/******************************************************************************* * * TRIQS: a Toolbox for Research in Interacting Quantum Systems * * Copyright (C) 2011 by O. Parcollet * * TRIQS is free software: you can redistribute it and/or modify it under the * terms of the GNU General Public License as published by the Free Software * Foundation, either version 3 of the License, or (at your option) any later * version. * * TRIQS is distributed in the hope that it will be useful, but WITHOUT ANY * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS * FOR A PARTICULAR PURPOSE. See the GNU General Public License for more * details. * * You should have received a copy of the GNU General Public License along with * TRIQS. If not, see . * ******************************************************************************/ #ifndef TRIQS_ARRAYS_LINALG_DET_INV_H #define TRIQS_ARRAYS_LINALG_DET_INV_H #include "../impl/common.hpp" #include "../matrix.hpp" #include "../blas_lapack/getrf.hpp" #include "../blas_lapack/getri.hpp" namespace triqs { namespace arrays { /// Error which occurs during the matrix inversion class matrix_inverse_exception : public triqs::runtime_error {}; /** * Lazy result of inverse(M) where M can be : * * a matrix, a matrix_view * * any matrix expression * The object is lazy, it does not do any computation. * It can be copied at no cost * It keeps view of the object A if it a matrix, a copy if it is a formal expression. */ template struct inverse_lazy; template inverse_lazy::type> inverse(A &&a) { return {std::forward(a)}; } // ----------------- implementation ----------------------------------------- // worker takes a contiguous view and compute the det and inverse in two steps. // it is separated in case of multiple use (no reallocation of ipvi, etc...) // A can be a matrix, a matrix_view template class det_and_inverse_worker { typedef typename A::value_type value_type; typedef matrix_view V_type; A a; int dim; triqs::arrays::vector ipiv; int step, info; value_type _det; public: det_and_inverse_worker(A a_) : a(std::move(a_)), dim(first_dim(a)), ipiv(dim), step(0) { if (first_dim(a) != second_dim(a)) TRIQS_RUNTIME_ERROR << "Inverse/Det error:non-square matrix. Dimensions are :(" << first_dim(a) << "," << second_dim(a) << ")\n "; if (!(has_contiguous_data(a))) TRIQS_RUNTIME_ERROR << "det_and_inverse_worker only takes a contiguous view"; } value_type det() { V_type W = fortran_view(a); _step1(W); _compute_det(W); return _det; } A const &inverse() { if (step < 2) { V_type W = fortran_view(a); _step1(W); _step2(W); } return a; } private: // no need of special traversal template V_type fortran_view(MT const &x) { if (x.indexmap().memory_layout_is_c()) return x.transpose(); else return x; } void _step1(V_type &W) { if (step > 0) return; step = 1; info = lapack::getrf(W, ipiv); if (info < 0) throw matrix_inverse_exception() << "Inverse/Det error : failure of getrf lapack routine "; } void _compute_det(V_type const &W) { if (step > 1) return; _det = 1; for (size_t i = 0; i < dim; i++) _det *= W(i, i); bool flip = false; // compute the sign of the permutation for (size_t i = 0; i < dim; i++) { if (ipiv(i) != int(i) + 1) flip = !(flip); } _det = (flip ? -_det : _det); } void _step2(V_type &W) { assert(step == 1); // if (step==1) return; step = 2; _compute_det(W); info = lapack::getri(W, ipiv); if (info != 0) throw matrix_inverse_exception() << "Inverse/Det error : matrix is not invertible"; } }; //----------------------------------------------------------- template struct inverse_lazy : TRIQS_CONCEPT_TAG_NAME(ImmutableMatrix) { typedef typename std::remove_reference::type A_t; typedef typename std::remove_const::type value_type; typedef typename A_t::domain_type domain_type; typedef matrix M_type; typedef matrix_view M_view_type; template inverse_lazy(AA &&a_) : a(std::forward(a_)), M{}, computed{false} { if (first_dim(a) != second_dim(a)) TRIQS_RUNTIME_ERROR << "Inverse : matrix is not square but of size " << first_dim(a) << " x " << second_dim(a); } domain_type domain() const { return a.domain(); } A const & input() const { return a;} template value_type const &operator()(K0 const &k0, K1 const &k1) const { activate(); return M(k0, k1); } M_type const &operator()() const { activate(); return M; } friend std::ostream &operator<<(std::ostream &out, inverse_lazy const &x) { return out << "inverse(" << x.a << ")"; } private: A a; mutable M_type M; mutable bool computed; void activate() const { if (computed) return; M = a; auto worker = det_and_inverse_worker {M}; worker.inverse(); computed = true; } }; // Optimized implementation of = // if M = inverse(M) with the SAME object, then we do not need to copy the data template ENABLE_IF(is_matrix_or_view::type>) triqs_arrays_assign_delegation(MT &lhs, inverse_lazy const &rhs) { static_assert(is_matrix_or_view::value, "Can only assign an inverse matrix to a matrix or a matrix_view"); bool M_eq_inverse_M = ((lhs.indexmap().memory_indices_layout() == rhs.input().indexmap().memory_indices_layout()) && (lhs.data_start() == rhs.input().data_start()) && (has_contiguous_data(lhs))); if (!M_eq_inverse_M) { lhs = rhs(); } else { blas_lapack_tools::reflexive_qcache C(lhs); // a reflexive cache will use a temporary "regrouping" copy if and only if needed det_and_inverse_worker W(C()); // the worker to make the inversion of the lhs... W.inverse(); // worker is working ... } } //------------------- det ---------------------------------------- template typename std::remove_reference::type::value_type determinant(A &&a) { // makes a temporary copy of A if A is a const & // If a is a matrix &&, it is moved into the worker. auto worker = det_and_inverse_worker::type::value_type>>(std::forward(a)); return worker.det(); } } namespace clef { TRIQS_CLEF_MAKE_FNT_LAZY(determinant); } } // namespace triqs::arrays #endif