[Vulkan] Keep reductions the reduce shaders cannot handle on CPU - #23206
Closed
mergennachin wants to merge 1 commit into
Closed
mergennachin wants to merge 1 commit into
mergennachin wants to merge 1 commit into
Conversation
The general reduce implementation handles one or two reduced dims, but the partitioner accepted empty dim lists and full sum/mean reductions over N-D inputs, which then failed to build or computed the wrong reduction. Texture storage also folds batch into channels, so 4D reductions over dim 0, or over dim 1 with batch above 1, cannot be computed. These cases now stay on CPU. Authored with OpenAI Codex; split planned with Claude Code.
pytorch-bot
Bot
added
the
module: vulkan
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/23206
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit d0ff92f with merge base a318382 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This was referenced Sep 28, 2026
mergennachin
added a commit
that referenced
this pull request
Oct 4, 2026
) The general reduce implementation handles one or two reduced dims, but the partitioner accepted empty dim lists and full sum/mean reductions over N-D inputs, which then failed to build or computed the wrong reduction. Texture storage also folds batch into channels, so 4D reductions over dim 0, or over dim 1 with batch above 1, cannot be computed. These cases now stay on CPU. `argmax` and `argmin` are restricted to contiguous buffers and last-dimension reductions, matching `ArgReduce.cpp`. A missing or `None` dim is supported only for rank-1 inputs; rank-2 full reductions, including `[1, N]`, fall back to CPU for either `keepdim` value. Part 6/15 of the Vulkan transformer and operator-conformance stack. Depends on #23244; review against the selected base branch. Integration PR: #23254. Validation: Four focused native tests passed on MoltenVK with portable CPU kernels. The new test covers 28 argmax/argmin cases: rank-2 `dim=None`, non-last dimensions, rank-1 full reductions, and positive/negative last dimensions, with both keepdim settings. It checks actual Vulkan buffer storage for supported cases, CPU fallback for unsupported cases, and exact agreement with ATen. Existing 4D reduction, reduction-dimension fallback, and first-NaN tests also passed. The new test failed on the reported `[1, N]` case before the fix. Lintrunner and `git diff --check` pass; CI is rerunning. Recreates #23206 through ghstack. Prior review discussion remains on that PR. Authored with OpenAI Codex; split planned with Claude Code. cc @SS-JIA @manuelcandales @digantdesai @cbilgin
This branch was successfully deployed
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
1 participant
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.
Superseded by #23245, part 6/15 of the ghstack replacement series. The current integration PR is #23254. This PR is closed in favor of the replacement; its previous description and review history are retained.
The general reduce implementation handles one or two reduced dims, but the partitioner accepted empty dim lists and full sum/mean reductions over N-D inputs, which then failed to build or computed the wrong reduction. Texture storage also folds batch into channels, so 4D reductions over dim 0, or over dim 1 with batch above 1, cannot be computed. These cases now stay on CPU.
Part 6/15 of the Vulkan transformer and operator-conformance stack. Depends on #23205; review against the selected base branch. Integration PR: #23162.
Validation: 2 passed, 2 warnings in 70.88s (0:01:10). Lintrunner and git diff --check pass. Native tests use MoltenVK with portable CPU kernels; hardware and SwiftShader CI are pending.
Authored with OpenAI Codex; split planned with Claude Code.
cc @SS-JIA @manuelcandales @digantdesai @cbilgin