| 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 8 |
| 9 |
|
| 10 |
void NOINLINE |
| 11 |
emulate_vcvtsh2ss(V512 * dest, V512 op1, V512 op2, |
| 12 |
__mmask8 k, int zero_mask) |
| 13 |
{ |
| 14 |
V512 v1, v2, v3, v4, v5, v6, v7, v8; |
| 15 |
int i; |
| 16 |
|
| 17 |
unpack_ph_2twops(op2, &v3, &v4); |
| 18 |
if ((k&1) || !k) |
| 19 |
v5.f32[0] = v3.f32[0]; |
| 20 |
else if (zero_mask) |
| 21 |
v5.f32[0] = 0; |
| 22 |
else |
| 23 |
v5.f32[0] = dest->f32[0]; |
| 24 |
|
| 25 |
for (i = 1; i < 4; i++) |
| 26 |
v5.f32[i] = op1.f32[i]; |
| 27 |
|
| 28 |
*dest = v5; |
| 29 |
} |
| 30 |
|
| 31 |
void |
| 32 |
test_512 (void) |
| 33 |
{ |
| 34 |
V512 res; |
| 35 |
V512 exp; |
| 36 |
|
| 37 |
init_src(); |
| 38 |
emulate_vcvtsh2ss(&exp, src1, src2, 0x1, 0); |
| 39 |
res.xmm[0] = _mm_cvt_roundsh_ss(src1.xmm[0], src2.xmmh[0], |
| 40 |
_ROUND_NINT); |
| 41 |
check_results(&res, &exp, N_ELEMS, "_mm_cvt_roundsh_ss"); |
| 42 |
|
| 43 |
init_dest(&res, &exp); |
| 44 |
emulate_vcvtsh2ss(&exp, src1, src2, 0x1, 0); |
| 45 |
res.xmm[0] = _mm_mask_cvt_roundsh_ss(res.xmm[0], 0x1, src1.xmm[0], |
| 46 |
src2.xmmh[0], _ROUND_NINT); |
| 47 |
check_results(&res, &exp, N_ELEMS, "mm_mask_cvt_roundsh_ss"); |
| 48 |
|
| 49 |
emulate_vcvtsh2ss(&exp, src1, src2, 0x2, 1); |
| 50 |
res.xmm[0] = _mm_maskz_cvt_roundsh_ss(0x2, src1.xmm[0], |
| 51 |
src2.xmmh[0], _ROUND_NINT); |
| 52 |
check_results(&res, &exp, N_ELEMS, "mm_maskz_cvt_roundsh_ss"); |
| 53 |
|
| 54 |
if (n_errs != 0) { |
| 55 |
abort (); |
| 56 |
} |
| 57 |
} |
| 58 |
|
| 59 |
|