skip triton template precompilation in 311.0-3.11.7 to workaround 311 cpython bug #125446
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/125446
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 92efd2e with merge base 1bcbc91 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
ghstack-source-id: 46d53133c782a2fbfa74dfc9b8ffc256db2a91ae Pull Request resolved: #125446
eellison
changed the title
…karound 311 cpython bug" Fix for #125374. We dont have CI for this specific versions, but I verified locally. THere is a cpython bug from 3.11->3.17 where the ast parsing state is global, and errors with multiple threads. when dust settles a little around the new process based compilation we can look into migrating. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
ghstack-source-id: 1a256ddbc98ec5714140865516fab6aee0e70d8e Pull Request resolved: #125446
torch/_inductor/select_algorithm.py
Outdated
| @@ -64,6 +64,14 @@ class KernelNamespace: | |||
| extern_kernels = KernelNamespace() | |||
|
|
|||
|
|
|||
| @functools.lru_cache(None) | |||
| def warn_ast_parse_3_11(): | |||
| log.warning( | |||
There was a problem hiding this comment.
Do we actually need a warning here? It doesn't matter that much right - only some compilation time improvement? In that case, I think most users wouldn't care.
There was a problem hiding this comment.
Now that we have a ton more mm configs, it actually matters a good amount.
There was a problem hiding this comment.
I'd lean towards not having it (i think warning spew is really bad), but if you really want a warning, at least mention what's going to happen if they don't upgrade.
There was a problem hiding this comment.
Yea, fair. I'll remove and then after this is landed try to add a workaround in triton.
…karound 311 cpython bug" Fix for #125374. We dont have CI for this specific versions, but I verified locally. THere is a cpython bug from 3.11.0->3.11.7 where the ast parsing state is global, and errors with multiple threads. when dust settles a little around the new process based compilation we can look into migrating. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
ghstack-source-id: 15b720c5c7e6fb9c0e9a0901313065c32074565f Pull Request resolved: #125446
|
@pytorchbot merge |
eellison
added
topic: not user facing
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Stack from ghstack (oldest at bottom):
Fix for #125374. We dont have CI for this specific versions, but I verified locally. THere is a cpython bug from 3.11.0->3.11.7 where the ast parsing state is global, and errors with multiple threads. when dust settles a little around the new process based compilation we can look into migrating.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang