• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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_DATA_FORMAT_OPS_H_
17 #define TENSORFLOW_CORE_KERNELS_DATA_FORMAT_OPS_H_
18 // Functor definition for data format dim mapping ops, must be compilable
19 // by nvcc.
20 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
21 #include "tensorflow/core/framework/tensor_types.h"
22 
23 namespace tensorflow {
24 namespace functor {
25 
26 // Functor used by DataFormatDimMapOP to do the computations.
27 template <typename Device, typename T>
28 struct DataFormatDimMap {
operatorDataFormatDimMap29   void operator()(const Device& d, typename TTypes<T>::ConstFlat x,
30                   typename TTypes<T>::Flat y, const TTypes<int>::Vec dst) {
31     auto zero = x.constant(0);
32     auto one = x.constant(1);
33     auto two = x.constant(2);
34 
35     auto f_zero = x.constant(dst(0));
36     auto f_one = x.constant(dst(1));
37     auto f_two = x.constant(dst(2));
38     auto f_three = x.constant(dst(3));
39 
40     auto four = x.constant(4);
41     auto x_mod = (x + four) % 4;
42 
43     auto is_zero = (x_mod == zero);
44     auto is_one = (x_mod == one);
45     auto is_two = (x_mod == two);
46 
47     y.device(d) = is_zero.select(
48         f_zero, is_one.select(f_one, is_two.select(f_two, f_three)));
49   }
50 };
51 
52 template <typename T>
53 struct VecPermute {
VecPermuteVecPermute54   VecPermute(const Eigen::DSizes<Eigen::DenseIndex, 8>& dst) : dst_(dst) {}
dimensionsVecPermute55   Eigen::DSizes<Eigen::DenseIndex, 1> dimensions(
56       typename TTypes<T>::ConstFlat input) const {
57     Eigen::DSizes<Eigen::DenseIndex, 1> result;
58     result[0] = input.dimension(0);
59     return result;
60   }
61   template <typename Output, typename Device>
evalVecPermute62   void eval(typename TTypes<T>::ConstFlat input, Output& output,
63             const Device& d) const {
64     for (int i = 0; i < input.size(); ++i) {
65       output.template chip<0>(dst_[i]).device(d) = input.template chip<0>(i);
66     }
67   }
68 
69  private:
70   Eigen::DSizes<Eigen::DenseIndex, 8> dst_;
71 };
72 
73 // Functor used by DataFormatVecPermuteOp to do the computations.
74 template <typename Device, typename T>
75 struct DataFormatVecPermute {
operatorDataFormatVecPermute76   void operator()(const Device& d, typename TTypes<T>::ConstFlat x,
77                   typename TTypes<T>::Flat y,
78                   const Eigen::DSizes<Eigen::DenseIndex, 8>& dst) {
79     y.device(d) = x.customOp(VecPermute<T>(dst));
80   }
81 };
82 
83 }  // namespace functor
84 }  // namespace tensorflow
85 
86 #endif  // TENSORFLOW_CORE_KERNELS_DATA_FORMAT_OPS_H_
87