diff options
author | 2018-09-04 11:17:30 -0700 | |
---|---|---|
committer | 2018-09-04 11:41:30 -0700 | |
commit | 5d183ab7fc7b82f1dea0b9fa9c6412c39ade15a1 (patch) | |
tree | 79a4f6fcf270617fc56082702b0209240425ae8c /tensorflow/compiler/xla/service/shape_inference.cc | |
parent | 9ae8214229960c634c9f82c00f2c0df287c27a9d (diff) |
[XLA] Make kConvolution, kDot HLO attributes mandatory
HLO transformations would forget to propagate the feature depth attribute.
Making these attributes mandatory, while slightly less convenient for tests,
makes HLO transformations more robust.
PiperOrigin-RevId: 211490160
Diffstat (limited to 'tensorflow/compiler/xla/service/shape_inference.cc')
-rw-r--r-- | tensorflow/compiler/xla/service/shape_inference.cc | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/tensorflow/compiler/xla/service/shape_inference.cc b/tensorflow/compiler/xla/service/shape_inference.cc index 2611749862..7758a5dd4d 100644 --- a/tensorflow/compiler/xla/service/shape_inference.cc +++ b/tensorflow/compiler/xla/service/shape_inference.cc @@ -1552,8 +1552,8 @@ ShapeInference::InferDegenerateDimensionBroadcastShape(HloOpcode operation, } /* static */ StatusOr<Shape> ShapeInference::InferConvolveShape( - const Shape& lhs, const Shape& rhs, const Window& window, - const ConvolutionDimensionNumbers& dnums, int64 feature_group_count) { + const Shape& lhs, const Shape& rhs, int64 feature_group_count, + const Window& window, const ConvolutionDimensionNumbers& dnums) { TF_RETURN_IF_ERROR(ExpectArray(lhs, "lhs of convolution")); TF_RETURN_IF_ERROR(ExpectArray(rhs, "rhs of convolution")); |