GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-06-04 18:54:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

some performance fixes in Assign.h reported by Gael. Some doc update in

Browse Source 
Cwise.
This commit is contained in:
Benoit Jacob 2008-07-10 16:15:55 +00:00
parent 7b4c6b8862
commit 2b53fd4d53
2 changed files with 18 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
Eigen/src/Core/Assign.h
View File

@ -203,15 +203,15 @@ struct ei_assign_impl<Derived1, Derived2, NoVectorization, NoUnrolling>
{ {
static void run(Derived1 &dst, const Derived2 &src) static void run(Derived1 &dst, const Derived2 &src)
{ {
const bool rowMajor = int(Derived1::Flags)&RowMajorBit;
const int innerSize = dst.innerSize(); const int innerSize = dst.innerSize();
const int outerSize = dst.outerSize(); const int outerSize = dst.outerSize();
for(int j = 0; j < outerSize; j++) for(int j = 0; j < outerSize; j++)
for(int i = 0; i < innerSize; i++) for(int i = 0; i < innerSize; i++)
{ {
const int row = rowMajor ? j : i; if(int(Derived1::Flags)&RowMajorBit)
const int col = rowMajor ? i : j; dst.coeffRef(j, i) = src.coeff(j, i);
dst.coeffRef(row, col) = src.coeff(row, col); else
dst.coeffRef(i, j) = src.coeff(i, j);
} }
} }
}; };
@ -254,13 +254,12 @@ struct ei_assign_impl<Derived1, Derived2, InnerVectorization, NoUnrolling>
const int outerSize = dst.outerSize(); const int outerSize = dst.outerSize();
const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size; const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
for(int j = 0; j < outerSize; j++) for(int j = 0; j < outerSize; j++)
{
for(int i = 0; i < innerSize; i+=packetSize) for(int i = 0; i < innerSize; i+=packetSize)
{ {
const int row = rowMajor ? j : i; if(int(Derived1::Flags)&RowMajorBit)
const int col = rowMajor ? i : j; dst.template writePacket<Aligned>(j, i, src.template packet<Aligned>(j, i));
dst.template writePacket<Aligned>(row, col, src.template packet<Aligned>(row, col)); else
} dst.template writePacket<Aligned>(i, j, src.template packet<Aligned>(i, j));
} }
} }
}; };
@ -336,7 +335,6 @@ struct ei_assign_impl<Derived1, Derived2, SliceVectorization, NoUnrolling>
static void run(Derived1 &dst, const Derived2 &src) static void run(Derived1 &dst, const Derived2 &src)
{ {
const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size; const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
const bool rowMajor = Derived1::Flags&RowMajorBit;
const int innerSize = dst.innerSize(); const int innerSize = dst.innerSize();
const int outerSize = dst.outerSize(); const int outerSize = dst.outerSize();
const int alignedInnerSize = (innerSize/packetSize)*packetSize; const int alignedInnerSize = (innerSize/packetSize)*packetSize;
@ -346,17 +344,19 @@ struct ei_assign_impl<Derived1, Derived2, SliceVectorization, NoUnrolling>
// do the vectorizable part of the assignment // do the vectorizable part of the assignment
for (int index = 0; index<alignedInnerSize ; index+=packetSize) for (int index = 0; index<alignedInnerSize ; index+=packetSize)
{ {
const int row = rowMajor ? i : index; if(Derived1::Flags&RowMajorBit)
const int col = rowMajor ? index : i; dst.template writePacket<Unaligned>(i, index, src.template packet<Unaligned>(i, index));
dst.template writePacket<Unaligned>(row, col, src.template packet<Unaligned>(row, col)); else
dst.template writePacket<Unaligned>(index, i, src.template packet<Unaligned>(index, i));
} }
// do the non-vectorizable part of the assignment // do the non-vectorizable part of the assignment
for (int index = alignedInnerSize; index<innerSize ; index++) for (int index = alignedInnerSize; index<innerSize ; index++)
{ {
const int row = rowMajor ? i : index; if(Derived1::Flags&RowMajorBit)
const int col = rowMajor ? index : i; dst.coeffRef(i, index) = src.coeff(i, index);
dst.coeffRef(row, col) = src.coeff(row, col); else
dst.coeffRef(index, i) = src.coeff(index, i);
} }
} }
} }

8
Eigen/src/Core/Cwise.h
View File

@ -144,9 +144,7 @@ template<typename ExpressionType> class Cwise
ExpressionTypeNested m_matrix; ExpressionTypeNested m_matrix;
}; };
/** \array_module /** \returns a Cwise wrapper of *this providing additional coefficient-wise operations
*
* \returns a Cwise expression of *this providing additional coefficient-wise operations
* *
* \sa class Cwise * \sa class Cwise
*/ */
@ -157,9 +155,7 @@ MatrixBase<Derived>::cwise() const
return derived(); return derived();
} }
/** \array_module /** \returns a Cwise wrapper of *this providing additional coefficient-wise operations
*
* \returns a Cwise expression of *this providing additional coefficient-wise operations
* *
* \sa class Cwise * \sa class Cwise
*/ */