Force storing GC state until we see a labeled IG #114475
Merged
+18
−1
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
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
|
As imagined, no diffs because of this being a corner scenario. |
|
/azp run runtime-coreclr gcstress-extra |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
@dotnet/jit-contrib |
BruceForstall
approved these changes
Apr 11, 2025
|
format job failures are unrelated. |
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Outdated suggestions cannot be applied.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.
This is very odd and corner case scenario. We have a loop after epilog (with stress block layout, we might have something like that). If there are IGs after epilog, the epilog creates additional dummy IG to store GC state in it. We also update the GC vars global variables
emitPrevGCrefVarsandemitInitGCrefVarsto have most up to date information. Whenever their values are different, we mark such IGs withIGF_GC_VARSstating that the IG contains gc information that the gcinfo should record.The loop that is immediately after the epilog is marked for alignment, and so we create a new label IG for it. At this point, the GC vars information
emitPrevGCrefVars == emitInitGCrefVarsand so we skip marking this IG withIGF_GC_VARS. As a result, it inherits the GC var liveness information from the previous block (which is the dummy block after the epilog) which is wrong.The fix is to force store the GC state on all IGs until we come across a valid labeled IG.
Attached is the dump from failing test. Here,
IG14is the epilog andIG15is the dummy IG for which we add all the GC-vars likeV73etc. When we createIG16, we do not add any new GC vars and hence inherit those fromIG15617.dump.asm.zip
Here is the diff after the fix:
Thanks @BruceForstall for brainstorming this scenario with me.
Fixes: #113553