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
|
#if GOOGLE_CUDA
#define EIGEN_USE_GPU
#include "tensorflow/core/kernels/aggregate_ops.h"
#include "tensorflow/core/platform/port.h"
#include "tensorflow/core/framework/tensor_types.h"
namespace tensorflow {
typedef Eigen::GpuDevice GPUDevice;
// Partial specialization for a GPUDevice, that uses the Eigen implementation.
namespace functor {
template <typename T>
struct Add2Functor<GPUDevice, T> {
void operator()(const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2) {
Add2EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2);
}
};
template <typename T>
struct Add3Functor<GPUDevice, T> {
void operator()(const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3) {
Add3EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3);
}
};
template <typename T>
struct Add4Functor<GPUDevice, T> {
void operator()(const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4) {
Add4EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4);
}
};
template <typename T>
struct Add5Functor<GPUDevice, T> {
void operator()(const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5) {
Add5EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4, in5);
}
};
template <typename T>
struct Add6Functor<GPUDevice, T> {
void operator()(const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5,
typename TTypes<T>::ConstFlat in6) {
Add6EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6);
}
};
template <typename T>
struct Add7Functor<GPUDevice, T> {
void operator()(const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5,
typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7) {
Add7EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7);
}
};
template <typename T>
struct Add8Functor<GPUDevice, T> {
void operator()(
const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8) {
Add8EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7, in8);
}
};
template <typename T>
struct Add8pFunctor<GPUDevice, T> {
void operator()(
const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8) {
Add8pEigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7, in8);
}
};
template <typename T>
struct Add9Functor<GPUDevice, T> {
void operator()(
const GPUDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8,
typename TTypes<T>::ConstFlat in9) {
Add9EigenImpl<GPUDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7, in8, in9);
}
};
} // end namespace functor
// Instantiate the GPU implementation for float.
template struct functor::Add2Functor<GPUDevice, float>;
template struct functor::Add3Functor<GPUDevice, float>;
template struct functor::Add4Functor<GPUDevice, float>;
template struct functor::Add5Functor<GPUDevice, float>;
template struct functor::Add6Functor<GPUDevice, float>;
template struct functor::Add7Functor<GPUDevice, float>;
template struct functor::Add8Functor<GPUDevice, float>;
template struct functor::Add8pFunctor<GPUDevice, float>;
template struct functor::Add9Functor<GPUDevice, float>;
} // end namespace tensorflow
#endif // GOOGLE_CUDA
|
