| 1 |
/* { dg-do run { target openacc_nvidia_accel_selected } } */ |
| 2 |
/* { dg-additional-options "-lcuda" } */ |
| 3 |
/* { dg-require-effective-target openacc_cuda } */ |
| 4 |
|
| 5 |
#include <pthread.h> |
| 6 |
#include <stdio.h> |
| 7 |
#include <string.h> |
| 8 |
#include <stdlib.h> |
| 9 |
#include <unistd.h> |
| 10 |
#include <errno.h> |
| 11 |
#include <ctype.h> |
| 12 |
#include <openacc.h> |
| 13 |
#include <cuda.h> |
| 14 |
|
| 15 |
unsigned char **x; |
| 16 |
void **d_x; |
| 17 |
const int N = 16; |
| 18 |
const int NTHREADS = 32; |
| 19 |
|
| 20 |
static void * |
| 21 |
test (void *arg) |
| 22 |
{ |
| 23 |
int i; |
| 24 |
int tid; |
| 25 |
unsigned char *p; |
| 26 |
int devnum; |
| 27 |
|
| 28 |
tid = (int) (long) arg; |
| 29 |
|
| 30 |
devnum = acc_get_device_num (acc_device_nvidia); |
| 31 |
acc_set_device_num (devnum, acc_device_nvidia); |
| 32 |
|
| 33 |
if (acc_get_current_cuda_context () == NULL) |
| 34 |
abort (); |
| 35 |
|
| 36 |
p = (unsigned char *) malloc (N); |
| 37 |
|
| 38 |
for (i = 0; i < N; i++) |
| 39 |
{ |
| 40 |
p[i] = tid; |
| 41 |
} |
| 42 |
|
| 43 |
x[tid] = p; |
| 44 |
|
| 45 |
d_x[tid] = acc_copyin (p, N); |
| 46 |
|
| 47 |
acc_wait_all (); |
| 48 |
|
| 49 |
return 0; |
| 50 |
} |
| 51 |
|
| 52 |
int |
| 53 |
main (int argc, char **argv) |
| 54 |
{ |
| 55 |
int i; |
| 56 |
pthread_attr_t attr; |
| 57 |
pthread_t *tid; |
| 58 |
CUresult r; |
| 59 |
CUstream s; |
| 60 |
|
| 61 |
acc_init (acc_device_nvidia); |
| 62 |
|
| 63 |
x = (unsigned char **) malloc (NTHREADS * N); |
| 64 |
d_x = (void **) malloc (NTHREADS * N); |
| 65 |
|
| 66 |
if (pthread_attr_init (&attr) != 0) |
| 67 |
perror ("pthread_attr_init failed"); |
| 68 |
|
| 69 |
tid = (pthread_t *) malloc (NTHREADS * sizeof (pthread_t)); |
| 70 |
|
| 71 |
r = cuStreamCreate (&s, CU_STREAM_DEFAULT); |
| 72 |
if (r != CUDA_SUCCESS) |
| 73 |
{ |
| 74 |
fprintf (stderr, "cuStreamCreate failed: %d\n", r); |
| 75 |
abort (); |
| 76 |
} |
| 77 |
|
| 78 |
if (!acc_set_cuda_stream (0, s)) |
| 79 |
abort (); |
| 80 |
|
| 81 |
for (i = 0; i < NTHREADS; i++) |
| 82 |
{ |
| 83 |
if (pthread_create (&tid[i], &attr, &test, (void *) (unsigned long) (i)) |
| 84 |
!= 0) |
| 85 |
perror ("pthread_create failed"); |
| 86 |
} |
| 87 |
|
| 88 |
if (pthread_attr_destroy (&attr) != 0) |
| 89 |
perror ("pthread_attr_destroy failed"); |
| 90 |
|
| 91 |
for (i = 0; i < NTHREADS; i++) |
| 92 |
{ |
| 93 |
void *res; |
| 94 |
|
| 95 |
if (pthread_join (tid[i], &res) != 0) |
| 96 |
perror ("pthread join failed"); |
| 97 |
} |
| 98 |
|
| 99 |
|
| 100 |
for (i = 0; i < NTHREADS; i++) |
| 101 |
{ |
| 102 |
if (acc_is_present (x[i], N) != 1) |
| 103 |
abort (); |
| 104 |
} |
| 105 |
|
| 106 |
acc_get_cuda_stream (1); |
| 107 |
|
| 108 |
for (i = 0; i < NTHREADS; i++) |
| 109 |
{ |
| 110 |
memset (x[i], 0, N); |
| 111 |
acc_copyout (x[i], N); |
| 112 |
} |
| 113 |
|
| 114 |
acc_wait_all (); |
| 115 |
|
| 116 |
for (i = 0; i < NTHREADS; i++) |
| 117 |
{ |
| 118 |
unsigned char *p; |
| 119 |
int j; |
| 120 |
|
| 121 |
p = x[i]; |
| 122 |
|
| 123 |
for (j = 0; j < N; j++) |
| 124 |
{ |
| 125 |
if (p[j] != i) |
| 126 |
abort (); |
| 127 |
} |
| 128 |
|
| 129 |
if (acc_is_present (x[i], N) != 0) |
| 130 |
abort (); |
| 131 |
} |
| 132 |
|
| 133 |
acc_shutdown (acc_device_nvidia); |
| 134 |
|
| 135 |
return 0; |
| 136 |
} |
| 137 |
|
| 138 |
/* { dg-output "" } */ |