aboutsummaryrefslogtreecommitdiffhomepage
path: root/Eigen/src/Core/products/TriangularMatrixMatrix.h
diff options
context:
space:
mode:
Diffstat (limited to 'Eigen/src/Core/products/TriangularMatrixMatrix.h')
-rw-r--r--Eigen/src/Core/products/TriangularMatrixMatrix.h54
1 files changed, 30 insertions, 24 deletions
diff --git a/Eigen/src/Core/products/TriangularMatrixMatrix.h b/Eigen/src/Core/products/TriangularMatrixMatrix.h
index 85fd744b9..f0c60507a 100644
--- a/Eigen/src/Core/products/TriangularMatrixMatrix.h
+++ b/Eigen/src/Core/products/TriangularMatrixMatrix.h
@@ -45,22 +45,24 @@ template <typename Scalar, typename Index,
int Mode, bool LhsIsTriangular,
int LhsStorageOrder, bool ConjugateLhs,
int RhsStorageOrder, bool ConjugateRhs,
- int ResStorageOrder, int Version = Specialized>
+ int ResStorageOrder, int ResInnerStride,
+ int Version = Specialized>
struct product_triangular_matrix_matrix;
template <typename Scalar, typename Index,
int Mode, bool LhsIsTriangular,
int LhsStorageOrder, bool ConjugateLhs,
- int RhsStorageOrder, bool ConjugateRhs, int Version>
+ int RhsStorageOrder, bool ConjugateRhs,
+ int ResInnerStride, int Version>
struct product_triangular_matrix_matrix<Scalar,Index,Mode,LhsIsTriangular,
LhsStorageOrder,ConjugateLhs,
- RhsStorageOrder,ConjugateRhs,RowMajor,Version>
+ RhsStorageOrder,ConjugateRhs,RowMajor,ResInnerStride,Version>
{
static EIGEN_STRONG_INLINE void run(
Index rows, Index cols, Index depth,
const Scalar* lhs, Index lhsStride,
const Scalar* rhs, Index rhsStride,
- Scalar* res, Index resStride,
+ Scalar* res, Index resIncr, Index resStride,
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
{
product_triangular_matrix_matrix<Scalar, Index,
@@ -70,18 +72,19 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,LhsIsTriangular,
ConjugateRhs,
LhsStorageOrder==RowMajor ? ColMajor : RowMajor,
ConjugateLhs,
- ColMajor>
- ::run(cols, rows, depth, rhs, rhsStride, lhs, lhsStride, res, resStride, alpha, blocking);
+ ColMajor, ResInnerStride>
+ ::run(cols, rows, depth, rhs, rhsStride, lhs, lhsStride, res, resIncr, resStride, alpha, blocking);
}
};
// implements col-major += alpha * op(triangular) * op(general)
template <typename Scalar, typename Index, int Mode,
int LhsStorageOrder, bool ConjugateLhs,
- int RhsStorageOrder, bool ConjugateRhs, int Version>
+ int RhsStorageOrder, bool ConjugateRhs,
+ int ResInnerStride, int Version>
struct product_triangular_matrix_matrix<Scalar,Index,Mode,true,
LhsStorageOrder,ConjugateLhs,
- RhsStorageOrder,ConjugateRhs,ColMajor,Version>
+ RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>
{
typedef gebp_traits<Scalar,Scalar> Traits;
@@ -95,20 +98,21 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,true,
Index _rows, Index _cols, Index _depth,
const Scalar* _lhs, Index lhsStride,
const Scalar* _rhs, Index rhsStride,
- Scalar* res, Index resStride,
+ Scalar* res, Index resIncr, Index resStride,
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
};
template <typename Scalar, typename Index, int Mode,
int LhsStorageOrder, bool ConjugateLhs,
- int RhsStorageOrder, bool ConjugateRhs, int Version>
+ int RhsStorageOrder, bool ConjugateRhs,
+ int ResInnerStride, int Version>
EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
LhsStorageOrder,ConjugateLhs,
- RhsStorageOrder,ConjugateRhs,ColMajor,Version>::run(
+ RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>::run(
Index _rows, Index _cols, Index _depth,
const Scalar* _lhs, Index lhsStride,
const Scalar* _rhs, Index rhsStride,
- Scalar* _res, Index resStride,
+ Scalar* _res, Index resIncr, Index resStride,
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
{
// strip zeros
@@ -119,10 +123,10 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
- typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+ typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
LhsMapper lhs(_lhs,lhsStride);
RhsMapper rhs(_rhs,rhsStride);
- ResMapper res(_res, resStride);
+ ResMapper res(_res, resStride, resIncr);
Index kc = blocking.kc(); // cache block size along the K direction
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
@@ -235,10 +239,11 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
// implements col-major += alpha * op(general) * op(triangular)
template <typename Scalar, typename Index, int Mode,
int LhsStorageOrder, bool ConjugateLhs,
- int RhsStorageOrder, bool ConjugateRhs, int Version>
+ int RhsStorageOrder, bool ConjugateRhs,
+ int ResInnerStride, int Version>
struct product_triangular_matrix_matrix<Scalar,Index,Mode,false,
LhsStorageOrder,ConjugateLhs,
- RhsStorageOrder,ConjugateRhs,ColMajor,Version>
+ RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>
{
typedef gebp_traits<Scalar,Scalar> Traits;
enum {
@@ -251,20 +256,21 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,false,
Index _rows, Index _cols, Index _depth,
const Scalar* _lhs, Index lhsStride,
const Scalar* _rhs, Index rhsStride,
- Scalar* res, Index resStride,
+ Scalar* res, Index resIncr, Index resStride,
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
};
template <typename Scalar, typename Index, int Mode,
int LhsStorageOrder, bool ConjugateLhs,
- int RhsStorageOrder, bool ConjugateRhs, int Version>
+ int RhsStorageOrder, bool ConjugateRhs,
+ int ResInnerStride, int Version>
EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
LhsStorageOrder,ConjugateLhs,
- RhsStorageOrder,ConjugateRhs,ColMajor,Version>::run(
+ RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>::run(
Index _rows, Index _cols, Index _depth,
const Scalar* _lhs, Index lhsStride,
const Scalar* _rhs, Index rhsStride,
- Scalar* _res, Index resStride,
+ Scalar* _res, Index resIncr, Index resStride,
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
{
const Index PacketBytes = packet_traits<Scalar>::size*sizeof(Scalar);
@@ -276,10 +282,10 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
- typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+ typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
LhsMapper lhs(_lhs,lhsStride);
RhsMapper rhs(_rhs,rhsStride);
- ResMapper res(_res, resStride);
+ ResMapper res(_res, resStride, resIncr);
Index kc = blocking.kc(); // cache block size along the K direction
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
@@ -433,12 +439,12 @@ struct triangular_product_impl<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
Mode, LhsIsTriangular,
(internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
(internal::traits<ActualRhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
- (internal::traits<Dest >::Flags&RowMajorBit) ? RowMajor : ColMajor>
+ (internal::traits<Dest >::Flags&RowMajorBit) ? RowMajor : ColMajor, Dest::InnerStrideAtCompileTime>
::run(
stripedRows, stripedCols, stripedDepth, // sizes
&lhs.coeffRef(0,0), lhs.outerStride(), // lhs info
&rhs.coeffRef(0,0), rhs.outerStride(), // rhs info
- &dst.coeffRef(0,0), dst.outerStride(), // result info
+ &dst.coeffRef(0,0), dst.innerStride(), dst.outerStride(), // result info
actualAlpha, blocking
);
generated by cgit v1.2.3 (git 2.39.1) at 2024-05-23 11:06:34 +0000