ffmpeg / libavcodec / ac3enc.c @ cb6247cb
History | View | Annotate | Download (50.2 KB)
| 1 |
/*
|
|---|---|
| 2 |
* The simplest AC-3 encoder
|
| 3 |
* Copyright (c) 2000 Fabrice Bellard
|
| 4 |
*
|
| 5 |
* This file is part of FFmpeg.
|
| 6 |
*
|
| 7 |
* FFmpeg is free software; you can redistribute it and/or
|
| 8 |
* modify it under the terms of the GNU Lesser General Public
|
| 9 |
* License as published by the Free Software Foundation; either
|
| 10 |
* version 2.1 of the License, or (at your option) any later version.
|
| 11 |
*
|
| 12 |
* FFmpeg is distributed in the hope that it will be useful,
|
| 13 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 14 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 15 |
* Lesser General Public License for more details.
|
| 16 |
*
|
| 17 |
* You should have received a copy of the GNU Lesser General Public
|
| 18 |
* License along with FFmpeg; if not, write to the Free Software
|
| 19 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
| 20 |
*/
|
| 21 |
|
| 22 |
/**
|
| 23 |
* @file
|
| 24 |
* The simplest AC-3 encoder.
|
| 25 |
*/
|
| 26 |
|
| 27 |
//#define DEBUG
|
| 28 |
|
| 29 |
#include "libavcore/audioconvert.h" |
| 30 |
#include "libavutil/crc.h" |
| 31 |
#include "avcodec.h" |
| 32 |
#include "put_bits.h" |
| 33 |
#include "ac3.h" |
| 34 |
#include "audioconvert.h" |
| 35 |
|
| 36 |
|
| 37 |
#define MDCT_NBITS 9 |
| 38 |
#define MDCT_SAMPLES (1 << MDCT_NBITS) |
| 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 |
/** Scale a float value by 2^15, convert to an integer, and clip to int16_t range. */
|
| 47 |
#define FIX15(a) av_clip_int16(SCALE_FLOAT(a, 15)) |
| 48 |
|
| 49 |
|
| 50 |
/**
|
| 51 |
* Compex number.
|
| 52 |
* Used in fixed-point MDCT calculation.
|
| 53 |
*/
|
| 54 |
typedef struct IComplex { |
| 55 |
int16_t re,im; |
| 56 |
} IComplex; |
| 57 |
|
| 58 |
/**
|
| 59 |
* AC-3 encoder private context.
|
| 60 |
*/
|
| 61 |
typedef struct AC3EncodeContext { |
| 62 |
PutBitContext pb; ///< bitstream writer context
|
| 63 |
|
| 64 |
int bitstream_id; ///< bitstream id (bsid) |
| 65 |
int bitstream_mode; ///< bitstream mode (bsmod) |
| 66 |
|
| 67 |
int bit_rate; ///< target bit rate, in bits-per-second |
| 68 |
int sample_rate; ///< sampling frequency, in Hz |
| 69 |
|
| 70 |
int frame_size_min; ///< minimum frame size in case rounding is necessary |
| 71 |
int frame_size; ///< current frame size in bytes |
| 72 |
int frame_size_code; ///< frame size code (frmsizecod) |
| 73 |
int bits_written; ///< bit count (used to avg. bitrate) |
| 74 |
int samples_written; ///< sample count (used to avg. bitrate) |
| 75 |
|
| 76 |
int fbw_channels; ///< number of full-bandwidth channels (nfchans) |
| 77 |
int channels; ///< total number of channels (nchans) |
| 78 |
int lfe_on; ///< indicates if there is an LFE channel (lfeon) |
| 79 |
int lfe_channel; ///< channel index of the LFE channel |
| 80 |
int channel_mode; ///< channel mode (acmod) |
| 81 |
const uint8_t *channel_map; ///< channel map used to reorder channels |
| 82 |
|
| 83 |
int bandwidth_code[AC3_MAX_CHANNELS]; ///< bandwidth code (0 to 60) (chbwcod) |
| 84 |
int nb_coefs[AC3_MAX_CHANNELS];
|
| 85 |
|
| 86 |
/* bitrate allocation control */
|
| 87 |
int slow_gain_code; ///< slow gain code (sgaincod) |
| 88 |
int slow_decay_code; ///< slow decay code (sdcycod) |
| 89 |
int fast_decay_code; ///< fast decay code (fdcycod) |
| 90 |
int db_per_bit_code; ///< dB/bit code (dbpbcod) |
| 91 |
int floor_code; ///< floor code (floorcod) |
| 92 |
AC3BitAllocParameters bit_alloc; ///< bit allocation parameters
|
| 93 |
int coarse_snr_offset; ///< coarse SNR offsets (csnroffst) |
| 94 |
int fast_gain_code[AC3_MAX_CHANNELS]; ///< fast gain codes (signal-to-mask ratio) (fgaincod) |
| 95 |
int fine_snr_offset[AC3_MAX_CHANNELS]; ///< fine SNR offsets (fsnroffst) |
| 96 |
int frame_bits; ///< all frame bits except exponents and mantissas |
| 97 |
int exponent_bits; ///< number of bits used for exponents |
| 98 |
|
| 99 |
/* mantissa encoding */
|
| 100 |
int mant1_cnt, mant2_cnt, mant4_cnt; ///< mantissa counts for bap=1,2,4 |
| 101 |
uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; ///< mantissa pointers for bap=1,2,4
|
| 102 |
|
| 103 |
int16_t last_samples[AC3_MAX_CHANNELS][AC3_BLOCK_SIZE]; ///< last 256 samples from previous frame
|
| 104 |
int16_t planar_samples[AC3_MAX_CHANNELS][AC3_BLOCK_SIZE+AC3_FRAME_SIZE]; |
| 105 |
int16_t windowed_samples[AC3_WINDOW_SIZE]; |
| 106 |
int32_t mdct_coef[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 107 |
uint8_t exp[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 108 |
uint8_t exp_strategy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS]; |
| 109 |
uint8_t encoded_exp[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 110 |
uint8_t num_exp_groups[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS]; |
| 111 |
uint8_t grouped_exp[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_EXP_GROUPS]; |
| 112 |
int16_t psd[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 113 |
int16_t band_psd[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_CRITICAL_BANDS]; |
| 114 |
int16_t mask[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_CRITICAL_BANDS]; |
| 115 |
uint8_t bap[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 116 |
uint8_t bap1[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 117 |
int8_t exp_shift[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS]; |
| 118 |
uint16_t qmant[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS]; |
| 119 |
} AC3EncodeContext; |
| 120 |
|
| 121 |
|
| 122 |
/** MDCT and FFT tables */
|
| 123 |
static int16_t costab[64]; |
| 124 |
static int16_t sintab[64]; |
| 125 |
static int16_t xcos1[128]; |
| 126 |
static int16_t xsin1[128]; |
| 127 |
|
| 128 |
|
| 129 |
/**
|
| 130 |
* Adjust the frame size to make the average bit rate match the target bit rate.
|
| 131 |
* This is only needed for 11025, 22050, and 44100 sample rates.
|
| 132 |
*/
|
| 133 |
static void adjust_frame_size(AC3EncodeContext *s) |
| 134 |
{
|
| 135 |
while (s->bits_written >= s->bit_rate && s->samples_written >= s->sample_rate) {
|
| 136 |
s->bits_written -= s->bit_rate; |
| 137 |
s->samples_written -= s->sample_rate; |
| 138 |
} |
| 139 |
s->frame_size = s->frame_size_min + 2 * (s->bits_written * s->sample_rate < s->samples_written * s->bit_rate);
|
| 140 |
s->bits_written += s->frame_size * 8;
|
| 141 |
s->samples_written += AC3_FRAME_SIZE; |
| 142 |
} |
| 143 |
|
| 144 |
|
| 145 |
/**
|
| 146 |
* Deinterleave input samples.
|
| 147 |
* Channels are reordered from FFmpeg's default order to AC-3 order.
|
| 148 |
*/
|
| 149 |
static void deinterleave_input_samples(AC3EncodeContext *s, |
| 150 |
const int16_t *samples)
|
| 151 |
{
|
| 152 |
int ch, i;
|
| 153 |
|
| 154 |
/* deinterleave and remap input samples */
|
| 155 |
for (ch = 0; ch < s->channels; ch++) { |
| 156 |
const int16_t *sptr;
|
| 157 |
int sinc;
|
| 158 |
|
| 159 |
/* copy last 256 samples of previous frame to the start of the current frame */
|
| 160 |
memcpy(&s->planar_samples[ch][0], s->last_samples[ch],
|
| 161 |
AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0])); |
| 162 |
|
| 163 |
/* deinterleave */
|
| 164 |
sinc = s->channels; |
| 165 |
sptr = samples + s->channel_map[ch]; |
| 166 |
for (i = AC3_BLOCK_SIZE; i < AC3_FRAME_SIZE+AC3_BLOCK_SIZE; i++) {
|
| 167 |
s->planar_samples[ch][i] = *sptr; |
| 168 |
sptr += sinc; |
| 169 |
} |
| 170 |
|
| 171 |
/* save last 256 samples for next frame */
|
| 172 |
memcpy(s->last_samples[ch], &s->planar_samples[ch][6* AC3_BLOCK_SIZE],
|
| 173 |
AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0])); |
| 174 |
} |
| 175 |
} |
| 176 |
|
| 177 |
|
| 178 |
/**
|
| 179 |
* Initialize FFT tables.
|
| 180 |
* @param ln log2(FFT size)
|
| 181 |
*/
|
| 182 |
static av_cold void fft_init(int ln) |
| 183 |
{
|
| 184 |
int i, n, n2;
|
| 185 |
float alpha;
|
| 186 |
|
| 187 |
n = 1 << ln;
|
| 188 |
n2 = n >> 1;
|
| 189 |
|
| 190 |
for (i = 0; i < n2; i++) { |
| 191 |
alpha = 2.0 * M_PI * i / n; |
| 192 |
costab[i] = FIX15(cos(alpha)); |
| 193 |
sintab[i] = FIX15(sin(alpha)); |
| 194 |
} |
| 195 |
} |
| 196 |
|
| 197 |
|
| 198 |
/**
|
| 199 |
* Initialize MDCT tables.
|
| 200 |
* @param nbits log2(MDCT size)
|
| 201 |
*/
|
| 202 |
static av_cold void mdct_init(int nbits) |
| 203 |
{
|
| 204 |
int i, n, n4;
|
| 205 |
|
| 206 |
n = 1 << nbits;
|
| 207 |
n4 = n >> 2;
|
| 208 |
|
| 209 |
fft_init(nbits - 2);
|
| 210 |
|
| 211 |
for (i = 0; i < n4; i++) { |
| 212 |
float alpha = 2.0 * M_PI * (i + 1.0 / 8.0) / n; |
| 213 |
xcos1[i] = FIX15(-cos(alpha)); |
| 214 |
xsin1[i] = FIX15(-sin(alpha)); |
| 215 |
} |
| 216 |
} |
| 217 |
|
| 218 |
|
| 219 |
/** Butterfly op */
|
| 220 |
#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
|
| 221 |
{ \
|
| 222 |
int ax, ay, bx, by; \
|
| 223 |
bx = pre1; \ |
| 224 |
by = pim1; \ |
| 225 |
ax = qre1; \ |
| 226 |
ay = qim1; \ |
| 227 |
pre = (bx + ax) >> 1; \
|
| 228 |
pim = (by + ay) >> 1; \
|
| 229 |
qre = (bx - ax) >> 1; \
|
| 230 |
qim = (by - ay) >> 1; \
|
| 231 |
} |
| 232 |
|
| 233 |
|
| 234 |
/** Complex multiply */
|
| 235 |
#define CMUL(pre, pim, are, aim, bre, bim) \
|
| 236 |
{ \
|
| 237 |
pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15; \
|
| 238 |
pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15; \
|
| 239 |
} |
| 240 |
|
| 241 |
|
| 242 |
/**
|
| 243 |
* Calculate a 2^n point complex FFT on 2^ln points.
|
| 244 |
* @param z complex input/output samples
|
| 245 |
* @param ln log2(FFT size)
|
| 246 |
*/
|
| 247 |
static void fft(IComplex *z, int ln) |
| 248 |
{
|
| 249 |
int j, l, np, np2;
|
| 250 |
int nblocks, nloops;
|
| 251 |
register IComplex *p,*q;
|
| 252 |
int tmp_re, tmp_im;
|
| 253 |
|
| 254 |
np = 1 << ln;
|
| 255 |
|
| 256 |
/* reverse */
|
| 257 |
for (j = 0; j < np; j++) { |
| 258 |
int k = av_reverse[j] >> (8 - ln); |
| 259 |
if (k < j)
|
| 260 |
FFSWAP(IComplex, z[k], z[j]); |
| 261 |
} |
| 262 |
|
| 263 |
/* pass 0 */
|
| 264 |
|
| 265 |
p = &z[0];
|
| 266 |
j = np >> 1;
|
| 267 |
do {
|
| 268 |
BF(p[0].re, p[0].im, p[1].re, p[1].im, |
| 269 |
p[0].re, p[0].im, p[1].re, p[1].im); |
| 270 |
p += 2;
|
| 271 |
} while (--j);
|
| 272 |
|
| 273 |
/* pass 1 */
|
| 274 |
|
| 275 |
p = &z[0];
|
| 276 |
j = np >> 2;
|
| 277 |
do {
|
| 278 |
BF(p[0].re, p[0].im, p[2].re, p[2].im, |
| 279 |
p[0].re, p[0].im, p[2].re, p[2].im); |
| 280 |
BF(p[1].re, p[1].im, p[3].re, p[3].im, |
| 281 |
p[1].re, p[1].im, p[3].im, -p[3].re); |
| 282 |
p+=4;
|
| 283 |
} while (--j);
|
| 284 |
|
| 285 |
/* pass 2 .. ln-1 */
|
| 286 |
|
| 287 |
nblocks = np >> 3;
|
| 288 |
nloops = 1 << 2; |
| 289 |
np2 = np >> 1;
|
| 290 |
do {
|
| 291 |
p = z; |
| 292 |
q = z + nloops; |
| 293 |
for (j = 0; j < nblocks; j++) { |
| 294 |
BF(p->re, p->im, q->re, q->im, |
| 295 |
p->re, p->im, q->re, q->im); |
| 296 |
p++; |
| 297 |
q++; |
| 298 |
for(l = nblocks; l < np2; l += nblocks) {
|
| 299 |
CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im); |
| 300 |
BF(p->re, p->im, q->re, q->im, |
| 301 |
p->re, p->im, tmp_re, tmp_im); |
| 302 |
p++; |
| 303 |
q++; |
| 304 |
} |
| 305 |
p += nloops; |
| 306 |
q += nloops; |
| 307 |
} |
| 308 |
nblocks = nblocks >> 1;
|
| 309 |
nloops = nloops << 1;
|
| 310 |
} while (nblocks);
|
| 311 |
} |
| 312 |
|
| 313 |
|
| 314 |
/**
|
| 315 |
* Calculate a 512-point MDCT
|
| 316 |
* @param out 256 output frequency coefficients
|
| 317 |
* @param in 512 windowed input audio samples
|
| 318 |
*/
|
| 319 |
static void mdct512(int32_t *out, int16_t *in) |
| 320 |
{
|
| 321 |
int i, re, im, re1, im1;
|
| 322 |
int16_t rot[MDCT_SAMPLES]; |
| 323 |
IComplex x[MDCT_SAMPLES/4];
|
| 324 |
|
| 325 |
/* shift to simplify computations */
|
| 326 |
for (i = 0; i < MDCT_SAMPLES/4; i++) |
| 327 |
rot[i] = -in[i + 3*MDCT_SAMPLES/4]; |
| 328 |
for (;i < MDCT_SAMPLES; i++)
|
| 329 |
rot[i] = in[i - MDCT_SAMPLES/4];
|
| 330 |
|
| 331 |
/* pre rotation */
|
| 332 |
for (i = 0; i < MDCT_SAMPLES/4; i++) { |
| 333 |
re = ((int)rot[ 2*i] - (int)rot[MDCT_SAMPLES -1-2*i]) >> 1; |
| 334 |
im = -((int)rot[MDCT_SAMPLES/2+2*i] - (int)rot[MDCT_SAMPLES/2-1-2*i]) >> 1; |
| 335 |
CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]); |
| 336 |
} |
| 337 |
|
| 338 |
fft(x, MDCT_NBITS - 2);
|
| 339 |
|
| 340 |
/* post rotation */
|
| 341 |
for (i = 0; i < MDCT_SAMPLES/4; i++) { |
| 342 |
re = x[i].re; |
| 343 |
im = x[i].im; |
| 344 |
CMUL(re1, im1, re, im, xsin1[i], xcos1[i]); |
| 345 |
out[ 2*i] = im1;
|
| 346 |
out[MDCT_SAMPLES/2-1-2*i] = re1; |
| 347 |
} |
| 348 |
} |
| 349 |
|
| 350 |
|
| 351 |
/**
|
| 352 |
* Apply KBD window to input samples prior to MDCT.
|
| 353 |
*/
|
| 354 |
static void apply_window(int16_t *output, const int16_t *input, |
| 355 |
const int16_t *window, int n) |
| 356 |
{
|
| 357 |
int i;
|
| 358 |
int n2 = n >> 1; |
| 359 |
|
| 360 |
for (i = 0; i < n2; i++) { |
| 361 |
output[i] = MUL16(input[i], window[i]) >> 15;
|
| 362 |
output[n-i-1] = MUL16(input[n-i-1], window[i]) >> 15; |
| 363 |
} |
| 364 |
} |
| 365 |
|
| 366 |
|
| 367 |
/**
|
| 368 |
* Calculate the log2() of the maximum absolute value in an array.
|
| 369 |
* @param tab input array
|
| 370 |
* @param n number of values in the array
|
| 371 |
* @return log2(max(abs(tab[])))
|
| 372 |
*/
|
| 373 |
static int log2_tab(int16_t *tab, int n) |
| 374 |
{
|
| 375 |
int i, v;
|
| 376 |
|
| 377 |
v = 0;
|
| 378 |
for (i = 0; i < n; i++) |
| 379 |
v |= abs(tab[i]); |
| 380 |
|
| 381 |
return av_log2(v);
|
| 382 |
} |
| 383 |
|
| 384 |
|
| 385 |
/**
|
| 386 |
* Left-shift each value in an array by a specified amount.
|
| 387 |
* @param tab input array
|
| 388 |
* @param n number of values in the array
|
| 389 |
* @param lshift left shift amount. a negative value means right shift.
|
| 390 |
*/
|
| 391 |
static void lshift_tab(int16_t *tab, int n, int lshift) |
| 392 |
{
|
| 393 |
int i;
|
| 394 |
|
| 395 |
if (lshift > 0) { |
| 396 |
for(i = 0; i < n; i++) |
| 397 |
tab[i] <<= lshift; |
| 398 |
} else if (lshift < 0) { |
| 399 |
lshift = -lshift; |
| 400 |
for (i = 0; i < n; i++) |
| 401 |
tab[i] >>= lshift; |
| 402 |
} |
| 403 |
} |
| 404 |
|
| 405 |
|
| 406 |
/**
|
| 407 |
* Normalize the input samples to use the maximum available precision.
|
| 408 |
* This assumes signed 16-bit input samples. Exponents are reduced by 9 to
|
| 409 |
* match the 24-bit internal precision for MDCT coefficients.
|
| 410 |
*
|
| 411 |
* @return exponent shift
|
| 412 |
*/
|
| 413 |
static int normalize_samples(AC3EncodeContext *s) |
| 414 |
{
|
| 415 |
int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE); |
| 416 |
v = FFMAX(0, v);
|
| 417 |
lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v); |
| 418 |
return v - 9; |
| 419 |
} |
| 420 |
|
| 421 |
|
| 422 |
/**
|
| 423 |
* Apply the MDCT to input samples to generate frequency coefficients.
|
| 424 |
* This applies the KBD window and normalizes the input to reduce precision
|
| 425 |
* loss due to fixed-point calculations.
|
| 426 |
*/
|
| 427 |
static void apply_mdct(AC3EncodeContext *s) |
| 428 |
{
|
| 429 |
int blk, ch;
|
| 430 |
|
| 431 |
for (ch = 0; ch < s->channels; ch++) { |
| 432 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 433 |
const int16_t *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
|
| 434 |
|
| 435 |
apply_window(s->windowed_samples, input_samples, ff_ac3_window, AC3_WINDOW_SIZE); |
| 436 |
|
| 437 |
s->exp_shift[blk][ch] = normalize_samples(s); |
| 438 |
|
| 439 |
mdct512(s->mdct_coef[blk][ch], s->windowed_samples); |
| 440 |
} |
| 441 |
} |
| 442 |
} |
| 443 |
|
| 444 |
|
| 445 |
/**
|
| 446 |
* Extract exponents from the MDCT coefficients.
|
| 447 |
* This takes into account the normalization that was done to the input samples
|
| 448 |
* by adjusting the exponents by the exponent shift values.
|
| 449 |
*/
|
| 450 |
static void extract_exponents(AC3EncodeContext *s) |
| 451 |
{
|
| 452 |
int blk, ch, i;
|
| 453 |
|
| 454 |
/* extract exponents */
|
| 455 |
for (ch = 0; ch < s->channels; ch++) { |
| 456 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 457 |
/* compute "exponents". We take into account the normalization there */
|
| 458 |
for (i = 0; i < AC3_MAX_COEFS; i++) { |
| 459 |
int e;
|
| 460 |
int v = abs(s->mdct_coef[blk][ch][i]);
|
| 461 |
if (v == 0) |
| 462 |
e = 24;
|
| 463 |
else {
|
| 464 |
e = 23 - av_log2(v) + s->exp_shift[blk][ch];
|
| 465 |
if (e >= 24) { |
| 466 |
e = 24;
|
| 467 |
s->mdct_coef[blk][ch][i] = 0;
|
| 468 |
} |
| 469 |
} |
| 470 |
s->exp[blk][ch][i] = e; |
| 471 |
} |
| 472 |
} |
| 473 |
} |
| 474 |
} |
| 475 |
|
| 476 |
|
| 477 |
/**
|
| 478 |
* Calculate the sum of absolute differences (SAD) between 2 sets of exponents.
|
| 479 |
*/
|
| 480 |
static int calc_exp_diff(uint8_t *exp1, uint8_t *exp2, int n) |
| 481 |
{
|
| 482 |
int sum, i;
|
| 483 |
sum = 0;
|
| 484 |
for (i = 0; i < n; i++) |
| 485 |
sum += abs(exp1[i] - exp2[i]); |
| 486 |
return sum;
|
| 487 |
} |
| 488 |
|
| 489 |
|
| 490 |
/**
|
| 491 |
* Exponent Difference Threshold.
|
| 492 |
* New exponents are sent if their SAD exceed this number.
|
| 493 |
*/
|
| 494 |
#define EXP_DIFF_THRESHOLD 1000 |
| 495 |
|
| 496 |
|
| 497 |
/**
|
| 498 |
* Calculate exponent strategies for all blocks in a single channel.
|
| 499 |
*/
|
| 500 |
static void compute_exp_strategy_ch(uint8_t *exp_strategy, uint8_t **exp) |
| 501 |
{
|
| 502 |
int blk, blk1;
|
| 503 |
int exp_diff;
|
| 504 |
|
| 505 |
/* estimate if the exponent variation & decide if they should be
|
| 506 |
reused in the next frame */
|
| 507 |
exp_strategy[0] = EXP_NEW;
|
| 508 |
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) { |
| 509 |
exp_diff = calc_exp_diff(exp[blk], exp[blk-1], AC3_MAX_COEFS);
|
| 510 |
if (exp_diff > EXP_DIFF_THRESHOLD)
|
| 511 |
exp_strategy[blk] = EXP_NEW; |
| 512 |
else
|
| 513 |
exp_strategy[blk] = EXP_REUSE; |
| 514 |
} |
| 515 |
|
| 516 |
/* now select the encoding strategy type : if exponents are often
|
| 517 |
recoded, we use a coarse encoding */
|
| 518 |
blk = 0;
|
| 519 |
while (blk < AC3_MAX_BLOCKS) {
|
| 520 |
blk1 = blk + 1;
|
| 521 |
while (blk1 < AC3_MAX_BLOCKS && exp_strategy[blk1] == EXP_REUSE)
|
| 522 |
blk1++; |
| 523 |
switch (blk1 - blk) {
|
| 524 |
case 1: exp_strategy[blk] = EXP_D45; break; |
| 525 |
case 2: |
| 526 |
case 3: exp_strategy[blk] = EXP_D25; break; |
| 527 |
default: exp_strategy[blk] = EXP_D15; break; |
| 528 |
} |
| 529 |
blk = blk1; |
| 530 |
} |
| 531 |
} |
| 532 |
|
| 533 |
|
| 534 |
/**
|
| 535 |
* Calculate exponent strategies for all channels.
|
| 536 |
* Array arrangement is reversed to simplify the per-channel calculation.
|
| 537 |
*/
|
| 538 |
static void compute_exp_strategy(AC3EncodeContext *s) |
| 539 |
{
|
| 540 |
uint8_t *exp1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; |
| 541 |
uint8_t exp_str1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; |
| 542 |
int ch, blk;
|
| 543 |
|
| 544 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 545 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 546 |
exp1[ch][blk] = s->exp[blk][ch]; |
| 547 |
exp_str1[ch][blk] = s->exp_strategy[blk][ch]; |
| 548 |
} |
| 549 |
|
| 550 |
compute_exp_strategy_ch(exp_str1[ch], exp1[ch]); |
| 551 |
|
| 552 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) |
| 553 |
s->exp_strategy[blk][ch] = exp_str1[ch][blk]; |
| 554 |
} |
| 555 |
if (s->lfe_on) {
|
| 556 |
ch = s->lfe_channel; |
| 557 |
s->exp_strategy[0][ch] = EXP_D15;
|
| 558 |
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) |
| 559 |
s->exp_strategy[blk][ch] = EXP_REUSE; |
| 560 |
} |
| 561 |
} |
| 562 |
|
| 563 |
|
| 564 |
/**
|
| 565 |
* Set each encoded exponent in a block to the minimum of itself and the
|
| 566 |
* exponent in the same frequency bin of a following block.
|
| 567 |
* exp[i] = min(exp[i], exp1[i]
|
| 568 |
*/
|
| 569 |
static void exponent_min(uint8_t exp[AC3_MAX_COEFS], uint8_t exp1[AC3_MAX_COEFS], int n) |
| 570 |
{
|
| 571 |
int i;
|
| 572 |
for (i = 0; i < n; i++) { |
| 573 |
if (exp1[i] < exp[i])
|
| 574 |
exp[i] = exp1[i]; |
| 575 |
} |
| 576 |
} |
| 577 |
|
| 578 |
|
| 579 |
/**
|
| 580 |
* Update the exponents so that they are the ones the decoder will decode.
|
| 581 |
*/
|
| 582 |
static void encode_exponents_blk_ch(uint8_t encoded_exp[AC3_MAX_COEFS], |
| 583 |
uint8_t exp[AC3_MAX_COEFS], |
| 584 |
int nb_exps, int exp_strategy, |
| 585 |
uint8_t *num_exp_groups) |
| 586 |
{
|
| 587 |
int group_size, nb_groups, i, j, k, exp_min;
|
| 588 |
uint8_t exp1[AC3_MAX_COEFS]; |
| 589 |
|
| 590 |
group_size = exp_strategy + (exp_strategy == EXP_D45); |
| 591 |
*num_exp_groups = (nb_exps + (group_size * 3) - 4) / (3 * group_size); |
| 592 |
nb_groups = *num_exp_groups * 3;
|
| 593 |
|
| 594 |
/* for each group, compute the minimum exponent */
|
| 595 |
exp1[0] = exp[0]; /* DC exponent is handled separately */ |
| 596 |
k = 1;
|
| 597 |
for (i = 1; i <= nb_groups; i++) { |
| 598 |
exp_min = exp[k]; |
| 599 |
assert(exp_min >= 0 && exp_min <= 24); |
| 600 |
for (j = 1; j < group_size; j++) { |
| 601 |
if (exp[k+j] < exp_min)
|
| 602 |
exp_min = exp[k+j]; |
| 603 |
} |
| 604 |
exp1[i] = exp_min; |
| 605 |
k += group_size; |
| 606 |
} |
| 607 |
|
| 608 |
/* constraint for DC exponent */
|
| 609 |
if (exp1[0] > 15) |
| 610 |
exp1[0] = 15; |
| 611 |
|
| 612 |
/* decrease the delta between each groups to within 2 so that they can be
|
| 613 |
differentially encoded */
|
| 614 |
for (i = 1; i <= nb_groups; i++) |
| 615 |
exp1[i] = FFMIN(exp1[i], exp1[i-1] + 2); |
| 616 |
for (i = nb_groups-1; i >= 0; i--) |
| 617 |
exp1[i] = FFMIN(exp1[i], exp1[i+1] + 2); |
| 618 |
|
| 619 |
/* now we have the exponent values the decoder will see */
|
| 620 |
encoded_exp[0] = exp1[0]; |
| 621 |
k = 1;
|
| 622 |
for (i = 1; i <= nb_groups; i++) { |
| 623 |
for (j = 0; j < group_size; j++) |
| 624 |
encoded_exp[k+j] = exp1[i]; |
| 625 |
k += group_size; |
| 626 |
} |
| 627 |
} |
| 628 |
|
| 629 |
|
| 630 |
/**
|
| 631 |
* Encode exponents from original extracted form to what the decoder will see.
|
| 632 |
* This copies and groups exponents based on exponent strategy and reduces
|
| 633 |
* deltas between adjacent exponent groups so that they can be differentially
|
| 634 |
* encoded.
|
| 635 |
*/
|
| 636 |
static void encode_exponents(AC3EncodeContext *s) |
| 637 |
{
|
| 638 |
int blk, blk1, blk2, ch;
|
| 639 |
|
| 640 |
for (ch = 0; ch < s->channels; ch++) { |
| 641 |
/* for the EXP_REUSE case we select the min of the exponents */
|
| 642 |
blk = 0;
|
| 643 |
while (blk < AC3_MAX_BLOCKS) {
|
| 644 |
blk1 = blk + 1;
|
| 645 |
while (blk1 < AC3_MAX_BLOCKS && s->exp_strategy[blk1][ch] == EXP_REUSE) {
|
| 646 |
exponent_min(s->exp[blk][ch], s->exp[blk1][ch], s->nb_coefs[ch]); |
| 647 |
blk1++; |
| 648 |
} |
| 649 |
encode_exponents_blk_ch(s->encoded_exp[blk][ch], |
| 650 |
s->exp[blk][ch], s->nb_coefs[ch], |
| 651 |
s->exp_strategy[blk][ch], |
| 652 |
&s->num_exp_groups[blk][ch]); |
| 653 |
/* copy encoded exponents for reuse case */
|
| 654 |
for (blk2 = blk+1; blk2 < blk1; blk2++) { |
| 655 |
memcpy(s->encoded_exp[blk2][ch], s->encoded_exp[blk][ch], |
| 656 |
s->nb_coefs[ch] * sizeof(uint8_t));
|
| 657 |
} |
| 658 |
blk = blk1; |
| 659 |
} |
| 660 |
} |
| 661 |
} |
| 662 |
|
| 663 |
|
| 664 |
/**
|
| 665 |
* Group exponents.
|
| 666 |
* 3 delta-encoded exponents are in each 7-bit group. The number of groups
|
| 667 |
* varies depending on exponent strategy and bandwidth.
|
| 668 |
*/
|
| 669 |
static void group_exponents(AC3EncodeContext *s) |
| 670 |
{
|
| 671 |
int blk, ch, i;
|
| 672 |
int group_size, bit_count;
|
| 673 |
uint8_t *p; |
| 674 |
int delta0, delta1, delta2;
|
| 675 |
int exp0, exp1;
|
| 676 |
|
| 677 |
bit_count = 0;
|
| 678 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 679 |
for (ch = 0; ch < s->channels; ch++) { |
| 680 |
if (s->exp_strategy[blk][ch] == EXP_REUSE) {
|
| 681 |
s->num_exp_groups[blk][ch] = 0;
|
| 682 |
continue;
|
| 683 |
} |
| 684 |
group_size = s->exp_strategy[blk][ch] + (s->exp_strategy[blk][ch] == EXP_D45); |
| 685 |
bit_count += 4 + (s->num_exp_groups[blk][ch] * 7); |
| 686 |
p = s->encoded_exp[blk][ch]; |
| 687 |
|
| 688 |
/* DC exponent */
|
| 689 |
exp1 = *p++; |
| 690 |
s->grouped_exp[blk][ch][0] = exp1;
|
| 691 |
|
| 692 |
/* remaining exponents are delta encoded */
|
| 693 |
for (i = 1; i <= s->num_exp_groups[blk][ch]; i++) { |
| 694 |
/* merge three delta in one code */
|
| 695 |
exp0 = exp1; |
| 696 |
exp1 = p[0];
|
| 697 |
p += group_size; |
| 698 |
delta0 = exp1 - exp0 + 2;
|
| 699 |
|
| 700 |
exp0 = exp1; |
| 701 |
exp1 = p[0];
|
| 702 |
p += group_size; |
| 703 |
delta1 = exp1 - exp0 + 2;
|
| 704 |
|
| 705 |
exp0 = exp1; |
| 706 |
exp1 = p[0];
|
| 707 |
p += group_size; |
| 708 |
delta2 = exp1 - exp0 + 2;
|
| 709 |
|
| 710 |
s->grouped_exp[blk][ch][i] = ((delta0 * 5 + delta1) * 5) + delta2; |
| 711 |
} |
| 712 |
} |
| 713 |
} |
| 714 |
|
| 715 |
s->exponent_bits = bit_count; |
| 716 |
} |
| 717 |
|
| 718 |
|
| 719 |
/**
|
| 720 |
* Calculate final exponents from the supplied MDCT coefficients and exponent shift.
|
| 721 |
* Extract exponents from MDCT coefficients, calculate exponent strategies,
|
| 722 |
* and encode final exponents.
|
| 723 |
*/
|
| 724 |
static void process_exponents(AC3EncodeContext *s) |
| 725 |
{
|
| 726 |
extract_exponents(s); |
| 727 |
|
| 728 |
compute_exp_strategy(s); |
| 729 |
|
| 730 |
encode_exponents(s); |
| 731 |
|
| 732 |
group_exponents(s); |
| 733 |
} |
| 734 |
|
| 735 |
|
| 736 |
/**
|
| 737 |
* Initialize bit allocation.
|
| 738 |
* Set default parameter codes and calculate parameter values.
|
| 739 |
*/
|
| 740 |
static void bit_alloc_init(AC3EncodeContext *s) |
| 741 |
{
|
| 742 |
int ch;
|
| 743 |
|
| 744 |
/* init default parameters */
|
| 745 |
s->slow_decay_code = 2;
|
| 746 |
s->fast_decay_code = 1;
|
| 747 |
s->slow_gain_code = 1;
|
| 748 |
s->db_per_bit_code = 2;
|
| 749 |
s->floor_code = 4;
|
| 750 |
for (ch = 0; ch < s->channels; ch++) |
| 751 |
s->fast_gain_code[ch] = 4;
|
| 752 |
|
| 753 |
/* initial snr offset */
|
| 754 |
s->coarse_snr_offset = 40;
|
| 755 |
|
| 756 |
/* compute real values */
|
| 757 |
/* currently none of these values change during encoding, so we can just
|
| 758 |
set them once at initialization */
|
| 759 |
s->bit_alloc.slow_decay = ff_ac3_slow_decay_tab[s->slow_decay_code] >> s->bit_alloc.sr_shift; |
| 760 |
s->bit_alloc.fast_decay = ff_ac3_fast_decay_tab[s->fast_decay_code] >> s->bit_alloc.sr_shift; |
| 761 |
s->bit_alloc.slow_gain = ff_ac3_slow_gain_tab[s->slow_gain_code]; |
| 762 |
s->bit_alloc.db_per_bit = ff_ac3_db_per_bit_tab[s->db_per_bit_code]; |
| 763 |
s->bit_alloc.floor = ff_ac3_floor_tab[s->floor_code]; |
| 764 |
} |
| 765 |
|
| 766 |
|
| 767 |
/**
|
| 768 |
* Count the bits used to encode the frame, minus exponents and mantissas.
|
| 769 |
*/
|
| 770 |
static void count_frame_bits(AC3EncodeContext *s) |
| 771 |
{
|
| 772 |
static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 }; |
| 773 |
int blk, ch;
|
| 774 |
int frame_bits;
|
| 775 |
|
| 776 |
/* header size */
|
| 777 |
frame_bits = 65;
|
| 778 |
frame_bits += frame_bits_inc[s->channel_mode]; |
| 779 |
|
| 780 |
/* audio blocks */
|
| 781 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 782 |
frame_bits += s->fbw_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */ |
| 783 |
if (s->channel_mode == AC3_CHMODE_STEREO) {
|
| 784 |
frame_bits++; /* rematstr */
|
| 785 |
if (!blk)
|
| 786 |
frame_bits += 4;
|
| 787 |
} |
| 788 |
frame_bits += 2 * s->fbw_channels; /* chexpstr[2] * c */ |
| 789 |
if (s->lfe_on)
|
| 790 |
frame_bits++; /* lfeexpstr */
|
| 791 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 792 |
if (s->exp_strategy[blk][ch] != EXP_REUSE)
|
| 793 |
frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */ |
| 794 |
} |
| 795 |
frame_bits++; /* baie */
|
| 796 |
frame_bits++; /* snr */
|
| 797 |
frame_bits += 2; /* delta / skip */ |
| 798 |
} |
| 799 |
frame_bits++; /* cplinu for block 0 */
|
| 800 |
/* bit alloc info */
|
| 801 |
/* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
|
| 802 |
/* csnroffset[6] */
|
| 803 |
/* (fsnoffset[4] + fgaincod[4]) * c */
|
| 804 |
frame_bits += 2*4 + 3 + 6 + s->channels * (4 + 3); |
| 805 |
|
| 806 |
/* auxdatae, crcrsv */
|
| 807 |
frame_bits += 2;
|
| 808 |
|
| 809 |
/* CRC */
|
| 810 |
frame_bits += 16;
|
| 811 |
|
| 812 |
s->frame_bits = frame_bits; |
| 813 |
} |
| 814 |
|
| 815 |
|
| 816 |
/**
|
| 817 |
* Calculate the number of bits needed to encode a set of mantissas.
|
| 818 |
*/
|
| 819 |
static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs) |
| 820 |
{
|
| 821 |
int bits, mant, i;
|
| 822 |
|
| 823 |
bits = 0;
|
| 824 |
for (i = 0; i < nb_coefs; i++) { |
| 825 |
mant = m[i]; |
| 826 |
switch (mant) {
|
| 827 |
case 0: |
| 828 |
/* nothing */
|
| 829 |
break;
|
| 830 |
case 1: |
| 831 |
/* 3 mantissa in 5 bits */
|
| 832 |
if (s->mant1_cnt == 0) |
| 833 |
bits += 5;
|
| 834 |
if (++s->mant1_cnt == 3) |
| 835 |
s->mant1_cnt = 0;
|
| 836 |
break;
|
| 837 |
case 2: |
| 838 |
/* 3 mantissa in 7 bits */
|
| 839 |
if (s->mant2_cnt == 0) |
| 840 |
bits += 7;
|
| 841 |
if (++s->mant2_cnt == 3) |
| 842 |
s->mant2_cnt = 0;
|
| 843 |
break;
|
| 844 |
case 3: |
| 845 |
bits += 3;
|
| 846 |
break;
|
| 847 |
case 4: |
| 848 |
/* 2 mantissa in 7 bits */
|
| 849 |
if (s->mant4_cnt == 0) |
| 850 |
bits += 7;
|
| 851 |
if (++s->mant4_cnt == 2) |
| 852 |
s->mant4_cnt = 0;
|
| 853 |
break;
|
| 854 |
case 14: |
| 855 |
bits += 14;
|
| 856 |
break;
|
| 857 |
case 15: |
| 858 |
bits += 16;
|
| 859 |
break;
|
| 860 |
default:
|
| 861 |
bits += mant - 1;
|
| 862 |
break;
|
| 863 |
} |
| 864 |
} |
| 865 |
return bits;
|
| 866 |
} |
| 867 |
|
| 868 |
|
| 869 |
/**
|
| 870 |
* Calculate masking curve based on the final exponents.
|
| 871 |
* Also calculate the power spectral densities to use in future calculations.
|
| 872 |
*/
|
| 873 |
static void bit_alloc_masking(AC3EncodeContext *s) |
| 874 |
{
|
| 875 |
int blk, ch;
|
| 876 |
|
| 877 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 878 |
for (ch = 0; ch < s->channels; ch++) { |
| 879 |
if (s->exp_strategy[blk][ch] == EXP_REUSE) {
|
| 880 |
memcpy(s->psd[blk][ch], s->psd[blk-1][ch], AC3_MAX_COEFS*sizeof(s->psd[0][0][0])); |
| 881 |
memcpy(s->mask[blk][ch], s->mask[blk-1][ch], AC3_CRITICAL_BANDS*sizeof(s->mask[0][0][0])); |
| 882 |
} else {
|
| 883 |
ff_ac3_bit_alloc_calc_psd(s->encoded_exp[blk][ch], 0,
|
| 884 |
s->nb_coefs[ch], |
| 885 |
s->psd[blk][ch], s->band_psd[blk][ch]); |
| 886 |
ff_ac3_bit_alloc_calc_mask(&s->bit_alloc, s->band_psd[blk][ch], |
| 887 |
0, s->nb_coefs[ch],
|
| 888 |
ff_ac3_fast_gain_tab[s->fast_gain_code[ch]], |
| 889 |
ch == s->lfe_channel, |
| 890 |
DBA_NONE, 0, NULL, NULL, NULL, |
| 891 |
s->mask[blk][ch]); |
| 892 |
} |
| 893 |
} |
| 894 |
} |
| 895 |
} |
| 896 |
|
| 897 |
|
| 898 |
/**
|
| 899 |
* Run the bit allocation with a given SNR offset.
|
| 900 |
* This calculates the bit allocation pointers that will be used to determine
|
| 901 |
* the quantization of each mantissa.
|
| 902 |
* @return the number of bits needed for mantissas if the given SNR offset is
|
| 903 |
* is used.
|
| 904 |
*/
|
| 905 |
static int bit_alloc(AC3EncodeContext *s, |
| 906 |
uint8_t bap[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][AC3_MAX_COEFS], |
| 907 |
int snr_offset)
|
| 908 |
{
|
| 909 |
int blk, ch;
|
| 910 |
int mantissa_bits;
|
| 911 |
|
| 912 |
snr_offset = (snr_offset - 240) << 2; |
| 913 |
|
| 914 |
mantissa_bits = 0;
|
| 915 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 916 |
s->mant1_cnt = 0;
|
| 917 |
s->mant2_cnt = 0;
|
| 918 |
s->mant4_cnt = 0;
|
| 919 |
for (ch = 0; ch < s->channels; ch++) { |
| 920 |
ff_ac3_bit_alloc_calc_bap(s->mask[blk][ch], s->psd[blk][ch], 0,
|
| 921 |
s->nb_coefs[ch], snr_offset, |
| 922 |
s->bit_alloc.floor, ff_ac3_bap_tab, |
| 923 |
bap[blk][ch]); |
| 924 |
mantissa_bits += compute_mantissa_size(s, bap[blk][ch], s->nb_coefs[ch]); |
| 925 |
} |
| 926 |
} |
| 927 |
return mantissa_bits;
|
| 928 |
} |
| 929 |
|
| 930 |
|
| 931 |
/**
|
| 932 |
* Constant bitrate bit allocation search.
|
| 933 |
* Find the largest SNR offset that will allow data to fit in the frame.
|
| 934 |
*/
|
| 935 |
static int cbr_bit_allocation(AC3EncodeContext *s) |
| 936 |
{
|
| 937 |
int ch;
|
| 938 |
int bits_left;
|
| 939 |
int snr_offset;
|
| 940 |
|
| 941 |
bits_left = 8 * s->frame_size - (s->frame_bits + s->exponent_bits);
|
| 942 |
|
| 943 |
snr_offset = s->coarse_snr_offset << 4;
|
| 944 |
|
| 945 |
while (snr_offset >= 0 && |
| 946 |
bit_alloc(s, s->bap, snr_offset) > bits_left) {
|
| 947 |
snr_offset -= 64;
|
| 948 |
} |
| 949 |
if (snr_offset < 0) { |
| 950 |
return AVERROR(EINVAL);
|
| 951 |
} |
| 952 |
|
| 953 |
while (snr_offset + 64 <= 1023 && |
| 954 |
bit_alloc(s, s->bap1, snr_offset + 64) <= bits_left) {
|
| 955 |
snr_offset += 64;
|
| 956 |
memcpy(s->bap, s->bap1, sizeof(s->bap1));
|
| 957 |
} |
| 958 |
while (snr_offset + 16 <= 1023 && |
| 959 |
bit_alloc(s, s->bap1, snr_offset + 16) <= bits_left) {
|
| 960 |
snr_offset += 16;
|
| 961 |
memcpy(s->bap, s->bap1, sizeof(s->bap1));
|
| 962 |
} |
| 963 |
while (snr_offset + 4 <= 1023 && |
| 964 |
bit_alloc(s, s->bap1, snr_offset + 4) <= bits_left) {
|
| 965 |
snr_offset += 4;
|
| 966 |
memcpy(s->bap, s->bap1, sizeof(s->bap1));
|
| 967 |
} |
| 968 |
while (snr_offset + 1 <= 1023 && |
| 969 |
bit_alloc(s, s->bap1, snr_offset + 1) <= bits_left) {
|
| 970 |
snr_offset++; |
| 971 |
memcpy(s->bap, s->bap1, sizeof(s->bap1));
|
| 972 |
} |
| 973 |
|
| 974 |
s->coarse_snr_offset = snr_offset >> 4;
|
| 975 |
for (ch = 0; ch < s->channels; ch++) |
| 976 |
s->fine_snr_offset[ch] = snr_offset & 0xF;
|
| 977 |
|
| 978 |
return 0; |
| 979 |
} |
| 980 |
|
| 981 |
|
| 982 |
/**
|
| 983 |
* Perform bit allocation search.
|
| 984 |
* Finds the SNR offset value that maximizes quality and fits in the specified
|
| 985 |
* frame size. Output is the SNR offset and a set of bit allocation pointers
|
| 986 |
* used to quantize the mantissas.
|
| 987 |
*/
|
| 988 |
static int compute_bit_allocation(AC3EncodeContext *s) |
| 989 |
{
|
| 990 |
/* count frame bits other than exponents and mantissas */
|
| 991 |
count_frame_bits(s); |
| 992 |
|
| 993 |
/* calculate psd and masking curve before doing bit allocation */
|
| 994 |
bit_alloc_masking(s); |
| 995 |
|
| 996 |
return cbr_bit_allocation(s);
|
| 997 |
} |
| 998 |
|
| 999 |
|
| 1000 |
/**
|
| 1001 |
* Symmetric quantization on 'levels' levels.
|
| 1002 |
*/
|
| 1003 |
static inline int sym_quant(int c, int e, int levels) |
| 1004 |
{
|
| 1005 |
int v;
|
| 1006 |
|
| 1007 |
if (c >= 0) { |
| 1008 |
v = (levels * (c << e)) >> 24;
|
| 1009 |
v = (v + 1) >> 1; |
| 1010 |
v = (levels >> 1) + v;
|
| 1011 |
} else {
|
| 1012 |
v = (levels * ((-c) << e)) >> 24;
|
| 1013 |
v = (v + 1) >> 1; |
| 1014 |
v = (levels >> 1) - v;
|
| 1015 |
} |
| 1016 |
assert (v >= 0 && v < levels);
|
| 1017 |
return v;
|
| 1018 |
} |
| 1019 |
|
| 1020 |
|
| 1021 |
/**
|
| 1022 |
* Asymmetric quantization on 2^qbits levels.
|
| 1023 |
*/
|
| 1024 |
static inline int asym_quant(int c, int e, int qbits) |
| 1025 |
{
|
| 1026 |
int lshift, m, v;
|
| 1027 |
|
| 1028 |
lshift = e + qbits - 24;
|
| 1029 |
if (lshift >= 0) |
| 1030 |
v = c << lshift; |
| 1031 |
else
|
| 1032 |
v = c >> (-lshift); |
| 1033 |
/* rounding */
|
| 1034 |
v = (v + 1) >> 1; |
| 1035 |
m = (1 << (qbits-1)); |
| 1036 |
if (v >= m)
|
| 1037 |
v = m - 1;
|
| 1038 |
assert(v >= -m); |
| 1039 |
return v & ((1 << qbits)-1); |
| 1040 |
} |
| 1041 |
|
| 1042 |
|
| 1043 |
/**
|
| 1044 |
* Quantize a set of mantissas for a single channel in a single block.
|
| 1045 |
*/
|
| 1046 |
static void quantize_mantissas_blk_ch(AC3EncodeContext *s, |
| 1047 |
int32_t *mdct_coef, int8_t exp_shift, |
| 1048 |
uint8_t *encoded_exp, uint8_t *bap, |
| 1049 |
uint16_t *qmant, int n)
|
| 1050 |
{
|
| 1051 |
int i;
|
| 1052 |
|
| 1053 |
for (i = 0; i < n; i++) { |
| 1054 |
int v;
|
| 1055 |
int c = mdct_coef[i];
|
| 1056 |
int e = encoded_exp[i] - exp_shift;
|
| 1057 |
int b = bap[i];
|
| 1058 |
switch (b) {
|
| 1059 |
case 0: |
| 1060 |
v = 0;
|
| 1061 |
break;
|
| 1062 |
case 1: |
| 1063 |
v = sym_quant(c, e, 3);
|
| 1064 |
switch (s->mant1_cnt) {
|
| 1065 |
case 0: |
| 1066 |
s->qmant1_ptr = &qmant[i]; |
| 1067 |
v = 9 * v;
|
| 1068 |
s->mant1_cnt = 1;
|
| 1069 |
break;
|
| 1070 |
case 1: |
| 1071 |
*s->qmant1_ptr += 3 * v;
|
| 1072 |
s->mant1_cnt = 2;
|
| 1073 |
v = 128;
|
| 1074 |
break;
|
| 1075 |
default:
|
| 1076 |
*s->qmant1_ptr += v; |
| 1077 |
s->mant1_cnt = 0;
|
| 1078 |
v = 128;
|
| 1079 |
break;
|
| 1080 |
} |
| 1081 |
break;
|
| 1082 |
case 2: |
| 1083 |
v = sym_quant(c, e, 5);
|
| 1084 |
switch (s->mant2_cnt) {
|
| 1085 |
case 0: |
| 1086 |
s->qmant2_ptr = &qmant[i]; |
| 1087 |
v = 25 * v;
|
| 1088 |
s->mant2_cnt = 1;
|
| 1089 |
break;
|
| 1090 |
case 1: |
| 1091 |
*s->qmant2_ptr += 5 * v;
|
| 1092 |
s->mant2_cnt = 2;
|
| 1093 |
v = 128;
|
| 1094 |
break;
|
| 1095 |
default:
|
| 1096 |
*s->qmant2_ptr += v; |
| 1097 |
s->mant2_cnt = 0;
|
| 1098 |
v = 128;
|
| 1099 |
break;
|
| 1100 |
} |
| 1101 |
break;
|
| 1102 |
case 3: |
| 1103 |
v = sym_quant(c, e, 7);
|
| 1104 |
break;
|
| 1105 |
case 4: |
| 1106 |
v = sym_quant(c, e, 11);
|
| 1107 |
switch (s->mant4_cnt) {
|
| 1108 |
case 0: |
| 1109 |
s->qmant4_ptr = &qmant[i]; |
| 1110 |
v = 11 * v;
|
| 1111 |
s->mant4_cnt = 1;
|
| 1112 |
break;
|
| 1113 |
default:
|
| 1114 |
*s->qmant4_ptr += v; |
| 1115 |
s->mant4_cnt = 0;
|
| 1116 |
v = 128;
|
| 1117 |
break;
|
| 1118 |
} |
| 1119 |
break;
|
| 1120 |
case 5: |
| 1121 |
v = sym_quant(c, e, 15);
|
| 1122 |
break;
|
| 1123 |
case 14: |
| 1124 |
v = asym_quant(c, e, 14);
|
| 1125 |
break;
|
| 1126 |
case 15: |
| 1127 |
v = asym_quant(c, e, 16);
|
| 1128 |
break;
|
| 1129 |
default:
|
| 1130 |
v = asym_quant(c, e, b - 1);
|
| 1131 |
break;
|
| 1132 |
} |
| 1133 |
qmant[i] = v; |
| 1134 |
} |
| 1135 |
} |
| 1136 |
|
| 1137 |
|
| 1138 |
/**
|
| 1139 |
* Quantize mantissas using coefficients, exponents, and bit allocation pointers.
|
| 1140 |
*/
|
| 1141 |
static void quantize_mantissas(AC3EncodeContext *s) |
| 1142 |
{
|
| 1143 |
int blk, ch;
|
| 1144 |
|
| 1145 |
|
| 1146 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1147 |
s->mant1_cnt = s->mant2_cnt = s->mant4_cnt = 0;
|
| 1148 |
s->qmant1_ptr = s->qmant2_ptr = s->qmant4_ptr = NULL;
|
| 1149 |
|
| 1150 |
for (ch = 0; ch < s->channels; ch++) { |
| 1151 |
quantize_mantissas_blk_ch(s, s->mdct_coef[blk][ch], s->exp_shift[blk][ch], |
| 1152 |
s->encoded_exp[blk][ch], s->bap[blk][ch], |
| 1153 |
s->qmant[blk][ch], s->nb_coefs[ch]); |
| 1154 |
} |
| 1155 |
} |
| 1156 |
} |
| 1157 |
|
| 1158 |
|
| 1159 |
/**
|
| 1160 |
* Write the AC-3 frame header to the output bitstream.
|
| 1161 |
*/
|
| 1162 |
static void output_frame_header(AC3EncodeContext *s) |
| 1163 |
{
|
| 1164 |
put_bits(&s->pb, 16, 0x0b77); /* frame header */ |
| 1165 |
put_bits(&s->pb, 16, 0); /* crc1: will be filled later */ |
| 1166 |
put_bits(&s->pb, 2, s->bit_alloc.sr_code);
|
| 1167 |
put_bits(&s->pb, 6, s->frame_size_code + (s->frame_size - s->frame_size_min) / 2); |
| 1168 |
put_bits(&s->pb, 5, s->bitstream_id);
|
| 1169 |
put_bits(&s->pb, 3, s->bitstream_mode);
|
| 1170 |
put_bits(&s->pb, 3, s->channel_mode);
|
| 1171 |
if ((s->channel_mode & 0x01) && s->channel_mode != AC3_CHMODE_MONO) |
| 1172 |
put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */ |
| 1173 |
if (s->channel_mode & 0x04) |
| 1174 |
put_bits(&s->pb, 2, 1); /* XXX -6 dB */ |
| 1175 |
if (s->channel_mode == AC3_CHMODE_STEREO)
|
| 1176 |
put_bits(&s->pb, 2, 0); /* surround not indicated */ |
| 1177 |
put_bits(&s->pb, 1, s->lfe_on); /* LFE */ |
| 1178 |
put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */ |
| 1179 |
put_bits(&s->pb, 1, 0); /* no compression control word */ |
| 1180 |
put_bits(&s->pb, 1, 0); /* no lang code */ |
| 1181 |
put_bits(&s->pb, 1, 0); /* no audio production info */ |
| 1182 |
put_bits(&s->pb, 1, 0); /* no copyright */ |
| 1183 |
put_bits(&s->pb, 1, 1); /* original bitstream */ |
| 1184 |
put_bits(&s->pb, 1, 0); /* no time code 1 */ |
| 1185 |
put_bits(&s->pb, 1, 0); /* no time code 2 */ |
| 1186 |
put_bits(&s->pb, 1, 0); /* no additional bit stream info */ |
| 1187 |
} |
| 1188 |
|
| 1189 |
|
| 1190 |
/**
|
| 1191 |
* Write one audio block to the output bitstream.
|
| 1192 |
*/
|
| 1193 |
static void output_audio_block(AC3EncodeContext *s, |
| 1194 |
int block_num)
|
| 1195 |
{
|
| 1196 |
int ch, i, baie, rbnd;
|
| 1197 |
|
| 1198 |
for (ch = 0; ch < s->fbw_channels; ch++) |
| 1199 |
put_bits(&s->pb, 1, 0); /* no block switching */ |
| 1200 |
for (ch = 0; ch < s->fbw_channels; ch++) |
| 1201 |
put_bits(&s->pb, 1, 1); /* no dither */ |
| 1202 |
put_bits(&s->pb, 1, 0); /* no dynamic range */ |
| 1203 |
if (!block_num) {
|
| 1204 |
put_bits(&s->pb, 1, 1); /* coupling strategy present */ |
| 1205 |
put_bits(&s->pb, 1, 0); /* no coupling strategy */ |
| 1206 |
} else {
|
| 1207 |
put_bits(&s->pb, 1, 0); /* no new coupling strategy */ |
| 1208 |
} |
| 1209 |
|
| 1210 |
if (s->channel_mode == AC3_CHMODE_STEREO) {
|
| 1211 |
if (!block_num) {
|
| 1212 |
/* first block must define rematrixing (rematstr) */
|
| 1213 |
put_bits(&s->pb, 1, 1); |
| 1214 |
|
| 1215 |
/* dummy rematrixing rematflg(1:4)=0 */
|
| 1216 |
for (rbnd = 0; rbnd < 4; rbnd++) |
| 1217 |
put_bits(&s->pb, 1, 0); |
| 1218 |
} else {
|
| 1219 |
/* no matrixing (but should be used in the future) */
|
| 1220 |
put_bits(&s->pb, 1, 0); |
| 1221 |
} |
| 1222 |
} |
| 1223 |
|
| 1224 |
/* exponent strategy */
|
| 1225 |
for (ch = 0; ch < s->fbw_channels; ch++) |
| 1226 |
put_bits(&s->pb, 2, s->exp_strategy[block_num][ch]);
|
| 1227 |
|
| 1228 |
if (s->lfe_on)
|
| 1229 |
put_bits(&s->pb, 1, s->exp_strategy[block_num][s->lfe_channel]);
|
| 1230 |
|
| 1231 |
/* bandwidth */
|
| 1232 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1233 |
if (s->exp_strategy[block_num][ch] != EXP_REUSE)
|
| 1234 |
put_bits(&s->pb, 6, s->bandwidth_code[ch]);
|
| 1235 |
} |
| 1236 |
|
| 1237 |
/* exponents */
|
| 1238 |
for (ch = 0; ch < s->channels; ch++) { |
| 1239 |
if (s->exp_strategy[block_num][ch] == EXP_REUSE)
|
| 1240 |
continue;
|
| 1241 |
|
| 1242 |
/* first exponent */
|
| 1243 |
put_bits(&s->pb, 4, s->grouped_exp[block_num][ch][0]); |
| 1244 |
|
| 1245 |
/* next ones are delta-encoded and grouped */
|
| 1246 |
for (i = 1; i <= s->num_exp_groups[block_num][ch]; i++) |
| 1247 |
put_bits(&s->pb, 7, s->grouped_exp[block_num][ch][i]);
|
| 1248 |
|
| 1249 |
if (ch != s->lfe_channel)
|
| 1250 |
put_bits(&s->pb, 2, 0); /* no gain range info */ |
| 1251 |
} |
| 1252 |
|
| 1253 |
/* bit allocation info */
|
| 1254 |
baie = (block_num == 0);
|
| 1255 |
put_bits(&s->pb, 1, baie);
|
| 1256 |
if (baie) {
|
| 1257 |
put_bits(&s->pb, 2, s->slow_decay_code);
|
| 1258 |
put_bits(&s->pb, 2, s->fast_decay_code);
|
| 1259 |
put_bits(&s->pb, 2, s->slow_gain_code);
|
| 1260 |
put_bits(&s->pb, 2, s->db_per_bit_code);
|
| 1261 |
put_bits(&s->pb, 3, s->floor_code);
|
| 1262 |
} |
| 1263 |
|
| 1264 |
/* snr offset */
|
| 1265 |
put_bits(&s->pb, 1, baie);
|
| 1266 |
if (baie) {
|
| 1267 |
put_bits(&s->pb, 6, s->coarse_snr_offset);
|
| 1268 |
for (ch = 0; ch < s->channels; ch++) { |
| 1269 |
put_bits(&s->pb, 4, s->fine_snr_offset[ch]);
|
| 1270 |
put_bits(&s->pb, 3, s->fast_gain_code[ch]);
|
| 1271 |
} |
| 1272 |
} |
| 1273 |
|
| 1274 |
put_bits(&s->pb, 1, 0); /* no delta bit allocation */ |
| 1275 |
put_bits(&s->pb, 1, 0); /* no data to skip */ |
| 1276 |
|
| 1277 |
/* mantissa encoding */
|
| 1278 |
for (ch = 0; ch < s->channels; ch++) { |
| 1279 |
int b, q;
|
| 1280 |
|
| 1281 |
for (i = 0; i < s->nb_coefs[ch]; i++) { |
| 1282 |
q = s->qmant[block_num][ch][i]; |
| 1283 |
b = s->bap[block_num][ch][i]; |
| 1284 |
switch (b) {
|
| 1285 |
case 0: break; |
| 1286 |
case 1: if (q != 128) put_bits(&s->pb, 5, q); break; |
| 1287 |
case 2: if (q != 128) put_bits(&s->pb, 7, q); break; |
| 1288 |
case 3: put_bits(&s->pb, 3, q); break; |
| 1289 |
case 4: if (q != 128) put_bits(&s->pb, 7, q); break; |
| 1290 |
case 14: put_bits(&s->pb, 14, q); break; |
| 1291 |
case 15: put_bits(&s->pb, 16, q); break; |
| 1292 |
default: put_bits(&s->pb, b-1, q); break; |
| 1293 |
} |
| 1294 |
} |
| 1295 |
} |
| 1296 |
} |
| 1297 |
|
| 1298 |
|
| 1299 |
/** CRC-16 Polynomial */
|
| 1300 |
#define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16)) |
| 1301 |
|
| 1302 |
|
| 1303 |
static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly) |
| 1304 |
{
|
| 1305 |
unsigned int c; |
| 1306 |
|
| 1307 |
c = 0;
|
| 1308 |
while (a) {
|
| 1309 |
if (a & 1) |
| 1310 |
c ^= b; |
| 1311 |
a = a >> 1;
|
| 1312 |
b = b << 1;
|
| 1313 |
if (b & (1 << 16)) |
| 1314 |
b ^= poly; |
| 1315 |
} |
| 1316 |
return c;
|
| 1317 |
} |
| 1318 |
|
| 1319 |
|
| 1320 |
static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly) |
| 1321 |
{
|
| 1322 |
unsigned int r; |
| 1323 |
r = 1;
|
| 1324 |
while (n) {
|
| 1325 |
if (n & 1) |
| 1326 |
r = mul_poly(r, a, poly); |
| 1327 |
a = mul_poly(a, a, poly); |
| 1328 |
n >>= 1;
|
| 1329 |
} |
| 1330 |
return r;
|
| 1331 |
} |
| 1332 |
|
| 1333 |
|
| 1334 |
/**
|
| 1335 |
* Fill the end of the frame with 0's and compute the two CRCs.
|
| 1336 |
*/
|
| 1337 |
static void output_frame_end(AC3EncodeContext *s) |
| 1338 |
{
|
| 1339 |
int frame_size, frame_size_58, pad_bytes, crc1, crc2, crc_inv;
|
| 1340 |
uint8_t *frame; |
| 1341 |
|
| 1342 |
frame_size = s->frame_size; /* frame size in words */
|
| 1343 |
/* align to 8 bits */
|
| 1344 |
flush_put_bits(&s->pb); |
| 1345 |
/* add zero bytes to reach the frame size */
|
| 1346 |
frame = s->pb.buf; |
| 1347 |
pad_bytes = s->frame_size - (put_bits_ptr(&s->pb) - frame) - 2;
|
| 1348 |
assert(pad_bytes >= 0);
|
| 1349 |
if (pad_bytes > 0) |
| 1350 |
memset(put_bits_ptr(&s->pb), 0, pad_bytes);
|
| 1351 |
|
| 1352 |
/* Now we must compute both crcs : this is not so easy for crc1
|
| 1353 |
because it is at the beginning of the data... */
|
| 1354 |
frame_size_58 = ((frame_size >> 2) + (frame_size >> 4)) << 1; |
| 1355 |
|
| 1356 |
crc1 = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
|
| 1357 |
frame + 4, frame_size_58 - 4)); |
| 1358 |
|
| 1359 |
/* XXX: could precompute crc_inv */
|
| 1360 |
crc_inv = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY); |
| 1361 |
crc1 = mul_poly(crc_inv, crc1, CRC16_POLY); |
| 1362 |
AV_WB16(frame + 2, crc1);
|
| 1363 |
|
| 1364 |
crc2 = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
|
| 1365 |
frame + frame_size_58, |
| 1366 |
frame_size - frame_size_58 - 2));
|
| 1367 |
AV_WB16(frame + frame_size - 2, crc2);
|
| 1368 |
} |
| 1369 |
|
| 1370 |
|
| 1371 |
/**
|
| 1372 |
* Write the frame to the output bitstream.
|
| 1373 |
*/
|
| 1374 |
static void output_frame(AC3EncodeContext *s, |
| 1375 |
unsigned char *frame) |
| 1376 |
{
|
| 1377 |
int blk;
|
| 1378 |
|
| 1379 |
init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE); |
| 1380 |
|
| 1381 |
output_frame_header(s); |
| 1382 |
|
| 1383 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
| 1384 |
output_audio_block(s, blk); |
| 1385 |
} |
| 1386 |
|
| 1387 |
output_frame_end(s); |
| 1388 |
} |
| 1389 |
|
| 1390 |
|
| 1391 |
/**
|
| 1392 |
* Encode a single AC-3 frame.
|
| 1393 |
*/
|
| 1394 |
static int ac3_encode_frame(AVCodecContext *avctx, |
| 1395 |
unsigned char *frame, int buf_size, void *data) |
| 1396 |
{
|
| 1397 |
AC3EncodeContext *s = avctx->priv_data; |
| 1398 |
const int16_t *samples = data;
|
| 1399 |
int ret;
|
| 1400 |
|
| 1401 |
if (s->bit_alloc.sr_code == 1) |
| 1402 |
adjust_frame_size(s); |
| 1403 |
|
| 1404 |
deinterleave_input_samples(s, samples); |
| 1405 |
|
| 1406 |
apply_mdct(s); |
| 1407 |
|
| 1408 |
process_exponents(s); |
| 1409 |
|
| 1410 |
ret = compute_bit_allocation(s); |
| 1411 |
if (ret) {
|
| 1412 |
av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
|
| 1413 |
return ret;
|
| 1414 |
} |
| 1415 |
|
| 1416 |
quantize_mantissas(s); |
| 1417 |
|
| 1418 |
output_frame(s, frame); |
| 1419 |
|
| 1420 |
return s->frame_size;
|
| 1421 |
} |
| 1422 |
|
| 1423 |
|
| 1424 |
/**
|
| 1425 |
* Finalize encoding and free any memory allocated by the encoder.
|
| 1426 |
*/
|
| 1427 |
static av_cold int ac3_encode_close(AVCodecContext *avctx) |
| 1428 |
{
|
| 1429 |
av_freep(&avctx->coded_frame); |
| 1430 |
return 0; |
| 1431 |
} |
| 1432 |
|
| 1433 |
|
| 1434 |
/**
|
| 1435 |
* Set channel information during initialization.
|
| 1436 |
*/
|
| 1437 |
static av_cold int set_channel_info(AC3EncodeContext *s, int channels, |
| 1438 |
int64_t *channel_layout) |
| 1439 |
{
|
| 1440 |
int ch_layout;
|
| 1441 |
|
| 1442 |
if (channels < 1 || channels > AC3_MAX_CHANNELS) |
| 1443 |
return AVERROR(EINVAL);
|
| 1444 |
if ((uint64_t)*channel_layout > 0x7FF) |
| 1445 |
return AVERROR(EINVAL);
|
| 1446 |
ch_layout = *channel_layout; |
| 1447 |
if (!ch_layout)
|
| 1448 |
ch_layout = avcodec_guess_channel_layout(channels, CODEC_ID_AC3, NULL);
|
| 1449 |
if (av_get_channel_layout_nb_channels(ch_layout) != channels)
|
| 1450 |
return AVERROR(EINVAL);
|
| 1451 |
|
| 1452 |
s->lfe_on = !!(ch_layout & AV_CH_LOW_FREQUENCY); |
| 1453 |
s->channels = channels; |
| 1454 |
s->fbw_channels = channels - s->lfe_on; |
| 1455 |
s->lfe_channel = s->lfe_on ? s->fbw_channels : -1;
|
| 1456 |
if (s->lfe_on)
|
| 1457 |
ch_layout -= AV_CH_LOW_FREQUENCY; |
| 1458 |
|
| 1459 |
switch (ch_layout) {
|
| 1460 |
case AV_CH_LAYOUT_MONO: s->channel_mode = AC3_CHMODE_MONO; break; |
| 1461 |
case AV_CH_LAYOUT_STEREO: s->channel_mode = AC3_CHMODE_STEREO; break; |
| 1462 |
case AV_CH_LAYOUT_SURROUND: s->channel_mode = AC3_CHMODE_3F; break; |
| 1463 |
case AV_CH_LAYOUT_2_1: s->channel_mode = AC3_CHMODE_2F1R; break; |
| 1464 |
case AV_CH_LAYOUT_4POINT0: s->channel_mode = AC3_CHMODE_3F1R; break; |
| 1465 |
case AV_CH_LAYOUT_QUAD:
|
| 1466 |
case AV_CH_LAYOUT_2_2: s->channel_mode = AC3_CHMODE_2F2R; break; |
| 1467 |
case AV_CH_LAYOUT_5POINT0:
|
| 1468 |
case AV_CH_LAYOUT_5POINT0_BACK: s->channel_mode = AC3_CHMODE_3F2R; break; |
| 1469 |
default:
|
| 1470 |
return AVERROR(EINVAL);
|
| 1471 |
} |
| 1472 |
|
| 1473 |
s->channel_map = ff_ac3_enc_channel_map[s->channel_mode][s->lfe_on]; |
| 1474 |
*channel_layout = ch_layout; |
| 1475 |
if (s->lfe_on)
|
| 1476 |
*channel_layout |= AV_CH_LOW_FREQUENCY; |
| 1477 |
|
| 1478 |
return 0; |
| 1479 |
} |
| 1480 |
|
| 1481 |
|
| 1482 |
static av_cold int validate_options(AVCodecContext *avctx, AC3EncodeContext *s) |
| 1483 |
{
|
| 1484 |
int i, ret;
|
| 1485 |
|
| 1486 |
/* validate channel layout */
|
| 1487 |
if (!avctx->channel_layout) {
|
| 1488 |
av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
|
| 1489 |
"encoder will guess the layout, but it "
|
| 1490 |
"might be incorrect.\n");
|
| 1491 |
} |
| 1492 |
ret = set_channel_info(s, avctx->channels, &avctx->channel_layout); |
| 1493 |
if (ret) {
|
| 1494 |
av_log(avctx, AV_LOG_ERROR, "invalid channel layout\n");
|
| 1495 |
return ret;
|
| 1496 |
} |
| 1497 |
|
| 1498 |
/* validate sample rate */
|
| 1499 |
for (i = 0; i < 9; i++) { |
| 1500 |
if ((ff_ac3_sample_rate_tab[i / 3] >> (i % 3)) == avctx->sample_rate) |
| 1501 |
break;
|
| 1502 |
} |
| 1503 |
if (i == 9) { |
| 1504 |
av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
|
| 1505 |
return AVERROR(EINVAL);
|
| 1506 |
} |
| 1507 |
s->sample_rate = avctx->sample_rate; |
| 1508 |
s->bit_alloc.sr_shift = i % 3;
|
| 1509 |
s->bit_alloc.sr_code = i / 3;
|
| 1510 |
|
| 1511 |
/* validate bit rate */
|
| 1512 |
for (i = 0; i < 19; i++) { |
| 1513 |
if ((ff_ac3_bitrate_tab[i] >> s->bit_alloc.sr_shift)*1000 == avctx->bit_rate) |
| 1514 |
break;
|
| 1515 |
} |
| 1516 |
if (i == 19) { |
| 1517 |
av_log(avctx, AV_LOG_ERROR, "invalid bit rate\n");
|
| 1518 |
return AVERROR(EINVAL);
|
| 1519 |
} |
| 1520 |
s->bit_rate = avctx->bit_rate; |
| 1521 |
s->frame_size_code = i << 1;
|
| 1522 |
|
| 1523 |
return 0; |
| 1524 |
} |
| 1525 |
|
| 1526 |
|
| 1527 |
/**
|
| 1528 |
* Set bandwidth for all channels.
|
| 1529 |
* The user can optionally supply a cutoff frequency. Otherwise an appropriate
|
| 1530 |
* default value will be used.
|
| 1531 |
*/
|
| 1532 |
static av_cold void set_bandwidth(AC3EncodeContext *s, int cutoff) |
| 1533 |
{
|
| 1534 |
int ch, bw_code;
|
| 1535 |
|
| 1536 |
if (cutoff) {
|
| 1537 |
/* calculate bandwidth based on user-specified cutoff frequency */
|
| 1538 |
int fbw_coeffs;
|
| 1539 |
cutoff = av_clip(cutoff, 1, s->sample_rate >> 1); |
| 1540 |
fbw_coeffs = cutoff * 2 * AC3_MAX_COEFS / s->sample_rate;
|
| 1541 |
bw_code = av_clip((fbw_coeffs - 73) / 3, 0, 60); |
| 1542 |
} else {
|
| 1543 |
/* use default bandwidth setting */
|
| 1544 |
/* XXX: should compute the bandwidth according to the frame
|
| 1545 |
size, so that we avoid annoying high frequency artifacts */
|
| 1546 |
bw_code = 50;
|
| 1547 |
} |
| 1548 |
|
| 1549 |
/* set number of coefficients for each channel */
|
| 1550 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
| 1551 |
s->bandwidth_code[ch] = bw_code; |
| 1552 |
s->nb_coefs[ch] = bw_code * 3 + 73; |
| 1553 |
} |
| 1554 |
if (s->lfe_on)
|
| 1555 |
s->nb_coefs[s->lfe_channel] = 7; /* LFE channel always has 7 coefs */ |
| 1556 |
} |
| 1557 |
|
| 1558 |
|
| 1559 |
/**
|
| 1560 |
* Initialize the encoder.
|
| 1561 |
*/
|
| 1562 |
static av_cold int ac3_encode_init(AVCodecContext *avctx) |
| 1563 |
{
|
| 1564 |
AC3EncodeContext *s = avctx->priv_data; |
| 1565 |
int ret;
|
| 1566 |
|
| 1567 |
avctx->frame_size = AC3_FRAME_SIZE; |
| 1568 |
|
| 1569 |
ac3_common_init(); |
| 1570 |
|
| 1571 |
ret = validate_options(avctx, s); |
| 1572 |
if (ret)
|
| 1573 |
return ret;
|
| 1574 |
|
| 1575 |
s->bitstream_id = 8 + s->bit_alloc.sr_shift;
|
| 1576 |
s->bitstream_mode = 0; /* complete main audio service */ |
| 1577 |
|
| 1578 |
s->frame_size_min = 2 * ff_ac3_frame_size_tab[s->frame_size_code][s->bit_alloc.sr_code];
|
| 1579 |
s->bits_written = 0;
|
| 1580 |
s->samples_written = 0;
|
| 1581 |
s->frame_size = s->frame_size_min; |
| 1582 |
|
| 1583 |
set_bandwidth(s, avctx->cutoff); |
| 1584 |
|
| 1585 |
bit_alloc_init(s); |
| 1586 |
|
| 1587 |
mdct_init(9);
|
| 1588 |
|
| 1589 |
avctx->coded_frame= avcodec_alloc_frame(); |
| 1590 |
avctx->coded_frame->key_frame= 1;
|
| 1591 |
|
| 1592 |
return 0; |
| 1593 |
} |
| 1594 |
|
| 1595 |
|
| 1596 |
#ifdef TEST
|
| 1597 |
/*************************************************************************/
|
| 1598 |
/* TEST */
|
| 1599 |
|
| 1600 |
#include "libavutil/lfg.h" |
| 1601 |
|
| 1602 |
#define FN (MDCT_SAMPLES/4) |
| 1603 |
|
| 1604 |
|
| 1605 |
static void fft_test(AVLFG *lfg) |
| 1606 |
{
|
| 1607 |
IComplex in[FN], in1[FN]; |
| 1608 |
int k, n, i;
|
| 1609 |
float sum_re, sum_im, a;
|
| 1610 |
|
| 1611 |
for (i = 0; i < FN; i++) { |
| 1612 |
in[i].re = av_lfg_get(lfg) % 65535 - 32767; |
| 1613 |
in[i].im = av_lfg_get(lfg) % 65535 - 32767; |
| 1614 |
in1[i] = in[i]; |
| 1615 |
} |
| 1616 |
fft(in, 7);
|
| 1617 |
|
| 1618 |
/* do it by hand */
|
| 1619 |
for (k = 0; k < FN; k++) { |
| 1620 |
sum_re = 0;
|
| 1621 |
sum_im = 0;
|
| 1622 |
for (n = 0; n < FN; n++) { |
| 1623 |
a = -2 * M_PI * (n * k) / FN;
|
| 1624 |
sum_re += in1[n].re * cos(a) - in1[n].im * sin(a); |
| 1625 |
sum_im += in1[n].re * sin(a) + in1[n].im * cos(a); |
| 1626 |
} |
| 1627 |
av_log(NULL, AV_LOG_DEBUG, "%3d: %6d,%6d %6.0f,%6.0f\n", |
| 1628 |
k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); |
| 1629 |
} |
| 1630 |
} |
| 1631 |
|
| 1632 |
|
| 1633 |
static void mdct_test(AVLFG *lfg) |
| 1634 |
{
|
| 1635 |
int16_t input[MDCT_SAMPLES]; |
| 1636 |
int32_t output[AC3_MAX_COEFS]; |
| 1637 |
float input1[MDCT_SAMPLES];
|
| 1638 |
float output1[AC3_MAX_COEFS];
|
| 1639 |
float s, a, err, e, emax;
|
| 1640 |
int i, k, n;
|
| 1641 |
|
| 1642 |
for (i = 0; i < MDCT_SAMPLES; i++) { |
| 1643 |
input[i] = (av_lfg_get(lfg) % 65535 - 32767) * 9 / 10; |
| 1644 |
input1[i] = input[i]; |
| 1645 |
} |
| 1646 |
|
| 1647 |
mdct512(output, input); |
| 1648 |
|
| 1649 |
/* do it by hand */
|
| 1650 |
for (k = 0; k < AC3_MAX_COEFS; k++) { |
| 1651 |
s = 0;
|
| 1652 |
for (n = 0; n < MDCT_SAMPLES; n++) { |
| 1653 |
a = (2*M_PI*(2*n+1+MDCT_SAMPLES/2)*(2*k+1) / (4 * MDCT_SAMPLES)); |
| 1654 |
s += input1[n] * cos(a); |
| 1655 |
} |
| 1656 |
output1[k] = -2 * s / MDCT_SAMPLES;
|
| 1657 |
} |
| 1658 |
|
| 1659 |
err = 0;
|
| 1660 |
emax = 0;
|
| 1661 |
for (i = 0; i < AC3_MAX_COEFS; i++) { |
| 1662 |
av_log(NULL, AV_LOG_DEBUG, "%3d: %7d %7.0f\n", i, output[i], output1[i]); |
| 1663 |
e = output[i] - output1[i]; |
| 1664 |
if (e > emax)
|
| 1665 |
emax = e; |
| 1666 |
err += e * e; |
| 1667 |
} |
| 1668 |
av_log(NULL, AV_LOG_DEBUG, "err2=%f emax=%f\n", err / AC3_MAX_COEFS, emax); |
| 1669 |
} |
| 1670 |
|
| 1671 |
|
| 1672 |
int main(void) |
| 1673 |
{
|
| 1674 |
AVLFG lfg; |
| 1675 |
|
| 1676 |
av_log_set_level(AV_LOG_DEBUG); |
| 1677 |
mdct_init(9);
|
| 1678 |
|
| 1679 |
fft_test(&lfg); |
| 1680 |
mdct_test(&lfg); |
| 1681 |
|
| 1682 |
return 0; |
| 1683 |
} |
| 1684 |
#endif /* TEST */ |
| 1685 |
|
| 1686 |
|
| 1687 |
AVCodec ac3_encoder = {
|
| 1688 |
"ac3",
|
| 1689 |
AVMEDIA_TYPE_AUDIO, |
| 1690 |
CODEC_ID_AC3, |
| 1691 |
sizeof(AC3EncodeContext),
|
| 1692 |
ac3_encode_init, |
| 1693 |
ac3_encode_frame, |
| 1694 |
ac3_encode_close, |
| 1695 |
NULL,
|
| 1696 |
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, |
| 1697 |
.long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
|
| 1698 |
.channel_layouts = (const int64_t[]){
|
| 1699 |
AV_CH_LAYOUT_MONO, |
| 1700 |
AV_CH_LAYOUT_STEREO, |
| 1701 |
AV_CH_LAYOUT_2_1, |
| 1702 |
AV_CH_LAYOUT_SURROUND, |
| 1703 |
AV_CH_LAYOUT_2_2, |
| 1704 |
AV_CH_LAYOUT_QUAD, |
| 1705 |
AV_CH_LAYOUT_4POINT0, |
| 1706 |
AV_CH_LAYOUT_5POINT0, |
| 1707 |
AV_CH_LAYOUT_5POINT0_BACK, |
| 1708 |
(AV_CH_LAYOUT_MONO | AV_CH_LOW_FREQUENCY), |
| 1709 |
(AV_CH_LAYOUT_STEREO | AV_CH_LOW_FREQUENCY), |
| 1710 |
(AV_CH_LAYOUT_2_1 | AV_CH_LOW_FREQUENCY), |
| 1711 |
(AV_CH_LAYOUT_SURROUND | AV_CH_LOW_FREQUENCY), |
| 1712 |
(AV_CH_LAYOUT_2_2 | AV_CH_LOW_FREQUENCY), |
| 1713 |
(AV_CH_LAYOUT_QUAD | AV_CH_LOW_FREQUENCY), |
| 1714 |
(AV_CH_LAYOUT_4POINT0 | AV_CH_LOW_FREQUENCY), |
| 1715 |
AV_CH_LAYOUT_5POINT1, |
| 1716 |
AV_CH_LAYOUT_5POINT1_BACK, |
| 1717 |
0 },
|
| 1718 |
}; |