In order to help to progress with A2, I can think of two solutions :
A2.1 : We could register a sub_target string (similiar to a mtriple in LLVM) and a decoder (e.g. : _register_sub_target_decoder(ultra_trail_target_decoder(str)) --> AttrDict/Dict. We could register the attr_dict as uma_attrs to the templated UMA target registration. A2.2 : We could directly register an attr_dict. i.e. _register_target_attr(AttrDict/Dict) I have a slight preference for A2.1 as it will only expose a string to the user in the actual compilation run. PS : @Mousius highlights the motivation behind making the target attributes discoverable in this new recent RFC : https://github.com/apache/tvm-rfcs/blob/1901b7361891fc7235b3bb8e80c89130cfe2c91d/rfcs/0070-target-preprocessing.md -- Reply to this email directly or view it on GitHub: https://github.com/apache/tvm-rfcs/pull/60#issuecomment-1123579692 You are receiving this because you are subscribed to this thread. Message ID: <apache/tvm-rfcs/pull/60/c1123579...@github.com>
