[4.1.x] pml/cm,mtl/ofi: fix datatype offsetting #11907
Conversation
…thin one" This reverts commit 6250f54. The revert fixes a segfault revealed by mtt ibm bsend test. bot:notacherrypick Signed-off-by: Wenduo Wang <[email protected]>
|
If the patch caused segfault, you would need to fix the root cause. |
|
@wzamazon Thanks for the reminder! Let me dig more into the segfault. For the MPAS issue, do you happen to have the reproducer? |
|
so should this PR be closed then? |
|
@hppritcha I'm looking into a fix today. I plan to reuse this PR. We need to fix bsend for the next 4.1 release IMO. |
|
The reproducer is attached Can you add it to one of the test suite in |
|
@wenduwan are you going to add the test case or do you want me to do this? |
|
@hppritcha I'm still looking into the bug. The reproducer uses MPI_Alltoallw but I have a hunch that I can reduce it to send/recv. Anyways I will add a test - which test suite do you recommend, ibm? |
|
please add to ibm. |
This change picks 17b09d9 pml/cm and mtl/ofi on 4.x have diverged further from main. This change manually applies the logic to handle datatype offsetting. It addresses open-mpi#11751 Signed-off-by: Wenduo Wang <[email protected]>
wenduwan
changed the title
wenduwan
changed the title
Thank you! Approved! |
|
@bwbarrett @jsquyres Can we include this change for 4.1.6 release? |
|
Reproducer PR https://github.com/open-mpi/ompi-tests/pull/181 Wei's original example used Alltoallw, which happens to trigger the pml_cm_isend_init path. I find it simpler to use |
This reverts commit 6250f54 and manually picks 17b09d9
It fixes a segfault revealed by mtt ibm bsend test.
bot:notacherrypick