1 /*
2 * Copyright (c) 2012 Andrew D'Addesio
3 * Copyright (c) 2013-2014 Mozilla Corporation
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * Opus decoder/parser shared code
25 */
26
27 #include <stdint.h>
28
29 #include "libavutil/error.h"
30 #include "libavutil/ffmath.h"
31
32 #include "opus_celt.h"
33 #include "opustab.h"
34 #include "vorbis.h"
35
36 static const uint16_t opus_frame_duration[32] = {
37 480, 960, 1920, 2880,
38 480, 960, 1920, 2880,
39 480, 960, 1920, 2880,
40 480, 960,
41 480, 960,
42 120, 240, 480, 960,
43 120, 240, 480, 960,
44 120, 240, 480, 960,
45 120, 240, 480, 960,
46 };
47
48 /**
49 * Read a 1- or 2-byte frame length
50 */
xiph_lacing_16bit(const uint8_t ** ptr,const uint8_t * end)51 static inline int xiph_lacing_16bit(const uint8_t **ptr, const uint8_t *end)
52 {
53 int val;
54
55 if (*ptr >= end)
56 return AVERROR_INVALIDDATA;
57 val = *(*ptr)++;
58 if (val >= 252) {
59 if (*ptr >= end)
60 return AVERROR_INVALIDDATA;
61 val += 4 * *(*ptr)++;
62 }
63 return val;
64 }
65
66 /**
67 * Read a multi-byte length (used for code 3 packet padding size)
68 */
xiph_lacing_full(const uint8_t ** ptr,const uint8_t * end)69 static inline int xiph_lacing_full(const uint8_t **ptr, const uint8_t *end)
70 {
71 int val = 0;
72 int next;
73
74 while (1) {
75 if (*ptr >= end || val > INT_MAX - 254)
76 return AVERROR_INVALIDDATA;
77 next = *(*ptr)++;
78 val += next;
79 if (next < 255)
80 break;
81 else
82 val--;
83 }
84 return val;
85 }
86
87 /**
88 * Parse Opus packet info from raw packet data
89 */
ff_opus_parse_packet(OpusPacket * pkt,const uint8_t * buf,int buf_size,int self_delimiting)90 int ff_opus_parse_packet(OpusPacket *pkt, const uint8_t *buf, int buf_size,
91 int self_delimiting)
92 {
93 const uint8_t *ptr = buf;
94 const uint8_t *end = buf + buf_size;
95 int padding = 0;
96 int frame_bytes, i;
97
98 if (buf_size < 1)
99 goto fail;
100
101 /* TOC byte */
102 i = *ptr++;
103 pkt->code = (i ) & 0x3;
104 pkt->stereo = (i >> 2) & 0x1;
105 pkt->config = (i >> 3) & 0x1F;
106
107 /* code 2 and code 3 packets have at least 1 byte after the TOC */
108 if (pkt->code >= 2 && buf_size < 2)
109 goto fail;
110
111 switch (pkt->code) {
112 case 0:
113 /* 1 frame */
114 pkt->frame_count = 1;
115 pkt->vbr = 0;
116
117 if (self_delimiting) {
118 int len = xiph_lacing_16bit(&ptr, end);
119 if (len < 0 || len > end - ptr)
120 goto fail;
121 end = ptr + len;
122 buf_size = end - buf;
123 }
124
125 frame_bytes = end - ptr;
126 if (frame_bytes > MAX_FRAME_SIZE)
127 goto fail;
128 pkt->frame_offset[0] = ptr - buf;
129 pkt->frame_size[0] = frame_bytes;
130 break;
131 case 1:
132 /* 2 frames, equal size */
133 pkt->frame_count = 2;
134 pkt->vbr = 0;
135
136 if (self_delimiting) {
137 int len = xiph_lacing_16bit(&ptr, end);
138 if (len < 0 || 2 * len > end - ptr)
139 goto fail;
140 end = ptr + 2 * len;
141 buf_size = end - buf;
142 }
143
144 frame_bytes = end - ptr;
145 if (frame_bytes & 1 || frame_bytes >> 1 > MAX_FRAME_SIZE)
146 goto fail;
147 pkt->frame_offset[0] = ptr - buf;
148 pkt->frame_size[0] = frame_bytes >> 1;
149 pkt->frame_offset[1] = pkt->frame_offset[0] + pkt->frame_size[0];
150 pkt->frame_size[1] = frame_bytes >> 1;
151 break;
152 case 2:
153 /* 2 frames, different sizes */
154 pkt->frame_count = 2;
155 pkt->vbr = 1;
156
157 /* read 1st frame size */
158 frame_bytes = xiph_lacing_16bit(&ptr, end);
159 if (frame_bytes < 0)
160 goto fail;
161
162 if (self_delimiting) {
163 int len = xiph_lacing_16bit(&ptr, end);
164 if (len < 0 || len + frame_bytes > end - ptr)
165 goto fail;
166 end = ptr + frame_bytes + len;
167 buf_size = end - buf;
168 }
169
170 pkt->frame_offset[0] = ptr - buf;
171 pkt->frame_size[0] = frame_bytes;
172
173 /* calculate 2nd frame size */
174 frame_bytes = end - ptr - pkt->frame_size[0];
175 if (frame_bytes < 0 || frame_bytes > MAX_FRAME_SIZE)
176 goto fail;
177 pkt->frame_offset[1] = pkt->frame_offset[0] + pkt->frame_size[0];
178 pkt->frame_size[1] = frame_bytes;
179 break;
180 case 3:
181 /* 1 to 48 frames, can be different sizes */
182 i = *ptr++;
183 pkt->frame_count = (i ) & 0x3F;
184 padding = (i >> 6) & 0x01;
185 pkt->vbr = (i >> 7) & 0x01;
186
187 if (pkt->frame_count == 0 || pkt->frame_count > MAX_FRAMES)
188 goto fail;
189
190 /* read padding size */
191 if (padding) {
192 padding = xiph_lacing_full(&ptr, end);
193 if (padding < 0)
194 goto fail;
195 }
196
197 /* read frame sizes */
198 if (pkt->vbr) {
199 /* for VBR, all frames except the final one have their size coded
200 in the bitstream. the last frame size is implicit. */
201 int total_bytes = 0;
202 for (i = 0; i < pkt->frame_count - 1; i++) {
203 frame_bytes = xiph_lacing_16bit(&ptr, end);
204 if (frame_bytes < 0)
205 goto fail;
206 pkt->frame_size[i] = frame_bytes;
207 total_bytes += frame_bytes;
208 }
209
210 if (self_delimiting) {
211 int len = xiph_lacing_16bit(&ptr, end);
212 if (len < 0 || len + total_bytes + padding > end - ptr)
213 goto fail;
214 end = ptr + total_bytes + len + padding;
215 buf_size = end - buf;
216 }
217
218 frame_bytes = end - ptr - padding;
219 if (total_bytes > frame_bytes)
220 goto fail;
221 pkt->frame_offset[0] = ptr - buf;
222 for (i = 1; i < pkt->frame_count; i++)
223 pkt->frame_offset[i] = pkt->frame_offset[i-1] + pkt->frame_size[i-1];
224 pkt->frame_size[pkt->frame_count-1] = frame_bytes - total_bytes;
225 } else {
226 /* for CBR, the remaining packet bytes are divided evenly between
227 the frames */
228 if (self_delimiting) {
229 frame_bytes = xiph_lacing_16bit(&ptr, end);
230 if (frame_bytes < 0 || pkt->frame_count * frame_bytes + padding > end - ptr)
231 goto fail;
232 end = ptr + pkt->frame_count * frame_bytes + padding;
233 buf_size = end - buf;
234 } else {
235 frame_bytes = end - ptr - padding;
236 if (frame_bytes % pkt->frame_count ||
237 frame_bytes / pkt->frame_count > MAX_FRAME_SIZE)
238 goto fail;
239 frame_bytes /= pkt->frame_count;
240 }
241
242 pkt->frame_offset[0] = ptr - buf;
243 pkt->frame_size[0] = frame_bytes;
244 for (i = 1; i < pkt->frame_count; i++) {
245 pkt->frame_offset[i] = pkt->frame_offset[i-1] + pkt->frame_size[i-1];
246 pkt->frame_size[i] = frame_bytes;
247 }
248 }
249 }
250
251 pkt->packet_size = buf_size;
252 pkt->data_size = pkt->packet_size - padding;
253
254 /* total packet duration cannot be larger than 120ms */
255 pkt->frame_duration = opus_frame_duration[pkt->config];
256 if (pkt->frame_duration * pkt->frame_count > MAX_PACKET_DUR)
257 goto fail;
258
259 /* set mode and bandwidth */
260 if (pkt->config < 12) {
261 pkt->mode = OPUS_MODE_SILK;
262 pkt->bandwidth = pkt->config >> 2;
263 } else if (pkt->config < 16) {
264 pkt->mode = OPUS_MODE_HYBRID;
265 pkt->bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND + (pkt->config >= 14);
266 } else {
267 pkt->mode = OPUS_MODE_CELT;
268 pkt->bandwidth = (pkt->config - 16) >> 2;
269 /* skip medium band */
270 if (pkt->bandwidth)
271 pkt->bandwidth++;
272 }
273
274 return 0;
275
276 fail:
277 memset(pkt, 0, sizeof(*pkt));
278 return AVERROR_INVALIDDATA;
279 }
280
channel_reorder_vorbis(int nb_channels,int channel_idx)281 static int channel_reorder_vorbis(int nb_channels, int channel_idx)
282 {
283 return ff_vorbis_channel_layout_offsets[nb_channels - 1][channel_idx];
284 }
285
channel_reorder_unknown(int nb_channels,int channel_idx)286 static int channel_reorder_unknown(int nb_channels, int channel_idx)
287 {
288 return channel_idx;
289 }
290
ff_opus_parse_extradata(AVCodecContext * avctx,OpusContext * s)291 av_cold int ff_opus_parse_extradata(AVCodecContext *avctx,
292 OpusContext *s)
293 {
294 static const uint8_t default_channel_map[2] = { 0, 1 };
295
296 int (*channel_reorder)(int, int) = channel_reorder_unknown;
297
298 const uint8_t *extradata, *channel_map;
299 int extradata_size;
300 int version, channels, map_type, streams, stereo_streams, i, j;
301 uint64_t layout;
302
303 if (!avctx->extradata) {
304 if (avctx->channels > 2) {
305 av_log(avctx, AV_LOG_ERROR,
306 "Multichannel configuration without extradata.\n");
307 return AVERROR(EINVAL);
308 }
309 extradata = opus_default_extradata;
310 extradata_size = sizeof(opus_default_extradata);
311 } else {
312 extradata = avctx->extradata;
313 extradata_size = avctx->extradata_size;
314 }
315
316 if (extradata_size < 19) {
317 av_log(avctx, AV_LOG_ERROR, "Invalid extradata size: %d\n",
318 extradata_size);
319 return AVERROR_INVALIDDATA;
320 }
321
322 version = extradata[8];
323 if (version > 15) {
324 avpriv_request_sample(avctx, "Extradata version %d", version);
325 return AVERROR_PATCHWELCOME;
326 }
327
328 avctx->delay = AV_RL16(extradata + 10);
329
330 channels = avctx->extradata ? extradata[9] : (avctx->channels == 1) ? 1 : 2;
331 if (!channels) {
332 av_log(avctx, AV_LOG_ERROR, "Zero channel count specified in the extradata\n");
333 return AVERROR_INVALIDDATA;
334 }
335
336 s->gain_i = AV_RL16(extradata + 16);
337 if (s->gain_i)
338 s->gain = ff_exp10(s->gain_i / (20.0 * 256));
339
340 map_type = extradata[18];
341 if (!map_type) {
342 if (channels > 2) {
343 av_log(avctx, AV_LOG_ERROR,
344 "Channel mapping 0 is only specified for up to 2 channels\n");
345 return AVERROR_INVALIDDATA;
346 }
347 layout = (channels == 1) ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
348 streams = 1;
349 stereo_streams = channels - 1;
350 channel_map = default_channel_map;
351 } else if (map_type == 1 || map_type == 2 || map_type == 255) {
352 if (extradata_size < 21 + channels) {
353 av_log(avctx, AV_LOG_ERROR, "Invalid extradata size: %d\n",
354 extradata_size);
355 return AVERROR_INVALIDDATA;
356 }
357
358 streams = extradata[19];
359 stereo_streams = extradata[20];
360 if (!streams || stereo_streams > streams ||
361 streams + stereo_streams > 255) {
362 av_log(avctx, AV_LOG_ERROR,
363 "Invalid stream/stereo stream count: %d/%d\n", streams, stereo_streams);
364 return AVERROR_INVALIDDATA;
365 }
366
367 if (map_type == 1) {
368 if (channels > 8) {
369 av_log(avctx, AV_LOG_ERROR,
370 "Channel mapping 1 is only specified for up to 8 channels\n");
371 return AVERROR_INVALIDDATA;
372 }
373 layout = ff_vorbis_channel_layouts[channels - 1];
374 channel_reorder = channel_reorder_vorbis;
375 } else if (map_type == 2) {
376 int ambisonic_order = ff_sqrt(channels) - 1;
377 if (channels != ((ambisonic_order + 1) * (ambisonic_order + 1)) &&
378 channels != ((ambisonic_order + 1) * (ambisonic_order + 1) + 2)) {
379 av_log(avctx, AV_LOG_ERROR,
380 "Channel mapping 2 is only specified for channel counts"
381 " which can be written as (n + 1)^2 or (n + 1)^2 + 2"
382 " for nonnegative integer n\n");
383 return AVERROR_INVALIDDATA;
384 }
385 if (channels > 227) {
386 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
387 return AVERROR_INVALIDDATA;
388 }
389 layout = 0;
390 } else
391 layout = 0;
392
393 channel_map = extradata + 21;
394 } else {
395 avpriv_request_sample(avctx, "Mapping type %d", map_type);
396 return AVERROR_PATCHWELCOME;
397 }
398
399 s->channel_maps = av_mallocz_array(channels, sizeof(*s->channel_maps));
400 if (!s->channel_maps)
401 return AVERROR(ENOMEM);
402
403 for (i = 0; i < channels; i++) {
404 ChannelMap *map = &s->channel_maps[i];
405 uint8_t idx = channel_map[channel_reorder(channels, i)];
406
407 if (idx == 255) {
408 map->silence = 1;
409 continue;
410 } else if (idx >= streams + stereo_streams) {
411 av_log(avctx, AV_LOG_ERROR,
412 "Invalid channel map for output channel %d: %d\n", i, idx);
413 av_freep(&s->channel_maps);
414 return AVERROR_INVALIDDATA;
415 }
416
417 /* check that we did not see this index yet */
418 map->copy = 0;
419 for (j = 0; j < i; j++)
420 if (channel_map[channel_reorder(channels, j)] == idx) {
421 map->copy = 1;
422 map->copy_idx = j;
423 break;
424 }
425
426 if (idx < 2 * stereo_streams) {
427 map->stream_idx = idx / 2;
428 map->channel_idx = idx & 1;
429 } else {
430 map->stream_idx = idx - stereo_streams;
431 map->channel_idx = 0;
432 }
433 }
434
435 avctx->channels = channels;
436 avctx->channel_layout = layout;
437 s->nb_streams = streams;
438 s->nb_stereo_streams = stereo_streams;
439
440 return 0;
441 }
442
ff_celt_quant_bands(CeltFrame * f,OpusRangeCoder * rc)443 void ff_celt_quant_bands(CeltFrame *f, OpusRangeCoder *rc)
444 {
445 float lowband_scratch[8 * 22];
446 float norm1[2 * 8 * 100];
447 float *norm2 = norm1 + 8 * 100;
448
449 int totalbits = (f->framebits << 3) - f->anticollapse_needed;
450
451 int update_lowband = 1;
452 int lowband_offset = 0;
453
454 int i, j;
455
456 for (i = f->start_band; i < f->end_band; i++) {
457 uint32_t cm[2] = { (1 << f->blocks) - 1, (1 << f->blocks) - 1 };
458 int band_offset = ff_celt_freq_bands[i] << f->size;
459 int band_size = ff_celt_freq_range[i] << f->size;
460 float *X = f->block[0].coeffs + band_offset;
461 float *Y = (f->channels == 2) ? f->block[1].coeffs + band_offset : NULL;
462 float *norm_loc1, *norm_loc2;
463
464 int consumed = opus_rc_tell_frac(rc);
465 int effective_lowband = -1;
466 int b = 0;
467
468 /* Compute how many bits we want to allocate to this band */
469 if (i != f->start_band)
470 f->remaining -= consumed;
471 f->remaining2 = totalbits - consumed - 1;
472 if (i <= f->coded_bands - 1) {
473 int curr_balance = f->remaining / FFMIN(3, f->coded_bands-i);
474 b = av_clip_uintp2(FFMIN(f->remaining2 + 1, f->pulses[i] + curr_balance), 14);
475 }
476
477 if ((ff_celt_freq_bands[i] - ff_celt_freq_range[i] >= ff_celt_freq_bands[f->start_band] ||
478 i == f->start_band + 1) && (update_lowband || lowband_offset == 0))
479 lowband_offset = i;
480
481 if (i == f->start_band + 1) {
482 /* Special Hybrid Folding (RFC 8251 section 9). Copy the first band into
483 the second to ensure the second band never has to use the LCG. */
484 int count = (ff_celt_freq_range[i] - ff_celt_freq_range[i-1]) << f->size;
485
486 memcpy(&norm1[band_offset], &norm1[band_offset - count], count * sizeof(float));
487
488 if (f->channels == 2)
489 memcpy(&norm2[band_offset], &norm2[band_offset - count], count * sizeof(float));
490 }
491
492 /* Get a conservative estimate of the collapse_mask's for the bands we're
493 going to be folding from. */
494 if (lowband_offset != 0 && (f->spread != CELT_SPREAD_AGGRESSIVE ||
495 f->blocks > 1 || f->tf_change[i] < 0)) {
496 int foldstart, foldend;
497
498 /* This ensures we never repeat spectral content within one band */
499 effective_lowband = FFMAX(ff_celt_freq_bands[f->start_band],
500 ff_celt_freq_bands[lowband_offset] - ff_celt_freq_range[i]);
501 foldstart = lowband_offset;
502 while (ff_celt_freq_bands[--foldstart] > effective_lowband);
503 foldend = lowband_offset - 1;
504 while (++foldend < i && ff_celt_freq_bands[foldend] < effective_lowband + ff_celt_freq_range[i]);
505
506 cm[0] = cm[1] = 0;
507 for (j = foldstart; j < foldend; j++) {
508 cm[0] |= f->block[0].collapse_masks[j];
509 cm[1] |= f->block[f->channels - 1].collapse_masks[j];
510 }
511 }
512
513 if (f->dual_stereo && i == f->intensity_stereo) {
514 /* Switch off dual stereo to do intensity */
515 f->dual_stereo = 0;
516 for (j = ff_celt_freq_bands[f->start_band] << f->size; j < band_offset; j++)
517 norm1[j] = (norm1[j] + norm2[j]) / 2;
518 }
519
520 norm_loc1 = effective_lowband != -1 ? norm1 + (effective_lowband << f->size) : NULL;
521 norm_loc2 = effective_lowband != -1 ? norm2 + (effective_lowband << f->size) : NULL;
522
523 if (f->dual_stereo) {
524 cm[0] = f->pvq->quant_band(f->pvq, f, rc, i, X, NULL, band_size, b >> 1,
525 f->blocks, norm_loc1, f->size,
526 norm1 + band_offset, 0, 1.0f,
527 lowband_scratch, cm[0]);
528
529 cm[1] = f->pvq->quant_band(f->pvq, f, rc, i, Y, NULL, band_size, b >> 1,
530 f->blocks, norm_loc2, f->size,
531 norm2 + band_offset, 0, 1.0f,
532 lowband_scratch, cm[1]);
533 } else {
534 cm[0] = f->pvq->quant_band(f->pvq, f, rc, i, X, Y, band_size, b >> 0,
535 f->blocks, norm_loc1, f->size,
536 norm1 + band_offset, 0, 1.0f,
537 lowband_scratch, cm[0] | cm[1]);
538 cm[1] = cm[0];
539 }
540
541 f->block[0].collapse_masks[i] = (uint8_t)cm[0];
542 f->block[f->channels - 1].collapse_masks[i] = (uint8_t)cm[1];
543 f->remaining += f->pulses[i] + consumed;
544
545 /* Update the folding position only as long as we have 1 bit/sample depth */
546 update_lowband = (b > band_size << 3);
547 }
548 }
549
550 #define NORMC(bits) ((bits) << (f->channels - 1) << f->size >> 2)
551
ff_celt_bitalloc(CeltFrame * f,OpusRangeCoder * rc,int encode)552 void ff_celt_bitalloc(CeltFrame *f, OpusRangeCoder *rc, int encode)
553 {
554 int i, j, low, high, total, done, bandbits, remaining, tbits_8ths;
555 int skip_startband = f->start_band;
556 int skip_bit = 0;
557 int intensitystereo_bit = 0;
558 int dualstereo_bit = 0;
559 int dynalloc = 6;
560 int extrabits = 0;
561
562 int boost[CELT_MAX_BANDS] = { 0 };
563 int trim_offset[CELT_MAX_BANDS];
564 int threshold[CELT_MAX_BANDS];
565 int bits1[CELT_MAX_BANDS];
566 int bits2[CELT_MAX_BANDS];
567
568 /* Spread */
569 if (opus_rc_tell(rc) + 4 <= f->framebits) {
570 if (encode)
571 ff_opus_rc_enc_cdf(rc, f->spread, ff_celt_model_spread);
572 else
573 f->spread = ff_opus_rc_dec_cdf(rc, ff_celt_model_spread);
574 } else {
575 f->spread = CELT_SPREAD_NORMAL;
576 }
577
578 /* Initialize static allocation caps */
579 for (i = 0; i < CELT_MAX_BANDS; i++)
580 f->caps[i] = NORMC((ff_celt_static_caps[f->size][f->channels - 1][i] + 64) * ff_celt_freq_range[i]);
581
582 /* Band boosts */
583 tbits_8ths = f->framebits << 3;
584 for (i = f->start_band; i < f->end_band; i++) {
585 int quanta = ff_celt_freq_range[i] << (f->channels - 1) << f->size;
586 int b_dynalloc = dynalloc;
587 int boost_amount = f->alloc_boost[i];
588 quanta = FFMIN(quanta << 3, FFMAX(6 << 3, quanta));
589
590 while (opus_rc_tell_frac(rc) + (b_dynalloc << 3) < tbits_8ths && boost[i] < f->caps[i]) {
591 int is_boost;
592 if (encode) {
593 is_boost = boost_amount--;
594 ff_opus_rc_enc_log(rc, is_boost, b_dynalloc);
595 } else {
596 is_boost = ff_opus_rc_dec_log(rc, b_dynalloc);
597 }
598
599 if (!is_boost)
600 break;
601
602 boost[i] += quanta;
603 tbits_8ths -= quanta;
604
605 b_dynalloc = 1;
606 }
607
608 if (boost[i])
609 dynalloc = FFMAX(dynalloc - 1, 2);
610 }
611
612 /* Allocation trim */
613 if (opus_rc_tell_frac(rc) + (6 << 3) <= tbits_8ths)
614 if (encode)
615 ff_opus_rc_enc_cdf(rc, f->alloc_trim, ff_celt_model_alloc_trim);
616 else
617 f->alloc_trim = ff_opus_rc_dec_cdf(rc, ff_celt_model_alloc_trim);
618
619 /* Anti-collapse bit reservation */
620 tbits_8ths = (f->framebits << 3) - opus_rc_tell_frac(rc) - 1;
621 f->anticollapse_needed = 0;
622 if (f->transient && f->size >= 2 && tbits_8ths >= ((f->size + 2) << 3))
623 f->anticollapse_needed = 1 << 3;
624 tbits_8ths -= f->anticollapse_needed;
625
626 /* Band skip bit reservation */
627 if (tbits_8ths >= 1 << 3)
628 skip_bit = 1 << 3;
629 tbits_8ths -= skip_bit;
630
631 /* Intensity/dual stereo bit reservation */
632 if (f->channels == 2) {
633 intensitystereo_bit = ff_celt_log2_frac[f->end_band - f->start_band];
634 if (intensitystereo_bit <= tbits_8ths) {
635 tbits_8ths -= intensitystereo_bit;
636 if (tbits_8ths >= 1 << 3) {
637 dualstereo_bit = 1 << 3;
638 tbits_8ths -= 1 << 3;
639 }
640 } else {
641 intensitystereo_bit = 0;
642 }
643 }
644
645 /* Trim offsets */
646 for (i = f->start_band; i < f->end_band; i++) {
647 int trim = f->alloc_trim - 5 - f->size;
648 int band = ff_celt_freq_range[i] * (f->end_band - i - 1);
649 int duration = f->size + 3;
650 int scale = duration + f->channels - 1;
651
652 /* PVQ minimum allocation threshold, below this value the band is
653 * skipped */
654 threshold[i] = FFMAX(3 * ff_celt_freq_range[i] << duration >> 4,
655 f->channels << 3);
656
657 trim_offset[i] = trim * (band << scale) >> 6;
658
659 if (ff_celt_freq_range[i] << f->size == 1)
660 trim_offset[i] -= f->channels << 3;
661 }
662
663 /* Bisection */
664 low = 1;
665 high = CELT_VECTORS - 1;
666 while (low <= high) {
667 int center = (low + high) >> 1;
668 done = total = 0;
669
670 for (i = f->end_band - 1; i >= f->start_band; i--) {
671 bandbits = NORMC(ff_celt_freq_range[i] * ff_celt_static_alloc[center][i]);
672
673 if (bandbits)
674 bandbits = FFMAX(bandbits + trim_offset[i], 0);
675 bandbits += boost[i];
676
677 if (bandbits >= threshold[i] || done) {
678 done = 1;
679 total += FFMIN(bandbits, f->caps[i]);
680 } else if (bandbits >= f->channels << 3) {
681 total += f->channels << 3;
682 }
683 }
684
685 if (total > tbits_8ths)
686 high = center - 1;
687 else
688 low = center + 1;
689 }
690 high = low--;
691
692 /* Bisection */
693 for (i = f->start_band; i < f->end_band; i++) {
694 bits1[i] = NORMC(ff_celt_freq_range[i] * ff_celt_static_alloc[low][i]);
695 bits2[i] = high >= CELT_VECTORS ? f->caps[i] :
696 NORMC(ff_celt_freq_range[i] * ff_celt_static_alloc[high][i]);
697
698 if (bits1[i])
699 bits1[i] = FFMAX(bits1[i] + trim_offset[i], 0);
700 if (bits2[i])
701 bits2[i] = FFMAX(bits2[i] + trim_offset[i], 0);
702
703 if (low)
704 bits1[i] += boost[i];
705 bits2[i] += boost[i];
706
707 if (boost[i])
708 skip_startband = i;
709 bits2[i] = FFMAX(bits2[i] - bits1[i], 0);
710 }
711
712 /* Bisection */
713 low = 0;
714 high = 1 << CELT_ALLOC_STEPS;
715 for (i = 0; i < CELT_ALLOC_STEPS; i++) {
716 int center = (low + high) >> 1;
717 done = total = 0;
718
719 for (j = f->end_band - 1; j >= f->start_band; j--) {
720 bandbits = bits1[j] + (center * bits2[j] >> CELT_ALLOC_STEPS);
721
722 if (bandbits >= threshold[j] || done) {
723 done = 1;
724 total += FFMIN(bandbits, f->caps[j]);
725 } else if (bandbits >= f->channels << 3)
726 total += f->channels << 3;
727 }
728 if (total > tbits_8ths)
729 high = center;
730 else
731 low = center;
732 }
733
734 /* Bisection */
735 done = total = 0;
736 for (i = f->end_band - 1; i >= f->start_band; i--) {
737 bandbits = bits1[i] + (low * bits2[i] >> CELT_ALLOC_STEPS);
738
739 if (bandbits >= threshold[i] || done)
740 done = 1;
741 else
742 bandbits = (bandbits >= f->channels << 3) ?
743 f->channels << 3 : 0;
744
745 bandbits = FFMIN(bandbits, f->caps[i]);
746 f->pulses[i] = bandbits;
747 total += bandbits;
748 }
749
750 /* Band skipping */
751 for (f->coded_bands = f->end_band; ; f->coded_bands--) {
752 int allocation;
753 j = f->coded_bands - 1;
754
755 if (j == skip_startband) {
756 /* all remaining bands are not skipped */
757 tbits_8ths += skip_bit;
758 break;
759 }
760
761 /* determine the number of bits available for coding "do not skip" markers */
762 remaining = tbits_8ths - total;
763 bandbits = remaining / (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[f->start_band]);
764 remaining -= bandbits * (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[f->start_band]);
765 allocation = f->pulses[j] + bandbits * ff_celt_freq_range[j];
766 allocation += FFMAX(remaining - (ff_celt_freq_bands[j] - ff_celt_freq_bands[f->start_band]), 0);
767
768 /* a "do not skip" marker is only coded if the allocation is
769 * above the chosen threshold */
770 if (allocation >= FFMAX(threshold[j], (f->channels + 1) << 3)) {
771 int do_not_skip;
772 if (encode) {
773 do_not_skip = f->coded_bands <= f->skip_band_floor;
774 ff_opus_rc_enc_log(rc, do_not_skip, 1);
775 } else {
776 do_not_skip = ff_opus_rc_dec_log(rc, 1);
777 }
778
779 if (do_not_skip)
780 break;
781
782 total += 1 << 3;
783 allocation -= 1 << 3;
784 }
785
786 /* the band is skipped, so reclaim its bits */
787 total -= f->pulses[j];
788 if (intensitystereo_bit) {
789 total -= intensitystereo_bit;
790 intensitystereo_bit = ff_celt_log2_frac[j - f->start_band];
791 total += intensitystereo_bit;
792 }
793
794 total += f->pulses[j] = (allocation >= f->channels << 3) ? f->channels << 3 : 0;
795 }
796
797 /* IS start band */
798 if (encode) {
799 if (intensitystereo_bit) {
800 f->intensity_stereo = FFMIN(f->intensity_stereo, f->coded_bands);
801 ff_opus_rc_enc_uint(rc, f->intensity_stereo, f->coded_bands + 1 - f->start_band);
802 }
803 } else {
804 f->intensity_stereo = f->dual_stereo = 0;
805 if (intensitystereo_bit)
806 f->intensity_stereo = f->start_band + ff_opus_rc_dec_uint(rc, f->coded_bands + 1 - f->start_band);
807 }
808
809 /* DS flag */
810 if (f->intensity_stereo <= f->start_band)
811 tbits_8ths += dualstereo_bit; /* no intensity stereo means no dual stereo */
812 else if (dualstereo_bit)
813 if (encode)
814 ff_opus_rc_enc_log(rc, f->dual_stereo, 1);
815 else
816 f->dual_stereo = ff_opus_rc_dec_log(rc, 1);
817
818 /* Supply the remaining bits in this frame to lower bands */
819 remaining = tbits_8ths - total;
820 bandbits = remaining / (ff_celt_freq_bands[f->coded_bands] - ff_celt_freq_bands[f->start_band]);
821 remaining -= bandbits * (ff_celt_freq_bands[f->coded_bands] - ff_celt_freq_bands[f->start_band]);
822 for (i = f->start_band; i < f->coded_bands; i++) {
823 const int bits = FFMIN(remaining, ff_celt_freq_range[i]);
824 f->pulses[i] += bits + bandbits * ff_celt_freq_range[i];
825 remaining -= bits;
826 }
827
828 /* Finally determine the allocation */
829 for (i = f->start_band; i < f->coded_bands; i++) {
830 int N = ff_celt_freq_range[i] << f->size;
831 int prev_extra = extrabits;
832 f->pulses[i] += extrabits;
833
834 if (N > 1) {
835 int dof; /* degrees of freedom */
836 int temp; /* dof * channels * log(dof) */
837 int fine_bits;
838 int max_bits;
839 int offset; /* fine energy quantization offset, i.e.
840 * extra bits assigned over the standard
841 * totalbits/dof */
842
843 extrabits = FFMAX(f->pulses[i] - f->caps[i], 0);
844 f->pulses[i] -= extrabits;
845
846 /* intensity stereo makes use of an extra degree of freedom */
847 dof = N * f->channels + (f->channels == 2 && N > 2 && !f->dual_stereo && i < f->intensity_stereo);
848 temp = dof * (ff_celt_log_freq_range[i] + (f->size << 3));
849 offset = (temp >> 1) - dof * CELT_FINE_OFFSET;
850 if (N == 2) /* dof=2 is the only case that doesn't fit the model */
851 offset += dof << 1;
852
853 /* grant an additional bias for the first and second pulses */
854 if (f->pulses[i] + offset < 2 * (dof << 3))
855 offset += temp >> 2;
856 else if (f->pulses[i] + offset < 3 * (dof << 3))
857 offset += temp >> 3;
858
859 fine_bits = (f->pulses[i] + offset + (dof << 2)) / (dof << 3);
860 max_bits = FFMIN((f->pulses[i] >> 3) >> (f->channels - 1), CELT_MAX_FINE_BITS);
861 max_bits = FFMAX(max_bits, 0);
862 f->fine_bits[i] = av_clip(fine_bits, 0, max_bits);
863
864 /* If fine_bits was rounded down or capped,
865 * give priority for the final fine energy pass */
866 f->fine_priority[i] = (f->fine_bits[i] * (dof << 3) >= f->pulses[i] + offset);
867
868 /* the remaining bits are assigned to PVQ */
869 f->pulses[i] -= f->fine_bits[i] << (f->channels - 1) << 3;
870 } else {
871 /* all bits go to fine energy except for the sign bit */
872 extrabits = FFMAX(f->pulses[i] - (f->channels << 3), 0);
873 f->pulses[i] -= extrabits;
874 f->fine_bits[i] = 0;
875 f->fine_priority[i] = 1;
876 }
877
878 /* hand back a limited number of extra fine energy bits to this band */
879 if (extrabits > 0) {
880 int fineextra = FFMIN(extrabits >> (f->channels + 2),
881 CELT_MAX_FINE_BITS - f->fine_bits[i]);
882 f->fine_bits[i] += fineextra;
883
884 fineextra <<= f->channels + 2;
885 f->fine_priority[i] = (fineextra >= extrabits - prev_extra);
886 extrabits -= fineextra;
887 }
888 }
889 f->remaining = extrabits;
890
891 /* skipped bands dedicate all of their bits for fine energy */
892 for (; i < f->end_band; i++) {
893 f->fine_bits[i] = f->pulses[i] >> (f->channels - 1) >> 3;
894 f->pulses[i] = 0;
895 f->fine_priority[i] = f->fine_bits[i] < 1;
896 }
897 }
898