SF: need to sync the stream before MPI send even when there is nothing to send but has something to receive

There might be pending gpu operations on the receive buffer. Without synchronization, say

SF: need to sync the stream before MPI send even when there is nothing to send but has something to receive

There might be pending gpu operations on the receive buffer. Without synchronization, say we proceed to MPI_Waitall(). MPI might stage on host and do a H2D copy on an internal stream on the receive side. Previous gpu operations COULD happen after the H2D copy, causing a write-after-write reorder violation!

show more ...

VecNestSetSubVec(): reference the input vector rather than explicitly duplicating it

This is consistent with what is done in VecCreateNest()

Update sowing bfort for automatic generation of XXXDestroy() code

Fix some Fortran stub code to work with generation of XXXDestroy()

Add some runtime tests for XXXCreate() calls

Also update sowing

Update sowing bfort for automatic generation of XXXDestroy() code

Fix some Fortran stub code to work with generation of XXXDestroy()

Add some runtime tests for XXXCreate() calls

Also update sowing and fortranimpl.h to use PETSC_NULLPTR instead of NULL

show more ...

Add some error checking for use of PETSC_NULL_XXX in Fortran interface

Minor cleanup of Fortran error messages

Vec: make lock available in optimized mode

Having the lock available is important to take runtime decisions
on how to export the vector data in python

Merge branch 'stefanozampini/vecis-collective' into 'main'

VecISXXX: require same communicator between Vecs and IS

See merge request petsc/petsc!7564

Merge remote-tracking branch 'origin/release'

Merge branch 'jolivet/fix-docs' into 'release'

Docs: fix double backticks in seealso

See merge request petsc/petsc!7560

VecISXXX: require same communicator between Vecs and IS

this allows to skip operations if the global size of the IS is zero

Docs: fix double backticks in seealso

Remove unneeded include <petsc/private/fortranimpl.h>

Update to sowing version that supports PetscViewer and PetscObject

Remove manual Fortran stubs for PetscViewer and PetscObject arguments since now handled by sowing

Also handles the passing of null

Update to sowing version that supports PetscViewer and PetscObject

Remove manual Fortran stubs for PetscViewer and PetscObject arguments since now handled by sowing

Also handles the passing of null objects correctly

show more ...

Merge remote-tracking branch 'origin/release'

VecKokkos: inherit the origin's MDot setting in VecDuplicateVecs

So that the duplicated vectors can use GEMV for VecMDot

Vec: use smaller alignment in VecDuplicateVecs GEMV

cupm: fix visibility to build without warnings

Thanks-to: Lawrence Mitchell

Merge remote-tracking branch 'origin/release'

Add more details on Vec/MatGetOwnershipRange() and how the values are determined.

Reported-by: Sayan Bhattacharjee

Merge branch 'barry/2024-04-13/fix-sowing-strings' into 'main'

Update PETSc to use sowing bfort that handles string arguments and PETSC_NULL_INT

See merge request petsc/petsc!7480

Merge remote-tracking branch 'origin/release'

docs fixes for release

Fix for defined but unused variables under certain configurations

Update source code removing all unneeded /*@C and associated manual stubs and interfaces

Merge remote-tracking branch 'origin/release'

Fix unused variable message from certain compiler/configuration situations.

Plus this violates the style rule about flop variables.