Recent Commits to gpuweb:main

Set adapter.state to "consumed" (#10470)

2026-09-23T17:49:33Z

Set adapter.state to "consumed" (#10470)

Prose already says:
> if a device is returned successfully, the adapter becomes
["consumed"](https://www.w3.org/TR/webgpu/#dom-adapter-state-consumed).

This PR updates steps to reflect that. CTS already expects this and dawn
implements this.

Fixes #10469

Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>

Serve PR previews through the GPUWeb Pages viewer (#10961)

2026-09-21T22:26:54Z

Serve PR previews through the GPUWeb Pages viewer (#10961)

Replace links with commit-pinned Pages viewer links for WebGPU, WGSL,
Explainer, and Correspondence. To avoid misreferences, changed HEAD to
refs/heads/main and added --exit-code to avoid silent fails. After
merge, new comments should work with this viewer than CDN. Also replaced
CDN_ prefix with PREVIEW_ prefix to reflect restricted nature of this
new approach compared to before. Retrospective comment updates will be
handled separately.

[bindless] Update proposal for WGSL aliasing extension. (#10345)

2026-09-18T19:35:36Z

[bindless] Update proposal for WGSL aliasing extension. (#10345)

Update the bindless proposal to state that the WGSL alias relaxation
extension is required for bindless support.

Fixes #5581

[wgsl] Add proposal for WGSL namespaces. (#7310)

2026-09-16T22:27:08Z

[wgsl] Add proposal for WGSL namespaces. (#7310)

Add a proposal for namespaces in WGSL.

Issue #777

Spec changes for atomic vec2u (#5610)

2026-09-15T22:52:04Z

Spec changes for atomic vec2u (#5610)

Original approved proposal

https://github.com/gpuweb/gpuweb/blob/main/proposals/atomic-64-min-max.md

This feature is extremely limited to side effect/mutation in storage
buffers.
There is no return value so no way of creating synchronize-with
sequencing.

---------

Co-authored-by: Peter McNeeley <petermcneeley@google.com>
Co-authored-by: Alan Baker <alanbaker@google.com>
Co-authored-by: David Neto <dneto@google.com>
Co-authored-by: Jim Blandy <jimb@red-bean.com>

Add `texture-compression-unaligned` feature to relax texture size blo…

2026-09-01T22:17:27Z

Add `texture-compression-unaligned` feature to relax texture size block alignment requirement. (#6312)

This PR updates the spec with the feature discussed in
https://github.com/gpuweb/gpuweb/issues/2006

Update to the validation test suite is in the following PR:
https://github.com/gpuweb/cts/pull/4676

---------

Co-authored-by: Kai Ninomiya <kainino@chromium.org>

wgsl: swizzle view support indexing: delete stale note (#7065)

2026-08-31T18:56:42Z

wgsl: swizzle view support indexing: delete stale note (#7065)

Explicitly list swizzle views in the cases listed for indexing
expressions, and for in-bounds index.

Issue: #6941

wgsl: support swizzle assignment (#5268)

2026-08-25T00:04:48Z

wgsl: support swizzle assignment (#5268)

Add language extension 'swizzle_assignment'

Example:

   var v:vec4u;
   v.yz = vec2(1,2);

Issue: #737

---------

Co-authored-by: Jim Blandy <jimb@red-bean.com>

wgsl: convert markdown blockquote to explicit HTML (#6878)

2026-08-24T23:50:07Z

wgsl: convert markdown blockquote to explicit HTML (#6878)

Avoids a warning in newer bikeshed versions.

Adds the `snorm10-10-10-2` vertex format to the WebGPU specification …

2026-08-20T01:43:35Z

Adds the `snorm10-10-10-2` vertex format to the WebGPU specification and WebIDL definition. (#6491)

* Adds `"snorm10-10-10-2"` to `enum GPUVertexFormat`.
* Adds `snorm10-10-10-2` row to the Vertex Formats table (signed
normalized, 4 components, 4 bytes,`vec4<f32>`).

Fixes #6292.

[wgsl] Integrate the fragment_depth proposal into the spec (#6299)

2026-08-17T17:03:30Z

[wgsl] Integrate the fragment_depth proposal into the spec (#6299)

Update the WGSL spec to include the fragment_depth proposal text.

Closes #5342

---------

Co-authored-by: alan-baker <alanbaker@google.com>
Co-authored-by: Jim Blandy <jimb@red-bean.com>

[wgsl] Add a note about alignments (#6412)

2026-08-14T15:27:08Z

[wgsl] Add a note about alignments (#6412)

* @align can only satisfy the value if the variable is aligned enough
  * it is satisfied relative to the start of the variable

[wgsl] Auto-enable dependent features (#6366)

2026-08-14T15:17:26Z

[wgsl] Auto-enable dependent features (#6366)

Fixes #6347

* Auto-enable subgroups when subgroup_size_control is enabled
* Update subgroup_matrix proposal for similar enablement of subgroups

[subgroup-matrix] Add stride restriction (#6365)

2026-08-12T19:10:57Z

[subgroup-matrix] Add stride restriction (#6365)

[editorial] Editing on Feature Index (#6351)

2026-08-12T17:25:42Z

[editorial] Editing on Feature Index (#6351)

Editorial tangent to #6344

- Make the Feature Index non-normative - it's a summary of stuff in
other parts of the spec.
- Make the list of API changes a list, so that other notes about the
feature are visually distinct from them.
- Note this still includes dependency info (but not reverse-dependency
info) as otherwise it's hard to understand.

[editorial/bugfix] Dedup feature dependencies, fix missing auto-enabl…

2026-08-10T20:35:56Z

[editorial/bugfix] Dedup feature dependencies, fix missing auto-enablements (#6344)

Previously this was defined separately in "Adapter Capability
Guarantees", "a new device", and the "Feature Index".

This fixes a bunch of stuff that was missing:
- "a new device" missing:
  - `texture-compression-bc-sliced-3d`&rarr;`texture-compression-bc`
  - `texture-compression-astc-sliced-3d`&rarr;`texture-compression-astc`
- "Adapter Capability Guarantees" missing:
-
`texture-formats-tier2`&rarr;`texture-formats-tier1`&rarr;`rg11b10ufloat-renderable`

Also moves the definition of `GPUFeatureName` to the "Optional
Capabilities" section to be consistent with limits.

[wgsl]: clarify where attributes can be used on function params and r…

2026-08-06T13:52:08Z

[wgsl]: clarify where attributes can be used on function params and r… (#6349)

…eturns

Avoid listing IO attributes in the user-declared function section. The
IO attributes are already limited to being applied to entry point
params, entry point return type specification, or members of structs. So
in the user-defined section we don't need to list the IO attributes
again.

Instead, mention that *entry point* params and return types may have
attributes to specify them as shader stage inputs and outputs, then link
to those terms and let the other section do the work of listing the IO
attributes.

Fixed: #5552

clarify "T is concrete numeric scalar or concrete numeric vector" (#6…

2026-08-05T20:31:50Z

clarify "T is concrete numeric scalar or concrete numeric vector" (#6348)

Fixed: #6346

[subgroup-matrix] Relax load/store array requirements (#6345)

2026-08-05T20:31:14Z

[subgroup-matrix] Relax load/store array requirements (#6345)

* Relax the allowed array types for use with subgroupMatrixLoad and
subgroupMatrixStore
  * allow any numeric scalar or vector type as the array element type
* Fix some typos

[buffer_view] Expand proposal motivation (#6342)

2026-07-27T20:55:08Z

[buffer_view] Expand proposal motivation (#6342)