Add more vlogging to op level estimator.
authorYao Zhang <yaozhang@google.com>
Thu, 8 Feb 2018 23:06:54 +0000 (15:06 -0800)
committerTensorFlower Gardener <gardener@tensorflow.org>
Thu, 8 Feb 2018 23:15:01 +0000 (15:15 -0800)
PiperOrigin-RevId: 185058989

tensorflow/core/grappler/costs/op_level_cost_estimator.cc

index 1af9738..76db1af 100644 (file)
@@ -457,10 +457,15 @@ OpLevelCostEstimator::ConvolutionDimensionsFromInputs(
     const TensorShapeProto& original_image_shape,
     const TensorShapeProto& original_filter_shape, const OpInfo& op_features,
     bool* found_unknown_shapes) {
+  VLOG(2) << "op features: " << op_features.DebugString();
+  VLOG(2) << "Original image shape: " << original_image_shape.DebugString();
+  VLOG(2) << "Original filter shape: " << original_filter_shape.DebugString();
   auto image_shape =
       MaybeGetMinimumShape(original_image_shape, 4, found_unknown_shapes);
   auto filter_shape =
       MaybeGetMinimumShape(original_filter_shape, 4, found_unknown_shapes);
+  VLOG(2) << "Image shape: " << image_shape.DebugString();
+  VLOG(2) << "Filter shape: " << filter_shape.DebugString();
 
   int x_index, y_index, channel_index;
   const string& data_format = GetDataFormat(op_features);