ffmpeg / libavcodec / ac3enc.c @ 5128842e
History | View | Annotate | Download (62.8 KB)
| 1 |
/*
|
|---|---|
| 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>
|
| 6 |
*
|
| 7 |
* This file is part of FFmpeg.
|
| 8 |
*
|
| 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.
|
| 13 |
*
|
| 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.
|
| 18 |
*
|
| 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
|
| 22 |
*/
|
| 23 |
|
| 24 |
/**
|
| 25 |
* @file
|
| 26 |
* The simplest AC-3 encoder.
|
| 27 |
*/
|
| 28 |
|
| 29 |
//#define DEBUG
|
| 30 |
|
| 31 |
#include "libavcore/audioconvert.h" |
| 32 |
#include "libavutil/crc.h" |
| 33 |
#include "avcodec.h" |
| 34 |
#include "put_bits.h" |
| 35 |
#include "dsputil.h" |
| 36 |
#include "ac3.h" |
| 37 |
#include "audioconvert.h" |
| 38 |
|
| 39 |
|
| 40 |
/** Maximum number of exponent groups. +1 for separate DC exponent. */
|
| 41 |
#define AC3_MAX_EXP_GROUPS 85 |
| 42 |
|
| 43 |
/** Scale a float value by 2^bits and convert to an integer. */
|
| 44 |
#define SCALE_FLOAT(a, bits) lrintf((a) * (float)(1 << (bits))) |
| 45 |
|
| 46 |
typedef int16_t SampleType;
|
| 47 |
typedef int32_t CoefType;
|
| 48 |
|
| 49 |
#define SCALE_COEF(a) (a)
|
| 50 |
|
| 51 |
/** Scale a float value by 2^15, convert to an integer, and clip to range -32767..32767. */
|
| 52 |
#define FIX15(a) av_clip(SCALE_FLOAT(a, 15), -32767, 32767) |
| 53 |
|
| 54 |
|
| 55 |
/**
|
| 56 |
* Compex number.
|
| 57 |
* Used in fixed-point MDCT calculation.
|
| 58 |
*/
|
| 59 |
typedef struct IComplex { |
| 60 |
int16_t re,im; |
| 61 |
} IComplex; |
| 62 |
|
| 63 |
typedef struct AC3MDCTContext { |
| 64 |
const int16_t *window; ///< MDCT window function |
| 65 |
int nbits; ///< log2(transform size) |
| 66 |
int16_t *costab; ///< FFT cos table
|
| 67 |
int16_t *sintab; ///< FFT sin table
|
| 68 |
int16_t *xcos1; ///< MDCT cos table
|
| 69 |
int16_t *xsin1; ///< MDCT sin table
|
| 70 |
int16_t *rot_tmp; ///< temp buffer for pre-rotated samples
|
| 71 |
IComplex *cplx_tmp; ///< temp buffer for complex pre-rotated samples
|
| 72 |
} AC3MDCTContext; |
| 73 |
|
| 74 |
/**
|
| 75 |
* Data for a single audio block.
|
| 76 |
*/
|
| 77 |
typedef struct AC3Block { |
| 78 |
uint8_t **bap; ///< bit allocation pointers (bap)
|
| 79 |
CoefType **mdct_coef; ///< MDCT coefficients
|
| 80 |
uint8_t **exp; ///< original exponents
|
| 81 |
uint8_t **grouped_exp; ///< grouped exponents
|
| 82 |
int16_t **psd; ///< psd per frequency bin
|
| 83 |
int16_t **band_psd; ///< psd per critical band
|
| 84 |
int16_t **mask; ///< masking curve
|
| 85 |
uint16_t **qmant; ///< quantized mantissas
|
| 86 |
uint8_t exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
|
| 87 |
int8_t exp_shift[AC3_MAX_CHANNELS]; ///< exponent shift values
|
| 88 |
} AC3Block; |
| 89 |
|
| 90 |
/**
|
| 91 |
* AC-3 encoder private context.
|
| 92 |
*/
|
| 93 |
typedef struct AC3EncodeContext { |
| 94 |
PutBitContext pb; ///< bitstream writer context
|
| 95 |
DSPContext dsp; |
| 96 |
AC3MDCTContext mdct; ///< MDCT context
|
| 97 |
|
| 98 |
AC3Block blocks[AC3_MAX_BLOCKS]; ///< per-block info
|
| 99 |
|
| 100 |
int bitstream_id; ///< bitstream id (bsid) |
| 101 |
int bitstream_mode; ///< bitstream mode (bsmod) |
| 102 |
|
| 103 |
int bit_rate; ///< target bit rate, in bits-per-second |
| 104 |
int sample_rate; ///< sampling frequency, in Hz |
| 105 |
|
| 106 |
int frame_size_min; ///< minimum frame size in case rounding is necessary |
| 107 |
int frame_size; ///< current frame size in bytes |
| 108 |
int frame_size_code; ///< frame size code (frmsizecod) |
| 109 |
uint16_t crc_inv[2];
|
| 110 |
int bits_written; ///< bit count (used to avg. bitrate) |
| 111 |
int samples_written; ///< sample count (used to avg. bitrate) |
| 112 |
|
| 113 |
int fbw_channels; ///< number of full-bandwidth channels (nfchans) |
| 114 |
int channels; ///< total number of channels (nchans) |
| 115 |
int lfe_on; ///< indicates if there is an LFE channel (lfeon) |
| 116 |
int lfe_channel; ///< channel index of the LFE channel |
| 117 |
int channel_mode; ///< channel mode (acmod) |
| 118 |
const uint8_t *channel_map; ///< channel map used to reorder channels |
| 119 |
|
| 120 |
int cutoff; ///< user-specified cutoff frequency, in Hz |
| 121 |
int bandwidth_code[AC3_MAX_CHANNELS]; ///< bandwidth code (0 to 60) (chbwcod) |
| 122 |
int nb_coefs[AC3_MAX_CHANNELS];
|
| 123 |
|
| 124 |
/* bitrate allocation control */
|
| 125 |
int slow_gain_code; ///< slow gain code (sgaincod) |
| 126 |
int slow_decay_code; ///< slow decay code (sdcycod) |
| 127 |
int fast_decay_code; ///< fast decay code (fdcycod) |
| 128 |
int db_per_bit_code; ///< dB/bit code (dbpbcod) |
| 129 |
int floor_code; ///< floor code (floorcod) |
| 130 |
AC3BitAllocParameters bit_alloc; ///< bit allocation parameters
|
| 131 |
int coarse_snr_offset; ///< coarse SNR offsets (csnroffst) |
| 132 |
int fast_gain_code[AC3_MAX_CHANNELS]; ///< fast gain codes (signal-to-mask ratio) (fgaincod) |
| 133 |
int fine_snr_offset[AC3_MAX_CHANNELS]; ///< fine SNR offsets (fsnroffst) |
| 134 |
int frame_bits_fixed; ///< number of non-coefficient bits for fixed parameters |
| 135 |
int frame_bits; ///< all frame bits except exponents and mantissas |
| 136 |
int exponent_bits; ///< number of bits used for exponents |
| 137 |
|
| 138 |
/* mantissa encoding */
|
| 139 |
int mant1_cnt, mant2_cnt, mant4_cnt; ///< mantissa counts for bap=1,2,4 |
| 140 |
uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; ///< mantissa pointers for bap=1,2,4
|
| 141 |
|
| 142 |
SampleType **planar_samples; |
| 143 |
uint8_t *bap_buffer; |
| 144 |
uint8_t *bap1_buffer; |
| 145 |
CoefType *mdct_coef_buffer; |
| 146 |
uint8_t *exp_buffer; |
| 147 |
uint8_t *grouped_exp_buffer; |
| 148 |
int16_t *psd_buffer; |
| 149 |
int16_t *band_psd_buffer; |
| 150 |
int16_t *mask_buffer; |
| 151 |
uint16_t *qmant_buffer; |
| 152 |
|
| 153 |
DECLARE_ALIGNED(16, SampleType, windowed_samples)[AC3_WINDOW_SIZE];
|
| 154 |
} AC3EncodeContext; |
| 155 |
|
| 156 |
|
| 157 |
/**
|
| 158 |
* LUT for number of exponent groups.
|
| 159 |
* exponent_group_tab[exponent strategy-1][number of coefficients]
|
| 160 |
*/
|
| 161 |
static uint8_t exponent_group_tab[3][256]; |
| 162 |
|
| 163 |
|
| 164 |
/**
|
| 165 |
* List of supported channel layouts.
|
| 166 |
*/
|
| 167 |
static const int64_t ac3_channel_layouts[] = { |
| 168 |
AV_CH_LAYOUT_MONO, |
| 169 |
AV_CH_LAYOUT_STEREO, |
| 170 |
AV_CH_LAYOUT_2_1, |
| 171 |
AV_CH_LAYOUT_SURROUND, |
| 172 |
AV_CH_LAYOUT_2_2, |
| 173 |
AV_CH_LAYOUT_QUAD, |
| 174 |
AV_CH_LAYOUT_4POINT0, |
| 175 |
AV_CH_LAYOUT_5POINT0, |
| 176 |
AV_CH_LAYOUT_5POINT0_BACK, |
| 177 |
(AV_CH_LAYOUT_MONO | AV_CH_LOW_FREQUENCY), |
| 178 |
(AV_CH_LAYOUT_STEREO | AV_CH_LOW_FREQUENCY), |
| 179 |
(AV_CH_LAYOUT_2_1 | AV_CH_LOW_FREQUENCY), |
| 180 |
(AV_CH_LAYOUT_SURROUND | AV_CH_LOW_FREQUENCY), |
| 181 |
(AV_CH_LAYOUT_2_2 | AV_CH_LOW_FREQUENCY), |
| 182 |
(AV_CH_LAYOUT_QUAD | AV_CH_LOW_FREQUENCY), |
| 183 |
(AV_CH_LAYOUT_4POINT0 | AV_CH_LOW_FREQUENCY), |
| 184 |
AV_CH_LAYOUT_5POINT1, |
| 185 |
AV_CH_LAYOUT_5POINT1_BACK, |
| 186 |
0
|
| 187 |
}; |
| 188 |
|
| 189 |
|
| 190 |
/**
|
| 191 |
* Adjust the frame size to make the average bit rate match the target bit rate.
|
| 192 |
* This is only needed for 11025, 22050, and 44100 sample rates.
|
| 193 |
*/
|
| 194 |
static void adjust_frame_size(AC3EncodeContext *s) |
| 195 |
{
|
| 196 |
while (s->bits_written >= s->bit_rate && s->samples_written >= s->sample_rate) {
|
| 197 |
s->bits_written -= s->bit_rate; |
| 198 |
s->samples_written -= s->sample_rate; |
| 199 |
} |
| 200 |
s->frame_size = s->frame_size_min + |
| 201 |
2 * (s->bits_written * s->sample_rate < s->samples_written * s->bit_rate);
|
| 202 |
s->bits_written += s->frame_size * 8;
|
| 203 |
s->samples_written += AC3_FRAME_SIZE; |
| 204 |
} |
| 205 |
|
| 206 |
|
| 207 |
/**
|
| 208 |
* Deinterleave input samples.
|
| 209 |
* Channels are reordered from FFmpeg's default order to AC-3 order.
|
| 210 |
*/
|
| 211 |
static void deinterleave_input_samples(AC3EncodeContext *s, |
| 212 |
const SampleType *samples)
|
| 213 |
{
|
| 214 |
int ch, i;
|
| 215 |
|
| 216 |
/* deinterleave and remap input samples */
|
| 217 |
for (ch = 0; ch < s->channels; ch++) { |
| 218 |
const SampleType *sptr;
|
| 219 |
int sinc;
|
| 220 |
|
| 221 |
/* copy last 256 samples of previous frame to the start of the current frame */
|
| 222 |
memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_FRAME_SIZE],
|
| 223 |
AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0])); |
| 224 |
|
| 225 |
/* deinterleave */
|
| 226 |
sinc = s->channels; |
| 227 |
sptr = samples + s->channel_map[ch]; |
| 228 |
for (i = AC3_BLOCK_SIZE; i < AC3_FRAME_SIZE+AC3_BLOCK_SIZE; i++) {
|
| 229 |
s->planar_samples[ch][i] = *sptr; |
| 230 |
sptr += sinc; |
| 231 |
} |
| 232 |
} |
| 233 |
} |
| 234 |
|
| 235 |
|
| 236 |
/**
|
| 237 |
* Finalize MDCT and free allocated memory.
|
| 238 |
*/
|
| 239 |
static av_cold void mdct_end(AC3MDCTContext *mdct) |
| 240 |
{
|
| 241 |
mdct->nbits = 0;
|
| 242 |
av_freep(&mdct->costab); |
| 243 |
av_freep(&mdct->sintab); |
| 244 |
av_freep(&mdct->xcos1); |
| 245 |
av_freep(&mdct->xsin1); |
| 246 |
av_freep(&mdct->rot_tmp); |
| 247 |
av_freep(&mdct->cplx_tmp); |
| 248 |
} |
| 249 |
|
| 250 |
|
| 251 |
/**
|
| 252 |
* Initialize FFT tables.
|
| 253 |
* @param ln log2(FFT size)
|
| 254 |
*/
|
| 255 |
static av_cold int fft_init(AVCodecContext *avctx, AC3MDCTContext *mdct, int ln) |
| 256 |
{
|
| 257 |
int i, n, n2;
|
| 258 |
float alpha;
|
| 259 |
|
| 260 |
n = 1 << ln;
|
| 261 |
n2 = n >> 1;
|
| 262 |
|
| 263 |
FF_ALLOC_OR_GOTO(avctx, mdct->costab, n2 * sizeof(*mdct->costab), fft_alloc_fail);
|
| 264 |
FF_ALLOC_OR_GOTO(avctx, mdct->sintab, n2 * sizeof(*mdct->sintab), fft_alloc_fail);
|
| 265 |
|
| 266 |
for (i = 0; i < n2; i++) { |
| 267 |
alpha = 2.0 * M_PI * i / n; |
| 268 |
mdct->costab[i] = FIX15(cos(alpha)); |
| 269 |
mdct->sintab[i] = FIX15(sin(alpha)); |
| 270 |
} |
| 271 |
|
| 272 |
return 0; |
| 273 |
fft_alloc_fail:
|
| 274 |
mdct_end(mdct); |
| 275 |
return AVERROR(ENOMEM);
|
| 276 |
} |
| 277 |
|
| 278 |
|
| 279 |
/**
|
| 280 |
* Initialize MDCT tables.
|
| 281 |
* @param nbits log2(MDCT size)
|
| 282 |
*/
|
| 283 |
static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct, |
| 284 |
int nbits)
|
| 285 |
{
|
| 286 |
int i, n, n4, ret;
|
| 287 |
|
| 288 |
n = 1 << nbits;
|
| 289 |
n4 = n >> 2;
|
| 290 |
|
| 291 |
mdct->nbits = nbits; |
| 292 |
|
| 293 |
ret = fft_init(avctx, mdct, nbits - 2);
|
| 294 |
if (ret)
|
| 295 |
return ret;
|
| 296 |
|
| 297 |
mdct->window = ff_ac3_window; |
| 298 |
|
| 299 |
FF_ALLOC_OR_GOTO(avctx, mdct->xcos1, n4 * sizeof(*mdct->xcos1), mdct_alloc_fail);
|
| 300 |
FF_ALLOC_OR_GOTO(avctx, mdct->xsin1, n4 * sizeof(*mdct->xsin1), mdct_alloc_fail);
|
| 301 |
FF_ALLOC_OR_GOTO(avctx, mdct->rot_tmp, n * sizeof(*mdct->rot_tmp), mdct_alloc_fail);
|
| 302 |
FF_ALLOC_OR_GOTO(avctx, mdct->cplx_tmp, n4 * sizeof(*mdct->cplx_tmp), mdct_alloc_fail);
|
| 303 |
|
| 304 |
for (i = 0; i < n4; i++) { |
| 305 |
float alpha = 2.0 * M_PI * (i + 1.0 / 8.0) / n; |
| 306 |
mdct->xcos1[i] = FIX15(-cos(alpha)); |
| 307 |
mdct->xsin1[i] = FIX15(-sin(alpha)); |
| 308 |
} |
| 309 |
|
| 310 |
return 0; |
| 311 |
mdct_alloc_fail:
|
| 312 |
mdct_end(mdct); |
| 313 |
return AVERROR(ENOMEM);
|
| 314 |
} |
| 315 |
|
| 316 |
|
| 317 |
/** Butterfly op */
|
| 318 |
#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
|
| 319 |
{ \
|
| 320 |
int ax, ay, bx, by; \
|
| 321 |
bx = pre1; \ |
| 322 |
by = pim1; \ |
| 323 |
ax = qre1; \ |
| 324 |
ay = qim1; \ |
| 325 |
pre = (bx + ax) >> 1; \
|
| 326 |
pim = (by + ay) >> 1; \
|
| 327 |
qre = (bx - ax) >> 1; \
|
| 328 |
qim = (by - ay) >> 1; \
|
| 329 |
} |
| 330 |
|
| 331 |
|
| 332 |
/** Complex multiply */
|
| 333 |
#define CMUL(pre, pim, are, aim, bre, bim) \
|
| 334 |
{ \
|
| 335 |
pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15; \
|
| 336 |
pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15; \
|
| 337 |
} |
| 338 |
|
| 339 |
|
| 340 |
/**
|
| 341 |
* Calculate a 2^n point complex FFT on 2^ln points.
|
| 342 |
* @param z complex input/output samples
|
| 343 |
* @param ln log2(FFT size)
|
| 344 |
*/
|
| 345 |
static void fft(AC3MDCTContext *mdct, IComplex *z, int ln) |
| 346 |
{
|
| 347 |
int j, l, np, np2;
|
| 348 |
int nblocks, nloops;
|
| 349 |
register IComplex *p,*q;
|
| 350 |
int tmp_re, tmp_im;
|
| 351 |
|
| 352 |
np = 1 << ln;
|
| 353 |
|
| 354 |
/* reverse */
|
| 355 |
for (j = 0; j < np; j++) { |
| 356 |
int k = av_reverse[j] >> (8 - ln); |
| 357 |
if (k < j)
|
| 358 |
FFSWAP(IComplex, z[k], z[j]); |
| 359 |
} |
| 360 |
|
| 361 |
/* pass 0 */
|
| 362 |
|
| 363 |
p = &z[0];
|
| 364 |
j = np >> 1;
|
| 365 |
do {
|
| 366 |
BF(p[0].re, p[0].im, p[1].re, p[1].im, |
| 367 |
p[0].re, p[0].im, p[1].re, p[1].im); |
| 368 |
p += 2;
|
| 369 |
} while (--j);
|
| 370 |
|
| 371 |
/* pass 1 */
|
| 372 |
|
| 373 |
p = &z[0];
|
| 374 |
j = np >> 2;
|
| 375 |
do {
|
| 376 |
BF(p[0].re, p[0].im, p[2].re, p[2].im, |
| 377 |
p[0].re, p[0].im, p[2].re, p[2].im); |
| 378 |
BF(p[1].re, p[1].im, p[3].re, p[3].im, |
| 379 |
p[1].re, p[1].im, p[3].im, -p[3].re); |
| 380 |
p+=4;
|
| 381 |
} while (--j);
|
| 382 |
|
| 383 |
/* pass 2 .. ln-1 */
|
| 384 |
|
| 385 |
nblocks = np >> 3;
|
| 386 |
nloops = 1 << 2; |
| 387 |
np2 = np >> 1;
|
| 388 |
do {
|
| 389 |
p = z; |
| 390 |
q = z + nloops; |
| 391 |
for (j = 0; j < nblocks; j++) { |
| 392 |
BF(p->re, p->im, q->re, q->im, |
| 393 |
p->re, p->im, q->re, q->im); |
| 394 |
p++; |
| 395 |
q++; |
| 396 |
for(l = nblocks; l < np2; l += nblocks) {
|
| 397 |
CMUL(tmp_re, tmp_im, mdct->costab[l], -mdct->sintab[l], q->re, q->im); |
| 398 |
BF(p->re, p->im, q->re, q->im, |
| 399 |
p->re, p->im, tmp_re, tmp_im); |
| 400 |
p++; |
| 401 |
q++; |
| 402 |
} |
| 403 |
p += nloops; |
| 404 |
q += nloops; |
| 405 |
} |
| 406 |
nblocks = nblocks >> 1;
|
| 407 |
nloops = nloops << 1;
|
| 408 |
} while (nblocks);
|
| 409 |
} |
| 410 |
|
| 411 |
|
| 412 |
/**
|
| 413 |
* Calculate a 512-point MDCT
|
| 414 |
* @param out 256 output frequency coefficients
|
| 415 |
* @param in 512 windowed input audio samples
|
| 416 |
*/
|
| 417 |
static void mdct512(AC3MDCTContext *mdct, int32_t *out, int16_t *in) |
| 418 |
{
|
| 419 |
int i, re, im, n, n2, n4;
|
| 420 |
int16_t *rot = mdct->rot_tmp; |
| 421 |
IComplex *x = mdct->cplx_tmp; |
| 422 |
|
| 423 |
n = 1 << mdct->nbits;
|
| 424 |
n2 = n >> 1;
|
| 425 |
n4 = n >> 2;
|
| 426 |
|
| 427 |
/* shift to simplify computations */
|
| 428 |
for (i = 0; i <n4; i++) |
| 429 |
rot[i] = -in[i + 3*n4];
|
| 430 |
memcpy(&rot[n4], &in[0], 3*n4*sizeof(*in)); |
| 431 |
|
| 432 |
/* pre rotation */
|
| 433 |
for (i = 0; i < n4; i++) { |
| 434 |
re = ((int)rot[ 2*i] - (int)rot[ n-1-2*i]) >> 1; |
| 435 |
im = -((int)rot[n2+2*i] - (int)rot[n2-1-2*i]) >> 1; |
| 436 |
CMUL(x[i].re, x[i].im, re, im, -mdct->xcos1[i], mdct->xsin1[i]); |
| 437 |
} |
| 438 |
|
| 439 |
fft(mdct, x, mdct->nbits - 2);
|
| 440 |
|
| 441 |
/* post rotation */
|
| 442 |
for (i = 0; i < n4; i++) { |
| 443 |
re = x[i].re; |
| 444 |
im = x[i].im; |
| 445 |
CMUL(out[n2-1-2*i], out[2*i], re, im, mdct->xsin1[i], mdct->xcos1[i]); |
| 446 |
} |
| 447 |
} |
| 448 |
|
| 449 |
|
| 450 |
/**
|
| 451 |
* Apply KBD window to input samples prior to MDCT.
|
| 452 |
*/
|
| 453 |
static void apply_window(int16_t *output, const int16_t *input, |
| 454 |
const int16_t *window, int n) |
| 455 |
{
|
| 456 |
int i;
|
| 457 |
int n2 = n >> 1; |
| 458 |
|
| 459 |
for (i = 0; i < n2; i++) { |
| 460 |
output[i] = MUL16(input[i], window[i]) >> 15;
|
| 461 |
output[n-i-1] = MUL16(input[n-i-1], window[i]) >> 15; |
| 462 |
} |
| 463 |
} |
| 464 |
|
| 465 |
|
| 466 |
/**
|
| 467 |
* Calculate the log2() of the maximum absolute value in an array.
|
| 468 |
* @param tab input array
|
| 469 |
* @param n number of values in the array
|
| 470 |
* @return log2(max(abs(tab[])))
|
| 471 |
*/
|
| 472 |
static int log2_tab(int16_t *tab, int n) |
| 473 |
{
|
| 474 |
int i, v;
|
| 475 |
|
| 476 |
v = 0;
|
| 477 |
for (i = 0; i < n; i++) |
| 478 |
v |= abs(tab[i]); |
| 479 |
|
| 480 |
return av_log2(v);
|
| 481 |
} |
| 482 |
|
| 483 |
|
| 484 |
/**
|
| 485 |
* Left-shift each value in an array by a specified amount.
|
| 486 |
* @param tab input array
|
| 487 |
* @param n number of values in the array
|
| 488 |
* @param lshift left shift amount. a negative value means right shift.
|
| 489 |
*/
|
| 490 |
static void lshift_tab(int16_t *tab, int n, int lshift) |
| 491 |
{
|
| 492 |
int i;
|
| 493 |
|
| 494 |
if (lshift > 0) { |
| 495 |
for (i = 0; i < n; i++) |
| 496 |
tab[i] <<= lshift; |
| 497 |
} else if (lshift < 0) { |
| 498 |
lshift = -lshift; |
| 499 |
for (i = 0; i < n; i++) |
| 500 |
tab[i] >>= lshift; |
| 501 |
} |
| 502 |
} |
| 503 |
|
| 504 |
|
| 505 |
/**
|
| 506 |
* Normalize the input samples to use the maximum available precision.
|
| 507 |
* This assumes signed 16-bit input samples. Exponents are reduced by 9 to
|
| 508 |
* match the 24-bit internal precision for MDCT coefficients.
|
| 509 |
*
|
| 510 |
* @return exponent shift
|
| 511 |
*/
|
| 512 |
static int normalize_samples(AC3EncodeContext *s) |
| 513 |
{
|
| 514 |
int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE); |
| 515 |
v = FFMAX(0, v);
|
| 516 |
lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v); |
| 517 |
return v - 9; |
| 518 |
} |
| 519 |
|
| 520 |
|
| 521 |
/**
|
| 522 |
* Apply the MDCT to input samples to generate frequency coefficients.
|
| 523 |
* This applies the KBD window and normalizes the input to reduce precision
|
| 524 |
* loss due to fixed-point calculations.
|
| 525 |
*/
|
| 526 |
static void apply_mdct(AC3EncodeContext *s) |
| 527 |
{
|
| 528 |
int blk, ch;
|
| 529 |
|
| 530 |
for (ch = 0; ch < s->channels; ch++) { |
| 531 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 532 |
AC3Block *block = &s->blocks[blk]; |
| 533 |
const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
|
| 534 |
|
| 535 |
apply_window(s->windowed_samples, input_samples, s->mdct.window, AC3_WINDOW_SIZE); |
| 536 |
|
| 537 |
block->exp_shift[ch] = normalize_samples(s); |
| 538 |
|
| 539 |
mdct512(&s->mdct, block->mdct_coef[ch], s->windowed_samples); |
| 540 |
} |
| 541 |
} |
| 542 |
} |
| 543 |
|
| 544 |
|
| 545 |
/**
|
| 546 |
* Initialize exponent tables.
|
| 547 |
*/
|
| 548 |
static av_cold void exponent_init(AC3EncodeContext *s) |
| 549 |
{
|
| 550 |
int i;
|
| 551 |
for (i = 73; i < 256; i++) { |
| 552 |
exponent_group_tab[0][i] = (i - 1) / 3; |
| 553 |
exponent_group_tab[1][i] = (i + 2) / 6; |
| 554 |
exponent_group_tab[2][i] = (i + 8) / 12; |
| 555 |
} |
| 556 |
/* LFE */
|
| 557 |
exponent_group_tab[0][7] = 2; |
| 558 |
} |
| 559 |
|
| 560 |
|
| 561 |
/**
|
| 562 |
* Extract exponents from the MDCT coefficients.
|
| 563 |
* This takes into account the normalization that was done to the input samples
|
| 564 |
* by adjusting the exponents by the exponent shift values.
|
| 565 |
*/
|
| 566 |
static void extract_exponents(AC3EncodeContext *s) |
| 567 |
{
|
| 568 |
int blk, ch, i;
|
| 569 |
|
| 570 |
for (ch = 0; ch < s->channels; ch++) { |
| 571 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 572 |
AC3Block *block = &s->blocks[blk]; |
| 573 |
for (i = 0; i < AC3_MAX_COEFS; i++) { |
| 574 |
int e;
|
| 575 |
int v = abs(SCALE_COEF(block->mdct_coef[ch][i]));
|
| 576 |
if (v == 0) |
| 577 |
e = 24;
|
| 578 |
else {
|
| 579 |
e = 23 - av_log2(v) + block->exp_shift[ch];
|
| 580 |
if (e >= 24) { |
| 581 |
e = 24;
|
| 582 |
block->mdct_coef[ch][i] = 0;
|
| 583 |
} |
| 584 |
} |
| 585 |
block->exp[ch][i] = e; |
| 586 |
} |
| 587 |
} |
| 588 |
} |
| 589 |
} |
| 590 |
|
| 591 |
|
| 592 |
/**
|
| 593 |
* Exponent Difference Threshold.
|
| 594 |
* New exponents are sent if their SAD exceed this number.
|
| 595 |
*/
|
| 596 |
#define EXP_DIFF_THRESHOLD 1000 |
| 597 |
|
| 598 |
|
| 599 |
/**
|
| 600 |
* Calculate exponent strategies for all blocks in a single channel.
|
| 601 |
*/
|
| 602 |
static void compute_exp_strategy_ch(AC3EncodeContext *s, uint8_t *exp_strategy, |
| 603 |
uint8_t **exp) |
| 604 |
{
|
| 605 |
int blk, blk1;
|
| 606 |
int exp_diff;
|
| 607 |
|
| 608 |
/* estimate if the exponent variation & decide if they should be
|
| 609 |
reused in the next frame */
|
| 610 |
exp_strategy[0] = EXP_NEW;
|
| 611 |
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) { |
| 612 |
exp_diff = s->dsp.sad[0](NULL, exp[blk], exp[blk-1], 16, 16); |
| 613 |
if (exp_diff > EXP_DIFF_THRESHOLD)
|
| 614 |
exp_strategy[blk] = EXP_NEW; |
| 615 |
else
|
| 616 |
exp_strategy[blk] = EXP_REUSE; |
| 617 |
} |
| 618 |
emms_c(); |
| 619 |
|
| 620 |
/* now select the encoding strategy type : if exponents are often
|
| 621 |
recoded, we use a coarse encoding */
|
| 622 |
blk = 0;
|
| 623 |
while (blk < AC3_MAX_BLOCKS) {
|
| 624 |
blk1 = blk + 1;
|
| 625 |
while (blk1 < AC3_MAX_BLOCKS && exp_strategy[blk1] == EXP_REUSE)
|
| 626 |
blk1++; |
| 627 |
switch (blk1 - blk) {
|
| 628 |
case 1: exp_strategy[blk] = EXP_D45; break; |
| 629 |
case 2: |
| 630 |
case 3: exp_strategy[blk] = EXP_D25; break; |
| 631 |
default: exp_strategy[blk] = EXP_D15; break; |
| 632 |
} |
| 633 |
blk = blk1; |
| 634 |
} |
| 635 |
} |
| 636 |
|
| 637 |
|
| 638 |
/**
|
| 639 |
* Calculate exponent strategies for all channels.
|
| 640 |
* Array arrangement is reversed to simplify the per-channel calculation.
|
| 641 |
*/
|
| 642 |
static void compute_exp_strategy(AC3EncodeContext *s) |
| 643 |
{
|
| 644 |
uint8_t *exp1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; |
| 645 |
uint8_t exp_str1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; |
| 646 |
int ch, blk;
|
| 647 |
|
| 648 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 649 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 650 |
exp1[ch][blk] = s->blocks[blk].exp[ch]; |
| 651 |
exp_str1[ch][blk] = s->blocks[blk].exp_strategy[ch]; |
| 652 |
} |
| 653 |
|
| 654 |
compute_exp_strategy_ch(s, exp_str1[ch], exp1[ch]); |
| 655 |
|
| 656 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) |
| 657 |
s->blocks[blk].exp_strategy[ch] = exp_str1[ch][blk]; |
| 658 |
} |
| 659 |
if (s->lfe_on) {
|
| 660 |
ch = s->lfe_channel; |
| 661 |
s->blocks[0].exp_strategy[ch] = EXP_D15;
|
| 662 |
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) |
| 663 |
s->blocks[blk].exp_strategy[ch] = EXP_REUSE; |
| 664 |
} |
| 665 |
} |
| 666 |
|
| 667 |
|
| 668 |
/**
|
| 669 |
* Set each encoded exponent in a block to the minimum of itself and the
|
| 670 |
* exponent in the same frequency bin of a following block.
|
| 671 |
* exp[i] = min(exp[i], exp1[i]
|
| 672 |
*/
|
| 673 |
static void exponent_min(uint8_t *exp, uint8_t *exp1, int n) |
| 674 |
{
|
| 675 |
int i;
|
| 676 |
for (i = 0; i < n; i++) { |
| 677 |
if (exp1[i] < exp[i])
|
| 678 |
exp[i] = exp1[i]; |
| 679 |
} |
| 680 |
} |
| 681 |
|
| 682 |
|
| 683 |
/**
|
| 684 |
* Update the exponents so that they are the ones the decoder will decode.
|
| 685 |
*/
|
| 686 |
static void encode_exponents_blk_ch(uint8_t *exp, int nb_exps, int exp_strategy) |
| 687 |
{
|
| 688 |
int nb_groups, i, k;
|
| 689 |
|
| 690 |
nb_groups = exponent_group_tab[exp_strategy-1][nb_exps] * 3; |
| 691 |
|
| 692 |
/* for each group, compute the minimum exponent */
|
| 693 |
switch(exp_strategy) {
|
| 694 |
case EXP_D25:
|
| 695 |
for (i = 1, k = 1; i <= nb_groups; i++) { |
| 696 |
uint8_t exp_min = exp[k]; |
| 697 |
if (exp[k+1] < exp_min) |
| 698 |
exp_min = exp[k+1];
|
| 699 |
exp[i] = exp_min; |
| 700 |
k += 2;
|
| 701 |
} |
| 702 |
break;
|
| 703 |
case EXP_D45:
|
| 704 |
for (i = 1, k = 1; i <= nb_groups; i++) { |
| 705 |
uint8_t exp_min = exp[k]; |
| 706 |
if (exp[k+1] < exp_min) |
| 707 |
exp_min = exp[k+1];
|
| 708 |
if (exp[k+2] < exp_min) |
| 709 |
exp_min = exp[k+2];
|
| 710 |
if (exp[k+3] < exp_min) |
| 711 |
exp_min = exp[k+3];
|
| 712 |
exp[i] = exp_min; |
| 713 |
k += 4;
|
| 714 |
} |
| 715 |
break;
|
| 716 |
} |
| 717 |
|
| 718 |
/* constraint for DC exponent */
|
| 719 |
if (exp[0] > 15) |
| 720 |
exp[0] = 15; |
| 721 |
|
| 722 |
/* decrease the delta between each groups to within 2 so that they can be
|
| 723 |
differentially encoded */
|
| 724 |
for (i = 1; i <= nb_groups; i++) |
| 725 |
exp[i] = FFMIN(exp[i], exp[i-1] + 2); |
| 726 |
i--; |
| 727 |
while (--i >= 0) |
| 728 |
exp[i] = FFMIN(exp[i], exp[i+1] + 2); |
| 729 |
|
| 730 |
/* now we have the exponent values the decoder will see */
|
| 731 |
switch (exp_strategy) {
|
| 732 |
case EXP_D25:
|
| 733 |
for (i = nb_groups, k = nb_groups * 2; i > 0; i--) { |
| 734 |
uint8_t exp1 = exp[i]; |
| 735 |
exp[k--] = exp1; |
| 736 |
exp[k--] = exp1; |
| 737 |
} |
| 738 |
break;
|
| 739 |
case EXP_D45:
|
| 740 |
for (i = nb_groups, k = nb_groups * 4; i > 0; i--) { |
| 741 |
exp[k] = exp[k-1] = exp[k-2] = exp[k-3] = exp[i]; |
| 742 |
k -= 4;
|
| 743 |
} |
| 744 |
break;
|
| 745 |
} |
| 746 |
} |
| 747 |
|
| 748 |
|
| 749 |
/**
|
| 750 |
* Encode exponents from original extracted form to what the decoder will see.
|
| 751 |
* This copies and groups exponents based on exponent strategy and reduces
|
| 752 |
* deltas between adjacent exponent groups so that they can be differentially
|
| 753 |
* encoded.
|
| 754 |
*/
|
| 755 |
static void encode_exponents(AC3EncodeContext *s) |
| 756 |
{
|
| 757 |
int blk, blk1, blk2, ch;
|
| 758 |
AC3Block *block, *block1, *block2; |
| 759 |
|
| 760 |
for (ch = 0; ch < s->channels; ch++) { |
| 761 |
blk = 0;
|
| 762 |
block = &s->blocks[0];
|
| 763 |
while (blk < AC3_MAX_BLOCKS) {
|
| 764 |
blk1 = blk + 1;
|
| 765 |
block1 = block + 1;
|
| 766 |
/* for the EXP_REUSE case we select the min of the exponents */
|
| 767 |
while (blk1 < AC3_MAX_BLOCKS && block1->exp_strategy[ch] == EXP_REUSE) {
|
| 768 |
exponent_min(block->exp[ch], block1->exp[ch], s->nb_coefs[ch]); |
| 769 |
blk1++; |
| 770 |
block1++; |
| 771 |
} |
| 772 |
encode_exponents_blk_ch(block->exp[ch], s->nb_coefs[ch], |
| 773 |
block->exp_strategy[ch]); |
| 774 |
/* copy encoded exponents for reuse case */
|
| 775 |
block2 = block + 1;
|
| 776 |
for (blk2 = blk+1; blk2 < blk1; blk2++, block2++) { |
| 777 |
memcpy(block2->exp[ch], block->exp[ch], |
| 778 |
s->nb_coefs[ch] * sizeof(uint8_t));
|
| 779 |
} |
| 780 |
blk = blk1; |
| 781 |
block = block1; |
| 782 |
} |
| 783 |
} |
| 784 |
} |
| 785 |
|
| 786 |
|
| 787 |
/**
|
| 788 |
* Group exponents.
|
| 789 |
* 3 delta-encoded exponents are in each 7-bit group. The number of groups
|
| 790 |
* varies depending on exponent strategy and bandwidth.
|
| 791 |
*/
|
| 792 |
static void group_exponents(AC3EncodeContext *s) |
| 793 |
{
|
| 794 |
int blk, ch, i;
|
| 795 |
int group_size, nb_groups, bit_count;
|
| 796 |
uint8_t *p; |
| 797 |
int delta0, delta1, delta2;
|
| 798 |
int exp0, exp1;
|
| 799 |
|
| 800 |
bit_count = 0;
|
| 801 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 802 |
AC3Block *block = &s->blocks[blk]; |
| 803 |
for (ch = 0; ch < s->channels; ch++) { |
| 804 |
if (block->exp_strategy[ch] == EXP_REUSE) {
|
| 805 |
continue;
|
| 806 |
} |
| 807 |
group_size = block->exp_strategy[ch] + (block->exp_strategy[ch] == EXP_D45); |
| 808 |
nb_groups = exponent_group_tab[block->exp_strategy[ch]-1][s->nb_coefs[ch]];
|
| 809 |
bit_count += 4 + (nb_groups * 7); |
| 810 |
p = block->exp[ch]; |
| 811 |
|
| 812 |
/* DC exponent */
|
| 813 |
exp1 = *p++; |
| 814 |
block->grouped_exp[ch][0] = exp1;
|
| 815 |
|
| 816 |
/* remaining exponents are delta encoded */
|
| 817 |
for (i = 1; i <= nb_groups; i++) { |
| 818 |
/* merge three delta in one code */
|
| 819 |
exp0 = exp1; |
| 820 |
exp1 = p[0];
|
| 821 |
p += group_size; |
| 822 |
delta0 = exp1 - exp0 + 2;
|
| 823 |
|
| 824 |
exp0 = exp1; |
| 825 |
exp1 = p[0];
|
| 826 |
p += group_size; |
| 827 |
delta1 = exp1 - exp0 + 2;
|
| 828 |
|
| 829 |
exp0 = exp1; |
| 830 |
exp1 = p[0];
|
| 831 |
p += group_size; |
| 832 |
delta2 = exp1 - exp0 + 2;
|
| 833 |
|
| 834 |
block->grouped_exp[ch][i] = ((delta0 * 5 + delta1) * 5) + delta2; |
| 835 |
} |
| 836 |
} |
| 837 |
} |
| 838 |
|
| 839 |
s->exponent_bits = bit_count; |
| 840 |
} |
| 841 |
|
| 842 |
|
| 843 |
/**
|
| 844 |
* Calculate final exponents from the supplied MDCT coefficients and exponent shift.
|
| 845 |
* Extract exponents from MDCT coefficients, calculate exponent strategies,
|
| 846 |
* and encode final exponents.
|
| 847 |
*/
|
| 848 |
static void process_exponents(AC3EncodeContext *s) |
| 849 |
{
|
| 850 |
extract_exponents(s); |
| 851 |
|
| 852 |
compute_exp_strategy(s); |
| 853 |
|
| 854 |
encode_exponents(s); |
| 855 |
|
| 856 |
group_exponents(s); |
| 857 |
} |
| 858 |
|
| 859 |
|
| 860 |
/**
|
| 861 |
* Count frame bits that are based solely on fixed parameters.
|
| 862 |
* This only has to be run once when the encoder is initialized.
|
| 863 |
*/
|
| 864 |
static void count_frame_bits_fixed(AC3EncodeContext *s) |
| 865 |
{
|
| 866 |
static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 }; |
| 867 |
int blk;
|
| 868 |
int frame_bits;
|
| 869 |
|
| 870 |
/* assumptions:
|
| 871 |
* no dynamic range codes
|
| 872 |
* no channel coupling
|
| 873 |
* no rematrixing
|
| 874 |
* bit allocation parameters do not change between blocks
|
| 875 |
* SNR offsets do not change between blocks
|
| 876 |
* no delta bit allocation
|
| 877 |
* no skipped data
|
| 878 |
* no auxilliary data
|
| 879 |
*/
|
| 880 |
|
| 881 |
/* header size */
|
| 882 |
frame_bits = 65;
|
| 883 |
frame_bits += frame_bits_inc[s->channel_mode]; |
| 884 |
|
| 885 |
/* audio blocks */
|
| 886 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 887 |
frame_bits += s->fbw_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */ |
| 888 |
if (s->channel_mode == AC3_CHMODE_STEREO) {
|
| 889 |
frame_bits++; /* rematstr */
|
| 890 |
if (!blk)
|
| 891 |
frame_bits += 4;
|
| 892 |
} |
| 893 |
frame_bits += 2 * s->fbw_channels; /* chexpstr[2] * c */ |
| 894 |
if (s->lfe_on)
|
| 895 |
frame_bits++; /* lfeexpstr */
|
| 896 |
frame_bits++; /* baie */
|
| 897 |
frame_bits++; /* snr */
|
| 898 |
frame_bits += 2; /* delta / skip */ |
| 899 |
} |
| 900 |
frame_bits++; /* cplinu for block 0 */
|
| 901 |
/* bit alloc info */
|
| 902 |
/* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
|
| 903 |
/* csnroffset[6] */
|
| 904 |
/* (fsnoffset[4] + fgaincod[4]) * c */
|
| 905 |
frame_bits += 2*4 + 3 + 6 + s->channels * (4 + 3); |
| 906 |
|
| 907 |
/* auxdatae, crcrsv */
|
| 908 |
frame_bits += 2;
|
| 909 |
|
| 910 |
/* CRC */
|
| 911 |
frame_bits += 16;
|
| 912 |
|
| 913 |
s->frame_bits_fixed = frame_bits; |
| 914 |
} |
| 915 |
|
| 916 |
|
| 917 |
/**
|
| 918 |
* Initialize bit allocation.
|
| 919 |
* Set default parameter codes and calculate parameter values.
|
| 920 |
*/
|
| 921 |
static void bit_alloc_init(AC3EncodeContext *s) |
| 922 |
{
|
| 923 |
int ch;
|
| 924 |
|
| 925 |
/* init default parameters */
|
| 926 |
s->slow_decay_code = 2;
|
| 927 |
s->fast_decay_code = 1;
|
| 928 |
s->slow_gain_code = 1;
|
| 929 |
s->db_per_bit_code = 3;
|
| 930 |
s->floor_code = 4;
|
| 931 |
for (ch = 0; ch < s->channels; ch++) |
| 932 |
s->fast_gain_code[ch] = 4;
|
| 933 |
|
| 934 |
/* initial snr offset */
|
| 935 |
s->coarse_snr_offset = 40;
|
| 936 |
|
| 937 |
/* compute real values */
|
| 938 |
/* currently none of these values change during encoding, so we can just
|
| 939 |
set them once at initialization */
|
| 940 |
s->bit_alloc.slow_decay = ff_ac3_slow_decay_tab[s->slow_decay_code] >> s->bit_alloc.sr_shift; |
| 941 |
s->bit_alloc.fast_decay = ff_ac3_fast_decay_tab[s->fast_decay_code] >> s->bit_alloc.sr_shift; |
| 942 |
s->bit_alloc.slow_gain = ff_ac3_slow_gain_tab[s->slow_gain_code]; |
| 943 |
s->bit_alloc.db_per_bit = ff_ac3_db_per_bit_tab[s->db_per_bit_code]; |
| 944 |
s->bit_alloc.floor = ff_ac3_floor_tab[s->floor_code]; |
| 945 |
|
| 946 |
count_frame_bits_fixed(s); |
| 947 |
} |
| 948 |
|
| 949 |
|
| 950 |
/**
|
| 951 |
* Count the bits used to encode the frame, minus exponents and mantissas.
|
| 952 |
* Bits based on fixed parameters have already been counted, so now we just
|
| 953 |
* have to add the bits based on parameters that change during encoding.
|
| 954 |
*/
|
| 955 |
static void count_frame_bits(AC3EncodeContext *s) |
| 956 |
{
|
| 957 |
int blk, ch;
|
| 958 |
int frame_bits = 0; |
| 959 |
|
| 960 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 961 |
uint8_t *exp_strategy = s->blocks[blk].exp_strategy; |
| 962 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 963 |
if (exp_strategy[ch] != EXP_REUSE)
|
| 964 |
frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */ |
| 965 |
} |
| 966 |
} |
| 967 |
s->frame_bits = s->frame_bits_fixed + frame_bits; |
| 968 |
} |
| 969 |
|
| 970 |
|
| 971 |
/**
|
| 972 |
* Calculate the number of bits needed to encode a set of mantissas.
|
| 973 |
*/
|
| 974 |
static int compute_mantissa_size(int mant_cnt[5], uint8_t *bap, int nb_coefs) |
| 975 |
{
|
| 976 |
int bits, b, i;
|
| 977 |
|
| 978 |
bits = 0;
|
| 979 |
for (i = 0; i < nb_coefs; i++) { |
| 980 |
b = bap[i]; |
| 981 |
if (b <= 4) { |
| 982 |
// bap=1 to bap=4 will be counted in compute_mantissa_size_final
|
| 983 |
mant_cnt[b]++; |
| 984 |
} else if (b <= 13) { |
| 985 |
// bap=5 to bap=13 use (bap-1) bits
|
| 986 |
bits += b - 1;
|
| 987 |
} else {
|
| 988 |
// bap=14 uses 14 bits and bap=15 uses 16 bits
|
| 989 |
bits += (b == 14) ? 14 : 16; |
| 990 |
} |
| 991 |
} |
| 992 |
return bits;
|
| 993 |
} |
| 994 |
|
| 995 |
|
| 996 |
/**
|
| 997 |
* Finalize the mantissa bit count by adding in the grouped mantissas.
|
| 998 |
*/
|
| 999 |
static int compute_mantissa_size_final(int mant_cnt[5]) |
| 1000 |
{
|
| 1001 |
// bap=1 : 3 mantissas in 5 bits
|
| 1002 |
int bits = (mant_cnt[1] / 3) * 5; |
| 1003 |
// bap=2 : 3 mantissas in 7 bits
|
| 1004 |
// bap=4 : 2 mantissas in 7 bits
|
| 1005 |
bits += ((mant_cnt[2] / 3) + (mant_cnt[4] >> 1)) * 7; |
| 1006 |
// bap=3 : each mantissa is 3 bits
|
| 1007 |
bits += mant_cnt[3] * 3; |
| 1008 |
return bits;
|
| 1009 |
} |
| 1010 |
|
| 1011 |
|
| 1012 |
/**
|
| 1013 |
* Calculate masking curve based on the final exponents.
|
| 1014 |
* Also calculate the power spectral densities to use in future calculations.
|
| 1015 |
*/
|
| 1016 |
static void bit_alloc_masking(AC3EncodeContext *s) |
| 1017 |
{
|
| 1018 |
int blk, ch;
|
| 1019 |
|
| 1020 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1021 |
AC3Block *block = &s->blocks[blk]; |
| 1022 |
for (ch = 0; ch < s->channels; ch++) { |
| 1023 |
/* We only need psd and mask for calculating bap.
|
| 1024 |
Since we currently do not calculate bap when exponent
|
| 1025 |
strategy is EXP_REUSE we do not need to calculate psd or mask. */
|
| 1026 |
if (block->exp_strategy[ch] != EXP_REUSE) {
|
| 1027 |
ff_ac3_bit_alloc_calc_psd(block->exp[ch], 0,
|
| 1028 |
s->nb_coefs[ch], |
| 1029 |
block->psd[ch], block->band_psd[ch]); |
| 1030 |
ff_ac3_bit_alloc_calc_mask(&s->bit_alloc, block->band_psd[ch], |
| 1031 |
0, s->nb_coefs[ch],
|
| 1032 |
ff_ac3_fast_gain_tab[s->fast_gain_code[ch]], |
| 1033 |
ch == s->lfe_channel, |
| 1034 |
DBA_NONE, 0, NULL, NULL, NULL, |
| 1035 |
block->mask[ch]); |
| 1036 |
} |
| 1037 |
} |
| 1038 |
} |
| 1039 |
} |
| 1040 |
|
| 1041 |
|
| 1042 |
/**
|
| 1043 |
* Ensure that bap for each block and channel point to the current bap_buffer.
|
| 1044 |
* They may have been switched during the bit allocation search.
|
| 1045 |
*/
|
| 1046 |
static void reset_block_bap(AC3EncodeContext *s) |
| 1047 |
{
|
| 1048 |
int blk, ch;
|
| 1049 |
if (s->blocks[0].bap[0] == s->bap_buffer) |
| 1050 |
return;
|
| 1051 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1052 |
for (ch = 0; ch < s->channels; ch++) { |
| 1053 |
s->blocks[blk].bap[ch] = &s->bap_buffer[AC3_MAX_COEFS * (blk * s->channels + ch)]; |
| 1054 |
} |
| 1055 |
} |
| 1056 |
} |
| 1057 |
|
| 1058 |
|
| 1059 |
/**
|
| 1060 |
* Run the bit allocation with a given SNR offset.
|
| 1061 |
* This calculates the bit allocation pointers that will be used to determine
|
| 1062 |
* the quantization of each mantissa.
|
| 1063 |
* @return the number of bits needed for mantissas if the given SNR offset is
|
| 1064 |
* is used.
|
| 1065 |
*/
|
| 1066 |
static int bit_alloc(AC3EncodeContext *s, int snr_offset) |
| 1067 |
{
|
| 1068 |
int blk, ch;
|
| 1069 |
int mantissa_bits;
|
| 1070 |
int mant_cnt[5]; |
| 1071 |
|
| 1072 |
snr_offset = (snr_offset - 240) << 2; |
| 1073 |
|
| 1074 |
reset_block_bap(s); |
| 1075 |
mantissa_bits = 0;
|
| 1076 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1077 |
AC3Block *block = &s->blocks[blk]; |
| 1078 |
// initialize grouped mantissa counts. these are set so that they are
|
| 1079 |
// padded to the next whole group size when bits are counted in
|
| 1080 |
// compute_mantissa_size_final
|
| 1081 |
mant_cnt[0] = mant_cnt[3] = 0; |
| 1082 |
mant_cnt[1] = mant_cnt[2] = 2; |
| 1083 |
mant_cnt[4] = 1; |
| 1084 |
for (ch = 0; ch < s->channels; ch++) { |
| 1085 |
/* Currently the only bit allocation parameters which vary across
|
| 1086 |
blocks within a frame are the exponent values. We can take
|
| 1087 |
advantage of that by reusing the bit allocation pointers
|
| 1088 |
whenever we reuse exponents. */
|
| 1089 |
if (block->exp_strategy[ch] == EXP_REUSE) {
|
| 1090 |
memcpy(block->bap[ch], s->blocks[blk-1].bap[ch], AC3_MAX_COEFS);
|
| 1091 |
} else {
|
| 1092 |
ff_ac3_bit_alloc_calc_bap(block->mask[ch], block->psd[ch], 0,
|
| 1093 |
s->nb_coefs[ch], snr_offset, |
| 1094 |
s->bit_alloc.floor, ff_ac3_bap_tab, |
| 1095 |
block->bap[ch]); |
| 1096 |
} |
| 1097 |
mantissa_bits += compute_mantissa_size(mant_cnt, block->bap[ch], s->nb_coefs[ch]); |
| 1098 |
} |
| 1099 |
mantissa_bits += compute_mantissa_size_final(mant_cnt); |
| 1100 |
} |
| 1101 |
return mantissa_bits;
|
| 1102 |
} |
| 1103 |
|
| 1104 |
|
| 1105 |
/**
|
| 1106 |
* Constant bitrate bit allocation search.
|
| 1107 |
* Find the largest SNR offset that will allow data to fit in the frame.
|
| 1108 |
*/
|
| 1109 |
static int cbr_bit_allocation(AC3EncodeContext *s) |
| 1110 |
{
|
| 1111 |
int ch;
|
| 1112 |
int bits_left;
|
| 1113 |
int snr_offset, snr_incr;
|
| 1114 |
|
| 1115 |
bits_left = 8 * s->frame_size - (s->frame_bits + s->exponent_bits);
|
| 1116 |
|
| 1117 |
snr_offset = s->coarse_snr_offset << 4;
|
| 1118 |
|
| 1119 |
while (snr_offset >= 0 && |
| 1120 |
bit_alloc(s, snr_offset) > bits_left) {
|
| 1121 |
snr_offset -= 64;
|
| 1122 |
} |
| 1123 |
if (snr_offset < 0) |
| 1124 |
return AVERROR(EINVAL);
|
| 1125 |
|
| 1126 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
| 1127 |
for (snr_incr = 64; snr_incr > 0; snr_incr >>= 2) { |
| 1128 |
while (snr_offset + snr_incr <= 1023 && |
| 1129 |
bit_alloc(s, snr_offset + snr_incr) <= bits_left) {
|
| 1130 |
snr_offset += snr_incr; |
| 1131 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
| 1132 |
} |
| 1133 |
} |
| 1134 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
| 1135 |
reset_block_bap(s); |
| 1136 |
|
| 1137 |
s->coarse_snr_offset = snr_offset >> 4;
|
| 1138 |
for (ch = 0; ch < s->channels; ch++) |
| 1139 |
s->fine_snr_offset[ch] = snr_offset & 0xF;
|
| 1140 |
|
| 1141 |
return 0; |
| 1142 |
} |
| 1143 |
|
| 1144 |
|
| 1145 |
/**
|
| 1146 |
* Downgrade exponent strategies to reduce the bits used by the exponents.
|
| 1147 |
* This is a fallback for when bit allocation fails with the normal exponent
|
| 1148 |
* strategies. Each time this function is run it only downgrades the
|
| 1149 |
* strategy in 1 channel of 1 block.
|
| 1150 |
* @return non-zero if downgrade was unsuccessful
|
| 1151 |
*/
|
| 1152 |
static int downgrade_exponents(AC3EncodeContext *s) |
| 1153 |
{
|
| 1154 |
int ch, blk;
|
| 1155 |
|
| 1156 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1157 |
for (blk = AC3_MAX_BLOCKS-1; blk >= 0; blk--) { |
| 1158 |
if (s->blocks[blk].exp_strategy[ch] == EXP_D15) {
|
| 1159 |
s->blocks[blk].exp_strategy[ch] = EXP_D25; |
| 1160 |
return 0; |
| 1161 |
} |
| 1162 |
} |
| 1163 |
} |
| 1164 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1165 |
for (blk = AC3_MAX_BLOCKS-1; blk >= 0; blk--) { |
| 1166 |
if (s->blocks[blk].exp_strategy[ch] == EXP_D25) {
|
| 1167 |
s->blocks[blk].exp_strategy[ch] = EXP_D45; |
| 1168 |
return 0; |
| 1169 |
} |
| 1170 |
} |
| 1171 |
} |
| 1172 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1173 |
/* block 0 cannot reuse exponents, so only downgrade D45 to REUSE if
|
| 1174 |
the block number > 0 */
|
| 1175 |
for (blk = AC3_MAX_BLOCKS-1; blk > 0; blk--) { |
| 1176 |
if (s->blocks[blk].exp_strategy[ch] > EXP_REUSE) {
|
| 1177 |
s->blocks[blk].exp_strategy[ch] = EXP_REUSE; |
| 1178 |
return 0; |
| 1179 |
} |
| 1180 |
} |
| 1181 |
} |
| 1182 |
return -1; |
| 1183 |
} |
| 1184 |
|
| 1185 |
|
| 1186 |
/**
|
| 1187 |
* Reduce the bandwidth to reduce the number of bits used for a given SNR offset.
|
| 1188 |
* This is a second fallback for when bit allocation still fails after exponents
|
| 1189 |
* have been downgraded.
|
| 1190 |
* @return non-zero if bandwidth reduction was unsuccessful
|
| 1191 |
*/
|
| 1192 |
static int reduce_bandwidth(AC3EncodeContext *s, int min_bw_code) |
| 1193 |
{
|
| 1194 |
int ch;
|
| 1195 |
|
| 1196 |
if (s->bandwidth_code[0] > min_bw_code) { |
| 1197 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1198 |
s->bandwidth_code[ch]--; |
| 1199 |
s->nb_coefs[ch] = s->bandwidth_code[ch] * 3 + 73; |
| 1200 |
} |
| 1201 |
return 0; |
| 1202 |
} |
| 1203 |
return -1; |
| 1204 |
} |
| 1205 |
|
| 1206 |
|
| 1207 |
/**
|
| 1208 |
* Perform bit allocation search.
|
| 1209 |
* Finds the SNR offset value that maximizes quality and fits in the specified
|
| 1210 |
* frame size. Output is the SNR offset and a set of bit allocation pointers
|
| 1211 |
* used to quantize the mantissas.
|
| 1212 |
*/
|
| 1213 |
static int compute_bit_allocation(AC3EncodeContext *s) |
| 1214 |
{
|
| 1215 |
int ret;
|
| 1216 |
|
| 1217 |
count_frame_bits(s); |
| 1218 |
|
| 1219 |
bit_alloc_masking(s); |
| 1220 |
|
| 1221 |
ret = cbr_bit_allocation(s); |
| 1222 |
while (ret) {
|
| 1223 |
/* fallback 1: downgrade exponents */
|
| 1224 |
if (!downgrade_exponents(s)) {
|
| 1225 |
extract_exponents(s); |
| 1226 |
encode_exponents(s); |
| 1227 |
group_exponents(s); |
| 1228 |
ret = compute_bit_allocation(s); |
| 1229 |
continue;
|
| 1230 |
} |
| 1231 |
|
| 1232 |
/* fallback 2: reduce bandwidth */
|
| 1233 |
/* only do this if the user has not specified a specific cutoff
|
| 1234 |
frequency */
|
| 1235 |
if (!s->cutoff && !reduce_bandwidth(s, 0)) { |
| 1236 |
process_exponents(s); |
| 1237 |
ret = compute_bit_allocation(s); |
| 1238 |
continue;
|
| 1239 |
} |
| 1240 |
|
| 1241 |
/* fallbacks were not enough... */
|
| 1242 |
break;
|
| 1243 |
} |
| 1244 |
|
| 1245 |
return ret;
|
| 1246 |
} |
| 1247 |
|
| 1248 |
|
| 1249 |
/**
|
| 1250 |
* Symmetric quantization on 'levels' levels.
|
| 1251 |
*/
|
| 1252 |
static inline int sym_quant(int c, int e, int levels) |
| 1253 |
{
|
| 1254 |
int v;
|
| 1255 |
|
| 1256 |
if (c >= 0) { |
| 1257 |
v = (levels * (c << e)) >> 24;
|
| 1258 |
v = (v + 1) >> 1; |
| 1259 |
v = (levels >> 1) + v;
|
| 1260 |
} else {
|
| 1261 |
v = (levels * ((-c) << e)) >> 24;
|
| 1262 |
v = (v + 1) >> 1; |
| 1263 |
v = (levels >> 1) - v;
|
| 1264 |
} |
| 1265 |
assert(v >= 0 && v < levels);
|
| 1266 |
return v;
|
| 1267 |
} |
| 1268 |
|
| 1269 |
|
| 1270 |
/**
|
| 1271 |
* Asymmetric quantization on 2^qbits levels.
|
| 1272 |
*/
|
| 1273 |
static inline int asym_quant(int c, int e, int qbits) |
| 1274 |
{
|
| 1275 |
int lshift, m, v;
|
| 1276 |
|
| 1277 |
lshift = e + qbits - 24;
|
| 1278 |
if (lshift >= 0) |
| 1279 |
v = c << lshift; |
| 1280 |
else
|
| 1281 |
v = c >> (-lshift); |
| 1282 |
/* rounding */
|
| 1283 |
v = (v + 1) >> 1; |
| 1284 |
m = (1 << (qbits-1)); |
| 1285 |
if (v >= m)
|
| 1286 |
v = m - 1;
|
| 1287 |
assert(v >= -m); |
| 1288 |
return v & ((1 << qbits)-1); |
| 1289 |
} |
| 1290 |
|
| 1291 |
|
| 1292 |
/**
|
| 1293 |
* Quantize a set of mantissas for a single channel in a single block.
|
| 1294 |
*/
|
| 1295 |
static void quantize_mantissas_blk_ch(AC3EncodeContext *s, CoefType *mdct_coef, |
| 1296 |
int8_t exp_shift, uint8_t *exp, |
| 1297 |
uint8_t *bap, uint16_t *qmant, int n)
|
| 1298 |
{
|
| 1299 |
int i;
|
| 1300 |
|
| 1301 |
for (i = 0; i < n; i++) { |
| 1302 |
int v;
|
| 1303 |
int c = SCALE_COEF(mdct_coef[i]);
|
| 1304 |
int e = exp[i] - exp_shift;
|
| 1305 |
int b = bap[i];
|
| 1306 |
switch (b) {
|
| 1307 |
case 0: |
| 1308 |
v = 0;
|
| 1309 |
break;
|
| 1310 |
case 1: |
| 1311 |
v = sym_quant(c, e, 3);
|
| 1312 |
switch (s->mant1_cnt) {
|
| 1313 |
case 0: |
| 1314 |
s->qmant1_ptr = &qmant[i]; |
| 1315 |
v = 9 * v;
|
| 1316 |
s->mant1_cnt = 1;
|
| 1317 |
break;
|
| 1318 |
case 1: |
| 1319 |
*s->qmant1_ptr += 3 * v;
|
| 1320 |
s->mant1_cnt = 2;
|
| 1321 |
v = 128;
|
| 1322 |
break;
|
| 1323 |
default:
|
| 1324 |
*s->qmant1_ptr += v; |
| 1325 |
s->mant1_cnt = 0;
|
| 1326 |
v = 128;
|
| 1327 |
break;
|
| 1328 |
} |
| 1329 |
break;
|
| 1330 |
case 2: |
| 1331 |
v = sym_quant(c, e, 5);
|
| 1332 |
switch (s->mant2_cnt) {
|
| 1333 |
case 0: |
| 1334 |
s->qmant2_ptr = &qmant[i]; |
| 1335 |
v = 25 * v;
|
| 1336 |
s->mant2_cnt = 1;
|
| 1337 |
break;
|
| 1338 |
case 1: |
| 1339 |
*s->qmant2_ptr += 5 * v;
|
| 1340 |
s->mant2_cnt = 2;
|
| 1341 |
v = 128;
|
| 1342 |
break;
|
| 1343 |
default:
|
| 1344 |
*s->qmant2_ptr += v; |
| 1345 |
s->mant2_cnt = 0;
|
| 1346 |
v = 128;
|
| 1347 |
break;
|
| 1348 |
} |
| 1349 |
break;
|
| 1350 |
case 3: |
| 1351 |
v = sym_quant(c, e, 7);
|
| 1352 |
break;
|
| 1353 |
case 4: |
| 1354 |
v = sym_quant(c, e, 11);
|
| 1355 |
switch (s->mant4_cnt) {
|
| 1356 |
case 0: |
| 1357 |
s->qmant4_ptr = &qmant[i]; |
| 1358 |
v = 11 * v;
|
| 1359 |
s->mant4_cnt = 1;
|
| 1360 |
break;
|
| 1361 |
default:
|
| 1362 |
*s->qmant4_ptr += v; |
| 1363 |
s->mant4_cnt = 0;
|
| 1364 |
v = 128;
|
| 1365 |
break;
|
| 1366 |
} |
| 1367 |
break;
|
| 1368 |
case 5: |
| 1369 |
v = sym_quant(c, e, 15);
|
| 1370 |
break;
|
| 1371 |
case 14: |
| 1372 |
v = asym_quant(c, e, 14);
|
| 1373 |
break;
|
| 1374 |
case 15: |
| 1375 |
v = asym_quant(c, e, 16);
|
| 1376 |
break;
|
| 1377 |
default:
|
| 1378 |
v = asym_quant(c, e, b - 1);
|
| 1379 |
break;
|
| 1380 |
} |
| 1381 |
qmant[i] = v; |
| 1382 |
} |
| 1383 |
} |
| 1384 |
|
| 1385 |
|
| 1386 |
/**
|
| 1387 |
* Quantize mantissas using coefficients, exponents, and bit allocation pointers.
|
| 1388 |
*/
|
| 1389 |
static void quantize_mantissas(AC3EncodeContext *s) |
| 1390 |
{
|
| 1391 |
int blk, ch;
|
| 1392 |
|
| 1393 |
|
| 1394 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1395 |
AC3Block *block = &s->blocks[blk]; |
| 1396 |
s->mant1_cnt = s->mant2_cnt = s->mant4_cnt = 0;
|
| 1397 |
s->qmant1_ptr = s->qmant2_ptr = s->qmant4_ptr = NULL;
|
| 1398 |
|
| 1399 |
for (ch = 0; ch < s->channels; ch++) { |
| 1400 |
quantize_mantissas_blk_ch(s, block->mdct_coef[ch], block->exp_shift[ch], |
| 1401 |
block->exp[ch], block->bap[ch], |
| 1402 |
block->qmant[ch], s->nb_coefs[ch]); |
| 1403 |
} |
| 1404 |
} |
| 1405 |
} |
| 1406 |
|
| 1407 |
|
| 1408 |
/**
|
| 1409 |
* Write the AC-3 frame header to the output bitstream.
|
| 1410 |
*/
|
| 1411 |
static void output_frame_header(AC3EncodeContext *s) |
| 1412 |
{
|
| 1413 |
put_bits(&s->pb, 16, 0x0b77); /* frame header */ |
| 1414 |
put_bits(&s->pb, 16, 0); /* crc1: will be filled later */ |
| 1415 |
put_bits(&s->pb, 2, s->bit_alloc.sr_code);
|
| 1416 |
put_bits(&s->pb, 6, s->frame_size_code + (s->frame_size - s->frame_size_min) / 2); |
| 1417 |
put_bits(&s->pb, 5, s->bitstream_id);
|
| 1418 |
put_bits(&s->pb, 3, s->bitstream_mode);
|
| 1419 |
put_bits(&s->pb, 3, s->channel_mode);
|
| 1420 |
if ((s->channel_mode & 0x01) && s->channel_mode != AC3_CHMODE_MONO) |
| 1421 |
put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */ |
| 1422 |
if (s->channel_mode & 0x04) |
| 1423 |
put_bits(&s->pb, 2, 1); /* XXX -6 dB */ |
| 1424 |
if (s->channel_mode == AC3_CHMODE_STEREO)
|
| 1425 |
put_bits(&s->pb, 2, 0); /* surround not indicated */ |
| 1426 |
put_bits(&s->pb, 1, s->lfe_on); /* LFE */ |
| 1427 |
put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */ |
| 1428 |
put_bits(&s->pb, 1, 0); /* no compression control word */ |
| 1429 |
put_bits(&s->pb, 1, 0); /* no lang code */ |
| 1430 |
put_bits(&s->pb, 1, 0); /* no audio production info */ |
| 1431 |
put_bits(&s->pb, 1, 0); /* no copyright */ |
| 1432 |
put_bits(&s->pb, 1, 1); /* original bitstream */ |
| 1433 |
put_bits(&s->pb, 1, 0); /* no time code 1 */ |
| 1434 |
put_bits(&s->pb, 1, 0); /* no time code 2 */ |
| 1435 |
put_bits(&s->pb, 1, 0); /* no additional bit stream info */ |
| 1436 |
} |
| 1437 |
|
| 1438 |
|
| 1439 |
/**
|
| 1440 |
* Write one audio block to the output bitstream.
|
| 1441 |
*/
|
| 1442 |
static void output_audio_block(AC3EncodeContext *s, int block_num) |
| 1443 |
{
|
| 1444 |
int ch, i, baie, rbnd;
|
| 1445 |
AC3Block *block = &s->blocks[block_num]; |
| 1446 |
|
| 1447 |
/* block switching */
|
| 1448 |
for (ch = 0; ch < s->fbw_channels; ch++) |
| 1449 |
put_bits(&s->pb, 1, 0); |
| 1450 |
|
| 1451 |
/* dither flags */
|
| 1452 |
for (ch = 0; ch < s->fbw_channels; ch++) |
| 1453 |
put_bits(&s->pb, 1, 1); |
| 1454 |
|
| 1455 |
/* dynamic range codes */
|
| 1456 |
put_bits(&s->pb, 1, 0); |
| 1457 |
|
| 1458 |
/* channel coupling */
|
| 1459 |
if (!block_num) {
|
| 1460 |
put_bits(&s->pb, 1, 1); /* coupling strategy present */ |
| 1461 |
put_bits(&s->pb, 1, 0); /* no coupling strategy */ |
| 1462 |
} else {
|
| 1463 |
put_bits(&s->pb, 1, 0); /* no new coupling strategy */ |
| 1464 |
} |
| 1465 |
|
| 1466 |
/* stereo rematrixing */
|
| 1467 |
if (s->channel_mode == AC3_CHMODE_STEREO) {
|
| 1468 |
if (!block_num) {
|
| 1469 |
/* first block must define rematrixing (rematstr) */
|
| 1470 |
put_bits(&s->pb, 1, 1); |
| 1471 |
|
| 1472 |
/* dummy rematrixing rematflg(1:4)=0 */
|
| 1473 |
for (rbnd = 0; rbnd < 4; rbnd++) |
| 1474 |
put_bits(&s->pb, 1, 0); |
| 1475 |
} else {
|
| 1476 |
/* no matrixing (but should be used in the future) */
|
| 1477 |
put_bits(&s->pb, 1, 0); |
| 1478 |
} |
| 1479 |
} |
| 1480 |
|
| 1481 |
/* exponent strategy */
|
| 1482 |
for (ch = 0; ch < s->fbw_channels; ch++) |
| 1483 |
put_bits(&s->pb, 2, block->exp_strategy[ch]);
|
| 1484 |
if (s->lfe_on)
|
| 1485 |
put_bits(&s->pb, 1, block->exp_strategy[s->lfe_channel]);
|
| 1486 |
|
| 1487 |
/* bandwidth */
|
| 1488 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1489 |
if (block->exp_strategy[ch] != EXP_REUSE)
|
| 1490 |
put_bits(&s->pb, 6, s->bandwidth_code[ch]);
|
| 1491 |
} |
| 1492 |
|
| 1493 |
/* exponents */
|
| 1494 |
for (ch = 0; ch < s->channels; ch++) { |
| 1495 |
int nb_groups;
|
| 1496 |
|
| 1497 |
if (block->exp_strategy[ch] == EXP_REUSE)
|
| 1498 |
continue;
|
| 1499 |
|
| 1500 |
/* DC exponent */
|
| 1501 |
put_bits(&s->pb, 4, block->grouped_exp[ch][0]); |
| 1502 |
|
| 1503 |
/* exponent groups */
|
| 1504 |
nb_groups = exponent_group_tab[block->exp_strategy[ch]-1][s->nb_coefs[ch]];
|
| 1505 |
for (i = 1; i <= nb_groups; i++) |
| 1506 |
put_bits(&s->pb, 7, block->grouped_exp[ch][i]);
|
| 1507 |
|
| 1508 |
/* gain range info */
|
| 1509 |
if (ch != s->lfe_channel)
|
| 1510 |
put_bits(&s->pb, 2, 0); |
| 1511 |
} |
| 1512 |
|
| 1513 |
/* bit allocation info */
|
| 1514 |
baie = (block_num == 0);
|
| 1515 |
put_bits(&s->pb, 1, baie);
|
| 1516 |
if (baie) {
|
| 1517 |
put_bits(&s->pb, 2, s->slow_decay_code);
|
| 1518 |
put_bits(&s->pb, 2, s->fast_decay_code);
|
| 1519 |
put_bits(&s->pb, 2, s->slow_gain_code);
|
| 1520 |
put_bits(&s->pb, 2, s->db_per_bit_code);
|
| 1521 |
put_bits(&s->pb, 3, s->floor_code);
|
| 1522 |
} |
| 1523 |
|
| 1524 |
/* snr offset */
|
| 1525 |
put_bits(&s->pb, 1, baie);
|
| 1526 |
if (baie) {
|
| 1527 |
put_bits(&s->pb, 6, s->coarse_snr_offset);
|
| 1528 |
for (ch = 0; ch < s->channels; ch++) { |
| 1529 |
put_bits(&s->pb, 4, s->fine_snr_offset[ch]);
|
| 1530 |
put_bits(&s->pb, 3, s->fast_gain_code[ch]);
|
| 1531 |
} |
| 1532 |
} |
| 1533 |
|
| 1534 |
put_bits(&s->pb, 1, 0); /* no delta bit allocation */ |
| 1535 |
put_bits(&s->pb, 1, 0); /* no data to skip */ |
| 1536 |
|
| 1537 |
/* mantissas */
|
| 1538 |
for (ch = 0; ch < s->channels; ch++) { |
| 1539 |
int b, q;
|
| 1540 |
for (i = 0; i < s->nb_coefs[ch]; i++) { |
| 1541 |
q = block->qmant[ch][i]; |
| 1542 |
b = block->bap[ch][i]; |
| 1543 |
switch (b) {
|
| 1544 |
case 0: break; |
| 1545 |
case 1: if (q != 128) put_bits(&s->pb, 5, q); break; |
| 1546 |
case 2: if (q != 128) put_bits(&s->pb, 7, q); break; |
| 1547 |
case 3: put_bits(&s->pb, 3, q); break; |
| 1548 |
case 4: if (q != 128) put_bits(&s->pb, 7, q); break; |
| 1549 |
case 14: put_bits(&s->pb, 14, q); break; |
| 1550 |
case 15: put_bits(&s->pb, 16, q); break; |
| 1551 |
default: put_bits(&s->pb, b-1, q); break; |
| 1552 |
} |
| 1553 |
} |
| 1554 |
} |
| 1555 |
} |
| 1556 |
|
| 1557 |
|
| 1558 |
/** CRC-16 Polynomial */
|
| 1559 |
#define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16)) |
| 1560 |
|
| 1561 |
|
| 1562 |
static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly) |
| 1563 |
{
|
| 1564 |
unsigned int c; |
| 1565 |
|
| 1566 |
c = 0;
|
| 1567 |
while (a) {
|
| 1568 |
if (a & 1) |
| 1569 |
c ^= b; |
| 1570 |
a = a >> 1;
|
| 1571 |
b = b << 1;
|
| 1572 |
if (b & (1 << 16)) |
| 1573 |
b ^= poly; |
| 1574 |
} |
| 1575 |
return c;
|
| 1576 |
} |
| 1577 |
|
| 1578 |
|
| 1579 |
static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly) |
| 1580 |
{
|
| 1581 |
unsigned int r; |
| 1582 |
r = 1;
|
| 1583 |
while (n) {
|
| 1584 |
if (n & 1) |
| 1585 |
r = mul_poly(r, a, poly); |
| 1586 |
a = mul_poly(a, a, poly); |
| 1587 |
n >>= 1;
|
| 1588 |
} |
| 1589 |
return r;
|
| 1590 |
} |
| 1591 |
|
| 1592 |
|
| 1593 |
/**
|
| 1594 |
* Fill the end of the frame with 0's and compute the two CRCs.
|
| 1595 |
*/
|
| 1596 |
static void output_frame_end(AC3EncodeContext *s) |
| 1597 |
{
|
| 1598 |
const AVCRC *crc_ctx = av_crc_get_table(AV_CRC_16_ANSI);
|
| 1599 |
int frame_size_58, pad_bytes, crc1, crc2_partial, crc2, crc_inv;
|
| 1600 |
uint8_t *frame; |
| 1601 |
|
| 1602 |
frame_size_58 = ((s->frame_size >> 2) + (s->frame_size >> 4)) << 1; |
| 1603 |
|
| 1604 |
/* pad the remainder of the frame with zeros */
|
| 1605 |
flush_put_bits(&s->pb); |
| 1606 |
frame = s->pb.buf; |
| 1607 |
pad_bytes = s->frame_size - (put_bits_ptr(&s->pb) - frame) - 2;
|
| 1608 |
assert(pad_bytes >= 0);
|
| 1609 |
if (pad_bytes > 0) |
| 1610 |
memset(put_bits_ptr(&s->pb), 0, pad_bytes);
|
| 1611 |
|
| 1612 |
/* compute crc1 */
|
| 1613 |
/* this is not so easy because it is at the beginning of the data... */
|
| 1614 |
crc1 = av_bswap16(av_crc(crc_ctx, 0, frame + 4, frame_size_58 - 4)); |
| 1615 |
crc_inv = s->crc_inv[s->frame_size > s->frame_size_min]; |
| 1616 |
crc1 = mul_poly(crc_inv, crc1, CRC16_POLY); |
| 1617 |
AV_WB16(frame + 2, crc1);
|
| 1618 |
|
| 1619 |
/* compute crc2 */
|
| 1620 |
crc2_partial = av_crc(crc_ctx, 0, frame + frame_size_58,
|
| 1621 |
s->frame_size - frame_size_58 - 3);
|
| 1622 |
crc2 = av_crc(crc_ctx, crc2_partial, frame + s->frame_size - 3, 1); |
| 1623 |
/* ensure crc2 does not match sync word by flipping crcrsv bit if needed */
|
| 1624 |
if (crc2 == 0x770B) { |
| 1625 |
frame[s->frame_size - 3] ^= 0x1; |
| 1626 |
crc2 = av_crc(crc_ctx, crc2_partial, frame + s->frame_size - 3, 1); |
| 1627 |
} |
| 1628 |
crc2 = av_bswap16(crc2); |
| 1629 |
AV_WB16(frame + s->frame_size - 2, crc2);
|
| 1630 |
} |
| 1631 |
|
| 1632 |
|
| 1633 |
/**
|
| 1634 |
* Write the frame to the output bitstream.
|
| 1635 |
*/
|
| 1636 |
static void output_frame(AC3EncodeContext *s, unsigned char *frame) |
| 1637 |
{
|
| 1638 |
int blk;
|
| 1639 |
|
| 1640 |
init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE); |
| 1641 |
|
| 1642 |
output_frame_header(s); |
| 1643 |
|
| 1644 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) |
| 1645 |
output_audio_block(s, blk); |
| 1646 |
|
| 1647 |
output_frame_end(s); |
| 1648 |
} |
| 1649 |
|
| 1650 |
|
| 1651 |
/**
|
| 1652 |
* Encode a single AC-3 frame.
|
| 1653 |
*/
|
| 1654 |
static int ac3_encode_frame(AVCodecContext *avctx, unsigned char *frame, |
| 1655 |
int buf_size, void *data) |
| 1656 |
{
|
| 1657 |
AC3EncodeContext *s = avctx->priv_data; |
| 1658 |
const SampleType *samples = data;
|
| 1659 |
int ret;
|
| 1660 |
|
| 1661 |
if (s->bit_alloc.sr_code == 1) |
| 1662 |
adjust_frame_size(s); |
| 1663 |
|
| 1664 |
deinterleave_input_samples(s, samples); |
| 1665 |
|
| 1666 |
apply_mdct(s); |
| 1667 |
|
| 1668 |
process_exponents(s); |
| 1669 |
|
| 1670 |
ret = compute_bit_allocation(s); |
| 1671 |
if (ret) {
|
| 1672 |
av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
|
| 1673 |
return ret;
|
| 1674 |
} |
| 1675 |
|
| 1676 |
quantize_mantissas(s); |
| 1677 |
|
| 1678 |
output_frame(s, frame); |
| 1679 |
|
| 1680 |
return s->frame_size;
|
| 1681 |
} |
| 1682 |
|
| 1683 |
|
| 1684 |
/**
|
| 1685 |
* Finalize encoding and free any memory allocated by the encoder.
|
| 1686 |
*/
|
| 1687 |
static av_cold int ac3_encode_close(AVCodecContext *avctx) |
| 1688 |
{
|
| 1689 |
int blk, ch;
|
| 1690 |
AC3EncodeContext *s = avctx->priv_data; |
| 1691 |
|
| 1692 |
for (ch = 0; ch < s->channels; ch++) |
| 1693 |
av_freep(&s->planar_samples[ch]); |
| 1694 |
av_freep(&s->planar_samples); |
| 1695 |
av_freep(&s->bap_buffer); |
| 1696 |
av_freep(&s->bap1_buffer); |
| 1697 |
av_freep(&s->mdct_coef_buffer); |
| 1698 |
av_freep(&s->exp_buffer); |
| 1699 |
av_freep(&s->grouped_exp_buffer); |
| 1700 |
av_freep(&s->psd_buffer); |
| 1701 |
av_freep(&s->band_psd_buffer); |
| 1702 |
av_freep(&s->mask_buffer); |
| 1703 |
av_freep(&s->qmant_buffer); |
| 1704 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1705 |
AC3Block *block = &s->blocks[blk]; |
| 1706 |
av_freep(&block->bap); |
| 1707 |
av_freep(&block->mdct_coef); |
| 1708 |
av_freep(&block->exp); |
| 1709 |
av_freep(&block->grouped_exp); |
| 1710 |
av_freep(&block->psd); |
| 1711 |
av_freep(&block->band_psd); |
| 1712 |
av_freep(&block->mask); |
| 1713 |
av_freep(&block->qmant); |
| 1714 |
} |
| 1715 |
|
| 1716 |
mdct_end(&s->mdct); |
| 1717 |
|
| 1718 |
av_freep(&avctx->coded_frame); |
| 1719 |
return 0; |
| 1720 |
} |
| 1721 |
|
| 1722 |
|
| 1723 |
/**
|
| 1724 |
* Set channel information during initialization.
|
| 1725 |
*/
|
| 1726 |
static av_cold int set_channel_info(AC3EncodeContext *s, int channels, |
| 1727 |
int64_t *channel_layout) |
| 1728 |
{
|
| 1729 |
int ch_layout;
|
| 1730 |
|
| 1731 |
if (channels < 1 || channels > AC3_MAX_CHANNELS) |
| 1732 |
return AVERROR(EINVAL);
|
| 1733 |
if ((uint64_t)*channel_layout > 0x7FF) |
| 1734 |
return AVERROR(EINVAL);
|
| 1735 |
ch_layout = *channel_layout; |
| 1736 |
if (!ch_layout)
|
| 1737 |
ch_layout = avcodec_guess_channel_layout(channels, CODEC_ID_AC3, NULL);
|
| 1738 |
if (av_get_channel_layout_nb_channels(ch_layout) != channels)
|
| 1739 |
return AVERROR(EINVAL);
|
| 1740 |
|
| 1741 |
s->lfe_on = !!(ch_layout & AV_CH_LOW_FREQUENCY); |
| 1742 |
s->channels = channels; |
| 1743 |
s->fbw_channels = channels - s->lfe_on; |
| 1744 |
s->lfe_channel = s->lfe_on ? s->fbw_channels : -1;
|
| 1745 |
if (s->lfe_on)
|
| 1746 |
ch_layout -= AV_CH_LOW_FREQUENCY; |
| 1747 |
|
| 1748 |
switch (ch_layout) {
|
| 1749 |
case AV_CH_LAYOUT_MONO: s->channel_mode = AC3_CHMODE_MONO; break; |
| 1750 |
case AV_CH_LAYOUT_STEREO: s->channel_mode = AC3_CHMODE_STEREO; break; |
| 1751 |
case AV_CH_LAYOUT_SURROUND: s->channel_mode = AC3_CHMODE_3F; break; |
| 1752 |
case AV_CH_LAYOUT_2_1: s->channel_mode = AC3_CHMODE_2F1R; break; |
| 1753 |
case AV_CH_LAYOUT_4POINT0: s->channel_mode = AC3_CHMODE_3F1R; break; |
| 1754 |
case AV_CH_LAYOUT_QUAD:
|
| 1755 |
case AV_CH_LAYOUT_2_2: s->channel_mode = AC3_CHMODE_2F2R; break; |
| 1756 |
case AV_CH_LAYOUT_5POINT0:
|
| 1757 |
case AV_CH_LAYOUT_5POINT0_BACK: s->channel_mode = AC3_CHMODE_3F2R; break; |
| 1758 |
default:
|
| 1759 |
return AVERROR(EINVAL);
|
| 1760 |
} |
| 1761 |
|
| 1762 |
s->channel_map = ff_ac3_enc_channel_map[s->channel_mode][s->lfe_on]; |
| 1763 |
*channel_layout = ch_layout; |
| 1764 |
if (s->lfe_on)
|
| 1765 |
*channel_layout |= AV_CH_LOW_FREQUENCY; |
| 1766 |
|
| 1767 |
return 0; |
| 1768 |
} |
| 1769 |
|
| 1770 |
|
| 1771 |
static av_cold int validate_options(AVCodecContext *avctx, AC3EncodeContext *s) |
| 1772 |
{
|
| 1773 |
int i, ret;
|
| 1774 |
|
| 1775 |
/* validate channel layout */
|
| 1776 |
if (!avctx->channel_layout) {
|
| 1777 |
av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
|
| 1778 |
"encoder will guess the layout, but it "
|
| 1779 |
"might be incorrect.\n");
|
| 1780 |
} |
| 1781 |
ret = set_channel_info(s, avctx->channels, &avctx->channel_layout); |
| 1782 |
if (ret) {
|
| 1783 |
av_log(avctx, AV_LOG_ERROR, "invalid channel layout\n");
|
| 1784 |
return ret;
|
| 1785 |
} |
| 1786 |
|
| 1787 |
/* validate sample rate */
|
| 1788 |
for (i = 0; i < 9; i++) { |
| 1789 |
if ((ff_ac3_sample_rate_tab[i / 3] >> (i % 3)) == avctx->sample_rate) |
| 1790 |
break;
|
| 1791 |
} |
| 1792 |
if (i == 9) { |
| 1793 |
av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
|
| 1794 |
return AVERROR(EINVAL);
|
| 1795 |
} |
| 1796 |
s->sample_rate = avctx->sample_rate; |
| 1797 |
s->bit_alloc.sr_shift = i % 3;
|
| 1798 |
s->bit_alloc.sr_code = i / 3;
|
| 1799 |
|
| 1800 |
/* validate bit rate */
|
| 1801 |
for (i = 0; i < 19; i++) { |
| 1802 |
if ((ff_ac3_bitrate_tab[i] >> s->bit_alloc.sr_shift)*1000 == avctx->bit_rate) |
| 1803 |
break;
|
| 1804 |
} |
| 1805 |
if (i == 19) { |
| 1806 |
av_log(avctx, AV_LOG_ERROR, "invalid bit rate\n");
|
| 1807 |
return AVERROR(EINVAL);
|
| 1808 |
} |
| 1809 |
s->bit_rate = avctx->bit_rate; |
| 1810 |
s->frame_size_code = i << 1;
|
| 1811 |
|
| 1812 |
/* validate cutoff */
|
| 1813 |
if (avctx->cutoff < 0) { |
| 1814 |
av_log(avctx, AV_LOG_ERROR, "invalid cutoff frequency\n");
|
| 1815 |
return AVERROR(EINVAL);
|
| 1816 |
} |
| 1817 |
s->cutoff = avctx->cutoff; |
| 1818 |
if (s->cutoff > (s->sample_rate >> 1)) |
| 1819 |
s->cutoff = s->sample_rate >> 1;
|
| 1820 |
|
| 1821 |
return 0; |
| 1822 |
} |
| 1823 |
|
| 1824 |
|
| 1825 |
/**
|
| 1826 |
* Set bandwidth for all channels.
|
| 1827 |
* The user can optionally supply a cutoff frequency. Otherwise an appropriate
|
| 1828 |
* default value will be used.
|
| 1829 |
*/
|
| 1830 |
static av_cold void set_bandwidth(AC3EncodeContext *s) |
| 1831 |
{
|
| 1832 |
int ch, bw_code;
|
| 1833 |
|
| 1834 |
if (s->cutoff) {
|
| 1835 |
/* calculate bandwidth based on user-specified cutoff frequency */
|
| 1836 |
int fbw_coeffs;
|
| 1837 |
fbw_coeffs = s->cutoff * 2 * AC3_MAX_COEFS / s->sample_rate;
|
| 1838 |
bw_code = av_clip((fbw_coeffs - 73) / 3, 0, 60); |
| 1839 |
} else {
|
| 1840 |
/* use default bandwidth setting */
|
| 1841 |
/* XXX: should compute the bandwidth according to the frame
|
| 1842 |
size, so that we avoid annoying high frequency artifacts */
|
| 1843 |
bw_code = 50;
|
| 1844 |
} |
| 1845 |
|
| 1846 |
/* set number of coefficients for each channel */
|
| 1847 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1848 |
s->bandwidth_code[ch] = bw_code; |
| 1849 |
s->nb_coefs[ch] = bw_code * 3 + 73; |
| 1850 |
} |
| 1851 |
if (s->lfe_on)
|
| 1852 |
s->nb_coefs[s->lfe_channel] = 7; /* LFE channel always has 7 coefs */ |
| 1853 |
} |
| 1854 |
|
| 1855 |
|
| 1856 |
static av_cold int allocate_buffers(AVCodecContext *avctx) |
| 1857 |
{
|
| 1858 |
int blk, ch;
|
| 1859 |
AC3EncodeContext *s = avctx->priv_data; |
| 1860 |
|
| 1861 |
FF_ALLOC_OR_GOTO(avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
|
| 1862 |
alloc_fail); |
| 1863 |
for (ch = 0; ch < s->channels; ch++) { |
| 1864 |
FF_ALLOCZ_OR_GOTO(avctx, s->planar_samples[ch], |
| 1865 |
(AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
|
| 1866 |
alloc_fail); |
| 1867 |
} |
| 1868 |
FF_ALLOC_OR_GOTO(avctx, s->bap_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1869 |
AC3_MAX_COEFS * sizeof(*s->bap_buffer), alloc_fail);
|
| 1870 |
FF_ALLOC_OR_GOTO(avctx, s->bap1_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1871 |
AC3_MAX_COEFS * sizeof(*s->bap1_buffer), alloc_fail);
|
| 1872 |
FF_ALLOC_OR_GOTO(avctx, s->mdct_coef_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1873 |
AC3_MAX_COEFS * sizeof(*s->mdct_coef_buffer), alloc_fail);
|
| 1874 |
FF_ALLOC_OR_GOTO(avctx, s->exp_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1875 |
AC3_MAX_COEFS * sizeof(*s->exp_buffer), alloc_fail);
|
| 1876 |
FF_ALLOC_OR_GOTO(avctx, s->grouped_exp_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1877 |
128 * sizeof(*s->grouped_exp_buffer), alloc_fail); |
| 1878 |
FF_ALLOC_OR_GOTO(avctx, s->psd_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1879 |
AC3_MAX_COEFS * sizeof(*s->psd_buffer), alloc_fail);
|
| 1880 |
FF_ALLOC_OR_GOTO(avctx, s->band_psd_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1881 |
64 * sizeof(*s->band_psd_buffer), alloc_fail); |
| 1882 |
FF_ALLOC_OR_GOTO(avctx, s->mask_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1883 |
64 * sizeof(*s->mask_buffer), alloc_fail); |
| 1884 |
FF_ALLOC_OR_GOTO(avctx, s->qmant_buffer, AC3_MAX_BLOCKS * s->channels * |
| 1885 |
AC3_MAX_COEFS * sizeof(*s->qmant_buffer), alloc_fail);
|
| 1886 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1887 |
AC3Block *block = &s->blocks[blk]; |
| 1888 |
FF_ALLOC_OR_GOTO(avctx, block->bap, s->channels * sizeof(*block->bap),
|
| 1889 |
alloc_fail); |
| 1890 |
FF_ALLOCZ_OR_GOTO(avctx, block->mdct_coef, s->channels * sizeof(*block->mdct_coef),
|
| 1891 |
alloc_fail); |
| 1892 |
FF_ALLOCZ_OR_GOTO(avctx, block->exp, s->channels * sizeof(*block->exp),
|
| 1893 |
alloc_fail); |
| 1894 |
FF_ALLOCZ_OR_GOTO(avctx, block->grouped_exp, s->channels * sizeof(*block->grouped_exp),
|
| 1895 |
alloc_fail); |
| 1896 |
FF_ALLOCZ_OR_GOTO(avctx, block->psd, s->channels * sizeof(*block->psd),
|
| 1897 |
alloc_fail); |
| 1898 |
FF_ALLOCZ_OR_GOTO(avctx, block->band_psd, s->channels * sizeof(*block->band_psd),
|
| 1899 |
alloc_fail); |
| 1900 |
FF_ALLOCZ_OR_GOTO(avctx, block->mask, s->channels * sizeof(*block->mask),
|
| 1901 |
alloc_fail); |
| 1902 |
FF_ALLOCZ_OR_GOTO(avctx, block->qmant, s->channels * sizeof(*block->qmant),
|
| 1903 |
alloc_fail); |
| 1904 |
|
| 1905 |
for (ch = 0; ch < s->channels; ch++) { |
| 1906 |
block->bap[ch] = &s->bap_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
| 1907 |
block->mdct_coef[ch] = &s->mdct_coef_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
| 1908 |
block->exp[ch] = &s->exp_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
| 1909 |
block->grouped_exp[ch] = &s->grouped_exp_buffer[128 * (blk * s->channels + ch)];
|
| 1910 |
block->psd[ch] = &s->psd_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
| 1911 |
block->band_psd[ch] = &s->band_psd_buffer [64 * (blk * s->channels + ch)];
|
| 1912 |
block->mask[ch] = &s->mask_buffer [64 * (blk * s->channels + ch)];
|
| 1913 |
block->qmant[ch] = &s->qmant_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
| 1914 |
} |
| 1915 |
} |
| 1916 |
|
| 1917 |
return 0; |
| 1918 |
alloc_fail:
|
| 1919 |
return AVERROR(ENOMEM);
|
| 1920 |
} |
| 1921 |
|
| 1922 |
|
| 1923 |
/**
|
| 1924 |
* Initialize the encoder.
|
| 1925 |
*/
|
| 1926 |
static av_cold int ac3_encode_init(AVCodecContext *avctx) |
| 1927 |
{
|
| 1928 |
AC3EncodeContext *s = avctx->priv_data; |
| 1929 |
int ret, frame_size_58;
|
| 1930 |
|
| 1931 |
avctx->frame_size = AC3_FRAME_SIZE; |
| 1932 |
|
| 1933 |
ac3_common_init(); |
| 1934 |
|
| 1935 |
ret = validate_options(avctx, s); |
| 1936 |
if (ret)
|
| 1937 |
return ret;
|
| 1938 |
|
| 1939 |
s->bitstream_id = 8 + s->bit_alloc.sr_shift;
|
| 1940 |
s->bitstream_mode = 0; /* complete main audio service */ |
| 1941 |
|
| 1942 |
s->frame_size_min = 2 * ff_ac3_frame_size_tab[s->frame_size_code][s->bit_alloc.sr_code];
|
| 1943 |
s->bits_written = 0;
|
| 1944 |
s->samples_written = 0;
|
| 1945 |
s->frame_size = s->frame_size_min; |
| 1946 |
|
| 1947 |
/* calculate crc_inv for both possible frame sizes */
|
| 1948 |
frame_size_58 = (( s->frame_size >> 2) + ( s->frame_size >> 4)) << 1; |
| 1949 |
s->crc_inv[0] = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY); |
| 1950 |
if (s->bit_alloc.sr_code == 1) { |
| 1951 |
frame_size_58 = (((s->frame_size+2) >> 2) + ((s->frame_size+2) >> 4)) << 1; |
| 1952 |
s->crc_inv[1] = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY); |
| 1953 |
} |
| 1954 |
|
| 1955 |
set_bandwidth(s); |
| 1956 |
|
| 1957 |
exponent_init(s); |
| 1958 |
|
| 1959 |
bit_alloc_init(s); |
| 1960 |
|
| 1961 |
ret = mdct_init(avctx, &s->mdct, 9);
|
| 1962 |
if (ret)
|
| 1963 |
goto init_fail;
|
| 1964 |
|
| 1965 |
ret = allocate_buffers(avctx); |
| 1966 |
if (ret)
|
| 1967 |
goto init_fail;
|
| 1968 |
|
| 1969 |
avctx->coded_frame= avcodec_alloc_frame(); |
| 1970 |
|
| 1971 |
dsputil_init(&s->dsp, avctx); |
| 1972 |
|
| 1973 |
return 0; |
| 1974 |
init_fail:
|
| 1975 |
ac3_encode_close(avctx); |
| 1976 |
return ret;
|
| 1977 |
} |
| 1978 |
|
| 1979 |
|
| 1980 |
#ifdef TEST
|
| 1981 |
/*************************************************************************/
|
| 1982 |
/* TEST */
|
| 1983 |
|
| 1984 |
#include "libavutil/lfg.h" |
| 1985 |
|
| 1986 |
#define MDCT_NBITS 9 |
| 1987 |
#define MDCT_SAMPLES (1 << MDCT_NBITS) |
| 1988 |
#define FN (MDCT_SAMPLES/4) |
| 1989 |
|
| 1990 |
|
| 1991 |
static void fft_test(AC3MDCTContext *mdct, AVLFG *lfg) |
| 1992 |
{
|
| 1993 |
IComplex in[FN], in1[FN]; |
| 1994 |
int k, n, i;
|
| 1995 |
float sum_re, sum_im, a;
|
| 1996 |
|
| 1997 |
for (i = 0; i < FN; i++) { |
| 1998 |
in[i].re = av_lfg_get(lfg) % 65535 - 32767; |
| 1999 |
in[i].im = av_lfg_get(lfg) % 65535 - 32767; |
| 2000 |
in1[i] = in[i]; |
| 2001 |
} |
| 2002 |
fft(mdct, in, 7);
|
| 2003 |
|
| 2004 |
/* do it by hand */
|
| 2005 |
for (k = 0; k < FN; k++) { |
| 2006 |
sum_re = 0;
|
| 2007 |
sum_im = 0;
|
| 2008 |
for (n = 0; n < FN; n++) { |
| 2009 |
a = -2 * M_PI * (n * k) / FN;
|
| 2010 |
sum_re += in1[n].re * cos(a) - in1[n].im * sin(a); |
| 2011 |
sum_im += in1[n].re * sin(a) + in1[n].im * cos(a); |
| 2012 |
} |
| 2013 |
av_log(NULL, AV_LOG_DEBUG, "%3d: %6d,%6d %6.0f,%6.0f\n", |
| 2014 |
k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); |
| 2015 |
} |
| 2016 |
} |
| 2017 |
|
| 2018 |
|
| 2019 |
static void mdct_test(AC3MDCTContext *mdct, AVLFG *lfg) |
| 2020 |
{
|
| 2021 |
int16_t input[MDCT_SAMPLES]; |
| 2022 |
int32_t output[AC3_MAX_COEFS]; |
| 2023 |
float input1[MDCT_SAMPLES];
|
| 2024 |
float output1[AC3_MAX_COEFS];
|
| 2025 |
float s, a, err, e, emax;
|
| 2026 |
int i, k, n;
|
| 2027 |
|
| 2028 |
for (i = 0; i < MDCT_SAMPLES; i++) { |
| 2029 |
input[i] = (av_lfg_get(lfg) % 65535 - 32767) * 9 / 10; |
| 2030 |
input1[i] = input[i]; |
| 2031 |
} |
| 2032 |
|
| 2033 |
mdct512(mdct, output, input); |
| 2034 |
|
| 2035 |
/* do it by hand */
|
| 2036 |
for (k = 0; k < AC3_MAX_COEFS; k++) { |
| 2037 |
s = 0;
|
| 2038 |
for (n = 0; n < MDCT_SAMPLES; n++) { |
| 2039 |
a = (2*M_PI*(2*n+1+MDCT_SAMPLES/2)*(2*k+1) / (4 * MDCT_SAMPLES)); |
| 2040 |
s += input1[n] * cos(a); |
| 2041 |
} |
| 2042 |
output1[k] = -2 * s / MDCT_SAMPLES;
|
| 2043 |
} |
| 2044 |
|
| 2045 |
err = 0;
|
| 2046 |
emax = 0;
|
| 2047 |
for (i = 0; i < AC3_MAX_COEFS; i++) { |
| 2048 |
av_log(NULL, AV_LOG_DEBUG, "%3d: %7d %7.0f\n", i, output[i], output1[i]); |
| 2049 |
e = output[i] - output1[i]; |
| 2050 |
if (e > emax)
|
| 2051 |
emax = e; |
| 2052 |
err += e * e; |
| 2053 |
} |
| 2054 |
av_log(NULL, AV_LOG_DEBUG, "err2=%f emax=%f\n", err / AC3_MAX_COEFS, emax); |
| 2055 |
} |
| 2056 |
|
| 2057 |
|
| 2058 |
int main(void) |
| 2059 |
{
|
| 2060 |
AVLFG lfg; |
| 2061 |
AC3MDCTContext mdct; |
| 2062 |
|
| 2063 |
mdct.avctx = NULL;
|
| 2064 |
av_log_set_level(AV_LOG_DEBUG); |
| 2065 |
mdct_init(&mdct, 9);
|
| 2066 |
|
| 2067 |
fft_test(&mdct, &lfg); |
| 2068 |
mdct_test(&mdct, &lfg); |
| 2069 |
|
| 2070 |
return 0; |
| 2071 |
} |
| 2072 |
#endif /* TEST */ |
| 2073 |
|
| 2074 |
|
| 2075 |
AVCodec ac3_encoder = {
|
| 2076 |
"ac3",
|
| 2077 |
AVMEDIA_TYPE_AUDIO, |
| 2078 |
CODEC_ID_AC3, |
| 2079 |
sizeof(AC3EncodeContext),
|
| 2080 |
ac3_encode_init, |
| 2081 |
ac3_encode_frame, |
| 2082 |
ac3_encode_close, |
| 2083 |
NULL,
|
| 2084 |
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, |
| 2085 |
.long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
|
| 2086 |
.channel_layouts = ac3_channel_layouts, |
| 2087 |
}; |