[webgpu] Set is_channels_last to true by default in ComputeMatMul

[Jiawei-Shao]

This patch sets is_channels_last to true by default in the parameter
of ComputeMatMul and ignores it in UseSplitK when there is no
bias.

[Jiawei-Shao]

@qjia7 PTAL, thanks!

[Copilot]

Pull request overview

This PR updates the WebGPU MatMul/Conv Split-K plumbing to treat is_channels_last as an optional signal (only meaningful when bias is present), reducing ambiguity for bias-less call sites and enabling the Split-K MatMul path.

Changes:

• Change is_channels_last parameters to std::optional<bool> across WebGPU MatMul helpers and Split-K configuration.
• Update shader-generation helpers (MatMulWriteFnSourceForMatMul) to accept std::optional<bool> and std::string_view.
• Adjust WebGPU Conv and WebGPU BERT Attention call sites to only pass is_channels_last when bias is used.

Reviewed changes

Copilot reviewed 11 out of 13 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
onnxruntime/core/providers/webgpu/webgpu_utils.h	Updates Split-K config API to accept optional is_channels_last.
onnxruntime/core/providers/webgpu/webgpu_utils.cc	Updates Split-K gating logic to only consult is_channels_last when provided.
onnxruntime/core/providers/webgpu/vendor/intel/math/matmul.h	Simplifies Intel MatMul subgroup program interface (bias removed).
onnxruntime/core/providers/webgpu/vendor/intel/math/matmul.cc	Updates Intel subgroup shader generation for new MatMul write helper signature.
onnxruntime/core/providers/webgpu/nn/conv.cc	Passes is_channels_last only when Conv MatMul path includes bias.
onnxruntime/core/providers/webgpu/math/matmul_packed.h	Updates MatMul program API to carry optional is_channels_last.
onnxruntime/core/providers/webgpu/math/matmul_packed.cc	Treats bias presence as is_channels_last_.has_value() and threads optional into shader generation.
onnxruntime/core/providers/webgpu/math/matmul.h	Makes ComputeMatMul accept optional is_channels_last with a default of {}.
onnxruntime/core/providers/webgpu/math/matmul.cc	Enforces consistency between bias presence and is_channels_last engagement; wires optional into Split-K selection and MatMul program creation.
onnxruntime/core/providers/webgpu/math/gemm_utils.h	Updates MatMul write helper signature to optional is_channels_last and string_view.
onnxruntime/core/providers/webgpu/math/gemm_utils.cc	Implements optional-aware bias handling in MatMul write helper.
onnxruntime/core/providers/webgpu/math/gemm_packed.cc	Updates Split-K selection call site to pass std::nullopt for is_channels_last.
onnxruntime/contrib_ops/webgpu/bert/attention.cc	Builds MatMul input list conditionally and passes optional is_channels_last only with bias.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

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

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Jiawei-Shao]

Hi @guschmue, could you take a look at this PR?

[guschmue]

/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline

[azure-pipelines]

Azure Pipelines successfully started running 4 pipeline(s).

[Jiawei-Shao]

The errors are not related to this PR:

The self-hosted runner lost communication with the server. Verify the machine is running and has a healthy network connection. Anything in your workflow that terminates the runner process, starves it for CPU/Memory, or blocks its network access can cause this error.

[Jiawei-Shao]

The failures on DirectML CI are not related to this PR.

[guschmue]

/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline

[azure-pipelines]

Azure Pipelines successfully started running 4 pipeline(s).