-
Notifications
You must be signed in to change notification settings - Fork 859
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
osc/rdma: Fix MPI_Win_start()/complete() with MPI_GROUP_EMPTY. #8779
Conversation
@hjelmn ping - can you take a quick look when you have some cycles. Thanks! |
@gpaulsen maybe you can ask someone else to review this PR? The fixes are rather trivial. |
The asymmetry proposed by the patch looks strange to me. Basically you always start an epoch but only increase the refcount on the group when there are processes in the group (group of size zero and not the MPI_GROUP_EMPTY). On the other side you release the refcount if the group is not MPI_GROUP_EMPTY, which indicate that 0-process groups will see their refcount reduced, while they did not go through the proper refcount increase. But I might misread the commit, and/or miss some other checks at the upper level that would provide a shortcut for groups with no processes (but that are not MPI_GROUP_EMPTY) |
f91f31d
to
23bccf9
Compare
@bosilca I agree - I restored the symmetry. |
- Make sure the epoch type is set before returning from MPI_Win_start(). - Make sure the group is only free'd if it is valid in MPI_Win_complete(). - Fix possible double free() of the group. Co-authored-by: Lisandro Dalcin <dalcinl@gmail.com> Signed-off-by: Austen Lauria <awlauria@us.ibm.com>
f91335b
to
ec0d6ff
Compare
bot:aws:retest |
Fixes #8677
Signed-off-by: Austen Lauria awlauria@us.ibm.com