chunit-quic commented on code in PR #13402:
URL: https://github.com/apache/tvm/pull/13402#discussion_r1030068767
##########
python/tvm/relay/frontend/common.py:
##########
@@ -997,3 +1002,167 @@ def try_resolve_var_to_const(x, graph_params):
return _op.const(value, dtype)
return x
+
+
+class _SpanFiller(ExprMutator):
Review Comment:
I see. Currently we don't have plan to migrate them to a C++ pass. There are
two reasons.
1. First of all, considering the profiling result in the preRFC, the
frontend conversion time is not increased too much.
2. Second, the fill() and \_\_init\_\_() function heavily interact with
different frontends. We would like to keep the flexibility to modify them via
python script quickly. For example, in the [min_max_common
](https://github.com/apache/tvm/blob/b2058f4dd2e0ae1fc5ab51ac9f84b372a389a65a/python/tvm/relay/frontend/pytorch.py#L275)conversion
of PyTorch, it might end up being a None, which is not common in other
frontends.
However, if the performance of frontend conversion is significantly
decreased by SpanFiller in some cases, we can arrange a patch to move the
mutator part (visitor funcions) to C++ pass. :)
--
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
