search

webmachinelearning / webnn-native

🧠⚙️ Standalone native implementation of the Web Neural Network API
Apache License 2.0
50 stars 21 forks source link

FP16 datatype support for chromium DML prototype #359

Closed mingmingtasd closed 1 year ago

mingmingtasd commented 2 years ago

We plan to support FP16 datatype especially for VPU acceleration on DML. After investigation, I found the https://learn.microsoft.com/en-us/windows/win32/dxcore/dxcore-source-code gives an example using dxcore to enumerate the devices (including GPUs and VPU), and it works. All of the available devices will be in the list: // From the factory, retrieve a list of all the Direct3D 12 Core Compute adapters. winrt::com_ptr<IDXCoreAdapterList> d3D12CoreComputeAdapters; But how to select the correct device VPU from the list is not clear. Do you know any flags/preferrence/properties that can indicate which device is VPU? Any suggestions? Thanks a lot! @fdwr

mingmingtasd commented 2 years ago

Sorry to bother you too early, please ignore this question temporarily, I will try FP16 on GPU firstly. @fdwr . We can discuss for VPU in WebNN-DirectML Teams Channel in the future.