[mpiwg-rma] discuss MPI_WIN_SHARED_QUERY

[Jeff Hammond]

I would like to discuss MPI_WIN_SHARED_QUERY (
https://github.com/mpi-forum/mpi-issues/issues/23) in a future call.  In
the past, Bill had issues with this because it would encourage (but not
require) implementations to allocate contiguous shared memory in
MPI_Win_allocate, whereas today it might allocate noncontiguous shared
memory.  I'm not sure if any implementation is actually doing the latter,
but I believe this is not a serious issue, since (1) implementations are
not required to change (2) Linux NUMA balancing is a thing now.

I will also note that I started sketching out a proposal to construct
windows better - https://github.com/mpiwg-rma/rma-issues/issues/22 - that
might be a way to escape a lot of the current problems with RMA having way
too many features.

Jeff
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-rma/attachments/20220704/6152b0ca/attachment.html>