/ecc_on_gpu

Primary LanguageCuda

ECC CUDA Project

This project implements an error correction (ECC) algorithm combined with a Rodinia benchamrks on a GPU using CUDA.

Prerequisites

  • CUDA Toolkit (with nvcc)
  • NVIDIA Nsight Systems (nsys)
  • Rodinia Benchmark Suite (specifically the BFS and Gaussian Elimination dataset)
  • An NVIDIA GPU supporting architecture sm_75 or higher

Setting up NVIDIA MPS

Enable NVIDIA Multi-Process Service (MPS). MPS allows multiple CUDA applications to run concurrently on the same GPU.

  1. Start the MPS control daemon:

    nvidia-cuda-mps-control -d

  2. Set the pipe and log directories to temporary locations:

    export CUDA_MPS_PIPE_DIRECTORY=/tmp/nvidia-mps
export CUDA_MPS_LOG_DIRECTORY=/tmp/nvidia-log

  3. Restrict your application to a specific GPU (e.g., GPU 0):

    export CUDA_VISIBLE_DEVICES=0

After completing these steps, MPS will be enabled and ready for use with your CUDA app

Compilation

To compile the CUDA files for this project, use the following command.

nvcc -o ecc_bfs ecc.cu ecc_main.cu bfs_new.cu kernel.cu kernel2.cu -arch=sm_75

Profiling

nsys profile -o report ./ecc_bfs ../rodinia_3.1/data/bfs/graph1MW_6.txt &
./bfs ../rodinia_3.1/data/bfs/graph1MW_6.txt &

Utilize the same flow for Gaussian Elimination