offload/test/offloading/ompx_bare_ballot_sync.c

   1 // RUN: %libomptarget-compilexx-run-and-check-generic
   2 //
   3 // REQUIRES: gpu
   4
   5 #include <assert.h>
   6 #include <ompx.h>
   7 #include <stdint.h>
   8 #include <stdio.h>
   9 #include <stdlib.h>
  10
  11 int main(int argc, char *argv[]) {
  12   const int num_blocks = 1;
  13   const int block_size = 256;
  14   const int N = num_blocks * block_size;
  15   int *res = (int *)malloc(N * sizeof(int));
  16
  17 #pragma omp target teams ompx_bare num_teams(num_blocks) thread_limit(block_size) \
  18         map(from: res[0:N])
  19   {
  20     int tid = ompx_thread_id_x();
  21     uint64_t mask = ompx_ballot_sync(~0LU, tid & 0x1);
  22 #if defined __AMDGCN_WAVEFRONT_SIZE && __AMDGCN_WAVEFRONT_SIZE == 64
  23     res[tid] = mask == 0xaaaaaaaaaaaaaaaa;
  24 #else
  25     res[tid] = mask == 0xaaaaaaaa;
  26 #endif
  27   }
  28
  29   for (int i = 0; i < N; ++i)
  30     assert(res[i]);
  31
  32   // CHECK: PASS
  33   printf("PASS\n");
  34
  35   free(res);
  36
  37   return 0;
  38 }