mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-03-31 19:00:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use vectorization when packing row-major rhs matrices. (bug #717)

Browse Source
This commit is contained in:
Benoit Steiner 2013-12-17 10:49:43 -08:00
parent 033ee7f6d9
commit ce99b502ce
1 changed files with 13 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@ -1261,6 +1261,7 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, Pan
template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
struct gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
{
typedef typename packet_traits<Scalar>::type Packet;
enum { PacketSize = packet_traits<Scalar>::size };
EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
};
@ -1282,12 +1283,18 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
if(PanelMode) count += nr * offset;
for(Index k=0; k<depth; k++)
{
const Scalar* b0 = &rhs[k*rhsStride + j2];
blockB[count+0] = cj(b0[0]);
blockB[count+1] = cj(b0[1]);
if(nr==4) blockB[count+2] = cj(b0[2]);
if(nr==4) blockB[count+3] = cj(b0[3]);
count += nr;
if (nr == PacketSize) {
Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
pstoreu(blockB+count, cj.pconj(A));
count += PacketSize;
} else {
const Scalar* b0 = &rhs[k*rhsStride + j2];
blockB[count+0] = cj(b0[0]);
blockB[count+1] = cj(b0[1]);
if(nr==4) blockB[count+2] = cj(b0[2]);
if(nr==4) blockB[count+3] = cj(b0[3]);
count += nr;
}
}
// skip what we have after
if(PanelMode) count += nr * (stride-offset-depth);