devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

add support for vectorized conjugated products

Browse Source
This commit is contained in:
Gael Guennebaud
2010-07-06 19:10:24 +02:00
parent 291fef5760
commit d6454788d9
12 changed files with 246 additions and 193 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
Eigen/src/Jacobi/Jacobi.h
View File

@@ -324,7 +324,7 @@ void /*EIGEN_DONT_INLINE*/ ei_apply_rotation_in_the_plane(VectorX& _x, VectorY&
const Packet pc = ei_pset1(Scalar(j.c()));
const Packet ps = ei_pset1(Scalar(j.s()));
ei_conj_helper<NumTraits<Scalar>::IsComplex,false> cj;
ei_conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
for(Index i=0; i<alignedStart; ++i)
{
@@ -343,7 +343,7 @@ void /*EIGEN_DONT_INLINE*/ ei_apply_rotation_in_the_plane(VectorX& _x, VectorY&
{
Packet xi = ei_pload(px);
Packet yi = ei_pload(py);
ei_pstore(px, ei_padd(ei_pmul(pc,xi),cj.pmul(ps,yi)));
ei_pstore(px, ei_padd(ei_pmul(pc,xi),pcj.pmul(ps,yi)));
ei_pstore(py, ei_psub(ei_pmul(pc,yi),ei_pmul(ps,xi)));
px += PacketSize;
py += PacketSize;
@@ -358,8 +358,8 @@ void /*EIGEN_DONT_INLINE*/ ei_apply_rotation_in_the_plane(VectorX& _x, VectorY&
Packet xi1 = ei_ploadu(px+PacketSize);
Packet yi = ei_pload (py);
Packet yi1 = ei_pload (py+PacketSize);
ei_pstoreu(px, ei_padd(ei_pmul(pc,xi),cj.pmul(ps,yi)));
ei_pstoreu(px+PacketSize, ei_padd(ei_pmul(pc,xi1),cj.pmul(ps,yi1)));
ei_pstoreu(px, ei_padd(ei_pmul(pc,xi),pcj.pmul(ps,yi)));
ei_pstoreu(px+PacketSize, ei_padd(ei_pmul(pc,xi1),pcj.pmul(ps,yi1)));
ei_pstore (py, ei_psub(ei_pmul(pc,yi),ei_pmul(ps,xi)));
ei_pstore (py+PacketSize, ei_psub(ei_pmul(pc,yi1),ei_pmul(ps,xi1)));
px += Peeling*PacketSize;
@@ -369,7 +369,7 @@ void /*EIGEN_DONT_INLINE*/ ei_apply_rotation_in_the_plane(VectorX& _x, VectorY&
{
Packet xi = ei_ploadu(x+peelingEnd);
Packet yi = ei_pload (y+peelingEnd);
ei_pstoreu(x+peelingEnd, ei_padd(ei_pmul(pc,xi),cj.pmul(ps,yi)));
ei_pstoreu(x+peelingEnd, ei_padd(ei_pmul(pc,xi),pcj.pmul(ps,yi)));
ei_pstore (y+peelingEnd, ei_psub(ei_pmul(pc,yi),ei_pmul(ps,xi)));
}
}