gcc/testsuite/c-c++-common/goacc/kernels-loop.c

   1 /* { dg-additional-options "-O2" } */
   2 /* { dg-additional-options "-fdump-tree-parloops1-all" } */
   3 /* { dg-additional-options "-fdump-tree-optimized" } */
   4
   5 #include <stdlib.h>
   6
   7 #define N (1024 * 512)
   8 #define COUNTERTYPE unsigned int
   9
  10 int
  11 main (void)
  12 {
  13   unsigned int *__restrict a;
  14   unsigned int *__restrict b;
  15   unsigned int *__restrict c;
  16
  17   a = (unsigned int *)malloc (N * sizeof (unsigned int));
  18   b = (unsigned int *)malloc (N * sizeof (unsigned int));
  19   c = (unsigned int *)malloc (N * sizeof (unsigned int));
  20
  21   for (COUNTERTYPE i = 0; i < N; i++)
  22     a[i] = i * 2;
  23
  24   for (COUNTERTYPE i = 0; i < N; i++)
  25     b[i] = i * 4;
  26
  27 #pragma acc kernels copyin (a[0:N], b[0:N]) copyout (c[0:N])
  28   {
  29 #ifdef ACC_LOOP
  30     #pragma acc loop
  31 #endif
  32     for (COUNTERTYPE ii = 0; ii < N; ii++)
  33       c[ii] = a[ii] + b[ii];
  34   }
  35
  36   for (COUNTERTYPE i = 0; i < N; i++)
  37     if (c[i] != a[i] + b[i])
  38       abort ();
  39
  40   free (a);
  41   free (b);
  42   free (c);
  43
  44   return 0;
  45 }
  46
  47 /* Check that only one loop is analyzed, and that it can be parallelized.  */
  48 /* { dg-final { scan-tree-dump-times "SUCCESS: may be parallelized" 1 "parloops1" } } */
  49 /* { dg-final { scan-tree-dump-times "(?n)__attribute__\\(\\(oacc kernels parallelized, oacc function \\(, , \\), oacc kernels, omp target entrypoint\\)\\)" 1 "parloops1" } } */
  50 /* { dg-final { scan-tree-dump-not "FAILED:" "parloops1" } } */
  51
  52 /* Check that the loop has been split off into a function.  */
  53 /* { dg-final { scan-tree-dump-times "(?n);; Function .*main._omp_fn.0" 1 "optimized" } } */