1 /*
2 * Copyright (c) 2010, Michael Lehn
3 *
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 *
10 * 1) Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2) Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the
15 * distribution.
16 * 3) Neither the name of the FLENS development group nor the names of
17 * its contributors may be used to endorse or promote products derived
18 * from this software without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
30 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 #ifndef CXXBLAS_LEVEL2_HPMV_TCC
34 #define CXXBLAS_LEVEL2_HPMV_TCC 1
35
36 #include <complex>
37 #include "xflens/cxxblas/cxxblas.h"
38
39 namespace cxxblas {
40
41 template <typename IndexType, typename ALPHA, typename MA, typename VX,
42 typename BETA, typename VY>
43 void
hpmv_generic(StorageOrder order,StorageUpLo upLo,Transpose conjugateA,IndexType n,const ALPHA & alpha,const MA * A,const VX * x,IndexType incX,const BETA & beta,VY * y,IndexType incY)44 hpmv_generic(StorageOrder order, StorageUpLo upLo, Transpose conjugateA,
45 IndexType n,
46 const ALPHA &alpha,
47 const MA *A,
48 const VX *x, IndexType incX,
49 const BETA &beta,
50 VY *y, IndexType incY)
51 {
52 if (order==ColMajor) {
53 upLo = (upLo==Upper) ? Lower : Upper;
54 conjugateA = Transpose(conjugateA^Conj);
55 }
56 scal_init_generic(n, beta, y, incY);
57 if (upLo==Upper) {
58 if (conjugateA==Conj) {
59 for (IndexType i=0, iY=0, iX=0; i<n; ++i, iX+=incX, iY+=incY) {
60 y[iY] += alpha*cxxblas::real(A[i*(2*n-i+1)/2]) * x[iX];
61
62 VY y_ = VY(0);
63 dot_generic(n-i-1, A+i*(2*n-i+1)/2+1, IndexType(1),
64 x+iX+incX, incX, y_);
65 y[iY] += alpha*y_;
66 axpy_generic(n-i-1, alpha*x[iX],
67 A+i*(2*n-i+1)/2+1, IndexType(1),
68 y+iY+incY, incY);
69 }
70 } else {
71 for (IndexType i=0, iY=0, iX=0; i<n; ++i, iX+=incX, iY+=incY) {
72 y[iY] += alpha*cxxblas::real(A[i*(2*n-i+1)/2]) * x[iX];
73
74 VY y_ = VY(0);
75 dotu_generic(n-i-1, A+i*(2*n-i+1)/2+1, IndexType(1),
76 x+iX+incX, incX, y_);
77 y[iY] += alpha*y_;
78 acxpy_generic(n-i-1, alpha*x[iX],
79 A+i*(2*n-i+1)/2+1, IndexType(1),
80 y+iY+incY, incY);
81 }
82 }
83 } else {
84 if (conjugateA==Conj) {
85 for (IndexType i=0, iY=0, iX=0; i<n; ++i, iX+=incX, iY+=incY) {
86 y[iY] += alpha*cxxblas::real(A[i+i*(i+1)/2]) * x[iX];
87
88 VY y_ = VY(0);
89 dot_generic(i, A+i*(i+1)/2, IndexType(1), x, incX, y_);
90 y[iY] += alpha*y_;
91 axpy_generic(i, alpha*x[iX],
92 A+i*(i+1)/2, IndexType(1),
93 y, incY);
94 }
95 } else {
96 for (IndexType i=0, iY=0, iX=0; i<n; ++i, iX+=incX, iY+=incY) {
97 y[iY] += alpha*cxxblas::real(A[i+i*(i+1)/2]) * x[iX];
98
99 VY y_ = VY(0);
100 dotu_generic(i, A+i*(i+1)/2, IndexType(1), x, incX, y_);
101 y[iY] += alpha*y_;
102 acxpy_generic(i, alpha*x[iX],
103 A+i*(i+1)/2, IndexType(1),
104 y, incY);
105 }
106 }
107 }
108 }
109
110 //------------------------------------------------------------------------------
111
112 template <typename IndexType, typename ALPHA, typename MA, typename VX,
113 typename BETA, typename VY>
114 void
hpmv(StorageOrder order,StorageUpLo upLo,IndexType n,const ALPHA & alpha,const MA * A,const VX * x,IndexType incX,const BETA & beta,VY * y,IndexType incY)115 hpmv(StorageOrder order, StorageUpLo upLo,
116 IndexType n,
117 const ALPHA &alpha,
118 const MA *A,
119 const VX *x, IndexType incX,
120 const BETA &beta,
121 VY *y, IndexType incY)
122 {
123 CXXBLAS_DEBUG_OUT("hpmv_generic");
124
125 if (incX<0) {
126 x -= incX*(n-1);
127 }
128 if (incY<0) {
129 y -= incY*(n-1);
130 }
131 hpmv_generic(order, upLo, NoTrans, n, alpha, A, x, incX, beta, y, incY);
132 }
133
134
135 #ifdef HAVE_CBLAS
136
137 // chpmv
138 template <typename IndexType>
139 typename If<IndexType>::isBlasCompatibleInteger
hpmv(StorageOrder order,StorageUpLo upLo,IndexType n,const ComplexFloat & alpha,const ComplexFloat * A,const ComplexFloat * x,IndexType incX,const ComplexFloat & beta,ComplexFloat * y,IndexType incY)140 hpmv(StorageOrder order, StorageUpLo upLo,
141 IndexType n,
142 const ComplexFloat &alpha,
143 const ComplexFloat *A,
144 const ComplexFloat *x, IndexType incX,
145 const ComplexFloat &beta,
146 ComplexFloat *y, IndexType incY)
147 {
148 CXXBLAS_DEBUG_OUT("[" BLAS_IMPL "] cblas_chpmv");
149
150 cblas_chpmv(CBLAS::getCblasType(order), CBLAS::getCblasType(upLo), n,
151 reinterpret_cast<const float *>(&alpha),
152 reinterpret_cast<const float *>(A),
153 reinterpret_cast<const float *>(x), incX,
154 reinterpret_cast<const float *>(&beta),
155 reinterpret_cast<float *>(y), incY);
156 }
157
158 // zhpmv
159 template <typename IndexType>
160 typename If<IndexType>::isBlasCompatibleInteger
hpmv(StorageOrder order,StorageUpLo upLo,IndexType n,const ComplexDouble & alpha,const ComplexDouble * A,const ComplexDouble * x,IndexType incX,const ComplexDouble & beta,ComplexDouble * y,IndexType incY)161 hpmv(StorageOrder order, StorageUpLo upLo,
162 IndexType n,
163 const ComplexDouble &alpha,
164 const ComplexDouble *A,
165 const ComplexDouble *x, IndexType incX,
166 const ComplexDouble &beta,
167 ComplexDouble *y, IndexType incY)
168 {
169 CXXBLAS_DEBUG_OUT("[" BLAS_IMPL "] cblas_zhpmv");
170
171 cblas_zhpmv(CBLAS::getCblasType(order), CBLAS::getCblasType(upLo), n,
172 reinterpret_cast<const double *>(&alpha),
173 reinterpret_cast<const double *>(A),
174 reinterpret_cast<const double *>(x), incX,
175 reinterpret_cast<const double *>(&beta),
176 reinterpret_cast<double *>(y), incY);
177 }
178
179 #endif // HAVE_CBLAS
180
181
182 } // namespace cxxblas
183
184 #endif // CXXBLAS_LEVEL2_HPMV_TCC
185