File size: 3,634 Bytes
477da44
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
#ifndef _RKNPU2_RKNN_MATMUL_API_DEMO_H_
#define _RKNPU2_RKNN_MATMUL_API_DEMO_H_

namespace rknpu2 {

using ushort = unsigned short;

typedef union suf32
{
  int      i;
  unsigned u;
  float    f;
} suf32;

class float16
{
public:
  float16() {}
  explicit float16(float x) { w = bits(x); }

  operator float() const
  {
    suf32 out;

    unsigned t    = ((w & 0x7fff) << 13) + 0x38000000;
    unsigned sign = (w & 0x8000) << 16;
    unsigned e    = w & 0x7c00;

    out.u = t + (1 << 23);
    out.u = (e >= 0x7c00 ? t + 0x38000000 : e == 0 ? (static_cast<void>(out.f -= 6.103515625e-05f), out.u) : t) | sign;
    return out.f;
  }

  static float16 fromBits(ushort b)
  {
    float16 result;
    result.w = b;
    return result;
  }
  static float16 zero()
  {
    float16 result;
    result.w = (ushort)0;
    return result;
  }
  ushort bits() const { return w; }

  static ushort bits(float x)
  {
    suf32 in;
    in.f          = x;
    unsigned sign = in.u & 0x80000000;
    in.u ^= sign;
    ushort w;

    if (in.u >= 0x47800000)
      w = (ushort)(in.u > 0x7f800000 ? 0x7e00 : 0x7c00);
    else {
      if (in.u < 0x38800000) {
        in.f += 0.5f;
        w = (ushort)(in.u - 0x3f000000);
      } else {
        unsigned t = in.u + 0xc8000fff;
        w          = (ushort)((t + ((in.u >> 13) & 1)) >> 13);
      }
    }

    w = (ushort)(w | (sign >> 16));

    return w;
  }

  float16& operator=(float x)
  {
    w = bits(x);
    return *this;
  }

  float16& operator+=(float x)
  {
    w = bits(float() + x);
    return *this;
  }

  float16& operator/(float x)
  {
    w = bits(float() / x);
    return *this;
  }

  inline bool is_nan() const { return ((w & 0x7c00u) == 0x7c00u) && ((w & 0x03ffu) != 0x0000u); }

  inline bool greater(const float16& x) const
  {
    bool sign   = w & 0x8000;
    bool sign_x = x.w & 0x8000;
    if (sign) {
      if (sign_x)
        return w < x.w;
      else
        return false;
    } else {
      if (sign_x)
        /* Signed zeros are equal, have to check for it */
        return (w != 0 || x.w != 0x8000);
      else
        return w > x.w;
    }
    return false;
  }

  inline bool less(const float16& x) const
  {
    bool sign   = w & 0x8000;
    bool sign_x = x.w & 0x8000;
    if (sign) {
      if (sign_x)
        return w > x.w;
      else
        /* Signed zeros are equal, have to check for it */
        return (w != 0x8000 || x.w != 0);
    } else {
      if (sign_x)
        return false;
      else
        return w < x.w;
    }
    return false;
  }

  inline bool operator>(const float16& x) const
  {
    if (is_nan() || x.is_nan()) {
      return false;
    }
    return greater(x);
  }

  inline bool operator<(const float16& x) const
  {
    if (is_nan() || x.is_nan()) {
      return false;
    }
    return less(x);
  }

  inline bool operator>=(const float16& x) const
  {
    if (is_nan() || x.is_nan()) {
      return false;
    }
    return !less(x);
  }

  inline bool operator<=(const float16& x) const
  {
    if (is_nan() || x.is_nan()) {
      return false;
    }
    return !greater(x);
  }

  inline bool operator==(const float16& x) const
  {
    /*
     * The equality cases are as follows:
     *   - If either value is NaN, never equal.
     *   - If the values are equal, equal.
     *   - If the values are both signed zeros, equal.
     */
    if (is_nan() || x.is_nan()) {
      return false;
    }
    return (w == x.w || ((w | x.w) & 0x7fff) == 0);
  }

  inline bool operator!=(const float16& x) const { return !((*this) == x); }

protected:
  ushort w = 0;
};

} // namespace rknn

#endif /* _RKNPU2_RKNN_MATMUL_API_DEMO_H_ */