Multi GPU Programming Models

This project implements the well known multi GPU Jacobi solver with different multi GPU Programming Models:

• single_threaded_copy           Single Threaded using cudaMemcpy for inter GPU communication

• multi_threaded_copy            Multi Threaded with OpenMP using cudaMemcpy for inter GPU communication

• multi_threaded_copy_overlapp   Multi Threaded with OpenMP using cudaMemcpy for itner GPU communication with overlapping communication

• multi_threaded_p2p             Multi Threaded with OpenMP using GPUDirect P2P mappings for inter GPU communication

• multi_threaded_p2p_opt         Multi Threaded with OpenMP using GPUDirect P2P mappings for inter GPU communication with delayed norm execution

• multi_threaded_um              Multi Threaded with OpenMP relying on transparent peer mappings with Unified Memory for inter GPU communication

• mpi                            Multi Process with MPI using CUDA-aware MPI for inter GPU communication

• mpi_overlapp                   Multi Process with MPI using CUDA-aware MPI for inter GPU communication with overlapping communication

• nvshmem                        Multi Process with MPI and NVSHMEM using NVSHMEM for inter GPU communication. Other approach, nvshmem_opt, might be better for portable performance.

• nvshmem_opt                    Multi Process with MPI and NVSHMEM using NVSHMEM for inter GPU communication with NVSHMEM extension API

Each variant is a stand alone Makefile project and all variants have been described in the GTC EU 2018 Talk Multi GPU Programming Models

Requirements

• CUDA: verison 9.2 or later is required by all variants.

• OpenMP capable compiler: Required by the Multi Threaded variants. The examples have been developed and tested with gcc.

• CUDA-aware MPI: Required by the MPI and NVSHMEM variants. The examples have been developed and tested with OpenMPI.

• CUB: Optional for optimized residual reductions. Set CUB_HOME to your cub installation directory. The examples have been developed and tested with cub 1.8.0.

• NVSHMEM: Required by the NVSHMEM variant. Please reach out to nvshmem@nvidia.com for an early access to NVSHMEM.

Building

Each variant come with a Makefile and can be build by simply issuing make, e.g.

multi-gpu-programming-models$ cd multi_threaded_copy
multi_threaded_copy$ make CUB_HOME=../cub
nvcc -DHAVE_CUB -I../cub -Xcompiler -fopenmp -lineinfo -DUSE_NVTX -lnvToolsExt -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_70,code=compute_70  -std=c++11 jacobi.cu -o jacobi
multi_threaded_copy$ ls jacobi
jacobi

Run instructions

All variant have the following command line options

• -niter: How many iterations to carry out (default 1000)

• -nccheck: How often to check for convergence (default 1)

• -nx: Size of the domain in x direction (default 7168)

• -ny: Size of the domain in y direction (default 7168)

• -csv: Print performance results as -csv

The provided script bench.sh contains some examples executing all the benchmarks presented in the GTC EU 2018 Talk Multi GPU Programming Models.

Developers guide

The code applies the style guide implemented in .clang-format file. clang-format version 7 or later should be used to format the code prior to submitting it. E.g. with

multi-gpu-programming-models$ cd multi_threaded_copy
multi_threaded_copy$ clang-format -style=file -i jacobi.cu