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 if (!FF_ALLOC_TYPED_ARRAY(s->windowed_samples, AC3_WINDOW_SIZE) ||
45 !FF_ALLOCZ_TYPED_ARRAY(s->planar_samples, s->channels))
46 return AVERROR(ENOMEM);
48 for (ch = 0; ch < s->channels; ch++) {
49 if (!(s->planar_samples[ch] = av_mallocz((AC3_FRAME_SIZE + AC3_BLOCK_SIZE) *
50 sizeof(**s->planar_samples))))
51 return AVERROR(ENOMEM);
59 * Channels are reordered from FFmpeg's default order to AC-3 order.
61 static void copy_input_samples(AC3EncodeContext *s, SampleType **samples)
65 /* copy and remap input samples */
66 for (ch = 0; ch < s->channels; ch++) {
67 /* copy last 256 samples of previous frame to the start of the current frame */
68 memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
69 AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
71 /* copy new samples for current frame */
72 memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE],
73 samples[s->channel_map[ch]],
74 AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0]));
80 * Apply the MDCT to input samples to generate frequency coefficients.
81 * This applies the KBD window and normalizes the input to reduce precision
82 * loss due to fixed-point calculations.
84 static void apply_mdct(AC3EncodeContext *s)
88 for (ch = 0; ch < s->channels; ch++) {
89 for (blk = 0; blk < s->num_blocks; blk++) {
90 AC3Block *block = &s->blocks[blk];
91 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
93 #if CONFIG_AC3ENC_FLOAT
94 s->fdsp->vector_fmul(s->windowed_samples, input_samples,
95 s->mdct_window, AC3_WINDOW_SIZE);
97 s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples,
98 s->mdct_window, AC3_WINDOW_SIZE);
101 block->coeff_shift[ch+1] = normalize_samples(s);
104 s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
105 s->windowed_samples);
112 * Calculate coupling channel and coupling coordinates.
114 static void apply_channel_coupling(AC3EncodeContext *s)
116 LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
117 #if CONFIG_AC3ENC_FLOAT
118 LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
120 int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
122 int av_uninit(blk), ch, bnd, i, j;
123 CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
124 int cpl_start, num_cpl_coefs;
126 memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
127 #if CONFIG_AC3ENC_FLOAT
128 memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
131 /* align start to 16-byte boundary. align length to multiple of 32.
132 note: coupling start bin % 4 will always be 1 */
133 cpl_start = s->start_freq[CPL_CH] - 1;
134 num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
135 cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
137 /* calculate coupling channel from fbw channels */
138 for (blk = 0; blk < s->num_blocks; blk++) {
139 AC3Block *block = &s->blocks[blk];
140 CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
141 if (!block->cpl_in_use)
143 memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
144 for (ch = 1; ch <= s->fbw_channels; ch++) {
145 CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
146 if (!block->channel_in_cpl[ch])
148 for (i = 0; i < num_cpl_coefs; i++)
149 cpl_coef[i] += ch_coef[i];
152 /* coefficients must be clipped in order to be encoded */
153 clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs);
156 /* calculate energy in each band in coupling channel and each fbw channel */
157 /* TODO: possibly use SIMD to speed up energy calculation */
159 i = s->start_freq[CPL_CH];
160 while (i < s->cpl_end_freq) {
161 int band_size = s->cpl_band_sizes[bnd];
162 for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
163 for (blk = 0; blk < s->num_blocks; blk++) {
164 AC3Block *block = &s->blocks[blk];
165 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
167 for (j = 0; j < band_size; j++) {
168 CoefType v = block->mdct_coef[ch][i+j];
169 MAC_COEF(energy[blk][ch][bnd], v, v);
177 /* calculate coupling coordinates for all blocks for all channels */
178 for (blk = 0; blk < s->num_blocks; blk++) {
179 AC3Block *block = &s->blocks[blk];
180 if (!block->cpl_in_use)
182 for (ch = 1; ch <= s->fbw_channels; ch++) {
183 if (!block->channel_in_cpl[ch])
185 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
186 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
187 energy[blk][CPL_CH][bnd]);
192 /* determine which blocks to send new coupling coordinates for */
193 for (blk = 0; blk < s->num_blocks; blk++) {
194 AC3Block *block = &s->blocks[blk];
195 AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
197 memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
199 if (block->cpl_in_use) {
200 /* send new coordinates if this is the first block, if previous
201 * block did not use coupling but this block does, the channels
202 * using coupling has changed from the previous block, or the
203 * coordinate difference from the last block for any channel is
204 * greater than a threshold value. */
205 if (blk == 0 || !block0->cpl_in_use) {
206 for (ch = 1; ch <= s->fbw_channels; ch++)
207 block->new_cpl_coords[ch] = 1;
209 for (ch = 1; ch <= s->fbw_channels; ch++) {
210 if (!block->channel_in_cpl[ch])
212 if (!block0->channel_in_cpl[ch]) {
213 block->new_cpl_coords[ch] = 1;
215 CoefSumType coord_diff = 0;
216 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
217 coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
218 cpl_coords[blk ][ch][bnd]);
220 coord_diff /= s->num_cpl_bands;
221 if (coord_diff > NEW_CPL_COORD_THRESHOLD)
222 block->new_cpl_coords[ch] = 1;
229 /* calculate final coupling coordinates, taking into account reusing of
230 coordinates in successive blocks */
231 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
233 while (blk < s->num_blocks) {
235 AC3Block *block = &s->blocks[blk];
237 if (!block->cpl_in_use) {
242 for (ch = 1; ch <= s->fbw_channels; ch++) {
243 CoefSumType energy_ch, energy_cpl;
244 if (!block->channel_in_cpl[ch])
246 energy_cpl = energy[blk][CPL_CH][bnd];
247 energy_ch = energy[blk][ch][bnd];
249 while (blk1 < s->num_blocks && !s->blocks[blk1].new_cpl_coords[ch]) {
250 if (s->blocks[blk1].cpl_in_use) {
251 energy_cpl += energy[blk1][CPL_CH][bnd];
252 energy_ch += energy[blk1][ch][bnd];
256 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
262 /* calculate exponents/mantissas for coupling coordinates */
263 for (blk = 0; blk < s->num_blocks; blk++) {
264 AC3Block *block = &s->blocks[blk];
265 if (!block->cpl_in_use)
268 #if CONFIG_AC3ENC_FLOAT
269 s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
271 s->fbw_channels * 16);
273 s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
274 fixed_cpl_coords[blk][1],
275 s->fbw_channels * 16);
277 for (ch = 1; ch <= s->fbw_channels; ch++) {
278 int bnd, min_exp, max_exp, master_exp;
280 if (!block->new_cpl_coords[ch])
283 /* determine master exponent */
284 min_exp = max_exp = block->cpl_coord_exp[ch][0];
285 for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
286 int exp = block->cpl_coord_exp[ch][bnd];
287 min_exp = FFMIN(exp, min_exp);
288 max_exp = FFMAX(exp, max_exp);
290 master_exp = ((max_exp - 15) + 2) / 3;
291 master_exp = FFMAX(master_exp, 0);
292 while (min_exp < master_exp * 3)
294 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
295 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
296 master_exp * 3, 0, 15);
298 block->cpl_master_exp[ch] = master_exp;
300 /* quantize mantissas */
301 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
302 int cpl_exp = block->cpl_coord_exp[ch][bnd];
303 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
309 block->cpl_coord_mant[ch][bnd] = cpl_mant;
314 if (CONFIG_EAC3_ENCODER && s->eac3)
315 ff_eac3_set_cpl_states(s);
320 * Determine rematrixing flags for each block and band.
322 static void compute_rematrixing_strategy(AC3EncodeContext *s)
326 AC3Block *block, *block0 = NULL;
328 if (s->channel_mode != AC3_CHMODE_STEREO)
331 for (blk = 0; blk < s->num_blocks; blk++) {
332 block = &s->blocks[blk];
333 block->new_rematrixing_strategy = !blk;
335 block->num_rematrixing_bands = 4;
336 if (block->cpl_in_use) {
337 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
338 block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
339 if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
340 block->new_rematrixing_strategy = 1;
342 nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
344 if (!s->rematrixing_enabled) {
349 for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
350 /* calculate sum of squared coeffs for one band in one block */
351 int start = ff_ac3_rematrix_band_tab[bnd];
352 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
354 sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
355 block->mdct_coef[2] + start, end - start);
357 /* compare sums to determine if rematrixing will be used for this band */
358 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
359 block->rematrixing_flags[bnd] = 1;
361 block->rematrixing_flags[bnd] = 0;
363 /* determine if new rematrixing flags will be sent */
365 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
366 block->new_rematrixing_strategy = 1;
374 int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
375 const AVFrame *frame, int *got_packet_ptr)
377 AC3EncodeContext *s = avctx->priv_data;
380 if (s->options.allow_per_frame_metadata) {
381 ret = ff_ac3_validate_metadata(s);
386 if (s->bit_alloc.sr_code == 1 || s->eac3)
387 ff_ac3_adjust_frame_size(s);
389 copy_input_samples(s, (SampleType **)frame->extended_data);
394 scale_coefficients(s);
396 clip_coefficients(&s->adsp, s->blocks[0].mdct_coef[1],
397 AC3_MAX_COEFS * s->num_blocks * s->channels);
399 s->cpl_on = s->cpl_enabled;
400 ff_ac3_compute_coupling_strategy(s);
403 apply_channel_coupling(s);
405 compute_rematrixing_strategy(s);
408 scale_coefficients(s);
410 ff_ac3_apply_rematrixing(s);
412 ff_ac3_process_exponents(s);
414 ret = ff_ac3_compute_bit_allocation(s);
416 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
420 ff_ac3_group_exponents(s);
422 ff_ac3_quantize_mantissas(s);
424 if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size, 0)) < 0)
426 ff_ac3_output_frame(s, avpkt->data);
428 if (frame->pts != AV_NOPTS_VALUE)
429 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding);