masahi commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487717415
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -1230,6 +1230,45 @@ def _impl(inputs, input_types):
return _impl
+def _pixel_shuffle(prelude):
+ def _impl(inputs, input_types):
+ data = inputs[0]
+ upscale_factor = inputs[1]
+ upscale_squared = upscale_factor * upscale_factor
+ b, c, h, w = _infer_shape(data)
+ assert c % upscale_squared == 0, \
+ "input channel should be divisible by square of upscale_factor"
+
+ import torch
+ if isinstance(data, _expr.Expr):
+ ndims = len(_infer_shape(data, prelude.mod))
+ elif isinstance(data, list):
+ ndims = data
+ elif isinstance(data, (torch.Tensor, np.ndarray)):
+ ndims = _infer_shape(data)
+ else:
+ msg = "Data type %s could not be parsed in transpose op" %
(type(data))
+ raise AssertionError(msg)
+
+ if isinstance(data, tvm.runtime.NDArray):
+ ndims = len(_infer_shape(data))
Review comment:
I mean, you have three `if isinstance(...)` right? You shouldn't need
`isintance`, just get ndims directly (figure out which code path actually hit
in your test). I guess all you need is jut ` ndims = data`.
----------------------------------------------------------------
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]
