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
|
#include "caffe2/operators/relu_op.h"
#include <algorithm>
#include <functional>
#include <string>
#include "caffe2/utils/eigen_utils.h"
namespace caffe2 {
template <>
template <typename T>
bool ReluFunctor<CPUContext>::
operator()(const int N, const T* X, T* Y, CPUContext* /* context */) const {
EigenVectorMap<T>(Y, N) = ConstEigenVectorMap<float>(X, N).cwiseMax(T(0));
return true;
}
#ifdef CAFFE2_USE_ACCELERATE
template <>
template <>
bool ReluFunctor<CPUContext>::operator()<float>(
const int N,
const float* X,
float* Y,
CPUContext* /* context */) const {
const float zero = 0.0f;
vDSP_vthres(X, 1, &zero, Y, 1, N);
return true;
}
#endif // CAFFE2_USE_ACCELERATE
template <>
template <typename T>
bool ReluGradientFunctor<CPUContext>::Forward(
const std::vector<int>& Y_dims,
const std::vector<int>& /* dY_dims */,
const T* Y,
const T* dY,
T* dX,
CPUContext* /* context */) const {
const int size = std::accumulate(
// NOLINTNEXTLINE(modernize-use-transparent-functors)
Y_dims.cbegin(), Y_dims.cend(), 1, std::multiplies<int>());
EigenVectorArrayMap<T>(dX, size) =
(ConstEigenVectorArrayMap<T>(Y, size) > T(0))
.select(ConstEigenVectorArrayMap<T>(dY, size), T(0));
return true;
}
namespace {
OpSchema::Cost CostInferenceForRelu(
const OperatorDef& def,
const vector<TensorShape>& in) {
struct OpSchema::Cost cost = PointwiseCostInference<0>(def, in);
cost.params_bytes = 0;
return cost;
}
} // namespace
REGISTER_CPU_OPERATOR(
Relu,
UnaryElementwiseOp<
TensorTypes<float>,
CPUContext,
ReluFunctor<CPUContext>>);
REGISTER_CPU_GRADIENT_OPERATOR(
ReluGradient,
BinaryElementwiseOp<
TensorTypes<float>,
CPUContext,
ReluGradientFunctor<CPUContext>>);
// Input: X, output: Y
OPERATOR_SCHEMA(Relu)
.NumInputs(1)
.NumOutputs(1)
.AllowInplace({{0, 0}})
.CostInferenceFunction(CostInferenceForRelu)
.IdenticalTypeAndShape()
.SetDoc(R"DOC(
Applies rectified linear unit operation to the input data element-wise. The Relu operation takes one input $X$, produces one output $Y$, and is defined as:
$$Y = max(0,X)$$
Github Links:
- https://github.com/pytorch/pytorch/blob/master/caffe2/operators/relu_op.h
- https://github.com/pytorch/pytorch/blob/master/caffe2/operators/relu_op.cc
<details>
<summary> <b>Example</b> </summary>
**Code**
```
workspace.ResetWorkspace()
op = core.CreateOperator(
"Relu",
["X"],
["Y"]
)
workspace.FeedBlob("X", np.random.randn(4, 4).astype(np.float32)) // NCHW
print("X:\n", workspace.FetchBlob("X"), "\n")
workspace.RunOperatorOnce(op)
print("Y:\n", workspace.FetchBlob("Y"))
```
**Result**
```
X:
[[-1.4655551 0.64575136 0.7921748 0.4150579 ]
[ 0.41085166 -0.2837964 0.9881425 -1.9300346 ]
[ 0.39705405 0.44639114 0.9940703 0.2926532 ]
[-0.6726489 0.01330667 1.101319 0.33858967]]
Y:
[[0. 0.64575136 0.7921748 0.4150579 ]
[0.41085166 0. 0.9881425 0. ]
[0.39705405 0.44639114 0.9940703 0.2926532 ]
[0. 0.01330667 1.101319 0.33858967]]
```
</details>
)DOC")
.Input(0, "X", "1D input tensor")
.Output(0, "Y", "1D output tensor with same shape as input")
.InheritOnnxSchema();
// Input: Y, dY, output: dX
GRADIENT_OPERATOR_SCHEMA(ReluGradient)
.NumInputs(2)
.NumOutputs(1)
.AllowInplace({{1, 0}})
.IdenticalTypeAndShapeOfInput(1)
.SetDoc(R"DOC(
ReluGradient takes both Y and dY and uses this to update dX according to the
chain rule and derivatives of the rectified linear function.
)DOC");
namespace {
class GetReluGradient : public GradientMakerBase {
using GradientMakerBase::GradientMakerBase;
std::vector<OperatorDef> GetGradientDefs() override {
return SingleGradientDef(
def_.type() + "Gradient",
"",
std::vector<std::string>{O(0), GO(0)},
std::vector<std::string>{GI(0)});
}
};
} // namespace
REGISTER_GRADIENT(Relu, GetReluGradient);
} // namespace caffe2
|
