Thanks, Barry and Jeff! I will pass the word on about MPI-4. I was at the
last FASTMath meeting and I do recall that you were at least wondering whether
a shared memory model was needed, but I didn’t realize that you had ripped all
of the threading code out of PETSc. For what it’s worth, I
Rob:
You might be interested in the MPI-4 endpoints proposal. It should allow
MPI communicators to act as thread communicators.
Jeff
On Tue, Jan 12, 2016 at 12:33 PM, Barry Smith wrote:
>
> Rob,
>
> I guess you missed the last FASTMath meeting. We have stopped all
Rob,
I guess you missed the last FASTMath meeting. We have stopped all work on
using threads a long time ago because we never saw an improvement in
performance and it was a pain to maintain. We even ripped the thread code out
of PETSc so currently we are back to being an all "MPI"