libgomp/testsuite/libgomp.oacc-c-c++-common/lib-77.c

   1 /* { dg-do run { target openacc_nvidia_accel_selected } } */
   2 /* { dg-additional-options "-lcuda" } */
   3
   4 #include <stdio.h>
   5 #include <stdlib.h>
   6 #include <unistd.h>
   7 #include <openacc.h>
   8 #include <cuda.h>
   9 #include "timer.h"
  10
  11 int
  12 main (int argc, char **argv)
  13 {
  14   CUdevice dev;
  15   CUfunction delay;
  16   CUmodule module;
  17   CUresult r;
  18   CUstream stream;
  19   unsigned long *a, *d_a, dticks;
  20   int nbytes;
  21   float atime, dtime;
  22   void *kargs[2];
  23   int clkrate;
  24   int devnum, nprocs;
  25
  26   acc_init (acc_device_nvidia);
  27
  28   devnum = acc_get_device_num (acc_device_nvidia);
  29
  30   r = cuDeviceGet (&dev, devnum);
  31   if (r != CUDA_SUCCESS)
  32     {
  33       fprintf (stderr, "cuDeviceGet failed: %d\n", r);
  34       abort ();
  35     }
  36
  37   r =
  38     cuDeviceGetAttribute (&nprocs, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
  39                           dev);
  40   if (r != CUDA_SUCCESS)
  41     {
  42       fprintf (stderr, "cuDeviceGetAttribute failed: %d\n", r);
  43       abort ();
  44     }
  45
  46   r = cuDeviceGetAttribute (&clkrate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, dev);
  47   if (r != CUDA_SUCCESS)
  48     {
  49       fprintf (stderr, "cuDeviceGetAttribute failed: %d\n", r);
  50       abort ();
  51     }
  52
  53   r = cuModuleLoad (&module, "subr.ptx");
  54   if (r != CUDA_SUCCESS)
  55     {
  56       fprintf (stderr, "cuModuleLoad failed: %d\n", r);
  57       abort ();
  58     }
  59
  60   r = cuModuleGetFunction (&delay, module, "delay");
  61   if (r != CUDA_SUCCESS)
  62     {
  63       fprintf (stderr, "cuModuleGetFunction failed: %d\n", r);
  64       abort ();
  65     }
  66
  67   nbytes = nprocs * sizeof (unsigned long);
  68
  69   dtime = 200.0;
  70
  71   dticks = (unsigned long) (dtime * clkrate);
  72
  73   a = (unsigned long *) malloc (nbytes);
  74   d_a = (unsigned long *) acc_malloc (nbytes);
  75
  76   acc_map_data (a, d_a, nbytes);
  77
  78   kargs[0] = (void *) &d_a;
  79   kargs[1] = (void *) &dticks;
  80
  81   r = cuStreamCreate (&stream, CU_STREAM_DEFAULT);
  82   if (r != CUDA_SUCCESS)
  83         {
  84           fprintf (stderr, "cuStreamCreate failed: %d\n", r);
  85           abort ();
  86         }
  87
  88   acc_set_cuda_stream (0, stream);
  89
  90   init_timers (1);
  91
  92   start_timer (0);
  93
  94   r = cuLaunchKernel (delay, 1, 1, 1, 1, 1, 1, 0, stream, kargs, 0);
  95   if (r != CUDA_SUCCESS)
  96     {
  97       fprintf (stderr, "cuLaunchKernel failed: %d\n", r);
  98       abort ();
  99     }
 100
 101   fprintf (stderr, "CheCKpOInT\n");
 102   acc_wait (1);
 103
 104   atime = stop_timer (0);
 105
 106   if (atime < dtime)
 107     {
 108       fprintf (stderr, "actual time < delay time\n");
 109       abort ();
 110     }
 111
 112   start_timer (0);
 113
 114   acc_wait (1);
 115
 116   atime = stop_timer (0);
 117
 118   if (0.010 < atime)
 119     {
 120       fprintf (stderr, "actual time < delay time\n");
 121       abort ();
 122     }
 123
 124   acc_unmap_data (a);
 125
 126   fini_timers ();
 127
 128   free (a);
 129   acc_free (d_a);
 130
 131   acc_shutdown (acc_device_nvidia);
 132
 133   return 0;
 134 }
 135
 136 /* { dg-output "CheCKpOInT(\n|\r\n|\r).*" } */
 137 /* { dg-output "unknown async \[0-9\]+" } */
 138 /* { dg-shouldfail "" } */