blob: d9f347159b143f7a99330e5ddd2f79d952bc2810 (
plain)
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
|
#include "THCUNN.h"
#include "THCHalf.h"
#include "THCHalfAutoNumerics.cuh"
#include "common.h"
template <typename Dtype, typename Acctype>
struct gatedLinearCSigMul_functor
{
__device__ void operator()(Dtype *target, const Dtype *sigTensor, const Dtype *mulTensor) const
{
const Acctype sigNum = Acctype(1)/(Acctype(1)+ exp(ScalarConvert<Dtype, Acctype>::to(-*sigTensor)));
const Dtype mulNum = *mulTensor;
*target = ScalarConvert<Acctype, Dtype>::to(sigNum * mulNum);
}
};
template <typename Dtype, typename Acctype>
struct gatedLinearDerivativeSecondHalf_functor
{
__device__ void operator()(Dtype *target, const Dtype *sigTensor, const Dtype *mulTensor) const
{
const Acctype sigNum = Acctype(1)/(Acctype(1)+ exp(ScalarConvert<Dtype, Acctype>::to(-*sigTensor)));
const Dtype mulNum = *mulTensor;
*target *= ScalarConvert<Acctype, Dtype>::to((Acctype(1) - sigNum) * sigNum * mulNum);
}
};
#include "generic/GatedLinearUnit.cu"
#include "THCGenerateFloatTypes.h"
|
