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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
|
/********************************************************************
* *
* THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. *
* *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 *
* BY THE Xiph.Org FOUNDATION http://www.xiph.org/ *
* *
********************************************************************
function: miscellaneous math and prototypes
********************************************************************/
#include "config-tremor.h"
#ifndef _V_RANDOM_H_
#define _V_RANDOM_H_
#include "ivorbiscodec.h"
#include "os_types.h"
#include "asm_arm.h"
#include "asm_mcf5249.h"
/* Some prototypes that were not defined elsewhere */
void *_vorbis_block_alloc(vorbis_block *vb,long bytes);
void _vorbis_block_ripcord(vorbis_block *vb);
extern int _ilog(unsigned int v);
#ifndef _V_WIDE_MATH
#define _V_WIDE_MATH
#ifndef _LOW_ACCURACY_
/* 64 bit multiply */
/* #include <sys/types.h> */
#if BYTE_ORDER==LITTLE_ENDIAN
union magic {
struct {
ogg_int32_t lo;
ogg_int32_t hi;
} halves;
ogg_int64_t whole;
};
#elif BYTE_ORDER==BIG_ENDIAN
union magic {
struct {
ogg_int32_t hi;
ogg_int32_t lo;
} halves;
ogg_int64_t whole;
};
#endif
static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
union magic magic;
magic.whole = (ogg_int64_t)x * y;
return magic.halves.hi;
}
static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
return MULT32(x,y)<<1;
}
static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
union magic magic;
magic.whole = (ogg_int64_t)x * y;
return ((ogg_uint32_t)(magic.halves.lo)>>15) | ((magic.halves.hi)<<17);
}
#else
/* 32 bit multiply, more portable but less accurate */
/*
* Note: Precision is biased towards the first argument therefore ordering
* is important. Shift values were chosen for the best sound quality after
* many listening tests.
*/
/*
* For MULT32 and MULT31: The second argument is always a lookup table
* value already preshifted from 31 to 8 bits. We therefore take the
* opportunity to save on text space and use unsigned char for those
* tables in this case.
*/
static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
return (x >> 9) * y; /* y preshifted >>23 */
}
static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
return (x >> 8) * y; /* y preshifted >>23 */
}
static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
return (x >> 6) * y; /* y preshifted >>9 */
}
#endif
/*
* This should be used as a memory barrier, forcing all cached values in
* registers to wr writen back to memory. Might or might not be beneficial
* depending on the architecture and compiler.
*/
#define MB()
/*
* The XPROD functions are meant to optimize the cross products found all
* over the place in mdct.c by forcing memory operation ordering to avoid
* unnecessary register reloads as soon as memory is being written to.
* However this is only beneficial on CPUs with a sane number of general
* purpose registers which exclude the Intel x86. On Intel, better let the
* compiler actually reload registers directly from original memory by using
* macros.
*/
/* replaced XPROD32 with a macro to avoid memory reference
_x, _y are the results (must be l-values) */
#define XPROD32(_a, _b, _t, _v, _x, _y) \
{ (_x)=MULT32(_a,_t)+MULT32(_b,_v); \
(_y)=MULT32(_b,_t)-MULT32(_a,_v); }
#ifdef __i386__
#define XPROD31(_a, _b, _t, _v, _x, _y) \
{ *(_x)=MULT31(_a,_t)+MULT31(_b,_v); \
*(_y)=MULT31(_b,_t)-MULT31(_a,_v); }
#define XNPROD31(_a, _b, _t, _v, _x, _y) \
{ *(_x)=MULT31(_a,_t)-MULT31(_b,_v); \
*(_y)=MULT31(_b,_t)+MULT31(_a,_v); }
#else
static inline void XPROD31(ogg_int32_t a, ogg_int32_t b,
ogg_int32_t t, ogg_int32_t v,
ogg_int32_t *x, ogg_int32_t *y)
{
*x = MULT31(a, t) + MULT31(b, v);
*y = MULT31(b, t) - MULT31(a, v);
}
static inline void XNPROD31(ogg_int32_t a, ogg_int32_t b,
ogg_int32_t t, ogg_int32_t v,
ogg_int32_t *x, ogg_int32_t *y)
{
*x = MULT31(a, t) - MULT31(b, v);
*y = MULT31(b, t) + MULT31(a, v);
}
#endif
#ifndef _V_VECT_OPS
#define _V_VECT_OPS
/* generic misc.h has symmetrical versions of vect_add_right_left
and vect_add_left_right (since symmetrical versions of
vect_mult_fw and vect_mult_bw i.e. both use MULT31) */
static inline
void vect_add_right_left(ogg_int32_t *x, const ogg_int32_t *y, int n)
{
while (n>0) {
*x++ += *y++;
n--;
}
}
static inline
void vect_add_left_right(ogg_int32_t *x, const ogg_int32_t *y, int n)
{
vect_add_right_left(x,y,n);
}
static inline
void vect_mult_fw(ogg_int32_t *data, LOOKUP_T *window, int n)
{
while(n>0) {
*data = MULT31(*data, *window);
data++;
window++;
n--;
}
}
static inline
void vect_mult_bw(ogg_int32_t *data, LOOKUP_T *window, int n)
{
while(n>0) {
*data = MULT31(*data, *window);
data++;
window--;
n--;
}
}
/* generic memcpy is probably optimal */
static inline void vect_copy(ogg_int32_t *x, const ogg_int32_t *y, int n)
{
memcpy(x,y,n*sizeof(ogg_int32_t));
}
#endif
#endif
static inline ogg_int32_t VFLOAT_MULT(ogg_int32_t a,ogg_int32_t ap,
ogg_int32_t b,ogg_int32_t bp,
ogg_int32_t *p){
if(a && b){
#ifndef _LOW_ACCURACY_
*p=ap+bp+32;
return MULT32(a,b);
#else
*p=ap+bp+31;
return (a>>15)*(b>>16);
#endif
}else
return 0;
}
static inline ogg_int32_t VFLOAT_MULTI(ogg_int32_t a,ogg_int32_t ap,
ogg_int32_t i,
ogg_int32_t *p){
int ip=_ilog(abs(i))-31;
return VFLOAT_MULT(a,ap,i<<-ip,ip,p);
}
static inline ogg_int32_t VFLOAT_ADD(ogg_int32_t a,ogg_int32_t ap,
ogg_int32_t b,ogg_int32_t bp,
ogg_int32_t *p){
if(!a){
*p=bp;
return b;
}else if(!b){
*p=ap;
return a;
}
/* yes, this can leak a bit. */
if(ap>bp){
int shift=ap-bp+1;
*p=ap+1;
a>>=1;
if(shift<32){
b=(b+(1<<(shift-1)))>>shift;
}else{
b=0;
}
}else{
int shift=bp-ap+1;
*p=bp+1;
b>>=1;
if(shift<32){
a=(a+(1<<(shift-1)))>>shift;
}else{
a=0;
}
}
a+=b;
if((a&0xc0000000)==0xc0000000 ||
(a&0xc0000000)==0){
a<<=1;
(*p)--;
}
return(a);
}
#endif
|
