| 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 (AVX512F_LEN / 16) |
| 9 |
|
| 10 |
void NOINLINE |
| 11 |
EMULATE(fmsubadd_ph) (V512 * dest, V512 op1, V512 op2, |
| 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 |
m2 = (k >> 16) & 0xffff; |
| 20 |
|
| 21 |
unpack_ph_2twops(op1, &v1, &v2); |
| 22 |
unpack_ph_2twops(op2, &v3, &v4); |
| 23 |
unpack_ph_2twops(*dest, &v7, &v8); |
| 24 |
|
| 25 |
for (i = 0; i < 16; i++) { |
| 26 |
if (((1 << i) & m1) == 0) { |
| 27 |
if (zero_mask) { |
| 28 |
v5.f32[i] = 0; |
| 29 |
} |
| 30 |
else { |
| 31 |
v5.u32[i] = v7.u32[i]; |
| 32 |
} |
| 33 |
} |
| 34 |
else { |
| 35 |
if (i % 2 == 1) { |
| 36 |
v5.f32[i] = v1.f32[i] * v3.f32[i] - v7.f32[i]; |
| 37 |
} |
| 38 |
else { |
| 39 |
v5.f32[i] = v1.f32[i] * v3.f32[i] + v7.f32[i]; |
| 40 |
} |
| 41 |
} |
| 42 |
|
| 43 |
if (((1 << i) & m2) == 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 == 1) { |
| 53 |
v6.f32[i] = v2.f32[i] * v4.f32[i] - v8.f32[i]; |
| 54 |
} |
| 55 |
else { |
| 56 |
v6.f32[i] = v2.f32[i] * v4.f32[i] + v8.f32[i]; |
| 57 |
} |
| 58 |
} |
| 59 |
} |
| 60 |
*dest = pack_twops_2ph(v5, v6); |
| 61 |
} |
| 62 |
|
| 63 |
void NOINLINE |
| 64 |
EMULATE(m_fmsubadd_ph) (V512 * dest, V512 op1, V512 op2, |
| 65 |
__mmask32 k, int zero_mask) |
| 66 |
{ |
| 67 |
V512 v1, v2, v3, v4, v5, v6, v7, v8; |
| 68 |
int i; |
| 69 |
__mmask16 m1, m2; |
| 70 |
|
| 71 |
m1 = k & 0xffff; |
| 72 |
m2 = (k >> 16) & 0xffff; |
| 73 |
|
| 74 |
unpack_ph_2twops(op1, &v1, &v2); |
| 75 |
unpack_ph_2twops(op2, &v3, &v4); |
| 76 |
unpack_ph_2twops(*dest, &v7, &v8); |
| 77 |
|
| 78 |
for (i = 0; i < 16; i++) { |
| 79 |
if (((1 << i) & m1) == 0) { |
| 80 |
if (zero_mask) { |
| 81 |
v5.f32[i] = 0; |
| 82 |
} |
| 83 |
else { |
| 84 |
v5.u32[i] = v7.u32[i]; |
| 85 |
} |
| 86 |
} |
| 87 |
else { |
| 88 |
if (i % 2 == 1) { |
| 89 |
v5.f32[i] = v1.f32[i] * v7.f32[i] - v3.f32[i]; |
| 90 |
} |
| 91 |
else { |
| 92 |
v5.f32[i] = v1.f32[i] * v7.f32[i] + v3.f32[i]; |
| 93 |
} |
| 94 |
} |
| 95 |
|
| 96 |
if (((1 << i) & m2) == 0) { |
| 97 |
if (zero_mask) { |
| 98 |
v6.f32[i] = 0; |
| 99 |
} |
| 100 |
else { |
| 101 |
v6.u32[i] = v8.u32[i]; |
| 102 |
} |
| 103 |
} |
| 104 |
else { |
| 105 |
if (i % 2 == 1) { |
| 106 |
v6.f32[i] = v2.f32[i] * v8.f32[i] - v4.f32[i]; |
| 107 |
} |
| 108 |
else { |
| 109 |
v6.f32[i] = v2.f32[i] * v8.f32[i] + v4.f32[i]; |
| 110 |
} |
| 111 |
} |
| 112 |
} |
| 113 |
*dest = pack_twops_2ph(v5, v6); |
| 114 |
} |
| 115 |
|
| 116 |
void |
| 117 |
TEST (void) |
| 118 |
{ |
| 119 |
V512 res; |
| 120 |
V512 exp; |
| 121 |
|
| 122 |
init_src(); |
| 123 |
|
| 124 |
init_dest(&res, &exp); |
| 125 |
EMULATE(fmsubadd_ph)(&exp, src1, src2, NET_MASK, 0); |
| 126 |
HF(res) = INTRINSIC (_fmsubadd_ph) (HF(src1), HF(src2), HF(res)); |
| 127 |
CHECK_RESULT (&res, &exp, N_ELEMS, _fmsubadd_ph); |
| 128 |
init_dest(&res, &exp); |
| 129 |
EMULATE(fmsubadd_ph)(&exp, src1, src2, MASK_VALUE, 0); |
| 130 |
HF(res) = INTRINSIC (_mask3_fmsubadd_ph) (HF(src1), HF(src2), |
| 131 |
HF(res), MASK_VALUE); |
| 132 |
CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fmsubadd_ph); |
| 133 |
init_dest(&res, &exp); |
| 134 |
EMULATE(m_fmsubadd_ph)(&exp, src1, src2, MASK_VALUE, 0); |
| 135 |
HF(res) = INTRINSIC (_mask_fmsubadd_ph) (HF(res), MASK_VALUE, |
| 136 |
HF(src1), HF(src2)); |
| 137 |
CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmsubadd_ph); |
| 138 |
init_dest(&res, &exp); |
| 139 |
EMULATE(fmsubadd_ph)(&exp, src1, src2, ZMASK_VALUE, 1); |
| 140 |
HF(res) = INTRINSIC (_maskz_fmsubadd_ph) (ZMASK_VALUE, HF(src1), |
| 141 |
HF(src2), HF(res)); |
| 142 |
CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmsubadd_ph); |
| 143 |
|
| 144 |
init_dest(&res, &exp); |
| 145 |
#if AVX512F_LEN == 512 |
| 146 |
EMULATE(fmsubadd_ph)(&exp, src1, src2, NET_MASK, 0); |
| 147 |
HF(res) = INTRINSIC (_fmsubadd_round_ph) (HF(src1), HF(src2), |
| 148 |
HF(res), _ROUND_NINT); |
| 149 |
CHECK_RESULT (&res, &exp, N_ELEMS, _fmsubadd_ph); |
| 150 |
init_dest(&res, &exp); |
| 151 |
EMULATE(fmsubadd_ph)(&exp, src1, src2, MASK_VALUE, 0); |
| 152 |
HF(res) = INTRINSIC (_mask3_fmsubadd_round_ph) (HF(src1), HF(src2), |
| 153 |
HF(res), MASK_VALUE, |
| 154 |
_ROUND_NINT); |
| 155 |
CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fmsubadd_ph); |
| 156 |
init_dest(&res, &exp); |
| 157 |
EMULATE(m_fmsubadd_ph)(&exp, src1, src2, MASK_VALUE, 0); |
| 158 |
HF(res) = INTRINSIC (_mask_fmsubadd_round_ph) (HF(res), MASK_VALUE, |
| 159 |
HF(src1), HF(src2), |
| 160 |
_ROUND_NINT); |
| 161 |
CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmsubadd_ph); |
| 162 |
init_dest(&res, &exp); |
| 163 |
EMULATE(fmsubadd_ph)(&exp, src1, src2, ZMASK_VALUE, 1); |
| 164 |
HF(res) = INTRINSIC (_maskz_fmsubadd_round_ph) (ZMASK_VALUE, HF(src1), |
| 165 |
HF(src2), HF(res), |
| 166 |
_ROUND_NINT); |
| 167 |
CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmsubadd_ph); |
| 168 |
#endif |
| 169 |
|
| 170 |
if (n_errs != 0) { |
| 171 |
abort (); |
| 172 |
} |
| 173 |
} |
| 174 |
|
| 175 |
|