GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-05-22 12:37:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix nesting sub-expression in outer-products

Browse Source
This commit is contained in:
Gael Guennebaud 2015-10-08 21:41:53 +02:00
parent 4140ee039d
commit c9718514f5
2 changed files with 18 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
Eigen/src/Core/ProductEvaluators.h
View File

@ -218,12 +218,12 @@ template<typename Dst, typename Lhs, typename Rhs, typename Func>
EIGEN_DONT_INLINE void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&) EIGEN_DONT_INLINE void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&)
{ {
evaluator<Rhs> rhsEval(rhs); evaluator<Rhs> rhsEval(rhs);
// FIXME make sure lhs is sequentially stored typename nested_eval<Lhs,Rhs::SizeAtCompileTime>::type actual_lhs(lhs);
// FIXME if cols is large enough, then it might be useful to make sure that lhs is sequentially stored
// FIXME not very good if rhs is real and lhs complex while alpha is real too // FIXME not very good if rhs is real and lhs complex while alpha is real too
// FIXME we should probably build an evaluator for dst
const Index cols = dst.cols(); const Index cols = dst.cols();
for (Index j=0; j<cols; ++j) for (Index j=0; j<cols; ++j)
func(dst.col(j), rhsEval.coeff(0,j) * lhs); func(dst.col(j), rhsEval.coeff(0,j) * actual_lhs);
} }
// Row major result // Row major result
@ -231,12 +231,12 @@ template<typename Dst, typename Lhs, typename Rhs, typename Func>
EIGEN_DONT_INLINE void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&) EIGEN_DONT_INLINE void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&)
{ {
evaluator<Lhs> lhsEval(lhs); evaluator<Lhs> lhsEval(lhs);
// FIXME make sure rhs is sequentially stored typename nested_eval<Rhs,Lhs::SizeAtCompileTime>::type actual_rhs(rhs);
// FIXME if rows is large enough, then it might be useful to make sure that rhs is sequentially stored
// FIXME not very good if lhs is real and rhs complex while alpha is real too // FIXME not very good if lhs is real and rhs complex while alpha is real too
// FIXME we should probably build an evaluator for dst
const Index rows = dst.rows(); const Index rows = dst.rows();
for (Index i=0; i<rows; ++i) for (Index i=0; i<rows; ++i)
func(dst.row(i), lhsEval.coeff(i,0) * rhs); func(dst.row(i), lhsEval.coeff(i,0) * actual_rhs);
} }
template<typename Lhs, typename Rhs> template<typename Lhs, typename Rhs>

13
test/product_notemporary.cpp
View File

@ -62,7 +62,7 @@ template<typename MatrixType> void product_notemporary(const MatrixType& m)
VERIFY_EVALUATION_COUNT( m3.noalias() -= (s1 * m1).template triangularView<Lower>() * m2, 0); VERIFY_EVALUATION_COUNT( m3.noalias() -= (s1 * m1).template triangularView<Lower>() * m2, 0);
VERIFY_EVALUATION_COUNT( rm3.noalias() = (s1 * m1.adjoint()).template triangularView<Upper>() * (m2+m2), 1); VERIFY_EVALUATION_COUNT( rm3.noalias() = (s1 * m1.adjoint()).template triangularView<Upper>() * (m2+m2), 1);
VERIFY_EVALUATION_COUNT( rm3.noalias() = (s1 * m1.adjoint()).template triangularView<UnitUpper>() * m2.adjoint(), 0); VERIFY_EVALUATION_COUNT( rm3.noalias() = (s1 * m1.adjoint()).template triangularView<UnitUpper>() * m2.adjoint(), 0);
VERIFY_EVALUATION_COUNT( m3.template triangularView<Upper>() = (m1 * m2.adjoint()), 0); VERIFY_EVALUATION_COUNT( m3.template triangularView<Upper>() = (m1 * m2.adjoint()), 0);
VERIFY_EVALUATION_COUNT( m3.template triangularView<Upper>() -= (m1 * m2.adjoint()), 0); VERIFY_EVALUATION_COUNT( m3.template triangularView<Upper>() -= (m1 * m2.adjoint()), 0);
@ -112,6 +112,17 @@ template<typename MatrixType> void product_notemporary(const MatrixType& m)
VERIFY_EVALUATION_COUNT( cvres.noalias() = (rm3+rm3) * cv1, 0 ); VERIFY_EVALUATION_COUNT( cvres.noalias() = (rm3+rm3) * cv1, 0 );
VERIFY_EVALUATION_COUNT( cvres.noalias() = (m1+m1) * (m1*cv1), 1 ); VERIFY_EVALUATION_COUNT( cvres.noalias() = (m1+m1) * (m1*cv1), 1 );
VERIFY_EVALUATION_COUNT( cvres.noalias() = (rm3+rm3) * (m1*cv1), 1 ); VERIFY_EVALUATION_COUNT( cvres.noalias() = (rm3+rm3) * (m1*cv1), 1 );
// Check outer products
m3 = cv1 * rv1;
VERIFY_EVALUATION_COUNT( m3.noalias() = cv1 * rv1, 0 );
VERIFY_EVALUATION_COUNT( m3.noalias() = (cv1+cv1) * (rv1+rv1), 1 );
VERIFY_EVALUATION_COUNT( m3.noalias() = (m1*cv1) * (rv1), 1 );
VERIFY_EVALUATION_COUNT( m3.noalias() += (m1*cv1) * (rv1), 1 );
VERIFY_EVALUATION_COUNT( rm3.noalias() = (cv1) * (rv1 * m1), 1 );
VERIFY_EVALUATION_COUNT( rm3.noalias() -= (cv1) * (rv1 * m1), 1 );
VERIFY_EVALUATION_COUNT( rm3.noalias() = (m1*cv1) * (rv1 * m1), 2 );
VERIFY_EVALUATION_COUNT( rm3.noalias() += (m1*cv1) * (rv1 * m1), 2 );
} }
void test_product_notemporary() void test_product_notemporary()