Conversation
Member
Author
|
For information @liamsemeria, update such that with TVM 0.26 we hoist 0-initialization correctly. |
Codecov Report❌ Patch coverage is 📢 Thoughts on this report? Let us know! |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Motivation
TVM 0.26 keeps reduction initialization inside the update loop, which can prevent LLVM from automatically vectorizing otherwise vectorizable loops and cause significant performance regressions.
Description
Add a final TVM schedule transform that separates reduction initialization from update loops.
Use TVM's
decompose_reductionprimitive for regular reductions and a structural fallback for reductions hidden below XTC loop-partition scopes. Preserve spatial loop annotations, support cache-write buffers, and leave already externalized reductions unchanged because they no longer contain an initialization statement.Emit the transform directly in generated TVM schedule fragments and update the corresponding FileCheck expectations.
Commits
tvm: decompose reduction initializersDiscussion
This PR is stacked on #142, which itself depends on #133.
TODO:
Testing
uv run pytest tests/pytest/tvm -q— 11 passeduv run lit -j 8 tests/filecheck/backends tests/filecheck/schedules tests/filecheck/search— 113 passed, 3 expected failures, 24 unsupportedmake check-formatcurrently reports unrelated pre-existing formatting issues inMlirOps.pyandTVMOps.py