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.

465 lines
16KB

  1. /*
  2. * AC-3 encoder float/fixed template
  3. * Copyright (c) 2000 Fabrice Bellard
  4. * Copyright (c) 2006-2011 Justin Ruggles <justin.ruggles@gmail.com>
  5. * Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de>
  6. *
  7. * This file is part of Libav.
  8. *
  9. * Libav is free software; you can redistribute it and/or
  10. * modify it under the terms of the GNU Lesser General Public
  11. * License as published by the Free Software Foundation; either
  12. * version 2.1 of the License, or (at your option) any later version.
  13. *
  14. * Libav is distributed in the hope that it will be useful,
  15. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  17. * Lesser General Public License for more details.
  18. *
  19. * You should have received a copy of the GNU Lesser General Public
  20. * License along with Libav; if not, write to the Free Software
  21. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22. */
  23. /**
  24. * @file
  25. * AC-3 encoder float/fixed template
  26. */
  27. #include <stdint.h>
  28. /* prototypes for static functions in ac3enc_fixed.c and ac3enc_float.c */
  29. static void scale_coefficients(AC3EncodeContext *s);
  30. static void apply_window(DSPContext *dsp, SampleType *output,
  31. const SampleType *input, const SampleType *window,
  32. unsigned int len);
  33. static int normalize_samples(AC3EncodeContext *s);
  34. static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
  35. int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
  36. {
  37. int ch;
  38. FF_ALLOC_OR_GOTO(s->avctx, s->windowed_samples, AC3_WINDOW_SIZE *
  39. sizeof(*s->windowed_samples), alloc_fail);
  40. FF_ALLOC_OR_GOTO(s->avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
  41. alloc_fail);
  42. for (ch = 0; ch < s->channels; ch++) {
  43. FF_ALLOCZ_OR_GOTO(s->avctx, s->planar_samples[ch],
  44. (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
  45. alloc_fail);
  46. }
  47. return 0;
  48. alloc_fail:
  49. return AVERROR(ENOMEM);
  50. }
  51. /**
  52. * Deinterleave input samples.
  53. * Channels are reordered from Libav's default order to AC-3 order.
  54. */
  55. static void deinterleave_input_samples(AC3EncodeContext *s,
  56. const SampleType *samples)
  57. {
  58. int ch, i;
  59. /* deinterleave and remap input samples */
  60. for (ch = 0; ch < s->channels; ch++) {
  61. const SampleType *sptr;
  62. int sinc;
  63. /* copy last 256 samples of previous frame to the start of the current frame */
  64. memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
  65. AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
  66. /* deinterleave */
  67. sinc = s->channels;
  68. sptr = samples + s->channel_map[ch];
  69. for (i = AC3_BLOCK_SIZE; i < AC3_BLOCK_SIZE * (s->num_blocks + 1); i++) {
  70. s->planar_samples[ch][i] = *sptr;
  71. sptr += sinc;
  72. }
  73. }
  74. }
  75. /**
  76. * Apply the MDCT to input samples to generate frequency coefficients.
  77. * This applies the KBD window and normalizes the input to reduce precision
  78. * loss due to fixed-point calculations.
  79. */
  80. static void apply_mdct(AC3EncodeContext *s)
  81. {
  82. int blk, ch;
  83. for (ch = 0; ch < s->channels; ch++) {
  84. for (blk = 0; blk < s->num_blocks; blk++) {
  85. AC3Block *block = &s->blocks[blk];
  86. const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
  87. apply_window(&s->dsp, s->windowed_samples, input_samples,
  88. s->mdct_window, AC3_WINDOW_SIZE);
  89. if (s->fixed_point)
  90. block->coeff_shift[ch+1] = normalize_samples(s);
  91. s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
  92. s->windowed_samples);
  93. }
  94. }
  95. }
  96. /**
  97. * Calculate a single coupling coordinate.
  98. */
  99. static inline float calc_cpl_coord(float energy_ch, float energy_cpl)
  100. {
  101. float coord = 0.125;
  102. if (energy_cpl > 0)
  103. coord *= sqrtf(energy_ch / energy_cpl);
  104. return FFMIN(coord, COEF_MAX);
  105. }
  106. /**
  107. * Calculate coupling channel and coupling coordinates.
  108. * TODO: Currently this is only used for the floating-point encoder. I was
  109. * able to make it work for the fixed-point encoder, but quality was
  110. * generally lower in most cases than not using coupling. If a more
  111. * adaptive coupling strategy were to be implemented it might be useful
  112. * at that time to use coupling for the fixed-point encoder as well.
  113. */
  114. static void apply_channel_coupling(AC3EncodeContext *s)
  115. {
  116. #if CONFIG_AC3ENC_FLOAT
  117. LOCAL_ALIGNED_16(float, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
  118. LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
  119. int blk, ch, bnd, i, j;
  120. CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
  121. int cpl_start, num_cpl_coefs;
  122. memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
  123. memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*fixed_cpl_coords));
  124. /* align start to 16-byte boundary. align length to multiple of 32.
  125. note: coupling start bin % 4 will always be 1 */
  126. cpl_start = s->start_freq[CPL_CH] - 1;
  127. num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
  128. cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
  129. /* calculate coupling channel from fbw channels */
  130. for (blk = 0; blk < s->num_blocks; blk++) {
  131. AC3Block *block = &s->blocks[blk];
  132. CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
  133. if (!block->cpl_in_use)
  134. continue;
  135. memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
  136. for (ch = 1; ch <= s->fbw_channels; ch++) {
  137. CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
  138. if (!block->channel_in_cpl[ch])
  139. continue;
  140. for (i = 0; i < num_cpl_coefs; i++)
  141. cpl_coef[i] += ch_coef[i];
  142. }
  143. /* coefficients must be clipped in order to be encoded */
  144. clip_coefficients(&s->dsp, cpl_coef, num_cpl_coefs);
  145. /* scale coupling coefficients from float to 24-bit fixed-point */
  146. s->ac3dsp.float_to_fixed24(&block->fixed_coef[CPL_CH][cpl_start],
  147. cpl_coef, num_cpl_coefs);
  148. }
  149. /* calculate energy in each band in coupling channel and each fbw channel */
  150. /* TODO: possibly use SIMD to speed up energy calculation */
  151. bnd = 0;
  152. i = s->start_freq[CPL_CH];
  153. while (i < s->cpl_end_freq) {
  154. int band_size = s->cpl_band_sizes[bnd];
  155. for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
  156. for (blk = 0; blk < s->num_blocks; blk++) {
  157. AC3Block *block = &s->blocks[blk];
  158. if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
  159. continue;
  160. for (j = 0; j < band_size; j++) {
  161. CoefType v = block->mdct_coef[ch][i+j];
  162. MAC_COEF(energy[blk][ch][bnd], v, v);
  163. }
  164. }
  165. }
  166. i += band_size;
  167. bnd++;
  168. }
  169. /* calculate coupling coordinates for all blocks for all channels */
  170. for (blk = 0; blk < s->num_blocks; blk++) {
  171. AC3Block *block = &s->blocks[blk];
  172. if (!block->cpl_in_use)
  173. continue;
  174. for (ch = 1; ch <= s->fbw_channels; ch++) {
  175. if (!block->channel_in_cpl[ch])
  176. continue;
  177. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  178. cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
  179. energy[blk][CPL_CH][bnd]);
  180. }
  181. }
  182. }
  183. /* determine which blocks to send new coupling coordinates for */
  184. for (blk = 0; blk < s->num_blocks; blk++) {
  185. AC3Block *block = &s->blocks[blk];
  186. AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
  187. memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
  188. if (block->cpl_in_use) {
  189. /* send new coordinates if this is the first block, if previous
  190. * block did not use coupling but this block does, the channels
  191. * using coupling has changed from the previous block, or the
  192. * coordinate difference from the last block for any channel is
  193. * greater than a threshold value. */
  194. if (blk == 0 || !block0->cpl_in_use) {
  195. for (ch = 1; ch <= s->fbw_channels; ch++)
  196. block->new_cpl_coords[ch] = 1;
  197. } else {
  198. for (ch = 1; ch <= s->fbw_channels; ch++) {
  199. if (!block->channel_in_cpl[ch])
  200. continue;
  201. if (!block0->channel_in_cpl[ch]) {
  202. block->new_cpl_coords[ch] = 1;
  203. } else {
  204. CoefSumType coord_diff = 0;
  205. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  206. coord_diff += fabs(cpl_coords[blk-1][ch][bnd] -
  207. cpl_coords[blk ][ch][bnd]);
  208. }
  209. coord_diff /= s->num_cpl_bands;
  210. if (coord_diff > 0.03)
  211. block->new_cpl_coords[ch] = 1;
  212. }
  213. }
  214. }
  215. }
  216. }
  217. /* calculate final coupling coordinates, taking into account reusing of
  218. coordinates in successive blocks */
  219. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  220. blk = 0;
  221. while (blk < s->num_blocks) {
  222. int av_uninit(blk1);
  223. AC3Block *block = &s->blocks[blk];
  224. if (!block->cpl_in_use) {
  225. blk++;
  226. continue;
  227. }
  228. for (ch = 1; ch <= s->fbw_channels; ch++) {
  229. CoefSumType energy_ch, energy_cpl;
  230. if (!block->channel_in_cpl[ch])
  231. continue;
  232. energy_cpl = energy[blk][CPL_CH][bnd];
  233. energy_ch = energy[blk][ch][bnd];
  234. blk1 = blk+1;
  235. while (!s->blocks[blk1].new_cpl_coords[ch] && blk1 < s->num_blocks) {
  236. if (s->blocks[blk1].cpl_in_use) {
  237. energy_cpl += energy[blk1][CPL_CH][bnd];
  238. energy_ch += energy[blk1][ch][bnd];
  239. }
  240. blk1++;
  241. }
  242. cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
  243. }
  244. blk = blk1;
  245. }
  246. }
  247. /* calculate exponents/mantissas for coupling coordinates */
  248. for (blk = 0; blk < s->num_blocks; blk++) {
  249. AC3Block *block = &s->blocks[blk];
  250. if (!block->cpl_in_use)
  251. continue;
  252. s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
  253. cpl_coords[blk][1],
  254. s->fbw_channels * 16);
  255. s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
  256. fixed_cpl_coords[blk][1],
  257. s->fbw_channels * 16);
  258. for (ch = 1; ch <= s->fbw_channels; ch++) {
  259. int bnd, min_exp, max_exp, master_exp;
  260. if (!block->new_cpl_coords[ch])
  261. continue;
  262. /* determine master exponent */
  263. min_exp = max_exp = block->cpl_coord_exp[ch][0];
  264. for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
  265. int exp = block->cpl_coord_exp[ch][bnd];
  266. min_exp = FFMIN(exp, min_exp);
  267. max_exp = FFMAX(exp, max_exp);
  268. }
  269. master_exp = ((max_exp - 15) + 2) / 3;
  270. master_exp = FFMAX(master_exp, 0);
  271. while (min_exp < master_exp * 3)
  272. master_exp--;
  273. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  274. block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
  275. master_exp * 3, 0, 15);
  276. }
  277. block->cpl_master_exp[ch] = master_exp;
  278. /* quantize mantissas */
  279. for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
  280. int cpl_exp = block->cpl_coord_exp[ch][bnd];
  281. int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
  282. if (cpl_exp == 15)
  283. cpl_mant >>= 1;
  284. else
  285. cpl_mant -= 16;
  286. block->cpl_coord_mant[ch][bnd] = cpl_mant;
  287. }
  288. }
  289. }
  290. if (CONFIG_EAC3_ENCODER && s->eac3)
  291. ff_eac3_set_cpl_states(s);
  292. #endif /* CONFIG_AC3ENC_FLOAT */
  293. }
  294. /**
  295. * Determine rematrixing flags for each block and band.
  296. */
  297. static void compute_rematrixing_strategy(AC3EncodeContext *s)
  298. {
  299. int nb_coefs;
  300. int blk, bnd, i;
  301. AC3Block *block, *av_uninit(block0);
  302. if (s->channel_mode != AC3_CHMODE_STEREO)
  303. return;
  304. for (blk = 0; blk < s->num_blocks; blk++) {
  305. block = &s->blocks[blk];
  306. block->new_rematrixing_strategy = !blk;
  307. if (!s->rematrixing_enabled) {
  308. block0 = block;
  309. continue;
  310. }
  311. block->num_rematrixing_bands = 4;
  312. if (block->cpl_in_use) {
  313. block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
  314. block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
  315. if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
  316. block->new_rematrixing_strategy = 1;
  317. }
  318. nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
  319. for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
  320. /* calculate calculate sum of squared coeffs for one band in one block */
  321. int start = ff_ac3_rematrix_band_tab[bnd];
  322. int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
  323. CoefSumType sum[4] = {0,};
  324. for (i = start; i < end; i++) {
  325. CoefType lt = block->mdct_coef[1][i];
  326. CoefType rt = block->mdct_coef[2][i];
  327. CoefType md = lt + rt;
  328. CoefType sd = lt - rt;
  329. MAC_COEF(sum[0], lt, lt);
  330. MAC_COEF(sum[1], rt, rt);
  331. MAC_COEF(sum[2], md, md);
  332. MAC_COEF(sum[3], sd, sd);
  333. }
  334. /* compare sums to determine if rematrixing will be used for this band */
  335. if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
  336. block->rematrixing_flags[bnd] = 1;
  337. else
  338. block->rematrixing_flags[bnd] = 0;
  339. /* determine if new rematrixing flags will be sent */
  340. if (blk &&
  341. block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
  342. block->new_rematrixing_strategy = 1;
  343. }
  344. }
  345. block0 = block;
  346. }
  347. }
  348. /**
  349. * Encode a single AC-3 frame.
  350. */
  351. int AC3_NAME(encode_frame)(AVCodecContext *avctx, unsigned char *frame,
  352. int buf_size, void *data)
  353. {
  354. AC3EncodeContext *s = avctx->priv_data;
  355. const SampleType *samples = data;
  356. int ret;
  357. if (s->options.allow_per_frame_metadata) {
  358. ret = ff_ac3_validate_metadata(s);
  359. if (ret)
  360. return ret;
  361. }
  362. if (s->bit_alloc.sr_code == 1 || s->eac3)
  363. ff_ac3_adjust_frame_size(s);
  364. deinterleave_input_samples(s, samples);
  365. apply_mdct(s);
  366. if (s->fixed_point)
  367. scale_coefficients(s);
  368. clip_coefficients(&s->dsp, s->blocks[0].mdct_coef[1],
  369. AC3_MAX_COEFS * s->num_blocks * s->channels);
  370. s->cpl_on = s->cpl_enabled;
  371. ff_ac3_compute_coupling_strategy(s);
  372. if (s->cpl_on)
  373. apply_channel_coupling(s);
  374. compute_rematrixing_strategy(s);
  375. if (!s->fixed_point)
  376. scale_coefficients(s);
  377. ff_ac3_apply_rematrixing(s);
  378. ff_ac3_process_exponents(s);
  379. ret = ff_ac3_compute_bit_allocation(s);
  380. if (ret) {
  381. av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
  382. return ret;
  383. }
  384. ff_ac3_group_exponents(s);
  385. ff_ac3_quantize_mantissas(s);
  386. ff_ac3_output_frame(s, frame);
  387. return s->frame_size;
  388. }