1 /* 2 Copyright (c) 2011, Intel Corporation. All rights reserved. 3 4 Redistribution and use in source and binary forms, with or without modification, 5 are permitted provided that the following conditions are met: 6 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. 15 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. 26 27 ******************************************************************************** 28 * Content : Eigen bindings to Intel(R) MKL 29 * Triangular matrix-vector product functionality based on ?TRMV. 30 ******************************************************************************** 31 */ 32 33 #ifndef EIGEN_TRIANGULAR_MATRIX_VECTOR_MKL_H 34 #define EIGEN_TRIANGULAR_MATRIX_VECTOR_MKL_H 35 36 namespace Eigen { 37 38 namespace internal { 39 40 /********************************************************************** 41 * This file implements triangular matrix-vector multiplication using BLAS 42 **********************************************************************/ 43 44 // trmv/hemv specialization 45 46 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs, int StorageOrder> 47 struct triangular_matrix_vector_product_trmv : 48 triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,StorageOrder,BuiltIn> {}; 49 50 #define EIGEN_MKL_TRMV_SPECIALIZE(Scalar) \ 51 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \ 52 struct triangular_matrix_vector_product<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,ColMajor,Specialized> { \ 53 static void run(Index _rows, Index _cols, const Scalar* _lhs, Index lhsStride, \ 54 const Scalar* _rhs, Index rhsIncr, Scalar* _res, Index resIncr, Scalar alpha) { \ 55 triangular_matrix_vector_product_trmv<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,ColMajor>::run( \ 56 _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \ 57 } \ 58 }; \ 59 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \ 60 struct triangular_matrix_vector_product<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,RowMajor,Specialized> { \ 61 static void run(Index _rows, Index _cols, const Scalar* _lhs, Index lhsStride, \ 62 const Scalar* _rhs, Index rhsIncr, Scalar* _res, Index resIncr, Scalar alpha) { \ 63 triangular_matrix_vector_product_trmv<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,RowMajor>::run( \ 64 _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \ 65 } \ 66 }; 67 68 EIGEN_MKL_TRMV_SPECIALIZE(double) 69 EIGEN_MKL_TRMV_SPECIALIZE(float) 70 EIGEN_MKL_TRMV_SPECIALIZE(dcomplex) 71 EIGEN_MKL_TRMV_SPECIALIZE(scomplex) 72 73 // implements col-major: res += alpha * op(triangular) * vector 74 #define EIGEN_MKL_TRMV_CM(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \ 75 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \ 76 struct triangular_matrix_vector_product_trmv<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,ColMajor> { \ 77 enum { \ 78 IsLower = (Mode&Lower) == Lower, \ 79 SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \ 80 IsUnitDiag = (Mode&UnitDiag) ? 1 : 0, \ 81 IsZeroDiag = (Mode&ZeroDiag) ? 1 : 0, \ 82 LowUp = IsLower ? Lower : Upper \ 83 }; \ 84 static void run(Index _rows, Index _cols, const EIGTYPE* _lhs, Index lhsStride, \ 85 const EIGTYPE* _rhs, Index rhsIncr, EIGTYPE* _res, Index resIncr, EIGTYPE alpha) \ 86 { \ 87 if (ConjLhs || IsZeroDiag) { \ 88 triangular_matrix_vector_product<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,ColMajor,BuiltIn>::run( \ 89 _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \ 90 return; \ 91 }\ 92 Index size = (std::min)(_rows,_cols); \ 93 Index rows = IsLower ? _rows : size; \ 94 Index cols = IsLower ? size : _cols; \ 95 \ 96 typedef VectorX##EIGPREFIX VectorRhs; \ 97 EIGTYPE *x, *y;\ 98 \ 99 /* Set x*/ \ 100 Map<const VectorRhs, 0, InnerStride<> > rhs(_rhs,cols,InnerStride<>(rhsIncr)); \ 101 VectorRhs x_tmp; \ 102 if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \ 103 x = x_tmp.data(); \ 104 \ 105 /* Square part handling */\ 106 \ 107 char trans, uplo, diag; \ 108 MKL_INT m, n, lda, incx, incy; \ 109 EIGTYPE const *a; \ 110 MKLTYPE alpha_, beta_; \ 111 assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \ 112 assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(beta_, EIGTYPE(1)); \ 113 \ 114 /* Set m, n */ \ 115 n = (MKL_INT)size; \ 116 lda = lhsStride; \ 117 incx = 1; \ 118 incy = resIncr; \ 119 \ 120 /* Set uplo, trans and diag*/ \ 121 trans = 'N'; \ 122 uplo = IsLower ? 'L' : 'U'; \ 123 diag = IsUnitDiag ? 'U' : 'N'; \ 124 \ 125 /* call ?TRMV*/ \ 126 MKLPREFIX##trmv(&uplo, &trans, &diag, &n, (const MKLTYPE*)_lhs, &lda, (MKLTYPE*)x, &incx); \ 127 \ 128 /* Add op(a_tr)rhs into res*/ \ 129 MKLPREFIX##axpy(&n, &alpha_,(const MKLTYPE*)x, &incx, (MKLTYPE*)_res, &incy); \ 130 /* Non-square case - doesn't fit to MKL ?TRMV. Fall to default triangular product*/ \ 131 if (size<(std::max)(rows,cols)) { \ 132 typedef Matrix<EIGTYPE, Dynamic, Dynamic> MatrixLhs; \ 133 if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \ 134 x = x_tmp.data(); \ 135 if (size<rows) { \ 136 y = _res + size*resIncr; \ 137 a = _lhs + size; \ 138 m = rows-size; \ 139 n = size; \ 140 } \ 141 else { \ 142 x += size; \ 143 y = _res; \ 144 a = _lhs + size*lda; \ 145 m = size; \ 146 n = cols-size; \ 147 } \ 148 MKLPREFIX##gemv(&trans, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)x, &incx, &beta_, (MKLTYPE*)y, &incy); \ 149 } \ 150 } \ 151 }; 152 153 EIGEN_MKL_TRMV_CM(double, double, d, d) 154 EIGEN_MKL_TRMV_CM(dcomplex, MKL_Complex16, cd, z) 155 EIGEN_MKL_TRMV_CM(float, float, f, s) 156 EIGEN_MKL_TRMV_CM(scomplex, MKL_Complex8, cf, c) 157 158 // implements row-major: res += alpha * op(triangular) * vector 159 #define EIGEN_MKL_TRMV_RM(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \ 160 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \ 161 struct triangular_matrix_vector_product_trmv<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,RowMajor> { \ 162 enum { \ 163 IsLower = (Mode&Lower) == Lower, \ 164 SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \ 165 IsUnitDiag = (Mode&UnitDiag) ? 1 : 0, \ 166 IsZeroDiag = (Mode&ZeroDiag) ? 1 : 0, \ 167 LowUp = IsLower ? Lower : Upper \ 168 }; \ 169 static void run(Index _rows, Index _cols, const EIGTYPE* _lhs, Index lhsStride, \ 170 const EIGTYPE* _rhs, Index rhsIncr, EIGTYPE* _res, Index resIncr, EIGTYPE alpha) \ 171 { \ 172 if (IsZeroDiag) { \ 173 triangular_matrix_vector_product<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,RowMajor,BuiltIn>::run( \ 174 _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \ 175 return; \ 176 }\ 177 Index size = (std::min)(_rows,_cols); \ 178 Index rows = IsLower ? _rows : size; \ 179 Index cols = IsLower ? size : _cols; \ 180 \ 181 typedef VectorX##EIGPREFIX VectorRhs; \ 182 EIGTYPE *x, *y;\ 183 \ 184 /* Set x*/ \ 185 Map<const VectorRhs, 0, InnerStride<> > rhs(_rhs,cols,InnerStride<>(rhsIncr)); \ 186 VectorRhs x_tmp; \ 187 if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \ 188 x = x_tmp.data(); \ 189 \ 190 /* Square part handling */\ 191 \ 192 char trans, uplo, diag; \ 193 MKL_INT m, n, lda, incx, incy; \ 194 EIGTYPE const *a; \ 195 MKLTYPE alpha_, beta_; \ 196 assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \ 197 assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(beta_, EIGTYPE(1)); \ 198 \ 199 /* Set m, n */ \ 200 n = (MKL_INT)size; \ 201 lda = lhsStride; \ 202 incx = 1; \ 203 incy = resIncr; \ 204 \ 205 /* Set uplo, trans and diag*/ \ 206 trans = ConjLhs ? 'C' : 'T'; \ 207 uplo = IsLower ? 'U' : 'L'; \ 208 diag = IsUnitDiag ? 'U' : 'N'; \ 209 \ 210 /* call ?TRMV*/ \ 211 MKLPREFIX##trmv(&uplo, &trans, &diag, &n, (const MKLTYPE*)_lhs, &lda, (MKLTYPE*)x, &incx); \ 212 \ 213 /* Add op(a_tr)rhs into res*/ \ 214 MKLPREFIX##axpy(&n, &alpha_,(const MKLTYPE*)x, &incx, (MKLTYPE*)_res, &incy); \ 215 /* Non-square case - doesn't fit to MKL ?TRMV. Fall to default triangular product*/ \ 216 if (size<(std::max)(rows,cols)) { \ 217 typedef Matrix<EIGTYPE, Dynamic, Dynamic> MatrixLhs; \ 218 if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \ 219 x = x_tmp.data(); \ 220 if (size<rows) { \ 221 y = _res + size*resIncr; \ 222 a = _lhs + size*lda; \ 223 m = rows-size; \ 224 n = size; \ 225 } \ 226 else { \ 227 x += size; \ 228 y = _res; \ 229 a = _lhs + size; \ 230 m = size; \ 231 n = cols-size; \ 232 } \ 233 MKLPREFIX##gemv(&trans, &n, &m, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)x, &incx, &beta_, (MKLTYPE*)y, &incy); \ 234 } \ 235 } \ 236 }; 237 238 EIGEN_MKL_TRMV_RM(double, double, d, d) 239 EIGEN_MKL_TRMV_RM(dcomplex, MKL_Complex16, cd, z) 240 EIGEN_MKL_TRMV_RM(float, float, f, s) 241 EIGEN_MKL_TRMV_RM(scomplex, MKL_Complex8, cf, c) 242 243 } // end namespase internal 244 245 } // end namespace Eigen 246 247 #endif // EIGEN_TRIANGULAR_MATRIX_VECTOR_MKL_H 248