1/* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
2
3Licensed under the Apache License, Version 2.0 (the "License");
4you may not use this file except in compliance with the License.
5You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9Unless required by applicable law or agreed to in writing, software
10distributed under the License is distributed on an "AS IS" BASIS,
11WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12See the License for the specific language governing permissions and
13limitations under the License.
14==============================================================================*/
15
16#ifndef TENSORFLOW_CORE_KERNELS_BIAS_OP_H_
17#define TENSORFLOW_CORE_KERNELS_BIAS_OP_H_
18// Functor definition for BiasOp, must be compilable by nvcc.
19
20#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
21#include "tensorflow/core/framework/tensor_types.h"
22
23namespace tensorflow {
24namespace functor {
25
26// Functor used by BiasOp to do the computations.
27template <typename Device, typename T>
28struct Bias {
29 // Add "bias" to "input", repeating "bias".
30 void operator()(const Device& d, typename TTypes<T>::ConstFlat input,
31 typename TTypes<T>::ConstVec bias,
32 typename TTypes<T>::Flat output) {
33 const Eigen::Index rest_size = input.size() / bias.dimension(0);
34 Eigen::DSizes<Eigen::Index, 1> bcast(rest_size);
35 MaybeWith32BitIndexing<Device>(
36 [&](auto input32, auto bias32, auto output32, const auto& bcast32) {
37 output32.device(d) = input32 + bias32.broadcast(bcast32);
38 },
39 input, bias, output, bcast);
40 }
41
42 // NCHW layout, repeating on the first dimension, broadcasting on the last
43 // dimension.
44 void operator()(const Device& d, typename TTypes<T>::ConstMatrix input,
45 typename TTypes<T>::ConstMatrix bias1, // shape [C, 1].
46 typename TTypes<T>::Matrix output) {
47 const Eigen::Index rest_size = input.dimension(0) / bias1.dimension(0);
48 Eigen::DSizes<Eigen::Index, 2> bcast(rest_size, input.dimension(1));
49 MaybeWith32BitIndexing<Device>(
50 [&](auto input32, auto bias32, auto output32, const auto& bcast32) {
51 output32.device(d) = input32 + bias32.broadcast(bcast32);
52 },
53 input, bias1, output, bcast);
54 }
55};
56
57} // namespace functor
58} // namespace tensorflow
59
60#endif // TENSORFLOW_CORE_KERNELS_BIAS_OP_H_
61