blob: b00e9e10cdce244ba073e84f54690ec95bb5e698 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*---------------------------------------------------------------------------+
2 | poly_2xm1.c |
3 | |
4 | Function to compute 2^x-1 by a polynomial approximation. |
5 | |
6 | Copyright (C) 1992,1993,1994,1997 |
7 | W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
8 | E-mail billm@suburbia.net |
9 | |
10 | |
11 +---------------------------------------------------------------------------*/
12
13#include "exception.h"
14#include "reg_constant.h"
15#include "fpu_emu.h"
16#include "fpu_system.h"
17#include "control_w.h"
18#include "poly.h"
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#define HIPOWER 11
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010021static const unsigned long long lterms[HIPOWER] = {
22 0x0000000000000000LL, /* This term done separately as 12 bytes */
23 0xf5fdeffc162c7543LL,
24 0x1c6b08d704a0bfa6LL,
25 0x0276556df749cc21LL,
26 0x002bb0ffcf14f6b8LL,
27 0x0002861225ef751cLL,
28 0x00001ffcbfcd5422LL,
29 0x00000162c005d5f1LL,
30 0x0000000da96ccb1bLL,
31 0x0000000078d1b897LL,
32 0x000000000422b029LL
Linus Torvalds1da177e2005-04-16 15:20:36 -070033};
34
35static const Xsig hiterm = MK_XSIG(0xb17217f7, 0xd1cf79ab, 0xc8a39194);
36
37/* Four slices: 0.0 : 0.25 : 0.50 : 0.75 : 1.0,
38 These numbers are 2^(1/4), 2^(1/2), and 2^(3/4)
39 */
40static const Xsig shiftterm0 = MK_XSIG(0, 0, 0);
41static const Xsig shiftterm1 = MK_XSIG(0x9837f051, 0x8db8a96f, 0x46ad2318);
42static const Xsig shiftterm2 = MK_XSIG(0xb504f333, 0xf9de6484, 0x597d89b3);
43static const Xsig shiftterm3 = MK_XSIG(0xd744fcca, 0xd69d6af4, 0x39a68bb9);
44
45static const Xsig *shiftterm[] = { &shiftterm0, &shiftterm1,
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010046 &shiftterm2, &shiftterm3
47};
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*--- poly_2xm1() -----------------------------------------------------------+
50 | Requires st(0) which is TAG_Valid and < 1. |
51 +---------------------------------------------------------------------------*/
Ingo Molnare8d591d2008-01-30 13:30:12 +010052int poly_2xm1(u_char sign, FPU_REG *arg, FPU_REG *result)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010054 long int exponent, shift;
55 unsigned long long Xll;
56 Xsig accumulator, Denom, argSignif;
57 u_char tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010059 exponent = exponent16(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61#ifdef PARANOID
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010062 if (exponent >= 0) { /* Don't want a |number| >= 1.0 */
63 /* Number negative, too large, or not Valid. */
64 EXCEPTION(EX_INTERNAL | 0x127);
65 return 1;
66 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#endif /* PARANOID */
68
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010069 argSignif.lsw = 0;
70 XSIG_LL(argSignif) = Xll = significand(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010072 if (exponent == -1) {
73 shift = (argSignif.msw & 0x40000000) ? 3 : 2;
74 /* subtract 0.5 or 0.75 */
75 exponent -= 2;
76 XSIG_LL(argSignif) <<= 2;
77 Xll <<= 2;
78 } else if (exponent == -2) {
79 shift = 1;
80 /* subtract 0.25 */
81 exponent--;
82 XSIG_LL(argSignif) <<= 1;
83 Xll <<= 1;
84 } else
85 shift = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010087 if (exponent < -2) {
88 /* Shift the argument right by the required places. */
89 if (FPU_shrx(&Xll, -2 - exponent) >= 0x80000000U)
90 Xll++; /* round up */
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010093 accumulator.lsw = accumulator.midw = accumulator.msw = 0;
94 polynomial_Xsig(&accumulator, &Xll, lterms, HIPOWER - 1);
95 mul_Xsig_Xsig(&accumulator, &argSignif);
96 shr_Xsig(&accumulator, 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Ingo Molnar3d0d14f2008-01-30 13:30:11 +010098 mul_Xsig_Xsig(&argSignif, &hiterm); /* The leading term */
99 add_two_Xsig(&accumulator, &argSignif, &exponent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
Ingo Molnar3d0d14f2008-01-30 13:30:11 +0100101 if (shift) {
102 /* The argument is large, use the identity:
103 f(x+a) = f(a) * (f(x) + 1) - 1;
104 */
105 shr_Xsig(&accumulator, -exponent);
106 accumulator.msw |= 0x80000000; /* add 1.0 */
107 mul_Xsig_Xsig(&accumulator, shiftterm[shift]);
108 accumulator.msw &= 0x3fffffff; /* subtract 1.0 */
109 exponent = 1;
110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Ingo Molnar3d0d14f2008-01-30 13:30:11 +0100112 if (sign != SIGN_POS) {
113 /* The argument is negative, use the identity:
114 f(-x) = -f(x) / (1 + f(x))
115 */
116 Denom.lsw = accumulator.lsw;
117 XSIG_LL(Denom) = XSIG_LL(accumulator);
118 if (exponent < 0)
119 shr_Xsig(&Denom, -exponent);
120 else if (exponent > 0) {
121 /* exponent must be 1 here */
122 XSIG_LL(Denom) <<= 1;
123 if (Denom.lsw & 0x80000000)
124 XSIG_LL(Denom) |= 1;
125 (Denom.lsw) <<= 1;
126 }
127 Denom.msw |= 0x80000000; /* add 1.0 */
128 div_Xsig(&accumulator, &Denom, &accumulator);
129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Ingo Molnar3d0d14f2008-01-30 13:30:11 +0100131 /* Convert to 64 bit signed-compatible */
132 exponent += round_Xsig(&accumulator);
133
134 result = &st(0);
135 significand(result) = XSIG_LL(accumulator);
136 setexponent16(result, exponent);
137
138 tag = FPU_round(result, 1, 0, FULL_PRECISION, sign);
139
140 setsign(result, sign);
141 FPU_settag0(tag);
142
143 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145}