0d2dcbbdb29cbeda72a318c5fbe800c8fb23f324
[libav.git] / libavcodec / takdec.c
1 /*
2 * TAK decoder
3 * Copyright (c) 2012 Paul B Mahol
4 *
5 * This file is part of Libav.
6 *
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * TAK (Tom's lossless Audio Kompressor) decoder
25 * @author Paul B Mahol
26 */
27
28 #include "libavutil/internal.h"
29 #include "libavutil/samplefmt.h"
30 #include "tak.h"
31 #include "avcodec.h"
32 #include "dsputil.h"
33 #include "internal.h"
34 #include "unary.h"
35
36 #define MAX_SUBFRAMES 8 // max number of subframes per channel
37 #define MAX_PREDICTORS 256
38
39 typedef struct MCDParam {
40 int8_t present; // decorrelation parameter availability for this channel
41 int8_t index; // index into array of decorrelation types
42 int8_t chan1;
43 int8_t chan2;
44 } MCDParam;
45
46 typedef struct TAKDecContext {
47 AVCodecContext *avctx; // parent AVCodecContext
48 DSPContext dsp;
49 TAKStreamInfo ti;
50 GetBitContext gb; // bitstream reader initialized to start at the current frame
51
52 int uval;
53 int nb_samples; // number of samples in the current frame
54 uint8_t *decode_buffer;
55 unsigned int decode_buffer_size;
56 int32_t *decoded[TAK_MAX_CHANNELS]; // decoded samples for each channel
57
58 int8_t lpc_mode[TAK_MAX_CHANNELS];
59 int8_t sample_shift[TAK_MAX_CHANNELS]; // shift applied to every sample in the channel
60 int subframe_scale;
61
62 int8_t dmode; // channel decorrelation type in the current frame
63
64 MCDParam mcdparams[TAK_MAX_CHANNELS]; // multichannel decorrelation parameters
65
66 int16_t *residues;
67 unsigned int residues_buf_size;
68 } TAKDecContext;
69
70 static const int8_t mc_dmodes[] = { 1, 3, 4, 6, };
71
72 static const uint16_t predictor_sizes[] = {
73 4, 8, 12, 16, 24, 32, 48, 64, 80, 96, 128, 160, 192, 224, 256, 0,
74 };
75
76 static const struct CParam {
77 int init;
78 int escape;
79 int scale;
80 int aescape;
81 int bias;
82 } xcodes[50] = {
83 { 0x01, 0x0000001, 0x0000001, 0x0000003, 0x0000008 },
84 { 0x02, 0x0000003, 0x0000001, 0x0000007, 0x0000006 },
85 { 0x03, 0x0000005, 0x0000002, 0x000000E, 0x000000D },
86 { 0x03, 0x0000003, 0x0000003, 0x000000D, 0x0000018 },
87 { 0x04, 0x000000B, 0x0000004, 0x000001C, 0x0000019 },
88 { 0x04, 0x0000006, 0x0000006, 0x000001A, 0x0000030 },
89 { 0x05, 0x0000016, 0x0000008, 0x0000038, 0x0000032 },
90 { 0x05, 0x000000C, 0x000000C, 0x0000034, 0x0000060 },
91 { 0x06, 0x000002C, 0x0000010, 0x0000070, 0x0000064 },
92 { 0x06, 0x0000018, 0x0000018, 0x0000068, 0x00000C0 },
93 { 0x07, 0x0000058, 0x0000020, 0x00000E0, 0x00000C8 },
94 { 0x07, 0x0000030, 0x0000030, 0x00000D0, 0x0000180 },
95 { 0x08, 0x00000B0, 0x0000040, 0x00001C0, 0x0000190 },
96 { 0x08, 0x0000060, 0x0000060, 0x00001A0, 0x0000300 },
97 { 0x09, 0x0000160, 0x0000080, 0x0000380, 0x0000320 },
98 { 0x09, 0x00000C0, 0x00000C0, 0x0000340, 0x0000600 },
99 { 0x0A, 0x00002C0, 0x0000100, 0x0000700, 0x0000640 },
100 { 0x0A, 0x0000180, 0x0000180, 0x0000680, 0x0000C00 },
101 { 0x0B, 0x0000580, 0x0000200, 0x0000E00, 0x0000C80 },
102 { 0x0B, 0x0000300, 0x0000300, 0x0000D00, 0x0001800 },
103 { 0x0C, 0x0000B00, 0x0000400, 0x0001C00, 0x0001900 },
104 { 0x0C, 0x0000600, 0x0000600, 0x0001A00, 0x0003000 },
105 { 0x0D, 0x0001600, 0x0000800, 0x0003800, 0x0003200 },
106 { 0x0D, 0x0000C00, 0x0000C00, 0x0003400, 0x0006000 },
107 { 0x0E, 0x0002C00, 0x0001000, 0x0007000, 0x0006400 },
108 { 0x0E, 0x0001800, 0x0001800, 0x0006800, 0x000C000 },
109 { 0x0F, 0x0005800, 0x0002000, 0x000E000, 0x000C800 },
110 { 0x0F, 0x0003000, 0x0003000, 0x000D000, 0x0018000 },
111 { 0x10, 0x000B000, 0x0004000, 0x001C000, 0x0019000 },
112 { 0x10, 0x0006000, 0x0006000, 0x001A000, 0x0030000 },
113 { 0x11, 0x0016000, 0x0008000, 0x0038000, 0x0032000 },
114 { 0x11, 0x000C000, 0x000C000, 0x0034000, 0x0060000 },
115 { 0x12, 0x002C000, 0x0010000, 0x0070000, 0x0064000 },
116 { 0x12, 0x0018000, 0x0018000, 0x0068000, 0x00C0000 },
117 { 0x13, 0x0058000, 0x0020000, 0x00E0000, 0x00C8000 },
118 { 0x13, 0x0030000, 0x0030000, 0x00D0000, 0x0180000 },
119 { 0x14, 0x00B0000, 0x0040000, 0x01C0000, 0x0190000 },
120 { 0x14, 0x0060000, 0x0060000, 0x01A0000, 0x0300000 },
121 { 0x15, 0x0160000, 0x0080000, 0x0380000, 0x0320000 },
122 { 0x15, 0x00C0000, 0x00C0000, 0x0340000, 0x0600000 },
123 { 0x16, 0x02C0000, 0x0100000, 0x0700000, 0x0640000 },
124 { 0x16, 0x0180000, 0x0180000, 0x0680000, 0x0C00000 },
125 { 0x17, 0x0580000, 0x0200000, 0x0E00000, 0x0C80000 },
126 { 0x17, 0x0300000, 0x0300000, 0x0D00000, 0x1800000 },
127 { 0x18, 0x0B00000, 0x0400000, 0x1C00000, 0x1900000 },
128 { 0x18, 0x0600000, 0x0600000, 0x1A00000, 0x3000000 },
129 { 0x19, 0x1600000, 0x0800000, 0x3800000, 0x3200000 },
130 { 0x19, 0x0C00000, 0x0C00000, 0x3400000, 0x6000000 },
131 { 0x1A, 0x2C00000, 0x1000000, 0x7000000, 0x6400000 },
132 { 0x1A, 0x1800000, 0x1800000, 0x6800000, 0xC000000 },
133 };
134
135 static av_cold void tak_init_static_data(AVCodec *codec)
136 {
137 ff_tak_init_crc();
138 }
139
140 static int set_bps_params(AVCodecContext *avctx)
141 {
142 switch (avctx->bits_per_coded_sample) {
143 case 8:
144 avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
145 break;
146 case 16:
147 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
148 break;
149 case 24:
150 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
151 break;
152 default:
153 av_log(avctx, AV_LOG_ERROR, "unsupported bits per sample: %d\n",
154 avctx->bits_per_coded_sample);
155 return AVERROR_INVALIDDATA;
156 }
157 avctx->bits_per_raw_sample = avctx->bits_per_coded_sample;
158
159 return 0;
160 }
161
162 static void set_sample_rate_params(AVCodecContext *avctx)
163 {
164 TAKDecContext *s = avctx->priv_data;
165 int shift = 3 - (avctx->sample_rate / 11025);
166 shift = FFMAX(0, shift);
167 s->uval = FFALIGN(avctx->sample_rate + 511 >> 9, 4) << shift;
168 s->subframe_scale = FFALIGN(avctx->sample_rate + 511 >> 9, 4) << 1;
169 }
170
171 static av_cold int tak_decode_init(AVCodecContext *avctx)
172 {
173 TAKDecContext *s = avctx->priv_data;
174
175 ff_dsputil_init(&s->dsp, avctx);
176
177 s->avctx = avctx;
178
179 set_sample_rate_params(avctx);
180
181 return set_bps_params(avctx);
182 }
183
184 static void decode_lpc(int32_t *coeffs, int mode, int length)
185 {
186 int i;
187
188 if (length < 2)
189 return;
190
191 if (mode == 1) {
192 int a1 = *coeffs++;
193 for (i = 0; i < length - 1 >> 1; i++) {
194 *coeffs += a1;
195 coeffs[1] += *coeffs;
196 a1 = coeffs[1];
197 coeffs += 2;
198 }
199 if (length - 1 & 1)
200 *coeffs += a1;
201 } else if (mode == 2) {
202 int a1 = coeffs[1];
203 int a2 = a1 + *coeffs;
204 coeffs[1] = a2;
205 if (length > 2) {
206 coeffs += 2;
207 for (i = 0; i < length - 2 >> 1; i++) {
208 int a3 = *coeffs + a1;
209 int a4 = a3 + a2;
210 *coeffs = a4;
211 a1 = coeffs[1] + a3;
212 a2 = a1 + a4;
213 coeffs[1] = a2;
214 coeffs += 2;
215 }
216 if (length & 1)
217 *coeffs += a1 + a2;
218 }
219 } else if (mode == 3) {
220 int a1 = coeffs[1];
221 int a2 = a1 + *coeffs;
222 coeffs[1] = a2;
223 if (length > 2) {
224 int a3 = coeffs[2];
225 int a4 = a3 + a1;
226 int a5 = a4 + a2;
227 coeffs += 3;
228 for (i = 0; i < length - 3; i++) {
229 a3 += *coeffs;
230 a4 += a3;
231 a5 += a4;
232 *coeffs = a5;
233 coeffs++;
234 }
235 }
236 }
237 }
238
239 static int decode_segment(GetBitContext *gb, int mode, int32_t *decoded,
240 int len)
241 {
242 struct CParam code;
243 int i;
244
245 if (!mode) {
246 memset(decoded, 0, len * sizeof(*decoded));
247 return 0;
248 }
249
250 if (mode > FF_ARRAY_ELEMS(xcodes))
251 return AVERROR_INVALIDDATA;
252 code = xcodes[mode - 1];
253
254 for (i = 0; i < len; i++) {
255 int x = get_bits_long(gb, code.init);
256 if (x >= code.escape && get_bits1(gb)) {
257 x |= 1 << code.init;
258 if (x >= code.aescape) {
259 int scale = get_unary(gb, 1, 9);
260 if (scale == 9) {
261 int scale_bits = get_bits(gb, 3);
262 if (scale_bits > 0) {
263 if (scale_bits == 7) {
264 scale_bits += get_bits(gb, 5);
265 if (scale_bits > 29)
266 return AVERROR_INVALIDDATA;
267 }
268 scale = get_bits_long(gb, scale_bits) + 1;
269 x += code.scale * scale;
270 }
271 x += code.bias;
272 } else
273 x += code.scale * scale - code.escape;
274 } else
275 x -= code.escape;
276 }
277 decoded[i] = (x >> 1) ^ -(x & 1);
278 }
279
280 return 0;
281 }
282
283 static int decode_residues(TAKDecContext *s, int32_t *decoded, int length)
284 {
285 GetBitContext *gb = &s->gb;
286 int i, mode, ret;
287
288 if (length > s->nb_samples)
289 return AVERROR_INVALIDDATA;
290
291 if (get_bits1(gb)) {
292 int wlength, rval;
293 int coding_mode[128];
294
295 wlength = length / s->uval;
296
297 rval = length - (wlength * s->uval);
298
299 if (rval < s->uval / 2)
300 rval += s->uval;
301 else
302 wlength++;
303
304 if (wlength <= 1 || wlength > 128)
305 return AVERROR_INVALIDDATA;
306
307 coding_mode[0] = mode = get_bits(gb, 6);
308
309 for (i = 1; i < wlength; i++) {
310 int c = get_unary(gb, 1, 6);
311
312 switch (c) {
313 case 6:
314 mode = get_bits(gb, 6);
315 break;
316 case 5:
317 case 4:
318 case 3: {
319 /* mode += sign ? (1 - c) : (c - 1) */
320 int sign = get_bits1(gb);
321 mode += (-sign ^ (c - 1)) + sign;
322 break;
323 }
324 case 2:
325 mode++;
326 break;
327 case 1:
328 mode--;
329 break;
330 }
331 coding_mode[i] = mode;
332 }
333
334 i = 0;
335 while (i < wlength) {
336 int len = 0;
337
338 mode = coding_mode[i];
339 do {
340 if (i >= wlength - 1)
341 len += rval;
342 else
343 len += s->uval;
344 i++;
345
346 if (i == wlength)
347 break;
348 } while (coding_mode[i] == mode);
349
350 if ((ret = decode_segment(gb, mode, decoded, len)) < 0)
351 return ret;
352 decoded += len;
353 }
354 } else {
355 mode = get_bits(gb, 6);
356 if ((ret = decode_segment(gb, mode, decoded, length)) < 0)
357 return ret;
358 }
359
360 return 0;
361 }
362
363 static int get_bits_esc4(GetBitContext *gb)
364 {
365 if (get_bits1(gb))
366 return get_bits(gb, 4) + 1;
367 else
368 return 0;
369 }
370
371 static void decode_filter_coeffs(TAKDecContext *s, int filter_order, int size,
372 int filter_quant, int16_t *filter)
373 {
374 GetBitContext *gb = &s->gb;
375 int i, j, a, b;
376 int filter_tmp[MAX_PREDICTORS];
377 int16_t predictors[MAX_PREDICTORS];
378
379 predictors[0] = get_sbits(gb, 10);
380 predictors[1] = get_sbits(gb, 10);
381 predictors[2] = get_sbits(gb, size) << (10 - size);
382 predictors[3] = get_sbits(gb, size) << (10 - size);
383 if (filter_order > 4) {
384 int av_uninit(code_size);
385 int code_size_base = size - get_bits1(gb);
386
387 for (i = 4; i < filter_order; i++) {
388 if (!(i & 3))
389 code_size = code_size_base - get_bits(gb, 2);
390 predictors[i] = get_sbits(gb, code_size) << (10 - size);
391 }
392 }
393
394 filter_tmp[0] = predictors[0] << 6;
395 for (i = 1; i < filter_order; i++) {
396 int *p1 = &filter_tmp[0];
397 int *p2 = &filter_tmp[i - 1];
398
399 for (j = 0; j < (i + 1) / 2; j++) {
400 int tmp = *p1 + (predictors[i] * *p2 + 256 >> 9);
401 *p2 = *p2 + (predictors[i] * *p1 + 256 >> 9);
402 *p1 = tmp;
403 p1++;
404 p2--;
405 }
406
407 filter_tmp[i] = predictors[i] << 6;
408 }
409
410 a = 1 << (32 - (15 - filter_quant));
411 b = 1 << ((15 - filter_quant) - 1);
412 for (i = 0, j = filter_order - 1; i < filter_order / 2; i++, j--) {
413 filter[j] = a - ((filter_tmp[i] + b) >> (15 - filter_quant));
414 filter[i] = a - ((filter_tmp[j] + b) >> (15 - filter_quant));
415 }
416 }
417
418 static int decode_subframe(TAKDecContext *s, int32_t *decoded,
419 int subframe_size, int prev_subframe_size)
420 {
421 LOCAL_ALIGNED_16(int16_t, filter, [MAX_PREDICTORS]);
422 GetBitContext *gb = &s->gb;
423 int i, ret;
424 int dshift, size, filter_quant, filter_order;
425
426 memset(filter, 0, MAX_PREDICTORS * sizeof(*filter));
427
428 if (!get_bits1(gb))
429 return decode_residues(s, decoded, subframe_size);
430
431 filter_order = predictor_sizes[get_bits(gb, 4)];
432
433 if (prev_subframe_size > 0 && get_bits1(gb)) {
434 if (filter_order > prev_subframe_size)
435 return AVERROR_INVALIDDATA;
436
437 decoded -= filter_order;
438 subframe_size += filter_order;
439
440 if (filter_order > subframe_size)
441 return AVERROR_INVALIDDATA;
442 } else {
443 int lpc_mode;
444
445 if (filter_order > subframe_size)
446 return AVERROR_INVALIDDATA;
447
448 lpc_mode = get_bits(gb, 2);
449 if (lpc_mode > 2)
450 return AVERROR_INVALIDDATA;
451
452 if ((ret = decode_residues(s, decoded, filter_order)) < 0)
453 return ret;
454
455 if (lpc_mode)
456 decode_lpc(decoded, lpc_mode, filter_order);
457 }
458
459 dshift = get_bits_esc4(gb);
460 size = get_bits1(gb) + 6;
461
462 filter_quant = 10;
463 if (get_bits1(gb)) {
464 filter_quant -= get_bits(gb, 3) + 1;
465 if (filter_quant < 3)
466 return AVERROR_INVALIDDATA;
467 }
468
469 decode_filter_coeffs(s, filter_order, size, filter_quant, filter);
470
471 if ((ret = decode_residues(s, &decoded[filter_order],
472 subframe_size - filter_order)) < 0)
473 return ret;
474
475 av_fast_malloc(&s->residues, &s->residues_buf_size,
476 FFALIGN(subframe_size + 16, 16) * sizeof(*s->residues));
477 if (!s->residues)
478 return AVERROR(ENOMEM);
479 memset(s->residues, 0, s->residues_buf_size);
480
481 for (i = 0; i < filter_order; i++)
482 s->residues[i] = *decoded++ >> dshift;
483
484 for (i = 0; i < subframe_size - filter_order; i++) {
485 int v = 1 << (filter_quant - 1);
486
487 v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
488 FFALIGN(filter_order, 16));
489
490 v = (av_clip(v >> filter_quant, -8192, 8191) << dshift) - *decoded;
491 *decoded++ = v;
492 s->residues[filter_order + i] = v >> dshift;
493 }
494
495 emms_c();
496
497 return 0;
498 }
499
500 static int decode_channel(TAKDecContext *s, int chan)
501 {
502 AVCodecContext *avctx = s->avctx;
503 GetBitContext *gb = &s->gb;
504 int32_t *decoded = s->decoded[chan];
505 int left = s->nb_samples - 1;
506 int i, prev, ret, nb_subframes;
507 int subframe_len[MAX_SUBFRAMES];
508
509 s->sample_shift[chan] = get_bits_esc4(gb);
510 if (s->sample_shift[chan] >= avctx->bits_per_coded_sample)
511 return AVERROR_INVALIDDATA;
512
513 /* NOTE: TAK 2.2.0 appears to set the sample value to 0 if
514 * bits_per_coded_sample - sample_shift is 1, but this produces
515 * non-bit-exact output. Reading the 1 bit using get_sbits() instead
516 * of skipping it produces bit-exact output. This has been reported
517 * to the TAK author. */
518 *decoded++ = get_sbits(gb,
519 avctx->bits_per_coded_sample -
520 s->sample_shift[chan]);
521 s->lpc_mode[chan] = get_bits(gb, 2);
522 nb_subframes = get_bits(gb, 3) + 1;
523
524 i = 0;
525 if (nb_subframes > 1) {
526 if (get_bits_left(gb) < (nb_subframes - 1) * 6)
527 return AVERROR_INVALIDDATA;
528
529 prev = 0;
530 for (; i < nb_subframes - 1; i++) {
531 int subframe_end = get_bits(gb, 6) * s->subframe_scale;
532 if (subframe_end <= prev)
533 return AVERROR_INVALIDDATA;
534 subframe_len[i] = subframe_end - prev;
535 left -= subframe_len[i];
536 prev = subframe_end;
537 }
538
539 if (left <= 0)
540 return AVERROR_INVALIDDATA;
541 }
542 subframe_len[i] = left;
543
544 prev = 0;
545 for (i = 0; i < nb_subframes; i++) {
546 if ((ret = decode_subframe(s, decoded, subframe_len[i], prev)) < 0)
547 return ret;
548 decoded += subframe_len[i];
549 prev = subframe_len[i];
550 }
551
552 return 0;
553 }
554
555 static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
556 {
557 GetBitContext *gb = &s->gb;
558 int32_t *p1 = s->decoded[c1] + 1;
559 int32_t *p2 = s->decoded[c2] + 1;
560 int i;
561 int dshift, dfactor;
562
563 switch (s->dmode) {
564 case 1: /* left/side */
565 for (i = 0; i < length; i++) {
566 int32_t a = p1[i];
567 int32_t b = p2[i];
568 p2[i] = a + b;
569 }
570 break;
571 case 2: /* side/right */
572 for (i = 0; i < length; i++) {
573 int32_t a = p1[i];
574 int32_t b = p2[i];
575 p1[i] = b - a;
576 }
577 break;
578 case 3: /* side/mid */
579 for (i = 0; i < length; i++) {
580 int32_t a = p1[i];
581 int32_t b = p2[i];
582 a -= b >> 1;
583 p1[i] = a;
584 p2[i] = a + b;
585 }
586 break;
587 case 4: /* side/left with scale factor */
588 FFSWAP(int32_t*, p1, p2);
589 case 5: /* side/right with scale factor */
590 dshift = get_bits_esc4(gb);
591 dfactor = get_sbits(gb, 10);
592 for (i = 0; i < length; i++) {
593 int32_t a = p1[i];
594 int32_t b = p2[i];
595 b = dfactor * (b >> dshift) + 128 >> 8 << dshift;
596 p1[i] = b - a;
597 }
598 break;
599 case 6:
600 FFSWAP(int32_t*, p1, p2);
601 case 7: {
602 LOCAL_ALIGNED_16(int16_t, filter, [MAX_PREDICTORS]);
603 int length2, order_half, filter_order, dval1, dval2;
604 int av_uninit(code_size);
605
606 memset(filter, 0, MAX_PREDICTORS * sizeof(*filter));
607
608 if (length < 256)
609 return AVERROR_INVALIDDATA;
610
611 dshift = get_bits_esc4(gb);
612 filter_order = 8 << get_bits1(gb);
613 dval1 = get_bits1(gb);
614 dval2 = get_bits1(gb);
615
616 for (i = 0; i < filter_order; i++) {
617 if (!(i & 3))
618 code_size = 14 - get_bits(gb, 3);
619 filter[i] = get_sbits(gb, code_size);
620 }
621
622 order_half = filter_order / 2;
623 length2 = length - (filter_order - 1);
624
625 /* decorrelate beginning samples */
626 if (dval1) {
627 for (i = 0; i < order_half; i++) {
628 int32_t a = p1[i];
629 int32_t b = p2[i];
630 p1[i] = a + b;
631 }
632 }
633
634 /* decorrelate ending samples */
635 if (dval2) {
636 for (i = length2 + order_half; i < length; i++) {
637 int32_t a = p1[i];
638 int32_t b = p2[i];
639 p1[i] = a + b;
640 }
641 }
642
643 av_fast_malloc(&s->residues, &s->residues_buf_size,
644 FFALIGN(length + 16, 16) * sizeof(*s->residues));
645 if (!s->residues)
646 return AVERROR(ENOMEM);
647 memset(s->residues, 0, s->residues_buf_size);
648
649 for (i = 0; i < length; i++)
650 s->residues[i] = p2[i] >> dshift;
651
652 p1 += order_half;
653
654 for (i = 0; i < length2; i++) {
655 int v = 1 << 9;
656
657 v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
658 FFALIGN(filter_order, 16));
659
660 p1[i] = (av_clip(v >> 10, -8192, 8191) << dshift) - p1[i];
661 }
662
663 emms_c();
664 break;
665 }
666 }
667
668 return 0;
669 }
670
671 static int tak_decode_frame(AVCodecContext *avctx, void *data,
672 int *got_frame_ptr, AVPacket *pkt)
673 {
674 TAKDecContext *s = avctx->priv_data;
675 AVFrame *frame = data;
676 GetBitContext *gb = &s->gb;
677 int chan, i, ret, hsize;
678
679 if (pkt->size < TAK_MIN_FRAME_HEADER_BYTES)
680 return AVERROR_INVALIDDATA;
681
682 init_get_bits(gb, pkt->data, pkt->size * 8);
683
684 if ((ret = ff_tak_decode_frame_header(avctx, gb, &s->ti, 0)) < 0)
685 return ret;
686
687 if (s->ti.flags & TAK_FRAME_FLAG_HAS_METADATA) {
688 avpriv_request_sample(avctx, "Frame metadata");
689 return AVERROR_PATCHWELCOME;
690 }
691
692 hsize = get_bits_count(gb) / 8;
693 if (avctx->err_recognition & AV_EF_CRCCHECK) {
694 if (ff_tak_check_crc(pkt->data, hsize)) {
695 av_log(avctx, AV_LOG_ERROR, "CRC error\n");
696 if (avctx->err_recognition & AV_EF_EXPLODE)
697 return AVERROR_INVALIDDATA;
698 }
699 }
700
701 if (s->ti.codec != TAK_CODEC_MONO_STEREO &&
702 s->ti.codec != TAK_CODEC_MULTICHANNEL) {
703 av_log(avctx, AV_LOG_ERROR, "unsupported codec: %d\n", s->ti.codec);
704 return AVERROR_PATCHWELCOME;
705 }
706 if (s->ti.data_type) {
707 av_log(avctx, AV_LOG_ERROR,
708 "unsupported data type: %d\n", s->ti.data_type);
709 return AVERROR_INVALIDDATA;
710 }
711 if (s->ti.codec == TAK_CODEC_MONO_STEREO && s->ti.channels > 2) {
712 av_log(avctx, AV_LOG_ERROR,
713 "invalid number of channels: %d\n", s->ti.channels);
714 return AVERROR_INVALIDDATA;
715 }
716 if (s->ti.channels > 6) {
717 av_log(avctx, AV_LOG_ERROR,
718 "unsupported number of channels: %d\n", s->ti.channels);
719 return AVERROR_INVALIDDATA;
720 }
721
722 if (s->ti.frame_samples <= 0) {
723 av_log(avctx, AV_LOG_ERROR, "unsupported/invalid number of samples\n");
724 return AVERROR_INVALIDDATA;
725 }
726
727 if (s->ti.bps != avctx->bits_per_coded_sample) {
728 avctx->bits_per_coded_sample = s->ti.bps;
729 if ((ret = set_bps_params(avctx)) < 0)
730 return ret;
731 }
732 if (s->ti.sample_rate != avctx->sample_rate) {
733 avctx->sample_rate = s->ti.sample_rate;
734 set_sample_rate_params(avctx);
735 }
736 if (s->ti.ch_layout)
737 avctx->channel_layout = s->ti.ch_layout;
738 avctx->channels = s->ti.channels;
739
740 s->nb_samples = s->ti.last_frame_samples ? s->ti.last_frame_samples
741 : s->ti.frame_samples;
742
743 frame->nb_samples = s->nb_samples;
744 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
745 return ret;
746
747 if (avctx->bits_per_coded_sample <= 16) {
748 int buf_size = av_samples_get_buffer_size(NULL, avctx->channels,
749 s->nb_samples,
750 AV_SAMPLE_FMT_S32P, 0);
751 av_fast_malloc(&s->decode_buffer, &s->decode_buffer_size, buf_size);
752 if (!s->decode_buffer)
753 return AVERROR(ENOMEM);
754 ret = av_samples_fill_arrays((uint8_t **)s->decoded, NULL,
755 s->decode_buffer, avctx->channels,
756 s->nb_samples, AV_SAMPLE_FMT_S32P, 0);
757 if (ret < 0)
758 return ret;
759 } else {
760 for (chan = 0; chan < avctx->channels; chan++)
761 s->decoded[chan] = (int32_t *)frame->extended_data[chan];
762 }
763
764 if (s->nb_samples < 16) {
765 for (chan = 0; chan < avctx->channels; chan++) {
766 int32_t *decoded = s->decoded[chan];
767 for (i = 0; i < s->nb_samples; i++)
768 decoded[i] = get_sbits(gb, avctx->bits_per_coded_sample);
769 }
770 } else {
771 if (s->ti.codec == TAK_CODEC_MONO_STEREO) {
772 for (chan = 0; chan < avctx->channels; chan++)
773 if (ret = decode_channel(s, chan))
774 return ret;
775
776 if (avctx->channels == 2) {
777 if (get_bits1(gb)) {
778 // some kind of subframe length, but it seems to be unused
779 skip_bits(gb, 6);
780 }
781
782 s->dmode = get_bits(gb, 3);
783 if (ret = decorrelate(s, 0, 1, s->nb_samples - 1))
784 return ret;
785 }
786 } else if (s->ti.codec == TAK_CODEC_MULTICHANNEL) {
787 if (get_bits1(gb)) {
788 int ch_mask = 0;
789
790 chan = get_bits(gb, 4) + 1;
791 if (chan > avctx->channels)
792 return AVERROR_INVALIDDATA;
793
794 for (i = 0; i < chan; i++) {
795 int nbit = get_bits(gb, 4);
796
797 if (nbit >= avctx->channels)
798 return AVERROR_INVALIDDATA;
799
800 if (ch_mask & 1 << nbit)
801 return AVERROR_INVALIDDATA;
802
803 s->mcdparams[i].present = get_bits1(gb);
804 if (s->mcdparams[i].present) {
805 s->mcdparams[i].index = get_bits(gb, 2);
806 s->mcdparams[i].chan2 = get_bits(gb, 4);
807 if (s->mcdparams[i].index == 1) {
808 if ((nbit == s->mcdparams[i].chan2) ||
809 (ch_mask & 1 << s->mcdparams[i].chan2))
810 return AVERROR_INVALIDDATA;
811
812 ch_mask |= 1 << s->mcdparams[i].chan2;
813 } else if (!(ch_mask & 1 << s->mcdparams[i].chan2)) {
814 return AVERROR_INVALIDDATA;
815 }
816 }
817 s->mcdparams[i].chan1 = nbit;
818
819 ch_mask |= 1 << nbit;
820 }
821 } else {
822 chan = avctx->channels;
823 for (i = 0; i < chan; i++) {
824 s->mcdparams[i].present = 0;
825 s->mcdparams[i].chan1 = i;
826 }
827 }
828
829 for (i = 0; i < chan; i++) {
830 if (s->mcdparams[i].present && s->mcdparams[i].index == 1)
831 if (ret = decode_channel(s, s->mcdparams[i].chan2))
832 return ret;
833
834 if (ret = decode_channel(s, s->mcdparams[i].chan1))
835 return ret;
836
837 if (s->mcdparams[i].present) {
838 s->dmode = mc_dmodes[s->mcdparams[i].index];
839 if (ret = decorrelate(s,
840 s->mcdparams[i].chan2,
841 s->mcdparams[i].chan1,
842 s->nb_samples - 1))
843 return ret;
844 }
845 }
846 }
847
848 for (chan = 0; chan < avctx->channels; chan++) {
849 int32_t *decoded = s->decoded[chan];
850
851 if (s->lpc_mode[chan])
852 decode_lpc(decoded, s->lpc_mode[chan], s->nb_samples);
853
854 if (s->sample_shift[chan] > 0)
855 for (i = 0; i < s->nb_samples; i++)
856 decoded[i] <<= s->sample_shift[chan];
857 }
858 }
859
860 align_get_bits(gb);
861 skip_bits(gb, 24);
862 if (get_bits_left(gb) < 0)
863 av_log(avctx, AV_LOG_DEBUG, "overread\n");
864 else if (get_bits_left(gb) > 0)
865 av_log(avctx, AV_LOG_DEBUG, "underread\n");
866
867 if (avctx->err_recognition & AV_EF_CRCCHECK) {
868 if (ff_tak_check_crc(pkt->data + hsize,
869 get_bits_count(gb) / 8 - hsize)) {
870 av_log(avctx, AV_LOG_ERROR, "CRC error\n");
871 if (avctx->err_recognition & AV_EF_EXPLODE)
872 return AVERROR_INVALIDDATA;
873 }
874 }
875
876 /* convert to output buffer */
877 switch (avctx->sample_fmt) {
878 case AV_SAMPLE_FMT_U8P:
879 for (chan = 0; chan < avctx->channels; chan++) {
880 uint8_t *samples = (uint8_t *)frame->extended_data[chan];
881 int32_t *decoded = s->decoded[chan];
882 for (i = 0; i < s->nb_samples; i++)
883 samples[i] = decoded[i] + 0x80;
884 }
885 break;
886 case AV_SAMPLE_FMT_S16P:
887 for (chan = 0; chan < avctx->channels; chan++) {
888 int16_t *samples = (int16_t *)frame->extended_data[chan];
889 int32_t *decoded = s->decoded[chan];
890 for (i = 0; i < s->nb_samples; i++)
891 samples[i] = decoded[i];
892 }
893 break;
894 case AV_SAMPLE_FMT_S32P:
895 for (chan = 0; chan < avctx->channels; chan++) {
896 int32_t *samples = (int32_t *)frame->extended_data[chan];
897 for (i = 0; i < s->nb_samples; i++)
898 samples[i] <<= 8;
899 }
900 break;
901 }
902
903 *got_frame_ptr = 1;
904
905 return pkt->size;
906 }
907
908 static av_cold int tak_decode_close(AVCodecContext *avctx)
909 {
910 TAKDecContext *s = avctx->priv_data;
911
912 av_freep(&s->decode_buffer);
913 av_freep(&s->residues);
914
915 return 0;
916 }
917
918 AVCodec ff_tak_decoder = {
919 .name = "tak",
920 .long_name = NULL_IF_CONFIG_SMALL("TAK (Tom's lossless Audio Kompressor)"),
921 .type = AVMEDIA_TYPE_AUDIO,
922 .id = AV_CODEC_ID_TAK,
923 .priv_data_size = sizeof(TAKDecContext),
924 .init = tak_decode_init,
925 .init_static_data = tak_init_static_data,
926 .close = tak_decode_close,
927 .decode = tak_decode_frame,
928 .capabilities = CODEC_CAP_DR1,
929 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
930 AV_SAMPLE_FMT_S16P,
931 AV_SAMPLE_FMT_S32P,
932 AV_SAMPLE_FMT_NONE },
933 };