[LAYOUTS] Implement generically getUniqueContigPerThread #5840
+152
−187
Conversation
This file contains 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
lezcano
force-pushed
the
layouts
branch
3 times, most recently
from
04fb39a to
18d4265
Compare
lezcano
commented
Feb 11, 2025
| @@ -44,29 +44,30 @@ module attributes {"ttg.num-warps" = 8 : i32, "ttg.threads-per-warp" = 64 : i32} | |||
| } | |||
| } | |||
|
|
|||
| // ----- | |||
| // // ----- | |||
There was a problem hiding this comment.
@antiagainst This PR is breaking these two lit tests. I think it's coming from the changes in Allocation.cpp, but I have not track them down. I assume the fix would be something like using LinearEncodingAttr::getOrder instead of getOrder somewhere or so.
There was a problem hiding this comment.
Would it be alright to merge this PR as-is and you guys could then send a forward-fix?
There was a problem hiding this comment.
Feel free to land; we can fix later. @binarman FYI
|
@ThomasRaoux @Jokeren this one's ready for another review. I ended up having to fix a ton of things in the scan lowering, but hopefully all of them are reasonable enough. |
Jokeren
reviewed
Feb 12, 2025
This comment was marked as spam.
This comment was marked as spam.
…ectorisation lenghts
ThomasRaoux
reviewed
Feb 12, 2025
ThomasRaoux
approved these changes
Feb 13, 2025
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Outdated suggestions cannot be applied.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.
This allows vectorisation on global loads and smem in some cases we didn't use it before, as we now compute the order of the elements looking at the actual LinearLayout associated to the given shape of the tensor, which is quite neat.
We end up touching a few things in the Scan lowering as BlockedLayouts when converted to LinearEncodings may not have the same order on elems/threads/warps. This is a feature, not a bug, as it takes us closer to supporting arbitrary LinearEncodings within the tt.scan op.