Skip to content

[ET-VK][ez] Make q8ta_conv2d use 4C1W layout#17390

Merged
meta-codesync[bot] merged 8 commits intogh/SS-JIA/421/basefrom
gh/SS-JIA/421/head
Feb 13, 2026
Merged

[ET-VK][ez] Make q8ta_conv2d use 4C1W layout#17390
meta-codesync[bot] merged 8 commits intogh/SS-JIA/421/basefrom
gh/SS-JIA/421/head

Conversation

@SS-JIA
Copy link
Copy Markdown
Contributor

@SS-JIA SS-JIA commented Feb 11, 2026 •
edited
Loading

Stack from ghstack (oldest at bottom):

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit outputs_storage declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: D93000165

[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
f21bfb9 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
@pytorch-bot
Copy link
Copy Markdown

pytorch-bot Bot commented Feb 11, 2026 •
edited
Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17390

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

✅ No Failures

As of commit a575c84 with merge base dcfd12d (image):
💚 Looks good so far! There are no failures yet. 💚

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

This was referenced Feb 11, 2026
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
@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 Feb 11, 2026
@github-actions
Copy link
Copy Markdown

github-actions Bot commented Feb 11, 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.

Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
0578684 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
@SS-JIA SS-JIA mentioned this pull request Feb 11, 2026
Merged
ssjia and others added 2 commits February 11, 2026 15:13
Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
03aec14 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
85e5541 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
@SS-JIA SS-JIA mentioned this pull request Feb 12, 2026
Merged
Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
c11b3d5 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
ssjia added 3 commits February 12, 2026 15:34
Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
c715909 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
4e071b7 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
Update on "[ET-VK][ez] Make q8ta_conv2d use 4C1W layout"
a575c84 
This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)

[ghstack-poisoned]
@meta-codesync meta-codesync Bot merged commit 27d0e53 into gh/SS-JIA/421/base Feb 13, 2026
195 of 197 checks passed
@meta-codesync meta-codesync Bot deleted the gh/SS-JIA/421/head branch February 13, 2026 16:14
@pytorchbot pytorchbot mentioned this pull request Feb 13, 2026
Merged
SS-JIA pushed a commit that referenced this pull request Feb 13, 2026
@SS-JIA
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
0d0d713 
Pull Request resolved: #17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
SS-JIA pushed a commit that referenced this pull request Feb 13, 2026
@SS-JIA
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
7acb2b9 
Pull Request resolved: #17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
SS-JIA pushed a commit that referenced this pull request Feb 13, 2026
@SS-JIA
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
5a07a0a 
Pull Request resolved: #17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
SS-JIA pushed a commit that referenced this pull request Feb 13, 2026
@SS-JIA
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
ffcada2 
Pull Request resolved: #17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
SS-JIA pushed a commit that referenced this pull request Feb 13, 2026
@SS-JIA
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
f5443c8 
Pull Request resolved: #17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
SS-JIA pushed a commit that referenced this pull request Feb 13, 2026
@SS-JIA
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
911ed78 
Pull Request resolved: #17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
chizkiyahu pushed a commit to chizkiyahu/executorch that referenced this pull request Feb 23, 2026
@chizkiyahu
[ET-VK][ez] Make q8ta_conv2d use 4C1W layout
727e84e 
Pull Request resolved: pytorch#17390

This changes the q8ta_conv2d and q8ta_conv2d_dw operators' input layout from PackedInt8_4W4C to PackedInt8_4C1W in the op registry. The 4C1W layout aligns with the natural output format of channel-packed convolutions, avoiding unnecessary layout conversions between consecutive conv layers.

Also adds explicit `outputs_storage` declarations (PACKED_INT8_CHANNELS_PACKED_BUFFER) to both the PW and general q8ta_conv2d op registrations, ensuring the layout propagation pass can correctly determine output layouts.
ghstack-source-id: 340983082
@exported-using-ghexport

Differential Revision: [D93000165](https://our.internmc.facebook.com/intern/diff/D93000165/)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@manuelcandales manuelcandales manuelcandales approved these changes

Assignees

No one assigned

Labels

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

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@SS-JIA @manuelcandales