| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-options "-mavxneconvert -O2" } */ |
| 3 |
/* { dg-require-effective-target avxneconvert } */ |
| 4 |
#define AVXNECONVERT |
| 5 |
#include <stdint.h> |
| 6 |
|
| 7 |
#ifndef CHECK |
| 8 |
#define CHECK "avx-check.h" |
| 9 |
#endif |
| 10 |
|
| 11 |
#ifndef TEST |
| 12 |
#define TEST avx_test |
| 13 |
#endif |
| 14 |
|
| 15 |
#include CHECK |
| 16 |
|
| 17 |
typedef union |
| 18 |
{ |
| 19 |
uint32_t int32; |
| 20 |
float flt; |
| 21 |
} float_int_t; |
| 22 |
|
| 23 |
typedef union |
| 24 |
{ |
| 25 |
__m128bh x; |
| 26 |
unsigned short a[8]; |
| 27 |
} union128bf16; |
| 28 |
|
| 29 |
void TEST (void) |
| 30 |
{ |
| 31 |
union128 src_128; |
| 32 |
union256 src_256; |
| 33 |
union128bf16 dst_128, dst_256; |
| 34 |
uint16_t res_ref_128[8] = {0}, res_ref_256[8]; |
| 35 |
float_int_t fp32; |
| 36 |
for (int i = 0; i < 4; i++) |
| 37 |
{ |
| 38 |
fp32.flt = (float) 2 * i + 7 + i * 0.25; |
| 39 |
src_128.a[i] = fp32.flt; |
| 40 |
res_ref_128[i] = fp32.int32 >> 16; |
| 41 |
dst_128.a[i] = 117; |
| 42 |
} |
| 43 |
|
| 44 |
for (int i = 0; i < 8; i++) |
| 45 |
{ |
| 46 |
fp32.flt = (float) 2 * i + 7 + i * 0.25; |
| 47 |
src_256.a[i] = fp32.flt; |
| 48 |
res_ref_256[i] = fp32.int32 >> 16; |
| 49 |
dst_256.a[i] = 117; |
| 50 |
} |
| 51 |
dst_128.x = _mm_cvtneps_avx_pbh (src_128.x); |
| 52 |
dst_256.x = _mm256_cvtneps_avx_pbh (src_256.x); |
| 53 |
|
| 54 |
if (checkVus (dst_128.a, res_ref_128, 8)) |
| 55 |
abort(); |
| 56 |
if (checkVus (dst_128.a, res_ref_128, 8)) |
| 57 |
abort(); |
| 58 |
} |