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#if !GOOGLE_CUDA
17#error This file must only be included when building with Cuda support
18#endif
19
20#ifndef TENSORFLOW_KERNELS_CWISE_OPS_GPU_GRADIENTS_CU_H_
21#define TENSORFLOW_KERNELS_CWISE_OPS_GPU_GRADIENTS_CU_H_
22
23#define EIGEN_USE_GPU
24
25#include <complex>
26
27#include "tensorflow/core/framework/tensor_types.h"
28#include "tensorflow/core/kernels/cwise_ops.h"
29#include "tensorflow/core/kernels/cwise_ops_gradients.h"
30#include "tensorflow/core/platform/types.h"
31
32#include "tensorflow/core/platform/logging.h"
33namespace tensorflow {
34namespace functor {
35
36typedef Eigen::GpuDevice GPUDevice;
37typedef std::complex<float> complex64;
38typedef std::complex<double> complex128;
39
40// Partial specialization of SimpleBinaryFunctor<Device=GPUDevice, Functor>.
41template <typename Functor>
42struct SimpleBinaryFunctor<GPUDevice, Functor> {
43  void operator()(const GPUDevice& d, typename Functor::tout_type out,
44                  typename Functor::tin_type in1,
45                  typename Functor::tin_type in2) {
46    To32Bit(out).device(d) =
47        To32Bit(in1).binaryExpr(in2, typename Functor::func());
48  }
49};
50
51// Macros to explicitly instantiate kernels on GPU for multiple types
52// (T0, T1, etc.) for SimpleBiaryFunctor (e.g., functor::tanh_grad).
53#define DEFINE_SIMPLE_BINARY1(F, T) \
54  template struct SimpleBinaryFunctor<GPUDevice, F<T> >
55#define DEFINE_SIMPLE_BINARY2(F, T0, T1) \
56  DEFINE_SIMPLE_BINARY1(F, T0);          \
57  DEFINE_SIMPLE_BINARY1(F, T1)
58#define DEFINE_SIMPLE_BINARY3(F, T0, T1, T2) \
59  DEFINE_SIMPLE_BINARY2(F, T0, T1);          \
60  DEFINE_SIMPLE_BINARY1(F, T2)
61#define DEFINE_SIMPLE_BINARY4(F, T0, T1, T2, T3) \
62  DEFINE_SIMPLE_BINARY2(F, T0, T1);              \
63  DEFINE_SIMPLE_BINARY2(F, T2, T3)
64#define DEFINE_SIMPLE_BINARY5(F, T0, T1, T2, T3, T4) \
65  DEFINE_SIMPLE_BINARY2(F, T0, T1);                  \
66  DEFINE_SIMPLE_BINARY3(F, T2, T3, T4)
67
68}  // end namespace functor
69}  // end namespace tensorflow
70
71#endif  // TENSORFLOW_KERNELS_CWISE_OPS_GPU_GRADIENTS_CU_H_
72