amamory-ampere/openmp-offloading

repository to test the OpenMP offloading capabilities to an NVIDIA GPU

Introduction

This is a repo used to save tests to check the use of OpenMP with GPU offloading. Since I had a new computing environment, 1st i checked it running CUDA examples, then I tested OpenMP examples for the host, and finally I tested OpenMP offloading (target omp keyword).

GPU monitors

Use them to 'see' the examples running.

• nsight-sys: for GPU
• stacer; for CPU

Used compilers

• cuda examples: NVCC v10.2
• openmp_host and openmp_gpu: Using clang 10, available in Ubuntu 18.04;

More information about the Clang compiler setup for GPU offloading is present in the dir openmp_gpu.

Dir contents

• cuda examples: example used to test the cuda environment install. Gencode adapted to sm_30 (Quadro K4000). Check the makefiles for 'SMS' to replace Gencode for other GPUs.
• openmp_host: examples used to test the openmp environment install. Using multithread capability;
• openmp_gpu: examples used to test OpenMP with GPUs (target keyword).

References to build these tests

• CUDA examples: some examples distributed with CUDA toolkit;
• OpenMP;
• OpenMP GPU (target):
  • OpenMP Application Programming Interface Examples - Version 5.0.0
  • http://cacs.usc.edu/education/cs596/OMPtarget.pdf
  • http://cacs.usc.edu/education/cs653/OpenMP4.5_3-20-19.pdf
  • https://on-demand.gputechconf.com/gtc/2016/presentation/s6510-jeff-larkin-targeting-gpus-openmp.pdf