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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
/*
* rpf.c: Conversion of float to reduced precision format values
*
* Written by: Stefan Frank
* Richard Krampfl
* Ullrich Hafner
*
* This file is part of FIASCO («F»ractal «I»mage «A»nd «S»equence «CO»dec)
* Copyright (C) 1994-2000 Ullrich Hafner <hafner@bigfoot.de>
*/
/*
* $Date: 2000/06/14 20:49:37 $
* $Author: hafner $
* $Revision: 5.1 $
* $State: Exp $
*/
#include "pm_config.h"
#include "config.h"
#include "types.h"
#include "macros.h"
#include "error.h"
#include "misc.h"
#include "rpf.h"
/*
* CAUTION: The IEEE float format must be used by your compiler,
* or all following code is void!
*/
#if BYTE_ORDER == BIG_ENDIAN
/*
* Big-Endian Architecture (e.g. SUN, Motorola)
* Memory representation of integer 0x00112233 is 00,11,22,33
*/
enum real_bytes {BYTE_0, BYTE_1, BYTE_2, BYTE_3};
#else
/*
* Little-Endian Architecture (e.g. Intel, VAX, Alpha)
* Memory representation of integer 0x00112233 is 33,22,11,00
*/
enum real_bytes {BYTE_3, BYTE_2, BYTE_1, BYTE_0};
#endif
const int RPF_ZERO = -1;
/*****************************************************************************
private code
*****************************************************************************/
int
rtob (real_t f, const rpf_t *rpf)
/*
* Convert real number 'f' into fixed point format.
* The real number in [-'range'; +'range'] is scaled to [-1 ; +1].
* Sign and the first 'precision' - 1 bits of the mantissa are
* packed into one integer.
*
* Return value:
* real value in reduced precision format
*/
{
unsigned int mantissa;
int exponent, sign;
union
{
float f;
unsigned char c[4];
} v; /* conversion dummy */
f /= rpf->range; /* scale f to [-1,+1] */
v.f = f;
/*
* Extract mantissa (23 Bits), exponent (8 Bits) and sign (1 Bit)
*/
mantissa = ((((v.c[BYTE_1] & 127) << 8 ) | v.c[BYTE_2]) << 8) | v.c[BYTE_3];
exponent = (((v.c[BYTE_0] & 127) << 1) | (v.c[BYTE_1] & 128 ? 1 : 0)) - 126;
sign = v.c[BYTE_0] & 128 ? 1 : 0;
/*
* Generate reduced precision mantissa.
*/
mantissa >>= 1; /* shift 1 into from left */
mantissa |= (1 << 22);
if (exponent > 0)
mantissa <<= exponent;
else
mantissa >>= -exponent;
mantissa >>= (23 - rpf->mantissa_bits - 1);
mantissa += 1; /* Round last bit. */
mantissa >>= 1;
if (mantissa == 0) /* close to zero */
return RPF_ZERO;
else if (mantissa >= (1U << rpf->mantissa_bits)) /* overflow */
return sign;
else
return ((mantissa & ((1U << rpf->mantissa_bits) - 1)) << 1) | sign;
}
float
btor (int binary, const rpf_t *rpf)
/*
* Convert value 'binary' in reduced precision format to a real value.
* For more information refer to function lin_rtob() above.
*
* Return value:
* converted value
*/
{
unsigned int mantissa;
int sign, exponent;
union
{
float f;
unsigned char c[4];
} value;
if (binary == RPF_ZERO)
return 0;
if (binary < 0 || binary >= 1 << (rpf->mantissa_bits + 1))
error ("Reduced precision format: value %d out of range.", binary);
/*
* Restore IEEE float format:
* mantissa (23 Bits), exponent (8 Bits) and sign (1 Bit)
*/
sign = binary & 1;
mantissa = (binary & ((1 << (rpf->mantissa_bits + 1)) - 1)) >> 1;
mantissa <<= (23 - rpf->mantissa_bits);
exponent = 0;
if (mantissa == 0)
{
value.f = (sign ? -1.0 : 1.0);
}
else
{
while (!(mantissa & (1 << 22))) /* normalize mantissa */
{
exponent--;
mantissa <<= 1;
}
mantissa <<= 1;
value.c[BYTE_0] = (sign << 7) | ((exponent + 126) >> 1);
value.c[BYTE_1] = (((exponent + 126) & 1) << 7)
| ((mantissa >> 16) & 127);
value.c[BYTE_2] = (mantissa >> 8) & 255;
value.c[BYTE_3] = mantissa & 255;
}
return value.f * rpf->range; /* expand [ -1 ; +1 ] to
[ -range ; +range ] */
}
rpf_t *
alloc_rpf (unsigned mantissa, fiasco_rpf_range_e range)
/*
* Reduced precision format constructor.
* Allocate memory for the rpf_t structure.
* Number of mantissa bits is given by `mantissa'.
* The range of the real values is in the interval [-`range', +`range'].
* In case of invalid parameters, a structure with default values is
* returned.
*
* Return value
* pointer to the new rpf structure
*/
{
rpf_t *rpf = Calloc (1, sizeof (rpf_t));
if (mantissa < 2)
{
warning (_("Size of RPF mantissa has to be in the interval [2,8]. "
"Using minimum value 2.\n"));
mantissa = 2;
}
else if (mantissa > 8)
{
warning (_("Size of RPF mantissa has to be in the interval [2,8]. "
"Using maximum value 8.\n"));
mantissa = 2;
}
rpf->mantissa_bits = mantissa;
rpf->range_e = range;
switch (range)
{
case FIASCO_RPF_RANGE_0_75:
rpf->range = 0.75;
break;
case FIASCO_RPF_RANGE_1_50:
rpf->range = 1.50;
break;
case FIASCO_RPF_RANGE_2_00:
rpf->range = 2.00;
break;
case FIASCO_RPF_RANGE_1_00:
rpf->range = 1.00;
break;
default:
warning (_("Invalid RPF range specified. Using default value 1.0."));
rpf->range = 1.00;
rpf->range_e = FIASCO_RPF_RANGE_1_00;
break;
}
return rpf;
}
|
