GPU Benchmark Suite

Based on Rodinia 3.1

Contributors:

• Alankrith Krishnan

• Alisha Sawant

Benchmarks Used

• backprop
• cfd
• gaussian
• heartwall
• lavaMD
• nn
• particlefilter
• streamcluster

Steps to run

• Note that these scripts automatically set the GPU device to 1, if you want to use a different GPU, please change export CUDA_VISIBLE_DEVICES=1 to the corresponding GPU number in the cluster.

• Run setup_benchmark.sh to pull CUDA samples and install them, and compile all the benchmark files

chmod +x setup_benchmark.sh
./setup_benchmark.sh

• Then run run_benchmark.sh with any number of parameters of programs (folder names in cuda or cuda_optimized) or alternatively, use all to run all of them:

chmod +x run_benchmark.sh
./run_benchmark.sh all

• Alternatively, you can run make in the outer directory or within any of the specific benchmarks to compile the programs. The command for running code is in a file called run, in each directory.