Usecase that makes use of OpenMP+GPU offload

openmp_offload.c (text/x-csrc), 881 bytes, created by Adrien Dessemond on 2023-04-11 12:42:04 UTC

#include <stdio.h>
#include <omp.h>
int main(int argc, char const * argv[]) {
    int ndev;
    int nteams;
    int thread_lmt;
    int on_host;
 
    ndev = omp_get_num_devices();
    printf("number of devices: %d\n", ndev);
    for (int i = 0; i < omp_get_num_devices(); ++i) {
        #pragma omp target teams \
                           device(i) \
                           map(from:on_host) map(from:nteams) \
                           map(from:thread_lmt)
        #pragma omp parallel
        #pragma omp master
        if (0 == omp_get_team_num()) {
            on_host = omp_is_initial_device();
            nteams = omp_get_num_teams();
            thread_lmt = omp_get_thread_limit();
        }
        printf("ran on GPU %d: %s, %d teams, "
               "limit of %d threads\n",
               i, on_host ? "no" : "yes", nteams, thread_lmt);
    }
    return 0;
}