mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-01-24 14:45:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix support of Intel's VML

Browse Source
This commit is contained in:
Gael Guennebaud 2016-07-06 14:07:32 +02:00
parent 8ec4d6480d
commit c3b23d7dbf
1 changed files with 14 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
Eigen/src/Core/Assign_MKL.h Normal file → Executable file
View File

@ -81,10 +81,10 @@ class vml_assign_traits
#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE) \
template< typename DstXprType, typename SrcXprNested> \
struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE>, \
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml,EIGENTYPE>::type> { \
struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>, \
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> { \
typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType; \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE> &/*func*/) { \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &/*func*/) { \
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \
if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) { \
VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(), \
@ -138,22 +138,24 @@ EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor, _)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil, Ceil, _)
#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE) \
template< typename DstXprType, typename SrcXprNested> \
struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE>, \
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml,EIGENTYPE>::type> { \
typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType; \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE> &/*func*/) { \
template< typename DstXprType, typename SrcXprNested, typename Plain> \
struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested, \
const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>, \
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> { \
typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested, \
const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType; \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &/*func*/) { \
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \
VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.functor().m_exponent); \
VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other); \
if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) \
{ \
VMLOP( dst.size(), (const VMLTYPE*)src.nestedExpression().data(), exponent, \
VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent, \
(VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) ); \
} else { \
const Index outerSize = dst.outerSize(); \
for(Index outer = 0; outer < outerSize; ++outer) { \
const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) : \
&(src.nestedExpression().coeffRef(0, outer)); \
const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) : \
&(src.lhs().coeffRef(0, outer)); \
EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer)); \
VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent, \
(VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE)); \