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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
|
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#ifndef EIGEN_GPU_SPECIALFUNCTIONS_H
#define EIGEN_GPU_SPECIALFUNCTIONS_H
namespace Eigen {
namespace internal {
// Make sure this is only available when targeting a GPU: we don't want to
// introduce conflicts between these packet_traits definitions and the ones
// we'll use on the host side (SSE, AVX, ...)
#if defined(EIGEN_GPUCC) && defined(EIGEN_USE_GPU)
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 plgamma<float4>(const float4& a)
{
return make_float4(lgammaf(a.x), lgammaf(a.y), lgammaf(a.z), lgammaf(a.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 plgamma<double2>(const double2& a)
{
using numext::lgamma;
return make_double2(lgamma(a.x), lgamma(a.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pdigamma<float4>(const float4& a)
{
using numext::digamma;
return make_float4(digamma(a.x), digamma(a.y), digamma(a.z), digamma(a.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pdigamma<double2>(const double2& a)
{
using numext::digamma;
return make_double2(digamma(a.x), digamma(a.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pzeta<float4>(const float4& x, const float4& q)
{
using numext::zeta;
return make_float4(zeta(x.x, q.x), zeta(x.y, q.y), zeta(x.z, q.z), zeta(x.w, q.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pzeta<double2>(const double2& x, const double2& q)
{
using numext::zeta;
return make_double2(zeta(x.x, q.x), zeta(x.y, q.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 ppolygamma<float4>(const float4& n, const float4& x)
{
using numext::polygamma;
return make_float4(polygamma(n.x, x.x), polygamma(n.y, x.y), polygamma(n.z, x.z), polygamma(n.w, x.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 ppolygamma<double2>(const double2& n, const double2& x)
{
using numext::polygamma;
return make_double2(polygamma(n.x, x.x), polygamma(n.y, x.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 perf<float4>(const float4& a)
{
return make_float4(erff(a.x), erff(a.y), erff(a.z), erff(a.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 perf<double2>(const double2& a)
{
using numext::erf;
return make_double2(erf(a.x), erf(a.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 perfc<float4>(const float4& a)
{
using numext::erfc;
return make_float4(erfc(a.x), erfc(a.y), erfc(a.z), erfc(a.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 perfc<double2>(const double2& a)
{
using numext::erfc;
return make_double2(erfc(a.x), erfc(a.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pndtri<float4>(const float4& a)
{
using numext::ndtri;
return make_float4(ndtri(a.x), ndtri(a.y), ndtri(a.z), ndtri(a.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pndtri<double2>(const double2& a)
{
using numext::ndtri;
return make_double2(ndtri(a.x), ndtri(a.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pigamma<float4>(const float4& a, const float4& x)
{
using numext::igamma;
return make_float4(
igamma(a.x, x.x),
igamma(a.y, x.y),
igamma(a.z, x.z),
igamma(a.w, x.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pigamma<double2>(const double2& a, const double2& x)
{
using numext::igamma;
return make_double2(igamma(a.x, x.x), igamma(a.y, x.y));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pigamma_der_a<float4>(
const float4& a, const float4& x) {
using numext::igamma_der_a;
return make_float4(igamma_der_a(a.x, x.x), igamma_der_a(a.y, x.y),
igamma_der_a(a.z, x.z), igamma_der_a(a.w, x.w));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2
pigamma_der_a<double2>(const double2& a, const double2& x) {
using numext::igamma_der_a;
return make_double2(igamma_der_a(a.x, x.x), igamma_der_a(a.y, x.y));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pgamma_sample_der_alpha<float4>(
const float4& alpha, const float4& sample) {
using numext::gamma_sample_der_alpha;
return make_float4(
gamma_sample_der_alpha(alpha.x, sample.x),
gamma_sample_der_alpha(alpha.y, sample.y),
gamma_sample_der_alpha(alpha.z, sample.z),
gamma_sample_der_alpha(alpha.w, sample.w));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2
pgamma_sample_der_alpha<double2>(const double2& alpha, const double2& sample) {
using numext::gamma_sample_der_alpha;
return make_double2(
gamma_sample_der_alpha(alpha.x, sample.x),
gamma_sample_der_alpha(alpha.y, sample.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pigammac<float4>(const float4& a, const float4& x)
{
using numext::igammac;
return make_float4(
igammac(a.x, x.x),
igammac(a.y, x.y),
igammac(a.z, x.z),
igammac(a.w, x.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pigammac<double2>(const double2& a, const double2& x)
{
using numext::igammac;
return make_double2(igammac(a.x, x.x), igammac(a.y, x.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pbetainc<float4>(const float4& a, const float4& b, const float4& x)
{
using numext::betainc;
return make_float4(
betainc(a.x, b.x, x.x),
betainc(a.y, b.y, x.y),
betainc(a.z, b.z, x.z),
betainc(a.w, b.w, x.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pbetainc<double2>(const double2& a, const double2& b, const double2& x)
{
using numext::betainc;
return make_double2(betainc(a.x, b.x, x.x), betainc(a.y, b.y, x.y));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pi0e<float4>(const float4& x) {
using numext::i0e;
return make_float4(i0e(x.x), i0e(x.y), i0e(x.z), i0e(x.w));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2
pi0e<double2>(const double2& x) {
using numext::i0e;
return make_double2(i0e(x.x), i0e(x.y));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pi1e<float4>(const float4& x) {
using numext::i1e;
return make_float4(i1e(x.x), i1e(x.y), i1e(x.z), i1e(x.w));
}
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2
pi1e<double2>(const double2& x) {
using numext::i1e;
return make_double2(i1e(x.x), i1e(x.y));
}
#endif
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_GPU_SPECIALFUNCTIONS_H
|
