devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

GEMM: catch all scalar-multiple variants when falling-back to a coeff-based product.

Browse Source 
Before only s*A*B was caught which was both inconsistent with GEMM, sub-optimal,
and could even lead to compilation-errors (https://stackoverflow.com/questions/54738495).
This commit is contained in:
Gael Guennebaud
2019-02-18 11:47:54 +01:00
parent ec032ac03b
commit 512b74aaa1
3 changed files with 92 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
Eigen/src/Core/util/BlasUtil.h
View File

@@ -274,7 +274,8 @@ template<typename XprType> struct blas_traits
HasUsableDirectAccess = ( (int(XprType::Flags)&DirectAccessBit)
&& ( bool(XprType::IsVectorAtCompileTime)
|| int(inner_stride_at_compile_time<XprType>::ret) == 1)
) ? 1 : 0
) ? 1 : 0,
HasScalarFactor = false
};
typedef typename conditional<bool(HasUsableDirectAccess),
ExtractType,
@@ -306,6 +307,9 @@ template<typename Scalar, typename NestedXpr, typename Plain>
struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> >
: blas_traits<NestedXpr>
{
enum {
HasScalarFactor = true
};
typedef blas_traits<NestedXpr> Base;
typedef CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> XprType;
typedef typename Base::ExtractType ExtractType;
@@ -317,6 +321,9 @@ template<typename Scalar, typename NestedXpr, typename Plain>
struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > >
: blas_traits<NestedXpr>
{
enum {
HasScalarFactor = true
};
typedef blas_traits<NestedXpr> Base;
typedef CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > XprType;
typedef typename Base::ExtractType ExtractType;
@@ -335,6 +342,9 @@ template<typename Scalar, typename NestedXpr>
struct blas_traits<CwiseUnaryOp<scalar_opposite_op<Scalar>, NestedXpr> >
: blas_traits<NestedXpr>
{
enum {
HasScalarFactor = true
};
typedef blas_traits<NestedXpr> Base;
typedef CwiseUnaryOp<scalar_opposite_op<Scalar>, NestedXpr> XprType;
typedef typename Base::ExtractType ExtractType;
@@ -358,7 +368,7 @@ struct blas_traits<Transpose<NestedXpr> >
typename ExtractType::PlainObject
>::type DirectLinearAccessType;
enum {
IsTransposed = Base::IsTransposed ? 0 : 1
IsTransposed = Base::IsTransposed ? 0 : 1,
};
static inline ExtractType extract(const XprType& x) { return ExtractType(Base::extract(x.nestedExpression())); }
static inline Scalar extractScalarFactor(const XprType& x) { return Base::extractScalarFactor(x.nestedExpression()); }