Plex: Now DMPlexMarkBoundaryFaces() can determine the marker from associated vertices

args: * ${PETSC_DIR} * need to use ${wPETSC_DIR}

On Window's, the -f argument needs to use windows style
paths even though the configure line uses a cygwin style path.

Reset the random seed for each partition that PTScotch may perform so as to get consistent results.

Note this does not resolve the reason for the difference between PetscPartitioner and MatPartition

Reset the random seed for each partition that PTScotch may perform so as to get consistent results.

Note this does not resolve the reason for the difference between PetscPartitioner and MatPartitioning

I found the problem! In plexpartition.c (line 1750) there is this horrible special case where if it determines a single process owns the entire graph (in this case MPI_Comm_rank 0) it calls a different sequential partitioner instead of the parallel preconditioner. MatPartitioning does not handle such special cases differently than others so, of course, uses the parallel partitioner and generates a different partition. @Matthew Knepley We could introduce this special case into the MatPartitioning for PTScotch and thus reproduce the PetscPartitioner results or we could say that handling the special case is goofy and live with the difference. Or we could remove the special case from PetscPartitioner and get identical results for both.

Commit-type: portability-fix, testing-fix

show more ...

try to reset the random seed for second run to get common results for PTScotch

this did not work! Used debugger and confirmed that seed is being reset.

no idea why the results are different.

Commi

try to reset the random seed for second run to get common results for PTScotch

this did not work! Used debugger and confirmed that seed is being reset.

no idea why the results are different.

Commit-type: portability-fix, testing-fix

show more ...

dm_impls_plex_tests-ex24_3 and its output updated

ex24: set type for repartitioning in the same way

dm_impls_plex_tests-ex24_4 small fix

Chaco fails in test on size 3,4, and 8

Commit-type: testing-fix

make -f gmakefile test globsearch="dm*impls*plex*tests*ex24_4"

produces same results with chaco, parametis or ptscotch

Commit-type: feature

Add MatMPIAdjToSeq() that creates a copy of the MPIAdj matrix on each process

ex24.c with the Chaco option crashes with this even though it appears the sequantial matrices are dorrect. Possible caus

Add MatMPIAdjToSeq() that creates a copy of the MPIAdj matrix on each process

ex24.c with the Chaco option crashes with this even though it appears the sequantial matrices are dorrect. Possible causes

Chaco needs columns sorted in matrix?
Results from Chaco are not properly put back into parallel form?

Commit-type: feature

show more ...

fix warning

fix for triangle that requires double input regardless of PETSc precision

Commit-type: bug-fix, portability-fix
Reported-by: nightly test

TODO test for chaco

Plex test 24: group prefixes, more comments, all parallel

Plex test 24: elaborated test 6

PetscPartitionerSetType: fix stale methods on type change

add forgot file to branch

Commit-type: bug-fix
Reported-by: travis

Converting DMPlex grid generators and refiners to registration process

Commit-type: feature
Time: 6 hours

Plex test 24: prefix dp2_ was set to wrong object

deleted src/dm/impls/plex/petscpartmatpart_snes_tests

Plex test 24: better handling of -faces

Plex test 24: TODO test for ptscotch

Revert plexadapt changes in "example requires hypre, why was it not listed?"

This partially reverts commit d2cd29b77a0e01176b6af98c18aba4d1aaafbb3e.

example requires hypre, why was it not listed?

Commit-type: bug-fix, portability-fix, testing-fix

Fix outputs after rebasing