lavc: G.723.1 encoder
[libav.git] / libavcodec / atrac3.c
1 /*
2 * ATRAC3 compatible decoder
3 * Copyright (c) 2006-2008 Maxim Poliakovski
4 * Copyright (c) 2006-2008 Benjamin Larsson
5 *
6 * This file is part of Libav.
7 *
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 /**
24 * @file
25 * ATRAC3 compatible decoder.
26 * This decoder handles Sony's ATRAC3 data.
27 *
28 * Container formats used to store ATRAC3 data:
29 * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3).
30 *
31 * To use this decoder, a calling application must supply the extradata
32 * bytes provided in the containers above.
33 */
34
35 #include <math.h>
36 #include <stddef.h>
37 #include <stdio.h>
38
39 #include "libavutil/attributes.h"
40 #include "libavutil/float_dsp.h"
41 #include "avcodec.h"
42 #include "bytestream.h"
43 #include "fft.h"
44 #include "get_bits.h"
45 #include "internal.h"
46
47 #include "atrac.h"
48 #include "atrac3data.h"
49
50 #define JOINT_STEREO 0x12
51 #define STEREO 0x2
52
53 #define SAMPLES_PER_FRAME 1024
54 #define MDCT_SIZE 512
55
56 typedef struct GainBlock {
57 AtracGainInfo g_block[4];
58 } GainBlock;
59
60 typedef struct TonalComponent {
61 int pos;
62 int num_coefs;
63 float coef[8];
64 } TonalComponent;
65
66 typedef struct ChannelUnit {
67 int bands_coded;
68 int num_components;
69 float prev_frame[SAMPLES_PER_FRAME];
70 int gc_blk_switch;
71 TonalComponent components[64];
72 GainBlock gain_block[2];
73
74 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
75 DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
76
77 float delay_buf1[46]; ///<qmf delay buffers
78 float delay_buf2[46];
79 float delay_buf3[46];
80 } ChannelUnit;
81
82 typedef struct ATRAC3Context {
83 GetBitContext gb;
84 //@{
85 /** stream data */
86 int coding_mode;
87
88 ChannelUnit *units;
89 //@}
90 //@{
91 /** joint-stereo related variables */
92 int matrix_coeff_index_prev[4];
93 int matrix_coeff_index_now[4];
94 int matrix_coeff_index_next[4];
95 int weighting_delay[6];
96 //@}
97 //@{
98 /** data buffers */
99 uint8_t *decoded_bytes_buffer;
100 float temp_buf[1070];
101 //@}
102 //@{
103 /** extradata */
104 int scrambled_stream;
105 //@}
106
107 AtracGCContext gainc_ctx;
108 FFTContext mdct_ctx;
109 AVFloatDSPContext fdsp;
110 } ATRAC3Context;
111
112 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
113 static VLC_TYPE atrac3_vlc_table[4096][2];
114 static VLC spectral_coeff_tab[7];
115
116 /**
117 * Regular 512 points IMDCT without overlapping, with the exception of the
118 * swapping of odd bands caused by the reverse spectra of the QMF.
119 *
120 * @param odd_band 1 if the band is an odd band
121 */
122 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
123 {
124 int i;
125
126 if (odd_band) {
127 /**
128 * Reverse the odd bands before IMDCT, this is an effect of the QMF
129 * transform or it gives better compression to do it this way.
130 * FIXME: It should be possible to handle this in imdct_calc
131 * for that to happen a modification of the prerotation step of
132 * all SIMD code and C code is needed.
133 * Or fix the functions before so they generate a pre reversed spectrum.
134 */
135 for (i = 0; i < 128; i++)
136 FFSWAP(float, input[i], input[255 - i]);
137 }
138
139 q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
140
141 /* Perform windowing on the output. */
142 q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE);
143 }
144
145 /*
146 * indata descrambling, only used for data coming from the rm container
147 */
148 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
149 {
150 int i, off;
151 uint32_t c;
152 const uint32_t *buf;
153 uint32_t *output = (uint32_t *)out;
154
155 off = (intptr_t)input & 3;
156 buf = (const uint32_t *)(input - off);
157 if (off)
158 c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
159 else
160 c = av_be2ne32(0x537F6103U);
161 bytes += 3 + off;
162 for (i = 0; i < bytes / 4; i++)
163 output[i] = c ^ buf[i];
164
165 if (off)
166 avpriv_request_sample(NULL, "Offset of %d", off);
167
168 return off;
169 }
170
171 static av_cold void init_imdct_window(void)
172 {
173 int i, j;
174
175 /* generate the mdct window, for details see
176 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
177 for (i = 0, j = 255; i < 128; i++, j--) {
178 float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
179 float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
180 float w = 0.5 * (wi * wi + wj * wj);
181 mdct_window[i] = mdct_window[511 - i] = wi / w;
182 mdct_window[j] = mdct_window[511 - j] = wj / w;
183 }
184 }
185
186 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
187 {
188 ATRAC3Context *q = avctx->priv_data;
189
190 av_free(q->units);
191 av_free(q->decoded_bytes_buffer);
192
193 ff_mdct_end(&q->mdct_ctx);
194
195 return 0;
196 }
197
198 /**
199 * Mantissa decoding
200 *
201 * @param selector which table the output values are coded with
202 * @param coding_flag constant length coding or variable length coding
203 * @param mantissas mantissa output table
204 * @param num_codes number of values to get
205 */
206 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
207 int coding_flag, int *mantissas,
208 int num_codes)
209 {
210 int i, code, huff_symb;
211
212 if (selector == 1)
213 num_codes /= 2;
214
215 if (coding_flag != 0) {
216 /* constant length coding (CLC) */
217 int num_bits = clc_length_tab[selector];
218
219 if (selector > 1) {
220 for (i = 0; i < num_codes; i++) {
221 if (num_bits)
222 code = get_sbits(gb, num_bits);
223 else
224 code = 0;
225 mantissas[i] = code;
226 }
227 } else {
228 for (i = 0; i < num_codes; i++) {
229 if (num_bits)
230 code = get_bits(gb, num_bits); // num_bits is always 4 in this case
231 else
232 code = 0;
233 mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
234 mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
235 }
236 }
237 } else {
238 /* variable length coding (VLC) */
239 if (selector != 1) {
240 for (i = 0; i < num_codes; i++) {
241 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
242 spectral_coeff_tab[selector-1].bits, 3);
243 huff_symb += 1;
244 code = huff_symb >> 1;
245 if (huff_symb & 1)
246 code = -code;
247 mantissas[i] = code;
248 }
249 } else {
250 for (i = 0; i < num_codes; i++) {
251 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
252 spectral_coeff_tab[selector - 1].bits, 3);
253 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
254 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
255 }
256 }
257 }
258 }
259
260 /**
261 * Restore the quantized band spectrum coefficients
262 *
263 * @return subband count, fix for broken specification/files
264 */
265 static int decode_spectrum(GetBitContext *gb, float *output)
266 {
267 int num_subbands, coding_mode, i, j, first, last, subband_size;
268 int subband_vlc_index[32], sf_index[32];
269 int mantissas[128];
270 float scale_factor;
271
272 num_subbands = get_bits(gb, 5); // number of coded subbands
273 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
274
275 /* get the VLC selector table for the subbands, 0 means not coded */
276 for (i = 0; i <= num_subbands; i++)
277 subband_vlc_index[i] = get_bits(gb, 3);
278
279 /* read the scale factor indexes from the stream */
280 for (i = 0; i <= num_subbands; i++) {
281 if (subband_vlc_index[i] != 0)
282 sf_index[i] = get_bits(gb, 6);
283 }
284
285 for (i = 0; i <= num_subbands; i++) {
286 first = subband_tab[i ];
287 last = subband_tab[i + 1];
288
289 subband_size = last - first;
290
291 if (subband_vlc_index[i] != 0) {
292 /* decode spectral coefficients for this subband */
293 /* TODO: This can be done faster is several blocks share the
294 * same VLC selector (subband_vlc_index) */
295 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
296 mantissas, subband_size);
297
298 /* decode the scale factor for this subband */
299 scale_factor = ff_atrac_sf_table[sf_index[i]] *
300 inv_max_quant[subband_vlc_index[i]];
301
302 /* inverse quantize the coefficients */
303 for (j = 0; first < last; first++, j++)
304 output[first] = mantissas[j] * scale_factor;
305 } else {
306 /* this subband was not coded, so zero the entire subband */
307 memset(output + first, 0, subband_size * sizeof(*output));
308 }
309 }
310
311 /* clear the subbands that were not coded */
312 first = subband_tab[i];
313 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
314 return num_subbands;
315 }
316
317 /**
318 * Restore the quantized tonal components
319 *
320 * @param components tonal components
321 * @param num_bands number of coded bands
322 */
323 static int decode_tonal_components(GetBitContext *gb,
324 TonalComponent *components, int num_bands)
325 {
326 int i, b, c, m;
327 int nb_components, coding_mode_selector, coding_mode;
328 int band_flags[4], mantissa[8];
329 int component_count = 0;
330
331 nb_components = get_bits(gb, 5);
332
333 /* no tonal components */
334 if (nb_components == 0)
335 return 0;
336
337 coding_mode_selector = get_bits(gb, 2);
338 if (coding_mode_selector == 2)
339 return AVERROR_INVALIDDATA;
340
341 coding_mode = coding_mode_selector & 1;
342
343 for (i = 0; i < nb_components; i++) {
344 int coded_values_per_component, quant_step_index;
345
346 for (b = 0; b <= num_bands; b++)
347 band_flags[b] = get_bits1(gb);
348
349 coded_values_per_component = get_bits(gb, 3);
350
351 quant_step_index = get_bits(gb, 3);
352 if (quant_step_index <= 1)
353 return AVERROR_INVALIDDATA;
354
355 if (coding_mode_selector == 3)
356 coding_mode = get_bits1(gb);
357
358 for (b = 0; b < (num_bands + 1) * 4; b++) {
359 int coded_components;
360
361 if (band_flags[b >> 2] == 0)
362 continue;
363
364 coded_components = get_bits(gb, 3);
365
366 for (c = 0; c < coded_components; c++) {
367 TonalComponent *cmp = &components[component_count];
368 int sf_index, coded_values, max_coded_values;
369 float scale_factor;
370
371 sf_index = get_bits(gb, 6);
372 if (component_count >= 64)
373 return AVERROR_INVALIDDATA;
374
375 cmp->pos = b * 64 + get_bits(gb, 6);
376
377 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
378 coded_values = coded_values_per_component + 1;
379 coded_values = FFMIN(max_coded_values, coded_values);
380
381 scale_factor = ff_atrac_sf_table[sf_index] *
382 inv_max_quant[quant_step_index];
383
384 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
385 mantissa, coded_values);
386
387 cmp->num_coefs = coded_values;
388
389 /* inverse quant */
390 for (m = 0; m < coded_values; m++)
391 cmp->coef[m] = mantissa[m] * scale_factor;
392
393 component_count++;
394 }
395 }
396 }
397
398 return component_count;
399 }
400
401 /**
402 * Decode gain parameters for the coded bands
403 *
404 * @param block the gainblock for the current band
405 * @param num_bands amount of coded bands
406 */
407 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
408 int num_bands)
409 {
410 int i, j;
411 int *level, *loc;
412
413 AtracGainInfo *gain = block->g_block;
414
415 for (i = 0; i <= num_bands; i++) {
416 gain[i].num_points = get_bits(gb, 3);
417 level = gain[i].lev_code;
418 loc = gain[i].loc_code;
419
420 for (j = 0; j < gain[i].num_points; j++) {
421 level[j] = get_bits(gb, 4);
422 loc[j] = get_bits(gb, 5);
423 if (j && loc[j] <= loc[j - 1])
424 return AVERROR_INVALIDDATA;
425 }
426 }
427
428 /* Clear the unused blocks. */
429 for (; i < 4 ; i++)
430 gain[i].num_points = 0;
431
432 return 0;
433 }
434
435 /**
436 * Combine the tonal band spectrum and regular band spectrum
437 *
438 * @param spectrum output spectrum buffer
439 * @param num_components number of tonal components
440 * @param components tonal components for this band
441 * @return position of the last tonal coefficient
442 */
443 static int add_tonal_components(float *spectrum, int num_components,
444 TonalComponent *components)
445 {
446 int i, j, last_pos = -1;
447 float *input, *output;
448
449 for (i = 0; i < num_components; i++) {
450 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
451 input = components[i].coef;
452 output = &spectrum[components[i].pos];
453
454 for (j = 0; j < components[i].num_coefs; j++)
455 output[j] += input[j];
456 }
457
458 return last_pos;
459 }
460
461 #define INTERPOLATE(old, new, nsample) \
462 ((old) + (nsample) * 0.125 * ((new) - (old)))
463
464 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
465 int *curr_code)
466 {
467 int i, nsample, band;
468 float mc1_l, mc1_r, mc2_l, mc2_r;
469
470 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
471 int s1 = prev_code[i];
472 int s2 = curr_code[i];
473 nsample = band;
474
475 if (s1 != s2) {
476 /* Selector value changed, interpolation needed. */
477 mc1_l = matrix_coeffs[s1 * 2 ];
478 mc1_r = matrix_coeffs[s1 * 2 + 1];
479 mc2_l = matrix_coeffs[s2 * 2 ];
480 mc2_r = matrix_coeffs[s2 * 2 + 1];
481
482 /* Interpolation is done over the first eight samples. */
483 for (; nsample < band + 8; nsample++) {
484 float c1 = su1[nsample];
485 float c2 = su2[nsample];
486 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
487 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
488 su1[nsample] = c2;
489 su2[nsample] = c1 * 2.0 - c2;
490 }
491 }
492
493 /* Apply the matrix without interpolation. */
494 switch (s2) {
495 case 0: /* M/S decoding */
496 for (; nsample < band + 256; nsample++) {
497 float c1 = su1[nsample];
498 float c2 = su2[nsample];
499 su1[nsample] = c2 * 2.0;
500 su2[nsample] = (c1 - c2) * 2.0;
501 }
502 break;
503 case 1:
504 for (; nsample < band + 256; nsample++) {
505 float c1 = su1[nsample];
506 float c2 = su2[nsample];
507 su1[nsample] = (c1 + c2) * 2.0;
508 su2[nsample] = c2 * -2.0;
509 }
510 break;
511 case 2:
512 case 3:
513 for (; nsample < band + 256; nsample++) {
514 float c1 = su1[nsample];
515 float c2 = su2[nsample];
516 su1[nsample] = c1 + c2;
517 su2[nsample] = c1 - c2;
518 }
519 break;
520 default:
521 assert(0);
522 }
523 }
524 }
525
526 static void get_channel_weights(int index, int flag, float ch[2])
527 {
528 if (index == 7) {
529 ch[0] = 1.0;
530 ch[1] = 1.0;
531 } else {
532 ch[0] = (index & 7) / 7.0;
533 ch[1] = sqrt(2 - ch[0] * ch[0]);
534 if (flag)
535 FFSWAP(float, ch[0], ch[1]);
536 }
537 }
538
539 static void channel_weighting(float *su1, float *su2, int *p3)
540 {
541 int band, nsample;
542 /* w[x][y] y=0 is left y=1 is right */
543 float w[2][2];
544
545 if (p3[1] != 7 || p3[3] != 7) {
546 get_channel_weights(p3[1], p3[0], w[0]);
547 get_channel_weights(p3[3], p3[2], w[1]);
548
549 for (band = 256; band < 4 * 256; band += 256) {
550 for (nsample = band; nsample < band + 8; nsample++) {
551 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
552 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
553 }
554 for(; nsample < band + 256; nsample++) {
555 su1[nsample] *= w[1][0];
556 su2[nsample] *= w[1][1];
557 }
558 }
559 }
560 }
561
562 /**
563 * Decode a Sound Unit
564 *
565 * @param snd the channel unit to be used
566 * @param output the decoded samples before IQMF in float representation
567 * @param channel_num channel number
568 * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
569 */
570 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
571 ChannelUnit *snd, float *output,
572 int channel_num, int coding_mode)
573 {
574 int band, ret, num_subbands, last_tonal, num_bands;
575 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
576 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
577
578 if (coding_mode == JOINT_STEREO && channel_num == 1) {
579 if (get_bits(gb, 2) != 3) {
580 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
581 return AVERROR_INVALIDDATA;
582 }
583 } else {
584 if (get_bits(gb, 6) != 0x28) {
585 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
586 return AVERROR_INVALIDDATA;
587 }
588 }
589
590 /* number of coded QMF bands */
591 snd->bands_coded = get_bits(gb, 2);
592
593 ret = decode_gain_control(gb, gain2, snd->bands_coded);
594 if (ret)
595 return ret;
596
597 snd->num_components = decode_tonal_components(gb, snd->components,
598 snd->bands_coded);
599 if (snd->num_components < 0)
600 return snd->num_components;
601
602 num_subbands = decode_spectrum(gb, snd->spectrum);
603
604 /* Merge the decoded spectrum and tonal components. */
605 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
606 snd->components);
607
608
609 /* calculate number of used MLT/QMF bands according to the amount of coded
610 spectral lines */
611 num_bands = (subband_tab[num_subbands] - 1) >> 8;
612 if (last_tonal >= 0)
613 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
614
615
616 /* Reconstruct time domain samples. */
617 for (band = 0; band < 4; band++) {
618 /* Perform the IMDCT step without overlapping. */
619 if (band <= num_bands)
620 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
621 else
622 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
623
624 /* gain compensation and overlapping */
625 ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf,
626 &snd->prev_frame[band * 256],
627 &gain1->g_block[band], &gain2->g_block[band],
628 256, &output[band * 256]);
629 }
630
631 /* Swap the gain control buffers for the next frame. */
632 snd->gc_blk_switch ^= 1;
633
634 return 0;
635 }
636
637 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
638 float **out_samples)
639 {
640 ATRAC3Context *q = avctx->priv_data;
641 int ret, i;
642 uint8_t *ptr1;
643
644 if (q->coding_mode == JOINT_STEREO) {
645 /* channel coupling mode */
646 /* decode Sound Unit 1 */
647 init_get_bits(&q->gb, databuf, avctx->block_align * 8);
648
649 ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
650 JOINT_STEREO);
651 if (ret != 0)
652 return ret;
653
654 /* Framedata of the su2 in the joint-stereo mode is encoded in
655 * reverse byte order so we need to swap it first. */
656 if (databuf == q->decoded_bytes_buffer) {
657 uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
658 ptr1 = q->decoded_bytes_buffer;
659 for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
660 FFSWAP(uint8_t, *ptr1, *ptr2);
661 } else {
662 const uint8_t *ptr2 = databuf + avctx->block_align - 1;
663 for (i = 0; i < avctx->block_align; i++)
664 q->decoded_bytes_buffer[i] = *ptr2--;
665 }
666
667 /* Skip the sync codes (0xF8). */
668 ptr1 = q->decoded_bytes_buffer;
669 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
670 if (i >= avctx->block_align)
671 return AVERROR_INVALIDDATA;
672 }
673
674
675 /* set the bitstream reader at the start of the second Sound Unit*/
676 init_get_bits(&q->gb, ptr1, (avctx->block_align - i) * 8);
677
678 /* Fill the Weighting coeffs delay buffer */
679 memmove(q->weighting_delay, &q->weighting_delay[2],
680 4 * sizeof(*q->weighting_delay));
681 q->weighting_delay[4] = get_bits1(&q->gb);
682 q->weighting_delay[5] = get_bits(&q->gb, 3);
683
684 for (i = 0; i < 4; i++) {
685 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
686 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
687 q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
688 }
689
690 /* Decode Sound Unit 2. */
691 ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
692 out_samples[1], 1, JOINT_STEREO);
693 if (ret != 0)
694 return ret;
695
696 /* Reconstruct the channel coefficients. */
697 reverse_matrixing(out_samples[0], out_samples[1],
698 q->matrix_coeff_index_prev,
699 q->matrix_coeff_index_now);
700
701 channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
702 } else {
703 /* normal stereo mode or mono */
704 /* Decode the channel sound units. */
705 for (i = 0; i < avctx->channels; i++) {
706 /* Set the bitstream reader at the start of a channel sound unit. */
707 init_get_bits(&q->gb,
708 databuf + i * avctx->block_align / avctx->channels,
709 avctx->block_align * 8 / avctx->channels);
710
711 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
712 out_samples[i], i, q->coding_mode);
713 if (ret != 0)
714 return ret;
715 }
716 }
717
718 /* Apply the iQMF synthesis filter. */
719 for (i = 0; i < avctx->channels; i++) {
720 float *p1 = out_samples[i];
721 float *p2 = p1 + 256;
722 float *p3 = p2 + 256;
723 float *p4 = p3 + 256;
724 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
725 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
726 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
727 }
728
729 return 0;
730 }
731
732 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
733 int *got_frame_ptr, AVPacket *avpkt)
734 {
735 AVFrame *frame = data;
736 const uint8_t *buf = avpkt->data;
737 int buf_size = avpkt->size;
738 ATRAC3Context *q = avctx->priv_data;
739 int ret;
740 const uint8_t *databuf;
741
742 if (buf_size < avctx->block_align) {
743 av_log(avctx, AV_LOG_ERROR,
744 "Frame too small (%d bytes). Truncated file?\n", buf_size);
745 return AVERROR_INVALIDDATA;
746 }
747
748 /* get output buffer */
749 frame->nb_samples = SAMPLES_PER_FRAME;
750 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
751 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
752 return ret;
753 }
754
755 /* Check if we need to descramble and what buffer to pass on. */
756 if (q->scrambled_stream) {
757 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
758 databuf = q->decoded_bytes_buffer;
759 } else {
760 databuf = buf;
761 }
762
763 ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
764 if (ret) {
765 av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
766 return ret;
767 }
768
769 *got_frame_ptr = 1;
770
771 return avctx->block_align;
772 }
773
774 static av_cold void atrac3_init_static_data(AVCodec *codec)
775 {
776 int i;
777
778 init_imdct_window();
779 ff_atrac_generate_tables();
780
781 /* Initialize the VLC tables. */
782 for (i = 0; i < 7; i++) {
783 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
784 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
785 atrac3_vlc_offs[i ];
786 init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
787 huff_bits[i], 1, 1,
788 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
789 }
790 }
791
792 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
793 {
794 int i, ret;
795 int version, delay, samples_per_frame, frame_factor;
796 const uint8_t *edata_ptr = avctx->extradata;
797 ATRAC3Context *q = avctx->priv_data;
798
799 if (avctx->channels <= 0 || avctx->channels > 2) {
800 av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
801 return AVERROR(EINVAL);
802 }
803
804 /* Take care of the codec-specific extradata. */
805 if (avctx->extradata_size == 14) {
806 /* Parse the extradata, WAV format */
807 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
808 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
809 edata_ptr += 4; // samples per channel
810 q->coding_mode = bytestream_get_le16(&edata_ptr);
811 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
812 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
813 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
814 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
815 bytestream_get_le16(&edata_ptr)); // Unknown always 0
816
817 /* setup */
818 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
819 version = 4;
820 delay = 0x88E;
821 q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO;
822 q->scrambled_stream = 0;
823
824 if (avctx->block_align != 96 * avctx->channels * frame_factor &&
825 avctx->block_align != 152 * avctx->channels * frame_factor &&
826 avctx->block_align != 192 * avctx->channels * frame_factor) {
827 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
828 "configuration %d/%d/%d\n", avctx->block_align,
829 avctx->channels, frame_factor);
830 return AVERROR_INVALIDDATA;
831 }
832 } else if (avctx->extradata_size == 10) {
833 /* Parse the extradata, RM format. */
834 version = bytestream_get_be32(&edata_ptr);
835 samples_per_frame = bytestream_get_be16(&edata_ptr);
836 delay = bytestream_get_be16(&edata_ptr);
837 q->coding_mode = bytestream_get_be16(&edata_ptr);
838 q->scrambled_stream = 1;
839
840 } else {
841 av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
842 avctx->extradata_size);
843 return AVERROR(EINVAL);
844 }
845
846 /* Check the extradata */
847
848 if (version != 4) {
849 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
850 return AVERROR_INVALIDDATA;
851 }
852
853 if (samples_per_frame != SAMPLES_PER_FRAME &&
854 samples_per_frame != SAMPLES_PER_FRAME * 2) {
855 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
856 samples_per_frame);
857 return AVERROR_INVALIDDATA;
858 }
859
860 if (delay != 0x88E) {
861 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
862 delay);
863 return AVERROR_INVALIDDATA;
864 }
865
866 if (q->coding_mode == STEREO)
867 av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
868 else if (q->coding_mode == JOINT_STEREO) {
869 if (avctx->channels != 2)
870 return AVERROR_INVALIDDATA;
871 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
872 } else {
873 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
874 q->coding_mode);
875 return AVERROR_INVALIDDATA;
876 }
877
878 if (avctx->block_align >= UINT_MAX / 2)
879 return AVERROR(EINVAL);
880
881 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
882 AV_INPUT_BUFFER_PADDING_SIZE);
883 if (!q->decoded_bytes_buffer)
884 return AVERROR(ENOMEM);
885
886 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
887
888 /* initialize the MDCT transform */
889 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
890 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
891 av_freep(&q->decoded_bytes_buffer);
892 return ret;
893 }
894
895 /* init the joint-stereo decoding data */
896 q->weighting_delay[0] = 0;
897 q->weighting_delay[1] = 7;
898 q->weighting_delay[2] = 0;
899 q->weighting_delay[3] = 7;
900 q->weighting_delay[4] = 0;
901 q->weighting_delay[5] = 7;
902
903 for (i = 0; i < 4; i++) {
904 q->matrix_coeff_index_prev[i] = 3;
905 q->matrix_coeff_index_now[i] = 3;
906 q->matrix_coeff_index_next[i] = 3;
907 }
908
909 ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3);
910 avpriv_float_dsp_init(&q->fdsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
911
912 q->units = av_mallocz(sizeof(*q->units) * avctx->channels);
913 if (!q->units) {
914 atrac3_decode_close(avctx);
915 return AVERROR(ENOMEM);
916 }
917
918 return 0;
919 }
920
921 AVCodec ff_atrac3_decoder = {
922 .name = "atrac3",
923 .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
924 .type = AVMEDIA_TYPE_AUDIO,
925 .id = AV_CODEC_ID_ATRAC3,
926 .priv_data_size = sizeof(ATRAC3Context),
927 .init = atrac3_decode_init,
928 .init_static_data = atrac3_init_static_data,
929 .close = atrac3_decode_close,
930 .decode = atrac3_decode_frame,
931 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
932 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
933 AV_SAMPLE_FMT_NONE },
934 };