Repository navigation
Vulkan: keep a reshape that changes the last dim before a quantized linear - #24116
Merged
SS-JIA merged 2 commits intoNov 3, 2026
Merged
Conversation
pytorch-bot
Bot
added
the
module: vulkan
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24116
Note: Links to docs will display an error until the docs builds have been completed. ⏳ 3 Pending, 1 Unrelated FailureAs of commit 1bf0cc8 with merge base 1a0ffe8 ( BROKEN TRUNK - The following job 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. |
meta-cla
Bot
added
the
CLA Signed
nil-is-all
added
the
release notes: vulkan
executorch-triage
Bot
added
the
community: contribution
SS-JIA
approved these changes
Oct 26, 2026
Contributor
|
@msluszniak changes LGTM but lintrunner is failing, so can't merge just yet! |
Contributor
Author
|
@SS-JIA Done ;) |
msluszniak
force-pushed
the
ms/vulkan-qlinear-keep-input-view
branch
from
October 30, 2026 13:37
f9ce1ae to
156f422
Compare
…ntized linear QuantizedLinearMatch skipped any view_copy feeding the linear, assuming it is the leading-dim flatten from aten.linear's N-D decomposition. A reshape that merges attention heads ([S, H, D] -> [S, H * D]) was skipped too, so the fused et_vk.linear_q4gsw received the [S, H, D] tensor and lowering failed in the fake kernel. Skip the view only when it keeps the last dim.
SS-JIA
force-pushed
the
ms/vulkan-qlinear-keep-input-view
branch
from
November 3, 2026 00:10
156f422 to
1bf0cc8
Compare
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.
Summary
QuantizedLinearMatchskipped anyview_copyfeeding a quantized linear. A reshape that changes the last dim (merging attention heads,[S, H, D] -> [S, H * D]) was skipped too, solinear_q4gswgot the[S, H, D]tensor andFusePatternsPassfailed. Now only a view that keeps the last dim is skipped.Test plan
python -m unittest backends.vulkan.test.test_vulkan_passes(newtest_fuse_q4gsw_linear_keeps_preceding_reshapefails without the fix).cc @SS-JIA @manuelcandales @digantdesai @cbilgin @nil-is-all