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
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
|
/***************************************************************************
* Copyright (c) Wolf Vollprecht, Johan Mabille and Sylvain Corlay *
* Copyright (c) QuantStack *
* *
* Distributed under the terms of the BSD 3-Clause License. *
* *
* The full license is in the file LICENSE, distributed with this software. *
****************************************************************************/
#ifndef XBLAS_HPP
#define XBLAS_HPP
#include <algorithm>
#include "xtensor/containers/xarray.hpp"
#include "xtensor/containers/xtensor.hpp"
#include "xtensor/io/xio.hpp"
#include "xtensor/misc/xcomplex.hpp"
#include "xtensor/utils/xutils.hpp"
#include "xflens/cxxblas/cxxblas.cxx"
#include "xtensor-blas/xblas_config.hpp"
#include "xtensor-blas/xblas_utils.hpp"
namespace xt
{
namespace blas
{
/**
* Calculate the 1-norm of a vector
*
* @param a vector of n elements
* @returns scalar result
*/
template <class E, class R>
void asum(const xexpression<E>& a, R& result)
{
auto&& ad = view_eval<E::static_layout>(a.derived_cast());
XTENSOR_ASSERT(ad.dimension() == 1);
cxxblas::asum<blas_index_t>(
static_cast<blas_index_t>(ad.shape()[0]),
ad.data() + ad.data_offset(),
stride_front(ad),
result
);
}
/**
* Calculate the 2-norm of a vector
*
* @param a vector of n elements
* @returns scalar result
*/
template <class E, class R>
void nrm2(const xexpression<E>& a, R& result)
{
auto&& ad = view_eval<E::static_layout>(a.derived_cast());
XTENSOR_ASSERT(ad.dimension() == 1);
cxxblas::nrm2<blas_index_t>(
static_cast<blas_index_t>(ad.shape()[0]),
ad.data() + ad.data_offset(),
stride_front(ad),
result
);
}
/**
* Calculate the dot product between two vectors, conjugating
* the first argument \em a in the case of complex vectors.
*
* @param a vector of n elements
* @param b vector of n elements
* @returns scalar result
*/
template <class E1, class E2, class R>
void dot(const xexpression<E1>& a, const xexpression<E2>& b, R& result)
{
auto&& ad = view_eval<E1::static_layout>(a.derived_cast());
auto&& bd = view_eval<E2::static_layout>(b.derived_cast());
XTENSOR_ASSERT(ad.dimension() == 1);
blas_index_t stride_a = stride_front(ad);
blas_index_t stride_b = stride_front(bd);
auto* adt = ad.data() + ad.data_offset();
auto* bdt = bd.data() + bd.data_offset();
// we need to have a pointer that points to the "real" start of the memory
// not to the first element (BLAS is doing that transformation itself)
if (stride_a < 0)
{
adt += (static_cast<blas_index_t>(ad.shape()[0]) - 1) * stride_a; // go back to the start
}
if (stride_b < 0)
{
bdt += (static_cast<blas_index_t>(ad.shape()[0]) - 1) * stride_b; // go back to the start
}
cxxblas::dot<blas_index_t>(static_cast<blas_index_t>(ad.shape()[0]), adt, stride_a, bdt, stride_b, result);
}
/**
* Calculate the dot product between two complex vectors, not conjugating the
* first argument \em a.
*
* @param a vector of n elements
* @param b vector of n elements
* @returns scalar result
*/
template <class E1, class E2, class R>
void dotu(const xexpression<E1>& a, const xexpression<E2>& b, R& result)
{
auto&& ad = view_eval<E1::static_layout>(a.derived_cast());
auto&& bd = view_eval<E2::static_layout>(b.derived_cast());
XTENSOR_ASSERT(ad.dimension() == 1);
blas_index_t stride_a = stride_front(ad);
blas_index_t stride_b = stride_front(bd);
auto* adt = ad.data() + ad.data_offset();
auto* bdt = bd.data() + bd.data_offset();
// we need to have a pointer that points to the "real" start of the memory
// not to the first element (BLAS is doing that transformation itself)
if (stride_a < 0)
{
adt += (static_cast<blas_index_t>(ad.shape()[0]) - 1) * stride_a; // go back to the start
}
if (stride_b < 0)
{
bdt += (static_cast<blas_index_t>(ad.shape()[0]) - 1) * stride_b; // go back to the start
}
cxxblas::dotu<blas_index_t>(static_cast<blas_index_t>(ad.shape()[0]), adt, stride_a, bdt, stride_b, result);
}
/**
* Calculate the general matrix times vector product according to
* ``y := alpha * A * x + beta * y``.
*
* @param A matrix of n x m elements
* @param x vector of n elements
* @param transpose select if A should be transposed
* @param alpha scalar scale factor
* @returns the resulting vector
*/
template <class E1, class E2, class R, class value_type = typename E1::value_type>
void gemv(
const xexpression<E1>& A,
const xexpression<E2>& x,
R& result,
bool transpose_A = false,
const value_type& alpha = value_type(1.0),
const value_type& beta = value_type(0.0)
)
{
auto&& dA = view_eval<E1::static_layout>(A.derived_cast());
auto&& dx = view_eval<E2::static_layout>(x.derived_cast());
cxxblas::gemv<blas_index_t>(
get_blas_storage_order(result),
transpose_A ? cxxblas::Transpose::Trans : cxxblas::Transpose::NoTrans,
static_cast<blas_index_t>(dA.shape()[0]),
static_cast<blas_index_t>(dA.shape()[1]),
alpha,
dA.data() + dA.data_offset(),
get_leading_stride(dA),
dx.data() + dx.data_offset(),
get_leading_stride(dx),
beta,
result.data() + result.data_offset(),
get_leading_stride(result)
);
}
/**
* Calculate the matrix-matrix product of matrix @A and matrix @B
*
* C := alpha * A * B + beta * C
*
* @param A matrix of m-by-n elements
* @param B matrix of n-by-k elements
* @param transpose_A transpose A on the fly
* @param transpose_B transpose B on the fly
* @param alpha scale factor for A * B (defaults to 1)
* @param beta scale factor for C (defaults to 0)
*/
template <class E, class F, class R, class value_type = typename E::value_type>
void gemm(
const xexpression<E>& A,
const xexpression<F>& B,
R& result,
char transpose_A = false,
char transpose_B = false,
const value_type& alpha = value_type(1.0),
const value_type& beta = value_type(0.0)
)
{
static_assert(R::static_layout != layout_type::dynamic, "GEMM result layout cannot be dynamic.");
auto&& dA = view_eval<R::static_layout>(A.derived_cast());
auto&& dB = view_eval<R::static_layout>(B.derived_cast());
XTENSOR_ASSERT(dA.layout() == dB.layout());
XTENSOR_ASSERT(result.layout() == dA.layout());
XTENSOR_ASSERT(dA.dimension() == 2);
XTENSOR_ASSERT(dB.dimension() == 2);
cxxblas::gemm<blas_index_t>(
get_blas_storage_order(result),
transpose_A ? cxxblas::Transpose::Trans : cxxblas::Transpose::NoTrans,
transpose_B ? cxxblas::Transpose::Trans : cxxblas::Transpose::NoTrans,
static_cast<blas_index_t>(transpose_A ? dA.shape()[1] : dA.shape()[0]),
static_cast<blas_index_t>(transpose_B ? dB.shape()[0] : dB.shape()[1]),
static_cast<blas_index_t>(transpose_B ? dB.shape()[1] : dB.shape()[0]),
alpha,
dA.data() + dA.data_offset(),
get_leading_stride(dA),
dB.data() + dB.data_offset(),
get_leading_stride(dB),
beta,
result.data() + result.data_offset(),
get_leading_stride(result)
);
}
/**
* Calculate the outer product of vector x and y.
* According to A:= alpha * x * y' + A
*
* @param x vector of n elements
* @param y vector of m elements
* @param alpha scalar scale factor
* @returns matrix of n-by-m elements
*/
template <class E1, class E2, class R, class value_type = typename E1::value_type>
void
ger(const xexpression<E1>& x,
const xexpression<E2>& y,
R& result,
const value_type& alpha = value_type(1.0))
{
auto&& dx = view_eval(x.derived_cast());
auto&& dy = view_eval(y.derived_cast());
XTENSOR_ASSERT(dx.dimension() == 1);
XTENSOR_ASSERT(dy.dimension() == 1);
cxxblas::ger<blas_index_t>(
get_blas_storage_order(result),
static_cast<blas_index_t>(dx.shape()[0]),
static_cast<blas_index_t>(dy.shape()[0]),
alpha,
dx.data() + dx.data_offset(),
get_leading_stride(dx),
dy.data() + dy.data_offset(),
get_leading_stride(dy),
result.data() + result.data_offset(),
get_leading_stride(result)
);
}
} // namespace blas
} // namespace xt
#endif
|
