2
Copyright (c) 2011, Intel Corporation. All rights reserved.
4
Redistribution and use in source and binary forms, with or without modification,
5
are permitted provided that the following conditions are met:
7
* Redistributions of source code must retain the above copyright notice, this
8
list of conditions and the following disclaimer.
9
* Redistributions in binary form must reproduce the above copyright notice,
10
this list of conditions and the following disclaimer in the documentation
11
and/or other materials provided with the distribution.
12
* Neither the name of Intel Corporation nor the names of its contributors may
13
be used to endorse or promote products derived from this software without
14
specific prior written permission.
16
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
17
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
20
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
22
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
23
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27
********************************************************************************
28
* Content : Eigen bindings to Intel(R) MKL
29
* Triangular matrix * matrix product functionality based on ?TRMM.
30
********************************************************************************
33
#ifndef EIGEN_TRIANGULAR_MATRIX_MATRIX_MKL_H
34
#define EIGEN_TRIANGULAR_MATRIX_MATRIX_MKL_H
41
template <typename Scalar, typename Index,
42
int Mode, bool LhsIsTriangular,
43
int LhsStorageOrder, bool ConjugateLhs,
44
int RhsStorageOrder, bool ConjugateRhs,
46
struct product_triangular_matrix_matrix_trmm :
47
product_triangular_matrix_matrix<Scalar,Index,Mode,
48
LhsIsTriangular,LhsStorageOrder,ConjugateLhs,
49
RhsStorageOrder, ConjugateRhs, ResStorageOrder, BuiltIn> {};
52
// try to go to BLAS specialization
53
#define EIGEN_MKL_TRMM_SPECIALIZE(Scalar, LhsIsTriangular) \
54
template <typename Index, int Mode, \
55
int LhsStorageOrder, bool ConjugateLhs, \
56
int RhsStorageOrder, bool ConjugateRhs> \
57
struct product_triangular_matrix_matrix<Scalar,Index, Mode, LhsIsTriangular, \
58
LhsStorageOrder,ConjugateLhs, RhsStorageOrder,ConjugateRhs,ColMajor,Specialized> { \
59
static inline void run(Index _rows, Index _cols, Index _depth, const Scalar* _lhs, Index lhsStride,\
60
const Scalar* _rhs, Index rhsStride, Scalar* res, Index resStride, Scalar alpha) { \
61
product_triangular_matrix_matrix_trmm<Scalar,Index,Mode, \
62
LhsIsTriangular,LhsStorageOrder,ConjugateLhs, \
63
RhsStorageOrder, ConjugateRhs, ColMajor>::run( \
64
_rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha); \
68
EIGEN_MKL_TRMM_SPECIALIZE(double, true)
69
EIGEN_MKL_TRMM_SPECIALIZE(double, false)
70
EIGEN_MKL_TRMM_SPECIALIZE(dcomplex, true)
71
EIGEN_MKL_TRMM_SPECIALIZE(dcomplex, false)
72
EIGEN_MKL_TRMM_SPECIALIZE(float, true)
73
EIGEN_MKL_TRMM_SPECIALIZE(float, false)
74
EIGEN_MKL_TRMM_SPECIALIZE(scomplex, true)
75
EIGEN_MKL_TRMM_SPECIALIZE(scomplex, false)
77
// implements col-major += alpha * op(triangular) * op(general)
78
#define EIGEN_MKL_TRMM_L(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
79
template <typename Index, int Mode, \
80
int LhsStorageOrder, bool ConjugateLhs, \
81
int RhsStorageOrder, bool ConjugateRhs> \
82
struct product_triangular_matrix_matrix_trmm<EIGTYPE,Index,Mode,true, \
83
LhsStorageOrder,ConjugateLhs,RhsStorageOrder,ConjugateRhs,ColMajor> \
86
IsLower = (Mode&Lower) == Lower, \
87
SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \
88
IsUnitDiag = (Mode&UnitDiag) ? 1 : 0, \
89
IsZeroDiag = (Mode&ZeroDiag) ? 1 : 0, \
90
LowUp = IsLower ? Lower : Upper, \
91
conjA = ((LhsStorageOrder==ColMajor) && ConjugateLhs) ? 1 : 0 \
94
static EIGEN_DONT_INLINE void run( \
95
Index _rows, Index _cols, Index _depth, \
96
const EIGTYPE* _lhs, Index lhsStride, \
97
const EIGTYPE* _rhs, Index rhsStride, \
98
EIGTYPE* res, Index resStride, \
101
Index diagSize = (std::min)(_rows,_depth); \
102
Index rows = IsLower ? _rows : diagSize; \
103
Index depth = IsLower ? diagSize : _depth; \
104
Index cols = _cols; \
106
typedef Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> MatrixLhs; \
107
typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs; \
109
/* Non-square case - doesn't fit to MKL ?TRMM. Fall to default triangular product or call MKL ?GEMM*/ \
110
if (rows != depth) { \
112
int nthr = mkl_domain_get_max_threads(MKL_BLAS); \
114
if (((nthr==1) && (((std::max)(rows,depth)-diagSize)/(double)diagSize < 0.5))) { \
115
/* Most likely no benefit to call TRMM or GEMM from MKL*/ \
116
product_triangular_matrix_matrix<EIGTYPE,Index,Mode,true, \
117
LhsStorageOrder,ConjugateLhs, RhsStorageOrder, ConjugateRhs, ColMajor, BuiltIn>::run( \
118
_rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha); \
119
/*std::cout << "TRMM_L: A is not square! Go to Eigen TRMM implementation!\n";*/ \
121
/* Make sense to call GEMM */ \
122
Map<const MatrixLhs, 0, OuterStride<> > lhsMap(_lhs,rows,depth,OuterStride<>(lhsStride)); \
123
MatrixLhs aa_tmp=lhsMap.template triangularView<Mode>(); \
124
MKL_INT aStride = aa_tmp.outerStride(); \
125
gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> blocking(_rows,_cols,_depth); \
126
general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor>::run( \
127
rows, cols, depth, aa_tmp.data(), aStride, _rhs, rhsStride, res, resStride, alpha, blocking, 0); \
129
/*std::cout << "TRMM_L: A is not square! Go to MKL GEMM implementation! " << nthr<<" \n";*/ \
133
char side = 'L', transa, uplo, diag = 'N'; \
136
MKL_INT m, n, lda, ldb; \
140
assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
143
m = (MKL_INT)diagSize; \
147
transa = (LhsStorageOrder==RowMajor) ? ((ConjugateLhs) ? 'C' : 'T') : 'N'; \
150
Map<const MatrixRhs, 0, OuterStride<> > rhs(_rhs,depth,cols,OuterStride<>(rhsStride)); \
151
MatrixX##EIGPREFIX b_tmp; \
153
if (ConjugateRhs) b_tmp = rhs.conjugate(); else b_tmp = rhs; \
155
ldb = b_tmp.outerStride(); \
158
uplo = IsLower ? 'L' : 'U'; \
159
if (LhsStorageOrder==RowMajor) uplo = (uplo == 'L') ? 'U' : 'L'; \
161
Map<const MatrixLhs, 0, OuterStride<> > lhs(_lhs,rows,depth,OuterStride<>(lhsStride)); \
164
if ((conjA!=0) || (SetDiag==0)) { \
165
if (conjA) a_tmp = lhs.conjugate(); else a_tmp = lhs; \
167
a_tmp.diagonal().setZero(); \
168
else if (IsUnitDiag) \
169
a_tmp.diagonal().setOnes();\
171
lda = a_tmp.outerStride(); \
176
/*std::cout << "TRMM_L: A is square! Go to MKL TRMM implementation! \n";*/ \
178
MKLPREFIX##trmm(&side, &uplo, &transa, &diag, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (MKLTYPE*)b, &ldb); \
180
/* Add op(a_triangular)*b into res*/ \
181
Map<MatrixX##EIGPREFIX, 0, OuterStride<> > res_tmp(res,rows,cols,OuterStride<>(resStride)); \
182
res_tmp=res_tmp+b_tmp; \
186
EIGEN_MKL_TRMM_L(double, double, d, d)
187
EIGEN_MKL_TRMM_L(dcomplex, MKL_Complex16, cd, z)
188
EIGEN_MKL_TRMM_L(float, float, f, s)
189
EIGEN_MKL_TRMM_L(scomplex, MKL_Complex8, cf, c)
191
// implements col-major += alpha * op(general) * op(triangular)
192
#define EIGEN_MKL_TRMM_R(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
193
template <typename Index, int Mode, \
194
int LhsStorageOrder, bool ConjugateLhs, \
195
int RhsStorageOrder, bool ConjugateRhs> \
196
struct product_triangular_matrix_matrix_trmm<EIGTYPE,Index,Mode,false, \
197
LhsStorageOrder,ConjugateLhs,RhsStorageOrder,ConjugateRhs,ColMajor> \
200
IsLower = (Mode&Lower) == Lower, \
201
SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \
202
IsUnitDiag = (Mode&UnitDiag) ? 1 : 0, \
203
IsZeroDiag = (Mode&ZeroDiag) ? 1 : 0, \
204
LowUp = IsLower ? Lower : Upper, \
205
conjA = ((RhsStorageOrder==ColMajor) && ConjugateRhs) ? 1 : 0 \
208
static EIGEN_DONT_INLINE void run( \
209
Index _rows, Index _cols, Index _depth, \
210
const EIGTYPE* _lhs, Index lhsStride, \
211
const EIGTYPE* _rhs, Index rhsStride, \
212
EIGTYPE* res, Index resStride, \
215
Index diagSize = (std::min)(_cols,_depth); \
216
Index rows = _rows; \
217
Index depth = IsLower ? _depth : diagSize; \
218
Index cols = IsLower ? diagSize : _cols; \
220
typedef Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> MatrixLhs; \
221
typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs; \
223
/* Non-square case - doesn't fit to MKL ?TRMM. Fall to default triangular product or call MKL ?GEMM*/ \
224
if (cols != depth) { \
226
int nthr = mkl_domain_get_max_threads(MKL_BLAS); \
228
if ((nthr==1) && (((std::max)(cols,depth)-diagSize)/(double)diagSize < 0.5)) { \
229
/* Most likely no benefit to call TRMM or GEMM from MKL*/ \
230
product_triangular_matrix_matrix<EIGTYPE,Index,Mode,false, \
231
LhsStorageOrder,ConjugateLhs, RhsStorageOrder, ConjugateRhs, ColMajor, BuiltIn>::run( \
232
_rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha); \
233
/*std::cout << "TRMM_R: A is not square! Go to Eigen TRMM implementation!\n";*/ \
235
/* Make sense to call GEMM */ \
236
Map<const MatrixRhs, 0, OuterStride<> > rhsMap(_rhs,depth,cols, OuterStride<>(rhsStride)); \
237
MatrixRhs aa_tmp=rhsMap.template triangularView<Mode>(); \
238
MKL_INT aStride = aa_tmp.outerStride(); \
239
gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> blocking(_rows,_cols,_depth); \
240
general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor>::run( \
241
rows, cols, depth, _lhs, lhsStride, aa_tmp.data(), aStride, res, resStride, alpha, blocking, 0); \
243
/*std::cout << "TRMM_R: A is not square! Go to MKL GEMM implementation! " << nthr<<" \n";*/ \
247
char side = 'R', transa, uplo, diag = 'N'; \
250
MKL_INT m, n, lda, ldb; \
254
assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
258
n = (MKL_INT)diagSize; \
261
transa = (RhsStorageOrder==RowMajor) ? ((ConjugateRhs) ? 'C' : 'T') : 'N'; \
264
Map<const MatrixLhs, 0, OuterStride<> > lhs(_lhs,rows,depth,OuterStride<>(lhsStride)); \
265
MatrixX##EIGPREFIX b_tmp; \
267
if (ConjugateLhs) b_tmp = lhs.conjugate(); else b_tmp = lhs; \
269
ldb = b_tmp.outerStride(); \
272
uplo = IsLower ? 'L' : 'U'; \
273
if (RhsStorageOrder==RowMajor) uplo = (uplo == 'L') ? 'U' : 'L'; \
275
Map<const MatrixRhs, 0, OuterStride<> > rhs(_rhs,depth,cols, OuterStride<>(rhsStride)); \
278
if ((conjA!=0) || (SetDiag==0)) { \
279
if (conjA) a_tmp = rhs.conjugate(); else a_tmp = rhs; \
281
a_tmp.diagonal().setZero(); \
282
else if (IsUnitDiag) \
283
a_tmp.diagonal().setOnes();\
285
lda = a_tmp.outerStride(); \
290
/*std::cout << "TRMM_R: A is square! Go to MKL TRMM implementation! \n";*/ \
292
MKLPREFIX##trmm(&side, &uplo, &transa, &diag, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (MKLTYPE*)b, &ldb); \
294
/* Add op(a_triangular)*b into res*/ \
295
Map<MatrixX##EIGPREFIX, 0, OuterStride<> > res_tmp(res,rows,cols,OuterStride<>(resStride)); \
296
res_tmp=res_tmp+b_tmp; \
300
EIGEN_MKL_TRMM_R(double, double, d, d)
301
EIGEN_MKL_TRMM_R(dcomplex, MKL_Complex16, cd, z)
302
EIGEN_MKL_TRMM_R(float, float, f, s)
303
EIGEN_MKL_TRMM_R(scomplex, MKL_Complex8, cf, c)
305
} // end namespace internal
307
} // end namespace Eigen
309
#endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_MKL_H