void ColStridedColumnPack
( Int height,
Int colAlign, Int colStride,
const T* A,
T* BPortions, Int portionSize )
{
for( Int k=0; k<colStride; ++k )
{
const Int colShift = Shift_( k, colAlign, colStride );
const Int localHeight = Length_( height, colShift, colStride );
StridedMemCopy
( &BPortions[k*portionSize], 1,
&A[colShift], colStride, localHeight );
}
}
void InterleaveMatrix
( Int height, Int width,
const T* A, Int colStrideA, Int rowStrideA,
T* B, Int colStrideB, Int rowStrideB )
{
// TODO: Add OpenMP parallelization and/or optimize
if( colStrideA == 1 && colStrideB == 1 )
{
lapack::Copy( 'F', height, width, A, rowStrideA, B, rowStrideB );
}
else
{
for( Int j=0; j<width; ++j )
StridedMemCopy
( &B[j*rowStrideB], colStrideB,
&A[j*rowStrideA], colStrideA, height );
}
}
void PartialColStridedColumnUnpack
( Int height,
Int colAlign, Int colStride,
Int colStrideUnion, Int colStridePart, Int colRankPart,
Int colShiftB,
const T* APortions, Int portionSize,
T* B )
{
for( Int k=0; k<colStrideUnion; ++k )
{
const Int colShift =
Shift_( colRankPart+k*colStridePart, colAlign, colStride );
const Int colOffset = (colShift-colShiftB) / colStridePart;
const Int localHeight = Length_( height, colShift, colStride );
StridedMemCopy
( &B[colOffset], colStrideUnion,
&APortions[k*portionSize], 1, localHeight );
}
}
void InterleaveMatrix
( Int height, Int width,
const T* A, Int colStrideA, Int rowStrideA,
T* B, Int colStrideB, Int rowStrideB )
{
if( colStrideA == 1 && colStrideB == 1 )
{
lapack::Copy( 'F', height, width, A, rowStrideA, B, rowStrideB );
}
else
{
#ifdef EL_HAVE_MKL
mkl::omatcopy
( NORMAL, height, width, T(1),
A, rowStrideA, colStrideA,
B, rowStrideB, colStrideB );
#else
for( Int j=0; j<width; ++j )
StridedMemCopy
( &B[j*rowStrideB], colStrideB,
&A[j*rowStrideA], colStrideA, height );
#endif
}
}
