[Hexagon] Fix LWP assembly handler (predicate register)#17204
Merged
quic-sanirudh merged 2 commits intoapache:mainfrom Jul 28, 2024
Merged
[Hexagon] Fix LWP assembly handler (predicate register)#17204quic-sanirudh merged 2 commits intoapache:mainfrom
quic-sanirudh merged 2 commits intoapache:mainfrom
Conversation
This solved the issue with LWP that appears with maxpool. The problem was that the LWP handler was forgetting to save p0 (used by the handler). This predicate register needs to be saved too, just like r0-r5, as it had been decided that it was the responsibility of the handler to save everything (even these theoretically caller-saved registers). Said differently, since it had been decided that calling the LWP handler would not follow the normal ABI, and that the LWP handler would save everything it touches (even normally caller-saved registers like r0-r15 and p0-3), then it absolutely needs to save the predicate registers too (in particular p0, which was causing the issue). The issue appeared only with maxpool because it's the only one that had a state saved in p0 before calling the LWP handler. And this call destroyed the content of what it had saved, making it subsequently branch to different portions of the code. Fix: Allocate 32 bytes (instead of 24 previously), in order to save p3:0, and I save those at the bottom of the stack. Restore it at the end of the LWP handler.
tqchen
approved these changes
Jul 26, 2024
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 PR solves a known issue with Lightweight Profiling (LWP) on Hexagon that we have fixed in our downstream repo and that I'd like to upstream for the community. The issue appeared with the operator maxpool. When using LWP it gave incorrect results (i.e. with the accuracy terribly bad).
The problem was that the LWP handler was forgetting to save p0 (used by the handler). This predicate register needs to be saved too, just like r0-r5, as it had been decided that it was the responsibility of the handler to save everything (even these theoretically caller-saved registers). Said differently, since it had been decided that calling the LWP handler would not follow the normal ABI, and that the LWP handler would save everything it touches (even normally caller-saved registers like r0-r15 and p0-3), then it absolutely needs to save the predicate registers too (in particular p0, which was causing the issue).
The issue appeared only with maxpool because it's the only one that had a state saved in p0 before calling the LWP handler. And this call destroyed the content of what it had saved, making it subsequently branch to different portions of the code.
Fix: Allocate 32 bytes (instead of 24 previously), in order to save p3:0, and I save those at the bottom of the stack. Restore it at the end of the LWP handler.