GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 11:49:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Made sure that the version of gemm_pack_rhs specialized for row major matrices is vectorized when nr == 2*PacketSize (which is the case for SSE when compiling in 64bit mode).

Browse Source
This commit is contained in:
Benoit Steiner 2014-03-26 17:35:18 -07:00
parent e45a6bed45
commit 14bc4b9704
1 changed files with 6 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@ -1091,7 +1091,11 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
if (nr == PacketSize) { if (nr == PacketSize) {
Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]); Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
pstoreu(blockB+count, cj.pconj(A)); pstoreu(blockB+count, cj.pconj(A));
count += PacketSize; } else if (nr == 2*PacketSize) {
Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
Packet B = ploadu<Packet>(&rhs[k*rhsStride + j2 + PacketSize]);
pstoreu(blockB+count, cj.pconj(A));
pstoreu(blockB+count+PacketSize, cj.pconj(B));
} else { } else {
const Scalar* b0 = &rhs[k*rhsStride + j2]; const Scalar* b0 = &rhs[k*rhsStride + j2];
blockB[count+0] = cj(b0[0]); blockB[count+0] = cj(b0[0]);
@ -1102,8 +1106,8 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
if(nr>=8) blockB[count+5] = cj(b0[5]); if(nr>=8) blockB[count+5] = cj(b0[5]);
if(nr>=8) blockB[count+6] = cj(b0[6]); if(nr>=8) blockB[count+6] = cj(b0[6]);
if(nr>=8) blockB[count+7] = cj(b0[7]); if(nr>=8) blockB[count+7] = cj(b0[7]);
count += nr;
} }
count += nr;
} }
// skip what we have after // skip what we have after
if(PanelMode) count += nr * (stride-offset-depth); if(PanelMode) count += nr * (stride-offset-depth);