| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-options "-O2 -mavx512fp16 -mavx512vl -ftree-vectorize -fvect-cost-model=unlimited -mprefer-vector-width=256 -fdump-tree-slp-details" } */ |
| 3 |
/* { dg-require-effective-target avx512fp16 } */ |
| 4 |
/* { dg-final { scan-tree-dump-times {(?n)add new stmt:.*VEC_PERM_EXPR.*\{ 14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1 \}} 2 "slp2" } } */ |
| 5 |
/* { dg-final { scan-tree-dump-times "basic block part vectorized using (?:32|64) byte vectors" 1 "slp2" } } */ |
| 6 |
|
| 7 |
#include <string.h> |
| 8 |
|
| 9 |
static void do_test (void); |
| 10 |
#define DO_TEST do_test |
| 11 |
#define AVX512FP16 |
| 12 |
#include "avx512-check.h" |
| 13 |
|
| 14 |
void |
| 15 |
__attribute__((noipa)) |
| 16 |
foo_ph (_Complex _Float16* a, _Complex _Float16* __restrict b) |
| 17 |
{ |
| 18 |
a[0] = b[15]; |
| 19 |
a[1] = b[14]; |
| 20 |
a[2] = b[13]; |
| 21 |
a[3] = b[12]; |
| 22 |
a[4] = b[11]; |
| 23 |
a[5] = b[10]; |
| 24 |
a[6] = b[9]; |
| 25 |
a[7] = b[8]; |
| 26 |
a[8] = b[7]; |
| 27 |
a[9] = b[6]; |
| 28 |
a[10] = b[5]; |
| 29 |
a[11] = b[4]; |
| 30 |
a[12] = b[3]; |
| 31 |
a[13] = b[2]; |
| 32 |
a[14] = b[1]; |
| 33 |
a[15] = b[0]; |
| 34 |
} |
| 35 |
|
| 36 |
void |
| 37 |
do_test (void) |
| 38 |
{ |
| 39 |
_Complex _Float16* ph_src = (_Complex _Float16*) malloc (64); |
| 40 |
_Complex _Float16* ph_dst = (_Complex _Float16*) malloc (64); |
| 41 |
_Complex _Float16* ph_exp = (_Complex _Float16*) malloc (64); |
| 42 |
char* p = (char* ) malloc (64); |
| 43 |
char* q = (char* ) malloc (64); |
| 44 |
|
| 45 |
__builtin_memset (ph_dst, 0, 64); |
| 46 |
|
| 47 |
for (int i = 0; i != 64; i++) |
| 48 |
p[i] = i; |
| 49 |
|
| 50 |
__builtin_memcpy (ph_src, p, 64); |
| 51 |
|
| 52 |
for (int i = 0; i != 4; i++) |
| 53 |
{ |
| 54 |
q[i] = i + 60; |
| 55 |
q[i + 4] = i + 56; |
| 56 |
q[i + 8] = i + 52; |
| 57 |
q[i + 12] = i + 48; |
| 58 |
q[i + 16] = i + 44; |
| 59 |
q[i + 20] = i + 40; |
| 60 |
q[i + 24] = i + 36; |
| 61 |
q[i + 28] = i + 32; |
| 62 |
q[i + 32] = i + 28; |
| 63 |
q[i + 36] = i + 24; |
| 64 |
q[i + 40] = i + 20; |
| 65 |
q[i + 44] = i + 16; |
| 66 |
q[i + 48] = i + 12; |
| 67 |
q[i + 52] = i + 8; |
| 68 |
q[i + 56] = i + 4; |
| 69 |
q[i + 60] = i; |
| 70 |
} |
| 71 |
|
| 72 |
__builtin_memcpy (ph_exp, q, 64); |
| 73 |
|
| 74 |
foo_ph (ph_dst, ph_src); |
| 75 |
|
| 76 |
if (__builtin_memcmp (ph_dst, ph_exp, 64) != 0) |
| 77 |
__builtin_abort (); |
| 78 |
|
| 79 |
return; |
| 80 |
} |