GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/aacenc.c Lines: 559 658 85.0 %
Date: 2020-09-21 17:35:45 Branches: 357 486 73.5 %

Line Branch Exec Source
1
/*
2
 * AAC encoder
3
 * Copyright (C) 2008 Konstantin Shishkov
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21
22
/**
23
 * @file
24
 * AAC encoder
25
 */
26
27
/***********************************
28
 *              TODOs:
29
 * add sane pulse detection
30
 ***********************************/
31
32
#include "libavutil/libm.h"
33
#include "libavutil/thread.h"
34
#include "libavutil/float_dsp.h"
35
#include "libavutil/opt.h"
36
#include "avcodec.h"
37
#include "put_bits.h"
38
#include "internal.h"
39
#include "mpeg4audio.h"
40
#include "kbdwin.h"
41
#include "sinewin.h"
42
#include "profiles.h"
43
44
#include "aac.h"
45
#include "aactab.h"
46
#include "aacenc.h"
47
#include "aacenctab.h"
48
#include "aacenc_utils.h"
49
50
#include "psymodel.h"
51
52
static AVOnce aac_table_init = AV_ONCE_INIT;
53
54
static void put_pce(PutBitContext *pb, AVCodecContext *avctx)
55
{
56
    int i, j;
57
    AACEncContext *s = avctx->priv_data;
58
    AACPCEInfo *pce = &s->pce;
59
    const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
60
    const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
61
62
    put_bits(pb, 4, 0);
63
64
    put_bits(pb, 2, avctx->profile);
65
    put_bits(pb, 4, s->samplerate_index);
66
67
    put_bits(pb, 4, pce->num_ele[0]); /* Front */
68
    put_bits(pb, 4, pce->num_ele[1]); /* Side */
69
    put_bits(pb, 4, pce->num_ele[2]); /* Back */
70
    put_bits(pb, 2, pce->num_ele[3]); /* LFE */
71
    put_bits(pb, 3, 0); /* Assoc data */
72
    put_bits(pb, 4, 0); /* CCs */
73
74
    put_bits(pb, 1, 0); /* Stereo mixdown */
75
    put_bits(pb, 1, 0); /* Mono mixdown */
76
    put_bits(pb, 1, 0); /* Something else */
77
78
    for (i = 0; i < 4; i++) {
79
        for (j = 0; j < pce->num_ele[i]; j++) {
80
            if (i < 3)
81
                put_bits(pb, 1, pce->pairing[i][j]);
82
            put_bits(pb, 4, pce->index[i][j]);
83
        }
84
    }
85
86
    avpriv_align_put_bits(pb);
87
    put_bits(pb, 8, strlen(aux_data));
88
    avpriv_put_string(pb, aux_data, 0);
89
}
90
91
/**
92
 * Make AAC audio config object.
93
 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
94
 */
95
11
static int put_audio_specific_config(AVCodecContext *avctx)
96
{
97
    PutBitContext pb;
98
11
    AACEncContext *s = avctx->priv_data;
99
11
    int channels = (!s->needs_pce)*(s->channels - (s->channels == 8 ? 1 : 0));
100
11
    const int max_size = 32;
101
102
11
    avctx->extradata = av_mallocz(max_size);
103
11
    if (!avctx->extradata)
104
        return AVERROR(ENOMEM);
105
106
11
    init_put_bits(&pb, avctx->extradata, max_size);
107
11
    put_bits(&pb, 5, s->profile+1); //profile
108
11
    put_bits(&pb, 4, s->samplerate_index); //sample rate index
109
11
    put_bits(&pb, 4, channels);
110
    //GASpecificConfig
111
11
    put_bits(&pb, 1, 0); //frame length - 1024 samples
112
11
    put_bits(&pb, 1, 0); //does not depend on core coder
113
11
    put_bits(&pb, 1, 0); //is not extension
114
11
    if (s->needs_pce)
115
        put_pce(&pb, avctx);
116
117
    //Explicitly Mark SBR absent
118
11
    put_bits(&pb, 11, 0x2b7); //sync extension
119
11
    put_bits(&pb, 5,  AOT_SBR);
120
11
    put_bits(&pb, 1,  0);
121
11
    flush_put_bits(&pb);
122
11
    avctx->extradata_size = put_bits_count(&pb) >> 3;
123
124
11
    return 0;
125
}
126
127
11647
void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
128
{
129
11647
    ++s->quantize_band_cost_cache_generation;
130
11647
    if (s->quantize_band_cost_cache_generation == 0) {
131
        memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
132
        s->quantize_band_cost_cache_generation = 1;
133
    }
134
11647
}
135
136
#define WINDOW_FUNC(type) \
137
static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
138
                                    SingleChannelElement *sce, \
139
                                    const float *audio)
140
141
6899
WINDOW_FUNC(only_long)
142
{
143
6899
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
144
6899
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
145
6899
    float *out = sce->ret_buf;
146
147
6899
    fdsp->vector_fmul        (out,        audio,        lwindow, 1024);
148
6899
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
149
6899
}
150
151
119
WINDOW_FUNC(long_start)
152
{
153
119
    const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
154
119
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
155
119
    float *out = sce->ret_buf;
156
157
119
    fdsp->vector_fmul(out, audio, lwindow, 1024);
158
119
    memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
159
119
    fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
160
119
    memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
161
119
}
162
163
102
WINDOW_FUNC(long_stop)
164
{
165
102
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
166
102
    const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
167
102
    float *out = sce->ret_buf;
168
169
102
    memset(out, 0, sizeof(out[0]) * 448);
170
102
    fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
171
102
    memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
172
102
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
173
102
}
174
175
164
WINDOW_FUNC(eight_short)
176
{
177
164
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
178
164
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
179
164
    const float *in = audio + 448;
180
164
    float *out = sce->ret_buf;
181
    int w;
182
183
1476
    for (w = 0; w < 8; w++) {
184
1312
        fdsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
185
1312
        out += 128;
186
1312
        in  += 128;
187
1312
        fdsp->vector_fmul_reverse(out, in, swindow, 128);
188
1312
        out += 128;
189
    }
190
164
}
191
192
static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
193
                                     SingleChannelElement *sce,
194
                                     const float *audio) = {
195
    [ONLY_LONG_SEQUENCE]   = apply_only_long_window,
196
    [LONG_START_SEQUENCE]  = apply_long_start_window,
197
    [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
198
    [LONG_STOP_SEQUENCE]   = apply_long_stop_window
199
};
200
201
7284
static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
202
                                  float *audio)
203
{
204
    int i;
205
7284
    const float *output = sce->ret_buf;
206
207
7284
    apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
208
209
7284
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
210
7120
        s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
211
    else
212
1476
        for (i = 0; i < 1024; i += 128)
213
1312
            s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
214
7284
    memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
215
7284
    memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
216
7284
}
217
218
/**
219
 * Encode ics_info element.
220
 * @see Table 4.6 (syntax of ics_info)
221
 */
222
6714
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
223
{
224
    int w;
225
226
6714
    put_bits(&s->pb, 1, 0);                // ics_reserved bit
227
6714
    put_bits(&s->pb, 2, info->window_sequence[0]);
228
6714
    put_bits(&s->pb, 1, info->use_kb_window[0]);
229
6714
    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
230
6527
        put_bits(&s->pb, 6, info->max_sfb);
231
6527
        put_bits(&s->pb, 1, !!info->predictor_present);
232
    } else {
233
187
        put_bits(&s->pb, 4, info->max_sfb);
234
1496
        for (w = 1; w < 8; w++)
235
1309
            put_bits(&s->pb, 1, !info->group_len[w]);
236
    }
237
6714
}
238
239
/**
240
 * Encode MS data.
241
 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
242
 */
243
5029
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
244
{
245
    int i, w;
246
247
5029
    put_bits(pb, 2, cpe->ms_mode);
248
5029
    if (cpe->ms_mode == 1)
249
1812
        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
250
40817
            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
251
39892
                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
252
5029
}
253
254
/**
255
 * Produce integer coefficients from scalefactors provided by the model.
256
 */
257
6469
static void adjust_frame_information(ChannelElement *cpe, int chans)
258
{
259
    int i, w, w2, g, ch;
260
    int maxsfb, cmaxsfb;
261
262
18212
    for (ch = 0; ch < chans; ch++) {
263
11743
        IndividualChannelStream *ics = &cpe->ch[ch].ics;
264
11743
        maxsfb = 0;
265
11743
        cpe->ch[ch].pulse.num_pulse = 0;
266
24173
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
267
26049
            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
268

41764
                for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
269
                    ;
270
13619
                maxsfb = FFMAX(maxsfb, cmaxsfb);
271
            }
272
        }
273
11743
        ics->max_sfb = maxsfb;
274
275
        //adjust zero bands for window groups
276
24173
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
277
563969
            for (g = 0; g < ics->max_sfb; g++) {
278
551539
                i = 1;
279
567030
                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
280
551879
                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
281
536388
                        i = 0;
282
536388
                        break;
283
                    }
284
                }
285
551539
                cpe->ch[ch].zeroes[w*16 + g] = i;
286
            }
287
        }
288
    }
289
290

6469
    if (chans > 1 && cpe->common_window) {
291
5029
        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
292
5029
        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
293
5029
        int msc = 0;
294
5029
        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
295
5029
        ics1->max_sfb = ics0->max_sfb;
296
10625
        for (w = 0; w < ics0->num_windows*16; w += 16)
297
247534
            for (i = 0; i < ics0->max_sfb; i++)
298
241938
                if (cpe->ms_mask[w+i])
299
30077
                    msc++;
300

5029
        if (msc == 0 || ics0->max_sfb == 0)
301
3749
            cpe->ms_mode = 0;
302
        else
303
1280
            cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
304
    }
305
6469
}
306
307
2355
static void apply_intensity_stereo(ChannelElement *cpe)
308
{
309
    int w, w2, g, i;
310
2355
    IndividualChannelStream *ics = &cpe->ch[0].ics;
311
2355
    if (!cpe->common_window)
312
1237
        return;
313
2289
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
314
2443
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
315
1272
            int start = (w+w2) * 128;
316
57440
            for (g = 0; g < ics->num_swb; g++) {
317
56168
                int p  = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
318
56168
                float scale = cpe->ch[0].is_ener[w*16+g];
319
56168
                if (!cpe->is_mask[w*16 + g]) {
320
47882
                    start += ics->swb_sizes[g];
321
47882
                    continue;
322
                }
323
8286
                if (cpe->ms_mask[w*16 + g])
324
2383
                    p *= -1;
325
288446
                for (i = 0; i < ics->swb_sizes[g]; i++) {
326
280160
                    float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
327
280160
                    cpe->ch[0].coeffs[start+i] = sum;
328
280160
                    cpe->ch[1].coeffs[start+i] = 0.0f;
329
                }
330
8286
                start += ics->swb_sizes[g];
331
            }
332
        }
333
    }
334
}
335
336
2071
static void apply_mid_side_stereo(ChannelElement *cpe)
337
{
338
    int w, w2, g, i;
339
2071
    IndividualChannelStream *ics = &cpe->ch[0].ics;
340
2071
    if (!cpe->common_window)
341
1078
        return;
342
2039
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
343
2193
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
344
1147
            int start = (w+w2) * 128;
345
51190
            for (g = 0; g < ics->num_swb; g++) {
346
                /* ms_mask can be used for other purposes in PNS and I/S,
347
                 * so must not apply M/S if any band uses either, even if
348
                 * ms_mask is set.
349
                 */
350

50043
                if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]
351
28910
                    || cpe->ch[0].band_type[w*16 + g] >= NOISE_BT
352
28910
                    || cpe->ch[1].band_type[w*16 + g] >= NOISE_BT) {
353
21133
                    start += ics->swb_sizes[g];
354
21133
                    continue;
355
                }
356
650946
                for (i = 0; i < ics->swb_sizes[g]; i++) {
357
622036
                    float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
358
622036
                    float R = L - cpe->ch[1].coeffs[start+i];
359
622036
                    cpe->ch[0].coeffs[start+i] = L;
360
622036
                    cpe->ch[1].coeffs[start+i] = R;
361
                }
362
28910
                start += ics->swb_sizes[g];
363
            }
364
        }
365
    }
366
}
367
368
/**
369
 * Encode scalefactor band coding type.
370
 */
371
11743
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
372
{
373
    int w;
374
375
11743
    if (s->coder->set_special_band_scalefactors)
376
11743
        s->coder->set_special_band_scalefactors(s, sce);
377
378
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
379
12430
        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
380
11743
}
381
382
/**
383
 * Encode scalefactors.
384
 */
385
11743
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
386
                                 SingleChannelElement *sce)
387
{
388
11743
    int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
389
11743
    int off_is = 0, noise_flag = 1;
390
    int i, w;
391
392
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
393
563974
        for (i = 0; i < sce->ics.max_sfb; i++) {
394
551544
            if (!sce->zeroes[w*16 + i]) {
395
521143
                if (sce->band_type[w*16 + i] == NOISE_BT) {
396
17340
                    diff = sce->sf_idx[w*16 + i] - off_pns;
397
17340
                    off_pns = sce->sf_idx[w*16 + i];
398
17340
                    if (noise_flag-- > 0) {
399
2242
                        put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
400
2242
                        continue;
401
                    }
402
503803
                } else if (sce->band_type[w*16 + i] == INTENSITY_BT  ||
403
497850
                           sce->band_type[w*16 + i] == INTENSITY_BT2) {
404
8191
                    diff = sce->sf_idx[w*16 + i] - off_is;
405
8191
                    off_is = sce->sf_idx[w*16 + i];
406
                } else {
407
495612
                    diff = sce->sf_idx[w*16 + i] - off_sf;
408
495612
                    off_sf = sce->sf_idx[w*16 + i];
409
                }
410
518901
                diff += SCALE_DIFF_ZERO;
411

518901
                av_assert0(diff >= 0 && diff <= 120);
412
518901
                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
413
            }
414
        }
415
    }
416
11743
}
417
418
/**
419
 * Encode pulse data.
420
 */
421
11743
static void encode_pulses(AACEncContext *s, Pulse *pulse)
422
{
423
    int i;
424
425
11743
    put_bits(&s->pb, 1, !!pulse->num_pulse);
426
11743
    if (!pulse->num_pulse)
427
11743
        return;
428
429
    put_bits(&s->pb, 2, pulse->num_pulse - 1);
430
    put_bits(&s->pb, 6, pulse->start);
431
    for (i = 0; i < pulse->num_pulse; i++) {
432
        put_bits(&s->pb, 5, pulse->pos[i]);
433
        put_bits(&s->pb, 4, pulse->amp[i]);
434
    }
435
}
436
437
/**
438
 * Encode spectral coefficients processed by psychoacoustic model.
439
 */
440
11743
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
441
{
442
    int start, i, w, w2;
443
444
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
445
12430
        start = 0;
446
563974
        for (i = 0; i < sce->ics.max_sfb; i++) {
447
551544
            if (sce->zeroes[w*16 + i]) {
448
30401
                start += sce->ics.swb_sizes[i];
449
30401
                continue;
450
            }
451
1050888
            for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
452
529745
                s->coder->quantize_and_encode_band(s, &s->pb,
453
529745
                                                   &sce->coeffs[start + w2*128],
454
529745
                                                   NULL, sce->ics.swb_sizes[i],
455
529745
                                                   sce->sf_idx[w*16 + i],
456
529745
                                                   sce->band_type[w*16 + i],
457
                                                   s->lambda,
458
529745
                                                   sce->ics.window_clipping[w]);
459
            }
460
521143
            start += sce->ics.swb_sizes[i];
461
        }
462
    }
463
11743
}
464
465
/**
466
 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
467
 */
468
7284
static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
469
{
470
    int start, i, j, w;
471
472
7284
    if (sce->ics.clip_avoidance_factor < 1.0f) {
473
392
        for (w = 0; w < sce->ics.num_windows; w++) {
474
224
            start = 0;
475
8645
            for (i = 0; i < sce->ics.max_sfb; i++) {
476
8421
                float *swb_coeffs = &sce->coeffs[start + w*128];
477
167173
                for (j = 0; j < sce->ics.swb_sizes[i]; j++)
478
158752
                    swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
479
8421
                start += sce->ics.swb_sizes[i];
480
            }
481
        }
482
    }
483
7284
}
484
485
/**
486
 * Encode one channel of audio data.
487
 */
488
11743
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
489
                                     SingleChannelElement *sce,
490
                                     int common_window)
491
{
492
11743
    put_bits(&s->pb, 8, sce->sf_idx[0]);
493
11743
    if (!common_window) {
494
1685
        put_ics_info(s, &sce->ics);
495
1685
        if (s->coder->encode_main_pred)
496
1685
            s->coder->encode_main_pred(s, sce);
497
1685
        if (s->coder->encode_ltp_info)
498
1685
            s->coder->encode_ltp_info(s, sce, 0);
499
    }
500
11743
    encode_band_info(s, sce);
501
11743
    encode_scale_factors(avctx, s, sce);
502
11743
    encode_pulses(s, &sce->pulse);
503
11743
    put_bits(&s->pb, 1, !!sce->tns.present);
504
11743
    if (s->coder->encode_tns_info)
505
11743
        s->coder->encode_tns_info(s, sce);
506
11743
    put_bits(&s->pb, 1, 0); //ssr
507
11743
    encode_spectral_coeffs(s, sce);
508
11743
    return 0;
509
}
510
511
/**
512
 * Write some auxiliary information about the created AAC file.
513
 */
514
static void put_bitstream_info(AACEncContext *s, const char *name)
515
{
516
    int i, namelen, padbits;
517
518
    namelen = strlen(name) + 2;
519
    put_bits(&s->pb, 3, TYPE_FIL);
520
    put_bits(&s->pb, 4, FFMIN(namelen, 15));
521
    if (namelen >= 15)
522
        put_bits(&s->pb, 8, namelen - 14);
523
    put_bits(&s->pb, 4, 0); //extension type - filler
524
    padbits = -put_bits_count(&s->pb) & 7;
525
    avpriv_align_put_bits(&s->pb);
526
    for (i = 0; i < namelen - 2; i++)
527
        put_bits(&s->pb, 8, name[i]);
528
    put_bits(&s->pb, 12 - padbits, 0);
529
}
530
531
/*
532
 * Copy input samples.
533
 * Channels are reordered from libavcodec's default order to AAC order.
534
 */
535
3816
static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
536
{
537
    int ch;
538
3816
    int end = 2048 + (frame ? frame->nb_samples : 0);
539
3816
    const uint8_t *channel_map = s->reorder_map;
540
541
    /* copy and remap input samples */
542
11125
    for (ch = 0; ch < s->channels; ch++) {
543
        /* copy last 1024 samples of previous frame to the start of the current frame */
544
7309
        memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
545
546
        /* copy new samples and zero any remaining samples */
547
7309
        if (frame) {
548
7259
            memcpy(&s->planar_samples[ch][2048],
549
7259
                   frame->extended_data[channel_map[ch]],
550
7259
                   frame->nb_samples * sizeof(s->planar_samples[0][0]));
551
        }
552
7309
        memset(&s->planar_samples[ch][end], 0,
553
7309
               (3072 - end) * sizeof(s->planar_samples[0][0]));
554
    }
555
3816
}
556
557
3827
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
558
                            const AVFrame *frame, int *got_packet_ptr)
559
{
560
3827
    AACEncContext *s = avctx->priv_data;
561
3827
    float **samples = s->planar_samples, *samples2, *la, *overlap;
562
    ChannelElement *cpe;
563
    SingleChannelElement *sce;
564
    IndividualChannelStream *ics;
565
    int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
566
    int target_bits, rate_bits, too_many_bits, too_few_bits;
567
3827
    int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
568
    int chan_el_counter[4];
569
    FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
570
571
    /* add current frame to queue */
572
3827
    if (frame) {
573
3794
        if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
574
            return ret;
575
    } else {
576

33
        if (!s->afq.remaining_samples || (!s->afq.frame_alloc && !s->afq.frame_count))
577
11
            return 0;
578
    }
579
580
3816
    copy_input_samples(s, frame);
581
3816
    if (s->psypp)
582
3816
        ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
583
584
3816
    if (!avctx->frame_number)
585
11
        return 0;
586
587
3805
    start_ch = 0;
588
7754
    for (i = 0; i < s->chan_map[0]; i++) {
589
3949
        FFPsyWindowInfo* wi = windows + start_ch;
590
3949
        tag      = s->chan_map[i+1];
591
3949
        chans    = tag == TYPE_CPE ? 2 : 1;
592
3949
        cpe      = &s->cpe[i];
593
11233
        for (ch = 0; ch < chans; ch++) {
594
            int k;
595
            float clip_avoidance_factor;
596
7284
            sce = &cpe->ch[ch];
597
7284
            ics = &sce->ics;
598
7284
            s->cur_channel = start_ch + ch;
599
7284
            overlap  = &samples[s->cur_channel][0];
600
7284
            samples2 = overlap + 1024;
601
7284
            la       = samples2 + (448+64);
602
7284
            if (!frame)
603
50
                la = NULL;
604
7284
            if (tag == TYPE_LFE) {
605
48
                wi[ch].window_type[0] = wi[ch].window_type[1] = ONLY_LONG_SEQUENCE;
606
48
                wi[ch].window_shape   = 0;
607
48
                wi[ch].num_windows    = 1;
608
48
                wi[ch].grouping[0]    = 1;
609
48
                wi[ch].clipping[0]    = 0;
610
611
                /* Only the lowest 12 coefficients are used in a LFE channel.
612
                 * The expression below results in only the bottom 8 coefficients
613
                 * being used for 11.025kHz to 16kHz sample rates.
614
                 */
615
48
                ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
616
            } else {
617
7236
                wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
618
7236
                                              ics->window_sequence[0]);
619
            }
620
7284
            ics->window_sequence[1] = ics->window_sequence[0];
621
7284
            ics->window_sequence[0] = wi[ch].window_type[0];
622
7284
            ics->use_kb_window[1]   = ics->use_kb_window[0];
623
7284
            ics->use_kb_window[0]   = wi[ch].window_shape;
624
7284
            ics->num_windows        = wi[ch].num_windows;
625
7284
            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
626

7284
            ics->num_swb            = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
627
7284
            ics->max_sfb            = FFMIN(ics->max_sfb, ics->num_swb);
628
14568
            ics->swb_offset         = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
629
7284
                                        ff_swb_offset_128 [s->samplerate_index]:
630
7120
                                        ff_swb_offset_1024[s->samplerate_index];
631
14568
            ics->tns_max_bands      = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
632
7284
                                        ff_tns_max_bands_128 [s->samplerate_index]:
633
7120
                                        ff_tns_max_bands_1024[s->samplerate_index];
634
635
15716
            for (w = 0; w < ics->num_windows; w++)
636
8432
                ics->group_len[w] = wi[ch].grouping[w];
637
638
            /* Calculate input sample maximums and evaluate clipping risk */
639
7284
            clip_avoidance_factor = 0.0f;
640
15716
            for (w = 0; w < ics->num_windows; w++) {
641
8432
                const float *wbuf = overlap + w * 128;
642
8432
                const int wlen = 2048 / ics->num_windows;
643
8432
                float max = 0;
644
                int j;
645
                /* mdct input is 2 * output */
646
14926064
                for (j = 0; j < wlen; j++)
647
14917632
                    max = FFMAX(max, fabsf(wbuf[j]));
648
8432
                wi[ch].clipping[w] = max;
649
            }
650
15716
            for (w = 0; w < ics->num_windows; w++) {
651
8432
                if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
652
176
                    ics->window_clipping[w] = 1;
653
176
                    clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
654
                } else {
655
8256
                    ics->window_clipping[w] = 0;
656
                }
657
            }
658
7284
            if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
659
168
                ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
660
            } else {
661
7116
                ics->clip_avoidance_factor = 1.0f;
662
            }
663
664
7284
            apply_window_and_mdct(s, sce, overlap);
665
666

7284
            if (s->options.ltp && s->coder->update_ltp) {
667
                s->coder->update_ltp(s, sce);
668
                apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
669
                s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);
670
            }
671
672
7466100
            for (k = 0; k < 1024; k++) {
673
7458816
                if (!(fabs(cpe->ch[ch].coeffs[k]) < 1E16)) { // Ensure headroom for energy calculation
674
                    av_log(avctx, AV_LOG_ERROR, "Input contains (near) NaN/+-Inf\n");
675
                    return AVERROR(EINVAL);
676
                }
677
            }
678
7284
            avoid_clipping(s, sce);
679
        }
680
3949
        start_ch += chans;
681
    }
682
3805
    if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
683
        return ret;
684
3805
    frame_bits = its = 0;
685
    do {
686
6253
        init_put_bits(&s->pb, avpkt->data, avpkt->size);
687
688

6253
        if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
689
            put_bitstream_info(s, LIBAVCODEC_IDENT);
690
6253
        start_ch = 0;
691
6253
        target_bits = 0;
692
6253
        memset(chan_el_counter, 0, sizeof(chan_el_counter));
693
12722
        for (i = 0; i < s->chan_map[0]; i++) {
694
6469
            FFPsyWindowInfo* wi = windows + start_ch;
695
            const float *coeffs[2];
696
6469
            tag      = s->chan_map[i+1];
697
6469
            chans    = tag == TYPE_CPE ? 2 : 1;
698
6469
            cpe      = &s->cpe[i];
699
6469
            cpe->common_window = 0;
700
6469
            memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
701
6469
            memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
702
6469
            put_bits(&s->pb, 3, tag);
703
6469
            put_bits(&s->pb, 4, chan_el_counter[tag]++);
704
18212
            for (ch = 0; ch < chans; ch++) {
705
11743
                sce = &cpe->ch[ch];
706
11743
                coeffs[ch] = sce->coeffs;
707
11743
                sce->ics.predictor_present = 0;
708
11743
                sce->ics.ltp.present = 0;
709
11743
                memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
710
11743
                memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
711
11743
                memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
712
1514847
                for (w = 0; w < 128; w++)
713
1503104
                    if (sce->band_type[w] > RESERVED_BT)
714
25469
                        sce->band_type[w] = 0;
715
            }
716
6469
            s->psy.bitres.alloc = -1;
717
6469
            s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
718
6469
            s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
719
6469
            if (s->psy.bitres.alloc > 0) {
720
                /* Lambda unused here on purpose, we need to take psy's unscaled allocation */
721
12938
                target_bits += s->psy.bitres.alloc
722
6469
                    * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
723
6469
                s->psy.bitres.alloc /= chans;
724
            }
725
6469
            s->cur_type = tag;
726
18212
            for (ch = 0; ch < chans; ch++) {
727
11743
                s->cur_channel = start_ch + ch;
728

11743
                if (s->options.pns && s->coder->mark_pns)
729
3515
                    s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
730
11743
                s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
731
            }
732
6469
            if (chans > 1
733
5274
                && wi[0].window_type[0] == wi[1].window_type[0]
734
5038
                && wi[0].window_shape   == wi[1].window_shape) {
735
736
5038
                cpe->common_window = 1;
737
10634
                for (w = 0; w < wi[0].num_windows; w++) {
738
5605
                    if (wi[0].grouping[w] != wi[1].grouping[w]) {
739
9
                        cpe->common_window = 0;
740
9
                        break;
741
                    }
742
                }
743
            }
744
18212
            for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
745
11743
                sce = &cpe->ch[ch];
746
11743
                s->cur_channel = start_ch + ch;
747

11743
                if (s->options.tns && s->coder->search_for_tns)
748
3515
                    s->coder->search_for_tns(s, sce);
749

11743
                if (s->options.tns && s->coder->apply_tns_filt)
750
3515
                    s->coder->apply_tns_filt(s, sce);
751
11743
                if (sce->tns.present)
752
46
                    tns_mode = 1;
753

11743
                if (s->options.pns && s->coder->search_for_pns)
754
3515
                    s->coder->search_for_pns(s, avctx, sce);
755
            }
756
6469
            s->cur_channel = start_ch;
757
6469
            if (s->options.intensity_stereo) { /* Intensity Stereo */
758
2355
                if (s->coder->search_for_is)
759
2355
                    s->coder->search_for_is(s, avctx, cpe);
760
2355
                if (cpe->is_mode) is_mode = 1;
761
2355
                apply_intensity_stereo(cpe);
762
            }
763
6469
            if (s->options.pred) { /* Prediction */
764
1248
                for (ch = 0; ch < chans; ch++) {
765
832
                    sce = &cpe->ch[ch];
766
832
                    s->cur_channel = start_ch + ch;
767

832
                    if (s->options.pred && s->coder->search_for_pred)
768
832
                        s->coder->search_for_pred(s, sce);
769
832
                    if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
770
                }
771
416
                if (s->coder->adjust_common_pred)
772
416
                    s->coder->adjust_common_pred(s, cpe);
773
1248
                for (ch = 0; ch < chans; ch++) {
774
832
                    sce = &cpe->ch[ch];
775
832
                    s->cur_channel = start_ch + ch;
776

832
                    if (s->options.pred && s->coder->apply_main_pred)
777
832
                        s->coder->apply_main_pred(s, sce);
778
                }
779
416
                s->cur_channel = start_ch;
780
            }
781
6469
            if (s->options.mid_side) { /* Mid/Side stereo */
782

2071
                if (s->options.mid_side == -1 && s->coder->search_for_ms)
783
1651
                    s->coder->search_for_ms(s, cpe);
784
420
                else if (cpe->common_window)
785
393
                    memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
786
2071
                apply_mid_side_stereo(cpe);
787
            }
788
6469
            adjust_frame_information(cpe, chans);
789
6469
            if (s->options.ltp) { /* LTP */
790
                for (ch = 0; ch < chans; ch++) {
791
                    sce = &cpe->ch[ch];
792
                    s->cur_channel = start_ch + ch;
793
                    if (s->coder->search_for_ltp)
794
                        s->coder->search_for_ltp(s, sce, cpe->common_window);
795
                    if (sce->ics.ltp.present) pred_mode = 1;
796
                }
797
                s->cur_channel = start_ch;
798
                if (s->coder->adjust_common_ltp)
799
                    s->coder->adjust_common_ltp(s, cpe);
800
            }
801
6469
            if (chans == 2) {
802
5274
                put_bits(&s->pb, 1, cpe->common_window);
803
5274
                if (cpe->common_window) {
804
5029
                    put_ics_info(s, &cpe->ch[0].ics);
805
5029
                    if (s->coder->encode_main_pred)
806
5029
                        s->coder->encode_main_pred(s, &cpe->ch[0]);
807
5029
                    if (s->coder->encode_ltp_info)
808
5029
                        s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
809
5029
                    encode_ms_info(&s->pb, cpe);
810
5029
                    if (cpe->ms_mode) ms_mode = 1;
811
                }
812
            }
813
18212
            for (ch = 0; ch < chans; ch++) {
814
11743
                s->cur_channel = start_ch + ch;
815
11743
                encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
816
            }
817
6469
            start_ch += chans;
818
        }
819
820
6253
        if (avctx->flags & AV_CODEC_FLAG_QSCALE) {
821
            /* When using a constant Q-scale, don't mess with lambda */
822
            break;
823
        }
824
825
        /* rate control stuff
826
         * allow between the nominal bitrate, and what psy's bit reservoir says to target
827
         * but drift towards the nominal bitrate always
828
         */
829
6253
        frame_bits = put_bits_count(&s->pb);
830
6253
        rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
831
6253
        rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
832
6253
        too_many_bits = FFMAX(target_bits, rate_bits);
833
6253
        too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
834
6253
        too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
835
836
        /* When using ABR, be strict (but only for increasing) */
837
6253
        too_few_bits = too_few_bits - too_few_bits/8;
838
6253
        too_many_bits = too_many_bits + too_many_bits/2;
839
840
6253
        if (   its == 0 /* for steady-state Q-scale tracking */
841

2448
            || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
842
497
            || frame_bits >= 6144 * s->channels - 3  )
843
        {
844
5756
            float ratio = ((float)rate_bits) / frame_bits;
845
846

5756
            if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
847
                /*
848
                 * This path is for steady-state Q-scale tracking
849
                 * When frame bits fall within the stable range, we still need to adjust
850
                 * lambda to maintain it like so in a stable fashion (large jumps in lambda
851
                 * create artifacts and should be avoided), but slowly
852
                 */
853
2859
                ratio = sqrtf(sqrtf(ratio));
854
2859
                ratio = av_clipf(ratio, 0.9f, 1.1f);
855
            } else {
856
                /* Not so fast though */
857
2897
                ratio = sqrtf(ratio);
858
            }
859
5756
            s->lambda = FFMIN(s->lambda * ratio, 65536.f);
860
861
            /* Keep iterating if we must reduce and lambda is in the sky */
862

5756
            if (ratio > 0.9f && ratio < 1.1f) {
863
                break;
864
            } else {
865


2448
                if (is_mode || ms_mode || tns_mode || pred_mode) {
866
1027
                    for (i = 0; i < s->chan_map[0]; i++) {
867
                        // Must restore coeffs
868
521
                        chans = tag == TYPE_CPE ? 2 : 1;
869
521
                        cpe = &s->cpe[i];
870
1543
                        for (ch = 0; ch < chans; ch++)
871
1022
                            memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
872
                    }
873
                }
874
2448
                its++;
875
            }
876
        } else {
877
            break;
878
        }
879
    } while (1);
880
881

3805
    if (s->options.ltp && s->coder->ltp_insert_new_frame)
882
        s->coder->ltp_insert_new_frame(s);
883
884
3805
    put_bits(&s->pb, 3, TYPE_END);
885
3805
    flush_put_bits(&s->pb);
886
887
3805
    s->last_frame_pb_count = put_bits_count(&s->pb);
888
889
3805
    s->lambda_sum += s->lambda;
890
3805
    s->lambda_count++;
891
892
3805
    ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
893
                       &avpkt->duration);
894
895
3805
    avpkt->size = put_bits_count(&s->pb) >> 3;
896
3805
    *got_packet_ptr = 1;
897
3805
    return 0;
898
}
899
900
11
static av_cold int aac_encode_end(AVCodecContext *avctx)
901
{
902
11
    AACEncContext *s = avctx->priv_data;
903
904
11
    av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);
905
906
11
    ff_mdct_end(&s->mdct1024);
907
11
    ff_mdct_end(&s->mdct128);
908
11
    ff_psy_end(&s->psy);
909
11
    ff_lpc_end(&s->lpc);
910
11
    if (s->psypp)
911
11
        ff_psy_preprocess_end(s->psypp);
912
11
    av_freep(&s->buffer.samples);
913
11
    av_freep(&s->cpe);
914
11
    av_freep(&s->fdsp);
915
11
    ff_af_queue_close(&s->afq);
916
11
    return 0;
917
}
918
919
11
static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
920
{
921
11
    int ret = 0;
922
923
11
    s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
924
11
    if (!s->fdsp)
925
        return AVERROR(ENOMEM);
926
927
    // window init
928
11
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
929
11
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
930
11
    ff_init_ff_sine_windows(10);
931
11
    ff_init_ff_sine_windows(7);
932
933
11
    if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
934
        return ret;
935
11
    if ((ret = ff_mdct_init(&s->mdct128,   8, 0, 32768.0)) < 0)
936
        return ret;
937
938
11
    return 0;
939
}
940
941
11
static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
942
{
943
    int ch;
944
11
    if (!FF_ALLOCZ_TYPED_ARRAY(s->buffer.samples, s->channels * 3 * 1024) ||
945
11
        !FF_ALLOCZ_TYPED_ARRAY(s->cpe,            s->chan_map[0]))
946
        return AVERROR(ENOMEM);
947
948
36
    for(ch = 0; ch < s->channels; ch++)
949
25
        s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
950
951
11
    return 0;
952
}
953
954
11
static av_cold void aac_encode_init_tables(void)
955
{
956
11
    ff_aac_tableinit();
957
11
}
958
959
11
static av_cold int aac_encode_init(AVCodecContext *avctx)
960
{
961
11
    AACEncContext *s = avctx->priv_data;
962
11
    int i, ret = 0;
963
    const uint8_t *sizes[2];
964
    uint8_t grouping[AAC_MAX_CHANNELS];
965
    int lengths[2];
966
967
    /* Constants */
968
11
    s->last_frame_pb_count = 0;
969
11
    avctx->frame_size = 1024;
970
11
    avctx->initial_padding = 1024;
971
11
    s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
972
973
    /* Channel map and unspecified bitrate guessing */
974
11
    s->channels = avctx->channels;
975
976
11
    s->needs_pce = 1;
977
25
    for (i = 0; i < FF_ARRAY_ELEMS(aac_normal_chan_layouts); i++) {
978
25
        if (avctx->channel_layout == aac_normal_chan_layouts[i]) {
979
11
            s->needs_pce = s->options.pce;
980
11
            break;
981
        }
982
    }
983
984
11
    if (s->needs_pce) {
985
        char buf[64];
986
        for (i = 0; i < FF_ARRAY_ELEMS(aac_pce_configs); i++)
987
            if (avctx->channel_layout == aac_pce_configs[i].layout)
988
                break;
989
        av_get_channel_layout_string(buf, sizeof(buf), -1, avctx->channel_layout);
990
        ERROR_IF(i == FF_ARRAY_ELEMS(aac_pce_configs), "Unsupported channel layout \"%s\"\n", buf);
991
        av_log(avctx, AV_LOG_INFO, "Using a PCE to encode channel layout \"%s\"\n", buf);
992
        s->pce = aac_pce_configs[i];
993
        s->reorder_map = s->pce.reorder_map;
994
        s->chan_map = s->pce.config_map;
995
    } else {
996
11
        s->reorder_map = aac_chan_maps[s->channels - 1];
997
11
        s->chan_map = aac_chan_configs[s->channels - 1];
998
    }
999
1000
11
    if (!avctx->bit_rate) {
1001
9
        for (i = 1; i <= s->chan_map[0]; i++) {
1002
9
            avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */
1003
3
                               s->chan_map[i] == TYPE_LFE ? 16000  : /* LFE  */
1004
                                                            69000  ; /* SCE  */
1005
        }
1006
    }
1007
1008
    /* Samplerate */
1009
54
    for (i = 0; i < 16; i++)
1010
54
        if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
1011
11
            break;
1012
11
    s->samplerate_index = i;
1013

11
    ERROR_IF(s->samplerate_index == 16 ||
1014
             s->samplerate_index >= ff_aac_swb_size_1024_len ||
1015
             s->samplerate_index >= ff_aac_swb_size_128_len,
1016
             "Unsupported sample rate %d\n", avctx->sample_rate);
1017
1018
    /* Bitrate limiting */
1019
11
    WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
1020
             "Too many bits %f > %d per frame requested, clamping to max\n",
1021
             1024.0 * avctx->bit_rate / avctx->sample_rate,
1022
             6144 * s->channels);
1023
11
    avctx->bit_rate = (int64_t)FFMIN(6144 * s->channels / 1024.0 * avctx->sample_rate,
1024
                                     avctx->bit_rate);
1025
1026
    /* Profile and option setting */
1027
11
    avctx->profile = avctx->profile == FF_PROFILE_UNKNOWN ? FF_PROFILE_AAC_LOW :
1028
                     avctx->profile;
1029
21
    for (i = 0; i < FF_ARRAY_ELEMS(aacenc_profiles); i++)
1030
21
        if (avctx->profile == aacenc_profiles[i])
1031
11
            break;
1032
11
    if (avctx->profile == FF_PROFILE_MPEG2_AAC_LOW) {
1033
        avctx->profile = FF_PROFILE_AAC_LOW;
1034
        ERROR_IF(s->options.pred,
1035
                 "Main prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1036
        ERROR_IF(s->options.ltp,
1037
                 "LTP prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1038
        WARN_IF(s->options.pns,
1039
                "PNS unavailable in the \"mpeg2_aac_low\" profile, turning off\n");
1040
        s->options.pns = 0;
1041
11
    } else if (avctx->profile == FF_PROFILE_AAC_LTP) {
1042
        s->options.ltp = 1;
1043
        ERROR_IF(s->options.pred,
1044
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1045
11
    } else if (avctx->profile == FF_PROFILE_AAC_MAIN) {
1046
1
        s->options.pred = 1;
1047
1
        ERROR_IF(s->options.ltp,
1048
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1049
10
    } else if (s->options.ltp) {
1050
        avctx->profile = FF_PROFILE_AAC_LTP;
1051
        WARN_IF(1,
1052
                "Chainging profile to \"aac_ltp\"\n");
1053
        ERROR_IF(s->options.pred,
1054
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1055
10
    } else if (s->options.pred) {
1056
        avctx->profile = FF_PROFILE_AAC_MAIN;
1057
        WARN_IF(1,
1058
                "Chainging profile to \"aac_main\"\n");
1059
        ERROR_IF(s->options.ltp,
1060
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1061
    }
1062
11
    s->profile = avctx->profile;
1063
1064
    /* Coder limitations */
1065
11
    s->coder = &ff_aac_coders[s->options.coder];
1066
11
    if (s->options.coder == AAC_CODER_ANMR) {
1067
        ERROR_IF(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1068
                 "The ANMR coder is considered experimental, add -strict -2 to enable!\n");
1069
        s->options.intensity_stereo = 0;
1070
        s->options.pns = 0;
1071
    }
1072

11
    ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1073
             "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");
1074
1075
    /* M/S introduces horrible artifacts with multichannel files, this is temporary */
1076
11
    if (s->channels > 3)
1077
1
        s->options.mid_side = 0;
1078
1079
11
    if ((ret = dsp_init(avctx, s)) < 0)
1080
        return ret;
1081
1082
11
    if ((ret = alloc_buffers(avctx, s)) < 0)
1083
        return ret;
1084
1085
11
    if ((ret = put_audio_specific_config(avctx)))
1086
        return ret;
1087
1088
11
    sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];
1089
11
    sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];
1090
11
    lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];
1091
11
    lengths[1] = ff_aac_num_swb_128[s->samplerate_index];
1092
25
    for (i = 0; i < s->chan_map[0]; i++)
1093
14
        grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
1094
11
    if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
1095
11
                           s->chan_map[0], grouping)) < 0)
1096
        return ret;
1097
11
    s->psypp = ff_psy_preprocess_init(avctx);
1098
11
    ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
1099
11
    s->random_state = 0x1f2e3d4c;
1100
1101
11
    s->abs_pow34   = abs_pow34_v;
1102
11
    s->quant_bands = quantize_bands;
1103
1104
    if (ARCH_X86)
1105
11
        ff_aac_dsp_init_x86(s);
1106
1107
    if (HAVE_MIPSDSP)
1108
        ff_aac_coder_init_mips(s);
1109
1110
11
    if ((ret = ff_thread_once(&aac_table_init, &aac_encode_init_tables)) != 0)
1111
        return AVERROR_UNKNOWN;
1112
1113
11
    ff_af_queue_init(avctx, &s->afq);
1114
1115
11
    return 0;
1116
}
1117
1118
#define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1119
static const AVOption aacenc_options[] = {
1120
    {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_FAST}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},
1121
        {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1122
        {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1123
        {"fast",     "Default fast search",       0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1124
    {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},
1125
    {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1126
    {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1127
    {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1128
    {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1129
    {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1130
    {"aac_pce", "Forces the use of PCEs", offsetof(AACEncContext, options.pce), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1131
    FF_AAC_PROFILE_OPTS
1132
    {NULL}
1133
};
1134
1135
static const AVClass aacenc_class = {
1136
    .class_name = "AAC encoder",
1137
    .item_name  = av_default_item_name,
1138
    .option     = aacenc_options,
1139
    .version    = LIBAVUTIL_VERSION_INT,
1140
};
1141
1142
static const AVCodecDefault aac_encode_defaults[] = {
1143
    { "b", "0" },
1144
    { NULL }
1145
};
1146
1147
AVCodec ff_aac_encoder = {
1148
    .name           = "aac",
1149
    .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
1150
    .type           = AVMEDIA_TYPE_AUDIO,
1151
    .id             = AV_CODEC_ID_AAC,
1152
    .priv_data_size = sizeof(AACEncContext),
1153
    .init           = aac_encode_init,
1154
    .encode2        = aac_encode_frame,
1155
    .close          = aac_encode_end,
1156
    .defaults       = aac_encode_defaults,
1157
    .supported_samplerates = mpeg4audio_sample_rates,
1158
    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1159
    .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,
1160
    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
1161
                                                     AV_SAMPLE_FMT_NONE },
1162
    .priv_class     = &aacenc_class,
1163
};