libgomp/testsuite/libgomp.oacc-c-c++-common/lib-80.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   int N;
  20   int i;
  21   unsigned long *a, *d_a, dticks;
  22   int nbytes;
  23   float atime, dtime;
  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   r = cuStreamCreate (&stream, CU_STREAM_DEFAULT);
  83   if (r != CUDA_SUCCESS)
  84         {
  85           fprintf (stderr, "cuStreamCreate failed: %d\n", r);
  86           abort ();
  87         }
  88
  89   acc_set_cuda_stream (1, stream);
  90
  91   init_timers (1);
  92
  93   kargs[0] = (void *) &d_a;
  94   kargs[1] = (void *) &dticks;
  95
  96   start_timer (0);
  97
  98   for (i = 0; i < N; i++)
  99     {
 100       r = cuLaunchKernel (delay, 1, 1, 1, 1, 1, 1, 0, stream, kargs, 0);
 101       if (r != CUDA_SUCCESS)
 102         {
 103           fprintf (stderr, "cuLaunchKernel failed: %d\n", r);
 104           abort ();
 105         }
 106     }
 107
 108   fprintf (stderr, "CheCKpOInT\n");
 109   acc_wait_async (1, 1);
 110
 111   acc_wait (1);
 112
 113   atime = stop_timer (0);
 114
 115   if (atime < dtime)
 116     {
 117       fprintf (stderr, "actual time < delay time\n");
 118       abort ();
 119     }
 120
 121   acc_unmap_data (a);
 122
 123   fini_timers ();
 124
 125   free (a);
 126   acc_free (d_a);
 127
 128   acc_shutdown (acc_device_nvidia);
 129
 130   return 0;
 131 }
 132
 133 /* { dg-output "CheCKpOInT(\n|\r\n|\r).*" } */
 134 /* { dg-output "identical parameters" } */
 135 /* { dg-shouldfail "" } */