| 1 |
/* { dg-do run { target avx512fp16 } } */ |
| 2 |
/* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */ |
| 3 |
|
| 4 |
#define AVX512FP16 |
| 5 |
#include "avx512fp16-helper.h" |
| 6 |
|
| 7 |
#define N_ELEMS (AVX512F_LEN / 16) |
| 8 |
|
| 9 |
void NOINLINE |
| 10 |
EMULATE(c_fmul_pch) (V512 * dest, V512 op1, V512 op2, |
| 11 |
__mmask16 k, int zero_mask, int c_flag) |
| 12 |
{ |
| 13 |
V512 v1, v2, v3, v4, v5, v6, v7, v8; |
| 14 |
int i; |
| 15 |
int invert = 1; |
| 16 |
if (c_flag == 1) |
| 17 |
invert = -1; |
| 18 |
|
| 19 |
unpack_ph_2twops(op1, &v1, &v2); |
| 20 |
unpack_ph_2twops(op2, &v3, &v4); |
| 21 |
unpack_ph_2twops(*dest, &v7, &v8); |
| 22 |
|
| 23 |
for (i = 0; i < 16; i++) { |
| 24 |
if (((1 << (i / 2)) & k) == 0) { |
| 25 |
if (zero_mask) { |
| 26 |
v5.f32[i] = 0; |
| 27 |
} |
| 28 |
else { |
| 29 |
v5.u32[i] = v7.u32[i]; |
| 30 |
} |
| 31 |
} |
| 32 |
else { |
| 33 |
if ((i % 2) == 0) { |
| 34 |
v5.f32[i] = v1.f32[i] * v3.f32[i] |
| 35 |
- invert * (v1.f32[i+1] * v3.f32[i+1]); |
| 36 |
} |
| 37 |
else { |
| 38 |
v5.f32[i] = v1.f32[i] * v3.f32[i-1] |
| 39 |
+ invert * (v1.f32[i-1] * v3.f32[i]); |
| 40 |
|
| 41 |
} |
| 42 |
} |
| 43 |
if (((1 << (i / 2 + 8)) & k) == 0) { |
| 44 |
if (zero_mask) { |
| 45 |
v6.f32[i] = 0; |
| 46 |
} |
| 47 |
else { |
| 48 |
v6.u32[i] = v8.u32[i]; |
| 49 |
} |
| 50 |
} |
| 51 |
else { |
| 52 |
if ((i % 2) == 0) { |
| 53 |
v6.f32[i] = v2.f32[i] * v4.f32[i] |
| 54 |
- invert * (v2.f32[i+1] * v4.f32[i+1]); |
| 55 |
} |
| 56 |
else { |
| 57 |
v6.f32[i] = v2.f32[i] * v4.f32[i-1] |
| 58 |
+ invert * (v2.f32[i-1] * v4.f32[i]); |
| 59 |
} |
| 60 |
|
| 61 |
} |
| 62 |
} |
| 63 |
|
| 64 |
*dest = pack_twops_2ph(v5, v6); |
| 65 |
} |
| 66 |
|
| 67 |
void |
| 68 |
TEST (void) |
| 69 |
{ |
| 70 |
V512 res; |
| 71 |
V512 exp; |
| 72 |
|
| 73 |
init_src(); |
| 74 |
|
| 75 |
EMULATE(c_fmul_pch)(&exp, src1, src2, NET_CMASK, 0, 1); |
| 76 |
HF(res) = INTRINSIC (_fcmul_pch) (HF(src1), HF(src2)); |
| 77 |
CHECK_RESULT (&res, &exp, N_ELEMS, _fcmul_pch); |
| 78 |
|
| 79 |
init_dest(&res, &exp); |
| 80 |
EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 0, 1); |
| 81 |
HF(res) = INTRINSIC (_mask_fcmul_pch) (HF(res) ,HALF_MASK, HF(src1), |
| 82 |
HF(src2)); |
| 83 |
CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fcmul_pch); |
| 84 |
|
| 85 |
EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 1, 1); |
| 86 |
HF(res) = INTRINSIC (_maskz_fcmul_pch) ( HALF_MASK, HF(src1), |
| 87 |
HF(src2)); |
| 88 |
CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fcmul_pch); |
| 89 |
|
| 90 |
#if AVX512F_LEN == 512 |
| 91 |
EMULATE(c_fmul_pch)(&exp, src1, src2, NET_CMASK, 0, 1); |
| 92 |
HF(res) = INTRINSIC (_fcmul_round_pch) (HF(src1), HF(src2), _ROUND_NINT); |
| 93 |
CHECK_RESULT (&res, &exp, N_ELEMS, _fcmul_round_pch); |
| 94 |
|
| 95 |
init_dest(&res, &exp); |
| 96 |
EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 0, 1); |
| 97 |
HF(res) = INTRINSIC (_mask_fcmul_round_pch) (HF(res) ,HALF_MASK, HF(src1), |
| 98 |
HF(src2), _ROUND_NINT); |
| 99 |
CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fcmul_round_pch); |
| 100 |
|
| 101 |
EMULATE(c_fmul_pch)(&exp, src1, src2, HALF_MASK, 1, 1); |
| 102 |
HF(res) = INTRINSIC (_maskz_fcmul_round_pch) ( HALF_MASK, HF(src1), |
| 103 |
HF(src2), _ROUND_NINT); |
| 104 |
CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fcmul_round_pch); |
| 105 |
#endif |
| 106 |
|
| 107 |
if (n_errs != 0) { |
| 108 |
abort (); |
| 109 |
} |
| 110 |
} |
| 111 |
|