Imported Debian version 2.4.3~trusty1
[deb_ffmpeg.git] / ffmpeg / libavcodec / vorbisenc.c
CommitLineData
2ba45a60
DM
1/*
2 * copyright (c) 2006 Oded Shimon <ods15@ods15.dyndns.org>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21/**
22 * @file
23 * Native Vorbis encoder.
24 * @author Oded Shimon <ods15@ods15.dyndns.org>
25 */
26
27#include <float.h>
28
29#include "avcodec.h"
30#include "internal.h"
31#include "fft.h"
32#include "mathops.h"
33#include "vorbis.h"
34#include "vorbis_enc_data.h"
35
36#define BITSTREAM_WRITER_LE
37#include "put_bits.h"
38
39#undef NDEBUG
40#include <assert.h>
41
42typedef struct {
43 int nentries;
44 uint8_t *lens;
45 uint32_t *codewords;
46 int ndimensions;
47 float min;
48 float delta;
49 int seq_p;
50 int lookup;
51 int *quantlist;
52 float *dimensions;
53 float *pow2;
54} vorbis_enc_codebook;
55
56typedef struct {
57 int dim;
58 int subclass;
59 int masterbook;
60 int *books;
61} vorbis_enc_floor_class;
62
63typedef struct {
64 int partitions;
65 int *partition_to_class;
66 int nclasses;
67 vorbis_enc_floor_class *classes;
68 int multiplier;
69 int rangebits;
70 int values;
71 vorbis_floor1_entry *list;
72} vorbis_enc_floor;
73
74typedef struct {
75 int type;
76 int begin;
77 int end;
78 int partition_size;
79 int classifications;
80 int classbook;
81 int8_t (*books)[8];
82 float (*maxes)[2];
83} vorbis_enc_residue;
84
85typedef struct {
86 int submaps;
87 int *mux;
88 int *floor;
89 int *residue;
90 int coupling_steps;
91 int *magnitude;
92 int *angle;
93} vorbis_enc_mapping;
94
95typedef struct {
96 int blockflag;
97 int mapping;
98} vorbis_enc_mode;
99
100typedef struct {
101 int channels;
102 int sample_rate;
103 int log2_blocksize[2];
104 FFTContext mdct[2];
105 const float *win[2];
106 int have_saved;
107 float *saved;
108 float *samples;
109 float *floor; // also used for tmp values for mdct
110 float *coeffs; // also used for residue after floor
111 float quality;
112
113 int ncodebooks;
114 vorbis_enc_codebook *codebooks;
115
116 int nfloors;
117 vorbis_enc_floor *floors;
118
119 int nresidues;
120 vorbis_enc_residue *residues;
121
122 int nmappings;
123 vorbis_enc_mapping *mappings;
124
125 int nmodes;
126 vorbis_enc_mode *modes;
127
128 int64_t next_pts;
129} vorbis_enc_context;
130
131#define MAX_CHANNELS 2
132#define MAX_CODEBOOK_DIM 8
133
134#define MAX_FLOOR_CLASS_DIM 4
135#define NUM_FLOOR_PARTITIONS 8
136#define MAX_FLOOR_VALUES (MAX_FLOOR_CLASS_DIM*NUM_FLOOR_PARTITIONS+2)
137
138#define RESIDUE_SIZE 1600
139#define RESIDUE_PART_SIZE 32
140#define NUM_RESIDUE_PARTITIONS (RESIDUE_SIZE/RESIDUE_PART_SIZE)
141
142static inline int put_codeword(PutBitContext *pb, vorbis_enc_codebook *cb,
143 int entry)
144{
145 av_assert2(entry >= 0);
146 av_assert2(entry < cb->nentries);
147 av_assert2(cb->lens[entry]);
148 if (pb->size_in_bits - put_bits_count(pb) < cb->lens[entry])
149 return AVERROR(EINVAL);
150 put_bits(pb, cb->lens[entry], cb->codewords[entry]);
151 return 0;
152}
153
154static int cb_lookup_vals(int lookup, int dimensions, int entries)
155{
156 if (lookup == 1)
157 return ff_vorbis_nth_root(entries, dimensions);
158 else if (lookup == 2)
159 return dimensions *entries;
160 return 0;
161}
162
163static int ready_codebook(vorbis_enc_codebook *cb)
164{
165 int i;
166
167 ff_vorbis_len2vlc(cb->lens, cb->codewords, cb->nentries);
168
169 if (!cb->lookup) {
170 cb->pow2 = cb->dimensions = NULL;
171 } else {
172 int vals = cb_lookup_vals(cb->lookup, cb->ndimensions, cb->nentries);
173 cb->dimensions = av_malloc_array(cb->nentries, sizeof(float) * cb->ndimensions);
174 cb->pow2 = av_mallocz_array(cb->nentries, sizeof(float));
175 if (!cb->dimensions || !cb->pow2)
176 return AVERROR(ENOMEM);
177 for (i = 0; i < cb->nentries; i++) {
178 float last = 0;
179 int j;
180 int div = 1;
181 for (j = 0; j < cb->ndimensions; j++) {
182 int off;
183 if (cb->lookup == 1)
184 off = (i / div) % vals; // lookup type 1
185 else
186 off = i * cb->ndimensions + j; // lookup type 2
187
188 cb->dimensions[i * cb->ndimensions + j] = last + cb->min + cb->quantlist[off] * cb->delta;
189 if (cb->seq_p)
190 last = cb->dimensions[i * cb->ndimensions + j];
191 cb->pow2[i] += cb->dimensions[i * cb->ndimensions + j] * cb->dimensions[i * cb->ndimensions + j];
192 div *= vals;
193 }
194 cb->pow2[i] /= 2.0;
195 }
196 }
197 return 0;
198}
199
200static int ready_residue(vorbis_enc_residue *rc, vorbis_enc_context *venc)
201{
202 int i;
203 av_assert0(rc->type == 2);
204 rc->maxes = av_mallocz_array(rc->classifications, sizeof(float[2]));
205 if (!rc->maxes)
206 return AVERROR(ENOMEM);
207 for (i = 0; i < rc->classifications; i++) {
208 int j;
209 vorbis_enc_codebook * cb;
210 for (j = 0; j < 8; j++)
211 if (rc->books[i][j] != -1)
212 break;
213 if (j == 8) // zero
214 continue;
215 cb = &venc->codebooks[rc->books[i][j]];
216 assert(cb->ndimensions >= 2);
217 assert(cb->lookup);
218
219 for (j = 0; j < cb->nentries; j++) {
220 float a;
221 if (!cb->lens[j])
222 continue;
223 a = fabs(cb->dimensions[j * cb->ndimensions]);
224 if (a > rc->maxes[i][0])
225 rc->maxes[i][0] = a;
226 a = fabs(cb->dimensions[j * cb->ndimensions + 1]);
227 if (a > rc->maxes[i][1])
228 rc->maxes[i][1] = a;
229 }
230 }
231 // small bias
232 for (i = 0; i < rc->classifications; i++) {
233 rc->maxes[i][0] += 0.8;
234 rc->maxes[i][1] += 0.8;
235 }
236 return 0;
237}
238
239static int create_vorbis_context(vorbis_enc_context *venc,
240 AVCodecContext *avctx)
241{
242 vorbis_enc_floor *fc;
243 vorbis_enc_residue *rc;
244 vorbis_enc_mapping *mc;
245 int i, book, ret;
246
247 venc->channels = avctx->channels;
248 venc->sample_rate = avctx->sample_rate;
249 venc->log2_blocksize[0] = venc->log2_blocksize[1] = 11;
250
251 venc->ncodebooks = FF_ARRAY_ELEMS(cvectors);
252 venc->codebooks = av_malloc(sizeof(vorbis_enc_codebook) * venc->ncodebooks);
253 if (!venc->codebooks)
254 return AVERROR(ENOMEM);
255
256 // codebook 0..14 - floor1 book, values 0..255
257 // codebook 15 residue masterbook
258 // codebook 16..29 residue
259 for (book = 0; book < venc->ncodebooks; book++) {
260 vorbis_enc_codebook *cb = &venc->codebooks[book];
261 int vals;
262 cb->ndimensions = cvectors[book].dim;
263 cb->nentries = cvectors[book].real_len;
264 cb->min = cvectors[book].min;
265 cb->delta = cvectors[book].delta;
266 cb->lookup = cvectors[book].lookup;
267 cb->seq_p = 0;
268
269 cb->lens = av_malloc_array(cb->nentries, sizeof(uint8_t));
270 cb->codewords = av_malloc_array(cb->nentries, sizeof(uint32_t));
271 if (!cb->lens || !cb->codewords)
272 return AVERROR(ENOMEM);
273 memcpy(cb->lens, cvectors[book].clens, cvectors[book].len);
274 memset(cb->lens + cvectors[book].len, 0, cb->nentries - cvectors[book].len);
275
276 if (cb->lookup) {
277 vals = cb_lookup_vals(cb->lookup, cb->ndimensions, cb->nentries);
278 cb->quantlist = av_malloc_array(vals, sizeof(int));
279 if (!cb->quantlist)
280 return AVERROR(ENOMEM);
281 for (i = 0; i < vals; i++)
282 cb->quantlist[i] = cvectors[book].quant[i];
283 } else {
284 cb->quantlist = NULL;
285 }
286 if ((ret = ready_codebook(cb)) < 0)
287 return ret;
288 }
289
290 venc->nfloors = 1;
291 venc->floors = av_malloc(sizeof(vorbis_enc_floor) * venc->nfloors);
292 if (!venc->floors)
293 return AVERROR(ENOMEM);
294
295 // just 1 floor
296 fc = &venc->floors[0];
297 fc->partitions = NUM_FLOOR_PARTITIONS;
298 fc->partition_to_class = av_malloc(sizeof(int) * fc->partitions);
299 if (!fc->partition_to_class)
300 return AVERROR(ENOMEM);
301 fc->nclasses = 0;
302 for (i = 0; i < fc->partitions; i++) {
303 static const int a[] = {0, 1, 2, 2, 3, 3, 4, 4};
304 fc->partition_to_class[i] = a[i];
305 fc->nclasses = FFMAX(fc->nclasses, fc->partition_to_class[i]);
306 }
307 fc->nclasses++;
308 fc->classes = av_malloc_array(fc->nclasses, sizeof(vorbis_enc_floor_class));
309 if (!fc->classes)
310 return AVERROR(ENOMEM);
311 for (i = 0; i < fc->nclasses; i++) {
312 vorbis_enc_floor_class * c = &fc->classes[i];
313 int j, books;
314 c->dim = floor_classes[i].dim;
315 c->subclass = floor_classes[i].subclass;
316 c->masterbook = floor_classes[i].masterbook;
317 books = (1 << c->subclass);
318 c->books = av_malloc_array(books, sizeof(int));
319 if (!c->books)
320 return AVERROR(ENOMEM);
321 for (j = 0; j < books; j++)
322 c->books[j] = floor_classes[i].nbooks[j];
323 }
324 fc->multiplier = 2;
325 fc->rangebits = venc->log2_blocksize[0] - 1;
326
327 fc->values = 2;
328 for (i = 0; i < fc->partitions; i++)
329 fc->values += fc->classes[fc->partition_to_class[i]].dim;
330
331 fc->list = av_malloc_array(fc->values, sizeof(vorbis_floor1_entry));
332 if (!fc->list)
333 return AVERROR(ENOMEM);
334 fc->list[0].x = 0;
335 fc->list[1].x = 1 << fc->rangebits;
336 for (i = 2; i < fc->values; i++) {
337 static const int a[] = {
338 93, 23,372, 6, 46,186,750, 14, 33, 65,
339 130,260,556, 3, 10, 18, 28, 39, 55, 79,
340 111,158,220,312,464,650,850
341 };
342 fc->list[i].x = a[i - 2];
343 }
344 if (ff_vorbis_ready_floor1_list(avctx, fc->list, fc->values))
345 return AVERROR_BUG;
346
347 venc->nresidues = 1;
348 venc->residues = av_malloc(sizeof(vorbis_enc_residue) * venc->nresidues);
349 if (!venc->residues)
350 return AVERROR(ENOMEM);
351
352 // single residue
353 rc = &venc->residues[0];
354 rc->type = 2;
355 rc->begin = 0;
356 rc->end = 1600;
357 rc->partition_size = 32;
358 rc->classifications = 10;
359 rc->classbook = 15;
360 rc->books = av_malloc(sizeof(*rc->books) * rc->classifications);
361 if (!rc->books)
362 return AVERROR(ENOMEM);
363 {
364 static const int8_t a[10][8] = {
365 { -1, -1, -1, -1, -1, -1, -1, -1, },
366 { -1, -1, 16, -1, -1, -1, -1, -1, },
367 { -1, -1, 17, -1, -1, -1, -1, -1, },
368 { -1, -1, 18, -1, -1, -1, -1, -1, },
369 { -1, -1, 19, -1, -1, -1, -1, -1, },
370 { -1, -1, 20, -1, -1, -1, -1, -1, },
371 { -1, -1, 21, -1, -1, -1, -1, -1, },
372 { 22, 23, -1, -1, -1, -1, -1, -1, },
373 { 24, 25, -1, -1, -1, -1, -1, -1, },
374 { 26, 27, 28, -1, -1, -1, -1, -1, },
375 };
376 memcpy(rc->books, a, sizeof a);
377 }
378 if ((ret = ready_residue(rc, venc)) < 0)
379 return ret;
380
381 venc->nmappings = 1;
382 venc->mappings = av_malloc(sizeof(vorbis_enc_mapping) * venc->nmappings);
383 if (!venc->mappings)
384 return AVERROR(ENOMEM);
385
386 // single mapping
387 mc = &venc->mappings[0];
388 mc->submaps = 1;
389 mc->mux = av_malloc(sizeof(int) * venc->channels);
390 if (!mc->mux)
391 return AVERROR(ENOMEM);
392 for (i = 0; i < venc->channels; i++)
393 mc->mux[i] = 0;
394 mc->floor = av_malloc(sizeof(int) * mc->submaps);
395 mc->residue = av_malloc(sizeof(int) * mc->submaps);
396 if (!mc->floor || !mc->residue)
397 return AVERROR(ENOMEM);
398 for (i = 0; i < mc->submaps; i++) {
399 mc->floor[i] = 0;
400 mc->residue[i] = 0;
401 }
402 mc->coupling_steps = venc->channels == 2 ? 1 : 0;
403 mc->magnitude = av_malloc(sizeof(int) * mc->coupling_steps);
404 mc->angle = av_malloc(sizeof(int) * mc->coupling_steps);
405 if (!mc->magnitude || !mc->angle)
406 return AVERROR(ENOMEM);
407 if (mc->coupling_steps) {
408 mc->magnitude[0] = 0;
409 mc->angle[0] = 1;
410 }
411
412 venc->nmodes = 1;
413 venc->modes = av_malloc(sizeof(vorbis_enc_mode) * venc->nmodes);
414 if (!venc->modes)
415 return AVERROR(ENOMEM);
416
417 // single mode
418 venc->modes[0].blockflag = 0;
419 venc->modes[0].mapping = 0;
420
421 venc->have_saved = 0;
422 venc->saved = av_malloc_array(sizeof(float) * venc->channels, (1 << venc->log2_blocksize[1]) / 2);
423 venc->samples = av_malloc_array(sizeof(float) * venc->channels, (1 << venc->log2_blocksize[1]));
424 venc->floor = av_malloc_array(sizeof(float) * venc->channels, (1 << venc->log2_blocksize[1]) / 2);
425 venc->coeffs = av_malloc_array(sizeof(float) * venc->channels, (1 << venc->log2_blocksize[1]) / 2);
426 if (!venc->saved || !venc->samples || !venc->floor || !venc->coeffs)
427 return AVERROR(ENOMEM);
428
429 venc->win[0] = ff_vorbis_vwin[venc->log2_blocksize[0] - 6];
430 venc->win[1] = ff_vorbis_vwin[venc->log2_blocksize[1] - 6];
431
432 if ((ret = ff_mdct_init(&venc->mdct[0], venc->log2_blocksize[0], 0, 1.0)) < 0)
433 return ret;
434 if ((ret = ff_mdct_init(&venc->mdct[1], venc->log2_blocksize[1], 0, 1.0)) < 0)
435 return ret;
436
437 return 0;
438}
439
440static void put_float(PutBitContext *pb, float f)
441{
442 int exp, mant;
443 uint32_t res = 0;
444 mant = (int)ldexp(frexp(f, &exp), 20);
445 exp += 788 - 20;
446 if (mant < 0) {
447 res |= (1U << 31);
448 mant = -mant;
449 }
450 res |= mant | (exp << 21);
451 put_bits32(pb, res);
452}
453
454static void put_codebook_header(PutBitContext *pb, vorbis_enc_codebook *cb)
455{
456 int i;
457 int ordered = 0;
458
459 put_bits(pb, 24, 0x564342); //magic
460 put_bits(pb, 16, cb->ndimensions);
461 put_bits(pb, 24, cb->nentries);
462
463 for (i = 1; i < cb->nentries; i++)
464 if (cb->lens[i] < cb->lens[i-1])
465 break;
466 if (i == cb->nentries)
467 ordered = 1;
468
469 put_bits(pb, 1, ordered);
470 if (ordered) {
471 int len = cb->lens[0];
472 put_bits(pb, 5, len - 1);
473 i = 0;
474 while (i < cb->nentries) {
475 int j;
476 for (j = 0; j+i < cb->nentries; j++)
477 if (cb->lens[j+i] != len)
478 break;
479 put_bits(pb, ilog(cb->nentries - i), j);
480 i += j;
481 len++;
482 }
483 } else {
484 int sparse = 0;
485 for (i = 0; i < cb->nentries; i++)
486 if (!cb->lens[i])
487 break;
488 if (i != cb->nentries)
489 sparse = 1;
490 put_bits(pb, 1, sparse);
491
492 for (i = 0; i < cb->nentries; i++) {
493 if (sparse)
494 put_bits(pb, 1, !!cb->lens[i]);
495 if (cb->lens[i])
496 put_bits(pb, 5, cb->lens[i] - 1);
497 }
498 }
499
500 put_bits(pb, 4, cb->lookup);
501 if (cb->lookup) {
502 int tmp = cb_lookup_vals(cb->lookup, cb->ndimensions, cb->nentries);
503 int bits = ilog(cb->quantlist[0]);
504
505 for (i = 1; i < tmp; i++)
506 bits = FFMAX(bits, ilog(cb->quantlist[i]));
507
508 put_float(pb, cb->min);
509 put_float(pb, cb->delta);
510
511 put_bits(pb, 4, bits - 1);
512 put_bits(pb, 1, cb->seq_p);
513
514 for (i = 0; i < tmp; i++)
515 put_bits(pb, bits, cb->quantlist[i]);
516 }
517}
518
519static void put_floor_header(PutBitContext *pb, vorbis_enc_floor *fc)
520{
521 int i;
522
523 put_bits(pb, 16, 1); // type, only floor1 is supported
524
525 put_bits(pb, 5, fc->partitions);
526
527 for (i = 0; i < fc->partitions; i++)
528 put_bits(pb, 4, fc->partition_to_class[i]);
529
530 for (i = 0; i < fc->nclasses; i++) {
531 int j, books;
532
533 put_bits(pb, 3, fc->classes[i].dim - 1);
534 put_bits(pb, 2, fc->classes[i].subclass);
535
536 if (fc->classes[i].subclass)
537 put_bits(pb, 8, fc->classes[i].masterbook);
538
539 books = (1 << fc->classes[i].subclass);
540
541 for (j = 0; j < books; j++)
542 put_bits(pb, 8, fc->classes[i].books[j] + 1);
543 }
544
545 put_bits(pb, 2, fc->multiplier - 1);
546 put_bits(pb, 4, fc->rangebits);
547
548 for (i = 2; i < fc->values; i++)
549 put_bits(pb, fc->rangebits, fc->list[i].x);
550}
551
552static void put_residue_header(PutBitContext *pb, vorbis_enc_residue *rc)
553{
554 int i;
555
556 put_bits(pb, 16, rc->type);
557
558 put_bits(pb, 24, rc->begin);
559 put_bits(pb, 24, rc->end);
560 put_bits(pb, 24, rc->partition_size - 1);
561 put_bits(pb, 6, rc->classifications - 1);
562 put_bits(pb, 8, rc->classbook);
563
564 for (i = 0; i < rc->classifications; i++) {
565 int j, tmp = 0;
566 for (j = 0; j < 8; j++)
567 tmp |= (rc->books[i][j] != -1) << j;
568
569 put_bits(pb, 3, tmp & 7);
570 put_bits(pb, 1, tmp > 7);
571
572 if (tmp > 7)
573 put_bits(pb, 5, tmp >> 3);
574 }
575
576 for (i = 0; i < rc->classifications; i++) {
577 int j;
578 for (j = 0; j < 8; j++)
579 if (rc->books[i][j] != -1)
580 put_bits(pb, 8, rc->books[i][j]);
581 }
582}
583
584static int put_main_header(vorbis_enc_context *venc, uint8_t **out)
585{
586 int i;
587 PutBitContext pb;
588 int len, hlens[3];
589 int buffer_len = 50000;
590 uint8_t *buffer = av_mallocz(buffer_len), *p = buffer;
591 if (!buffer)
592 return AVERROR(ENOMEM);
593
594 // identification header
595 init_put_bits(&pb, p, buffer_len);
596 put_bits(&pb, 8, 1); //magic
597 for (i = 0; "vorbis"[i]; i++)
598 put_bits(&pb, 8, "vorbis"[i]);
599 put_bits32(&pb, 0); // version
600 put_bits(&pb, 8, venc->channels);
601 put_bits32(&pb, venc->sample_rate);
602 put_bits32(&pb, 0); // bitrate
603 put_bits32(&pb, 0); // bitrate
604 put_bits32(&pb, 0); // bitrate
605 put_bits(&pb, 4, venc->log2_blocksize[0]);
606 put_bits(&pb, 4, venc->log2_blocksize[1]);
607 put_bits(&pb, 1, 1); // framing
608
609 flush_put_bits(&pb);
610 hlens[0] = put_bits_count(&pb) >> 3;
611 buffer_len -= hlens[0];
612 p += hlens[0];
613
614 // comment header
615 init_put_bits(&pb, p, buffer_len);
616 put_bits(&pb, 8, 3); //magic
617 for (i = 0; "vorbis"[i]; i++)
618 put_bits(&pb, 8, "vorbis"[i]);
619 put_bits32(&pb, 0); // vendor length TODO
620 put_bits32(&pb, 0); // amount of comments
621 put_bits(&pb, 1, 1); // framing
622
623 flush_put_bits(&pb);
624 hlens[1] = put_bits_count(&pb) >> 3;
625 buffer_len -= hlens[1];
626 p += hlens[1];
627
628 // setup header
629 init_put_bits(&pb, p, buffer_len);
630 put_bits(&pb, 8, 5); //magic
631 for (i = 0; "vorbis"[i]; i++)
632 put_bits(&pb, 8, "vorbis"[i]);
633
634 // codebooks
635 put_bits(&pb, 8, venc->ncodebooks - 1);
636 for (i = 0; i < venc->ncodebooks; i++)
637 put_codebook_header(&pb, &venc->codebooks[i]);
638
639 // time domain, reserved, zero
640 put_bits(&pb, 6, 0);
641 put_bits(&pb, 16, 0);
642
643 // floors
644 put_bits(&pb, 6, venc->nfloors - 1);
645 for (i = 0; i < venc->nfloors; i++)
646 put_floor_header(&pb, &venc->floors[i]);
647
648 // residues
649 put_bits(&pb, 6, venc->nresidues - 1);
650 for (i = 0; i < venc->nresidues; i++)
651 put_residue_header(&pb, &venc->residues[i]);
652
653 // mappings
654 put_bits(&pb, 6, venc->nmappings - 1);
655 for (i = 0; i < venc->nmappings; i++) {
656 vorbis_enc_mapping *mc = &venc->mappings[i];
657 int j;
658 put_bits(&pb, 16, 0); // mapping type
659
660 put_bits(&pb, 1, mc->submaps > 1);
661 if (mc->submaps > 1)
662 put_bits(&pb, 4, mc->submaps - 1);
663
664 put_bits(&pb, 1, !!mc->coupling_steps);
665 if (mc->coupling_steps) {
666 put_bits(&pb, 8, mc->coupling_steps - 1);
667 for (j = 0; j < mc->coupling_steps; j++) {
668 put_bits(&pb, ilog(venc->channels - 1), mc->magnitude[j]);
669 put_bits(&pb, ilog(venc->channels - 1), mc->angle[j]);
670 }
671 }
672
673 put_bits(&pb, 2, 0); // reserved
674
675 if (mc->submaps > 1)
676 for (j = 0; j < venc->channels; j++)
677 put_bits(&pb, 4, mc->mux[j]);
678
679 for (j = 0; j < mc->submaps; j++) {
680 put_bits(&pb, 8, 0); // reserved time configuration
681 put_bits(&pb, 8, mc->floor[j]);
682 put_bits(&pb, 8, mc->residue[j]);
683 }
684 }
685
686 // modes
687 put_bits(&pb, 6, venc->nmodes - 1);
688 for (i = 0; i < venc->nmodes; i++) {
689 put_bits(&pb, 1, venc->modes[i].blockflag);
690 put_bits(&pb, 16, 0); // reserved window type
691 put_bits(&pb, 16, 0); // reserved transform type
692 put_bits(&pb, 8, venc->modes[i].mapping);
693 }
694
695 put_bits(&pb, 1, 1); // framing
696
697 flush_put_bits(&pb);
698 hlens[2] = put_bits_count(&pb) >> 3;
699
700 len = hlens[0] + hlens[1] + hlens[2];
701 p = *out = av_mallocz(64 + len + len/255);
702 if (!p)
703 return AVERROR(ENOMEM);
704
705 *p++ = 2;
706 p += av_xiphlacing(p, hlens[0]);
707 p += av_xiphlacing(p, hlens[1]);
708 buffer_len = 0;
709 for (i = 0; i < 3; i++) {
710 memcpy(p, buffer + buffer_len, hlens[i]);
711 p += hlens[i];
712 buffer_len += hlens[i];
713 }
714
715 av_freep(&buffer);
716 return p - *out;
717}
718
719static float get_floor_average(vorbis_enc_floor * fc, float *coeffs, int i)
720{
721 int begin = fc->list[fc->list[FFMAX(i-1, 0)].sort].x;
722 int end = fc->list[fc->list[FFMIN(i+1, fc->values - 1)].sort].x;
723 int j;
724 float average = 0;
725
726 for (j = begin; j < end; j++)
727 average += fabs(coeffs[j]);
728 return average / (end - begin);
729}
730
731static void floor_fit(vorbis_enc_context *venc, vorbis_enc_floor *fc,
732 float *coeffs, uint16_t *posts, int samples)
733{
734 int range = 255 / fc->multiplier + 1;
735 int i;
736 float tot_average = 0.0;
737 float averages[MAX_FLOOR_VALUES];
738 for (i = 0; i < fc->values; i++) {
739 averages[i] = get_floor_average(fc, coeffs, i);
740 tot_average += averages[i];
741 }
742 tot_average /= fc->values;
743 tot_average /= venc->quality;
744
745 for (i = 0; i < fc->values; i++) {
746 int position = fc->list[fc->list[i].sort].x;
747 float average = averages[i];
748 int j;
749
750 average = sqrt(tot_average * average) * pow(1.25f, position*0.005f); // MAGIC!
751 for (j = 0; j < range - 1; j++)
752 if (ff_vorbis_floor1_inverse_db_table[j * fc->multiplier] > average)
753 break;
754 posts[fc->list[i].sort] = j;
755 }
756}
757
758static int render_point(int x0, int y0, int x1, int y1, int x)
759{
760 return y0 + (x - x0) * (y1 - y0) / (x1 - x0);
761}
762
763static int floor_encode(vorbis_enc_context *venc, vorbis_enc_floor *fc,
764 PutBitContext *pb, uint16_t *posts,
765 float *floor, int samples)
766{
767 int range = 255 / fc->multiplier + 1;
768 int coded[MAX_FLOOR_VALUES]; // first 2 values are unused
769 int i, counter;
770
771 if (pb->size_in_bits - put_bits_count(pb) < 1 + 2 * ilog(range - 1))
772 return AVERROR(EINVAL);
773 put_bits(pb, 1, 1); // non zero
774 put_bits(pb, ilog(range - 1), posts[0]);
775 put_bits(pb, ilog(range - 1), posts[1]);
776 coded[0] = coded[1] = 1;
777
778 for (i = 2; i < fc->values; i++) {
779 int predicted = render_point(fc->list[fc->list[i].low].x,
780 posts[fc->list[i].low],
781 fc->list[fc->list[i].high].x,
782 posts[fc->list[i].high],
783 fc->list[i].x);
784 int highroom = range - predicted;
785 int lowroom = predicted;
786 int room = FFMIN(highroom, lowroom);
787 if (predicted == posts[i]) {
788 coded[i] = 0; // must be used later as flag!
789 continue;
790 } else {
791 if (!coded[fc->list[i].low ])
792 coded[fc->list[i].low ] = -1;
793 if (!coded[fc->list[i].high])
794 coded[fc->list[i].high] = -1;
795 }
796 if (posts[i] > predicted) {
797 if (posts[i] - predicted > room)
798 coded[i] = posts[i] - predicted + lowroom;
799 else
800 coded[i] = (posts[i] - predicted) << 1;
801 } else {
802 if (predicted - posts[i] > room)
803 coded[i] = predicted - posts[i] + highroom - 1;
804 else
805 coded[i] = ((predicted - posts[i]) << 1) - 1;
806 }
807 }
808
809 counter = 2;
810 for (i = 0; i < fc->partitions; i++) {
811 vorbis_enc_floor_class * c = &fc->classes[fc->partition_to_class[i]];
812 int k, cval = 0, csub = 1<<c->subclass;
813 if (c->subclass) {
814 vorbis_enc_codebook * book = &venc->codebooks[c->masterbook];
815 int cshift = 0;
816 for (k = 0; k < c->dim; k++) {
817 int l;
818 for (l = 0; l < csub; l++) {
819 int maxval = 1;
820 if (c->books[l] != -1)
821 maxval = venc->codebooks[c->books[l]].nentries;
822 // coded could be -1, but this still works, cause that is 0
823 if (coded[counter + k] < maxval)
824 break;
825 }
826 assert(l != csub);
827 cval |= l << cshift;
828 cshift += c->subclass;
829 }
830 if (put_codeword(pb, book, cval))
831 return AVERROR(EINVAL);
832 }
833 for (k = 0; k < c->dim; k++) {
834 int book = c->books[cval & (csub-1)];
835 int entry = coded[counter++];
836 cval >>= c->subclass;
837 if (book == -1)
838 continue;
839 if (entry == -1)
840 entry = 0;
841 if (put_codeword(pb, &venc->codebooks[book], entry))
842 return AVERROR(EINVAL);
843 }
844 }
845
846 ff_vorbis_floor1_render_list(fc->list, fc->values, posts, coded,
847 fc->multiplier, floor, samples);
848
849 return 0;
850}
851
852static float *put_vector(vorbis_enc_codebook *book, PutBitContext *pb,
853 float *num)
854{
855 int i, entry = -1;
856 float distance = FLT_MAX;
857 assert(book->dimensions);
858 for (i = 0; i < book->nentries; i++) {
859 float * vec = book->dimensions + i * book->ndimensions, d = book->pow2[i];
860 int j;
861 if (!book->lens[i])
862 continue;
863 for (j = 0; j < book->ndimensions; j++)
864 d -= vec[j] * num[j];
865 if (distance > d) {
866 entry = i;
867 distance = d;
868 }
869 }
870 if (put_codeword(pb, book, entry))
871 return NULL;
872 return &book->dimensions[entry * book->ndimensions];
873}
874
875static int residue_encode(vorbis_enc_context *venc, vorbis_enc_residue *rc,
876 PutBitContext *pb, float *coeffs, int samples,
877 int real_ch)
878{
879 int pass, i, j, p, k;
880 int psize = rc->partition_size;
881 int partitions = (rc->end - rc->begin) / psize;
882 int channels = (rc->type == 2) ? 1 : real_ch;
883 int classes[MAX_CHANNELS][NUM_RESIDUE_PARTITIONS];
884 int classwords = venc->codebooks[rc->classbook].ndimensions;
885
886 av_assert0(rc->type == 2);
887 av_assert0(real_ch == 2);
888 for (p = 0; p < partitions; p++) {
889 float max1 = 0.0, max2 = 0.0;
890 int s = rc->begin + p * psize;
891 for (k = s; k < s + psize; k += 2) {
892 max1 = FFMAX(max1, fabs(coeffs[ k / real_ch]));
893 max2 = FFMAX(max2, fabs(coeffs[samples + k / real_ch]));
894 }
895
896 for (i = 0; i < rc->classifications - 1; i++)
897 if (max1 < rc->maxes[i][0] && max2 < rc->maxes[i][1])
898 break;
899 classes[0][p] = i;
900 }
901
902 for (pass = 0; pass < 8; pass++) {
903 p = 0;
904 while (p < partitions) {
905 if (pass == 0)
906 for (j = 0; j < channels; j++) {
907 vorbis_enc_codebook * book = &venc->codebooks[rc->classbook];
908 int entry = 0;
909 for (i = 0; i < classwords; i++) {
910 entry *= rc->classifications;
911 entry += classes[j][p + i];
912 }
913 if (put_codeword(pb, book, entry))
914 return AVERROR(EINVAL);
915 }
916 for (i = 0; i < classwords && p < partitions; i++, p++) {
917 for (j = 0; j < channels; j++) {
918 int nbook = rc->books[classes[j][p]][pass];
919 vorbis_enc_codebook * book = &venc->codebooks[nbook];
920 float *buf = coeffs + samples*j + rc->begin + p*psize;
921 if (nbook == -1)
922 continue;
923
924 assert(rc->type == 0 || rc->type == 2);
925 assert(!(psize % book->ndimensions));
926
927 if (rc->type == 0) {
928 for (k = 0; k < psize; k += book->ndimensions) {
929 int l;
930 float *a = put_vector(book, pb, &buf[k]);
931 if (!a)
932 return AVERROR(EINVAL);
933 for (l = 0; l < book->ndimensions; l++)
934 buf[k + l] -= a[l];
935 }
936 } else {
937 int s = rc->begin + p * psize, a1, b1;
938 a1 = (s % real_ch) * samples;
939 b1 = s / real_ch;
940 s = real_ch * samples;
941 for (k = 0; k < psize; k += book->ndimensions) {
942 int dim, a2 = a1, b2 = b1;
943 float vec[MAX_CODEBOOK_DIM], *pv = vec;
944 for (dim = book->ndimensions; dim--; ) {
945 *pv++ = coeffs[a2 + b2];
946 if ((a2 += samples) == s) {
947 a2 = 0;
948 b2++;
949 }
950 }
951 pv = put_vector(book, pb, vec);
952 if (!pv)
953 return AVERROR(EINVAL);
954 for (dim = book->ndimensions; dim--; ) {
955 coeffs[a1 + b1] -= *pv++;
956 if ((a1 += samples) == s) {
957 a1 = 0;
958 b1++;
959 }
960 }
961 }
962 }
963 }
964 }
965 }
966 }
967 return 0;
968}
969
970static int apply_window_and_mdct(vorbis_enc_context *venc,
971 float **audio, int samples)
972{
973 int i, channel;
974 const float * win = venc->win[0];
975 int window_len = 1 << (venc->log2_blocksize[0] - 1);
976 float n = (float)(1 << venc->log2_blocksize[0]) / 4.0;
977 // FIXME use dsp
978
979 if (!venc->have_saved && !samples)
980 return 0;
981
982 if (venc->have_saved) {
983 for (channel = 0; channel < venc->channels; channel++)
984 memcpy(venc->samples + channel * window_len * 2,
985 venc->saved + channel * window_len, sizeof(float) * window_len);
986 } else {
987 for (channel = 0; channel < venc->channels; channel++)
988 memset(venc->samples + channel * window_len * 2, 0,
989 sizeof(float) * window_len);
990 }
991
992 if (samples) {
993 for (channel = 0; channel < venc->channels; channel++) {
994 float * offset = venc->samples + channel*window_len*2 + window_len;
995 for (i = 0; i < samples; i++)
996 offset[i] = audio[channel][i] / n * win[window_len - i - 1];
997 }
998 } else {
999 for (channel = 0; channel < venc->channels; channel++)
1000 memset(venc->samples + channel * window_len * 2 + window_len,
1001 0, sizeof(float) * window_len);
1002 }
1003
1004 for (channel = 0; channel < venc->channels; channel++)
1005 venc->mdct[0].mdct_calc(&venc->mdct[0], venc->coeffs + channel * window_len,
1006 venc->samples + channel * window_len * 2);
1007
1008 if (samples) {
1009 for (channel = 0; channel < venc->channels; channel++) {
1010 float *offset = venc->saved + channel * window_len;
1011 for (i = 0; i < samples; i++)
1012 offset[i] = audio[channel][i] / n * win[i];
1013 }
1014 venc->have_saved = 1;
1015 } else {
1016 venc->have_saved = 0;
1017 }
1018 return 1;
1019}
1020
1021static int vorbis_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
1022 const AVFrame *frame, int *got_packet_ptr)
1023{
1024 vorbis_enc_context *venc = avctx->priv_data;
1025 float **audio = frame ? (float **)frame->extended_data : NULL;
1026 int samples = frame ? frame->nb_samples : 0;
1027 vorbis_enc_mode *mode;
1028 vorbis_enc_mapping *mapping;
1029 PutBitContext pb;
1030 int i, ret;
1031
1032 if (!apply_window_and_mdct(venc, audio, samples))
1033 return 0;
1034 samples = 1 << (venc->log2_blocksize[0] - 1);
1035
1036 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192)) < 0)
1037 return ret;
1038
1039 init_put_bits(&pb, avpkt->data, avpkt->size);
1040
1041 if (pb.size_in_bits - put_bits_count(&pb) < 1 + ilog(venc->nmodes - 1)) {
1042 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n");
1043 return AVERROR(EINVAL);
1044 }
1045
1046 put_bits(&pb, 1, 0); // magic bit
1047
1048 put_bits(&pb, ilog(venc->nmodes - 1), 0); // 0 bits, the mode
1049
1050 mode = &venc->modes[0];
1051 mapping = &venc->mappings[mode->mapping];
1052 if (mode->blockflag) {
1053 put_bits(&pb, 1, 0);
1054 put_bits(&pb, 1, 0);
1055 }
1056
1057 for (i = 0; i < venc->channels; i++) {
1058 vorbis_enc_floor *fc = &venc->floors[mapping->floor[mapping->mux[i]]];
1059 uint16_t posts[MAX_FLOOR_VALUES];
1060 floor_fit(venc, fc, &venc->coeffs[i * samples], posts, samples);
1061 if (floor_encode(venc, fc, &pb, posts, &venc->floor[i * samples], samples)) {
1062 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n");
1063 return AVERROR(EINVAL);
1064 }
1065 }
1066
1067 for (i = 0; i < venc->channels * samples; i++)
1068 venc->coeffs[i] /= venc->floor[i];
1069
1070 for (i = 0; i < mapping->coupling_steps; i++) {
1071 float *mag = venc->coeffs + mapping->magnitude[i] * samples;
1072 float *ang = venc->coeffs + mapping->angle[i] * samples;
1073 int j;
1074 for (j = 0; j < samples; j++) {
1075 float a = ang[j];
1076 ang[j] -= mag[j];
1077 if (mag[j] > 0)
1078 ang[j] = -ang[j];
1079 if (ang[j] < 0)
1080 mag[j] = a;
1081 }
1082 }
1083
1084 if (residue_encode(venc, &venc->residues[mapping->residue[mapping->mux[0]]],
1085 &pb, venc->coeffs, samples, venc->channels)) {
1086 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n");
1087 return AVERROR(EINVAL);
1088 }
1089
1090 flush_put_bits(&pb);
1091 avpkt->size = put_bits_count(&pb) >> 3;
1092
1093 avpkt->duration = ff_samples_to_time_base(avctx, avctx->frame_size);
1094 if (frame) {
1095 if (frame->pts != AV_NOPTS_VALUE)
1096 avpkt->pts = ff_samples_to_time_base(avctx, frame->pts);
1097 } else
1098 avpkt->pts = venc->next_pts;
1099 if (avpkt->pts != AV_NOPTS_VALUE)
1100 venc->next_pts = avpkt->pts + avpkt->duration;
1101
1102 *got_packet_ptr = 1;
1103 return 0;
1104}
1105
1106
1107static av_cold int vorbis_encode_close(AVCodecContext *avctx)
1108{
1109 vorbis_enc_context *venc = avctx->priv_data;
1110 int i;
1111
1112 if (venc->codebooks)
1113 for (i = 0; i < venc->ncodebooks; i++) {
1114 av_freep(&venc->codebooks[i].lens);
1115 av_freep(&venc->codebooks[i].codewords);
1116 av_freep(&venc->codebooks[i].quantlist);
1117 av_freep(&venc->codebooks[i].dimensions);
1118 av_freep(&venc->codebooks[i].pow2);
1119 }
1120 av_freep(&venc->codebooks);
1121
1122 if (venc->floors)
1123 for (i = 0; i < venc->nfloors; i++) {
1124 int j;
1125 if (venc->floors[i].classes)
1126 for (j = 0; j < venc->floors[i].nclasses; j++)
1127 av_freep(&venc->floors[i].classes[j].books);
1128 av_freep(&venc->floors[i].classes);
1129 av_freep(&venc->floors[i].partition_to_class);
1130 av_freep(&venc->floors[i].list);
1131 }
1132 av_freep(&venc->floors);
1133
1134 if (venc->residues)
1135 for (i = 0; i < venc->nresidues; i++) {
1136 av_freep(&venc->residues[i].books);
1137 av_freep(&venc->residues[i].maxes);
1138 }
1139 av_freep(&venc->residues);
1140
1141 if (venc->mappings)
1142 for (i = 0; i < venc->nmappings; i++) {
1143 av_freep(&venc->mappings[i].mux);
1144 av_freep(&venc->mappings[i].floor);
1145 av_freep(&venc->mappings[i].residue);
1146 av_freep(&venc->mappings[i].magnitude);
1147 av_freep(&venc->mappings[i].angle);
1148 }
1149 av_freep(&venc->mappings);
1150
1151 av_freep(&venc->modes);
1152
1153 av_freep(&venc->saved);
1154 av_freep(&venc->samples);
1155 av_freep(&venc->floor);
1156 av_freep(&venc->coeffs);
1157
1158 ff_mdct_end(&venc->mdct[0]);
1159 ff_mdct_end(&venc->mdct[1]);
1160
1161 av_freep(&avctx->extradata);
1162
1163 return 0 ;
1164}
1165
1166static av_cold int vorbis_encode_init(AVCodecContext *avctx)
1167{
1168 vorbis_enc_context *venc = avctx->priv_data;
1169 int ret;
1170
1171 if (avctx->channels != 2) {
1172 av_log(avctx, AV_LOG_ERROR, "Current FFmpeg Vorbis encoder only supports 2 channels.\n");
1173 return -1;
1174 }
1175
1176 if ((ret = create_vorbis_context(venc, avctx)) < 0)
1177 goto error;
1178
1179 avctx->bit_rate = 0;
1180 if (avctx->flags & CODEC_FLAG_QSCALE)
1181 venc->quality = avctx->global_quality / (float)FF_QP2LAMBDA;
1182 else
1183 venc->quality = 8;
1184 venc->quality *= venc->quality;
1185
1186 if ((ret = put_main_header(venc, (uint8_t**)&avctx->extradata)) < 0)
1187 goto error;
1188 avctx->extradata_size = ret;
1189
1190 avctx->frame_size = 1 << (venc->log2_blocksize[0] - 1);
1191
1192 return 0;
1193error:
1194 vorbis_encode_close(avctx);
1195 return ret;
1196}
1197
1198AVCodec ff_vorbis_encoder = {
1199 .name = "vorbis",
1200 .long_name = NULL_IF_CONFIG_SMALL("Vorbis"),
1201 .type = AVMEDIA_TYPE_AUDIO,
1202 .id = AV_CODEC_ID_VORBIS,
1203 .priv_data_size = sizeof(vorbis_enc_context),
1204 .init = vorbis_encode_init,
1205 .encode2 = vorbis_encode_frame,
1206 .close = vorbis_encode_close,
1207 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
1208 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
1209 AV_SAMPLE_FMT_NONE },
1210};