Skip to content

[ET-VK] Fix pack_fp_linear_weight for devices without VK_KHR_16bit_storage#18653

Merged
Gasoonjia merged 1 commit intomainfrom
gh/SS-JIA/515/orig
Apr 1, 2026
Merged

[ET-VK] Fix pack_fp_linear_weight for devices without VK_KHR_16bit_storage#18653
Gasoonjia merged 1 commit intomainfrom
gh/SS-JIA/515/orig

Conversation

@pytorchbot
Copy link
Copy Markdown
Collaborator

This PR was created by the merge bot to help merge the original PR into the main branch.
ghstack PR number: #18642 by @SS-JIA
^ Please use this as the source of truth for the PR details, comments, and reviews
ghstack PR base: https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/base
ghstack PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/main
Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/orig
Differential Revision: D99133993
@diff-train-skip-merge

…orage

The `pack_fp_linear_weight` prepack shader crashes on devices that lack
`VK_KHR_16bit_storage` support because the half-precision variant reads from a
`float16_t[]` staging buffer, which requires that extension.

This applies the same two-dtype pattern used by `nchw_to_image` and
`conv2d_dw_prepack_weights`: a new `BUF_DTYPE` shader parameter allows the
staging buffer to use float32 (`[half, float]` combo) while the packed output
remains half-precision. The runtime selects the correct variant via
`get_staging_dtype_for()`, which returns `kFloat` when the device lacks fp16
buffer support.

All three call sites that construct the `pack_fp_linear_weight` shader name
(Linear.cpp, Conv1dPW.cpp, Conv2dPW.cpp) are updated to append the staging
dtype suffix.

Authored with Claude.

Differential Revision: [D99133993](https://our.internmc.facebook.com/intern/diff/D99133993/)

ghstack-source-id: 361148853
Pull Request resolved: #18642
@pytorchbot pytorchbot requested a review from SS-JIA as a code owner April 1, 2026 21:16
@pytorch-bot
Copy link
Copy Markdown

pytorch-bot Bot commented Apr 1, 2026

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18653

Note: Links to docs will display an error until the docs builds have been completed.

⏳ No Failures, 115 Pending

As of commit 0c0e540 with merge base ad235f8 (image):
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Apr 1, 2026
@github-actions
Copy link
Copy Markdown

github-actions Bot commented Apr 1, 2026

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

@Gasoonjia Gasoonjia merged commit 6bd9bca into main Apr 1, 2026
159 of 163 checks passed
@Gasoonjia Gasoonjia deleted the gh/SS-JIA/515/orig branch April 1, 2026 21:30
Jiseong-oh pushed a commit to Jiseong-oh/executorch that referenced this pull request Apr 2, 2026
…orage (pytorch#18653)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: pytorch#18642 by
@SS-JIA
^ Please use this as the source of truth for the PR details, comments,
and reviews
ghstack PR base:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/main
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/orig
Differential Revision:
[D99133993](https://our.internmc.facebook.com/intern/diff/D99133993/)
@diff-train-skip-merge

Co-authored-by: ssjia <ssjia@devvm26340.ftw0.facebook.com>
Jiseong-oh pushed a commit that referenced this pull request Apr 2, 2026
…orage (#18653)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: #18642 by
@SS-JIA
^ Please use this as the source of truth for the PR details, comments,
and reviews
ghstack PR base:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/main
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/orig
Differential Revision:
[D99133993](https://our.internmc.facebook.com/intern/diff/D99133993/)
@diff-train-skip-merge

Co-authored-by: ssjia <ssjia@devvm26340.ftw0.facebook.com>
Jiseong-oh pushed a commit that referenced this pull request Apr 7, 2026
…orage (#18653)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: #18642 by
@SS-JIA
^ Please use this as the source of truth for the PR details, comments,
and reviews
ghstack PR base:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/main
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/SS-JIA/515/orig
Differential Revision:
[D99133993](https://our.internmc.facebook.com/intern/diff/D99133993/)
@diff-train-skip-merge

Co-authored-by: ssjia <ssjia@devvm26340.ftw0.facebook.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants