rewrite the TF lite binary conv. op #88
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
|
What was the speed difference that you got? 😄 |
Tombana
reviewed
Nov 7, 2019
Tombana
reviewed
Nov 7, 2019
| @pytest.mark.parametrize("kernel_size", [3, 5]) | ||
| @pytest.mark.parametrize("strides", [[1, 1], [2, 2]]) | ||
| @pytest.mark.parametrize("padding", ["VALID", "SAME"]) | ||
| def test_bconv2d_op_inference( |
There was a problem hiding this comment.
Nice new testing setup!
Once our zero-padding-correction-functor supports dilations, we should also add that here.
Tombana
reviewed
Nov 7, 2019
Comment on lines
305
to
307
| PaddingFunctor padding_functor; | ||
| padding_functor(params->batch, params->input_height, params->input_width, | ||
| params->channels_in, filter->data.f, |
There was a problem hiding this comment.
I think the padding functor requires special care when there are dilations. Now that we partially support dilations (since we use TFlite's im2col) we should add a TODO here or in the padding functor itself, and open a GitHub issue.
Tombana
reviewed
Nov 7, 2019
Tombana
reviewed
Nov 7, 2019
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Outdated suggestions cannot be applied.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.
closes #73
In this PR, I separate the impl. path of bconv op in TF and TF lite. Now the TF lite bconv op uses the default memory layout of TF lite for filters and also(dilated)im2col functions and TF lite internal types (MatrixParams, ...). This simplifies the integrating the Op to RUY framework and might give us also some performance improvements.