libgomp/testsuite/libgomp.oacc-c-c++-common/data-3.c

   1 /* { dg-do run } */
   2
   3 #include <stdlib.h>
   4
   5 int
   6 main (int argc, char **argv)
   7 {
   8   int N = 128; //1024 * 1024;
   9   float *a, *b, *c, *d, *e;
  10   int i;
  11   int nbytes;
  12
  13   nbytes = N * sizeof (float);
  14
  15   a = (float *) malloc (nbytes);
  16   b = (float *) malloc (nbytes);
  17   c = (float *) malloc (nbytes);
  18   d = (float *) malloc (nbytes);
  19   e = (float *) malloc (nbytes);
  20
  21   for (i = 0; i < N; i++)
  22     {
  23       a[i] = 3.0;
  24       b[i] = 0.0;
  25     }
  26
  27 #pragma acc enter data copyin (a[0:N]) copyin (b[0:N]) copyin (N) async
  28 #pragma acc parallel async wait
  29 #pragma acc loop
  30   for (i = 0; i < N; i++)
  31     b[i] = a[i];
  32
  33 #pragma acc update host (a[0:N], b[0:N]) async wait
  34 #pragma acc wait
  35
  36   for (i = 0; i < N; i++)
  37     {
  38       if (a[i] != 3.0)
  39         abort ();
  40
  41       if (b[i] != 3.0)
  42         abort ();
  43     }
  44
  45   for (i = 0; i < N; i++)
  46     {
  47       a[i] = 2.0;
  48       b[i] = 0.0;
  49     }
  50
  51 #pragma acc update device (a[0:N], b[0:N]) async (1)
  52 #pragma acc parallel async (1)
  53 #pragma acc loop
  54   for (i = 0; i < N; i++)
  55     b[i] = a[i];
  56
  57 #pragma acc update host (a[0:N], b[0:N]) async (1) wait (1)
  58 #pragma acc wait (1)
  59
  60   for (i = 0; i < N; i++)
  61     {
  62       if (a[i] != 2.0)
  63         abort ();
  64
  65       if (b[i] != 2.0)
  66         abort ();
  67     }
  68
  69   for (i = 0; i < N; i++)
  70     {
  71       a[i] = 3.0;
  72       b[i] = 0.0;
  73       c[i] = 0.0;
  74       d[i] = 0.0;
  75     }
  76
  77 #pragma acc update device (a[0:N]) async (1)
  78 #pragma acc update device (b[0:N]) async (2)
  79 #pragma acc enter data copyin (c[0:N], d[0:N]) async (3)
  80
  81 #pragma acc parallel async (1) wait (1,2)
  82 #pragma acc loop
  83   for (i = 0; i < N; i++)
  84     b[i] = (a[i] * a[i] * a[i]) / a[i];
  85
  86 #pragma acc parallel async (2) wait (1,3)
  87 #pragma acc loop
  88   for (i = 0; i < N; i++)
  89     c[i] = (a[i] + a[i] + a[i] + a[i]) / a[i];
  90
  91 #pragma acc parallel async (3) wait (1,3)
  92 #pragma acc loop
  93   for (i = 0; i < N; i++)
  94     d[i] = ((a[i] * a[i] + a[i]) / a[i]) - a[i];
  95
  96 #pragma acc update host (a[0:N], b[0:N], c[0:N], d[0:N]) async (1) wait (1,2,3)
  97 #pragma acc wait (1)
  98
  99   for (i = 0; i < N; i++)
 100     {
 101       if (a[i] != 3.0)
 102         abort ();
 103
 104       if (b[i] != 9.0)
 105         abort ();
 106
 107       if (c[i] != 4.0)
 108         abort ();
 109
 110       if (d[i] != 1.0)
 111         abort ();
 112     }
 113
 114   for (i = 0; i < N; i++)
 115     {
 116       a[i] = 2.0;
 117       b[i] = 0.0;
 118       c[i] = 0.0;
 119       d[i] = 0.0;
 120       e[i] = 0.0;
 121     }
 122
 123 #pragma acc update device (a[0:N], b[0:N], c[0:N], d[0:N]) async (1)
 124 #pragma acc enter data copyin (e[0:N]) async (5)
 125
 126 #pragma acc parallel async (1) wait (1)
 127   for (int ii = 0; ii < N; ii++)
 128     b[ii] = (a[ii] * a[ii] * a[ii]) / a[ii];
 129
 130 #pragma acc parallel async (2) wait (1)
 131   for (int ii = 0; ii < N; ii++)
 132     c[ii] = (a[ii] + a[ii] + a[ii] + a[ii]) / a[ii];
 133
 134 #pragma acc parallel async (3) wait (1)
 135   for (int ii = 0; ii < N; ii++)
 136     d[ii] = ((a[ii] * a[ii] + a[ii]) / a[ii]) - a[ii];
 137
 138 #pragma acc parallel wait (1,5) async (4)
 139   for (int ii = 0; ii < N; ii++)
 140     e[ii] = a[ii] + b[ii] + c[ii] + d[ii];
 141
 142 #pragma acc exit data copyout (a[0:N]) copyout (b[0:N]) copyout (c[0:N]) copyout (d[0:N]) copyout (e[0:N]) wait (1, 2, 3, 4) async (1)
 143 #pragma acc exit data delete (N)
 144 #pragma acc wait (1)
 145
 146
 147   for (i = 0; i < N; i++)
 148     {
 149       if (a[i] != 2.0)
 150         abort ();
 151
 152       if (b[i] != 4.0)
 153         abort ();
 154
 155       if (c[i] != 4.0)
 156         abort ();
 157
 158       if (d[i] != 1.0)
 159         abort ();
 160
 161       if (e[i] != 11.0)
 162         abort ();
 163     }
 164
 165   return 0;
 166 }