GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/aacenc.c Lines: 556 654 85.0 %
Date: 2020-11-28 20:53:16 Branches: 356 484 73.6 %

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/float_dsp.h"
34
#include "libavutil/opt.h"
35
#include "avcodec.h"
36
#include "put_bits.h"
37
#include "internal.h"
38
#include "mpeg4audio.h"
39
#include "kbdwin.h"
40
#include "sinewin.h"
41
#include "profiles.h"
42
43
#include "aac.h"
44
#include "aactab.h"
45
#include "aacenc.h"
46
#include "aacenctab.h"
47
#include "aacenc_utils.h"
48
49
#include "psymodel.h"
50
51
static void put_pce(PutBitContext *pb, AVCodecContext *avctx)
52
{
53
    int i, j;
54
    AACEncContext *s = avctx->priv_data;
55
    AACPCEInfo *pce = &s->pce;
56
    const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
57
    const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
58
59
    put_bits(pb, 4, 0);
60
61
    put_bits(pb, 2, avctx->profile);
62
    put_bits(pb, 4, s->samplerate_index);
63
64
    put_bits(pb, 4, pce->num_ele[0]); /* Front */
65
    put_bits(pb, 4, pce->num_ele[1]); /* Side */
66
    put_bits(pb, 4, pce->num_ele[2]); /* Back */
67
    put_bits(pb, 2, pce->num_ele[3]); /* LFE */
68
    put_bits(pb, 3, 0); /* Assoc data */
69
    put_bits(pb, 4, 0); /* CCs */
70
71
    put_bits(pb, 1, 0); /* Stereo mixdown */
72
    put_bits(pb, 1, 0); /* Mono mixdown */
73
    put_bits(pb, 1, 0); /* Something else */
74
75
    for (i = 0; i < 4; i++) {
76
        for (j = 0; j < pce->num_ele[i]; j++) {
77
            if (i < 3)
78
                put_bits(pb, 1, pce->pairing[i][j]);
79
            put_bits(pb, 4, pce->index[i][j]);
80
        }
81
    }
82
83
    align_put_bits(pb);
84
    put_bits(pb, 8, strlen(aux_data));
85
    ff_put_string(pb, aux_data, 0);
86
}
87
88
/**
89
 * Make AAC audio config object.
90
 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
91
 */
92
11
static int put_audio_specific_config(AVCodecContext *avctx)
93
{
94
    PutBitContext pb;
95
11
    AACEncContext *s = avctx->priv_data;
96
11
    int channels = (!s->needs_pce)*(s->channels - (s->channels == 8 ? 1 : 0));
97
11
    const int max_size = 32;
98
99
11
    avctx->extradata = av_mallocz(max_size);
100
11
    if (!avctx->extradata)
101
        return AVERROR(ENOMEM);
102
103
11
    init_put_bits(&pb, avctx->extradata, max_size);
104
11
    put_bits(&pb, 5, s->profile+1); //profile
105
11
    put_bits(&pb, 4, s->samplerate_index); //sample rate index
106
11
    put_bits(&pb, 4, channels);
107
    //GASpecificConfig
108
11
    put_bits(&pb, 1, 0); //frame length - 1024 samples
109
11
    put_bits(&pb, 1, 0); //does not depend on core coder
110
11
    put_bits(&pb, 1, 0); //is not extension
111
11
    if (s->needs_pce)
112
        put_pce(&pb, avctx);
113
114
    //Explicitly Mark SBR absent
115
11
    put_bits(&pb, 11, 0x2b7); //sync extension
116
11
    put_bits(&pb, 5,  AOT_SBR);
117
11
    put_bits(&pb, 1,  0);
118
11
    flush_put_bits(&pb);
119
11
    avctx->extradata_size = put_bits_count(&pb) >> 3;
120
121
11
    return 0;
122
}
123
124
11647
void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
125
{
126
11647
    ++s->quantize_band_cost_cache_generation;
127
11647
    if (s->quantize_band_cost_cache_generation == 0) {
128
        memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
129
        s->quantize_band_cost_cache_generation = 1;
130
    }
131
11647
}
132
133
#define WINDOW_FUNC(type) \
134
static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
135
                                    SingleChannelElement *sce, \
136
                                    const float *audio)
137
138
6899
WINDOW_FUNC(only_long)
139
{
140
6899
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
141
6899
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
142
6899
    float *out = sce->ret_buf;
143
144
6899
    fdsp->vector_fmul        (out,        audio,        lwindow, 1024);
145
6899
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
146
6899
}
147
148
119
WINDOW_FUNC(long_start)
149
{
150
119
    const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
151
119
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
152
119
    float *out = sce->ret_buf;
153
154
119
    fdsp->vector_fmul(out, audio, lwindow, 1024);
155
119
    memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
156
119
    fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
157
119
    memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
158
119
}
159
160
102
WINDOW_FUNC(long_stop)
161
{
162
102
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
163
102
    const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
164
102
    float *out = sce->ret_buf;
165
166
102
    memset(out, 0, sizeof(out[0]) * 448);
167
102
    fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
168
102
    memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
169
102
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
170
102
}
171
172
164
WINDOW_FUNC(eight_short)
173
{
174
164
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
175
164
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
176
164
    const float *in = audio + 448;
177
164
    float *out = sce->ret_buf;
178
    int w;
179
180
1476
    for (w = 0; w < 8; w++) {
181
1312
        fdsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
182
1312
        out += 128;
183
1312
        in  += 128;
184
1312
        fdsp->vector_fmul_reverse(out, in, swindow, 128);
185
1312
        out += 128;
186
    }
187
164
}
188
189
static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
190
                                     SingleChannelElement *sce,
191
                                     const float *audio) = {
192
    [ONLY_LONG_SEQUENCE]   = apply_only_long_window,
193
    [LONG_START_SEQUENCE]  = apply_long_start_window,
194
    [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
195
    [LONG_STOP_SEQUENCE]   = apply_long_stop_window
196
};
197
198
7284
static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
199
                                  float *audio)
200
{
201
    int i;
202
7284
    const float *output = sce->ret_buf;
203
204
7284
    apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
205
206
7284
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
207
7120
        s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
208
    else
209
1476
        for (i = 0; i < 1024; i += 128)
210
1312
            s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
211
7284
    memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
212
7284
    memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
213
7284
}
214
215
/**
216
 * Encode ics_info element.
217
 * @see Table 4.6 (syntax of ics_info)
218
 */
219
6714
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
220
{
221
    int w;
222
223
6714
    put_bits(&s->pb, 1, 0);                // ics_reserved bit
224
6714
    put_bits(&s->pb, 2, info->window_sequence[0]);
225
6714
    put_bits(&s->pb, 1, info->use_kb_window[0]);
226
6714
    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
227
6527
        put_bits(&s->pb, 6, info->max_sfb);
228
6527
        put_bits(&s->pb, 1, !!info->predictor_present);
229
    } else {
230
187
        put_bits(&s->pb, 4, info->max_sfb);
231
1496
        for (w = 1; w < 8; w++)
232
1309
            put_bits(&s->pb, 1, !info->group_len[w]);
233
    }
234
6714
}
235
236
/**
237
 * Encode MS data.
238
 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
239
 */
240
5029
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
241
{
242
    int i, w;
243
244
5029
    put_bits(pb, 2, cpe->ms_mode);
245
5029
    if (cpe->ms_mode == 1)
246
1812
        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
247
40817
            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
248
39892
                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
249
5029
}
250
251
/**
252
 * Produce integer coefficients from scalefactors provided by the model.
253
 */
254
6469
static void adjust_frame_information(ChannelElement *cpe, int chans)
255
{
256
    int i, w, w2, g, ch;
257
    int maxsfb, cmaxsfb;
258
259
18212
    for (ch = 0; ch < chans; ch++) {
260
11743
        IndividualChannelStream *ics = &cpe->ch[ch].ics;
261
11743
        maxsfb = 0;
262
11743
        cpe->ch[ch].pulse.num_pulse = 0;
263
24173
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
264
26049
            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
265

41764
                for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
266
                    ;
267
13619
                maxsfb = FFMAX(maxsfb, cmaxsfb);
268
            }
269
        }
270
11743
        ics->max_sfb = maxsfb;
271
272
        //adjust zero bands for window groups
273
24173
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
274
563969
            for (g = 0; g < ics->max_sfb; g++) {
275
551539
                i = 1;
276
567030
                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
277
551879
                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
278
536388
                        i = 0;
279
536388
                        break;
280
                    }
281
                }
282
551539
                cpe->ch[ch].zeroes[w*16 + g] = i;
283
            }
284
        }
285
    }
286
287

6469
    if (chans > 1 && cpe->common_window) {
288
5029
        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
289
5029
        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
290
5029
        int msc = 0;
291
5029
        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
292
5029
        ics1->max_sfb = ics0->max_sfb;
293
10625
        for (w = 0; w < ics0->num_windows*16; w += 16)
294
247534
            for (i = 0; i < ics0->max_sfb; i++)
295
241938
                if (cpe->ms_mask[w+i])
296
30077
                    msc++;
297

5029
        if (msc == 0 || ics0->max_sfb == 0)
298
3749
            cpe->ms_mode = 0;
299
        else
300
1280
            cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
301
    }
302
6469
}
303
304
2355
static void apply_intensity_stereo(ChannelElement *cpe)
305
{
306
    int w, w2, g, i;
307
2355
    IndividualChannelStream *ics = &cpe->ch[0].ics;
308
2355
    if (!cpe->common_window)
309
1237
        return;
310
2289
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
311
2443
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
312
1272
            int start = (w+w2) * 128;
313
57440
            for (g = 0; g < ics->num_swb; g++) {
314
56168
                int p  = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
315
56168
                float scale = cpe->ch[0].is_ener[w*16+g];
316
56168
                if (!cpe->is_mask[w*16 + g]) {
317
47882
                    start += ics->swb_sizes[g];
318
47882
                    continue;
319
                }
320
8286
                if (cpe->ms_mask[w*16 + g])
321
2383
                    p *= -1;
322
288446
                for (i = 0; i < ics->swb_sizes[g]; i++) {
323
280160
                    float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
324
280160
                    cpe->ch[0].coeffs[start+i] = sum;
325
280160
                    cpe->ch[1].coeffs[start+i] = 0.0f;
326
                }
327
8286
                start += ics->swb_sizes[g];
328
            }
329
        }
330
    }
331
}
332
333
2071
static void apply_mid_side_stereo(ChannelElement *cpe)
334
{
335
    int w, w2, g, i;
336
2071
    IndividualChannelStream *ics = &cpe->ch[0].ics;
337
2071
    if (!cpe->common_window)
338
1078
        return;
339
2039
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
340
2193
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
341
1147
            int start = (w+w2) * 128;
342
51190
            for (g = 0; g < ics->num_swb; g++) {
343
                /* ms_mask can be used for other purposes in PNS and I/S,
344
                 * so must not apply M/S if any band uses either, even if
345
                 * ms_mask is set.
346
                 */
347

50043
                if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]
348
28910
                    || cpe->ch[0].band_type[w*16 + g] >= NOISE_BT
349
28910
                    || cpe->ch[1].band_type[w*16 + g] >= NOISE_BT) {
350
21133
                    start += ics->swb_sizes[g];
351
21133
                    continue;
352
                }
353
650946
                for (i = 0; i < ics->swb_sizes[g]; i++) {
354
622036
                    float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
355
622036
                    float R = L - cpe->ch[1].coeffs[start+i];
356
622036
                    cpe->ch[0].coeffs[start+i] = L;
357
622036
                    cpe->ch[1].coeffs[start+i] = R;
358
                }
359
28910
                start += ics->swb_sizes[g];
360
            }
361
        }
362
    }
363
}
364
365
/**
366
 * Encode scalefactor band coding type.
367
 */
368
11743
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
369
{
370
    int w;
371
372
11743
    if (s->coder->set_special_band_scalefactors)
373
11743
        s->coder->set_special_band_scalefactors(s, sce);
374
375
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
376
12430
        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
377
11743
}
378
379
/**
380
 * Encode scalefactors.
381
 */
382
11743
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
383
                                 SingleChannelElement *sce)
384
{
385
11743
    int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
386
11743
    int off_is = 0, noise_flag = 1;
387
    int i, w;
388
389
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
390
563974
        for (i = 0; i < sce->ics.max_sfb; i++) {
391
551544
            if (!sce->zeroes[w*16 + i]) {
392
521143
                if (sce->band_type[w*16 + i] == NOISE_BT) {
393
17340
                    diff = sce->sf_idx[w*16 + i] - off_pns;
394
17340
                    off_pns = sce->sf_idx[w*16 + i];
395
17340
                    if (noise_flag-- > 0) {
396
2242
                        put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
397
2242
                        continue;
398
                    }
399
503803
                } else if (sce->band_type[w*16 + i] == INTENSITY_BT  ||
400
497850
                           sce->band_type[w*16 + i] == INTENSITY_BT2) {
401
8191
                    diff = sce->sf_idx[w*16 + i] - off_is;
402
8191
                    off_is = sce->sf_idx[w*16 + i];
403
                } else {
404
495612
                    diff = sce->sf_idx[w*16 + i] - off_sf;
405
495612
                    off_sf = sce->sf_idx[w*16 + i];
406
                }
407
518901
                diff += SCALE_DIFF_ZERO;
408

518901
                av_assert0(diff >= 0 && diff <= 120);
409
518901
                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
410
            }
411
        }
412
    }
413
11743
}
414
415
/**
416
 * Encode pulse data.
417
 */
418
11743
static void encode_pulses(AACEncContext *s, Pulse *pulse)
419
{
420
    int i;
421
422
11743
    put_bits(&s->pb, 1, !!pulse->num_pulse);
423
11743
    if (!pulse->num_pulse)
424
11743
        return;
425
426
    put_bits(&s->pb, 2, pulse->num_pulse - 1);
427
    put_bits(&s->pb, 6, pulse->start);
428
    for (i = 0; i < pulse->num_pulse; i++) {
429
        put_bits(&s->pb, 5, pulse->pos[i]);
430
        put_bits(&s->pb, 4, pulse->amp[i]);
431
    }
432
}
433
434
/**
435
 * Encode spectral coefficients processed by psychoacoustic model.
436
 */
437
11743
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
438
{
439
    int start, i, w, w2;
440
441
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
442
12430
        start = 0;
443
563974
        for (i = 0; i < sce->ics.max_sfb; i++) {
444
551544
            if (sce->zeroes[w*16 + i]) {
445
30401
                start += sce->ics.swb_sizes[i];
446
30401
                continue;
447
            }
448
1050888
            for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
449
529745
                s->coder->quantize_and_encode_band(s, &s->pb,
450
529745
                                                   &sce->coeffs[start + w2*128],
451
529745
                                                   NULL, sce->ics.swb_sizes[i],
452
529745
                                                   sce->sf_idx[w*16 + i],
453
529745
                                                   sce->band_type[w*16 + i],
454
                                                   s->lambda,
455
529745
                                                   sce->ics.window_clipping[w]);
456
            }
457
521143
            start += sce->ics.swb_sizes[i];
458
        }
459
    }
460
11743
}
461
462
/**
463
 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
464
 */
465
7284
static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
466
{
467
    int start, i, j, w;
468
469
7284
    if (sce->ics.clip_avoidance_factor < 1.0f) {
470
392
        for (w = 0; w < sce->ics.num_windows; w++) {
471
224
            start = 0;
472
8645
            for (i = 0; i < sce->ics.max_sfb; i++) {
473
8421
                float *swb_coeffs = &sce->coeffs[start + w*128];
474
167173
                for (j = 0; j < sce->ics.swb_sizes[i]; j++)
475
158752
                    swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
476
8421
                start += sce->ics.swb_sizes[i];
477
            }
478
        }
479
    }
480
7284
}
481
482
/**
483
 * Encode one channel of audio data.
484
 */
485
11743
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
486
                                     SingleChannelElement *sce,
487
                                     int common_window)
488
{
489
11743
    put_bits(&s->pb, 8, sce->sf_idx[0]);
490
11743
    if (!common_window) {
491
1685
        put_ics_info(s, &sce->ics);
492
1685
        if (s->coder->encode_main_pred)
493
1685
            s->coder->encode_main_pred(s, sce);
494
1685
        if (s->coder->encode_ltp_info)
495
1685
            s->coder->encode_ltp_info(s, sce, 0);
496
    }
497
11743
    encode_band_info(s, sce);
498
11743
    encode_scale_factors(avctx, s, sce);
499
11743
    encode_pulses(s, &sce->pulse);
500
11743
    put_bits(&s->pb, 1, !!sce->tns.present);
501
11743
    if (s->coder->encode_tns_info)
502
11743
        s->coder->encode_tns_info(s, sce);
503
11743
    put_bits(&s->pb, 1, 0); //ssr
504
11743
    encode_spectral_coeffs(s, sce);
505
11743
    return 0;
506
}
507
508
/**
509
 * Write some auxiliary information about the created AAC file.
510
 */
511
static void put_bitstream_info(AACEncContext *s, const char *name)
512
{
513
    int i, namelen, padbits;
514
515
    namelen = strlen(name) + 2;
516
    put_bits(&s->pb, 3, TYPE_FIL);
517
    put_bits(&s->pb, 4, FFMIN(namelen, 15));
518
    if (namelen >= 15)
519
        put_bits(&s->pb, 8, namelen - 14);
520
    put_bits(&s->pb, 4, 0); //extension type - filler
521
    padbits = -put_bits_count(&s->pb) & 7;
522
    align_put_bits(&s->pb);
523
    for (i = 0; i < namelen - 2; i++)
524
        put_bits(&s->pb, 8, name[i]);
525
    put_bits(&s->pb, 12 - padbits, 0);
526
}
527
528
/*
529
 * Copy input samples.
530
 * Channels are reordered from libavcodec's default order to AAC order.
531
 */
532
3816
static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
533
{
534
    int ch;
535
3816
    int end = 2048 + (frame ? frame->nb_samples : 0);
536
3816
    const uint8_t *channel_map = s->reorder_map;
537
538
    /* copy and remap input samples */
539
11125
    for (ch = 0; ch < s->channels; ch++) {
540
        /* copy last 1024 samples of previous frame to the start of the current frame */
541
7309
        memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
542
543
        /* copy new samples and zero any remaining samples */
544
7309
        if (frame) {
545
7259
            memcpy(&s->planar_samples[ch][2048],
546
7259
                   frame->extended_data[channel_map[ch]],
547
7259
                   frame->nb_samples * sizeof(s->planar_samples[0][0]));
548
        }
549
7309
        memset(&s->planar_samples[ch][end], 0,
550
7309
               (3072 - end) * sizeof(s->planar_samples[0][0]));
551
    }
552
3816
}
553
554
3827
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
555
                            const AVFrame *frame, int *got_packet_ptr)
556
{
557
3827
    AACEncContext *s = avctx->priv_data;
558
3827
    float **samples = s->planar_samples, *samples2, *la, *overlap;
559
    ChannelElement *cpe;
560
    SingleChannelElement *sce;
561
    IndividualChannelStream *ics;
562
    int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
563
    int target_bits, rate_bits, too_many_bits, too_few_bits;
564
3827
    int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
565
    int chan_el_counter[4];
566
    FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
567
568
    /* add current frame to queue */
569
3827
    if (frame) {
570
3794
        if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
571
            return ret;
572
    } else {
573

33
        if (!s->afq.remaining_samples || (!s->afq.frame_alloc && !s->afq.frame_count))
574
11
            return 0;
575
    }
576
577
3816
    copy_input_samples(s, frame);
578
3816
    if (s->psypp)
579
3816
        ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
580
581
3816
    if (!avctx->frame_number)
582
11
        return 0;
583
584
3805
    start_ch = 0;
585
7754
    for (i = 0; i < s->chan_map[0]; i++) {
586
3949
        FFPsyWindowInfo* wi = windows + start_ch;
587
3949
        tag      = s->chan_map[i+1];
588
3949
        chans    = tag == TYPE_CPE ? 2 : 1;
589
3949
        cpe      = &s->cpe[i];
590
11233
        for (ch = 0; ch < chans; ch++) {
591
            int k;
592
            float clip_avoidance_factor;
593
7284
            sce = &cpe->ch[ch];
594
7284
            ics = &sce->ics;
595
7284
            s->cur_channel = start_ch + ch;
596
7284
            overlap  = &samples[s->cur_channel][0];
597
7284
            samples2 = overlap + 1024;
598
7284
            la       = samples2 + (448+64);
599
7284
            if (!frame)
600
50
                la = NULL;
601
7284
            if (tag == TYPE_LFE) {
602
48
                wi[ch].window_type[0] = wi[ch].window_type[1] = ONLY_LONG_SEQUENCE;
603
48
                wi[ch].window_shape   = 0;
604
48
                wi[ch].num_windows    = 1;
605
48
                wi[ch].grouping[0]    = 1;
606
48
                wi[ch].clipping[0]    = 0;
607
608
                /* Only the lowest 12 coefficients are used in a LFE channel.
609
                 * The expression below results in only the bottom 8 coefficients
610
                 * being used for 11.025kHz to 16kHz sample rates.
611
                 */
612
48
                ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
613
            } else {
614
7236
                wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
615
7236
                                              ics->window_sequence[0]);
616
            }
617
7284
            ics->window_sequence[1] = ics->window_sequence[0];
618
7284
            ics->window_sequence[0] = wi[ch].window_type[0];
619
7284
            ics->use_kb_window[1]   = ics->use_kb_window[0];
620
7284
            ics->use_kb_window[0]   = wi[ch].window_shape;
621
7284
            ics->num_windows        = wi[ch].num_windows;
622
7284
            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
623

7284
            ics->num_swb            = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
624
7284
            ics->max_sfb            = FFMIN(ics->max_sfb, ics->num_swb);
625
14568
            ics->swb_offset         = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
626
7284
                                        ff_swb_offset_128 [s->samplerate_index]:
627
7120
                                        ff_swb_offset_1024[s->samplerate_index];
628
14568
            ics->tns_max_bands      = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
629
7284
                                        ff_tns_max_bands_128 [s->samplerate_index]:
630
7120
                                        ff_tns_max_bands_1024[s->samplerate_index];
631
632
15716
            for (w = 0; w < ics->num_windows; w++)
633
8432
                ics->group_len[w] = wi[ch].grouping[w];
634
635
            /* Calculate input sample maximums and evaluate clipping risk */
636
7284
            clip_avoidance_factor = 0.0f;
637
15716
            for (w = 0; w < ics->num_windows; w++) {
638
8432
                const float *wbuf = overlap + w * 128;
639
8432
                const int wlen = 2048 / ics->num_windows;
640
8432
                float max = 0;
641
                int j;
642
                /* mdct input is 2 * output */
643
14926064
                for (j = 0; j < wlen; j++)
644
14917632
                    max = FFMAX(max, fabsf(wbuf[j]));
645
8432
                wi[ch].clipping[w] = max;
646
            }
647
15716
            for (w = 0; w < ics->num_windows; w++) {
648
8432
                if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
649
176
                    ics->window_clipping[w] = 1;
650
176
                    clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
651
                } else {
652
8256
                    ics->window_clipping[w] = 0;
653
                }
654
            }
655
7284
            if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
656
168
                ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
657
            } else {
658
7116
                ics->clip_avoidance_factor = 1.0f;
659
            }
660
661
7284
            apply_window_and_mdct(s, sce, overlap);
662
663

7284
            if (s->options.ltp && s->coder->update_ltp) {
664
                s->coder->update_ltp(s, sce);
665
                apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
666
                s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);
667
            }
668
669
7466100
            for (k = 0; k < 1024; k++) {
670
7458816
                if (!(fabs(cpe->ch[ch].coeffs[k]) < 1E16)) { // Ensure headroom for energy calculation
671
                    av_log(avctx, AV_LOG_ERROR, "Input contains (near) NaN/+-Inf\n");
672
                    return AVERROR(EINVAL);
673
                }
674
            }
675
7284
            avoid_clipping(s, sce);
676
        }
677
3949
        start_ch += chans;
678
    }
679
3805
    if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
680
        return ret;
681
3805
    frame_bits = its = 0;
682
    do {
683
6253
        init_put_bits(&s->pb, avpkt->data, avpkt->size);
684
685

6253
        if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
686
            put_bitstream_info(s, LIBAVCODEC_IDENT);
687
6253
        start_ch = 0;
688
6253
        target_bits = 0;
689
6253
        memset(chan_el_counter, 0, sizeof(chan_el_counter));
690
12722
        for (i = 0; i < s->chan_map[0]; i++) {
691
6469
            FFPsyWindowInfo* wi = windows + start_ch;
692
            const float *coeffs[2];
693
6469
            tag      = s->chan_map[i+1];
694
6469
            chans    = tag == TYPE_CPE ? 2 : 1;
695
6469
            cpe      = &s->cpe[i];
696
6469
            cpe->common_window = 0;
697
6469
            memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
698
6469
            memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
699
6469
            put_bits(&s->pb, 3, tag);
700
6469
            put_bits(&s->pb, 4, chan_el_counter[tag]++);
701
18212
            for (ch = 0; ch < chans; ch++) {
702
11743
                sce = &cpe->ch[ch];
703
11743
                coeffs[ch] = sce->coeffs;
704
11743
                sce->ics.predictor_present = 0;
705
11743
                sce->ics.ltp.present = 0;
706
11743
                memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
707
11743
                memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
708
11743
                memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
709
1514847
                for (w = 0; w < 128; w++)
710
1503104
                    if (sce->band_type[w] > RESERVED_BT)
711
25469
                        sce->band_type[w] = 0;
712
            }
713
6469
            s->psy.bitres.alloc = -1;
714
6469
            s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
715
6469
            s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
716
6469
            if (s->psy.bitres.alloc > 0) {
717
                /* Lambda unused here on purpose, we need to take psy's unscaled allocation */
718
12938
                target_bits += s->psy.bitres.alloc
719
6469
                    * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
720
6469
                s->psy.bitres.alloc /= chans;
721
            }
722
6469
            s->cur_type = tag;
723
18212
            for (ch = 0; ch < chans; ch++) {
724
11743
                s->cur_channel = start_ch + ch;
725

11743
                if (s->options.pns && s->coder->mark_pns)
726
3515
                    s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
727
11743
                s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
728
            }
729
6469
            if (chans > 1
730
5274
                && wi[0].window_type[0] == wi[1].window_type[0]
731
5038
                && wi[0].window_shape   == wi[1].window_shape) {
732
733
5038
                cpe->common_window = 1;
734
10634
                for (w = 0; w < wi[0].num_windows; w++) {
735
5605
                    if (wi[0].grouping[w] != wi[1].grouping[w]) {
736
9
                        cpe->common_window = 0;
737
9
                        break;
738
                    }
739
                }
740
            }
741
18212
            for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
742
11743
                sce = &cpe->ch[ch];
743
11743
                s->cur_channel = start_ch + ch;
744

11743
                if (s->options.tns && s->coder->search_for_tns)
745
3515
                    s->coder->search_for_tns(s, sce);
746

11743
                if (s->options.tns && s->coder->apply_tns_filt)
747
3515
                    s->coder->apply_tns_filt(s, sce);
748
11743
                if (sce->tns.present)
749
46
                    tns_mode = 1;
750

11743
                if (s->options.pns && s->coder->search_for_pns)
751
3515
                    s->coder->search_for_pns(s, avctx, sce);
752
            }
753
6469
            s->cur_channel = start_ch;
754
6469
            if (s->options.intensity_stereo) { /* Intensity Stereo */
755
2355
                if (s->coder->search_for_is)
756
2355
                    s->coder->search_for_is(s, avctx, cpe);
757
2355
                if (cpe->is_mode) is_mode = 1;
758
2355
                apply_intensity_stereo(cpe);
759
            }
760
6469
            if (s->options.pred) { /* Prediction */
761
1248
                for (ch = 0; ch < chans; ch++) {
762
832
                    sce = &cpe->ch[ch];
763
832
                    s->cur_channel = start_ch + ch;
764

832
                    if (s->options.pred && s->coder->search_for_pred)
765
832
                        s->coder->search_for_pred(s, sce);
766
832
                    if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
767
                }
768
416
                if (s->coder->adjust_common_pred)
769
416
                    s->coder->adjust_common_pred(s, cpe);
770
1248
                for (ch = 0; ch < chans; ch++) {
771
832
                    sce = &cpe->ch[ch];
772
832
                    s->cur_channel = start_ch + ch;
773

832
                    if (s->options.pred && s->coder->apply_main_pred)
774
832
                        s->coder->apply_main_pred(s, sce);
775
                }
776
416
                s->cur_channel = start_ch;
777
            }
778
6469
            if (s->options.mid_side) { /* Mid/Side stereo */
779

2071
                if (s->options.mid_side == -1 && s->coder->search_for_ms)
780
1651
                    s->coder->search_for_ms(s, cpe);
781
420
                else if (cpe->common_window)
782
393
                    memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
783
2071
                apply_mid_side_stereo(cpe);
784
            }
785
6469
            adjust_frame_information(cpe, chans);
786
6469
            if (s->options.ltp) { /* LTP */
787
                for (ch = 0; ch < chans; ch++) {
788
                    sce = &cpe->ch[ch];
789
                    s->cur_channel = start_ch + ch;
790
                    if (s->coder->search_for_ltp)
791
                        s->coder->search_for_ltp(s, sce, cpe->common_window);
792
                    if (sce->ics.ltp.present) pred_mode = 1;
793
                }
794
                s->cur_channel = start_ch;
795
                if (s->coder->adjust_common_ltp)
796
                    s->coder->adjust_common_ltp(s, cpe);
797
            }
798
6469
            if (chans == 2) {
799
5274
                put_bits(&s->pb, 1, cpe->common_window);
800
5274
                if (cpe->common_window) {
801
5029
                    put_ics_info(s, &cpe->ch[0].ics);
802
5029
                    if (s->coder->encode_main_pred)
803
5029
                        s->coder->encode_main_pred(s, &cpe->ch[0]);
804
5029
                    if (s->coder->encode_ltp_info)
805
5029
                        s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
806
5029
                    encode_ms_info(&s->pb, cpe);
807
5029
                    if (cpe->ms_mode) ms_mode = 1;
808
                }
809
            }
810
18212
            for (ch = 0; ch < chans; ch++) {
811
11743
                s->cur_channel = start_ch + ch;
812
11743
                encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
813
            }
814
6469
            start_ch += chans;
815
        }
816
817
6253
        if (avctx->flags & AV_CODEC_FLAG_QSCALE) {
818
            /* When using a constant Q-scale, don't mess with lambda */
819
            break;
820
        }
821
822
        /* rate control stuff
823
         * allow between the nominal bitrate, and what psy's bit reservoir says to target
824
         * but drift towards the nominal bitrate always
825
         */
826
6253
        frame_bits = put_bits_count(&s->pb);
827
6253
        rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
828
6253
        rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
829
6253
        too_many_bits = FFMAX(target_bits, rate_bits);
830
6253
        too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
831
6253
        too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
832
833
        /* When using ABR, be strict (but only for increasing) */
834
6253
        too_few_bits = too_few_bits - too_few_bits/8;
835
6253
        too_many_bits = too_many_bits + too_many_bits/2;
836
837
6253
        if (   its == 0 /* for steady-state Q-scale tracking */
838

2448
            || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
839
497
            || frame_bits >= 6144 * s->channels - 3  )
840
        {
841
5756
            float ratio = ((float)rate_bits) / frame_bits;
842
843

5756
            if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
844
                /*
845
                 * This path is for steady-state Q-scale tracking
846
                 * When frame bits fall within the stable range, we still need to adjust
847
                 * lambda to maintain it like so in a stable fashion (large jumps in lambda
848
                 * create artifacts and should be avoided), but slowly
849
                 */
850
2859
                ratio = sqrtf(sqrtf(ratio));
851
2859
                ratio = av_clipf(ratio, 0.9f, 1.1f);
852
            } else {
853
                /* Not so fast though */
854
2897
                ratio = sqrtf(ratio);
855
            }
856
5756
            s->lambda = FFMIN(s->lambda * ratio, 65536.f);
857
858
            /* Keep iterating if we must reduce and lambda is in the sky */
859

5756
            if (ratio > 0.9f && ratio < 1.1f) {
860
                break;
861
            } else {
862


2448
                if (is_mode || ms_mode || tns_mode || pred_mode) {
863
1027
                    for (i = 0; i < s->chan_map[0]; i++) {
864
                        // Must restore coeffs
865
521
                        chans = tag == TYPE_CPE ? 2 : 1;
866
521
                        cpe = &s->cpe[i];
867
1543
                        for (ch = 0; ch < chans; ch++)
868
1022
                            memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
869
                    }
870
                }
871
2448
                its++;
872
            }
873
        } else {
874
            break;
875
        }
876
    } while (1);
877
878

3805
    if (s->options.ltp && s->coder->ltp_insert_new_frame)
879
        s->coder->ltp_insert_new_frame(s);
880
881
3805
    put_bits(&s->pb, 3, TYPE_END);
882
3805
    flush_put_bits(&s->pb);
883
884
3805
    s->last_frame_pb_count = put_bits_count(&s->pb);
885
886
3805
    s->lambda_sum += s->lambda;
887
3805
    s->lambda_count++;
888
889
3805
    ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
890
                       &avpkt->duration);
891
892
3805
    avpkt->size = put_bits_count(&s->pb) >> 3;
893
3805
    *got_packet_ptr = 1;
894
3805
    return 0;
895
}
896
897
11
static av_cold int aac_encode_end(AVCodecContext *avctx)
898
{
899
11
    AACEncContext *s = avctx->priv_data;
900
901
11
    av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);
902
903
11
    ff_mdct_end(&s->mdct1024);
904
11
    ff_mdct_end(&s->mdct128);
905
11
    ff_psy_end(&s->psy);
906
11
    ff_lpc_end(&s->lpc);
907
11
    if (s->psypp)
908
11
        ff_psy_preprocess_end(s->psypp);
909
11
    av_freep(&s->buffer.samples);
910
11
    av_freep(&s->cpe);
911
11
    av_freep(&s->fdsp);
912
11
    ff_af_queue_close(&s->afq);
913
11
    return 0;
914
}
915
916
11
static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
917
{
918
11
    int ret = 0;
919
920
11
    s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
921
11
    if (!s->fdsp)
922
        return AVERROR(ENOMEM);
923
924
    // window init
925
11
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
926
11
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
927
11
    ff_init_ff_sine_windows(10);
928
11
    ff_init_ff_sine_windows(7);
929
930
11
    if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
931
        return ret;
932
11
    if ((ret = ff_mdct_init(&s->mdct128,   8, 0, 32768.0)) < 0)
933
        return ret;
934
935
11
    return 0;
936
}
937
938
11
static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
939
{
940
    int ch;
941
11
    if (!FF_ALLOCZ_TYPED_ARRAY(s->buffer.samples, s->channels * 3 * 1024) ||
942
11
        !FF_ALLOCZ_TYPED_ARRAY(s->cpe,            s->chan_map[0]))
943
        return AVERROR(ENOMEM);
944
945
36
    for(ch = 0; ch < s->channels; ch++)
946
25
        s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
947
948
11
    return 0;
949
}
950
951
11
static av_cold int aac_encode_init(AVCodecContext *avctx)
952
{
953
11
    AACEncContext *s = avctx->priv_data;
954
11
    int i, ret = 0;
955
    const uint8_t *sizes[2];
956
    uint8_t grouping[AAC_MAX_CHANNELS];
957
    int lengths[2];
958
959
    /* Constants */
960
11
    s->last_frame_pb_count = 0;
961
11
    avctx->frame_size = 1024;
962
11
    avctx->initial_padding = 1024;
963
11
    s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
964
965
    /* Channel map and unspecified bitrate guessing */
966
11
    s->channels = avctx->channels;
967
968
11
    s->needs_pce = 1;
969
25
    for (i = 0; i < FF_ARRAY_ELEMS(aac_normal_chan_layouts); i++) {
970
25
        if (avctx->channel_layout == aac_normal_chan_layouts[i]) {
971
11
            s->needs_pce = s->options.pce;
972
11
            break;
973
        }
974
    }
975
976
11
    if (s->needs_pce) {
977
        char buf[64];
978
        for (i = 0; i < FF_ARRAY_ELEMS(aac_pce_configs); i++)
979
            if (avctx->channel_layout == aac_pce_configs[i].layout)
980
                break;
981
        av_get_channel_layout_string(buf, sizeof(buf), -1, avctx->channel_layout);
982
        ERROR_IF(i == FF_ARRAY_ELEMS(aac_pce_configs), "Unsupported channel layout \"%s\"\n", buf);
983
        av_log(avctx, AV_LOG_INFO, "Using a PCE to encode channel layout \"%s\"\n", buf);
984
        s->pce = aac_pce_configs[i];
985
        s->reorder_map = s->pce.reorder_map;
986
        s->chan_map = s->pce.config_map;
987
    } else {
988
11
        s->reorder_map = aac_chan_maps[s->channels - 1];
989
11
        s->chan_map = aac_chan_configs[s->channels - 1];
990
    }
991
992
11
    if (!avctx->bit_rate) {
993
9
        for (i = 1; i <= s->chan_map[0]; i++) {
994
9
            avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */
995
3
                               s->chan_map[i] == TYPE_LFE ? 16000  : /* LFE  */
996
                                                            69000  ; /* SCE  */
997
        }
998
    }
999
1000
    /* Samplerate */
1001
54
    for (i = 0; i < 16; i++)
1002
54
        if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
1003
11
            break;
1004
11
    s->samplerate_index = i;
1005

11
    ERROR_IF(s->samplerate_index == 16 ||
1006
             s->samplerate_index >= ff_aac_swb_size_1024_len ||
1007
             s->samplerate_index >= ff_aac_swb_size_128_len,
1008
             "Unsupported sample rate %d\n", avctx->sample_rate);
1009
1010
    /* Bitrate limiting */
1011
11
    WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
1012
             "Too many bits %f > %d per frame requested, clamping to max\n",
1013
             1024.0 * avctx->bit_rate / avctx->sample_rate,
1014
             6144 * s->channels);
1015
11
    avctx->bit_rate = (int64_t)FFMIN(6144 * s->channels / 1024.0 * avctx->sample_rate,
1016
                                     avctx->bit_rate);
1017
1018
    /* Profile and option setting */
1019
11
    avctx->profile = avctx->profile == FF_PROFILE_UNKNOWN ? FF_PROFILE_AAC_LOW :
1020
                     avctx->profile;
1021
21
    for (i = 0; i < FF_ARRAY_ELEMS(aacenc_profiles); i++)
1022
21
        if (avctx->profile == aacenc_profiles[i])
1023
11
            break;
1024
11
    if (avctx->profile == FF_PROFILE_MPEG2_AAC_LOW) {
1025
        avctx->profile = FF_PROFILE_AAC_LOW;
1026
        ERROR_IF(s->options.pred,
1027
                 "Main prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1028
        ERROR_IF(s->options.ltp,
1029
                 "LTP prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1030
        WARN_IF(s->options.pns,
1031
                "PNS unavailable in the \"mpeg2_aac_low\" profile, turning off\n");
1032
        s->options.pns = 0;
1033
11
    } else if (avctx->profile == FF_PROFILE_AAC_LTP) {
1034
        s->options.ltp = 1;
1035
        ERROR_IF(s->options.pred,
1036
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1037
11
    } else if (avctx->profile == FF_PROFILE_AAC_MAIN) {
1038
1
        s->options.pred = 1;
1039
1
        ERROR_IF(s->options.ltp,
1040
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1041
10
    } else if (s->options.ltp) {
1042
        avctx->profile = FF_PROFILE_AAC_LTP;
1043
        WARN_IF(1,
1044
                "Chainging profile to \"aac_ltp\"\n");
1045
        ERROR_IF(s->options.pred,
1046
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1047
10
    } else if (s->options.pred) {
1048
        avctx->profile = FF_PROFILE_AAC_MAIN;
1049
        WARN_IF(1,
1050
                "Chainging profile to \"aac_main\"\n");
1051
        ERROR_IF(s->options.ltp,
1052
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1053
    }
1054
11
    s->profile = avctx->profile;
1055
1056
    /* Coder limitations */
1057
11
    s->coder = &ff_aac_coders[s->options.coder];
1058
11
    if (s->options.coder == AAC_CODER_ANMR) {
1059
        ERROR_IF(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1060
                 "The ANMR coder is considered experimental, add -strict -2 to enable!\n");
1061
        s->options.intensity_stereo = 0;
1062
        s->options.pns = 0;
1063
    }
1064

11
    ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1065
             "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");
1066
1067
    /* M/S introduces horrible artifacts with multichannel files, this is temporary */
1068
11
    if (s->channels > 3)
1069
1
        s->options.mid_side = 0;
1070
1071
11
    if ((ret = dsp_init(avctx, s)) < 0)
1072
        return ret;
1073
1074
11
    if ((ret = alloc_buffers(avctx, s)) < 0)
1075
        return ret;
1076
1077
11
    if ((ret = put_audio_specific_config(avctx)))
1078
        return ret;
1079
1080
11
    sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];
1081
11
    sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];
1082
11
    lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];
1083
11
    lengths[1] = ff_aac_num_swb_128[s->samplerate_index];
1084
25
    for (i = 0; i < s->chan_map[0]; i++)
1085
14
        grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
1086
11
    if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
1087
11
                           s->chan_map[0], grouping)) < 0)
1088
        return ret;
1089
11
    s->psypp = ff_psy_preprocess_init(avctx);
1090
11
    ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
1091
11
    s->random_state = 0x1f2e3d4c;
1092
1093
11
    s->abs_pow34   = abs_pow34_v;
1094
11
    s->quant_bands = quantize_bands;
1095
1096
    if (ARCH_X86)
1097
11
        ff_aac_dsp_init_x86(s);
1098
1099
    if (HAVE_MIPSDSP)
1100
        ff_aac_coder_init_mips(s);
1101
1102
11
    ff_af_queue_init(avctx, &s->afq);
1103
11
    ff_aac_tableinit();
1104
1105
11
    return 0;
1106
}
1107
1108
#define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1109
static const AVOption aacenc_options[] = {
1110
    {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_FAST}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},
1111
        {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1112
        {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1113
        {"fast",     "Default fast search",       0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1114
    {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},
1115
    {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1116
    {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1117
    {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1118
    {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1119
    {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1120
    {"aac_pce", "Forces the use of PCEs", offsetof(AACEncContext, options.pce), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1121
    FF_AAC_PROFILE_OPTS
1122
    {NULL}
1123
};
1124
1125
static const AVClass aacenc_class = {
1126
    .class_name = "AAC encoder",
1127
    .item_name  = av_default_item_name,
1128
    .option     = aacenc_options,
1129
    .version    = LIBAVUTIL_VERSION_INT,
1130
};
1131
1132
static const AVCodecDefault aac_encode_defaults[] = {
1133
    { "b", "0" },
1134
    { NULL }
1135
};
1136
1137
AVCodec ff_aac_encoder = {
1138
    .name           = "aac",
1139
    .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
1140
    .type           = AVMEDIA_TYPE_AUDIO,
1141
    .id             = AV_CODEC_ID_AAC,
1142
    .priv_data_size = sizeof(AACEncContext),
1143
    .init           = aac_encode_init,
1144
    .encode2        = aac_encode_frame,
1145
    .close          = aac_encode_end,
1146
    .defaults       = aac_encode_defaults,
1147
    .supported_samplerates = mpeg4audio_sample_rates,
1148
    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1149
    .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,
1150
    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
1151
                                                     AV_SAMPLE_FMT_NONE },
1152
    .priv_class     = &aacenc_class,
1153
};