fix(llvm): Restore fast lowering for f64.min in LLVM #5796
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
Replaces f64 min/max emulation.
The emulation was previously added because of codegen issues with LLVM 11,
but that does not seem to be an issue with LLVM 18 anymore.
- add llvm.minimum.* / llvm.maximum.* intrinsics to the LLVM translator
- lower scalar and vector f{32,64}.{min,max} directly to those intrinsics
- drop the legacy constrained fcmp/select emulation that was blocking native lowering
- rely on the existing canonicalise-nans path so Wasm’s NaN semantics stay intact
- restores native LLVM min/max codegen and removes the observed f64.min slowdown
|
The initial fix removed canonicalization. With those added back, we are almost as slow as before, even when using the intrinsics. So the real culprit seems to be the canonicalization system. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
Replaces f64 min/max emulation with the regular LLVM intrinsics.
The emulation was previously added because of codegen issues with LLVM 11,
but that does not seem to be an issue with LLVM 18 anymore
Closes #5771
This change drastically improves the performance of the test case in issue #5771 .
Before:
After:
NOTE: Still need to fix preserving nan canonicalization!