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
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#include "../imf.h"
#pragma OPENCL FP_CONTRACT OFF
#pragma float_control(precise,on)
static __constant int_double __ssincos_la_dp_reduction_tbl[] = {
{0x3fe45f306e000000UL}, {0xbdfb1bbead603d8bUL},
{0x3fe45f306dc9c883UL}, {0xbc86b01ec5417056UL},
{0x3f87cc1b727220a9UL}, {0x3c23f84eafa3ea6aUL},
{0x3ef836e4e44152a0UL}, {0xbb8ec54170565912UL},
{0x3e8b727220a94fe1UL}, {0x3b1d5f47d4d37703UL},
{0x3e027220a94fe13bUL}, {0xbaa05c1596447e49UL},
{0x3d8220a94fe13abfUL}, {0xba2c1596447e493bUL},
{0x3cb529fc2757d1f5UL}, {0x394a6ee06db14acdUL},
{0x3c729fc2757d1f53UL}, {0x391377036d8a5665UL},
{0x3bffc2757d1f534eUL}, {0xb881f924eb53361eUL},
{0x3b43abe8fa9a6ee0UL}, {0x37eb6c52b3278872UL},
{0x3b0abe8fa9a6ee07UL}, {0xb79275a99b0ef1bfUL},
{0x3a8e8fa9a6ee06dbUL}, {0x3704acc9e21c8210UL},
{0x39ff534ddc0db629UL}, {0x369664f10e41080eUL},
{0x39734ddc0db6295aUL}, {0xb61b0ef1bef7fac6UL},
{0x38ebb81b6c52b328UL}, {0xb58de37defff0eceUL},
{0x387c0db6295993c4UL}, {0x350c820ff0d954bbUL},
};
static __constant int_double __ssincos_la_dp_invpi_h = { 0x3fe45f306e000000UL };
static __constant int_double __ssincos_la_dp_invpi_l = { 0xbdfb1bbead603d8bUL };
static __constant unsigned int __ssincos_la_dp_AbsMask = 0x7fffffffu;
static __constant int_double __ssincos_la_dp_Shifter = { 0x4338000000000000UL };
static __constant int_double __ssincos_la_dp_s_coeff[] = {
{0x3fe243f6a5f0c8e5UL}, {0xbfe4abbb99e5a6dbUL},
{0x3fb465ec65afe9deUL}, {0xbf72d9bb7b168b2bUL},
};
static __constant int_double __ssincos_la_dp_c_coeff[] = {
{0x3feFFFFFFFF97C47UL}, {0xbff3BD3CC7323529UL},
{0x3fd03C1DC1BAFA81UL}, {0xbf955C57B06D183CUL},
{0x3f4D9C364E89E2CCUL}
};
static __constant int_float __ssincos_la_dp_invpi_s = { 0x3f22F983u };
static __constant int_float __ssincos_la_dp_two19 = { 0x49000000u };
static __constant int_float __ssincos_la_dp_fShifter = { 0x4b400000u };
static __constant int_float __ssincos_la_dp_Pi2_h = { 0xbfc90fdbu };
static __constant int_float __ssincos_la_dp_Pi2_m = { 0x333bbd2eu };
static __constant int_float __ssincos_la_dp_Pi2_l = { 0x26f72cedu };
static __constant int_float __ssincos_la_dp_fc3 = { 0xbab212bfu };
static __constant int_float __ssincos_la_dp_fc2 = { 0x3d2a9e76u };
static __constant int_float __ssincos_la_dp_fc1 = { 0xbeffffd0u };
static __constant int_float __ssincos_la_dp_fc0 = { 0x3f800000u };
static __constant int_float __ssincos_la_dp_fs3 = { 0xb94c83b6u };
static __constant int_float __ssincos_la_dp_fs2 = { 0x3c088318u };
static __constant int_float __ssincos_la_dp_fs1 = { 0xbe2aaaa0u };
static __constant int_float __ssincos_la_dp_fs0 = { 0x3f800000u };
__attribute__((always_inline))
inline int __internal_ssincos_cout_dp (float *a, float *psin, float *pcos)
{
int nRet = 0;
float xin = *a;
unsigned int sgn_x, expon;
int k, index, c_sgn;
double xd, dN, dR, dRh, dRl, dR2;
unsigned int R_sgn;
int_double S, s_poly, c_poly;
int_float fres0, fres1, sin_res, cos_res, x0, x, fS, fN;
float fR, fR2, fspoly, fcpoly;
x0.f = xin;
x.w = x0.w & __ssincos_la_dp_AbsMask;
sgn_x = x0.w ^ x.w;
if (x.f < __ssincos_la_dp_two19.f)
{
fS.f = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (x.f, __ssincos_la_dp_invpi_s.f, __ssincos_la_dp_fShifter.f);
fN.f = fS.f - __ssincos_la_dp_fShifter.f;
fR = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fN.f, __ssincos_la_dp_Pi2_h.f, x.f);
fR = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fN.f, __ssincos_la_dp_Pi2_m.f, fR);
fR = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fN.f, __ssincos_la_dp_Pi2_l.f, fR);
R_sgn = (fS.w >> 1) << 31;
index = fS.w << 31;
fR2 = fR * fR;
fspoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fR2, __ssincos_la_dp_fs3.f, __ssincos_la_dp_fs2.f);
fcpoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fR2, __ssincos_la_dp_fc3.f, __ssincos_la_dp_fc2.f);
fspoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fspoly, fR2, __ssincos_la_dp_fs1.f);
fcpoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fcpoly, fR2, __ssincos_la_dp_fc1.f);
fspoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fspoly, fR2, 0.0f);
fcpoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fcpoly, fR2, __ssincos_la_dp_fc0.f);
fspoly = SPIRV_OCL_BUILTIN(fma, _f32_f32_f32, ) (fspoly, fR, fR);
sgn_x ^= R_sgn;
c_sgn = index ^ R_sgn;
sin_res.f = (index == 0) ? fspoly : fcpoly;
cos_res.f = (index == 0) ? fcpoly : fspoly;
sin_res.w ^= sgn_x;
cos_res.w ^= c_sgn;
*pcos = cos_res.f;
*psin = sin_res.f;
}
else
{
if (x.w == 0x7f800000)
{
sin_res.w = 0xffc00000;
nRet = 1;
*psin = *pcos = sin_res.f;
return nRet;
}
xd = (double) x.f;
expon = x.w >> 23;
k = (expon - (23 - 8 + 0x7f)) >> 3;
k += k;
dRh = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (xd, (__ssincos_la_dp_reduction_tbl[k]).f, 0.0);
dRl = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (xd, (__ssincos_la_dp_reduction_tbl[k]).f, -dRh);
S.f = (__ssincos_la_dp_Shifter).f + dRh;
dN = S.f - (__ssincos_la_dp_Shifter).f;
dR = dRh - dN;
dR = dR + dRl;
dR = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (xd, (__ssincos_la_dp_reduction_tbl[k + 1]).f, dR);
R_sgn = (S.w32[0] >> 1);
index = S.w32[0];
dR2 = dR * dR;
R_sgn <<= 31;
index <<= 31;
c_sgn = index;
sgn_x ^= R_sgn;
c_sgn ^= R_sgn;
c_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, (__ssincos_la_dp_c_coeff[4]).f, (__ssincos_la_dp_c_coeff[3]).f);
s_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, (__ssincos_la_dp_s_coeff[3]).f, (__ssincos_la_dp_s_coeff[2]).f);
c_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, c_poly.f, (__ssincos_la_dp_c_coeff[2]).f);
s_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, s_poly.f, (__ssincos_la_dp_s_coeff[1]).f);
c_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, c_poly.f, (__ssincos_la_dp_c_coeff[1]).f);
s_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, s_poly.f, (__ssincos_la_dp_s_coeff[0]).f);
c_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR2, c_poly.f, (__ssincos_la_dp_c_coeff[0]).f);
s_poly.f = SPIRV_OCL_BUILTIN(fma, _f64_f64_f64, ) (dR, s_poly.f, dR);
fres0.f = (float) s_poly.f;
fres1.f = (float) c_poly.f;
sin_res.w = (index == 0) ? fres0.w : fres1.w;
cos_res.w = (index == 0) ? fres1.w : fres0.w;
sin_res.w ^= sgn_x;
cos_res.w ^= c_sgn;
*pcos = cos_res.f;
*psin = sin_res.f;
}
return nRet;
}
void __ocl_svml_sincosf_dp (float a, __private float *b, __private float *c)
{
float va1;
float vr1;
float vr2;
va1 = a;
__internal_ssincos_cout_dp (&va1, &vr1, &vr2);
((float *) b)[0] = vr1;
((float *) c)[0] = vr2;
return;
}
|
