GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-11 19:29:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

rename indices to a common scheme

Browse Source
This commit is contained in:
Gael Guennebaud 2010-07-22 13:17:39 +02:00
parent 0916d69ca5
commit bec3f9bfe4
1 changed files with 11 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
Eigen/src/Core/products/GeneralMatrixVector.h
View File

@ -91,7 +91,7 @@ EIGEN_DONT_INLINE static void run(
const Index ResPacketAlignedMask = ResPacketSize-1; const Index ResPacketAlignedMask = ResPacketSize-1;
const Index PeelAlignedMask = ResPacketSize*peels-1; const Index PeelAlignedMask = ResPacketSize*peels-1;
const Index size = rows; const Index size = rows;
// How many coeffs of the result do we have to skip to be aligned. // How many coeffs of the result do we have to skip to be aligned.
// Here we assume data are at least aligned on the base scalar type. // Here we assume data are at least aligned on the base scalar type.
Index alignedStart = ei_first_aligned(res,size); Index alignedStart = ei_first_aligned(res,size);
@ -244,30 +244,29 @@ EIGEN_DONT_INLINE static void run(
Index start = columnBound; Index start = columnBound;
do do
{ {
for (Index i=start; i<end; ++i) for (Index k=start; k<end; ++k)
{ {
RhsPacket ptmp0 = ei_pset1<RhsPacket>(alpha*rhs[i*rhsIncr]); RhsPacket ptmp0 = ei_pset1<RhsPacket>(alpha*rhs[k*rhsIncr]);
const LhsScalar* lhs0 = lhs + i*lhsStride; const LhsScalar* lhs0 = lhs + k*lhsStride;
if (Vectorizable) if (Vectorizable)
{ {
/* explicit vectorization */ /* explicit vectorization */
// process first unaligned result's coeffs // process first unaligned result's coeffs
for (Index j=0; j<alignedStart; ++j) for (Index j=0; j<alignedStart; ++j)
res[j] += cj.pmul(lhs0[j], ei_pfirst(ptmp0)); res[j] += cj.pmul(lhs0[j], rhs[k*rhsIncr]/*ei_pfirst(ptmp0)*/);
// process aligned result's coeffs // process aligned result's coeffs
if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0) if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0)
for (Index j = alignedStart;j<alignedSize;j+=ResPacketSize) for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
ei_pstore(&res[j], pcj.pmadd(ei_pload<LhsPacket>(&lhs0[j]), ptmp0, ei_pload<ResPacket>(&res[j]))); ei_pstore(&res[i], pcj.pmadd(ei_ploadu<LhsPacket>(&lhs0[i]), ptmp0, ei_pload<ResPacket>(&res[i])));
else else
for (Index j = alignedStart;j<alignedSize;j+=ResPacketSize) for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
ei_pstore(&res[j], pcj.pmadd(ei_ploadu<LhsPacket>(&lhs0[j]), ptmp0, ei_pload<ResPacket>(&res[j]))); ei_pstore(&res[i], pcj.pmadd(ei_ploadu<LhsPacket>(&lhs0[i]), ptmp0, ei_pload<ResPacket>(&res[i])));
} }
// process remaining scalars (or all if no explicit vectorization) // process remaining scalars (or all if no explicit vectorization)
for (Index j=alignedSize; j<size; ++j) for (Index i=alignedSize; i<size; ++i)
res[j] += cj.pmul(lhs0[j], ei_pfirst(ptmp0)); res[i] += cj.pmul(lhs0[i], ei_pfirst(ptmp0));
} }
if (skipColumns) if (skipColumns)
{ {