Cortex-M: make quantized add and multiply layout-agnostic - #21966
Draft
rascani wants to merge 9 commits into
Draft
Cortex-M: make quantized add and multiply layout-agnostic#21966rascani wants to merge 9 commits into
rascani wants to merge 9 commits into
Conversation
Contributor
Author
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21966
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Cancelled Job, 5 Unrelated FailuresAs of commit 33d1db0 with merge base 8b93850 ( CANCELLED JOB - The following job was cancelled. Please retry:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Aug 20, 2026
This was referenced Aug 21, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Derive the channel-broadcast repeat length from the broadcast operand's element
count instead of a hard-coded channel dimension, so one pair of operators serves
both the legacy NCHW-logical channels-last contract and the NHWC-logical
contiguous one.
The kernels already read identical bytes under either contract; only the shape
metadata placed channels at a different index. Consult the dim order of the
larger operand rather than sniffing the shape, which stays correct when the
broadcast operand's unit dimensions leave its own dim order ambiguous. The
quantizer keeps asking the logical NCHW question it asked before, since it runs
ahead of any layout transform.
Cortex-M55 FVP add and multiply suites, the explicit-NHWC runtime cases, host
layout-parity coverage comparing both contracts through one operator, and
lintrunner.
Authored with Codex.