[asm] Handle PackOp inputs in regalloc liveness and linear scan#1144
Open
Hardcode84 wants to merge 2 commits intoiree-org:mainfrom
Open
[asm] Handle PackOp inputs in regalloc liveness and linear scan#1144Hardcode84 wants to merge 2 commits intoiree-org:mainfrom
Hardcode84 wants to merge 2 commits intoiree-org:mainfrom
Conversation
Hardcode84
changed the title
Hardcode84
force-pushed
the
waveasm-pack-regalloc
branch
from
8582ac3 to
dbe05c6
Compare
harsh-amd
reviewed
Mar 23, 2026
waveasm/lib/Transforms/Liveness.cpp
Outdated
| "pack result must have a live range"); | ||
|
|
||
| for (Value input : packOp.getElements()) { | ||
| // Extend the pack result's range start to cover this input's def. |
Contributor
There was a problem hiding this comment.
Is it guaranteed that pack inputs have no uses other than through the pack result? If a pack input value is also used independently after the pack op, then erasing its live range and only extending the pack result's start could miss extending the end.
Contributor
Author
There was a problem hiding this comment.
That's a good point, we need to extend pack op live range to the entire input lifetime.
PackOp is a register allocation directive: its N inputs must form a contiguous register block matching the pack result. Previously, pack inputs got independent allocations to arbitrary registers while the result got a correct contiguous allocation, leaving downstream consumers reading uninitialized physical registers. Fix by treating pack inputs as sub-registers of the pack result: - Liveness: extend the pack result's live range backwards to cover input defs, then remove inputs from allocation worklists. - LinearScanPass: post-pass assigns input[i].physReg = result + i, mirroring the existing ExtractOp post-pass. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
Previously only the start of pack inputs' live ranges was merged into the pack result. If a pack input had independent uses after the pack op, the allocator could reuse its physical register prematurely. Fix by extending the pack result's end to the max of all input ends. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
Hardcode84
force-pushed
the
waveasm-pack-regalloc
branch
from
dbe05c6 to
9f1be23
Compare
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
2 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.
PackOp is a register allocation directive: its N inputs must form a contiguous register block matching the pack result. Previously, pack inputs got independent allocations to arbitrary registers while the result got a correct contiguous allocation, leaving downstream consumers reading uninitialized physical registers.
Fix by treating pack inputs as sub-registers of the pack result:
This is prerequisite for properly handling SRD construction instead of using raw asm ops.