| 1 |
/* { dg-do run { target aarch64_sve256_hw } } */ |
| 2 |
/* { dg-options "-msve-vector-bits=256 -fopenmp -O2" } */ |
| 3 |
|
| 4 |
#pragma GCC target "+sve" |
| 5 |
|
| 6 |
#include <arm_sve.h> |
| 7 |
#include <stdint.h> |
| 8 |
|
| 9 |
#define N 256 |
| 10 |
|
| 11 |
int a[N] __attribute__ ((aligned (64))); |
| 12 |
int b[N] __attribute__ ((aligned (64))); |
| 13 |
|
| 14 |
void __attribute__ ((noipa)) |
| 15 |
foo (int *p, int *q, svint32_t *onesp) |
| 16 |
{ |
| 17 |
svint32_t va, vc; |
| 18 |
int i; |
| 19 |
uint64_t sz = svcntw (); |
| 20 |
|
| 21 |
#pragma omp simd aligned(p, q : 64) aligned (onesp : 128) \ |
| 22 |
private (va, vc) nontemporal (va, vc) |
| 23 |
for (i = 0; i < N; i++) |
| 24 |
{ |
| 25 |
if (i % sz == 0) |
| 26 |
{ |
| 27 |
va = svld1_s32 (svptrue_b32 (), p); |
| 28 |
vc = svadd_s32_z (svptrue_b32 (), va, *onesp); |
| 29 |
svst1_s32 (svptrue_b32 (), q, vc); |
| 30 |
q += sz; |
| 31 |
} |
| 32 |
} |
| 33 |
} |
| 34 |
|
| 35 |
int |
| 36 |
main () |
| 37 |
{ |
| 38 |
svint32_t ones __attribute__ ((aligned(128))) = svindex_s32 (1, 0); |
| 39 |
|
| 40 |
for (int i = 0; i < N; i++) |
| 41 |
{ |
| 42 |
a[i] = 1; |
| 43 |
b[i] = 0; |
| 44 |
} |
| 45 |
|
| 46 |
foo (a, b, &ones); |
| 47 |
|
| 48 |
for (int i = 0; i < N; i++) |
| 49 |
if (b[i] != 2) |
| 50 |
__builtin_abort (); |
| 51 |
} |