Support zero-sized Hopper MX scale layouts#10275
Merged
Merged
Conversation
roman-openai
force-pushed
the
roman/hopper-zero-sized-mx-layouts
branch
4 times, most recently
from
e332c8a to
6453855
Compare
roman-openai
force-pushed
the
roman/hopper-zero-sized-mx-layouts
branch
from
6453855 to
b8df62b
Compare
ThomasRaoux
approved these changes
May 11, 2026
ptillet
approved these changes
May 11, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fix Hopper MX scale layout conversion for zero-sized scale tensors.
torch.nn.functional.padrejects some valid zero-numel outputs. This shows up in Hopper scale swizzle after RHS transpose: for example,(0, 64)becomes(64, 0), and row padding still leaves a zero-width output.The fix mirrors Blackwell scale layout: skip
padwhen there are no elements, updateM/Kto the padded extents, and let the existing zero-element reshape produce the swizzled storage shape.Also relax only
StridedLayoutTransformation.swizzle_datafor zero-numel input. Hopper scale roundtrip can produce a valid empty canonical tensor such as(2, 0)withstride(-1) == 2; there are no elements for the packed-stride invariant to constrain.Validation:
PYTHONPATH=$PWD/python/triton_kernels python -m pytest -q python/triton_kernels/tests/test_tensor_details/test_layout_hopper.py::test_mxfp4_scale_zero_sized_roundtrip