build Clang/LLVM with OpenMP+ GPU support

[chunhualiao]

part of the process to revise the steps, ask the official steps.

https://hpc-wiki.info/hpc/Building_LLVM/Clang_with_OpenMP_Offloading_to_NVIDIA_GPUs

Later deployed in AWS gpu instance

[chunhualiao]

//https://lc.llnl.gov/confluence/display/LC/Clang+OpenMP+4.5+with+GPU+support#space-menu-link-content
#include <stdio.h>
#include <omp.h>

int main()
{
  int runningOnGPU = 0;
  /* Test if GPU is available using OpenMP4.5 */
#pragma omp target map(from:runningOnGPU)
  {
    if (omp_is_initial_device() == 0)
      runningOnGPU = 1;
  }
  /* If still running on CPU, GPU must not be available */
  if (runningOnGPU)
    printf("### Able to use the GPU! ### \n");
  else
    printf("### Unable to use the GPU, using CPU! ###\n");

  return 0;
}

results

liao6@ip-128-115-246-7:~$ clang -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda hasGPU.c
clang-10: warning: Unknown CUDA version 10.2. Assuming the latest supported version 10.1 [-Wunknown-cuda-version]
clang-10: warning: No library 'libomptarget-nvptx-sm_70.bc' found in the default clang lib directory or in LIBRARY_PATH. Expect degraded performance due to no inlining of runtime functions on target devices. [-Wopenmp-target]
liao6@ip-128-115-246-7:~$ ./a.out 
./a.out: error while loading shared libraries: libomp.so: cannot open shared object file: No such file or directory
liao6@ip-128-115-246-7:~$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/liao6/opt/llvm-10.0.0/lib
liao6@ip-128-115-246-7:~$ ./a.out 
### Unable to use the GPU, using CPU! ###

[chunhualiao]

More instructions:

• https://github.com/devitocodes/devito/wiki/OPENMP-5-GPU-offloading,-docker-image-%5BMathias%5D
• https://github.com/devitocodes/devito/wiki/OpenMP-5-GPU-offloading-instructions--%5BFabio%5D

[chunhualiao]

No more warning about libomptarget lib. But the code still does not behave as expected.

//https://lc.llnl.gov/confluence/display/LC/Clang+OpenMP+4.5+with+GPU+support#space-menu-link-content
// Revised a bit
#include <stdio.h>
#include <omp.h>

int main()
{
  int runningOnGPU = 0;

  printf ("The number of target devices =%d\n", omp_get_num_devices());
  /* Test if GPU is available using OpenMP4.5 */
#pragma omp target map(from:runningOnGPU)
  {
    // This function returns true if currently running on the host device, false otherwise.
    if (!omp_is_initial_device())
      runningOnGPU = 1;
  }
  /* If still running on CPU, GPU must not be available */
  if (runningOnGPU == 1)
    printf("### Able to use the GPU! ### \n");
  else
    printf("### Unable to use the GPU, using CPU! ###\n");

  return 0;
}

liao6@ip-128-115-246-7:~$ clang -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda hasGPU.c
clang-10: warning: Unknown CUDA version 10.2. Assuming the latest supported version 10.1 [-Wunknown-cuda-version]

liao6@ip-128-115-246-7:~$ ./a.out 
The number of target devices =0
### Unable to use the GPU, using CPU! ###

[chunhualiao]

@ouankou Can you test the instructions on http://freecompilercamp.org/llvm-openmp-build/ using the GPU machine? I updated them a few weeks back and I need someone to review/test them.