xref: /reactos/sdk/lib/3rdparty/libmpg123/tabinit.c (revision 1734f297)
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 "debug.h"
11 
12 /* That altivec alignment part here should not hurt generic code, I hope */
13 #ifdef OPT_ALTIVEC
14 static ALIGNED(16) real cos64[16];
15 static ALIGNED(16) real cos32[8];
16 static ALIGNED(16) real cos16[4];
17 static ALIGNED(16) real cos8[2];
18 static ALIGNED(16) real cos4[1];
19 #elif defined(REAL_IS_FIXED) && defined(PRECALC_TABLES)
20 static real cos64[16] =
21 {
22 	8398725,8480395,8647771,8909416,9279544,9780026,10443886,11321405,
23 	12491246,14081950,16316987,19619946,24900150,34523836,57170182,170959967
24 };
25 static real cos32[8] =
26 {
27 	8429197,8766072,9511743,10851869,13223040,17795219,28897867,85583072
28 };
29 static real cos16[4] =
30 {
31 	8552951,10088893,15099095,42998586
32 };
33 static real cos8[2] =
34 {
35 	9079764,21920489
36 };
37 static real cos4[1] =
38 {
39 	11863283
40 };
41 #else
42 static real cos64[16],cos32[8],cos16[4],cos8[2],cos4[1];
43 #endif
44 
45 real *pnts[] = { cos64,cos32,cos16,cos8,cos4 };
46 
47 
48 static long intwinbase[] = {
49      0,    -1,    -1,    -1,    -1,    -1,    -1,    -2,    -2,    -2,
50     -2,    -3,    -3,    -4,    -4,    -5,    -5,    -6,    -7,    -7,
51     -8,    -9,   -10,   -11,   -13,   -14,   -16,   -17,   -19,   -21,
52    -24,   -26,   -29,   -31,   -35,   -38,   -41,   -45,   -49,   -53,
53    -58,   -63,   -68,   -73,   -79,   -85,   -91,   -97,  -104,  -111,
54   -117,  -125,  -132,  -139,  -147,  -154,  -161,  -169,  -176,  -183,
55   -190,  -196,  -202,  -208,  -213,  -218,  -222,  -225,  -227,  -228,
56   -228,  -227,  -224,  -221,  -215,  -208,  -200,  -189,  -177,  -163,
57   -146,  -127,  -106,   -83,   -57,   -29,     2,    36,    72,   111,
58    153,   197,   244,   294,   347,   401,   459,   519,   581,   645,
59    711,   779,   848,   919,   991,  1064,  1137,  1210,  1283,  1356,
60   1428,  1498,  1567,  1634,  1698,  1759,  1817,  1870,  1919,  1962,
61   2001,  2032,  2057,  2075,  2085,  2087,  2080,  2063,  2037,  2000,
62   1952,  1893,  1822,  1739,  1644,  1535,  1414,  1280,  1131,   970,
63    794,   605,   402,   185,   -45,  -288,  -545,  -814, -1095, -1388,
64  -1692, -2006, -2330, -2663, -3004, -3351, -3705, -4063, -4425, -4788,
65  -5153, -5517, -5879, -6237, -6589, -6935, -7271, -7597, -7910, -8209,
66  -8491, -8755, -8998, -9219, -9416, -9585, -9727, -9838, -9916, -9959,
67  -9966, -9935, -9863, -9750, -9592, -9389, -9139, -8840, -8492, -8092,
68  -7640, -7134, -6574, -5959, -5288, -4561, -3776, -2935, -2037, -1082,
69    -70,   998,  2122,  3300,  4533,  5818,  7154,  8540,  9975, 11455,
70  12980, 14548, 16155, 17799, 19478, 21189, 22929, 24694, 26482, 28289,
71  30112, 31947, 33791, 35640, 37489, 39336, 41176, 43006, 44821, 46617,
72  48390, 50137, 51853, 53534, 55178, 56778, 58333, 59838, 61289, 62684,
73  64019, 65290, 66494, 67629, 68692, 69679, 70590, 71420, 72169, 72835,
74  73415, 73908, 74313, 74630, 74856, 74992, 75038 };
75 
76 void prepare_decode_tables()
77 {
78 #if !defined(REAL_IS_FIXED) || !defined(PRECALC_TABLES)
79   int i,k,kr,divv;
80   real *costab;
81 
82   for(i=0;i<5;i++)
83   {
84     kr=0x10>>i; divv=0x40>>i;
85     costab = pnts[i];
86     for(k=0;k<kr;k++)
87       costab[k] = DOUBLE_TO_REAL(1.0 / (2.0 * cos(M_PI * ((double) k * 2.0 + 1.0) / (double) divv)));
88   }
89 #endif
90 }
91 
92 #ifdef OPT_MMXORSSE
93 #if !defined(OPT_X86_64) && !defined(OPT_NEON) && !defined(OPT_NEON64) && !defined(OPT_AVX)
94 void make_decode_tables_mmx_asm(long scaleval, float* decwin_mmx, float *decwins);
95 void make_decode_tables_mmx(mpg123_handle *fr)
96 {
97 	debug("MMX decode tables");
98 	/* Take care: The scale should be like before, when we didn't have float output all around. */
99 	make_decode_tables_mmx_asm((long)((fr->lastscale < 0 ? fr->p.outscale : fr->lastscale)*SHORT_SCALE), fr->decwin_mmx, fr->decwins);
100 	debug("MMX decode tables done");
101 }
102 #else
103 
104 /* This mimics round() as defined in C99. We stay C89. */
105 static int rounded(double f)
106 {
107 	return (int)(f>0 ? floor(f+0.5) : ceil(f-0.5));
108 }
109 
110 /* x86-64 doesn't use asm version */
111 void make_decode_tables_mmx(mpg123_handle *fr)
112 {
113 	int i,j,val;
114 	int idx = 0;
115 	short *ptr = (short *)fr->decwins;
116 	/* Scale is always based on 1.0 . */
117 	double scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
118 	debug1("MMX decode tables with scaleval %g", scaleval);
119 	for(i=0,j=0;i<256;i++,j++,idx+=32)
120 	{
121 		if(idx < 512+16)
122 		fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
123 
124 		if(i % 32 == 31)
125 		idx -= 1023;
126 		if(i % 64 == 63)
127 		scaleval = - scaleval;
128 	}
129 
130 	for( /* i=256 */ ;i<512;i++,j--,idx+=32)
131 	{
132 		if(idx < 512+16)
133 		fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
134 
135 		if(i % 32 == 31)
136 		idx -= 1023;
137 		if(i % 64 == 63)
138 		scaleval = - scaleval;
139 	}
140 
141 	for(i=0; i<512; i++) {
142 		if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
143 		else val = rounded(fr->decwin_mmx[i]*-0.5);
144 		if(val > 32767) val = 32767;
145 		else if(val < -32768) val = -32768;
146 		ptr[i] = val;
147 	}
148 	for(i=512; i<512+32; i++) {
149 		if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
150 		else val = 0;
151 		if(val > 32767) val = 32767;
152 		else if(val < -32768) val = -32768;
153 		ptr[i] = val;
154 	}
155 	for(i=0; i<512; i++) {
156 		val = rounded(fr->decwin_mmx[511-i]*-0.5);
157 		if(val > 32767) val = 32767;
158 		else if(val < -32768) val = -32768;
159 		ptr[512+32+i] = val;
160 	}
161 	debug("decode tables done");
162 }
163 #endif
164 #endif
165 
166 #ifdef REAL_IS_FIXED
167 /* Need saturating multiplication that keeps table values in 32 bit range,
168    with the option to swap sign at will (so -2**31 is out).
169    This code is far from the decoder core and so assembly optimization might
170    be overkill. */
171 static int32_t sat_mul32(int32_t a, int32_t b)
172 {
173 	int64_t prod = (int64_t)a * (int64_t)b;
174 	/* TODO: record the clipping? An extra flag? */
175 	if(prod >  2147483647L) return  2147483647L;
176 	if(prod < -2147483647L) return -2147483647L;
177 	return (int32_t)prod;
178 }
179 #endif
180 
181 void make_decode_tables(mpg123_handle *fr)
182 {
183 	int i,j;
184 	int idx = 0;
185 	double scaleval;
186 #ifdef REAL_IS_FIXED
187 	real scaleval_long;
188 #endif
189 	/* Scale is always based on 1.0 . */
190 	scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
191 	debug1("decode tables with scaleval %g", scaleval);
192 #ifdef REAL_IS_FIXED
193 	scaleval_long = DOUBLE_TO_REAL_15(scaleval);
194 	debug1("decode table with fixed scaleval %li", (long)scaleval_long);
195 	if(scaleval_long > 28618 || scaleval_long < -28618)
196 	{
197 		/* TODO: Limit the scaleval itself or limit the multiplication afterwards?
198 		   The former basically disables significant amplification for fixed-point
199 		   decoders, but avoids (possibly subtle) distortion. */
200 		/* This would limit the amplification instead:
201 		   scaleval_long = scaleval_long < 0 ? -28618 : 28618; */
202 		if(NOQUIET) warning("Desired amplification may introduce distortion.");
203 	}
204 #endif
205 	for(i=0,j=0;i<256;i++,j++,idx+=32)
206 	{
207 		if(idx < 512+16)
208 #ifdef REAL_IS_FIXED
209 		fr->decwin[idx+16] = fr->decwin[idx] =
210 			REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
211 #else
212 		fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
213 #endif
214 
215 		if(i % 32 == 31)
216 		idx -= 1023;
217 		if(i % 64 == 63)
218 #ifdef REAL_IS_FIXED
219 		scaleval_long = - scaleval_long;
220 #else
221 		scaleval = - scaleval;
222 #endif
223 	}
224 
225 	for( /* i=256 */ ;i<512;i++,j--,idx+=32)
226 	{
227 		if(idx < 512+16)
228 #ifdef REAL_IS_FIXED
229 		fr->decwin[idx+16] = fr->decwin[idx] =
230 			REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
231 #else
232 		fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
233 #endif
234 
235 		if(i % 32 == 31)
236 		idx -= 1023;
237 		if(i % 64 == 63)
238 #ifdef REAL_IS_FIXED
239 		scaleval_long = - scaleval_long;
240 #else
241 		scaleval = - scaleval;
242 #endif
243 	}
244 #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)
245 	if(  fr->cpu_opts.type == x86_64
246 	  || fr->cpu_opts.type == altivec
247 	  || fr->cpu_opts.type == sse
248 	  || fr->cpu_opts.type == sse_vintage
249 	  || fr->cpu_opts.type == arm
250 	  || fr->cpu_opts.type == neon
251 	  || fr->cpu_opts.type == neon64
252 	  || fr->cpu_opts.type == avx )
253 	{ /* for float SSE / AltiVec / ARM decoder */
254 		for(i=512; i<512+32; i++)
255 		{
256 			fr->decwin[i] = (i&1) ? fr->decwin[i] : 0;
257 		}
258 		for(i=0; i<512; i++)
259 		{
260 			fr->decwin[512+32+i] = -fr->decwin[511-i];
261 		}
262 #if defined(OPT_NEON) || defined(OPT_NEON64)
263 		if(fr->cpu_opts.type == neon || fr->cpu_opts.type == neon64)
264 		{
265 			for(i=0; i<512; i+=2)
266 			{
267 				fr->decwin[i] = -fr->decwin[i];
268 			}
269 		}
270 #endif
271 	}
272 #endif
273 	debug("decode tables done");
274 }
275 
276 #ifndef NO_8BIT
277 int make_conv16to8_table(mpg123_handle *fr)
278 {
279   int i;
280 	int mode = fr->af.dec_enc;
281 
282   /*
283    * ????: 8.0 is right but on SB cards '2.0' is a better value ???
284    */
285   const double mul = 8.0;
286 
287   if(!fr->conv16to8_buf){
288     fr->conv16to8_buf = (unsigned char *) malloc(8192);
289     if(!fr->conv16to8_buf) {
290       fr->err = MPG123_ERR_16TO8TABLE;
291       if(NOQUIET) error("Can't allocate 16 to 8 converter table!");
292       return -1;
293     }
294     fr->conv16to8 = fr->conv16to8_buf + 4096;
295   }
296 
297 	switch(mode)
298 	{
299 	case MPG123_ENC_ULAW_8:
300 	{
301 		double m=127.0 / log(256.0);
302 		int c1;
303 
304 		for(i=-4096;i<4096;i++)
305 		{
306 			/* dunno whether this is a valid transformation rule ?!?!? */
307 			if(i < 0)
308 			c1 = 127 - (int) (log( 1.0 - 255.0 * (double) i*mul / 32768.0 ) * m);
309 			else
310 			c1 = 255 - (int) (log( 1.0 + 255.0 * (double) i*mul / 32768.0 ) * m);
311 			if(c1 < 0 || c1 > 255)
312 			{
313 				if(NOQUIET) error2("Converror %d %d",i,c1);
314 				return -1;
315 			}
316 			if(c1 == 0)
317 			c1 = 2;
318 			fr->conv16to8[i] = (unsigned char) c1;
319 		}
320 	}
321 	break;
322 	case MPG123_ENC_SIGNED_8:
323 		for(i=-4096;i<4096;i++)
324 		fr->conv16to8[i] = i>>5;
325 	break;
326 	case MPG123_ENC_UNSIGNED_8:
327 		for(i=-4096;i<4096;i++)
328 		fr->conv16to8[i] = (i>>5)+128;
329 	break;
330 	case MPG123_ENC_ALAW_8:
331 	{
332 		/*
333 			Let's believe Wikipedia (http://en.wikipedia.org/wiki/G.711) that this
334 			is the correct table:
335 
336 			s0000000wxyza... 	n000wxyz  [0-31] -> [0-15]
337 			s0000001wxyza... 	n001wxyz  [32-63] -> [16-31]
338 			s000001wxyzab... 	n010wxyz  [64-127] -> [32-47]
339 			s00001wxyzabc... 	n011wxyz  [128-255] -> [48-63]
340 			s0001wxyzabcd... 	n100wxyz  [256-511] -> [64-79]
341 			s001wxyzabcde... 	n101wxyz  [512-1023] -> [80-95]
342 			s01wxyzabcdef... 	n110wxyz  [1024-2047] -> [96-111]
343 			s1wxyzabcdefg... 	n111wxyz  [2048-4095] -> [112-127]
344 
345 			Let's extend to -4096, too.
346 			Also, bytes are xored with 0x55 for transmission.
347 
348 			Since it sounds OK, I assume it is fine;-)
349 		*/
350 		for(i=0; i<64; ++i)
351 		fr->conv16to8[i] = ((unsigned int)i)>>1;
352 		for(i=64; i<128; ++i)
353 		fr->conv16to8[i] = ((((unsigned int)i)>>2) & 0xf) | (2<<4);
354 		for(i=128; i<256; ++i)
355 		fr->conv16to8[i] = ((((unsigned int)i)>>3) & 0xf) | (3<<4);
356 		for(i=256; i<512; ++i)
357 		fr->conv16to8[i] = ((((unsigned int)i)>>4) & 0xf) | (4<<4);
358 		for(i=512; i<1024; ++i)
359 		fr->conv16to8[i] = ((((unsigned int)i)>>5) & 0xf) | (5<<4);
360 		for(i=1024; i<2048; ++i)
361 		fr->conv16to8[i] = ((((unsigned int)i)>>6) & 0xf) | (6<<4);
362 		for(i=2048; i<4096; ++i)
363 		fr->conv16to8[i] = ((((unsigned int)i)>>7) & 0xf) | (7<<4);
364 
365 		for(i=-4095; i<0; ++i)
366 		fr->conv16to8[i] = fr->conv16to8[-i] | 0x80;
367 
368 		fr->conv16to8[-4096] = fr->conv16to8[-4095];
369 
370 		for(i=-4096;i<4096;i++)
371 		{
372 			/* fr->conv16to8[i] = - i>>5; */
373 			/* fprintf(stderr, "table %i %i\n", i<<AUSHIFT, fr->conv16to8[i]); */
374 			fr->conv16to8[i] ^= 0x55;
375 		}
376 	}
377 	break;
378 	default:
379 		fr->err = MPG123_ERR_16TO8TABLE;
380 		if(NOQUIET) error("Unknown 8 bit encoding choice.");
381 		return -1;
382 	break;
383 	}
384 
385 	return 0;
386 }
387 #endif
388 
389