takdec: fix initialisation of LOCAL_ALIGNED array
[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/samplefmt.h"
29 #include "tak.h"
30 #include "avcodec.h"
31 #include "dsputil.h"
32 #include "internal.h"
33 #include "unary.h"
34
35 #define MAX_SUBFRAMES 8 // max number of subframes per channel
36 #define MAX_PREDICTORS 256
37
38 typedef struct MCDParam {
39 int8_t present; // decorrelation parameter availability for this channel
40 int8_t index; // index into array of decorrelation types
41 int8_t chan1;
42 int8_t chan2;
43 } MCDParam;
44
45 typedef struct TAKDecContext {
46 AVCodecContext *avctx; // parent AVCodecContext
47 AVFrame frame; // AVFrame for decoded output
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 avcodec_get_frame_defaults(&s->frame);
179 avctx->coded_frame = &s->frame;
180
181 set_sample_rate_params(avctx);
182
183 return set_bps_params(avctx);
184 }
185
186 static void decode_lpc(int32_t *coeffs, int mode, int length)
187 {
188 int i;
189
190 if (length < 2)
191 return;
192
193 if (mode == 1) {
194 int a1 = *coeffs++;
195 for (i = 0; i < length - 1 >> 1; i++) {
196 *coeffs += a1;
197 coeffs[1] += *coeffs;
198 a1 = coeffs[1];
199 coeffs += 2;
200 }
201 if (length - 1 & 1)
202 *coeffs += a1;
203 } else if (mode == 2) {
204 int a1 = coeffs[1];
205 int a2 = a1 + *coeffs;
206 coeffs[1] = a2;
207 if (length > 2) {
208 coeffs += 2;
209 for (i = 0; i < length - 2 >> 1; i++) {
210 int a3 = *coeffs + a1;
211 int a4 = a3 + a2;
212 *coeffs = a4;
213 a1 = coeffs[1] + a3;
214 a2 = a1 + a4;
215 coeffs[1] = a2;
216 coeffs += 2;
217 }
218 if (length & 1)
219 *coeffs += a1 + a2;
220 }
221 } else if (mode == 3) {
222 int a1 = coeffs[1];
223 int a2 = a1 + *coeffs;
224 coeffs[1] = a2;
225 if (length > 2) {
226 int a3 = coeffs[2];
227 int a4 = a3 + a1;
228 int a5 = a4 + a2;
229 coeffs += 3;
230 for (i = 0; i < length - 3; i++) {
231 a3 += *coeffs;
232 a4 += a3;
233 a5 += a4;
234 *coeffs = a5;
235 coeffs++;
236 }
237 }
238 }
239 }
240
241 static int decode_segment(GetBitContext *gb, int mode, int32_t *decoded,
242 int len)
243 {
244 struct CParam code;
245 int i;
246
247 if (!mode) {
248 memset(decoded, 0, len * sizeof(*decoded));
249 return 0;
250 }
251
252 if (mode > FF_ARRAY_ELEMS(xcodes))
253 return AVERROR_INVALIDDATA;
254 code = xcodes[mode - 1];
255
256 for (i = 0; i < len; i++) {
257 int x = get_bits_long(gb, code.init);
258 if (x >= code.escape && get_bits1(gb)) {
259 x |= 1 << code.init;
260 if (x >= code.aescape) {
261 int scale = get_unary(gb, 1, 9);
262 if (scale == 9) {
263 int scale_bits = get_bits(gb, 3);
264 if (scale_bits > 0) {
265 if (scale_bits == 7) {
266 scale_bits += get_bits(gb, 5);
267 if (scale_bits > 29)
268 return AVERROR_INVALIDDATA;
269 }
270 scale = get_bits_long(gb, scale_bits) + 1;
271 x += code.scale * scale;
272 }
273 x += code.bias;
274 } else
275 x += code.scale * scale - code.escape;
276 } else
277 x -= code.escape;
278 }
279 decoded[i] = (x >> 1) ^ -(x & 1);
280 }
281
282 return 0;
283 }
284
285 static int decode_residues(TAKDecContext *s, int32_t *decoded, int length)
286 {
287 GetBitContext *gb = &s->gb;
288 int i, mode, ret;
289
290 if (length > s->nb_samples)
291 return AVERROR_INVALIDDATA;
292
293 if (get_bits1(gb)) {
294 int wlength, rval;
295 int coding_mode[128];
296
297 wlength = length / s->uval;
298
299 rval = length - (wlength * s->uval);
300
301 if (rval < s->uval / 2)
302 rval += s->uval;
303 else
304 wlength++;
305
306 if (wlength <= 1 || wlength > 128)
307 return AVERROR_INVALIDDATA;
308
309 coding_mode[0] = mode = get_bits(gb, 6);
310
311 for (i = 1; i < wlength; i++) {
312 int c = get_unary(gb, 1, 6);
313
314 switch (c) {
315 case 6:
316 mode = get_bits(gb, 6);
317 break;
318 case 5:
319 case 4:
320 case 3: {
321 /* mode += sign ? (1 - c) : (c - 1) */
322 int sign = get_bits1(gb);
323 mode += (-sign ^ (c - 1)) + sign;
324 break;
325 }
326 case 2:
327 mode++;
328 break;
329 case 1:
330 mode--;
331 break;
332 }
333 coding_mode[i] = mode;
334 }
335
336 i = 0;
337 while (i < wlength) {
338 int len = 0;
339
340 mode = coding_mode[i];
341 do {
342 if (i >= wlength - 1)
343 len += rval;
344 else
345 len += s->uval;
346 i++;
347
348 if (i == wlength)
349 break;
350 } while (coding_mode[i] == mode);
351
352 if ((ret = decode_segment(gb, mode, decoded, len)) < 0)
353 return ret;
354 decoded += len;
355 }
356 } else {
357 mode = get_bits(gb, 6);
358 if ((ret = decode_segment(gb, mode, decoded, length)) < 0)
359 return ret;
360 }
361
362 return 0;
363 }
364
365 static int get_bits_esc4(GetBitContext *gb)
366 {
367 if (get_bits1(gb))
368 return get_bits(gb, 4) + 1;
369 else
370 return 0;
371 }
372
373 static void decode_filter_coeffs(TAKDecContext *s, int filter_order, int size,
374 int filter_quant, int16_t *filter)
375 {
376 GetBitContext *gb = &s->gb;
377 int i, j, a, b;
378 int filter_tmp[MAX_PREDICTORS];
379 int16_t predictors[MAX_PREDICTORS];
380
381 predictors[0] = get_sbits(gb, 10);
382 predictors[1] = get_sbits(gb, 10);
383 predictors[2] = get_sbits(gb, size) << (10 - size);
384 predictors[3] = get_sbits(gb, size) << (10 - size);
385 if (filter_order > 4) {
386 int av_uninit(code_size);
387 int code_size_base = size - get_bits1(gb);
388
389 for (i = 4; i < filter_order; i++) {
390 if (!(i & 3))
391 code_size = code_size_base - get_bits(gb, 2);
392 predictors[i] = get_sbits(gb, code_size) << (10 - size);
393 }
394 }
395
396 filter_tmp[0] = predictors[0] << 6;
397 for (i = 1; i < filter_order; i++) {
398 int *p1 = &filter_tmp[0];
399 int *p2 = &filter_tmp[i - 1];
400
401 for (j = 0; j < (i + 1) / 2; j++) {
402 int tmp = *p1 + (predictors[i] * *p2 + 256 >> 9);
403 *p2 = *p2 + (predictors[i] * *p1 + 256 >> 9);
404 *p1 = tmp;
405 p1++;
406 p2--;
407 }
408
409 filter_tmp[i] = predictors[i] << 6;
410 }
411
412 a = 1 << (32 - (15 - filter_quant));
413 b = 1 << ((15 - filter_quant) - 1);
414 for (i = 0, j = filter_order - 1; i < filter_order / 2; i++, j--) {
415 filter[j] = a - ((filter_tmp[i] + b) >> (15 - filter_quant));
416 filter[i] = a - ((filter_tmp[j] + b) >> (15 - filter_quant));
417 }
418 }
419
420 static int decode_subframe(TAKDecContext *s, int32_t *decoded,
421 int subframe_size, int prev_subframe_size)
422 {
423 LOCAL_ALIGNED_16(int16_t, filter, [MAX_PREDICTORS]);
424 GetBitContext *gb = &s->gb;
425 int i, ret;
426 int dshift, size, filter_quant, filter_order;
427
428 memset(filter, 0, MAX_PREDICTORS * sizeof(*filter));
429
430 if (!get_bits1(gb))
431 return decode_residues(s, decoded, subframe_size);
432
433 filter_order = predictor_sizes[get_bits(gb, 4)];
434
435 if (prev_subframe_size > 0 && get_bits1(gb)) {
436 if (filter_order > prev_subframe_size)
437 return AVERROR_INVALIDDATA;
438
439 decoded -= filter_order;
440 subframe_size += filter_order;
441
442 if (filter_order > subframe_size)
443 return AVERROR_INVALIDDATA;
444 } else {
445 int lpc_mode;
446
447 if (filter_order > subframe_size)
448 return AVERROR_INVALIDDATA;
449
450 lpc_mode = get_bits(gb, 2);
451 if (lpc_mode > 2)
452 return AVERROR_INVALIDDATA;
453
454 if ((ret = decode_residues(s, decoded, filter_order)) < 0)
455 return ret;
456
457 if (lpc_mode)
458 decode_lpc(decoded, lpc_mode, filter_order);
459 }
460
461 dshift = get_bits_esc4(gb);
462 size = get_bits1(gb) + 6;
463
464 filter_quant = 10;
465 if (get_bits1(gb)) {
466 filter_quant -= get_bits(gb, 3) + 1;
467 if (filter_quant < 3)
468 return AVERROR_INVALIDDATA;
469 }
470
471 decode_filter_coeffs(s, filter_order, size, filter_quant, filter);
472
473 if ((ret = decode_residues(s, &decoded[filter_order],
474 subframe_size - filter_order)) < 0)
475 return ret;
476
477 av_fast_malloc(&s->residues, &s->residues_buf_size,
478 FFALIGN(subframe_size + 16, 16) * sizeof(*s->residues));
479 if (!s->residues)
480 return AVERROR(ENOMEM);
481 memset(s->residues, 0, s->residues_buf_size);
482
483 for (i = 0; i < filter_order; i++)
484 s->residues[i] = *decoded++ >> dshift;
485
486 for (i = 0; i < subframe_size - filter_order; i++) {
487 int v = 1 << (filter_quant - 1);
488
489 v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
490 FFALIGN(filter_order, 16));
491
492 v = (av_clip(v >> filter_quant, -8192, 8191) << dshift) - *decoded;
493 *decoded++ = v;
494 s->residues[filter_order + i] = v >> dshift;
495 }
496
497 emms_c();
498
499 return 0;
500 }
501
502 static int decode_channel(TAKDecContext *s, int chan)
503 {
504 AVCodecContext *avctx = s->avctx;
505 GetBitContext *gb = &s->gb;
506 int32_t *decoded = s->decoded[chan];
507 int left = s->nb_samples - 1;
508 int i, prev, ret, nb_subframes;
509 int subframe_len[MAX_SUBFRAMES];
510
511 s->sample_shift[chan] = get_bits_esc4(gb);
512 if (s->sample_shift[chan] >= avctx->bits_per_coded_sample)
513 return AVERROR_INVALIDDATA;
514
515 /* NOTE: TAK 2.2.0 appears to set the sample value to 0 if
516 * bits_per_coded_sample - sample_shift is 1, but this produces
517 * non-bit-exact output. Reading the 1 bit using get_sbits() instead
518 * of skipping it produces bit-exact output. This has been reported
519 * to the TAK author. */
520 *decoded++ = get_sbits(gb,
521 avctx->bits_per_coded_sample -
522 s->sample_shift[chan]);
523 s->lpc_mode[chan] = get_bits(gb, 2);
524 nb_subframes = get_bits(gb, 3) + 1;
525
526 i = 0;
527 if (nb_subframes > 1) {
528 if (get_bits_left(gb) < (nb_subframes - 1) * 6)
529 return AVERROR_INVALIDDATA;
530
531 prev = 0;
532 for (; i < nb_subframes - 1; i++) {
533 int subframe_end = get_bits(gb, 6) * s->subframe_scale;
534 if (subframe_end <= prev)
535 return AVERROR_INVALIDDATA;
536 subframe_len[i] = subframe_end - prev;
537 left -= subframe_len[i];
538 prev = subframe_end;
539 }
540
541 if (left <= 0)
542 return AVERROR_INVALIDDATA;
543 }
544 subframe_len[i] = left;
545
546 prev = 0;
547 for (i = 0; i < nb_subframes; i++) {
548 if ((ret = decode_subframe(s, decoded, subframe_len[i], prev)) < 0)
549 return ret;
550 decoded += subframe_len[i];
551 prev = subframe_len[i];
552 }
553
554 return 0;
555 }
556
557 static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
558 {
559 GetBitContext *gb = &s->gb;
560 int32_t *p1 = s->decoded[c1] + 1;
561 int32_t *p2 = s->decoded[c2] + 1;
562 int i;
563 int dshift, dfactor;
564
565 switch (s->dmode) {
566 case 1: /* left/side */
567 for (i = 0; i < length; i++) {
568 int32_t a = p1[i];
569 int32_t b = p2[i];
570 p2[i] = a + b;
571 }
572 break;
573 case 2: /* side/right */
574 for (i = 0; i < length; i++) {
575 int32_t a = p1[i];
576 int32_t b = p2[i];
577 p1[i] = b - a;
578 }
579 break;
580 case 3: /* side/mid */
581 for (i = 0; i < length; i++) {
582 int32_t a = p1[i];
583 int32_t b = p2[i];
584 a -= b >> 1;
585 p1[i] = a;
586 p2[i] = a + b;
587 }
588 break;
589 case 4: /* side/left with scale factor */
590 FFSWAP(int32_t*, p1, p2);
591 case 5: /* side/right with scale factor */
592 dshift = get_bits_esc4(gb);
593 dfactor = get_sbits(gb, 10);
594 for (i = 0; i < length; i++) {
595 int32_t a = p1[i];
596 int32_t b = p2[i];
597 b = dfactor * (b >> dshift) + 128 >> 8 << dshift;
598 p1[i] = b - a;
599 }
600 break;
601 case 6:
602 FFSWAP(int32_t*, p1, p2);
603 case 7: {
604 LOCAL_ALIGNED_16(int16_t, filter, [MAX_PREDICTORS]);
605 int length2, order_half, filter_order, dval1, dval2;
606 int av_uninit(code_size);
607
608 memset(filter, 0, MAX_PREDICTORS * sizeof(*filter));
609
610 if (length < 256)
611 return AVERROR_INVALIDDATA;
612
613 dshift = get_bits_esc4(gb);
614 filter_order = 8 << get_bits1(gb);
615 dval1 = get_bits1(gb);
616 dval2 = get_bits1(gb);
617
618 for (i = 0; i < filter_order; i++) {
619 if (!(i & 3))
620 code_size = 14 - get_bits(gb, 3);
621 filter[i] = get_sbits(gb, code_size);
622 }
623
624 order_half = filter_order / 2;
625 length2 = length - (filter_order - 1);
626
627 /* decorrelate beginning samples */
628 if (dval1) {
629 for (i = 0; i < order_half; i++) {
630 int32_t a = p1[i];
631 int32_t b = p2[i];
632 p1[i] = a + b;
633 }
634 }
635
636 /* decorrelate ending samples */
637 if (dval2) {
638 for (i = length2 + order_half; i < length; i++) {
639 int32_t a = p1[i];
640 int32_t b = p2[i];
641 p1[i] = a + b;
642 }
643 }
644
645 av_fast_malloc(&s->residues, &s->residues_buf_size,
646 FFALIGN(length + 16, 16) * sizeof(*s->residues));
647 if (!s->residues)
648 return AVERROR(ENOMEM);
649 memset(s->residues, 0, s->residues_buf_size);
650
651 for (i = 0; i < length; i++)
652 s->residues[i] = p2[i] >> dshift;
653
654 p1 += order_half;
655
656 for (i = 0; i < length2; i++) {
657 int v = 1 << 9;
658
659 v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
660 FFALIGN(filter_order, 16));
661
662 p1[i] = (av_clip(v >> 10, -8192, 8191) << dshift) - p1[i];
663 }
664
665 emms_c();
666 break;
667 }
668 }
669
670 return 0;
671 }
672
673 static int tak_decode_frame(AVCodecContext *avctx, void *data,
674 int *got_frame_ptr, AVPacket *pkt)
675 {
676 TAKDecContext *s = avctx->priv_data;
677 GetBitContext *gb = &s->gb;
678 int chan, i, ret, hsize;
679
680 if (pkt->size < TAK_MIN_FRAME_HEADER_BYTES)
681 return AVERROR_INVALIDDATA;
682
683 init_get_bits(gb, pkt->data, pkt->size * 8);
684
685 if ((ret = ff_tak_decode_frame_header(avctx, gb, &s->ti, 0)) < 0)
686 return ret;
687
688 if (s->ti.flags & TAK_FRAME_FLAG_HAS_METADATA) {
689 av_log_missing_feature(avctx, "frame metadata", 1);
690 return AVERROR_PATCHWELCOME;
691 }
692
693 hsize = get_bits_count(gb) / 8;
694 if (avctx->err_recognition & AV_EF_CRCCHECK) {
695 if (ff_tak_check_crc(pkt->data, hsize)) {
696 av_log(avctx, AV_LOG_ERROR, "CRC error\n");
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 s->frame.nb_samples = s->nb_samples;
744 if ((ret = ff_get_buffer(avctx, &s->frame)) < 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 *)s->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 return AVERROR_INVALIDDATA;
872 }
873 }
874
875 /* convert to output buffer */
876 switch (avctx->sample_fmt) {
877 case AV_SAMPLE_FMT_U8P:
878 for (chan = 0; chan < avctx->channels; chan++) {
879 uint8_t *samples = (uint8_t *)s->frame.extended_data[chan];
880 int32_t *decoded = s->decoded[chan];
881 for (i = 0; i < s->nb_samples; i++)
882 samples[i] = decoded[i] + 0x80;
883 }
884 break;
885 case AV_SAMPLE_FMT_S16P:
886 for (chan = 0; chan < avctx->channels; chan++) {
887 int16_t *samples = (int16_t *)s->frame.extended_data[chan];
888 int32_t *decoded = s->decoded[chan];
889 for (i = 0; i < s->nb_samples; i++)
890 samples[i] = decoded[i];
891 }
892 break;
893 case AV_SAMPLE_FMT_S32P:
894 for (chan = 0; chan < avctx->channels; chan++) {
895 int32_t *samples = (int32_t *)s->frame.extended_data[chan];
896 for (i = 0; i < s->nb_samples; i++)
897 samples[i] <<= 8;
898 }
899 break;
900 }
901
902 *got_frame_ptr = 1;
903 *(AVFrame *)data = s->frame;
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 .type = AVMEDIA_TYPE_AUDIO,
921 .id = AV_CODEC_ID_TAK,
922 .priv_data_size = sizeof(TAKDecContext),
923 .init = tak_decode_init,
924 .init_static_data = tak_init_static_data,
925 .close = tak_decode_close,
926 .decode = tak_decode_frame,
927 .capabilities = CODEC_CAP_DR1,
928 .long_name = NULL_IF_CONFIG_SMALL("TAK (Tom's lossless Audio Kompressor)"),
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 };