2 tabinit.c: initialize tables...
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
9 #include "mpg123lib_intern.h"
12 // The precomputed cos tables.
14 const real *pnts[] = { cos64,cos32,cos16,cos8,cos4 };
16 static const long intwinbase[] = {
17 0, -1, -1, -1, -1, -1, -1, -2, -2, -2,
18 -2, -3, -3, -4, -4, -5, -5, -6, -7, -7,
19 -8, -9, -10, -11, -13, -14, -16, -17, -19, -21,
20 -24, -26, -29, -31, -35, -38, -41, -45, -49, -53,
21 -58, -63, -68, -73, -79, -85, -91, -97, -104, -111,
22 -117, -125, -132, -139, -147, -154, -161, -169, -176, -183,
23 -190, -196, -202, -208, -213, -218, -222, -225, -227, -228,
24 -228, -227, -224, -221, -215, -208, -200, -189, -177, -163,
25 -146, -127, -106, -83, -57, -29, 2, 36, 72, 111,
26 153, 197, 244, 294, 347, 401, 459, 519, 581, 645,
27 711, 779, 848, 919, 991, 1064, 1137, 1210, 1283, 1356,
28 1428, 1498, 1567, 1634, 1698, 1759, 1817, 1870, 1919, 1962,
29 2001, 2032, 2057, 2075, 2085, 2087, 2080, 2063, 2037, 2000,
30 1952, 1893, 1822, 1739, 1644, 1535, 1414, 1280, 1131, 970,
31 794, 605, 402, 185, -45, -288, -545, -814, -1095, -1388,
32 -1692, -2006, -2330, -2663, -3004, -3351, -3705, -4063, -4425, -4788,
33 -5153, -5517, -5879, -6237, -6589, -6935, -7271, -7597, -7910, -8209,
34 -8491, -8755, -8998, -9219, -9416, -9585, -9727, -9838, -9916, -9959,
35 -9966, -9935, -9863, -9750, -9592, -9389, -9139, -8840, -8492, -8092,
36 -7640, -7134, -6574, -5959, -5288, -4561, -3776, -2935, -2037, -1082,
37 -70, 998, 2122, 3300, 4533, 5818, 7154, 8540, 9975, 11455,
38 12980, 14548, 16155, 17799, 19478, 21189, 22929, 24694, 26482, 28289,
39 30112, 31947, 33791, 35640, 37489, 39336, 41176, 43006, 44821, 46617,
40 48390, 50137, 51853, 53534, 55178, 56778, 58333, 59838, 61289, 62684,
41 64019, 65290, 66494, 67629, 68692, 69679, 70590, 71420, 72169, 72835,
42 73415, 73908, 74313, 74630, 74856, 74992, 75038 };
45 #if !defined(OPT_X86_64) && !defined(OPT_NEON) && !defined(OPT_NEON64) && !defined(OPT_AVX)
46 void make_decode_tables_mmx_asm(long scaleval, float* decwin_mmx, float *decwins);
47 void make_decode_tables_mmx(mpg123_handle *fr)
49 debug("MMX decode tables");
50 /* Take care: The scale should be like before, when we didn't have float output all around. */
51 make_decode_tables_mmx_asm((long)((fr->lastscale < 0 ? fr->p.outscale : fr->lastscale)*SHORT_SCALE), fr->decwin_mmx, fr->decwins);
52 debug("MMX decode tables done");
56 /* This mimics round() as defined in C99. We stay C89. */
57 static int rounded(double f)
59 return (int)(f>0 ? floor(f+0.5) : ceil(f-0.5));
62 /* x86-64 doesn't use asm version */
63 void make_decode_tables_mmx(mpg123_handle *fr)
67 short *ptr = (short *)fr->decwins;
68 /* Scale is always based on 1.0 . */
69 double scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
70 debug1("MMX decode tables with scaleval %g", scaleval);
71 for(i=0,j=0;i<256;i++,j++,idx+=32)
74 fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
79 scaleval = - scaleval;
82 for( /* i=256 */ ;i<512;i++,j--,idx+=32)
85 fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
90 scaleval = - scaleval;
93 for(i=0; i<512; i++) {
94 if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
95 else val = rounded(fr->decwin_mmx[i]*-0.5);
96 if(val > 32767) val = 32767;
97 else if(val < -32768) val = -32768;
100 for(i=512; i<512+32; i++) {
101 if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
103 if(val > 32767) val = 32767;
104 else if(val < -32768) val = -32768;
107 for(i=0; i<512; i++) {
108 val = rounded(fr->decwin_mmx[511-i]*-0.5);
109 if(val > 32767) val = 32767;
110 else if(val < -32768) val = -32768;
113 debug("decode tables done");
119 /* Need saturating multiplication that keeps table values in 32 bit range,
120 with the option to swap sign at will (so -2**31 is out).
121 This code is far from the decoder core and so assembly optimization might
123 static int32_t sat_mul32(int32_t a, int32_t b)
125 int64_t prod = (int64_t)a * (int64_t)b;
126 /* TODO: record the clipping? An extra flag? */
127 if(prod > 2147483647L) return 2147483647L;
128 if(prod < -2147483647L) return -2147483647L;
129 return (int32_t)prod;
133 void make_decode_tables(mpg123_handle *fr)
141 /* Scale is always based on 1.0 . */
142 scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
143 debug1("decode tables with scaleval %g", scaleval);
145 scaleval_long = DOUBLE_TO_REAL_15(scaleval);
146 debug1("decode table with fixed scaleval %li", (long)scaleval_long);
147 if(scaleval_long > 28618 || scaleval_long < -28618)
149 /* TODO: Limit the scaleval itself or limit the multiplication afterwards?
150 The former basically disables significant amplification for fixed-point
151 decoders, but avoids (possibly subtle) distortion. */
152 /* This would limit the amplification instead:
153 scaleval_long = scaleval_long < 0 ? -28618 : 28618; */
154 if(NOQUIET) warning("Desired amplification may introduce distortion.");
157 for(i=0,j=0;i<256;i++,j++,idx+=32)
161 fr->decwin[idx+16] = fr->decwin[idx] =
162 REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
164 fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
171 scaleval_long = - scaleval_long;
173 scaleval = - scaleval;
177 for( /* i=256 */ ;i<512;i++,j--,idx+=32)
181 fr->decwin[idx+16] = fr->decwin[idx] =
182 REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
184 fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
191 scaleval_long = - scaleval_long;
193 scaleval = - scaleval;
196 #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)
197 if( fr->cpu_opts.type == x86_64
198 || fr->cpu_opts.type == altivec
199 || fr->cpu_opts.type == sse
200 || fr->cpu_opts.type == sse_vintage
201 || fr->cpu_opts.type == arm
202 || fr->cpu_opts.type == neon
203 || fr->cpu_opts.type == neon64
204 || fr->cpu_opts.type == avx )
205 { /* for float SSE / AltiVec / ARM decoder */
206 for(i=512; i<512+32; i++)
208 fr->decwin[i] = (i&1) ? fr->decwin[i] : 0;
212 fr->decwin[512+32+i] = -fr->decwin[511-i];
214 #if defined(OPT_NEON) || defined(OPT_NEON64)
215 if(fr->cpu_opts.type == neon || fr->cpu_opts.type == neon64)
217 for(i=0; i<512; i+=2)
219 fr->decwin[i] = -fr->decwin[i];
225 debug("decode tables done");
229 int make_conv16to8_table(mpg123_handle *fr)
232 int mode = fr->af.dec_enc;
235 * ????: 8.0 is right but on SB cards '2.0' is a better value ???
237 const double mul = 8.0;
239 if(!fr->conv16to8_buf){
240 fr->conv16to8_buf = (unsigned char *) malloc(8192);
241 if(!fr->conv16to8_buf) {
242 fr->err = MPG123_ERR_16TO8TABLE;
243 if(NOQUIET) error("Can't allocate 16 to 8 converter table!");
246 fr->conv16to8 = fr->conv16to8_buf + 4096;
251 case MPG123_ENC_ULAW_8:
253 double m=127.0 / log(256.0);
256 for(i=-4096;i<4096;i++)
258 /* dunno whether this is a valid transformation rule ?!?!? */
260 c1 = 127 - (int) (log( 1.0 - 255.0 * (double) i*mul / 32768.0 ) * m);
262 c1 = 255 - (int) (log( 1.0 + 255.0 * (double) i*mul / 32768.0 ) * m);
263 if(c1 < 0 || c1 > 255)
265 if(NOQUIET) error2("Converror %d %d",i,c1);
270 fr->conv16to8[i] = (unsigned char) c1;
274 case MPG123_ENC_SIGNED_8:
275 for(i=-4096;i<4096;i++)
276 fr->conv16to8[i] = i>>5;
278 case MPG123_ENC_UNSIGNED_8:
279 for(i=-4096;i<4096;i++)
280 fr->conv16to8[i] = (i>>5)+128;
282 case MPG123_ENC_ALAW_8:
285 Let's believe Wikipedia (http://en.wikipedia.org/wiki/G.711) that this
286 is the correct table:
288 s0000000wxyza... n000wxyz [0-31] -> [0-15]
289 s0000001wxyza... n001wxyz [32-63] -> [16-31]
290 s000001wxyzab... n010wxyz [64-127] -> [32-47]
291 s00001wxyzabc... n011wxyz [128-255] -> [48-63]
292 s0001wxyzabcd... n100wxyz [256-511] -> [64-79]
293 s001wxyzabcde... n101wxyz [512-1023] -> [80-95]
294 s01wxyzabcdef... n110wxyz [1024-2047] -> [96-111]
295 s1wxyzabcdefg... n111wxyz [2048-4095] -> [112-127]
297 Let's extend to -4096, too.
298 Also, bytes are xored with 0x55 for transmission.
300 Since it sounds OK, I assume it is fine;-)
303 fr->conv16to8[i] = ((unsigned int)i)>>1;
304 for(i=64; i<128; ++i)
305 fr->conv16to8[i] = ((((unsigned int)i)>>2) & 0xf) | (2<<4);
306 for(i=128; i<256; ++i)
307 fr->conv16to8[i] = ((((unsigned int)i)>>3) & 0xf) | (3<<4);
308 for(i=256; i<512; ++i)
309 fr->conv16to8[i] = ((((unsigned int)i)>>4) & 0xf) | (4<<4);
310 for(i=512; i<1024; ++i)
311 fr->conv16to8[i] = ((((unsigned int)i)>>5) & 0xf) | (5<<4);
312 for(i=1024; i<2048; ++i)
313 fr->conv16to8[i] = ((((unsigned int)i)>>6) & 0xf) | (6<<4);
314 for(i=2048; i<4096; ++i)
315 fr->conv16to8[i] = ((((unsigned int)i)>>7) & 0xf) | (7<<4);
317 for(i=-4095; i<0; ++i)
318 fr->conv16to8[i] = fr->conv16to8[-i] | 0x80;
320 fr->conv16to8[-4096] = fr->conv16to8[-4095];
322 for(i=-4096;i<4096;i++)
324 /* fr->conv16to8[i] = - i>>5; */
325 /* fprintf(stderr, "table %i %i\n", i<<AUSHIFT, fr->conv16to8[i]); */
326 fr->conv16to8[i] ^= 0x55;
331 fr->err = MPG123_ERR_16TO8TABLE;
332 if(NOQUIET) error("Unknown 8 bit encoding choice.");