| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-options "-mavx512f -O2" } */ |
| 3 |
/* { dg-require-effective-target avx512f } */ |
| 4 |
|
| 5 |
#include "avx512f-check.h" |
| 6 |
|
| 7 |
#define SCALE 2 |
| 8 |
|
| 9 |
static void |
| 10 |
compute_gatherdps (float *res, __mmask16 m16, int *idx, |
| 11 |
float *src, int scale, float *r) |
| 12 |
{ |
| 13 |
int i; |
| 14 |
|
| 15 |
for (i = 0; i < 16; i++) |
| 16 |
{ |
| 17 |
if (m16 & (1 << i)) |
| 18 |
r[i] = *(float *) (((unsigned char *) src) + idx[i] * scale); |
| 19 |
else |
| 20 |
r[i] = res[i]; |
| 21 |
} |
| 22 |
} |
| 23 |
|
| 24 |
static void |
| 25 |
avx512f_test (void) |
| 26 |
{ |
| 27 |
int i; |
| 28 |
union512 res; |
| 29 |
union512i_d idx; |
| 30 |
float src[16]; |
| 31 |
float res_ref[16]; |
| 32 |
__mmask16 m16 = 0xBC5D; |
| 33 |
|
| 34 |
res.x = _mm512_setzero_ps(); |
| 35 |
|
| 36 |
for (i = 0; i < 16; i++) |
| 37 |
{ |
| 38 |
src[i] = 2.718281828459045 * (i + 1) * (i + 2); |
| 39 |
|
| 40 |
/* About to gather in reverse order, |
| 41 |
divide by 2 to demonstrate scale */ |
| 42 |
idx.a[i] = (64 - (i + 1) * 4) >> 1; |
| 43 |
} |
| 44 |
|
| 45 |
res.x = _mm512_mask_i32gather_ps (res.x, m16, idx.x, src, SCALE); |
| 46 |
compute_gatherdps (res.a, m16, idx.a, src, SCALE, res_ref); |
| 47 |
|
| 48 |
if (check_union512 (res, res_ref)) |
| 49 |
abort (); |
| 50 |
|
| 51 |
res.x = _mm512_i32gather_ps (idx.x, src, SCALE); |
| 52 |
compute_gatherdps (res.a, 0xFFFF, idx.a, src, SCALE, res_ref); |
| 53 |
|
| 54 |
if (check_union512 (res, res_ref)) |
| 55 |
abort (); |
| 56 |
} |