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>
7 * This file is part of Libav.
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.
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.
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
26 * AC-3 encoder float/fixed template
34 /* prototypes for static functions in ac3enc_fixed.c and ac3enc_float.c */
36 static void scale_coefficients(AC3EncodeContext *s);
38 static void apply_window(DSPContext *dsp, SampleType *output,
39 const SampleType *input, const SampleType *window,
42 static int normalize_samples(AC3EncodeContext *s);
44 static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
46 static void sum_square_butterfly(AC3EncodeContext *s, CoefSumType sum[4],
47 const CoefType *coef0, const CoefType *coef1,
50 int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
54 FF_ALLOC_OR_GOTO(s->avctx, s->windowed_samples, AC3_WINDOW_SIZE *
55 sizeof(*s->windowed_samples), alloc_fail);
56 FF_ALLOC_OR_GOTO(s->avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
58 for (ch = 0; ch < s->channels; ch++) {
59 FF_ALLOCZ_OR_GOTO(s->avctx, s->planar_samples[ch],
60 (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
66 return AVERROR(ENOMEM);
71 * Deinterleave input samples.
72 * Channels are reordered from Libav's default order to AC-3 order.
74 static void deinterleave_input_samples(AC3EncodeContext *s,
75 const SampleType *samples)
79 /* deinterleave and remap input samples */
80 for (ch = 0; ch < s->channels; ch++) {
81 const SampleType *sptr;
84 /* copy last 256 samples of previous frame to the start of the current frame */
85 memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_FRAME_SIZE],
86 AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
90 sptr = samples + s->channel_map[ch];
91 for (i = AC3_BLOCK_SIZE; i < AC3_FRAME_SIZE+AC3_BLOCK_SIZE; i++) {
92 s->planar_samples[ch][i] = *sptr;
100 * Apply the MDCT to input samples to generate frequency coefficients.
101 * This applies the KBD window and normalizes the input to reduce precision
102 * loss due to fixed-point calculations.
104 static void apply_mdct(AC3EncodeContext *s)
108 for (ch = 0; ch < s->channels; ch++) {
109 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
110 AC3Block *block = &s->blocks[blk];
111 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
113 apply_window(&s->dsp, s->windowed_samples, input_samples,
114 s->mdct->window, AC3_WINDOW_SIZE);
117 block->coeff_shift[ch+1] = normalize_samples(s);
119 s->mdct->fft.mdct_calcw(&s->mdct->fft, block->mdct_coef[ch+1],
120 s->windowed_samples);
127 * Calculate a single coupling coordinate.
129 static inline float calc_cpl_coord(float energy_ch, float energy_cpl)
133 coord *= sqrtf(energy_ch / energy_cpl);
139 * Calculate coupling channel and coupling coordinates.
140 * TODO: Currently this is only used for the floating-point encoder. I was
141 * able to make it work for the fixed-point encoder, but quality was
142 * generally lower in most cases than not using coupling. If a more
143 * adaptive coupling strategy were to be implemented it might be useful
144 * at that time to use coupling for the fixed-point encoder as well.
146 static void apply_channel_coupling(AC3EncodeContext *s)
148 #if CONFIG_AC3ENC_FLOAT
149 LOCAL_ALIGNED_16(float, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
150 LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
151 int blk, ch, bnd, i, j;
152 CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
153 int cpl_start, num_cpl_coefs;
155 memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
156 memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*fixed_cpl_coords));
158 /* align start to 16-byte boundary. align length to multiple of 32.
159 note: coupling start bin % 4 will always be 1 */
160 cpl_start = s->start_freq[CPL_CH] - 1;
161 num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
162 cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
164 /* calculate coupling channel from fbw channels */
165 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
166 AC3Block *block = &s->blocks[blk];
167 CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
168 if (!block->cpl_in_use)
170 memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
171 for (ch = 1; ch <= s->fbw_channels; ch++) {
172 CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
173 if (!block->channel_in_cpl[ch])
175 for (i = 0; i < num_cpl_coefs; i++)
176 cpl_coef[i] += ch_coef[i];
179 /* coefficients must be clipped in order to be encoded */
180 clip_coefficients(&s->dsp, cpl_coef, num_cpl_coefs);
182 /* scale coupling coefficients from float to 24-bit fixed-point */
183 s->ac3dsp.float_to_fixed24(&block->fixed_coef[CPL_CH][cpl_start],
184 cpl_coef, num_cpl_coefs);
187 /* calculate energy in each band in coupling channel and each fbw channel */
188 /* TODO: possibly use SIMD to speed up energy calculation */
190 i = s->start_freq[CPL_CH];
191 while (i < s->cpl_end_freq) {
192 int band_size = s->cpl_band_sizes[bnd];
193 for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
194 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
195 AC3Block *block = &s->blocks[blk];
196 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
198 for (j = 0; j < band_size; j++) {
199 CoefType v = block->mdct_coef[ch][i+j];
200 MAC_COEF(energy[blk][ch][bnd], v, v);
208 /* determine which blocks to send new coupling coordinates for */
209 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
210 AC3Block *block = &s->blocks[blk];
211 AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
213 CoefSumType coord_diff[AC3_MAX_CHANNELS] = {0,};
215 if (block->cpl_in_use) {
216 /* calculate coupling coordinates for all blocks and calculate the
217 average difference between coordinates in successive blocks */
218 for (ch = 1; ch <= s->fbw_channels; ch++) {
219 if (!block->channel_in_cpl[ch])
222 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
223 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
224 energy[blk][CPL_CH][bnd]);
225 if (blk > 0 && block0->cpl_in_use &&
226 block0->channel_in_cpl[ch]) {
227 coord_diff[ch] += fabs(cpl_coords[blk-1][ch][bnd] -
228 cpl_coords[blk ][ch][bnd]);
231 coord_diff[ch] /= s->num_cpl_bands;
234 /* send new coordinates if this is the first block, if previous
235 * block did not use coupling but this block does, the channels
236 * using coupling has changed from the previous block, or the
237 * coordinate difference from the last block for any channel is
238 * greater than a threshold value. */
241 } else if (!block0->cpl_in_use) {
244 for (ch = 1; ch <= s->fbw_channels; ch++) {
245 if (block->channel_in_cpl[ch] && !block0->channel_in_cpl[ch]) {
251 for (ch = 1; ch <= s->fbw_channels; ch++) {
252 if (block->channel_in_cpl[ch] && coord_diff[ch] > 0.04) {
260 block->new_cpl_coords = new_coords;
263 /* calculate final coupling coordinates, taking into account reusing of
264 coordinates in successive blocks */
265 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
267 while (blk < AC3_MAX_BLOCKS) {
269 CoefSumType energy_cpl;
270 AC3Block *block = &s->blocks[blk];
272 if (!block->cpl_in_use) {
277 energy_cpl = energy[blk][CPL_CH][bnd];
279 while (!s->blocks[blk1].new_cpl_coords && blk1 < AC3_MAX_BLOCKS) {
280 if (s->blocks[blk1].cpl_in_use)
281 energy_cpl += energy[blk1][CPL_CH][bnd];
285 for (ch = 1; ch <= s->fbw_channels; ch++) {
287 if (!block->channel_in_cpl[ch])
289 energy_ch = energy[blk][ch][bnd];
291 while (!s->blocks[blk1].new_cpl_coords && blk1 < AC3_MAX_BLOCKS) {
292 if (s->blocks[blk1].cpl_in_use)
293 energy_ch += energy[blk1][ch][bnd];
296 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
302 /* calculate exponents/mantissas for coupling coordinates */
303 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
304 AC3Block *block = &s->blocks[blk];
305 if (!block->cpl_in_use || !block->new_cpl_coords)
308 clip_coefficients(&s->dsp, cpl_coords[blk][1], s->fbw_channels * 16);
309 s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
311 s->fbw_channels * 16);
312 s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
313 fixed_cpl_coords[blk][1],
314 s->fbw_channels * 16);
316 for (ch = 1; ch <= s->fbw_channels; ch++) {
317 int bnd, min_exp, max_exp, master_exp;
319 /* determine master exponent */
320 min_exp = max_exp = block->cpl_coord_exp[ch][0];
321 for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
322 int exp = block->cpl_coord_exp[ch][bnd];
323 min_exp = FFMIN(exp, min_exp);
324 max_exp = FFMAX(exp, max_exp);
326 master_exp = ((max_exp - 15) + 2) / 3;
327 master_exp = FFMAX(master_exp, 0);
328 while (min_exp < master_exp * 3)
330 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
331 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
332 master_exp * 3, 0, 15);
334 block->cpl_master_exp[ch] = master_exp;
336 /* quantize mantissas */
337 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
338 int cpl_exp = block->cpl_coord_exp[ch][bnd];
339 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
345 block->cpl_coord_mant[ch][bnd] = cpl_mant;
350 if (CONFIG_EAC3_ENCODER && s->eac3)
351 ff_eac3_set_cpl_states(s);
352 #endif /* CONFIG_AC3ENC_FLOAT */
357 * Determine rematrixing flags for each block and band.
359 static void compute_rematrixing_strategy(AC3EncodeContext *s)
363 AC3Block *block, *av_uninit(block0);
365 if (s->channel_mode != AC3_CHMODE_STEREO)
368 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
369 block = &s->blocks[blk];
370 block->new_rematrixing_strategy = !blk;
372 if (!s->rematrixing_enabled) {
377 block->num_rematrixing_bands = 4;
378 if (block->cpl_in_use) {
379 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
380 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
381 if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
382 block->new_rematrixing_strategy = 1;
384 nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
386 for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
387 /* calculate calculate sum of squared coeffs for one band in one block */
388 int start = ff_ac3_rematrix_band_tab[bnd];
389 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
391 sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
392 block->mdct_coef[2] + start, end - start);
394 /* compare sums to determine if rematrixing will be used for this band */
395 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
396 block->rematrixing_flags[bnd] = 1;
398 block->rematrixing_flags[bnd] = 0;
400 /* determine if new rematrixing flags will be sent */
402 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
403 block->new_rematrixing_strategy = 1;
412 * Encode a single AC-3 frame.
414 int AC3_NAME(encode_frame)(AVCodecContext *avctx, unsigned char *frame,
415 int buf_size, void *data)
417 AC3EncodeContext *s = avctx->priv_data;
418 const SampleType *samples = data;
421 if (!s->eac3 && s->options.allow_per_frame_metadata) {
422 ret = ff_ac3_validate_metadata(avctx);
427 if (s->bit_alloc.sr_code == 1 || s->eac3)
428 ff_ac3_adjust_frame_size(s);
430 deinterleave_input_samples(s, samples);
435 scale_coefficients(s);
437 clip_coefficients(&s->dsp, s->blocks[0].mdct_coef[1],
438 AC3_MAX_COEFS * AC3_MAX_BLOCKS * s->channels);
440 s->cpl_on = s->cpl_enabled;
441 ff_ac3_compute_coupling_strategy(s);
444 apply_channel_coupling(s);
446 compute_rematrixing_strategy(s);
449 scale_coefficients(s);
451 ff_ac3_apply_rematrixing(s);
453 ff_ac3_process_exponents(s);
455 ret = ff_ac3_compute_bit_allocation(s);
457 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
461 ff_ac3_quantize_mantissas(s);
463 ff_ac3_output_frame(s, frame);
465 return s->frame_size;