[ExecuTorch][WebGPU] Support byte-packed BOOL storage by JCNTH · Pull Request #21598 · pytorch/executorch

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

JCNTH · 2026-08-05T19:54:48Z

Stack from ghstack (oldest at bottom):

Support exact byte-packed BOOL tensors

WebGPU storage and transfers require four-byte alignment while serialized BOOL tensors use one logical byte per element. The runtime now preserves exact BOOL dtype and logical byte counts while padding physical storage, uploads, and readback.

Key changes:

Compare output — packs guarded tail lanes for arbitrary nonzero lengths.
BOOL-to-fp32 — decodes packed values numerically instead of reinterpreting bytes.
_to_copy routing — mirrors Vulkan exact-dtype conversion intent in backends/vulkan/runtime/graph/ops/impl/View.cpp:183 while keeping BOOL separate from INT8 and UINT8.
Generated sources — refresh WGSL registry output and drift digests.

Logical tensor sizes and existing non-BOOL routes are unchanged.

Co-authored-with: Claude Code.

Differential Revision: D114936143

[ghstack-poisoned]

pytorch-bot · 2026-08-05T19:54:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21598

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f741fdc with merge base ad3a71f ():
💚 Looks good so far! There are no failures yet. 💚

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

github-actions · 2026-08-05T19:56:33Z

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.

**Support exact byte-packed BOOL tensors** WebGPU storage and transfers require four-byte alignment while serialized `BOOL` tensors use one logical byte per element. The runtime now preserves exact `BOOL` dtype and logical byte counts while padding physical storage, uploads, and readback. Key changes: - Compare output — packs guarded tail lanes for arbitrary nonzero lengths. - `BOOL`-to-fp32 — decodes packed values numerically instead of reinterpreting bytes. - `_to_copy` routing — mirrors Vulkan exact-dtype conversion intent in `backends/vulkan/runtime/graph/ops/impl/View.cpp:183` while keeping `BOOL` separate from `INT8` and `UINT8`. - Generated sources — refresh WGSL registry output and drift digests. Logical tensor sizes and existing non-`BOOL` routes are unchanged. Co-authored-with: Claude Code. Differential Revision: [D114936143](https://our.internmc.facebook.com/intern/diff/D114936143/) ghstack-source-id: 411044625 Pull-Request: #21598

Update
f741fdc
[ghstack-poisoned]

JCNTH temporarily deployed to cadence August 5, 2026 19:54 — with GitHub Actions Inactive

meta-claBot 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 Aug 5, 2026

JCNTH temporarily deployed to cadence August 5, 2026 20:31 — with GitHub Actions Inactive

psiddh approved these changes Aug 6, 2026
View reviewed changes

meta-codesyncBot merged commit 4e006e2 into gh/JCNTH/205/baseAug 7, 2026
190 of 198 checks passed

meta-codesyncBot deleted the gh/JCNTH/205/head branch August 7, 2026 17:22

meta-codesyncBottemporarily deployed to cherry-pick-bot August 7, 2026 17:22 Inactive

pytorchbot mentioned this pull request Aug 7, 2026
[ExecuTorch][WebGPU] Support byte-packed BOOL storage #21663
Merged