[petsc-users] Regarding the status of VecSetValues(Blocked) for GPU vectors

[Mark Adams]

I saw "Mat" also ... we don't have device set values for vectors, I am
pretty sure. But it is not hard to do for setting local data.
I don't see an easy way to do off-processor data ... we have not gotten to
this.

See
https://petsc.org/main/docs/manualpages/Vec/VecGetArrayWriteAndMemType.html

You get a pointer to the raw data and pass that down to your device
functions. If it is a device vector it will give you the device data.

You then add to it yourself in your kernel.

On Thu, Mar 17, 2022 at 5:44 PM Jed Brown <jed at jedbrown.org> wrote:

> The question is about vectors. I think it will work, but haven't tested.
>
> Barry Smith <bsmith at petsc.dev> writes:
>
> >   We seem to be emphasizing using MatSetValuesCOO() for GPUs (can also
> be for CPUs); in the main branch you can find a simple example in
> src/mat/tutorials/ex18.c which demonstrates its use.
> >
> >   Barry
> >
> >
> >> On Mar 17, 2022, at 4:46 PM, Sajid Ali Syed <sasyed at fnal.gov> wrote:
> >>
> >> Hi PETSc-developers,
> >>
> >> Is it possible to use VecSetValues with distributed-memory CUDA &
> Kokkos vectors from the device, i.e. can I call VecSetValues with GPU
> memory pointers and expect PETSc to figure out how to stash on the device
> it until I call VecAssemblyBegin (at which point PETSc could use GPU-aware
> MPI to populate off-process values) ?
> >>
> >> If this is not currently supported, is supporting this on the roadmap?
> Thanks in advance!
> >>
> >> Thank You,
> >> Sajid Ali (he/him) | Research Associate
> >> Scientific Computing Division
> >> Fermi National Accelerator Laboratory
> >> s-sajid-ali.github.io <http://s-sajid-ali.github.io/>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220317/0d7171d0/attachment.html>