ffmpeg / libavcodec / mpegaudioenc.c @ 6bb6fb05
History | View | Annotate | Download (22.6 KB)
1 |
/*
|
---|---|
2 |
* The simplest mpeg audio layer 2 encoder
|
3 |
* Copyright (c) 2000, 2001 Fabrice Bellard
|
4 |
*
|
5 |
* This file is part of Libav.
|
6 |
*
|
7 |
* Libav 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 |
* Libav 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 Libav; 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 mpeg audio layer 2 encoder.
|
25 |
*/
|
26 |
|
27 |
#include "avcodec.h" |
28 |
#include "put_bits.h" |
29 |
|
30 |
#define FRAC_BITS 15 /* fractional bits for sb_samples and dct */ |
31 |
#define WFRAC_BITS 14 /* fractional bits for window */ |
32 |
|
33 |
#include "mpegaudio.h" |
34 |
|
35 |
/* currently, cannot change these constants (need to modify
|
36 |
quantization stage) */
|
37 |
#define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
|
38 |
|
39 |
#define SAMPLES_BUF_SIZE 4096 |
40 |
|
41 |
typedef struct MpegAudioContext { |
42 |
PutBitContext pb; |
43 |
int nb_channels;
|
44 |
int lsf; /* 1 if mpeg2 low bitrate selected */ |
45 |
int bitrate_index; /* bit rate */ |
46 |
int freq_index;
|
47 |
int frame_size; /* frame size, in bits, without padding */ |
48 |
/* padding computation */
|
49 |
int frame_frac, frame_frac_incr, do_padding;
|
50 |
short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ |
51 |
int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ |
52 |
int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; |
53 |
unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ |
54 |
/* code to group 3 scale factors */
|
55 |
unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; |
56 |
int sblimit; /* number of used subbands */ |
57 |
const unsigned char *alloc_table; |
58 |
} MpegAudioContext; |
59 |
|
60 |
/* define it to use floats in quantization (I don't like floats !) */
|
61 |
#define USE_FLOATS
|
62 |
|
63 |
#include "mpegaudiodata.h" |
64 |
#include "mpegaudiotab.h" |
65 |
|
66 |
static av_cold int MPA_encode_init(AVCodecContext *avctx) |
67 |
{ |
68 |
MpegAudioContext *s = avctx->priv_data; |
69 |
int freq = avctx->sample_rate;
|
70 |
int bitrate = avctx->bit_rate;
|
71 |
int channels = avctx->channels;
|
72 |
int i, v, table;
|
73 |
float a;
|
74 |
|
75 |
if (channels <= 0 || channels > 2){ |
76 |
av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels);
|
77 |
return -1; |
78 |
} |
79 |
bitrate = bitrate / 1000;
|
80 |
s->nb_channels = channels; |
81 |
avctx->frame_size = MPA_FRAME_SIZE; |
82 |
|
83 |
/* encoding freq */
|
84 |
s->lsf = 0;
|
85 |
for(i=0;i<3;i++) { |
86 |
if (ff_mpa_freq_tab[i] == freq)
|
87 |
break;
|
88 |
if ((ff_mpa_freq_tab[i] / 2) == freq) { |
89 |
s->lsf = 1;
|
90 |
break;
|
91 |
} |
92 |
} |
93 |
if (i == 3){ |
94 |
av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq);
|
95 |
return -1; |
96 |
} |
97 |
s->freq_index = i; |
98 |
|
99 |
/* encoding bitrate & frequency */
|
100 |
for(i=0;i<15;i++) { |
101 |
if (ff_mpa_bitrate_tab[s->lsf][1][i] == bitrate) |
102 |
break;
|
103 |
} |
104 |
if (i == 15){ |
105 |
av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate);
|
106 |
return -1; |
107 |
} |
108 |
s->bitrate_index = i; |
109 |
|
110 |
/* compute total header size & pad bit */
|
111 |
|
112 |
a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); |
113 |
s->frame_size = ((int)a) * 8; |
114 |
|
115 |
/* frame fractional size to compute padding */
|
116 |
s->frame_frac = 0;
|
117 |
s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); |
118 |
|
119 |
/* select the right allocation table */
|
120 |
table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf); |
121 |
|
122 |
/* number of used subbands */
|
123 |
s->sblimit = ff_mpa_sblimit_table[table]; |
124 |
s->alloc_table = ff_mpa_alloc_tables[table]; |
125 |
|
126 |
av_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
|
127 |
bitrate, freq, s->frame_size, table, s->frame_frac_incr); |
128 |
|
129 |
for(i=0;i<s->nb_channels;i++) |
130 |
s->samples_offset[i] = 0;
|
131 |
|
132 |
for(i=0;i<257;i++) { |
133 |
int v;
|
134 |
v = ff_mpa_enwindow[i]; |
135 |
#if WFRAC_BITS != 16 |
136 |
v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); |
137 |
#endif
|
138 |
filter_bank[i] = v; |
139 |
if ((i & 63) != 0) |
140 |
v = -v; |
141 |
if (i != 0) |
142 |
filter_bank[512 - i] = v;
|
143 |
} |
144 |
|
145 |
for(i=0;i<64;i++) { |
146 |
v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20)); |
147 |
if (v <= 0) |
148 |
v = 1;
|
149 |
scale_factor_table[i] = v; |
150 |
#ifdef USE_FLOATS
|
151 |
scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20); |
152 |
#else
|
153 |
#define P 15 |
154 |
scale_factor_shift[i] = 21 - P - (i / 3); |
155 |
scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0); |
156 |
#endif
|
157 |
} |
158 |
for(i=0;i<128;i++) { |
159 |
v = i - 64;
|
160 |
if (v <= -3) |
161 |
v = 0;
|
162 |
else if (v < 0) |
163 |
v = 1;
|
164 |
else if (v == 0) |
165 |
v = 2;
|
166 |
else if (v < 3) |
167 |
v = 3;
|
168 |
else
|
169 |
v = 4;
|
170 |
scale_diff_table[i] = v; |
171 |
} |
172 |
|
173 |
for(i=0;i<17;i++) { |
174 |
v = ff_mpa_quant_bits[i]; |
175 |
if (v < 0) |
176 |
v = -v; |
177 |
else
|
178 |
v = v * 3;
|
179 |
total_quant_bits[i] = 12 * v;
|
180 |
} |
181 |
|
182 |
avctx->coded_frame= avcodec_alloc_frame(); |
183 |
avctx->coded_frame->key_frame= 1;
|
184 |
|
185 |
return 0; |
186 |
} |
187 |
|
188 |
/* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */
|
189 |
static void idct32(int *out, int *tab) |
190 |
{ |
191 |
int i, j;
|
192 |
int *t, *t1, xr;
|
193 |
const int *xp = costab32; |
194 |
|
195 |
for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; |
196 |
|
197 |
t = tab + 30;
|
198 |
t1 = tab + 2;
|
199 |
do {
|
200 |
t[0] += t[-4]; |
201 |
t[1] += t[1 - 4]; |
202 |
t -= 4;
|
203 |
} while (t != t1);
|
204 |
|
205 |
t = tab + 28;
|
206 |
t1 = tab + 4;
|
207 |
do {
|
208 |
t[0] += t[-8]; |
209 |
t[1] += t[1-8]; |
210 |
t[2] += t[2-8]; |
211 |
t[3] += t[3-8]; |
212 |
t -= 8;
|
213 |
} while (t != t1);
|
214 |
|
215 |
t = tab; |
216 |
t1 = tab + 32;
|
217 |
do {
|
218 |
t[ 3] = -t[ 3]; |
219 |
t[ 6] = -t[ 6]; |
220 |
|
221 |
t[11] = -t[11]; |
222 |
t[12] = -t[12]; |
223 |
t[13] = -t[13]; |
224 |
t[15] = -t[15]; |
225 |
t += 16;
|
226 |
} while (t != t1);
|
227 |
|
228 |
|
229 |
t = tab; |
230 |
t1 = tab + 8;
|
231 |
do {
|
232 |
int x1, x2, x3, x4;
|
233 |
|
234 |
x3 = MUL(t[16], FIX(SQRT2*0.5)); |
235 |
x4 = t[0] - x3;
|
236 |
x3 = t[0] + x3;
|
237 |
|
238 |
x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5)); |
239 |
x1 = MUL((t[8] - x2), xp[0]); |
240 |
x2 = MUL((t[8] + x2), xp[1]); |
241 |
|
242 |
t[ 0] = x3 + x1;
|
243 |
t[ 8] = x4 - x2;
|
244 |
t[16] = x4 + x2;
|
245 |
t[24] = x3 - x1;
|
246 |
t++; |
247 |
} while (t != t1);
|
248 |
|
249 |
xp += 2;
|
250 |
t = tab; |
251 |
t1 = tab + 4;
|
252 |
do {
|
253 |
xr = MUL(t[28],xp[0]); |
254 |
t[28] = (t[0] - xr); |
255 |
t[0] = (t[0] + xr); |
256 |
|
257 |
xr = MUL(t[4],xp[1]); |
258 |
t[ 4] = (t[24] - xr); |
259 |
t[24] = (t[24] + xr); |
260 |
|
261 |
xr = MUL(t[20],xp[2]); |
262 |
t[20] = (t[8] - xr); |
263 |
t[ 8] = (t[8] + xr); |
264 |
|
265 |
xr = MUL(t[12],xp[3]); |
266 |
t[12] = (t[16] - xr); |
267 |
t[16] = (t[16] + xr); |
268 |
t++; |
269 |
} while (t != t1);
|
270 |
xp += 4;
|
271 |
|
272 |
for (i = 0; i < 4; i++) { |
273 |
xr = MUL(tab[30-i*4],xp[0]); |
274 |
tab[30-i*4] = (tab[i*4] - xr); |
275 |
tab[ i*4] = (tab[i*4] + xr); |
276 |
|
277 |
xr = MUL(tab[ 2+i*4],xp[1]); |
278 |
tab[ 2+i*4] = (tab[28-i*4] - xr); |
279 |
tab[28-i*4] = (tab[28-i*4] + xr); |
280 |
|
281 |
xr = MUL(tab[31-i*4],xp[0]); |
282 |
tab[31-i*4] = (tab[1+i*4] - xr); |
283 |
tab[ 1+i*4] = (tab[1+i*4] + xr); |
284 |
|
285 |
xr = MUL(tab[ 3+i*4],xp[1]); |
286 |
tab[ 3+i*4] = (tab[29-i*4] - xr); |
287 |
tab[29-i*4] = (tab[29-i*4] + xr); |
288 |
|
289 |
xp += 2;
|
290 |
} |
291 |
|
292 |
t = tab + 30;
|
293 |
t1 = tab + 1;
|
294 |
do {
|
295 |
xr = MUL(t1[0], *xp);
|
296 |
t1[0] = (t[0] - xr); |
297 |
t[0] = (t[0] + xr); |
298 |
t -= 2;
|
299 |
t1 += 2;
|
300 |
xp++; |
301 |
} while (t >= tab);
|
302 |
|
303 |
for(i=0;i<32;i++) { |
304 |
out[i] = tab[bitinv32[i]]; |
305 |
} |
306 |
} |
307 |
|
308 |
#define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS) |
309 |
|
310 |
static void filter(MpegAudioContext *s, int ch, const short *samples, int incr) |
311 |
{ |
312 |
short *p, *q;
|
313 |
int sum, offset, i, j;
|
314 |
int tmp[64]; |
315 |
int tmp1[32]; |
316 |
int *out;
|
317 |
|
318 |
// print_pow1(samples, 1152);
|
319 |
|
320 |
offset = s->samples_offset[ch]; |
321 |
out = &s->sb_samples[ch][0][0][0]; |
322 |
for(j=0;j<36;j++) { |
323 |
/* 32 samples at once */
|
324 |
for(i=0;i<32;i++) { |
325 |
s->samples_buf[ch][offset + (31 - i)] = samples[0]; |
326 |
samples += incr; |
327 |
} |
328 |
|
329 |
/* filter */
|
330 |
p = s->samples_buf[ch] + offset; |
331 |
q = filter_bank; |
332 |
/* maxsum = 23169 */
|
333 |
for(i=0;i<64;i++) { |
334 |
sum = p[0*64] * q[0*64]; |
335 |
sum += p[1*64] * q[1*64]; |
336 |
sum += p[2*64] * q[2*64]; |
337 |
sum += p[3*64] * q[3*64]; |
338 |
sum += p[4*64] * q[4*64]; |
339 |
sum += p[5*64] * q[5*64]; |
340 |
sum += p[6*64] * q[6*64]; |
341 |
sum += p[7*64] * q[7*64]; |
342 |
tmp[i] = sum; |
343 |
p++; |
344 |
q++; |
345 |
} |
346 |
tmp1[0] = tmp[16] >> WSHIFT; |
347 |
for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT; |
348 |
for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; |
349 |
|
350 |
idct32(out, tmp1); |
351 |
|
352 |
/* advance of 32 samples */
|
353 |
offset -= 32;
|
354 |
out += 32;
|
355 |
/* handle the wrap around */
|
356 |
if (offset < 0) { |
357 |
memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), |
358 |
s->samples_buf[ch], (512 - 32) * 2); |
359 |
offset = SAMPLES_BUF_SIZE - 512;
|
360 |
} |
361 |
} |
362 |
s->samples_offset[ch] = offset; |
363 |
|
364 |
// print_pow(s->sb_samples, 1152);
|
365 |
} |
366 |
|
367 |
static void compute_scale_factors(unsigned char scale_code[SBLIMIT], |
368 |
unsigned char scale_factors[SBLIMIT][3], |
369 |
int sb_samples[3][12][SBLIMIT], |
370 |
int sblimit)
|
371 |
{ |
372 |
int *p, vmax, v, n, i, j, k, code;
|
373 |
int index, d1, d2;
|
374 |
unsigned char *sf = &scale_factors[0][0]; |
375 |
|
376 |
for(j=0;j<sblimit;j++) { |
377 |
for(i=0;i<3;i++) { |
378 |
/* find the max absolute value */
|
379 |
p = &sb_samples[i][0][j];
|
380 |
vmax = abs(*p); |
381 |
for(k=1;k<12;k++) { |
382 |
p += SBLIMIT; |
383 |
v = abs(*p); |
384 |
if (v > vmax)
|
385 |
vmax = v; |
386 |
} |
387 |
/* compute the scale factor index using log 2 computations */
|
388 |
if (vmax > 1) { |
389 |
n = av_log2(vmax); |
390 |
/* n is the position of the MSB of vmax. now
|
391 |
use at most 2 compares to find the index */
|
392 |
index = (21 - n) * 3 - 3; |
393 |
if (index >= 0) { |
394 |
while (vmax <= scale_factor_table[index+1]) |
395 |
index++; |
396 |
} else {
|
397 |
index = 0; /* very unlikely case of overflow */ |
398 |
} |
399 |
} else {
|
400 |
index = 62; /* value 63 is not allowed */ |
401 |
} |
402 |
|
403 |
av_dlog(NULL, "%2d:%d in=%x %x %d\n", |
404 |
j, i, vmax, scale_factor_table[index], index); |
405 |
/* store the scale factor */
|
406 |
assert(index >=0 && index <= 63); |
407 |
sf[i] = index; |
408 |
} |
409 |
|
410 |
/* compute the transmission factor : look if the scale factors
|
411 |
are close enough to each other */
|
412 |
d1 = scale_diff_table[sf[0] - sf[1] + 64]; |
413 |
d2 = scale_diff_table[sf[1] - sf[2] + 64]; |
414 |
|
415 |
/* handle the 25 cases */
|
416 |
switch(d1 * 5 + d2) { |
417 |
case 0*5+0: |
418 |
case 0*5+4: |
419 |
case 3*5+4: |
420 |
case 4*5+0: |
421 |
case 4*5+4: |
422 |
code = 0;
|
423 |
break;
|
424 |
case 0*5+1: |
425 |
case 0*5+2: |
426 |
case 4*5+1: |
427 |
case 4*5+2: |
428 |
code = 3;
|
429 |
sf[2] = sf[1]; |
430 |
break;
|
431 |
case 0*5+3: |
432 |
case 4*5+3: |
433 |
code = 3;
|
434 |
sf[1] = sf[2]; |
435 |
break;
|
436 |
case 1*5+0: |
437 |
case 1*5+4: |
438 |
case 2*5+4: |
439 |
code = 1;
|
440 |
sf[1] = sf[0]; |
441 |
break;
|
442 |
case 1*5+1: |
443 |
case 1*5+2: |
444 |
case 2*5+0: |
445 |
case 2*5+1: |
446 |
case 2*5+2: |
447 |
code = 2;
|
448 |
sf[1] = sf[2] = sf[0]; |
449 |
break;
|
450 |
case 2*5+3: |
451 |
case 3*5+3: |
452 |
code = 2;
|
453 |
sf[0] = sf[1] = sf[2]; |
454 |
break;
|
455 |
case 3*5+0: |
456 |
case 3*5+1: |
457 |
case 3*5+2: |
458 |
code = 2;
|
459 |
sf[0] = sf[2] = sf[1]; |
460 |
break;
|
461 |
case 1*5+3: |
462 |
code = 2;
|
463 |
if (sf[0] > sf[2]) |
464 |
sf[0] = sf[2]; |
465 |
sf[1] = sf[2] = sf[0]; |
466 |
break;
|
467 |
default:
|
468 |
assert(0); //cannot happen |
469 |
code = 0; /* kill warning */ |
470 |
} |
471 |
|
472 |
av_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j, |
473 |
sf[0], sf[1], sf[2], d1, d2, code); |
474 |
scale_code[j] = code; |
475 |
sf += 3;
|
476 |
} |
477 |
} |
478 |
|
479 |
/* The most important function : psycho acoustic module. In this
|
480 |
encoder there is basically none, so this is the worst you can do,
|
481 |
but also this is the simpler. */
|
482 |
static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) |
483 |
{ |
484 |
int i;
|
485 |
|
486 |
for(i=0;i<s->sblimit;i++) { |
487 |
smr[i] = (int)(fixed_smr[i] * 10); |
488 |
} |
489 |
} |
490 |
|
491 |
|
492 |
#define SB_NOTALLOCATED 0 |
493 |
#define SB_ALLOCATED 1 |
494 |
#define SB_NOMORE 2 |
495 |
|
496 |
/* Try to maximize the smr while using a number of bits inferior to
|
497 |
the frame size. I tried to make the code simpler, faster and
|
498 |
smaller than other encoders :-) */
|
499 |
static void compute_bit_allocation(MpegAudioContext *s, |
500 |
short smr1[MPA_MAX_CHANNELS][SBLIMIT],
|
501 |
unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
502 |
int *padding)
|
503 |
{ |
504 |
int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
|
505 |
int incr;
|
506 |
short smr[MPA_MAX_CHANNELS][SBLIMIT];
|
507 |
unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; |
508 |
const unsigned char *alloc; |
509 |
|
510 |
memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); |
511 |
memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); |
512 |
memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
|
513 |
|
514 |
/* compute frame size and padding */
|
515 |
max_frame_size = s->frame_size; |
516 |
s->frame_frac += s->frame_frac_incr; |
517 |
if (s->frame_frac >= 65536) { |
518 |
s->frame_frac -= 65536;
|
519 |
s->do_padding = 1;
|
520 |
max_frame_size += 8;
|
521 |
} else {
|
522 |
s->do_padding = 0;
|
523 |
} |
524 |
|
525 |
/* compute the header + bit alloc size */
|
526 |
current_frame_size = 32;
|
527 |
alloc = s->alloc_table; |
528 |
for(i=0;i<s->sblimit;i++) { |
529 |
incr = alloc[0];
|
530 |
current_frame_size += incr * s->nb_channels; |
531 |
alloc += 1 << incr;
|
532 |
} |
533 |
for(;;) {
|
534 |
/* look for the subband with the largest signal to mask ratio */
|
535 |
max_sb = -1;
|
536 |
max_ch = -1;
|
537 |
max_smr = INT_MIN; |
538 |
for(ch=0;ch<s->nb_channels;ch++) { |
539 |
for(i=0;i<s->sblimit;i++) { |
540 |
if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
|
541 |
max_smr = smr[ch][i]; |
542 |
max_sb = i; |
543 |
max_ch = ch; |
544 |
} |
545 |
} |
546 |
} |
547 |
av_dlog(NULL, "current=%d max=%d max_sb=%d alloc=%d\n", |
548 |
current_frame_size, max_frame_size, max_sb, |
549 |
bit_alloc[max_sb]); |
550 |
if (max_sb < 0) |
551 |
break;
|
552 |
|
553 |
/* find alloc table entry (XXX: not optimal, should use
|
554 |
pointer table) */
|
555 |
alloc = s->alloc_table; |
556 |
for(i=0;i<max_sb;i++) { |
557 |
alloc += 1 << alloc[0]; |
558 |
} |
559 |
|
560 |
if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
|
561 |
/* nothing was coded for this band: add the necessary bits */
|
562 |
incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; |
563 |
incr += total_quant_bits[alloc[1]];
|
564 |
} else {
|
565 |
/* increments bit allocation */
|
566 |
b = bit_alloc[max_ch][max_sb]; |
567 |
incr = total_quant_bits[alloc[b + 1]] -
|
568 |
total_quant_bits[alloc[b]]; |
569 |
} |
570 |
|
571 |
if (current_frame_size + incr <= max_frame_size) {
|
572 |
/* can increase size */
|
573 |
b = ++bit_alloc[max_ch][max_sb]; |
574 |
current_frame_size += incr; |
575 |
/* decrease smr by the resolution we added */
|
576 |
smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; |
577 |
/* max allocation size reached ? */
|
578 |
if (b == ((1 << alloc[0]) - 1)) |
579 |
subband_status[max_ch][max_sb] = SB_NOMORE; |
580 |
else
|
581 |
subband_status[max_ch][max_sb] = SB_ALLOCATED; |
582 |
} else {
|
583 |
/* cannot increase the size of this subband */
|
584 |
subband_status[max_ch][max_sb] = SB_NOMORE; |
585 |
} |
586 |
} |
587 |
*padding = max_frame_size - current_frame_size; |
588 |
assert(*padding >= 0);
|
589 |
} |
590 |
|
591 |
/*
|
592 |
* Output the mpeg audio layer 2 frame. Note how the code is small
|
593 |
* compared to other encoders :-)
|
594 |
*/
|
595 |
static void encode_frame(MpegAudioContext *s, |
596 |
unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
597 |
int padding)
|
598 |
{ |
599 |
int i, j, k, l, bit_alloc_bits, b, ch;
|
600 |
unsigned char *sf; |
601 |
int q[3]; |
602 |
PutBitContext *p = &s->pb; |
603 |
|
604 |
/* header */
|
605 |
|
606 |
put_bits(p, 12, 0xfff); |
607 |
put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */ |
608 |
put_bits(p, 2, 4-2); /* layer 2 */ |
609 |
put_bits(p, 1, 1); /* no error protection */ |
610 |
put_bits(p, 4, s->bitrate_index);
|
611 |
put_bits(p, 2, s->freq_index);
|
612 |
put_bits(p, 1, s->do_padding); /* use padding */ |
613 |
put_bits(p, 1, 0); /* private_bit */ |
614 |
put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); |
615 |
put_bits(p, 2, 0); /* mode_ext */ |
616 |
put_bits(p, 1, 0); /* no copyright */ |
617 |
put_bits(p, 1, 1); /* original */ |
618 |
put_bits(p, 2, 0); /* no emphasis */ |
619 |
|
620 |
/* bit allocation */
|
621 |
j = 0;
|
622 |
for(i=0;i<s->sblimit;i++) { |
623 |
bit_alloc_bits = s->alloc_table[j]; |
624 |
for(ch=0;ch<s->nb_channels;ch++) { |
625 |
put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); |
626 |
} |
627 |
j += 1 << bit_alloc_bits;
|
628 |
} |
629 |
|
630 |
/* scale codes */
|
631 |
for(i=0;i<s->sblimit;i++) { |
632 |
for(ch=0;ch<s->nb_channels;ch++) { |
633 |
if (bit_alloc[ch][i])
|
634 |
put_bits(p, 2, s->scale_code[ch][i]);
|
635 |
} |
636 |
} |
637 |
|
638 |
/* scale factors */
|
639 |
for(i=0;i<s->sblimit;i++) { |
640 |
for(ch=0;ch<s->nb_channels;ch++) { |
641 |
if (bit_alloc[ch][i]) {
|
642 |
sf = &s->scale_factors[ch][i][0];
|
643 |
switch(s->scale_code[ch][i]) {
|
644 |
case 0: |
645 |
put_bits(p, 6, sf[0]); |
646 |
put_bits(p, 6, sf[1]); |
647 |
put_bits(p, 6, sf[2]); |
648 |
break;
|
649 |
case 3: |
650 |
case 1: |
651 |
put_bits(p, 6, sf[0]); |
652 |
put_bits(p, 6, sf[2]); |
653 |
break;
|
654 |
case 2: |
655 |
put_bits(p, 6, sf[0]); |
656 |
break;
|
657 |
} |
658 |
} |
659 |
} |
660 |
} |
661 |
|
662 |
/* quantization & write sub band samples */
|
663 |
|
664 |
for(k=0;k<3;k++) { |
665 |
for(l=0;l<12;l+=3) { |
666 |
j = 0;
|
667 |
for(i=0;i<s->sblimit;i++) { |
668 |
bit_alloc_bits = s->alloc_table[j]; |
669 |
for(ch=0;ch<s->nb_channels;ch++) { |
670 |
b = bit_alloc[ch][i]; |
671 |
if (b) {
|
672 |
int qindex, steps, m, sample, bits;
|
673 |
/* we encode 3 sub band samples of the same sub band at a time */
|
674 |
qindex = s->alloc_table[j+b]; |
675 |
steps = ff_mpa_quant_steps[qindex]; |
676 |
for(m=0;m<3;m++) { |
677 |
sample = s->sb_samples[ch][k][l + m][i]; |
678 |
/* divide by scale factor */
|
679 |
#ifdef USE_FLOATS
|
680 |
{ |
681 |
float a;
|
682 |
a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]];
|
683 |
q[m] = (int)((a + 1.0) * steps * 0.5); |
684 |
} |
685 |
#else
|
686 |
{ |
687 |
int q1, e, shift, mult;
|
688 |
e = s->scale_factors[ch][i][k]; |
689 |
shift = scale_factor_shift[e]; |
690 |
mult = scale_factor_mult[e]; |
691 |
|
692 |
/* normalize to P bits */
|
693 |
if (shift < 0) |
694 |
q1 = sample << (-shift); |
695 |
else
|
696 |
q1 = sample >> shift; |
697 |
q1 = (q1 * mult) >> P; |
698 |
q[m] = ((q1 + (1 << P)) * steps) >> (P + 1); |
699 |
} |
700 |
#endif
|
701 |
if (q[m] >= steps)
|
702 |
q[m] = steps - 1;
|
703 |
assert(q[m] >= 0 && q[m] < steps);
|
704 |
} |
705 |
bits = ff_mpa_quant_bits[qindex]; |
706 |
if (bits < 0) { |
707 |
/* group the 3 values to save bits */
|
708 |
put_bits(p, -bits, |
709 |
q[0] + steps * (q[1] + steps * q[2])); |
710 |
} else {
|
711 |
put_bits(p, bits, q[0]);
|
712 |
put_bits(p, bits, q[1]);
|
713 |
put_bits(p, bits, q[2]);
|
714 |
} |
715 |
} |
716 |
} |
717 |
/* next subband in alloc table */
|
718 |
j += 1 << bit_alloc_bits;
|
719 |
} |
720 |
} |
721 |
} |
722 |
|
723 |
/* padding */
|
724 |
for(i=0;i<padding;i++) |
725 |
put_bits(p, 1, 0); |
726 |
|
727 |
/* flush */
|
728 |
flush_put_bits(p); |
729 |
} |
730 |
|
731 |
static int MPA_encode_frame(AVCodecContext *avctx, |
732 |
unsigned char *frame, int buf_size, void *data) |
733 |
{ |
734 |
MpegAudioContext *s = avctx->priv_data; |
735 |
const short *samples = data; |
736 |
short smr[MPA_MAX_CHANNELS][SBLIMIT];
|
737 |
unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; |
738 |
int padding, i;
|
739 |
|
740 |
for(i=0;i<s->nb_channels;i++) { |
741 |
filter(s, i, samples + i, s->nb_channels); |
742 |
} |
743 |
|
744 |
for(i=0;i<s->nb_channels;i++) { |
745 |
compute_scale_factors(s->scale_code[i], s->scale_factors[i], |
746 |
s->sb_samples[i], s->sblimit); |
747 |
} |
748 |
for(i=0;i<s->nb_channels;i++) { |
749 |
psycho_acoustic_model(s, smr[i]); |
750 |
} |
751 |
compute_bit_allocation(s, smr, bit_alloc, &padding); |
752 |
|
753 |
init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE); |
754 |
|
755 |
encode_frame(s, bit_alloc, padding); |
756 |
|
757 |
return put_bits_ptr(&s->pb) - s->pb.buf;
|
758 |
} |
759 |
|
760 |
static av_cold int MPA_encode_close(AVCodecContext *avctx) |
761 |
{ |
762 |
av_freep(&avctx->coded_frame); |
763 |
return 0; |
764 |
} |
765 |
|
766 |
AVCodec ff_mp2_encoder = { |
767 |
"mp2",
|
768 |
AVMEDIA_TYPE_AUDIO, |
769 |
CODEC_ID_MP2, |
770 |
sizeof(MpegAudioContext),
|
771 |
MPA_encode_init, |
772 |
MPA_encode_frame, |
773 |
MPA_encode_close, |
774 |
NULL,
|
775 |
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE}, |
776 |
.supported_samplerates= (const int[]){44100, 48000, 32000, 22050, 24000, 16000, 0}, |
777 |
.long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
|
778 |
}; |
779 |
|
780 |
#undef FIX
|