[ET-VK][Ops] enabling double support for quantization and dequantization ops #11553

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Open

ahmtox wants to merge 9 commits into gh/ahmtox/18/base from gh/ahmtox/18/head

+96 −2

ahmtox commented Jun 11, 2025 •

edited

Loading

Stack from ghstack (oldest at bottom):

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: D76289197


          [ET-VK][Ops] enabling double support for quantization and dequantizat…

33e9b53

…ion ops

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

ahmtox requested a review from SS-JIA as a code owner

June 11, 2025 16:59

pytorch-bot bot commented Jun 11, 2025 •

edited

Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11553

📄 Preview Python docs built from this PR

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

❌ 2 New Failures, 4 Unrelated Failures

As of commit a553b06 with merge base 3d3cf2a ():

NEW FAILURES - The following jobs have failed:

Lint / lintrunner / linux-job (gh)
>>> Lint for backends/vulkan/runtime/graph/ops/impl/ChooseQParams.cpp:
pull / android / run-emulator (gh)
The process '/usr/bin/sh' failed with exit code 255

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

pull / test-models-linux-basic (mv3, portable, cmake, linux.2xlarge, executorch-ubuntu-22.04-clang12) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
pull / test-qnn-models-linux / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest / linux / linux-job (gh) (trunk failure)
/pytorch/executorch/backends/vulkan/runtime/api/containers/Tensor.cpp:651:17: error: no matching constructor for initialization of 'vkcompute::api::vTensor::TextureLimits'
pull / unittest-editable / linux / linux-job (gh) (trunk failure)
/pytorch/executorch/backends/vulkan/runtime/api/containers/Tensor.cpp:651:17: error: no matching constructor for initialization of 'vkcompute::api::vTensor::TextureLimits'

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

This was referenced Jun 9, 2025

[ET] enabling half dtype input for quantization #11479

Open

[ET-VK] double, short, and uint16 dtype runtime support #11365

Open

[ET-VK][Ops] quantize ops skeleton test framework #11366

Open

[ET-VK][Ops] quantize_per_token.default test setup #11367

Open

[ET-VK][Ops] quantize_per_tensor.default test setup #11368

Open

facebook-github-bot added the CLA Signed label

This was referenced Jun 11, 2025

[ET-VK][Ops] quantization op shaders and impl #11369

Open

[ET] enabling half dtype output for dequantization and making logic consistent #11552

Open

[ET-VK][Ops] dequantize ops skeleton test framework #11480

Open

[ET-VK][Ops] dequantize_per_tensor.default test setup #11481

Open

[ET-VK][Ops] dequantize_per_token.default test setup #11482

Open

[ET-VK][Ops] dequantization op shaders and impl #11483

Open

[ET-VK][Ops] choose_qparams ops skeleton test framework #11554

Open

[ET-VK][Ops] choose_qparams.tensor test setup #11555

Open

[ET-VK][Ops] choose_qparams_per_token_asymmetric.default test setup #11556

Open

[ET-VK][Ops] choose_qparams op shaders and impl #11557

Open

Contributor

facebook-github-bot commented Jun 11, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197

facebook-github-bot added the fb-exported label


          Update on "[ET-VK][Ops] enabling double support for quantization and …

26b834e

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Jun 11, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197

ahmtox added the release notes: vulkan label


          Update on "[ET-VK][Ops] enabling double support for quantization and …

7d22726

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

ahmtox mentioned this pull request

[ET-VK][Ops] common test utils for converting aten types to vulkan types #11575

Open

Contributor

facebook-github-bot commented Jun 11, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197


          Update on "[ET-VK][Ops] enabling double support for quantization and …

cd7c8d6

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Jun 12, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197


          Update on "[ET-VK][Ops] enabling double support for quantization and …

3015eec

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Jun 12, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197


          Update on "[ET-VK][Ops] enabling double support for quantization and …

8eb9125

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

Contributor

facebook-github-bot commented Jun 12, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197


          Update on "[ET-VK][Ops] enabling double support for quantization and …

57093c0

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

ahmtox pushed a commit that referenced this pull request


          [ET-VK][Ops] enabling double support for quantization and dequantizat…

01e4a06

…ion ops

Pull Request resolved: #11553

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit
ghstack-source-id: 290041469
@exported-using-ghexport

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

Contributor

facebook-github-bot commented Jun 12, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197


          Update on "[ET-VK][Ops] enabling double support for quantization and …

afe1f42

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

ahmtox pushed a commit that referenced this pull request


          [ET-VK][Ops] enabling double support for quantization and dequantizat…

8a595c8

…ion ops

Pull Request resolved: #11553

# Context
Since we enabled the possibility for double support in an earlier diff, this enables double support for quantization and dequantization. Since there are limitations to how 64bit can be supported, the expectation is that IO is to be downgraded to 32bit.

# Changes
We create additional test cases for double support and make sure to pass in the double if its permitted (it's only allowed in buffers), and we also make sure to include double variants in the corresponding YAML files for quantization and dequantization.
ghstack-source-id: 290156616
@exported-using-ghexport

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

Contributor

facebook-github-bot commented Jun 13, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197


          Update on "[ET-VK][Ops] enabling double support for quantization and …

a553b06

…dequantization ops"

With the added double support in the layout template, this diff is enabling it as input/output for dequantization. Since there are limitations with how 64bit can be supported, the expectation is that IO be downgraded to 32bit

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

[ghstack-poisoned]

ahmtox pushed a commit that referenced this pull request


          [ET-VK][Ops] enabling double support for quantization and dequantizat…

21ea435

…ion ops

Pull Request resolved: #11553

# Context
Since we enabled the possibility for double support in an earlier diff, this enables double support for quantization and dequantization. Since there are limitations to how 64bit can be supported, the expectation is that IO is to be downgraded to 32bit.

# Changes
We create additional test cases for double support and make sure to pass in the double if its permitted (it's only allowed in buffers), and we also make sure to include double variants in the corresponding YAML files for quantization and dequantization.
ghstack-source-id: 290156616
@exported-using-ghexport

Differential Revision: [D76289197](https://our.internmc.facebook.com/intern/diff/D76289197/)

Contributor

facebook-github-bot commented Jun 13, 2025

This pull request was exported from Phabricator. Differential Revision: D76289197

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed fb-exported release notes: vulkan