yaoyaoding commented on issue #292: URL: https://github.com/apache/tvm-ffi/issues/292#issuecomment-3599061875
Based on the above discussion (and some discussion in wechat group), here are the action items: * **API Control Macro:** Introduce a macro, such as `TVM_FFI_CUBIN_LAUNCHER_USE_DRIVER_API`, to explicitly control whether the **CUDA Driver API** or **CUDA Runtime API** is used for kernel launching. The default behavior (if the macro is unset) should remain the **Runtime API**. Maybe we should have another API-generic macro to check the error code (for runtime and driver API) for the return code of kernel launch. * **Driver API Example:** Add a new example demonstrating how to correctly use the cubin launcher when configured to use the **Driver API**. * **Documentation Update:** Update the relevant documentation to clearly mention and explain the purpose and usage of the new API control macro. * **Windows Support via `bin2c` (Optional):** Update the `embed_cubin.py` utility script to utilize `bin2c` to enable support for **Windows**. @oraluben please let me know if this sounds good to you. if so, I believe you can proceed to the implementation. Thanks! -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
