-
Notifications
You must be signed in to change notification settings - Fork 931
v4.1.x: Fix handling of large data in MPI_Sendrecv_replace #8897
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
v4.1.x: Fix handling of large data in MPI_Sendrecv_replace #8897
Conversation
bd2a35e to
f1099d1
Compare
|
:bot:aws:retest |
|
@jsquyres could you please verify the bot? this PR is made of two commits:
The first one contains the following messages the bot complained of a missing cherry pick message and did not notice the I would suggest to add the new |
|
The bot:notacherrypick goes in the PR body, not in the git commit message. We are opinionated that bot commands don’t belong in git history. I don’t get your back port comment. The bot doesn’t check that it was a clean cherry pick or that you changed things. Only that it is either a PR marked as not a cherry commit or there’s a cherry-commit line with a reference to a valid hash. |
|
Thanks @bwbarrett for the insight, I will update the PR accordingly. Don't the My point is that, regardless the bot, I like to make the distinction between clean cherry pick and backports. In the second case, I understand we have several options:
|
Because MPI_Sendrecv_replace() uses PMPI_Sendrecv() with MPI_PACKED under the hood, the data to be exchanged size = MPI_Type_size(datatype) * count must fit in a signed integer. Otherwise, PMPI_Sendrecv() - fails with an error message if (int)size < 0 - silently truncate the data if (int)size >= 0 Refs. open-mpi#8862 Thanks Jakub Benda for reporting this issue and suggesting a fix. Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp> (back-ported from commit 6a11873)
and fix a comment Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp> (cherry picked from commit 0b38190)
f1099d1 to
19c30a0
Compare
do ignore the status returned by opal_convertor_pack() since it is *not* MPI_SUCCESS in case of success. This fixes a regression introduced in open-mpi/ompi@0b38190 Refs. open-mpi#8907 Thanks Lisandro Dalcin for reporting this issue. Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp> (cherry picked from commit 34b764c)
bot:notacherrypick