lhutton1 commented on PR #104: URL: https://github.com/apache/tvm-rfcs/pull/104#issuecomment-1779468180
Regarding the changes required to support scalability in the data type, I've been prototyping adding a new `scalable_` attribute to `DataType` that wraps `DLDataType`. However, I've ran into what I believe is an issue when accessing data types at compile-time across the FFI boundary between python and c++. `TVMArgValue` and `TVMRetValue` may have a value stored as a `DLDataType`. Storing a scalable `DataType` as a `DLDataType` will mean that we lose information about the scalability (assuming we don't want to alter DLPack, or use the negative lanes < -1 approach). For the limited number of test cases I've written, I've worked around this limitation by forcing `DataType` to be stored as a string across the boundary. But this feels a bit wrong. I wonder if there could be something I've missed here or if there are any other suggestions? Are there any rules for using `string`, `DataType` and `DLDataType` interchangeably? -- 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: commits-unsubscr...@tvm.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
