You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1247 lines
44KB

  1. /*
  2. * AC-3 Audio Decoder
  3. * This code is developed as part of Google Summer of Code 2006 Program.
  4. *
  5. * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com).
  6. * Copyright (c) 2007 Justin Ruggles
  7. *
  8. * Portions of this code are derived from liba52
  9. * http://liba52.sourceforge.net
  10. * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
  11. * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
  12. *
  13. * This file is part of FFmpeg.
  14. *
  15. * FFmpeg is free software; you can redistribute it and/or
  16. * modify it under the terms of the GNU General Public
  17. * License as published by the Free Software Foundation; either
  18. * version 2 of the License, or (at your option) any later version.
  19. *
  20. * FFmpeg is distributed in the hope that it will be useful,
  21. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  23. * General Public License for more details.
  24. *
  25. * You should have received a copy of the GNU General Public
  26. * License along with FFmpeg; if not, write to the Free Software
  27. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  28. */
  29. #include <stdio.h>
  30. #include <stddef.h>
  31. #include <math.h>
  32. #include <string.h>
  33. #include "libavutil/crc.h"
  34. #include "libavutil/random.h"
  35. #include "avcodec.h"
  36. #include "ac3_parser.h"
  37. #include "bitstream.h"
  38. #include "dsputil.h"
  39. /** Maximum possible frame size when the specification limit is ignored */
  40. #define AC3_MAX_FRAME_SIZE 21695
  41. /**
  42. * Table of bin locations for rematrixing bands
  43. * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
  44. */
  45. static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
  46. /** table for grouping exponents */
  47. static uint8_t exp_ungroup_tab[128][3];
  48. /** tables for ungrouping mantissas */
  49. static int b1_mantissas[32][3];
  50. static int b2_mantissas[128][3];
  51. static int b3_mantissas[8];
  52. static int b4_mantissas[128][2];
  53. static int b5_mantissas[16];
  54. /**
  55. * Quantization table: levels for symmetric. bits for asymmetric.
  56. * reference: Table 7.18 Mapping of bap to Quantizer
  57. */
  58. static const uint8_t quantization_tab[16] = {
  59. 0, 3, 5, 7, 11, 15,
  60. 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
  61. };
  62. /** dynamic range table. converts codes to scale factors. */
  63. static float dynamic_range_tab[256];
  64. /** Adjustments in dB gain */
  65. #define LEVEL_MINUS_3DB 0.7071067811865476
  66. #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
  67. #define LEVEL_MINUS_6DB 0.5000000000000000
  68. #define LEVEL_MINUS_9DB 0.3535533905932738
  69. #define LEVEL_ZERO 0.0000000000000000
  70. #define LEVEL_ONE 1.0000000000000000
  71. static const float gain_levels[6] = {
  72. LEVEL_ZERO,
  73. LEVEL_ONE,
  74. LEVEL_MINUS_3DB,
  75. LEVEL_MINUS_4POINT5DB,
  76. LEVEL_MINUS_6DB,
  77. LEVEL_MINUS_9DB
  78. };
  79. /**
  80. * Table for default stereo downmixing coefficients
  81. * reference: Section 7.8.2 Downmixing Into Two Channels
  82. */
  83. static const uint8_t ac3_default_coeffs[8][5][2] = {
  84. { { 1, 0 }, { 0, 1 }, },
  85. { { 2, 2 }, },
  86. { { 1, 0 }, { 0, 1 }, },
  87. { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
  88. { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
  89. { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
  90. { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
  91. { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
  92. };
  93. /* override ac3.h to include coupling channel */
  94. #undef AC3_MAX_CHANNELS
  95. #define AC3_MAX_CHANNELS 7
  96. #define CPL_CH 0
  97. #define AC3_OUTPUT_LFEON 8
  98. typedef struct {
  99. int channel_mode; ///< channel mode (acmod)
  100. int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
  101. int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
  102. int dither_all; ///< true if all channels are dithered
  103. int cpl_in_use; ///< coupling in use
  104. int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
  105. int phase_flags_in_use; ///< phase flags in use
  106. int phase_flags[18]; ///< phase flags
  107. int cpl_band_struct[18]; ///< coupling band structure
  108. int num_rematrixing_bands; ///< number of rematrixing bands
  109. int rematrixing_flags[4]; ///< rematrixing flags
  110. int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
  111. int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
  112. int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
  113. int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
  114. int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
  115. uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
  116. uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
  117. uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
  118. int sample_rate; ///< sample frequency, in Hz
  119. int bit_rate; ///< stream bit rate, in bits-per-second
  120. int frame_size; ///< current frame size, in bytes
  121. int channels; ///< number of total channels
  122. int fbw_channels; ///< number of full-bandwidth channels
  123. int lfe_on; ///< lfe channel in use
  124. int lfe_ch; ///< index of LFE channel
  125. int output_mode; ///< output channel configuration
  126. int out_channels; ///< number of output channels
  127. int center_mix_level; ///< Center mix level index
  128. int surround_mix_level; ///< Surround mix level index
  129. float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
  130. float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
  131. float dynamic_range[2]; ///< dynamic range
  132. int cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
  133. int num_cpl_bands; ///< number of coupling bands
  134. int num_cpl_subbands; ///< number of coupling sub bands
  135. int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
  136. int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
  137. AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
  138. int num_exp_groups[AC3_MAX_CHANNELS]; ///< Number of exponent groups
  139. int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
  140. uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
  141. int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
  142. int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
  143. int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
  144. int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
  145. DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
  146. int downmixed; ///< indicates if coeffs are currently downmixed
  147. /* For IMDCT. */
  148. MDCTContext imdct_512; ///< for 512 sample IMDCT
  149. MDCTContext imdct_256; ///< for 256 sample IMDCT
  150. DSPContext dsp; ///< for optimization
  151. float add_bias; ///< offset for float_to_int16 conversion
  152. float mul_bias; ///< scaling for float_to_int16 conversion
  153. DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
  154. DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
  155. DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
  156. DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
  157. DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
  158. DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
  159. /* Miscellaneous. */
  160. GetBitContext gbc; ///< bitstream reader
  161. AVRandomState dith_state; ///< for dither generation
  162. AVCodecContext *avctx; ///< parent context
  163. uint8_t *input_buffer; ///< temp buffer to prevent overread
  164. } AC3DecodeContext;
  165. /**
  166. * Symmetrical Dequantization
  167. * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
  168. * Tables 7.19 to 7.23
  169. */
  170. static inline int
  171. symmetric_dequant(int code, int levels)
  172. {
  173. return ((code - (levels >> 1)) << 24) / levels;
  174. }
  175. /*
  176. * Initialize tables at runtime.
  177. */
  178. static av_cold void ac3_tables_init(void)
  179. {
  180. int i;
  181. /* generate grouped mantissa tables
  182. reference: Section 7.3.5 Ungrouping of Mantissas */
  183. for(i=0; i<32; i++) {
  184. /* bap=1 mantissas */
  185. b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
  186. b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
  187. b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
  188. }
  189. for(i=0; i<128; i++) {
  190. /* bap=2 mantissas */
  191. b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
  192. b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
  193. b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
  194. /* bap=4 mantissas */
  195. b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
  196. b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
  197. }
  198. /* generate ungrouped mantissa tables
  199. reference: Tables 7.21 and 7.23 */
  200. for(i=0; i<7; i++) {
  201. /* bap=3 mantissas */
  202. b3_mantissas[i] = symmetric_dequant(i, 7);
  203. }
  204. for(i=0; i<15; i++) {
  205. /* bap=5 mantissas */
  206. b5_mantissas[i] = symmetric_dequant(i, 15);
  207. }
  208. /* generate dynamic range table
  209. reference: Section 7.7.1 Dynamic Range Control */
  210. for(i=0; i<256; i++) {
  211. int v = (i >> 5) - ((i >> 7) << 3) - 5;
  212. dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
  213. }
  214. /* generate exponent tables
  215. reference: Section 7.1.3 Exponent Decoding */
  216. for(i=0; i<128; i++) {
  217. exp_ungroup_tab[i][0] = i / 25;
  218. exp_ungroup_tab[i][1] = (i % 25) / 5;
  219. exp_ungroup_tab[i][2] = (i % 25) % 5;
  220. }
  221. }
  222. /**
  223. * AVCodec initialization
  224. */
  225. static av_cold int ac3_decode_init(AVCodecContext *avctx)
  226. {
  227. AC3DecodeContext *s = avctx->priv_data;
  228. s->avctx = avctx;
  229. ac3_common_init();
  230. ac3_tables_init();
  231. ff_mdct_init(&s->imdct_256, 8, 1);
  232. ff_mdct_init(&s->imdct_512, 9, 1);
  233. ff_kbd_window_init(s->window, 5.0, 256);
  234. dsputil_init(&s->dsp, avctx);
  235. av_init_random(0, &s->dith_state);
  236. /* set bias values for float to int16 conversion */
  237. if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
  238. s->add_bias = 385.0f;
  239. s->mul_bias = 1.0f;
  240. } else {
  241. s->add_bias = 0.0f;
  242. s->mul_bias = 32767.0f;
  243. }
  244. /* allow downmixing to stereo or mono */
  245. if (avctx->channels > 0 && avctx->request_channels > 0 &&
  246. avctx->request_channels < avctx->channels &&
  247. avctx->request_channels <= 2) {
  248. avctx->channels = avctx->request_channels;
  249. }
  250. s->downmixed = 1;
  251. /* allocate context input buffer */
  252. if (avctx->error_resilience >= FF_ER_CAREFUL) {
  253. s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
  254. if (!s->input_buffer)
  255. return AVERROR_NOMEM;
  256. }
  257. return 0;
  258. }
  259. /**
  260. * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
  261. * GetBitContext within AC3DecodeContext must point to
  262. * start of the synchronized ac3 bitstream.
  263. */
  264. static int ac3_parse_header(AC3DecodeContext *s)
  265. {
  266. AC3HeaderInfo hdr;
  267. GetBitContext *gbc = &s->gbc;
  268. int err, i;
  269. err = ff_ac3_parse_header(gbc, &hdr);
  270. if(err)
  271. return err;
  272. if(hdr.bitstream_id > 10)
  273. return AC3_PARSE_ERROR_BSID;
  274. /* get decoding parameters from header info */
  275. s->bit_alloc_params.sr_code = hdr.sr_code;
  276. s->channel_mode = hdr.channel_mode;
  277. s->lfe_on = hdr.lfe_on;
  278. s->bit_alloc_params.sr_shift = hdr.sr_shift;
  279. s->sample_rate = hdr.sample_rate;
  280. s->bit_rate = hdr.bit_rate;
  281. s->channels = hdr.channels;
  282. s->fbw_channels = s->channels - s->lfe_on;
  283. s->lfe_ch = s->fbw_channels + 1;
  284. s->frame_size = hdr.frame_size;
  285. s->center_mix_level = hdr.center_mix_level;
  286. s->surround_mix_level = hdr.surround_mix_level;
  287. if(s->lfe_on) {
  288. s->start_freq[s->lfe_ch] = 0;
  289. s->end_freq[s->lfe_ch] = 7;
  290. s->num_exp_groups[s->lfe_ch] = 2;
  291. s->channel_in_cpl[s->lfe_ch] = 0;
  292. }
  293. /* read the rest of the bsi. read twice for dual mono mode. */
  294. i = !(s->channel_mode);
  295. do {
  296. skip_bits(gbc, 5); // skip dialog normalization
  297. if (get_bits1(gbc))
  298. skip_bits(gbc, 8); //skip compression
  299. if (get_bits1(gbc))
  300. skip_bits(gbc, 8); //skip language code
  301. if (get_bits1(gbc))
  302. skip_bits(gbc, 7); //skip audio production information
  303. } while (i--);
  304. skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
  305. /* skip the timecodes (or extra bitstream information for Alternate Syntax)
  306. TODO: read & use the xbsi1 downmix levels */
  307. if (get_bits1(gbc))
  308. skip_bits(gbc, 14); //skip timecode1 / xbsi1
  309. if (get_bits1(gbc))
  310. skip_bits(gbc, 14); //skip timecode2 / xbsi2
  311. /* skip additional bitstream info */
  312. if (get_bits1(gbc)) {
  313. i = get_bits(gbc, 6);
  314. do {
  315. skip_bits(gbc, 8);
  316. } while(i--);
  317. }
  318. return 0;
  319. }
  320. /**
  321. * Set stereo downmixing coefficients based on frame header info.
  322. * reference: Section 7.8.2 Downmixing Into Two Channels
  323. */
  324. static void set_downmix_coeffs(AC3DecodeContext *s)
  325. {
  326. int i;
  327. float cmix = gain_levels[s->center_mix_level];
  328. float smix = gain_levels[s->surround_mix_level];
  329. for(i=0; i<s->fbw_channels; i++) {
  330. s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
  331. s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
  332. }
  333. if(s->channel_mode > 1 && s->channel_mode & 1) {
  334. s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
  335. }
  336. if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
  337. int nf = s->channel_mode - 2;
  338. s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
  339. }
  340. if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
  341. int nf = s->channel_mode - 4;
  342. s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
  343. }
  344. /* calculate adjustment needed for each channel to avoid clipping */
  345. s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
  346. for(i=0; i<s->fbw_channels; i++) {
  347. s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
  348. s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
  349. }
  350. s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
  351. s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
  352. }
  353. /**
  354. * Decode the grouped exponents according to exponent strategy.
  355. * reference: Section 7.1.3 Exponent Decoding
  356. */
  357. static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
  358. uint8_t absexp, int8_t *dexps)
  359. {
  360. int i, j, grp, group_size;
  361. int dexp[256];
  362. int expacc, prevexp;
  363. /* unpack groups */
  364. group_size = exp_strategy + (exp_strategy == EXP_D45);
  365. for(grp=0,i=0; grp<ngrps; grp++) {
  366. expacc = get_bits(gbc, 7);
  367. dexp[i++] = exp_ungroup_tab[expacc][0];
  368. dexp[i++] = exp_ungroup_tab[expacc][1];
  369. dexp[i++] = exp_ungroup_tab[expacc][2];
  370. }
  371. /* convert to absolute exps and expand groups */
  372. prevexp = absexp;
  373. for(i=0; i<ngrps*3; i++) {
  374. prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
  375. for(j=0; j<group_size; j++) {
  376. dexps[(i*group_size)+j] = prevexp;
  377. }
  378. }
  379. }
  380. /**
  381. * Generate transform coefficients for each coupled channel in the coupling
  382. * range using the coupling coefficients and coupling coordinates.
  383. * reference: Section 7.4.3 Coupling Coordinate Format
  384. */
  385. static void uncouple_channels(AC3DecodeContext *s)
  386. {
  387. int i, j, ch, bnd, subbnd;
  388. subbnd = -1;
  389. i = s->start_freq[CPL_CH];
  390. for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
  391. do {
  392. subbnd++;
  393. for(j=0; j<12; j++) {
  394. for(ch=1; ch<=s->fbw_channels; ch++) {
  395. if(s->channel_in_cpl[ch]) {
  396. s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
  397. if (ch == 2 && s->phase_flags[bnd])
  398. s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
  399. }
  400. }
  401. i++;
  402. }
  403. } while(s->cpl_band_struct[subbnd]);
  404. }
  405. }
  406. /**
  407. * Grouped mantissas for 3-level 5-level and 11-level quantization
  408. */
  409. typedef struct {
  410. int b1_mant[3];
  411. int b2_mant[3];
  412. int b4_mant[2];
  413. int b1ptr;
  414. int b2ptr;
  415. int b4ptr;
  416. } mant_groups;
  417. /**
  418. * Get the transform coefficients for a particular channel
  419. * reference: Section 7.3 Quantization and Decoding of Mantissas
  420. */
  421. static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
  422. {
  423. GetBitContext *gbc = &s->gbc;
  424. int i, gcode, tbap, start, end;
  425. uint8_t *exps;
  426. uint8_t *bap;
  427. int *coeffs;
  428. exps = s->dexps[ch_index];
  429. bap = s->bap[ch_index];
  430. coeffs = s->fixed_coeffs[ch_index];
  431. start = s->start_freq[ch_index];
  432. end = s->end_freq[ch_index];
  433. for (i = start; i < end; i++) {
  434. tbap = bap[i];
  435. switch (tbap) {
  436. case 0:
  437. coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
  438. break;
  439. case 1:
  440. if(m->b1ptr > 2) {
  441. gcode = get_bits(gbc, 5);
  442. m->b1_mant[0] = b1_mantissas[gcode][0];
  443. m->b1_mant[1] = b1_mantissas[gcode][1];
  444. m->b1_mant[2] = b1_mantissas[gcode][2];
  445. m->b1ptr = 0;
  446. }
  447. coeffs[i] = m->b1_mant[m->b1ptr++];
  448. break;
  449. case 2:
  450. if(m->b2ptr > 2) {
  451. gcode = get_bits(gbc, 7);
  452. m->b2_mant[0] = b2_mantissas[gcode][0];
  453. m->b2_mant[1] = b2_mantissas[gcode][1];
  454. m->b2_mant[2] = b2_mantissas[gcode][2];
  455. m->b2ptr = 0;
  456. }
  457. coeffs[i] = m->b2_mant[m->b2ptr++];
  458. break;
  459. case 3:
  460. coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
  461. break;
  462. case 4:
  463. if(m->b4ptr > 1) {
  464. gcode = get_bits(gbc, 7);
  465. m->b4_mant[0] = b4_mantissas[gcode][0];
  466. m->b4_mant[1] = b4_mantissas[gcode][1];
  467. m->b4ptr = 0;
  468. }
  469. coeffs[i] = m->b4_mant[m->b4ptr++];
  470. break;
  471. case 5:
  472. coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
  473. break;
  474. default: {
  475. /* asymmetric dequantization */
  476. int qlevel = quantization_tab[tbap];
  477. coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
  478. break;
  479. }
  480. }
  481. coeffs[i] >>= exps[i];
  482. }
  483. }
  484. /**
  485. * Remove random dithering from coefficients with zero-bit mantissas
  486. * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
  487. */
  488. static void remove_dithering(AC3DecodeContext *s) {
  489. int ch, i;
  490. int end=0;
  491. int *coeffs;
  492. uint8_t *bap;
  493. for(ch=1; ch<=s->fbw_channels; ch++) {
  494. if(!s->dither_flag[ch]) {
  495. coeffs = s->fixed_coeffs[ch];
  496. bap = s->bap[ch];
  497. if(s->channel_in_cpl[ch])
  498. end = s->start_freq[CPL_CH];
  499. else
  500. end = s->end_freq[ch];
  501. for(i=0; i<end; i++) {
  502. if(!bap[i])
  503. coeffs[i] = 0;
  504. }
  505. if(s->channel_in_cpl[ch]) {
  506. bap = s->bap[CPL_CH];
  507. for(; i<s->end_freq[CPL_CH]; i++) {
  508. if(!bap[i])
  509. coeffs[i] = 0;
  510. }
  511. }
  512. }
  513. }
  514. }
  515. /**
  516. * Get the transform coefficients.
  517. */
  518. static void get_transform_coeffs(AC3DecodeContext *s)
  519. {
  520. int ch, end;
  521. int got_cplchan = 0;
  522. mant_groups m;
  523. m.b1ptr = m.b2ptr = m.b4ptr = 3;
  524. for (ch = 1; ch <= s->channels; ch++) {
  525. /* transform coefficients for full-bandwidth channel */
  526. get_transform_coeffs_ch(s, ch, &m);
  527. /* tranform coefficients for coupling channel come right after the
  528. coefficients for the first coupled channel*/
  529. if (s->channel_in_cpl[ch]) {
  530. if (!got_cplchan) {
  531. get_transform_coeffs_ch(s, CPL_CH, &m);
  532. uncouple_channels(s);
  533. got_cplchan = 1;
  534. }
  535. end = s->end_freq[CPL_CH];
  536. } else {
  537. end = s->end_freq[ch];
  538. }
  539. do
  540. s->fixed_coeffs[ch][end] = 0;
  541. while(++end < 256);
  542. }
  543. /* if any channel doesn't use dithering, zero appropriate coefficients */
  544. if(!s->dither_all)
  545. remove_dithering(s);
  546. }
  547. /**
  548. * Stereo rematrixing.
  549. * reference: Section 7.5.4 Rematrixing : Decoding Technique
  550. */
  551. static void do_rematrixing(AC3DecodeContext *s)
  552. {
  553. int bnd, i;
  554. int end, bndend;
  555. int tmp0, tmp1;
  556. end = FFMIN(s->end_freq[1], s->end_freq[2]);
  557. for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
  558. if(s->rematrixing_flags[bnd]) {
  559. bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
  560. for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
  561. tmp0 = s->fixed_coeffs[1][i];
  562. tmp1 = s->fixed_coeffs[2][i];
  563. s->fixed_coeffs[1][i] = tmp0 + tmp1;
  564. s->fixed_coeffs[2][i] = tmp0 - tmp1;
  565. }
  566. }
  567. }
  568. }
  569. /**
  570. * Perform the 256-point IMDCT
  571. */
  572. static void do_imdct_256(AC3DecodeContext *s, int chindex)
  573. {
  574. int i, k;
  575. DECLARE_ALIGNED_16(float, x[128]);
  576. FFTComplex z[2][64];
  577. float *o_ptr = s->tmp_output;
  578. for(i=0; i<2; i++) {
  579. /* de-interleave coefficients */
  580. for(k=0; k<128; k++) {
  581. x[k] = s->transform_coeffs[chindex][2*k+i];
  582. }
  583. /* run standard IMDCT */
  584. s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
  585. /* reverse the post-rotation & reordering from standard IMDCT */
  586. for(k=0; k<32; k++) {
  587. z[i][32+k].re = -o_ptr[128+2*k];
  588. z[i][32+k].im = -o_ptr[2*k];
  589. z[i][31-k].re = o_ptr[2*k+1];
  590. z[i][31-k].im = o_ptr[128+2*k+1];
  591. }
  592. }
  593. /* apply AC-3 post-rotation & reordering */
  594. for(k=0; k<64; k++) {
  595. o_ptr[ 2*k ] = -z[0][ k].im;
  596. o_ptr[ 2*k+1] = z[0][63-k].re;
  597. o_ptr[128+2*k ] = -z[0][ k].re;
  598. o_ptr[128+2*k+1] = z[0][63-k].im;
  599. o_ptr[256+2*k ] = -z[1][ k].re;
  600. o_ptr[256+2*k+1] = z[1][63-k].im;
  601. o_ptr[384+2*k ] = z[1][ k].im;
  602. o_ptr[384+2*k+1] = -z[1][63-k].re;
  603. }
  604. }
  605. /**
  606. * Inverse MDCT Transform.
  607. * Convert frequency domain coefficients to time-domain audio samples.
  608. * reference: Section 7.9.4 Transformation Equations
  609. */
  610. static inline void do_imdct(AC3DecodeContext *s, int channels)
  611. {
  612. int ch;
  613. for (ch=1; ch<=channels; ch++) {
  614. if (s->block_switch[ch]) {
  615. do_imdct_256(s, ch);
  616. } else {
  617. s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
  618. s->transform_coeffs[ch], s->tmp_imdct);
  619. }
  620. /* For the first half of the block, apply the window, add the delay
  621. from the previous block, and send to output */
  622. s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
  623. s->window, s->delay[ch-1], 0, 256, 1);
  624. /* For the second half of the block, apply the window and store the
  625. samples to delay, to be combined with the next block */
  626. s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
  627. s->window, 256);
  628. }
  629. }
  630. /**
  631. * Downmix the output to mono or stereo.
  632. */
  633. static void ac3_downmix(AC3DecodeContext *s,
  634. float samples[AC3_MAX_CHANNELS][256], int ch_offset)
  635. {
  636. int i, j;
  637. float v0, v1;
  638. for(i=0; i<256; i++) {
  639. v0 = v1 = 0.0f;
  640. for(j=0; j<s->fbw_channels; j++) {
  641. v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
  642. v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
  643. }
  644. v0 *= s->downmix_coeff_adjust[0];
  645. v1 *= s->downmix_coeff_adjust[1];
  646. if(s->output_mode == AC3_CHMODE_MONO) {
  647. samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
  648. } else if(s->output_mode == AC3_CHMODE_STEREO) {
  649. samples[ ch_offset][i] = v0;
  650. samples[1+ch_offset][i] = v1;
  651. }
  652. }
  653. }
  654. /**
  655. * Upmix delay samples from stereo to original channel layout.
  656. */
  657. static void ac3_upmix_delay(AC3DecodeContext *s)
  658. {
  659. int channel_data_size = sizeof(s->delay[0]);
  660. switch(s->channel_mode) {
  661. case AC3_CHMODE_DUALMONO:
  662. case AC3_CHMODE_STEREO:
  663. /* upmix mono to stereo */
  664. memcpy(s->delay[1], s->delay[0], channel_data_size);
  665. break;
  666. case AC3_CHMODE_2F2R:
  667. memset(s->delay[3], 0, channel_data_size);
  668. case AC3_CHMODE_2F1R:
  669. memset(s->delay[2], 0, channel_data_size);
  670. break;
  671. case AC3_CHMODE_3F2R:
  672. memset(s->delay[4], 0, channel_data_size);
  673. case AC3_CHMODE_3F1R:
  674. memset(s->delay[3], 0, channel_data_size);
  675. case AC3_CHMODE_3F:
  676. memcpy(s->delay[2], s->delay[1], channel_data_size);
  677. memset(s->delay[1], 0, channel_data_size);
  678. break;
  679. }
  680. }
  681. /**
  682. * Parse an audio block from AC-3 bitstream.
  683. */
  684. static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
  685. {
  686. int fbw_channels = s->fbw_channels;
  687. int channel_mode = s->channel_mode;
  688. int i, bnd, seg, ch;
  689. int different_transforms;
  690. int downmix_output;
  691. GetBitContext *gbc = &s->gbc;
  692. uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
  693. memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
  694. /* block switch flags */
  695. different_transforms = 0;
  696. for (ch = 1; ch <= fbw_channels; ch++) {
  697. s->block_switch[ch] = get_bits1(gbc);
  698. if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
  699. different_transforms = 1;
  700. }
  701. /* dithering flags */
  702. s->dither_all = 1;
  703. for (ch = 1; ch <= fbw_channels; ch++) {
  704. s->dither_flag[ch] = get_bits1(gbc);
  705. if(!s->dither_flag[ch])
  706. s->dither_all = 0;
  707. }
  708. /* dynamic range */
  709. i = !(s->channel_mode);
  710. do {
  711. if(get_bits1(gbc)) {
  712. s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
  713. s->avctx->drc_scale)+1.0;
  714. } else if(blk == 0) {
  715. s->dynamic_range[i] = 1.0f;
  716. }
  717. } while(i--);
  718. /* coupling strategy */
  719. if (get_bits1(gbc)) {
  720. memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
  721. s->cpl_in_use = get_bits1(gbc);
  722. if (s->cpl_in_use) {
  723. /* coupling in use */
  724. int cpl_begin_freq, cpl_end_freq;
  725. if (channel_mode < AC3_CHMODE_STEREO) {
  726. av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
  727. return -1;
  728. }
  729. /* determine which channels are coupled */
  730. for (ch = 1; ch <= fbw_channels; ch++)
  731. s->channel_in_cpl[ch] = get_bits1(gbc);
  732. /* phase flags in use */
  733. if (channel_mode == AC3_CHMODE_STEREO)
  734. s->phase_flags_in_use = get_bits1(gbc);
  735. /* coupling frequency range and band structure */
  736. cpl_begin_freq = get_bits(gbc, 4);
  737. cpl_end_freq = get_bits(gbc, 4);
  738. if (3 + cpl_end_freq - cpl_begin_freq < 0) {
  739. av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
  740. return -1;
  741. }
  742. s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
  743. s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
  744. s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
  745. for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
  746. if (get_bits1(gbc)) {
  747. s->cpl_band_struct[bnd] = 1;
  748. s->num_cpl_bands--;
  749. }
  750. }
  751. s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
  752. } else {
  753. /* coupling not in use */
  754. for (ch = 1; ch <= fbw_channels; ch++)
  755. s->channel_in_cpl[ch] = 0;
  756. }
  757. } else if (!blk) {
  758. av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
  759. return -1;
  760. }
  761. /* coupling coordinates */
  762. if (s->cpl_in_use) {
  763. int cpl_coords_exist = 0;
  764. for (ch = 1; ch <= fbw_channels; ch++) {
  765. if (s->channel_in_cpl[ch]) {
  766. if (get_bits1(gbc)) {
  767. int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
  768. cpl_coords_exist = 1;
  769. master_cpl_coord = 3 * get_bits(gbc, 2);
  770. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  771. cpl_coord_exp = get_bits(gbc, 4);
  772. cpl_coord_mant = get_bits(gbc, 4);
  773. if (cpl_coord_exp == 15)
  774. s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
  775. else
  776. s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
  777. s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
  778. }
  779. } else if (!blk) {
  780. av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
  781. return -1;
  782. }
  783. }
  784. }
  785. /* phase flags */
  786. if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
  787. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  788. s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
  789. }
  790. }
  791. }
  792. /* stereo rematrixing strategy and band structure */
  793. if (channel_mode == AC3_CHMODE_STEREO) {
  794. if (get_bits1(gbc)) {
  795. s->num_rematrixing_bands = 4;
  796. if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
  797. s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
  798. for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
  799. s->rematrixing_flags[bnd] = get_bits1(gbc);
  800. } else if (!blk) {
  801. av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
  802. return -1;
  803. }
  804. }
  805. /* exponent strategies for each channel */
  806. s->exp_strategy[CPL_CH] = EXP_REUSE;
  807. s->exp_strategy[s->lfe_ch] = EXP_REUSE;
  808. for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
  809. s->exp_strategy[ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
  810. if(s->exp_strategy[ch] != EXP_REUSE)
  811. bit_alloc_stages[ch] = 3;
  812. }
  813. /* channel bandwidth */
  814. for (ch = 1; ch <= fbw_channels; ch++) {
  815. s->start_freq[ch] = 0;
  816. if (s->exp_strategy[ch] != EXP_REUSE) {
  817. int group_size;
  818. int prev = s->end_freq[ch];
  819. if (s->channel_in_cpl[ch])
  820. s->end_freq[ch] = s->start_freq[CPL_CH];
  821. else {
  822. int bandwidth_code = get_bits(gbc, 6);
  823. if (bandwidth_code > 60) {
  824. av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
  825. return -1;
  826. }
  827. s->end_freq[ch] = bandwidth_code * 3 + 73;
  828. }
  829. group_size = 3 << (s->exp_strategy[ch] - 1);
  830. s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
  831. if(blk > 0 && s->end_freq[ch] != prev)
  832. memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
  833. }
  834. }
  835. if (s->cpl_in_use && s->exp_strategy[CPL_CH] != EXP_REUSE) {
  836. s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
  837. (3 << (s->exp_strategy[CPL_CH] - 1));
  838. }
  839. /* decode exponents for each channel */
  840. for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
  841. if (s->exp_strategy[ch] != EXP_REUSE) {
  842. s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
  843. decode_exponents(gbc, s->exp_strategy[ch],
  844. s->num_exp_groups[ch], s->dexps[ch][0],
  845. &s->dexps[ch][s->start_freq[ch]+!!ch]);
  846. if(ch != CPL_CH && ch != s->lfe_ch)
  847. skip_bits(gbc, 2); /* skip gainrng */
  848. }
  849. }
  850. /* bit allocation information */
  851. if (get_bits1(gbc)) {
  852. s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
  853. s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
  854. s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
  855. s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
  856. s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
  857. for(ch=!s->cpl_in_use; ch<=s->channels; ch++)
  858. bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
  859. } else if (!blk) {
  860. av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
  861. return -1;
  862. }
  863. /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
  864. if (get_bits1(gbc)) {
  865. int csnr;
  866. csnr = (get_bits(gbc, 6) - 15) << 4;
  867. for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
  868. s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
  869. s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
  870. }
  871. memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
  872. } else if (!blk) {
  873. av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
  874. return -1;
  875. }
  876. /* coupling leak information */
  877. if (s->cpl_in_use) {
  878. if (get_bits1(gbc)) {
  879. s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
  880. s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
  881. bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
  882. } else if (!blk) {
  883. av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
  884. return -1;
  885. }
  886. }
  887. /* delta bit allocation information */
  888. if (get_bits1(gbc)) {
  889. /* delta bit allocation exists (strategy) */
  890. for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
  891. s->dba_mode[ch] = get_bits(gbc, 2);
  892. if (s->dba_mode[ch] == DBA_RESERVED) {
  893. av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
  894. return -1;
  895. }
  896. bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
  897. }
  898. /* channel delta offset, len and bit allocation */
  899. for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
  900. if (s->dba_mode[ch] == DBA_NEW) {
  901. s->dba_nsegs[ch] = get_bits(gbc, 3);
  902. for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
  903. s->dba_offsets[ch][seg] = get_bits(gbc, 5);
  904. s->dba_lengths[ch][seg] = get_bits(gbc, 4);
  905. s->dba_values[ch][seg] = get_bits(gbc, 3);
  906. }
  907. /* run last 2 bit allocation stages if new dba values */
  908. bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
  909. }
  910. }
  911. } else if(blk == 0) {
  912. for(ch=0; ch<=s->channels; ch++) {
  913. s->dba_mode[ch] = DBA_NONE;
  914. }
  915. }
  916. /* Bit allocation */
  917. for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
  918. if(bit_alloc_stages[ch] > 2) {
  919. /* Exponent mapping into PSD and PSD integration */
  920. ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
  921. s->start_freq[ch], s->end_freq[ch],
  922. s->psd[ch], s->band_psd[ch]);
  923. }
  924. if(bit_alloc_stages[ch] > 1) {
  925. /* Compute excitation function, Compute masking curve, and
  926. Apply delta bit allocation */
  927. ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
  928. s->start_freq[ch], s->end_freq[ch],
  929. s->fast_gain[ch], (ch == s->lfe_ch),
  930. s->dba_mode[ch], s->dba_nsegs[ch],
  931. s->dba_offsets[ch], s->dba_lengths[ch],
  932. s->dba_values[ch], s->mask[ch]);
  933. }
  934. if(bit_alloc_stages[ch] > 0) {
  935. /* Compute bit allocation */
  936. ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
  937. s->start_freq[ch], s->end_freq[ch],
  938. s->snr_offset[ch],
  939. s->bit_alloc_params.floor,
  940. s->bap[ch]);
  941. }
  942. }
  943. /* unused dummy data */
  944. if (get_bits1(gbc)) {
  945. int skipl = get_bits(gbc, 9);
  946. while(skipl--)
  947. skip_bits(gbc, 8);
  948. }
  949. /* unpack the transform coefficients
  950. this also uncouples channels if coupling is in use. */
  951. get_transform_coeffs(s);
  952. /* recover coefficients if rematrixing is in use */
  953. if(s->channel_mode == AC3_CHMODE_STEREO)
  954. do_rematrixing(s);
  955. /* apply scaling to coefficients (headroom, dynrng) */
  956. for(ch=1; ch<=s->channels; ch++) {
  957. float gain = s->mul_bias / 4194304.0f;
  958. if(s->channel_mode == AC3_CHMODE_DUALMONO) {
  959. gain *= s->dynamic_range[ch-1];
  960. } else {
  961. gain *= s->dynamic_range[0];
  962. }
  963. for(i=0; i<256; i++) {
  964. s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
  965. }
  966. }
  967. /* downmix and MDCT. order depends on whether block switching is used for
  968. any channel in this block. this is because coefficients for the long
  969. and short transforms cannot be mixed. */
  970. downmix_output = s->channels != s->out_channels &&
  971. !((s->output_mode & AC3_OUTPUT_LFEON) &&
  972. s->fbw_channels == s->out_channels);
  973. if(different_transforms) {
  974. /* the delay samples have already been downmixed, so we upmix the delay
  975. samples in order to reconstruct all channels before downmixing. */
  976. if(s->downmixed) {
  977. s->downmixed = 0;
  978. ac3_upmix_delay(s);
  979. }
  980. do_imdct(s, s->channels);
  981. if(downmix_output) {
  982. ac3_downmix(s, s->output, 0);
  983. }
  984. } else {
  985. if(downmix_output) {
  986. ac3_downmix(s, s->transform_coeffs, 1);
  987. }
  988. if(!s->downmixed) {
  989. s->downmixed = 1;
  990. ac3_downmix(s, s->delay, 0);
  991. }
  992. do_imdct(s, s->out_channels);
  993. }
  994. /* convert float to 16-bit integer */
  995. for(ch=0; ch<s->out_channels; ch++) {
  996. for(i=0; i<256; i++) {
  997. s->output[ch][i] += s->add_bias;
  998. }
  999. s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
  1000. }
  1001. return 0;
  1002. }
  1003. /**
  1004. * Decode a single AC-3 frame.
  1005. */
  1006. static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
  1007. const uint8_t *buf, int buf_size)
  1008. {
  1009. AC3DecodeContext *s = avctx->priv_data;
  1010. int16_t *out_samples = (int16_t *)data;
  1011. int i, blk, ch, err;
  1012. /* initialize the GetBitContext with the start of valid AC-3 Frame */
  1013. if (s->input_buffer) {
  1014. /* copy input buffer to decoder context to avoid reading past the end
  1015. of the buffer, which can be caused by a damaged input stream. */
  1016. memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
  1017. init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
  1018. } else {
  1019. init_get_bits(&s->gbc, buf, buf_size * 8);
  1020. }
  1021. /* parse the syncinfo */
  1022. *data_size = 0;
  1023. err = ac3_parse_header(s);
  1024. /* check that reported frame size fits in input buffer */
  1025. if(s->frame_size > buf_size) {
  1026. av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
  1027. err = AC3_PARSE_ERROR_FRAME_SIZE;
  1028. }
  1029. /* check for crc mismatch */
  1030. if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
  1031. if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
  1032. av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
  1033. err = AC3_PARSE_ERROR_CRC;
  1034. }
  1035. }
  1036. if(err && err != AC3_PARSE_ERROR_CRC) {
  1037. switch(err) {
  1038. case AC3_PARSE_ERROR_SYNC:
  1039. av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
  1040. return -1;
  1041. case AC3_PARSE_ERROR_BSID:
  1042. av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
  1043. break;
  1044. case AC3_PARSE_ERROR_SAMPLE_RATE:
  1045. av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
  1046. break;
  1047. case AC3_PARSE_ERROR_FRAME_SIZE:
  1048. av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
  1049. break;
  1050. case AC3_PARSE_ERROR_FRAME_TYPE:
  1051. av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
  1052. break;
  1053. default:
  1054. av_log(avctx, AV_LOG_ERROR, "invalid header\n");
  1055. break;
  1056. }
  1057. }
  1058. /* if frame is ok, set audio parameters */
  1059. if (!err) {
  1060. avctx->sample_rate = s->sample_rate;
  1061. avctx->bit_rate = s->bit_rate;
  1062. /* channel config */
  1063. s->out_channels = s->channels;
  1064. s->output_mode = s->channel_mode;
  1065. if(s->lfe_on)
  1066. s->output_mode |= AC3_OUTPUT_LFEON;
  1067. if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
  1068. avctx->request_channels < s->channels) {
  1069. s->out_channels = avctx->request_channels;
  1070. s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
  1071. }
  1072. avctx->channels = s->out_channels;
  1073. /* set downmixing coefficients if needed */
  1074. if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
  1075. s->fbw_channels == s->out_channels)) {
  1076. set_downmix_coeffs(s);
  1077. }
  1078. } else if (!s->out_channels) {
  1079. s->out_channels = avctx->channels;
  1080. if(s->out_channels < s->channels)
  1081. s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
  1082. }
  1083. /* parse the audio blocks */
  1084. for (blk = 0; blk < NB_BLOCKS; blk++) {
  1085. if (!err && ac3_parse_audio_block(s, blk)) {
  1086. av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
  1087. }
  1088. /* interleave output samples */
  1089. for (i = 0; i < 256; i++)
  1090. for (ch = 0; ch < s->out_channels; ch++)
  1091. *(out_samples++) = s->int_output[ch][i];
  1092. }
  1093. *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
  1094. return s->frame_size;
  1095. }
  1096. /**
  1097. * Uninitialize the AC-3 decoder.
  1098. */
  1099. static av_cold int ac3_decode_end(AVCodecContext *avctx)
  1100. {
  1101. AC3DecodeContext *s = avctx->priv_data;
  1102. ff_mdct_end(&s->imdct_512);
  1103. ff_mdct_end(&s->imdct_256);
  1104. av_freep(&s->input_buffer);
  1105. return 0;
  1106. }
  1107. AVCodec ac3_decoder = {
  1108. .name = "ac3",
  1109. .type = CODEC_TYPE_AUDIO,
  1110. .id = CODEC_ID_AC3,
  1111. .priv_data_size = sizeof (AC3DecodeContext),
  1112. .init = ac3_decode_init,
  1113. .close = ac3_decode_end,
  1114. .decode = ac3_decode_frame,
  1115. .long_name = "ATSC A/52 / AC-3",
  1116. };