feat: support aten.pixel_shuffle dynamo converter #2596
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
004810f to
e2c7081
Compare
gs-olive
reviewed
Jan 22, 2024
Comment on lines
+33
to
+54
| shape = input.shape | ||
| in_channels, in_height, in_width = shape[-3:] | ||
| out_channels = in_channels // (upscale_factor**2) | ||
| out_height = in_height * upscale_factor | ||
| out_width = in_width * upscale_factor | ||
| new_shape = shape[:-3] + ( | ||
| out_channels, | ||
| upscale_factor, | ||
| upscale_factor, | ||
| in_height, | ||
| in_width, | ||
| ) | ||
| reshaped_tensor = reshape( | ||
| ctx, target, source_ir, f"{name}_reshape1", input, new_shape | ||
| ) | ||
| rank = len(shape) | ||
| permute_shape = list(range(rank)) | ||
| permute_shape.insert(-2, rank) | ||
| permute_shape.insert(-1, rank + 1) | ||
| permuted_tensor = impl.permutation.permute( | ||
| ctx, target, source_ir, f"{name}_permute", reshaped_tensor, permute_shape | ||
| ) |
There was a problem hiding this comment.
What is the reason for the intermediate reshape and permute here?
There was a problem hiding this comment.
Thanks for the comment. This is because if we directly use reshape, the output shape will be correct, but the values won't. For example:
>>> x = torch.randn((8,2,3))
>>> x.reshape((2,4,6))
tensor([[[-1.4204, -0.4205, -1.3309, -1.1576, -1.8777, -1.3462],
[-0.5689, -0.1234, -0.5276, 1.2325, 0.2859, 0.4005],
[-0.7908, -0.4946, -0.7183, 0.2497, -0.6588, -1.0771],
[ 0.5446, -0.0980, 0.9309, -2.9004, 1.9834, -0.2377]],
[[ 1.3769, 0.5741, -0.3463, 0.6038, -0.9376, 1.1402],
[-0.1754, 0.4850, -3.5597, -0.5911, 1.7931, -1.7492],
[ 0.9871, -0.2294, 0.7445, -0.0991, 0.0278, 0.6699],
[-0.1543, -1.4414, -0.6795, -0.0403, 0.4620, -1.2007]]])
>>> torch.nn.functional.pixel_shuffle(x, upscale_factor=2)
tensor([[[-1.4204, -0.5689, -0.4205, -0.1234, -1.3309, -0.5276],
[-0.7908, 0.5446, -0.4946, -0.0980, -0.7183, 0.9309],
[-1.1576, 1.2325, -1.8777, 0.2859, -1.3462, 0.4005],
[ 0.2497, -2.9004, -0.6588, 1.9834, -1.0771, -0.2377]],
[[ 1.3769, -0.1754, 0.5741, 0.4850, -0.3463, -3.5597],
[ 0.9871, -0.1543, -0.2294, -1.4414, 0.7445, -0.6795],
[ 0.6038, -0.5911, -0.9376, 1.7931, 1.1402, -1.7492],
[-0.0991, -0.0403, 0.0278, 0.4620, 0.6699, -1.2007]]])
There was a problem hiding this comment.
Thanks for the example - that's very helpful
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Support
aten.pixel_shuffledynamo converter.Fixes #2594
Type of change
Checklist: