libgomp/testsuite/libgomp.oacc-c-c++-common/vector-length-128-3.c

   1 /* { dg-do run { target openacc_nvidia_accel_selected } } */
   2 /* { dg-additional-options "-foffload=-fdump-tree-oaccloops" } */
   3 /* We default to warp size 32 for the vector length, so the GOMP_OPENACC_DIM has
   4    no effect.  */
   5 /* { dg-set-target-env-var "GOMP_OPENACC_DIM" "::128" } */
   6 /* { dg-set-target-env-var "GOMP_DEBUG" "1" } */
   7
   8
   9 #include <stdlib.h>
  10
  11 #define N 1024
  12
  13 unsigned int a[N];
  14 unsigned int b[N];
  15 unsigned int c[N];
  16 unsigned int n = N;
  17
  18 int
  19 main (void)
  20 {
  21   for (unsigned int i = 0; i < n; ++i)
  22     {
  23       a[i] = i % 3;
  24       b[i] = i % 5;
  25     }
  26
  27 #pragma acc parallel copyin (a,b) copyout (c)
  28   {
  29 #pragma acc loop vector
  30     for (unsigned int i = 0; i < n; i++)
  31       c[i] = a[i] + b[i];
  32   }
  33
  34   for (unsigned int i = 0; i < n; ++i)
  35     if (c[i] != (i % 3) + (i % 5))
  36       abort ();
  37
  38   return 0;
  39 }
  40
  41 /* { dg-final { scan-offload-tree-dump "__attribute__\\(\\(oacc function \\(1, 1, 32\\)" "oaccloops" } } */
  42 /* { dg-output "nvptx_exec: kernel main\\\$_omp_fn\\\$0: launch gangs=1, workers=1, vectors=32" } */