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
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
|
/* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#include "tensorflow/cc/ops/nn_ops.h"
#include "tensorflow/cc/ops/nn_ops_internal.h"
#include "tensorflow/cc/ops/standard_ops.h"
#include "tensorflow/cc/framework/grad_op_registry.h"
#include "tensorflow/cc/framework/gradients.h"
namespace tensorflow {
namespace ops {
namespace {
Status SoftmaxGrad(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
// Softmax gradient function.
// p = softmax(x) maps from [batch, n] to [batch, m]
// dp/dx = [dp0/dx0 ... dp0/dxn-1 ]
// [ ... ... ]
// [dpm-1/dx0 ... dpm-1/dxn-1]
// dL/dx = dp/dx * dL/dy
//
// Using alternative formula:
// dL/dx = dL/dy * y - sum(dL/dy * y) * y
// = (dL/dy - sum(dL/dy * y)) * y
auto y = op.output(0);
auto dyy = Mul(scope, grad_inputs[0], y);
auto sum = Reshape(scope, Sum(scope, dyy, {1}), {-1, 1});
auto sub = Sub(scope, grad_inputs[0], sum);
auto dx = Mul(scope, sub, y);
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Softmax", SoftmaxGrad);
Status LogSoftmaxGrad(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto softmax = Exp(scope, op.output(0));
auto sum = Sum(scope, grad_inputs[0], {1}, Sum::KeepDims(true));
auto mul = Mul(scope, sum, softmax);
auto dx = Sub(scope, grad_inputs[0], mul);
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("LogSoftmax", LogSoftmaxGrad);
Status ReluGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::ReluGrad(scope, grad_inputs[0], op.input(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Relu", ReluGradHelper);
Status Relu6GradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::Relu6Grad(scope, grad_inputs[0], op.input(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Relu6", Relu6GradHelper);
Status EluGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::EluGrad(scope, grad_inputs[0], op.output(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Elu", EluGradHelper);
Status SeluGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::SeluGrad(scope, grad_inputs[0], op.output(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Selu", SeluGradHelper);
Status L2LossGrad(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
grad_outputs->push_back(Mul(scope, op.input(0), grad_inputs[0]));
return scope.status();
}
REGISTER_GRADIENT_OP("L2Loss", L2LossGrad);
Status BiasAddGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
string data_format;
TF_RETURN_IF_ERROR(
GetNodeAttr(op.output(0).node()->attrs(), "data_format", &data_format));
auto dx_1 =
BiasAddGrad(scope, grad_inputs[0], BiasAddGrad::DataFormat(data_format));
grad_outputs->push_back(Identity(scope, grad_inputs[0]));
grad_outputs->push_back(dx_1);
return scope.status();
}
REGISTER_GRADIENT_OP("BiasAdd", BiasAddGradHelper);
Status Conv2DGrad(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
string data_format;
string padding;
std::vector<int32> strides;
bool use_cudnn_on_gpu;
auto attrs = op.output(0).node()->attrs();
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "data_format", &data_format));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "padding", &padding));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "strides", &strides));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "use_cudnn_on_gpu", &use_cudnn_on_gpu));
auto dx_1 = Conv2DBackpropInput(scope, Shape(scope, op.input(0)), op.input(1),
grad_inputs[0], strides, padding,
Conv2DBackpropInput::DataFormat(data_format)
.UseCudnnOnGpu(use_cudnn_on_gpu));
grad_outputs->push_back(dx_1);
auto dx_2 =
Conv2DBackpropFilter(scope, op.input(0), Shape(scope, op.input(1)),
grad_inputs[0], strides, padding,
Conv2DBackpropFilter::DataFormat(data_format)
.UseCudnnOnGpu(use_cudnn_on_gpu));
grad_outputs->push_back(dx_2);
return scope.status();
}
REGISTER_GRADIENT_OP("Conv2D", Conv2DGrad);
Status MaxPoolGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
string data_format;
string padding;
std::vector<int32> strides;
std::vector<int32> ksize;
auto attrs = op.output(0).node()->attrs();
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "data_format", &data_format));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "ksize", &ksize));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "padding", &padding));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "strides", &strides));
auto dx = internal::MaxPoolGrad(
scope, op.input(0), op.output(0), grad_inputs[0], ksize, strides, padding,
internal::MaxPoolGrad::DataFormat(data_format));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("MaxPool", MaxPoolGradHelper);
Status MaxPoolGradV2Helper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
string data_format;
string padding;
auto attrs = op.output(0).node()->attrs();
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "data_format", &data_format));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "padding", &padding));
auto dx = MaxPoolGradV2(scope, op.input(0), op.output(0), grad_inputs[0],
op.input(1), op.input(2), padding,
MaxPoolGradV2::DataFormat(data_format));
grad_outputs->push_back(dx);
grad_outputs->push_back(NoGradient());
grad_outputs->push_back(NoGradient());
return scope.status();
}
REGISTER_GRADIENT_OP("MaxPoolV2", MaxPoolGradV2Helper);
Status MaxPool3DGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
std::vector<int32> ksize;
std::vector<int32> strides;
string padding;
string data_format;
auto attrs = op.output(0).node()->attrs();
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "ksize", &ksize));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "strides", &strides));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "padding", &padding));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "data_format", &data_format));
MaxPool3DGrad::Attrs grad_attrs;
auto dx = MaxPool3DGrad(scope, op.input(0), op.output(0), grad_inputs[0],
ksize, strides, padding,
grad_attrs.DataFormat(data_format));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("MaxPool3D", MaxPool3DGradHelper);
Status AvgPoolGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
std::vector<int32> ksize;
std::vector<int32> strides;
string padding;
string data_format;
auto attrs = op.output(0).node()->attrs();
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "ksize", &ksize));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "strides", &strides));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "padding", &padding));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "data_format", &data_format));
internal::AvgPoolGrad::Attrs grad_attrs;
auto dx =
internal::AvgPoolGrad(scope, Shape(scope, op.input(0)), grad_inputs[0],
ksize, strides, padding,
grad_attrs.DataFormat(data_format));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("AvgPool", AvgPoolGradHelper);
Status AvgPool3DGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
std::vector<int32> ksize;
std::vector<int32> strides;
string padding;
string data_format;
auto attrs = op.output(0).node()->attrs();
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "ksize", &ksize));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "strides", &strides));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "padding", &padding));
TF_RETURN_IF_ERROR(GetNodeAttr(attrs, "data_format", &data_format));
AvgPool3DGrad::Attrs grad_attrs;
auto dx = AvgPool3DGrad(scope, Shape(scope, op.input(0)), grad_inputs[0],
ksize, strides, padding,
grad_attrs.DataFormat(data_format));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("AvgPool3D", AvgPool3DGradHelper);
Status LRNGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::LRNGrad(scope, grad_inputs[0], op.input(0), op.output(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("LRN", LRNGradHelper);
Status SoftplusGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::SoftplusGrad(scope, grad_inputs[0], op.input(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Softplus", SoftplusGradHelper);
Status SoftsignGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
auto dx = internal::SoftsignGrad(scope, grad_inputs[0], op.input(0));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("Softsign", SoftsignGradHelper);
Status FractionalAvgPoolGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
bool overlapping;
TF_RETURN_IF_ERROR(
GetNodeAttr(op.output(0).node()->attrs(), "overlapping", &overlapping));
auto dx = internal::FractionalAvgPoolGrad(
scope, Shape(scope, op.input(0), Shape::OutType(DT_INT64)),
grad_inputs[0], op.output(1), op.output(2),
internal::FractionalAvgPoolGrad::Overlapping(overlapping));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("FractionalAvgPool", FractionalAvgPoolGradHelper);
Status FractionalMaxPoolGradHelper(const Scope& scope, const Operation& op,
const std::vector<Output>& grad_inputs,
std::vector<Output>* grad_outputs) {
bool overlapping;
TF_RETURN_IF_ERROR(
GetNodeAttr(op.output(0).node()->attrs(), "overlapping", &overlapping));
auto dx = internal::FractionalMaxPoolGrad(
scope, op.input(0), op.output(0), grad_inputs[0], op.output(1),
op.output(2), internal::FractionalMaxPoolGrad::Overlapping(overlapping));
grad_outputs->push_back(dx);
return scope.status();
}
REGISTER_GRADIENT_OP("FractionalMaxPool", FractionalMaxPoolGradHelper);
} // anonymous namespace
} // namespace ops
} // namespace tensorflow
|
