Debian Codegraph @

src/libmpg123/tabinit.c

raw
1 /*
2 tabinit.c: initialize tables...
3
4 copyright ?-2008 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 Michael Hipp
7 */
8
9 #include "mpg123lib_intern.h"
10 #include "../common/debug.h"
11
12 // The (normally precomputed) cos tables.
13 #include "costabs.h"
14 #ifdef RUNTIME_TABLES
15 #include "init_costabs.h"
16 #else
17 const
18 #endif
19 real *INT123_pnts[] = { cos64,cos32,cos16,cos8,cos4 };
20
21 static const long intwinbase[] = {
22 0, -1, -1, -1, -1, -1, -1, -2, -2, -2,
23 -2, -3, -3, -4, -4, -5, -5, -6, -7, -7,
24 -8, -9, -10, -11, -13, -14, -16, -17, -19, -21,
25 -24, -26, -29, -31, -35, -38, -41, -45, -49, -53,
26 -58, -63, -68, -73, -79, -85, -91, -97, -104, -111,
27 -117, -125, -132, -139, -147, -154, -161, -169, -176, -183,
28 -190, -196, -202, -208, -213, -218, -222, -225, -227, -228,
29 -228, -227, -224, -221, -215, -208, -200, -189, -177, -163,
30 -146, -127, -106, -83, -57, -29, 2, 36, 72, 111,
31 153, 197, 244, 294, 347, 401, 459, 519, 581, 645,
32 711, 779, 848, 919, 991, 1064, 1137, 1210, 1283, 1356,
33 1428, 1498, 1567, 1634, 1698, 1759, 1817, 1870, 1919, 1962,
34 2001, 2032, 2057, 2075, 2085, 2087, 2080, 2063, 2037, 2000,
35 1952, 1893, 1822, 1739, 1644, 1535, 1414, 1280, 1131, 970,
36 794, 605, 402, 185, -45, -288, -545, -814, -1095, -1388,
37 -1692, -2006, -2330, -2663, -3004, -3351, -3705, -4063, -4425, -4788,
38 -5153, -5517, -5879, -6237, -6589, -6935, -7271, -7597, -7910, -8209,
39 -8491, -8755, -8998, -9219, -9416, -9585, -9727, -9838, -9916, -9959,
40 -9966, -9935, -9863, -9750, -9592, -9389, -9139, -8840, -8492, -8092,
41 -7640, -7134, -6574, -5959, -5288, -4561, -3776, -2935, -2037, -1082,
42 -70, 998, 2122, 3300, 4533, 5818, 7154, 8540, 9975, 11455,
43 12980, 14548, 16155, 17799, 19478, 21189, 22929, 24694, 26482, 28289,
44 30112, 31947, 33791, 35640, 37489, 39336, 41176, 43006, 44821, 46617,
45 48390, 50137, 51853, 53534, 55178, 56778, 58333, 59838, 61289, 62684,
46 64019, 65290, 66494, 67629, 68692, 69679, 70590, 71420, 72169, 72835,
47 73415, 73908, 74313, 74630, 74856, 74992, 75038 };
48
49 #ifdef OPT_MMXORSSE
50 #if !defined(OPT_X86_64) && !defined(OPT_NEON) && !defined(OPT_NEON64) && !defined(OPT_AVX)
51 void INT123_make_decode_tables_mmx_asm(long scaleval, float* decwin_mmx, float *decwins);
52 void INT123_make_decode_tables_mmx(mpg123_handle *fr)
53 {
54 debug("MMX decode tables");
55 /* Take care: The scale should be like before, when we didn't have float output all around. */
56 INT123_make_decode_tables_mmx_asm((long)((fr->lastscale < 0 ? fr->p.outscale : fr->lastscale)*SHORT_SCALE), fr->decwin_mmx, fr->decwins);
57 debug("MMX decode tables done");
58 }
59 #else
60
61 /* This mimics round() as defined in C99. We stay C89. */
62 static int rounded(double f)
63 {
64 return (int)(f>0 ? floor(f+0.5) : ceil(f-0.5));
65 }
66
67 /* x86-64 doesn't use asm version */
68 void INT123_make_decode_tables_mmx(mpg123_handle *fr)
69 {
70 int i,j,val;
71 int idx = 0;
72 short *ptr = (short *)fr->decwins;
73 /* Scale is always based on 1.0 . */
74 double scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
75 debug1("MMX decode tables with scaleval %g", scaleval);
76 for(i=0,j=0;i<256;i++,j++,idx+=32)
77 {
78 if(idx < 512+16)
79 fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
80
81 if(i % 32 == 31)
82 idx -= 1023;
83 if(i % 64 == 63)
84 scaleval = - scaleval;
85 }
86
87 for( /* i=256 */ ;i<512;i++,j--,idx+=32)
88 {
89 if(idx < 512+16)
90 fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
91
92 if(i % 32 == 31)
93 idx -= 1023;
94 if(i % 64 == 63)
95 scaleval = - scaleval;
96 }
97
98 for(i=0; i<512; i++) {
99 if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
100 else val = rounded(fr->decwin_mmx[i]*-0.5);
101 if(val > 32767) val = 32767;
102 else if(val < -32768) val = -32768;
103 ptr[i] = val;
104 }
105 for(i=512; i<512+32; i++) {
106 if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
107 else val = 0;
108 if(val > 32767) val = 32767;
109 else if(val < -32768) val = -32768;
110 ptr[i] = val;
111 }
112 for(i=0; i<512; i++) {
113 val = rounded(fr->decwin_mmx[511-i]*-0.5);
114 if(val > 32767) val = 32767;
115 else if(val < -32768) val = -32768;
116 ptr[512+32+i] = val;
117 }
118 debug("decode tables done");
119 }
120 #endif
121 #endif
122
123 #ifdef REAL_IS_FIXED
124 /* Need saturating multiplication that keeps table values in 32 bit range,
125 with the option to swap sign at will (so -2**31 is out).
126 This code is far from the decoder core and so assembly optimization might
127 be overkill. */
128 static int32_t sat_mul32(int32_t a, int32_t b)
129 {
130 int64_t prod = (int64_t)a * (int64_t)b;
131 /* TODO: record the clipping? An extra flag? */
132 if(prod > 2147483647L) return 2147483647L;
133 if(prod < -2147483647L) return -2147483647L;
134 return (int32_t)prod;
135 }
136 #endif
137
138 void INT123_make_decode_tables(mpg123_handle *fr)
139 {
140 int i,j;
141 int idx = 0;
142 double scaleval;
143 #ifdef REAL_IS_FIXED
144 real scaleval_long;
145 #endif
146 /* Scale is always based on 1.0 . */
147 scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
148 debug1("decode tables with scaleval %g", scaleval);
149 #ifdef REAL_IS_FIXED
150 scaleval_long = DOUBLE_TO_REAL_15(scaleval);
151 debug1("decode table with fixed scaleval %li", (long)scaleval_long);
152 if(scaleval_long > 28618 || scaleval_long < -28618)
153 {
154 /* TODO: Limit the scaleval itself or limit the multiplication afterwards?
155 The former basically disables significant amplification for fixed-point
156 decoders, but avoids (possibly subtle) distortion. */
157 /* This would limit the amplification instead:
158 scaleval_long = scaleval_long < 0 ? -28618 : 28618; */
159 if(NOQUIET) warning("Desired amplification may introduce distortion.");
160 }
161 #endif
162 for(i=0,j=0;i<256;i++,j++,idx+=32)
163 {
164 if(idx < 512+16)
165 #ifdef REAL_IS_FIXED
166 fr->decwin[idx+16] = fr->decwin[idx] =
167 REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
168 #else
169 fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
170 #endif
171
172 if(i % 32 == 31)
173 idx -= 1023;
174 if(i % 64 == 63)
175 #ifdef REAL_IS_FIXED
176 scaleval_long = - scaleval_long;
177 #else
178 scaleval = - scaleval;
179 #endif
180 }
181
182 for( /* i=256 */ ;i<512;i++,j--,idx+=32)
183 {
184 if(idx < 512+16)
185 #ifdef REAL_IS_FIXED
186 fr->decwin[idx+16] = fr->decwin[idx] =
187 REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
188 #else
189 fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
190 #endif
191
192 if(i % 32 == 31)
193 idx -= 1023;
194 if(i % 64 == 63)
195 #ifdef REAL_IS_FIXED
196 scaleval_long = - scaleval_long;
197 #else
198 scaleval = - scaleval;
199 #endif
200 }
201 #if defined(OPT_X86_64) || defined(OPT_ALTIVEC) || defined(OPT_SSE) || defined(OPT_SSE_VINTAGE) || defined(OPT_ARM) || defined(OPT_NEON) || defined(OPT_NEON64) || defined(OPT_AVX)
202 if( fr->cpu_opts.type == x86_64
203 || fr->cpu_opts.type == altivec
204 || fr->cpu_opts.type == sse
205 || fr->cpu_opts.type == sse_vintage
206 || fr->cpu_opts.type == arm
207 || fr->cpu_opts.type == neon
208 || fr->cpu_opts.type == neon64
209 || fr->cpu_opts.type == avx )
210 { /* for float SSE / AltiVec / ARM decoder */
211 for(i=512; i<512+32; i++)
212 {
213 fr->decwin[i] = (i&1) ? fr->decwin[i] : 0;
214 }
215 for(i=0; i<512; i++)
216 {
217 fr->decwin[512+32+i] = -fr->decwin[511-i];
218 }
219 #if defined(OPT_NEON) || defined(OPT_NEON64)
220 if(fr->cpu_opts.type == neon || fr->cpu_opts.type == neon64)
221 {
222 for(i=0; i<512; i+=2)
223 {
224 fr->decwin[i] = -fr->decwin[i];
225 }
226 }
227 #endif
228 }
229 #endif
230 debug("decode tables done");
231 }
232
233 #ifndef NO_8BIT
234 int INT123_make_conv16to8_table(mpg123_handle *fr)
235 {
236 int i;
237 int mode = fr->af.dec_enc;
238
239 /*
240 * ????: 8.0 is right but on SB cards '2.0' is a better value ???
241 */
242 const double mul = 8.0;
243
244 if(!fr->conv16to8_buf){
245 fr->conv16to8_buf = (unsigned char *) malloc(8192);
246 if(!fr->conv16to8_buf) {
247 fr->err = MPG123_ERR_16TO8TABLE;
248 if(NOQUIET) error("Can't allocate 16 to 8 converter table!");
249 return -1;
250 }
251 fr->conv16to8 = fr->conv16to8_buf + 4096;
252 }
253
254 switch(mode)
255 {
256 case MPG123_ENC_ULAW_8:
257 {
258 double m=127.0 / log(256.0);
259 int c1;
260
261 for(i=-4096;i<4096;i++)
262 {
263 /* dunno whether this is a valid transformation rule ?!?!? */
264 if(i < 0)
265 c1 = 127 - (int) (log( 1.0 - 255.0 * (double) i*mul / 32768.0 ) * m);
266 else
267 c1 = 255 - (int) (log( 1.0 + 255.0 * (double) i*mul / 32768.0 ) * m);
268 if(c1 < 0 || c1 > 255)
269 {
270 if(NOQUIET) error2("Converror %d %d",i,c1);
271 return -1;
272 }
273 if(c1 == 0)
274 c1 = 2;
275 fr->conv16to8[i] = (unsigned char) c1;
276 }
277 }
278 break;
279 case MPG123_ENC_SIGNED_8:
280 for(i=-4096;i<4096;i++)
281 fr->conv16to8[i] = i>>5;
282 break;
283 case MPG123_ENC_UNSIGNED_8:
284 for(i=-4096;i<4096;i++)
285 fr->conv16to8[i] = (i>>5)+128;
286 break;
287 case MPG123_ENC_ALAW_8:
288 {
289 /*
290 Let's believe Wikipedia (http://en.wikipedia.org/wiki/G.711) that this
291 is the correct table:
292
293 s0000000wxyza... n000wxyz [0-31] -> [0-15]
294 s0000001wxyza... n001wxyz [32-63] -> [16-31]
295 s000001wxyzab... n010wxyz [64-127] -> [32-47]
296 s00001wxyzabc... n011wxyz [128-255] -> [48-63]
297 s0001wxyzabcd... n100wxyz [256-511] -> [64-79]
298 s001wxyzabcde... n101wxyz [512-1023] -> [80-95]
299 s01wxyzabcdef... n110wxyz [1024-2047] -> [96-111]
300 s1wxyzabcdefg... n111wxyz [2048-4095] -> [112-127]
301
302 Let's extend to -4096, too.
303 Also, bytes are xored with 0x55 for transmission.
304
305 Since it sounds OK, I assume it is fine;-)
306 */
307 for(i=0; i<64; ++i)
308 fr->conv16to8[i] = ((unsigned int)i)>>1;
309 for(i=64; i<128; ++i)
310 fr->conv16to8[i] = ((((unsigned int)i)>>2) & 0xf) | (2<<4);
311 for(i=128; i<256; ++i)
312 fr->conv16to8[i] = ((((unsigned int)i)>>3) & 0xf) | (3<<4);
313 for(i=256; i<512; ++i)
314 fr->conv16to8[i] = ((((unsigned int)i)>>4) & 0xf) | (4<<4);
315 for(i=512; i<1024; ++i)
316 fr->conv16to8[i] = ((((unsigned int)i)>>5) & 0xf) | (5<<4);
317 for(i=1024; i<2048; ++i)
318 fr->conv16to8[i] = ((((unsigned int)i)>>6) & 0xf) | (6<<4);
319 for(i=2048; i<4096; ++i)
320 fr->conv16to8[i] = ((((unsigned int)i)>>7) & 0xf) | (7<<4);
321
322 for(i=-4095; i<0; ++i)
323 fr->conv16to8[i] = fr->conv16to8[-i] | 0x80;
324
325 fr->conv16to8[-4096] = fr->conv16to8[-4095];
326
327 for(i=-4096;i<4096;i++)
328 {
329 /* fr->conv16to8[i] = - i>>5; */
330 /* fprintf(stderr, "table %i %i\n", i<<AUSHIFT, fr->conv16to8[i]); */
331 fr->conv16to8[i] ^= 0x55;
332 }
333 }
334 break;
335 default:
336 fr->err = MPG123_ERR_16TO8TABLE;
337 if(NOQUIET) error("Unknown 8 bit encoding choice.");
338 return -1;
339 break;
340 }
341
342 return 0;
343 }
344 #endif
345