tqchen commented on code in PR #17216: URL: https://github.com/apache/tvm/pull/17216#discussion_r1697590272
########## src/relax/op/op.cc: ########## @@ -331,8 +331,133 @@ RELAY_REGISTER_OP("relax.call_tir") .set_attr<FNormalize>("FNormalize", NormalizeCallTIR) .set_attr<Bool>("FPurity", Bool(true)); -Expr MakeCallTIR(Expr func, Tuple args, Array<TensorStructInfo> out_sinfo_list, +static Array<TensorStructInfo> InferCallTIROutputStructInfo(Expr func, Tuple args, Review Comment: One thing to note is that this is not always possible to do such inference. Since it is possible to have tir functions like reshape, where the output shape is being explicitly specified via the destination. For the particular low-level call_tir op. I think it is safer to always ask for the sinfo, then explicitly checks the consistency to avoid error -- 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