mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert vec/y to vec*(1/y) in row-major TRSM:

Browse Source 
- div is extremely costly
- this is consistent with the column-major case
- this is consistent with all other BLAS implementations
This commit is contained in:
Gael Guennebaud 2016-12-06 15:04:50 +01:00
parent 8365c2c941
commit eb621413c1
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
Eigen/src/Core/products/TriangularSolverMatrix.h
View File

@ -183,7 +183,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
} }
} }
/* Optimized triangular solver with multiple left hand sides and the trinagular matrix on the right /* Optimized triangular solver with multiple left hand sides and the triangular matrix on the right
*/ */
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder> template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor> struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>
@ -202,6 +202,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
level3_blocking<Scalar,Scalar>& blocking) level3_blocking<Scalar,Scalar>& blocking)
{ {
Index rows = otherSize; Index rows = otherSize;
typedef typename NumTraits<Scalar>::Real RealScalar;
typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper; typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper;
typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper; typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper;
@ -306,9 +307,9 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
} }
if((Mode & UnitDiag)==0) if((Mode & UnitDiag)==0)
{ {
Scalar b = conj(rhs(j,j)); Scalar inv_rjj = RealScalar(1)/conj(rhs(j,j));
for (Index i=0; i<actual_mc; ++i) for (Index i=0; i<actual_mc; ++i)
r[i] /= b; r[i] *= inv_rjj;
} }
} }