| 1 |
/* { dg-do run { target avx512fp16 } } */ |
| 2 |
/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */ |
| 3 |
|
| 4 |
|
| 5 |
#define AVX512FP16 |
| 6 |
#include "avx512fp16-helper.h" |
| 7 |
|
| 8 |
#define N_ELEMS 2 |
| 9 |
|
| 10 |
void NOINLINE |
| 11 |
emulate_cvtph2_d(V512 * dest, V512 op1, |
| 12 |
__mmask32 k, int zero_mask) |
| 13 |
{ |
| 14 |
V512 v1, v2, v3, v4, v5, v6, v7, v8; |
| 15 |
int i; |
| 16 |
__mmask16 m1, m2; |
| 17 |
|
| 18 |
m1 = k & 0xffff; |
| 19 |
|
| 20 |
unpack_ph_2twops(op1, &v1, &v2); |
| 21 |
|
| 22 |
for (i = 0; i < 16; i++) { |
| 23 |
if (((1 << i) & m1) == 0) { |
| 24 |
if (zero_mask) { |
| 25 |
v5.u32[i] = 0; |
| 26 |
} |
| 27 |
else { |
| 28 |
v5.u32[i] = dest->u32[i]; |
| 29 |
} |
| 30 |
} |
| 31 |
else { |
| 32 |
v5.u32[i] = v1.f32[i]; |
| 33 |
|
| 34 |
} |
| 35 |
} |
| 36 |
*dest = v5; |
| 37 |
} |
| 38 |
|
| 39 |
void |
| 40 |
test_512 (void) |
| 41 |
{ |
| 42 |
V512 res; |
| 43 |
V512 exp; |
| 44 |
|
| 45 |
init_src(); |
| 46 |
emulate_cvtph2_d(&exp, src1, NET_MASK, 0); |
| 47 |
res.u32[0] = _mm_cvtt_roundsh_i32(src1.xmmh[0], _ROUND_NINT); |
| 48 |
check_results(&res, &exp, N_ELEMS, "_mm_cvtt_roundsh_u32"); |
| 49 |
|
| 50 |
if (n_errs != 0) { |
| 51 |
abort (); |
| 52 |
} |
| 53 |
} |
| 54 |
|