Fix: specializing symbols after runtime assertions added cause codegen issue #153661
+65
−12
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/153661
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 34 New FailuresAs of commit 65e98aa with merge base 004dad4 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Microve
reviewed
May 16, 2025
| r = self._find(s) | ||
|
|
||
| if only_specialized_backed and not (r.is_integer or r.is_real): |
There was a problem hiding this comment.
why do we check not (r.is_integer or r.is_real)?
There was a problem hiding this comment.
if rhs is not integer, its guaranteed the input it still in the graph.
the replacement is not needed for the runtime assert symbol. it could be that its ok if we relax this,idk if we would hit issues.
There was a problem hiding this comment.
actually this should be int float and not is_integer, is_real
…ause codegen issue" Fix #153650 if we have a runtime assertions u0+s1 =10 then s1 got specialized to say 8, then when s1 will no longer be part of the graph we shall replace it with 8. now here is an interesting situation, if we have assert(s0==10) and prefer runtime assertions over guards ! what should happen in that case? we do want the assertion to be emitted, meaning that specializing s0 should not lead to s0 being removed from the graph. Maybe the proper fix is that we should NEVER remove symbols used in runtime assertions from the graph ever? cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]
…ause codegen issue" Fix #153650 if we have a runtime assertions u0+s1 =10 then s1 got specialized to say 8, then when s1 will no longer be part of the graph we shall replace it with 8. now here is an interesting situation, if we have assert(s0==10) and prefer runtime assertions over guards ! what should happen in that case? we do want the assertion to be emitted, meaning that specializing s0 should not lead to s0 being removed from the graph. Maybe the proper fix is that we should NEVER remove symbols used in runtime assertions from the graph ever? cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]
…ause codegen issue" Fix #153650 if we have a runtime assertions u0+s1 =10 then s1 got specialized to say 8, then when s1 will no longer be part of the graph we shall replace it with 8. now here is an interesting situation, if we have assert(s0==10) and prefer runtime assertions over guards ! what should happen in that case? we do want the assertion to be emitted, meaning that specializing s0 should not lead to s0 being removed from the graph. Maybe the proper fix is that we should NEVER remove symbols used in runtime assertions from the graph ever? cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]
|
change the fix strategy trying the new one. |
…ause codegen issue" Fix #153650 Two things: 1) input nodes that represents symbols that are used in runtime assertions should NOT be deleted. 2) during codegen and runtime assertion insertion, do not apply the replacements to input nodes that represent symbols that are used in runtime assertions. cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv voznesenskym penguinwu Guobing-Chen XiaobingSuper zhuhaozhe blzheng jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov [ghstack-poisoned]
…ause codegen issue" Fix #153650 Two things: 1) input nodes that represents symbols that are used in runtime assertions should NOT be deleted. 2) during codegen and runtime assertion insertion, do not apply the replacements to input nodes that represent symbols that are used in runtime assertions. cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv voznesenskym penguinwu Guobing-Chen XiaobingSuper zhuhaozhe blzheng jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov [ghstack-poisoned]
…ause codegen issue" Fix #153650 Two things: 1) input nodes that represents symbols that are used in runtime assertions should NOT be deleted. During remove_unused_inputs if the runtime assertions were inserted correctly and non missed this wont happpen. 2) during codegen and runtime assertion insertion, do not apply the replacements to input nodes that represent symbols that are used in runtime assertions in order to avoid missing inserting runtime assertions depending on those symbols. and to prevent (in the case of inductor) generating invalid code. cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv voznesenskym penguinwu Guobing-Chen XiaobingSuper zhuhaozhe blzheng jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov [ghstack-poisoned]
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
Fix #153650
Two things:
input nodes that represents symbols that are used in runtime assertions should NOT be deleted. During remove_unused_inputs if the runtime assertions were inserted correctly and non missed this wont happpen.
during codegen and runtime assertion insertion, do not apply the replacements to input nodes that represent symbols that are used in runtime assertions in order to avoid missing inserting runtime assertions depending on those symbols.
and to prevent (in the case of inductor) generating invalid code.
cc @ezyang @SherlockNoMad @EikanWang @jgong5 @wenzhe-nrv @voznesenskym @penguinwu @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov