OpenMP Offloading double-build hanging (original) (raw)

February 27, 2024, 7:05pm 1

Our internal OpenMP Offloading double-build started hanging on AArch64 sometime between:

Feb 22 - bcbffd99c48ed0cabd1b94e9ff252680f0968fc3
Feb 23 - 13acb3af5ad48e850cf37dcf02270ede3f267bd4

Is this a known issue? If not, I will attempt to triage it.

jdoerfert February 27, 2024, 7:11pm 2

We were looking at a hang in on of the offloading tests in sollve_vv that started around that time.
It seems to be due to 7f71fa909a10be182b82b9dfaf0fade6eb84796c, but I am somewhat doubtful.
Could you check your setup before/after that commit?

Tag: @ajarmusch

mcinally February 27, 2024, 7:37pm 3

Thanks for the heads up. Running the experiment now and will report back shortly.

mcinally February 27, 2024, 10:36pm 4

7f71fa909a10be182b82b9dfaf0fade6eb84796c^ is hanging as well.

I only have 5 steps in a bisect from here. I’ll see if I can narrow it down. It will take a while, but I’ll report back when I know more.

mcinally February 29, 2024, 9:38pm 5

The build looks to have fixed itself before I could finish bisecting. Thanks for your help, Johannes.