Hoist the shared dense load as a let in stage_strided_loads#8964
Merged
Hoist the shared dense load as a let in stage_strided_loads#8964
Conversation
This will make it possible in future to transpose it as one thing.
alexreinking
approved these changes
Feb 25, 2026
Member
alexreinking
left a comment
alexreinking
left a comment
There was a problem hiding this comment.
Looks good with a couple clarifying questions and one comment-wording nit.
src/StageStridedLoads.cpp
Outdated
| replacer.replacements.emplace(std::make_pair(alloc, l), shuf); | ||
|
|
||
| // We can't lift the shared load further out than the scope over | ||
| // which the loads definition occur. If k.scope is null, the loads |
Member
There was a problem hiding this comment.
Grammar here is awkward. "over which the load's definition occurs"? "the load is valid everywhere"?
| return result; | ||
| } | ||
|
|
||
| bool can_hoist_shared_load(const IRNode *n, const std::string &buf, const Expr &idx) { |
Member
There was a problem hiding this comment.
Is there any risk of hoisting past an extern stage with a side-effect?
Member
Author
There was a problem hiding this comment.
I think there might be. I'll try to construct a failure
src/StageStridedLoads.cpp
Outdated
Comment on lines
+249
to
+250
| // within this stmt, and there are no stores to the given buffer in the | ||
| // stmt. |
Member
There was a problem hiding this comment.
Is there a test that covers the "don't hoist past a store" mistake?
Member
|
Failure is #8928 |
Also address other review comments
Member
|
correctness_stage_strided_loads failure is real. I'm going to clear ccache on arm64-linux-worker-4 |
Member
Author
|
Failures unrelated |
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
2 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.
Previously each strided load did the same dense load redundantly, and we relied on LLVM to dedup. Pulling it out as a let makes it possible in #8925 future to transpose it as one thing.