1 /*
2 	decode_i486.c: i486 decode
3 
4 	copyright 1998-2006 by the mpg123 project - free software under the terms of the LGPL 2.1
5 	see COPYING and AUTHORS files in distribution or http://mpg123.org
6 	initially written by Fabrice Bellard
7 
8 	One has to see if the modification for non-static memory kills this optimization (cache locality?).
9 */
10 
11 /*
12  * Subband Synthesis for MPEG Audio.
13  *
14  * Version optimized for 80486 by using integer arithmetic,
15  * multiplications by shift and add, and by increasing locality in
16  * order to fit the 8KB L1 cache. This code should be compiled with gcc
17  * 2.7.2 or higher.
18  *
19  * Note: this version does not guaranty a good accuracy. The filter
20  * coefficients are quantified on 14 bits.
21  *
22  * (c) 1998 Fabrice Bellard
23  */
24 
25 #include "mpg123lib_intern.h"
26 
27 #define FIR16_1(pos,c0,c1,c2,c3,c4,c5,c6,c7,c8,c9,c10,c11,c12,c13,c14,c15) \
28 {\
29   int sum;\
30   sum=(c0)*b0[0]+(c1)*b0[1]+(c2)*b0[2]+(c3)*b0[3]+\
31   (c4)*b0[4]+(c5)*b0[5]+(c6)*b0[6]+(c7)*b0[7]+\
32   (c8)*b0[8]+(c9)*b0[9]+(c10)*b0[10]+(c11)*b0[11]+\
33   (c12)*b0[12]+(c13)*b0[13]+(c14)*b0[14]+(c15)*b0[15];\
34   sum=(sum+(1 << 13))>>14;\
35   if (sum<-32768) sum=-32768;\
36   else if (sum>32767) sum=32767;\
37   samples[2*(pos)]=sum;\
38   b0+=FIR_BUFFER_SIZE;\
39 }
40 
41 #define FIR16_2(pos1,c0,c1,c2,c3,c4,c5,c6,c7,c8,c9,c10,c11,c12,c13,c14,c15,\
42               pos2,d0,d1,d2,d3,d4,d5,d6,d7,d8,d9,d10,d11,d12,d13,d14,d15) \
43 {\
44   int sum1,sum2,v;\
45 \
46   v=b0[0];\
47   sum1=(c0)*v;\
48   sum2=(d0)*v;\
49   v=b0[1];\
50   sum1+=(c1)*v;\
51   sum2+=(d1)*v;\
52   v=b0[2];\
53   sum1+=(c2)*v;\
54   sum2+=(d2)*v;\
55   v=b0[3];\
56   sum1+=(c3)*v;\
57   sum2+=(d3)*v;\
58   v=b0[4];\
59   sum1+=(c4)*v;\
60   sum2+=(d4)*v;\
61   v=b0[5];\
62   sum1+=(c5)*v;\
63   sum2+=(d5)*v;\
64   v=b0[6];\
65   sum1+=(c6)*v;\
66   sum2+=(d6)*v;\
67   v=b0[7];\
68   sum1+=(c7)*v;\
69   sum2+=(d7)*v;\
70   v=b0[8];\
71   sum1+=(c8)*v;\
72   sum2+=(d8)*v;\
73   v=b0[9];\
74   sum1+=(c9)*v;\
75   sum2+=(d9)*v;\
76   v=b0[10];\
77   sum1+=(c10)*v;\
78   sum2+=(d10)*v;\
79   v=b0[11];\
80   sum1+=(c11)*v;\
81   sum2+=(d11)*v;\
82   v=b0[12];\
83   sum1+=(c12)*v;\
84   sum2+=(d12)*v;\
85   v=b0[13];\
86   sum1+=(c13)*v;\
87   sum2+=(d13)*v;\
88   v=b0[14];\
89   sum1+=(c14)*v;\
90   sum2+=(d14)*v;\
91   v=b0[15];\
92   sum1+=(c15)*v;\
93   sum2+=(d15)*v;\
94 \
95   sum1=(sum1+(1<<13))>>14;\
96   sum2=(sum2+(1<<13))>>14;\
97 \
98   if (sum1<-32768) sum1=-32768;\
99   else if (sum1>32767) sum1=32767;\
100   samples[(pos1)*2]=sum1;\
101 \
102   if (sum2<-32768) sum2=-32768;\
103   else if (sum2>32767) sum2=32767;\
104   samples[(pos2)*2]=sum2;\
105   b0+=FIR_BUFFER_SIZE;\
106 }
107 
absynth_1to1_i486(real * bandPtr,int channel,mpg123_handle * fr,int nb_blocks)108 int absynth_1to1_i486(real *bandPtr, int channel, mpg123_handle *fr, int nb_blocks)
109 {
110   short *samples = (short *) (fr->buffer.data+fr->buffer.fill);
111   int *b0,**buf;
112   int clip = 0;
113   int block,b,bo_start;
114 
115   /* samples address */
116   samples+=channel;
117 
118   bo_start=fr->i486bo[channel];
119   buf = fr->int_buffs[channel];
120 
121   b=bo_start;
122   for(block=0;block<nb_blocks;block++) {
123 
124     /* FIR offset */
125     b++;
126     if (b >= FIR_BUFFER_SIZE) {
127       int *p,*q;
128       int c,i,j;
129 
130       /* we shift the buffers */
131       for(c=0;c<2;c++) {
132         p=&buf[c][0]+1;
133         q=p+(FIR_BUFFER_SIZE-FIR_SIZE);
134         for(i=0;i<17;i++) {
135           for(j=0;j<FIR_SIZE-1;j++) p[j]=q[j];
136           p+=FIR_BUFFER_SIZE;
137           q+=FIR_BUFFER_SIZE;
138         }
139       }
140       /* we update 'bo' accordingly */
141       b=fr->i486bo[channel]=FIR_SIZE;
142     }
143 
144     if(b & 1) {
145       dct64_i486(buf[1]+b,buf[0]+b,bandPtr);
146     } else {
147       dct64_i486(buf[0]+b,buf[1]+b,bandPtr);
148     }
149     bandPtr+=32;
150   }
151   fr->i486bo[channel]=b;
152 
153   /* filter bank: part 1 */
154   b=bo_start;
155   for(block=0;block<nb_blocks;block++) {
156     b++;
157     if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
158     if(b & 1) {
159       b0 = buf[0] + b - (FIR_SIZE-1);
160     } else {
161       b0 = buf[1] + b - (FIR_SIZE-1);
162     }
163 
164  FIR16_1(0,-7,53,-114,509,-1288,1643,-9372,18759,9372,1643,1288,509,114,53,7,0);
165  FIR16_2(1,-6,52,-100,515,-1197,1783,-8910,18748,9834,1489,1379,500,129,54,7,0,
166  31,0,-7,54,-129,500,-1379,1489,-9834,18748,8910,1783,1197,515,100,52,6);
167  FIR16_2(2,-6,50,-86,520,-1106,1910,-8447,18714,10294,1322,1469,488,145,55,8,0,
168  30,0,-8,55,-145,488,-1469,1322,-10294,18714,8447,1910,1106,520,86,50,6);
169  FIR16_2(3,-5,49,-73,521,-1015,2023,-7986,18657,10751,1140,1559,473,161,56,9,0,
170  29,0,-9,56,-161,473,-1559,1140,-10751,18657,7986,2023,1015,521,73,49,5);
171     samples+=64;
172   }
173   samples-=64*nb_blocks;
174 
175   /* filter bank: part 2 */
176 
177   b=bo_start;
178   for(block=0;block<nb_blocks;block++) {
179     b++;
180     if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
181     if(b & 1) {
182       b0 = buf[0] + b - (FIR_SIZE-1) + 4*FIR_BUFFER_SIZE;
183     } else {
184       b0 = buf[1] + b - (FIR_SIZE-1) + 4*FIR_BUFFER_SIZE;
185     }
186 
187  FIR16_2(4,-4,47,-61,521,-926,2123,-7528,18578,11205,944,1647,455,177,56,10,0,
188  28,0,-10,56,-177,455,-1647,944,-11205,18578,7528,2123,926,521,61,47,4);
189  FIR16_2(5,-4,45,-49,518,-837,2210,-7072,18477,11654,733,1733,434,194,57,11,0,
190  27,0,-11,57,-194,434,-1733,733,-11654,18477,7072,2210,837,518,49,45,4);
191  FIR16_2(6,-4,44,-38,514,-751,2284,-6620,18353,12097,509,1817,411,212,57,12,0,
192  26,0,-12,57,-212,411,-1817,509,-12097,18353,6620,2284,751,514,38,44,4);
193  FIR16_2(7,-3,42,-27,508,-665,2347,-6173,18208,12534,270,1899,383,229,56,13,0,
194  25,0,-13,56,-229,383,-1899,270,-12534,18208,6173,2347,665,508,27,42,3);
195 
196     samples+=64;
197   }
198   samples-=64*nb_blocks;
199 
200   /* filter bank: part 3 */
201 
202   b=bo_start;
203   for(block=0;block<nb_blocks;block++) {
204     b++;
205     if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
206     if(b & 1) {
207       b0 = buf[0] + b - (FIR_SIZE-1) + 8*FIR_BUFFER_SIZE;
208     } else {
209       b0 = buf[1] + b - (FIR_SIZE-1) + 8*FIR_BUFFER_SIZE;
210     }
211 
212  FIR16_2(8,-3,40,-18,500,-582,2398,-5732,18042,12963,17,1977,353,247,56,14,0,
213  24,0,-14,56,-247,353,-1977,17,-12963,18042,5732,2398,582,500,18,40,3);
214  FIR16_2(9,-2,38,-9,490,-501,2437,-5297,17855,13383,-249,2052,320,266,55,15,0,
215  23,0,-15,55,-266,320,-2052,-249,-13383,17855,5297,2437,501,490,9,38,2);
216  FIR16_2(10,-2,36,0,479,-423,2465,-4869,17647,13794,-530,2122,282,284,53,17,0,
217  22,0,-17,53,-284,282,-2122,-530,-13794,17647,4869,2465,423,479,0,36,2);
218  FIR16_2(11,-2,34,7,467,-347,2483,-4449,17419,14194,-825,2188,242,302,52,18,0,
219  21,0,-18,52,-302,242,-2188,-825,-14194,17419,4449,2483,347,467,-7,34,2);
220 
221     samples+=64;
222   }
223   samples-=64*nb_blocks;
224 
225   /* filter bank: part 4 */
226 
227   b=bo_start;
228   for(block=0;block<nb_blocks;block++) {
229     b++;
230     if (b >= FIR_BUFFER_SIZE) b=FIR_SIZE;
231     if(b & 1) {
232       b0 = buf[0] + b - (FIR_SIZE-1) + 12*FIR_BUFFER_SIZE;
233     } else {
234       b0 = buf[1] + b - (FIR_SIZE-1) + 12*FIR_BUFFER_SIZE;
235     }
236 
237  FIR16_2(12,-2,33,14,454,-273,2491,-4038,17173,14583,-1133,2249,198,320,50,19,0,
238  20,0,-19,50,-320,198,-2249,-1133,-14583,17173,4038,2491,273,454,-14,33,2);
239  FIR16_2(13,-1,31,20,439,-203,2489,-3637,16907,14959,-1454,2304,151,339,47,21,-1,
240  19,-1,-21,47,-339,151,-2304,-1454,-14959,16907,3637,2489,203,439,-20,31,1);
241  FIR16_2(14,-1,29,26,424,-136,2479,-3245,16623,15322,-1788,2354,100,357,44,22,-1,
242  18,-1,-22,44,-357,100,-2354,-1788,-15322,16623,3245,2479,136,424,-26,29,1);
243  FIR16_2(15,-1,27,31,408,-72,2459,-2863,16322,15671,-2135,2396,46,374,40,24,-1,
244  17,-1,-24,40,-374,46,-2396,-2135,-15671,16322,2863,2459,72,408,-31,27,1);
245  FIR16_1(16,-1,0,36,0,-11,0,-2493,0,16004,0,2431,0,391,0,26,0);
246 
247     samples+=64;
248   }
249 
250   return clip;
251 }
252 
253