mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-27 07:29:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

optimize LU_kernel_bmod for small cases, and add an important .noalias()

Browse Source
This commit is contained in:
Gael Guennebaud 2012-07-29 22:26:00 +02:00
parent ce30d50e3e
commit 8f6d5eacb4
3 changed files with 85 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
Eigen/src/SparseLU/SparseLU_column_bmod.h
View File

@ -122,7 +122,10 @@ int LU_column_bmod(const int jcol, const int nseg, BlockScalarVector& dense, Sca
// Perform a triangular solver and block update,
// then scatter the result of sup-col update to dense
no_zeros = kfnz - fst_col;
LU_kernel_bmod(segsize, dense, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
if(segsize==1)
LU_kernel_bmod<1>::run(segsize, dense, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
else
LU_kernel_bmod<Dynamic>::run(segsize, dense, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
} // end if jsupno
} // end for each segment

123
Eigen/src/SparseLU/SparseLU_kernel_bmod.h
View File

@ -39,54 +39,85 @@
* \param no_zeros Number of nonzeros elements before the diagonal part of the supernode
* \return 0 on success
*/
template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
int LU_kernel_bmod(const int segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, int& luptr, const int nsupr, const int nrow, IndexVector& lsub, const int lptr, const int no_zeros)
template <int SegSizeAtCompileTime> struct LU_kernel_bmod
{
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[*]
int isub = lptr + no_zeros;
int i, irow;
for (i = 0; i < segsize; i++)
template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
EIGEN_DONT_INLINE static void run(const int segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, int& luptr, const int nsupr, const int nrow, IndexVector& lsub, const int lptr, const int no_zeros)
{
irow = lsub(isub);
tempv(i) = dense(irow);
++isub;
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[*]
int isub = lptr + no_zeros;
int i, 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 += nsupr * no_zeros + no_zeros;
// Form Eigen matrix and vector
Map<Matrix<Scalar,SegSizeAtCompileTime,SegSizeAtCompileTime>, 0, OuterStride<> > A( &(lusup.data()[luptr]), segsize, segsize, OuterStride<>(nsupr) );
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;
Map<Matrix<Scalar,Dynamic,SegSizeAtCompileTime>, 0, OuterStride<> > B( &(lusup.data()[luptr]), nrow, segsize, OuterStride<>(nsupr) );
Map<Matrix<Scalar,Dynamic,1> > l(tempv.data()+segsize, nrow);
if(SegSizeAtCompileTime==2)
l = u(0) * B.col(0) + u(1) * B.col(1);
else if(SegSizeAtCompileTime==3)
l = u(0) * B.col(0) + u(1) * B.col(1) + u(2) * B.col(2);
else
l.noalias() = B * u;
// 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);
}
}
// Dense triangular solve -- start effective triangle
luptr += nsupr * no_zeros + no_zeros;
// Form Eigen matrix and vector
Map<Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > A( &(lusup.data()[luptr]), segsize, segsize, OuterStride<>(nsupr) );
VectorBlock<ScalarVector> u(tempv, 0, segsize);
u = A.template triangularView<UnitLower>().solve(u);
// Dense matrix-vector product y <-- A*x
luptr += segsize;
new (&A) Map<Matrix<Scalar,Dynamic, Dynamic>, 0, OuterStride<> > ( &(lusup.data()[luptr]), nrow, segsize, OuterStride<>(nsupr) );
VectorBlock<ScalarVector> l(tempv, segsize, nrow);
l= A * u;
// Scatter tempv[] into SPA dense[] as a temporary storage
isub = lptr + no_zeros;
for (i = 0; i < segsize; i++)
};
template <> struct LU_kernel_bmod<1>
{
template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
EIGEN_DONT_INLINE static void run(const int /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, int& luptr, const int nsupr, const int nrow, IndexVector& lsub, const int lptr, const int no_zeros)
{
irow = lsub(isub);
dense(irow) = tempv(i);
tempv(i) = Scalar(0.0);
++isub;
typedef typename ScalarVector::Scalar Scalar;
Scalar f = dense(lsub(lptr + no_zeros));
luptr += nsupr * no_zeros + no_zeros + 1;
const Scalar* a(lusup.data() + luptr);
const typename IndexVector::Scalar* irow(lsub.data()+lptr + no_zeros + 1);
int i = 0;
for (; i+1 < nrow; i+=2)
{
int i0 = *(irow++);
int i1 = *(irow++);
Scalar a0 = *(a++);
Scalar a1 = *(a++);
Scalar d0 = dense.coeff(i0);
Scalar d1 = dense.coeff(i1);
d0 -= f*a0;
d1 -= f*a1;
dense.coeffRef(i0) = d0;
dense.coeffRef(i1) = d1;
}
if(i<nrow)
dense.coeffRef(*(irow++)) -= f * *(a++);
}
// Scatter l into SPA dense[]
for (i = 0; i < nrow; i++)
{
irow = lsub(isub);
dense(irow) -= l(i);
l(i) = Scalar(0.0);
++isub;
}
return 0;
}
#endif
};
#endif

5
Eigen/src/SparseLU/SparseLU_panel_bmod.h
View File

@ -118,7 +118,10 @@ void LU_panel_bmod(const int m, const int w, const int jcol, const int nseg, Sca
// Perform a trianglar solve and block update,
// then scatter the result of sup-col update to dense[]
no_zeros = kfnz - fsupc;
LU_kernel_bmod(segsize, dense_col, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
if(segsize==1) LU_kernel_bmod<1>::run(segsize, dense_col, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
else if(segsize==2) LU_kernel_bmod<2>::run(segsize, dense_col, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
else if(segsize==3) LU_kernel_bmod<3>::run(segsize, dense_col, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
else LU_kernel_bmod<Dynamic>::run(segsize, dense_col, tempv, lusup, luptr, nsupr, nrow, lsub, lptr, no_zeros);
} // End for each column in the panel
} // End for each updating supernode