libgomp/testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-3.c

   1 #include <assert.h>
   2
   3 /* Test of reduction on both parallel and loop directives (workers and vectors
   4    together in gang-partitioned mode, float type).  */
   5
   6 int
   7 main (int argc, char *argv[])
   8 {
   9   int i, j;
  10   float arr[32768];
  11   float res = 0, hres = 0;
  12
  13   for (i = 0; i < 32768; i++)
  14     arr[i] = i;
  15
  16   #pragma acc parallel num_gangs(32) num_workers(32) vector_length(32) \
  17     reduction(+:res) copy(res)
  18   {
  19     #pragma acc loop gang
  20     for (j = 0; j < 32; j++)
  21       {
  22         #pragma acc loop worker vector reduction(+:res)
  23         for (i = 0; i < 1024; i++)
  24           res += arr[j * 1024 + i];
  25
  26         #pragma acc loop worker vector reduction(+:res)
  27         for (i = 0; i < 1024; i++)
  28           res += arr[j * 1024 + (1023 - i)];
  29       }
  30   }
  31
  32   for (j = 0; j < 32; j++)
  33     for (i = 0; i < 1024; i++)
  34       {
  35         hres += arr[j * 1024 + i];
  36         hres += arr[j * 1024 + (1023 - i)];
  37       }
  38
  39   assert (res == hres);
  40
  41   return 0;
  42 }