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
32 /* prototypes for static functions in ac3enc_fixed.c and ac3enc_float.c */
34 static void scale_coefficients(AC3EncodeContext *s);
36 static void apply_window(void *dsp, SampleType *output,
37 const SampleType *input, const SampleType *window,
40 static int normalize_samples(AC3EncodeContext *s);
42 static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
44 static CoefType calc_cpl_coord(CoefSumType energy_ch, CoefSumType energy_cpl);
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 FFmpeg'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_BLOCK_SIZE * s->num_blocks],
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_BLOCK_SIZE * (s->num_blocks + 1); 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 < s->num_blocks; blk++) {
110 AC3Block *block = &s->blocks[blk];
111 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
113 #if CONFIG_AC3ENC_FLOAT
114 apply_window(&s->fdsp, s->windowed_samples, input_samples,
115 s->mdct_window, AC3_WINDOW_SIZE);
117 apply_window(&s->dsp, s->windowed_samples, input_samples,
118 s->mdct_window, AC3_WINDOW_SIZE);
122 block->coeff_shift[ch+1] = normalize_samples(s);
124 s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
125 s->windowed_samples);
132 * Calculate coupling channel and coupling coordinates.
134 static void apply_channel_coupling(AC3EncodeContext *s)
136 LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
137 #if CONFIG_AC3ENC_FLOAT
138 LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
140 int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
142 int blk, ch, bnd, i, j;
143 CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
144 int cpl_start, num_cpl_coefs;
146 memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
147 #if CONFIG_AC3ENC_FLOAT
148 memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
151 /* align start to 16-byte boundary. align length to multiple of 32.
152 note: coupling start bin % 4 will always be 1 */
153 cpl_start = s->start_freq[CPL_CH] - 1;
154 num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
155 cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
157 /* calculate coupling channel from fbw channels */
158 for (blk = 0; blk < s->num_blocks; blk++) {
159 AC3Block *block = &s->blocks[blk];
160 CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
161 if (!block->cpl_in_use)
163 memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
164 for (ch = 1; ch <= s->fbw_channels; ch++) {
165 CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
166 if (!block->channel_in_cpl[ch])
168 for (i = 0; i < num_cpl_coefs; i++)
169 cpl_coef[i] += ch_coef[i];
172 /* coefficients must be clipped in order to be encoded */
173 clip_coefficients(&s->dsp, cpl_coef, num_cpl_coefs);
176 /* calculate energy in each band in coupling channel and each fbw channel */
177 /* TODO: possibly use SIMD to speed up energy calculation */
179 i = s->start_freq[CPL_CH];
180 while (i < s->cpl_end_freq) {
181 int band_size = s->cpl_band_sizes[bnd];
182 for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
183 for (blk = 0; blk < s->num_blocks; blk++) {
184 AC3Block *block = &s->blocks[blk];
185 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
187 for (j = 0; j < band_size; j++) {
188 CoefType v = block->mdct_coef[ch][i+j];
189 MAC_COEF(energy[blk][ch][bnd], v, v);
197 /* calculate coupling coordinates for all blocks for all channels */
198 for (blk = 0; blk < s->num_blocks; blk++) {
199 AC3Block *block = &s->blocks[blk];
200 if (!block->cpl_in_use)
202 for (ch = 1; ch <= s->fbw_channels; ch++) {
203 if (!block->channel_in_cpl[ch])
205 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
206 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
207 energy[blk][CPL_CH][bnd]);
212 /* determine which blocks to send new coupling coordinates for */
213 for (blk = 0; blk < s->num_blocks; blk++) {
214 AC3Block *block = &s->blocks[blk];
215 AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
217 memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
219 if (block->cpl_in_use) {
220 /* send new coordinates if this is the first block, if previous
221 * block did not use coupling but this block does, the channels
222 * using coupling has changed from the previous block, or the
223 * coordinate difference from the last block for any channel is
224 * greater than a threshold value. */
225 if (blk == 0 || !block0->cpl_in_use) {
226 for (ch = 1; ch <= s->fbw_channels; ch++)
227 block->new_cpl_coords[ch] = 1;
229 for (ch = 1; ch <= s->fbw_channels; ch++) {
230 if (!block->channel_in_cpl[ch])
232 if (!block0->channel_in_cpl[ch]) {
233 block->new_cpl_coords[ch] = 1;
235 CoefSumType coord_diff = 0;
236 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
237 coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
238 cpl_coords[blk ][ch][bnd]);
240 coord_diff /= s->num_cpl_bands;
241 if (coord_diff > NEW_CPL_COORD_THRESHOLD)
242 block->new_cpl_coords[ch] = 1;
249 /* calculate final coupling coordinates, taking into account reusing of
250 coordinates in successive blocks */
251 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
253 while (blk < s->num_blocks) {
255 AC3Block *block = &s->blocks[blk];
257 if (!block->cpl_in_use) {
262 for (ch = 1; ch <= s->fbw_channels; ch++) {
263 CoefSumType energy_ch, energy_cpl;
264 if (!block->channel_in_cpl[ch])
266 energy_cpl = energy[blk][CPL_CH][bnd];
267 energy_ch = energy[blk][ch][bnd];
269 while (!s->blocks[blk1].new_cpl_coords[ch] && blk1 < s->num_blocks) {
270 if (s->blocks[blk1].cpl_in_use) {
271 energy_cpl += energy[blk1][CPL_CH][bnd];
272 energy_ch += energy[blk1][ch][bnd];
276 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
282 /* calculate exponents/mantissas for coupling coordinates */
283 for (blk = 0; blk < s->num_blocks; blk++) {
284 AC3Block *block = &s->blocks[blk];
285 if (!block->cpl_in_use)
288 #if CONFIG_AC3ENC_FLOAT
289 s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
291 s->fbw_channels * 16);
293 s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
294 fixed_cpl_coords[blk][1],
295 s->fbw_channels * 16);
297 for (ch = 1; ch <= s->fbw_channels; ch++) {
298 int bnd, min_exp, max_exp, master_exp;
300 if (!block->new_cpl_coords[ch])
303 /* determine master exponent */
304 min_exp = max_exp = block->cpl_coord_exp[ch][0];
305 for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
306 int exp = block->cpl_coord_exp[ch][bnd];
307 min_exp = FFMIN(exp, min_exp);
308 max_exp = FFMAX(exp, max_exp);
310 master_exp = ((max_exp - 15) + 2) / 3;
311 master_exp = FFMAX(master_exp, 0);
312 while (min_exp < master_exp * 3)
314 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
315 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
316 master_exp * 3, 0, 15);
318 block->cpl_master_exp[ch] = master_exp;
320 /* quantize mantissas */
321 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
322 int cpl_exp = block->cpl_coord_exp[ch][bnd];
323 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
329 block->cpl_coord_mant[ch][bnd] = cpl_mant;
334 if (CONFIG_EAC3_ENCODER && s->eac3)
335 ff_eac3_set_cpl_states(s);
340 * Determine rematrixing flags for each block and band.
342 static void compute_rematrixing_strategy(AC3EncodeContext *s)
346 AC3Block *block, *block0;
348 if (s->channel_mode != AC3_CHMODE_STEREO)
351 for (blk = 0; blk < s->num_blocks; blk++) {
352 block = &s->blocks[blk];
353 block->new_rematrixing_strategy = !blk;
355 block->num_rematrixing_bands = 4;
356 if (block->cpl_in_use) {
357 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
358 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
359 if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
360 block->new_rematrixing_strategy = 1;
362 nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
364 if (!s->rematrixing_enabled) {
369 for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
370 /* calculate calculate sum of squared coeffs for one band in one block */
371 int start = ff_ac3_rematrix_band_tab[bnd];
372 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
374 sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
375 block->mdct_coef[2] + start, end - start);
377 /* compare sums to determine if rematrixing will be used for this band */
378 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
379 block->rematrixing_flags[bnd] = 1;
381 block->rematrixing_flags[bnd] = 0;
383 /* determine if new rematrixing flags will be sent */
385 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
386 block->new_rematrixing_strategy = 1;
394 int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
395 const AVFrame *frame, int *got_packet_ptr)
397 AC3EncodeContext *s = avctx->priv_data;
398 const SampleType *samples = (const SampleType *)frame->data[0];
401 if (s->options.allow_per_frame_metadata) {
402 ret = ff_ac3_validate_metadata(s);
407 if (s->bit_alloc.sr_code == 1 || s->eac3)
408 ff_ac3_adjust_frame_size(s);
410 deinterleave_input_samples(s, samples);
415 scale_coefficients(s);
417 clip_coefficients(&s->dsp, s->blocks[0].mdct_coef[1],
418 AC3_MAX_COEFS * s->num_blocks * s->channels);
420 s->cpl_on = s->cpl_enabled;
421 ff_ac3_compute_coupling_strategy(s);
424 apply_channel_coupling(s);
426 compute_rematrixing_strategy(s);
429 scale_coefficients(s);
431 ff_ac3_apply_rematrixing(s);
433 ff_ac3_process_exponents(s);
435 ret = ff_ac3_compute_bit_allocation(s);
437 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
441 ff_ac3_group_exponents(s);
443 ff_ac3_quantize_mantissas(s);
445 if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size)))
447 ff_ac3_output_frame(s, avpkt->data);
449 if (frame->pts != AV_NOPTS_VALUE)
450 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay);