Request for Guidance on Building a TensorRT Plugin (Windows) and Resolving MSVC build errors

[Lzf2640487101]

thank you for your great contributions to this project.

I am currently attempting to deploy my module on Windows and build a Windows-based TensorRT Plugin DLL using spconv, but I have encountered persistent compilation errors. Is there any established documentation or recommended workflow for achieving this successfully?

Specifically, when using spconv v2.3.6, the tensorview headers (such as bfloat16.h) cause explicit namespace conflicts (e.g., std::std) under MSVC 2022.

For my particular use case — I only need to build a pure TensorRT Plugin (IPluginV2DynamicExt) for the sm_89 like architecture on Windows — is there a supported CMake flag to disable non-essential data types (like bfloat16) to avoid these conflicts and streamline the build?

Additionally, I would like to ask:

As a potential workaround, would it be possible to use preprocessor macros (e.g., -DTV_CUDA_NO_FLOAT16=ON) to conditionally exclude conflicting or unnecessary parts of tensorview during compilation?

If such a CMake option does not currently exist, would it be feasible to request a minimal build target tailored for TensorRT Plugin deployment, allowing users to disable unnecessary features?

I believe this kind of lightweight configuration would be very valuable for deployment-focused scenarios. Any guidance or suggestions would be highly appreciated.

Thank you in advance for your support.