mirror of
https://github.com/darlinghq/darling-gdb.git
synced 2024-11-27 05:50:43 +00:00
6efd877de5
Removed some unneeded files. configure.in: Don't look for te-386bsd.h, which doesn't exist...
184 lines
6.8 KiB
C
184 lines
6.8 KiB
C
/* flonum_const.c - Useful Flonum constants
|
|
Copyright (C) 1987, 1990, 1991, 1992 Free Software Foundation, Inc.
|
|
|
|
This file is part of GAS, the GNU Assembler.
|
|
|
|
GAS is free software; you can redistribute it and/or modify
|
|
it under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation; either version 2, or (at your option)
|
|
any later version.
|
|
|
|
GAS is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with GAS; see the file COPYING. If not, write to
|
|
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */
|
|
|
|
#include "flonum.h"
|
|
/* JF: I added the last entry to this table, and I'm not
|
|
sure if its right or not. Could go either way. I wish
|
|
I really understood this stuff. */
|
|
|
|
|
|
const int table_size_of_flonum_powers_of_ten = 11;
|
|
|
|
static const LITTLENUM_TYPE zero[] =
|
|
{1};
|
|
|
|
/***********************************************************************\
|
|
* *
|
|
* Warning: the low order bits may be WRONG here. *
|
|
* I took this from a suspect bc(1) script. *
|
|
* "minus_X"[] is supposed to be 10^(2^-X) expressed in base 2^16. *
|
|
* The radix point is just AFTER the highest element of the [] *
|
|
* *
|
|
* Because bc rounds DOWN for printing (I think), the lowest *
|
|
* significance littlenums should probably have 1 added to them. *
|
|
* *
|
|
\***********************************************************************/
|
|
|
|
/* JF: If this equals 6553/(2^16)+39321/(2^32)+... it approaches .1 */
|
|
static const LITTLENUM_TYPE minus_1[] =
|
|
{
|
|
39321, 39321, 39321, 39321, 39321, 39321, 39321, 39321, 39321, 39321,
|
|
39321, 39321, 39321, 39321, 39321, 39321, 39321, 39321, 39321, 6553};
|
|
static const LITTLENUM_TYPE plus_1[] =
|
|
{10};
|
|
|
|
/* JF: If this equals 655/(2^16) + 23592/(2^32) + ... it approaches .01 */
|
|
static const LITTLENUM_TYPE minus_2[] =
|
|
{
|
|
10485, 36700, 62914, 23592, 49807, 10485, 36700, 62914, 23592, 49807,
|
|
10485, 36700, 62914, 23592, 49807, 10485, 36700, 62914, 23592, 655};
|
|
static const LITTLENUM_TYPE plus_2[] =
|
|
{100};
|
|
|
|
/* This approaches .0001 */
|
|
static const LITTLENUM_TYPE minus_3[] =
|
|
{
|
|
52533, 20027, 37329, 65116, 64067, 60397, 14784, 18979, 33659, 19503,
|
|
2726, 9542, 629, 2202, 40475, 10590, 4299, 47815, 36280, 6};
|
|
static const LITTLENUM_TYPE plus_3[] =
|
|
{10000};
|
|
|
|
/* JF: this approaches 1e-8 */
|
|
static const LITTLENUM_TYPE minus_4[] =
|
|
{
|
|
22516, 49501, 54293, 19424, 60699, 6716, 24348, 22618, 23904, 21327,
|
|
3919, 44703, 19149, 28803, 48959, 6259, 50273, 62237, 42};
|
|
/* This equals 1525 * 2^16 + 57600 */
|
|
static const LITTLENUM_TYPE plus_4[] =
|
|
{57600, 1525};
|
|
|
|
/* This approaches 1e-16 */
|
|
static const LITTLENUM_TYPE minus_5[] =
|
|
{
|
|
22199, 45957, 17005, 26266, 10526, 16260, 55017, 35680, 40443, 19789,
|
|
17356, 30195, 55905, 28426, 63010, 44197, 1844};
|
|
static const LITTLENUM_TYPE plus_5[] =
|
|
{28609, 34546, 35};
|
|
|
|
static const LITTLENUM_TYPE minus_6[] =
|
|
{
|
|
30926, 26518, 13110, 43018, 54982, 48258, 24658, 15209, 63366, 11929,
|
|
20069, 43857, 60487, 51};
|
|
static const LITTLENUM_TYPE plus_6[] =
|
|
{61313, 34220, 16731, 11629, 1262};
|
|
|
|
static const LITTLENUM_TYPE minus_7[] =
|
|
{
|
|
29819, 14733, 21490, 40602, 31315, 65186, 2695};
|
|
static const LITTLENUM_TYPE plus_7[] =
|
|
{
|
|
7937, 49002, 60772, 28216, 38893, 55975, 63988, 59711, 20227, 24};
|
|
|
|
static const LITTLENUM_TYPE minus_8[] =
|
|
{
|
|
45849, 19069, 18068, 36324, 37948, 48745, 10873, 64360, 15961, 20566,
|
|
24178, 15922, 59427, 110};
|
|
static const LITTLENUM_TYPE plus_8[] =
|
|
{
|
|
15873, 11925, 39177, 991, 14589, 19735, 25347, 65086, 53853, 938,
|
|
37209, 47086, 33626, 23253, 32586, 42547, 9731, 59679, 590};
|
|
|
|
static const LITTLENUM_TYPE minus_9[] =
|
|
{
|
|
63601, 55221, 43562, 33661, 29067, 28203, 65417, 64352, 22462, 41110,
|
|
12570, 28635, 23199, 50572, 28471, 27074, 46375, 64028, 13106, 63700,
|
|
32698, 17493, 32420, 34382, 22750, 20681, 12300};
|
|
static const LITTLENUM_TYPE plus_9[] =
|
|
{
|
|
63564, 61556, 29377, 54467, 18621, 28141, 36415, 61241, 47119, 30026,
|
|
19740, 46002, 13541, 61413, 30480, 38664, 32205, 50593, 51112, 48904,
|
|
48263, 43814, 286, 30826, 52813, 62575, 61390, 24540, 21495, 5};
|
|
|
|
static const LITTLENUM_TYPE minus_10[] =
|
|
{
|
|
50313, 34681, 1464, 25889, 19575, 41125, 17635, 4598, 49708, 13427,
|
|
17287, 56115, 53783, 38255, 32415, 17778, 31596, 7557, 20951, 18477,
|
|
40353, 1178, 44405, 11837, 11571, 50963, 15649, 11698, 40675, 2308,};
|
|
static const LITTLENUM_TYPE plus_10[] =
|
|
{
|
|
18520, 53764, 54535, 61910, 61962, 59843, 46270, 58053, 12473, 63785,
|
|
2449, 43230, 50044, 47595, 10403, 35766, 32607, 1124, 24966, 35044,
|
|
25524, 23631, 18826, 14518, 58448, 14562, 49618, 5588, 25396, 28};
|
|
|
|
static const LITTLENUM_TYPE minus_11[] =
|
|
{
|
|
6223, 59909, 62437, 59960, 14652, 45336, 48800, 7647, 51962, 37982,
|
|
60436, 58176, 26767, 8440, 9831, 48556, 20994, 14148, 6757, 17221,
|
|
60624, 46129, 53210, 44085, 54016, 24259, 11232, 21229, 21313, 81,};
|
|
static const LITTLENUM_TYPE plus_11[] =
|
|
{
|
|
36159, 2055, 33615, 61362, 23581, 62454, 9748, 15275, 39284, 58636,
|
|
16269, 42793, 47240, 45774, 50861, 48400, 9413, 40281, 4030, 9572,
|
|
7984, 33038, 59522, 19450, 40593, 24486, 54320, 6661, 55766, 805,};
|
|
|
|
/* Shut up complaints about differing pointer types. They only differ
|
|
in the const attribute, but there isn't any easy way to do this
|
|
*/
|
|
#define X (LITTLENUM_TYPE *)
|
|
|
|
const FLONUM_TYPE flonum_negative_powers_of_ten[] =
|
|
{
|
|
{X zero, X zero, X zero, 0, '+'},
|
|
{X minus_1, X minus_1 + 19, X minus_1 + 19, -20, '+'},
|
|
{X minus_2, X minus_2 + 19, X minus_2 + 19, -20, '+'},
|
|
{X minus_3, X minus_3 + 19, X minus_3 + 19, -20, '+'},
|
|
{X minus_4, X minus_4 + 18, X minus_4 + 18, -20, '+'},
|
|
{X minus_5, X minus_5 + 16, X minus_5 + 16, -20, '+'},
|
|
{X minus_6, X minus_6 + 13, X minus_6 + 13, -20, '+'},
|
|
{X minus_7, X minus_7 + 6, X minus_7 + 6, -20, '+'},
|
|
{X minus_8, X minus_8 + 13, X minus_8 + 13, -40, '+'},
|
|
{X minus_9, X minus_9 + 26, X minus_9 + 26, -80, '+'},
|
|
{X minus_10, X minus_10 + 29, X minus_10 + 29, -136, '+'},
|
|
{X minus_11, X minus_11 + 29, X minus_11 + 29, -242, '+'},
|
|
};
|
|
|
|
const FLONUM_TYPE flonum_positive_powers_of_ten[] =
|
|
{
|
|
{X zero, X zero, X zero, 0, '+'},
|
|
{X plus_1, X plus_1 + 0, X plus_1 + 0, 0, '+'},
|
|
{X plus_2, X plus_2 + 0, X plus_2 + 0, 0, '+'},
|
|
{X plus_3, X plus_3 + 0, X plus_3 + 0, 0, '+'},
|
|
{X plus_4, X plus_4 + 1, X plus_4 + 1, 0, '+'},
|
|
{X plus_5, X plus_5 + 2, X plus_5 + 2, 1, '+'},
|
|
{X plus_6, X plus_6 + 4, X plus_6 + 4, 2, '+'},
|
|
{X plus_7, X plus_7 + 9, X plus_7 + 9, 4, '+'},
|
|
{X plus_8, X plus_8 + 18, X plus_8 + 18, 8, '+'},
|
|
{X plus_9, X plus_9 + 29, X plus_9 + 29, 24, '+'},
|
|
{X plus_10, X plus_10 + 29, X plus_10 + 29, 77, '+'},
|
|
{X plus_11, X plus_11 + 29, X plus_11 + 29, 183, '+'},
|
|
};
|
|
|
|
#ifdef VMS
|
|
dummy1 ()
|
|
{
|
|
}
|
|
|
|
#endif
|
|
/* end of flonum_const.c */
|