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
|
/*---------------------------------------------------------------------------+
| poly.h |
| |
| Header file for the FPU-emu poly*.c source files. |
| |
| Copyright (C) 1994 |
| W. Metzenthen, 22 Parker St, Ormond, Vic 3163, |
| Australia. E-mail billm@vaxc.cc.monash.edu.au |
| |
| Declarations and definitions for functions operating on Xsig (12-byte |
| extended-significand) quantities. |
| |
+---------------------------------------------------------------------------*/
#ifndef _POLY_H
#define _POLY_H
/* This 12-byte structure is used to improve the accuracy of computation
of transcendental functions.
Intended to be used to get results better than 8-byte computation
allows. 9-byte would probably be sufficient.
*/
typedef struct {
unsigned long lsw;
unsigned long midw;
unsigned long msw;
} Xsig;
asmlinkage void mul64(unsigned long long const *a, unsigned long long const *b,
unsigned long long *result);
asmlinkage void polynomial_Xsig(Xsig *, const unsigned long long *x,
const unsigned long long terms[], const int n);
asmlinkage void mul32_Xsig(Xsig *, const unsigned long mult);
asmlinkage void mul64_Xsig(Xsig *, const unsigned long long *mult);
asmlinkage void mul_Xsig_Xsig(Xsig *dest, const Xsig *mult);
asmlinkage void shr_Xsig(Xsig *, const int n);
asmlinkage int round_Xsig(Xsig *);
asmlinkage int norm_Xsig(Xsig *);
asmlinkage void div_Xsig(Xsig *x1, const Xsig *x2, const Xsig *dest);
/* Macro to extract the most significant 32 bits from a long long */
#define LL_MSW(x) (((unsigned long *)&x)[1])
/* Macro to initialize an Xsig struct */
#define MK_XSIG(a,b,c) { c, b, a }
/* Macro to access the 8 ms bytes of an Xsig as a long long */
#define XSIG_LL(x) (*(unsigned long long *)&x.midw)
/*
Need to run gcc with optimizations on to get these to
actually be in-line.
*/
/* Multiply two fixed-point 32 bit numbers. */
extern inline void mul_32_32(const unsigned long arg1,
const unsigned long arg2,
unsigned long *out)
{
asm volatile ("movl %1,%%eax; mull %2; movl %%edx,%0" \
:"=g" (*out) \
:"g" (arg1), "g" (arg2) \
:"ax","dx");
}
/* Add the 12 byte Xsig x2 to Xsig dest, with no checks for overflow. */
extern inline void add_Xsig_Xsig(Xsig *dest, const Xsig *x2)
{
asm volatile ("movl %1,%%edi; movl %2,%%esi;
movl (%%esi),%%eax; addl %%eax,(%%edi);
movl 4(%%esi),%%eax; adcl %%eax,4(%%edi);
movl 8(%%esi),%%eax; adcl %%eax,8(%%edi);"
:"=g" (*dest):"g" (dest), "g" (x2)
:"ax","si","di");
}
/* Add the 12 byte Xsig x2 to Xsig dest, adjust exp if overflow occurs. */
/* Note: the constraints in the asm statement didn't always work properly
with gcc 2.5.8. Changing from using edi to using ecx got around the
problem, but keep fingers crossed! */
extern inline int add_two_Xsig(Xsig *dest, const Xsig *x2, long int *exp)
{
asm volatile ("movl %2,%%ecx; movl %3,%%esi;
movl (%%esi),%%eax; addl %%eax,(%%ecx);
movl 4(%%esi),%%eax; adcl %%eax,4(%%ecx);
movl 8(%%esi),%%eax; adcl %%eax,8(%%ecx);
jnc 0f;
rcrl 8(%%ecx); rcrl 4(%%ecx); rcrl (%%ecx)
movl %4,%%ecx; incl (%%ecx)
movl $1,%%eax; jmp 1f;
0: xorl %%eax,%%eax;
1:"
:"=g" (*exp), "=g" (*dest)
:"g" (dest), "g" (x2), "g" (exp)
:"cx","si","ax");
}
/* Negate (subtract from 1.0) the 12 byte Xsig */
/* This is faster in a loop on my 386 than using the "neg" instruction. */
extern inline void negate_Xsig(Xsig *x)
{
asm volatile("movl %1,%%esi; "
"xorl %%ecx,%%ecx; "
"movl %%ecx,%%eax; subl (%%esi),%%eax; movl %%eax,(%%esi); "
"movl %%ecx,%%eax; sbbl 4(%%esi),%%eax; movl %%eax,4(%%esi); "
"movl %%ecx,%%eax; sbbl 8(%%esi),%%eax; movl %%eax,8(%%esi); "
:"=g" (*x):"g" (x):"si","ax","cx");
}
#endif _POLY_H
|
