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

21
Eigen/src/Core/products/GeneralMatrixVector.h
View File

@ -244,30 +244,29 @@ EIGEN_DONT_INLINE static void run(
Index start = columnBound;
do
{
for (Index i=start; i<end; ++i)
for (Index k=start; k<end; ++k)
{
RhsPacket ptmp0 = ei_pset1<RhsPacket>(alpha*rhs[i*rhsIncr]);
const LhsScalar* lhs0 = lhs + i*lhsStride;
RhsPacket ptmp0 = ei_pset1<RhsPacket>(alpha*rhs[k*rhsIncr]);
const LhsScalar* lhs0 = lhs + k*lhsStride;
if (Vectorizable)
{
/* explicit vectorization */
// process first unaligned result's coeffs
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
if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0)
for (Index j = alignedStart;j<alignedSize;j+=ResPacketSize)
ei_pstore(&res[j], pcj.pmadd(ei_pload<LhsPacket>(&lhs0[j]), ptmp0, ei_pload<ResPacket>(&res[j])));
for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
ei_pstore(&res[i], pcj.pmadd(ei_ploadu<LhsPacket>(&lhs0[i]), ptmp0, ei_pload<ResPacket>(&res[i])));
else
for (Index j = alignedStart;j<alignedSize;j+=ResPacketSize)
ei_pstore(&res[j], pcj.pmadd(ei_ploadu<LhsPacket>(&lhs0[j]), ptmp0, ei_pload<ResPacket>(&res[j])));
for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
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)
for (Index j=alignedSize; j<size; ++j)
res[j] += cj.pmul(lhs0[j], ei_pfirst(ptmp0));
for (Index i=alignedSize; i<size; ++i)
res[i] += cj.pmul(lhs0[i], ei_pfirst(ptmp0));
}
if (skipColumns)
{