Jonathan Youett authored 10 years ago and Markus Blatt committed 10 years ago

When assigning a BCRSMatrix to another BCRSMatrix which was setup for
a different size/pattern, it segfaulted and I got the following
Exception:

"InvalidStateException [allocate:.../dune/istl/bcrsmatrix.hh:2108]:
Rows have already been allocated, cannot allocate a second time."

The problem was, that inside the assignment operator the rows (r) were
never deallocated. This commit fixes this and adds a testcase.

Testcase slightly modified by Markus.

This mode is not implemented and therefore throws
an assertion that makes the test fail everytime.

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

Namely we try to document the template parameters in more detail and
make sure that cross reference to OwnerOverlapCopyCommunication,
vectors, and matrix implemetation appear.

Fix broken handling of complex problems in MINRES and GMRes. With these
patches, both solvers now support complex matrices as well as complex
right-hand sides. Due to a couple of bugs, the convergence rate of
complex problems was off even before the current breakage.

* feature/fs1467-fix-gmres-and-minres-for-complex:
  [Documentation] Add file comment and formatting hints to complexdata.hh
  [complexmatrixtest] Rename assemble_random_matrix_vectors.hh -> complexdata.hh
  [CMake] Properly exclude header from headercheck.
  [Buildsystem] Ignore helper header for complexmatrixtest
  [Bugfix] GMRes now needs enable_if, so include common/typetraits.hh
  [Tests] Move tests to exercise MINRES and GMRes on complex matrices into separate executable
  [Cleanup][Tests] Rearranged the automatically parsed code for the assembling of the random matrices and random vectors in assemble_random_matrix_vectors.hh for better readability
  [Autotools] added new header file in the Makefile.am
  [Tests] added tests using complex nonsingular matrices and complex hermitian matrices to test the MinRes and GMRes solver on purely complex problems
  [Bugfix] Exploit analytical properties in MINRES to avoid special-casing for complex case
  [Bugfix] fixed order of the scalar product when applying the Arnoldi algorithm in GMRes solver, which is important in the complex case
  [Bugfix] implemented complex Givens rotations and fixed MinRes and GMRes for complex valued problems

As Christoph pointed out in FS1467, that's a rather unusual name for a
DUNE header file, so make it a little more consistent.

Add newline at and of file.

That header is not supposed to be used on its own, so it makes no sense
to check whether that is possible.

The recently added tests for complex matrices added to complexrhstest
made that file really large and unwieldy, so this patch moves those
tests to a new file complexmatrixtest.

[Cleanup][Tests] Rearranged the automatically parsed code for the assembling of the random matrices and random vectors in assemble_random_matrix_vectors.hh for better readability

[Tests] added tests using complex nonsingular matrices and complex hermitian matrices to test the MinRes and GMRes solver on purely complex problems

[Bugfix] fixed order of the scalar product when applying the Arnoldi algorithm in GMRes solver, which is important in the complex case

minor cleanups in the MinRes solver

[Bugfix] implemented complex Givens rotations and fixed MinRes and GMRes for complex valued problems

The type idxtype is used by header repartition even if parmetis is not
used. Therefore it has to be typedefed meaningful in this case,
too. This was not the case before and is accomplished by this commit.

This was introduced for debugging only and could have side effects
onto other headers. Removes the accidentally committed line.

* feature/fix-gmres-and-minres:
  [GMRes] Fix deprecation warning for old constructors
  made the distinction between field_type and real_type which is the same for real numbers but differs for std::complex
  added the old constructors called with the variable recalc_defect to RestartedGMResSolver and set them as deprecated, since they should not be used anymore
  changed the initialization of the arrays in the MinRes solver to remove the compiler warnings
  fixed MinRes with the following changes:
  fixed GMRes

Conflicts:
	dune/istl/solvers.hh

Function attribute have to be placed in front of the function name,
not behind it.

Marian Piatkowski authored 10 years ago and Steffen Müthing committed 10 years ago

made the distinction between field_type and real_type which is the same for real numbers but differs for std::complex

Marian Piatkowski authored 10 years ago and Steffen Müthing committed 10 years ago

added the old constructors called with the variable recalc_defect to RestartedGMResSolver and set them as deprecated, since they should not be used anymore

Marian Piatkowski authored 11 years ago and Steffen Müthing committed 10 years ago

replaced the usual arrays for the Givens-rotation coefficients and the search directions etc. with the arrays from Dune

implemented the BLAS-routine drotg for greater robustness which was a TODO note to compute the Givens rotation

Marian Piatkowski authored 11 years ago and Steffen Müthing committed 10 years ago

removed bool recalc_defect
reimplement the update function with respect to optimality
fixed strange behaviour when maxiter is reached and GMRes wants to restart infinitely many times
fixed strange behaviour when linear reduction is reached but GMRes wants to restart and continue calculation
fixed the printing of the results

Complained by autoconf 1.14, cf. FS#1462.

This reverts commit 901b5954.
It (METIS) does not seem needed for compilation and the patch
forced the user and downstream modules to add additional compile
 and linker flags.

This reverts commit 6132d0df.
It (METIS) does not seem needed for compilation and the patch
forced the user and downstream modules to add additional compile
 and linker flags.

Unlike the original METIS/ParMETIS scotch does not use extern "C"
in its headers. This lead to linker errors as descibed in
http://lists.dune-project.org/pipermail/dune/2014-March/012745.html.
This patch fixes this by guarding the includes with an extra extern "C"
directive.

Just like for the autotools.

For HAVE_METIS we do not use the ENABLE trick. It is therefore either defined or
not. Therfore using "#if HAVE_METIS" is a bug. This patch fixes this.

There were problems with repartition.hh when using the original parmetis. The cause was
that for scotch we need to require both parmetis.h and metis.h, which broke backwards
compatibility for the original parmetis (which worked without metis, too). With the patch
we check whether parmetis.h is from scotch and if that is the case also require metis. If not,
metis suffices.