Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aacenc.c @ b58e2985

History | View | Annotate | Download (22.8 KB)

1
/*
2
 * AAC encoder
3
 * Copyright (C) 2008 Konstantin Shishkov
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
 * AAC encoder
25
 */
26

    
27
/***********************************
28
 *              TODOs:
29
 * add sane pulse detection
30
 * add temporal noise shaping
31
 ***********************************/
32

    
33
#include "avcodec.h"
34
#include "put_bits.h"
35
#include "dsputil.h"
36
#include "mpeg4audio.h"
37
#include "kbdwin.h"
38
#include "sinewin.h"
39

    
40
#include "aac.h"
41
#include "aactab.h"
42
#include "aacenc.h"
43

    
44
#include "psymodel.h"
45

    
46
#define AAC_MAX_CHANNELS 6
47

    
48
static const uint8_t swb_size_1024_96[] = {
49
    4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8,
50
    12, 12, 12, 12, 12, 16, 16, 24, 28, 36, 44,
51
    64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
52
};
53

    
54
static const uint8_t swb_size_1024_64[] = {
55
    4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8,
56
    12, 12, 12, 16, 16, 16, 20, 24, 24, 28, 36,
57
    40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40
58
};
59

    
60
static const uint8_t swb_size_1024_48[] = {
61
    4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
62
    12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
63
    32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
64
    96
65
};
66

    
67
static const uint8_t swb_size_1024_32[] = {
68
    4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
69
    12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
70
    32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32
71
};
72

    
73
static const uint8_t swb_size_1024_24[] = {
74
    4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
75
    12, 12, 12, 12, 16, 16, 16, 20, 20, 24, 24, 28, 28,
76
    32, 36, 36, 40, 44, 48, 52, 52, 64, 64, 64, 64, 64
77
};
78

    
79
static const uint8_t swb_size_1024_16[] = {
80
    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
81
    12, 12, 12, 12, 12, 12, 12, 12, 12, 16, 16, 16, 16, 20, 20, 20, 24, 24, 28, 28,
82
    32, 36, 40, 40, 44, 48, 52, 56, 60, 64, 64, 64
83
};
84

    
85
static const uint8_t swb_size_1024_8[] = {
86
    12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
87
    16, 16, 16, 16, 16, 16, 16, 20, 20, 20, 20, 24, 24, 24, 28, 28,
88
    32, 36, 36, 40, 44, 48, 52, 56, 60, 64, 80
89
};
90

    
91
static const uint8_t *swb_size_1024[] = {
92
    swb_size_1024_96, swb_size_1024_96, swb_size_1024_64,
93
    swb_size_1024_48, swb_size_1024_48, swb_size_1024_32,
94
    swb_size_1024_24, swb_size_1024_24, swb_size_1024_16,
95
    swb_size_1024_16, swb_size_1024_16, swb_size_1024_8
96
};
97

    
98
static const uint8_t swb_size_128_96[] = {
99
    4, 4, 4, 4, 4, 4, 8, 8, 8, 16, 28, 36
100
};
101

    
102
static const uint8_t swb_size_128_48[] = {
103
    4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 12, 16, 16, 16
104
};
105

    
106
static const uint8_t swb_size_128_24[] = {
107
    4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 20
108
};
109

    
110
static const uint8_t swb_size_128_16[] = {
111
    4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 12, 12, 16, 20, 20
112
};
113

    
114
static const uint8_t swb_size_128_8[] = {
115
    4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 12, 16, 20, 20
116
};
117

    
118
static const uint8_t *swb_size_128[] = {
119
    /* the last entry on the following row is swb_size_128_64 but is a
120
       duplicate of swb_size_128_96 */
121
    swb_size_128_96, swb_size_128_96, swb_size_128_96,
122
    swb_size_128_48, swb_size_128_48, swb_size_128_48,
123
    swb_size_128_24, swb_size_128_24, swb_size_128_16,
124
    swb_size_128_16, swb_size_128_16, swb_size_128_8
125
};
126

    
127
/** default channel configurations */
128
static const uint8_t aac_chan_configs[6][5] = {
129
 {1, TYPE_SCE},                               // 1 channel  - single channel element
130
 {1, TYPE_CPE},                               // 2 channels - channel pair
131
 {2, TYPE_SCE, TYPE_CPE},                     // 3 channels - center + stereo
132
 {3, TYPE_SCE, TYPE_CPE, TYPE_SCE},           // 4 channels - front center + stereo + back center
133
 {3, TYPE_SCE, TYPE_CPE, TYPE_CPE},           // 5 channels - front center + stereo + back stereo
134
 {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE
135
};
136

    
137
/**
138
 * Make AAC audio config object.
139
 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
140
 */
141
static void put_audio_specific_config(AVCodecContext *avctx)
142
{
143
    PutBitContext pb;
144
    AACEncContext *s = avctx->priv_data;
145

    
146
    init_put_bits(&pb, avctx->extradata, avctx->extradata_size*8);
147
    put_bits(&pb, 5, 2); //object type - AAC-LC
148
    put_bits(&pb, 4, s->samplerate_index); //sample rate index
149
    put_bits(&pb, 4, avctx->channels);
150
    //GASpecificConfig
151
    put_bits(&pb, 1, 0); //frame length - 1024 samples
152
    put_bits(&pb, 1, 0); //does not depend on core coder
153
    put_bits(&pb, 1, 0); //is not extension
154

    
155
    //Explicitly Mark SBR absent
156
    put_bits(&pb, 11, 0x2b7); //sync extension
157
    put_bits(&pb, 5,  AOT_SBR);
158
    put_bits(&pb, 1,  0);
159
    flush_put_bits(&pb);
160
}
161

    
162
static av_cold int aac_encode_init(AVCodecContext *avctx)
163
{
164
    AACEncContext *s = avctx->priv_data;
165
    int i;
166
    const uint8_t *sizes[2];
167
    int lengths[2];
168

    
169
    avctx->frame_size = 1024;
170

    
171
    for (i = 0; i < 16; i++)
172
        if (avctx->sample_rate == ff_mpeg4audio_sample_rates[i])
173
            break;
174
    if (i == 16) {
175
        av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate %d\n", avctx->sample_rate);
176
        return -1;
177
    }
178
    if (avctx->channels > AAC_MAX_CHANNELS) {
179
        av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n", avctx->channels);
180
        return -1;
181
    }
182
    if (avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW) {
183
        av_log(avctx, AV_LOG_ERROR, "Unsupported profile %d\n", avctx->profile);
184
        return -1;
185
    }
186
    if (1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * avctx->channels) {
187
        av_log(avctx, AV_LOG_ERROR, "Too many bits per frame requested\n");
188
        return -1;
189
    }
190
    s->samplerate_index = i;
191

    
192
    dsputil_init(&s->dsp, avctx);
193
    ff_mdct_init(&s->mdct1024, 11, 0, 1.0);
194
    ff_mdct_init(&s->mdct128,   8, 0, 1.0);
195
    // window init
196
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
197
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
198
    ff_init_ff_sine_windows(10);
199
    ff_init_ff_sine_windows(7);
200

    
201
    s->samples            = av_malloc(2 * 1024 * avctx->channels * sizeof(s->samples[0]));
202
    s->cpe                = av_mallocz(sizeof(ChannelElement) * aac_chan_configs[avctx->channels-1][0]);
203
    avctx->extradata      = av_mallocz(5 + FF_INPUT_BUFFER_PADDING_SIZE);
204
    avctx->extradata_size = 5;
205
    put_audio_specific_config(avctx);
206

    
207
    sizes[0]   = swb_size_1024[i];
208
    sizes[1]   = swb_size_128[i];
209
    lengths[0] = ff_aac_num_swb_1024[i];
210
    lengths[1] = ff_aac_num_swb_128[i];
211
    ff_psy_init(&s->psy, avctx, 2, sizes, lengths);
212
    s->psypp = ff_psy_preprocess_init(avctx);
213
    s->coder = &ff_aac_coders[2];
214

    
215
    s->lambda = avctx->global_quality ? avctx->global_quality : 120;
216

    
217
    ff_aac_tableinit();
218

    
219
    return 0;
220
}
221

    
222
static void apply_window_and_mdct(AVCodecContext *avctx, AACEncContext *s,
223
                                  SingleChannelElement *sce, short *audio)
224
{
225
    int i, k;
226
    const int chans = avctx->channels;
227
    const float * lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
228
    const float * swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
229
    const float * pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
230
    float *output = sce->ret;
231

    
232
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
233
        memcpy(output, sce->saved, sizeof(float)*1024);
234
        if (sce->ics.window_sequence[0] == LONG_STOP_SEQUENCE) {
235
            memset(output, 0, sizeof(output[0]) * 448);
236
            for (i = 448; i < 576; i++)
237
                output[i] = sce->saved[i] * pwindow[i - 448];
238
            for (i = 576; i < 704; i++)
239
                output[i] = sce->saved[i];
240
        }
241
        if (sce->ics.window_sequence[0] != LONG_START_SEQUENCE) {
242
            for (i = 0; i < 1024; i++) {
243
                output[i+1024]         = audio[i * chans] * lwindow[1024 - i - 1];
244
                sce->saved[i] = audio[i * chans] * lwindow[i];
245
            }
246
        } else {
247
            for (i = 0; i < 448; i++)
248
                output[i+1024]         = audio[i * chans];
249
            for (; i < 576; i++)
250
                output[i+1024]         = audio[i * chans] * swindow[576 - i - 1];
251
            memset(output+1024+576, 0, sizeof(output[0]) * 448);
252
            for (i = 0; i < 1024; i++)
253
                sce->saved[i] = audio[i * chans];
254
        }
255
        s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
256
    } else {
257
        for (k = 0; k < 1024; k += 128) {
258
            for (i = 448 + k; i < 448 + k + 256; i++)
259
                output[i - 448 - k] = (i < 1024)
260
                                         ? sce->saved[i]
261
                                         : audio[(i-1024)*chans];
262
            s->dsp.vector_fmul        (output,     output, k ?  swindow : pwindow, 128);
263
            s->dsp.vector_fmul_reverse(output+128, output+128, swindow, 128);
264
            s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + k, output);
265
        }
266
        for (i = 0; i < 1024; i++)
267
            sce->saved[i] = audio[i * chans];
268
    }
269
}
270

    
271
/**
272
 * Encode ics_info element.
273
 * @see Table 4.6 (syntax of ics_info)
274
 */
275
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
276
{
277
    int w;
278

    
279
    put_bits(&s->pb, 1, 0);                // ics_reserved bit
280
    put_bits(&s->pb, 2, info->window_sequence[0]);
281
    put_bits(&s->pb, 1, info->use_kb_window[0]);
282
    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
283
        put_bits(&s->pb, 6, info->max_sfb);
284
        put_bits(&s->pb, 1, 0);            // no prediction
285
    } else {
286
        put_bits(&s->pb, 4, info->max_sfb);
287
        for (w = 1; w < 8; w++)
288
            put_bits(&s->pb, 1, !info->group_len[w]);
289
    }
290
}
291

    
292
/**
293
 * Encode MS data.
294
 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
295
 */
296
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
297
{
298
    int i, w;
299

    
300
    put_bits(pb, 2, cpe->ms_mode);
301
    if (cpe->ms_mode == 1)
302
        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
303
            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
304
                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
305
}
306

    
307
/**
308
 * Produce integer coefficients from scalefactors provided by the model.
309
 */
310
static void adjust_frame_information(AACEncContext *apc, ChannelElement *cpe, int chans)
311
{
312
    int i, w, w2, g, ch;
313
    int start, maxsfb, cmaxsfb;
314

    
315
    for (ch = 0; ch < chans; ch++) {
316
        IndividualChannelStream *ics = &cpe->ch[ch].ics;
317
        start = 0;
318
        maxsfb = 0;
319
        cpe->ch[ch].pulse.num_pulse = 0;
320
        for (w = 0; w < ics->num_windows*16; w += 16) {
321
            for (g = 0; g < ics->num_swb; g++) {
322
                //apply M/S
323
                if (cpe->common_window && !ch && cpe->ms_mask[w + g]) {
324
                    for (i = 0; i < ics->swb_sizes[g]; i++) {
325
                        cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0;
326
                        cpe->ch[1].coeffs[start+i] =  cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i];
327
                    }
328
                }
329
                start += ics->swb_sizes[g];
330
            }
331
            for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--)
332
                ;
333
            maxsfb = FFMAX(maxsfb, cmaxsfb);
334
        }
335
        ics->max_sfb = maxsfb;
336

    
337
        //adjust zero bands for window groups
338
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
339
            for (g = 0; g < ics->max_sfb; g++) {
340
                i = 1;
341
                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
342
                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
343
                        i = 0;
344
                        break;
345
                    }
346
                }
347
                cpe->ch[ch].zeroes[w*16 + g] = i;
348
            }
349
        }
350
    }
351

    
352
    if (chans > 1 && cpe->common_window) {
353
        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
354
        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
355
        int msc = 0;
356
        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
357
        ics1->max_sfb = ics0->max_sfb;
358
        for (w = 0; w < ics0->num_windows*16; w += 16)
359
            for (i = 0; i < ics0->max_sfb; i++)
360
                if (cpe->ms_mask[w+i])
361
                    msc++;
362
        if (msc == 0 || ics0->max_sfb == 0)
363
            cpe->ms_mode = 0;
364
        else
365
            cpe->ms_mode = msc < ics0->max_sfb ? 1 : 2;
366
    }
367
}
368

    
369
/**
370
 * Encode scalefactor band coding type.
371
 */
372
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
373
{
374
    int w;
375

    
376
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
377
        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
378
}
379

    
380
/**
381
 * Encode scalefactors.
382
 */
383
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
384
                                 SingleChannelElement *sce)
385
{
386
    int off = sce->sf_idx[0], diff;
387
    int i, w;
388

    
389
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
390
        for (i = 0; i < sce->ics.max_sfb; i++) {
391
            if (!sce->zeroes[w*16 + i]) {
392
                diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO;
393
                if (diff < 0 || diff > 120)
394
                    av_log(avctx, AV_LOG_ERROR, "Scalefactor difference is too big to be coded\n");
395
                off = sce->sf_idx[w*16 + i];
396
                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
397
            }
398
        }
399
    }
400
}
401

    
402
/**
403
 * Encode pulse data.
404
 */
405
static void encode_pulses(AACEncContext *s, Pulse *pulse)
406
{
407
    int i;
408

    
409
    put_bits(&s->pb, 1, !!pulse->num_pulse);
410
    if (!pulse->num_pulse)
411
        return;
412

    
413
    put_bits(&s->pb, 2, pulse->num_pulse - 1);
414
    put_bits(&s->pb, 6, pulse->start);
415
    for (i = 0; i < pulse->num_pulse; i++) {
416
        put_bits(&s->pb, 5, pulse->pos[i]);
417
        put_bits(&s->pb, 4, pulse->amp[i]);
418
    }
419
}
420

    
421
/**
422
 * Encode spectral coefficients processed by psychoacoustic model.
423
 */
424
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
425
{
426
    int start, i, w, w2;
427

    
428
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
429
        start = 0;
430
        for (i = 0; i < sce->ics.max_sfb; i++) {
431
            if (sce->zeroes[w*16 + i]) {
432
                start += sce->ics.swb_sizes[i];
433
                continue;
434
            }
435
            for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
436
                s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
437
                                                   sce->ics.swb_sizes[i],
438
                                                   sce->sf_idx[w*16 + i],
439
                                                   sce->band_type[w*16 + i],
440
                                                   s->lambda);
441
            start += sce->ics.swb_sizes[i];
442
        }
443
    }
444
}
445

    
446
/**
447
 * Encode one channel of audio data.
448
 */
449
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
450
                                     SingleChannelElement *sce,
451
                                     int common_window)
452
{
453
    put_bits(&s->pb, 8, sce->sf_idx[0]);
454
    if (!common_window)
455
        put_ics_info(s, &sce->ics);
456
    encode_band_info(s, sce);
457
    encode_scale_factors(avctx, s, sce);
458
    encode_pulses(s, &sce->pulse);
459
    put_bits(&s->pb, 1, 0); //tns
460
    put_bits(&s->pb, 1, 0); //ssr
461
    encode_spectral_coeffs(s, sce);
462
    return 0;
463
}
464

    
465
/**
466
 * Write some auxiliary information about the created AAC file.
467
 */
468
static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s,
469
                               const char *name)
470
{
471
    int i, namelen, padbits;
472

    
473
    namelen = strlen(name) + 2;
474
    put_bits(&s->pb, 3, TYPE_FIL);
475
    put_bits(&s->pb, 4, FFMIN(namelen, 15));
476
    if (namelen >= 15)
477
        put_bits(&s->pb, 8, namelen - 16);
478
    put_bits(&s->pb, 4, 0); //extension type - filler
479
    padbits = 8 - (put_bits_count(&s->pb) & 7);
480
    align_put_bits(&s->pb);
481
    for (i = 0; i < namelen - 2; i++)
482
        put_bits(&s->pb, 8, name[i]);
483
    put_bits(&s->pb, 12 - padbits, 0);
484
}
485

    
486
static int aac_encode_frame(AVCodecContext *avctx,
487
                            uint8_t *frame, int buf_size, void *data)
488
{
489
    AACEncContext *s = avctx->priv_data;
490
    int16_t *samples = s->samples, *samples2, *la;
491
    ChannelElement *cpe;
492
    int i, ch, w, chans, tag, start_ch;
493
    const uint8_t *chan_map = aac_chan_configs[avctx->channels-1];
494
    int chan_el_counter[4];
495
    FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
496

    
497
    if (s->last_frame)
498
        return 0;
499
    if (data) {
500
        if (!s->psypp) {
501
            memcpy(s->samples + 1024 * avctx->channels, data,
502
                   1024 * avctx->channels * sizeof(s->samples[0]));
503
        } else {
504
            start_ch = 0;
505
            samples2 = s->samples + 1024 * avctx->channels;
506
            for (i = 0; i < chan_map[0]; i++) {
507
                tag = chan_map[i+1];
508
                chans = tag == TYPE_CPE ? 2 : 1;
509
                ff_psy_preprocess(s->psypp, (uint16_t*)data + start_ch,
510
                                  samples2 + start_ch, start_ch, chans);
511
                start_ch += chans;
512
            }
513
        }
514
    }
515
    if (!avctx->frame_number) {
516
        memcpy(s->samples, s->samples + 1024 * avctx->channels,
517
               1024 * avctx->channels * sizeof(s->samples[0]));
518
        return 0;
519
    }
520

    
521
    start_ch = 0;
522
    for (i = 0; i < chan_map[0]; i++) {
523
        FFPsyWindowInfo* wi = windows + start_ch;
524
        tag      = chan_map[i+1];
525
        chans    = tag == TYPE_CPE ? 2 : 1;
526
        cpe      = &s->cpe[i];
527
        for (ch = 0; ch < chans; ch++) {
528
            IndividualChannelStream *ics = &cpe->ch[ch].ics;
529
            int cur_channel = start_ch + ch;
530
            samples2 = samples + cur_channel;
531
            la       = samples2 + (448+64) * avctx->channels;
532
            if (!data)
533
                la = NULL;
534
            if (tag == TYPE_LFE) {
535
                wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
536
                wi[ch].window_shape   = 0;
537
                wi[ch].num_windows    = 1;
538
                wi[ch].grouping[0]    = 1;
539
            } else {
540
                wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
541
                                              ics->window_sequence[0]);
542
            }
543
            ics->window_sequence[1] = ics->window_sequence[0];
544
            ics->window_sequence[0] = wi[ch].window_type[0];
545
            ics->use_kb_window[1]   = ics->use_kb_window[0];
546
            ics->use_kb_window[0]   = wi[ch].window_shape;
547
            ics->num_windows        = wi[ch].num_windows;
548
            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
549
            ics->num_swb            = tag == TYPE_LFE ? 12 : s->psy.num_bands[ics->num_windows == 8];
550
            for (w = 0; w < ics->num_windows; w++)
551
                ics->group_len[w] = wi[ch].grouping[w];
552

    
553
            apply_window_and_mdct(avctx, s, &cpe->ch[ch], samples2);
554
        }
555
        start_ch += chans;
556
    }
557
    do {
558
        int frame_bits;
559
        init_put_bits(&s->pb, frame, buf_size*8);
560
        if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT))
561
            put_bitstream_info(avctx, s, LIBAVCODEC_IDENT);
562
        start_ch = 0;
563
        memset(chan_el_counter, 0, sizeof(chan_el_counter));
564
        for (i = 0; i < chan_map[0]; i++) {
565
            FFPsyWindowInfo* wi = windows + start_ch;
566
            tag      = chan_map[i+1];
567
            chans    = tag == TYPE_CPE ? 2 : 1;
568
            cpe      = &s->cpe[i];
569
            put_bits(&s->pb, 3, tag);
570
            put_bits(&s->pb, 4, chan_el_counter[tag]++);
571
            for (ch = 0; ch < chans; ch++) {
572
                s->cur_channel = start_ch + ch;
573
                s->psy.model->analyze(&s->psy, s->cur_channel, cpe->ch[ch].coeffs, &wi[ch]);
574
                s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
575
            }
576
            cpe->common_window = 0;
577
            if (chans > 1
578
                && wi[0].window_type[0] == wi[1].window_type[0]
579
                && wi[0].window_shape   == wi[1].window_shape) {
580

    
581
                cpe->common_window = 1;
582
                for (w = 0; w < wi[0].num_windows; w++) {
583
                    if (wi[0].grouping[w] != wi[1].grouping[w]) {
584
                        cpe->common_window = 0;
585
                        break;
586
                    }
587
                }
588
            }
589
            s->cur_channel = start_ch;
590
            if (cpe->common_window && s->coder->search_for_ms)
591
                s->coder->search_for_ms(s, cpe, s->lambda);
592
            adjust_frame_information(s, cpe, chans);
593
            if (chans == 2) {
594
                put_bits(&s->pb, 1, cpe->common_window);
595
                if (cpe->common_window) {
596
                    put_ics_info(s, &cpe->ch[0].ics);
597
                    encode_ms_info(&s->pb, cpe);
598
                }
599
            }
600
            for (ch = 0; ch < chans; ch++) {
601
                s->cur_channel = start_ch + ch;
602
                encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
603
            }
604
            start_ch += chans;
605
        }
606

    
607
        frame_bits = put_bits_count(&s->pb);
608
        if (frame_bits <= 6144 * avctx->channels - 3) {
609
            s->psy.bitres.bits = frame_bits / avctx->channels;
610
            break;
611
        }
612

    
613
        s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
614

    
615
    } while (1);
616

    
617
    put_bits(&s->pb, 3, TYPE_END);
618
    flush_put_bits(&s->pb);
619
    avctx->frame_bits = put_bits_count(&s->pb);
620

    
621
    // rate control stuff
622
    if (!(avctx->flags & CODEC_FLAG_QSCALE)) {
623
        float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
624
        s->lambda *= ratio;
625
        s->lambda = FFMIN(s->lambda, 65536.f);
626
    }
627

    
628
    if (!data)
629
        s->last_frame = 1;
630
    memcpy(s->samples, s->samples + 1024 * avctx->channels,
631
           1024 * avctx->channels * sizeof(s->samples[0]));
632
    return put_bits_count(&s->pb)>>3;
633
}
634

    
635
static av_cold int aac_encode_end(AVCodecContext *avctx)
636
{
637
    AACEncContext *s = avctx->priv_data;
638

    
639
    ff_mdct_end(&s->mdct1024);
640
    ff_mdct_end(&s->mdct128);
641
    ff_psy_end(&s->psy);
642
    ff_psy_preprocess_end(s->psypp);
643
    av_freep(&s->samples);
644
    av_freep(&s->cpe);
645
    return 0;
646
}
647

    
648
AVCodec ff_aac_encoder = {
649
    "aac",
650
    AVMEDIA_TYPE_AUDIO,
651
    CODEC_ID_AAC,
652
    sizeof(AACEncContext),
653
    aac_encode_init,
654
    aac_encode_frame,
655
    aac_encode_end,
656
    .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
657
    .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
658
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
659
};