avformat/mpeg: demux ivtv captions
[ffmpeg.git] / libavcodec / aacenc.c
blob3ff61f788b7ce36322491f78e5becd869592f86f
1 /*
2 * AAC encoder
3 * Copyright (C) 2008 Konstantin Shishkov
5 * This file is part of FFmpeg.
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.
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.
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
22 /**
23 * @file
24 * AAC encoder
27 /***********************************
28 * TODOs:
29 * add sane pulse detection
30 ***********************************/
31 #include <float.h>
33 #include "libavutil/channel_layout.h"
34 #include "libavutil/libm.h"
35 #include "libavutil/float_dsp.h"
36 #include "libavutil/mem.h"
37 #include "libavutil/opt.h"
38 #include "avcodec.h"
39 #include "codec_internal.h"
40 #include "encode.h"
41 #include "put_bits.h"
42 #include "mpeg4audio.h"
43 #include "sinewin.h"
44 #include "profiles.h"
45 #include "version.h"
47 #include "aac.h"
48 #include "aactab.h"
49 #include "aacenc.h"
50 #include "aacenctab.h"
51 #include "aacenc_utils.h"
53 #include "psymodel.h"
55 /**
56 * List of PCE (Program Configuration Element) for the channel layouts listed
57 * in channel_layout.h
59 * For those wishing in the future to add other layouts:
61 * - num_ele: number of elements in each group of front, side, back, lfe channels
62 * (an element is of type SCE (single channel), CPE (channel pair) for
63 * the first 3 groups; and is LFE for LFE group).
65 * - pairing: 0 for an SCE element or 1 for a CPE; does not apply to LFE group
67 * - index: there are three independent indices for SCE, CPE and LFE;
68 * they are incremented irrespective of the group to which the element belongs;
69 * they are not reset when going from one group to another
71 * Example: for 7.0 channel layout,
72 * .pairing = { { 1, 0 }, { 1 }, { 1 }, }, (3 CPE and 1 SCE in front group)
73 * .index = { { 0, 0 }, { 1 }, { 2 }, },
74 * (index is 0 for the single SCE but goes from 0 to 2 for the CPEs)
76 * The index order impacts the channel ordering. But is otherwise arbitrary
77 * (the sequence could have been 2, 0, 1 instead of 0, 1, 2).
79 * Spec allows for discontinuous indices, e.g. if one has a total of two SCE,
80 * SCE.0 SCE.15 is OK per spec; BUT it won't be decoded by our AAC decoder
81 * which at this time requires that indices fully cover some range starting
82 * from 0 (SCE.1 SCE.0 is OK but not SCE.0 SCE.15).
84 * - config_map: total number of elements and their types. Beware, the way the
85 * types are ordered impacts the final channel ordering.
87 * - reorder_map: reorders the channels.
90 static const AACPCEInfo aac_pce_configs[] = {
92 .layout = AV_CHANNEL_LAYOUT_MONO,
93 .num_ele = { 1, 0, 0, 0 },
94 .pairing = { { 0 }, },
95 .index = { { 0 }, },
96 .config_map = { 1, TYPE_SCE, },
97 .reorder_map = { 0 },
100 .layout = AV_CHANNEL_LAYOUT_STEREO,
101 .num_ele = { 1, 0, 0, 0 },
102 .pairing = { { 1 }, },
103 .index = { { 0 }, },
104 .config_map = { 1, TYPE_CPE, },
105 .reorder_map = { 0, 1 },
108 .layout = AV_CHANNEL_LAYOUT_2POINT1,
109 .num_ele = { 1, 0, 0, 1 },
110 .pairing = { { 1 }, },
111 .index = { { 0 },{ 0 },{ 0 },{ 0 } },
112 .config_map = { 2, TYPE_CPE, TYPE_LFE },
113 .reorder_map = { 0, 1, 2 },
116 .layout = AV_CHANNEL_LAYOUT_2_1,
117 .num_ele = { 1, 0, 1, 0 },
118 .pairing = { { 1 },{ 0 },{ 0 } },
119 .index = { { 0 },{ 0 },{ 0 }, },
120 .config_map = { 2, TYPE_CPE, TYPE_SCE },
121 .reorder_map = { 0, 1, 2 },
124 .layout = AV_CHANNEL_LAYOUT_SURROUND,
125 .num_ele = { 2, 0, 0, 0 },
126 .pairing = { { 1, 0 }, },
127 .index = { { 0, 0 }, },
128 .config_map = { 2, TYPE_CPE, TYPE_SCE, },
129 .reorder_map = { 0, 1, 2 },
132 .layout = AV_CHANNEL_LAYOUT_3POINT1,
133 .num_ele = { 2, 0, 0, 1 },
134 .pairing = { { 1, 0 }, },
135 .index = { { 0, 0 }, { 0 }, { 0 }, { 0 }, },
136 .config_map = { 3, TYPE_CPE, TYPE_SCE, TYPE_LFE },
137 .reorder_map = { 0, 1, 2, 3 },
140 .layout = AV_CHANNEL_LAYOUT_4POINT0,
141 .num_ele = { 2, 0, 1, 0 },
142 .pairing = { { 1, 0 }, { 0 }, { 0 }, },
143 .index = { { 0, 0 }, { 0 }, { 1 } },
144 .config_map = { 3, TYPE_CPE, TYPE_SCE, TYPE_SCE },
145 .reorder_map = { 0, 1, 2, 3 },
148 .layout = AV_CHANNEL_LAYOUT_4POINT1,
149 .num_ele = { 2, 1, 1, 0 },
150 .pairing = { { 1, 0 }, { 0 }, { 0 }, },
151 .index = { { 0, 0 }, { 1 }, { 2 }, { 0 } },
152 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_SCE },
153 .reorder_map = { 0, 1, 2, 3, 4 },
156 .layout = AV_CHANNEL_LAYOUT_2_2,
157 .num_ele = { 1, 1, 0, 0 },
158 .pairing = { { 1 }, { 1 }, },
159 .index = { { 0 }, { 1 }, },
160 .config_map = { 2, TYPE_CPE, TYPE_CPE },
161 .reorder_map = { 0, 1, 2, 3 },
164 .layout = AV_CHANNEL_LAYOUT_QUAD,
165 .num_ele = { 1, 0, 1, 0 },
166 .pairing = { { 1 }, { 0 }, { 1 }, },
167 .index = { { 0 }, { 0 }, { 1 } },
168 .config_map = { 2, TYPE_CPE, TYPE_CPE },
169 .reorder_map = { 0, 1, 2, 3 },
172 .layout = AV_CHANNEL_LAYOUT_5POINT0,
173 .num_ele = { 2, 1, 0, 0 },
174 .pairing = { { 1, 0 }, { 1 }, },
175 .index = { { 0, 0 }, { 1 } },
176 .config_map = { 3, TYPE_CPE, TYPE_SCE, TYPE_CPE },
177 .reorder_map = { 0, 1, 2, 3, 4 },
180 .layout = AV_CHANNEL_LAYOUT_5POINT1,
181 .num_ele = { 2, 1, 1, 0 },
182 .pairing = { { 1, 0 }, { 0 }, { 1 }, },
183 .index = { { 0, 0 }, { 1 }, { 1 } },
184 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE },
185 .reorder_map = { 0, 1, 2, 3, 4, 5 },
188 .layout = AV_CHANNEL_LAYOUT_5POINT0_BACK,
189 .num_ele = { 2, 0, 1, 0 },
190 .pairing = { { 1, 0 }, { 0 }, { 1 } },
191 .index = { { 0, 0 }, { 0 }, { 1 } },
192 .config_map = { 3, TYPE_CPE, TYPE_SCE, TYPE_CPE },
193 .reorder_map = { 0, 1, 2, 3, 4 },
196 .layout = AV_CHANNEL_LAYOUT_5POINT1_BACK,
197 .num_ele = { 2, 1, 1, 0 },
198 .pairing = { { 1, 0 }, { 0 }, { 1 }, },
199 .index = { { 0, 0 }, { 1 }, { 1 } },
200 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE },
201 .reorder_map = { 0, 1, 2, 3, 4, 5 },
204 .layout = AV_CHANNEL_LAYOUT_6POINT0,
205 .num_ele = { 2, 1, 1, 0 },
206 .pairing = { { 1, 0 }, { 1 }, { 0 }, },
207 .index = { { 0, 0 }, { 1 }, { 1 } },
208 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
209 .reorder_map = { 0, 1, 2, 3, 4, 5 },
212 .layout = AV_CHANNEL_LAYOUT_6POINT0_FRONT,
213 .num_ele = { 2, 1, 0, 0 },
214 .pairing = { { 1, 1 }, { 1 } },
215 .index = { { 1, 0 }, { 2 }, },
216 .config_map = { 3, TYPE_CPE, TYPE_CPE, TYPE_CPE, },
217 .reorder_map = { 0, 1, 2, 3, 4, 5 },
220 .layout = AV_CHANNEL_LAYOUT_HEXAGONAL,
221 .num_ele = { 2, 0, 2, 0 },
222 .pairing = { { 1, 0 },{ 0 },{ 1, 0 }, },
223 .index = { { 0, 0 },{ 0 },{ 1, 1 } },
224 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE, },
225 .reorder_map = { 0, 1, 2, 3, 4, 5 },
228 .layout = AV_CHANNEL_LAYOUT_6POINT1,
229 .num_ele = { 2, 1, 2, 0 },
230 .pairing = { { 1, 0 },{ 0 },{ 1, 0 }, },
231 .index = { { 0, 0 },{ 1 },{ 1, 2 } },
232 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
233 .reorder_map = { 0, 1, 2, 3, 4, 5, 6 },
236 .layout = AV_CHANNEL_LAYOUT_6POINT1_BACK,
237 .num_ele = { 2, 1, 2, 0 },
238 .pairing = { { 1, 0 }, { 0 }, { 1, 0 }, },
239 .index = { { 0, 0 }, { 1 }, { 1, 2 } },
240 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
241 .reorder_map = { 0, 1, 2, 3, 4, 5, 6 },
244 .layout = AV_CHANNEL_LAYOUT_6POINT1_FRONT,
245 .num_ele = { 2, 1, 2, 0 },
246 .pairing = { { 1, 0 }, { 0 }, { 1, 0 }, },
247 .index = { { 0, 0 }, { 1 }, { 1, 2 } },
248 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
249 .reorder_map = { 0, 1, 2, 3, 4, 5, 6 },
252 .layout = AV_CHANNEL_LAYOUT_7POINT0,
253 .num_ele = { 2, 1, 1, 0 },
254 .pairing = { { 1, 0 }, { 1 }, { 1 }, },
255 .index = { { 0, 0 }, { 1 }, { 2 }, },
256 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_CPE },
257 .reorder_map = { 0, 1, 2, 3, 4, 5, 6 },
260 .layout = AV_CHANNEL_LAYOUT_7POINT0_FRONT,
261 .num_ele = { 2, 1, 1, 0 },
262 .pairing = { { 1, 0 }, { 1 }, { 1 }, },
263 .index = { { 0, 0 }, { 1 }, { 2 }, },
264 .config_map = { 4, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_CPE },
265 .reorder_map = { 0, 1, 2, 3, 4, 5, 6 },
268 .layout = AV_CHANNEL_LAYOUT_7POINT1,
269 .num_ele = { 2, 1, 2, 0 },
270 .pairing = { { 1, 0 }, { 0 }, { 1, 1 }, },
271 .index = { { 0, 0 }, { 1 }, { 1, 2 }, { 0 } },
272 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE, TYPE_CPE },
273 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7 },
276 .layout = AV_CHANNEL_LAYOUT_7POINT1_WIDE,
277 .num_ele = { 2, 1, 2, 0 },
278 .pairing = { { 1, 0 }, { 0 },{ 1, 1 }, },
279 .index = { { 0, 0 }, { 1 }, { 1, 2 }, { 0 } },
280 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE, TYPE_CPE },
281 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7 },
284 .layout = AV_CHANNEL_LAYOUT_7POINT1_WIDE_BACK,
285 .num_ele = { 2, 1, 2, 0 },
286 .pairing = { { 1, 0 }, { 0 }, { 1, 1 }, },
287 .index = { { 0, 0 }, { 1 }, { 1, 2 }, { 0 } },
288 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_SCE, TYPE_CPE, TYPE_CPE },
289 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7 },
292 .layout = AV_CHANNEL_LAYOUT_OCTAGONAL,
293 .num_ele = { 2, 1, 2, 0 },
294 .pairing = { { 1, 0 }, { 1 }, { 1, 0 }, },
295 .index = { { 0, 0 }, { 1 }, { 2, 1 } },
296 .config_map = { 5, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_SCE },
297 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7 },
299 { /* Meant for order 2/mixed ambisonics */
300 .layout = { .order = AV_CHANNEL_ORDER_NATIVE, .nb_channels = 9,
301 .u.mask = AV_CH_LAYOUT_OCTAGONAL | AV_CH_TOP_CENTER },
302 .num_ele = { 2, 2, 2, 0 },
303 .pairing = { { 1, 0 }, { 1, 0 }, { 1, 0 }, },
304 .index = { { 0, 0 }, { 1, 1 }, { 2, 2 } },
305 .config_map = { 6, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
306 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7, 8 },
308 { /* Meant for order 2/mixed ambisonics */
309 .layout = { .order = AV_CHANNEL_ORDER_NATIVE, .nb_channels = 10,
310 .u.mask = AV_CH_LAYOUT_6POINT0_FRONT | AV_CH_BACK_CENTER |
311 AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT | AV_CH_TOP_CENTER },
312 .num_ele = { 2, 2, 2, 0 },
313 .pairing = { { 1, 1 }, { 1, 0 }, { 1, 0 }, },
314 .index = { { 0, 1 }, { 2, 0 }, { 3, 1 } },
315 .config_map = { 6, TYPE_CPE, TYPE_CPE, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
316 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 },
319 .layout = AV_CHANNEL_LAYOUT_HEXADECAGONAL,
320 .num_ele = { 4, 2, 4, 0 },
321 .pairing = { { 1, 0, 1, 0 }, { 1, 1 }, { 1, 0, 1, 0 }, },
322 .index = { { 0, 0, 1, 1 }, { 2, 3 }, { 4, 2, 5, 3 } },
323 .config_map = { 10, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_CPE, TYPE_SCE, TYPE_CPE, TYPE_SCE },
324 .reorder_map = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
328 static void put_pce(PutBitContext *pb, AVCodecContext *avctx)
330 int i, j;
331 AACEncContext *s = avctx->priv_data;
332 AACPCEInfo *pce = &s->pce;
333 const int bitexact = avctx->flags & AV_CODEC_FLAG_BITEXACT;
334 const char *aux_data = bitexact ? "Lavc" : LIBAVCODEC_IDENT;
336 put_bits(pb, 4, 0);
338 put_bits(pb, 2, avctx->profile);
339 put_bits(pb, 4, s->samplerate_index);
341 put_bits(pb, 4, pce->num_ele[0]); /* Front */
342 put_bits(pb, 4, pce->num_ele[1]); /* Side */
343 put_bits(pb, 4, pce->num_ele[2]); /* Back */
344 put_bits(pb, 2, pce->num_ele[3]); /* LFE */
345 put_bits(pb, 3, 0); /* Assoc data */
346 put_bits(pb, 4, 0); /* CCs */
348 put_bits(pb, 1, 0); /* Stereo mixdown */
349 put_bits(pb, 1, 0); /* Mono mixdown */
350 put_bits(pb, 1, 0); /* Something else */
352 for (i = 0; i < 4; i++) {
353 for (j = 0; j < pce->num_ele[i]; j++) {
354 if (i < 3)
355 put_bits(pb, 1, pce->pairing[i][j]);
356 put_bits(pb, 4, pce->index[i][j]);
360 align_put_bits(pb);
361 put_bits(pb, 8, strlen(aux_data));
362 ff_put_string(pb, aux_data, 0);
366 * Make AAC audio config object.
367 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
369 static int put_audio_specific_config(AVCodecContext *avctx)
371 PutBitContext pb;
372 AACEncContext *s = avctx->priv_data;
373 int channels = (!s->needs_pce)*(s->channels - (s->channels == 8 ? 1 : 0));
374 const int max_size = 32;
376 avctx->extradata = av_mallocz(max_size);
377 if (!avctx->extradata)
378 return AVERROR(ENOMEM);
380 init_put_bits(&pb, avctx->extradata, max_size);
381 put_bits(&pb, 5, s->profile+1); //profile
382 put_bits(&pb, 4, s->samplerate_index); //sample rate index
383 put_bits(&pb, 4, channels);
384 //GASpecificConfig
385 put_bits(&pb, 1, 0); //frame length - 1024 samples
386 put_bits(&pb, 1, 0); //does not depend on core coder
387 put_bits(&pb, 1, 0); //is not extension
388 if (s->needs_pce)
389 put_pce(&pb, avctx);
391 //Explicitly Mark SBR absent
392 put_bits(&pb, 11, 0x2b7); //sync extension
393 put_bits(&pb, 5, AOT_SBR);
394 put_bits(&pb, 1, 0);
395 flush_put_bits(&pb);
396 avctx->extradata_size = put_bytes_output(&pb);
398 return 0;
401 void ff_quantize_band_cost_cache_init(struct AACEncContext *s)
403 ++s->quantize_band_cost_cache_generation;
404 if (s->quantize_band_cost_cache_generation == 0) {
405 memset(s->quantize_band_cost_cache, 0, sizeof(s->quantize_band_cost_cache));
406 s->quantize_band_cost_cache_generation = 1;
410 #define WINDOW_FUNC(type) \
411 static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
412 SingleChannelElement *sce, \
413 const float *audio)
415 WINDOW_FUNC(only_long)
417 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
418 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
419 float *out = sce->ret_buf;
421 fdsp->vector_fmul (out, audio, lwindow, 1024);
422 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
425 WINDOW_FUNC(long_start)
427 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
428 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
429 float *out = sce->ret_buf;
431 fdsp->vector_fmul(out, audio, lwindow, 1024);
432 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
433 fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
434 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
437 WINDOW_FUNC(long_stop)
439 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
440 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
441 float *out = sce->ret_buf;
443 memset(out, 0, sizeof(out[0]) * 448);
444 fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
445 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
446 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
449 WINDOW_FUNC(eight_short)
451 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
452 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
453 const float *in = audio + 448;
454 float *out = sce->ret_buf;
455 int w;
457 for (w = 0; w < 8; w++) {
458 fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
459 out += 128;
460 in += 128;
461 fdsp->vector_fmul_reverse(out, in, swindow, 128);
462 out += 128;
466 static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
467 SingleChannelElement *sce,
468 const float *audio) = {
469 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
470 [LONG_START_SEQUENCE] = apply_long_start_window,
471 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
472 [LONG_STOP_SEQUENCE] = apply_long_stop_window
475 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
476 float *audio)
478 int i;
479 float *output = sce->ret_buf;
481 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
483 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
484 s->mdct1024_fn(s->mdct1024, sce->coeffs, output, sizeof(float));
485 else
486 for (i = 0; i < 1024; i += 128)
487 s->mdct128_fn(s->mdct128, &sce->coeffs[i], output + i*2, sizeof(float));
488 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
489 memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
493 * Encode ics_info element.
494 * @see Table 4.6 (syntax of ics_info)
496 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
498 int w;
500 put_bits(&s->pb, 1, 0); // ics_reserved bit
501 put_bits(&s->pb, 2, info->window_sequence[0]);
502 put_bits(&s->pb, 1, info->use_kb_window[0]);
503 if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
504 put_bits(&s->pb, 6, info->max_sfb);
505 put_bits(&s->pb, 1, !!info->predictor_present);
506 } else {
507 put_bits(&s->pb, 4, info->max_sfb);
508 for (w = 1; w < 8; w++)
509 put_bits(&s->pb, 1, !info->group_len[w]);
514 * Encode MS data.
515 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
517 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
519 int i, w;
521 put_bits(pb, 2, cpe->ms_mode);
522 if (cpe->ms_mode == 1)
523 for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
524 for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
525 put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
529 * Produce integer coefficients from scalefactors provided by the model.
531 static void adjust_frame_information(ChannelElement *cpe, int chans)
533 int i, w, w2, g, ch;
534 int maxsfb, cmaxsfb;
536 for (ch = 0; ch < chans; ch++) {
537 IndividualChannelStream *ics = &cpe->ch[ch].ics;
538 maxsfb = 0;
539 cpe->ch[ch].pulse.num_pulse = 0;
540 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
541 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
543 maxsfb = FFMAX(maxsfb, cmaxsfb);
545 ics->max_sfb = maxsfb;
547 //adjust zero bands for window groups
548 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
549 for (g = 0; g < ics->max_sfb; g++) {
550 i = 1;
551 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
552 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
553 i = 0;
554 break;
557 cpe->ch[ch].zeroes[w*16 + g] = i;
562 if (chans > 1 && cpe->common_window) {
563 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
564 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
565 int msc = 0;
566 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
567 ics1->max_sfb = ics0->max_sfb;
568 for (w = 0; w < ics0->num_windows*16; w += 16)
569 for (i = 0; i < ics0->max_sfb; i++)
570 if (cpe->ms_mask[w+i])
571 msc++;
572 if (msc == 0 || ics0->max_sfb == 0)
573 cpe->ms_mode = 0;
574 else
575 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
579 static void apply_intensity_stereo(ChannelElement *cpe)
581 int w, w2, g, i;
582 IndividualChannelStream *ics = &cpe->ch[0].ics;
583 if (!cpe->common_window)
584 return;
585 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
586 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
587 int start = (w+w2) * 128;
588 for (g = 0; g < ics->num_swb; g++) {
589 int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
590 float scale = cpe->ch[0].is_ener[w*16+g];
591 if (!cpe->is_mask[w*16 + g]) {
592 start += ics->swb_sizes[g];
593 continue;
595 if (cpe->ms_mask[w*16 + g])
596 p *= -1;
597 for (i = 0; i < ics->swb_sizes[g]; i++) {
598 float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
599 cpe->ch[0].coeffs[start+i] = sum;
600 cpe->ch[1].coeffs[start+i] = 0.0f;
602 start += ics->swb_sizes[g];
608 static void apply_mid_side_stereo(ChannelElement *cpe)
610 int w, w2, g, i;
611 IndividualChannelStream *ics = &cpe->ch[0].ics;
612 if (!cpe->common_window)
613 return;
614 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
615 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
616 int start = (w+w2) * 128;
617 for (g = 0; g < ics->num_swb; g++) {
618 /* ms_mask can be used for other purposes in PNS and I/S,
619 * so must not apply M/S if any band uses either, even if
620 * ms_mask is set.
622 if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]
623 || cpe->ch[0].band_type[w*16 + g] >= NOISE_BT
624 || cpe->ch[1].band_type[w*16 + g] >= NOISE_BT) {
625 start += ics->swb_sizes[g];
626 continue;
628 for (i = 0; i < ics->swb_sizes[g]; i++) {
629 float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
630 float R = L - cpe->ch[1].coeffs[start+i];
631 cpe->ch[0].coeffs[start+i] = L;
632 cpe->ch[1].coeffs[start+i] = R;
634 start += ics->swb_sizes[g];
641 * Encode scalefactor band coding type.
643 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
645 int w;
647 if (s->coder->set_special_band_scalefactors)
648 s->coder->set_special_band_scalefactors(s, sce);
650 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
651 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
655 * Encode scalefactors.
657 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
658 SingleChannelElement *sce)
660 int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
661 int off_is = 0, noise_flag = 1;
662 int i, w;
664 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
665 for (i = 0; i < sce->ics.max_sfb; i++) {
666 if (!sce->zeroes[w*16 + i]) {
667 if (sce->band_type[w*16 + i] == NOISE_BT) {
668 diff = sce->sf_idx[w*16 + i] - off_pns;
669 off_pns = sce->sf_idx[w*16 + i];
670 if (noise_flag-- > 0) {
671 put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
672 continue;
674 } else if (sce->band_type[w*16 + i] == INTENSITY_BT ||
675 sce->band_type[w*16 + i] == INTENSITY_BT2) {
676 diff = sce->sf_idx[w*16 + i] - off_is;
677 off_is = sce->sf_idx[w*16 + i];
678 } else {
679 diff = sce->sf_idx[w*16 + i] - off_sf;
680 off_sf = sce->sf_idx[w*16 + i];
682 diff += SCALE_DIFF_ZERO;
683 av_assert0(diff >= 0 && diff <= 120);
684 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
691 * Encode pulse data.
693 static void encode_pulses(AACEncContext *s, Pulse *pulse)
695 int i;
697 put_bits(&s->pb, 1, !!pulse->num_pulse);
698 if (!pulse->num_pulse)
699 return;
701 put_bits(&s->pb, 2, pulse->num_pulse - 1);
702 put_bits(&s->pb, 6, pulse->start);
703 for (i = 0; i < pulse->num_pulse; i++) {
704 put_bits(&s->pb, 5, pulse->pos[i]);
705 put_bits(&s->pb, 4, pulse->amp[i]);
710 * Encode spectral coefficients processed by psychoacoustic model.
712 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
714 int start, i, w, w2;
716 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
717 start = 0;
718 for (i = 0; i < sce->ics.max_sfb; i++) {
719 if (sce->zeroes[w*16 + i]) {
720 start += sce->ics.swb_sizes[i];
721 continue;
723 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
724 s->coder->quantize_and_encode_band(s, &s->pb,
725 &sce->coeffs[start + w2*128],
726 NULL, sce->ics.swb_sizes[i],
727 sce->sf_idx[w*16 + i],
728 sce->band_type[w*16 + i],
729 s->lambda,
730 sce->ics.window_clipping[w]);
732 start += sce->ics.swb_sizes[i];
738 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
740 static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
742 int start, i, j, w;
744 if (sce->ics.clip_avoidance_factor < 1.0f) {
745 for (w = 0; w < sce->ics.num_windows; w++) {
746 start = 0;
747 for (i = 0; i < sce->ics.max_sfb; i++) {
748 float *swb_coeffs = &sce->coeffs[start + w*128];
749 for (j = 0; j < sce->ics.swb_sizes[i]; j++)
750 swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
751 start += sce->ics.swb_sizes[i];
758 * Encode one channel of audio data.
760 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
761 SingleChannelElement *sce,
762 int common_window)
764 put_bits(&s->pb, 8, sce->sf_idx[0]);
765 if (!common_window) {
766 put_ics_info(s, &sce->ics);
767 if (s->coder->encode_main_pred)
768 s->coder->encode_main_pred(s, sce);
769 if (s->coder->encode_ltp_info)
770 s->coder->encode_ltp_info(s, sce, 0);
772 encode_band_info(s, sce);
773 encode_scale_factors(avctx, s, sce);
774 encode_pulses(s, &sce->pulse);
775 put_bits(&s->pb, 1, !!sce->tns.present);
776 if (s->coder->encode_tns_info)
777 s->coder->encode_tns_info(s, sce);
778 put_bits(&s->pb, 1, 0); //ssr
779 encode_spectral_coeffs(s, sce);
780 return 0;
784 * Write some auxiliary information about the created AAC file.
786 static void put_bitstream_info(AACEncContext *s, const char *name)
788 int i, namelen, padbits;
790 namelen = strlen(name) + 2;
791 put_bits(&s->pb, 3, TYPE_FIL);
792 put_bits(&s->pb, 4, FFMIN(namelen, 15));
793 if (namelen >= 15)
794 put_bits(&s->pb, 8, namelen - 14);
795 put_bits(&s->pb, 4, 0); //extension type - filler
796 padbits = -put_bits_count(&s->pb) & 7;
797 align_put_bits(&s->pb);
798 for (i = 0; i < namelen - 2; i++)
799 put_bits(&s->pb, 8, name[i]);
800 put_bits(&s->pb, 12 - padbits, 0);
804 * Copy input samples.
805 * Channels are reordered from libavcodec's default order to AAC order.
807 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
809 int ch;
810 int end = 2048 + (frame ? frame->nb_samples : 0);
811 const uint8_t *channel_map = s->reorder_map;
813 /* copy and remap input samples */
814 for (ch = 0; ch < s->channels; ch++) {
815 /* copy last 1024 samples of previous frame to the start of the current frame */
816 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
818 /* copy new samples and zero any remaining samples */
819 if (frame) {
820 memcpy(&s->planar_samples[ch][2048],
821 frame->extended_data[channel_map[ch]],
822 frame->nb_samples * sizeof(s->planar_samples[0][0]));
824 memset(&s->planar_samples[ch][end], 0,
825 (3072 - end) * sizeof(s->planar_samples[0][0]));
829 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
830 const AVFrame *frame, int *got_packet_ptr)
832 AACEncContext *s = avctx->priv_data;
833 float **samples = s->planar_samples, *samples2, *la, *overlap;
834 ChannelElement *cpe;
835 SingleChannelElement *sce;
836 IndividualChannelStream *ics;
837 int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
838 int target_bits, rate_bits, too_many_bits, too_few_bits;
839 int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
840 int chan_el_counter[4];
841 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
843 /* add current frame to queue */
844 if (frame) {
845 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
846 return ret;
847 } else {
848 if (!s->afq.remaining_samples || (!s->afq.frame_alloc && !s->afq.frame_count))
849 return 0;
852 copy_input_samples(s, frame);
853 if (s->psypp)
854 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
856 if (!avctx->frame_num)
857 return 0;
859 start_ch = 0;
860 for (i = 0; i < s->chan_map[0]; i++) {
861 FFPsyWindowInfo* wi = windows + start_ch;
862 tag = s->chan_map[i+1];
863 chans = tag == TYPE_CPE ? 2 : 1;
864 cpe = &s->cpe[i];
865 for (ch = 0; ch < chans; ch++) {
866 int k;
867 float clip_avoidance_factor;
868 sce = &cpe->ch[ch];
869 ics = &sce->ics;
870 s->cur_channel = start_ch + ch;
871 overlap = &samples[s->cur_channel][0];
872 samples2 = overlap + 1024;
873 la = samples2 + (448+64);
874 if (!frame)
875 la = NULL;
876 if (tag == TYPE_LFE) {
877 wi[ch].window_type[0] = wi[ch].window_type[1] = ONLY_LONG_SEQUENCE;
878 wi[ch].window_shape = 0;
879 wi[ch].num_windows = 1;
880 wi[ch].grouping[0] = 1;
881 wi[ch].clipping[0] = 0;
883 /* Only the lowest 12 coefficients are used in a LFE channel.
884 * The expression below results in only the bottom 8 coefficients
885 * being used for 11.025kHz to 16kHz sample rates.
887 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
888 } else {
889 wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,
890 ics->window_sequence[0]);
892 ics->window_sequence[1] = ics->window_sequence[0];
893 ics->window_sequence[0] = wi[ch].window_type[0];
894 ics->use_kb_window[1] = ics->use_kb_window[0];
895 ics->use_kb_window[0] = wi[ch].window_shape;
896 ics->num_windows = wi[ch].num_windows;
897 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
898 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
899 ics->max_sfb = FFMIN(ics->max_sfb, ics->num_swb);
900 ics->swb_offset = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
901 ff_swb_offset_128 [s->samplerate_index]:
902 ff_swb_offset_1024[s->samplerate_index];
903 ics->tns_max_bands = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
904 ff_tns_max_bands_128 [s->samplerate_index]:
905 ff_tns_max_bands_1024[s->samplerate_index];
907 for (w = 0; w < ics->num_windows; w++)
908 ics->group_len[w] = wi[ch].grouping[w];
910 /* Calculate input sample maximums and evaluate clipping risk */
911 clip_avoidance_factor = 0.0f;
912 for (w = 0; w < ics->num_windows; w++) {
913 const float *wbuf = overlap + w * 128;
914 const int wlen = 2048 / ics->num_windows;
915 float max = 0;
916 int j;
917 /* mdct input is 2 * output */
918 for (j = 0; j < wlen; j++)
919 max = FFMAX(max, fabsf(wbuf[j]));
920 wi[ch].clipping[w] = max;
922 for (w = 0; w < ics->num_windows; w++) {
923 if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
924 ics->window_clipping[w] = 1;
925 clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
926 } else {
927 ics->window_clipping[w] = 0;
930 if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
931 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
932 } else {
933 ics->clip_avoidance_factor = 1.0f;
936 apply_window_and_mdct(s, sce, overlap);
938 if (s->options.ltp && s->coder->update_ltp) {
939 s->coder->update_ltp(s, sce);
940 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);
941 s->mdct1024_fn(s->mdct1024, sce->lcoeffs, sce->ret_buf, sizeof(float));
944 for (k = 0; k < 1024; k++) {
945 if (!(fabs(cpe->ch[ch].coeffs[k]) < 1E16)) { // Ensure headroom for energy calculation
946 av_log(avctx, AV_LOG_ERROR, "Input contains (near) NaN/+-Inf\n");
947 return AVERROR(EINVAL);
950 avoid_clipping(s, sce);
952 start_ch += chans;
954 if ((ret = ff_alloc_packet(avctx, avpkt, 8192 * s->channels)) < 0)
955 return ret;
956 frame_bits = its = 0;
957 do {
958 init_put_bits(&s->pb, avpkt->data, avpkt->size);
960 if ((avctx->frame_num & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
961 put_bitstream_info(s, LIBAVCODEC_IDENT);
962 start_ch = 0;
963 target_bits = 0;
964 memset(chan_el_counter, 0, sizeof(chan_el_counter));
965 for (i = 0; i < s->chan_map[0]; i++) {
966 FFPsyWindowInfo* wi = windows + start_ch;
967 const float *coeffs[2];
968 tag = s->chan_map[i+1];
969 chans = tag == TYPE_CPE ? 2 : 1;
970 cpe = &s->cpe[i];
971 cpe->common_window = 0;
972 memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
973 memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
974 put_bits(&s->pb, 3, tag);
975 put_bits(&s->pb, 4, chan_el_counter[tag]++);
976 for (ch = 0; ch < chans; ch++) {
977 sce = &cpe->ch[ch];
978 coeffs[ch] = sce->coeffs;
979 sce->ics.predictor_present = 0;
980 sce->ics.ltp.present = 0;
981 memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));
982 memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
983 memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
984 for (w = 0; w < 128; w++)
985 if (sce->band_type[w] > RESERVED_BT)
986 sce->band_type[w] = 0;
988 s->psy.bitres.alloc = -1;
989 s->psy.bitres.bits = s->last_frame_pb_count / s->channels;
990 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
991 if (s->psy.bitres.alloc > 0) {
992 /* Lambda unused here on purpose, we need to take psy's unscaled allocation */
993 target_bits += s->psy.bitres.alloc
994 * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
995 s->psy.bitres.alloc /= chans;
997 s->cur_type = tag;
998 for (ch = 0; ch < chans; ch++) {
999 s->cur_channel = start_ch + ch;
1000 if (s->options.pns && s->coder->mark_pns)
1001 s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
1002 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
1004 if (chans > 1
1005 && wi[0].window_type[0] == wi[1].window_type[0]
1006 && wi[0].window_shape == wi[1].window_shape) {
1008 cpe->common_window = 1;
1009 for (w = 0; w < wi[0].num_windows; w++) {
1010 if (wi[0].grouping[w] != wi[1].grouping[w]) {
1011 cpe->common_window = 0;
1012 break;
1016 for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
1017 sce = &cpe->ch[ch];
1018 s->cur_channel = start_ch + ch;
1019 if (s->options.tns && s->coder->search_for_tns)
1020 s->coder->search_for_tns(s, sce);
1021 if (s->options.tns && s->coder->apply_tns_filt)
1022 s->coder->apply_tns_filt(s, sce);
1023 if (sce->tns.present)
1024 tns_mode = 1;
1025 if (s->options.pns && s->coder->search_for_pns)
1026 s->coder->search_for_pns(s, avctx, sce);
1028 s->cur_channel = start_ch;
1029 if (s->options.intensity_stereo) { /* Intensity Stereo */
1030 if (s->coder->search_for_is)
1031 s->coder->search_for_is(s, avctx, cpe);
1032 if (cpe->is_mode) is_mode = 1;
1033 apply_intensity_stereo(cpe);
1035 if (s->options.pred) { /* Prediction */
1036 for (ch = 0; ch < chans; ch++) {
1037 sce = &cpe->ch[ch];
1038 s->cur_channel = start_ch + ch;
1039 if (s->options.pred && s->coder->search_for_pred)
1040 s->coder->search_for_pred(s, sce);
1041 if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
1043 if (s->coder->adjust_common_pred)
1044 s->coder->adjust_common_pred(s, cpe);
1045 for (ch = 0; ch < chans; ch++) {
1046 sce = &cpe->ch[ch];
1047 s->cur_channel = start_ch + ch;
1048 if (s->options.pred && s->coder->apply_main_pred)
1049 s->coder->apply_main_pred(s, sce);
1051 s->cur_channel = start_ch;
1053 if (s->options.mid_side) { /* Mid/Side stereo */
1054 if (s->options.mid_side == -1 && s->coder->search_for_ms)
1055 s->coder->search_for_ms(s, cpe);
1056 else if (cpe->common_window)
1057 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
1058 apply_mid_side_stereo(cpe);
1060 adjust_frame_information(cpe, chans);
1061 if (s->options.ltp) { /* LTP */
1062 for (ch = 0; ch < chans; ch++) {
1063 sce = &cpe->ch[ch];
1064 s->cur_channel = start_ch + ch;
1065 if (s->coder->search_for_ltp)
1066 s->coder->search_for_ltp(s, sce, cpe->common_window);
1067 if (sce->ics.ltp.present) pred_mode = 1;
1069 s->cur_channel = start_ch;
1070 if (s->coder->adjust_common_ltp)
1071 s->coder->adjust_common_ltp(s, cpe);
1073 if (chans == 2) {
1074 put_bits(&s->pb, 1, cpe->common_window);
1075 if (cpe->common_window) {
1076 put_ics_info(s, &cpe->ch[0].ics);
1077 if (s->coder->encode_main_pred)
1078 s->coder->encode_main_pred(s, &cpe->ch[0]);
1079 if (s->coder->encode_ltp_info)
1080 s->coder->encode_ltp_info(s, &cpe->ch[0], 1);
1081 encode_ms_info(&s->pb, cpe);
1082 if (cpe->ms_mode) ms_mode = 1;
1085 for (ch = 0; ch < chans; ch++) {
1086 s->cur_channel = start_ch + ch;
1087 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
1089 start_ch += chans;
1092 if (avctx->flags & AV_CODEC_FLAG_QSCALE) {
1093 /* When using a constant Q-scale, don't mess with lambda */
1094 break;
1097 /* rate control stuff
1098 * allow between the nominal bitrate, and what psy's bit reservoir says to target
1099 * but drift towards the nominal bitrate always
1101 frame_bits = put_bits_count(&s->pb);
1102 rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
1103 rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
1104 too_many_bits = FFMAX(target_bits, rate_bits);
1105 too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
1106 too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
1108 /* When strict bit-rate control is demanded */
1109 if (avctx->bit_rate_tolerance == 0) {
1110 if (rate_bits < frame_bits) {
1111 float ratio = ((float)rate_bits) / frame_bits;
1112 s->lambda *= FFMIN(0.9f, ratio);
1113 continue;
1115 /* reset lambda when solution is found */
1116 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
1117 break;
1120 /* When using ABR, be strict (but only for increasing) */
1121 too_few_bits = too_few_bits - too_few_bits/8;
1122 too_many_bits = too_many_bits + too_many_bits/2;
1124 if ( its == 0 /* for steady-state Q-scale tracking */
1125 || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
1126 || frame_bits >= 6144 * s->channels - 3 )
1128 float ratio = ((float)rate_bits) / frame_bits;
1130 if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
1132 * This path is for steady-state Q-scale tracking
1133 * When frame bits fall within the stable range, we still need to adjust
1134 * lambda to maintain it like so in a stable fashion (large jumps in lambda
1135 * create artifacts and should be avoided), but slowly
1137 ratio = sqrtf(sqrtf(ratio));
1138 ratio = av_clipf(ratio, 0.9f, 1.1f);
1139 } else {
1140 /* Not so fast though */
1141 ratio = sqrtf(ratio);
1143 s->lambda = av_clipf(s->lambda * ratio, FLT_EPSILON, 65536.f);
1145 /* Keep iterating if we must reduce and lambda is in the sky */
1146 if (ratio > 0.9f && ratio < 1.1f) {
1147 break;
1148 } else {
1149 if (is_mode || ms_mode || tns_mode || pred_mode) {
1150 for (i = 0; i < s->chan_map[0]; i++) {
1151 // Must restore coeffs
1152 chans = tag == TYPE_CPE ? 2 : 1;
1153 cpe = &s->cpe[i];
1154 for (ch = 0; ch < chans; ch++)
1155 memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
1158 its++;
1160 } else {
1161 break;
1163 } while (1);
1165 if (s->options.ltp && s->coder->ltp_insert_new_frame)
1166 s->coder->ltp_insert_new_frame(s);
1168 put_bits(&s->pb, 3, TYPE_END);
1169 flush_put_bits(&s->pb);
1171 s->last_frame_pb_count = put_bits_count(&s->pb);
1172 avpkt->size = put_bytes_output(&s->pb);
1174 s->lambda_sum += s->lambda;
1175 s->lambda_count++;
1177 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
1178 &avpkt->duration);
1180 avpkt->flags |= AV_PKT_FLAG_KEY;
1182 *got_packet_ptr = 1;
1183 return 0;
1186 static av_cold int aac_encode_end(AVCodecContext *avctx)
1188 AACEncContext *s = avctx->priv_data;
1190 av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_count ? s->lambda_sum / s->lambda_count : NAN);
1192 av_tx_uninit(&s->mdct1024);
1193 av_tx_uninit(&s->mdct128);
1194 ff_psy_end(&s->psy);
1195 ff_lpc_end(&s->lpc);
1196 if (s->psypp)
1197 ff_psy_preprocess_end(s->psypp);
1198 av_freep(&s->buffer.samples);
1199 av_freep(&s->cpe);
1200 av_freep(&s->fdsp);
1201 ff_af_queue_close(&s->afq);
1202 return 0;
1205 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
1207 int ret = 0;
1208 float scale = 32768.0f;
1210 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1211 if (!s->fdsp)
1212 return AVERROR(ENOMEM);
1214 if ((ret = av_tx_init(&s->mdct1024, &s->mdct1024_fn, AV_TX_FLOAT_MDCT, 0,
1215 1024, &scale, 0)) < 0)
1216 return ret;
1217 if ((ret = av_tx_init(&s->mdct128, &s->mdct128_fn, AV_TX_FLOAT_MDCT, 0,
1218 128, &scale, 0)) < 0)
1219 return ret;
1221 return 0;
1224 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
1226 int ch;
1227 if (!FF_ALLOCZ_TYPED_ARRAY(s->buffer.samples, s->channels * 3 * 1024) ||
1228 !FF_ALLOCZ_TYPED_ARRAY(s->cpe, s->chan_map[0]))
1229 return AVERROR(ENOMEM);
1231 for(ch = 0; ch < s->channels; ch++)
1232 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
1234 return 0;
1237 static av_cold int aac_encode_init(AVCodecContext *avctx)
1239 AACEncContext *s = avctx->priv_data;
1240 int i, ret = 0;
1241 const uint8_t *sizes[2];
1242 uint8_t grouping[AAC_MAX_CHANNELS];
1243 int lengths[2];
1245 /* Constants */
1246 s->last_frame_pb_count = 0;
1247 avctx->frame_size = 1024;
1248 avctx->initial_padding = 1024;
1249 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
1251 /* Channel map and unspecified bitrate guessing */
1252 s->channels = avctx->ch_layout.nb_channels;
1254 s->needs_pce = 1;
1255 for (i = 0; i < FF_ARRAY_ELEMS(aac_normal_chan_layouts); i++) {
1256 if (!av_channel_layout_compare(&avctx->ch_layout, &aac_normal_chan_layouts[i])) {
1257 s->needs_pce = s->options.pce;
1258 break;
1262 if (s->needs_pce) {
1263 char buf[64];
1264 for (i = 0; i < FF_ARRAY_ELEMS(aac_pce_configs); i++)
1265 if (!av_channel_layout_compare(&avctx->ch_layout, &aac_pce_configs[i].layout))
1266 break;
1267 av_channel_layout_describe(&avctx->ch_layout, buf, sizeof(buf));
1268 if (i == FF_ARRAY_ELEMS(aac_pce_configs)) {
1269 av_log(avctx, AV_LOG_ERROR, "Unsupported channel layout \"%s\"\n", buf);
1270 return AVERROR(EINVAL);
1272 av_log(avctx, AV_LOG_INFO, "Using a PCE to encode channel layout \"%s\"\n", buf);
1273 s->pce = aac_pce_configs[i];
1274 s->reorder_map = s->pce.reorder_map;
1275 s->chan_map = s->pce.config_map;
1276 } else {
1277 s->reorder_map = aac_chan_maps[s->channels - 1];
1278 s->chan_map = aac_chan_configs[s->channels - 1];
1281 if (!avctx->bit_rate) {
1282 for (i = 1; i <= s->chan_map[0]; i++) {
1283 avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */
1284 s->chan_map[i] == TYPE_LFE ? 16000 : /* LFE */
1285 69000 ; /* SCE */
1289 /* Samplerate */
1290 for (i = 0; i < 16; i++)
1291 if (avctx->sample_rate == ff_mpeg4audio_sample_rates[i])
1292 break;
1293 s->samplerate_index = i;
1294 ERROR_IF(s->samplerate_index == 16 ||
1295 s->samplerate_index >= ff_aac_swb_size_1024_len ||
1296 s->samplerate_index >= ff_aac_swb_size_128_len,
1297 "Unsupported sample rate %d\n", avctx->sample_rate);
1299 /* Bitrate limiting */
1300 WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
1301 "Too many bits %f > %d per frame requested, clamping to max\n",
1302 1024.0 * avctx->bit_rate / avctx->sample_rate,
1303 6144 * s->channels);
1304 avctx->bit_rate = (int64_t)FFMIN(6144 * s->channels / 1024.0 * avctx->sample_rate,
1305 avctx->bit_rate);
1307 /* Profile and option setting */
1308 avctx->profile = avctx->profile == AV_PROFILE_UNKNOWN ? AV_PROFILE_AAC_LOW :
1309 avctx->profile;
1310 for (i = 0; i < FF_ARRAY_ELEMS(aacenc_profiles); i++)
1311 if (avctx->profile == aacenc_profiles[i])
1312 break;
1313 if (avctx->profile == AV_PROFILE_MPEG2_AAC_LOW) {
1314 avctx->profile = AV_PROFILE_AAC_LOW;
1315 ERROR_IF(s->options.pred,
1316 "Main prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1317 ERROR_IF(s->options.ltp,
1318 "LTP prediction unavailable in the \"mpeg2_aac_low\" profile\n");
1319 WARN_IF(s->options.pns,
1320 "PNS unavailable in the \"mpeg2_aac_low\" profile, turning off\n");
1321 s->options.pns = 0;
1322 } else if (avctx->profile == AV_PROFILE_AAC_LTP) {
1323 s->options.ltp = 1;
1324 ERROR_IF(s->options.pred,
1325 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1326 } else if (avctx->profile == AV_PROFILE_AAC_MAIN) {
1327 s->options.pred = 1;
1328 ERROR_IF(s->options.ltp,
1329 "LTP prediction unavailable in the \"aac_main\" profile\n");
1330 } else if (s->options.ltp) {
1331 avctx->profile = AV_PROFILE_AAC_LTP;
1332 WARN_IF(1,
1333 "Chainging profile to \"aac_ltp\"\n");
1334 ERROR_IF(s->options.pred,
1335 "Main prediction unavailable in the \"aac_ltp\" profile\n");
1336 } else if (s->options.pred) {
1337 avctx->profile = AV_PROFILE_AAC_MAIN;
1338 WARN_IF(1,
1339 "Chainging profile to \"aac_main\"\n");
1340 ERROR_IF(s->options.ltp,
1341 "LTP prediction unavailable in the \"aac_main\" profile\n");
1343 s->profile = avctx->profile;
1345 /* Coder limitations */
1346 s->coder = &ff_aac_coders[s->options.coder];
1347 if (s->options.coder == AAC_CODER_ANMR) {
1348 ERROR_IF(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1349 "The ANMR coder is considered experimental, add -strict -2 to enable!\n");
1350 s->options.intensity_stereo = 0;
1351 s->options.pns = 0;
1353 ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,
1354 "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");
1356 /* M/S introduces horrible artifacts with multichannel files, this is temporary */
1357 if (s->channels > 3)
1358 s->options.mid_side = 0;
1360 // Initialize static tables
1361 ff_aac_float_common_init();
1363 if ((ret = dsp_init(avctx, s)) < 0)
1364 return ret;
1366 if ((ret = alloc_buffers(avctx, s)) < 0)
1367 return ret;
1369 if ((ret = put_audio_specific_config(avctx)))
1370 return ret;
1372 sizes[0] = ff_aac_swb_size_1024[s->samplerate_index];
1373 sizes[1] = ff_aac_swb_size_128[s->samplerate_index];
1374 lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];
1375 lengths[1] = ff_aac_num_swb_128[s->samplerate_index];
1376 for (i = 0; i < s->chan_map[0]; i++)
1377 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
1378 if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
1379 s->chan_map[0], grouping)) < 0)
1380 return ret;
1381 s->psypp = ff_psy_preprocess_init(avctx);
1382 ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
1383 s->random_state = 0x1f2e3d4c;
1385 ff_aacenc_dsp_init(&s->aacdsp);
1387 ff_af_queue_init(avctx, &s->afq);
1389 return 0;
1392 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
1393 static const AVOption aacenc_options[] = {
1394 {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, .unit = "coder"},
1395 {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, .unit = "coder"},
1396 {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, .unit = "coder"},
1397 {"fast", "Fast search", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, .unit = "coder"},
1398 {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},
1399 {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1400 {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1401 {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},
1402 {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1403 {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1404 {"aac_pce", "Forces the use of PCEs", offsetof(AACEncContext, options.pce), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},
1405 FF_AAC_PROFILE_OPTS
1406 {NULL}
1409 static const AVClass aacenc_class = {
1410 .class_name = "AAC encoder",
1411 .item_name = av_default_item_name,
1412 .option = aacenc_options,
1413 .version = LIBAVUTIL_VERSION_INT,
1416 static const FFCodecDefault aac_encode_defaults[] = {
1417 { "b", "0" },
1418 { NULL }
1421 const FFCodec ff_aac_encoder = {
1422 .p.name = "aac",
1423 CODEC_LONG_NAME("AAC (Advanced Audio Coding)"),
1424 .p.type = AVMEDIA_TYPE_AUDIO,
1425 .p.id = AV_CODEC_ID_AAC,
1426 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
1427 AV_CODEC_CAP_SMALL_LAST_FRAME,
1428 .priv_data_size = sizeof(AACEncContext),
1429 .init = aac_encode_init,
1430 FF_CODEC_ENCODE_CB(aac_encode_frame),
1431 .close = aac_encode_end,
1432 .defaults = aac_encode_defaults,
1433 .p.supported_samplerates = ff_mpeg4audio_sample_rates,
1434 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1435 .p.sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
1436 AV_SAMPLE_FMT_NONE },
1437 .p.priv_class = &aacenc_class,