Fix failure in Flex Attention UT #4031
Draft
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
chengjunlu
approved these changes
Apr 28, 2025
| // Case 4: Value is defined by arith::ExtSIOp, tt::AddPtrOp or | ||
| // arith::AddIOp operation. | ||
| // Case 4: Value is defined by arith::ExtSIOp, tt::AddPtrOp, | ||
| // arith::AddIOp or tt::BroadcastOp,tt::SplatOp operation |
There was a problem hiding this comment.
Suggested change
| // arith::AddIOp or tt::BroadcastOp,tt::SplatOp operation | |
| // arith::AddIOp, tt::BroadcastOp or tt::SplatOp operation. |
| @@ -147,6 +147,11 @@ struct TritonIntelGPUMaterializeBlockPointerPass | |||
|
|
|||
| LDBG("Considering tensor of pointer load op: " << loadOp); | |||
|
|
|||
| if (!ttgi::isDivisible(ptr, 4)) { | |||
There was a problem hiding this comment.
Why below code is not sufficient?
// Base pointer can be compensate by the offset and base width, where they
// each has restriction that it has to be 4 bytes aligned.
if (axisInfo->getDivisibility(fastChangeDim) % 4 != 0) {
LDBG(
"Found Non 4 bytes aligned base: " << axisInfo->getDivisibility(1));
return false;
}
There was a problem hiding this comment.
Check base ptr itself that just similar to check for base ptr created by make_tensor_ptr.
// Ensure the base ptr is 4-byte aligned.
// Note: the HW requires the address to be 64-byte aligned, however we will
// compensate by imposing restrictions on the offsetX and baseWidth.
TypedValue<tt::PointerType> base = makeTensorPtrOp.getBase();
if (!ttgi::isDivisible(base, 4)) {
LDBG("Found non 4-bytes aligned base: " << base);
return false;
}
There was a problem hiding this comment.
As discussed offline, root cause still need to be identified.
|
Please add a description explaining what changed and why (see https://cbea.ms/git-commit/#why-not-how for more info). |
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.
#3998