OSHMEM: Removing erroneous initialization check #1247
Conversation
Since the introduction of the on-demand proc allocation the check become erroneous and irrelevant. Moreover, it completely breaks OpenSHMEM support in OMPI. Signed-off-by: Pavel Shamis (Pasha) <pasharesearch@gmail.com>
|
This is a followup commit on the bugfix open-mpi/ompi#1817 |
|
@shamisp This breaks with UCX, right? I think we determined earlier today that it doesn't break OSHMEM with MXM because MXM disables on-demand addprocs (which is why we hadn't seen this before). @hppritcha We discussed this earlier today. If this passes CI (it might still run into the RDMACM problem at Mellanox Jenkins), I think we should merge so that we can get an MTT night in with it. |
|
I don't have MXM to try but I suspect it does. It breaks yoda for sure. I have not seen any special disable code in OSHMEM code. I only saw the code in the OMPI codebase which is not shared with OSHMEM. |
|
Test FAILed. |
|
I"m good with merging this. I'm still confused as to how the mlnx jenkins oshmem tests were working though. |
4 participants
Since the introduction of the on-demand proc allocation
the check become erroneous and irrelevant.
Moreover, it completely breaks OpenSHMEM support in OMPI.
Signed-off-by: Pavel Shamis (Pasha) pasharesearch@gmail.com