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
|
/*---------------------------------------------------------------------------+
| poly.h |
| $Id: poly.h,v 1.4 2001/10/06 03:53:46 bdenney Exp $
| |
| Header file for the FPU-emu poly*.c source files. |
| |
| Copyright (C) 1994,1999 |
| W. Metzenthen, 22 Parker St, Ormond, Vic 3163, |
| Australia. E-mail billm@melbpc.org.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 {
#ifdef EMU_BIG_ENDIAN
u32 msw;
u32 midw;
u32 lsw;
#else
u32 lsw;
u32 midw;
u32 msw;
#endif
} GCC_ATTRIBUTE((packed)) Xsig;
asmlinkage void mul64(u64 const *a, u64 const *b,
u64 *result);
asmlinkage void polynomial_Xsig(Xsig *, const u64 *x,
const u64 terms[], const int n);
asmlinkage void mul32_Xsig(Xsig *, const u32 mult);
asmlinkage void mul64_Xsig(Xsig *, const u64 *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(const Xsig *x1, const Xsig *x2, Xsig *dest);
/* Macro to extract the most significant 32 bits from a 64bit quantity */
#ifdef EMU_BIG_ENDIAN
#define LL_MSW(x) (((u32 *)&x)[0])
#else
#define LL_MSW(x) (((u32 *)&x)[1])
#endif
/* Macro to initialize an Xsig struct */
#ifdef EMU_BIG_ENDIAN
#define MK_XSIG(a,b,c) { a, b, c }
#else
#define MK_XSIG(a,b,c) { c, b, a }
#endif
/* Macro to access the 8 ms bytes of an Xsig as a 64bit quantity */
#ifdef EMU_BIG_ENDIAN
#define XSIG_LL(x) (*(u64 *)&x.msw)
#else
#define XSIG_LL(x) (*(u64 *)&x.midw)
#endif
/*
Need to run gcc with optimizations on to get these to
actually be in-line.
*/
/* Multiply two fixed-point 32 bit numbers, producing a 32 bit result.
The answer is the ms word of the product. */
BX_C_INLINE
u32 mul_32_32(const u32 arg1, const u32 arg2)
{
#ifdef NO_ASSEMBLER
return (((u64)arg1) * arg2) >> 32;
#else
/* Some versions of gcc make it difficult to stop eax from being clobbered.
Merely specifying that it is used doesn't work...
*/
int retval;
asm volatile ("mull %2; movl %%edx,%%eax" \
:"=a" (retval) \
:"0" (arg1), "g" (arg2) \
:"dx");
return retval;
#endif
}
/* Add the 12 byte Xsig x2 to Xsig dest, with no checks for overflow. */
BX_C_INLINE
void add_Xsig_Xsig(Xsig *dest, const Xsig *x2)
{
#ifdef NO_ASSEMBLER
dest->lsw += x2->lsw;
if ( dest->lsw < x2->lsw )
{
dest->midw ++;
if ( dest->midw == 0 )
dest->msw ++;
}
dest->midw += x2->midw;
if ( dest->midw < x2->midw )
{
dest->msw ++;
}
dest->msw += x2->msw;
#else
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");
#endif
}
/* Add the 12 byte Xsig x2 to Xsig dest, adjust exp if overflow occurs. */
BX_C_INLINE
void add_two_Xsig(Xsig *dest, const Xsig *x2, s32 *exp)
{
#ifdef NO_ASSEMBLER
int ovfl = 0;
dest->lsw += x2->lsw;
if ( dest->lsw < x2->lsw )
{
dest->midw ++;
if ( dest->midw == 0 )
{
dest->msw ++;
if ( dest->msw == 0 )
ovfl = 1;
}
}
dest->midw += x2->midw;
if ( dest->midw < x2->midw )
{
dest->msw ++;
if ( dest->msw == 0 )
ovfl = 1;
}
dest->msw += x2->msw;
if ( dest->msw < x2->msw )
ovfl = 1;
if ( ovfl )
{
(*exp) ++;
dest->lsw >>= 1;
if ( dest->midw & 1 )
dest->lsw |= 0x80000000;
dest->midw >>= 1;
if ( dest->msw & 1 )
dest->midw |= 0x80000000;
dest->msw >>= 1;
dest->msw |= 0x80000000;
}
#else
/* 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! */
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");
#endif
}
/* Negate the 12 byte Xsig */
BX_C_INLINE
void negate_Xsig(Xsig *x)
{
#ifdef NO_ASSEMBLER
x->lsw = ~x->lsw;
x->midw = ~x->midw;
x->msw = ~x->msw;
x->lsw ++;
if ( x->lsw == 0 )
{
x->midw ++;
if ( x->midw == 0 )
x->msw ++;
}
#else
/* Negate (subtract from 1.0) the 12 byte Xsig */
/* This is faster in a loop on my 386 than using the "neg" instruction. */
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
}
#endif /* _POLY_H */
|
