GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-30 07:44:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

Ease compiler job to generate clean and efficient code in mat*vec.

Browse Source
This commit is contained in:
Gael Guennebaud 2016-12-02 22:41:26 +01:00
parent fe696022ec
commit 66f65ccc36
1 changed files with 53 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

91
Eigen/src/Core/GeneralProduct.h
View File

@ -224,50 +224,65 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
// on, the other hand it is good for the cache to pack the vector anyways... // on, the other hand it is good for the cache to pack the vector anyways...
EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1), EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex), ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
MightCannotUseDest = (ActualDest::InnerStrideAtCompileTime!=1) || ComplexByReal MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
}; };
gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
evalToDest ? dest.data() : static_dest.data());
if(!evalToDest)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
Index size = dest.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
if(!alphaIsCompatible)
{
MappedDest(actualDestPtr, dest.size()).setZero();
compatibleAlpha = RhsScalar(1);
}
else
MappedDest(actualDestPtr, dest.size()) = dest;
}
typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper; typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper; typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
general_matrix_vector_product RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
<Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
LhsMapper(actualLhs.data(), actualLhs.outerStride()),
RhsMapper(actualRhs.data(), actualRhs.innerStride()),
actualDestPtr, 1,
compatibleAlpha);
if (!evalToDest) if(!MightCannotUseDest)
{ {
if(!alphaIsCompatible) // shortcut if we are sure to be able to use dest directly,
dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size()); // this ease the compiler to generate cleaner and more optimzized code for most common cases
else general_matrix_vector_product
dest = MappedDest(actualDestPtr, dest.size()); <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
LhsMapper(actualLhs.data(), actualLhs.outerStride()),
RhsMapper(actualRhs.data(), actualRhs.innerStride()),
dest.data(), 1,
compatibleAlpha);
}
else
{
gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
evalToDest ? dest.data() : static_dest.data());
if(!evalToDest)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
Index size = dest.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
if(!alphaIsCompatible)
{
MappedDest(actualDestPtr, dest.size()).setZero();
compatibleAlpha = RhsScalar(1);
}
else
MappedDest(actualDestPtr, dest.size()) = dest;
}
general_matrix_vector_product
<Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
LhsMapper(actualLhs.data(), actualLhs.outerStride()),
RhsMapper(actualRhs.data(), actualRhs.innerStride()),
actualDestPtr, 1,
compatibleAlpha);
if (!evalToDest)
{
if(!alphaIsCompatible)
dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
else
dest = MappedDest(actualDestPtr, dest.size());
}
} }
} }
}; };