yaoyaoding commented on issue #292: URL: https://github.com/apache/tvm-ffi/issues/292#issuecomment-3599450537
> One clarification - the problem of going directly all in to driver API is that we can't launch kernels via <<<>>> - is my understanding correct? There is no way to use `<<< >>>` syntax to launch a kernel given cubin. We have to use explicit runtime/driver API to launch the kernel (e.g., `cudaLaunchKernel(...)`). The main advantage of driver api is that we don't need the runtime dependency (i.e., `libcudart`). The advantage of runtime api (from my perspective) is that we don't need to take care of the cuda contexts explicitly, which is simpiler. > Do we want to test on Windows + cuda? I suppose we don't have such nodes :( We don't have windows + cuda ci node right now, sadly. I usually test locally on my windows machine used for gaming :) -- 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]
