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
|
#include "caffe2/operators/rsqrt_op.h"
#include "caffe2/utils/eigen_utils.h"
#include <algorithm>
#include <functional>
#include <string>
namespace caffe2 {
template <>
template <typename T>
bool RsqrtGradientFunctor<CPUContext>::Forward(
const std::vector<int>& dY_dims,
const std::vector<int>& /* Y_dims */,
const T* dY,
const T* Y,
T* dX,
CPUContext* /* context */) const {
const int size = std::accumulate(
// NOLINTNEXTLINE(modernize-use-transparent-functors)
dY_dims.cbegin(), dY_dims.cend(), 1, std::multiplies<int>());
EigenVectorMap<T>(dX, size) = ConstEigenVectorMap<T>(dY, size).array() *
ConstEigenVectorMap<T>(Y, size).array().cube() * static_cast<T>(-0.5);
return true;
}
REGISTER_CPU_OPERATOR(
Rsqrt,
UnaryElementwiseOp<
TensorTypes<float>,
CPUContext,
RsqrtFunctor<CPUContext>>);
REGISTER_CPU_OPERATOR(
RsqrtGradient,
BinaryElementwiseOp<
TensorTypes<float>,
CPUContext,
RsqrtGradientFunctor<CPUContext>>);
OPERATOR_SCHEMA(Rsqrt)
.NumInputs(1)
.NumOutputs(1)
.AllowInplace({{0, 0}})
.IdenticalTypeAndShape()
.SetDoc("Computes the element-wise rsqrt of the input.")
.Input(0, "X", "ND input tensor")
.Output(0, "Y", "ND output tensor");
OPERATOR_SCHEMA(RsqrtGradient)
.NumInputs(2)
.NumOutputs(1)
.AllowInplace({{0, 0}});
namespace {
class GetRsqrtGradient final : public GradientMakerBase {
using GradientMakerBase::GradientMakerBase;
std::vector<OperatorDef> GetGradientDefs() override {
return SingleGradientDef(
"RsqrtGradient",
"",
std::vector<std::string>{GO(0), O(0)},
std::vector<std::string>{GI(0)});
}
};
} // namespace
REGISTER_GRADIENT(Rsqrt, GetRsqrtGradient);
} // namespace caffe2
|
