1 |
282 |
jeremybenn |
/* Copyright (C) 2003, 2009 Free Software Foundation, Inc.
|
2 |
|
|
|
3 |
|
|
This file is free software; you can redistribute it and/or modify it
|
4 |
|
|
under the terms of the GNU General Public License as published by the
|
5 |
|
|
Free Software Foundation; either version 3, or (at your option) any
|
6 |
|
|
later version.
|
7 |
|
|
|
8 |
|
|
This file is distributed in the hope that it will be useful, but
|
9 |
|
|
WITHOUT ANY WARRANTY; without even the implied warranty of
|
10 |
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
11 |
|
|
General Public License for more details.
|
12 |
|
|
|
13 |
|
|
Under Section 7 of GPL version 3, you are granted additional
|
14 |
|
|
permissions described in the GCC Runtime Library Exception, version
|
15 |
|
|
3.1, as published by the Free Software Foundation.
|
16 |
|
|
|
17 |
|
|
You should have received a copy of the GNU General Public License and
|
18 |
|
|
a copy of the GCC Runtime Library Exception along with this program;
|
19 |
|
|
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
20 |
|
|
<http://www.gnu.org/licenses/>. */
|
21 |
|
|
|
22 |
|
|
|
23 |
|
|
/* Calculate division table for SH5Media integer division
|
24 |
|
|
Contributed by Joern Rennecke
|
25 |
|
|
joern.rennecke@superh.com */
|
26 |
|
|
|
27 |
|
|
#include <stdio.h>
|
28 |
|
|
#include <math.h>
|
29 |
|
|
|
30 |
|
|
#define BITS 5
|
31 |
|
|
#define N_ENTRIES (1 << BITS)
|
32 |
|
|
#define CUTOFF_BITS 20
|
33 |
|
|
|
34 |
|
|
#define BIAS (-330)
|
35 |
|
|
|
36 |
|
|
double max_defect = 0.;
|
37 |
|
|
double max_defect_x;
|
38 |
|
|
|
39 |
|
|
double min_defect = 1e9;
|
40 |
|
|
double min_defect_x;
|
41 |
|
|
|
42 |
|
|
double max_defect2 = 0.;
|
43 |
|
|
double max_defect2_x;
|
44 |
|
|
|
45 |
|
|
double min_defect2 = 0.;
|
46 |
|
|
double min_defect2_x;
|
47 |
|
|
|
48 |
|
|
double min_defect3 = 01e9;
|
49 |
|
|
double min_defect3_x;
|
50 |
|
|
int min_defect3_val;
|
51 |
|
|
|
52 |
|
|
double max_defect3 = 0.;
|
53 |
|
|
double max_defect3_x;
|
54 |
|
|
int max_defect3_val;
|
55 |
|
|
|
56 |
|
|
static double note_defect3 (int val, double d2, double y2d, double x)
|
57 |
|
|
{
|
58 |
|
|
int cutoff_val = val >> CUTOFF_BITS;
|
59 |
|
|
double cutoff;
|
60 |
|
|
double defect;
|
61 |
|
|
|
62 |
|
|
if (val < 0)
|
63 |
|
|
cutoff_val++;
|
64 |
|
|
cutoff = (cutoff_val * (1<<CUTOFF_BITS) - val) * y2d;
|
65 |
|
|
defect = cutoff + val * d2;
|
66 |
|
|
if (val < 0)
|
67 |
|
|
defect = - defect;
|
68 |
|
|
if (defect > max_defect3)
|
69 |
|
|
{
|
70 |
|
|
max_defect3 = defect;
|
71 |
|
|
max_defect3_x = x;
|
72 |
|
|
max_defect3_val = val;
|
73 |
|
|
}
|
74 |
|
|
if (defect < min_defect3)
|
75 |
|
|
{
|
76 |
|
|
min_defect3 = defect;
|
77 |
|
|
min_defect3_x = x;
|
78 |
|
|
min_defect3_val = val;
|
79 |
|
|
}
|
80 |
|
|
}
|
81 |
|
|
|
82 |
|
|
/* This function assumes 32-bit integers. */
|
83 |
|
|
static double
|
84 |
|
|
calc_defect (double x, int constant, int factor)
|
85 |
|
|
{
|
86 |
|
|
double y0 = (constant - (int) floor ((x * factor * 64.))) / 16384.;
|
87 |
|
|
double y1 = 2 * y0 -y0 * y0 * (x + BIAS / (1.*(1LL<<30)));
|
88 |
|
|
double y2d0, y2d;
|
89 |
|
|
int y2d1;
|
90 |
|
|
double d, d2;
|
91 |
|
|
|
92 |
|
|
y1 = floor (y1 * (1024 * 1024 * 1024)) / (1024 * 1024 * 1024);
|
93 |
|
|
d = y1 - 1 / x;
|
94 |
|
|
if (d > max_defect)
|
95 |
|
|
{
|
96 |
|
|
max_defect = d;
|
97 |
|
|
max_defect_x = x;
|
98 |
|
|
}
|
99 |
|
|
if (d < min_defect)
|
100 |
|
|
{
|
101 |
|
|
min_defect = d;
|
102 |
|
|
min_defect_x = x;
|
103 |
|
|
}
|
104 |
|
|
y2d0 = floor (y1 * x * (1LL << 60-16));
|
105 |
|
|
y2d1 = (int) (long long) y2d0;
|
106 |
|
|
y2d = - floor ((y1 - y0 / (1<<30-14)) * y2d1) / (1LL<<44);
|
107 |
|
|
d2 = y1 + y2d - 1/x;
|
108 |
|
|
if (d2 > max_defect2)
|
109 |
|
|
{
|
110 |
|
|
max_defect2 = d2;
|
111 |
|
|
max_defect2_x = x;
|
112 |
|
|
}
|
113 |
|
|
if (d2 < min_defect2)
|
114 |
|
|
{
|
115 |
|
|
min_defect2 = d2;
|
116 |
|
|
min_defect2_x = x;
|
117 |
|
|
}
|
118 |
|
|
/* zero times anything is trivially zero. */
|
119 |
|
|
note_defect3 ((1 << CUTOFF_BITS) - 1, d2, y2d, x);
|
120 |
|
|
note_defect3 (1 << CUTOFF_BITS, d2, y2d, x);
|
121 |
|
|
note_defect3 ((1U << 31) - (1 << CUTOFF_BITS), d2, y2d, x);
|
122 |
|
|
note_defect3 ((1U << 31) - 1, d2, y2d, x);
|
123 |
|
|
note_defect3 (-1, d2, y2d, x);
|
124 |
|
|
note_defect3 (-(1 << CUTOFF_BITS), d2, y2d, x);
|
125 |
|
|
note_defect3 ((1U << 31) - (1 << CUTOFF_BITS) + 1, d2, y2d, x);
|
126 |
|
|
note_defect3 (-(1U << 31), d2, y2d, x);
|
127 |
|
|
return d;
|
128 |
|
|
}
|
129 |
|
|
|
130 |
|
|
int
|
131 |
|
|
main ()
|
132 |
|
|
{
|
133 |
|
|
int i;
|
134 |
|
|
unsigned char factors[N_ENTRIES];
|
135 |
|
|
short constants[N_ENTRIES];
|
136 |
|
|
int steps = N_ENTRIES / 2;
|
137 |
|
|
double step = 1. / steps;
|
138 |
|
|
double eps30 = 1. / (1024 * 1024 * 1024);
|
139 |
|
|
|
140 |
|
|
for (i = 0; i < N_ENTRIES; i++)
|
141 |
|
|
{
|
142 |
|
|
double x_low = (i < steps ? 1. : -3.) + i * step;
|
143 |
|
|
double x_high = x_low + step - eps30;
|
144 |
|
|
double x_med;
|
145 |
|
|
int factor, constant;
|
146 |
|
|
double low_defect, med_defect, high_defect, max_defect;
|
147 |
|
|
|
148 |
|
|
factor = (1./x_low- 1./x_high) / step * 256. + 0.5;
|
149 |
|
|
if (factor == 256)
|
150 |
|
|
factor = 255;
|
151 |
|
|
factors[i] = factor;
|
152 |
|
|
/* Use minimum of error function for x_med. */
|
153 |
|
|
x_med = sqrt (256./factor);
|
154 |
|
|
if (x_low < 0)
|
155 |
|
|
x_med = - x_med;
|
156 |
|
|
low_defect = 1. / x_low + x_low * factor / 256.;
|
157 |
|
|
high_defect = 1. / x_high + x_high * factor / 256.;
|
158 |
|
|
med_defect = 1. / x_med + x_med * factor / 256.;
|
159 |
|
|
max_defect
|
160 |
|
|
= ((low_defect > high_defect) ^ (x_med < 0)) ? low_defect : high_defect;
|
161 |
|
|
constant = (med_defect + max_defect) * 0.5 * 16384. + 0.5;
|
162 |
|
|
if (constant < -32768 || constant > 32767)
|
163 |
|
|
abort ();
|
164 |
|
|
constants[i] = constant;
|
165 |
|
|
calc_defect (x_low, constant, factor);
|
166 |
|
|
calc_defect (x_med, constant, factor);
|
167 |
|
|
calc_defect (x_high, constant, factor);
|
168 |
|
|
}
|
169 |
|
|
printf ("/* This table has been generated by divtab.c .\n");
|
170 |
|
|
printf ("Defects for bias %d:\n", BIAS);
|
171 |
|
|
printf (" Max defect: %e at %e\n", max_defect, max_defect_x);
|
172 |
|
|
printf (" Min defect: %e at %e\n", min_defect, min_defect_x);
|
173 |
|
|
printf (" Max 2nd step defect: %e at %e\n", max_defect2, max_defect2_x);
|
174 |
|
|
printf (" Min 2nd step defect: %e at %e\n", min_defect2, min_defect2_x);
|
175 |
|
|
printf (" Max div defect: %e at %d:%e\n", max_defect3, max_defect3_val, max_defect3_x);
|
176 |
|
|
printf (" Min div defect: %e at %d:%e\n", min_defect3, min_defect3_val, min_defect3_x);
|
177 |
|
|
printf (" Defect at 1: %e\n",
|
178 |
|
|
calc_defect (1., constants[0], factors[0]));
|
179 |
|
|
printf (" Defect at -2: %e */\n",
|
180 |
|
|
calc_defect (-2., constants[steps], factors[steps]));
|
181 |
|
|
printf ("\t.section\t.rodata\n");
|
182 |
|
|
printf ("\t.balign 2\n");
|
183 |
|
|
printf ("/* negative division constants */\n");
|
184 |
|
|
for (i = steps; i < 2 * steps; i++)
|
185 |
|
|
printf ("\t.word\t%d\n", constants[i]);
|
186 |
|
|
printf ("/* negative division factors */\n");
|
187 |
|
|
for (i = steps; i < 2*steps; i++)
|
188 |
|
|
printf ("\t.byte\t%d\n", factors[i]);
|
189 |
|
|
printf ("\t.skip %d\n", steps);
|
190 |
|
|
printf ("\t.global GLOBAL(div_table):\n");
|
191 |
|
|
printf ("GLOBAL(div_table):\n");
|
192 |
|
|
printf ("\t.skip %d\n", steps);
|
193 |
|
|
printf ("/* positive division factors */\n");
|
194 |
|
|
for (i = 0; i < steps; i++)
|
195 |
|
|
printf ("\t.byte\t%d\n", factors[i]);
|
196 |
|
|
printf ("/* positive division constants */\n");
|
197 |
|
|
for (i = 0; i < steps; i++)
|
198 |
|
|
printf ("\t.word\t%d\n", constants[i]);
|
199 |
|
|
exit (0);
|
200 |
|
|
}
|