Add SetBatchSize pass to MLIR converter #690
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
lgeiger
reviewed
Dec 3, 2021
Comment on lines
20
to
21
| // CHECK: %arg0: tensor<1x6xf32> | ||
| // CHECK: %arg1: tensor<1x4xf32> |
There was a problem hiding this comment.
Should we also run shape inference (-tf-shape-inference) in this test and verify that the shapes are propagated correctly? Alternatively we can also keep it like and simplify the test cases further to only test the the actual change in the input tensor size.
Comment on lines
28
to
29
| %0 = "tf.ReadVariableOp"(%arg2) {device = ""} : (tensor<!tf.resource<tensor<6xf32>>>) -> tensor<6xf32> | ||
| %1 = "tf.ReadVariableOp"(%arg3) {device = ""} : (tensor<!tf.resource<tensor<4x6xf32>>>) -> tensor<4x6xf32> |
There was a problem hiding this comment.
Let's remove tf.ReadVariableOp and tf.Identity ops to improve readability of the test case.
Co-authored-by: Lukas Geiger <lukas.geiger94@gmail.com>
lgeiger
approved these changes
Dec 3, 2021
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Outdated suggestions cannot be applied.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.
What do these changes do?
This adds a pass to the converter that sets the dynamic batchsize to 1 on the input layer.
When
batch_sizeis not explicitly set to 1, it will remain a?(wildcard) throughout all MLIR passes. Only at the final MLIR to Flatbuffer conversion stage, are all the?simply converted to1:https://github.com/tensorflow/tensorflow/blob/v2.7.0/tensorflow/compiler/mlir/lite/flatbuffer_export.cc#L844
This causes broken Reshape and Concatenation ops: the MLIR passes will correctly infer the Reshape output shape to something like
?x?x2, and then the flatbuffer exporter simply overrides both with1. Other artifactsStridedSliceops in the graph (which should've been folded out)The graphdef converter code has a
batch_size=1fix in Python, but the saved model converter did not have something like this yet.How Has This Been Tested?
MLIR filecheck tests have been added.