Don't do an invalid mask conversion optimization for conditional select nodes #118154
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
There was a problem hiding this comment.
Pull Request Overview
This PR fixes a bug in the mask conversion optimization pass that was creating invalid IR by incorrectly handling conditional select nodes. The optimization was transforming ConditionalSelect(mask, simd, simd) when IR requires ConditionalSelect(simd, simd, simd) format.
Key changes:
- Removes invalid conditional select handling from mask conversion optimization
- Eliminates the
OperIsVectorConditionalSelectmethod that enabled the faulty transform - Cleans up whitespace in the optimization visitor code
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| src/coreclr/jit/optimizemaskconversions.cpp | Removes conditional select handling logic and cleans up whitespace |
| src/coreclr/jit/gentree.h | Removes OperIsVectorConditionalSelect method declaration |
| src/coreclr/jit/gentree.cpp | Removes OperIsVectorConditionalSelect method implementation |
github-actions
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
tannergooding
commented
Jul 29, 2025
EgorBo
approved these changes
Jul 29, 2025
|
Also, worth adding the original repro as a test case maybe? |
|
/backport to release/10.0-preview7 |
|
Started backporting to release/10.0-preview7: https://github.com/dotnet/runtime/actions/runs/16599485416 |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.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 resolves #118143
This was exposed in, but not actually introduced by, #116983. The actual bug was introduced a bit over 10 months ago (still unique to .NET 10), as part of enabling the
mask conversionoptimization pass for xarch.The general issue was that the transform was introducing invalid IR as it was creating a
ConditionalSelect(mask, simd, simd)node when in actuality IR requires this to be:ConditionalSelect(simd, simd, simd). Cases likeSve.ConditionalSelect(mask, simd, simd)andBlendVariableMask(simd, simd, mask)do exist. But they will already be in the right shape as the mask operand will either directly be taking a mask (no conversion necessary) or they will have something likeCvtSimdToMask(simd)for that argument position (which is handled by a different logical path in the mask conversion optimization phase).The side effect of this invalid IR is that we would end up with an incorrect mask, which was logically being treated as
zerowhen it should have been treated as a bitwise operation instead. This resulted in one path being dropped as dead code and us always returningop2ofConditionalSelect(mask, op1, op2).The intent of the handling was rather to look for cases that were logically
ConditionalSelect(MaskToSimd(mask), simd, simd)and say "this could instead beBlendVariableMask(simd, simd, mask), but that is a bit more involved and is largely irrelevant with #116983 as we're preferring importing things to be handled as masks natively where possible already, which would allow the normalCvtSimdToMask(lcl)andCvtMaskToSimd(lcl)handling to kick-in.