diff options
| author |  Blake Hechtman <blakehechtman@google.com> | 2018-09-20 15:42:23 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2018-09-20 15:58:45 -0700 |
| commit | 23a88ec5e913ba7086a9aef57875447ccf96e4b5 (patch) | |
| tree | 6e94f3f0fa29edb01c2cef29628e9a0d7f0b5034 /tensorflow/compiler/xla/service/shape_inference.cc | |
| parent | 1797aacbd8b910fb8c15577f66257b35af97cc1a (diff) | |
It is more computationally efficient to represent resize bilinear as a
depthwise convolution instead of a full convolution now that it exists in XLA.
PiperOrigin-RevId: 213896333
Diffstat (limited to 'tensorflow/compiler/xla/service/shape_inference.cc')
| -rw-r--r-- | tensorflow/compiler/xla/service/shape_inference.cc | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/tensorflow/compiler/xla/service/shape_inference.cc b/tensorflow/compiler/xla/service/shape_inference.cc index 74bdf2a2e3..7194b2cafd 100644 --- a/tensorflow/compiler/xla/service/shape_inference.cc +++ b/tensorflow/compiler/xla/service/shape_inference.cc @@ -1665,10 +1665,11 @@ ShapeInference::InferDegenerateDimensionBroadcastShape(HloOpcode operation, if (input_features != kernel_input_features * feature_group_count) { return InvalidArgument( "Expected LHS feature dimension (value %d) to match RHS " - "input feature dimension * feature_group_count (value %d); " + "input feature dimension * feature_group_count (value %d * %d = %d); " "got <conv>(%s, %s)\n" "Dimension numbers: {%s}.", - input_features, kernel_input_features * feature_group_count, + input_features, kernel_input_features, feature_group_count, + kernel_input_features * feature_group_count, ShapeUtil::HumanString(lhs), ShapeUtil::HumanString(rhs), dnums.DebugString()); } |