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
|
#include "matrix_wrapper.h"
namespace amunmt {
namespace GPU {
struct NthOut
{
uint ind;
float score;
__device__ __host__
NthOut() {}
__device__ __host__
NthOut(uint init)
:ind(init)
,score(init)
{}
__device__ __host__
NthOut(uint &vInd, float vScore)
:ind(vInd)
,score(vScore)
{}
__device__ __host__
NthOut& operator+=(const NthOut& rhs)
{
ind += rhs.ind;
score += rhs.score;
return *this;
}
};
/////////////////////////////////////////////////////////////////////////////////////////
inline std::ostream& operator<<(std::ostream &out, const NthOut &obj)
{
out << "(" << obj.ind << "," << obj.score << ")";
return out;
}
/////////////////////////////////////////////////////////////////////////////////////////
__global__ void gMaxElement(mblas::MatrixWrapper<NthOut> out,
const mblas::MatrixWrapper<float> probsWrap,
const mblas::MatrixWrapper<uint> batchPositionWrap,
uint numBatches);
__global__ void gMaxElementUpdate(mblas::MatrixWrapper<NthOut> out,
mblas::MatrixWrapper<float> probsWrap,
mblas::MatrixWrapper<NthOut> resNewWrap,
const mblas::MatrixWrapper<uint> batchPositionWrap,
const mblas::MatrixWrapper<uint> cumBeamSizesWrap,
uint numBlocks);
__global__ void gGetValueByKey(mblas::MatrixWrapper<float> out,
const mblas::MatrixWrapper<float> in,
uint* indices, uint n);
}
}
|
