Home | History | Annotate | Download | only in kernels
      1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
      2 
      3 Licensed under the Apache License, Version 2.0 (the "License");
      4 you may not use this file except in compliance with the License.
      5 You may obtain a copy of the License at
      6 
      7     http://www.apache.org/licenses/LICENSE-2.0
      8 
      9 Unless required by applicable law or agreed to in writing, software
     10 distributed under the License is distributed on an "AS IS" BASIS,
     11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     12 See the License for the specific language governing permissions and
     13 limitations under the License.
     14 ==============================================================================*/
     15 
     16 #ifndef TENSORFLOW_CORE_KERNELS_SPACETODEPTH_OP_H_
     17 #define TENSORFLOW_CORE_KERNELS_SPACETODEPTH_OP_H_
     18 // Functor definition for XentOp, must be compilable by nvcc.
     19 
     20 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
     21 #include "tensorflow/core/framework/tensor_types.h"
     22 #include "tensorflow/core/util/tensor_format.h"
     23 
     24 namespace tensorflow {
     25 namespace functor {
     26 
     27 // Functor used by SpaceToDepthOp to do the computations.
     28 // Implements a family of Space to Depth transforms for a 4D 'input' tensor
     29 // to a 4D 'output' tensor, both tensors use type 'T' and layout 'data_format'.
     30 // These transforms divide the vertical and horizontal image sizes by
     31 // 'block_size', and multiply the depth dimension size by
     32 // (block_size * block_size). The offset within each block_size * block_size
     33 // patch within the image is combined with the input channel index to form
     34 // the output channel index, with the Y, X coordinates within each block of
     35 // the input image used as the high order component of the output channel.
     36 // e.g. for data_format = NHWC:
     37 //      Each element in the input tensor can be specified via 6 coordinates,
     38 //      ordered by decreasing memory layout significance as:
     39 //      n,oY,bY,oX,bX,iC  (where n=batch index, oX, oY means X or Y coordinates
     40 //                         within the output image, bX, bY means coordinates
     41 //                         within the input block, iC means input channels).
     42 //      The output would be a transpose to the following layout:
     43 //      n,oY,oX,bY,bX,iC
     44 template <typename Device, typename T, TensorFormat data_format>
     45 struct SpaceToDepthOpFunctor {
     46   void operator()(const Device& d, typename TTypes<T, 4>::ConstTensor input,
     47                   int block_size, typename TTypes<T, 4>::Tensor output);
     48 
     49   // This 5-D version is to support NCHW_VECT_C.
     50   void operator()(const Device& d, typename TTypes<T, 5>::ConstTensor input,
     51                   int block_size, typename TTypes<T, 5>::Tensor output);
     52 };
     53 
     54 }  // namespace functor
     55 }  // namespace tensorflow
     56 
     57 #endif  // TENSORFLOW_CORE_KERNELS_SPACETODEPTH_OP_H_
     58