1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
|
/*
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
#if defined(TARGET_LINUX_POWER)
#error "Source cannot be compiled for POWER architectures"
#include "xmm2altivec.h"
#else
#include <immintrin.h>
#include "mth_avx512helper.h"
#endif
#include "rslog_defs.h"
extern "C" __m512 FCN_AVX512(__rvs_log_fma3)(__m512);
#ifndef PRECISION
#define PRECISION 0
#endif
__m512 FCN_AVX512(__rvs_log_fma3)(__m512 a) {
#if PRECISION == 0
__m512 const A = _mm512_set1_ps(A0);
__m512 const B = _mm512_set1_ps(B0);
__m512 const C = _mm512_set1_ps(C0);
__m512 const D = _mm512_set1_ps(D0);
__m512 const E = _mm512_set1_ps(E0);
#endif
#if PRECISION == 1
__m512 const A = _mm512_set1_ps(A1);
__m512 const B = _mm512_set1_ps(B1);
__m512 const C = _mm512_set1_ps(C1);
__m512 const D = _mm512_set1_ps(D1);
__m512 const E = _mm512_set1_ps(E1);
__m512 const F = _mm512_set1_ps(F1);
__m512 const G = _mm512_set1_ps(G1);
#endif
#if PRECISION == 2
__m512 const A = _mm512_set1_ps(A2);
__m512 const B = _mm512_set1_ps(B2);
__m512 const C = _mm512_set1_ps(C2);
__m512 const D = _mm512_set1_ps(D2);
__m512 const E = _mm512_set1_ps(E2);
__m512 const F = _mm512_set1_ps(F2);
__m512 const G = _mm512_set1_ps(G2);
__m512 const H = _mm512_set1_ps(H2);
#endif
__m512 const PARTITION_CONST = _mm512_set1_ps(PARTITION_CONST_F);
__m512 const TWO_TO_M126 = _mm512_set1_ps(TWO_TO_M126_F);
__m512 const LN2 = _mm512_set1_ps(LN2_F);
__m512i const N_INF = _mm512_set1_epi32(0xff800000);
__m512i const P_INF = _mm512_set1_epi32(0x7f800000);
__m512i const NINF2NAN = _mm512_set1_epi32(CANONICAL_NAN_I ^ 0xff800000);
__m512i const bit_mask2 = _mm512_set1_epi32(0x807fffff);
__m512i const offset = _mm512_set1_epi32(0x3f000000);
__m512i const exp_offset = _mm512_set1_epi32(126);
__m512 const ZERO = _mm512_set1_ps(0.0f);
__m512 const ONE = _mm512_set1_ps(1.0f);
__m512 e = (__m512)_mm512_srli_epi32((__m512i)a, 23);
e = (__m512)_mm512_sub_epi32((__m512i)e, exp_offset);
e = _mm512_cvtepi32_ps((__m512i)e);
__m512i im = _mm512_and_si512(bit_mask2, (__m512i)a);
__m512 m = (__m512)_mm512_add_epi32(im, offset);
__m512 cmp = (__m512)_MM512_CMPGT_EPI32((__m512i)PARTITION_CONST, (__m512i)m);
__m512 fixe = _MM512_AND_PS(cmp, LN2);
e = _mm512_fmsub_ps(e, LN2, fixe);
__m512i fixm = _mm512_and_si512((__m512i)cmp, _mm512_set1_epi32(0x00800000));
m = (__m512)_mm512_add_epi32((__m512i)m, fixm);
m = _mm512_sub_ps(m, ONE);
__m512 t = A;
t = _mm512_fmadd_ps(t, m, B);
t = _mm512_fmadd_ps(t, m, C);
t = _mm512_fmadd_ps(t, m, D);
t = _mm512_fmadd_ps(t, m, E);
#if PRECISION >= 1
t = _mm512_fmadd_ps(t, m, F);
t = _mm512_fmadd_ps(t, m, G);
#endif
#if PRECISION >= 2
t = _mm512_fmadd_ps(t, m, H);
#endif
t = _mm512_fmadd_ps(t, m, e);
__m512 mask0, mask1;
mask0 = _MM512_CMP_PS(a, TWO_TO_M126, _CMP_NGE_UQ);
mask1 = (__m512)_MM512_CMPEQ_EPI32((__m512i)a, P_INF);
if (__builtin_expect(_MM512_MOVEMASK_PS(_MM512_OR_PS(mask0, mask1)) ,0))
{
// [0.0, FLT_MIN) u nan -> -inf
__m512 spec = _MM512_AND_PS(mask0, (__m512)N_INF);
// (-oo, 0.0) -> nan
__m512 neg = _MM512_CMP_PS(a, ZERO, _CMP_LT_OQ);
neg = _MM512_AND_PS(neg, (__m512)NINF2NAN);
// nan -> nan, inf -> inf
__m512 unord = _MM512_CMP_PS(a, (__m512)P_INF, _CMP_NLT_UQ);
unord = _MM512_AND_PS(unord, a);
spec = _MM512_XOR_PS(spec, neg);
spec = _mm512_add_ps(spec, unord);
t = _mm512_add_ps(t, spec);
}
return t;
}
|
