Overview

Request 1221877 accepted

- update to version 2.10.0
* Improve testing support on Laplacian matrices with an optional
diagonal regularization parameter.
* Base the implementation of VariableBlockVector on std::vector
as the storage type. Note that this prevents from using bool
as block type that was possible before.
* A method BCRSMatrix::setIndicesNoSort() was added. Similar
to BCRSMatrix::setIndices() this allows to insert all indices
of a row at once, but - in contrast to the latter - does not
sort them. This allows to speed up insertion if indices are
already sorted.
* UMFPACK is extended to arbitrary blocked matrix structures.
This includes MultiTypeBlockMatrix. The external interface
is unchanged.
* The internal storage in MatrixIndexSet was changed from
std::set to a sorted std::vector to improve performance. The
stored index type was changed from std::size_t to uint32_t to
reduce memory consumption and improve performance. Hence,
MatrixIndexSet can no longer be used for very large matrices
with more than 2^32 columns.
* Added flag 'useFixedOrder' to the coarsen method of AMGs
ParallelIndicesCoarsener. If set to true, during the creation
of the coarser matrix (by accumulation and restriction
to fewer processes) the coarse matrix rows are ordered in a
fixed manner, making parallel runs reproducible; the runtime
is possibly not ideal though.

Loading...
Request History
Christoph G's avatar

mathletic created request

- update to version 2.10.0
* Improve testing support on Laplacian matrices with an optional
diagonal regularization parameter.
* Base the implementation of VariableBlockVector on std::vector
as the storage type. Note that this prevents from using bool
as block type that was possible before.
* A method BCRSMatrix::setIndicesNoSort() was added. Similar
to BCRSMatrix::setIndices() this allows to insert all indices
of a row at once, but - in contrast to the latter - does not
sort them. This allows to speed up insertion if indices are
already sorted.
* UMFPACK is extended to arbitrary blocked matrix structures.
This includes MultiTypeBlockMatrix. The external interface
is unchanged.
* The internal storage in MatrixIndexSet was changed from
std::set to a sorted std::vector to improve performance. The
stored index type was changed from std::size_t to uint32_t to
reduce memory consumption and improve performance. Hence,
MatrixIndexSet can no longer be used for very large matrices
with more than 2^32 columns.
* Added flag 'useFixedOrder' to the coarsen method of AMGs
ParallelIndicesCoarsener. If set to true, during the creation
of the coarser matrix (by accumulation and restriction
to fewer processes) the coarse matrix rows are ordered in a
fixed manner, making parallel runs reproducible; the runtime
is possibly not ideal though.


Dmitry Roshchin's avatar

Dmitry_R accepted request

openSUSE Build Service is sponsored by