Channels last more ops#20775
Merged
Merged
Conversation
Extends the channels_last dialect and DecomposeChannelsLastPass with three more ops from pytorch#20093: adaptive_avg_pool2d, upsample_bilinear2d.vec, and upsample_nearest2d.vec. Each follows the established pattern (permute-wrapped aten op in the dialect; permute + aten op + permute in the decomposition). The decomposition targets aten._adaptive_avg_pool2d (the form to_edge produces) and the upsample .vec overloads, all of which have portable runtime kernels. Part of pytorch#20093. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Completes the implementable pytorch#20093 op list. grid_sampler_2d follows the standard permute-wrapper pattern (only the input is channels-last; the grid passes through). max_pool2d_with_indices is multi-output: the dialect op returns (values, indices) and the decomposition permutes both back to NHWC — the indices are flat spatial positions (h*W+w) and so are layout-agnostic. Not added: max_pool2d.default (no portable runtime kernel; edge lowers maxpool to the _with_indices form) and t_copy/transpose_copy (per the issue, these normalize to permute_copy, which is conversion-side rather than new dialect ops). Part of pytorch#20093. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20775
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 1 New Failure, 1 Cancelled Job, 1 PendingAs of commit 443cfc7 with merge base 8f1abad ( NEW FAILURE - The following job has failed:
CANCELLED JOB - The following job was cancelled. Please retry:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
AdrianLundell
approved these changes
Jul 14, 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
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
Adds the remaining #20093 ops to the channels_last dialect and DecomposeChannelsLastPass:
adaptive_avg_pool2d, upsample_bilinear2d.vec, upsample_nearest2d.vec, max_pool2d_with_indices, and
grid_sampler_2d. Builds on the landed dialect (#20559) and decompose pass (#20615); this
completes the implementable op list.
Each op is a permute-wrapped aten op in the dialect and decomposes to permute_copy → edge.aten.X →
permute_copy for the CPU path. grid_sampler_2d permutes only the activation (grid passes
through); max_pool2d_with_indices is multi-output and permutes both values and indices
(layout-agnostic h*W+w positions).
Excluded per the issue: max_pool2d.default (no runtime kernel) and t_copy/transpose_copy
(normalize to permute_copy, conversion-side).
Part of #20093
Test plan
kernel).
to_executorch → run).