TCoarsen Class Reference

Implements coarsening, e.g. agglomeration of blocked matrices into dense or low-rank matrices if the later use less memory.

#include <TCoarsen.hh>

Public Member Functions
	TCoarsen (const bool arecompress_lr=true, const size_t amax_block_size=50, const size_t amax_dense_size=500)
	constructs coarsening object
virtual	~TCoarsen ()
	destructor
TCoarsen &	recompress_lr (const bool b)
	set recompression of low-rank blocks
TCoarsen &	only_offdiag_lr (const bool b)
	turn on/off low-rank compression of diagonal blocks
TCoarsen &	max_block_size (const size_t n)
	set maximal size of block-matrices, still worth to try coarsening
TCoarsen &	max_dense_size (const size_t n)
	set maximal size of dense matrices, still worth to try coarsening
virtual TMatrix *	rec_coarsen (const uint nprocs, TMatrix *A, const TTruncAcc &acc, TProgressBar *progress=NULL) const
virtual TMatrix *	coarsen (TMatrix *A, const TTruncAcc &acc) const

Protected Member Functions
size_t	byte_size (const TMatrix *A) const
	return "corrected" bytesizes, e.g. in symmetric case, of matrices for internal use

Protected Attributes
bool	_recompress_lr
	recompress low-rank blocks if true
size_t	_max_block_size
	maximal size of block-matrices, still worth to try coarsening
size_t	_max_dense_size
	maximal size of dense matrices, still worth to try coarsening
bool	_only_offdiag_lr
	do not convert diagonal blocks to low-rank if true

Member Function Documentation

virtual TMatrix* coarsen ( TMatrix *  A, const TTruncAcc &  acc  ) const

virtual

coarsen matrix A without recursion, e.g. coarsening of subblocks.

Parameters

A	matrix to coarsen
acc	accuracy to mantain during coarsening

virtual TMatrix* rec_coarsen ( const uint  nprocs, TMatrix *  A, const TTruncAcc &  acc, TProgressBar *  progress = NULL  ) const

virtual

coarsen matrix A with recursion, e.g. first try to coarsen subblocks.

Parameters

nprocs	number of threads to use
A	matrix to coarsen
acc	accuracy to mantain during coarsening
progress	progress bar