Document that PetscViewerGetSubViewer() is collective on the outer viewer and fix bugs.Reported-by: Pierre Jolivet
Add missing interface function MatProductGetAlgorithm()
MatNullSpaceTest: fix output
Modified ex88 to test MatMultHermitianTranspose_Shell and MatMultHermitianTransposeAdd_Shell and to test with complex numbers
Add support for MatMultHermitianTranspose with a shell matrix
Merge remote-tracking branch 'origin/release'
Fix bug if MatDuplicate() is called on a MatSetUp() but not MatAssembly matrixReported-by: Benjamin Sturdevant <bsturdev@pppl.gov>Reported-by: Mark Adams <mfadams@lbl.gov>
Fix src/mat/tests/ex183.c to use PetscViewerGetSubViewer() correctlyCommit-type: bug-fix
Add PetscBench with implementation based on HPL
Fix use of offdiagonal to off-diagonal
Mat: support Mat{Shift, DiagonalSet, DiagonalScale} for MATAIJKOKKOS
Remove some unneeded whitespaces
MatAssemblyEnd(): same logic for [S]BAIJ than for AIJ
matrix free in docs and comments should be matrix-free
Mat tests: add HYPRE COO testssupport COO with hacking permutation array of AIJ
MatView() is incorrectly called inside itself when -mat_view is used since some MatView assemble matrices that then get printed in the middle of the view of the previous
PetscInfo: print object type
Add more tests
Add the load-balancing kernel for MatMultAdd_SeqSELL and fine tune the heuristicKernel7 is significantly slower than kernel9x for the following two cases:- nrows is too small. Kernel7 uses 2 threa
Add the load-balancing kernel for MatMultAdd_SeqSELL and fine tune the heuristicKernel7 is significantly slower than kernel9x for the following two cases:- nrows is too small. Kernel7 uses 2 threads per row (assuming sliceheight=16), it does not fully utilize the GPU if nrows < 100K.- maxslicewidth is too big.Thanks-to: Peng Wang <penwang@nvidia.com>
show more ...
Fix sanitizer errors
Add support for Mat to allow skipping preallocation by having the first round of MatSetValues use hash* MATPREALLOCATOR can be deprecated as well as possibly some other preallocation macros/routine
Add support for Mat to allow skipping preallocation by having the first round of MatSetValues use hash* MATPREALLOCATOR can be deprecated as well as possibly some other preallocation macros/routines* Optimizations such as direct support for MatSetValuesBlocked() for *BAIJ can be added in the futureCommit-type: feature/spend 25h
SBAIJ should display entire matrix if common formatShould not just display upper triangular/spend 5m
Do not name an unnamed object unless necessary while viewing matricesCommit-type: bug, usability/spend 5m
123456