icemelon9 commented on a change in pull request #5429:
URL: https://github.com/apache/incubator-tvm/pull/5429#discussion_r422452792
##########
File path: python/tvm/relay/op/_transform.py
##########
@@ -189,10 +192,9 @@ def _reshape_shape_func(data_shape, newshape, ndim):
out[infer_idx] = old_size // new_size
return out
-@_reg.register_shape_func("reshape", False)
+@_reg.register_shape_func("reshape", True)
Review comment:
I still have a bit concern here because if `newshape` is not a tensor
input, we can still treat `reshape` as data independent shape func, which means
that we can fuse reshape with other ops. But now, if the output of `reshape` is
dynamic, it can no longer be fused any more.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
