[AMD] [FA] Hoist convert_layout to dotOp for Q out of the loop #6017
+176
−166
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
sjw36
approved these changes
Feb 25, 2025
This PR adds a new amd.pass that hoists conver_layout to dotOperand layout for the Q tensor out of the loop. Therefore, Q tensor is kept in registers instead of being loaded at every iteration of the loop. This PR is actually achieving the same thing as triton-lang#4901. However, triton-lang#4901 does not hoist local_load for Q in the epilogue, making Q tensor live in shared memory all the time. On the other hand, this PR does the trick before stream-pipeline pass. Therefore, the livessness of Q tensor in shared memory is limited in the prologue.
antiagainst
requested changes
Feb 26, 2025
third_party/amd/lib/TritonAMDGPUTransforms/HoistLayoutConversions.cpp
Outdated
Show resolved
Hide resolved
antiagainst
reviewed
Feb 26, 2025
third_party/amd/lib/TritonAMDGPUTransforms/HoistLayoutConversions.cpp
Outdated
Show resolved
Hide resolved
antiagainst
approved these changes
Feb 26, 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Outdated suggestions cannot be applied.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 PR adds a new amd.pass that hoists conver_layout to dotOperand layout for the Q tensor out of the loop. Therefore, Q tensor is kept in registers instead of being loaded at every iteration of the loop.
This PR is actually achieving the same thing as #4901. However, #4901 does not hoist local_load for Q in the epilogue, making Q tensor live in shared memory all the time.
On the other hand, this PR does the trick before stream-pipeline pass. Therefore, the livessness of Q tensor in shared memory is limited in the prologue.