aacenc.c 43.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
/*
 * AAC encoder
 * Copyright (C) 2008 Konstantin Shishkov
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
23
 * @file
24 25 26 27 28
 * AAC encoder
 */

/***********************************
 *              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 44

#include "aac.h"
#include "aactab.h"
45
#include "aacenc.h"
46
#include "aacenctab.h"
47
#include "aacenc_utils.h"
48 49

#include "psymodel.h"
50

51 52
static AVOnce aac_table_init = AV_ONCE_INIT;

53 54 55 56 57
static void put_pce(PutBitContext *pb, AVCodecContext *avctx)
{
    int i, j;
    AACEncContext *s = avctx->priv_data;
    AACPCEInfo *pce = &s->pce;
58 59
    const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
    const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85

    put_bits(pb, 4, 0);

    put_bits(pb, 2, avctx->profile);
    put_bits(pb, 4, s->samplerate_index);

    put_bits(pb, 4, pce->num_ele[0]); /* Front */
    put_bits(pb, 4, pce->num_ele[1]); /* Side */
    put_bits(pb, 4, pce->num_ele[2]); /* Back */
    put_bits(pb, 2, pce->num_ele[3]); /* LFE */
    put_bits(pb, 3, 0); /* Assoc data */
    put_bits(pb, 4, 0); /* CCs */

    put_bits(pb, 1, 0); /* Stereo mixdown */
    put_bits(pb, 1, 0); /* Mono mixdown */
    put_bits(pb, 1, 0); /* Something else */

    for (i = 0; i < 4; i++) {
        for (j = 0; j < pce->num_ele[i]; j++) {
            if (i < 3)
                put_bits(pb, 1, pce->pairing[i][j]);
            put_bits(pb, 4, pce->index[i][j]);
        }
    }

    avpriv_align_put_bits(pb);
86 87
    put_bits(pb, 8, strlen(aux_data));
    avpriv_put_string(pb, aux_data, 0);
88 89
}

90 91 92 93
/**
 * Make AAC audio config object.
 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
 */
94
static int put_audio_specific_config(AVCodecContext *avctx)
95 96 97
{
    PutBitContext pb;
    AACEncContext *s = avctx->priv_data;
98
    int channels = (!s->needs_pce)*(s->channels - (s->channels == 8 ? 1 : 0));
99
    const int max_size = 32;
100

101 102 103 104 105
    avctx->extradata = av_mallocz(max_size);
    if (!avctx->extradata)
        return AVERROR(ENOMEM);

    init_put_bits(&pb, avctx->extradata, max_size);
106
    put_bits(&pb, 5, s->profile+1); //profile
107
    put_bits(&pb, 4, s->samplerate_index); //sample rate index
108
    put_bits(&pb, 4, channels);
109 110 111 112
    //GASpecificConfig
    put_bits(&pb, 1, 0); //frame length - 1024 samples
    put_bits(&pb, 1, 0); //does not depend on core coder
    put_bits(&pb, 1, 0); //is not extension
113 114
    if (s->needs_pce)
        put_pce(&pb, avctx);
Alex Converse's avatar
Alex Converse committed
115 116

    //Explicitly Mark SBR absent
117
    put_bits(&pb, 11, 0x2b7); //sync extension
Alex Converse's avatar
Alex Converse committed
118 119
    put_bits(&pb, 5,  AOT_SBR);
    put_bits(&pb, 1,  0);
120
    flush_put_bits(&pb);
121 122 123
    avctx->extradata_size = put_bits_count(&pb) >> 3;

    return 0;
124 125
}

126 127
void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
{
128 129 130 131
    ++s->quantize_band_cost_cache_generation;
    if (s->quantize_band_cost_cache_generation == 0) {
        memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
        s->quantize_band_cost_cache_generation = 1;
132 133 134
    }
}

135
#define WINDOW_FUNC(type) \
136
static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
137 138
                                    SingleChannelElement *sce, \
                                    const float *audio)
139

140 141 142 143
WINDOW_FUNC(only_long)
{
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
144
    float *out = sce->ret_buf;
145

146 147
    fdsp->vector_fmul        (out,        audio,        lwindow, 1024);
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
148
}
149

150 151 152 153
WINDOW_FUNC(long_start)
{
    const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
154
    float *out = sce->ret_buf;
155

156
    fdsp->vector_fmul(out, audio, lwindow, 1024);
157
    memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
158
    fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
159 160
    memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
}
161

162 163 164 165
WINDOW_FUNC(long_stop)
{
    const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
    const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
166
    float *out = sce->ret_buf;
167 168

    memset(out, 0, sizeof(out[0]) * 448);
169
    fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
170
    memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
171
    fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
172
}
173

174 175 176 177 178
WINDOW_FUNC(eight_short)
{
    const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
    const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
    const float *in = audio + 448;
179
    float *out = sce->ret_buf;
180
    int w;
181

182
    for (w = 0; w < 8; w++) {
183
        fdsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
184 185
        out += 128;
        in  += 128;
186
        fdsp->vector_fmul_reverse(out, in, swindow, 128);
187 188
        out += 128;
    }
189 190
}

191
static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
192 193
                                     SingleChannelElement *sce,
                                     const float *audio) = {
194 195 196 197 198 199
    [ONLY_LONG_SEQUENCE]   = apply_only_long_window,
    [LONG_START_SEQUENCE]  = apply_long_start_window,
    [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
    [LONG_STOP_SEQUENCE]   = apply_long_stop_window
};

200 201
static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
                                  float *audio)
202
{
203
    int i;
204
    const float *output = sce->ret_buf;
205

206
    apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
207 208

    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
209
        s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
210 211
    else
        for (i = 0; i < 1024; i += 128)
212
            s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
213
    memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
214
    memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
215 216
}

217 218 219 220
/**
 * Encode ics_info element.
 * @see Table 4.6 (syntax of ics_info)
 */
221
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
222
{
223
    int w;
224 225 226 227

    put_bits(&s->pb, 1, 0);                // ics_reserved bit
    put_bits(&s->pb, 2, info->window_sequence[0]);
    put_bits(&s->pb, 1, info->use_kb_window[0]);
228
    if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
229
        put_bits(&s->pb, 6, info->max_sfb);
230
        put_bits(&s->pb, 1, !!info->predictor_present);
231
    } else {
232
        put_bits(&s->pb, 4, info->max_sfb);
233
        for (w = 1; w < 8; w++)
234
            put_bits(&s->pb, 1, !info->group_len[w]);
235 236 237
    }
}

238
/**
239 240
 * Encode MS data.
 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
241
 */
242
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
243 244
{
    int i, w;
245 246

    put_bits(pb, 2, cpe->ms_mode);
247 248
    if (cpe->ms_mode == 1)
        for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
249
            for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
250 251 252 253 254 255
                put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
}

/**
 * Produce integer coefficients from scalefactors provided by the model.
 */
256
static void adjust_frame_information(ChannelElement *cpe, int chans)
257 258
{
    int i, w, w2, g, ch;
259 260 261 262 263 264 265 266
    int maxsfb, cmaxsfb;

    for (ch = 0; ch < chans; ch++) {
        IndividualChannelStream *ics = &cpe->ch[ch].ics;
        maxsfb = 0;
        cpe->ch[ch].pulse.num_pulse = 0;
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
267 268 269
                for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
                    ;
                maxsfb = FFMAX(maxsfb, cmaxsfb);
270 271 272 273 274
            }
        }
        ics->max_sfb = maxsfb;

        //adjust zero bands for window groups
275 276
        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
            for (g = 0; g < ics->max_sfb; g++) {
277
                i = 1;
278 279
                for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
                    if (!cpe->ch[ch].zeroes[w2*16 + g]) {
280 281 282 283 284 285 286 287 288
                        i = 0;
                        break;
                    }
                }
                cpe->ch[ch].zeroes[w*16 + g] = i;
            }
        }
    }

289
    if (chans > 1 && cpe->common_window) {
290 291 292 293 294
        IndividualChannelStream *ics0 = &cpe->ch[0].ics;
        IndividualChannelStream *ics1 = &cpe->ch[1].ics;
        int msc = 0;
        ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
        ics1->max_sfb = ics0->max_sfb;
295 296
        for (w = 0; w < ics0->num_windows*16; w += 16)
            for (i = 0; i < ics0->max_sfb; i++)
297 298
                if (cpe->ms_mask[w+i])
                    msc++;
299 300 301
        if (msc == 0 || ics0->max_sfb == 0)
            cpe->ms_mode = 0;
        else
302
            cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
303 304 305
    }
}

306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
static void apply_intensity_stereo(ChannelElement *cpe)
{
    int w, w2, g, i;
    IndividualChannelStream *ics = &cpe->ch[0].ics;
    if (!cpe->common_window)
        return;
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
            int start = (w+w2) * 128;
            for (g = 0; g < ics->num_swb; g++) {
                int p  = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
                float scale = cpe->ch[0].is_ener[w*16+g];
                if (!cpe->is_mask[w*16 + g]) {
                    start += ics->swb_sizes[g];
                    continue;
                }
322 323
                if (cpe->ms_mask[w*16 + g])
                    p *= -1;
324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
                for (i = 0; i < ics->swb_sizes[g]; i++) {
                    float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
                    cpe->ch[0].coeffs[start+i] = sum;
                    cpe->ch[1].coeffs[start+i] = 0.0f;
                }
                start += ics->swb_sizes[g];
            }
        }
    }
}

static void apply_mid_side_stereo(ChannelElement *cpe)
{
    int w, w2, g, i;
    IndividualChannelStream *ics = &cpe->ch[0].ics;
    if (!cpe->common_window)
        return;
    for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
        for (w2 =  0; w2 < ics->group_len[w]; w2++) {
            int start = (w+w2) * 128;
            for (g = 0; g < ics->num_swb; g++) {
345 346 347 348 349
                /* ms_mask can be used for other purposes in PNS and I/S,
                 * so must not apply M/S if any band uses either, even if
                 * ms_mask is set.
                 */
                if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]
350 351
                    || cpe->ch[0].band_type[w*16 + g] >= NOISE_BT
                    || cpe->ch[1].band_type[w*16 + g] >= NOISE_BT) {
352 353 354 355 356 357 358 359 360 361 362 363 364 365 366
                    start += ics->swb_sizes[g];
                    continue;
                }
                for (i = 0; i < ics->swb_sizes[g]; i++) {
                    float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
                    float R = L - cpe->ch[1].coeffs[start+i];
                    cpe->ch[0].coeffs[start+i] = L;
                    cpe->ch[1].coeffs[start+i] = R;
                }
                start += ics->swb_sizes[g];
            }
        }
    }
}

367 368 369 370 371 372 373
/**
 * Encode scalefactor band coding type.
 */
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
{
    int w;

374 375 376
    if (s->coder->set_special_band_scalefactors)
        s->coder->set_special_band_scalefactors(s, sce);

377
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
378 379 380 381 382 383
        s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
}

/**
 * Encode scalefactors.
 */
384 385
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
                                 SingleChannelElement *sce)
386
{
387
    int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
388
    int off_is = 0, noise_flag = 1;
389 390
    int i, w;

391 392 393
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
        for (i = 0; i < sce->ics.max_sfb; i++) {
            if (!sce->zeroes[w*16 + i]) {
394 395 396 397 398 399 400
                if (sce->band_type[w*16 + i] == NOISE_BT) {
                    diff = sce->sf_idx[w*16 + i] - off_pns;
                    off_pns = sce->sf_idx[w*16 + i];
                    if (noise_flag-- > 0) {
                        put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
                        continue;
                    }
401 402 403 404
                } else if (sce->band_type[w*16 + i] == INTENSITY_BT  ||
                           sce->band_type[w*16 + i] == INTENSITY_BT2) {
                    diff = sce->sf_idx[w*16 + i] - off_is;
                    off_is = sce->sf_idx[w*16 + i];
405 406 407 408 409
                } else {
                    diff = sce->sf_idx[w*16 + i] - off_sf;
                    off_sf = sce->sf_idx[w*16 + i];
                }
                diff += SCALE_DIFF_ZERO;
410
                av_assert0(diff >= 0 && diff <= 120);
411 412
                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
            }
413 414 415 416
        }
    }
}

417 418 419
/**
 * Encode pulse data.
 */
420
static void encode_pulses(AACEncContext *s, Pulse *pulse)
421 422 423 424
{
    int i;

    put_bits(&s->pb, 1, !!pulse->num_pulse);
425 426
    if (!pulse->num_pulse)
        return;
427 428 429

    put_bits(&s->pb, 2, pulse->num_pulse - 1);
    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]);
432 433 434 435 436 437 438
        put_bits(&s->pb, 4, pulse->amp[i]);
    }
}

/**
 * Encode spectral coefficients processed by psychoacoustic model.
 */
439
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
440
{
441
    int start, i, w, w2;
442

443
    for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
444
        start = 0;
445 446
        for (i = 0; i < sce->ics.max_sfb; i++) {
            if (sce->zeroes[w*16 + i]) {
447
                start += sce->ics.swb_sizes[i];
448 449
                continue;
            }
450
            for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
451 452
                s->coder->quantize_and_encode_band(s, &s->pb,
                                                   &sce->coeffs[start + w2*128],
453
                                                   NULL, sce->ics.swb_sizes[i],
454 455
                                                   sce->sf_idx[w*16 + i],
                                                   sce->band_type[w*16 + i],
456 457
                                                   s->lambda,
                                                   sce->ics.window_clipping[w]);
458
            }
459
            start += sce->ics.swb_sizes[i];
460 461 462 463
        }
    }
}

464 465 466 467 468 469 470 471 472 473 474
/**
 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
 */
static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
{
    int start, i, j, w;

    if (sce->ics.clip_avoidance_factor < 1.0f) {
        for (w = 0; w < sce->ics.num_windows; w++) {
            start = 0;
            for (i = 0; i < sce->ics.max_sfb; i++) {
475
                float *swb_coeffs = &sce->coeffs[start + w*128];
476 477 478 479 480 481 482 483
                for (j = 0; j < sce->ics.swb_sizes[i]; j++)
                    swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
                start += sce->ics.swb_sizes[i];
            }
        }
    }
}

484 485 486
/**
 * Encode one channel of audio data.
 */
487 488 489
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
                                     SingleChannelElement *sce,
                                     int common_window)
490 491
{
    put_bits(&s->pb, 8, sce->sf_idx[0]);
492
    if (!common_window) {
493
        put_ics_info(s, &sce->ics);
494 495
        if (s->coder->encode_main_pred)
            s->coder->encode_main_pred(s, sce);
496 497
        if (s->coder->encode_ltp_info)
            s->coder->encode_ltp_info(s, sce, 0);
498
    }
499 500 501
    encode_band_info(s, sce);
    encode_scale_factors(avctx, s, sce);
    encode_pulses(s, &sce->pulse);
502
    put_bits(&s->pb, 1, !!sce->tns.present);
503 504
    if (s->coder->encode_tns_info)
        s->coder->encode_tns_info(s, sce);
505 506 507 508 509
    put_bits(&s->pb, 1, 0); //ssr
    encode_spectral_coeffs(s, sce);
    return 0;
}

510 511 512
/**
 * Write some auxiliary information about the created AAC file.
 */
513
static void put_bitstream_info(AACEncContext *s, const char *name)
514 515 516 517
{
    int i, namelen, padbits;

    namelen = strlen(name) + 2;
518
    put_bits(&s->pb, 3, TYPE_FIL);
519
    put_bits(&s->pb, 4, FFMIN(namelen, 15));
520
    if (namelen >= 15)
521
        put_bits(&s->pb, 8, namelen - 14);
522
    put_bits(&s->pb, 4, 0); //extension type - filler
523
    padbits = -put_bits_count(&s->pb) & 7;
524
    avpriv_align_put_bits(&s->pb);
525
    for (i = 0; i < namelen - 2; i++)
526 527 528 529
        put_bits(&s->pb, 8, name[i]);
    put_bits(&s->pb, 12 - padbits, 0);
}

530
/*
531
 * Copy input samples.
532
 * Channels are reordered from libavcodec's default order to AAC order.
533
 */
534
static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
535
{
536 537
    int ch;
    int end = 2048 + (frame ? frame->nb_samples : 0);
538
    const uint8_t *channel_map = s->reorder_map;
539

540 541
    /* copy and remap input samples */
    for (ch = 0; ch < s->channels; ch++) {
542
        /* copy last 1024 samples of previous frame to the start of the current frame */
543
        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
        if (frame) {
547 548 549
            memcpy(&s->planar_samples[ch][2048],
                   frame->extended_data[channel_map[ch]],
                   frame->nb_samples * sizeof(s->planar_samples[0][0]));
550
        }
551 552
        memset(&s->planar_samples[ch][end], 0,
               (3072 - end) * sizeof(s->planar_samples[0][0]));
553 554 555
    }
}

556 557
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
                            const AVFrame *frame, int *got_packet_ptr)
558 559
{
    AACEncContext *s = avctx->priv_data;
560
    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
    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

570 571
    /* add current frame to queue */
    if (frame) {
572
        if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
573
            return ret;
574 575 576
    } else {
        if (!s->afq.remaining_samples || (!s->afq.frame_alloc && !s->afq.frame_count))
            return 0;
577 578
    }

579
    copy_input_samples(s, frame);
580 581
    if (s->psypp)
        ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
582 583

    if (!avctx->frame_number)
584 585 586
        return 0;

    start_ch = 0;
587
    for (i = 0; i < s->chan_map[0]; i++) {
588
        FFPsyWindowInfo* wi = windows + start_ch;
589
        tag      = s->chan_map[i+1];
590 591
        chans    = tag == TYPE_CPE ? 2 : 1;
        cpe      = &s->cpe[i];
592
        for (ch = 0; ch < chans; ch++) {
593
            int k;
594
            float clip_avoidance_factor;
595 596 597 598
            sce = &cpe->ch[ch];
            ics = &sce->ics;
            s->cur_channel = start_ch + ch;
            overlap  = &samples[s->cur_channel][0];
599
            samples2 = overlap + 1024;
600
            la       = samples2 + (448+64);
601
            if (!frame)
602
                la = NULL;
603
            if (tag == TYPE_LFE) {
604
                wi[ch].window_type[0] = wi[ch].window_type[1] = ONLY_LONG_SEQUENCE;
605 606 607
                wi[ch].window_shape   = 0;
                wi[ch].num_windows    = 1;
                wi[ch].grouping[0]    = 1;
608
                wi[ch].clipping[0]    = 0;
609 610 611 612 613 614

                /* Only the lowest 12 coefficients are used in a LFE channel.
                 * The expression below results in only the bottom 8 coefficients
                 * being used for 11.025kHz to 16kHz sample rates.
                 */
                ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
615
            } else {
616
                wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
Alex Converse's avatar
Alex Converse committed
617
                                              ics->window_sequence[0]);
618
            }
619
            ics->window_sequence[1] = ics->window_sequence[0];
620
            ics->window_sequence[0] = wi[ch].window_type[0];
621
            ics->use_kb_window[1]   = ics->use_kb_window[0];
622 623
            ics->use_kb_window[0]   = wi[ch].window_shape;
            ics->num_windows        = wi[ch].num_windows;
624
            ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];
625
            ics->num_swb            = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
626
            ics->max_sfb            = FFMIN(ics->max_sfb, ics->num_swb);
627 628 629
            ics->swb_offset         = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
                                        ff_swb_offset_128 [s->samplerate_index]:
                                        ff_swb_offset_1024[s->samplerate_index];
630 631 632
            ics->tns_max_bands      = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
                                        ff_tns_max_bands_128 [s->samplerate_index]:
                                        ff_tns_max_bands_1024[s->samplerate_index];
633

634 635
            for (w = 0; w < ics->num_windows; w++)
                ics->group_len[w] = wi[ch].grouping[w];
636 637 638 639 640 641 642 643 644 645 646 647 648

            /* Calculate input sample maximums and evaluate clipping risk */
            clip_avoidance_factor = 0.0f;
            for (w = 0; w < ics->num_windows; w++) {
                const float *wbuf = overlap + w * 128;
                const int wlen = 2048 / ics->num_windows;
                float max = 0;
                int j;
                /* mdct input is 2 * output */
                for (j = 0; j < wlen; j++)
                    max = FFMAX(max, fabsf(wbuf[j]));
                wi[ch].clipping[w] = max;
            }
649 650 651 652 653 654 655 656 657 658 659 660 661
            for (w = 0; w < ics->num_windows; w++) {
                if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
                    ics->window_clipping[w] = 1;
                    clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
                } else {
                    ics->window_clipping[w] = 0;
                }
            }
            if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
                ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
            } else {
                ics->clip_avoidance_factor = 1.0f;
            }
662

663
            apply_window_and_mdct(s, sce, overlap);
664 665 666 667 668 669 670

            if (s->options.ltp && s->coder->update_ltp) {
                s->coder->update_ltp(s, sce);
                apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
                s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);
            }

671
            for (k = 0; k < 1024; k++) {
672 673
                if (!(fabs(cpe->ch[ch].coeffs[k]) < 1E16)) { // Ensure headroom for energy calculation
                    av_log(avctx, AV_LOG_ERROR, "Input contains (near) NaN/+-Inf\n");
674 675
                    return AVERROR(EINVAL);
                }
676
            }
677
            avoid_clipping(s, sce);
678 679 680
        }
        start_ch += chans;
    }
681
    if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
682
        return ret;
683
    frame_bits = its = 0;
684
    do {
685 686
        init_put_bits(&s->pb, avpkt->data, avpkt->size);

687
        if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
688
            put_bitstream_info(s, LIBAVCODEC_IDENT);
689
        start_ch = 0;
690
        target_bits = 0;
691
        memset(chan_el_counter, 0, sizeof(chan_el_counter));
692
        for (i = 0; i < s->chan_map[0]; i++) {
693
            FFPsyWindowInfo* wi = windows + start_ch;
694
            const float *coeffs[2];
695
            tag      = s->chan_map[i+1];
696 697
            chans    = tag == TYPE_CPE ? 2 : 1;
            cpe      = &s->cpe[i];
698
            cpe->common_window = 0;
699 700
            memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
            memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
701 702
            put_bits(&s->pb, 3, tag);
            put_bits(&s->pb, 4, chan_el_counter[tag]++);
703 704 705
            for (ch = 0; ch < chans; ch++) {
                sce = &cpe->ch[ch];
                coeffs[ch] = sce->coeffs;
706
                sce->ics.predictor_present = 0;
707 708
                sce->ics.ltp.present = 0;
                memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
709
                memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
710
                memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
711 712 713 714
                for (w = 0; w < 128; w++)
                    if (sce->band_type[w] > RESERVED_BT)
                        sce->band_type[w] = 0;
            }
715
            s->psy.bitres.alloc = -1;
716
            s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
717
            s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
718 719
            if (s->psy.bitres.alloc > 0) {
                /* Lambda unused here on purpose, we need to take psy's unscaled allocation */
720 721
                target_bits += s->psy.bitres.alloc
                    * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
722 723 724
                s->psy.bitres.alloc /= chans;
            }
            s->cur_type = tag;
725
            for (ch = 0; ch < chans; ch++) {
726
                s->cur_channel = start_ch + ch;
727 728
                if (s->options.pns && s->coder->mark_pns)
                    s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
729
                s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
730 731 732 733 734 735
            }
            if (chans > 1
                && wi[0].window_type[0] == wi[1].window_type[0]
                && wi[0].window_shape   == wi[1].window_shape) {

                cpe->common_window = 1;
736 737
                for (w = 0; w < wi[0].num_windows; w++) {
                    if (wi[0].grouping[w] != wi[1].grouping[w]) {
738 739 740
                        cpe->common_window = 0;
                        break;
                    }
741 742
                }
            }
743
            for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
744 745 746 747
                sce = &cpe->ch[ch];
                s->cur_channel = start_ch + ch;
                if (s->options.tns && s->coder->search_for_tns)
                    s->coder->search_for_tns(s, sce);
748
                if (s->options.tns && s->coder->apply_tns_filt)
749
                    s->coder->apply_tns_filt(s, sce);
750 751
                if (sce->tns.present)
                    tns_mode = 1;
752 753
                if (s->options.pns && s->coder->search_for_pns)
                    s->coder->search_for_pns(s, avctx, sce);
754
            }
755
            s->cur_channel = start_ch;
756 757 758
            if (s->options.intensity_stereo) { /* Intensity Stereo */
                if (s->coder->search_for_is)
                    s->coder->search_for_is(s, avctx, cpe);
759
                if (cpe->is_mode) is_mode = 1;
760
                apply_intensity_stereo(cpe);
761
            }
762 763 764 765 766 767 768 769
            if (s->options.pred) { /* Prediction */
                for (ch = 0; ch < chans; ch++) {
                    sce = &cpe->ch[ch];
                    s->cur_channel = start_ch + ch;
                    if (s->options.pred && s->coder->search_for_pred)
                        s->coder->search_for_pred(s, sce);
                    if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
                }
770 771
                if (s->coder->adjust_common_pred)
                    s->coder->adjust_common_pred(s, cpe);
772 773 774 775 776 777 778
                for (ch = 0; ch < chans; ch++) {
                    sce = &cpe->ch[ch];
                    s->cur_channel = start_ch + ch;
                    if (s->options.pred && s->coder->apply_main_pred)
                        s->coder->apply_main_pred(s, sce);
                }
                s->cur_channel = start_ch;
779
            }
780 781
            if (s->options.mid_side) { /* Mid/Side stereo */
                if (s->options.mid_side == -1 && s->coder->search_for_ms)
782 783 784 785
                    s->coder->search_for_ms(s, cpe);
                else if (cpe->common_window)
                    memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
                apply_mid_side_stereo(cpe);
786
            }
787
            adjust_frame_information(cpe, chans);
788 789 790 791 792 793 794 795 796 797 798 799
            if (s->options.ltp) { /* LTP */
                for (ch = 0; ch < chans; ch++) {
                    sce = &cpe->ch[ch];
                    s->cur_channel = start_ch + ch;
                    if (s->coder->search_for_ltp)
                        s->coder->search_for_ltp(s, sce, cpe->common_window);
                    if (sce->ics.ltp.present) pred_mode = 1;
                }
                s->cur_channel = start_ch;
                if (s->coder->adjust_common_ltp)
                    s->coder->adjust_common_ltp(s, cpe);
            }
800 801 802 803
            if (chans == 2) {
                put_bits(&s->pb, 1, cpe->common_window);
                if (cpe->common_window) {
                    put_ics_info(s, &cpe->ch[0].ics);
804 805
                    if (s->coder->encode_main_pred)
                        s->coder->encode_main_pred(s, &cpe->ch[0]);
806 807
                    if (s->coder->encode_ltp_info)
                        s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
808
                    encode_ms_info(&s->pb, cpe);
809
                    if (cpe->ms_mode) ms_mode = 1;
810
                }
811
            }
812 813 814
            for (ch = 0; ch < chans; ch++) {
                s->cur_channel = start_ch + ch;
                encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
815 816
            }
            start_ch += chans;
817 818
        }

819
        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
825 826
         * allow between the nominal bitrate, and what psy's bit reservoir says to target
         * but drift towards the nominal bitrate always
827 828
         */
        frame_bits = put_bits_count(&s->pb);
829 830 831 832 833
        rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
        rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
        too_many_bits = FFMAX(target_bits, rate_bits);
        too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
        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) */
836 837
        too_few_bits = too_few_bits - too_few_bits/8;
        too_many_bits = too_many_bits + too_many_bits/2;
838 839 840 841 842

        if (   its == 0 /* for steady-state Q-scale tracking */
            || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
            || frame_bits >= 6144 * s->channels - 3  )
        {
843
            float ratio = ((float)rate_bits) / frame_bits;
844 845 846 847 848 849 850 851 852 853 854 855 856 857 858

            if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
                /*
                 * This path is for steady-state Q-scale tracking
                 * When frame bits fall within the stable range, we still need to adjust
                 * lambda to maintain it like so in a stable fashion (large jumps in lambda
                 * create artifacts and should be avoided), but slowly
                 */
                ratio = sqrtf(sqrtf(ratio));
                ratio = av_clipf(ratio, 0.9f, 1.1f);
            } else {
                /* Not so fast though */
                ratio = sqrtf(ratio);
            }
            s->lambda = FFMIN(s->lambda * ratio, 65536.f);
859

860
            /* Keep iterating if we must reduce and lambda is in the sky */
861
            if (ratio > 0.9f && ratio < 1.1f) {
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877
                break;
            } else {
                if (is_mode || ms_mode || tns_mode || pred_mode) {
                    for (i = 0; i < s->chan_map[0]; i++) {
                        // Must restore coeffs
                        chans = tag == TYPE_CPE ? 2 : 1;
                        cpe = &s->cpe[i];
                        for (ch = 0; ch < chans; ch++)
                            memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
                    }
                }
                its++;
            }
        } else {
            break;
        }
878 879
    } while (1);

880 881 882
    if (s->options.ltp && s->coder->ltp_insert_new_frame)
        s->coder->ltp_insert_new_frame(s);

883 884
    put_bits(&s->pb, 3, TYPE_END);
    flush_put_bits(&s->pb);
885

886
    s->last_frame_pb_count = put_bits_count(&s->pb);
887

888 889
    s->lambda_sum += s->lambda;
    s->lambda_count++;
890

891 892 893 894 895 896
    ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
                       &avpkt->duration);

    avpkt->size = put_bits_count(&s->pb) >> 3;
    *got_packet_ptr = 1;
    return 0;
897 898
}

899 900 901 902
static av_cold int aac_encode_end(AVCodecContext *avctx)
{
    AACEncContext *s = avctx->priv_data;

903 904
    av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);

905 906
    ff_mdct_end(&s->mdct1024);
    ff_mdct_end(&s->mdct128);
907
    ff_psy_end(&s->psy);
908
    ff_lpc_end(&s->lpc);
909 910
    if (s->psypp)
        ff_psy_preprocess_end(s->psypp);
911
    av_freep(&s->buffer.samples);
912
    av_freep(&s->cpe);
913
    av_freep(&s->fdsp);
914
    ff_af_queue_close(&s->afq);
915 916 917
    return 0;
}

918 919 920 921
static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
{
    int ret = 0;

922
    s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
923 924
    if (!s->fdsp)
        return AVERROR(ENOMEM);
925 926 927 928 929 930 931

    // window init
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
    ff_init_ff_sine_windows(10);
    ff_init_ff_sine_windows(7);

932
    if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
933
        return ret;
934
    if ((ret = ff_mdct_init(&s->mdct128,   8, 0, 32768.0)) < 0)
935 936 937 938 939 940 941
        return ret;

    return 0;
}

static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
{
942
    int ch;
943 944
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
945

946
    for(ch = 0; ch < s->channels; ch++)
947
        s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
948

949 950 951 952 953
    return 0;
alloc_fail:
    return AVERROR(ENOMEM);
}

954 955 956 957 958
static av_cold void aac_encode_init_tables(void)
{
    ff_aac_tableinit();
}

959 960 961 962 963 964 965 966
static av_cold int aac_encode_init(AVCodecContext *avctx)
{
    AACEncContext *s = avctx->priv_data;
    int i, ret = 0;
    const uint8_t *sizes[2];
    uint8_t grouping[AAC_MAX_CHANNELS];
    int lengths[2];

967
    /* Constants */
968
    s->last_frame_pb_count = 0;
969
    avctx->frame_size = 1024;
970
    avctx->initial_padding = 1024;
971 972 973 974
    s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;

    /* Channel map and unspecified bitrate guessing */
    s->channels = avctx->channels;
975 976 977 978 979 980 981 982 983 984

    s->needs_pce = 1;
    for (i = 0; i < FF_ARRAY_ELEMS(aac_normal_chan_layouts); i++) {
        if (avctx->channel_layout == aac_normal_chan_layouts[i]) {
            s->needs_pce = s->options.pce;
            break;
        }
    }

    if (s->needs_pce) {
985
        char buf[64];
986 987 988
        for (i = 0; i < FF_ARRAY_ELEMS(aac_pce_configs); i++)
            if (avctx->channel_layout == aac_pce_configs[i].layout)
                break;
989 990 991
        av_get_channel_layout_string(buf, sizeof(buf), -1, avctx->channel_layout);
        ERROR_IF(i == FF_ARRAY_ELEMS(aac_pce_configs), "Unsupported channel layout \"%s\"\n", buf);
        av_log(avctx, AV_LOG_INFO, "Using a PCE to encode channel layout \"%s\"\n", buf);
992 993 994 995 996 997 998 999
        s->pce = aac_pce_configs[i];
        s->reorder_map = s->pce.reorder_map;
        s->chan_map = s->pce.config_map;
    } else {
        s->reorder_map = aac_chan_maps[s->channels - 1];
        s->chan_map = aac_chan_configs[s->channels - 1];
    }

1000 1001 1002 1003 1004 1005 1006
    if (!avctx->bit_rate) {
        for (i = 1; i <= s->chan_map[0]; i++) {
            avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */
                               s->chan_map[i] == TYPE_LFE ? 16000  : /* LFE  */
                                                            69000  ; /* SCE  */
        }
    }
1007

1008
    /* Samplerate */
1009 1010 1011
    for (i = 0; i < 16; i++)
        if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
            break;
1012 1013 1014 1015
    s->samplerate_index = i;
    ERROR_IF(s->samplerate_index == 16 ||
             s->samplerate_index >= ff_aac_swb_size_1024_len ||
             s->samplerate_index >= ff_aac_swb_size_128_len,
1016
             "Unsupported sample rate %d\n", avctx->sample_rate);
1017 1018

    /* Bitrate limiting */
1019
    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 1022
             1024.0 * avctx->bit_rate / avctx->sample_rate,
             6144 * s->channels);
1023 1024
    avctx->bit_rate = (int64_t)FFMIN(6144 * s->channels / 1024.0 * avctx->sample_rate,
                                     avctx->bit_rate);
1025

1026 1027 1028
    /* Profile and option setting */
    avctx->profile = avctx->profile == FF_PROFILE_UNKNOWN ? FF_PROFILE_AAC_LOW :
                     avctx->profile;
1029 1030
    for (i = 0; i < FF_ARRAY_ELEMS(aacenc_profiles); i++)
        if (avctx->profile == aacenc_profiles[i])
1031
            break;
1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058
    if (avctx->profile == FF_PROFILE_MPEG2_AAC_LOW) {
        avctx->profile = FF_PROFILE_AAC_LOW;
        ERROR_IF(s->options.pred,
                 "Main prediction unavailable in the \"mpeg2_aac_low\" profile\n");
        ERROR_IF(s->options.ltp,
                 "LTP prediction unavailable in the \"mpeg2_aac_low\" profile\n");
        WARN_IF(s->options.pns,
                "PNS unavailable in the \"mpeg2_aac_low\" profile, turning off\n");
        s->options.pns = 0;
    } else if (avctx->profile == FF_PROFILE_AAC_LTP) {
        s->options.ltp = 1;
        ERROR_IF(s->options.pred,
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
    } else if (avctx->profile == FF_PROFILE_AAC_MAIN) {
        s->options.pred = 1;
        ERROR_IF(s->options.ltp,
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
    } else if (s->options.ltp) {
        avctx->profile = FF_PROFILE_AAC_LTP;
        WARN_IF(1,
                "Chainging profile to \"aac_ltp\"\n");
        ERROR_IF(s->options.pred,
                 "Main prediction unavailable in the \"aac_ltp\" profile\n");
    } else if (s->options.pred) {
        avctx->profile = FF_PROFILE_AAC_MAIN;
        WARN_IF(1,
                "Chainging profile to \"aac_main\"\n");
1059
        ERROR_IF(s->options.ltp,
1060
                 "LTP prediction unavailable in the \"aac_main\" profile\n");
1061
    }
1062
    s->profile = avctx->profile;
1063

1064 1065
    /* Coder limitations */
    s->coder = &ff_aac_coders[s->options.coder];
1066
    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 1070 1071
        s->options.intensity_stereo = 0;
        s->options.pns = 0;
    }
1072 1073 1074
    ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
             "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");

1075 1076 1077 1078
    /* M/S introduces horrible artifacts with multichannel files, this is temporary */
    if (s->channels > 3)
        s->options.mid_side = 0;

1079
    if ((ret = dsp_init(avctx, s)) < 0)
1080 1081
        goto fail;

1082
    if ((ret = alloc_buffers(avctx, s)) < 0)
1083 1084
        goto fail;

1085 1086
    if ((ret = put_audio_specific_config(avctx)))
        goto fail;
1087

1088 1089 1090 1091
    sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];
    sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];
    lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];
    lengths[1] = ff_aac_num_swb_128[s->samplerate_index];
1092 1093
    for (i = 0; i < s->chan_map[0]; i++)
        grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
1094 1095
    if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
                           s->chan_map[0], grouping)) < 0)
1096 1097
        goto fail;
    s->psypp = ff_psy_preprocess_init(avctx);
1098
    ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
1099
    s->random_state = 0x1f2e3d4c;
1100

1101 1102 1103 1104 1105 1106
    s->abs_pow34   = abs_pow34_v;
    s->quant_bands = quantize_bands;

    if (ARCH_X86)
        ff_aac_dsp_init_x86(s);

1107
    if (HAVE_MIPSDSP)
1108 1109
        ff_aac_coder_init_mips(s);

1110
    if ((ret = ff_thread_once(&aac_table_init, &aac_encode_init_tables)) != 0)
1111
        return AVERROR_UNKNOWN;
1112

1113 1114
    ff_af_queue_init(avctx, &s->afq);

1115 1116 1117 1118 1119 1120
    return 0;
fail:
    aac_encode_end(avctx);
    return ret;
}

1121 1122
#define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
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 1125
        {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},
        {"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 1129
    {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
    {"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 1132
    {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
    {"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 1135 1136 1137
    {NULL}
};

static const AVClass aacenc_class = {
1138 1139 1140 1141
    .class_name = "AAC encoder",
    .item_name  = av_default_item_name,
    .option     = aacenc_options,
    .version    = LIBAVUTIL_VERSION_INT,
1142 1143
};

1144 1145 1146 1147 1148
static const AVCodecDefault aac_encode_defaults[] = {
    { "b", "0" },
    { NULL }
};

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 1155
    .priv_data_size = sizeof(AACEncContext),
    .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 1164
                                                     AV_SAMPLE_FMT_NONE },
    .priv_class     = &aacenc_class,
1165
};