Welcome to mirror list, hosted at ThFree Co, Russian Federation.

nth_element_kernels.h « mblas « gpu « amun « src - github.com/marian-nmt/marian.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
blob: aeefcdd7f688a76e76eef01e9b23d2275c8d4a40 (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
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
#pragma once

#include "matrix_wrapper.h"
#include "vector_wrapper.h"

namespace amunmt {
namespace GPU {

struct NthOut
{
  unsigned ind;
  float score;

  __device__ __host__
  NthOut() {}

  __device__ __host__
  NthOut(unsigned init)
  :ind(init)
  ,score(init)
  {}

  __device__ __host__
  NthOut(unsigned &vInd, float vScore)
  :ind(vInd)
  ,score(vScore)
  {}

  __device__ __host__
  NthOut& operator+=(const NthOut& rhs)
  {
    ind += rhs.ind;
    score += rhs.score;
    return *this;
  }
};

/////////////////////////////////////////////////////////////////////////////////////////

struct NthOutBatch
{
  unsigned ind;
  float score;
  //unsigned hypoInd;
  //unsigned vocabInd;

  __device__ __host__
  NthOutBatch(const float& rhs)
  {
    // only to be used to init variable in matrix.h gSum
    assert(rhs == 0.0f);
    ind = rhs;
    score = rhs;
    //hypoInd = rhs;
    //vocabInd = rhs;
  }

  __device__ __host__
  NthOutBatch() {}

  __device__ __host__
  NthOutBatch(unsigned vInd, float vScore, unsigned vHypoInd, unsigned vVocabInd)
  :ind(vInd)
  ,score(vScore)
  //,hypoInd(vHypoInd)
  //,vocabInd(vVocabInd)
  {}

  __device__ __host__
  NthOutBatch& operator=(const NthOutBatch& rhs)
  {
    ind = rhs.ind;
    score = rhs.score;
    //hypoInd = rhs.hypoInd;
    //vocabInd = rhs.vocabInd;
    return *this;
  }

  __device__ __host__
  NthOutBatch& operator+=(const NthOutBatch& rhs)
  {
    ind += rhs.ind;
    score += rhs.score;
    //hypoInd += rhs.hypoInd;
    //vocabInd += rhs.vocabInd;
    return *this;
  }

};


/////////////////////////////////////////////////////////////////////////////////////////

inline std::ostream& operator<<(std::ostream &out, const NthOut &obj)
{
  out << "(" << obj.ind << "," << obj.score << ")";
  return out;
}

inline std::ostream& operator<<(std::ostream &out, const NthOutBatch &obj)
{
  out << "("
      << obj.ind << ","
      << obj.score << ","
      //<< obj.hypoInd << ","
      //<< obj.vocabInd
      << ")";
  return out;
}

/////////////////////////////////////////////////////////////////////////////////////////

__global__ void gMaxElement(mblas::VectorWrapper<NthOut> out,
                            const mblas::MatrixWrapper<float> probsWrap,
                            const mblas::VectorWrapper<unsigned> batchPositionWrap,
                            unsigned numBatches);

__global__ void gMaxElementUpdate(mblas::VectorWrapper<NthOut> out,
                                  mblas::MatrixWrapper<float> probsWrap,
                                  mblas::VectorWrapper<NthOut> resWrap,
                                  const mblas::VectorWrapper<unsigned> batchPositionWrap,
                                  const mblas::VectorWrapper<unsigned> cumBeamSizesWrap,
                                  unsigned numBlocks);

__global__ void gGetValueByKey(mblas::MatrixWrapper<float> out,
                              const   mblas::MatrixWrapper<float> in,
                              unsigned* indices, unsigned n);

}
}