Enable MatMatSolve(F,RHS,RHS) for some packages
Requested-by: Jaysaval, Piyoosh <Piyoosh.Jaysaval@pnnl.gov>

Bug fix: computing MG complexity had integer overflow. Would only occur for very large problems

Also fixed code that computes parallel matrix storage information to communicate with PetscLogDouble i

Bug fix: computing MG complexity had integer overflow. Would only occur for very large problems

Also fixed code that computes parallel matrix storage information to communicate with PetscLogDouble instead of PetscReal.

Commit-type: bug-fix
Reported-by: Mark Lohry <mlohry@gmail.com>

show more ...

Propogate MUMPS return code within omp_comm. Also fixed the datatype and count in MPI_Bcast

Add error checking in mumps.c
Reported-by: Barry Smith

Merge branch 'maint'

Rename valid_GPU_array/matrix to offloadmask

Double semicolons

Support for MatMatMult with Nest and MPIDense

Add dummy initialize values and change code organization slightly to prevent incorrect compiler warnings about unitialized values

Fix a int that should be a size_t, fix handling of long

Add additio

Add dummy initialize values and change code organization slightly to prevent incorrect compiler warnings about unitialized values

Fix a int that should be a size_t, fix handling of long

Add additional details for PetscDataType to prevent possible confusion

Commit-type: portability-fix
Reported-by: Satish Balay <balay@mcs.anl.gov>
Thanks-to: Jed Brown <jed@jedbrown.org>

show more ...

ad hoc fix for cuda bug in mat-transpose-mult

bugfix for matmatmult and mattransposematmult for mpiaij and mpidense
Contributed-by: Pierre Jolivet <pierre.jolivet@enseeiht.fr>

bugfix for MatMatMultNumeric_MPIDense() which allows user-provided d

bugfix for matmatmult and mattransposematmult for mpiaij and mpidense
Contributed-by: Pierre Jolivet <pierre.jolivet@enseeiht.fr>

bugfix for MatMatMultNumeric_MPIDense() which allows user-provided dense matrix
Reported-by: Jose E. Roman <jroman@dsic.upv.es>

small fixes reported by prj- <gitlab@mg.gitlab.com>

fix typo; check matrix local size of user-provided C in MatMatMultSymbolic_MPIAIJ_MPIDense()

comm -> PETSC_COMM_SELF for local error checking.

convert dos format to unix; add LDA to help menu.

show more ...

Merge branch 'rmills/fix-aijmkl-help' into 'master'

Changed MATAIJMKL help messages to be more informative.

See merge request petsc/petsc!2094

Changed MATAIJMKL help messages to be more informative.

CUDA: Fixed compilation warning obtained with CUDA 10.1 and GCC 8.

CUDA: Fixed incorrect use of separate streams.

This solves synchronization problems that have arisen due to the incorrect use of multiple CUDA streams for vector and matrix operations (without using

CUDA: Fixed incorrect use of separate streams.

This solves synchronization problems that have arisen due to the incorrect use of multiple CUDA streams for vector and matrix operations (without using proper synchronization mechanisms).
In particular, SNES ex19 and ex56 now run reliably (no failure after 20+ reruns).
Instead, the default stream (NULL pointer) is now used for all CUDA operations.
I don't have performance comparisons at hand for the performance implications in this commit, but expect any changes to be small.
Correctness first :-)

show more ...

Fix compilation error for nvcc in optimized code with AVX-512 (march=native on my GPU workstation)

for some reason, the host compiler fails with this error message
/home/zampins/Devel/petsc/include/

Fix compilation error for nvcc in optimized code with AVX-512 (march=native on my GPU workstation)

for some reason, the host compiler fails with this error message
/home/zampins/Devel/petsc/include/../src/mat/impls/aij/seq/aij.h(535): error: identifier "_mm512_reduce_add_pd" is undefined

This optimized C kernel is not used in the GPU classes, so it is safe to skip its declaration

show more ...

MATMPIAIJ: need to implement MatPinToCPU for derived GPU classes

MatMatMult: support for {seqaijviennacl|seqaijcusparse}_seqdense

MATSEQAJIJCUSPARSE: multiple fixes

- Use MatMatSolve_Basic for multiple solves instead of the CPU version (need to write support for multiple solves from cusparse)
- when using Cholesky, column perm

MATSEQAJIJCUSPARSE: multiple fixes

- Use MatMatSolve_Basic for multiple solves instead of the CPU version (need to write support for multiple solves from cusparse)
- when using Cholesky, column permutation must be inverted
- remove unneeded extra copy in MatSolve_
- the class does not implement PinToCPU, so remove usage of the flag

show more ...

MATMKLPARDISO: improvements for GPU classes

Use accessor routines for SeqAIJ base classes
Schur complement need to set the valid flag to GPU
MatGetInfo needs to return size of the factors
Added log

MATMKLPARDISO: improvements for GPU classes

Use accessor routines for SeqAIJ base classes
Schur complement need to set the valid flag to GPU
MatGetInfo needs to return size of the factors
Added log flops for factorization phase

show more ...

MATSOLVERMUMPS: use accessors routines

this allows to use MUMPS with up-to-date VIENNACL matrices

MatPermute_SeqAIJ: propagate PinToCpu flag

MatSeqAIJ{Get|Restore}ArrayRead: first implementation

This allows to do few things, waiting for Karl to write the wrapper library...

MatCreateSubMatrix_SeqAIJ: propagate pinning information to submatrix

MATSEQAIJVIENNACL: implement MatSeqAIJGetArray