You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2020/09/14 01:49:31 UTC
[GitHub] [incubator-tvm] wjliu1998 opened a new pull request #6468: add aten::pixel_shuffle implementation (#6328)
wjliu1998 opened a new pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468
Following the implementation of aten::pixel_shuffle in pytorch. The implementation of pixel_shuffle has been added in pytorch frontend
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] wjliu1998 commented on pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
wjliu1998 commented on pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#issuecomment-691781602
> thanks, please add a test.
Thanks! The test has been added!
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] wjliu1998 commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
wjliu1998 commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487642925
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -1230,6 +1230,42 @@ 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))
+ axes = list(range(ndims))
+ num_inputs = len(inputs)
+ oc = c // upscale_squared
+ oh = h * upscale_factor
+ ow = w * upscale_factor
+
+ new_shape = [b, oc, upscale_factor, upscale_factor, h, w]
+ out_shape = [b, oc, oh, ow]
+
+ data = _op.transform.reshape(data, new_shape)
+ axes = [0, 1, 4, 2, 5, 3]
Review comment:
The comment has been added.
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487641161
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -1230,6 +1230,42 @@ 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))
+ axes = list(range(ndims))
+ num_inputs = len(inputs)
+ oc = c // upscale_squared
+ oh = h * upscale_factor
+ ow = w * upscale_factor
+
+ new_shape = [b, oc, upscale_factor, upscale_factor, h, w]
+ out_shape = [b, oc, oh, ow]
+
+ data = _op.transform.reshape(data, new_shape)
+ axes = [0, 1, 4, 2, 5, 3]
Review comment:
ok, please add your explanation above as a comment (so that people won't get confused).
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
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).
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] wjliu1998 commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
wjliu1998 commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487699190
##########
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:
Thanks! The code has been deleted
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] wjliu1998 commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
wjliu1998 commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487824170
##########
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:
Thank you for pointing out that. It seems `ndims = len(_infer_shape(data, prelude.mod))` is enough
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi merged pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi merged pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi commented on pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#issuecomment-691802319
@wjliu1998 lint is complaining
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
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 just ` 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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487630990
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -1230,6 +1230,42 @@ 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))
+ axes = list(range(ndims))
+ num_inputs = len(inputs)
+ oc = c // upscale_squared
+ oh = h * upscale_factor
+ ow = w * upscale_factor
+
+ new_shape = [b, oc, upscale_factor, upscale_factor, h, w]
+ out_shape = [b, oc, oh, ow]
+
+ data = _op.transform.reshape(data, new_shape)
+ axes = [0, 1, 4, 2, 5, 3]
Review comment:
Can you tell me what are these hardcoded axes?
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi commented on pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#issuecomment-692109522
Thanks @wjliu1998
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] wjliu1998 commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
wjliu1998 commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487640206
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -1230,6 +1230,42 @@ 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))
+ axes = list(range(ndims))
+ num_inputs = len(inputs)
+ oc = c // upscale_squared
+ oh = h * upscale_factor
+ ow = w * upscale_factor
+
+ new_shape = [b, oc, upscale_factor, upscale_factor, h, w]
+ out_shape = [b, oc, oh, ow]
+
+ data = _op.transform.reshape(data, new_shape)
+ axes = [0, 1, 4, 2, 5, 3]
Review comment:
The original shape order is [b, oc, upscale_factor, upscale_factor, h, w], the hardcoded axes transpose the order to [b, oc, h, upscale_factor, w, upscale_factor] for further reshape
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi commented on pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#issuecomment-691769965
thanks, please add a test.
----------------------------------------------------------------
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:
users@infra.apache.org
[GitHub] [incubator-tvm] masahi commented on a change in pull request #6468: add aten::pixel_shuffle implementation (#6328)
Posted by GitBox <gi...@apache.org>.
masahi commented on a change in pull request #6468:
URL: https://github.com/apache/incubator-tvm/pull/6468#discussion_r487684233
##########
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:
It seems you copy pasted the above code from `_transpose()`. That one is a legacy code we don't want to repeat. Please leave only what is necessary.
----------------------------------------------------------------
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:
users@infra.apache.org