ffmpeg / libavcodec / ac3enc.c @ 834550ea
History | View | Annotate | Download (56.8 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 |
#define MDCT_NBITS 9 |
41 |
#define MDCT_SAMPLES (1 << MDCT_NBITS) |
42 |
|
43 |
/** Maximum number of exponent groups. +1 for separate DC exponent. */
|
44 |
#define AC3_MAX_EXP_GROUPS 85 |
45 |
|
46 |
/** Scale a float value by 2^bits and convert to an integer. */
|
47 |
#define SCALE_FLOAT(a, bits) lrintf((a) * (float)(1 << (bits))) |
48 |
|
49 |
/** Scale a float value by 2^15, convert to an integer, and clip to int16_t range. */
|
50 |
#define FIX15(a) av_clip_int16(SCALE_FLOAT(a, 15)) |
51 |
|
52 |
|
53 |
/**
|
54 |
* Compex number.
|
55 |
* Used in fixed-point MDCT calculation.
|
56 |
*/
|
57 |
typedef struct IComplex { |
58 |
int16_t re,im; |
59 |
} IComplex; |
60 |
|
61 |
typedef struct AC3MDCTContext { |
62 |
AVCodecContext *avctx; ///< parent context for av_log()
|
63 |
int16_t *rot_tmp; ///< temp buffer for pre-rotated samples
|
64 |
IComplex *cplx_tmp; ///< temp buffer for complex pre-rotated samples
|
65 |
} AC3MDCTContext; |
66 |
|
67 |
/**
|
68 |
* Data for a single audio block.
|
69 |
*/
|
70 |
typedef struct AC3Block { |
71 |
uint8_t **bap; ///< bit allocation pointers (bap)
|
72 |
int32_t **mdct_coef; ///< MDCT coefficients
|
73 |
uint8_t **exp; ///< original exponents
|
74 |
uint8_t **grouped_exp; ///< grouped exponents
|
75 |
int16_t **psd; ///< psd per frequency bin
|
76 |
int16_t **band_psd; ///< psd per critical band
|
77 |
int16_t **mask; ///< masking curve
|
78 |
uint16_t **qmant; ///< quantized mantissas
|
79 |
uint8_t num_exp_groups[AC3_MAX_CHANNELS]; ///< number of exponent groups
|
80 |
uint8_t exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
|
81 |
int8_t exp_shift[AC3_MAX_CHANNELS]; ///< exponent shift values
|
82 |
} AC3Block; |
83 |
|
84 |
/**
|
85 |
* AC-3 encoder private context.
|
86 |
*/
|
87 |
typedef struct AC3EncodeContext { |
88 |
PutBitContext pb; ///< bitstream writer context
|
89 |
DSPContext dsp; |
90 |
AC3MDCTContext mdct; ///< MDCT context
|
91 |
|
92 |
AC3Block blocks[AC3_MAX_BLOCKS]; ///< per-block info
|
93 |
|
94 |
int bitstream_id; ///< bitstream id (bsid) |
95 |
int bitstream_mode; ///< bitstream mode (bsmod) |
96 |
|
97 |
int bit_rate; ///< target bit rate, in bits-per-second |
98 |
int sample_rate; ///< sampling frequency, in Hz |
99 |
|
100 |
int frame_size_min; ///< minimum frame size in case rounding is necessary |
101 |
int frame_size; ///< current frame size in bytes |
102 |
int frame_size_code; ///< frame size code (frmsizecod) |
103 |
int bits_written; ///< bit count (used to avg. bitrate) |
104 |
int samples_written; ///< sample count (used to avg. bitrate) |
105 |
|
106 |
int fbw_channels; ///< number of full-bandwidth channels (nfchans) |
107 |
int channels; ///< total number of channels (nchans) |
108 |
int lfe_on; ///< indicates if there is an LFE channel (lfeon) |
109 |
int lfe_channel; ///< channel index of the LFE channel |
110 |
int channel_mode; ///< channel mode (acmod) |
111 |
const uint8_t *channel_map; ///< channel map used to reorder channels |
112 |
|
113 |
int bandwidth_code[AC3_MAX_CHANNELS]; ///< bandwidth code (0 to 60) (chbwcod) |
114 |
int nb_coefs[AC3_MAX_CHANNELS];
|
115 |
|
116 |
/* bitrate allocation control */
|
117 |
int slow_gain_code; ///< slow gain code (sgaincod) |
118 |
int slow_decay_code; ///< slow decay code (sdcycod) |
119 |
int fast_decay_code; ///< fast decay code (fdcycod) |
120 |
int db_per_bit_code; ///< dB/bit code (dbpbcod) |
121 |
int floor_code; ///< floor code (floorcod) |
122 |
AC3BitAllocParameters bit_alloc; ///< bit allocation parameters
|
123 |
int coarse_snr_offset; ///< coarse SNR offsets (csnroffst) |
124 |
int fast_gain_code[AC3_MAX_CHANNELS]; ///< fast gain codes (signal-to-mask ratio) (fgaincod) |
125 |
int fine_snr_offset[AC3_MAX_CHANNELS]; ///< fine SNR offsets (fsnroffst) |
126 |
int frame_bits; ///< all frame bits except exponents and mantissas |
127 |
int exponent_bits; ///< number of bits used for exponents |
128 |
|
129 |
/* mantissa encoding */
|
130 |
int mant1_cnt, mant2_cnt, mant4_cnt; ///< mantissa counts for bap=1,2,4 |
131 |
uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr; ///< mantissa pointers for bap=1,2,4
|
132 |
|
133 |
int16_t **planar_samples; |
134 |
uint8_t *bap_buffer; |
135 |
uint8_t *bap1_buffer; |
136 |
int32_t *mdct_coef_buffer; |
137 |
uint8_t *exp_buffer; |
138 |
uint8_t *grouped_exp_buffer; |
139 |
int16_t *psd_buffer; |
140 |
int16_t *band_psd_buffer; |
141 |
int16_t *mask_buffer; |
142 |
uint16_t *qmant_buffer; |
143 |
|
144 |
DECLARE_ALIGNED(16, int16_t, windowed_samples)[AC3_WINDOW_SIZE];
|
145 |
} AC3EncodeContext; |
146 |
|
147 |
|
148 |
/** MDCT and FFT tables */
|
149 |
static int16_t costab[64]; |
150 |
static int16_t sintab[64]; |
151 |
static int16_t xcos1[128]; |
152 |
static int16_t xsin1[128]; |
153 |
|
154 |
|
155 |
/**
|
156 |
* Adjust the frame size to make the average bit rate match the target bit rate.
|
157 |
* This is only needed for 11025, 22050, and 44100 sample rates.
|
158 |
*/
|
159 |
static void adjust_frame_size(AC3EncodeContext *s) |
160 |
{ |
161 |
while (s->bits_written >= s->bit_rate && s->samples_written >= s->sample_rate) {
|
162 |
s->bits_written -= s->bit_rate; |
163 |
s->samples_written -= s->sample_rate; |
164 |
} |
165 |
s->frame_size = s->frame_size_min + |
166 |
2 * (s->bits_written * s->sample_rate < s->samples_written * s->bit_rate);
|
167 |
s->bits_written += s->frame_size * 8;
|
168 |
s->samples_written += AC3_FRAME_SIZE; |
169 |
} |
170 |
|
171 |
|
172 |
/**
|
173 |
* Deinterleave input samples.
|
174 |
* Channels are reordered from FFmpeg's default order to AC-3 order.
|
175 |
*/
|
176 |
static void deinterleave_input_samples(AC3EncodeContext *s, |
177 |
const int16_t *samples)
|
178 |
{ |
179 |
int ch, i;
|
180 |
|
181 |
/* deinterleave and remap input samples */
|
182 |
for (ch = 0; ch < s->channels; ch++) { |
183 |
const int16_t *sptr;
|
184 |
int sinc;
|
185 |
|
186 |
/* copy last 256 samples of previous frame to the start of the current frame */
|
187 |
memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_FRAME_SIZE],
|
188 |
AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0])); |
189 |
|
190 |
/* deinterleave */
|
191 |
sinc = s->channels; |
192 |
sptr = samples + s->channel_map[ch]; |
193 |
for (i = AC3_BLOCK_SIZE; i < AC3_FRAME_SIZE+AC3_BLOCK_SIZE; i++) {
|
194 |
s->planar_samples[ch][i] = *sptr; |
195 |
sptr += sinc; |
196 |
} |
197 |
} |
198 |
} |
199 |
|
200 |
|
201 |
/**
|
202 |
* Finalize MDCT and free allocated memory.
|
203 |
*/
|
204 |
static av_cold void mdct_end(AC3MDCTContext *mdct) |
205 |
{ |
206 |
av_freep(&mdct->rot_tmp); |
207 |
av_freep(&mdct->cplx_tmp); |
208 |
} |
209 |
|
210 |
|
211 |
|
212 |
/**
|
213 |
* Initialize FFT tables.
|
214 |
* @param ln log2(FFT size)
|
215 |
*/
|
216 |
static av_cold void fft_init(int ln) |
217 |
{ |
218 |
int i, n, n2;
|
219 |
float alpha;
|
220 |
|
221 |
n = 1 << ln;
|
222 |
n2 = n >> 1;
|
223 |
|
224 |
for (i = 0; i < n2; i++) { |
225 |
alpha = 2.0 * M_PI * i / n; |
226 |
costab[i] = FIX15(cos(alpha)); |
227 |
sintab[i] = FIX15(sin(alpha)); |
228 |
} |
229 |
} |
230 |
|
231 |
|
232 |
/**
|
233 |
* Initialize MDCT tables.
|
234 |
* @param nbits log2(MDCT size)
|
235 |
*/
|
236 |
static av_cold int mdct_init(AC3MDCTContext *mdct, int nbits) |
237 |
{ |
238 |
int i, n, n4;
|
239 |
|
240 |
n = 1 << nbits;
|
241 |
n4 = n >> 2;
|
242 |
|
243 |
fft_init(nbits - 2);
|
244 |
|
245 |
FF_ALLOC_OR_GOTO(mdct->avctx, mdct->rot_tmp, n * sizeof(*mdct->rot_tmp),
|
246 |
mdct_alloc_fail); |
247 |
FF_ALLOC_OR_GOTO(mdct->avctx, mdct->cplx_tmp, n4 * sizeof(*mdct->cplx_tmp),
|
248 |
mdct_alloc_fail); |
249 |
|
250 |
for (i = 0; i < n4; i++) { |
251 |
float alpha = 2.0 * M_PI * (i + 1.0 / 8.0) / n; |
252 |
xcos1[i] = FIX15(-cos(alpha)); |
253 |
xsin1[i] = FIX15(-sin(alpha)); |
254 |
} |
255 |
|
256 |
return 0; |
257 |
mdct_alloc_fail:
|
258 |
return AVERROR(ENOMEM);
|
259 |
} |
260 |
|
261 |
|
262 |
/** Butterfly op */
|
263 |
#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
|
264 |
{ \ |
265 |
int ax, ay, bx, by; \
|
266 |
bx = pre1; \ |
267 |
by = pim1; \ |
268 |
ax = qre1; \ |
269 |
ay = qim1; \ |
270 |
pre = (bx + ax) >> 1; \
|
271 |
pim = (by + ay) >> 1; \
|
272 |
qre = (bx - ax) >> 1; \
|
273 |
qim = (by - ay) >> 1; \
|
274 |
} |
275 |
|
276 |
|
277 |
/** Complex multiply */
|
278 |
#define CMUL(pre, pim, are, aim, bre, bim) \
|
279 |
{ \ |
280 |
pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15; \
|
281 |
pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15; \
|
282 |
} |
283 |
|
284 |
|
285 |
/**
|
286 |
* Calculate a 2^n point complex FFT on 2^ln points.
|
287 |
* @param z complex input/output samples
|
288 |
* @param ln log2(FFT size)
|
289 |
*/
|
290 |
static void fft(IComplex *z, int ln) |
291 |
{ |
292 |
int j, l, np, np2;
|
293 |
int nblocks, nloops;
|
294 |
register IComplex *p,*q;
|
295 |
int tmp_re, tmp_im;
|
296 |
|
297 |
np = 1 << ln;
|
298 |
|
299 |
/* reverse */
|
300 |
for (j = 0; j < np; j++) { |
301 |
int k = av_reverse[j] >> (8 - ln); |
302 |
if (k < j)
|
303 |
FFSWAP(IComplex, z[k], z[j]); |
304 |
} |
305 |
|
306 |
/* pass 0 */
|
307 |
|
308 |
p = &z[0];
|
309 |
j = np >> 1;
|
310 |
do {
|
311 |
BF(p[0].re, p[0].im, p[1].re, p[1].im, |
312 |
p[0].re, p[0].im, p[1].re, p[1].im); |
313 |
p += 2;
|
314 |
} while (--j);
|
315 |
|
316 |
/* pass 1 */
|
317 |
|
318 |
p = &z[0];
|
319 |
j = np >> 2;
|
320 |
do {
|
321 |
BF(p[0].re, p[0].im, p[2].re, p[2].im, |
322 |
p[0].re, p[0].im, p[2].re, p[2].im); |
323 |
BF(p[1].re, p[1].im, p[3].re, p[3].im, |
324 |
p[1].re, p[1].im, p[3].im, -p[3].re); |
325 |
p+=4;
|
326 |
} while (--j);
|
327 |
|
328 |
/* pass 2 .. ln-1 */
|
329 |
|
330 |
nblocks = np >> 3;
|
331 |
nloops = 1 << 2; |
332 |
np2 = np >> 1;
|
333 |
do {
|
334 |
p = z; |
335 |
q = z + nloops; |
336 |
for (j = 0; j < nblocks; j++) { |
337 |
BF(p->re, p->im, q->re, q->im, |
338 |
p->re, p->im, q->re, q->im); |
339 |
p++; |
340 |
q++; |
341 |
for(l = nblocks; l < np2; l += nblocks) {
|
342 |
CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im); |
343 |
BF(p->re, p->im, q->re, q->im, |
344 |
p->re, p->im, tmp_re, tmp_im); |
345 |
p++; |
346 |
q++; |
347 |
} |
348 |
p += nloops; |
349 |
q += nloops; |
350 |
} |
351 |
nblocks = nblocks >> 1;
|
352 |
nloops = nloops << 1;
|
353 |
} while (nblocks);
|
354 |
} |
355 |
|
356 |
|
357 |
/**
|
358 |
* Calculate a 512-point MDCT
|
359 |
* @param out 256 output frequency coefficients
|
360 |
* @param in 512 windowed input audio samples
|
361 |
*/
|
362 |
static void mdct512(AC3MDCTContext *mdct, int32_t *out, int16_t *in) |
363 |
{ |
364 |
int i, re, im;
|
365 |
int16_t *rot = mdct->rot_tmp; |
366 |
IComplex *x = mdct->cplx_tmp; |
367 |
|
368 |
/* shift to simplify computations */
|
369 |
for (i = 0; i < MDCT_SAMPLES/4; i++) |
370 |
rot[i] = -in[i + 3*MDCT_SAMPLES/4]; |
371 |
memcpy(&rot[MDCT_SAMPLES/4], &in[0], 3*MDCT_SAMPLES/4*sizeof(*in)); |
372 |
|
373 |
/* pre rotation */
|
374 |
for (i = 0; i < MDCT_SAMPLES/4; i++) { |
375 |
re = ((int)rot[ 2*i] - (int)rot[MDCT_SAMPLES -1-2*i]) >> 1; |
376 |
im = -((int)rot[MDCT_SAMPLES/2+2*i] - (int)rot[MDCT_SAMPLES/2-1-2*i]) >> 1; |
377 |
CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]); |
378 |
} |
379 |
|
380 |
fft(x, MDCT_NBITS - 2);
|
381 |
|
382 |
/* post rotation */
|
383 |
for (i = 0; i < MDCT_SAMPLES/4; i++) { |
384 |
re = x[i].re; |
385 |
im = x[i].im; |
386 |
CMUL(out[MDCT_SAMPLES/2-1-2*i], out[2*i], re, im, xsin1[i], xcos1[i]); |
387 |
} |
388 |
} |
389 |
|
390 |
|
391 |
/**
|
392 |
* Apply KBD window to input samples prior to MDCT.
|
393 |
*/
|
394 |
static void apply_window(int16_t *output, const int16_t *input, |
395 |
const int16_t *window, int n) |
396 |
{ |
397 |
int i;
|
398 |
int n2 = n >> 1; |
399 |
|
400 |
for (i = 0; i < n2; i++) { |
401 |
output[i] = MUL16(input[i], window[i]) >> 15;
|
402 |
output[n-i-1] = MUL16(input[n-i-1], window[i]) >> 15; |
403 |
} |
404 |
} |
405 |
|
406 |
|
407 |
/**
|
408 |
* Calculate the log2() of the maximum absolute value in an array.
|
409 |
* @param tab input array
|
410 |
* @param n number of values in the array
|
411 |
* @return log2(max(abs(tab[])))
|
412 |
*/
|
413 |
static int log2_tab(int16_t *tab, int n) |
414 |
{ |
415 |
int i, v;
|
416 |
|
417 |
v = 0;
|
418 |
for (i = 0; i < n; i++) |
419 |
v |= abs(tab[i]); |
420 |
|
421 |
return av_log2(v);
|
422 |
} |
423 |
|
424 |
|
425 |
/**
|
426 |
* Left-shift each value in an array by a specified amount.
|
427 |
* @param tab input array
|
428 |
* @param n number of values in the array
|
429 |
* @param lshift left shift amount. a negative value means right shift.
|
430 |
*/
|
431 |
static void lshift_tab(int16_t *tab, int n, int lshift) |
432 |
{ |
433 |
int i;
|
434 |
|
435 |
if (lshift > 0) { |
436 |
for (i = 0; i < n; i++) |
437 |
tab[i] <<= lshift; |
438 |
} else if (lshift < 0) { |
439 |
lshift = -lshift; |
440 |
for (i = 0; i < n; i++) |
441 |
tab[i] >>= lshift; |
442 |
} |
443 |
} |
444 |
|
445 |
|
446 |
/**
|
447 |
* Normalize the input samples to use the maximum available precision.
|
448 |
* This assumes signed 16-bit input samples. Exponents are reduced by 9 to
|
449 |
* match the 24-bit internal precision for MDCT coefficients.
|
450 |
*
|
451 |
* @return exponent shift
|
452 |
*/
|
453 |
static int normalize_samples(AC3EncodeContext *s) |
454 |
{ |
455 |
int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE); |
456 |
v = FFMAX(0, v);
|
457 |
lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v); |
458 |
return v - 9; |
459 |
} |
460 |
|
461 |
|
462 |
/**
|
463 |
* Apply the MDCT to input samples to generate frequency coefficients.
|
464 |
* This applies the KBD window and normalizes the input to reduce precision
|
465 |
* loss due to fixed-point calculations.
|
466 |
*/
|
467 |
static void apply_mdct(AC3EncodeContext *s) |
468 |
{ |
469 |
int blk, ch;
|
470 |
|
471 |
for (ch = 0; ch < s->channels; ch++) { |
472 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
473 |
AC3Block *block = &s->blocks[blk]; |
474 |
const int16_t *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
|
475 |
|
476 |
apply_window(s->windowed_samples, input_samples, ff_ac3_window, AC3_WINDOW_SIZE); |
477 |
|
478 |
block->exp_shift[ch] = normalize_samples(s); |
479 |
|
480 |
mdct512(&s->mdct, block->mdct_coef[ch], s->windowed_samples); |
481 |
} |
482 |
} |
483 |
} |
484 |
|
485 |
|
486 |
/**
|
487 |
* Extract exponents from the MDCT coefficients.
|
488 |
* This takes into account the normalization that was done to the input samples
|
489 |
* by adjusting the exponents by the exponent shift values.
|
490 |
*/
|
491 |
static void extract_exponents(AC3EncodeContext *s) |
492 |
{ |
493 |
int blk, ch, i;
|
494 |
|
495 |
for (ch = 0; ch < s->channels; ch++) { |
496 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
497 |
AC3Block *block = &s->blocks[blk]; |
498 |
for (i = 0; i < AC3_MAX_COEFS; i++) { |
499 |
int e;
|
500 |
int v = abs(block->mdct_coef[ch][i]);
|
501 |
if (v == 0) |
502 |
e = 24;
|
503 |
else {
|
504 |
e = 23 - av_log2(v) + block->exp_shift[ch];
|
505 |
if (e >= 24) { |
506 |
e = 24;
|
507 |
block->mdct_coef[ch][i] = 0;
|
508 |
} |
509 |
} |
510 |
block->exp[ch][i] = e; |
511 |
} |
512 |
} |
513 |
} |
514 |
} |
515 |
|
516 |
|
517 |
/**
|
518 |
* Exponent Difference Threshold.
|
519 |
* New exponents are sent if their SAD exceed this number.
|
520 |
*/
|
521 |
#define EXP_DIFF_THRESHOLD 1000 |
522 |
|
523 |
|
524 |
/**
|
525 |
* Calculate exponent strategies for all blocks in a single channel.
|
526 |
*/
|
527 |
static void compute_exp_strategy_ch(AC3EncodeContext *s, uint8_t *exp_strategy, uint8_t **exp) |
528 |
{ |
529 |
int blk, blk1;
|
530 |
int exp_diff;
|
531 |
|
532 |
/* estimate if the exponent variation & decide if they should be
|
533 |
reused in the next frame */
|
534 |
exp_strategy[0] = EXP_NEW;
|
535 |
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) { |
536 |
exp_diff = s->dsp.sad[0](NULL, exp[blk], exp[blk-1], 16, 16); |
537 |
if (exp_diff > EXP_DIFF_THRESHOLD)
|
538 |
exp_strategy[blk] = EXP_NEW; |
539 |
else
|
540 |
exp_strategy[blk] = EXP_REUSE; |
541 |
} |
542 |
|
543 |
/* now select the encoding strategy type : if exponents are often
|
544 |
recoded, we use a coarse encoding */
|
545 |
blk = 0;
|
546 |
while (blk < AC3_MAX_BLOCKS) {
|
547 |
blk1 = blk + 1;
|
548 |
while (blk1 < AC3_MAX_BLOCKS && exp_strategy[blk1] == EXP_REUSE)
|
549 |
blk1++; |
550 |
switch (blk1 - blk) {
|
551 |
case 1: exp_strategy[blk] = EXP_D45; break; |
552 |
case 2: |
553 |
case 3: exp_strategy[blk] = EXP_D25; break; |
554 |
default: exp_strategy[blk] = EXP_D15; break; |
555 |
} |
556 |
blk = blk1; |
557 |
} |
558 |
} |
559 |
|
560 |
|
561 |
/**
|
562 |
* Calculate exponent strategies for all channels.
|
563 |
* Array arrangement is reversed to simplify the per-channel calculation.
|
564 |
*/
|
565 |
static void compute_exp_strategy(AC3EncodeContext *s) |
566 |
{ |
567 |
uint8_t *exp1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; |
568 |
uint8_t exp_str1[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; |
569 |
int ch, blk;
|
570 |
|
571 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
572 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
573 |
exp1[ch][blk] = s->blocks[blk].exp[ch]; |
574 |
exp_str1[ch][blk] = s->blocks[blk].exp_strategy[ch]; |
575 |
} |
576 |
|
577 |
compute_exp_strategy_ch(s, exp_str1[ch], exp1[ch]); |
578 |
|
579 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) |
580 |
s->blocks[blk].exp_strategy[ch] = exp_str1[ch][blk]; |
581 |
} |
582 |
if (s->lfe_on) {
|
583 |
ch = s->lfe_channel; |
584 |
s->blocks[0].exp_strategy[ch] = EXP_D15;
|
585 |
for (blk = 1; blk < AC3_MAX_BLOCKS; blk++) |
586 |
s->blocks[blk].exp_strategy[ch] = EXP_REUSE; |
587 |
} |
588 |
} |
589 |
|
590 |
|
591 |
/**
|
592 |
* Set each encoded exponent in a block to the minimum of itself and the
|
593 |
* exponent in the same frequency bin of a following block.
|
594 |
* exp[i] = min(exp[i], exp1[i]
|
595 |
*/
|
596 |
static void exponent_min(uint8_t *exp, uint8_t *exp1, int n) |
597 |
{ |
598 |
int i;
|
599 |
for (i = 0; i < n; i++) { |
600 |
if (exp1[i] < exp[i])
|
601 |
exp[i] = exp1[i]; |
602 |
} |
603 |
} |
604 |
|
605 |
|
606 |
/**
|
607 |
* Update the exponents so that they are the ones the decoder will decode.
|
608 |
*/
|
609 |
static void encode_exponents_blk_ch(uint8_t *exp, |
610 |
int nb_exps, int exp_strategy, |
611 |
uint8_t *num_exp_groups) |
612 |
{ |
613 |
int group_size, nb_groups, i, k;
|
614 |
|
615 |
group_size = exp_strategy + (exp_strategy == EXP_D45); |
616 |
*num_exp_groups = (nb_exps + (group_size * 3) - 4) / (3 * group_size); |
617 |
nb_groups = *num_exp_groups * 3;
|
618 |
|
619 |
/* for each group, compute the minimum exponent */
|
620 |
switch(exp_strategy) {
|
621 |
case EXP_D25:
|
622 |
for (i = 1, k = 1; i <= nb_groups; i++) { |
623 |
uint8_t exp_min = exp[k]; |
624 |
if (exp[k+1] < exp_min) |
625 |
exp_min = exp[k+1];
|
626 |
exp[i] = exp_min; |
627 |
k += 2;
|
628 |
} |
629 |
break;
|
630 |
case EXP_D45:
|
631 |
for (i = 1, k = 1; i <= nb_groups; i++) { |
632 |
uint8_t exp_min = exp[k]; |
633 |
if (exp[k+1] < exp_min) |
634 |
exp_min = exp[k+1];
|
635 |
if (exp[k+2] < exp_min) |
636 |
exp_min = exp[k+2];
|
637 |
if (exp[k+3] < exp_min) |
638 |
exp_min = exp[k+3];
|
639 |
exp[i] = exp_min; |
640 |
k += 4;
|
641 |
} |
642 |
break;
|
643 |
} |
644 |
|
645 |
/* constraint for DC exponent */
|
646 |
if (exp[0] > 15) |
647 |
exp[0] = 15; |
648 |
|
649 |
/* decrease the delta between each groups to within 2 so that they can be
|
650 |
differentially encoded */
|
651 |
for (i = 1; i <= nb_groups; i++) |
652 |
exp[i] = FFMIN(exp[i], exp[i-1] + 2); |
653 |
i--; |
654 |
while (--i >= 0) |
655 |
exp[i] = FFMIN(exp[i], exp[i+1] + 2); |
656 |
|
657 |
/* now we have the exponent values the decoder will see */
|
658 |
switch (exp_strategy) {
|
659 |
case EXP_D25:
|
660 |
for (i = nb_groups, k = nb_groups * 2; i > 0; i--) { |
661 |
uint8_t exp1 = exp[i]; |
662 |
exp[k--] = exp1; |
663 |
exp[k--] = exp1; |
664 |
} |
665 |
break;
|
666 |
case EXP_D45:
|
667 |
for (i = nb_groups, k = nb_groups * 4; i > 0; i--) { |
668 |
exp[k] = exp[k-1] = exp[k-2] = exp[k-3] = exp[i]; |
669 |
k -= 4;
|
670 |
} |
671 |
break;
|
672 |
} |
673 |
} |
674 |
|
675 |
|
676 |
/**
|
677 |
* Encode exponents from original extracted form to what the decoder will see.
|
678 |
* This copies and groups exponents based on exponent strategy and reduces
|
679 |
* deltas between adjacent exponent groups so that they can be differentially
|
680 |
* encoded.
|
681 |
*/
|
682 |
static void encode_exponents(AC3EncodeContext *s) |
683 |
{ |
684 |
int blk, blk1, blk2, ch;
|
685 |
AC3Block *block, *block1, *block2; |
686 |
|
687 |
for (ch = 0; ch < s->channels; ch++) { |
688 |
blk = 0;
|
689 |
block = &s->blocks[0];
|
690 |
while (blk < AC3_MAX_BLOCKS) {
|
691 |
blk1 = blk + 1;
|
692 |
block1 = block + 1;
|
693 |
/* for the EXP_REUSE case we select the min of the exponents */
|
694 |
while (blk1 < AC3_MAX_BLOCKS && block1->exp_strategy[ch] == EXP_REUSE) {
|
695 |
exponent_min(block->exp[ch], block1->exp[ch], s->nb_coefs[ch]); |
696 |
blk1++; |
697 |
block1++; |
698 |
} |
699 |
encode_exponents_blk_ch(block->exp[ch], s->nb_coefs[ch], |
700 |
block->exp_strategy[ch], |
701 |
&block->num_exp_groups[ch]); |
702 |
/* copy encoded exponents for reuse case */
|
703 |
block2 = block + 1;
|
704 |
for (blk2 = blk+1; blk2 < blk1; blk2++, block2++) { |
705 |
memcpy(block2->exp[ch], block->exp[ch], |
706 |
s->nb_coefs[ch] * sizeof(uint8_t));
|
707 |
} |
708 |
blk = blk1; |
709 |
block = block1; |
710 |
} |
711 |
} |
712 |
} |
713 |
|
714 |
|
715 |
/**
|
716 |
* Group exponents.
|
717 |
* 3 delta-encoded exponents are in each 7-bit group. The number of groups
|
718 |
* varies depending on exponent strategy and bandwidth.
|
719 |
*/
|
720 |
static void group_exponents(AC3EncodeContext *s) |
721 |
{ |
722 |
int blk, ch, i;
|
723 |
int group_size, bit_count;
|
724 |
uint8_t *p; |
725 |
int delta0, delta1, delta2;
|
726 |
int exp0, exp1;
|
727 |
|
728 |
bit_count = 0;
|
729 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
730 |
AC3Block *block = &s->blocks[blk]; |
731 |
for (ch = 0; ch < s->channels; ch++) { |
732 |
if (block->exp_strategy[ch] == EXP_REUSE) {
|
733 |
block->num_exp_groups[ch] = 0;
|
734 |
continue;
|
735 |
} |
736 |
group_size = block->exp_strategy[ch] + (block->exp_strategy[ch] == EXP_D45); |
737 |
bit_count += 4 + (block->num_exp_groups[ch] * 7); |
738 |
p = block->exp[ch]; |
739 |
|
740 |
/* DC exponent */
|
741 |
exp1 = *p++; |
742 |
block->grouped_exp[ch][0] = exp1;
|
743 |
|
744 |
/* remaining exponents are delta encoded */
|
745 |
for (i = 1; i <= block->num_exp_groups[ch]; i++) { |
746 |
/* merge three delta in one code */
|
747 |
exp0 = exp1; |
748 |
exp1 = p[0];
|
749 |
p += group_size; |
750 |
delta0 = exp1 - exp0 + 2;
|
751 |
|
752 |
exp0 = exp1; |
753 |
exp1 = p[0];
|
754 |
p += group_size; |
755 |
delta1 = exp1 - exp0 + 2;
|
756 |
|
757 |
exp0 = exp1; |
758 |
exp1 = p[0];
|
759 |
p += group_size; |
760 |
delta2 = exp1 - exp0 + 2;
|
761 |
|
762 |
block->grouped_exp[ch][i] = ((delta0 * 5 + delta1) * 5) + delta2; |
763 |
} |
764 |
} |
765 |
} |
766 |
|
767 |
s->exponent_bits = bit_count; |
768 |
} |
769 |
|
770 |
|
771 |
/**
|
772 |
* Calculate final exponents from the supplied MDCT coefficients and exponent shift.
|
773 |
* Extract exponents from MDCT coefficients, calculate exponent strategies,
|
774 |
* and encode final exponents.
|
775 |
*/
|
776 |
static void process_exponents(AC3EncodeContext *s) |
777 |
{ |
778 |
extract_exponents(s); |
779 |
|
780 |
compute_exp_strategy(s); |
781 |
|
782 |
encode_exponents(s); |
783 |
|
784 |
group_exponents(s); |
785 |
} |
786 |
|
787 |
|
788 |
/**
|
789 |
* Initialize bit allocation.
|
790 |
* Set default parameter codes and calculate parameter values.
|
791 |
*/
|
792 |
static void bit_alloc_init(AC3EncodeContext *s) |
793 |
{ |
794 |
int ch;
|
795 |
|
796 |
/* init default parameters */
|
797 |
s->slow_decay_code = 2;
|
798 |
s->fast_decay_code = 1;
|
799 |
s->slow_gain_code = 1;
|
800 |
s->db_per_bit_code = 2;
|
801 |
s->floor_code = 4;
|
802 |
for (ch = 0; ch < s->channels; ch++) |
803 |
s->fast_gain_code[ch] = 4;
|
804 |
|
805 |
/* initial snr offset */
|
806 |
s->coarse_snr_offset = 40;
|
807 |
|
808 |
/* compute real values */
|
809 |
/* currently none of these values change during encoding, so we can just
|
810 |
set them once at initialization */
|
811 |
s->bit_alloc.slow_decay = ff_ac3_slow_decay_tab[s->slow_decay_code] >> s->bit_alloc.sr_shift; |
812 |
s->bit_alloc.fast_decay = ff_ac3_fast_decay_tab[s->fast_decay_code] >> s->bit_alloc.sr_shift; |
813 |
s->bit_alloc.slow_gain = ff_ac3_slow_gain_tab[s->slow_gain_code]; |
814 |
s->bit_alloc.db_per_bit = ff_ac3_db_per_bit_tab[s->db_per_bit_code]; |
815 |
s->bit_alloc.floor = ff_ac3_floor_tab[s->floor_code]; |
816 |
} |
817 |
|
818 |
|
819 |
/**
|
820 |
* Count the bits used to encode the frame, minus exponents and mantissas.
|
821 |
*/
|
822 |
static void count_frame_bits(AC3EncodeContext *s) |
823 |
{ |
824 |
static const int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 }; |
825 |
int blk, ch;
|
826 |
int frame_bits;
|
827 |
|
828 |
/* header size */
|
829 |
frame_bits = 65;
|
830 |
frame_bits += frame_bits_inc[s->channel_mode]; |
831 |
|
832 |
/* audio blocks */
|
833 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
834 |
frame_bits += s->fbw_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */ |
835 |
if (s->channel_mode == AC3_CHMODE_STEREO) {
|
836 |
frame_bits++; /* rematstr */
|
837 |
if (!blk)
|
838 |
frame_bits += 4;
|
839 |
} |
840 |
frame_bits += 2 * s->fbw_channels; /* chexpstr[2] * c */ |
841 |
if (s->lfe_on)
|
842 |
frame_bits++; /* lfeexpstr */
|
843 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
844 |
if (s->blocks[blk].exp_strategy[ch] != EXP_REUSE)
|
845 |
frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */ |
846 |
} |
847 |
frame_bits++; /* baie */
|
848 |
frame_bits++; /* snr */
|
849 |
frame_bits += 2; /* delta / skip */ |
850 |
} |
851 |
frame_bits++; /* cplinu for block 0 */
|
852 |
/* bit alloc info */
|
853 |
/* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
|
854 |
/* csnroffset[6] */
|
855 |
/* (fsnoffset[4] + fgaincod[4]) * c */
|
856 |
frame_bits += 2*4 + 3 + 6 + s->channels * (4 + 3); |
857 |
|
858 |
/* auxdatae, crcrsv */
|
859 |
frame_bits += 2;
|
860 |
|
861 |
/* CRC */
|
862 |
frame_bits += 16;
|
863 |
|
864 |
s->frame_bits = frame_bits; |
865 |
} |
866 |
|
867 |
|
868 |
/**
|
869 |
* Calculate the number of bits needed to encode a set of mantissas.
|
870 |
*/
|
871 |
static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *bap, int nb_coefs) |
872 |
{ |
873 |
int bits, b, i;
|
874 |
|
875 |
bits = 0;
|
876 |
for (i = 0; i < nb_coefs; i++) { |
877 |
b = bap[i]; |
878 |
switch (b) {
|
879 |
case 0: |
880 |
/* bap=0 mantissas are not encoded */
|
881 |
break;
|
882 |
case 1: |
883 |
/* 3 mantissas in 5 bits */
|
884 |
if (s->mant1_cnt == 0) |
885 |
bits += 5;
|
886 |
if (++s->mant1_cnt == 3) |
887 |
s->mant1_cnt = 0;
|
888 |
break;
|
889 |
case 2: |
890 |
/* 3 mantissas in 7 bits */
|
891 |
if (s->mant2_cnt == 0) |
892 |
bits += 7;
|
893 |
if (++s->mant2_cnt == 3) |
894 |
s->mant2_cnt = 0;
|
895 |
break;
|
896 |
case 3: |
897 |
bits += 3;
|
898 |
break;
|
899 |
case 4: |
900 |
/* 2 mantissas in 7 bits */
|
901 |
if (s->mant4_cnt == 0) |
902 |
bits += 7;
|
903 |
if (++s->mant4_cnt == 2) |
904 |
s->mant4_cnt = 0;
|
905 |
break;
|
906 |
case 14: |
907 |
bits += 14;
|
908 |
break;
|
909 |
case 15: |
910 |
bits += 16;
|
911 |
break;
|
912 |
default:
|
913 |
bits += b - 1;
|
914 |
break;
|
915 |
} |
916 |
} |
917 |
return bits;
|
918 |
} |
919 |
|
920 |
|
921 |
/**
|
922 |
* Calculate masking curve based on the final exponents.
|
923 |
* Also calculate the power spectral densities to use in future calculations.
|
924 |
*/
|
925 |
static void bit_alloc_masking(AC3EncodeContext *s) |
926 |
{ |
927 |
int blk, ch;
|
928 |
|
929 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
930 |
AC3Block *block = &s->blocks[blk]; |
931 |
for (ch = 0; ch < s->channels; ch++) { |
932 |
if (block->exp_strategy[ch] == EXP_REUSE) {
|
933 |
AC3Block *block1 = &s->blocks[blk-1];
|
934 |
memcpy(block->psd[ch], block1->psd[ch], AC3_MAX_COEFS*sizeof(block->psd[0][0])); |
935 |
memcpy(block->mask[ch], block1->mask[ch], AC3_CRITICAL_BANDS*sizeof(block->mask[0][0])); |
936 |
} else {
|
937 |
ff_ac3_bit_alloc_calc_psd(block->exp[ch], 0,
|
938 |
s->nb_coefs[ch], |
939 |
block->psd[ch], block->band_psd[ch]); |
940 |
ff_ac3_bit_alloc_calc_mask(&s->bit_alloc, block->band_psd[ch], |
941 |
0, s->nb_coefs[ch],
|
942 |
ff_ac3_fast_gain_tab[s->fast_gain_code[ch]], |
943 |
ch == s->lfe_channel, |
944 |
DBA_NONE, 0, NULL, NULL, NULL, |
945 |
block->mask[ch]); |
946 |
} |
947 |
} |
948 |
} |
949 |
} |
950 |
|
951 |
|
952 |
/**
|
953 |
* Ensure that bap for each block and channel point to the current bap_buffer.
|
954 |
* They may have been switched during the bit allocation search.
|
955 |
*/
|
956 |
static void reset_block_bap(AC3EncodeContext *s) |
957 |
{ |
958 |
int blk, ch;
|
959 |
if (s->blocks[0].bap[0] == s->bap_buffer) |
960 |
return;
|
961 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
962 |
for (ch = 0; ch < s->channels; ch++) { |
963 |
s->blocks[blk].bap[ch] = &s->bap_buffer[AC3_MAX_COEFS * (blk * s->channels + ch)]; |
964 |
} |
965 |
} |
966 |
} |
967 |
|
968 |
|
969 |
/**
|
970 |
* Run the bit allocation with a given SNR offset.
|
971 |
* This calculates the bit allocation pointers that will be used to determine
|
972 |
* the quantization of each mantissa.
|
973 |
* @return the number of bits needed for mantissas if the given SNR offset is
|
974 |
* is used.
|
975 |
*/
|
976 |
static int bit_alloc(AC3EncodeContext *s, |
977 |
int snr_offset)
|
978 |
{ |
979 |
int blk, ch;
|
980 |
int mantissa_bits;
|
981 |
|
982 |
snr_offset = (snr_offset - 240) << 2; |
983 |
|
984 |
reset_block_bap(s); |
985 |
mantissa_bits = 0;
|
986 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
987 |
AC3Block *block = &s->blocks[blk]; |
988 |
s->mant1_cnt = 0;
|
989 |
s->mant2_cnt = 0;
|
990 |
s->mant4_cnt = 0;
|
991 |
for (ch = 0; ch < s->channels; ch++) { |
992 |
ff_ac3_bit_alloc_calc_bap(block->mask[ch], block->psd[ch], 0,
|
993 |
s->nb_coefs[ch], snr_offset, |
994 |
s->bit_alloc.floor, ff_ac3_bap_tab, |
995 |
block->bap[ch]); |
996 |
mantissa_bits += compute_mantissa_size(s, block->bap[ch], s->nb_coefs[ch]); |
997 |
} |
998 |
} |
999 |
return mantissa_bits;
|
1000 |
} |
1001 |
|
1002 |
|
1003 |
/**
|
1004 |
* Constant bitrate bit allocation search.
|
1005 |
* Find the largest SNR offset that will allow data to fit in the frame.
|
1006 |
*/
|
1007 |
static int cbr_bit_allocation(AC3EncodeContext *s) |
1008 |
{ |
1009 |
int ch;
|
1010 |
int bits_left;
|
1011 |
int snr_offset;
|
1012 |
|
1013 |
bits_left = 8 * s->frame_size - (s->frame_bits + s->exponent_bits);
|
1014 |
|
1015 |
snr_offset = s->coarse_snr_offset << 4;
|
1016 |
|
1017 |
while (snr_offset >= 0 && |
1018 |
bit_alloc(s, snr_offset) > bits_left) { |
1019 |
snr_offset -= 64;
|
1020 |
} |
1021 |
if (snr_offset < 0) |
1022 |
return AVERROR(EINVAL);
|
1023 |
|
1024 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
1025 |
while (snr_offset + 64 <= 1023 && |
1026 |
bit_alloc(s, snr_offset + 64) <= bits_left) {
|
1027 |
snr_offset += 64;
|
1028 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
1029 |
} |
1030 |
while (snr_offset + 16 <= 1023 && |
1031 |
bit_alloc(s, snr_offset + 16) <= bits_left) {
|
1032 |
snr_offset += 16;
|
1033 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
1034 |
} |
1035 |
while (snr_offset + 4 <= 1023 && |
1036 |
bit_alloc(s, snr_offset + 4) <= bits_left) {
|
1037 |
snr_offset += 4;
|
1038 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
1039 |
} |
1040 |
while (snr_offset + 1 <= 1023 && |
1041 |
bit_alloc(s, snr_offset + 1) <= bits_left) {
|
1042 |
snr_offset++; |
1043 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
1044 |
} |
1045 |
FFSWAP(uint8_t *, s->bap_buffer, s->bap1_buffer); |
1046 |
reset_block_bap(s); |
1047 |
|
1048 |
s->coarse_snr_offset = snr_offset >> 4;
|
1049 |
for (ch = 0; ch < s->channels; ch++) |
1050 |
s->fine_snr_offset[ch] = snr_offset & 0xF;
|
1051 |
|
1052 |
return 0; |
1053 |
} |
1054 |
|
1055 |
|
1056 |
/**
|
1057 |
* Perform bit allocation search.
|
1058 |
* Finds the SNR offset value that maximizes quality and fits in the specified
|
1059 |
* frame size. Output is the SNR offset and a set of bit allocation pointers
|
1060 |
* used to quantize the mantissas.
|
1061 |
*/
|
1062 |
static int compute_bit_allocation(AC3EncodeContext *s) |
1063 |
{ |
1064 |
count_frame_bits(s); |
1065 |
|
1066 |
bit_alloc_masking(s); |
1067 |
|
1068 |
return cbr_bit_allocation(s);
|
1069 |
} |
1070 |
|
1071 |
|
1072 |
/**
|
1073 |
* Symmetric quantization on 'levels' levels.
|
1074 |
*/
|
1075 |
static inline int sym_quant(int c, int e, int levels) |
1076 |
{ |
1077 |
int v;
|
1078 |
|
1079 |
if (c >= 0) { |
1080 |
v = (levels * (c << e)) >> 24;
|
1081 |
v = (v + 1) >> 1; |
1082 |
v = (levels >> 1) + v;
|
1083 |
} else {
|
1084 |
v = (levels * ((-c) << e)) >> 24;
|
1085 |
v = (v + 1) >> 1; |
1086 |
v = (levels >> 1) - v;
|
1087 |
} |
1088 |
assert(v >= 0 && v < levels);
|
1089 |
return v;
|
1090 |
} |
1091 |
|
1092 |
|
1093 |
/**
|
1094 |
* Asymmetric quantization on 2^qbits levels.
|
1095 |
*/
|
1096 |
static inline int asym_quant(int c, int e, int qbits) |
1097 |
{ |
1098 |
int lshift, m, v;
|
1099 |
|
1100 |
lshift = e + qbits - 24;
|
1101 |
if (lshift >= 0) |
1102 |
v = c << lshift; |
1103 |
else
|
1104 |
v = c >> (-lshift); |
1105 |
/* rounding */
|
1106 |
v = (v + 1) >> 1; |
1107 |
m = (1 << (qbits-1)); |
1108 |
if (v >= m)
|
1109 |
v = m - 1;
|
1110 |
assert(v >= -m); |
1111 |
return v & ((1 << qbits)-1); |
1112 |
} |
1113 |
|
1114 |
|
1115 |
/**
|
1116 |
* Quantize a set of mantissas for a single channel in a single block.
|
1117 |
*/
|
1118 |
static void quantize_mantissas_blk_ch(AC3EncodeContext *s, |
1119 |
int32_t *mdct_coef, int8_t exp_shift, |
1120 |
uint8_t *exp, uint8_t *bap, |
1121 |
uint16_t *qmant, int n)
|
1122 |
{ |
1123 |
int i;
|
1124 |
|
1125 |
for (i = 0; i < n; i++) { |
1126 |
int v;
|
1127 |
int c = mdct_coef[i];
|
1128 |
int e = exp[i] - exp_shift;
|
1129 |
int b = bap[i];
|
1130 |
switch (b) {
|
1131 |
case 0: |
1132 |
v = 0;
|
1133 |
break;
|
1134 |
case 1: |
1135 |
v = sym_quant(c, e, 3);
|
1136 |
switch (s->mant1_cnt) {
|
1137 |
case 0: |
1138 |
s->qmant1_ptr = &qmant[i]; |
1139 |
v = 9 * v;
|
1140 |
s->mant1_cnt = 1;
|
1141 |
break;
|
1142 |
case 1: |
1143 |
*s->qmant1_ptr += 3 * v;
|
1144 |
s->mant1_cnt = 2;
|
1145 |
v = 128;
|
1146 |
break;
|
1147 |
default:
|
1148 |
*s->qmant1_ptr += v; |
1149 |
s->mant1_cnt = 0;
|
1150 |
v = 128;
|
1151 |
break;
|
1152 |
} |
1153 |
break;
|
1154 |
case 2: |
1155 |
v = sym_quant(c, e, 5);
|
1156 |
switch (s->mant2_cnt) {
|
1157 |
case 0: |
1158 |
s->qmant2_ptr = &qmant[i]; |
1159 |
v = 25 * v;
|
1160 |
s->mant2_cnt = 1;
|
1161 |
break;
|
1162 |
case 1: |
1163 |
*s->qmant2_ptr += 5 * v;
|
1164 |
s->mant2_cnt = 2;
|
1165 |
v = 128;
|
1166 |
break;
|
1167 |
default:
|
1168 |
*s->qmant2_ptr += v; |
1169 |
s->mant2_cnt = 0;
|
1170 |
v = 128;
|
1171 |
break;
|
1172 |
} |
1173 |
break;
|
1174 |
case 3: |
1175 |
v = sym_quant(c, e, 7);
|
1176 |
break;
|
1177 |
case 4: |
1178 |
v = sym_quant(c, e, 11);
|
1179 |
switch (s->mant4_cnt) {
|
1180 |
case 0: |
1181 |
s->qmant4_ptr = &qmant[i]; |
1182 |
v = 11 * v;
|
1183 |
s->mant4_cnt = 1;
|
1184 |
break;
|
1185 |
default:
|
1186 |
*s->qmant4_ptr += v; |
1187 |
s->mant4_cnt = 0;
|
1188 |
v = 128;
|
1189 |
break;
|
1190 |
} |
1191 |
break;
|
1192 |
case 5: |
1193 |
v = sym_quant(c, e, 15);
|
1194 |
break;
|
1195 |
case 14: |
1196 |
v = asym_quant(c, e, 14);
|
1197 |
break;
|
1198 |
case 15: |
1199 |
v = asym_quant(c, e, 16);
|
1200 |
break;
|
1201 |
default:
|
1202 |
v = asym_quant(c, e, b - 1);
|
1203 |
break;
|
1204 |
} |
1205 |
qmant[i] = v; |
1206 |
} |
1207 |
} |
1208 |
|
1209 |
|
1210 |
/**
|
1211 |
* Quantize mantissas using coefficients, exponents, and bit allocation pointers.
|
1212 |
*/
|
1213 |
static void quantize_mantissas(AC3EncodeContext *s) |
1214 |
{ |
1215 |
int blk, ch;
|
1216 |
|
1217 |
|
1218 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
1219 |
AC3Block *block = &s->blocks[blk]; |
1220 |
s->mant1_cnt = s->mant2_cnt = s->mant4_cnt = 0;
|
1221 |
s->qmant1_ptr = s->qmant2_ptr = s->qmant4_ptr = NULL;
|
1222 |
|
1223 |
for (ch = 0; ch < s->channels; ch++) { |
1224 |
quantize_mantissas_blk_ch(s, block->mdct_coef[ch], block->exp_shift[ch], |
1225 |
block->exp[ch], block->bap[ch], |
1226 |
block->qmant[ch], s->nb_coefs[ch]); |
1227 |
} |
1228 |
} |
1229 |
} |
1230 |
|
1231 |
|
1232 |
/**
|
1233 |
* Write the AC-3 frame header to the output bitstream.
|
1234 |
*/
|
1235 |
static void output_frame_header(AC3EncodeContext *s) |
1236 |
{ |
1237 |
put_bits(&s->pb, 16, 0x0b77); /* frame header */ |
1238 |
put_bits(&s->pb, 16, 0); /* crc1: will be filled later */ |
1239 |
put_bits(&s->pb, 2, s->bit_alloc.sr_code);
|
1240 |
put_bits(&s->pb, 6, s->frame_size_code + (s->frame_size - s->frame_size_min) / 2); |
1241 |
put_bits(&s->pb, 5, s->bitstream_id);
|
1242 |
put_bits(&s->pb, 3, s->bitstream_mode);
|
1243 |
put_bits(&s->pb, 3, s->channel_mode);
|
1244 |
if ((s->channel_mode & 0x01) && s->channel_mode != AC3_CHMODE_MONO) |
1245 |
put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */ |
1246 |
if (s->channel_mode & 0x04) |
1247 |
put_bits(&s->pb, 2, 1); /* XXX -6 dB */ |
1248 |
if (s->channel_mode == AC3_CHMODE_STEREO)
|
1249 |
put_bits(&s->pb, 2, 0); /* surround not indicated */ |
1250 |
put_bits(&s->pb, 1, s->lfe_on); /* LFE */ |
1251 |
put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */ |
1252 |
put_bits(&s->pb, 1, 0); /* no compression control word */ |
1253 |
put_bits(&s->pb, 1, 0); /* no lang code */ |
1254 |
put_bits(&s->pb, 1, 0); /* no audio production info */ |
1255 |
put_bits(&s->pb, 1, 0); /* no copyright */ |
1256 |
put_bits(&s->pb, 1, 1); /* original bitstream */ |
1257 |
put_bits(&s->pb, 1, 0); /* no time code 1 */ |
1258 |
put_bits(&s->pb, 1, 0); /* no time code 2 */ |
1259 |
put_bits(&s->pb, 1, 0); /* no additional bit stream info */ |
1260 |
} |
1261 |
|
1262 |
|
1263 |
/**
|
1264 |
* Write one audio block to the output bitstream.
|
1265 |
*/
|
1266 |
static void output_audio_block(AC3EncodeContext *s, |
1267 |
int block_num)
|
1268 |
{ |
1269 |
int ch, i, baie, rbnd;
|
1270 |
AC3Block *block = &s->blocks[block_num]; |
1271 |
|
1272 |
/* block switching */
|
1273 |
for (ch = 0; ch < s->fbw_channels; ch++) |
1274 |
put_bits(&s->pb, 1, 0); |
1275 |
|
1276 |
/* dither flags */
|
1277 |
for (ch = 0; ch < s->fbw_channels; ch++) |
1278 |
put_bits(&s->pb, 1, 1); |
1279 |
|
1280 |
/* dynamic range codes */
|
1281 |
put_bits(&s->pb, 1, 0); |
1282 |
|
1283 |
/* channel coupling */
|
1284 |
if (!block_num) {
|
1285 |
put_bits(&s->pb, 1, 1); /* coupling strategy present */ |
1286 |
put_bits(&s->pb, 1, 0); /* no coupling strategy */ |
1287 |
} else {
|
1288 |
put_bits(&s->pb, 1, 0); /* no new coupling strategy */ |
1289 |
} |
1290 |
|
1291 |
/* stereo rematrixing */
|
1292 |
if (s->channel_mode == AC3_CHMODE_STEREO) {
|
1293 |
if (!block_num) {
|
1294 |
/* first block must define rematrixing (rematstr) */
|
1295 |
put_bits(&s->pb, 1, 1); |
1296 |
|
1297 |
/* dummy rematrixing rematflg(1:4)=0 */
|
1298 |
for (rbnd = 0; rbnd < 4; rbnd++) |
1299 |
put_bits(&s->pb, 1, 0); |
1300 |
} else {
|
1301 |
/* no matrixing (but should be used in the future) */
|
1302 |
put_bits(&s->pb, 1, 0); |
1303 |
} |
1304 |
} |
1305 |
|
1306 |
/* exponent strategy */
|
1307 |
for (ch = 0; ch < s->fbw_channels; ch++) |
1308 |
put_bits(&s->pb, 2, block->exp_strategy[ch]);
|
1309 |
if (s->lfe_on)
|
1310 |
put_bits(&s->pb, 1, block->exp_strategy[s->lfe_channel]);
|
1311 |
|
1312 |
/* bandwidth */
|
1313 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
1314 |
if (block->exp_strategy[ch] != EXP_REUSE)
|
1315 |
put_bits(&s->pb, 6, s->bandwidth_code[ch]);
|
1316 |
} |
1317 |
|
1318 |
/* exponents */
|
1319 |
for (ch = 0; ch < s->channels; ch++) { |
1320 |
if (block->exp_strategy[ch] == EXP_REUSE)
|
1321 |
continue;
|
1322 |
|
1323 |
/* DC exponent */
|
1324 |
put_bits(&s->pb, 4, block->grouped_exp[ch][0]); |
1325 |
|
1326 |
/* exponent groups */
|
1327 |
for (i = 1; i <= block->num_exp_groups[ch]; i++) |
1328 |
put_bits(&s->pb, 7, block->grouped_exp[ch][i]);
|
1329 |
|
1330 |
/* gain range info */
|
1331 |
if (ch != s->lfe_channel)
|
1332 |
put_bits(&s->pb, 2, 0); |
1333 |
} |
1334 |
|
1335 |
/* bit allocation info */
|
1336 |
baie = (block_num == 0);
|
1337 |
put_bits(&s->pb, 1, baie);
|
1338 |
if (baie) {
|
1339 |
put_bits(&s->pb, 2, s->slow_decay_code);
|
1340 |
put_bits(&s->pb, 2, s->fast_decay_code);
|
1341 |
put_bits(&s->pb, 2, s->slow_gain_code);
|
1342 |
put_bits(&s->pb, 2, s->db_per_bit_code);
|
1343 |
put_bits(&s->pb, 3, s->floor_code);
|
1344 |
} |
1345 |
|
1346 |
/* snr offset */
|
1347 |
put_bits(&s->pb, 1, baie);
|
1348 |
if (baie) {
|
1349 |
put_bits(&s->pb, 6, s->coarse_snr_offset);
|
1350 |
for (ch = 0; ch < s->channels; ch++) { |
1351 |
put_bits(&s->pb, 4, s->fine_snr_offset[ch]);
|
1352 |
put_bits(&s->pb, 3, s->fast_gain_code[ch]);
|
1353 |
} |
1354 |
} |
1355 |
|
1356 |
put_bits(&s->pb, 1, 0); /* no delta bit allocation */ |
1357 |
put_bits(&s->pb, 1, 0); /* no data to skip */ |
1358 |
|
1359 |
/* mantissas */
|
1360 |
for (ch = 0; ch < s->channels; ch++) { |
1361 |
int b, q;
|
1362 |
for (i = 0; i < s->nb_coefs[ch]; i++) { |
1363 |
q = block->qmant[ch][i]; |
1364 |
b = block->bap[ch][i]; |
1365 |
switch (b) {
|
1366 |
case 0: break; |
1367 |
case 1: if (q != 128) put_bits(&s->pb, 5, q); break; |
1368 |
case 2: if (q != 128) put_bits(&s->pb, 7, q); break; |
1369 |
case 3: put_bits(&s->pb, 3, q); break; |
1370 |
case 4: if (q != 128) put_bits(&s->pb, 7, q); break; |
1371 |
case 14: put_bits(&s->pb, 14, q); break; |
1372 |
case 15: put_bits(&s->pb, 16, q); break; |
1373 |
default: put_bits(&s->pb, b-1, q); break; |
1374 |
} |
1375 |
} |
1376 |
} |
1377 |
} |
1378 |
|
1379 |
|
1380 |
/** CRC-16 Polynomial */
|
1381 |
#define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16)) |
1382 |
|
1383 |
|
1384 |
static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly) |
1385 |
{ |
1386 |
unsigned int c; |
1387 |
|
1388 |
c = 0;
|
1389 |
while (a) {
|
1390 |
if (a & 1) |
1391 |
c ^= b; |
1392 |
a = a >> 1;
|
1393 |
b = b << 1;
|
1394 |
if (b & (1 << 16)) |
1395 |
b ^= poly; |
1396 |
} |
1397 |
return c;
|
1398 |
} |
1399 |
|
1400 |
|
1401 |
static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly) |
1402 |
{ |
1403 |
unsigned int r; |
1404 |
r = 1;
|
1405 |
while (n) {
|
1406 |
if (n & 1) |
1407 |
r = mul_poly(r, a, poly); |
1408 |
a = mul_poly(a, a, poly); |
1409 |
n >>= 1;
|
1410 |
} |
1411 |
return r;
|
1412 |
} |
1413 |
|
1414 |
|
1415 |
/**
|
1416 |
* Fill the end of the frame with 0's and compute the two CRCs.
|
1417 |
*/
|
1418 |
static void output_frame_end(AC3EncodeContext *s) |
1419 |
{ |
1420 |
int frame_size, frame_size_58, pad_bytes, crc1, crc2, crc_inv;
|
1421 |
uint8_t *frame; |
1422 |
|
1423 |
frame_size = s->frame_size; |
1424 |
frame_size_58 = ((frame_size >> 2) + (frame_size >> 4)) << 1; |
1425 |
|
1426 |
/* pad the remainder of the frame with zeros */
|
1427 |
flush_put_bits(&s->pb); |
1428 |
frame = s->pb.buf; |
1429 |
pad_bytes = s->frame_size - (put_bits_ptr(&s->pb) - frame) - 2;
|
1430 |
assert(pad_bytes >= 0);
|
1431 |
if (pad_bytes > 0) |
1432 |
memset(put_bits_ptr(&s->pb), 0, pad_bytes);
|
1433 |
|
1434 |
/* compute crc1 */
|
1435 |
/* this is not so easy because it is at the beginning of the data... */
|
1436 |
crc1 = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
|
1437 |
frame + 4, frame_size_58 - 4)); |
1438 |
/* XXX: could precompute crc_inv */
|
1439 |
crc_inv = pow_poly((CRC16_POLY >> 1), (8 * frame_size_58) - 16, CRC16_POLY); |
1440 |
crc1 = mul_poly(crc_inv, crc1, CRC16_POLY); |
1441 |
AV_WB16(frame + 2, crc1);
|
1442 |
|
1443 |
/* compute crc2 */
|
1444 |
crc2 = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
|
1445 |
frame + frame_size_58, |
1446 |
frame_size - frame_size_58 - 2));
|
1447 |
AV_WB16(frame + frame_size - 2, crc2);
|
1448 |
} |
1449 |
|
1450 |
|
1451 |
/**
|
1452 |
* Write the frame to the output bitstream.
|
1453 |
*/
|
1454 |
static void output_frame(AC3EncodeContext *s, |
1455 |
unsigned char *frame) |
1456 |
{ |
1457 |
int blk;
|
1458 |
|
1459 |
init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE); |
1460 |
|
1461 |
output_frame_header(s); |
1462 |
|
1463 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) |
1464 |
output_audio_block(s, blk); |
1465 |
|
1466 |
output_frame_end(s); |
1467 |
} |
1468 |
|
1469 |
|
1470 |
/**
|
1471 |
* Encode a single AC-3 frame.
|
1472 |
*/
|
1473 |
static int ac3_encode_frame(AVCodecContext *avctx, |
1474 |
unsigned char *frame, int buf_size, void *data) |
1475 |
{ |
1476 |
AC3EncodeContext *s = avctx->priv_data; |
1477 |
const int16_t *samples = data;
|
1478 |
int ret;
|
1479 |
|
1480 |
if (s->bit_alloc.sr_code == 1) |
1481 |
adjust_frame_size(s); |
1482 |
|
1483 |
deinterleave_input_samples(s, samples); |
1484 |
|
1485 |
apply_mdct(s); |
1486 |
|
1487 |
process_exponents(s); |
1488 |
|
1489 |
ret = compute_bit_allocation(s); |
1490 |
if (ret) {
|
1491 |
av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
|
1492 |
return ret;
|
1493 |
} |
1494 |
|
1495 |
quantize_mantissas(s); |
1496 |
|
1497 |
output_frame(s, frame); |
1498 |
|
1499 |
return s->frame_size;
|
1500 |
} |
1501 |
|
1502 |
|
1503 |
/**
|
1504 |
* Finalize encoding and free any memory allocated by the encoder.
|
1505 |
*/
|
1506 |
static av_cold int ac3_encode_close(AVCodecContext *avctx) |
1507 |
{ |
1508 |
int blk, ch;
|
1509 |
AC3EncodeContext *s = avctx->priv_data; |
1510 |
|
1511 |
for (ch = 0; ch < s->channels; ch++) |
1512 |
av_freep(&s->planar_samples[ch]); |
1513 |
av_freep(&s->planar_samples); |
1514 |
av_freep(&s->bap_buffer); |
1515 |
av_freep(&s->bap1_buffer); |
1516 |
av_freep(&s->mdct_coef_buffer); |
1517 |
av_freep(&s->exp_buffer); |
1518 |
av_freep(&s->grouped_exp_buffer); |
1519 |
av_freep(&s->psd_buffer); |
1520 |
av_freep(&s->band_psd_buffer); |
1521 |
av_freep(&s->mask_buffer); |
1522 |
av_freep(&s->qmant_buffer); |
1523 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
1524 |
AC3Block *block = &s->blocks[blk]; |
1525 |
av_freep(&block->bap); |
1526 |
av_freep(&block->mdct_coef); |
1527 |
av_freep(&block->exp); |
1528 |
av_freep(&block->grouped_exp); |
1529 |
av_freep(&block->psd); |
1530 |
av_freep(&block->band_psd); |
1531 |
av_freep(&block->mask); |
1532 |
av_freep(&block->qmant); |
1533 |
} |
1534 |
|
1535 |
mdct_end(&s->mdct); |
1536 |
|
1537 |
av_freep(&avctx->coded_frame); |
1538 |
return 0; |
1539 |
} |
1540 |
|
1541 |
|
1542 |
/**
|
1543 |
* Set channel information during initialization.
|
1544 |
*/
|
1545 |
static av_cold int set_channel_info(AC3EncodeContext *s, int channels, |
1546 |
int64_t *channel_layout) |
1547 |
{ |
1548 |
int ch_layout;
|
1549 |
|
1550 |
if (channels < 1 || channels > AC3_MAX_CHANNELS) |
1551 |
return AVERROR(EINVAL);
|
1552 |
if ((uint64_t)*channel_layout > 0x7FF) |
1553 |
return AVERROR(EINVAL);
|
1554 |
ch_layout = *channel_layout; |
1555 |
if (!ch_layout)
|
1556 |
ch_layout = avcodec_guess_channel_layout(channels, CODEC_ID_AC3, NULL);
|
1557 |
if (av_get_channel_layout_nb_channels(ch_layout) != channels)
|
1558 |
return AVERROR(EINVAL);
|
1559 |
|
1560 |
s->lfe_on = !!(ch_layout & AV_CH_LOW_FREQUENCY); |
1561 |
s->channels = channels; |
1562 |
s->fbw_channels = channels - s->lfe_on; |
1563 |
s->lfe_channel = s->lfe_on ? s->fbw_channels : -1;
|
1564 |
if (s->lfe_on)
|
1565 |
ch_layout -= AV_CH_LOW_FREQUENCY; |
1566 |
|
1567 |
switch (ch_layout) {
|
1568 |
case AV_CH_LAYOUT_MONO: s->channel_mode = AC3_CHMODE_MONO; break; |
1569 |
case AV_CH_LAYOUT_STEREO: s->channel_mode = AC3_CHMODE_STEREO; break; |
1570 |
case AV_CH_LAYOUT_SURROUND: s->channel_mode = AC3_CHMODE_3F; break; |
1571 |
case AV_CH_LAYOUT_2_1: s->channel_mode = AC3_CHMODE_2F1R; break; |
1572 |
case AV_CH_LAYOUT_4POINT0: s->channel_mode = AC3_CHMODE_3F1R; break; |
1573 |
case AV_CH_LAYOUT_QUAD:
|
1574 |
case AV_CH_LAYOUT_2_2: s->channel_mode = AC3_CHMODE_2F2R; break; |
1575 |
case AV_CH_LAYOUT_5POINT0:
|
1576 |
case AV_CH_LAYOUT_5POINT0_BACK: s->channel_mode = AC3_CHMODE_3F2R; break; |
1577 |
default:
|
1578 |
return AVERROR(EINVAL);
|
1579 |
} |
1580 |
|
1581 |
s->channel_map = ff_ac3_enc_channel_map[s->channel_mode][s->lfe_on]; |
1582 |
*channel_layout = ch_layout; |
1583 |
if (s->lfe_on)
|
1584 |
*channel_layout |= AV_CH_LOW_FREQUENCY; |
1585 |
|
1586 |
return 0; |
1587 |
} |
1588 |
|
1589 |
|
1590 |
static av_cold int validate_options(AVCodecContext *avctx, AC3EncodeContext *s) |
1591 |
{ |
1592 |
int i, ret;
|
1593 |
|
1594 |
/* validate channel layout */
|
1595 |
if (!avctx->channel_layout) {
|
1596 |
av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
|
1597 |
"encoder will guess the layout, but it "
|
1598 |
"might be incorrect.\n");
|
1599 |
} |
1600 |
ret = set_channel_info(s, avctx->channels, &avctx->channel_layout); |
1601 |
if (ret) {
|
1602 |
av_log(avctx, AV_LOG_ERROR, "invalid channel layout\n");
|
1603 |
return ret;
|
1604 |
} |
1605 |
|
1606 |
/* validate sample rate */
|
1607 |
for (i = 0; i < 9; i++) { |
1608 |
if ((ff_ac3_sample_rate_tab[i / 3] >> (i % 3)) == avctx->sample_rate) |
1609 |
break;
|
1610 |
} |
1611 |
if (i == 9) { |
1612 |
av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
|
1613 |
return AVERROR(EINVAL);
|
1614 |
} |
1615 |
s->sample_rate = avctx->sample_rate; |
1616 |
s->bit_alloc.sr_shift = i % 3;
|
1617 |
s->bit_alloc.sr_code = i / 3;
|
1618 |
|
1619 |
/* validate bit rate */
|
1620 |
for (i = 0; i < 19; i++) { |
1621 |
if ((ff_ac3_bitrate_tab[i] >> s->bit_alloc.sr_shift)*1000 == avctx->bit_rate) |
1622 |
break;
|
1623 |
} |
1624 |
if (i == 19) { |
1625 |
av_log(avctx, AV_LOG_ERROR, "invalid bit rate\n");
|
1626 |
return AVERROR(EINVAL);
|
1627 |
} |
1628 |
s->bit_rate = avctx->bit_rate; |
1629 |
s->frame_size_code = i << 1;
|
1630 |
|
1631 |
return 0; |
1632 |
} |
1633 |
|
1634 |
|
1635 |
/**
|
1636 |
* Set bandwidth for all channels.
|
1637 |
* The user can optionally supply a cutoff frequency. Otherwise an appropriate
|
1638 |
* default value will be used.
|
1639 |
*/
|
1640 |
static av_cold void set_bandwidth(AC3EncodeContext *s, int cutoff) |
1641 |
{ |
1642 |
int ch, bw_code;
|
1643 |
|
1644 |
if (cutoff) {
|
1645 |
/* calculate bandwidth based on user-specified cutoff frequency */
|
1646 |
int fbw_coeffs;
|
1647 |
cutoff = av_clip(cutoff, 1, s->sample_rate >> 1); |
1648 |
fbw_coeffs = cutoff * 2 * AC3_MAX_COEFS / s->sample_rate;
|
1649 |
bw_code = av_clip((fbw_coeffs - 73) / 3, 0, 60); |
1650 |
} else {
|
1651 |
/* use default bandwidth setting */
|
1652 |
/* XXX: should compute the bandwidth according to the frame
|
1653 |
size, so that we avoid annoying high frequency artifacts */
|
1654 |
bw_code = 50;
|
1655 |
} |
1656 |
|
1657 |
/* set number of coefficients for each channel */
|
1658 |
for (ch = 0; ch < s->fbw_channels; ch++) { |
1659 |
s->bandwidth_code[ch] = bw_code; |
1660 |
s->nb_coefs[ch] = bw_code * 3 + 73; |
1661 |
} |
1662 |
if (s->lfe_on)
|
1663 |
s->nb_coefs[s->lfe_channel] = 7; /* LFE channel always has 7 coefs */ |
1664 |
} |
1665 |
|
1666 |
|
1667 |
static av_cold int allocate_buffers(AVCodecContext *avctx) |
1668 |
{ |
1669 |
int blk, ch;
|
1670 |
AC3EncodeContext *s = avctx->priv_data; |
1671 |
|
1672 |
FF_ALLOC_OR_GOTO(avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
|
1673 |
alloc_fail); |
1674 |
for (ch = 0; ch < s->channels; ch++) { |
1675 |
FF_ALLOCZ_OR_GOTO(avctx, s->planar_samples[ch], |
1676 |
(AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
|
1677 |
alloc_fail); |
1678 |
} |
1679 |
FF_ALLOC_OR_GOTO(avctx, s->bap_buffer, AC3_MAX_BLOCKS * s->channels * |
1680 |
AC3_MAX_COEFS * sizeof(*s->bap_buffer), alloc_fail);
|
1681 |
FF_ALLOC_OR_GOTO(avctx, s->bap1_buffer, AC3_MAX_BLOCKS * s->channels * |
1682 |
AC3_MAX_COEFS * sizeof(*s->bap1_buffer), alloc_fail);
|
1683 |
FF_ALLOC_OR_GOTO(avctx, s->mdct_coef_buffer, AC3_MAX_BLOCKS * s->channels * |
1684 |
AC3_MAX_COEFS * sizeof(*s->mdct_coef_buffer), alloc_fail);
|
1685 |
FF_ALLOC_OR_GOTO(avctx, s->exp_buffer, AC3_MAX_BLOCKS * s->channels * |
1686 |
AC3_MAX_COEFS * sizeof(*s->exp_buffer), alloc_fail);
|
1687 |
FF_ALLOC_OR_GOTO(avctx, s->grouped_exp_buffer, AC3_MAX_BLOCKS * s->channels * |
1688 |
128 * sizeof(*s->grouped_exp_buffer), alloc_fail); |
1689 |
FF_ALLOC_OR_GOTO(avctx, s->psd_buffer, AC3_MAX_BLOCKS * s->channels * |
1690 |
AC3_MAX_COEFS * sizeof(*s->psd_buffer), alloc_fail);
|
1691 |
FF_ALLOC_OR_GOTO(avctx, s->band_psd_buffer, AC3_MAX_BLOCKS * s->channels * |
1692 |
64 * sizeof(*s->band_psd_buffer), alloc_fail); |
1693 |
FF_ALLOC_OR_GOTO(avctx, s->mask_buffer, AC3_MAX_BLOCKS * s->channels * |
1694 |
64 * sizeof(*s->mask_buffer), alloc_fail); |
1695 |
FF_ALLOC_OR_GOTO(avctx, s->qmant_buffer, AC3_MAX_BLOCKS * s->channels * |
1696 |
AC3_MAX_COEFS * sizeof(*s->qmant_buffer), alloc_fail);
|
1697 |
for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { |
1698 |
AC3Block *block = &s->blocks[blk]; |
1699 |
FF_ALLOC_OR_GOTO(avctx, block->bap, s->channels * sizeof(*block->bap),
|
1700 |
alloc_fail); |
1701 |
FF_ALLOCZ_OR_GOTO(avctx, block->mdct_coef, s->channels * sizeof(*block->mdct_coef),
|
1702 |
alloc_fail); |
1703 |
FF_ALLOCZ_OR_GOTO(avctx, block->exp, s->channels * sizeof(*block->exp),
|
1704 |
alloc_fail); |
1705 |
FF_ALLOCZ_OR_GOTO(avctx, block->grouped_exp, s->channels * sizeof(*block->grouped_exp),
|
1706 |
alloc_fail); |
1707 |
FF_ALLOCZ_OR_GOTO(avctx, block->psd, s->channels * sizeof(*block->psd),
|
1708 |
alloc_fail); |
1709 |
FF_ALLOCZ_OR_GOTO(avctx, block->band_psd, s->channels * sizeof(*block->band_psd),
|
1710 |
alloc_fail); |
1711 |
FF_ALLOCZ_OR_GOTO(avctx, block->mask, s->channels * sizeof(*block->mask),
|
1712 |
alloc_fail); |
1713 |
FF_ALLOCZ_OR_GOTO(avctx, block->qmant, s->channels * sizeof(*block->qmant),
|
1714 |
alloc_fail); |
1715 |
|
1716 |
for (ch = 0; ch < s->channels; ch++) { |
1717 |
block->bap[ch] = &s->bap_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
1718 |
block->mdct_coef[ch] = &s->mdct_coef_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
1719 |
block->exp[ch] = &s->exp_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
1720 |
block->grouped_exp[ch] = &s->grouped_exp_buffer[128 * (blk * s->channels + ch)];
|
1721 |
block->psd[ch] = &s->psd_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
1722 |
block->band_psd[ch] = &s->band_psd_buffer [64 * (blk * s->channels + ch)];
|
1723 |
block->mask[ch] = &s->mask_buffer [64 * (blk * s->channels + ch)];
|
1724 |
block->qmant[ch] = &s->qmant_buffer [AC3_MAX_COEFS * (blk * s->channels + ch)]; |
1725 |
} |
1726 |
} |
1727 |
|
1728 |
return 0; |
1729 |
alloc_fail:
|
1730 |
return AVERROR(ENOMEM);
|
1731 |
} |
1732 |
|
1733 |
|
1734 |
/**
|
1735 |
* Initialize the encoder.
|
1736 |
*/
|
1737 |
static av_cold int ac3_encode_init(AVCodecContext *avctx) |
1738 |
{ |
1739 |
AC3EncodeContext *s = avctx->priv_data; |
1740 |
int ret;
|
1741 |
|
1742 |
avctx->frame_size = AC3_FRAME_SIZE; |
1743 |
|
1744 |
ac3_common_init(); |
1745 |
|
1746 |
ret = validate_options(avctx, s); |
1747 |
if (ret)
|
1748 |
return ret;
|
1749 |
|
1750 |
s->bitstream_id = 8 + s->bit_alloc.sr_shift;
|
1751 |
s->bitstream_mode = 0; /* complete main audio service */ |
1752 |
|
1753 |
s->frame_size_min = 2 * ff_ac3_frame_size_tab[s->frame_size_code][s->bit_alloc.sr_code];
|
1754 |
s->bits_written = 0;
|
1755 |
s->samples_written = 0;
|
1756 |
s->frame_size = s->frame_size_min; |
1757 |
|
1758 |
set_bandwidth(s, avctx->cutoff); |
1759 |
|
1760 |
bit_alloc_init(s); |
1761 |
|
1762 |
s->mdct.avctx = avctx; |
1763 |
ret = mdct_init(&s->mdct, 9);
|
1764 |
if (ret)
|
1765 |
goto init_fail;
|
1766 |
|
1767 |
ret = allocate_buffers(avctx); |
1768 |
if (ret)
|
1769 |
goto init_fail;
|
1770 |
|
1771 |
avctx->coded_frame= avcodec_alloc_frame(); |
1772 |
|
1773 |
dsputil_init(&s->dsp, avctx); |
1774 |
|
1775 |
return 0; |
1776 |
init_fail:
|
1777 |
ac3_encode_close(avctx); |
1778 |
return ret;
|
1779 |
} |
1780 |
|
1781 |
|
1782 |
#ifdef TEST
|
1783 |
/*************************************************************************/
|
1784 |
/* TEST */
|
1785 |
|
1786 |
#include "libavutil/lfg.h" |
1787 |
|
1788 |
#define FN (MDCT_SAMPLES/4) |
1789 |
|
1790 |
|
1791 |
static void fft_test(AVLFG *lfg) |
1792 |
{ |
1793 |
IComplex in[FN], in1[FN]; |
1794 |
int k, n, i;
|
1795 |
float sum_re, sum_im, a;
|
1796 |
|
1797 |
for (i = 0; i < FN; i++) { |
1798 |
in[i].re = av_lfg_get(lfg) % 65535 - 32767; |
1799 |
in[i].im = av_lfg_get(lfg) % 65535 - 32767; |
1800 |
in1[i] = in[i]; |
1801 |
} |
1802 |
fft(in, 7);
|
1803 |
|
1804 |
/* do it by hand */
|
1805 |
for (k = 0; k < FN; k++) { |
1806 |
sum_re = 0;
|
1807 |
sum_im = 0;
|
1808 |
for (n = 0; n < FN; n++) { |
1809 |
a = -2 * M_PI * (n * k) / FN;
|
1810 |
sum_re += in1[n].re * cos(a) - in1[n].im * sin(a); |
1811 |
sum_im += in1[n].re * sin(a) + in1[n].im * cos(a); |
1812 |
} |
1813 |
av_log(NULL, AV_LOG_DEBUG, "%3d: %6d,%6d %6.0f,%6.0f\n", |
1814 |
k, in[k].re, in[k].im, sum_re / FN, sum_im / FN); |
1815 |
} |
1816 |
} |
1817 |
|
1818 |
|
1819 |
static void mdct_test(AVLFG *lfg) |
1820 |
{ |
1821 |
int16_t input[MDCT_SAMPLES]; |
1822 |
int32_t output[AC3_MAX_COEFS]; |
1823 |
float input1[MDCT_SAMPLES];
|
1824 |
float output1[AC3_MAX_COEFS];
|
1825 |
float s, a, err, e, emax;
|
1826 |
int i, k, n;
|
1827 |
|
1828 |
for (i = 0; i < MDCT_SAMPLES; i++) { |
1829 |
input[i] = (av_lfg_get(lfg) % 65535 - 32767) * 9 / 10; |
1830 |
input1[i] = input[i]; |
1831 |
} |
1832 |
|
1833 |
mdct512(output, input); |
1834 |
|
1835 |
/* do it by hand */
|
1836 |
for (k = 0; k < AC3_MAX_COEFS; k++) { |
1837 |
s = 0;
|
1838 |
for (n = 0; n < MDCT_SAMPLES; n++) { |
1839 |
a = (2*M_PI*(2*n+1+MDCT_SAMPLES/2)*(2*k+1) / (4 * MDCT_SAMPLES)); |
1840 |
s += input1[n] * cos(a); |
1841 |
} |
1842 |
output1[k] = -2 * s / MDCT_SAMPLES;
|
1843 |
} |
1844 |
|
1845 |
err = 0;
|
1846 |
emax = 0;
|
1847 |
for (i = 0; i < AC3_MAX_COEFS; i++) { |
1848 |
av_log(NULL, AV_LOG_DEBUG, "%3d: %7d %7.0f\n", i, output[i], output1[i]); |
1849 |
e = output[i] - output1[i]; |
1850 |
if (e > emax)
|
1851 |
emax = e; |
1852 |
err += e * e; |
1853 |
} |
1854 |
av_log(NULL, AV_LOG_DEBUG, "err2=%f emax=%f\n", err / AC3_MAX_COEFS, emax); |
1855 |
} |
1856 |
|
1857 |
|
1858 |
int main(void) |
1859 |
{ |
1860 |
AVLFG lfg; |
1861 |
|
1862 |
av_log_set_level(AV_LOG_DEBUG); |
1863 |
mdct_init(9);
|
1864 |
|
1865 |
fft_test(&lfg); |
1866 |
mdct_test(&lfg); |
1867 |
|
1868 |
return 0; |
1869 |
} |
1870 |
#endif /* TEST */ |
1871 |
|
1872 |
|
1873 |
AVCodec ac3_encoder = { |
1874 |
"ac3",
|
1875 |
AVMEDIA_TYPE_AUDIO, |
1876 |
CODEC_ID_AC3, |
1877 |
sizeof(AC3EncodeContext),
|
1878 |
ac3_encode_init, |
1879 |
ac3_encode_frame, |
1880 |
ac3_encode_close, |
1881 |
NULL,
|
1882 |
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, |
1883 |
.long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
|
1884 |
.channel_layouts = (const int64_t[]){
|
1885 |
AV_CH_LAYOUT_MONO, |
1886 |
AV_CH_LAYOUT_STEREO, |
1887 |
AV_CH_LAYOUT_2_1, |
1888 |
AV_CH_LAYOUT_SURROUND, |
1889 |
AV_CH_LAYOUT_2_2, |
1890 |
AV_CH_LAYOUT_QUAD, |
1891 |
AV_CH_LAYOUT_4POINT0, |
1892 |
AV_CH_LAYOUT_5POINT0, |
1893 |
AV_CH_LAYOUT_5POINT0_BACK, |
1894 |
(AV_CH_LAYOUT_MONO | AV_CH_LOW_FREQUENCY), |
1895 |
(AV_CH_LAYOUT_STEREO | AV_CH_LOW_FREQUENCY), |
1896 |
(AV_CH_LAYOUT_2_1 | AV_CH_LOW_FREQUENCY), |
1897 |
(AV_CH_LAYOUT_SURROUND | AV_CH_LOW_FREQUENCY), |
1898 |
(AV_CH_LAYOUT_2_2 | AV_CH_LOW_FREQUENCY), |
1899 |
(AV_CH_LAYOUT_QUAD | AV_CH_LOW_FREQUENCY), |
1900 |
(AV_CH_LAYOUT_4POINT0 | AV_CH_LOW_FREQUENCY), |
1901 |
AV_CH_LAYOUT_5POINT1, |
1902 |
AV_CH_LAYOUT_5POINT1_BACK, |
1903 |
0 },
|
1904 |
}; |