Additional diagnostics for DML failure path

[adrastogi]

Description

In DmlGraphFusionHelper::ExecuteReusableCommandList, after ExecuteCommandList fails:

• Broaden the failure branch from just DXGI_ERROR_DEVICE_REMOVED to also catch DEVICE_HUNG, DEVICE_RESET, and
  DRIVER_INTERNAL_ERROR.
• Query GetDeviceRemovedReason on both the DML and D3D12 devices (matching the pattern in DmlCommandRecorder.cpp).
• Throw via ORT_THROW_HR_MSG with a clear message that names the failure as a TDR / device-removal event, calls out and includes all three HRESULTs for triage. Preserves the prior thrown-HRESULT for the existing DEVICE_REMOVED path

Motivation and Context

While investigating a WebNN sample failure on Chrome running Stable Diffusion 1.5 on an AMD Radeon 860M iGPU, ORT 1.23.4 surfaced this error:

DmlGraphFusionHelper.cpp(1078) ... 887A0006 The GPU will not respond to more commands, most likely because of an invalid command passed by the calling application.

0x887A0006 is DXGI_ERROR_DEVICE_HUNG. The text "...invalid command passed by the calling application" seems to be the FormatMessage string for that HRESULT.

The pre-existing code in DmlGraphFusionHelper::ExecuteReusableCommandList only special-cased DXGI_ERROR_DEVICE_REMOVED, so for DEVICE_HUNG / DEVICE_RESET / DRIVER_INTERNAL_ERROR HRESULTs the user just got the raw message. I wanted to add a little more diagnostic information to this.

[Copilot]

Pull request overview

Adds richer diagnostics in DmlGraphFusionHelper::ExecuteReusableCommandList when ExecuteCommandList fails with a device-lost/removal-class HRESULT, so that TDR-induced failures (DEVICE_HUNG, DEVICE_RESET, DRIVER_INTERNAL_ERROR) no longer surface only as the misleading "invalid command" FormatMessage text.

Changes:

• Broadens the failure branch from DXGI_ERROR_DEVICE_REMOVED only to also include DEVICE_HUNG, DEVICE_RESET, and DRIVER_INTERNAL_ERROR.
• Queries GetDeviceRemovedReason on both the DML and D3D12 devices and selects the most specific HRESULT to throw with.
• Uses ORT_THROW_HR_MSG to emit a clear message naming TDR/device-removal as the likely cause and including the original HRESULT and both removed-reason HRESULTs.

Show a summary per file

File	Description
onnxruntime/core/providers/dml/DmlExecutionProvider/src/DmlGraphFusionHelper.cpp	Expands the device-lost detection branch and emits a more diagnostic error message including DML/D3D12 GetDeviceRemovedReason values.

Copilot's findings

Tip

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

• Files reviewed: 1/1 changed files
• Comments generated: 0

[fdwr]

Note you might also get more info via the DML debug layer. It's probably already installed you have Visual Studio installed, but otherwise: https://learn.microsoft.com/en-us/windows/ai/directml/dml-debug-layer. Then:

• Start / Run / dxcpl.exe
• Add your process .exe path to the list.
• Then Force debug messages on.
• You should see additional output in the Visual Studio Output window, a different debugger of your choice, or via DebugView.

[fdwr]

👍