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 FLENS_BLAS_LEVEL3_R2K_TCC
 34 #define FLENS_BLAS_LEVEL3_R2K_TCC
 35 
 36 #include <flens/typedefs.h>
 37 
 38 namespace flens { namespace blas {
 39 
 40 //-- her2k
 41 template <typename ALPHA, typename MA, typename MB, typename BETA, typename MC>
 42 void
 43 r2k(Transpose trans,
 44     const ALPHA &alpha,
 45     const GeMatrix<MA> &A, const GeMatrix<MB> &B,
 46     const BETA &beta,
 47     HeMatrix<MC> &C)
 48 {
 49     typedef typename GeMatrix<MA>::IndexType IndexType;
 50 
 51     IndexType n = (trans==NoTrans) ? A.numRows()
 52                                    : A.numCols();
 53     IndexType k = (trans==NoTrans) ? A.numCols()
 54                                    : A.numRows();
 55 
 56     ASSERT(A.numRows()==B.numRows());
 57     ASSERT(A.numCols()==B.numCols());
 58     ASSERT((beta==static_cast<BETA>(0)) || (C.dim()==n));
 59 
 60     if (C.dim()!=n) {
 61         C.resize(n, n);
 62     }
 63 
 64 #   ifdef HAVE_CXXBLAS_HER2K
 65     cxxblas::her2k(MC::order, C.upLo(), trans,
 66                    n, k,
 67                    alpha,
 68                    A.data(), A.leadingDimension(),
 69                    B.data(), B.leadingDimension(),
 70                    beta,
 71                    C.data(), C.leadingDimension());
 72 #   else
 73     ASSERT(0);
 74 #   endif
 75 }
 76 
 77 //-- syr2k
 78 template <typename ALPHA, typename MA, typename MB, typename BETA, typename MC>
 79 void
 80 r2k(Transpose trans,
 81     const ALPHA &alpha,
 82     const GeMatrix<MA> &A, const GeMatrix<MB> &B,
 83     const BETA &beta,
 84     SyMatrix<MC> &C)
 85 {
 86     typedef typename GeMatrix<MA>::IndexType IndexType;
 87 
 88     IndexType n = (trans==NoTrans) ? A.numRows()
 89                                    : A.numCols();
 90     IndexType k = (trans==NoTrans) ? A.numCols()
 91                                    : A.numRows();
 92 
 93     ASSERT(A.numRows()==B.numRows());
 94     ASSERT(A.numCols()==B.numCols());
 95     ASSERT((beta==static_cast<BETA>(0)) || (C.dim()==n));
 96 
 97     if (C.dim()!=n) {
 98         C.resize(n, n);
 99     }
100 
101 #   ifdef HAVE_CXXBLAS_SYR2K
102     cxxblas::syr2k(MC::order, C.upLo(), trans,
103                    n, k,
104                    alpha,
105                    A.data(), A.leadingDimension(),
106                    B.data(), B.leadingDimension(),
107                    beta,
108                    C.data(), C.leadingDimension());
109 #   else
110     ASSERT(0);
111 #   endif
112 }
113 
114 } } // namespace blas, flens
115 
116 #endif // FLENS_BLAS_LEVEL3_R2K_TCC