// Copyright (C) 2009-2015 National ICT Australia (NICTA) // // This Source Code Form is subject to the terms of the Mozilla Public // License, v. 2.0. If a copy of the MPL was not distributed with this // file, You can obtain one at http://mozilla.org/MPL/2.0/. // ------------------------------------------------------------------- // // Written by Conrad Sanderson - http://conradsanderson.id.au //! \addtogroup glue_mixed //! @{ //! matrix multiplication with different element types template inline void glue_mixed_times::apply(Mat::eT>& out, const mtGlue::eT, T1, T2, glue_mixed_times>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; const unwrap_check_mixed tmp1(X.A, out); const unwrap_check_mixed tmp2(X.B, out); const Mat& A = tmp1.M; const Mat& B = tmp2.M; arma_debug_assert_mul_size(A, B, "matrix multiplication"); out.set_size(A.n_rows, B.n_cols); gemm_mixed<>::apply(out, A, B); } //! matrix addition with different element types template inline void glue_mixed_plus::apply(Mat::eT>& out, const mtGlue::eT, T1, T2, glue_mixed_plus>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const Proxy A(X.A); const Proxy B(X.B); arma_debug_assert_same_size(A, B, "addition"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); out.set_size(n_rows, n_cols); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (Proxy::prefer_at_accessor || Proxy::prefer_at_accessor); if(prefer_at_accessor == false) { typename Proxy::ea_type AA = A.get_ea(); typename Proxy::ea_type BB = B.get_ea(); if(memory::is_aligned(out_mem)) { memory::mark_as_aligned(out_mem); for(uword i=0; i::apply(AA[i]) + upgrade_val::apply(BB[i]); } } else { for(uword i=0; i::apply(AA[i]) + upgrade_val::apply(BB[i]); } } } else { for(uword col=0; col < n_cols; ++col) for(uword row=0; row < n_rows; ++row) { (*out_mem) = upgrade_val::apply(A.at(row,col)) + upgrade_val::apply(B.at(row,col)); out_mem++; } } } //! matrix subtraction with different element types template inline void glue_mixed_minus::apply(Mat::eT>& out, const mtGlue::eT, T1, T2, glue_mixed_minus>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const Proxy A(X.A); const Proxy B(X.B); arma_debug_assert_same_size(A, B, "subtraction"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); out.set_size(n_rows, n_cols); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (Proxy::prefer_at_accessor || Proxy::prefer_at_accessor); if(prefer_at_accessor == false) { typename Proxy::ea_type AA = A.get_ea(); typename Proxy::ea_type BB = B.get_ea(); if(memory::is_aligned(out_mem)) { memory::mark_as_aligned(out_mem); for(uword i=0; i::apply(AA[i]) - upgrade_val::apply(BB[i]); } } else { for(uword i=0; i::apply(AA[i]) - upgrade_val::apply(BB[i]); } } } else { for(uword col=0; col < n_cols; ++col) for(uword row=0; row < n_rows; ++row) { (*out_mem) = upgrade_val::apply(A.at(row,col)) - upgrade_val::apply(B.at(row,col)); out_mem++; } } } //! element-wise matrix division with different element types template inline void glue_mixed_div::apply(Mat::eT>& out, const mtGlue::eT, T1, T2, glue_mixed_div>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const Proxy A(X.A); const Proxy B(X.B); arma_debug_assert_same_size(A, B, "element-wise division"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); out.set_size(n_rows, n_cols); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (Proxy::prefer_at_accessor || Proxy::prefer_at_accessor); if(prefer_at_accessor == false) { typename Proxy::ea_type AA = A.get_ea(); typename Proxy::ea_type BB = B.get_ea(); if(memory::is_aligned(out_mem)) { memory::mark_as_aligned(out_mem); for(uword i=0; i::apply(AA[i]) / upgrade_val::apply(BB[i]); } } else { for(uword i=0; i::apply(AA[i]) / upgrade_val::apply(BB[i]); } } } else { for(uword col=0; col < n_cols; ++col) for(uword row=0; row < n_rows; ++row) { (*out_mem) = upgrade_val::apply(A.at(row,col)) / upgrade_val::apply(B.at(row,col)); out_mem++; } } } //! element-wise matrix multiplication with different element types template inline void glue_mixed_schur::apply(Mat::eT>& out, const mtGlue::eT, T1, T2, glue_mixed_schur>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const Proxy A(X.A); const Proxy B(X.B); arma_debug_assert_same_size(A, B, "element-wise multiplication"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); out.set_size(n_rows, n_cols); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (Proxy::prefer_at_accessor || Proxy::prefer_at_accessor); if(prefer_at_accessor == false) { typename Proxy::ea_type AA = A.get_ea(); typename Proxy::ea_type BB = B.get_ea(); if(memory::is_aligned(out_mem)) { memory::mark_as_aligned(out_mem); for(uword i=0; i::apply(AA[i]) * upgrade_val::apply(BB[i]); } } else { for(uword i=0; i::apply(AA[i]) * upgrade_val::apply(BB[i]); } } } else { for(uword col=0; col < n_cols; ++col) for(uword row=0; row < n_rows; ++row) { (*out_mem) = upgrade_val::apply(A.at(row,col)) * upgrade_val::apply(B.at(row,col)); out_mem++; } } } // // // //! cube addition with different element types template inline void glue_mixed_plus::apply(Cube::eT>& out, const mtGlueCube::eT, T1, T2, glue_mixed_plus>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const ProxyCube A(X.A); const ProxyCube B(X.B); arma_debug_assert_same_size(A, B, "addition"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); const uword n_slices = A.get_n_slices(); out.set_size(n_rows, n_cols, n_slices); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (ProxyCube::prefer_at_accessor || ProxyCube::prefer_at_accessor); if(prefer_at_accessor == false) { typename ProxyCube::ea_type AA = A.get_ea(); typename ProxyCube::ea_type BB = B.get_ea(); for(uword i=0; i::apply(AA[i]) + upgrade_val::apply(BB[i]); } } else { for(uword slice = 0; slice < n_slices; ++slice) for(uword col = 0; col < n_cols; ++col ) for(uword row = 0; row < n_rows; ++row ) { (*out_mem) = upgrade_val::apply(A.at(row,col,slice)) + upgrade_val::apply(B.at(row,col,slice)); out_mem++; } } } //! cube subtraction with different element types template inline void glue_mixed_minus::apply(Cube::eT>& out, const mtGlueCube::eT, T1, T2, glue_mixed_minus>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const ProxyCube A(X.A); const ProxyCube B(X.B); arma_debug_assert_same_size(A, B, "subtraction"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); const uword n_slices = A.get_n_slices(); out.set_size(n_rows, n_cols, n_slices); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (ProxyCube::prefer_at_accessor || ProxyCube::prefer_at_accessor); if(prefer_at_accessor == false) { typename ProxyCube::ea_type AA = A.get_ea(); typename ProxyCube::ea_type BB = B.get_ea(); for(uword i=0; i::apply(AA[i]) - upgrade_val::apply(BB[i]); } } else { for(uword slice = 0; slice < n_slices; ++slice) for(uword col = 0; col < n_cols; ++col ) for(uword row = 0; row < n_rows; ++row ) { (*out_mem) = upgrade_val::apply(A.at(row,col,slice)) - upgrade_val::apply(B.at(row,col,slice)); out_mem++; } } } //! element-wise cube division with different element types template inline void glue_mixed_div::apply(Cube::eT>& out, const mtGlueCube::eT, T1, T2, glue_mixed_div>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const ProxyCube A(X.A); const ProxyCube B(X.B); arma_debug_assert_same_size(A, B, "element-wise division"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); const uword n_slices = A.get_n_slices(); out.set_size(n_rows, n_cols, n_slices); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (ProxyCube::prefer_at_accessor || ProxyCube::prefer_at_accessor); if(prefer_at_accessor == false) { typename ProxyCube::ea_type AA = A.get_ea(); typename ProxyCube::ea_type BB = B.get_ea(); for(uword i=0; i::apply(AA[i]) / upgrade_val::apply(BB[i]); } } else { for(uword slice = 0; slice < n_slices; ++slice) for(uword col = 0; col < n_cols; ++col ) for(uword row = 0; row < n_rows; ++row ) { (*out_mem) = upgrade_val::apply(A.at(row,col,slice)) / upgrade_val::apply(B.at(row,col,slice)); out_mem++; } } } //! element-wise cube multiplication with different element types template inline void glue_mixed_schur::apply(Cube::eT>& out, const mtGlueCube::eT, T1, T2, glue_mixed_schur>& X) { arma_extra_debug_sigprint(); typedef typename T1::elem_type eT1; typedef typename T2::elem_type eT2; typedef typename promote_type::result out_eT; promote_type::check(); const ProxyCube A(X.A); const ProxyCube B(X.B); arma_debug_assert_same_size(A, B, "element-wise multiplication"); const uword n_rows = A.get_n_rows(); const uword n_cols = A.get_n_cols(); const uword n_slices = A.get_n_slices(); out.set_size(n_rows, n_cols, n_slices); out_eT* out_mem = out.memptr(); const uword n_elem = out.n_elem; const bool prefer_at_accessor = (ProxyCube::prefer_at_accessor || ProxyCube::prefer_at_accessor); if(prefer_at_accessor == false) { typename ProxyCube::ea_type AA = A.get_ea(); typename ProxyCube::ea_type BB = B.get_ea(); for(uword i=0; i::apply(AA[i]) * upgrade_val::apply(BB[i]); } } else { for(uword slice = 0; slice < n_slices; ++slice) for(uword col = 0; col < n_cols; ++col ) for(uword row = 0; row < n_rows; ++row ) { (*out_mem) = upgrade_val::apply(A.at(row,col,slice)) * upgrade_val::apply(B.at(row,col,slice)); out_mem++; } } } //! @}