fix: propagate NaN in arg reductions - #4291
Open
atirna wants to merge 13 commits into
Open
Conversation
Apply the repository clang-format (v21) to the CUDA arg-reduce changes so pre-commit CI passes on the PR.
zcbenz
added
the
await verification
Contributor
|
Sorry about that, I checked for prs before setting up a fork CI and didn't check after. Just tested yours and it's solid. Good work! |
Author
|
hey @erwinzhang7 thank you! for taking time to check this pr. sorry it ended up being competing pr |
zcbenz
added
low priority
and removed
await verification
zcbenz
requested changes
Aug 25, 2026
zcbenz
left a comment
zcbenz
left a comment
Member
There was a problem hiding this comment.
std::isnan and cuda::std::isnan are guaranteed to simply return false for integers so there is no need to add a wrapper. For metal isnan is probably not overloaded for integers but I would prefer adding overloads ourselves than adding a confusing is_nan wrapper.
zcbenz
added
the
await response
std::isnan and cuda::std::isnan already return false for integers. Metal gets integral isnan overloads instead of a named wrapper.
std::isnan(complex64_t) only sees real(), so argmin/argmax skipped an imag NaN while min/max already propagate it via x != x.
Author
|
dropped the is_nan wrappers. cpu uses |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Proposed changes
Fixes #4274 by making
argminandargmaxselect the firstNaN, matching MLX min/max propagation and NumPy/PyTorch behavior. The implementation covers CPU, Metal, and CUDA reduction paths, including pairwise GPU reductions where separate partials contain NaNs.Checklist
pre-commit run --all-filesto format my code / installed pre-commit prior to committing changes (passed on the formatting commit; see 03694f5)Tests
cmake --build build-metal --target tests -j2./build-metal/tests/tests --test-case='test arg reduce NaN'(14 assertions)./build-metal/tests/tests --test-case='test arg reduce*'(190 assertions)