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
|
/*
* Mesa 3-D graphics library
*
* Copyright (C) 1999-2007 Brian Paul All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include <math.h>
#include <assert.h>
#include "half_float.h"
#include "rounding.h"
typedef union { float f; int32_t i; uint32_t u; } fi_type;
/**
* Convert a 4-byte float to a 2-byte half float.
*
* Not all float32 values can be represented exactly as a float16 value. We
* round such intermediate float32 values to the nearest float16. When the
* float32 lies exactly between to float16 values, we round to the one with
* an even mantissa.
*
* This rounding behavior has several benefits:
* - It has no sign bias.
*
* - It reproduces the behavior of real hardware: opcode F32TO16 in Intel's
* GPU ISA.
*
* - By reproducing the behavior of the GPU (at least on Intel hardware),
* compile-time evaluation of constant packHalf2x16 GLSL expressions will
* result in the same value as if the expression were executed on the GPU.
*/
uint16_t
_mesa_float_to_half(float val)
{
const fi_type fi = {val};
const int flt_m = fi.i & 0x7fffff;
const int flt_e = (fi.i >> 23) & 0xff;
const int flt_s = (fi.i >> 31) & 0x1;
int s, e, m = 0;
uint16_t result;
/* sign bit */
s = flt_s;
/* handle special cases */
if ((flt_e == 0) && (flt_m == 0)) {
/* zero */
/* m = 0; - already set */
e = 0;
}
else if ((flt_e == 0) && (flt_m != 0)) {
/* denorm -- denorm float maps to 0 half */
/* m = 0; - already set */
e = 0;
}
else if ((flt_e == 0xff) && (flt_m == 0)) {
/* infinity */
/* m = 0; - already set */
e = 31;
}
else if ((flt_e == 0xff) && (flt_m != 0)) {
/* NaN */
m = 1;
e = 31;
}
else {
/* regular number */
const int new_exp = flt_e - 127;
if (new_exp < -14) {
/* The float32 lies in the range (0.0, min_normal16) and is rounded
* to a nearby float16 value. The result will be either zero, subnormal,
* or normal.
*/
e = 0;
m = _mesa_lroundevenf((1 << 24) * fabsf(fi.f));
}
else if (new_exp > 15) {
/* map this value to infinity */
/* m = 0; - already set */
e = 31;
}
else {
/* The float32 lies in the range
* [min_normal16, max_normal16 + max_step16)
* and is rounded to a nearby float16 value. The result will be
* either normal or infinite.
*/
e = new_exp + 15;
m = _mesa_lroundevenf(flt_m / (float) (1 << 13));
}
}
assert(0 <= m && m <= 1024);
if (m == 1024) {
/* The float32 was rounded upwards into the range of the next exponent,
* so bump the exponent. This correctly handles the case where f32
* should be rounded up to float16 infinity.
*/
++e;
m = 0;
}
result = (s << 15) | (e << 10) | m;
return result;
}
/**
* Convert a 2-byte half float to a 4-byte float.
* Based on code from:
* http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008786.html
*/
float
_mesa_half_to_float(uint16_t val)
{
/* XXX could also use a 64K-entry lookup table */
const int m = val & 0x3ff;
const int e = (val >> 10) & 0x1f;
const int s = (val >> 15) & 0x1;
int flt_m, flt_e, flt_s;
fi_type fi;
float result;
/* sign bit */
flt_s = s;
/* handle special cases */
if ((e == 0) && (m == 0)) {
/* zero */
flt_m = 0;
flt_e = 0;
}
else if ((e == 0) && (m != 0)) {
/* denorm -- denorm half will fit in non-denorm single */
const float half_denorm = 1.0f / 16384.0f; /* 2^-14 */
float mantissa = ((float) (m)) / 1024.0f;
float sign = s ? -1.0f : 1.0f;
return sign * mantissa * half_denorm;
}
else if ((e == 31) && (m == 0)) {
/* infinity */
flt_e = 0xff;
flt_m = 0;
}
else if ((e == 31) && (m != 0)) {
/* NaN */
flt_e = 0xff;
flt_m = 1;
}
else {
/* regular */
flt_e = e + 112;
flt_m = m << 13;
}
fi.i = (flt_s << 31) | (flt_e << 23) | flt_m;
result = fi.f;
return result;
}
|
