lixiaoquan opened a new issue, #13480: URL: https://github.com/apache/tvm/issues/13480
Thanks for participating in the TVM community! We use https://discuss.tvm.ai for any general usage questions and discussions. The issue tracker is used for actionable items such as feature proposals discussion, roadmaps, and bug tracking. You are always welcomed to post on the forum first :smile_cat: Issues that are inactive for a period of time may get closed. We adopt this policy so that we won't lose track of actionable issues that may fall at the bottom of the pile. Feel free to reopen a new one if you feel there is an additional problem that needs attention when an old one gets closed. Consider this case: def test_bilayout_with_any(): bilayout = tvm.tir.bijective_layout("NCHW", "NHWC") assert isinstance(bilayout, tvm.tir.BijectiveLayout) dst_shape = bilayout.forward_shape((Any(), 32, 7, Any())) print(dst_shape) # output is [?, 7, ?, ?] src_shape = bilayout.backward_shape(dst_shape) print(src_shape) # output is [?, ?, ?, ?] ### Expected behavior forward_shape's output should be [?, 7, ?, 32] backward_shape's output should be [?, 32, 7, ?] ### Actual behavior forward_shape's output should be [?, 7, ?, ?] backward_shape's output should be [?, ?, ?, ?] As you can see, all dims become Any(). This is caused by https://github.com/apache/tvm/blob/main/src/tir/ir/data_layout.cc#L372 If original dim is Any(), dim in transformed shape will be Any(). I'm not sure this is the expected behavior. Is there any reason behind this code? It is introduced by https://github.com/apache/tvm/pull/4179 @kevinthesun Could you please comment? ### Environment Latest TVM ### Steps to reproduce Please check the test above ### Triage Please refer to the list of label tags [here](https://github.com/apache/tvm/wiki/Issue-Triage-Labels) to find the relevant tags and add them below in a bullet format (example below). * needs-triage -- 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]
