Implement WebGPU subgroup-size-control infrastructure

[Jie Chen (jchen10)]

Add support for the WebGPU subgroup-size-control extension, which allows compute pipelines to request a fixed subgroup size via WGSL @subgroup_size(N) attribute. As an example, the MatMulNBits kernel has been enforced to use a subgroup size of 32 on Intel.

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Jie Chen (jchen10)]

Jiawei Shao (@Jiawei-Shao) PTAL

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Jie Chen (jchen10)]

Jiajia Qin (@qjia7) Hariharan Seshadri (@hariharans29) PTAL

[Copilot]

Pull request overview

This PR adds support in the WebGPU EP for the subgroup-size-control feature so WGSL compute entry points can request a fixed subgroup size via @subgroup_size(N), and wires that through program configuration, shader generation, and cache keys. It also applies the new capability to enforce subgroup size 32 for the WebGPU MatMulNBits subgroup-matrix path on Intel adapters.

Changes:

• Request wgpu::FeatureName::SubgroupSizeControl during device creation when available.
• Add ProgramBase::SetSubgroupSize() / SubgroupSize() and include subgroup size in shader generation (enable subgroup_size_control; + @subgroup_size(...)) and the program cache key.
• Apply Intel-specific subgroup size 32 selection in subgroup_matrix_matmul_nbits.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
onnxruntime/core/providers/webgpu/webgpu_context.cc	Requests SubgroupSizeControl feature when the adapter supports it.
onnxruntime/core/providers/webgpu/shader_helper.cc	Validates subgroup-size requests and emits WGSL feature enable + @subgroup_size.
onnxruntime/core/providers/webgpu/program.h	Adds subgroup-size API surface to ProgramBase.
onnxruntime/core/providers/webgpu/program.cc	Implements SetSubgroupSize and initializes the new member.
onnxruntime/core/providers/webgpu/program_cache_key.cc	Extends program cache key to include subgroup-size overrides.
onnxruntime/contrib_ops/webgpu/quantization/subgroup_matrix_matmul_nbits.cc	Enforces subgroup size 32 on Intel when supported.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.