CUDA: PetscCUBLASGetHandle() is now public.

Required to access to CUBLAS handle in user code.

Reported-by: Jose Roman <jroman@dsic.upv.es>
Alejandro Lamas Daviña <alejandro.lamas@dsic.

CUDA: PetscCUBLASGetHandle() is now public.

Required to access to CUBLAS handle in user code.

Reported-by: Jose Roman <jroman@dsic.upv.es>
Alejandro Lamas Daviña <alejandro.lamas@dsic.upv.es>

Fix provided by Alejandro. Thanks!

show more ...

ISSTRIDE: fix bug in ISView for MATLAB format

fix complex problem

Commit-type: bug-fix
Reported-by: nightly tests

Refactorization of MATSHELL now supports matrix free version of ts ex50.c

which uses composition of matrix free matrices.

Commit-type: feature

update tests; small fix for complex build

Fix issues reported by nightly builds

I really don't know why the testing bddc code was included in the commit

xlf90: compile error

"/gpfs5r/workgpfs/idris/sos/ssos455/TuringTests/F_Alet/petsc/src/vec/f90-mod/petscvecmod.F", line 155.47: 1513-213 (S) A derived-type object that appears as a dummy argument in

xlf90: compile error

"/gpfs5r/workgpfs/idris/sos/ssos455/TuringTests/F_Alet/petsc/src/vec/f90-mod/petscvecmod.F", line 155.47: 1513-213 (S) A derived-type object that appears as a dummy argument in an external procedu\
re definition must be of a sequence type or a type with the BIND(C) attribute, unless it is of a type that is use-associated.

Reported-by: Serge Van Criekingen <serge.van.criekingen@idris.fr>

show more ...

VecScatter: reorganization

implement several vecnode routines
rm mpi3 from _p_VecScatter struct -- nolonger useful
rm using native shared memory from src/ksp/ksp/examples/tests/ex56.c

VecScatter: reorganization

implement several vecnode routines
rm mpi3 from _p_VecScatter struct -- nolonger useful
rm using native shared memory from src/ksp/ksp/examples/tests/ex56.c
fix memory bug in VecScatterCreateLocal_PtoS_MPI3()
bugfix for VecScatterCopy() -- copy scattertype
add flg is_duplicate to _p_VecScatter; use it to prevent using StoP in VecScatterEndMPI3Node()
rename VecScatterCreateGeneral_ -> VecScatterCreate_
rename VecScatterCreate_vectype_mpitype() -> VecScatterCreateLocal_vectype_mpitype()
check for duplicate indices when mpi3node scatter is created; update tests
undo the commit bb8026e126c43792cc81c094ed5610e9a92280e3: add an example for testing MatMult() with node vectors
add MPI_Barrier(mscomm) in VecScatterEndMPI3Node for safe, but would lose correctness due to wright to same variable (see run2_5 and run3_5). Until finding an efficient way to deal with variable racing, we may give up this work :-(
add test output files
bugfix for VecScatterEndMPI3Node; update ex2.c
Merge branch 'hzhang/vec_node/master-rebased' of bitbucket.org:petsc/petsc into hzhang/vec_node/master-rebased
add an example for testing MatMult() with node vectors
bugfix for VecScatterEndMPI3Node()
rm checking vec_type in VecScatterCreate_mpi3node()
add winarray to Vec_node struct to avoid calls to MPI_Win_shared_query() in scatterEnd()
optionize a loop for xv -> yv
small bugfix and optimization
implement several Vec routines for node type
using object state counter for vector to rm 2 MPI_Barrier()
small optimization and cleanup
small build fix
add few vector routines for node type
add VecScatterCreate_vectype_private() for cleanup
replace MPI_Barrier(comm) by MPI_Barrier(mscomm)
add tests for StoP scatter
bugfix: src/vec/vscat/examples/ex1 and ex2 are working well now
update ex1 and ex2
reverse scatter PtoS works for bs=1
cleanup; fixed memory bug for bs>1
PtoP and PtoS(forward) scatter work for bs>1
mpi3node works for PtoP
forward scatter, I can read on-core xvalues in VecScatterEndMPI3Node
check vectypes for VecScatterCreate_MPI3Node()
add VecScatterCreate_MPI2Node() -- not done yet
rename VecScatterCreate_StoS() -> xxx_Seq(); add VecScatterCreate_type(); reorganize
cleanup and organization -- all seems working well
cleanup; update VecScatterSetFromOptions()
add VecScatterSetFromOptions()
rm VecScatterCreate_MPI1(); cleanup
cleanup VecScatterCreateGeneral_StoP()
add VecScatterCreateGeneral_StoP()
cleanup VecScatterCreateGeneral_PtoS()
VecScatterCreateGeneral_PtoS() works
bugfix after rebase -- passed local tests for both mpi1 and mpi3

add VecScatterCreate_StoS()
add GetInputISType_private()
update src/vec/vscat/examples/ex1.c and its output
new VecScatterCreate() as an interface function
add src/vec/vscat/interface/dlregisvecscat.c
mv vecscatter routines from src/vec/vec/ to src/vec/vscat/
add src/vec/vscat/interface/vscreate.c
reorganize files: create src/vec/vscat
small fix for mmaij.c
add VecScatterType
rename VecScatterType -> VecScatterFormat
add new VecType VECNODE for creating vector over on-node shared memory

show more ...

Merge branch 'barry/scalable-assembly'

Merge branch 'karlrupp/feature-lazy-cublas-init'

IS: added MATLAB format to ISView_{General|Stride|Block}

PetscSection: doc "seealso" fix

need custom Fortran stubs for these static inline functions

Commit-type: bug-fix
Reported-by: nightly tests

Cannot use _BTS Vec and Mat assembly with MPIUni

Commit-type: portability-fix
Reported-by: nightly tests

Fix faulty logic for turning on VecAssembly legacy

Commit-type: bug-fix
Reported-by: nightly tests

Small fix for documentation

Work towards making BTS version of Vec and Mat Assembly the default

Commit-type: style-fix
Reported-by: Fande Kong <fande.kong@inl.gov>

CUSP: Fixed access to CUBLAS handle.

This duplicates a bit of code from VECCUDA.
However, as VECCUSP will be removed soon, this is acceptable.

Merge branch 'jed/fortran-seterr'

Merge branch 'barry/add-fortran-custom-interfaces'

Merge branch 'maint'

Updates for viewing Load Balance information

Don't display for sequential objects. Label all values with "Load Balance"
Want to display load balance for petsc ordering for vectors, not using natural

Updates for viewing Load Balance information

Don't display for sequential objects. Label all values with "Load Balance"
Want to display load balance for petsc ordering for vectors, not using natural ordering

Commit-type: feature

show more ...

CUDA: Internal CUBLAS management functions are now static.

CUDA: Encapsulated formerly non-namespaced global cublasv2handle.

Commit-type: bug-fix

Add viewer format object for displaying load imbalance of vectors and nonzeros in sparse MPIAIJ matrices

Could be extended to DMs include DMDA and DMPlex

Commit-type: feature
Thanks-to: Fande Kong

Add viewer format object for displaying load imbalance of vectors and nonzeros in sparse MPIAIJ matrices

Could be extended to DMs include DMDA and DMPlex

Commit-type: feature
Thanks-to: Fande Kong <fande.kong@inl.gov>

show more ...