libgomp/testsuite/libgomp.oacc-c-c++-common/lib-72.c

   1 /* { dg-do run { target openacc_nvidia_accel_selected } } */
   2 /* { dg-additional-options "-lcuda" } */
   3
   4 #include <stdio.h>
   5 #include <unistd.h>
   6 #include <stdlib.h>
   7 #include <openacc.h>
   8 #include <cuda.h>
   9
  10 int
  11 main (int argc, char **argv)
  12 {
  13   CUdevice dev;
  14   CUfunction delay;
  15   CUmodule module;
  16   CUresult r;
  17   CUstream stream;
  18   unsigned long *a, *d_a, dticks;
  19   int nbytes;
  20   float dtime;
  21   void *kargs[2];
  22   int clkrate;
  23   int devnum, nprocs;
  24
  25   acc_init (acc_device_nvidia);
  26
  27   devnum = acc_get_device_num (acc_device_nvidia);
  28
  29   r = cuDeviceGet (&dev, devnum);
  30   if (r != CUDA_SUCCESS)
  31     {
  32       fprintf (stderr, "cuDeviceGet failed: %d\n", r);
  33       abort ();
  34     }
  35
  36   r =
  37     cuDeviceGetAttribute (&nprocs, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
  38                           dev);
  39   if (r != CUDA_SUCCESS)
  40     {
  41       fprintf (stderr, "cuDeviceGetAttribute failed: %d\n", r);
  42       abort ();
  43     }
  44
  45   r = cuDeviceGetAttribute (&clkrate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, dev);
  46   if (r != CUDA_SUCCESS)
  47     {
  48       fprintf (stderr, "cuDeviceGetAttribute failed: %d\n", r);
  49       abort ();
  50     }
  51
  52   r = cuModuleLoad (&module, "subr.ptx");
  53   if (r != CUDA_SUCCESS)
  54     {
  55       fprintf (stderr, "cuModuleLoad failed: %d\n", r);
  56       abort ();
  57     }
  58
  59   r = cuModuleGetFunction (&delay, module, "delay");
  60   if (r != CUDA_SUCCESS)
  61     {
  62       fprintf (stderr, "cuModuleGetFunction failed: %d\n", r);
  63       abort ();
  64     }
  65
  66   nbytes = nprocs * sizeof (unsigned long);
  67
  68   dtime = 200.0;
  69
  70   dticks = (unsigned long) (dtime * clkrate);
  71
  72   a = (unsigned long *) malloc (nbytes);
  73   d_a = (unsigned long *) acc_malloc (nbytes);
  74
  75   acc_map_data (a, d_a, nbytes);
  76
  77   kargs[0] = (void *) &d_a;
  78   kargs[1] = (void *) &dticks;
  79
  80   r = cuStreamCreate (&stream, CU_STREAM_DEFAULT);
  81   if (r != CUDA_SUCCESS)
  82     {
  83       fprintf (stderr, "cuStreamCreate failed: %d\n", r);
  84       abort ();
  85     }
  86
  87   if (!acc_set_cuda_stream (0, stream))
  88     abort ();
  89
  90   r = cuLaunchKernel (delay, 1, 1, 1, 1, 1, 1, 0, stream, kargs, 0);
  91   if (r != CUDA_SUCCESS)
  92     {
  93       fprintf (stderr, "cuLaunchKernel failed: %d\n", r);
  94       abort ();
  95     }
  96
  97   if (acc_async_test_all () != 0)
  98     {
  99       fprintf (stderr, "asynchronous operation not running\n");
 100       abort ();
 101     }
 102
 103   sleep ((int) (dtime / 1000.f) + 1);
 104
 105   if (acc_async_test_all () != 1)
 106     {
 107       fprintf (stderr, "found asynchronous operation still running\n");
 108       abort ();
 109     }
 110
 111   acc_unmap_data (a);
 112
 113   free (a);
 114   acc_free (d_a);
 115
 116   acc_shutdown (acc_device_nvidia);
 117
 118   exit (0);
 119 }
 120
 121 /* { dg-output "" } */