Get PetscVector::type() from Vec sizes, not type#4402
Conversation
Otherwise we think that e.g. a VECKOKKOS in parallel should be SERIAL.
|
This isn't just a theoretical problem - see https://civet.inl.gov/job/3585537/ for some example failures. Our callbacks from PETSc SNES operate by using this constructor, swapping the constructed |
|
Damn. This isn't a fix for us yet. I get to |
PETSc doesn't even want to let us call these functions on e.g. a Kokkos vector without throwing an error, so we'll hardcode for now that other vectors just can't be ghosted.
|
Yea that's https://gitlab.com/petsc/petsc/-/issues/1096. It's one of the things we're hoping to have the PETSc team work on under Genesis ties |
|
Job Coverage, step Generate coverage on 52745d4 wanted to post the following: CoverageNo coverage report(s) were found for the base commit 20eccfe. This comment will be updated on new commits. |
| Vec localrep; | ||
| LibmeshPetscCall(VecGhostGetLocalForm(_vec,&localrep)); |
There was a problem hiding this comment.
Was this just an unused variable previously?
There was a problem hiding this comment.
Nah; I just moved it (to get the declaration and thus scope out of the new if()), despite the diff looking like I removed it.
|
For the record, in my own testing that last commit was enough to fix the |
|
When are you anticipating next submodule update? |
|
We're in master as of yesterday afternoon, and IMHO ready to go now. @loganharbour said yesterday he'd "need an update for some python formatting stuff I did" and I figured I'd let him combine us with that, but if he's too swamped and the python update isn't urgent then I'll get a PR up myself this afternoon. |
|
We should be getting this downstream in idaholab/moose#32379; cross your fingers. |
5 participants
Otherwise we think that e.g. a VECKOKKOS in parallel should be SERIAL.