How to use mpi to compile mpi+openacc C++ programs

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

I compiled an openacc program using the nvc++ compiler, and when I wanted to multimachine it, I compiled it with

mpicc-o test-ac-gpu = cc87-cuda-cudalib test.c,

But I find that this seems to be a c compiler, which can’t implement function overloading and variable references

2

I used the following command for the new version of openmpi

./configure CC=/opt/nvidia/hpc_sdk/Linux_ aarch64/23.1/compilers/bin/pgcc CXX=pgCC FC=/opt/nvidia/hpc_sdk/Linux_aarch64/23.1/compilers/bin/pgf77 F90=/opt/nvidia/hpc_sdk/Linux_ aarch64/23.1/compilers/bin/pgf90 --prefix=/usr/local/openmpi

I think it’s probably not the new version of openmpi. What command should I try to compile the code for mpi+openacc C++？

3

The MPI driver for C++ is “mpicxx” or “mpic++”. “mpicc” is the C driver.

Example:

mpicxx -o test -acc -gpu=cc87 -cuda -cudalib test.cpp

4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.