Lunderberg commented on PR #17216: URL: https://github.com/apache/tvm/pull/17216#issuecomment-2260649071
> For example, the above code is a valid tir call, but needs the output sinfo to be explicitly specified. Because we have such cases, and `call_tir` is a lower level function, it is safer to always ask for sinfo, but checks its consistency with the corresponding prim_func signature if needed That's a good point, and I agree that we should always be able to explicitly specify the output struct info, as output tensor shapes in TIR may define symbolic shapes. However, I don't think it should a required argument. I've added a new test case, based on your example with `reshape`, to validate the behavior when the output shape cannot be inferred. While the initial implementation did identify this failure and throw an error, the error message wasn't ideal. I've added an earlier check for non-inferable output shapes, so that the error message can direct the user to provide the `out_sinfo` field. Does the udpated check/error messages address your concerns for this PR? -- 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]
