1 /*
2 * TAK decoder
3 * Copyright (c) 2012 Paul B Mahol
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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/mem_internal.h"
30 #include "libavutil/samplefmt.h"
31
32 #define BITSTREAM_READER_LE
33 #include "audiodsp.h"
34 #include "thread.h"
35 #include "avcodec.h"
36 #include "codec_internal.h"
37 #include "unary.h"
38 #include "tak.h"
39 #include "takdsp.h"
40
41 #define MAX_SUBFRAMES 8 ///< max number of subframes per channel
42 #define MAX_PREDICTORS 256
43
44 typedef struct MCDParam {
45 int8_t present; ///< decorrelation parameter availability for this channel
46 int8_t index; ///< index into array of decorrelation types
47 int8_t chan1;
48 int8_t chan2;
49 } MCDParam;
50
51 typedef struct TAKDecContext {
52 AVCodecContext *avctx; ///< parent AVCodecContext
53 AudioDSPContext adsp;
54 TAKDSPContext tdsp;
55 TAKStreamInfo ti;
56 GetBitContext gb; ///< bitstream reader initialized to start at the current frame
57
58 int uval;
59 int nb_samples; ///< number of samples in the current frame
60 uint8_t *decode_buffer;
61 unsigned int decode_buffer_size;
62 int32_t *decoded[TAK_MAX_CHANNELS]; ///< decoded samples for each channel
63
64 int8_t lpc_mode[TAK_MAX_CHANNELS];
65 int8_t sample_shift[TAK_MAX_CHANNELS]; ///< shift applied to every sample in the channel
66 int16_t predictors[MAX_PREDICTORS];
67 int nb_subframes; ///< number of subframes in the current frame
68 int16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
69 int subframe_scale;
70
71 int8_t dmode; ///< channel decorrelation type in the current frame
72
73 MCDParam mcdparams[TAK_MAX_CHANNELS]; ///< multichannel decorrelation parameters
74
75 int8_t coding_mode[128];
76 DECLARE_ALIGNED(16, int16_t, filter)[MAX_PREDICTORS];
77 DECLARE_ALIGNED(16, int16_t, residues)[544];
78 } TAKDecContext;
79
80 static const int8_t mc_dmodes[] = { 1, 3, 4, 6, };
81
82 static const uint16_t predictor_sizes[] = {
83 4, 8, 12, 16, 24, 32, 48, 64, 80, 96, 128, 160, 192, 224, 256, 0,
84 };
85
86 static const struct CParam {
87 int init;
88 int escape;
89 int scale;
90 int aescape;
91 int bias;
92 } xcodes[50] = {
93 { 0x01, 0x0000001, 0x0000001, 0x0000003, 0x0000008 },
94 { 0x02, 0x0000003, 0x0000001, 0x0000007, 0x0000006 },
95 { 0x03, 0x0000005, 0x0000002, 0x000000E, 0x000000D },
96 { 0x03, 0x0000003, 0x0000003, 0x000000D, 0x0000018 },
97 { 0x04, 0x000000B, 0x0000004, 0x000001C, 0x0000019 },
98 { 0x04, 0x0000006, 0x0000006, 0x000001A, 0x0000030 },
99 { 0x05, 0x0000016, 0x0000008, 0x0000038, 0x0000032 },
100 { 0x05, 0x000000C, 0x000000C, 0x0000034, 0x0000060 },
101 { 0x06, 0x000002C, 0x0000010, 0x0000070, 0x0000064 },
102 { 0x06, 0x0000018, 0x0000018, 0x0000068, 0x00000C0 },
103 { 0x07, 0x0000058, 0x0000020, 0x00000E0, 0x00000C8 },
104 { 0x07, 0x0000030, 0x0000030, 0x00000D0, 0x0000180 },
105 { 0x08, 0x00000B0, 0x0000040, 0x00001C0, 0x0000190 },
106 { 0x08, 0x0000060, 0x0000060, 0x00001A0, 0x0000300 },
107 { 0x09, 0x0000160, 0x0000080, 0x0000380, 0x0000320 },
108 { 0x09, 0x00000C0, 0x00000C0, 0x0000340, 0x0000600 },
109 { 0x0A, 0x00002C0, 0x0000100, 0x0000700, 0x0000640 },
110 { 0x0A, 0x0000180, 0x0000180, 0x0000680, 0x0000C00 },
111 { 0x0B, 0x0000580, 0x0000200, 0x0000E00, 0x0000C80 },
112 { 0x0B, 0x0000300, 0x0000300, 0x0000D00, 0x0001800 },
113 { 0x0C, 0x0000B00, 0x0000400, 0x0001C00, 0x0001900 },
114 { 0x0C, 0x0000600, 0x0000600, 0x0001A00, 0x0003000 },
115 { 0x0D, 0x0001600, 0x0000800, 0x0003800, 0x0003200 },
116 { 0x0D, 0x0000C00, 0x0000C00, 0x0003400, 0x0006000 },
117 { 0x0E, 0x0002C00, 0x0001000, 0x0007000, 0x0006400 },
118 { 0x0E, 0x0001800, 0x0001800, 0x0006800, 0x000C000 },
119 { 0x0F, 0x0005800, 0x0002000, 0x000E000, 0x000C800 },
120 { 0x0F, 0x0003000, 0x0003000, 0x000D000, 0x0018000 },
121 { 0x10, 0x000B000, 0x0004000, 0x001C000, 0x0019000 },
122 { 0x10, 0x0006000, 0x0006000, 0x001A000, 0x0030000 },
123 { 0x11, 0x0016000, 0x0008000, 0x0038000, 0x0032000 },
124 { 0x11, 0x000C000, 0x000C000, 0x0034000, 0x0060000 },
125 { 0x12, 0x002C000, 0x0010000, 0x0070000, 0x0064000 },
126 { 0x12, 0x0018000, 0x0018000, 0x0068000, 0x00C0000 },
127 { 0x13, 0x0058000, 0x0020000, 0x00E0000, 0x00C8000 },
128 { 0x13, 0x0030000, 0x0030000, 0x00D0000, 0x0180000 },
129 { 0x14, 0x00B0000, 0x0040000, 0x01C0000, 0x0190000 },
130 { 0x14, 0x0060000, 0x0060000, 0x01A0000, 0x0300000 },
131 { 0x15, 0x0160000, 0x0080000, 0x0380000, 0x0320000 },
132 { 0x15, 0x00C0000, 0x00C0000, 0x0340000, 0x0600000 },
133 { 0x16, 0x02C0000, 0x0100000, 0x0700000, 0x0640000 },
134 { 0x16, 0x0180000, 0x0180000, 0x0680000, 0x0C00000 },
135 { 0x17, 0x0580000, 0x0200000, 0x0E00000, 0x0C80000 },
136 { 0x17, 0x0300000, 0x0300000, 0x0D00000, 0x1800000 },
137 { 0x18, 0x0B00000, 0x0400000, 0x1C00000, 0x1900000 },
138 { 0x18, 0x0600000, 0x0600000, 0x1A00000, 0x3000000 },
139 { 0x19, 0x1600000, 0x0800000, 0x3800000, 0x3200000 },
140 { 0x19, 0x0C00000, 0x0C00000, 0x3400000, 0x6000000 },
141 { 0x1A, 0x2C00000, 0x1000000, 0x7000000, 0x6400000 },
142 { 0x1A, 0x1800000, 0x1800000, 0x6800000, 0xC000000 },
143 };
144
set_bps_params(AVCodecContext * avctx)145 static int set_bps_params(AVCodecContext *avctx)
146 {
147 switch (avctx->bits_per_raw_sample) {
148 case 8:
149 avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
150 break;
151 case 16:
152 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
153 break;
154 case 24:
155 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
156 break;
157 default:
158 av_log(avctx, AV_LOG_ERROR, "invalid/unsupported bits per sample: %d\n",
159 avctx->bits_per_raw_sample);
160 return AVERROR_INVALIDDATA;
161 }
162
163 return 0;
164 }
165
set_sample_rate_params(AVCodecContext * avctx)166 static void set_sample_rate_params(AVCodecContext *avctx)
167 {
168 TAKDecContext *s = avctx->priv_data;
169 int shift;
170
171 if (avctx->sample_rate < 11025) {
172 shift = 3;
173 } else if (avctx->sample_rate < 22050) {
174 shift = 2;
175 } else if (avctx->sample_rate < 44100) {
176 shift = 1;
177 } else {
178 shift = 0;
179 }
180 s->uval = FFALIGN(avctx->sample_rate + 511LL >> 9, 4) << shift;
181 s->subframe_scale = FFALIGN(avctx->sample_rate + 511LL >> 9, 4) << 1;
182 }
183
tak_decode_init(AVCodecContext * avctx)184 static av_cold int tak_decode_init(AVCodecContext *avctx)
185 {
186 TAKDecContext *s = avctx->priv_data;
187
188 ff_audiodsp_init(&s->adsp);
189 ff_takdsp_init(&s->tdsp);
190
191 s->avctx = avctx;
192 avctx->bits_per_raw_sample = avctx->bits_per_coded_sample;
193
194 set_sample_rate_params(avctx);
195
196 return set_bps_params(avctx);
197 }
198
decode_lpc(int32_t * coeffs,int mode,int length)199 static void decode_lpc(int32_t *coeffs, int mode, int length)
200 {
201 int i;
202
203 if (length < 2)
204 return;
205
206 if (mode == 1) {
207 unsigned a1 = *coeffs++;
208 for (i = 0; i < length - 1 >> 1; i++) {
209 *coeffs += a1;
210 coeffs[1] += (unsigned)*coeffs;
211 a1 = coeffs[1];
212 coeffs += 2;
213 }
214 if (length - 1 & 1)
215 *coeffs += a1;
216 } else if (mode == 2) {
217 unsigned a1 = coeffs[1];
218 unsigned a2 = a1 + *coeffs;
219 coeffs[1] = a2;
220 if (length > 2) {
221 coeffs += 2;
222 for (i = 0; i < length - 2 >> 1; i++) {
223 unsigned a3 = *coeffs + a1;
224 unsigned a4 = a3 + a2;
225 *coeffs = a4;
226 a1 = coeffs[1] + a3;
227 a2 = a1 + a4;
228 coeffs[1] = a2;
229 coeffs += 2;
230 }
231 if (length & 1)
232 *coeffs += a1 + a2;
233 }
234 } else if (mode == 3) {
235 unsigned a1 = coeffs[1];
236 unsigned a2 = a1 + *coeffs;
237 coeffs[1] = a2;
238 if (length > 2) {
239 unsigned a3 = coeffs[2];
240 unsigned a4 = a3 + a1;
241 unsigned a5 = a4 + a2;
242 coeffs[2] = a5;
243 coeffs += 3;
244 for (i = 0; i < length - 3; i++) {
245 a3 += *coeffs;
246 a4 += a3;
247 a5 += a4;
248 *coeffs = a5;
249 coeffs++;
250 }
251 }
252 }
253 }
254
decode_segment(TAKDecContext * s,int8_t mode,int32_t * decoded,int len)255 static int decode_segment(TAKDecContext *s, int8_t mode, int32_t *decoded, int len)
256 {
257 struct CParam code;
258 GetBitContext *gb = &s->gb;
259 int i;
260
261 if (!mode) {
262 memset(decoded, 0, len * sizeof(*decoded));
263 return 0;
264 }
265
266 if (mode > FF_ARRAY_ELEMS(xcodes))
267 return AVERROR_INVALIDDATA;
268 code = xcodes[mode - 1];
269
270 for (i = 0; i < len; i++) {
271 unsigned x = get_bits_long(gb, code.init);
272 if (x >= code.escape && get_bits1(gb)) {
273 x |= 1 << code.init;
274 if (x >= code.aescape) {
275 unsigned scale = get_unary(gb, 1, 9);
276 if (scale == 9) {
277 int scale_bits = get_bits(gb, 3);
278 if (scale_bits > 0) {
279 if (scale_bits == 7) {
280 scale_bits += get_bits(gb, 5);
281 if (scale_bits > 29)
282 return AVERROR_INVALIDDATA;
283 }
284 scale = get_bits_long(gb, scale_bits) + 1;
285 x += code.scale * scale;
286 }
287 x += code.bias;
288 } else
289 x += code.scale * scale - code.escape;
290 } else
291 x -= code.escape;
292 }
293 decoded[i] = (x >> 1) ^ -(x & 1);
294 }
295
296 return 0;
297 }
298
decode_residues(TAKDecContext * s,int32_t * decoded,int length)299 static int decode_residues(TAKDecContext *s, int32_t *decoded, int length)
300 {
301 GetBitContext *gb = &s->gb;
302 int i, mode, ret;
303
304 if (length > s->nb_samples)
305 return AVERROR_INVALIDDATA;
306
307 if (get_bits1(gb)) {
308 int wlength, rval;
309
310 wlength = length / s->uval;
311
312 rval = length - (wlength * s->uval);
313
314 if (rval < s->uval / 2)
315 rval += s->uval;
316 else
317 wlength++;
318
319 if (wlength <= 1 || wlength > 128)
320 return AVERROR_INVALIDDATA;
321
322 s->coding_mode[0] = mode = get_bits(gb, 6);
323
324 for (i = 1; i < wlength; i++) {
325 int c = get_unary(gb, 1, 6);
326
327 switch (c) {
328 case 6:
329 mode = get_bits(gb, 6);
330 break;
331 case 5:
332 case 4:
333 case 3: {
334 /* mode += sign ? (1 - c) : (c - 1) */
335 int sign = get_bits1(gb);
336 mode += (-sign ^ (c - 1)) + sign;
337 break;
338 }
339 case 2:
340 mode++;
341 break;
342 case 1:
343 mode--;
344 break;
345 }
346 s->coding_mode[i] = mode;
347 }
348
349 i = 0;
350 while (i < wlength) {
351 int len = 0;
352
353 mode = s->coding_mode[i];
354 do {
355 if (i >= wlength - 1)
356 len += rval;
357 else
358 len += s->uval;
359 i++;
360
361 if (i == wlength)
362 break;
363 } while (s->coding_mode[i] == mode);
364
365 if ((ret = decode_segment(s, mode, decoded, len)) < 0)
366 return ret;
367 decoded += len;
368 }
369 } else {
370 mode = get_bits(gb, 6);
371 if ((ret = decode_segment(s, mode, decoded, length)) < 0)
372 return ret;
373 }
374
375 return 0;
376 }
377
get_bits_esc4(GetBitContext * gb)378 static int get_bits_esc4(GetBitContext *gb)
379 {
380 if (get_bits1(gb))
381 return get_bits(gb, 4) + 1;
382 else
383 return 0;
384 }
385
decode_subframe(TAKDecContext * s,int32_t * decoded,int subframe_size,int prev_subframe_size)386 static int decode_subframe(TAKDecContext *s, int32_t *decoded,
387 int subframe_size, int prev_subframe_size)
388 {
389 GetBitContext *gb = &s->gb;
390 int x, y, i, j, ret = 0;
391 int dshift, size, filter_quant, filter_order;
392 int tfilter[MAX_PREDICTORS];
393
394 if (!get_bits1(gb))
395 return decode_residues(s, decoded, subframe_size);
396
397 filter_order = predictor_sizes[get_bits(gb, 4)];
398
399 if (prev_subframe_size > 0 && get_bits1(gb)) {
400 if (filter_order > prev_subframe_size)
401 return AVERROR_INVALIDDATA;
402
403 decoded -= filter_order;
404 subframe_size += filter_order;
405
406 if (filter_order > subframe_size)
407 return AVERROR_INVALIDDATA;
408 } else {
409 int lpc_mode;
410
411 if (filter_order > subframe_size)
412 return AVERROR_INVALIDDATA;
413
414 lpc_mode = get_bits(gb, 2);
415 if (lpc_mode > 2)
416 return AVERROR_INVALIDDATA;
417
418 if ((ret = decode_residues(s, decoded, filter_order)) < 0)
419 return ret;
420
421 if (lpc_mode)
422 decode_lpc(decoded, lpc_mode, filter_order);
423 }
424
425 dshift = get_bits_esc4(gb);
426 size = get_bits1(gb) + 6;
427
428 filter_quant = 10;
429 if (get_bits1(gb)) {
430 filter_quant -= get_bits(gb, 3) + 1;
431 if (filter_quant < 3)
432 return AVERROR_INVALIDDATA;
433 }
434
435 if (get_bits_left(gb) < 2*10 + 2*size)
436 return AVERROR_INVALIDDATA;
437
438 s->predictors[0] = get_sbits(gb, 10);
439 s->predictors[1] = get_sbits(gb, 10);
440 s->predictors[2] = get_sbits(gb, size) * (1 << (10 - size));
441 s->predictors[3] = get_sbits(gb, size) * (1 << (10 - size));
442 if (filter_order > 4) {
443 int tmp = size - get_bits1(gb);
444
445 for (i = 4; i < filter_order; i++) {
446 if (!(i & 3))
447 x = tmp - get_bits(gb, 2);
448 s->predictors[i] = get_sbits(gb, x) * (1 << (10 - size));
449 }
450 }
451
452 tfilter[0] = s->predictors[0] * 64;
453 for (i = 1; i < filter_order; i++) {
454 uint32_t *p1 = &tfilter[0];
455 uint32_t *p2 = &tfilter[i - 1];
456
457 for (j = 0; j < (i + 1) / 2; j++) {
458 x = *p1 + ((int32_t)(s->predictors[i] * *p2 + 256) >> 9);
459 *p2 += (int32_t)(s->predictors[i] * *p1 + 256) >> 9;
460 *p1++ = x;
461 p2--;
462 }
463
464 tfilter[i] = s->predictors[i] * 64;
465 }
466
467 x = 1 << (32 - (15 - filter_quant));
468 y = 1 << ((15 - filter_quant) - 1);
469 for (i = 0, j = filter_order - 1; i < filter_order / 2; i++, j--) {
470 s->filter[j] = x - ((tfilter[i] + y) >> (15 - filter_quant));
471 s->filter[i] = x - ((tfilter[j] + y) >> (15 - filter_quant));
472 }
473
474 if ((ret = decode_residues(s, &decoded[filter_order],
475 subframe_size - filter_order)) < 0)
476 return ret;
477
478 for (i = 0; i < filter_order; i++)
479 s->residues[i] = *decoded++ >> dshift;
480
481 y = FF_ARRAY_ELEMS(s->residues) - filter_order;
482 x = subframe_size - filter_order;
483 while (x > 0) {
484 int tmp = FFMIN(y, x);
485
486 for (i = 0; i < tmp; i++) {
487 int v = 1 << (filter_quant - 1);
488
489 if (filter_order & -16)
490 v += (unsigned)s->adsp.scalarproduct_int16(&s->residues[i], s->filter,
491 filter_order & -16);
492 for (j = filter_order & -16; j < filter_order; j += 4) {
493 v += s->residues[i + j + 3] * (unsigned)s->filter[j + 3] +
494 s->residues[i + j + 2] * (unsigned)s->filter[j + 2] +
495 s->residues[i + j + 1] * (unsigned)s->filter[j + 1] +
496 s->residues[i + j ] * (unsigned)s->filter[j ];
497 }
498 v = (av_clip_intp2(v >> filter_quant, 13) * (1 << dshift)) - (unsigned)*decoded;
499 *decoded++ = v;
500 s->residues[filter_order + i] = v >> dshift;
501 }
502
503 x -= tmp;
504 if (x > 0)
505 memcpy(s->residues, &s->residues[y], 2 * filter_order);
506 }
507
508 emms_c();
509
510 return 0;
511 }
512
decode_channel(TAKDecContext * s,int chan)513 static int decode_channel(TAKDecContext *s, int chan)
514 {
515 AVCodecContext *avctx = s->avctx;
516 GetBitContext *gb = &s->gb;
517 int32_t *decoded = s->decoded[chan];
518 int left = s->nb_samples - 1;
519 int i = 0, ret, prev = 0;
520
521 s->sample_shift[chan] = get_bits_esc4(gb);
522 if (s->sample_shift[chan] >= avctx->bits_per_raw_sample)
523 return AVERROR_INVALIDDATA;
524
525 *decoded++ = get_sbits(gb, avctx->bits_per_raw_sample - s->sample_shift[chan]);
526 s->lpc_mode[chan] = get_bits(gb, 2);
527 s->nb_subframes = get_bits(gb, 3) + 1;
528
529 if (s->nb_subframes > 1) {
530 if (get_bits_left(gb) < (s->nb_subframes - 1) * 6)
531 return AVERROR_INVALIDDATA;
532
533 for (; i < s->nb_subframes - 1; i++) {
534 int v = get_bits(gb, 6);
535
536 s->subframe_len[i] = (v - prev) * s->subframe_scale;
537 if (s->subframe_len[i] <= 0)
538 return AVERROR_INVALIDDATA;
539
540 left -= s->subframe_len[i];
541 prev = v;
542 }
543
544 if (left <= 0)
545 return AVERROR_INVALIDDATA;
546 }
547 s->subframe_len[i] = left;
548
549 prev = 0;
550 for (i = 0; i < s->nb_subframes; i++) {
551 if ((ret = decode_subframe(s, decoded, s->subframe_len[i], prev)) < 0)
552 return ret;
553 decoded += s->subframe_len[i];
554 prev = s->subframe_len[i];
555 }
556
557 return 0;
558 }
559
decorrelate(TAKDecContext * s,int c1,int c2,int length)560 static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
561 {
562 GetBitContext *gb = &s->gb;
563 int32_t *p1 = s->decoded[c1] + (s->dmode > 5);
564 int32_t *p2 = s->decoded[c2] + (s->dmode > 5);
565 int32_t bp1 = p1[0];
566 int32_t bp2 = p2[0];
567 int i;
568 int dshift, dfactor;
569
570 length += s->dmode < 6;
571
572 switch (s->dmode) {
573 case 1: /* left/side */
574 s->tdsp.decorrelate_ls(p1, p2, length);
575 break;
576 case 2: /* side/right */
577 s->tdsp.decorrelate_sr(p1, p2, length);
578 break;
579 case 3: /* side/mid */
580 s->tdsp.decorrelate_sm(p1, p2, length);
581 break;
582 case 4: /* side/left with scale factor */
583 FFSWAP(int32_t*, p1, p2);
584 FFSWAP(int32_t, bp1, bp2);
585 case 5: /* side/right with scale factor */
586 dshift = get_bits_esc4(gb);
587 dfactor = get_sbits(gb, 10);
588 s->tdsp.decorrelate_sf(p1, p2, length, dshift, dfactor);
589 break;
590 case 6:
591 FFSWAP(int32_t*, p1, p2);
592 case 7: {
593 int length2, order_half, filter_order, dval1, dval2;
594 int tmp, x, code_size;
595
596 if (length < 256)
597 return AVERROR_INVALIDDATA;
598
599 dshift = get_bits_esc4(gb);
600 filter_order = 8 << get_bits1(gb);
601 dval1 = get_bits1(gb);
602 dval2 = get_bits1(gb);
603
604 for (i = 0; i < filter_order; i++) {
605 if (!(i & 3))
606 code_size = 14 - get_bits(gb, 3);
607 s->filter[i] = get_sbits(gb, code_size);
608 }
609
610 order_half = filter_order / 2;
611 length2 = length - (filter_order - 1);
612
613 /* decorrelate beginning samples */
614 if (dval1) {
615 for (i = 0; i < order_half; i++) {
616 int32_t a = p1[i];
617 int32_t b = p2[i];
618 p1[i] = a + b;
619 }
620 }
621
622 /* decorrelate ending samples */
623 if (dval2) {
624 for (i = length2 + order_half; i < length; i++) {
625 int32_t a = p1[i];
626 int32_t b = p2[i];
627 p1[i] = a + b;
628 }
629 }
630
631
632 for (i = 0; i < filter_order; i++)
633 s->residues[i] = *p2++ >> dshift;
634
635 p1 += order_half;
636 x = FF_ARRAY_ELEMS(s->residues) - filter_order;
637 for (; length2 > 0; length2 -= tmp) {
638 tmp = FFMIN(length2, x);
639
640 for (i = 0; i < tmp - (tmp == length2); i++)
641 s->residues[filter_order + i] = *p2++ >> dshift;
642
643 for (i = 0; i < tmp; i++) {
644 int v = 1 << 9;
645
646 if (filter_order == 16) {
647 v += s->adsp.scalarproduct_int16(&s->residues[i], s->filter,
648 filter_order);
649 } else {
650 v += s->residues[i + 7] * s->filter[7] +
651 s->residues[i + 6] * s->filter[6] +
652 s->residues[i + 5] * s->filter[5] +
653 s->residues[i + 4] * s->filter[4] +
654 s->residues[i + 3] * s->filter[3] +
655 s->residues[i + 2] * s->filter[2] +
656 s->residues[i + 1] * s->filter[1] +
657 s->residues[i ] * s->filter[0];
658 }
659
660 v = av_clip_intp2(v >> 10, 13) * (1U << dshift) - *p1;
661 *p1++ = v;
662 }
663
664 memmove(s->residues, &s->residues[tmp], 2 * filter_order);
665 }
666
667 emms_c();
668 break;
669 }
670 }
671
672 if (s->dmode > 0 && s->dmode < 6) {
673 p1[0] = bp1;
674 p2[0] = bp2;
675 }
676
677 return 0;
678 }
679
tak_decode_frame(AVCodecContext * avctx,AVFrame * frame,int * got_frame_ptr,AVPacket * pkt)680 static int tak_decode_frame(AVCodecContext *avctx, AVFrame *frame,
681 int *got_frame_ptr, AVPacket *pkt)
682 {
683 TAKDecContext *s = avctx->priv_data;
684 GetBitContext *gb = &s->gb;
685 int chan, i, ret, hsize;
686
687 if (pkt->size < TAK_MIN_FRAME_HEADER_BYTES)
688 return AVERROR_INVALIDDATA;
689
690 if ((ret = init_get_bits8(gb, pkt->data, pkt->size)) < 0)
691 return ret;
692
693 if ((ret = ff_tak_decode_frame_header(avctx, gb, &s->ti, 0)) < 0)
694 return ret;
695
696 hsize = get_bits_count(gb) / 8;
697 if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_COMPLIANT)) {
698 if (ff_tak_check_crc(pkt->data, hsize)) {
699 av_log(avctx, AV_LOG_ERROR, "CRC error\n");
700 if (avctx->err_recognition & AV_EF_EXPLODE)
701 return AVERROR_INVALIDDATA;
702 }
703 }
704
705 if (s->ti.codec != TAK_CODEC_MONO_STEREO &&
706 s->ti.codec != TAK_CODEC_MULTICHANNEL) {
707 avpriv_report_missing_feature(avctx, "TAK codec type %d", s->ti.codec);
708 return AVERROR_PATCHWELCOME;
709 }
710 if (s->ti.data_type) {
711 av_log(avctx, AV_LOG_ERROR,
712 "unsupported data type: %d\n", s->ti.data_type);
713 return AVERROR_INVALIDDATA;
714 }
715 if (s->ti.codec == TAK_CODEC_MONO_STEREO && s->ti.channels > 2) {
716 av_log(avctx, AV_LOG_ERROR,
717 "invalid number of channels: %d\n", s->ti.channels);
718 return AVERROR_INVALIDDATA;
719 }
720 if (s->ti.channels > 6) {
721 av_log(avctx, AV_LOG_ERROR,
722 "unsupported number of channels: %d\n", s->ti.channels);
723 return AVERROR_INVALIDDATA;
724 }
725
726 if (s->ti.frame_samples <= 0) {
727 av_log(avctx, AV_LOG_ERROR, "unsupported/invalid number of samples\n");
728 return AVERROR_INVALIDDATA;
729 }
730
731 avctx->bits_per_raw_sample = s->ti.bps;
732 if ((ret = set_bps_params(avctx)) < 0)
733 return ret;
734 if (s->ti.sample_rate != avctx->sample_rate) {
735 avctx->sample_rate = s->ti.sample_rate;
736 set_sample_rate_params(avctx);
737 }
738
739 av_channel_layout_uninit(&avctx->ch_layout);
740 if (s->ti.ch_layout) {
741 av_channel_layout_from_mask(&avctx->ch_layout, s->ti.ch_layout);
742 } else {
743 avctx->ch_layout.order = AV_CHANNEL_ORDER_UNSPEC;
744 avctx->ch_layout.nb_channels = s->ti.channels;
745 }
746
747 s->nb_samples = s->ti.last_frame_samples ? s->ti.last_frame_samples
748 : s->ti.frame_samples;
749
750 frame->nb_samples = s->nb_samples;
751 if ((ret = ff_thread_get_buffer(avctx, frame, 0)) < 0)
752 return ret;
753 ff_thread_finish_setup(avctx);
754
755 if (avctx->bits_per_raw_sample <= 16) {
756 int buf_size = av_samples_get_buffer_size(NULL, avctx->ch_layout.nb_channels,
757 s->nb_samples,
758 AV_SAMPLE_FMT_S32P, 0);
759 if (buf_size < 0)
760 return buf_size;
761 av_fast_malloc(&s->decode_buffer, &s->decode_buffer_size, buf_size);
762 if (!s->decode_buffer)
763 return AVERROR(ENOMEM);
764 ret = av_samples_fill_arrays((uint8_t **)s->decoded, NULL,
765 s->decode_buffer, avctx->ch_layout.nb_channels,
766 s->nb_samples, AV_SAMPLE_FMT_S32P, 0);
767 if (ret < 0)
768 return ret;
769 } else {
770 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++)
771 s->decoded[chan] = (int32_t *)frame->extended_data[chan];
772 }
773
774 if (s->nb_samples < 16) {
775 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++) {
776 int32_t *decoded = s->decoded[chan];
777 for (i = 0; i < s->nb_samples; i++)
778 decoded[i] = get_sbits(gb, avctx->bits_per_raw_sample);
779 }
780 } else {
781 if (s->ti.codec == TAK_CODEC_MONO_STEREO) {
782 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++)
783 if (ret = decode_channel(s, chan))
784 return ret;
785
786 if (avctx->ch_layout.nb_channels == 2) {
787 s->nb_subframes = get_bits(gb, 1) + 1;
788 if (s->nb_subframes > 1) {
789 s->subframe_len[1] = get_bits(gb, 6);
790 }
791
792 s->dmode = get_bits(gb, 3);
793 if (ret = decorrelate(s, 0, 1, s->nb_samples - 1))
794 return ret;
795 }
796 } else if (s->ti.codec == TAK_CODEC_MULTICHANNEL) {
797 if (get_bits1(gb)) {
798 int ch_mask = 0;
799
800 chan = get_bits(gb, 4) + 1;
801 if (chan > avctx->ch_layout.nb_channels)
802 return AVERROR_INVALIDDATA;
803
804 for (i = 0; i < chan; i++) {
805 int nbit = get_bits(gb, 4);
806
807 if (nbit >= avctx->ch_layout.nb_channels)
808 return AVERROR_INVALIDDATA;
809
810 if (ch_mask & 1 << nbit)
811 return AVERROR_INVALIDDATA;
812
813 s->mcdparams[i].present = get_bits1(gb);
814 if (s->mcdparams[i].present) {
815 s->mcdparams[i].index = get_bits(gb, 2);
816 s->mcdparams[i].chan2 = get_bits(gb, 4);
817 if (s->mcdparams[i].chan2 >= avctx->ch_layout.nb_channels) {
818 av_log(avctx, AV_LOG_ERROR,
819 "invalid channel 2 (%d) for %d channel(s)\n",
820 s->mcdparams[i].chan2, avctx->ch_layout.nb_channels);
821 return AVERROR_INVALIDDATA;
822 }
823 if (s->mcdparams[i].index == 1) {
824 if ((nbit == s->mcdparams[i].chan2) ||
825 (ch_mask & 1 << s->mcdparams[i].chan2))
826 return AVERROR_INVALIDDATA;
827
828 ch_mask |= 1 << s->mcdparams[i].chan2;
829 } else if (!(ch_mask & 1 << s->mcdparams[i].chan2)) {
830 return AVERROR_INVALIDDATA;
831 }
832 }
833 s->mcdparams[i].chan1 = nbit;
834
835 ch_mask |= 1 << nbit;
836 }
837 } else {
838 chan = avctx->ch_layout.nb_channels;
839 for (i = 0; i < chan; i++) {
840 s->mcdparams[i].present = 0;
841 s->mcdparams[i].chan1 = i;
842 }
843 }
844
845 for (i = 0; i < chan; i++) {
846 if (s->mcdparams[i].present && s->mcdparams[i].index == 1)
847 if (ret = decode_channel(s, s->mcdparams[i].chan2))
848 return ret;
849
850 if (ret = decode_channel(s, s->mcdparams[i].chan1))
851 return ret;
852
853 if (s->mcdparams[i].present) {
854 s->dmode = mc_dmodes[s->mcdparams[i].index];
855 if (ret = decorrelate(s,
856 s->mcdparams[i].chan2,
857 s->mcdparams[i].chan1,
858 s->nb_samples - 1))
859 return ret;
860 }
861 }
862 }
863
864 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++) {
865 int32_t *decoded = s->decoded[chan];
866
867 if (s->lpc_mode[chan])
868 decode_lpc(decoded, s->lpc_mode[chan], s->nb_samples);
869
870 if (s->sample_shift[chan] > 0)
871 for (i = 0; i < s->nb_samples; i++)
872 decoded[i] *= 1U << s->sample_shift[chan];
873 }
874 }
875
876 align_get_bits(gb);
877 skip_bits(gb, 24);
878 if (get_bits_left(gb) < 0)
879 av_log(avctx, AV_LOG_DEBUG, "overread\n");
880 else if (get_bits_left(gb) > 0)
881 av_log(avctx, AV_LOG_DEBUG, "underread\n");
882
883 if (avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_COMPLIANT)) {
884 if (ff_tak_check_crc(pkt->data + hsize,
885 get_bits_count(gb) / 8 - hsize)) {
886 av_log(avctx, AV_LOG_ERROR, "CRC error\n");
887 if (avctx->err_recognition & AV_EF_EXPLODE)
888 return AVERROR_INVALIDDATA;
889 }
890 }
891
892 /* convert to output buffer */
893 switch (avctx->sample_fmt) {
894 case AV_SAMPLE_FMT_U8P:
895 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++) {
896 uint8_t *samples = (uint8_t *)frame->extended_data[chan];
897 int32_t *decoded = s->decoded[chan];
898 for (i = 0; i < s->nb_samples; i++)
899 samples[i] = decoded[i] + 0x80U;
900 }
901 break;
902 case AV_SAMPLE_FMT_S16P:
903 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++) {
904 int16_t *samples = (int16_t *)frame->extended_data[chan];
905 int32_t *decoded = s->decoded[chan];
906 for (i = 0; i < s->nb_samples; i++)
907 samples[i] = decoded[i];
908 }
909 break;
910 case AV_SAMPLE_FMT_S32P:
911 for (chan = 0; chan < avctx->ch_layout.nb_channels; chan++) {
912 int32_t *samples = (int32_t *)frame->extended_data[chan];
913 for (i = 0; i < s->nb_samples; i++)
914 samples[i] *= 1U << 8;
915 }
916 break;
917 }
918
919 *got_frame_ptr = 1;
920
921 return pkt->size;
922 }
923
924 #if HAVE_THREADS
update_thread_context(AVCodecContext * dst,const AVCodecContext * src)925 static int update_thread_context(AVCodecContext *dst,
926 const AVCodecContext *src)
927 {
928 TAKDecContext *tsrc = src->priv_data;
929 TAKDecContext *tdst = dst->priv_data;
930
931 if (dst == src)
932 return 0;
933 memcpy(&tdst->ti, &tsrc->ti, sizeof(TAKStreamInfo));
934 return 0;
935 }
936 #endif
937
tak_decode_close(AVCodecContext * avctx)938 static av_cold int tak_decode_close(AVCodecContext *avctx)
939 {
940 TAKDecContext *s = avctx->priv_data;
941
942 av_freep(&s->decode_buffer);
943
944 return 0;
945 }
946
947 const FFCodec ff_tak_decoder = {
948 .p.name = "tak",
949 .p.long_name = NULL_IF_CONFIG_SMALL("TAK (Tom's lossless Audio Kompressor)"),
950 .p.type = AVMEDIA_TYPE_AUDIO,
951 .p.id = AV_CODEC_ID_TAK,
952 .priv_data_size = sizeof(TAKDecContext),
953 .init = tak_decode_init,
954 .close = tak_decode_close,
955 FF_CODEC_DECODE_CB(tak_decode_frame),
956 .update_thread_context = ONLY_IF_THREADS_ENABLED(update_thread_context),
957 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_CHANNEL_CONF,
958 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
959 AV_SAMPLE_FMT_S16P,
960 AV_SAMPLE_FMT_S32P,
961 AV_SAMPLE_FMT_NONE },
962 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
963 };
964