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 FFmpeg.
7 *
8 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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 "libavutil/libm.h"
42 #include "libavutil/mem_internal.h"
43 #include "libavutil/thread.h"
44
45 #include "avcodec.h"
46 #include "bytestream.h"
47 #include "codec_internal.h"
48 #include "fft.h"
49 #include "get_bits.h"
50 #include "internal.h"
51
52 #include "atrac.h"
53 #include "atrac3data.h"
54
55 #define MIN_CHANNELS 1
56 #define MAX_CHANNELS 8
57 #define MAX_JS_PAIRS 8 / 2
58
59 #define JOINT_STEREO 0x12
60 #define SINGLE 0x2
61
62 #define SAMPLES_PER_FRAME 1024
63 #define MDCT_SIZE 512
64
65 #define ATRAC3_VLC_BITS 8
66
67 typedef struct GainBlock {
68 AtracGainInfo g_block[4];
69 } GainBlock;
70
71 typedef struct TonalComponent {
72 int pos;
73 int num_coefs;
74 float coef[8];
75 } TonalComponent;
76
77 typedef struct ChannelUnit {
78 int bands_coded;
79 int num_components;
80 float prev_frame[SAMPLES_PER_FRAME];
81 int gc_blk_switch;
82 TonalComponent components[64];
83 GainBlock gain_block[2];
84
85 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
86 DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
87
88 float delay_buf1[46]; ///<qmf delay buffers
89 float delay_buf2[46];
90 float delay_buf3[46];
91 } ChannelUnit;
92
93 typedef struct ATRAC3Context {
94 GetBitContext gb;
95 //@{
96 /** stream data */
97 int coding_mode;
98
99 ChannelUnit *units;
100 //@}
101 //@{
102 /** joint-stereo related variables */
103 int matrix_coeff_index_prev[MAX_JS_PAIRS][4];
104 int matrix_coeff_index_now[MAX_JS_PAIRS][4];
105 int matrix_coeff_index_next[MAX_JS_PAIRS][4];
106 int weighting_delay[MAX_JS_PAIRS][6];
107 //@}
108 //@{
109 /** data buffers */
110 uint8_t *decoded_bytes_buffer;
111 float temp_buf[1070];
112 //@}
113 //@{
114 /** extradata */
115 int scrambled_stream;
116 //@}
117
118 AtracGCContext gainc_ctx;
119 FFTContext mdct_ctx;
120 void (*vector_fmul)(float *dst, const float *src0, const float *src1,
121 int len);
122 } ATRAC3Context;
123
124 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
125 static VLCElem atrac3_vlc_table[7 * 1 << ATRAC3_VLC_BITS];
126 static VLC spectral_coeff_tab[7];
127
128 /**
129 * Regular 512 points IMDCT without overlapping, with the exception of the
130 * swapping of odd bands caused by the reverse spectra of the QMF.
131 *
132 * @param odd_band 1 if the band is an odd band
133 */
imlt(ATRAC3Context *q, float *input, float *output, int odd_band)134 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
135 {
136 int i;
137
138 if (odd_band) {
139 /**
140 * Reverse the odd bands before IMDCT, this is an effect of the QMF
141 * transform or it gives better compression to do it this way.
142 * FIXME: It should be possible to handle this in imdct_calc
143 * for that to happen a modification of the prerotation step of
144 * all SIMD code and C code is needed.
145 * Or fix the functions before so they generate a pre reversed spectrum.
146 */
147 for (i = 0; i < 128; i++)
148 FFSWAP(float, input[i], input[255 - i]);
149 }
150
151 q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
152
153 /* Perform windowing on the output. */
154 q->vector_fmul(output, output, mdct_window, MDCT_SIZE);
155 }
156
157 /*
158 * indata descrambling, only used for data coming from the rm container
159 */
decode_bytes(const uint8_t *input, uint8_t *out, int bytes)160 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
161 {
162 int i, off;
163 uint32_t c;
164 const uint32_t *buf;
165 uint32_t *output = (uint32_t *)out;
166
167 off = (intptr_t)input & 3;
168 buf = (const uint32_t *)(input - off);
169 if (off)
170 c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
171 else
172 c = av_be2ne32(0x537F6103U);
173 bytes += 3 + off;
174 for (i = 0; i < bytes / 4; i++)
175 output[i] = c ^ buf[i];
176
177 if (off)
178 avpriv_request_sample(NULL, "Offset of %d", off);
179
180 return off;
181 }
182
init_imdct_window(void)183 static av_cold void init_imdct_window(void)
184 {
185 int i, j;
186
187 /* generate the mdct window, for details see
188 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
189 for (i = 0, j = 255; i < 128; i++, j--) {
190 float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
191 float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
192 float w = 0.5 * (wi * wi + wj * wj);
193 mdct_window[i] = mdct_window[511 - i] = wi / w;
194 mdct_window[j] = mdct_window[511 - j] = wj / w;
195 }
196 }
197
atrac3_decode_close(AVCodecContext *avctx)198 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
199 {
200 ATRAC3Context *q = avctx->priv_data;
201
202 av_freep(&q->units);
203 av_freep(&q->decoded_bytes_buffer);
204
205 ff_mdct_end(&q->mdct_ctx);
206
207 return 0;
208 }
209
210 /**
211 * Mantissa decoding
212 *
213 * @param selector which table the output values are coded with
214 * @param coding_flag constant length coding or variable length coding
215 * @param mantissas mantissa output table
216 * @param num_codes number of values to get
217 */
read_quant_spectral_coeffs(GetBitContext *gb, int selector, int coding_flag, int *mantissas, int num_codes)218 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
219 int coding_flag, int *mantissas,
220 int num_codes)
221 {
222 int i, code, huff_symb;
223
224 if (selector == 1)
225 num_codes /= 2;
226
227 if (coding_flag != 0) {
228 /* constant length coding (CLC) */
229 int num_bits = clc_length_tab[selector];
230
231 if (selector > 1) {
232 for (i = 0; i < num_codes; i++) {
233 if (num_bits)
234 code = get_sbits(gb, num_bits);
235 else
236 code = 0;
237 mantissas[i] = code;
238 }
239 } else {
240 for (i = 0; i < num_codes; i++) {
241 if (num_bits)
242 code = get_bits(gb, num_bits); // num_bits is always 4 in this case
243 else
244 code = 0;
245 mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
246 mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
247 }
248 }
249 } else {
250 /* variable length coding (VLC) */
251 if (selector != 1) {
252 for (i = 0; i < num_codes; i++) {
253 mantissas[i] = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
254 ATRAC3_VLC_BITS, 1);
255 }
256 } else {
257 for (i = 0; i < num_codes; i++) {
258 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
259 ATRAC3_VLC_BITS, 1);
260 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
261 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
262 }
263 }
264 }
265 }
266
267 /**
268 * Restore the quantized band spectrum coefficients
269 *
270 * @return subband count, fix for broken specification/files
271 */
decode_spectrum(GetBitContext *gb, float *output)272 static int decode_spectrum(GetBitContext *gb, float *output)
273 {
274 int num_subbands, coding_mode, i, j, first, last, subband_size;
275 int subband_vlc_index[32], sf_index[32];
276 int mantissas[128];
277 float scale_factor;
278
279 num_subbands = get_bits(gb, 5); // number of coded subbands
280 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
281
282 /* get the VLC selector table for the subbands, 0 means not coded */
283 for (i = 0; i <= num_subbands; i++)
284 subband_vlc_index[i] = get_bits(gb, 3);
285
286 /* read the scale factor indexes from the stream */
287 for (i = 0; i <= num_subbands; i++) {
288 if (subband_vlc_index[i] != 0)
289 sf_index[i] = get_bits(gb, 6);
290 }
291
292 for (i = 0; i <= num_subbands; i++) {
293 first = subband_tab[i ];
294 last = subband_tab[i + 1];
295
296 subband_size = last - first;
297
298 if (subband_vlc_index[i] != 0) {
299 /* decode spectral coefficients for this subband */
300 /* TODO: This can be done faster is several blocks share the
301 * same VLC selector (subband_vlc_index) */
302 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
303 mantissas, subband_size);
304
305 /* decode the scale factor for this subband */
306 scale_factor = ff_atrac_sf_table[sf_index[i]] *
307 inv_max_quant[subband_vlc_index[i]];
308
309 /* inverse quantize the coefficients */
310 for (j = 0; first < last; first++, j++)
311 output[first] = mantissas[j] * scale_factor;
312 } else {
313 /* this subband was not coded, so zero the entire subband */
314 memset(output + first, 0, subband_size * sizeof(*output));
315 }
316 }
317
318 /* clear the subbands that were not coded */
319 first = subband_tab[i];
320 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
321 return num_subbands;
322 }
323
324 /**
325 * Restore the quantized tonal components
326 *
327 * @param components tonal components
328 * @param num_bands number of coded bands
329 */
decode_tonal_components(GetBitContext *gb, TonalComponent *components, int num_bands)330 static int decode_tonal_components(GetBitContext *gb,
331 TonalComponent *components, int num_bands)
332 {
333 int i, b, c, m;
334 int nb_components, coding_mode_selector, coding_mode;
335 int band_flags[4], mantissa[8];
336 int component_count = 0;
337
338 nb_components = get_bits(gb, 5);
339
340 /* no tonal components */
341 if (nb_components == 0)
342 return 0;
343
344 coding_mode_selector = get_bits(gb, 2);
345 if (coding_mode_selector == 2)
346 return AVERROR_INVALIDDATA;
347
348 coding_mode = coding_mode_selector & 1;
349
350 for (i = 0; i < nb_components; i++) {
351 int coded_values_per_component, quant_step_index;
352
353 for (b = 0; b <= num_bands; b++)
354 band_flags[b] = get_bits1(gb);
355
356 coded_values_per_component = get_bits(gb, 3);
357
358 quant_step_index = get_bits(gb, 3);
359 if (quant_step_index <= 1)
360 return AVERROR_INVALIDDATA;
361
362 if (coding_mode_selector == 3)
363 coding_mode = get_bits1(gb);
364
365 for (b = 0; b < (num_bands + 1) * 4; b++) {
366 int coded_components;
367
368 if (band_flags[b >> 2] == 0)
369 continue;
370
371 coded_components = get_bits(gb, 3);
372
373 for (c = 0; c < coded_components; c++) {
374 TonalComponent *cmp = &components[component_count];
375 int sf_index, coded_values, max_coded_values;
376 float scale_factor;
377
378 sf_index = get_bits(gb, 6);
379 if (component_count >= 64)
380 return AVERROR_INVALIDDATA;
381
382 cmp->pos = b * 64 + get_bits(gb, 6);
383
384 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
385 coded_values = coded_values_per_component + 1;
386 coded_values = FFMIN(max_coded_values, coded_values);
387
388 scale_factor = ff_atrac_sf_table[sf_index] *
389 inv_max_quant[quant_step_index];
390
391 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
392 mantissa, coded_values);
393
394 cmp->num_coefs = coded_values;
395
396 /* inverse quant */
397 for (m = 0; m < coded_values; m++)
398 cmp->coef[m] = mantissa[m] * scale_factor;
399
400 component_count++;
401 }
402 }
403 }
404
405 return component_count;
406 }
407
408 /**
409 * Decode gain parameters for the coded bands
410 *
411 * @param block the gainblock for the current band
412 * @param num_bands amount of coded bands
413 */
decode_gain_control(GetBitContext *gb, GainBlock *block, int num_bands)414 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
415 int num_bands)
416 {
417 int b, j;
418 int *level, *loc;
419
420 AtracGainInfo *gain = block->g_block;
421
422 for (b = 0; b <= num_bands; b++) {
423 gain[b].num_points = get_bits(gb, 3);
424 level = gain[b].lev_code;
425 loc = gain[b].loc_code;
426
427 for (j = 0; j < gain[b].num_points; j++) {
428 level[j] = get_bits(gb, 4);
429 loc[j] = get_bits(gb, 5);
430 if (j && loc[j] <= loc[j - 1])
431 return AVERROR_INVALIDDATA;
432 }
433 }
434
435 /* Clear the unused blocks. */
436 for (; b < 4 ; b++)
437 gain[b].num_points = 0;
438
439 return 0;
440 }
441
442 /**
443 * Combine the tonal band spectrum and regular band spectrum
444 *
445 * @param spectrum output spectrum buffer
446 * @param num_components number of tonal components
447 * @param components tonal components for this band
448 * @return position of the last tonal coefficient
449 */
add_tonal_components(float *spectrum, int num_components, TonalComponent *components)450 static int add_tonal_components(float *spectrum, int num_components,
451 TonalComponent *components)
452 {
453 int i, j, last_pos = -1;
454 float *input, *output;
455
456 for (i = 0; i < num_components; i++) {
457 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
458 input = components[i].coef;
459 output = &spectrum[components[i].pos];
460
461 for (j = 0; j < components[i].num_coefs; j++)
462 output[j] += input[j];
463 }
464
465 return last_pos;
466 }
467
468 #define INTERPOLATE(old, new, nsample) \
469 ((old) + (nsample) * 0.125 * ((new) - (old)))
470
reverse_matrixing(float *su1, float *su2, int *prev_code, int *curr_code)471 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
472 int *curr_code)
473 {
474 int i, nsample, band;
475 float mc1_l, mc1_r, mc2_l, mc2_r;
476
477 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
478 int s1 = prev_code[i];
479 int s2 = curr_code[i];
480 nsample = band;
481
482 if (s1 != s2) {
483 /* Selector value changed, interpolation needed. */
484 mc1_l = matrix_coeffs[s1 * 2 ];
485 mc1_r = matrix_coeffs[s1 * 2 + 1];
486 mc2_l = matrix_coeffs[s2 * 2 ];
487 mc2_r = matrix_coeffs[s2 * 2 + 1];
488
489 /* Interpolation is done over the first eight samples. */
490 for (; nsample < band + 8; nsample++) {
491 float c1 = su1[nsample];
492 float c2 = su2[nsample];
493 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
494 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
495 su1[nsample] = c2;
496 su2[nsample] = c1 * 2.0 - c2;
497 }
498 }
499
500 /* Apply the matrix without interpolation. */
501 switch (s2) {
502 case 0: /* M/S decoding */
503 for (; nsample < band + 256; nsample++) {
504 float c1 = su1[nsample];
505 float c2 = su2[nsample];
506 su1[nsample] = c2 * 2.0;
507 su2[nsample] = (c1 - c2) * 2.0;
508 }
509 break;
510 case 1:
511 for (; nsample < band + 256; nsample++) {
512 float c1 = su1[nsample];
513 float c2 = su2[nsample];
514 su1[nsample] = (c1 + c2) * 2.0;
515 su2[nsample] = c2 * -2.0;
516 }
517 break;
518 case 2:
519 case 3:
520 for (; nsample < band + 256; nsample++) {
521 float c1 = su1[nsample];
522 float c2 = su2[nsample];
523 su1[nsample] = c1 + c2;
524 su2[nsample] = c1 - c2;
525 }
526 break;
527 default:
528 av_assert1(0);
529 }
530 }
531 }
532
get_channel_weights(int index, int flag, float ch[2])533 static void get_channel_weights(int index, int flag, float ch[2])
534 {
535 if (index == 7) {
536 ch[0] = 1.0;
537 ch[1] = 1.0;
538 } else {
539 ch[0] = (index & 7) / 7.0;
540 ch[1] = sqrt(2 - ch[0] * ch[0]);
541 if (flag)
542 FFSWAP(float, ch[0], ch[1]);
543 }
544 }
545
channel_weighting(float *su1, float *su2, int *p3)546 static void channel_weighting(float *su1, float *su2, int *p3)
547 {
548 int band, nsample;
549 /* w[x][y] y=0 is left y=1 is right */
550 float w[2][2];
551
552 if (p3[1] != 7 || p3[3] != 7) {
553 get_channel_weights(p3[1], p3[0], w[0]);
554 get_channel_weights(p3[3], p3[2], w[1]);
555
556 for (band = 256; band < 4 * 256; band += 256) {
557 for (nsample = band; nsample < band + 8; nsample++) {
558 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
559 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
560 }
561 for(; nsample < band + 256; nsample++) {
562 su1[nsample] *= w[1][0];
563 su2[nsample] *= w[1][1];
564 }
565 }
566 }
567 }
568
569 /**
570 * Decode a Sound Unit
571 *
572 * @param snd the channel unit to be used
573 * @param output the decoded samples before IQMF in float representation
574 * @param channel_num channel number
575 * @param coding_mode the coding mode (JOINT_STEREO or single channels)
576 */
decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb, ChannelUnit *snd, float *output, int channel_num, int coding_mode)577 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
578 ChannelUnit *snd, float *output,
579 int channel_num, int coding_mode)
580 {
581 int band, ret, num_subbands, last_tonal, num_bands;
582 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
583 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
584
585 if (coding_mode == JOINT_STEREO && (channel_num % 2) == 1) {
586 if (get_bits(gb, 2) != 3) {
587 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
588 return AVERROR_INVALIDDATA;
589 }
590 } else {
591 if (get_bits(gb, 6) != 0x28) {
592 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
593 return AVERROR_INVALIDDATA;
594 }
595 }
596
597 /* number of coded QMF bands */
598 snd->bands_coded = get_bits(gb, 2);
599
600 ret = decode_gain_control(gb, gain2, snd->bands_coded);
601 if (ret)
602 return ret;
603
604 snd->num_components = decode_tonal_components(gb, snd->components,
605 snd->bands_coded);
606 if (snd->num_components < 0)
607 return snd->num_components;
608
609 num_subbands = decode_spectrum(gb, snd->spectrum);
610
611 /* Merge the decoded spectrum and tonal components. */
612 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
613 snd->components);
614
615
616 /* calculate number of used MLT/QMF bands according to the amount of coded
617 spectral lines */
618 num_bands = (subband_tab[num_subbands] - 1) >> 8;
619 if (last_tonal >= 0)
620 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
621
622
623 /* Reconstruct time domain samples. */
624 for (band = 0; band < 4; band++) {
625 /* Perform the IMDCT step without overlapping. */
626 if (band <= num_bands)
627 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
628 else
629 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
630
631 /* gain compensation and overlapping */
632 ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf,
633 &snd->prev_frame[band * 256],
634 &gain1->g_block[band], &gain2->g_block[band],
635 256, &output[band * 256]);
636 }
637
638 /* Swap the gain control buffers for the next frame. */
639 snd->gc_blk_switch ^= 1;
640
641 return 0;
642 }
643
decode_frame(AVCodecContext *avctx, const uint8_t *databuf, float **out_samples)644 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
645 float **out_samples)
646 {
647 ATRAC3Context *q = avctx->priv_data;
648 int ret, i, ch;
649 uint8_t *ptr1;
650 int channels = avctx->ch_layout.nb_channels;
651
652 if (q->coding_mode == JOINT_STEREO) {
653 /* channel coupling mode */
654
655 /* Decode sound unit pairs (channels are expected to be even).
656 * Multichannel joint stereo interleaves pairs (6ch: 2ch + 2ch + 2ch) */
657 const uint8_t *js_databuf;
658 int js_pair, js_block_align;
659
660 js_block_align = (avctx->block_align / channels) * 2; /* block pair */
661
662 for (ch = 0; ch < channels; ch = ch + 2) {
663 js_pair = ch/2;
664 js_databuf = databuf + js_pair * js_block_align; /* align to current pair */
665
666 /* Set the bitstream reader at the start of first channel sound unit. */
667 init_get_bits(&q->gb,
668 js_databuf, js_block_align * 8);
669
670 /* decode Sound Unit 1 */
671 ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch],
672 out_samples[ch], ch, JOINT_STEREO);
673 if (ret != 0)
674 return ret;
675
676 /* Framedata of the su2 in the joint-stereo mode is encoded in
677 * reverse byte order so we need to swap it first. */
678 if (js_databuf == q->decoded_bytes_buffer) {
679 uint8_t *ptr2 = q->decoded_bytes_buffer + js_block_align - 1;
680 ptr1 = q->decoded_bytes_buffer;
681 for (i = 0; i < js_block_align / 2; i++, ptr1++, ptr2--)
682 FFSWAP(uint8_t, *ptr1, *ptr2);
683 } else {
684 const uint8_t *ptr2 = js_databuf + js_block_align - 1;
685 for (i = 0; i < js_block_align; i++)
686 q->decoded_bytes_buffer[i] = *ptr2--;
687 }
688
689 /* Skip the sync codes (0xF8). */
690 ptr1 = q->decoded_bytes_buffer;
691 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
692 if (i >= js_block_align)
693 return AVERROR_INVALIDDATA;
694 }
695
696
697 /* set the bitstream reader at the start of the second Sound Unit */
698 ret = init_get_bits8(&q->gb,
699 ptr1, q->decoded_bytes_buffer + js_block_align - ptr1);
700 if (ret < 0)
701 return ret;
702
703 /* Fill the Weighting coeffs delay buffer */
704 memmove(q->weighting_delay[js_pair], &q->weighting_delay[js_pair][2],
705 4 * sizeof(*q->weighting_delay[js_pair]));
706 q->weighting_delay[js_pair][4] = get_bits1(&q->gb);
707 q->weighting_delay[js_pair][5] = get_bits(&q->gb, 3);
708
709 for (i = 0; i < 4; i++) {
710 q->matrix_coeff_index_prev[js_pair][i] = q->matrix_coeff_index_now[js_pair][i];
711 q->matrix_coeff_index_now[js_pair][i] = q->matrix_coeff_index_next[js_pair][i];
712 q->matrix_coeff_index_next[js_pair][i] = get_bits(&q->gb, 2);
713 }
714
715 /* Decode Sound Unit 2. */
716 ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch+1],
717 out_samples[ch+1], ch+1, JOINT_STEREO);
718 if (ret != 0)
719 return ret;
720
721 /* Reconstruct the channel coefficients. */
722 reverse_matrixing(out_samples[ch], out_samples[ch+1],
723 q->matrix_coeff_index_prev[js_pair],
724 q->matrix_coeff_index_now[js_pair]);
725
726 channel_weighting(out_samples[ch], out_samples[ch+1], q->weighting_delay[js_pair]);
727 }
728 } else {
729 /* single channels */
730 /* Decode the channel sound units. */
731 for (i = 0; i < channels; i++) {
732 /* Set the bitstream reader at the start of a channel sound unit. */
733 init_get_bits(&q->gb,
734 databuf + i * avctx->block_align / channels,
735 avctx->block_align * 8 / channels);
736
737 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
738 out_samples[i], i, q->coding_mode);
739 if (ret != 0)
740 return ret;
741 }
742 }
743
744 /* Apply the iQMF synthesis filter. */
745 for (i = 0; i < channels; i++) {
746 float *p1 = out_samples[i];
747 float *p2 = p1 + 256;
748 float *p3 = p2 + 256;
749 float *p4 = p3 + 256;
750 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
751 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
752 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
753 }
754
755 return 0;
756 }
757
al_decode_frame(AVCodecContext *avctx, const uint8_t *databuf, int size, float **out_samples)758 static int al_decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
759 int size, float **out_samples)
760 {
761 ATRAC3Context *q = avctx->priv_data;
762 int channels = avctx->ch_layout.nb_channels;
763 int ret, i;
764
765 /* Set the bitstream reader at the start of a channel sound unit. */
766 init_get_bits(&q->gb, databuf, size * 8);
767 /* single channels */
768 /* Decode the channel sound units. */
769 for (i = 0; i < channels; i++) {
770 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
771 out_samples[i], i, q->coding_mode);
772 if (ret != 0)
773 return ret;
774 while (i < channels && get_bits_left(&q->gb) > 6 && show_bits(&q->gb, 6) != 0x28) {
775 skip_bits(&q->gb, 1);
776 }
777 }
778
779 /* Apply the iQMF synthesis filter. */
780 for (i = 0; i < channels; i++) {
781 float *p1 = out_samples[i];
782 float *p2 = p1 + 256;
783 float *p3 = p2 + 256;
784 float *p4 = p3 + 256;
785 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
786 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
787 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
788 }
789
790 return 0;
791 }
792
atrac3_decode_frame(AVCodecContext *avctx, AVFrame *frame, int *got_frame_ptr, AVPacket *avpkt)793 static int atrac3_decode_frame(AVCodecContext *avctx, AVFrame *frame,
794 int *got_frame_ptr, AVPacket *avpkt)
795 {
796 const uint8_t *buf = avpkt->data;
797 int buf_size = avpkt->size;
798 ATRAC3Context *q = avctx->priv_data;
799 int ret;
800 const uint8_t *databuf;
801
802 if (buf_size < avctx->block_align) {
803 av_log(avctx, AV_LOG_ERROR,
804 "Frame too small (%d bytes). Truncated file?\n", buf_size);
805 return AVERROR_INVALIDDATA;
806 }
807
808 /* get output buffer */
809 frame->nb_samples = SAMPLES_PER_FRAME;
810 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
811 return ret;
812
813 /* Check if we need to descramble and what buffer to pass on. */
814 if (q->scrambled_stream) {
815 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
816 databuf = q->decoded_bytes_buffer;
817 } else {
818 databuf = buf;
819 }
820
821 ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
822 if (ret) {
823 av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
824 return ret;
825 }
826
827 *got_frame_ptr = 1;
828
829 return avctx->block_align;
830 }
831
atrac3al_decode_frame(AVCodecContext *avctx, AVFrame *frame, int *got_frame_ptr, AVPacket *avpkt)832 static int atrac3al_decode_frame(AVCodecContext *avctx, AVFrame *frame,
833 int *got_frame_ptr, AVPacket *avpkt)
834 {
835 int ret;
836
837 frame->nb_samples = SAMPLES_PER_FRAME;
838 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
839 return ret;
840
841 ret = al_decode_frame(avctx, avpkt->data, avpkt->size,
842 (float **)frame->extended_data);
843 if (ret) {
844 av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
845 return ret;
846 }
847
848 *got_frame_ptr = 1;
849
850 return avpkt->size;
851 }
852
atrac3_init_static_data(void)853 static av_cold void atrac3_init_static_data(void)
854 {
855 VLCElem *table = atrac3_vlc_table;
856 const uint8_t (*hufftabs)[2] = atrac3_hufftabs;
857 int i;
858
859 init_imdct_window();
860 ff_atrac_generate_tables();
861
862 /* Initialize the VLC tables. */
863 for (i = 0; i < 7; i++) {
864 spectral_coeff_tab[i].table = table;
865 spectral_coeff_tab[i].table_allocated = 256;
866 ff_init_vlc_from_lengths(&spectral_coeff_tab[i], ATRAC3_VLC_BITS, huff_tab_sizes[i],
867 &hufftabs[0][1], 2,
868 &hufftabs[0][0], 2, 1,
869 -31, INIT_VLC_USE_NEW_STATIC, NULL);
870 hufftabs += huff_tab_sizes[i];
871 table += 256;
872 }
873 }
874
atrac3_decode_init(AVCodecContext *avctx)875 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
876 {
877 static AVOnce init_static_once = AV_ONCE_INIT;
878 int i, js_pair, ret;
879 int version, delay, samples_per_frame, frame_factor;
880 const uint8_t *edata_ptr = avctx->extradata;
881 ATRAC3Context *q = avctx->priv_data;
882 AVFloatDSPContext *fdsp;
883 int channels = avctx->ch_layout.nb_channels;
884
885 if (channels < MIN_CHANNELS || channels > MAX_CHANNELS) {
886 av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
887 return AVERROR(EINVAL);
888 }
889
890 /* Take care of the codec-specific extradata. */
891 if (avctx->codec_id == AV_CODEC_ID_ATRAC3AL) {
892 version = 4;
893 samples_per_frame = SAMPLES_PER_FRAME * channels;
894 delay = 0x88E;
895 q->coding_mode = SINGLE;
896 } else if (avctx->extradata_size == 14) {
897 /* Parse the extradata, WAV format */
898 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
899 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
900 edata_ptr += 4; // samples per channel
901 q->coding_mode = bytestream_get_le16(&edata_ptr);
902 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
903 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
904 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
905 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
906 bytestream_get_le16(&edata_ptr)); // Unknown always 0
907
908 /* setup */
909 samples_per_frame = SAMPLES_PER_FRAME * channels;
910 version = 4;
911 delay = 0x88E;
912 q->coding_mode = q->coding_mode ? JOINT_STEREO : SINGLE;
913 q->scrambled_stream = 0;
914
915 if (avctx->block_align != 96 * channels * frame_factor &&
916 avctx->block_align != 152 * channels * frame_factor &&
917 avctx->block_align != 192 * channels * frame_factor) {
918 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
919 "configuration %d/%d/%d\n", avctx->block_align,
920 channels, frame_factor);
921 return AVERROR_INVALIDDATA;
922 }
923 } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
924 /* Parse the extradata, RM format. */
925 version = bytestream_get_be32(&edata_ptr);
926 samples_per_frame = bytestream_get_be16(&edata_ptr);
927 delay = bytestream_get_be16(&edata_ptr);
928 q->coding_mode = bytestream_get_be16(&edata_ptr);
929 q->scrambled_stream = 1;
930
931 } else {
932 av_log(avctx, AV_LOG_ERROR, "Unknown extradata size %d.\n",
933 avctx->extradata_size);
934 return AVERROR(EINVAL);
935 }
936
937 /* Check the extradata */
938
939 if (version != 4) {
940 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
941 return AVERROR_INVALIDDATA;
942 }
943
944 if (samples_per_frame != SAMPLES_PER_FRAME * channels) {
945 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
946 samples_per_frame);
947 return AVERROR_INVALIDDATA;
948 }
949
950 if (delay != 0x88E) {
951 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
952 delay);
953 return AVERROR_INVALIDDATA;
954 }
955
956 if (q->coding_mode == SINGLE)
957 av_log(avctx, AV_LOG_DEBUG, "Single channels detected.\n");
958 else if (q->coding_mode == JOINT_STEREO) {
959 if (channels % 2 == 1) { /* Joint stereo channels must be even */
960 av_log(avctx, AV_LOG_ERROR, "Invalid joint stereo channel configuration.\n");
961 return AVERROR_INVALIDDATA;
962 }
963 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
964 } else {
965 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
966 q->coding_mode);
967 return AVERROR_INVALIDDATA;
968 }
969
970 if (avctx->block_align > 4096 || avctx->block_align <= 0)
971 return AVERROR(EINVAL);
972
973 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
974 AV_INPUT_BUFFER_PADDING_SIZE);
975 if (!q->decoded_bytes_buffer)
976 return AVERROR(ENOMEM);
977
978 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
979
980 /* initialize the MDCT transform */
981 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
982 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
983 return ret;
984 }
985
986 /* init the joint-stereo decoding data */
987 for (js_pair = 0; js_pair < MAX_JS_PAIRS; js_pair++) {
988 q->weighting_delay[js_pair][0] = 0;
989 q->weighting_delay[js_pair][1] = 7;
990 q->weighting_delay[js_pair][2] = 0;
991 q->weighting_delay[js_pair][3] = 7;
992 q->weighting_delay[js_pair][4] = 0;
993 q->weighting_delay[js_pair][5] = 7;
994
995 for (i = 0; i < 4; i++) {
996 q->matrix_coeff_index_prev[js_pair][i] = 3;
997 q->matrix_coeff_index_now[js_pair][i] = 3;
998 q->matrix_coeff_index_next[js_pair][i] = 3;
999 }
1000 }
1001
1002 ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3);
1003 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1004 if (!fdsp)
1005 return AVERROR(ENOMEM);
1006 q->vector_fmul = fdsp->vector_fmul;
1007 av_free(fdsp);
1008
1009 q->units = av_calloc(channels, sizeof(*q->units));
1010 if (!q->units)
1011 return AVERROR(ENOMEM);
1012
1013 ff_thread_once(&init_static_once, atrac3_init_static_data);
1014
1015 return 0;
1016 }
1017
1018 const FFCodec ff_atrac3_decoder = {
1019 .p.name = "atrac3",
1020 .p.long_name = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
1021 .p.type = AVMEDIA_TYPE_AUDIO,
1022 .p.id = AV_CODEC_ID_ATRAC3,
1023 .priv_data_size = sizeof(ATRAC3Context),
1024 .init = atrac3_decode_init,
1025 .close = atrac3_decode_close,
1026 FF_CODEC_DECODE_CB(atrac3_decode_frame),
1027 .p.capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1028 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1029 AV_SAMPLE_FMT_NONE },
1030 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1031 };
1032
1033 const FFCodec ff_atrac3al_decoder = {
1034 .p.name = "atrac3al",
1035 .p.long_name = NULL_IF_CONFIG_SMALL("ATRAC3 AL (Adaptive TRansform Acoustic Coding 3 Advanced Lossless)"),
1036 .p.type = AVMEDIA_TYPE_AUDIO,
1037 .p.id = AV_CODEC_ID_ATRAC3AL,
1038 .priv_data_size = sizeof(ATRAC3Context),
1039 .init = atrac3_decode_init,
1040 .close = atrac3_decode_close,
1041 FF_CODEC_DECODE_CB(atrac3al_decode_frame),
1042 .p.capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1043 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1044 AV_SAMPLE_FMT_NONE },
1045 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1046 };
1047