libgomp/testsuite/libgomp.oacc-c-c++-common/lib-76.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   int N;
  19   int i;
  20   CUstream *streams;
  21   unsigned long *a, *d_a, dticks;
  22   int nbytes;
  23   float atime, dtime, hitime, lotime;
  24   void *kargs[2];
  25   int clkrate;
  26   int devnum, nprocs;
  27
  28   acc_init (acc_device_nvidia);
  29
  30   devnum = acc_get_device_num (acc_device_nvidia);
  31
  32   r = cuDeviceGet (&dev, devnum);
  33   if (r != CUDA_SUCCESS)
  34     {
  35       fprintf (stderr, "cuDeviceGet failed: %d\n", r);
  36       abort ();
  37     }
  38
  39   r =
  40     cuDeviceGetAttribute (&nprocs, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
  41                           dev);
  42   if (r != CUDA_SUCCESS)
  43     {
  44       fprintf (stderr, "cuDeviceGetAttribute failed: %d\n", r);
  45       abort ();
  46     }
  47
  48   r = cuDeviceGetAttribute (&clkrate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, dev);
  49   if (r != CUDA_SUCCESS)
  50     {
  51       fprintf (stderr, "cuDeviceGetAttribute failed: %d\n", r);
  52       abort ();
  53     }
  54
  55   r = cuModuleLoad (&module, "subr.ptx");
  56   if (r != CUDA_SUCCESS)
  57     {
  58       fprintf (stderr, "cuModuleLoad failed: %d\n", r);
  59       abort ();
  60     }
  61
  62   r = cuModuleGetFunction (&delay, module, "delay");
  63   if (r != CUDA_SUCCESS)
  64     {
  65       fprintf (stderr, "cuModuleGetFunction failed: %d\n", r);
  66       abort ();
  67     }
  68
  69   nbytes = nprocs * sizeof (unsigned long);
  70
  71   dtime = 200.0;
  72
  73   dticks = (unsigned long) (dtime * clkrate);
  74
  75   N = nprocs;
  76
  77   a = (unsigned long *) malloc (nbytes);
  78   d_a = (unsigned long *) acc_malloc (nbytes);
  79
  80   acc_map_data (a, d_a, nbytes);
  81
  82   streams = (CUstream *) malloc (N * sizeof (void *));
  83
  84   for (i = 0; i < N; i++)
  85     {
  86       streams[i] = (CUstream) acc_get_cuda_stream (i);
  87       if (streams[i] != NULL)
  88         abort ();
  89
  90       r = cuStreamCreate (&streams[i], CU_STREAM_DEFAULT);
  91       if (r != CUDA_SUCCESS)
  92         {
  93           fprintf (stderr, "cuStreamCreate failed: %d\n", r);
  94           abort ();
  95         }
  96
  97         if (!acc_set_cuda_stream (i, streams[i]))
  98           abort ();
  99     }
 100
 101   init_timers (1);
 102
 103   kargs[0] = (void *) &d_a;
 104   kargs[1] = (void *) &dticks;
 105
 106   start_timer (0);
 107
 108   for (i = 0; i < N; i++)
 109     {
 110       r = cuLaunchKernel (delay, 1, 1, 1, 1, 1, 1, 0, streams[i], kargs, 0);
 111       if (r != CUDA_SUCCESS)
 112         {
 113           fprintf (stderr, "cuLaunchKernel failed: %d\n", r);
 114           abort ();
 115         }
 116
 117       acc_wait (i);
 118     }
 119
 120   atime = stop_timer (0);
 121
 122   hitime = dtime * N;
 123   hitime += hitime * 0.02;
 124
 125   lotime = dtime * N;
 126   lotime -= lotime * 0.02;
 127
 128   if (atime > hitime || atime < lotime)
 129     {
 130       fprintf (stderr, "actual time < delay time\n");
 131       abort ();
 132     }
 133
 134   acc_unmap_data (a);
 135
 136   fini_timers ();
 137
 138   free (streams);
 139   free (a);
 140   acc_free (d_a);
 141
 142   acc_shutdown (acc_device_nvidia);
 143
 144   exit (0);
 145 }
 146
 147 /* { dg-output "" } */