diff options
author | Gael Guennebaud <g.gael@free.fr> | 2009-08-14 00:17:14 +0200 |
---|---|---|
committer | Gael Guennebaud <g.gael@free.fr> | 2009-08-14 00:17:14 +0200 |
commit | 13e95f7f68f67ee0b66cfa37d9cdc9fdd71eeca8 (patch) | |
tree | 9b6c4291e2b3abcfbc50666fa174ec95cb4d06e2 /Eigen/src/Jacobi | |
parent | f2536416da990f12e98d01806331ad8d78545863 (diff) |
optimize "apply Jacobi" for small sizes, and move it to Jacobi.h
Diffstat (limited to 'Eigen/src/Jacobi')
-rw-r--r-- | Eigen/src/Jacobi/Jacobi.h | 97 |
1 files changed, 97 insertions, 0 deletions
diff --git a/Eigen/src/Jacobi/Jacobi.h b/Eigen/src/Jacobi/Jacobi.h index 40181cd08..b5940c74b 100644 --- a/Eigen/src/Jacobi/Jacobi.h +++ b/Eigen/src/Jacobi/Jacobi.h @@ -2,6 +2,7 @@ // for linear algebra. // // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com> +// Copyright (C) 2009 Gael Guennebaud <g.gael@free.fr> // // Eigen is free software; you can redistribute it and/or // modify it under the terms of the GNU Lesser General Public @@ -25,6 +26,9 @@ #ifndef EIGEN_JACOBI_H #define EIGEN_JACOBI_H +template<typename VectorX, typename VectorY> +void ei_apply_rotation_in_the_plane(VectorX& _x, VectorY& _y, typename VectorX::Scalar c, typename VectorY::Scalar s); + template<typename Derived> inline void MatrixBase<Derived>::applyJacobiOnTheLeft(int p, int q, Scalar c, Scalar s) { @@ -101,5 +105,98 @@ inline void ei_normalizeJacobi(Scalar *c, Scalar *s, const Scalar& x, const Scal } } +template<typename VectorX, typename VectorY> +void /*EIGEN_DONT_INLINE*/ ei_apply_rotation_in_the_plane(VectorX& _x, VectorY& _y, typename VectorX::Scalar c, typename VectorY::Scalar s) +{ + typedef typename VectorX::Scalar Scalar; + ei_assert(_x.size() == _y.size()); + int size = _x.size(); + int incrx = size ==1 ? 1 : &_x.coeffRef(1) - &_x.coeffRef(0); + int incry = size ==1 ? 1 : &_y.coeffRef(1) - &_y.coeffRef(0); + + Scalar* EIGEN_RESTRICT x = &_x.coeffRef(0); + Scalar* EIGEN_RESTRICT y = &_y.coeffRef(0); + + if (incrx==1 && incry==1) + { + // both vectors are sequentially stored in memory => vectorization + typedef typename ei_packet_traits<Scalar>::type Packet; + enum { PacketSize = ei_packet_traits<Scalar>::size, Peeling = 2 }; + + int alignedStart = ei_alignmentOffset(y, size); + int alignedEnd = alignedStart + ((size-alignedStart)/PacketSize)*PacketSize; + + const Packet pc = ei_pset1(c); + const Packet ps = ei_pset1(s); + + for(int i=0; i<alignedStart; ++i) + { + Scalar xi = x[i]; + Scalar yi = y[i]; + x[i] = c * xi - s * yi; + y[i] = s * xi + c * yi; + } + + Scalar* px = x + alignedStart; + Scalar* py = y + alignedStart; + + if(ei_alignmentOffset(x, size)==alignedStart) + { + for(int i=alignedStart; i<alignedEnd; i+=PacketSize) + { + Packet xi = ei_pload(px); + Packet yi = ei_pload(py); + ei_pstore(px, ei_psub(ei_pmul(pc,xi),ei_pmul(ps,yi))); + ei_pstore(py, ei_padd(ei_pmul(ps,xi),ei_pmul(pc,yi))); + px += PacketSize; + py += PacketSize; + } + } + else + { + int peelingEnd = alignedStart + ((size-alignedStart)/(Peeling*PacketSize))*(Peeling*PacketSize); + for(int i=alignedStart; i<peelingEnd; i+=Peeling*PacketSize) + { + Packet xi = ei_ploadu(px); + Packet xi1 = ei_ploadu(px+PacketSize); + Packet yi = ei_pload (py); + Packet yi1 = ei_pload (py+PacketSize); + ei_pstoreu(px, ei_psub(ei_pmul(pc,xi),ei_pmul(ps,yi))); + ei_pstoreu(px+PacketSize, ei_psub(ei_pmul(pc,xi1),ei_pmul(ps,yi1))); + ei_pstore (py, ei_padd(ei_pmul(ps,xi),ei_pmul(pc,yi))); + ei_pstore (py+PacketSize, ei_padd(ei_pmul(ps,xi1),ei_pmul(pc,yi1))); + px += Peeling*PacketSize; + py += Peeling*PacketSize; + } + if(alignedEnd!=peelingEnd) + { + Packet xi = ei_ploadu(x+peelingEnd); + Packet yi = ei_pload (y+peelingEnd); + ei_pstoreu(x+peelingEnd, ei_psub(ei_pmul(pc,xi),ei_pmul(ps,yi))); + ei_pstore (y+peelingEnd, ei_padd(ei_pmul(ps,xi),ei_pmul(pc,yi))); + } + } + + for(int i=alignedEnd; i<size; ++i) + { + Scalar xi = x[i]; + Scalar yi = y[i]; + x[i] = c * xi - s * yi; + y[i] = s * xi + c * yi; + } + } + else + { + for(int i=0; i<size; ++i) + { + Scalar xi = *x; + Scalar yi = *y; + *x = c * xi - s * yi; + *y = s * xi + c * yi; + x += incrx; + y += incry; + } + } +} #endif // EIGEN_JACOBI_H |