2 * The simplest AC-3 encoder
3 * Copyright (c) 2000 Fabrice Bellard
4 * Copyright (c) 2006-2010 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 * The simplest AC-3 encoder.
30 //#define ASSERT_LEVEL 2
34 #include "libavutil/audioconvert.h"
35 #include "libavutil/avassert.h"
36 #include "libavutil/crc.h"
37 #include "libavutil/opt.h"
43 #include "audioconvert.h"
47 #ifndef CONFIG_AC3ENC_FLOAT
48 #define CONFIG_AC3ENC_FLOAT 0
52 /** Maximum number of exponent groups. +1 for separate DC exponent. */
53 #define AC3_MAX_EXP_GROUPS 85
55 #if CONFIG_AC3ENC_FLOAT
56 #define MAC_COEF(d,a,b) ((d)+=(a)*(b))
57 typedef float SampleType;
58 typedef float CoefType;
59 typedef float CoefSumType;
61 #define MAC_COEF(d,a,b) MAC64(d,a,b)
62 typedef int16_t SampleType;
63 typedef int32_t CoefType;
64 typedef int64_t CoefSumType;
67 typedef struct AC3MDCTContext {
68 const SampleType *window; ///< MDCT window function
69 FFTContext fft; ///< FFT context for MDCT calculation
73 * Encoding Options used by AVOption.
75 typedef struct AC3EncOptions {
76 /* AC-3 metadata options*/
79 float center_mix_level;
80 float surround_mix_level;
81 int dolby_surround_mode;
82 int audio_production_info;
88 int preferred_stereo_downmix;
89 float ltrt_center_mix_level;
90 float ltrt_surround_mix_level;
91 float loro_center_mix_level;
92 float loro_surround_mix_level;
94 int dolby_surround_ex_mode;
95 int dolby_headphone_mode;
96 int ad_converter_type;
98 /* other encoding options */
99 int allow_per_frame_metadata;
100 int stereo_rematrixing;
104 * Data for a single audio block.
106 typedef struct AC3Block {
107 uint8_t **bap; ///< bit allocation pointers (bap)
108 CoefType **mdct_coef; ///< MDCT coefficients
109 int32_t **fixed_coef; ///< fixed-point MDCT coefficients
110 uint8_t **exp; ///< original exponents
111 uint8_t **grouped_exp; ///< grouped exponents
112 int16_t **psd; ///< psd per frequency bin
113 int16_t **band_psd; ///< psd per critical band
114 int16_t **mask; ///< masking curve
115 uint16_t **qmant; ///< quantized mantissas
116 uint8_t coeff_shift[AC3_MAX_CHANNELS]; ///< fixed-point coefficient shift values
117 uint8_t new_rematrixing_strategy; ///< send new rematrixing flags in this block
118 uint8_t rematrixing_flags[4]; ///< rematrixing flags
119 struct AC3Block *exp_ref_block[AC3_MAX_CHANNELS]; ///< reference blocks for EXP_REUSE
123 * AC-3 encoder private context.
125 typedef struct AC3EncodeContext {
126 AVClass *av_class; ///< AVClass used for AVOption
127 AC3EncOptions options; ///< encoding options
128 PutBitContext pb; ///< bitstream writer context
130 AC3DSPContext ac3dsp; ///< AC-3 optimized functions
131 AC3MDCTContext mdct; ///< MDCT context
133 AC3Block blocks[AC3_MAX_BLOCKS]; ///< per-block info
135 int bitstream_id; ///< bitstream id (bsid)
136 int bitstream_mode; ///< bitstream mode (bsmod)
138 int bit_rate; ///< target bit rate, in bits-per-second
139 int sample_rate; ///< sampling frequency, in Hz
141 int frame_size_min; ///< minimum frame size in case rounding is necessary
142 int frame_size; ///< current frame size in bytes
143 int frame_size_code; ///< frame size code (frmsizecod)
145 int bits_written; ///< bit count (used to avg. bitrate)
146 int samples_written; ///< sample count (used to avg. bitrate)
148 int fbw_channels; ///< number of full-bandwidth channels (nfchans)
149 int channels; ///< total number of channels (nchans)
150 int lfe_on; ///< indicates if there is an LFE channel (lfeon)
151 int lfe_channel; ///< channel index of the LFE channel
152 int has_center; ///< indicates if there is a center channel
153 int has_surround; ///< indicates if there are one or more surround channels
154 int channel_mode; ///< channel mode (acmod)
155 const uint8_t *channel_map; ///< channel map used to reorder channels
157 int center_mix_level; ///< center mix level code
158 int surround_mix_level; ///< surround mix level code
159 int ltrt_center_mix_level; ///< Lt/Rt center mix level code
160 int ltrt_surround_mix_level; ///< Lt/Rt surround mix level code
161 int loro_center_mix_level; ///< Lo/Ro center mix level code
162 int loro_surround_mix_level; ///< Lo/Ro surround mix level code
164 int cutoff; ///< user-specified cutoff frequency, in Hz
165 int bandwidth_code[AC3_MAX_CHANNELS]; ///< bandwidth code (0 to 60) (chbwcod)
166 int nb_coefs[AC3_MAX_CHANNELS];
168 int rematrixing_enabled; ///< stereo rematrixing enabled
169 int num_rematrixing_bands; ///< number of rematrixing bands
171 /* bitrate allocation control */
172 int slow_gain_code; ///< slow gain code (sgaincod)
173 int slow_decay_code; ///< slow decay code (sdcycod)
174 int fast_decay_code; ///< fast decay code (fdcycod)
175 int db_per_bit_code; ///< dB/bit code (dbpbcod)
176 int floor_code; ///< floor code (floorcod)
177 AC3BitAllocParameters bit_alloc; ///< bit allocation parameters
178 int coarse_snr_offset; ///< coarse SNR offsets (csnroffst)
179 int fast_gain_code[AC3_MAX_CHANNELS]; ///< fast gain codes (signal-to-mask ratio) (fgaincod)
180 int fine_snr_offset[AC3_MAX_CHANNELS]; ///< fine SNR offsets (fsnroffst)
181 int frame_bits_fixed; ///< number of non-coefficient bits for fixed parameters
182 int frame_bits; ///< all frame bits except exponents and mantissas
183 int exponent_bits; ///< number of bits used for exponents
185 SampleType **planar_samples;
187 uint8_t *bap1_buffer;
188 CoefType *mdct_coef_buffer;
189 int32_t *fixed_coef_buffer;
191 uint8_t *grouped_exp_buffer;
193 int16_t *band_psd_buffer;
194 int16_t *mask_buffer;
195 uint16_t *qmant_buffer;
197 uint8_t exp_strategy[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; ///< exponent strategies
199 DECLARE_ALIGNED(32, SampleType, windowed_samples)[AC3_WINDOW_SIZE];
202 typedef struct AC3Mant {
203 uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; ///< mantissa pointers for bap=1,2,4
204 int mant1_cnt, mant2_cnt, mant4_cnt; ///< mantissa counts for bap=1,2,4
207 #define CMIXLEV_NUM_OPTIONS 3
208 static const float cmixlev_options[CMIXLEV_NUM_OPTIONS] = {
209 LEVEL_MINUS_3DB, LEVEL_MINUS_4POINT5DB, LEVEL_MINUS_6DB
212 #define SURMIXLEV_NUM_OPTIONS 3
213 static const float surmixlev_options[SURMIXLEV_NUM_OPTIONS] = {
214 LEVEL_MINUS_3DB, LEVEL_MINUS_6DB, LEVEL_ZERO
217 #define EXTMIXLEV_NUM_OPTIONS 8
218 static const float extmixlev_options[EXTMIXLEV_NUM_OPTIONS] = {
219 LEVEL_PLUS_3DB, LEVEL_PLUS_1POINT5DB, LEVEL_ONE, LEVEL_MINUS_4POINT5DB,
220 LEVEL_MINUS_3DB, LEVEL_MINUS_4POINT5DB, LEVEL_MINUS_6DB, LEVEL_ZERO
224 #define OFFSET(param) offsetof(AC3EncodeContext, options.param)
225 #define AC3ENC_PARAM (AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM)
227 static const AVOption options[] = {
228 /* Metadata Options */
229 {"per_frame_metadata", "Allow Changing Metadata Per-Frame", OFFSET(allow_per_frame_metadata), FF_OPT_TYPE_INT, 0, 0, 1, AC3ENC_PARAM},
231 {"center_mixlev", "Center Mix Level", OFFSET(center_mix_level), FF_OPT_TYPE_FLOAT, LEVEL_MINUS_4POINT5DB, 0.0, 1.0, AC3ENC_PARAM},
232 {"surround_mixlev", "Surround Mix Level", OFFSET(surround_mix_level), FF_OPT_TYPE_FLOAT, LEVEL_MINUS_6DB, 0.0, 1.0, AC3ENC_PARAM},
233 /* audio production information */
234 {"mixing_level", "Mixing Level", OFFSET(mixing_level), FF_OPT_TYPE_INT, -1, -1, 111, AC3ENC_PARAM},
235 {"room_type", "Room Type", OFFSET(room_type), FF_OPT_TYPE_INT, -1, -1, 2, AC3ENC_PARAM, "room_type"},
236 {"notindicated", "Not Indicated (default)", 0, FF_OPT_TYPE_CONST, 0, INT_MIN, INT_MAX, AC3ENC_PARAM, "room_type"},
237 {"large", "Large Room", 0, FF_OPT_TYPE_CONST, 1, INT_MIN, INT_MAX, AC3ENC_PARAM, "room_type"},
238 {"small", "Small Room", 0, FF_OPT_TYPE_CONST, 2, INT_MIN, INT_MAX, AC3ENC_PARAM, "room_type"},
239 /* other metadata options */
240 {"copyright", "Copyright Bit", OFFSET(copyright), FF_OPT_TYPE_INT, 0, 0, 1, AC3ENC_PARAM},
241 {"dialnorm", "Dialogue Level (dB)", OFFSET(dialogue_level), FF_OPT_TYPE_INT, -31, -31, -1, AC3ENC_PARAM},
242 {"dsur_mode", "Dolby Surround Mode", OFFSET(dolby_surround_mode), FF_OPT_TYPE_INT, 0, 0, 2, AC3ENC_PARAM, "dsur_mode"},
243 {"notindicated", "Not Indicated (default)", 0, FF_OPT_TYPE_CONST, 0, INT_MIN, INT_MAX, AC3ENC_PARAM, "dsur_mode"},
244 {"on", "Dolby Surround Encoded", 0, FF_OPT_TYPE_CONST, 1, INT_MIN, INT_MAX, AC3ENC_PARAM, "dsur_mode"},
245 {"off", "Not Dolby Surround Encoded", 0, FF_OPT_TYPE_CONST, 2, INT_MIN, INT_MAX, AC3ENC_PARAM, "dsur_mode"},
246 {"original", "Original Bit Stream", OFFSET(original), FF_OPT_TYPE_INT, 1, 0, 1, AC3ENC_PARAM},
247 /* extended bitstream information */
248 {"dmix_mode", "Preferred Stereo Downmix Mode", OFFSET(preferred_stereo_downmix), FF_OPT_TYPE_INT, -1, -1, 2, AC3ENC_PARAM, "dmix_mode"},
249 {"notindicated", "Not Indicated (default)", 0, FF_OPT_TYPE_CONST, 0, INT_MIN, INT_MAX, AC3ENC_PARAM, "dmix_mode"},
250 {"ltrt", "Lt/Rt Downmix Preferred", 0, FF_OPT_TYPE_CONST, 1, INT_MIN, INT_MAX, AC3ENC_PARAM, "dmix_mode"},
251 {"loro", "Lo/Ro Downmix Preferred", 0, FF_OPT_TYPE_CONST, 2, INT_MIN, INT_MAX, AC3ENC_PARAM, "dmix_mode"},
252 {"ltrt_cmixlev", "Lt/Rt Center Mix Level", OFFSET(ltrt_center_mix_level), FF_OPT_TYPE_FLOAT, -1.0, -1.0, 2.0, AC3ENC_PARAM},
253 {"ltrt_surmixlev", "Lt/Rt Surround Mix Level", OFFSET(ltrt_surround_mix_level), FF_OPT_TYPE_FLOAT, -1.0, -1.0, 2.0, AC3ENC_PARAM},
254 {"loro_cmixlev", "Lo/Ro Center Mix Level", OFFSET(loro_center_mix_level), FF_OPT_TYPE_FLOAT, -1.0, -1.0, 2.0, AC3ENC_PARAM},
255 {"loro_surmixlev", "Lo/Ro Surround Mix Level", OFFSET(loro_surround_mix_level), FF_OPT_TYPE_FLOAT, -1.0, -1.0, 2.0, AC3ENC_PARAM},
256 {"dsurex_mode", "Dolby Surround EX Mode", OFFSET(dolby_surround_ex_mode), FF_OPT_TYPE_INT, -1, -1, 2, AC3ENC_PARAM, "dsurex_mode"},
257 {"notindicated", "Not Indicated (default)", 0, FF_OPT_TYPE_CONST, 0, INT_MIN, INT_MAX, AC3ENC_PARAM, "dsurex_mode"},
258 {"on", "Dolby Surround EX Encoded", 0, FF_OPT_TYPE_CONST, 1, INT_MIN, INT_MAX, AC3ENC_PARAM, "dsurex_mode"},
259 {"off", "Not Dolby Surround EX Encoded", 0, FF_OPT_TYPE_CONST, 2, INT_MIN, INT_MAX, AC3ENC_PARAM, "dsurex_mode"},
260 {"dheadphone_mode", "Dolby Headphone Mode", OFFSET(dolby_headphone_mode), FF_OPT_TYPE_INT, -1, -1, 2, AC3ENC_PARAM, "dheadphone_mode"},
261 {"notindicated", "Not Indicated (default)", 0, FF_OPT_TYPE_CONST, 0, INT_MIN, INT_MAX, AC3ENC_PARAM, "dheadphone_mode"},
262 {"on", "Dolby Headphone Encoded", 0, FF_OPT_TYPE_CONST, 1, INT_MIN, INT_MAX, AC3ENC_PARAM, "dheadphone_mode"},
263 {"off", "Not Dolby Headphone Encoded", 0, FF_OPT_TYPE_CONST, 2, INT_MIN, INT_MAX, AC3ENC_PARAM, "dheadphone_mode"},
264 {"ad_conv_type", "A/D Converter Type", OFFSET(ad_converter_type), FF_OPT_TYPE_INT, -1, -1, 1, AC3ENC_PARAM, "ad_conv_type"},
265 {"standard", "Standard (default)", 0, FF_OPT_TYPE_CONST, 0, INT_MIN, INT_MAX, AC3ENC_PARAM, "ad_conv_type"},
266 {"hdcd", "HDCD", 0, FF_OPT_TYPE_CONST, 1, INT_MIN, INT_MAX, AC3ENC_PARAM, "ad_conv_type"},
267 /* Other Encoding Options */
268 {"stereo_rematrixing", "Stereo Rematrixing", OFFSET(stereo_rematrixing), FF_OPT_TYPE_INT, 1, 0, 1, AC3ENC_PARAM},
272 #if CONFIG_AC3ENC_FLOAT
273 static AVClass ac3enc_class = { "AC-3 Encoder", av_default_item_name,
274 options, LIBAVUTIL_VERSION_INT };
276 static AVClass ac3enc_class = { "Fixed-Point AC-3 Encoder", av_default_item_name,
277 options, LIBAVUTIL_VERSION_INT };
281 /* prototypes for functions in ac3enc_fixed.c and ac3enc_float.c */
283 static av_cold void mdct_end(AC3MDCTContext *mdct);
285 static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct,
288 static void apply_window(DSPContext *dsp, SampleType *output, const SampleType *input,
289 const SampleType *window, unsigned int len);
291 static int normalize_samples(AC3EncodeContext *s);
293 static void scale_coefficients(AC3EncodeContext *s);
297 * LUT for number of exponent groups.
298 * exponent_group_tab[exponent strategy-1][number of coefficients]
300 static uint8_t exponent_group_tab[3][256];
304 * List of supported channel layouts.
306 static const int64_t ac3_channel_layouts[] = {
310 AV_CH_LAYOUT_SURROUND,
313 AV_CH_LAYOUT_4POINT0,
314 AV_CH_LAYOUT_5POINT0,
315 AV_CH_LAYOUT_5POINT0_BACK,
316 (AV_CH_LAYOUT_MONO | AV_CH_LOW_FREQUENCY),
317 (AV_CH_LAYOUT_STEREO | AV_CH_LOW_FREQUENCY),
318 (AV_CH_LAYOUT_2_1 | AV_CH_LOW_FREQUENCY),
319 (AV_CH_LAYOUT_SURROUND | AV_CH_LOW_FREQUENCY),
320 (AV_CH_LAYOUT_2_2 | AV_CH_LOW_FREQUENCY),
321 (AV_CH_LAYOUT_QUAD | AV_CH_LOW_FREQUENCY),
322 (AV_CH_LAYOUT_4POINT0 | AV_CH_LOW_FREQUENCY),
323 AV_CH_LAYOUT_5POINT1,
324 AV_CH_LAYOUT_5POINT1_BACK,
330 * LUT to select the bandwidth code based on the bit rate, sample rate, and
331 * number of full-bandwidth channels.
332 * bandwidth_tab[fbw_channels-1][sample rate code][bit rate code]
334 static const uint8_t ac3_bandwidth_tab[5][3][19] = {
335 // 32 40 48 56 64 80 96 112 128 160 192 224 256 320 384 448 512 576 640
337 { { 0, 0, 0, 12, 16, 32, 48, 48, 48, 48, 48, 48, 48, 48, 48, 48, 48, 48, 48 },
338 { 0, 0, 0, 16, 20, 36, 56, 56, 56, 56, 56, 56, 56, 56, 56, 56, 56, 56, 56 },
339 { 0, 0, 0, 32, 40, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60 } },
341 { { 0, 0, 0, 0, 0, 0, 0, 20, 24, 32, 48, 48, 48, 48, 48, 48, 48, 48, 48 },
342 { 0, 0, 0, 0, 0, 0, 4, 24, 28, 36, 56, 56, 56, 56, 56, 56, 56, 56, 56 },
343 { 0, 0, 0, 0, 0, 0, 20, 44, 52, 60, 60, 60, 60, 60, 60, 60, 60, 60, 60 } },
345 { { 0, 0, 0, 0, 0, 0, 0, 0, 0, 16, 24, 32, 40, 48, 48, 48, 48, 48, 48 },
346 { 0, 0, 0, 0, 0, 0, 0, 0, 4, 20, 28, 36, 44, 56, 56, 56, 56, 56, 56 },
347 { 0, 0, 0, 0, 0, 0, 0, 0, 20, 40, 48, 60, 60, 60, 60, 60, 60, 60, 60 } },
349 { { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 12, 24, 32, 48, 48, 48, 48, 48, 48 },
350 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 16, 28, 36, 56, 56, 56, 56, 56, 56 },
351 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 32, 48, 60, 60, 60, 60, 60, 60, 60 } },
353 { { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8, 20, 32, 40, 48, 48, 48, 48 },
354 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 12, 24, 36, 44, 56, 56, 56, 56 },
355 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 28, 44, 60, 60, 60, 60, 60, 60 } }
360 * Adjust the frame size to make the average bit rate match the target bit rate.
361 * This is only needed for 11025, 22050, and 44100 sample rates.
363 static void adjust_frame_size(AC3EncodeContext *s)
365 while (s->bits_written >= s->bit_rate && s->samples_written >= s->sample_rate) {
366 s->bits_written -= s->bit_rate;
367 s->samples_written -= s->sample_rate;
369 s->frame_size = s->frame_size_min +
370 2 * (s->bits_written * s->sample_rate < s->samples_written * s->bit_rate);
371 s->bits_written += s->frame_size * 8;
372 s->samples_written += AC3_FRAME_SIZE;
377 * Deinterleave input samples.
378 * Channels are reordered from Libav's default order to AC-3 order.
380 static void deinterleave_input_samples(AC3EncodeContext *s,
381 const SampleType *samples)
385 /* deinterleave and remap input samples */
386 for (ch = 0; ch < s->channels; ch++) {
387 const SampleType *sptr;
390 /* copy last 256 samples of previous frame to the start of the current frame */
391 memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_FRAME_SIZE],
392 AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
396 sptr = samples + s->channel_map[ch];
397 for (i = AC3_BLOCK_SIZE; i < AC3_FRAME_SIZE+AC3_BLOCK_SIZE; i++) {
398 s->planar_samples[ch][i] = *sptr;
406 * Apply the MDCT to input samples to generate frequency coefficients.
407 * This applies the KBD window and normalizes the input to reduce precision
408 * loss due to fixed-point calculations.
410 static void apply_mdct(AC3EncodeContext *s)
414 for (ch = 0; ch < s->channels; ch++) {
415 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
416 AC3Block *block = &s->blocks[blk];
417 const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
419 apply_window(&s->dsp, s->windowed_samples, input_samples, s->mdct.window, AC3_WINDOW_SIZE);
421 block->coeff_shift[ch] = normalize_samples(s);
423 s->mdct.fft.mdct_calcw(&s->mdct.fft, block->mdct_coef[ch],
424 s->windowed_samples);
431 * Determine rematrixing flags for each block and band.
433 static void compute_rematrixing_strategy(AC3EncodeContext *s)
437 AC3Block *block, *block0;
439 if (s->channel_mode != AC3_CHMODE_STEREO)
442 s->num_rematrixing_bands = 4;
444 nb_coefs = FFMIN(s->nb_coefs[0], s->nb_coefs[1]);
446 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
447 block = &s->blocks[blk];
448 block->new_rematrixing_strategy = !blk;
449 if (!s->rematrixing_enabled)
451 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
452 /* calculate calculate sum of squared coeffs for one band in one block */
453 int start = ff_ac3_rematrix_band_tab[bnd];
454 int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
455 CoefSumType sum[4] = {0,};
456 for (i = start; i < end; i++) {
457 CoefType lt = block->mdct_coef[0][i];
458 CoefType rt = block->mdct_coef[1][i];
459 CoefType md = lt + rt;
460 CoefType sd = lt - rt;
461 MAC_COEF(sum[0], lt, lt);
462 MAC_COEF(sum[1], rt, rt);
463 MAC_COEF(sum[2], md, md);
464 MAC_COEF(sum[3], sd, sd);
467 /* compare sums to determine if rematrixing will be used for this band */
468 if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
469 block->rematrixing_flags[bnd] = 1;
471 block->rematrixing_flags[bnd] = 0;
473 /* determine if new rematrixing flags will be sent */
475 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
476 block->new_rematrixing_strategy = 1;
485 * Apply stereo rematrixing to coefficients based on rematrixing flags.
487 static void apply_rematrixing(AC3EncodeContext *s)
494 if (!s->rematrixing_enabled)
497 nb_coefs = FFMIN(s->nb_coefs[0], s->nb_coefs[1]);
499 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
500 AC3Block *block = &s->blocks[blk];
501 if (block->new_rematrixing_strategy)
502 flags = block->rematrixing_flags;
503 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
505 start = ff_ac3_rematrix_band_tab[bnd];
506 end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
507 for (i = start; i < end; i++) {
508 int32_t lt = block->fixed_coef[0][i];
509 int32_t rt = block->fixed_coef[1][i];
510 block->fixed_coef[0][i] = (lt + rt) >> 1;
511 block->fixed_coef[1][i] = (lt - rt) >> 1;
520 * Initialize exponent tables.
522 static av_cold void exponent_init(AC3EncodeContext *s)
524 int expstr, i, grpsize;
526 for (expstr = EXP_D15-1; expstr <= EXP_D45-1; expstr++) {
527 grpsize = 3 << expstr;
528 for (i = 73; i < 256; i++) {
529 exponent_group_tab[expstr][i] = (i + grpsize - 4) / grpsize;
533 exponent_group_tab[0][7] = 2;
538 * Extract exponents from the MDCT coefficients.
539 * This takes into account the normalization that was done to the input samples
540 * by adjusting the exponents by the exponent shift values.
542 static void extract_exponents(AC3EncodeContext *s)
546 for (ch = 0; ch < s->channels; ch++) {
547 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
548 AC3Block *block = &s->blocks[blk];
549 s->ac3dsp.extract_exponents(block->exp[ch], block->fixed_coef[ch],
557 * Exponent Difference Threshold.
558 * New exponents are sent if their SAD exceed this number.
560 #define EXP_DIFF_THRESHOLD 500
564 * Calculate exponent strategies for all blocks in a single channel.
566 static void compute_exp_strategy_ch(AC3EncodeContext *s, uint8_t *exp_strategy,
572 /* estimate if the exponent variation & decide if they should be
573 reused in the next frame */
574 exp_strategy[0] = EXP_NEW;
575 exp += AC3_MAX_COEFS;
576 for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) {
577 exp_diff = s->dsp.sad[0](NULL, exp, exp - AC3_MAX_COEFS, 16, 16);
578 if (exp_diff > EXP_DIFF_THRESHOLD)
579 exp_strategy[blk] = EXP_NEW;
581 exp_strategy[blk] = EXP_REUSE;
582 exp += AC3_MAX_COEFS;
585 /* now select the encoding strategy type : if exponents are often
586 recoded, we use a coarse encoding */
588 while (blk < AC3_MAX_BLOCKS) {
590 while (blk1 < AC3_MAX_BLOCKS && exp_strategy[blk1] == EXP_REUSE)
592 switch (blk1 - blk) {
593 case 1: exp_strategy[blk] = EXP_D45; break;
595 case 3: exp_strategy[blk] = EXP_D25; break;
596 default: exp_strategy[blk] = EXP_D15; break;
604 * Calculate exponent strategies for all channels.
605 * Array arrangement is reversed to simplify the per-channel calculation.
607 static void compute_exp_strategy(AC3EncodeContext *s)
611 for (ch = 0; ch < s->fbw_channels; ch++) {
612 compute_exp_strategy_ch(s, s->exp_strategy[ch], s->blocks[0].exp[ch]);
616 s->exp_strategy[ch][0] = EXP_D15;
617 for (blk = 1; blk < AC3_MAX_BLOCKS; blk++)
618 s->exp_strategy[ch][blk] = EXP_REUSE;
624 * Update the exponents so that they are the ones the decoder will decode.
626 static void encode_exponents_blk_ch(uint8_t *exp, int nb_exps, int exp_strategy)
630 nb_groups = exponent_group_tab[exp_strategy-1][nb_exps] * 3;
632 /* for each group, compute the minimum exponent */
633 switch(exp_strategy) {
635 for (i = 1, k = 1; i <= nb_groups; i++) {
636 uint8_t exp_min = exp[k];
637 if (exp[k+1] < exp_min)
644 for (i = 1, k = 1; i <= nb_groups; i++) {
645 uint8_t exp_min = exp[k];
646 if (exp[k+1] < exp_min)
648 if (exp[k+2] < exp_min)
650 if (exp[k+3] < exp_min)
658 /* constraint for DC exponent */
662 /* decrease the delta between each groups to within 2 so that they can be
663 differentially encoded */
664 for (i = 1; i <= nb_groups; i++)
665 exp[i] = FFMIN(exp[i], exp[i-1] + 2);
668 exp[i] = FFMIN(exp[i], exp[i+1] + 2);
670 /* now we have the exponent values the decoder will see */
671 switch (exp_strategy) {
673 for (i = nb_groups, k = nb_groups * 2; i > 0; i--) {
674 uint8_t exp1 = exp[i];
680 for (i = nb_groups, k = nb_groups * 4; i > 0; i--) {
681 exp[k] = exp[k-1] = exp[k-2] = exp[k-3] = exp[i];
690 * Encode exponents from original extracted form to what the decoder will see.
691 * This copies and groups exponents based on exponent strategy and reduces
692 * deltas between adjacent exponent groups so that they can be differentially
695 static void encode_exponents(AC3EncodeContext *s)
698 uint8_t *exp, *exp_strategy;
699 int nb_coefs, num_reuse_blocks;
701 for (ch = 0; ch < s->channels; ch++) {
702 exp = s->blocks[0].exp[ch];
703 exp_strategy = s->exp_strategy[ch];
704 nb_coefs = s->nb_coefs[ch];
707 while (blk < AC3_MAX_BLOCKS) {
710 /* count the number of EXP_REUSE blocks after the current block
711 and set exponent reference block pointers */
712 s->blocks[blk].exp_ref_block[ch] = &s->blocks[blk];
713 while (blk1 < AC3_MAX_BLOCKS && exp_strategy[blk1] == EXP_REUSE) {
714 s->blocks[blk1].exp_ref_block[ch] = &s->blocks[blk];
717 num_reuse_blocks = blk1 - blk - 1;
719 /* for the EXP_REUSE case we select the min of the exponents */
720 s->ac3dsp.ac3_exponent_min(exp, num_reuse_blocks, nb_coefs);
722 encode_exponents_blk_ch(exp, nb_coefs, exp_strategy[blk]);
724 exp += AC3_MAX_COEFS * (num_reuse_blocks + 1);
733 * 3 delta-encoded exponents are in each 7-bit group. The number of groups
734 * varies depending on exponent strategy and bandwidth.
736 static void group_exponents(AC3EncodeContext *s)
739 int group_size, nb_groups, bit_count;
741 int delta0, delta1, delta2;
745 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
746 AC3Block *block = &s->blocks[blk];
747 for (ch = 0; ch < s->channels; ch++) {
748 int exp_strategy = s->exp_strategy[ch][blk];
749 if (exp_strategy == EXP_REUSE)
751 group_size = exp_strategy + (exp_strategy == EXP_D45);
752 nb_groups = exponent_group_tab[exp_strategy-1][s->nb_coefs[ch]];
753 bit_count += 4 + (nb_groups * 7);
758 block->grouped_exp[ch][0] = exp1;
760 /* remaining exponents are delta encoded */
761 for (i = 1; i <= nb_groups; i++) {
762 /* merge three delta in one code */
766 delta0 = exp1 - exp0 + 2;
767 av_assert2(delta0 >= 0 && delta0 <= 4);
772 delta1 = exp1 - exp0 + 2;
773 av_assert2(delta1 >= 0 && delta1 <= 4);
778 delta2 = exp1 - exp0 + 2;
779 av_assert2(delta2 >= 0 && delta2 <= 4);
781 block->grouped_exp[ch][i] = ((delta0 * 5 + delta1) * 5) + delta2;
786 s->exponent_bits = bit_count;
791 * Calculate final exponents from the supplied MDCT coefficients and exponent shift.
792 * Extract exponents from MDCT coefficients, calculate exponent strategies,
793 * and encode final exponents.
795 static void process_exponents(AC3EncodeContext *s)
797 extract_exponents(s);
799 compute_exp_strategy(s);
810 * Count frame bits that are based solely on fixed parameters.
811 * This only has to be run once when the encoder is initialized.
813 static void count_frame_bits_fixed(AC3EncodeContext *s)
815 static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
820 * no dynamic range codes
821 * no channel coupling
822 * bit allocation parameters do not change between blocks
823 * SNR offsets do not change between blocks
824 * no delta bit allocation
831 frame_bits += frame_bits_inc[s->channel_mode];
834 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
835 frame_bits += s->fbw_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
836 if (s->channel_mode == AC3_CHMODE_STEREO) {
837 frame_bits++; /* rematstr */
839 frame_bits += 2 * s->fbw_channels; /* chexpstr[2] * c */
841 frame_bits++; /* lfeexpstr */
842 frame_bits++; /* baie */
843 frame_bits++; /* snr */
844 frame_bits += 2; /* delta / skip */
846 frame_bits++; /* cplinu for block 0 */
848 /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
850 /* (fsnoffset[4] + fgaincod[4]) * c */
851 frame_bits += 2*4 + 3 + 6 + s->channels * (4 + 3);
853 /* auxdatae, crcrsv */
859 s->frame_bits_fixed = frame_bits;
864 * Initialize bit allocation.
865 * Set default parameter codes and calculate parameter values.
867 static void bit_alloc_init(AC3EncodeContext *s)
871 /* init default parameters */
872 s->slow_decay_code = 2;
873 s->fast_decay_code = 1;
874 s->slow_gain_code = 1;
875 s->db_per_bit_code = 3;
877 for (ch = 0; ch < s->channels; ch++)
878 s->fast_gain_code[ch] = 4;
880 /* initial snr offset */
881 s->coarse_snr_offset = 40;
883 /* compute real values */
884 /* currently none of these values change during encoding, so we can just
885 set them once at initialization */
886 s->bit_alloc.slow_decay = ff_ac3_slow_decay_tab[s->slow_decay_code] >> s->bit_alloc.sr_shift;
887 s->bit_alloc.fast_decay = ff_ac3_fast_decay_tab[s->fast_decay_code] >> s->bit_alloc.sr_shift;
888 s->bit_alloc.slow_gain = ff_ac3_slow_gain_tab[s->slow_gain_code];
889 s->bit_alloc.db_per_bit = ff_ac3_db_per_bit_tab[s->db_per_bit_code];
890 s->bit_alloc.floor = ff_ac3_floor_tab[s->floor_code];
892 count_frame_bits_fixed(s);
897 * Count the bits used to encode the frame, minus exponents and mantissas.
898 * Bits based on fixed parameters have already been counted, so now we just
899 * have to add the bits based on parameters that change during encoding.
901 static void count_frame_bits(AC3EncodeContext *s)
903 AC3EncOptions *opt = &s->options;
907 if (opt->audio_production_info)
909 if (s->bitstream_id == 6) {
910 if (opt->extended_bsi_1)
912 if (opt->extended_bsi_2)
916 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
917 /* stereo rematrixing */
918 if (s->channel_mode == AC3_CHMODE_STEREO &&
919 s->blocks[blk].new_rematrixing_strategy) {
920 frame_bits += s->num_rematrixing_bands;
923 for (ch = 0; ch < s->fbw_channels; ch++) {
924 if (s->exp_strategy[ch][blk] != EXP_REUSE)
925 frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */
928 s->frame_bits = s->frame_bits_fixed + frame_bits;
933 * Finalize the mantissa bit count by adding in the grouped mantissas.
935 static int compute_mantissa_size_final(int mant_cnt[5])
937 // bap=1 : 3 mantissas in 5 bits
938 int bits = (mant_cnt[1] / 3) * 5;
939 // bap=2 : 3 mantissas in 7 bits
940 // bap=4 : 2 mantissas in 7 bits
941 bits += ((mant_cnt[2] / 3) + (mant_cnt[4] >> 1)) * 7;
942 // bap=3 : each mantissa is 3 bits
943 bits += mant_cnt[3] * 3;
949 * Calculate masking curve based on the final exponents.
950 * Also calculate the power spectral densities to use in future calculations.
952 static void bit_alloc_masking(AC3EncodeContext *s)
956 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
957 AC3Block *block = &s->blocks[blk];
958 for (ch = 0; ch < s->channels; ch++) {
959 /* We only need psd and mask for calculating bap.
960 Since we currently do not calculate bap when exponent
961 strategy is EXP_REUSE we do not need to calculate psd or mask. */
962 if (s->exp_strategy[ch][blk] != EXP_REUSE) {
963 ff_ac3_bit_alloc_calc_psd(block->exp[ch], 0,
965 block->psd[ch], block->band_psd[ch]);
966 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc, block->band_psd[ch],
968 ff_ac3_fast_gain_tab[s->fast_gain_code[ch]],
969 ch == s->lfe_channel,
970 DBA_NONE, 0, NULL, NULL, NULL,
979 * Ensure that bap for each block and channel point to the current bap_buffer.
980 * They may have been switched during the bit allocation search.
982 static void reset_block_bap(AC3EncodeContext *s)
985 if (s->blocks[0].bap[0] == s->bap_buffer)
987 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
988 for (ch = 0; ch < s->channels; ch++) {
989 s->blocks[blk].bap[ch] = &s->bap_buffer[AC3_MAX_COEFS * (blk * s->channels + ch)];
996 * Run the bit allocation with a given SNR offset.
997 * This calculates the bit allocation pointers that will be used to determine
998 * the quantization of each mantissa.
999 * @return the number of bits needed for mantissas if the given SNR offset is
1002 static int bit_alloc(AC3EncodeContext *s, int snr_offset)
1008 snr_offset = (snr_offset - 240) << 2;
1012 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
1013 AC3Block *block = &s->blocks[blk];
1014 AC3Block *ref_block;
1015 // initialize grouped mantissa counts. these are set so that they are
1016 // padded to the next whole group size when bits are counted in
1017 // compute_mantissa_size_final
1018 mant_cnt[0] = mant_cnt[3] = 0;
1019 mant_cnt[1] = mant_cnt[2] = 2;
1021 for (ch = 0; ch < s->channels; ch++) {
1022 /* Currently the only bit allocation parameters which vary across
1023 blocks within a frame are the exponent values. We can take
1024 advantage of that by reusing the bit allocation pointers
1025 whenever we reuse exponents. */
1026 ref_block = block->exp_ref_block[ch];
1027 if (s->exp_strategy[ch][blk] != EXP_REUSE) {
1028 s->ac3dsp.bit_alloc_calc_bap(ref_block->mask[ch],
1029 ref_block->psd[ch], 0,
1030 s->nb_coefs[ch], snr_offset,
1031 s->bit_alloc.floor, ff_ac3_bap_tab,
1032 ref_block->bap[ch]);
1034 mantissa_bits += s->ac3dsp.compute_mantissa_size(mant_cnt,
1038 mantissa_bits += compute_mantissa_size_final(mant_cnt);
1040 return mantissa_bits;
1045 * Constant bitrate bit allocation search.
1046 * Find the largest SNR offset that will allow data to fit in the frame.
1048 static int cbr_bit_allocation(AC3EncodeContext *s)
1052 int snr_offset, snr_incr;
1054 bits_left = 8 * s->frame_size - (s->frame_bits + s->exponent_bits);
1056 return AVERROR(EINVAL);
1058 snr_offset = s->coarse_snr_offset << 4;
1060 /* if previous frame SNR offset was 1023, check if current frame can also
1061 use SNR offset of 1023. if so, skip the search. */
1062 if ((snr_offset | s->fine_snr_offset[0]) == 1023) {
1063 if (bit_alloc(s, 1023) <= bits_left)
1067 while (snr_offset >= 0 &&
1068 bit_alloc(s, snr_offset) > bits_left) {
1072 return AVERROR(EINVAL);
1074 FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
1075 for (snr_incr = 64; snr_incr > 0; snr_incr >>= 2) {
1076 while (snr_offset + snr_incr <= 1023 &&
1077 bit_alloc(s, snr_offset + snr_incr) <= bits_left) {
1078 snr_offset += snr_incr;
1079 FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
1082 FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer);
1085 s->coarse_snr_offset = snr_offset >> 4;
1086 for (ch = 0; ch < s->channels; ch++)
1087 s->fine_snr_offset[ch] = snr_offset & 0xF;
1094 * Downgrade exponent strategies to reduce the bits used by the exponents.
1095 * This is a fallback for when bit allocation fails with the normal exponent
1096 * strategies. Each time this function is run it only downgrades the
1097 * strategy in 1 channel of 1 block.
1098 * @return non-zero if downgrade was unsuccessful
1100 static int downgrade_exponents(AC3EncodeContext *s)
1104 for (ch = 0; ch < s->fbw_channels; ch++) {
1105 for (blk = AC3_MAX_BLOCKS-1; blk >= 0; blk--) {
1106 if (s->exp_strategy[ch][blk] == EXP_D15) {
1107 s->exp_strategy[ch][blk] = EXP_D25;
1112 for (ch = 0; ch < s->fbw_channels; ch++) {
1113 for (blk = AC3_MAX_BLOCKS-1; blk >= 0; blk--) {
1114 if (s->exp_strategy[ch][blk] == EXP_D25) {
1115 s->exp_strategy[ch][blk] = EXP_D45;
1120 for (ch = 0; ch < s->fbw_channels; ch++) {
1121 /* block 0 cannot reuse exponents, so only downgrade D45 to REUSE if
1122 the block number > 0 */
1123 for (blk = AC3_MAX_BLOCKS-1; blk > 0; blk--) {
1124 if (s->exp_strategy[ch][blk] > EXP_REUSE) {
1125 s->exp_strategy[ch][blk] = EXP_REUSE;
1135 * Reduce the bandwidth to reduce the number of bits used for a given SNR offset.
1136 * This is a second fallback for when bit allocation still fails after exponents
1137 * have been downgraded.
1138 * @return non-zero if bandwidth reduction was unsuccessful
1140 static int reduce_bandwidth(AC3EncodeContext *s, int min_bw_code)
1144 if (s->bandwidth_code[0] > min_bw_code) {
1145 for (ch = 0; ch < s->fbw_channels; ch++) {
1146 s->bandwidth_code[ch]--;
1147 s->nb_coefs[ch] = s->bandwidth_code[ch] * 3 + 73;
1156 * Perform bit allocation search.
1157 * Finds the SNR offset value that maximizes quality and fits in the specified
1158 * frame size. Output is the SNR offset and a set of bit allocation pointers
1159 * used to quantize the mantissas.
1161 static int compute_bit_allocation(AC3EncodeContext *s)
1165 count_frame_bits(s);
1167 bit_alloc_masking(s);
1169 ret = cbr_bit_allocation(s);
1171 /* fallback 1: downgrade exponents */
1172 if (!downgrade_exponents(s)) {
1173 extract_exponents(s);
1174 encode_exponents(s);
1176 ret = compute_bit_allocation(s);
1180 /* fallback 2: reduce bandwidth */
1181 /* only do this if the user has not specified a specific cutoff
1183 if (!s->cutoff && !reduce_bandwidth(s, 0)) {
1184 process_exponents(s);
1185 ret = compute_bit_allocation(s);
1189 /* fallbacks were not enough... */
1198 * Symmetric quantization on 'levels' levels.
1200 static inline int sym_quant(int c, int e, int levels)
1202 int v = (((levels * c) >> (24 - e)) + levels) >> 1;
1203 av_assert2(v >= 0 && v < levels);
1209 * Asymmetric quantization on 2^qbits levels.
1211 static inline int asym_quant(int c, int e, int qbits)
1215 lshift = e + qbits - 24;
1222 m = (1 << (qbits-1));
1225 av_assert2(v >= -m);
1226 return v & ((1 << qbits)-1);
1231 * Quantize a set of mantissas for a single channel in a single block.
1233 static void quantize_mantissas_blk_ch(AC3Mant *s, int32_t *fixed_coef,
1235 uint8_t *bap, uint16_t *qmant, int n)
1239 for (i = 0; i < n; i++) {
1241 int c = fixed_coef[i];
1249 v = sym_quant(c, e, 3);
1250 switch (s->mant1_cnt) {
1252 s->qmant1_ptr = &qmant[i];
1257 *s->qmant1_ptr += 3 * v;
1262 *s->qmant1_ptr += v;
1269 v = sym_quant(c, e, 5);
1270 switch (s->mant2_cnt) {
1272 s->qmant2_ptr = &qmant[i];
1277 *s->qmant2_ptr += 5 * v;
1282 *s->qmant2_ptr += v;
1289 v = sym_quant(c, e, 7);
1292 v = sym_quant(c, e, 11);
1293 switch (s->mant4_cnt) {
1295 s->qmant4_ptr = &qmant[i];
1300 *s->qmant4_ptr += v;
1307 v = sym_quant(c, e, 15);
1310 v = asym_quant(c, e, 14);
1313 v = asym_quant(c, e, 16);
1316 v = asym_quant(c, e, b - 1);
1325 * Quantize mantissas using coefficients, exponents, and bit allocation pointers.
1327 static void quantize_mantissas(AC3EncodeContext *s)
1332 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
1333 AC3Block *block = &s->blocks[blk];
1334 AC3Block *ref_block;
1337 for (ch = 0; ch < s->channels; ch++) {
1338 ref_block = block->exp_ref_block[ch];
1339 quantize_mantissas_blk_ch(&m, block->fixed_coef[ch],
1340 ref_block->exp[ch], ref_block->bap[ch],
1341 block->qmant[ch], s->nb_coefs[ch]);
1348 * Write the AC-3 frame header to the output bitstream.
1350 static void output_frame_header(AC3EncodeContext *s)
1352 AC3EncOptions *opt = &s->options;
1354 put_bits(&s->pb, 16, 0x0b77); /* frame header */
1355 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */
1356 put_bits(&s->pb, 2, s->bit_alloc.sr_code);
1357 put_bits(&s->pb, 6, s->frame_size_code + (s->frame_size - s->frame_size_min) / 2);
1358 put_bits(&s->pb, 5, s->bitstream_id);
1359 put_bits(&s->pb, 3, s->bitstream_mode);
1360 put_bits(&s->pb, 3, s->channel_mode);
1361 if ((s->channel_mode & 0x01) && s->channel_mode != AC3_CHMODE_MONO)
1362 put_bits(&s->pb, 2, s->center_mix_level);
1363 if (s->channel_mode & 0x04)
1364 put_bits(&s->pb, 2, s->surround_mix_level);
1365 if (s->channel_mode == AC3_CHMODE_STEREO)
1366 put_bits(&s->pb, 2, opt->dolby_surround_mode);
1367 put_bits(&s->pb, 1, s->lfe_on); /* LFE */
1368 put_bits(&s->pb, 5, -opt->dialogue_level);
1369 put_bits(&s->pb, 1, 0); /* no compression control word */
1370 put_bits(&s->pb, 1, 0); /* no lang code */
1371 put_bits(&s->pb, 1, opt->audio_production_info);
1372 if (opt->audio_production_info) {
1373 put_bits(&s->pb, 5, opt->mixing_level - 80);
1374 put_bits(&s->pb, 2, opt->room_type);
1376 put_bits(&s->pb, 1, opt->copyright);
1377 put_bits(&s->pb, 1, opt->original);
1378 if (s->bitstream_id == 6) {
1379 /* alternate bit stream syntax */
1380 put_bits(&s->pb, 1, opt->extended_bsi_1);
1381 if (opt->extended_bsi_1) {
1382 put_bits(&s->pb, 2, opt->preferred_stereo_downmix);
1383 put_bits(&s->pb, 3, s->ltrt_center_mix_level);
1384 put_bits(&s->pb, 3, s->ltrt_surround_mix_level);
1385 put_bits(&s->pb, 3, s->loro_center_mix_level);
1386 put_bits(&s->pb, 3, s->loro_surround_mix_level);
1388 put_bits(&s->pb, 1, opt->extended_bsi_2);
1389 if (opt->extended_bsi_2) {
1390 put_bits(&s->pb, 2, opt->dolby_surround_ex_mode);
1391 put_bits(&s->pb, 2, opt->dolby_headphone_mode);
1392 put_bits(&s->pb, 1, opt->ad_converter_type);
1393 put_bits(&s->pb, 9, 0); /* xbsi2 and encinfo : reserved */
1396 put_bits(&s->pb, 1, 0); /* no time code 1 */
1397 put_bits(&s->pb, 1, 0); /* no time code 2 */
1399 put_bits(&s->pb, 1, 0); /* no additional bit stream info */
1404 * Write one audio block to the output bitstream.
1406 static void output_audio_block(AC3EncodeContext *s, int blk)
1408 int ch, i, baie, rbnd;
1409 AC3Block *block = &s->blocks[blk];
1411 /* block switching */
1412 for (ch = 0; ch < s->fbw_channels; ch++)
1413 put_bits(&s->pb, 1, 0);
1416 for (ch = 0; ch < s->fbw_channels; ch++)
1417 put_bits(&s->pb, 1, 1);
1419 /* dynamic range codes */
1420 put_bits(&s->pb, 1, 0);
1422 /* channel coupling */
1424 put_bits(&s->pb, 1, 1); /* coupling strategy present */
1425 put_bits(&s->pb, 1, 0); /* no coupling strategy */
1427 put_bits(&s->pb, 1, 0); /* no new coupling strategy */
1430 /* stereo rematrixing */
1431 if (s->channel_mode == AC3_CHMODE_STEREO) {
1432 put_bits(&s->pb, 1, block->new_rematrixing_strategy);
1433 if (block->new_rematrixing_strategy) {
1434 /* rematrixing flags */
1435 for (rbnd = 0; rbnd < s->num_rematrixing_bands; rbnd++)
1436 put_bits(&s->pb, 1, block->rematrixing_flags[rbnd]);
1440 /* exponent strategy */
1441 for (ch = 0; ch < s->fbw_channels; ch++)
1442 put_bits(&s->pb, 2, s->exp_strategy[ch][blk]);
1444 put_bits(&s->pb, 1, s->exp_strategy[s->lfe_channel][blk]);
1447 for (ch = 0; ch < s->fbw_channels; ch++) {
1448 if (s->exp_strategy[ch][blk] != EXP_REUSE)
1449 put_bits(&s->pb, 6, s->bandwidth_code[ch]);
1453 for (ch = 0; ch < s->channels; ch++) {
1456 if (s->exp_strategy[ch][blk] == EXP_REUSE)
1460 put_bits(&s->pb, 4, block->grouped_exp[ch][0]);
1462 /* exponent groups */
1463 nb_groups = exponent_group_tab[s->exp_strategy[ch][blk]-1][s->nb_coefs[ch]];
1464 for (i = 1; i <= nb_groups; i++)
1465 put_bits(&s->pb, 7, block->grouped_exp[ch][i]);
1467 /* gain range info */
1468 if (ch != s->lfe_channel)
1469 put_bits(&s->pb, 2, 0);
1472 /* bit allocation info */
1474 put_bits(&s->pb, 1, baie);
1476 put_bits(&s->pb, 2, s->slow_decay_code);
1477 put_bits(&s->pb, 2, s->fast_decay_code);
1478 put_bits(&s->pb, 2, s->slow_gain_code);
1479 put_bits(&s->pb, 2, s->db_per_bit_code);
1480 put_bits(&s->pb, 3, s->floor_code);
1484 put_bits(&s->pb, 1, baie);
1486 put_bits(&s->pb, 6, s->coarse_snr_offset);
1487 for (ch = 0; ch < s->channels; ch++) {
1488 put_bits(&s->pb, 4, s->fine_snr_offset[ch]);
1489 put_bits(&s->pb, 3, s->fast_gain_code[ch]);
1493 put_bits(&s->pb, 1, 0); /* no delta bit allocation */
1494 put_bits(&s->pb, 1, 0); /* no data to skip */
1497 for (ch = 0; ch < s->channels; ch++) {
1499 AC3Block *ref_block = block->exp_ref_block[ch];
1500 for (i = 0; i < s->nb_coefs[ch]; i++) {
1501 q = block->qmant[ch][i];
1502 b = ref_block->bap[ch][i];
1505 case 1: if (q != 128) put_bits(&s->pb, 5, q); break;
1506 case 2: if (q != 128) put_bits(&s->pb, 7, q); break;
1507 case 3: put_bits(&s->pb, 3, q); break;
1508 case 4: if (q != 128) put_bits(&s->pb, 7, q); break;
1509 case 14: put_bits(&s->pb, 14, q); break;
1510 case 15: put_bits(&s->pb, 16, q); break;
1511 default: put_bits(&s->pb, b-1, q); break;
1518 /** CRC-16 Polynomial */
1519 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
1522 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly)
1539 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly)
1545 r = mul_poly(r, a, poly);
1546 a = mul_poly(a, a, poly);
1554 * Fill the end of the frame with 0's and compute the two CRCs.
1556 static void output_frame_end(AC3EncodeContext *s)
1558 const AVCRC *crc_ctx = av_crc_get_table(AV_CRC_16_ANSI);
1559 int frame_size_58, pad_bytes, crc1, crc2_partial, crc2, crc_inv;
1562 frame_size_58 = ((s->frame_size >> 2) + (s->frame_size >> 4)) << 1;
1564 /* pad the remainder of the frame with zeros */
1565 av_assert2(s->frame_size * 8 - put_bits_count(&s->pb) >= 18);
1566 flush_put_bits(&s->pb);
1568 pad_bytes = s->frame_size - (put_bits_ptr(&s->pb) - frame) - 2;
1569 av_assert2(pad_bytes >= 0);
1571 memset(put_bits_ptr(&s->pb), 0, pad_bytes);
1574 /* this is not so easy because it is at the beginning of the data... */
1575 crc1 = av_bswap16(av_crc(crc_ctx, 0, frame + 4, frame_size_58 - 4));
1576 crc_inv = s->crc_inv[s->frame_size > s->frame_size_min];
1577 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
1578 AV_WB16(frame + 2, crc1);
1581 crc2_partial = av_crc(crc_ctx, 0, frame + frame_size_58,
1582 s->frame_size - frame_size_58 - 3);
1583 crc2 = av_crc(crc_ctx, crc2_partial, frame + s->frame_size - 3, 1);
1584 /* ensure crc2 does not match sync word by flipping crcrsv bit if needed */
1585 if (crc2 == 0x770B) {
1586 frame[s->frame_size - 3] ^= 0x1;
1587 crc2 = av_crc(crc_ctx, crc2_partial, frame + s->frame_size - 3, 1);
1589 crc2 = av_bswap16(crc2);
1590 AV_WB16(frame + s->frame_size - 2, crc2);
1595 * Write the frame to the output bitstream.
1597 static void output_frame(AC3EncodeContext *s, unsigned char *frame)
1601 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE);
1603 output_frame_header(s);
1605 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
1606 output_audio_block(s, blk);
1608 output_frame_end(s);
1612 static void dprint_options(AVCodecContext *avctx)
1615 AC3EncodeContext *s = avctx->priv_data;
1616 AC3EncOptions *opt = &s->options;
1619 switch (s->bitstream_id) {
1620 case 6: strncpy(strbuf, "AC-3 (alt syntax)", 32); break;
1621 case 8: strncpy(strbuf, "AC-3 (standard)", 32); break;
1622 case 9: strncpy(strbuf, "AC-3 (dnet half-rate)", 32); break;
1623 case 10: strncpy(strbuf, "AC-3 (dnet quater-rate", 32); break;
1624 default: snprintf(strbuf, 32, "ERROR");
1626 av_dlog(avctx, "bitstream_id: %s (%d)\n", strbuf, s->bitstream_id);
1627 av_dlog(avctx, "sample_fmt: %s\n", av_get_sample_fmt_name(avctx->sample_fmt));
1628 av_get_channel_layout_string(strbuf, 32, s->channels, avctx->channel_layout);
1629 av_dlog(avctx, "channel_layout: %s\n", strbuf);
1630 av_dlog(avctx, "sample_rate: %d\n", s->sample_rate);
1631 av_dlog(avctx, "bit_rate: %d\n", s->bit_rate);
1633 av_dlog(avctx, "cutoff: %d\n", s->cutoff);
1635 av_dlog(avctx, "per_frame_metadata: %s\n",
1636 opt->allow_per_frame_metadata?"on":"off");
1638 av_dlog(avctx, "center_mixlev: %0.3f (%d)\n", opt->center_mix_level,
1639 s->center_mix_level);
1641 av_dlog(avctx, "center_mixlev: {not written}\n");
1642 if (s->has_surround)
1643 av_dlog(avctx, "surround_mixlev: %0.3f (%d)\n", opt->surround_mix_level,
1644 s->surround_mix_level);
1646 av_dlog(avctx, "surround_mixlev: {not written}\n");
1647 if (opt->audio_production_info) {
1648 av_dlog(avctx, "mixing_level: %ddB\n", opt->mixing_level);
1649 switch (opt->room_type) {
1650 case 0: strncpy(strbuf, "notindicated", 32); break;
1651 case 1: strncpy(strbuf, "large", 32); break;
1652 case 2: strncpy(strbuf, "small", 32); break;
1653 default: snprintf(strbuf, 32, "ERROR (%d)", opt->room_type);
1655 av_dlog(avctx, "room_type: %s\n", strbuf);
1657 av_dlog(avctx, "mixing_level: {not written}\n");
1658 av_dlog(avctx, "room_type: {not written}\n");
1660 av_dlog(avctx, "copyright: %s\n", opt->copyright?"on":"off");
1661 av_dlog(avctx, "dialnorm: %ddB\n", opt->dialogue_level);
1662 if (s->channel_mode == AC3_CHMODE_STEREO) {
1663 switch (opt->dolby_surround_mode) {
1664 case 0: strncpy(strbuf, "notindicated", 32); break;
1665 case 1: strncpy(strbuf, "on", 32); break;
1666 case 2: strncpy(strbuf, "off", 32); break;
1667 default: snprintf(strbuf, 32, "ERROR (%d)", opt->dolby_surround_mode);
1669 av_dlog(avctx, "dsur_mode: %s\n", strbuf);
1671 av_dlog(avctx, "dsur_mode: {not written}\n");
1673 av_dlog(avctx, "original: %s\n", opt->original?"on":"off");
1675 if (s->bitstream_id == 6) {
1676 if (opt->extended_bsi_1) {
1677 switch (opt->preferred_stereo_downmix) {
1678 case 0: strncpy(strbuf, "notindicated", 32); break;
1679 case 1: strncpy(strbuf, "ltrt", 32); break;
1680 case 2: strncpy(strbuf, "loro", 32); break;
1681 default: snprintf(strbuf, 32, "ERROR (%d)", opt->preferred_stereo_downmix);
1683 av_dlog(avctx, "dmix_mode: %s\n", strbuf);
1684 av_dlog(avctx, "ltrt_cmixlev: %0.3f (%d)\n",
1685 opt->ltrt_center_mix_level, s->ltrt_center_mix_level);
1686 av_dlog(avctx, "ltrt_surmixlev: %0.3f (%d)\n",
1687 opt->ltrt_surround_mix_level, s->ltrt_surround_mix_level);
1688 av_dlog(avctx, "loro_cmixlev: %0.3f (%d)\n",
1689 opt->loro_center_mix_level, s->loro_center_mix_level);
1690 av_dlog(avctx, "loro_surmixlev: %0.3f (%d)\n",
1691 opt->loro_surround_mix_level, s->loro_surround_mix_level);
1693 av_dlog(avctx, "extended bitstream info 1: {not written}\n");
1695 if (opt->extended_bsi_2) {
1696 switch (opt->dolby_surround_ex_mode) {
1697 case 0: strncpy(strbuf, "notindicated", 32); break;
1698 case 1: strncpy(strbuf, "on", 32); break;
1699 case 2: strncpy(strbuf, "off", 32); break;
1700 default: snprintf(strbuf, 32, "ERROR (%d)", opt->dolby_surround_ex_mode);
1702 av_dlog(avctx, "dsurex_mode: %s\n", strbuf);
1703 switch (opt->dolby_headphone_mode) {
1704 case 0: strncpy(strbuf, "notindicated", 32); break;
1705 case 1: strncpy(strbuf, "on", 32); break;
1706 case 2: strncpy(strbuf, "off", 32); break;
1707 default: snprintf(strbuf, 32, "ERROR (%d)", opt->dolby_headphone_mode);
1709 av_dlog(avctx, "dheadphone_mode: %s\n", strbuf);
1711 switch (opt->ad_converter_type) {
1712 case 0: strncpy(strbuf, "standard", 32); break;
1713 case 1: strncpy(strbuf, "hdcd", 32); break;
1714 default: snprintf(strbuf, 32, "ERROR (%d)", opt->ad_converter_type);
1716 av_dlog(avctx, "ad_conv_type: %s\n", strbuf);
1718 av_dlog(avctx, "extended bitstream info 2: {not written}\n");
1725 #define FLT_OPTION_THRESHOLD 0.01
1727 static int validate_float_option(float v, const float *v_list, int v_list_size)
1731 for (i = 0; i < v_list_size; i++) {
1732 if (v < (v_list[i] + FLT_OPTION_THRESHOLD) &&
1733 v > (v_list[i] - FLT_OPTION_THRESHOLD))
1736 if (i == v_list_size)
1743 static void validate_mix_level(void *log_ctx, const char *opt_name,
1744 float *opt_param, const float *list,
1745 int list_size, int default_value, int min_value,
1748 int mixlev = validate_float_option(*opt_param, list, list_size);
1749 if (mixlev < min_value) {
1750 mixlev = default_value;
1751 if (*opt_param >= 0.0) {
1752 av_log(log_ctx, AV_LOG_WARNING, "requested %s is not valid. using "
1753 "default value: %0.3f\n", opt_name, list[mixlev]);
1756 *opt_param = list[mixlev];
1757 *ctx_param = mixlev;
1762 * Validate metadata options as set by AVOption system.
1763 * These values can optionally be changed per-frame.
1765 static int validate_metadata(AVCodecContext *avctx)
1767 AC3EncodeContext *s = avctx->priv_data;
1768 AC3EncOptions *opt = &s->options;
1770 /* validate mixing levels */
1771 if (s->has_center) {
1772 validate_mix_level(avctx, "center_mix_level", &opt->center_mix_level,
1773 cmixlev_options, CMIXLEV_NUM_OPTIONS, 1, 0,
1774 &s->center_mix_level);
1776 if (s->has_surround) {
1777 validate_mix_level(avctx, "surround_mix_level", &opt->surround_mix_level,
1778 surmixlev_options, SURMIXLEV_NUM_OPTIONS, 1, 0,
1779 &s->surround_mix_level);
1782 /* set audio production info flag */
1783 if (opt->mixing_level >= 0 || opt->room_type >= 0) {
1784 if (opt->mixing_level < 0) {
1785 av_log(avctx, AV_LOG_ERROR, "mixing_level must be set if "
1786 "room_type is set\n");
1787 return AVERROR(EINVAL);
1789 if (opt->mixing_level < 80) {
1790 av_log(avctx, AV_LOG_ERROR, "invalid mixing level. must be between "
1791 "80dB and 111dB\n");
1792 return AVERROR(EINVAL);
1794 /* default room type */
1795 if (opt->room_type < 0)
1797 opt->audio_production_info = 1;
1799 opt->audio_production_info = 0;
1802 /* set extended bsi 1 flag */
1803 if ((s->has_center || s->has_surround) &&
1804 (opt->preferred_stereo_downmix >= 0 ||
1805 opt->ltrt_center_mix_level >= 0 ||
1806 opt->ltrt_surround_mix_level >= 0 ||
1807 opt->loro_center_mix_level >= 0 ||
1808 opt->loro_surround_mix_level >= 0)) {
1809 /* default preferred stereo downmix */
1810 if (opt->preferred_stereo_downmix < 0)
1811 opt->preferred_stereo_downmix = 0;
1812 /* validate Lt/Rt center mix level */
1813 validate_mix_level(avctx, "ltrt_center_mix_level",
1814 &opt->ltrt_center_mix_level, extmixlev_options,
1815 EXTMIXLEV_NUM_OPTIONS, 5, 0,
1816 &s->ltrt_center_mix_level);
1817 /* validate Lt/Rt surround mix level */
1818 validate_mix_level(avctx, "ltrt_surround_mix_level",
1819 &opt->ltrt_surround_mix_level, extmixlev_options,
1820 EXTMIXLEV_NUM_OPTIONS, 6, 3,
1821 &s->ltrt_surround_mix_level);
1822 /* validate Lo/Ro center mix level */
1823 validate_mix_level(avctx, "loro_center_mix_level",
1824 &opt->loro_center_mix_level, extmixlev_options,
1825 EXTMIXLEV_NUM_OPTIONS, 5, 0,
1826 &s->loro_center_mix_level);
1827 /* validate Lo/Ro surround mix level */
1828 validate_mix_level(avctx, "loro_surround_mix_level",
1829 &opt->loro_surround_mix_level, extmixlev_options,
1830 EXTMIXLEV_NUM_OPTIONS, 6, 3,
1831 &s->loro_surround_mix_level);
1832 opt->extended_bsi_1 = 1;
1834 opt->extended_bsi_1 = 0;
1837 /* set extended bsi 2 flag */
1838 if (opt->dolby_surround_ex_mode >= 0 ||
1839 opt->dolby_headphone_mode >= 0 ||
1840 opt->ad_converter_type >= 0) {
1841 /* default dolby surround ex mode */
1842 if (opt->dolby_surround_ex_mode < 0)
1843 opt->dolby_surround_ex_mode = 0;
1844 /* default dolby headphone mode */
1845 if (opt->dolby_headphone_mode < 0)
1846 opt->dolby_headphone_mode = 0;
1847 /* default A/D converter type */
1848 if (opt->ad_converter_type < 0)
1849 opt->ad_converter_type = 0;
1850 opt->extended_bsi_2 = 1;
1852 opt->extended_bsi_2 = 0;
1855 /* set bitstream id for alternate bitstream syntax */
1856 if (opt->extended_bsi_1 || opt->extended_bsi_2) {
1857 if (s->bitstream_id > 8 && s->bitstream_id < 11) {
1858 static int warn_once = 1;
1860 av_log(avctx, AV_LOG_WARNING, "alternate bitstream syntax is "
1861 "not compatible with reduced samplerates. writing of "
1862 "extended bitstream information will be disabled.\n");
1866 s->bitstream_id = 6;
1875 * Encode a single AC-3 frame.
1877 static int ac3_encode_frame(AVCodecContext *avctx, unsigned char *frame,
1878 int buf_size, void *data)
1880 AC3EncodeContext *s = avctx->priv_data;
1881 const SampleType *samples = data;
1884 if (s->options.allow_per_frame_metadata) {
1885 ret = validate_metadata(avctx);
1890 if (s->bit_alloc.sr_code == 1)
1891 adjust_frame_size(s);
1893 deinterleave_input_samples(s, samples);
1897 scale_coefficients(s);
1899 compute_rematrixing_strategy(s);
1901 apply_rematrixing(s);
1903 process_exponents(s);
1905 ret = compute_bit_allocation(s);
1907 av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
1911 quantize_mantissas(s);
1913 output_frame(s, frame);
1915 return s->frame_size;
1920 * Finalize encoding and free any memory allocated by the encoder.
1922 static av_cold int ac3_encode_close(AVCodecContext *avctx)
1925 AC3EncodeContext *s = avctx->priv_data;
1927 for (ch = 0; ch < s->channels; ch++)
1928 av_freep(&s->planar_samples[ch]);
1929 av_freep(&s->planar_samples);
1930 av_freep(&s->bap_buffer);
1931 av_freep(&s->bap1_buffer);
1932 av_freep(&s->mdct_coef_buffer);
1933 av_freep(&s->fixed_coef_buffer);
1934 av_freep(&s->exp_buffer);
1935 av_freep(&s->grouped_exp_buffer);
1936 av_freep(&s->psd_buffer);
1937 av_freep(&s->band_psd_buffer);
1938 av_freep(&s->mask_buffer);
1939 av_freep(&s->qmant_buffer);
1940 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
1941 AC3Block *block = &s->blocks[blk];
1942 av_freep(&block->bap);
1943 av_freep(&block->mdct_coef);
1944 av_freep(&block->fixed_coef);
1945 av_freep(&block->exp);
1946 av_freep(&block->grouped_exp);
1947 av_freep(&block->psd);
1948 av_freep(&block->band_psd);
1949 av_freep(&block->mask);
1950 av_freep(&block->qmant);
1955 av_freep(&avctx->coded_frame);
1961 * Set channel information during initialization.
1963 static av_cold int set_channel_info(AC3EncodeContext *s, int channels,
1964 int64_t *channel_layout)
1968 if (channels < 1 || channels > AC3_MAX_CHANNELS)
1969 return AVERROR(EINVAL);
1970 if ((uint64_t)*channel_layout > 0x7FF)
1971 return AVERROR(EINVAL);
1972 ch_layout = *channel_layout;
1974 ch_layout = avcodec_guess_channel_layout(channels, CODEC_ID_AC3, NULL);
1976 s->lfe_on = !!(ch_layout & AV_CH_LOW_FREQUENCY);
1977 s->channels = channels;
1978 s->fbw_channels = channels - s->lfe_on;
1979 s->lfe_channel = s->lfe_on ? s->fbw_channels : -1;
1981 ch_layout -= AV_CH_LOW_FREQUENCY;
1983 switch (ch_layout) {
1984 case AV_CH_LAYOUT_MONO: s->channel_mode = AC3_CHMODE_MONO; break;
1985 case AV_CH_LAYOUT_STEREO: s->channel_mode = AC3_CHMODE_STEREO; break;
1986 case AV_CH_LAYOUT_SURROUND: s->channel_mode = AC3_CHMODE_3F; break;
1987 case AV_CH_LAYOUT_2_1: s->channel_mode = AC3_CHMODE_2F1R; break;
1988 case AV_CH_LAYOUT_4POINT0: s->channel_mode = AC3_CHMODE_3F1R; break;
1989 case AV_CH_LAYOUT_QUAD:
1990 case AV_CH_LAYOUT_2_2: s->channel_mode = AC3_CHMODE_2F2R; break;
1991 case AV_CH_LAYOUT_5POINT0:
1992 case AV_CH_LAYOUT_5POINT0_BACK: s->channel_mode = AC3_CHMODE_3F2R; break;
1994 return AVERROR(EINVAL);
1996 s->has_center = (s->channel_mode & 0x01) && s->channel_mode != AC3_CHMODE_MONO;
1997 s->has_surround = s->channel_mode & 0x04;
1999 s->channel_map = ff_ac3_enc_channel_map[s->channel_mode][s->lfe_on];
2000 *channel_layout = ch_layout;
2002 *channel_layout |= AV_CH_LOW_FREQUENCY;
2008 static av_cold int validate_options(AVCodecContext *avctx, AC3EncodeContext *s)
2012 /* validate channel layout */
2013 if (!avctx->channel_layout) {
2014 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
2015 "encoder will guess the layout, but it "
2016 "might be incorrect.\n");
2018 ret = set_channel_info(s, avctx->channels, &avctx->channel_layout);
2020 av_log(avctx, AV_LOG_ERROR, "invalid channel layout\n");
2024 /* validate sample rate */
2025 for (i = 0; i < 9; i++) {
2026 if ((ff_ac3_sample_rate_tab[i / 3] >> (i % 3)) == avctx->sample_rate)
2030 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
2031 return AVERROR(EINVAL);
2033 s->sample_rate = avctx->sample_rate;
2034 s->bit_alloc.sr_shift = i % 3;
2035 s->bit_alloc.sr_code = i / 3;
2036 s->bitstream_id = 8 + s->bit_alloc.sr_shift;
2038 /* validate bit rate */
2039 for (i = 0; i < 19; i++) {
2040 if ((ff_ac3_bitrate_tab[i] >> s->bit_alloc.sr_shift)*1000 == avctx->bit_rate)
2044 av_log(avctx, AV_LOG_ERROR, "invalid bit rate\n");
2045 return AVERROR(EINVAL);
2047 s->bit_rate = avctx->bit_rate;
2048 s->frame_size_code = i << 1;
2050 /* validate cutoff */
2051 if (avctx->cutoff < 0) {
2052 av_log(avctx, AV_LOG_ERROR, "invalid cutoff frequency\n");
2053 return AVERROR(EINVAL);
2055 s->cutoff = avctx->cutoff;
2056 if (s->cutoff > (s->sample_rate >> 1))
2057 s->cutoff = s->sample_rate >> 1;
2059 /* validate audio service type / channels combination */
2060 if ((avctx->audio_service_type == AV_AUDIO_SERVICE_TYPE_KARAOKE &&
2061 avctx->channels == 1) ||
2062 ((avctx->audio_service_type == AV_AUDIO_SERVICE_TYPE_COMMENTARY ||
2063 avctx->audio_service_type == AV_AUDIO_SERVICE_TYPE_EMERGENCY ||
2064 avctx->audio_service_type == AV_AUDIO_SERVICE_TYPE_VOICE_OVER)
2065 && avctx->channels > 1)) {
2066 av_log(avctx, AV_LOG_ERROR, "invalid audio service type for the "
2067 "specified number of channels\n");
2068 return AVERROR(EINVAL);
2071 ret = validate_metadata(avctx);
2075 s->rematrixing_enabled = s->options.stereo_rematrixing &&
2076 (s->channel_mode == AC3_CHMODE_STEREO);
2083 * Set bandwidth for all channels.
2084 * The user can optionally supply a cutoff frequency. Otherwise an appropriate
2085 * default value will be used.
2087 static av_cold void set_bandwidth(AC3EncodeContext *s)
2092 /* calculate bandwidth based on user-specified cutoff frequency */
2094 fbw_coeffs = s->cutoff * 2 * AC3_MAX_COEFS / s->sample_rate;
2095 bw_code = av_clip((fbw_coeffs - 73) / 3, 0, 60);
2097 /* use default bandwidth setting */
2098 bw_code = ac3_bandwidth_tab[s->fbw_channels-1][s->bit_alloc.sr_code][s->frame_size_code/2];
2101 /* set number of coefficients for each channel */
2102 for (ch = 0; ch < s->fbw_channels; ch++) {
2103 s->bandwidth_code[ch] = bw_code;
2104 s->nb_coefs[ch] = bw_code * 3 + 73;
2107 s->nb_coefs[s->lfe_channel] = 7; /* LFE channel always has 7 coefs */
2111 static av_cold int allocate_buffers(AVCodecContext *avctx)
2114 AC3EncodeContext *s = avctx->priv_data;
2116 FF_ALLOC_OR_GOTO(avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
2118 for (ch = 0; ch < s->channels; ch++) {
2119 FF_ALLOCZ_OR_GOTO(avctx, s->planar_samples[ch],
2120 (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
2123 FF_ALLOC_OR_GOTO(avctx, s->bap_buffer, AC3_MAX_BLOCKS * s->channels *
2124 AC3_MAX_COEFS * sizeof(*s->bap_buffer), alloc_fail);
2125 FF_ALLOC_OR_GOTO(avctx, s->bap1_buffer, AC3_MAX_BLOCKS * s->channels *
2126 AC3_MAX_COEFS * sizeof(*s->bap1_buffer), alloc_fail);
2127 FF_ALLOC_OR_GOTO(avctx, s->mdct_coef_buffer, AC3_MAX_BLOCKS * s->channels *
2128 AC3_MAX_COEFS * sizeof(*s->mdct_coef_buffer), alloc_fail);
2129 FF_ALLOC_OR_GOTO(avctx, s->exp_buffer, AC3_MAX_BLOCKS * s->channels *
2130 AC3_MAX_COEFS * sizeof(*s->exp_buffer), alloc_fail);
2131 FF_ALLOC_OR_GOTO(avctx, s->grouped_exp_buffer, AC3_MAX_BLOCKS * s->channels *
2132 128 * sizeof(*s->grouped_exp_buffer), alloc_fail);
2133 FF_ALLOC_OR_GOTO(avctx, s->psd_buffer, AC3_MAX_BLOCKS * s->channels *
2134 AC3_MAX_COEFS * sizeof(*s->psd_buffer), alloc_fail);
2135 FF_ALLOC_OR_GOTO(avctx, s->band_psd_buffer, AC3_MAX_BLOCKS * s->channels *
2136 64 * sizeof(*s->band_psd_buffer), alloc_fail);
2137 FF_ALLOC_OR_GOTO(avctx, s->mask_buffer, AC3_MAX_BLOCKS * s->channels *
2138 64 * sizeof(*s->mask_buffer), alloc_fail);
2139 FF_ALLOC_OR_GOTO(avctx, s->qmant_buffer, AC3_MAX_BLOCKS * s->channels *
2140 AC3_MAX_COEFS * sizeof(*s->qmant_buffer), alloc_fail);
2141 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
2142 AC3Block *block = &s->blocks[blk];
2143 FF_ALLOC_OR_GOTO(avctx, block->bap, s->channels * sizeof(*block->bap),
2145 FF_ALLOCZ_OR_GOTO(avctx, block->mdct_coef, s->channels * sizeof(*block->mdct_coef),
2147 FF_ALLOCZ_OR_GOTO(avctx, block->exp, s->channels * sizeof(*block->exp),
2149 FF_ALLOCZ_OR_GOTO(avctx, block->grouped_exp, s->channels * sizeof(*block->grouped_exp),
2151 FF_ALLOCZ_OR_GOTO(avctx, block->psd, s->channels * sizeof(*block->psd),
2153 FF_ALLOCZ_OR_GOTO(avctx, block->band_psd, s->channels * sizeof(*block->band_psd),
2155 FF_ALLOCZ_OR_GOTO(avctx, block->mask, s->channels * sizeof(*block->mask),
2157 FF_ALLOCZ_OR_GOTO(avctx, block->qmant, s->channels * sizeof(*block->qmant),
2160 for (ch = 0; ch < s->channels; ch++) {
2161 /* arrangement: block, channel, coeff */
2162 block->bap[ch] = &s->bap_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
2163 block->mdct_coef[ch] = &s->mdct_coef_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
2164 block->grouped_exp[ch] = &s->grouped_exp_buffer[128 * (blk * s->channels + ch)];
2165 block->psd[ch] = &s->psd_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
2166 block->band_psd[ch] = &s->band_psd_buffer [64 * (blk * s->channels + ch)];
2167 block->mask[ch] = &s->mask_buffer [64 * (blk * s->channels + ch)];
2168 block->qmant[ch] = &s->qmant_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)];
2170 /* arrangement: channel, block, coeff */
2171 block->exp[ch] = &s->exp_buffer [AC3_MAX_COEFS * (AC3_MAX_BLOCKS * ch + blk)];
2175 if (CONFIG_AC3ENC_FLOAT) {
2176 FF_ALLOC_OR_GOTO(avctx, s->fixed_coef_buffer, AC3_MAX_BLOCKS * s->channels *
2177 AC3_MAX_COEFS * sizeof(*s->fixed_coef_buffer), alloc_fail);
2178 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
2179 AC3Block *block = &s->blocks[blk];
2180 FF_ALLOCZ_OR_GOTO(avctx, block->fixed_coef, s->channels *
2181 sizeof(*block->fixed_coef), alloc_fail);
2182 for (ch = 0; ch < s->channels; ch++)
2183 block->fixed_coef[ch] = &s->fixed_coef_buffer[AC3_MAX_COEFS * (blk * s->channels + ch)];
2186 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
2187 AC3Block *block = &s->blocks[blk];
2188 FF_ALLOCZ_OR_GOTO(avctx, block->fixed_coef, s->channels *
2189 sizeof(*block->fixed_coef), alloc_fail);
2190 for (ch = 0; ch < s->channels; ch++)
2191 block->fixed_coef[ch] = (int32_t *)block->mdct_coef[ch];
2197 return AVERROR(ENOMEM);
2202 * Initialize the encoder.
2204 static av_cold int ac3_encode_init(AVCodecContext *avctx)
2206 AC3EncodeContext *s = avctx->priv_data;
2207 int ret, frame_size_58;
2209 avctx->frame_size = AC3_FRAME_SIZE;
2211 ff_ac3_common_init();
2213 ret = validate_options(avctx, s);
2217 s->bitstream_mode = avctx->audio_service_type;
2218 if (s->bitstream_mode == AV_AUDIO_SERVICE_TYPE_KARAOKE)
2219 s->bitstream_mode = 0x7;
2221 s->frame_size_min = 2 * ff_ac3_frame_size_tab[s->frame_size_code][s->bit_alloc.sr_code];
2222 s->bits_written = 0;
2223 s->samples_written = 0;
2224 s->frame_size = s->frame_size_min;
2226 /* calculate crc_inv for both possible frame sizes */
2227 frame_size_58 = (( s->frame_size >> 2) + ( s->frame_size >> 4)) << 1;
2228 s->crc_inv[0] = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY);
2229 if (s->bit_alloc.sr_code == 1) {
2230 frame_size_58 = (((s->frame_size+2) >> 2) + ((s->frame_size+2) >> 4)) << 1;
2231 s->crc_inv[1] = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY);
2240 ret = mdct_init(avctx, &s->mdct, 9);
2244 ret = allocate_buffers(avctx);
2248 avctx->coded_frame= avcodec_alloc_frame();
2250 dsputil_init(&s->dsp, avctx);
2251 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & CODEC_FLAG_BITEXACT);
2253 dprint_options(avctx);
2257 ac3_encode_close(avctx);