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"
40 int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
44 FF_ALLOC_OR_GOTO(s->avctx, s->windowed_samples, AC3_WINDOW_SIZE *
45 sizeof(*s->windowed_samples), alloc_fail);
46 FF_ALLOC_ARRAY_OR_GOTO(s->avctx, s->planar_samples, s->channels, sizeof(*s->planar_samples),
48 for (ch = 0; ch < s->channels; ch++) {
49 FF_ALLOCZ_OR_GOTO(s->avctx, s->planar_samples[ch],
50 (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
56 return AVERROR(ENOMEM);
62 * Channels are reordered from FFmpeg's default order to AC-3 order.
64 static void copy_input_samples(AC3EncodeContext *s, SampleType **samples)
68 /* copy and remap input samples */
69 for (ch = 0; ch < s->channels; ch++) {
70 /* copy last 256 samples of previous frame to the start of the current frame */
71 memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
72 AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
74 /* copy new samples for current frame */
75 memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE],
76 samples[s->channel_map[ch]],
77 AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0]));
83 * Apply the MDCT to input samples to generate frequency coefficients.
84 * This applies the KBD window and normalizes the input to reduce precision
85 * loss due to fixed-point calculations.
87 static void apply_mdct(AC3EncodeContext *s)
91 for (ch = 0; ch < s->channels; ch++) {
92 for (blk = 0; blk < s->num_blocks; blk++) {
93 AC3Block *block = &s->blocks[blk];
94 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
96 #if CONFIG_AC3ENC_FLOAT
97 s->fdsp->vector_fmul(s->windowed_samples, input_samples,
98 s->mdct_window, AC3_WINDOW_SIZE);
100 s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples,
101 s->mdct_window, AC3_WINDOW_SIZE);
104 block->coeff_shift[ch+1] = normalize_samples(s);
107 s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
108 s->windowed_samples);
115 * Calculate coupling channel and coupling coordinates.
117 static void apply_channel_coupling(AC3EncodeContext *s)
119 LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
120 #if CONFIG_AC3ENC_FLOAT
121 LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
123 int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
125 int av_uninit(blk), ch, bnd, i, j;
126 CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
127 int cpl_start, num_cpl_coefs;
129 memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
130 #if CONFIG_AC3ENC_FLOAT
131 memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
134 /* align start to 16-byte boundary. align length to multiple of 32.
135 note: coupling start bin % 4 will always be 1 */
136 cpl_start = s->start_freq[CPL_CH] - 1;
137 num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
138 cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
140 /* calculate coupling channel from fbw channels */
141 for (blk = 0; blk < s->num_blocks; blk++) {
142 AC3Block *block = &s->blocks[blk];
143 CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
144 if (!block->cpl_in_use)
146 memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
147 for (ch = 1; ch <= s->fbw_channels; ch++) {
148 CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
149 if (!block->channel_in_cpl[ch])
151 for (i = 0; i < num_cpl_coefs; i++)
152 cpl_coef[i] += ch_coef[i];
155 /* coefficients must be clipped in order to be encoded */
156 clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs);
159 /* calculate energy in each band in coupling channel and each fbw channel */
160 /* TODO: possibly use SIMD to speed up energy calculation */
162 i = s->start_freq[CPL_CH];
163 while (i < s->cpl_end_freq) {
164 int band_size = s->cpl_band_sizes[bnd];
165 for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
166 for (blk = 0; blk < s->num_blocks; blk++) {
167 AC3Block *block = &s->blocks[blk];
168 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
170 for (j = 0; j < band_size; j++) {
171 CoefType v = block->mdct_coef[ch][i+j];
172 MAC_COEF(energy[blk][ch][bnd], v, v);
180 /* calculate coupling coordinates for all blocks for all channels */
181 for (blk = 0; blk < s->num_blocks; blk++) {
182 AC3Block *block = &s->blocks[blk];
183 if (!block->cpl_in_use)
185 for (ch = 1; ch <= s->fbw_channels; ch++) {
186 if (!block->channel_in_cpl[ch])
188 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
189 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
190 energy[blk][CPL_CH][bnd]);
195 /* determine which blocks to send new coupling coordinates for */
196 for (blk = 0; blk < s->num_blocks; blk++) {
197 AC3Block *block = &s->blocks[blk];
198 AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
200 memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
202 if (block->cpl_in_use) {
203 /* send new coordinates if this is the first block, if previous
204 * block did not use coupling but this block does, the channels
205 * using coupling has changed from the previous block, or the
206 * coordinate difference from the last block for any channel is
207 * greater than a threshold value. */
208 if (blk == 0 || !block0->cpl_in_use) {
209 for (ch = 1; ch <= s->fbw_channels; ch++)
210 block->new_cpl_coords[ch] = 1;
212 for (ch = 1; ch <= s->fbw_channels; ch++) {
213 if (!block->channel_in_cpl[ch])
215 if (!block0->channel_in_cpl[ch]) {
216 block->new_cpl_coords[ch] = 1;
218 CoefSumType coord_diff = 0;
219 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
220 coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
221 cpl_coords[blk ][ch][bnd]);
223 coord_diff /= s->num_cpl_bands;
224 if (coord_diff > NEW_CPL_COORD_THRESHOLD)
225 block->new_cpl_coords[ch] = 1;
232 /* calculate final coupling coordinates, taking into account reusing of
233 coordinates in successive blocks */
234 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
236 while (blk < s->num_blocks) {
238 AC3Block *block = &s->blocks[blk];
240 if (!block->cpl_in_use) {
245 for (ch = 1; ch <= s->fbw_channels; ch++) {
246 CoefSumType energy_ch, energy_cpl;
247 if (!block->channel_in_cpl[ch])
249 energy_cpl = energy[blk][CPL_CH][bnd];
250 energy_ch = energy[blk][ch][bnd];
252 while (blk1 < s->num_blocks && !s->blocks[blk1].new_cpl_coords[ch]) {
253 if (s->blocks[blk1].cpl_in_use) {
254 energy_cpl += energy[blk1][CPL_CH][bnd];
255 energy_ch += energy[blk1][ch][bnd];
259 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
265 /* calculate exponents/mantissas for coupling coordinates */
266 for (blk = 0; blk < s->num_blocks; blk++) {
267 AC3Block *block = &s->blocks[blk];
268 if (!block->cpl_in_use)
271 #if CONFIG_AC3ENC_FLOAT
272 s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
274 s->fbw_channels * 16);
276 s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
277 fixed_cpl_coords[blk][1],
278 s->fbw_channels * 16);
280 for (ch = 1; ch <= s->fbw_channels; ch++) {
281 int bnd, min_exp, max_exp, master_exp;
283 if (!block->new_cpl_coords[ch])
286 /* determine master exponent */
287 min_exp = max_exp = block->cpl_coord_exp[ch][0];
288 for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
289 int exp = block->cpl_coord_exp[ch][bnd];
290 min_exp = FFMIN(exp, min_exp);
291 max_exp = FFMAX(exp, max_exp);
293 master_exp = ((max_exp - 15) + 2) / 3;
294 master_exp = FFMAX(master_exp, 0);
295 while (min_exp < master_exp * 3)
297 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
298 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
299 master_exp * 3, 0, 15);
301 block->cpl_master_exp[ch] = master_exp;
303 /* quantize mantissas */
304 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
305 int cpl_exp = block->cpl_coord_exp[ch][bnd];
306 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
312 block->cpl_coord_mant[ch][bnd] = cpl_mant;
317 if (CONFIG_EAC3_ENCODER && s->eac3)
318 ff_eac3_set_cpl_states(s);
323 * Determine rematrixing flags for each block and band.
325 static void compute_rematrixing_strategy(AC3EncodeContext *s)
329 AC3Block *block, *block0 = NULL;
331 if (s->channel_mode != AC3_CHMODE_STEREO)
334 for (blk = 0; blk < s->num_blocks; blk++) {
335 block = &s->blocks[blk];
336 block->new_rematrixing_strategy = !blk;
338 block->num_rematrixing_bands = 4;
339 if (block->cpl_in_use) {
340 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
341 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
342 if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
343 block->new_rematrixing_strategy = 1;
345 nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
347 if (!s->rematrixing_enabled) {
352 for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
353 /* calculate sum of squared coeffs for one band in one block */
354 int start = ff_ac3_rematrix_band_tab[bnd];
355 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
357 sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
358 block->mdct_coef[2] + start, end - start);
360 /* compare sums to determine if rematrixing will be used for this band */
361 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
362 block->rematrixing_flags[bnd] = 1;
364 block->rematrixing_flags[bnd] = 0;
366 /* determine if new rematrixing flags will be sent */
368 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
369 block->new_rematrixing_strategy = 1;
377 int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
378 const AVFrame *frame, int *got_packet_ptr)
380 AC3EncodeContext *s = avctx->priv_data;
383 if (s->options.allow_per_frame_metadata) {
384 ret = ff_ac3_validate_metadata(s);
389 if (s->bit_alloc.sr_code == 1 || s->eac3)
390 ff_ac3_adjust_frame_size(s);
392 copy_input_samples(s, (SampleType **)frame->extended_data);
397 scale_coefficients(s);
399 clip_coefficients(&s->adsp, s->blocks[0].mdct_coef[1],
400 AC3_MAX_COEFS * s->num_blocks * s->channels);
402 s->cpl_on = s->cpl_enabled;
403 ff_ac3_compute_coupling_strategy(s);
406 apply_channel_coupling(s);
408 compute_rematrixing_strategy(s);
411 scale_coefficients(s);
413 ff_ac3_apply_rematrixing(s);
415 ff_ac3_process_exponents(s);
417 ret = ff_ac3_compute_bit_allocation(s);
419 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
423 ff_ac3_group_exponents(s);
425 ff_ac3_quantize_mantissas(s);
427 if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size, 0)) < 0)
429 ff_ac3_output_frame(s, avpkt->data);
431 if (frame->pts != AV_NOPTS_VALUE)
432 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding);