GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/aacenc.c Lines: 553 651 84.9 %
Date: 2021-03-05 01:16:01 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 "sinewin.h"
40
#include "profiles.h"
41
42
#include "aac.h"
43
#include "aactab.h"
44
#include "aacenc.h"
45
#include "aacenctab.h"
46
#include "aacenc_utils.h"
47
48
#include "psymodel.h"
49
50
static void put_pce(PutBitContext *pb, AVCodecContext *avctx)
51
{
52
    int i, j;
53
    AACEncContext *s = avctx->priv_data;
54
    AACPCEInfo *pce = &s->pce;
55
    const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
56
    const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
57
58
    put_bits(pb, 4, 0);
59
60
    put_bits(pb, 2, avctx->profile);
61
    put_bits(pb, 4, s->samplerate_index);
62
63
    put_bits(pb, 4, pce->num_ele[0]); /* Front */
64
    put_bits(pb, 4, pce->num_ele[1]); /* Side */
65
    put_bits(pb, 4, pce->num_ele[2]); /* Back */
66
    put_bits(pb, 2, pce->num_ele[3]); /* LFE */
67
    put_bits(pb, 3, 0); /* Assoc data */
68
    put_bits(pb, 4, 0); /* CCs */
69
70
    put_bits(pb, 1, 0); /* Stereo mixdown */
71
    put_bits(pb, 1, 0); /* Mono mixdown */
72
    put_bits(pb, 1, 0); /* Something else */
73
74
    for (i = 0; i < 4; i++) {
75
        for (j = 0; j < pce->num_ele[i]; j++) {
76
            if (i < 3)
77
                put_bits(pb, 1, pce->pairing[i][j]);
78
            put_bits(pb, 4, pce->index[i][j]);
79
        }
80
    }
81
82
    align_put_bits(pb);
83
    put_bits(pb, 8, strlen(aux_data));
84
    ff_put_string(pb, aux_data, 0);
85
}
86
87
/**
88
 * Make AAC audio config object.
89
 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
90
 */
91
11
static int put_audio_specific_config(AVCodecContext *avctx)
92
{
93
    PutBitContext pb;
94
11
    AACEncContext *s = avctx->priv_data;
95
11
    int channels = (!s->needs_pce)*(s->channels - (s->channels == 8 ? 1 : 0));
96
11
    const int max_size = 32;
97
98
11
    avctx->extradata = av_mallocz(max_size);
99
11
    if (!avctx->extradata)
100
        return AVERROR(ENOMEM);
101
102
11
    init_put_bits(&pb, avctx->extradata, max_size);
103
11
    put_bits(&pb, 5, s->profile+1); //profile
104
11
    put_bits(&pb, 4, s->samplerate_index); //sample rate index
105
11
    put_bits(&pb, 4, channels);
106
    //GASpecificConfig
107
11
    put_bits(&pb, 1, 0); //frame length - 1024 samples
108
11
    put_bits(&pb, 1, 0); //does not depend on core coder
109
11
    put_bits(&pb, 1, 0); //is not extension
110
11
    if (s->needs_pce)
111
        put_pce(&pb, avctx);
112
113
    //Explicitly Mark SBR absent
114
11
    put_bits(&pb, 11, 0x2b7); //sync extension
115
11
    put_bits(&pb, 5,  AOT_SBR);
116
11
    put_bits(&pb, 1,  0);
117
11
    flush_put_bits(&pb);
118
11
    avctx->extradata_size = put_bits_count(&pb) >> 3;
119
120
11
    return 0;
121
}
122
123
11647
void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
124
{
125
11647
    ++s->quantize_band_cost_cache_generation;
126
11647
    if (s->quantize_band_cost_cache_generation == 0) {
127
        memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
128
        s->quantize_band_cost_cache_generation = 1;
129
    }
130
11647
}
131
132
#define WINDOW_FUNC(type) \
133
static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
134
                                    SingleChannelElement *sce, \
135
                                    const float *audio)
136
137
6899
WINDOW_FUNC(only_long)
138
{
139
6899
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
140
6899
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
141
6899
    float *out = sce->ret_buf;
142
143
6899
    fdsp->vector_fmul        (out,        audio,        lwindow, 1024);
144
6899
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
145
6899
}
146
147
119
WINDOW_FUNC(long_start)
148
{
149
119
    const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
150
119
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
151
119
    float *out = sce->ret_buf;
152
153
119
    fdsp->vector_fmul(out, audio, lwindow, 1024);
154
119
    memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
155
119
    fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
156
119
    memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
157
119
}
158
159
102
WINDOW_FUNC(long_stop)
160
{
161
102
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
162
102
    const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
163
102
    float *out = sce->ret_buf;
164
165
102
    memset(out, 0, sizeof(out[0]) * 448);
166
102
    fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
167
102
    memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
168
102
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
169
102
}
170
171
164
WINDOW_FUNC(eight_short)
172
{
173
164
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
174
164
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
175
164
    const float *in = audio + 448;
176
164
    float *out = sce->ret_buf;
177
    int w;
178
179
1476
    for (w = 0; w < 8; w++) {
180
1312
        fdsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
181
1312
        out += 128;
182
1312
        in  += 128;
183
1312
        fdsp->vector_fmul_reverse(out, in, swindow, 128);
184
1312
        out += 128;
185
    }
186
164
}
187
188
static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
189
                                     SingleChannelElement *sce,
190
                                     const float *audio) = {
191
    [ONLY_LONG_SEQUENCE]   = apply_only_long_window,
192
    [LONG_START_SEQUENCE]  = apply_long_start_window,
193
    [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
194
    [LONG_STOP_SEQUENCE]   = apply_long_stop_window
195
};
196
197
7284
static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
198
                                  float *audio)
199
{
200
    int i;
201
7284
    const float *output = sce->ret_buf;
202
203
7284
    apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
204
205
7284
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
206
7120
        s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
207
    else
208
1476
        for (i = 0; i < 1024; i += 128)
209
1312
            s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
210
7284
    memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
211
7284
    memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
212
7284
}
213
214
/**
215
 * Encode ics_info element.
216
 * @see Table 4.6 (syntax of ics_info)
217
 */
218
6714
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
219
{
220
    int w;
221
222
6714
    put_bits(&s->pb, 1, 0);                // ics_reserved bit
223
6714
    put_bits(&s->pb, 2, info->window_sequence[0]);
224
6714
    put_bits(&s->pb, 1, info->use_kb_window[0]);
225
6714
    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
226
6527
        put_bits(&s->pb, 6, info->max_sfb);
227
6527
        put_bits(&s->pb, 1, !!info->predictor_present);
228
    } else {
229
187
        put_bits(&s->pb, 4, info->max_sfb);
230
1496
        for (w = 1; w < 8; w++)
231
1309
            put_bits(&s->pb, 1, !info->group_len[w]);
232
    }
233
6714
}
234
235
/**
236
 * Encode MS data.
237
 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
238
 */
239
5029
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
240
{
241
    int i, w;
242
243
5029
    put_bits(pb, 2, cpe->ms_mode);
244
5029
    if (cpe->ms_mode == 1)
245
1812
        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
246
40817
            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
247
39892
                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
248
5029
}
249
250
/**
251
 * Produce integer coefficients from scalefactors provided by the model.
252
 */
253
6469
static void adjust_frame_information(ChannelElement *cpe, int chans)
254
{
255
    int i, w, w2, g, ch;
256
    int maxsfb, cmaxsfb;
257
258
18212
    for (ch = 0; ch < chans; ch++) {
259
11743
        IndividualChannelStream *ics = &cpe->ch[ch].ics;
260
11743
        maxsfb = 0;
261
11743
        cpe->ch[ch].pulse.num_pulse = 0;
262
24173
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
263
26049
            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
264

41764
                for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
265
                    ;
266
13619
                maxsfb = FFMAX(maxsfb, cmaxsfb);
267
            }
268
        }
269
11743
        ics->max_sfb = maxsfb;
270
271
        //adjust zero bands for window groups
272
24173
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
273
563969
            for (g = 0; g < ics->max_sfb; g++) {
274
551539
                i = 1;
275
567030
                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
276
551879
                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
277
536388
                        i = 0;
278
536388
                        break;
279
                    }
280
                }
281
551539
                cpe->ch[ch].zeroes[w*16 + g] = i;
282
            }
283
        }
284
    }
285
286

6469
    if (chans > 1 && cpe->common_window) {
287
5029
        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
288
5029
        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
289
5029
        int msc = 0;
290
5029
        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
291
5029
        ics1->max_sfb = ics0->max_sfb;
292
10625
        for (w = 0; w < ics0->num_windows*16; w += 16)
293
247534
            for (i = 0; i < ics0->max_sfb; i++)
294
241938
                if (cpe->ms_mask[w+i])
295
30077
                    msc++;
296

5029
        if (msc == 0 || ics0->max_sfb == 0)
297
3749
            cpe->ms_mode = 0;
298
        else
299
1280
            cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
300
    }
301
6469
}
302
303
2355
static void apply_intensity_stereo(ChannelElement *cpe)
304
{
305
    int w, w2, g, i;
306
2355
    IndividualChannelStream *ics = &cpe->ch[0].ics;
307
2355
    if (!cpe->common_window)
308
1237
        return;
309
2289
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
310
2443
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
311
1272
            int start = (w+w2) * 128;
312
57440
            for (g = 0; g < ics->num_swb; g++) {
313
56168
                int p  = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
314
56168
                float scale = cpe->ch[0].is_ener[w*16+g];
315
56168
                if (!cpe->is_mask[w*16 + g]) {
316
47882
                    start += ics->swb_sizes[g];
317
47882
                    continue;
318
                }
319
8286
                if (cpe->ms_mask[w*16 + g])
320
2383
                    p *= -1;
321
288446
                for (i = 0; i < ics->swb_sizes[g]; i++) {
322
280160
                    float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
323
280160
                    cpe->ch[0].coeffs[start+i] = sum;
324
280160
                    cpe->ch[1].coeffs[start+i] = 0.0f;
325
                }
326
8286
                start += ics->swb_sizes[g];
327
            }
328
        }
329
    }
330
}
331
332
2071
static void apply_mid_side_stereo(ChannelElement *cpe)
333
{
334
    int w, w2, g, i;
335
2071
    IndividualChannelStream *ics = &cpe->ch[0].ics;
336
2071
    if (!cpe->common_window)
337
1078
        return;
338
2039
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
339
2193
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
340
1147
            int start = (w+w2) * 128;
341
51190
            for (g = 0; g < ics->num_swb; g++) {
342
                /* ms_mask can be used for other purposes in PNS and I/S,
343
                 * so must not apply M/S if any band uses either, even if
344
                 * ms_mask is set.
345
                 */
346

50043
                if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]
347
28910
                    || cpe->ch[0].band_type[w*16 + g] >= NOISE_BT
348
28910
                    || cpe->ch[1].band_type[w*16 + g] >= NOISE_BT) {
349
21133
                    start += ics->swb_sizes[g];
350
21133
                    continue;
351
                }
352
650946
                for (i = 0; i < ics->swb_sizes[g]; i++) {
353
622036
                    float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
354
622036
                    float R = L - cpe->ch[1].coeffs[start+i];
355
622036
                    cpe->ch[0].coeffs[start+i] = L;
356
622036
                    cpe->ch[1].coeffs[start+i] = R;
357
                }
358
28910
                start += ics->swb_sizes[g];
359
            }
360
        }
361
    }
362
}
363
364
/**
365
 * Encode scalefactor band coding type.
366
 */
367
11743
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
368
{
369
    int w;
370
371
11743
    if (s->coder->set_special_band_scalefactors)
372
11743
        s->coder->set_special_band_scalefactors(s, sce);
373
374
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
375
12430
        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
376
11743
}
377
378
/**
379
 * Encode scalefactors.
380
 */
381
11743
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
382
                                 SingleChannelElement *sce)
383
{
384
11743
    int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
385
11743
    int off_is = 0, noise_flag = 1;
386
    int i, w;
387
388
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
389
563974
        for (i = 0; i < sce->ics.max_sfb; i++) {
390
551544
            if (!sce->zeroes[w*16 + i]) {
391
521143
                if (sce->band_type[w*16 + i] == NOISE_BT) {
392
17340
                    diff = sce->sf_idx[w*16 + i] - off_pns;
393
17340
                    off_pns = sce->sf_idx[w*16 + i];
394
17340
                    if (noise_flag-- > 0) {
395
2242
                        put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
396
2242
                        continue;
397
                    }
398
503803
                } else if (sce->band_type[w*16 + i] == INTENSITY_BT  ||
399
497850
                           sce->band_type[w*16 + i] == INTENSITY_BT2) {
400
8191
                    diff = sce->sf_idx[w*16 + i] - off_is;
401
8191
                    off_is = sce->sf_idx[w*16 + i];
402
                } else {
403
495612
                    diff = sce->sf_idx[w*16 + i] - off_sf;
404
495612
                    off_sf = sce->sf_idx[w*16 + i];
405
                }
406
518901
                diff += SCALE_DIFF_ZERO;
407

518901
                av_assert0(diff >= 0 && diff <= 120);
408
518901
                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
409
            }
410
        }
411
    }
412
11743
}
413
414
/**
415
 * Encode pulse data.
416
 */
417
11743
static void encode_pulses(AACEncContext *s, Pulse *pulse)
418
{
419
    int i;
420
421
11743
    put_bits(&s->pb, 1, !!pulse->num_pulse);
422
11743
    if (!pulse->num_pulse)
423
11743
        return;
424
425
    put_bits(&s->pb, 2, pulse->num_pulse - 1);
426
    put_bits(&s->pb, 6, pulse->start);
427
    for (i = 0; i < pulse->num_pulse; i++) {
428
        put_bits(&s->pb, 5, pulse->pos[i]);
429
        put_bits(&s->pb, 4, pulse->amp[i]);
430
    }
431
}
432
433
/**
434
 * Encode spectral coefficients processed by psychoacoustic model.
435
 */
436
11743
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
437
{
438
    int start, i, w, w2;
439
440
24173
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
441
12430
        start = 0;
442
563974
        for (i = 0; i < sce->ics.max_sfb; i++) {
443
551544
            if (sce->zeroes[w*16 + i]) {
444
30401
                start += sce->ics.swb_sizes[i];
445
30401
                continue;
446
            }
447
1050888
            for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
448
529745
                s->coder->quantize_and_encode_band(s, &s->pb,
449
529745
                                                   &sce->coeffs[start + w2*128],
450
529745
                                                   NULL, sce->ics.swb_sizes[i],
451
529745
                                                   sce->sf_idx[w*16 + i],
452
529745
                                                   sce->band_type[w*16 + i],
453
                                                   s->lambda,
454
529745
                                                   sce->ics.window_clipping[w]);
455
            }
456
521143
            start += sce->ics.swb_sizes[i];
457
        }
458
    }
459
11743
}
460
461
/**
462
 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
463
 */
464
7284
static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
465
{
466
    int start, i, j, w;
467
468
7284
    if (sce->ics.clip_avoidance_factor < 1.0f) {
469
392
        for (w = 0; w < sce->ics.num_windows; w++) {
470
224
            start = 0;
471
8645
            for (i = 0; i < sce->ics.max_sfb; i++) {
472
8421
                float *swb_coeffs = &sce->coeffs[start + w*128];
473
167173
                for (j = 0; j < sce->ics.swb_sizes[i]; j++)
474
158752
                    swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
475
8421
                start += sce->ics.swb_sizes[i];
476
            }
477
        }
478
    }
479
7284
}
480
481
/**
482
 * Encode one channel of audio data.
483
 */
484
11743
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
485
                                     SingleChannelElement *sce,
486
                                     int common_window)
487
{
488
11743
    put_bits(&s->pb, 8, sce->sf_idx[0]);
489
11743
    if (!common_window) {
490
1685
        put_ics_info(s, &sce->ics);
491
1685
        if (s->coder->encode_main_pred)
492
1685
            s->coder->encode_main_pred(s, sce);
493
1685
        if (s->coder->encode_ltp_info)
494
1685
            s->coder->encode_ltp_info(s, sce, 0);
495
    }
496
11743
    encode_band_info(s, sce);
497
11743
    encode_scale_factors(avctx, s, sce);
498
11743
    encode_pulses(s, &sce->pulse);
499
11743
    put_bits(&s->pb, 1, !!sce->tns.present);
500
11743
    if (s->coder->encode_tns_info)
501
11743
        s->coder->encode_tns_info(s, sce);
502
11743
    put_bits(&s->pb, 1, 0); //ssr
503
11743
    encode_spectral_coeffs(s, sce);
504
11743
    return 0;
505
}
506
507
/**
508
 * Write some auxiliary information about the created AAC file.
509
 */
510
static void put_bitstream_info(AACEncContext *s, const char *name)
511
{
512
    int i, namelen, padbits;
513
514
    namelen = strlen(name) + 2;
515
    put_bits(&s->pb, 3, TYPE_FIL);
516
    put_bits(&s->pb, 4, FFMIN(namelen, 15));
517
    if (namelen >= 15)
518
        put_bits(&s->pb, 8, namelen - 14);
519
    put_bits(&s->pb, 4, 0); //extension type - filler
520
    padbits = -put_bits_count(&s->pb) & 7;
521
    align_put_bits(&s->pb);
522
    for (i = 0; i < namelen - 2; i++)
523
        put_bits(&s->pb, 8, name[i]);
524
    put_bits(&s->pb, 12 - padbits, 0);
525
}
526
527
/*
528
 * Copy input samples.
529
 * Channels are reordered from libavcodec's default order to AAC order.
530
 */
531
3816
static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
532
{
533
    int ch;
534
3816
    int end = 2048 + (frame ? frame->nb_samples : 0);
535
3816
    const uint8_t *channel_map = s->reorder_map;
536
537
    /* copy and remap input samples */
538
11125
    for (ch = 0; ch < s->channels; ch++) {
539
        /* copy last 1024 samples of previous frame to the start of the current frame */
540
7309
        memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
541
542
        /* copy new samples and zero any remaining samples */
543
7309
        if (frame) {
544
7259
            memcpy(&s->planar_samples[ch][2048],
545
7259
                   frame->extended_data[channel_map[ch]],
546
7259
                   frame->nb_samples * sizeof(s->planar_samples[0][0]));
547
        }
548
7309
        memset(&s->planar_samples[ch][end], 0,
549
7309
               (3072 - end) * sizeof(s->planar_samples[0][0]));
550
    }
551
3816
}
552
553
3827
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
554
                            const AVFrame *frame, int *got_packet_ptr)
555
{
556
3827
    AACEncContext *s = avctx->priv_data;
557
3827
    float **samples = s->planar_samples, *samples2, *la, *overlap;
558
    ChannelElement *cpe;
559
    SingleChannelElement *sce;
560
    IndividualChannelStream *ics;
561
    int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
562
    int target_bits, rate_bits, too_many_bits, too_few_bits;
563
3827
    int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
564
    int chan_el_counter[4];
565
    FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
566
567
    /* add current frame to queue */
568
3827
    if (frame) {
569
3794
        if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
570
            return ret;
571
    } else {
572

33
        if (!s->afq.remaining_samples || (!s->afq.frame_alloc && !s->afq.frame_count))
573
11
            return 0;
574
    }
575
576
3816
    copy_input_samples(s, frame);
577
3816
    if (s->psypp)
578
3816
        ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
579
580
3816
    if (!avctx->frame_number)
581
11
        return 0;
582
583
3805
    start_ch = 0;
584
7754
    for (i = 0; i < s->chan_map[0]; i++) {
585
3949
        FFPsyWindowInfo* wi = windows + start_ch;
586
3949
        tag      = s->chan_map[i+1];
587
3949
        chans    = tag == TYPE_CPE ? 2 : 1;
588
3949
        cpe      = &s->cpe[i];
589
11233
        for (ch = 0; ch < chans; ch++) {
590
            int k;
591
            float clip_avoidance_factor;
592
7284
            sce = &cpe->ch[ch];
593
7284
            ics = &sce->ics;
594
7284
            s->cur_channel = start_ch + ch;
595
7284
            overlap  = &samples[s->cur_channel][0];
596
7284
            samples2 = overlap + 1024;
597
7284
            la       = samples2 + (448+64);
598
7284
            if (!frame)
599
50
                la = NULL;
600
7284
            if (tag == TYPE_LFE) {
601
48
                wi[ch].window_type[0] = wi[ch].window_type[1] = ONLY_LONG_SEQUENCE;
602
48
                wi[ch].window_shape   = 0;
603
48
                wi[ch].num_windows    = 1;
604
48
                wi[ch].grouping[0]    = 1;
605
48
                wi[ch].clipping[0]    = 0;
606
607
                /* Only the lowest 12 coefficients are used in a LFE channel.
608
                 * The expression below results in only the bottom 8 coefficients
609
                 * being used for 11.025kHz to 16kHz sample rates.
610
                 */
611
48
                ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
612
            } else {
613
7236
                wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
614
7236
                                              ics->window_sequence[0]);
615
            }
616
7284
            ics->window_sequence[1] = ics->window_sequence[0];
617
7284
            ics->window_sequence[0] = wi[ch].window_type[0];
618
7284
            ics->use_kb_window[1]   = ics->use_kb_window[0];
619
7284
            ics->use_kb_window[0]   = wi[ch].window_shape;
620
7284
            ics->num_windows        = wi[ch].num_windows;
621
7284
            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
622

7284
            ics->num_swb            = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
623
7284
            ics->max_sfb            = FFMIN(ics->max_sfb, ics->num_swb);
624
14568
            ics->swb_offset         = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
625
7284
                                        ff_swb_offset_128 [s->samplerate_index]:
626
7120
                                        ff_swb_offset_1024[s->samplerate_index];
627
14568
            ics->tns_max_bands      = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
628
7284
                                        ff_tns_max_bands_128 [s->samplerate_index]:
629
7120
                                        ff_tns_max_bands_1024[s->samplerate_index];
630
631
15716
            for (w = 0; w < ics->num_windows; w++)
632
8432
                ics->group_len[w] = wi[ch].grouping[w];
633
634
            /* Calculate input sample maximums and evaluate clipping risk */
635
7284
            clip_avoidance_factor = 0.0f;
636
15716
            for (w = 0; w < ics->num_windows; w++) {
637
8432
                const float *wbuf = overlap + w * 128;
638
8432
                const int wlen = 2048 / ics->num_windows;
639
8432
                float max = 0;
640
                int j;
641
                /* mdct input is 2 * output */
642
14926064
                for (j = 0; j < wlen; j++)
643
14917632
                    max = FFMAX(max, fabsf(wbuf[j]));
644
8432
                wi[ch].clipping[w] = max;
645
            }
646
15716
            for (w = 0; w < ics->num_windows; w++) {
647
8432
                if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
648
176
                    ics->window_clipping[w] = 1;
649
176
                    clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
650
                } else {
651
8256
                    ics->window_clipping[w] = 0;
652
                }
653
            }
654
7284
            if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
655
168
                ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
656
            } else {
657
7116
                ics->clip_avoidance_factor = 1.0f;
658
            }
659
660
7284
            apply_window_and_mdct(s, sce, overlap);
661
662

7284
            if (s->options.ltp && s->coder->update_ltp) {
663
                s->coder->update_ltp(s, sce);
664
                apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
665
                s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);
666
            }
667
668
7466100
            for (k = 0; k < 1024; k++) {
669
7458816
                if (!(fabs(cpe->ch[ch].coeffs[k]) < 1E16)) { // Ensure headroom for energy calculation
670
                    av_log(avctx, AV_LOG_ERROR, "Input contains (near) NaN/+-Inf\n");
671
                    return AVERROR(EINVAL);
672
                }
673
            }
674
7284
            avoid_clipping(s, sce);
675
        }
676
3949
        start_ch += chans;
677
    }
678
3805
    if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
679
        return ret;
680
3805
    frame_bits = its = 0;
681
    do {
682
6253
        init_put_bits(&s->pb, avpkt->data, avpkt->size);
683
684

6253
        if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
685
            put_bitstream_info(s, LIBAVCODEC_IDENT);
686
6253
        start_ch = 0;
687
6253
        target_bits = 0;
688
6253
        memset(chan_el_counter, 0, sizeof(chan_el_counter));
689
12722
        for (i = 0; i < s->chan_map[0]; i++) {
690
6469
            FFPsyWindowInfo* wi = windows + start_ch;
691
            const float *coeffs[2];
692
6469
            tag      = s->chan_map[i+1];
693
6469
            chans    = tag == TYPE_CPE ? 2 : 1;
694
6469
            cpe      = &s->cpe[i];
695
6469
            cpe->common_window = 0;
696
6469
            memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
697
6469
            memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
698
6469
            put_bits(&s->pb, 3, tag);
699
6469
            put_bits(&s->pb, 4, chan_el_counter[tag]++);
700
18212
            for (ch = 0; ch < chans; ch++) {
701
11743
                sce = &cpe->ch[ch];
702
11743
                coeffs[ch] = sce->coeffs;
703
11743
                sce->ics.predictor_present = 0;
704
11743
                sce->ics.ltp.present = 0;
705
11743
                memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
706
11743
                memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
707
11743
                memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
708
1514847
                for (w = 0; w < 128; w++)
709
1503104
                    if (sce->band_type[w] > RESERVED_BT)
710
25469
                        sce->band_type[w] = 0;
711
            }
712
6469
            s->psy.bitres.alloc = -1;
713
6469
            s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
714
6469
            s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
715
6469
            if (s->psy.bitres.alloc > 0) {
716
                /* Lambda unused here on purpose, we need to take psy's unscaled allocation */
717
12938
                target_bits += s->psy.bitres.alloc
718
6469
                    * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
719
6469
                s->psy.bitres.alloc /= chans;
720
            }
721
6469
            s->cur_type = tag;
722
18212
            for (ch = 0; ch < chans; ch++) {
723
11743
                s->cur_channel = start_ch + ch;
724

11743
                if (s->options.pns && s->coder->mark_pns)
725
3515
                    s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
726
11743
                s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
727
            }
728
6469
            if (chans > 1
729
5274
                && wi[0].window_type[0] == wi[1].window_type[0]
730
5038
                && wi[0].window_shape   == wi[1].window_shape) {
731
732
5038
                cpe->common_window = 1;
733
10634
                for (w = 0; w < wi[0].num_windows; w++) {
734
5605
                    if (wi[0].grouping[w] != wi[1].grouping[w]) {
735
9
                        cpe->common_window = 0;
736
9
                        break;
737
                    }
738
                }
739
            }
740
18212
            for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
741
11743
                sce = &cpe->ch[ch];
742
11743
                s->cur_channel = start_ch + ch;
743

11743
                if (s->options.tns && s->coder->search_for_tns)
744
3515
                    s->coder->search_for_tns(s, sce);
745

11743
                if (s->options.tns && s->coder->apply_tns_filt)
746
3515
                    s->coder->apply_tns_filt(s, sce);
747
11743
                if (sce->tns.present)
748
46
                    tns_mode = 1;
749

11743
                if (s->options.pns && s->coder->search_for_pns)
750
3515
                    s->coder->search_for_pns(s, avctx, sce);
751
            }
752
6469
            s->cur_channel = start_ch;
753
6469
            if (s->options.intensity_stereo) { /* Intensity Stereo */
754
2355
                if (s->coder->search_for_is)
755
2355
                    s->coder->search_for_is(s, avctx, cpe);
756
2355
                if (cpe->is_mode) is_mode = 1;
757
2355
                apply_intensity_stereo(cpe);
758
            }
759
6469
            if (s->options.pred) { /* Prediction */
760
1248
                for (ch = 0; ch < chans; ch++) {
761
832
                    sce = &cpe->ch[ch];
762
832
                    s->cur_channel = start_ch + ch;
763

832
                    if (s->options.pred && s->coder->search_for_pred)
764
832
                        s->coder->search_for_pred(s, sce);
765
832
                    if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
766
                }
767
416
                if (s->coder->adjust_common_pred)
768
416
                    s->coder->adjust_common_pred(s, cpe);
769
1248
                for (ch = 0; ch < chans; ch++) {
770
832
                    sce = &cpe->ch[ch];
771
832
                    s->cur_channel = start_ch + ch;
772

832
                    if (s->options.pred && s->coder->apply_main_pred)
773
832
                        s->coder->apply_main_pred(s, sce);
774
                }
775
416
                s->cur_channel = start_ch;
776
            }
777
6469
            if (s->options.mid_side) { /* Mid/Side stereo */
778

2071
                if (s->options.mid_side == -1 && s->coder->search_for_ms)
779
1651
                    s->coder->search_for_ms(s, cpe);
780
420
                else if (cpe->common_window)
781
393
                    memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
782
2071
                apply_mid_side_stereo(cpe);
783
            }
784
6469
            adjust_frame_information(cpe, chans);
785
6469
            if (s->options.ltp) { /* LTP */
786
                for (ch = 0; ch < chans; ch++) {
787
                    sce = &cpe->ch[ch];
788
                    s->cur_channel = start_ch + ch;
789
                    if (s->coder->search_for_ltp)
790
                        s->coder->search_for_ltp(s, sce, cpe->common_window);
791
                    if (sce->ics.ltp.present) pred_mode = 1;
792
                }
793
                s->cur_channel = start_ch;
794
                if (s->coder->adjust_common_ltp)
795
                    s->coder->adjust_common_ltp(s, cpe);
796
            }
797
6469
            if (chans == 2) {
798
5274
                put_bits(&s->pb, 1, cpe->common_window);
799
5274
                if (cpe->common_window) {
800
5029
                    put_ics_info(s, &cpe->ch[0].ics);
801
5029
                    if (s->coder->encode_main_pred)
802
5029
                        s->coder->encode_main_pred(s, &cpe->ch[0]);
803
5029
                    if (s->coder->encode_ltp_info)
804
5029
                        s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
805
5029
                    encode_ms_info(&s->pb, cpe);
806
5029
                    if (cpe->ms_mode) ms_mode = 1;
807
                }
808
            }
809
18212
            for (ch = 0; ch < chans; ch++) {
810
11743
                s->cur_channel = start_ch + ch;
811
11743
                encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
812
            }
813
6469
            start_ch += chans;
814
        }
815
816
6253
        if (avctx->flags & AV_CODEC_FLAG_QSCALE) {
817
            /* When using a constant Q-scale, don't mess with lambda */
818
            break;
819
        }
820
821
        /* rate control stuff
822
         * allow between the nominal bitrate, and what psy's bit reservoir says to target
823
         * but drift towards the nominal bitrate always
824
         */
825
6253
        frame_bits = put_bits_count(&s->pb);
826
6253
        rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
827
6253
        rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
828
6253
        too_many_bits = FFMAX(target_bits, rate_bits);
829
6253
        too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
830
6253
        too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
831
832
        /* When using ABR, be strict (but only for increasing) */
833
6253
        too_few_bits = too_few_bits - too_few_bits/8;
834
6253
        too_many_bits = too_many_bits + too_many_bits/2;
835
836
6253
        if (   its == 0 /* for steady-state Q-scale tracking */
837

2448
            || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
838
497
            || frame_bits >= 6144 * s->channels - 3  )
839
        {
840
5756
            float ratio = ((float)rate_bits) / frame_bits;
841
842

5756
            if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
843
                /*
844
                 * This path is for steady-state Q-scale tracking
845
                 * When frame bits fall within the stable range, we still need to adjust
846
                 * lambda to maintain it like so in a stable fashion (large jumps in lambda
847
                 * create artifacts and should be avoided), but slowly
848
                 */
849
2859
                ratio = sqrtf(sqrtf(ratio));
850
2859
                ratio = av_clipf(ratio, 0.9f, 1.1f);
851
            } else {
852
                /* Not so fast though */
853
2897
                ratio = sqrtf(ratio);
854
            }
855
5756
            s->lambda = FFMIN(s->lambda * ratio, 65536.f);
856
857
            /* Keep iterating if we must reduce and lambda is in the sky */
858

5756
            if (ratio > 0.9f && ratio < 1.1f) {
859
                break;
860
            } else {
861


2448
                if (is_mode || ms_mode || tns_mode || pred_mode) {
862
1027
                    for (i = 0; i < s->chan_map[0]; i++) {
863
                        // Must restore coeffs
864
521
                        chans = tag == TYPE_CPE ? 2 : 1;
865
521
                        cpe = &s->cpe[i];
866
1543
                        for (ch = 0; ch < chans; ch++)
867
1022
                            memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
868
                    }
869
                }
870
2448
                its++;
871
            }
872
        } else {
873
            break;
874
        }
875
    } while (1);
876
877

3805
    if (s->options.ltp && s->coder->ltp_insert_new_frame)
878
        s->coder->ltp_insert_new_frame(s);
879
880
3805
    put_bits(&s->pb, 3, TYPE_END);
881
3805
    flush_put_bits(&s->pb);
882
883
3805
    s->last_frame_pb_count = put_bits_count(&s->pb);
884
885
3805
    s->lambda_sum += s->lambda;
886
3805
    s->lambda_count++;
887
888
3805
    ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
889
                       &avpkt->duration);
890
891
3805
    avpkt->size = put_bits_count(&s->pb) >> 3;
892
3805
    *got_packet_ptr = 1;
893
3805
    return 0;
894
}
895
896
11
static av_cold int aac_encode_end(AVCodecContext *avctx)
897
{
898
11
    AACEncContext *s = avctx->priv_data;
899
900
11
    av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);
901
902
11
    ff_mdct_end(&s->mdct1024);
903
11
    ff_mdct_end(&s->mdct128);
904
11
    ff_psy_end(&s->psy);
905
11
    ff_lpc_end(&s->lpc);
906
11
    if (s->psypp)
907
11
        ff_psy_preprocess_end(s->psypp);
908
11
    av_freep(&s->buffer.samples);
909
11
    av_freep(&s->cpe);
910
11
    av_freep(&s->fdsp);
911
11
    ff_af_queue_close(&s->afq);
912
11
    return 0;
913
}
914
915
11
static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
916
{
917
11
    int ret = 0;
918
919
11
    s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
920
11
    if (!s->fdsp)
921
        return AVERROR(ENOMEM);
922
923
    // window init
924
11
    ff_aac_float_common_init();
925
926
11
    if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
927
        return ret;
928
11
    if ((ret = ff_mdct_init(&s->mdct128,   8, 0, 32768.0)) < 0)
929
        return ret;
930
931
11
    return 0;
932
}
933
934
11
static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
935
{
936
    int ch;
937
11
    if (!FF_ALLOCZ_TYPED_ARRAY(s->buffer.samples, s->channels * 3 * 1024) ||
938
11
        !FF_ALLOCZ_TYPED_ARRAY(s->cpe,            s->chan_map[0]))
939
        return AVERROR(ENOMEM);
940
941
36
    for(ch = 0; ch < s->channels; ch++)
942
25
        s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
943
944
11
    return 0;
945
}
946
947
11
static av_cold int aac_encode_init(AVCodecContext *avctx)
948
{
949
11
    AACEncContext *s = avctx->priv_data;
950
11
    int i, ret = 0;
951
    const uint8_t *sizes[2];
952
    uint8_t grouping[AAC_MAX_CHANNELS];
953
    int lengths[2];
954
955
    /* Constants */
956
11
    s->last_frame_pb_count = 0;
957
11
    avctx->frame_size = 1024;
958
11
    avctx->initial_padding = 1024;
959
11
    s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
960
961
    /* Channel map and unspecified bitrate guessing */
962
11
    s->channels = avctx->channels;
963
964
11
    s->needs_pce = 1;
965
25
    for (i = 0; i < FF_ARRAY_ELEMS(aac_normal_chan_layouts); i++) {
966
25
        if (avctx->channel_layout == aac_normal_chan_layouts[i]) {
967
11
            s->needs_pce = s->options.pce;
968
11
            break;
969
        }
970
    }
971
972
11
    if (s->needs_pce) {
973
        char buf[64];
974
        for (i = 0; i < FF_ARRAY_ELEMS(aac_pce_configs); i++)
975
            if (avctx->channel_layout == aac_pce_configs[i].layout)
976
                break;
977
        av_get_channel_layout_string(buf, sizeof(buf), -1, avctx->channel_layout);
978
        ERROR_IF(i == FF_ARRAY_ELEMS(aac_pce_configs), "Unsupported channel layout \"%s\"\n", buf);
979
        av_log(avctx, AV_LOG_INFO, "Using a PCE to encode channel layout \"%s\"\n", buf);
980
        s->pce = aac_pce_configs[i];
981
        s->reorder_map = s->pce.reorder_map;
982
        s->chan_map = s->pce.config_map;
983
    } else {
984
11
        s->reorder_map = aac_chan_maps[s->channels - 1];
985
11
        s->chan_map = aac_chan_configs[s->channels - 1];
986
    }
987
988
11
    if (!avctx->bit_rate) {
989
9
        for (i = 1; i <= s->chan_map[0]; i++) {
990
9
            avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */
991
3
                               s->chan_map[i] == TYPE_LFE ? 16000  : /* LFE  */
992
                                                            69000  ; /* SCE  */
993
        }
994
    }
995
996
    /* Samplerate */
997
54
    for (i = 0; i < 16; i++)
998
54
        if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
999
11
            break;
1000
11
    s->samplerate_index = i;
1001

11
    ERROR_IF(s->samplerate_index == 16 ||
1002
             s->samplerate_index >= ff_aac_swb_size_1024_len ||
1003
             s->samplerate_index >= ff_aac_swb_size_128_len,
1004
             "Unsupported sample rate %d\n", avctx->sample_rate);
1005
1006
    /* Bitrate limiting */
1007
11
    WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
1008
             "Too many bits %f > %d per frame requested, clamping to max\n",
1009
             1024.0 * avctx->bit_rate / avctx->sample_rate,
1010
             6144 * s->channels);
1011
11
    avctx->bit_rate = (int64_t)FFMIN(6144 * s->channels / 1024.0 * avctx->sample_rate,
1012
                                     avctx->bit_rate);
1013
1014
    /* Profile and option setting */
1015
11
    avctx->profile = avctx->profile == FF_PROFILE_UNKNOWN ? FF_PROFILE_AAC_LOW :
1016
                     avctx->profile;
1017
21
    for (i = 0; i < FF_ARRAY_ELEMS(aacenc_profiles); i++)
1018
21
        if (avctx->profile == aacenc_profiles[i])
1019
11
            break;
1020
11
    if (avctx->profile == FF_PROFILE_MPEG2_AAC_LOW) {
1021
        avctx->profile = FF_PROFILE_AAC_LOW;
1022
        ERROR_IF(s->options.pred,
1023
                 "Main prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1024
        ERROR_IF(s->options.ltp,
1025
                 "LTP prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1026
        WARN_IF(s->options.pns,
1027
                "PNS unavailable in the \"mpeg2_aac_low\" profile, turning off\n");
1028
        s->options.pns = 0;
1029
11
    } else if (avctx->profile == FF_PROFILE_AAC_LTP) {
1030
        s->options.ltp = 1;
1031
        ERROR_IF(s->options.pred,
1032
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1033
11
    } else if (avctx->profile == FF_PROFILE_AAC_MAIN) {
1034
1
        s->options.pred = 1;
1035
1
        ERROR_IF(s->options.ltp,
1036
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1037
10
    } else if (s->options.ltp) {
1038
        avctx->profile = FF_PROFILE_AAC_LTP;
1039
        WARN_IF(1,
1040
                "Chainging profile to \"aac_ltp\"\n");
1041
        ERROR_IF(s->options.pred,
1042
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1043
10
    } else if (s->options.pred) {
1044
        avctx->profile = FF_PROFILE_AAC_MAIN;
1045
        WARN_IF(1,
1046
                "Chainging profile to \"aac_main\"\n");
1047
        ERROR_IF(s->options.ltp,
1048
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1049
    }
1050
11
    s->profile = avctx->profile;
1051
1052
    /* Coder limitations */
1053
11
    s->coder = &ff_aac_coders[s->options.coder];
1054
11
    if (s->options.coder == AAC_CODER_ANMR) {
1055
        ERROR_IF(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1056
                 "The ANMR coder is considered experimental, add -strict -2 to enable!\n");
1057
        s->options.intensity_stereo = 0;
1058
        s->options.pns = 0;
1059
    }
1060

11
    ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1061
             "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");
1062
1063
    /* M/S introduces horrible artifacts with multichannel files, this is temporary */
1064
11
    if (s->channels > 3)
1065
1
        s->options.mid_side = 0;
1066
1067
11
    if ((ret = dsp_init(avctx, s)) < 0)
1068
        return ret;
1069
1070
11
    if ((ret = alloc_buffers(avctx, s)) < 0)
1071
        return ret;
1072
1073
11
    if ((ret = put_audio_specific_config(avctx)))
1074
        return ret;
1075
1076
11
    sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];
1077
11
    sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];
1078
11
    lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];
1079
11
    lengths[1] = ff_aac_num_swb_128[s->samplerate_index];
1080
25
    for (i = 0; i < s->chan_map[0]; i++)
1081
14
        grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
1082
11
    if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
1083
11
                           s->chan_map[0], grouping)) < 0)
1084
        return ret;
1085
11
    s->psypp = ff_psy_preprocess_init(avctx);
1086
11
    ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
1087
11
    s->random_state = 0x1f2e3d4c;
1088
1089
11
    s->abs_pow34   = abs_pow34_v;
1090
11
    s->quant_bands = quantize_bands;
1091
1092
    if (ARCH_X86)
1093
11
        ff_aac_dsp_init_x86(s);
1094
1095
    if (HAVE_MIPSDSP)
1096
        ff_aac_coder_init_mips(s);
1097
1098
11
    ff_af_queue_init(avctx, &s->afq);
1099
11
    ff_aac_tableinit();
1100
1101
11
    return 0;
1102
}
1103
1104
#define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1105
static const AVOption aacenc_options[] = {
1106
    {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_FAST}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},
1107
        {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1108
        {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1109
        {"fast",     "Default fast search",       0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1110
    {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},
1111
    {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1112
    {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1113
    {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1114
    {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1115
    {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1116
    {"aac_pce", "Forces the use of PCEs", offsetof(AACEncContext, options.pce), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1117
    FF_AAC_PROFILE_OPTS
1118
    {NULL}
1119
};
1120
1121
static const AVClass aacenc_class = {
1122
    .class_name = "AAC encoder",
1123
    .item_name  = av_default_item_name,
1124
    .option     = aacenc_options,
1125
    .version    = LIBAVUTIL_VERSION_INT,
1126
};
1127
1128
static const AVCodecDefault aac_encode_defaults[] = {
1129
    { "b", "0" },
1130
    { NULL }
1131
};
1132
1133
AVCodec ff_aac_encoder = {
1134
    .name           = "aac",
1135
    .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
1136
    .type           = AVMEDIA_TYPE_AUDIO,
1137
    .id             = AV_CODEC_ID_AAC,
1138
    .priv_data_size = sizeof(AACEncContext),
1139
    .init           = aac_encode_init,
1140
    .encode2        = aac_encode_frame,
1141
    .close          = aac_encode_end,
1142
    .defaults       = aac_encode_defaults,
1143
    .supported_samplerates = mpeg4audio_sample_rates,
1144
    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1145
    .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,
1146
    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
1147
                                                     AV_SAMPLE_FMT_NONE },
1148
    .priv_class     = &aacenc_class,
1149
};