Arm backend: Keep INT delegate boundaries quantized - #22966
Merged
zingo merged 1 commit intoSep 22, 2026
Merged
Conversation
INT-only TOSA partitions could keep dequantize and floating-point operations inside a delegate at an output boundary. The delegate then wrote integer data into a buffer ExecuTorch described as FP32. Without: delegate [INT16 -> DQ -> split FP32] -> portable FP32 With: delegate INT16 -> portable [DQ -> split FP32] Remove Q/DQ and floating-point boundary nodes until an INT-only partition exposes integer buffers. This handles the supported Q/DQ variants for both INT8 and INT16 graphs. Add a regression based on the shared split found in YOLO. Force BMM outside the delegate so the boundary remains stable if U55 gains INT16 BMM support. Run the regression on FVP with the portable runner. Add the portable split and BMM kernels needed by the test. Signed-off-by: Zingo Andersen <Zingo.Andersen@arm.com> Change-Id: I51fb0e916fcf2b8a77d86d832c6acfcfddd63335
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22966
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New Failure, 1 PendingAs of commit f556a7b with merge base 3ffd297 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
zingo
added
partner: arm
Collaborator
Author
|
Failed jobs are unrelated |
oscarandersson8218
approved these changes
Sep 22, 2026
zingo
deleted the
Arm-backend-Keep-INT-delegate-boundaries-quantized
branch
September 23, 2026 07:19
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Summary
INT-only TOSA partitions could keep dequantize and floating-point operations inside a delegate at an output boundary. The delegate then wrote integer data into a buffer ExecuTorch described as FP32.
Without:
delegate [INT16 -> DQ -> split FP32] -> portable FP32
With:
delegate INT16 -> portable [DQ -> split FP32]
Remove Q/DQ and floating-point boundary nodes until an INT-only partition exposes integer buffers. This handles the supported Q/DQ variants for both INT8 and INT16 graphs.
Add a regression based on the shared split found in YOLO. Force BMM outside the delegate so the boundary remains stable if U55 gains INT16 BMM support.
Test plan
Run the regression on FVP with the portable runner. Add the portable split and BMM kernels needed by the test.
cc @digantdesai @freddan80 @per @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani