How to detect GPU support at compile-time

[gzagaris]

Hello,

Is there a good way to detect whether MPICH is compiled with GPU support at compile time?

For example, with OpenMPI, I can do the following:

  #include<mpi-ext.h>
  #if defined(MPIX_CUDA_AWARE_SUPPORT) && MPIX_CUDA_AWARE_SUPPORT 
    // OpenMPI is compiled with cuda-aware support
  #endif 

Is there something equivalent for MPICH?

Is the following sufficient:

#if defined(MPIX_GPU_SUPPORT_CUDA) 
     // MPICH is compiled with cuda-aware support
#endif 

Thank you for all your help.

[hzhou]

No. But we support runtime queries: https://www.mpich.org/static/docs/v4.3.1/www3/MPIX_GPU_query_support.html

You can test MPI support in your configure and define compile-time macros at configure time.

[gzagaris]

Thank you for the quick response & clarification @hzhou.

I was hoping to avoid the runtime check since we may be cross-compiling, but, I can add a small test program that makes the runtime calls and returns the output to CMake.