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 FFmpeg.
9 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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
31 #include "libavutil/attributes.h"
32 #include "libavutil/internal.h"
33 #include "libavutil/mem_internal.h"
41 int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
45 if (!FF_ALLOC_TYPED_ARRAY(s->windowed_samples, AC3_WINDOW_SIZE) ||
46 !FF_ALLOCZ_TYPED_ARRAY(s->planar_samples, s->channels))
47 return AVERROR(ENOMEM);
49 for (ch = 0; ch < s->channels; ch++) {
50 if (!(s->planar_samples[ch] = av_mallocz((AC3_FRAME_SIZE + AC3_BLOCK_SIZE) *
51 sizeof(**s->planar_samples))))
52 return AVERROR(ENOMEM);
60 * Channels are reordered from FFmpeg's default order to AC-3 order.
62 static void copy_input_samples(AC3EncodeContext *s, SampleType **samples)
66 /* copy and remap input samples */
67 for (ch = 0; ch < s->channels; ch++) {
68 /* copy last 256 samples of previous frame to the start of the current frame */
69 memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
70 AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
72 /* copy new samples for current frame */
73 memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE],
74 samples[s->channel_map[ch]],
75 AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0]));
81 * Apply the MDCT to input samples to generate frequency coefficients.
82 * This applies the KBD window and normalizes the input to reduce precision
83 * loss due to fixed-point calculations.
85 static void apply_mdct(AC3EncodeContext *s)
89 for (ch = 0; ch < s->channels; ch++) {
90 for (blk = 0; blk < s->num_blocks; blk++) {
91 AC3Block *block = &s->blocks[blk];
92 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
94 #if CONFIG_AC3ENC_FLOAT
95 s->fdsp->vector_fmul(s->windowed_samples, input_samples,
96 s->mdct_window, AC3_WINDOW_SIZE);
98 s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples,
99 s->mdct_window, AC3_WINDOW_SIZE);
102 block->coeff_shift[ch+1] = normalize_samples(s);
105 s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
106 s->windowed_samples);
113 * Calculate coupling channel and coupling coordinates.
115 static void apply_channel_coupling(AC3EncodeContext *s)
117 LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
118 #if CONFIG_AC3ENC_FLOAT
119 LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
121 int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
123 int av_uninit(blk), ch, bnd, i, j;
124 CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
125 int cpl_start, num_cpl_coefs;
127 memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
128 #if CONFIG_AC3ENC_FLOAT
129 memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
132 /* align start to 16-byte boundary. align length to multiple of 32.
133 note: coupling start bin % 4 will always be 1 */
134 cpl_start = s->start_freq[CPL_CH] - 1;
135 num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
136 cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
138 /* calculate coupling channel from fbw channels */
139 for (blk = 0; blk < s->num_blocks; blk++) {
140 AC3Block *block = &s->blocks[blk];
141 CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
142 if (!block->cpl_in_use)
144 memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
145 for (ch = 1; ch <= s->fbw_channels; ch++) {
146 CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
147 if (!block->channel_in_cpl[ch])
149 for (i = 0; i < num_cpl_coefs; i++)
150 cpl_coef[i] += ch_coef[i];
153 /* coefficients must be clipped in order to be encoded */
154 clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs);
157 /* calculate energy in each band in coupling channel and each fbw channel */
158 /* TODO: possibly use SIMD to speed up energy calculation */
160 i = s->start_freq[CPL_CH];
161 while (i < s->cpl_end_freq) {
162 int band_size = s->cpl_band_sizes[bnd];
163 for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
164 for (blk = 0; blk < s->num_blocks; blk++) {
165 AC3Block *block = &s->blocks[blk];
166 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
168 for (j = 0; j < band_size; j++) {
169 CoefType v = block->mdct_coef[ch][i+j];
170 MAC_COEF(energy[blk][ch][bnd], v, v);
178 /* calculate coupling coordinates for all blocks for all channels */
179 for (blk = 0; blk < s->num_blocks; blk++) {
180 AC3Block *block = &s->blocks[blk];
181 if (!block->cpl_in_use)
183 for (ch = 1; ch <= s->fbw_channels; ch++) {
184 if (!block->channel_in_cpl[ch])
186 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
187 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
188 energy[blk][CPL_CH][bnd]);
193 /* determine which blocks to send new coupling coordinates for */
194 for (blk = 0; blk < s->num_blocks; blk++) {
195 AC3Block *block = &s->blocks[blk];
196 AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
198 memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
200 if (block->cpl_in_use) {
201 /* send new coordinates if this is the first block, if previous
202 * block did not use coupling but this block does, the channels
203 * using coupling has changed from the previous block, or the
204 * coordinate difference from the last block for any channel is
205 * greater than a threshold value. */
206 if (blk == 0 || !block0->cpl_in_use) {
207 for (ch = 1; ch <= s->fbw_channels; ch++)
208 block->new_cpl_coords[ch] = 1;
210 for (ch = 1; ch <= s->fbw_channels; ch++) {
211 if (!block->channel_in_cpl[ch])
213 if (!block0->channel_in_cpl[ch]) {
214 block->new_cpl_coords[ch] = 1;
216 CoefSumType coord_diff = 0;
217 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
218 coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
219 cpl_coords[blk ][ch][bnd]);
221 coord_diff /= s->num_cpl_bands;
222 if (coord_diff > NEW_CPL_COORD_THRESHOLD)
223 block->new_cpl_coords[ch] = 1;
230 /* calculate final coupling coordinates, taking into account reusing of
231 coordinates in successive blocks */
232 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
234 while (blk < s->num_blocks) {
236 AC3Block *block = &s->blocks[blk];
238 if (!block->cpl_in_use) {
243 for (ch = 1; ch <= s->fbw_channels; ch++) {
244 CoefSumType energy_ch, energy_cpl;
245 if (!block->channel_in_cpl[ch])
247 energy_cpl = energy[blk][CPL_CH][bnd];
248 energy_ch = energy[blk][ch][bnd];
250 while (blk1 < s->num_blocks && !s->blocks[blk1].new_cpl_coords[ch]) {
251 if (s->blocks[blk1].cpl_in_use) {
252 energy_cpl += energy[blk1][CPL_CH][bnd];
253 energy_ch += energy[blk1][ch][bnd];
257 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
263 /* calculate exponents/mantissas for coupling coordinates */
264 for (blk = 0; blk < s->num_blocks; blk++) {
265 AC3Block *block = &s->blocks[blk];
266 if (!block->cpl_in_use)
269 #if CONFIG_AC3ENC_FLOAT
270 s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
272 s->fbw_channels * 16);
274 s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
275 fixed_cpl_coords[blk][1],
276 s->fbw_channels * 16);
278 for (ch = 1; ch <= s->fbw_channels; ch++) {
279 int bnd, min_exp, max_exp, master_exp;
281 if (!block->new_cpl_coords[ch])
284 /* determine master exponent */
285 min_exp = max_exp = block->cpl_coord_exp[ch][0];
286 for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
287 int exp = block->cpl_coord_exp[ch][bnd];
288 min_exp = FFMIN(exp, min_exp);
289 max_exp = FFMAX(exp, max_exp);
291 master_exp = ((max_exp - 15) + 2) / 3;
292 master_exp = FFMAX(master_exp, 0);
293 while (min_exp < master_exp * 3)
295 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
296 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
297 master_exp * 3, 0, 15);
299 block->cpl_master_exp[ch] = master_exp;
301 /* quantize mantissas */
302 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
303 int cpl_exp = block->cpl_coord_exp[ch][bnd];
304 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
310 block->cpl_coord_mant[ch][bnd] = cpl_mant;
315 if (CONFIG_EAC3_ENCODER && s->eac3)
316 ff_eac3_set_cpl_states(s);
321 * Determine rematrixing flags for each block and band.
323 static void compute_rematrixing_strategy(AC3EncodeContext *s)
327 AC3Block *block, *block0 = NULL;
329 if (s->channel_mode != AC3_CHMODE_STEREO)
332 for (blk = 0; blk < s->num_blocks; blk++) {
333 block = &s->blocks[blk];
334 block->new_rematrixing_strategy = !blk;
336 block->num_rematrixing_bands = 4;
337 if (block->cpl_in_use) {
338 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
339 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
340 if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
341 block->new_rematrixing_strategy = 1;
343 nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
345 if (!s->rematrixing_enabled) {
350 for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
351 /* calculate sum of squared coeffs for one band in one block */
352 int start = ff_ac3_rematrix_band_tab[bnd];
353 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
355 sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
356 block->mdct_coef[2] + start, end - start);
358 /* compare sums to determine if rematrixing will be used for this band */
359 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
360 block->rematrixing_flags[bnd] = 1;
362 block->rematrixing_flags[bnd] = 0;
364 /* determine if new rematrixing flags will be sent */
366 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
367 block->new_rematrixing_strategy = 1;
375 int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
376 const AVFrame *frame, int *got_packet_ptr)
378 AC3EncodeContext *s = avctx->priv_data;
381 if (s->options.allow_per_frame_metadata) {
382 ret = ff_ac3_validate_metadata(s);
387 if (s->bit_alloc.sr_code == 1 || s->eac3)
388 ff_ac3_adjust_frame_size(s);
390 copy_input_samples(s, (SampleType **)frame->extended_data);
395 scale_coefficients(s);
397 clip_coefficients(&s->adsp, s->blocks[0].mdct_coef[1],
398 AC3_MAX_COEFS * s->num_blocks * s->channels);
400 s->cpl_on = s->cpl_enabled;
401 ff_ac3_compute_coupling_strategy(s);
404 apply_channel_coupling(s);
406 compute_rematrixing_strategy(s);
409 scale_coefficients(s);
411 ff_ac3_apply_rematrixing(s);
413 ff_ac3_process_exponents(s);
415 ret = ff_ac3_compute_bit_allocation(s);
417 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
421 ff_ac3_group_exponents(s);
423 ff_ac3_quantize_mantissas(s);
425 if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size, 0)) < 0)
427 ff_ac3_output_frame(s, avpkt->data);
429 if (frame->pts != AV_NOPTS_VALUE)
430 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding);