]> git.saurik.com Git - wxWidgets.git/blob - contrib/src/mmedia/g72x.cpp
Misc release-related mods
[wxWidgets.git] / contrib / src / mmedia / g72x.cpp
1 /*
2 * This source code is a product of Sun Microsystems, Inc. and is provided
3 * for unrestricted use. Users may copy or modify this source code without
4 * charge.
5 *
6 * SUN SOURCE CODE IS PROVIDED AS IS WITH NO WARRANTIES OF ANY KIND INCLUDING
7 * THE WARRANTIES OF DESIGN, MERCHANTIBILITY AND FITNESS FOR A PARTICULAR
8 * PURPOSE, OR ARISING FROM A COURSE OF DEALING, USAGE OR TRADE PRACTICE.
9 *
10 * Sun source code is provided with no support and without any obligation on
11 * the part of Sun Microsystems, Inc. to assist in its use, correction,
12 * modification or enhancement.
13 *
14 * SUN MICROSYSTEMS, INC. SHALL HAVE NO LIABILITY WITH RESPECT TO THE
15 * INFRINGEMENT OF COPYRIGHTS, TRADE SECRETS OR ANY PATENTS BY THIS SOFTWARE
16 * OR ANY PART THEREOF.
17 *
18 * In no event will Sun Microsystems, Inc. be liable for any lost revenue
19 * or profits or other special, indirect and consequential damages, even if
20 * Sun has been advised of the possibility of such damages.
21 *
22 * Sun Microsystems, Inc.
23 * 2550 Garcia Avenue
24 * Mountain View, California 94043
25 */
26
27 /*
28 * g72x.c
29 *
30 * Common routines for G.721 and G.723 conversions.
31 */
32
33 #include <wx/wxprec.h>
34 #include <stdlib.h>
35 #include "wx/mmedia/internal/g72x.h"
36
37 static short power2[15] = {1, 2, 4, 8, 0x10, 0x20, 0x40, 0x80,
38 0x100, 0x200, 0x400, 0x800, 0x1000, 0x2000, 0x4000};
39
40 /*
41 * quan()
42 *
43 * quantizes the input val against the table of size short integers.
44 * It returns i if table[i - 1] <= val < table[i].
45 *
46 * Using linear search for simple coding.
47 */
48 static int
49 quan(
50 int val,
51 short *table,
52 int size)
53 {
54 int i;
55
56 for (i = 0; i < size; i++)
57 if (val < *table++)
58 break;
59 return (i);
60 }
61
62 static char quan2_tab[65536];
63 static short base2_tab[65536];
64 static int init_tabs_done = 0;
65
66 inline char quan2 (unsigned short val)
67 {
68 return quan2_tab[val];
69 }
70
71 inline short base2 (unsigned short val)
72 {
73 return base2_tab[val];
74 }
75
76 static void init_quan2_tab (void)
77 {
78 long i;
79
80 for (i = 0; i < 65536; i++) {
81 quan2_tab[i] = quan (i, power2, 15);
82 };
83 }
84
85 static void init_base2_tab (void)
86 {
87 long i;
88 short exp;
89
90 for (i = 0; i < 65536; i++) {
91 exp = quan2 (short (i));
92 base2_tab[i] = short ((exp << 6) + ((i << 6) >> exp));
93 };
94 }
95
96 static void init_tabs (void)
97 {
98 if (init_tabs_done) return;
99
100 init_quan2_tab();
101 init_base2_tab();
102
103 init_tabs_done = 1;
104 }
105
106 /*
107 * fmult()
108 *
109 * returns the integer product of the 14-bit integer "an" and
110 * "floating point" representation (4-bit exponent, 6-bit mantessa) "srn".
111 */
112 static int
113 fmult(
114 int an,
115 int srn)
116 {
117 short anmag, anexp, anmant;
118 short wanexp, wanmant;
119 short retval;
120
121 anmag = (an > 0) ? an : ((-an) & 0x1FFF);
122 anexp = quan2(anmag) - 6;
123 anmant = (anmag == 0) ? 32 :
124 (anexp >= 0) ? anmag >> anexp : anmag << -anexp;
125 wanexp = anexp + ((srn >> 6) & 0xF) - 13;
126
127 wanmant = (anmant * (srn & 077) + 0x30) >> 4;
128 retval = (wanexp >= 0) ? ((wanmant << wanexp) & 0x7FFF) :
129 (wanmant >> -wanexp);
130
131 return (((an ^ srn) < 0) ? -retval : retval);
132 }
133
134 /*
135 * g72x_init_state()
136 *
137 * This routine initializes and/or resets the g72x_state structure
138 * pointed to by 'state_ptr'.
139 * All the initial state values are specified in the CCITT G.721 document.
140 */
141 void
142 g72x_init_state(
143 struct g72x_state *state_ptr)
144 {
145 int cnta;
146
147 init_tabs ();
148
149 state_ptr->yl = 34816;
150 state_ptr->yu = 544;
151 state_ptr->dms = 0;
152 state_ptr->dml = 0;
153 state_ptr->ap = 0;
154 for (cnta = 0; cnta < 2; cnta++) {
155 state_ptr->a[cnta] = 0;
156 state_ptr->pk[cnta] = 0;
157 state_ptr->sr[cnta] = 32;
158 }
159 for (cnta = 0; cnta < 6; cnta++) {
160 state_ptr->b[cnta] = 0;
161 state_ptr->dq[cnta] = 32;
162 }
163 state_ptr->td = 0;
164 }
165
166 /*
167 * predictor_zero()
168 *
169 * computes the estimated signal from 6-zero predictor.
170 *
171 */
172 int
173 predictor_zero(
174 struct g72x_state *state_ptr)
175 {
176 int i;
177 int sezi;
178
179 sezi = fmult(state_ptr->b[0] >> 2, state_ptr->dq[0]);
180 for (i = 1; i < 6; i++) /* ACCUM */
181 sezi += fmult(state_ptr->b[i] >> 2, state_ptr->dq[i]);
182 return (sezi);
183 }
184 /*
185 * predictor_pole()
186 *
187 * computes the estimated signal from 2-pole predictor.
188 *
189 */
190 int
191 predictor_pole(
192 struct g72x_state *state_ptr)
193 {
194 return (fmult(state_ptr->a[1] >> 2, state_ptr->sr[1]) +
195 fmult(state_ptr->a[0] >> 2, state_ptr->sr[0]));
196 }
197 /*
198 * step_size()
199 *
200 * computes the quantization step size of the adaptive quantizer.
201 *
202 */
203 int
204 step_size(
205 struct g72x_state *state_ptr)
206 {
207 int y;
208 int dif;
209 int al;
210
211 if (state_ptr->ap >= 256)
212 return (state_ptr->yu);
213 else {
214 y = state_ptr->yl >> 6;
215 dif = state_ptr->yu - y;
216 al = state_ptr->ap >> 2;
217 if (dif > 0)
218 y += (dif * al) >> 6;
219 else if (dif < 0)
220 y += (dif * al + 0x3F) >> 6;
221 return (y);
222 }
223 }
224
225 /*
226 * quantize()
227 *
228 * Given a raw sample, 'd', of the difference signal and a
229 * quantization step size scale factor, 'y', this routine returns the
230 * ADPCM codeword to which that sample gets quantized. The step
231 * size scale factor division operation is done in the log base 2 domain
232 * as a subtraction.
233 */
234 int
235 quantize(
236 int d, /* Raw difference signal sample */
237 int y, /* Step size multiplier */
238 short *table, /* quantization table */
239 int size) /* table size of short integers */
240 {
241 short dqm; /* Magnitude of 'd' */
242 short exp; /* Integer part of base 2 log of 'd' */
243 short mant; /* Fractional part of base 2 log */
244 short dl; /* Log of magnitude of 'd' */
245 short dln; /* Step size scale factor normalized log */
246 int i;
247
248 /*
249 * LOG
250 *
251 * Compute base 2 log of 'd', and store in 'dl'.
252 */
253 dqm = abs(d);
254 exp = quan2(dqm >> 1);
255 mant = ((dqm << 7) >> exp) & 0x7F; /* Fractional portion. */
256 dl = (exp << 7) + mant;
257
258 /*
259 * SUBTB
260 *
261 * "Divide" by step size multiplier.
262 */
263 dln = dl - (y >> 2);
264
265 /*
266 * QUAN
267 *
268 * Obtain codword i for 'd'.
269 */
270 i = quan(dln, table, size);
271 if (d < 0) /* take 1's complement of i */
272 return ((size << 1) + 1 - i);
273 else if (i == 0) /* take 1's complement of 0 */
274 return ((size << 1) + 1); /* new in 1988 */
275 else
276 return (i);
277 }
278 /*
279 * reconstruct()
280 *
281 * Returns reconstructed difference signal 'dq' obtained from
282 * codeword 'i' and quantization step size scale factor 'y'.
283 * Multiplication is performed in log base 2 domain as addition.
284 */
285 int
286 reconstruct(
287 int sign, /* 0 for non-negative value */
288 int dqln, /* G.72x codeword */
289 int y) /* Step size multiplier */
290 {
291 short dql; /* Log of 'dq' magnitude */
292 short dex; /* Integer part of log */
293 short dqt;
294 short dq; /* Reconstructed difference signal sample */
295
296 dql = dqln + (y >> 2); /* ADDA */
297
298 if (dql < 0) {
299 return ((sign) ? -0x8000 : 0);
300 } else { /* ANTILOG */
301 dex = (dql >> 7) & 15;
302 dqt = 128 + (dql & 127);
303 dq = (dqt << 7) >> (14 - dex);
304 return ((sign) ? (dq - 0x8000) : dq);
305 }
306 }
307
308
309 /*
310 * update()
311 *
312 * updates the state variables for each output code
313 */
314 void
315 update(
316 int code_size, /* distinguish 723_40 with others */
317 int y, /* quantizer step size */
318 int wi, /* scale factor multiplier */
319 int fi, /* for long/short term energies */
320 int dq, /* quantized prediction difference */
321 int sr, /* reconstructed signal */
322 int dqsez, /* difference from 2-pole predictor */
323 struct g72x_state *state_ptr) /* coder state pointer */
324 {
325 int cnt;
326 short mag; /* Adaptive predictor, FLOAT A */
327 short a2p; /* LIMC */
328 short a1ul; /* UPA1 */
329 short pks1; /* UPA2 */
330 short fa1;
331 char tr; /* tone/transition detector */
332 short ylint, thr2, dqthr;
333 short ylfrac, thr1;
334 short pk0;
335
336 pk0 = (dqsez < 0) ? 1 : 0; /* needed in updating predictor poles */
337
338 mag = dq & 0x7FFF; /* prediction difference magnitude */
339 /* TRANS */
340 ylint = short (state_ptr->yl >> 15); /* exponent part of yl */
341 ylfrac = (state_ptr->yl >> 10) & 0x1F; /* fractional part of yl */
342 thr1 = (32 + ylfrac) << ylint; /* threshold */
343 thr2 = (ylint > 9) ? 31 << 10 : thr1; /* limit thr2 to 31 << 10 */
344 dqthr = (thr2 + (thr2 >> 1)) >> 1; /* dqthr = 0.75 * thr2 */
345 if (state_ptr->td == 0) /* signal supposed voice */
346 tr = 0;
347 else if (mag <= dqthr) /* supposed data, but small mag */
348 tr = 0; /* treated as voice */
349 else /* signal is data (modem) */
350 tr = 1;
351
352 /*
353 * Quantizer scale factor adaptation.
354 */
355
356 /* FUNCTW & FILTD & DELAY */
357 /* update non-steady state step size multiplier */
358 state_ptr->yu = y + ((wi - y) >> 5);
359
360 /* LIMB */
361 if (state_ptr->yu < 544) /* 544 <= yu <= 5120 */
362 state_ptr->yu = 544;
363 else if (state_ptr->yu > 5120)
364 state_ptr->yu = 5120;
365
366 /* FILTE & DELAY */
367 /* update steady state step size multiplier */
368 state_ptr->yl += state_ptr->yu + ((-state_ptr->yl) >> 6);
369
370 /*
371 * Adaptive predictor coefficients.
372 */
373 if (tr == 1) { /* reset a's and b's for modem signal */
374 state_ptr->a[0] = 0;
375 state_ptr->a[1] = 0;
376 state_ptr->b[0] = 0;
377 state_ptr->b[1] = 0;
378 state_ptr->b[2] = 0;
379 state_ptr->b[3] = 0;
380 state_ptr->b[4] = 0;
381 state_ptr->b[5] = 0;
382
383 a2p = 0; /* eliminate Compiler Warnings */
384 } else { /* update a's and b's */
385 pks1 = pk0 ^ state_ptr->pk[0]; /* UPA2 */
386
387 /* update predictor pole a[1] */
388 a2p = state_ptr->a[1] - (state_ptr->a[1] >> 7);
389 if (dqsez != 0) {
390 fa1 = (pks1) ? state_ptr->a[0] : -state_ptr->a[0];
391 if (fa1 < -8191) /* a2p = function of fa1 */
392 a2p -= 0x100;
393 else if (fa1 > 8191)
394 a2p += 0xFF;
395 else
396 a2p += fa1 >> 5;
397
398 if (pk0 ^ state_ptr->pk[1])
399 /* LIMC */
400 if (a2p <= -12160)
401 a2p = -12288;
402 else if (a2p >= 12416)
403 a2p = 12288;
404 else
405 a2p -= 0x80;
406 else if (a2p <= -12416)
407 a2p = -12288;
408 else if (a2p >= 12160)
409 a2p = 12288;
410 else
411 a2p += 0x80;
412 }
413
414 /* TRIGB & DELAY */
415 state_ptr->a[1] = a2p;
416
417 /* UPA1 */
418 /* update predictor pole a[0] */
419 state_ptr->a[0] -= state_ptr->a[0] >> 8;
420 if (dqsez != 0)
421 if (pks1 == 0)
422 state_ptr->a[0] += 192;
423 else
424 state_ptr->a[0] -= 192;
425
426 /* LIMD */
427 a1ul = 15360 - a2p;
428 if (state_ptr->a[0] < -a1ul)
429 state_ptr->a[0] = -a1ul;
430 else if (state_ptr->a[0] > a1ul)
431 state_ptr->a[0] = a1ul;
432
433 /* UPB : update predictor zeros b[6] */
434 for (cnt = 0; cnt < 6; cnt++) {
435 if (code_size == 5) /* for 40Kbps G.723 */
436 state_ptr->b[cnt] -= state_ptr->b[cnt] >> 9;
437 else /* for G.721 and 24Kbps G.723 */
438 state_ptr->b[cnt] -= state_ptr->b[cnt] >> 8;
439 if (dq & 0x7FFF) { /* XOR */
440 if ((dq ^ state_ptr->dq[cnt]) >= 0)
441 state_ptr->b[cnt] += 128;
442 else
443 state_ptr->b[cnt] -= 128;
444 }
445 }
446 }
447
448 for (cnt = 5; cnt > 0; cnt--)
449 state_ptr->dq[cnt] = state_ptr->dq[cnt-1];
450 /* FLOAT A : convert dq[0] to 4-bit exp, 6-bit mantissa f.p. */
451 if (mag == 0) {
452 state_ptr->dq[0] = (dq >= 0) ? 0x20 : 0xFC20;
453 } else {
454 state_ptr->dq[0] = (dq >= 0) ?
455 base2 (mag) : base2 (mag) - 0x400;
456 }
457
458 state_ptr->sr[1] = state_ptr->sr[0];
459 /* FLOAT B : convert sr to 4-bit exp., 6-bit mantissa f.p. */
460 if (sr == 0) {
461 state_ptr->sr[0] = 0x20;
462 } else if (sr > 0) {
463 state_ptr->sr[0] = base2(sr);
464 } else if (sr > -32768) {
465 mag = -sr;
466 state_ptr->sr[0] = base2(mag) - 0x400;
467 } else
468 state_ptr->sr[0] = short (0xFC20);
469
470 /* DELAY A */
471 state_ptr->pk[1] = state_ptr->pk[0];
472 state_ptr->pk[0] = pk0;
473
474 /* TONE */
475 if (tr == 1) /* this sample has been treated as data */
476 state_ptr->td = 0; /* next one will be treated as voice */
477 else if (a2p < -11776) /* small sample-to-sample correlation */
478 state_ptr->td = 1; /* signal may be data */
479 else /* signal is voice */
480 state_ptr->td = 0;
481
482 /*
483 * Adaptation speed control.
484 */
485 state_ptr->dms += (fi - state_ptr->dms) >> 5; /* FILTA */
486 state_ptr->dml += (((fi << 2) - state_ptr->dml) >> 7); /* FILTB */
487
488 if (tr == 1)
489 state_ptr->ap = 256;
490 else if (y < 1536) /* SUBTC */
491 state_ptr->ap += (0x200 - state_ptr->ap) >> 4;
492 else if (state_ptr->td == 1)
493 state_ptr->ap += (0x200 - state_ptr->ap) >> 4;
494 else if (abs((state_ptr->dms << 2) - state_ptr->dml) >=
495 (state_ptr->dml >> 3))
496 state_ptr->ap += (0x200 - state_ptr->ap) >> 4;
497 else
498 state_ptr->ap += (-state_ptr->ap) >> 4;
499 }
500
501 /*
502 * tandem_adjust(sr, se, y, i, sign)
503 *
504 * At the end of ADPCM decoding, it simulates an encoder which may be receiving
505 * the output of this decoder as a tandem process. If the output of the
506 * simulated encoder differs from the input to this decoder, the decoder output
507 * is adjusted by one level of A-law or u-law codes.
508 *
509 * Input:
510 * sr decoder output linear PCM sample,
511 * se predictor estimate sample,
512 * y quantizer step size,
513 * i decoder input code,
514 * sign sign bit of code i
515 *
516 * Return:
517 * adjusted A-law or u-law compressed sample.
518 */
519 int
520 tandem_adjust_alaw(
521 int sr, /* decoder output linear PCM sample */
522 int se, /* predictor estimate sample */
523 int y, /* quantizer step size */
524 int i, /* decoder input code */
525 int sign,
526 short *qtab)
527 {
528 unsigned char sp; /* A-law compressed 8-bit code */
529 short dx; /* prediction error */
530 char id; /* quantized prediction error */
531 int sd; /* adjusted A-law decoded sample value */
532 int im; /* biased magnitude of i */
533 int imx; /* biased magnitude of id */
534
535 if (sr <= -32768)
536 sr = -1;
537 sp = linear2alaw((sr >> 1) << 3); /* short to A-law compression */
538 dx = (alaw2linear(sp) >> 2) - se; /* 16-bit prediction error */
539 id = quantize(dx, y, qtab, sign - 1);
540
541 if (id == i) { /* no adjustment on sp */
542 return (sp);
543 } else { /* sp adjustment needed */
544 /* ADPCM codes : 8, 9, ... F, 0, 1, ... , 6, 7 */
545 im = i ^ sign; /* 2's complement to biased unsigned */
546 imx = id ^ sign;
547
548 if (imx > im) { /* sp adjusted to next lower value */
549 if (sp & 0x80) {
550 sd = (sp == 0xD5) ? 0x55 :
551 ((sp ^ 0x55) - 1) ^ 0x55;
552 } else {
553 sd = (sp == 0x2A) ? 0x2A :
554 ((sp ^ 0x55) + 1) ^ 0x55;
555 }
556 } else { /* sp adjusted to next higher value */
557 if (sp & 0x80)
558 sd = (sp == 0xAA) ? 0xAA :
559 ((sp ^ 0x55) + 1) ^ 0x55;
560 else
561 sd = (sp == 0x55) ? 0xD5 :
562 ((sp ^ 0x55) - 1) ^ 0x55;
563 }
564 return (sd);
565 }
566 }
567
568 int
569 tandem_adjust_ulaw(
570 int sr, /* decoder output linear PCM sample */
571 int se, /* predictor estimate sample */
572 int y, /* quantizer step size */
573 int i, /* decoder input code */
574 int sign,
575 short *qtab)
576 {
577 unsigned char sp; /* u-law compressed 8-bit code */
578 short dx; /* prediction error */
579 char id; /* quantized prediction error */
580 int sd; /* adjusted u-law decoded sample value */
581 int im; /* biased magnitude of i */
582 int imx; /* biased magnitude of id */
583
584 if (sr <= -32768)
585 sr = 0;
586 sp = linear2ulaw(sr << 2); /* short to u-law compression */
587 dx = (ulaw2linear(sp) >> 2) - se; /* 16-bit prediction error */
588 id = quantize(dx, y, qtab, sign - 1);
589 if (id == i) {
590 return (sp);
591 } else {
592 /* ADPCM codes : 8, 9, ... F, 0, 1, ... , 6, 7 */
593 im = i ^ sign; /* 2's complement to biased unsigned */
594 imx = id ^ sign;
595 if (imx > im) { /* sp adjusted to next lower value */
596 if (sp & 0x80)
597 sd = (sp == 0xFF) ? 0x7E : sp + 1;
598 else
599 sd = (sp == 0) ? 0 : sp - 1;
600
601 } else { /* sp adjusted to next higher value */
602 if (sp & 0x80)
603 sd = (sp == 0x80) ? 0x80 : sp - 1;
604 else
605 sd = (sp == 0x7F) ? 0xFE : sp + 1;
606 }
607 return (sd);
608 }
609 }