1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
|
// See docs in ../ops/nn_ops.cc.
#define EIGEN_USE_THREADS
#include "tensorflow/core/framework/numeric_op.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/framework/register_types.h"
#include "tensorflow/core/kernels/relu_op.h"
#include "tensorflow/core/public/tensor.h"
#include "tensorflow/core/lib/core/errors.h"
#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
template <typename Device, typename T>
class ReluOp : public UnaryElementWiseOp<T, ReluOp<Device, T>> {
public:
using UnaryElementWiseOp<T, ReluOp<Device, T>>::UnaryElementWiseOp;
void Operate(OpKernelContext* context, const Tensor& input, Tensor* output) {
functor::Relu<Device, T> functor;
functor(context->eigen_device<Device>(), input.flat<T>(),
output->flat<T>());
}
};
template <typename Device, typename T>
class Relu6Op : public UnaryElementWiseOp<T, Relu6Op<Device, T>> {
public:
using UnaryElementWiseOp<T, Relu6Op<Device, T>>::UnaryElementWiseOp;
void Operate(OpKernelContext* context, const Tensor& input, Tensor* output) {
functor::Relu6<Device, T> functor;
functor(context->eigen_device<Device>(), input.flat<T>(),
output->flat<T>());
}
};
template <typename Device, typename T>
class ReluGradOp : public BinaryElementWiseOp<T, ReluGradOp<Device, T>> {
public:
using BinaryElementWiseOp<T, ReluGradOp<Device, T>>::BinaryElementWiseOp;
// INPUTS:
// g (gradients): backpropagated gradients
// a (inputs): inputs that were passed to ReluOp()
// OUTPUT:
// gradients to backprop
template <int NDIMS>
void Operate(OpKernelContext* context, const Tensor& g, const Tensor& a,
Tensor* output) {
OP_REQUIRES(context, a.IsSameSize(g),
errors::InvalidArgument("g and a must be the same size"));
functor::ReluGrad<Device, T> functor;
functor(context->eigen_device<Device>(), g.flat<T>(), a.flat<T>(),
output->flat<T>());
}
};
template <typename Device, typename T>
class Relu6GradOp : public BinaryElementWiseOp<T, Relu6GradOp<Device, T>> {
public:
using BinaryElementWiseOp<T, Relu6GradOp<Device, T>>::BinaryElementWiseOp;
// INPUTS:
// g (gradients): backpropagated gradients
// a (inputs): inputs that were passed to Relu6Op()
// OUTPUT:
// gradients to backprop
template <int NDIMS>
void Operate(OpKernelContext* context, const Tensor& g, const Tensor& a,
Tensor* output) {
OP_REQUIRES(context, a.IsSameSize(g),
errors::InvalidArgument("g and a must be the same size"));
functor::Relu6Grad<Device, T> functor;
functor(context->eigen_device<Device>(), g.flat<T>(), a.flat<T>(),
output->flat<T>());
}
};
#define REGISTER_KERNELS(type) \
REGISTER_KERNEL_BUILDER( \
Name("Relu").Device(DEVICE_CPU).TypeConstraint<type>("T"), \
ReluOp<CPUDevice, type>); \
REGISTER_KERNEL_BUILDER( \
Name("Relu6").Device(DEVICE_CPU).TypeConstraint<type>("T"), \
Relu6Op<CPUDevice, type>); \
REGISTER_KERNEL_BUILDER( \
Name("ReluGrad").Device(DEVICE_CPU).TypeConstraint<type>("T"), \
ReluGradOp<CPUDevice, type>); \
REGISTER_KERNEL_BUILDER( \
Name("Relu6Grad").Device(DEVICE_CPU).TypeConstraint<type>("T"), \
Relu6GradOp<CPUDevice, type>)
TF_CALL_REAL_NUMBER_TYPES(REGISTER_KERNELS);
#undef REGISTER_KERNELS
#if GOOGLE_CUDA
// Forward declarations of the functor specializations for GPU.
namespace functor {
#define DECLARE_GPU_SPEC(T) \
template <> \
void Relu<GPUDevice, T>::operator()( \
const GPUDevice& d, typename TTypes<T>::ConstTensor features, \
typename TTypes<T>::Tensor activations); \
extern template struct Relu<GPUDevice, T>; \
\
template <> \
void ReluGrad<GPUDevice, T>::operator()( \
const GPUDevice& d, typename TTypes<T>::ConstTensor gradients, \
typename TTypes<T>::ConstTensor features, \
typename TTypes<T>::Tensor backprops); \
\
extern template struct ReluGrad<GPUDevice, T>; \
template <> \
void Relu6<GPUDevice, T>::operator()( \
const GPUDevice& d, typename TTypes<T>::ConstTensor features, \
typename TTypes<T>::Tensor activations); \
extern template struct Relu6<GPUDevice, T>; \
\
template <> \
void Relu6Grad<GPUDevice, T>::operator()( \
const GPUDevice& d, typename TTypes<T>::ConstTensor gradients, \
typename TTypes<T>::ConstTensor features, \
typename TTypes<T>::Tensor backprops); \
extern template struct Relu6Grad<GPUDevice, T>;
TF_CALL_GPU_NUMBER_TYPES(DECLARE_GPU_SPEC);
} // namespace functor
// Registration of the GPU implementations.
#define REGISTER_GPU_KERNELS(type) \
REGISTER_KERNEL_BUILDER( \
Name("Relu").Device(DEVICE_GPU).TypeConstraint<type>("T"), \
ReluOp<GPUDevice, type>); \
REGISTER_KERNEL_BUILDER( \
Name("Relu6").Device(DEVICE_GPU).TypeConstraint<type>("T"), \
Relu6Op<GPUDevice, type>); \
REGISTER_KERNEL_BUILDER( \
Name("ReluGrad").Device(DEVICE_GPU).TypeConstraint<type>("T"), \
ReluGradOp<GPUDevice, type>); \
REGISTER_KERNEL_BUILDER( \
Name("Relu6Grad").Device(DEVICE_GPU).TypeConstraint<type>("T"), \
Relu6GradOp<GPUDevice, type>)
TF_CALL_GPU_NUMBER_TYPES(REGISTER_GPU_KERNELS);
#undef REGISTER_GPU_KERNELS
#endif // GOOGLE_CUDA
} // namespace tensorflow
|
