MOAB
4.9.3pre
|
#include <SparseLU_kernel_bmod.h>
Static Public Member Functions | |
template<typename BlockScalarVector , typename ScalarVector , typename IndexVector > | |
static EIGEN_DONT_INLINE void | run (const Index segsize, BlockScalarVector &dense, ScalarVector &tempv, ScalarVector &lusup, Index &luptr, const Index lda, const Index nrow, IndexVector &lsub, const Index lptr, const Index no_zeros) |
Performs numeric block updates from a given supernode to a single column. |
Definition at line 17 of file SparseLU_kernel_bmod.h.
EIGEN_DONT_INLINE void Eigen::internal::LU_kernel_bmod< SegSizeAtCompileTime >::run | ( | const Index | segsize, |
BlockScalarVector & | dense, | ||
ScalarVector & | tempv, | ||
ScalarVector & | lusup, | ||
Index & | luptr, | ||
const Index | lda, | ||
const Index | nrow, | ||
IndexVector & | lsub, | ||
const Index | lptr, | ||
const Index | no_zeros | ||
) | [static] |
Performs numeric block updates from a given supernode to a single column.
segsize | Size of the segment (and blocks ) to use for updates | |
[in,out] | dense | Packed values of the original matrix |
tempv | temporary vector to use for updates | |
lusup | array containing the supernodes | |
lda | Leading dimension in the supernode | |
nrow | Number of rows in the rectangular part of the supernode | |
lsub | compressed row subscripts of supernodes | |
lptr | pointer to the first column of the current supernode in lsub | |
no_zeros | Number of nonzeros elements before the diagonal part of the supernode |
Definition at line 39 of file SparseLU_kernel_bmod.h.
{ typedef typename ScalarVector::Scalar Scalar; // First, copy U[*,j] segment from dense(*) to tempv(*) // The result of triangular solve is in tempv[*]; // The result of matric-vector update is in dense[*] Index isub = lptr + no_zeros; Index i; Index irow; for (i = 0; i < ((SegSizeAtCompileTime==Dynamic)?segsize:SegSizeAtCompileTime); i++) { irow = lsub(isub); tempv(i) = dense(irow); ++isub; } // Dense triangular solve -- start effective triangle luptr += lda * no_zeros + no_zeros; // Form Eigen matrix and vector Map<Matrix<Scalar,SegSizeAtCompileTime,SegSizeAtCompileTime, ColMajor>, 0, OuterStride<> > A( &(lusup.data()[luptr]), segsize, segsize, OuterStride<>(lda) ); Map<Matrix<Scalar,SegSizeAtCompileTime,1> > u(tempv.data(), segsize); u = A.template triangularView<UnitLower>().solve(u); // Dense matrix-vector product y <-- B*x luptr += segsize; const Index PacketSize = internal::packet_traits<Scalar>::size; Index ldl = internal::first_multiple(nrow, PacketSize); Map<Matrix<Scalar,Dynamic,SegSizeAtCompileTime, ColMajor>, 0, OuterStride<> > B( &(lusup.data()[luptr]), nrow, segsize, OuterStride<>(lda) ); Index aligned_offset = internal::first_default_aligned(tempv.data()+segsize, PacketSize); Index aligned_with_B_offset = (PacketSize-internal::first_default_aligned(B.data(), PacketSize))%PacketSize; Map<Matrix<Scalar,Dynamic,1>, 0, OuterStride<> > l(tempv.data()+segsize+aligned_offset+aligned_with_B_offset, nrow, OuterStride<>(ldl) ); l.setZero(); internal::sparselu_gemm<Scalar>(l.rows(), l.cols(), B.cols(), B.data(), B.outerStride(), u.data(), u.outerStride(), l.data(), l.outerStride()); // Scatter tempv[] into SPA dense[] as a temporary storage isub = lptr + no_zeros; for (i = 0; i < ((SegSizeAtCompileTime==Dynamic)?segsize:SegSizeAtCompileTime); i++) { irow = lsub(isub++); dense(irow) = tempv(i); } // Scatter l into SPA dense[] for (i = 0; i < nrow; i++) { irow = lsub(isub++); dense(irow) -= l(i); } }