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_AGGREGATE_OPS_H_ |
17 | #define TENSORFLOW_CORE_KERNELS_AGGREGATE_OPS_H_ |
18 | |
19 | #include <numeric> |
20 | |
21 | #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor" |
22 | #include "tensorflow/core/framework/tensor_types.h" |
23 | |
24 | namespace tensorflow { |
25 | namespace functor { |
26 | |
27 | // Functor definitions for Aggregate ops, must be compilable by nvcc. |
28 | template <typename Device, typename T> |
29 | struct Add2Functor { |
30 | void operator()(const Device& d, typename TTypes<T>::Flat out, |
31 | typename TTypes<T>::ConstFlat in1, |
32 | typename TTypes<T>::ConstFlat in2); |
33 | }; |
34 | |
35 | template <typename Device, typename T> |
36 | struct Add2EigenImpl { |
37 | static void Compute(const Device& d, typename TTypes<T>::Flat out, |
38 | typename TTypes<T>::ConstFlat in1, |
39 | typename TTypes<T>::ConstFlat in2) { |
40 | out.device(d) = in1 + in2; |
41 | } |
42 | }; |
43 | |
44 | template <typename Device, typename T> |
45 | struct Add3Functor { |
46 | void operator()(const Device& d, typename TTypes<T>::Flat out, |
47 | typename TTypes<T>::ConstFlat in1, |
48 | typename TTypes<T>::ConstFlat in2, |
49 | typename TTypes<T>::ConstFlat in3); |
50 | }; |
51 | |
52 | template <typename Device, typename T> |
53 | struct Add3EigenImpl { |
54 | static void Compute(const Device& d, typename TTypes<T>::Flat out, |
55 | typename TTypes<T>::ConstFlat in1, |
56 | typename TTypes<T>::ConstFlat in2, |
57 | typename TTypes<T>::ConstFlat in3) { |
58 | out.device(d) = in1 + in2 + in3; |
59 | } |
60 | }; |
61 | |
62 | template <typename Device, typename T> |
63 | struct Add4Functor { |
64 | void operator()(const Device& d, typename TTypes<T>::Flat out, |
65 | typename TTypes<T>::ConstFlat in1, |
66 | typename TTypes<T>::ConstFlat in2, |
67 | typename TTypes<T>::ConstFlat in3, |
68 | typename TTypes<T>::ConstFlat in4); |
69 | }; |
70 | |
71 | template <typename Device, typename T> |
72 | struct Add4EigenImpl { |
73 | static void Compute(const Device& d, typename TTypes<T>::Flat out, |
74 | typename TTypes<T>::ConstFlat in1, |
75 | typename TTypes<T>::ConstFlat in2, |
76 | typename TTypes<T>::ConstFlat in3, |
77 | typename TTypes<T>::ConstFlat in4) { |
78 | out.device(d) = in1 + in2 + in3 + in4; |
79 | } |
80 | }; |
81 | |
82 | template <typename Device, typename T> |
83 | struct Add5Functor { |
84 | void operator()(const Device& d, typename TTypes<T>::Flat out, |
85 | typename TTypes<T>::ConstFlat in1, |
86 | typename TTypes<T>::ConstFlat in2, |
87 | typename TTypes<T>::ConstFlat in3, |
88 | typename TTypes<T>::ConstFlat in4, |
89 | typename TTypes<T>::ConstFlat in5); |
90 | }; |
91 | |
92 | template <typename Device, typename T> |
93 | struct Add5EigenImpl { |
94 | static void Compute(const Device& d, typename TTypes<T>::Flat out, |
95 | typename TTypes<T>::ConstFlat in1, |
96 | typename TTypes<T>::ConstFlat in2, |
97 | typename TTypes<T>::ConstFlat in3, |
98 | typename TTypes<T>::ConstFlat in4, |
99 | typename TTypes<T>::ConstFlat in5) { |
100 | out.device(d) = in1 + in2 + in3 + in4 + in5; |
101 | } |
102 | }; |
103 | |
104 | template <typename Device, typename T> |
105 | struct Add6Functor { |
106 | void operator()(const Device& d, typename TTypes<T>::Flat out, |
107 | typename TTypes<T>::ConstFlat in1, |
108 | typename TTypes<T>::ConstFlat in2, |
109 | typename TTypes<T>::ConstFlat in3, |
110 | typename TTypes<T>::ConstFlat in4, |
111 | typename TTypes<T>::ConstFlat in5, |
112 | typename TTypes<T>::ConstFlat in6); |
113 | }; |
114 | |
115 | template <typename Device, typename T> |
116 | struct Add6EigenImpl { |
117 | static void Compute(const Device& d, typename TTypes<T>::Flat out, |
118 | typename TTypes<T>::ConstFlat in1, |
119 | typename TTypes<T>::ConstFlat in2, |
120 | typename TTypes<T>::ConstFlat in3, |
121 | typename TTypes<T>::ConstFlat in4, |
122 | typename TTypes<T>::ConstFlat in5, |
123 | typename TTypes<T>::ConstFlat in6) { |
124 | out.device(d) = in1 + in2 + in3 + in4 + in5 + in6; |
125 | } |
126 | }; |
127 | |
128 | template <typename Device, typename T> |
129 | struct Add7Functor { |
130 | void operator()(const Device& d, typename TTypes<T>::Flat out, |
131 | typename TTypes<T>::ConstFlat in1, |
132 | typename TTypes<T>::ConstFlat in2, |
133 | typename TTypes<T>::ConstFlat in3, |
134 | typename TTypes<T>::ConstFlat in4, |
135 | typename TTypes<T>::ConstFlat in5, |
136 | typename TTypes<T>::ConstFlat in6, |
137 | typename TTypes<T>::ConstFlat in7); |
138 | }; |
139 | |
140 | template <typename Device, typename T> |
141 | struct Add7EigenImpl { |
142 | static void Compute(const Device& d, typename TTypes<T>::Flat out, |
143 | typename TTypes<T>::ConstFlat in1, |
144 | typename TTypes<T>::ConstFlat in2, |
145 | typename TTypes<T>::ConstFlat in3, |
146 | typename TTypes<T>::ConstFlat in4, |
147 | typename TTypes<T>::ConstFlat in5, |
148 | typename TTypes<T>::ConstFlat in6, |
149 | typename TTypes<T>::ConstFlat in7) { |
150 | out.device(d) = in1 + in2 + in3 + in4 + in5 + in6 + in7; |
151 | } |
152 | }; |
153 | |
154 | template <typename Device, typename T> |
155 | struct Add8Functor { |
156 | void operator()( |
157 | const Device& d, typename TTypes<T>::Flat out, |
158 | typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2, |
159 | typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4, |
160 | typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6, |
161 | typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8); |
162 | }; |
163 | |
164 | template <typename Device, typename T> |
165 | struct Add8EigenImpl { |
166 | static void Compute( |
167 | const Device& d, typename TTypes<T>::Flat out, |
168 | typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2, |
169 | typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4, |
170 | typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6, |
171 | typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8) { |
172 | out.device(d) = in1 + in2 + in3 + in4 + in5 + in6 + in7 + in8; |
173 | } |
174 | }; |
175 | |
176 | // Add8p is like Add8 except the underlying implementation should += |
177 | // rather than assign to the output. |
178 | template <typename Device, typename T> |
179 | struct Add8pFunctor { |
180 | void operator()( |
181 | const Device& d, typename TTypes<T>::Flat out, |
182 | typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2, |
183 | typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4, |
184 | typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6, |
185 | typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8); |
186 | }; |
187 | |
188 | template <typename Device, typename T> |
189 | struct Add8pEigenImpl { |
190 | static void Compute( |
191 | const Device& d, typename TTypes<T>::Flat out, |
192 | typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2, |
193 | typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4, |
194 | typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6, |
195 | typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8) { |
196 | out.device(d) += in1 + in2 + in3 + in4 + in5 + in6 + in7 + in8; |
197 | } |
198 | }; |
199 | |
200 | template <typename Device, typename T> |
201 | struct Add9Functor { |
202 | void operator()( |
203 | const Device& d, typename TTypes<T>::Flat out, |
204 | typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2, |
205 | typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4, |
206 | typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6, |
207 | typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8, |
208 | typename TTypes<T>::ConstFlat in9); |
209 | }; |
210 | |
211 | template <typename Device, typename T> |
212 | struct Add9EigenImpl { |
213 | static void Compute( |
214 | const Device& d, typename TTypes<T>::Flat out, |
215 | typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2, |
216 | typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4, |
217 | typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6, |
218 | typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8, |
219 | typename TTypes<T>::ConstFlat in9) { |
220 | out.device(d) = in1 + in2 + in3 + in4 + in5 + in6 + in7 + in8 + in9; |
221 | } |
222 | }; |
223 | } // namespace functor |
224 | } // namespace tensorflow |
225 | |
226 | #endif // TENSORFLOW_CORE_KERNELS_AGGREGATE_OPS_H_ |
227 | |