[mpich-discuss] How expensive is MPI_Win_create() compared to memcpy()?

[Jed Brown]

On Sun, Jan 1, 2012 at 22:16, Dave Goodell <goodell at mcs.anl.gov> wrote:

> Jed and I previously talked about this offline.  I don't think that
> dynamic windows are a good match for his problem because all that it would
> do is shift the burden of determining buffer address starting offsets from
> the MPI implementation to his code.


Right, so long as my code would have to know the memory address of the
newly attached buffer on the remote process.

The ideal thing from my perspective would be to re-seat a window with new
memory and have the new base address be communicated (if necessary) during
the next MPI_Win_post/start. That seems like something that should be
possible without minimal overhead relative to reusing existing memory, but
maybe I'm missing some details of the implementation.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20120101/88e3eeb1/attachment.htm>