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
|
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cmath>
#include <cstring>
#include "common/common_types.h"
namespace Pica {
/**
* Template class for converting arbitrary Pica float types to IEEE 754 32-bit single-precision
* floating point.
*
* When decoding, format is as follows:
* - The first `M` bits are the mantissa
* - The next `E` bits are the exponent
* - The last bit is the sign bit
*
* @todo Verify on HW if this conversion is sufficiently accurate.
*/
template <unsigned M, unsigned E>
struct Float {
public:
static Float<M, E> FromFloat32(float val) {
Float<M, E> ret;
ret.value = val;
return ret;
}
static Float<M, E> FromRaw(u32 hex) {
Float<M, E> res;
const int width = M + E + 1;
const int bias = 128 - (1 << (E - 1));
const int exponent = (hex >> M) & ((1 << E) - 1);
const unsigned mantissa = hex & ((1 << M) - 1);
if (hex & ((1 << (width - 1)) - 1))
hex = ((hex >> (E + M)) << 31) | (mantissa << (23 - M)) | ((exponent + bias) << 23);
else
hex = ((hex >> (E + M)) << 31);
std::memcpy(&res.value, &hex, sizeof(float));
return res;
}
static Float<M, E> Zero() {
return FromFloat32(0.f);
}
// Not recommended for anything but logging
float ToFloat32() const {
return value;
}
Float<M, E> operator*(const Float<M, E>& flt) const {
if ((this->value == 0.f && !std::isnan(flt.value)) ||
(flt.value == 0.f && !std::isnan(this->value)))
// PICA gives 0 instead of NaN when multiplying by inf
return Zero();
return Float<M, E>::FromFloat32(ToFloat32() * flt.ToFloat32());
}
Float<M, E> operator/(const Float<M, E>& flt) const {
return Float<M, E>::FromFloat32(ToFloat32() / flt.ToFloat32());
}
Float<M, E> operator+(const Float<M, E>& flt) const {
return Float<M, E>::FromFloat32(ToFloat32() + flt.ToFloat32());
}
Float<M, E> operator-(const Float<M, E>& flt) const {
return Float<M, E>::FromFloat32(ToFloat32() - flt.ToFloat32());
}
Float<M, E>& operator*=(const Float<M, E>& flt) {
if ((this->value == 0.f && !std::isnan(flt.value)) ||
(flt.value == 0.f && !std::isnan(this->value)))
// PICA gives 0 instead of NaN when multiplying by inf
*this = Zero();
else
value *= flt.ToFloat32();
return *this;
}
Float<M, E>& operator/=(const Float<M, E>& flt) {
value /= flt.ToFloat32();
return *this;
}
Float<M, E>& operator+=(const Float<M, E>& flt) {
value += flt.ToFloat32();
return *this;
}
Float<M, E>& operator-=(const Float<M, E>& flt) {
value -= flt.ToFloat32();
return *this;
}
Float<M, E> operator-() const {
return Float<M, E>::FromFloat32(-ToFloat32());
}
bool operator<(const Float<M, E>& flt) const {
return ToFloat32() < flt.ToFloat32();
}
bool operator>(const Float<M, E>& flt) const {
return ToFloat32() > flt.ToFloat32();
}
bool operator>=(const Float<M, E>& flt) const {
return ToFloat32() >= flt.ToFloat32();
}
bool operator<=(const Float<M, E>& flt) const {
return ToFloat32() <= flt.ToFloat32();
}
bool operator==(const Float<M, E>& flt) const {
return ToFloat32() == flt.ToFloat32();
}
bool operator!=(const Float<M, E>& flt) const {
return ToFloat32() != flt.ToFloat32();
}
private:
static const unsigned MASK = (1 << (M + E + 1)) - 1;
static const unsigned MANTISSA_MASK = (1 << M) - 1;
static const unsigned EXPONENT_MASK = (1 << E) - 1;
// Stored as a regular float, merely for convenience
// TODO: Perform proper arithmetic on this!
float value;
};
using float24 = Float<16, 7>;
using float20 = Float<12, 7>;
using float16 = Float<10, 5>;
} // namespace Pica
|
