coll/basic: fix MPI_Alltoallw(MPI_IN_PLACE) gap handling #9330
Conversation
The temporary buffer must be shifted by the true_extent on a per type basis (since the various datatypes might have different true_extent). Thanks Heiko Bauke for reporting this. Refs. open-mpi#9329 Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
|
bot:ibm:retest |
|
bot:ibm:retest |
This function can be used to compute the packed size of a datatype on a target architecture. Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Dont copy the datatype into a buffer with the same extent, but instead pack it and send it to the peer as packed. Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
|
A more memory-aware approach. I still don't like the double loop in the call, but I haven't had enough time to figure out a better solution yet. Once we agree we should apply the same patch to all in_place functions. |
bosilca
force-pushed
the
topic/a2aw_inplace
branch
from
173d94f
to
da90df7
Compare
Provide optimized variant for the homogeneous case. Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
bosilca
force-pushed
the
topic/a2aw_inplace
branch
from
da90df7
to
b9012a3
Compare
|
@jsquyres and @bwbarrett we need this in the 5.0 |
|
@bosilca Is there a reason you don't review/approve this PR? You're far more qualified to review it than me or @bwbarrett... |
|
@bosilca Never mind; I see you added commits beyond @ggouaillardet's original commits. Let me at least prove that this PR fixes the code posted in the original #9329 issue. |
|
@bosilca I tried to cherry pick these commits to v5.0.x and failed -- there's some conflicts. Are there some fixes that should be over on v5.0.x that aren't there (such that these fixes would apply cleanly)? |
The temporary buffer must be shifted by the true_extent on a
per type basis (since the various datatypes might have different
true_extent).
Thanks Heiko Bauke for reporting this.
Refs. #9329
Signed-off-by: Gilles Gouaillardet gilles@rist.or.jp