Robert K authored 2 years ago and Andreas Dedner committed 2 years ago

use using and SizeType to size_type to adhere to istl conventions

See https://reuse.software/ for a description.

This makes the naming consistent again.

This is the name and semantics used by BCRSMatrix.
As BCCSMatrix is currently only used with scalar blocks
anyway, changing the meaning of the nonzeroes method
doesn't make a difference.

It is nothing but a thin wrapper around a matrix.  This patch removes
the MatrixRowSet class and hands over matrices directly.  This
simplifies the code a bit.  More importantly, it paves the way
for setup of BCCSMatrix objects (needed by the SuiteSparse solvers)
from MultiTypeBlockMatrix objects.

The classes ColCompMatrix and ColCompMatrixInitializer were never
intended to be interface.  Let's make this explicit by including
them in the namespace Impl.  In fact, this patch moves them into
Dune::ISTL::Impl, to avoid potential name clashes with other modules.

Previously, the first template parameter was the matrix type that
the BCCSMatrix was a copy of.  Weird, and very closely tied to
the fact that BCCSMatrix/ColCompMatrix is currently only used
to get matrix data into SuiteSparse solvers.

This patch makes the first template parameter the type of
matrix entries.  This is what all other dune-istl matrices do,
and hence the code again gets easier to understand.

Remove operator= and setMatrix methods.  The code of these methods
is copied to all locations were operator= or setMatrix were
previously used.  This makes the calling code a bit longer,
but it simplifies the overall code structure considerably.

In particular, with these changes the BCCRSMatrix class does not
have 'virtual' methods anymore.

Several members of `std::allocator` were already deprecated in C++17
and have now been removed in C++20. This patch uses the appropriate
members of `std::allocator_traits` instead.

with LoopSIMD

solver or preconditioner

Christian Engwer authored 5 years ago and Nils-Arne Dreier committed 5 years ago

the issue is that suitesparse solvers can only be used with real_type
double. We now use a SFINAE to inect a runtime exception into the factory,
which throws an error when trying to create a suitesparse solver over
anything except double.

THe index type used for UMFPack was recently switched to long int
by @kilian.weishaupt. However, the specialization of the method
dispatcher for std::complex was not adapted.

Kilian Weishaupt authored 5 years ago and Christoph Grüninger committed 5 years ago

With the *_dl_* versions instead of the *_di_* versions UMFPACK will
not have a memory limit of just 2GB.

is selected. Since UMFPack does not support this, it cannot be used in this
case.

Marco Agnese authored 10 years ago and Christoph Grüninger committed 9 years ago

Cherry-picked changes
from 7de2a9a70ef3523f0d3c88097d6eab237d51e910
and c3775bc0253b3e1d6dd63d7944a1b554c7f232f3
Removed one occurence more (gruenich).

Signed-off-by: Christoph Grüninger <gruenich@dune-project.org>

Marco Agnese authored 10 years ago and Christoph Grüninger committed 9 years ago

Cherry-picked changes, left out white-space changes,
from 857fdd91dcea7af2ab5f1c1147a59a187cb253db
and 8c21e316e4b72b38972b0c1a93254171566979c6

Signed-off-by: Christoph Grüninger <gruenich@dune-project.org>

Marco Agnese authored 10 years ago and Christoph Grüninger committed 9 years ago

Cherry-picked changes, left out white-space changes
and variable renaming, from 36cebfb6d30703a9db4253e4ca4f7d7f3b4e92d3

Signed-off-by: Christoph Grüninger <gruenich@dune-project.org>

This patch fixes the errornous commits
e8d16384
1a2f4d0f

The introduced range checks were plain wring, as they neglected the nested structure of the
BlockVector. We now use `b.dim()` to get the flat size.

This fixes flyspray/FS#1724.

This avoids strange bugs when using the setSubMatrix method: that method
automatically removes rows and columns from the matrix, and expects _the user_
to do the same for the x and b vectors.

The elapsed wallclock time is readily available by UMFPack
and can be written to InverseOperatorResult.elapsed!

Arne Morten Kvarving authored 11 years ago and Markus Blatt committed 11 years ago

Since revision 8df1d996 overlappingschwarz
with UmfPack was broken due to renaming an internal variable accessed by a friend.
This patch fixes this by adding a new private method to access the internal matrix
of solver and uses that for the assembly.