oraluben commented on code in PR #300: URL: https://github.com/apache/tvm-ffi/pull/300#discussion_r2633416386
########## include/tvm/ffi/extra/cuda/unify_api.h: ########## @@ -0,0 +1,183 @@ +/* Review Comment: After reviewing the code again, I believe we still need a unified compatibility layer. Currently, we have `CUDADeviceGuard`, which works as `aten::DeviceGuard` - this could also be implemented using the driver API. On the other hand, if `CUDADeviceGuard` only works with the runtime API while `cubin_launcher` can switch between runtime/driver APIs, it introduces some inconsistency. I suggest keeping a single internal compatibility header (perhaps renamed to `_unify_api.h` to reflect its internal use). For APIs like `CUDADeviceGuard` that are beyond the current PR’s scope, we can temporarily mark them as “not implemented” in `_unify_api.h` and throw an error when driver API is requested. Looking forward to your thoughts. cc @cyx-6 @junrushao @tqchen -- 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]
