computeSliceParameters: use full slice if affine exprs are non-monotonic #407
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
cferry-AMD
reviewed
Dec 3, 2024
cferry-AMD
reviewed
Dec 3, 2024
cferry-AMD
reviewed
Dec 3, 2024
cferry-AMD
reviewed
Dec 3, 2024
cferry-AMD
reviewed
Dec 3, 2024
cferry-AMD
approved these changes
Dec 3, 2024
Collaborator
cferry-AMD
left a comment
cferry-AMD
left a comment
There was a problem hiding this comment.
LGTM, thanks for solving! That's much simpler than using the LexMin and LexMax which I imagined as a fix for that.
josel-amd
approved these changes
Dec 3, 2024
josel-amd
added a commit
that referenced
this pull request
Feb 17, 2025
…monotonic_slice_params"" This reverts commit 018230a.
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Outdated suggestions cannot be applied.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.
computeSliceParametersmaps the loop range (given by lower boundslbsand upper boundsubs) into the operand tensor to compute the input slices that contains all indexed values.The computation currently assumes that the affine expressions to map from loop iteration domain to tensor domain are monotonic, and use that information by just evaluating those at
lbsandubsto obtain the slice o the operand tensor.For non-monotonic expressions, the maximum and minimum values might not be at the boundaries.
Detect that case, and then use the full slice to be safe.
Fixes llvm#111830
If shapes are static, we could be cleverer: sometimes we can prove that even though the affine expression is not monotonic in itself, it is monotonic on the range
(lbs, ubs). For example when the affine expression contains ad0 mod 8, and the range is(0, 4).