

1. GPGPU-Sim bundle

   $ cd ispass2009
   $ make -f Makefile.ispass-2009
   

2. Rodinia 3.1

   $ cd rodinia_3.1
   $ ./compile_rodinia.sh
   $ ./compile_rodinia_nvidia.sh
   

3. Mars

   $ cd Mars
   $ ./compile_mars.sh
   

4. PolyBench

   $ cd PolyBench-ACC
   $ ./compile_cuda.sh
   

5. SHOC

   $ cd shoc
   $ ./configure
   $ ./make
   $ ./make install
   

For other benchmark suites, refer readme.txt

---