[MLIR][XeGPU] Add unroll patterns and blocking pass for XeGPU (1/N) #137010
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
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
Can you please add some justification/explanation regarding those unroll patterns |
Thanks @fschlimb, I made the changes according to your feedback. I hope I addressed all of your concerns. |
adam-smnk
reviewed
May 8, 2025
There was a problem hiding this comment.
It'll be interesting to later see if we could generalize and reuse vector unrolling to achieve the same. For now, I think it's a good addition to xegpu infrastructure and we'll see in practice how it holds up.
I take it depends on #138701?
adam-smnk
reviewed
May 8, 2025
Yeah. these patterns are supposed to be companions to vector unrolling patterns. They share the same idea, one is for XeGPU ops only, and one is for vector ops. A pass are supposed to use both of them. |
Garra1980
reviewed
May 8, 2025
| } | ||
|
|
||
| private: | ||
| const char *const packAttrName = "__xetile_blocking_pack__"; |
There was a problem hiding this comment.
xetile->xegpu I guess, here and in tests
There was a problem hiding this comment.
good catch. fixed it
Garra1980
reviewed
May 8, 2025
| /// provide a way to customize the native shape of the operation. | ||
| struct UnrollOptions { | ||
| using FilterConstraintFnType = std::function<LogicalResult(Operation *op)>; | ||
| /// Callback function that indicates whether vector unrolling should be |
There was a problem hiding this comment.
nit: let's place this comment above "using" to have uniform look :)
|
Hi @fschlimb and @adam-smnk, do you have more suggestions? |
no, LGTM! |
|
Hi @adam-smnk, I am going to merge this first. If you have more suggestions, feel free to let me know. Thanks for your help! |
|
I'm seeing a regression on 32-bit platforms from this change: |
Jianhui-Li
reviewed
May 17, 2025
| auto shape = vecTy.getShape(); | ||
| SmallVector<Value> results; | ||
| for (SmallVector<int64_t> offsets : | ||
| StaticTileOffsetRange(shape, blockSize)) { |
There was a problem hiding this comment.
The PR seems not handling xegpu.order attribute at this stage. load_nd on PVC support load with transpose, which takes order as [0,1].
StaticTileOffsetRange() can take extra loopOrder parameter which can be used to support this feature.
There was a problem hiding this comment.
The PR seems not handling xegpu.order attribute at this stage. load_nd on PVC support load with transpose, which takes order as [0,1]. StaticTileOffsetRange() can take extra loopOrder parameter which can be used to support this feature.
I feel this belongs to the handling of transpose op, not the handling of xegpu.order attribute.
Hi @mgorny, could you help to share your build instructions, so I can reproduce it in my local machine. |
|
|
A relative simple reproducer would be: For us, Though I suppose |
Thanks, it should have been fixed in #140567, please take a look. |
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.
Similar to vector ops, XeGPU ops need to be unrolled into smaller shapes such that they can be dispatched into a hardware instruction. This PR marks the initial phase of a series dedicated to incorporating unroll patterns for XeGPU operations. In this installment, we introduce patterns for the following operations: