GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/aacenc.c Lines: 573 663 86.4 %
Date: 2019-11-22 03:34:36 Branches: 361 486 74.3 %

Line Branch Exec Source
1
/*
2
 * AAC encoder
3
 * Copyright (C) 2008 Konstantin Shishkov
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21
22
/**
23
 * @file
24
 * AAC encoder
25
 */
26
27
/***********************************
28
 *              TODOs:
29
 * add sane pulse detection
30
 ***********************************/
31
32
#include "libavutil/libm.h"
33
#include "libavutil/thread.h"
34
#include "libavutil/float_dsp.h"
35
#include "libavutil/opt.h"
36
#include "avcodec.h"
37
#include "put_bits.h"
38
#include "internal.h"
39
#include "mpeg4audio.h"
40
#include "kbdwin.h"
41
#include "sinewin.h"
42
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 AVOnce aac_table_init = AV_ONCE_INIT;
52
53
static void put_pce(PutBitContext *pb, AVCodecContext *avctx)
54
{
55
    int i, j;
56
    AACEncContext *s = avctx->priv_data;
57
    AACPCEInfo *pce = &s->pce;
58
    const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
59
    const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
60
61
    put_bits(pb, 4, 0);
62
63
    put_bits(pb, 2, avctx->profile);
64
    put_bits(pb, 4, s->samplerate_index);
65
66
    put_bits(pb, 4, pce->num_ele[0]); /* Front */
67
    put_bits(pb, 4, pce->num_ele[1]); /* Side */
68
    put_bits(pb, 4, pce->num_ele[2]); /* Back */
69
    put_bits(pb, 2, pce->num_ele[3]); /* LFE */
70
    put_bits(pb, 3, 0); /* Assoc data */
71
    put_bits(pb, 4, 0); /* CCs */
72
73
    put_bits(pb, 1, 0); /* Stereo mixdown */
74
    put_bits(pb, 1, 0); /* Mono mixdown */
75
    put_bits(pb, 1, 0); /* Something else */
76
77
    for (i = 0; i < 4; i++) {
78
        for (j = 0; j < pce->num_ele[i]; j++) {
79
            if (i < 3)
80
                put_bits(pb, 1, pce->pairing[i][j]);
81
            put_bits(pb, 4, pce->index[i][j]);
82
        }
83
    }
84
85
    avpriv_align_put_bits(pb);
86
    put_bits(pb, 8, strlen(aux_data));
87
    avpriv_put_string(pb, aux_data, 0);
88
}
89
90
/**
91
 * Make AAC audio config object.
92
 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
93
 */
94
11
static int put_audio_specific_config(AVCodecContext *avctx)
95
{
96
    PutBitContext pb;
97
11
    AACEncContext *s = avctx->priv_data;
98
11
    int channels = (!s->needs_pce)*(s->channels - (s->channels == 8 ? 1 : 0));
99
11
    const int max_size = 32;
100
101
11
    avctx->extradata = av_mallocz(max_size);
102
11
    if (!avctx->extradata)
103
        return AVERROR(ENOMEM);
104
105
11
    init_put_bits(&pb, avctx->extradata, max_size);
106
11
    put_bits(&pb, 5, s->profile+1); //profile
107
11
    put_bits(&pb, 4, s->samplerate_index); //sample rate index
108
11
    put_bits(&pb, 4, channels);
109
    //GASpecificConfig
110
11
    put_bits(&pb, 1, 0); //frame length - 1024 samples
111
11
    put_bits(&pb, 1, 0); //does not depend on core coder
112
11
    put_bits(&pb, 1, 0); //is not extension
113
11
    if (s->needs_pce)
114
        put_pce(&pb, avctx);
115
116
    //Explicitly Mark SBR absent
117
11
    put_bits(&pb, 11, 0x2b7); //sync extension
118
11
    put_bits(&pb, 5,  AOT_SBR);
119
11
    put_bits(&pb, 1,  0);
120
11
    flush_put_bits(&pb);
121
11
    avctx->extradata_size = put_bits_count(&pb) >> 3;
122
123
11
    return 0;
124
}
125
126
11643
void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
127
{
128
11643
    ++s->quantize_band_cost_cache_generation;
129
11643
    if (s->quantize_band_cost_cache_generation == 0) {
130
        memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
131
        s->quantize_band_cost_cache_generation = 1;
132
    }
133
11643
}
134
135
#define WINDOW_FUNC(type) \
136
static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
137
                                    SingleChannelElement *sce, \
138
                                    const float *audio)
139
140
6899
WINDOW_FUNC(only_long)
141
{
142
6899
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
143
6899
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
144
6899
    float *out = sce->ret_buf;
145
146
6899
    fdsp->vector_fmul        (out,        audio,        lwindow, 1024);
147
6899
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
148
6899
}
149
150
119
WINDOW_FUNC(long_start)
151
{
152
119
    const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
153
119
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
154
119
    float *out = sce->ret_buf;
155
156
119
    fdsp->vector_fmul(out, audio, lwindow, 1024);
157
119
    memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
158
119
    fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
159
119
    memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
160
119
}
161
162
102
WINDOW_FUNC(long_stop)
163
{
164
102
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
165
102
    const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
166
102
    float *out = sce->ret_buf;
167
168
102
    memset(out, 0, sizeof(out[0]) * 448);
169
102
    fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
170
102
    memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
171
102
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
172
102
}
173
174
164
WINDOW_FUNC(eight_short)
175
{
176
164
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
177
164
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
178
164
    const float *in = audio + 448;
179
164
    float *out = sce->ret_buf;
180
    int w;
181
182
1476
    for (w = 0; w < 8; w++) {
183
1312
        fdsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
184
1312
        out += 128;
185
1312
        in  += 128;
186
1312
        fdsp->vector_fmul_reverse(out, in, swindow, 128);
187
1312
        out += 128;
188
    }
189
164
}
190
191
static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
192
                                     SingleChannelElement *sce,
193
                                     const float *audio) = {
194
    [ONLY_LONG_SEQUENCE]   = apply_only_long_window,
195
    [LONG_START_SEQUENCE]  = apply_long_start_window,
196
    [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
197
    [LONG_STOP_SEQUENCE]   = apply_long_stop_window
198
};
199
200
7284
static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
201
                                  float *audio)
202
{
203
    int i;
204
7284
    const float *output = sce->ret_buf;
205
206
7284
    apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
207
208
7284
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
209
7120
        s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
210
    else
211
1476
        for (i = 0; i < 1024; i += 128)
212
1312
            s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
213
7284
    memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
214
7284
    memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
215
7284
}
216
217
/**
218
 * Encode ics_info element.
219
 * @see Table 4.6 (syntax of ics_info)
220
 */
221
6711
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
222
{
223
    int w;
224
225
6711
    put_bits(&s->pb, 1, 0);                // ics_reserved bit
226
6711
    put_bits(&s->pb, 2, info->window_sequence[0]);
227
6711
    put_bits(&s->pb, 1, info->use_kb_window[0]);
228
6711
    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
229
6526
        put_bits(&s->pb, 6, info->max_sfb);
230
6526
        put_bits(&s->pb, 1, !!info->predictor_present);
231
    } else {
232
185
        put_bits(&s->pb, 4, info->max_sfb);
233
1480
        for (w = 1; w < 8; w++)
234
1295
            put_bits(&s->pb, 1, !info->group_len[w]);
235
    }
236
6711
}
237
238
/**
239
 * Encode MS data.
240
 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
241
 */
242
5028
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
243
{
244
    int i, w;
245
246
5028
    put_bits(pb, 2, cpe->ms_mode);
247
5028
    if (cpe->ms_mode == 1)
248
1826
        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
249
41222
            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
250
40291
                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
251
5028
}
252
253
/**
254
 * Produce integer coefficients from scalefactors provided by the model.
255
 */
256
6467
static void adjust_frame_information(ChannelElement *cpe, int chans)
257
{
258
    int i, w, w2, g, ch;
259
    int maxsfb, cmaxsfb;
260
261
18206
    for (ch = 0; ch < chans; ch++) {
262
11739
        IndividualChannelStream *ics = &cpe->ch[ch].ics;
263
11739
        maxsfb = 0;
264
11739
        cpe->ch[ch].pulse.num_pulse = 0;
265
24159
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
266
26021
            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
267

41746
                for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
268
                    ;
269
13601
                maxsfb = FFMAX(maxsfb, cmaxsfb);
270
            }
271
        }
272
11739
        ics->max_sfb = maxsfb;
273
274
        //adjust zero bands for window groups
275
24159
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
276
563749
            for (g = 0; g < ics->max_sfb; g++) {
277
551329
                i = 1;
278
566819
                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
279
551669
                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
280
536179
                        i = 0;
281
536179
                        break;
282
                    }
283
                }
284
551329
                cpe->ch[ch].zeroes[w*16 + g] = i;
285
            }
286
        }
287
    }
288
289

6467
    if (chans > 1 && cpe->common_window) {
290
5028
        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
291
5028
        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
292
5028
        int msc = 0;
293
5028
        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
294
5028
        ics1->max_sfb = ics0->max_sfb;
295
10623
        for (w = 0; w < ics0->num_windows*16; w += 16)
296
247484
            for (i = 0; i < ics0->max_sfb; i++)
297
241889
                if (cpe->ms_mask[w+i])
298
30094
                    msc++;
299

5028
        if (msc == 0 || ics0->max_sfb == 0)
300
3741
            cpe->ms_mode = 0;
301
        else
302
1287
            cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
303
    }
304
6467
}
305
306
2355
static void apply_intensity_stereo(ChannelElement *cpe)
307
{
308
    int w, w2, g, i;
309
2355
    IndividualChannelStream *ics = &cpe->ch[0].ics;
310
2355
    if (!cpe->common_window)
311
1237
        return;
312
2289
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
313
2443
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
314
1272
            int start = (w+w2) * 128;
315
57440
            for (g = 0; g < ics->num_swb; g++) {
316
56168
                int p  = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
317
56168
                float scale = cpe->ch[0].is_ener[w*16+g];
318
56168
                if (!cpe->is_mask[w*16 + g]) {
319
47945
                    start += ics->swb_sizes[g];
320
47945
                    continue;
321
                }
322
8223
                if (cpe->ms_mask[w*16 + g])
323
2441
                    p *= -1;
324
290175
                for (i = 0; i < ics->swb_sizes[g]; i++) {
325
281952
                    float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
326
281952
                    cpe->ch[0].coeffs[start+i] = sum;
327
281952
                    cpe->ch[1].coeffs[start+i] = 0.0f;
328
                }
329
8223
                start += ics->swb_sizes[g];
330
            }
331
        }
332
    }
333
}
334
335
2069
static void apply_mid_side_stereo(ChannelElement *cpe)
336
{
337
    int w, w2, g, i;
338
2069
    IndividualChannelStream *ics = &cpe->ch[0].ics;
339
2069
    if (!cpe->common_window)
340
1077
        return;
341
2037
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
342
2191
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
343
1146
            int start = (w+w2) * 128;
344
51140
            for (g = 0; g < ics->num_swb; g++) {
345
                /* ms_mask can be used for other purposes in PNS and I/S,
346
                 * so must not apply M/S if any band uses either, even if
347
                 * ms_mask is set.
348
                 */
349

49994
                if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]
350
28861
                    || cpe->ch[0].band_type[w*16 + g] >= NOISE_BT
351
28861
                    || cpe->ch[1].band_type[w*16 + g] >= NOISE_BT) {
352
21133
                    start += ics->swb_sizes[g];
353
21133
                    continue;
354
                }
355
649873
                for (i = 0; i < ics->swb_sizes[g]; i++) {
356
621012
                    float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
357
621012
                    float R = L - cpe->ch[1].coeffs[start+i];
358
621012
                    cpe->ch[0].coeffs[start+i] = L;
359
621012
                    cpe->ch[1].coeffs[start+i] = R;
360
                }
361
28861
                start += ics->swb_sizes[g];
362
            }
363
        }
364
    }
365
}
366
367
/**
368
 * Encode scalefactor band coding type.
369
 */
370
11739
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
371
{
372
    int w;
373
374
11739
    if (s->coder->set_special_band_scalefactors)
375
11739
        s->coder->set_special_band_scalefactors(s, sce);
376
377
24159
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
378
12420
        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
379
11739
}
380
381
/**
382
 * Encode scalefactors.
383
 */
384
11739
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
385
                                 SingleChannelElement *sce)
386
{
387
11739
    int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
388
11739
    int off_is = 0, noise_flag = 1;
389
    int i, w;
390
391
24159
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
392
563754
        for (i = 0; i < sce->ics.max_sfb; i++) {
393
551334
            if (!sce->zeroes[w*16 + i]) {
394
521193
                if (sce->band_type[w*16 + i] == NOISE_BT) {
395
17340
                    diff = sce->sf_idx[w*16 + i] - off_pns;
396
17340
                    off_pns = sce->sf_idx[w*16 + i];
397
17340
                    if (noise_flag-- > 0) {
398
2242
                        put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
399
2242
                        continue;
400
                    }
401
503853
                } else if (sce->band_type[w*16 + i] == INTENSITY_BT  ||
402
498009
                           sce->band_type[w*16 + i] == INTENSITY_BT2) {
403
8132
                    diff = sce->sf_idx[w*16 + i] - off_is;
404
8132
                    off_is = sce->sf_idx[w*16 + i];
405
                } else {
406
495721
                    diff = sce->sf_idx[w*16 + i] - off_sf;
407
495721
                    off_sf = sce->sf_idx[w*16 + i];
408
                }
409
518951
                diff += SCALE_DIFF_ZERO;
410

518951
                av_assert0(diff >= 0 && diff <= 120);
411
518951
                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
412
            }
413
        }
414
    }
415
11739
}
416
417
/**
418
 * Encode pulse data.
419
 */
420
11739
static void encode_pulses(AACEncContext *s, Pulse *pulse)
421
{
422
    int i;
423
424
11739
    put_bits(&s->pb, 1, !!pulse->num_pulse);
425
11739
    if (!pulse->num_pulse)
426
11739
        return;
427
428
    put_bits(&s->pb, 2, pulse->num_pulse - 1);
429
    put_bits(&s->pb, 6, pulse->start);
430
    for (i = 0; i < pulse->num_pulse; i++) {
431
        put_bits(&s->pb, 5, pulse->pos[i]);
432
        put_bits(&s->pb, 4, pulse->amp[i]);
433
    }
434
}
435
436
/**
437
 * Encode spectral coefficients processed by psychoacoustic model.
438
 */
439
11739
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
440
{
441
    int start, i, w, w2;
442
443
24159
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
444
12420
        start = 0;
445
563754
        for (i = 0; i < sce->ics.max_sfb; i++) {
446
551334
            if (sce->zeroes[w*16 + i]) {
447
30141
                start += sce->ics.swb_sizes[i];
448
30141
                continue;
449
            }
450
1050921
            for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
451
529728
                s->coder->quantize_and_encode_band(s, &s->pb,
452
529728
                                                   &sce->coeffs[start + w2*128],
453
529728
                                                   NULL, sce->ics.swb_sizes[i],
454
529728
                                                   sce->sf_idx[w*16 + i],
455
529728
                                                   sce->band_type[w*16 + i],
456
                                                   s->lambda,
457
529728
                                                   sce->ics.window_clipping[w]);
458
            }
459
521193
            start += sce->ics.swb_sizes[i];
460
        }
461
    }
462
11739
}
463
464
/**
465
 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
466
 */
467
7284
static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
468
{
469
    int start, i, j, w;
470
471
7284
    if (sce->ics.clip_avoidance_factor < 1.0f) {
472
392
        for (w = 0; w < sce->ics.num_windows; w++) {
473
224
            start = 0;
474
8645
            for (i = 0; i < sce->ics.max_sfb; i++) {
475
8421
                float *swb_coeffs = &sce->coeffs[start + w*128];
476
167173
                for (j = 0; j < sce->ics.swb_sizes[i]; j++)
477
158752
                    swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
478
8421
                start += sce->ics.swb_sizes[i];
479
            }
480
        }
481
    }
482
7284
}
483
484
/**
485
 * Encode one channel of audio data.
486
 */
487
11739
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
488
                                     SingleChannelElement *sce,
489
                                     int common_window)
490
{
491
11739
    put_bits(&s->pb, 8, sce->sf_idx[0]);
492
11739
    if (!common_window) {
493
1683
        put_ics_info(s, &sce->ics);
494
1683
        if (s->coder->encode_main_pred)
495
1683
            s->coder->encode_main_pred(s, sce);
496
1683
        if (s->coder->encode_ltp_info)
497
1683
            s->coder->encode_ltp_info(s, sce, 0);
498
    }
499
11739
    encode_band_info(s, sce);
500
11739
    encode_scale_factors(avctx, s, sce);
501
11739
    encode_pulses(s, &sce->pulse);
502
11739
    put_bits(&s->pb, 1, !!sce->tns.present);
503
11739
    if (s->coder->encode_tns_info)
504
11739
        s->coder->encode_tns_info(s, sce);
505
11739
    put_bits(&s->pb, 1, 0); //ssr
506
11739
    encode_spectral_coeffs(s, sce);
507
11739
    return 0;
508
}
509
510
/**
511
 * Write some auxiliary information about the created AAC file.
512
 */
513
17
static void put_bitstream_info(AACEncContext *s, const char *name)
514
{
515
    int i, namelen, padbits;
516
517
17
    namelen = strlen(name) + 2;
518
17
    put_bits(&s->pb, 3, TYPE_FIL);
519
17
    put_bits(&s->pb, 4, FFMIN(namelen, 15));
520
17
    if (namelen >= 15)
521
17
        put_bits(&s->pb, 8, namelen - 14);
522
17
    put_bits(&s->pb, 4, 0); //extension type - filler
523
17
    padbits = -put_bits_count(&s->pb) & 7;
524
17
    avpriv_align_put_bits(&s->pb);
525
238
    for (i = 0; i < namelen - 2; i++)
526
221
        put_bits(&s->pb, 8, name[i]);
527
17
    put_bits(&s->pb, 12 - padbits, 0);
528
17
}
529
530
/*
531
 * Copy input samples.
532
 * Channels are reordered from libavcodec's default order to AAC order.
533
 */
534
3816
static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
535
{
536
    int ch;
537
3816
    int end = 2048 + (frame ? frame->nb_samples : 0);
538
3816
    const uint8_t *channel_map = s->reorder_map;
539
540
    /* copy and remap input samples */
541
11125
    for (ch = 0; ch < s->channels; ch++) {
542
        /* copy last 1024 samples of previous frame to the start of the current frame */
543
7309
        memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
544
545
        /* copy new samples and zero any remaining samples */
546
7309
        if (frame) {
547
7259
            memcpy(&s->planar_samples[ch][2048],
548
7259
                   frame->extended_data[channel_map[ch]],
549
7259
                   frame->nb_samples * sizeof(s->planar_samples[0][0]));
550
        }
551
7309
        memset(&s->planar_samples[ch][end], 0,
552
7309
               (3072 - end) * sizeof(s->planar_samples[0][0]));
553
    }
554
3816
}
555
556
3827
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
557
                            const AVFrame *frame, int *got_packet_ptr)
558
{
559
3827
    AACEncContext *s = avctx->priv_data;
560
3827
    float **samples = s->planar_samples, *samples2, *la, *overlap;
561
    ChannelElement *cpe;
562
    SingleChannelElement *sce;
563
    IndividualChannelStream *ics;
564
    int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
565
    int target_bits, rate_bits, too_many_bits, too_few_bits;
566
3827
    int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
567
    int chan_el_counter[4];
568
    FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
569
570
    /* add current frame to queue */
571
3827
    if (frame) {
572
3794
        if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
573
            return ret;
574
    } else {
575

33
        if (!s->afq.remaining_samples || (!s->afq.frame_alloc && !s->afq.frame_count))
576
11
            return 0;
577
    }
578
579
3816
    copy_input_samples(s, frame);
580
3816
    if (s->psypp)
581
3816
        ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
582
583
3816
    if (!avctx->frame_number)
584
11
        return 0;
585
586
3805
    start_ch = 0;
587
7754
    for (i = 0; i < s->chan_map[0]; i++) {
588
3949
        FFPsyWindowInfo* wi = windows + start_ch;
589
3949
        tag      = s->chan_map[i+1];
590
3949
        chans    = tag == TYPE_CPE ? 2 : 1;
591
3949
        cpe      = &s->cpe[i];
592
11233
        for (ch = 0; ch < chans; ch++) {
593
            int k;
594
            float clip_avoidance_factor;
595
7284
            sce = &cpe->ch[ch];
596
7284
            ics = &sce->ics;
597
7284
            s->cur_channel = start_ch + ch;
598
7284
            overlap  = &samples[s->cur_channel][0];
599
7284
            samples2 = overlap + 1024;
600
7284
            la       = samples2 + (448+64);
601
7284
            if (!frame)
602
50
                la = NULL;
603
7284
            if (tag == TYPE_LFE) {
604
48
                wi[ch].window_type[0] = wi[ch].window_type[1] = ONLY_LONG_SEQUENCE;
605
48
                wi[ch].window_shape   = 0;
606
48
                wi[ch].num_windows    = 1;
607
48
                wi[ch].grouping[0]    = 1;
608
48
                wi[ch].clipping[0]    = 0;
609
610
                /* Only the lowest 12 coefficients are used in a LFE channel.
611
                 * The expression below results in only the bottom 8 coefficients
612
                 * being used for 11.025kHz to 16kHz sample rates.
613
                 */
614
48
                ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
615
            } else {
616
7236
                wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
617
7236
                                              ics->window_sequence[0]);
618
            }
619
7284
            ics->window_sequence[1] = ics->window_sequence[0];
620
7284
            ics->window_sequence[0] = wi[ch].window_type[0];
621
7284
            ics->use_kb_window[1]   = ics->use_kb_window[0];
622
7284
            ics->use_kb_window[0]   = wi[ch].window_shape;
623
7284
            ics->num_windows        = wi[ch].num_windows;
624
7284
            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
625

7284
            ics->num_swb            = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
626
7284
            ics->max_sfb            = FFMIN(ics->max_sfb, ics->num_swb);
627
14568
            ics->swb_offset         = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
628
7284
                                        ff_swb_offset_128 [s->samplerate_index]:
629
7120
                                        ff_swb_offset_1024[s->samplerate_index];
630
14568
            ics->tns_max_bands      = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
631
7284
                                        ff_tns_max_bands_128 [s->samplerate_index]:
632
7120
                                        ff_tns_max_bands_1024[s->samplerate_index];
633
634
15716
            for (w = 0; w < ics->num_windows; w++)
635
8432
                ics->group_len[w] = wi[ch].grouping[w];
636
637
            /* Calculate input sample maximums and evaluate clipping risk */
638
7284
            clip_avoidance_factor = 0.0f;
639
15716
            for (w = 0; w < ics->num_windows; w++) {
640
8432
                const float *wbuf = overlap + w * 128;
641
8432
                const int wlen = 2048 / ics->num_windows;
642
8432
                float max = 0;
643
                int j;
644
                /* mdct input is 2 * output */
645
14926064
                for (j = 0; j < wlen; j++)
646
14917632
                    max = FFMAX(max, fabsf(wbuf[j]));
647
8432
                wi[ch].clipping[w] = max;
648
            }
649
15716
            for (w = 0; w < ics->num_windows; w++) {
650
8432
                if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
651
176
                    ics->window_clipping[w] = 1;
652
176
                    clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
653
                } else {
654
8256
                    ics->window_clipping[w] = 0;
655
                }
656
            }
657
7284
            if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
658
168
                ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
659
            } else {
660
7116
                ics->clip_avoidance_factor = 1.0f;
661
            }
662
663
7284
            apply_window_and_mdct(s, sce, overlap);
664
665

7284
            if (s->options.ltp && s->coder->update_ltp) {
666
                s->coder->update_ltp(s, sce);
667
                apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
668
                s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);
669
            }
670
671
7466100
            for (k = 0; k < 1024; k++) {
672
7458816
                if (!(fabs(cpe->ch[ch].coeffs[k]) < 1E16)) { // Ensure headroom for energy calculation
673
                    av_log(avctx, AV_LOG_ERROR, "Input contains (near) NaN/+-Inf\n");
674
                    return AVERROR(EINVAL);
675
                }
676
            }
677
7284
            avoid_clipping(s, sce);
678
        }
679
3949
        start_ch += chans;
680
    }
681
3805
    if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
682
        return ret;
683
3805
    frame_bits = its = 0;
684
    do {
685
6251
        init_put_bits(&s->pb, avpkt->data, avpkt->size);
686
687

6251
        if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
688
17
            put_bitstream_info(s, LIBAVCODEC_IDENT);
689
6251
        start_ch = 0;
690
6251
        target_bits = 0;
691
6251
        memset(chan_el_counter, 0, sizeof(chan_el_counter));
692
12718
        for (i = 0; i < s->chan_map[0]; i++) {
693
6467
            FFPsyWindowInfo* wi = windows + start_ch;
694
            const float *coeffs[2];
695
6467
            tag      = s->chan_map[i+1];
696
6467
            chans    = tag == TYPE_CPE ? 2 : 1;
697
6467
            cpe      = &s->cpe[i];
698
6467
            cpe->common_window = 0;
699
6467
            memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
700
6467
            memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
701
6467
            put_bits(&s->pb, 3, tag);
702
6467
            put_bits(&s->pb, 4, chan_el_counter[tag]++);
703
18206
            for (ch = 0; ch < chans; ch++) {
704
11739
                sce = &cpe->ch[ch];
705
11739
                coeffs[ch] = sce->coeffs;
706
11739
                sce->ics.predictor_present = 0;
707
11739
                sce->ics.ltp.present = 0;
708
11739
                memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
709
11739
                memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
710
11739
                memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
711
1514331
                for (w = 0; w < 128; w++)
712
1502592
                    if (sce->band_type[w] > RESERVED_BT)
713
25410
                        sce->band_type[w] = 0;
714
            }
715
6467
            s->psy.bitres.alloc = -1;
716
6467
            s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
717
6467
            s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
718
6467
            if (s->psy.bitres.alloc > 0) {
719
                /* Lambda unused here on purpose, we need to take psy's unscaled allocation */
720
12934
                target_bits += s->psy.bitres.alloc
721
6467
                    * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
722
6467
                s->psy.bitres.alloc /= chans;
723
            }
724
6467
            s->cur_type = tag;
725
18206
            for (ch = 0; ch < chans; ch++) {
726
11739
                s->cur_channel = start_ch + ch;
727

11739
                if (s->options.pns && s->coder->mark_pns)
728
3515
                    s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
729
11739
                s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
730
            }
731
6467
            if (chans > 1
732
5272
                && wi[0].window_type[0] == wi[1].window_type[0]
733
5036
                && wi[0].window_shape   == wi[1].window_shape) {
734
735
5036
                cpe->common_window = 1;
736
10631
                for (w = 0; w < wi[0].num_windows; w++) {
737
5603
                    if (wi[0].grouping[w] != wi[1].grouping[w]) {
738
8
                        cpe->common_window = 0;
739
8
                        break;
740
                    }
741
                }
742
            }
743
18206
            for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
744
11739
                sce = &cpe->ch[ch];
745
11739
                s->cur_channel = start_ch + ch;
746

11739
                if (s->options.tns && s->coder->search_for_tns)
747
3515
                    s->coder->search_for_tns(s, sce);
748

11739
                if (s->options.tns && s->coder->apply_tns_filt)
749
3515
                    s->coder->apply_tns_filt(s, sce);
750
11739
                if (sce->tns.present)
751
46
                    tns_mode = 1;
752

11739
                if (s->options.pns && s->coder->search_for_pns)
753
3515
                    s->coder->search_for_pns(s, avctx, sce);
754
            }
755
6467
            s->cur_channel = start_ch;
756
6467
            if (s->options.intensity_stereo) { /* Intensity Stereo */
757
2355
                if (s->coder->search_for_is)
758
2355
                    s->coder->search_for_is(s, avctx, cpe);
759
2355
                if (cpe->is_mode) is_mode = 1;
760
2355
                apply_intensity_stereo(cpe);
761
            }
762
6467
            if (s->options.pred) { /* Prediction */
763
1248
                for (ch = 0; ch < chans; ch++) {
764
832
                    sce = &cpe->ch[ch];
765
832
                    s->cur_channel = start_ch + ch;
766

832
                    if (s->options.pred && s->coder->search_for_pred)
767
832
                        s->coder->search_for_pred(s, sce);
768
832
                    if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
769
                }
770
416
                if (s->coder->adjust_common_pred)
771
416
                    s->coder->adjust_common_pred(s, cpe);
772
1248
                for (ch = 0; ch < chans; ch++) {
773
832
                    sce = &cpe->ch[ch];
774
832
                    s->cur_channel = start_ch + ch;
775

832
                    if (s->options.pred && s->coder->apply_main_pred)
776
832
                        s->coder->apply_main_pred(s, sce);
777
                }
778
416
                s->cur_channel = start_ch;
779
            }
780
6467
            if (s->options.mid_side) { /* Mid/Side stereo */
781

2069
                if (s->options.mid_side == -1 && s->coder->search_for_ms)
782
1651
                    s->coder->search_for_ms(s, cpe);
783
418
                else if (cpe->common_window)
784
392
                    memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
785
2069
                apply_mid_side_stereo(cpe);
786
            }
787
6467
            adjust_frame_information(cpe, chans);
788
6467
            if (s->options.ltp) { /* LTP */
789
                for (ch = 0; ch < chans; ch++) {
790
                    sce = &cpe->ch[ch];
791
                    s->cur_channel = start_ch + ch;
792
                    if (s->coder->search_for_ltp)
793
                        s->coder->search_for_ltp(s, sce, cpe->common_window);
794
                    if (sce->ics.ltp.present) pred_mode = 1;
795
                }
796
                s->cur_channel = start_ch;
797
                if (s->coder->adjust_common_ltp)
798
                    s->coder->adjust_common_ltp(s, cpe);
799
            }
800
6467
            if (chans == 2) {
801
5272
                put_bits(&s->pb, 1, cpe->common_window);
802
5272
                if (cpe->common_window) {
803
5028
                    put_ics_info(s, &cpe->ch[0].ics);
804
5028
                    if (s->coder->encode_main_pred)
805
5028
                        s->coder->encode_main_pred(s, &cpe->ch[0]);
806
5028
                    if (s->coder->encode_ltp_info)
807
5028
                        s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
808
5028
                    encode_ms_info(&s->pb, cpe);
809
5028
                    if (cpe->ms_mode) ms_mode = 1;
810
                }
811
            }
812
18206
            for (ch = 0; ch < chans; ch++) {
813
11739
                s->cur_channel = start_ch + ch;
814
11739
                encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
815
            }
816
6467
            start_ch += chans;
817
        }
818
819
6251
        if (avctx->flags & AV_CODEC_FLAG_QSCALE) {
820
            /* When using a constant Q-scale, don't mess with lambda */
821
            break;
822
        }
823
824
        /* rate control stuff
825
         * allow between the nominal bitrate, and what psy's bit reservoir says to target
826
         * but drift towards the nominal bitrate always
827
         */
828
6251
        frame_bits = put_bits_count(&s->pb);
829
6251
        rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
830
6251
        rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
831
6251
        too_many_bits = FFMAX(target_bits, rate_bits);
832
6251
        too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
833
6251
        too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
834
835
        /* When using ABR, be strict (but only for increasing) */
836
6251
        too_few_bits = too_few_bits - too_few_bits/8;
837
6251
        too_many_bits = too_many_bits + too_many_bits/2;
838
839
6251
        if (   its == 0 /* for steady-state Q-scale tracking */
840

2446
            || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
841
495
            || frame_bits >= 6144 * s->channels - 3  )
842
        {
843
5756
            float ratio = ((float)rate_bits) / frame_bits;
844
845

5756
            if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
846
                /*
847
                 * This path is for steady-state Q-scale tracking
848
                 * When frame bits fall within the stable range, we still need to adjust
849
                 * lambda to maintain it like so in a stable fashion (large jumps in lambda
850
                 * create artifacts and should be avoided), but slowly
851
                 */
852
2861
                ratio = sqrtf(sqrtf(ratio));
853
2861
                ratio = av_clipf(ratio, 0.9f, 1.1f);
854
            } else {
855
                /* Not so fast though */
856
2895
                ratio = sqrtf(ratio);
857
            }
858
5756
            s->lambda = FFMIN(s->lambda * ratio, 65536.f);
859
860
            /* Keep iterating if we must reduce and lambda is in the sky */
861

5756
            if (ratio > 0.9f && ratio < 1.1f) {
862
                break;
863
            } else {
864


2446
                if (is_mode || ms_mode || tns_mode || pred_mode) {
865
1025
                    for (i = 0; i < s->chan_map[0]; i++) {
866
                        // Must restore coeffs
867
520
                        chans = tag == TYPE_CPE ? 2 : 1;
868
520
                        cpe = &s->cpe[i];
869
1540
                        for (ch = 0; ch < chans; ch++)
870
1020
                            memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
871
                    }
872
                }
873
2446
                its++;
874
            }
875
        } else {
876
            break;
877
        }
878
    } while (1);
879
880

3805
    if (s->options.ltp && s->coder->ltp_insert_new_frame)
881
        s->coder->ltp_insert_new_frame(s);
882
883
3805
    put_bits(&s->pb, 3, TYPE_END);
884
3805
    flush_put_bits(&s->pb);
885
886
3805
    s->last_frame_pb_count = put_bits_count(&s->pb);
887
888
3805
    s->lambda_sum += s->lambda;
889
3805
    s->lambda_count++;
890
891
3805
    ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
892
                       &avpkt->duration);
893
894
3805
    avpkt->size = put_bits_count(&s->pb) >> 3;
895
3805
    *got_packet_ptr = 1;
896
3805
    return 0;
897
}
898
899
11
static av_cold int aac_encode_end(AVCodecContext *avctx)
900
{
901
11
    AACEncContext *s = avctx->priv_data;
902
903
11
    av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);
904
905
11
    ff_mdct_end(&s->mdct1024);
906
11
    ff_mdct_end(&s->mdct128);
907
11
    ff_psy_end(&s->psy);
908
11
    ff_lpc_end(&s->lpc);
909
11
    if (s->psypp)
910
11
        ff_psy_preprocess_end(s->psypp);
911
11
    av_freep(&s->buffer.samples);
912
11
    av_freep(&s->cpe);
913
11
    av_freep(&s->fdsp);
914
11
    ff_af_queue_close(&s->afq);
915
11
    return 0;
916
}
917
918
11
static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
919
{
920
11
    int ret = 0;
921
922
11
    s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
923
11
    if (!s->fdsp)
924
        return AVERROR(ENOMEM);
925
926
    // window init
927
11
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
928
11
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
929
11
    ff_init_ff_sine_windows(10);
930
11
    ff_init_ff_sine_windows(7);
931
932
11
    if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
933
        return ret;
934
11
    if ((ret = ff_mdct_init(&s->mdct128,   8, 0, 32768.0)) < 0)
935
        return ret;
936
937
11
    return 0;
938
}
939
940
11
static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
941
{
942
    int ch;
943
11
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
944
11
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
945
946
36
    for(ch = 0; ch < s->channels; ch++)
947
25
        s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
948
949
11
    return 0;
950
alloc_fail:
951
    return AVERROR(ENOMEM);
952
}
953
954
11
static av_cold void aac_encode_init_tables(void)
955
{
956
11
    ff_aac_tableinit();
957
11
}
958
959
11
static av_cold int aac_encode_init(AVCodecContext *avctx)
960
{
961
11
    AACEncContext *s = avctx->priv_data;
962
11
    int i, ret = 0;
963
    const uint8_t *sizes[2];
964
    uint8_t grouping[AAC_MAX_CHANNELS];
965
    int lengths[2];
966
967
    /* Constants */
968
11
    s->last_frame_pb_count = 0;
969
11
    avctx->frame_size = 1024;
970
11
    avctx->initial_padding = 1024;
971
11
    s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
972
973
    /* Channel map and unspecified bitrate guessing */
974
11
    s->channels = avctx->channels;
975
976
11
    s->needs_pce = 1;
977
25
    for (i = 0; i < FF_ARRAY_ELEMS(aac_normal_chan_layouts); i++) {
978
25
        if (avctx->channel_layout == aac_normal_chan_layouts[i]) {
979
11
            s->needs_pce = s->options.pce;
980
11
            break;
981
        }
982
    }
983
984
11
    if (s->needs_pce) {
985
        char buf[64];
986
        for (i = 0; i < FF_ARRAY_ELEMS(aac_pce_configs); i++)
987
            if (avctx->channel_layout == aac_pce_configs[i].layout)
988
                break;
989
        av_get_channel_layout_string(buf, sizeof(buf), -1, avctx->channel_layout);
990
        ERROR_IF(i == FF_ARRAY_ELEMS(aac_pce_configs), "Unsupported channel layout \"%s\"\n", buf);
991
        av_log(avctx, AV_LOG_INFO, "Using a PCE to encode channel layout \"%s\"\n", buf);
992
        s->pce = aac_pce_configs[i];
993
        s->reorder_map = s->pce.reorder_map;
994
        s->chan_map = s->pce.config_map;
995
    } else {
996
11
        s->reorder_map = aac_chan_maps[s->channels - 1];
997
11
        s->chan_map = aac_chan_configs[s->channels - 1];
998
    }
999
1000
11
    if (!avctx->bit_rate) {
1001
9
        for (i = 1; i <= s->chan_map[0]; i++) {
1002
9
            avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */
1003
3
                               s->chan_map[i] == TYPE_LFE ? 16000  : /* LFE  */
1004
                                                            69000  ; /* SCE  */
1005
        }
1006
    }
1007
1008
    /* Samplerate */
1009
54
    for (i = 0; i < 16; i++)
1010
54
        if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
1011
11
            break;
1012
11
    s->samplerate_index = i;
1013

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

11
    ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1073
             "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");
1074
1075
    /* M/S introduces horrible artifacts with multichannel files, this is temporary */
1076
11
    if (s->channels > 3)
1077
1
        s->options.mid_side = 0;
1078
1079
11
    if ((ret = dsp_init(avctx, s)) < 0)
1080
        goto fail;
1081
1082
11
    if ((ret = alloc_buffers(avctx, s)) < 0)
1083
        goto fail;
1084
1085
11
    if ((ret = put_audio_specific_config(avctx)))
1086
        goto fail;
1087
1088
11
    sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];
1089
11
    sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];
1090
11
    lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];
1091
11
    lengths[1] = ff_aac_num_swb_128[s->samplerate_index];
1092
25
    for (i = 0; i < s->chan_map[0]; i++)
1093
14
        grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
1094
11
    if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
1095
11
                           s->chan_map[0], grouping)) < 0)
1096
        goto fail;
1097
11
    s->psypp = ff_psy_preprocess_init(avctx);
1098
11
    ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
1099
11
    s->random_state = 0x1f2e3d4c;
1100
1101
11
    s->abs_pow34   = abs_pow34_v;
1102
11
    s->quant_bands = quantize_bands;
1103
1104
    if (ARCH_X86)
1105
11
        ff_aac_dsp_init_x86(s);
1106
1107
    if (HAVE_MIPSDSP)
1108
        ff_aac_coder_init_mips(s);
1109
1110
11
    if ((ret = ff_thread_once(&aac_table_init, &aac_encode_init_tables)) != 0)
1111
        return AVERROR_UNKNOWN;
1112
1113
11
    ff_af_queue_init(avctx, &s->afq);
1114
1115
11
    return 0;
1116
fail:
1117
    aac_encode_end(avctx);
1118
    return ret;
1119
}
1120
1121
#define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1122
static const AVOption aacenc_options[] = {
1123
    {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_FAST}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},
1124
        {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1125
        {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1126
        {"fast",     "Default fast search",       0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
1127
    {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},
1128
    {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1129
    {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1130
    {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1131
    {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1132
    {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1133
    {"aac_pce", "Forces the use of PCEs", offsetof(AACEncContext, options.pce), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1134
    {NULL}
1135
};
1136
1137
static const AVClass aacenc_class = {
1138
    .class_name = "AAC encoder",
1139
    .item_name  = av_default_item_name,
1140
    .option     = aacenc_options,
1141
    .version    = LIBAVUTIL_VERSION_INT,
1142
};
1143
1144
static const AVCodecDefault aac_encode_defaults[] = {
1145
    { "b", "0" },
1146
    { NULL }
1147
};
1148
1149
AVCodec ff_aac_encoder = {
1150
    .name           = "aac",
1151
    .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
1152
    .type           = AVMEDIA_TYPE_AUDIO,
1153
    .id             = AV_CODEC_ID_AAC,
1154
    .priv_data_size = sizeof(AACEncContext),
1155
    .init           = aac_encode_init,
1156
    .encode2        = aac_encode_frame,
1157
    .close          = aac_encode_end,
1158
    .defaults       = aac_encode_defaults,
1159
    .supported_samplerates = mpeg4audio_sample_rates,
1160
    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,
1161
    .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,
1162
    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
1163
                                                     AV_SAMPLE_FMT_NONE },
1164
    .priv_class     = &aacenc_class,
1165
};