| 1 |
/* { dg-do run { target openacc_nvidia_accel_selected } } */ |
| 2 |
/* { dg-additional-options "-DVECTOR_LENGTH=vector_length(128)" } */ |
| 3 |
|
| 4 |
/* { dg-additional-options "-foffload=-fdump-tree-oaccloops" } */ |
| 5 |
/* { dg-set-target-env-var "GOMP_DEBUG" "1" } */ |
| 6 |
|
| 7 |
/* Minimized from ref-1.C. */ |
| 8 |
|
| 9 |
#include <stdio.h> |
| 10 |
|
| 11 |
#pragma acc routine vector |
| 12 |
void __attribute__((noinline, noclone)) |
| 13 |
Vector (int *ptr, int n, const int inc) |
| 14 |
{ |
| 15 |
#pragma acc loop vector |
| 16 |
for (unsigned ix = 0; ix < n; ix++) |
| 17 |
ptr[ix] += inc; |
| 18 |
} |
| 19 |
|
| 20 |
int |
| 21 |
main (void) |
| 22 |
{ |
| 23 |
const int n = 32, m=32; |
| 24 |
|
| 25 |
int ary[m][n]; |
| 26 |
unsigned ix, iy; |
| 27 |
|
| 28 |
for (ix = m; ix--;) |
| 29 |
for (iy = n; iy--;) |
| 30 |
ary[ix][iy] = (1 << 16) + (ix << 8) + iy; |
| 31 |
|
| 32 |
int err = 0; |
| 33 |
|
| 34 |
#pragma acc parallel copy (ary) VECTOR_LENGTH /* { dg-warning "using .vector_length \\(32\\). due to call to vector-partitionable routine, ignoring 128" } */ |
| 35 |
{ |
| 36 |
Vector (&ary[0][0], m * n, (1 << 24) - (1 << 16)); |
| 37 |
} |
| 38 |
|
| 39 |
for (ix = m; ix--;) |
| 40 |
for (iy = n; iy--;) |
| 41 |
if (ary[ix][iy] != ((1 << 24) + (ix << 8) + iy)) |
| 42 |
{ |
| 43 |
printf ("ary[%u][%u] = %x expected %x\n", |
| 44 |
ix, iy, ary[ix][iy], ((1 << 24) + (ix << 8) + iy)); |
| 45 |
err++; |
| 46 |
} |
| 47 |
|
| 48 |
if (err) |
| 49 |
{ |
| 50 |
printf ("%d failed\n", err); |
| 51 |
return 1; |
| 52 |
} |
| 53 |
|
| 54 |
return 0; |
| 55 |
} |
| 56 |
|
| 57 |
/* { dg-final { scan-offload-tree-dump "__attribute__\\(\\(oacc function \\(1, 1, 32\\)" "oaccloops" } } */ |
| 58 |
/* { dg-output "nvptx_exec: kernel main\\\$_omp_fn\\\$0: launch gangs=1, workers=1, vectors=32" } */ |