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
132
133
|
/*
* 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)
#include "xmm2altivec.h"
#elif defined(TARGET_LINUX_ARM64)
#include "arm64intrin.h"
#else
#include <immintrin.h>
#endif
#include "rslog_defs.h"
extern "C" __m128 __rvs_log_fma3(__m128);
#ifndef PRECISION
#define PRECISION 0
#endif
__m128 __rvs_log_fma3(__m128 a) {
#if PRECISION == 0
__m128 const A = _mm_set1_ps(A0);
__m128 const B = _mm_set1_ps(B0);
__m128 const C = _mm_set1_ps(C0);
__m128 const D = _mm_set1_ps(D0);
__m128 const E = _mm_set1_ps(E0);
#endif
#if PRECISION == 1
__m128 const A = _mm_set1_ps(A1);
__m128 const B = _mm_set1_ps(B1);
__m128 const C = _mm_set1_ps(C1);
__m128 const D = _mm_set1_ps(D1);
__m128 const E = _mm_set1_ps(E1);
__m128 const F = _mm_set1_ps(F1);
__m128 const G = _mm_set1_ps(G1);
#endif
#if PRECISION == 2
__m128 const A = _mm_set1_ps(A2);
__m128 const B = _mm_set1_ps(B2);
__m128 const C = _mm_set1_ps(C2);
__m128 const D = _mm_set1_ps(D2);
__m128 const E = _mm_set1_ps(E2);
__m128 const F = _mm_set1_ps(F2);
__m128 const G = _mm_set1_ps(G2);
__m128 const H = _mm_set1_ps(H2);
#endif
__m128 const PARTITION_CONST = _mm_set1_ps(PARTITION_CONST_F);
__m128 const TWO_TO_M126 = _mm_set1_ps(TWO_TO_M126_F);
__m128 const LN2 = _mm_set1_ps(LN2_F);
__m128i const N_INF = _mm_set1_epi32(0xff800000);
__m128i const P_INF = _mm_set1_epi32(0x7f800000);
__m128i const NINF2NAN = _mm_set1_epi32(CANONICAL_NAN_I ^ 0xff800000);
__m128i const bit_mask2 = _mm_set1_epi32(0x807fffff);
__m128i const offset = _mm_set1_epi32(0x3f000000);
__m128i const exp_offset = _mm_set1_epi32(126);
__m128 const ZERO = _mm_set1_ps(0.0f);
__m128 const ONE = _mm_set1_ps(1.0f);
__m128 e = (__m128)_mm_srli_epi32((__m128i)a, 23);
e = (__m128)_mm_sub_epi32((__m128i)e, exp_offset);
e = _mm_cvtepi32_ps((__m128i)e);
__m128i im = _mm_and_si128(bit_mask2, (__m128i)a);
__m128 m = (__m128)_mm_add_epi32(im, offset);
__m128 cmp = (__m128)_mm_cmpgt_epi32((__m128i)PARTITION_CONST, (__m128i)m);
__m128 fixe = _mm_and_ps(cmp, LN2);
e = _mm_fmsub_ps(e, LN2, fixe);
__m128i fixm = _mm_and_si128((__m128i)cmp, _mm_set1_epi32(0x00800000));
m = (__m128)_mm_add_epi32((__m128i)m, fixm);
m = _mm_sub_ps(m, ONE);
__m128 t = A;
t = _mm_fmadd_ps(t, m, B);
t = _mm_fmadd_ps(t, m, C);
t = _mm_fmadd_ps(t, m, D);
t = _mm_fmadd_ps(t, m, E);
#if PRECISION >= 1
t = _mm_fmadd_ps(t, m, F);
t = _mm_fmadd_ps(t, m, G);
#endif
#if PRECISION >= 2
t = _mm_fmadd_ps(t, m, H);
#endif
t = _mm_fmadd_ps(t, m, e);
__m128 mask0, mask1;
mask0 = _mm_cmp_ps(a, TWO_TO_M126, _CMP_NGE_UQ);
mask1 = (__m128)_mm_cmpeq_epi32((__m128i)a, P_INF);
#if defined(TARGET_LINUX_POWER)
if (__builtin_expect(_vec_any_nz((__m128i)_mm_or_ps(mask0, mask1)) ,0))
#else
if (__builtin_expect(_mm_movemask_ps(_mm_or_ps(mask0, mask1)) ,0))
#endif
{
// [0.0, FLT_MIN) u nan -> -inf
__m128 spec = _mm_and_ps(mask0, (__m128)N_INF);
// (-oo, 0.0) -> nan
__m128 neg = _mm_cmp_ps(a, ZERO, _CMP_LT_OQ);
neg = _mm_and_ps(neg, (__m128)NINF2NAN);
// nan -> nan, inf -> inf
__m128 unord = _mm_cmp_ps(a, (__m128)P_INF, _CMP_NLT_UQ);
unord = _mm_and_ps(unord, a);
spec = _mm_xor_ps(spec, neg);
spec = _mm_add_ps(spec, unord);
t = _mm_add_ps(t, spec);
}
return t;
}
|
