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(sizeof(float) * cb->nentries * cb->ndimensions); 174 cb->pow2 = av_mallocz(sizeof(float) * cb->nentries); 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(sizeof(float[2]) * rc->classifications); 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(sizeof(uint8_t) * cb->nentries); 270 cb->codewords = av_malloc(sizeof(uint32_t) * cb->nentries); 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(sizeof(int) * vals); 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(sizeof(vorbis_enc_floor_class) * fc->nclasses); 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(sizeof(int) * books); 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(sizeof(vorbis_floor1_entry) * fc->values); 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(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1]) / 2); 423 venc->samples = av_malloc(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1])); 424 venc->floor = av_malloc(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1]) / 2); 425 venc->coeffs = av_malloc(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 uint8_t buffer[50000] = {0}, *p = buffer; 589 int buffer_len = sizeof buffer; 590 int len, hlens[3]; 591 592 // identification header 593 init_put_bits(&pb, p, buffer_len); 594 put_bits(&pb, 8, 1); //magic 595 for (i = 0; "vorbis"[i]; i++) 596 put_bits(&pb, 8, "vorbis"[i]); 597 put_bits32(&pb, 0); // version 598 put_bits(&pb, 8, venc->channels); 599 put_bits32(&pb, venc->sample_rate); 600 put_bits32(&pb, 0); // bitrate 601 put_bits32(&pb, 0); // bitrate 602 put_bits32(&pb, 0); // bitrate 603 put_bits(&pb, 4, venc->log2_blocksize[0]); 604 put_bits(&pb, 4, venc->log2_blocksize[1]); 605 put_bits(&pb, 1, 1); // framing 606 607 flush_put_bits(&pb); 608 hlens[0] = put_bits_count(&pb) >> 3; 609 buffer_len -= hlens[0]; 610 p += hlens[0]; 611 612 // comment header 613 init_put_bits(&pb, p, buffer_len); 614 put_bits(&pb, 8, 3); //magic 615 for (i = 0; "vorbis"[i]; i++) 616 put_bits(&pb, 8, "vorbis"[i]); 617 put_bits32(&pb, 0); // vendor length TODO 618 put_bits32(&pb, 0); // amount of comments 619 put_bits(&pb, 1, 1); // framing 620 621 flush_put_bits(&pb); 622 hlens[1] = put_bits_count(&pb) >> 3; 623 buffer_len -= hlens[1]; 624 p += hlens[1]; 625 626 // setup header 627 init_put_bits(&pb, p, buffer_len); 628 put_bits(&pb, 8, 5); //magic 629 for (i = 0; "vorbis"[i]; i++) 630 put_bits(&pb, 8, "vorbis"[i]); 631 632 // codebooks 633 put_bits(&pb, 8, venc->ncodebooks - 1); 634 for (i = 0; i < venc->ncodebooks; i++) 635 put_codebook_header(&pb, &venc->codebooks[i]); 636 637 // time domain, reserved, zero 638 put_bits(&pb, 6, 0); 639 put_bits(&pb, 16, 0); 640 641 // floors 642 put_bits(&pb, 6, venc->nfloors - 1); 643 for (i = 0; i < venc->nfloors; i++) 644 put_floor_header(&pb, &venc->floors[i]); 645 646 // residues 647 put_bits(&pb, 6, venc->nresidues - 1); 648 for (i = 0; i < venc->nresidues; i++) 649 put_residue_header(&pb, &venc->residues[i]); 650 651 // mappings 652 put_bits(&pb, 6, venc->nmappings - 1); 653 for (i = 0; i < venc->nmappings; i++) { 654 vorbis_enc_mapping *mc = &venc->mappings[i]; 655 int j; 656 put_bits(&pb, 16, 0); // mapping type 657 658 put_bits(&pb, 1, mc->submaps > 1); 659 if (mc->submaps > 1) 660 put_bits(&pb, 4, mc->submaps - 1); 661 662 put_bits(&pb, 1, !!mc->coupling_steps); 663 if (mc->coupling_steps) { 664 put_bits(&pb, 8, mc->coupling_steps - 1); 665 for (j = 0; j < mc->coupling_steps; j++) { 666 put_bits(&pb, ilog(venc->channels - 1), mc->magnitude[j]); 667 put_bits(&pb, ilog(venc->channels - 1), mc->angle[j]); 668 } 669 } 670 671 put_bits(&pb, 2, 0); // reserved 672 673 if (mc->submaps > 1) 674 for (j = 0; j < venc->channels; j++) 675 put_bits(&pb, 4, mc->mux[j]); 676 677 for (j = 0; j < mc->submaps; j++) { 678 put_bits(&pb, 8, 0); // reserved time configuration 679 put_bits(&pb, 8, mc->floor[j]); 680 put_bits(&pb, 8, mc->residue[j]); 681 } 682 } 683 684 // modes 685 put_bits(&pb, 6, venc->nmodes - 1); 686 for (i = 0; i < venc->nmodes; i++) { 687 put_bits(&pb, 1, venc->modes[i].blockflag); 688 put_bits(&pb, 16, 0); // reserved window type 689 put_bits(&pb, 16, 0); // reserved transform type 690 put_bits(&pb, 8, venc->modes[i].mapping); 691 } 692 693 put_bits(&pb, 1, 1); // framing 694 695 flush_put_bits(&pb); 696 hlens[2] = put_bits_count(&pb) >> 3; 697 698 len = hlens[0] + hlens[1] + hlens[2]; 699 p = *out = av_mallocz(64 + len + len/255); 700 if (!p) 701 return AVERROR(ENOMEM); 702 703 *p++ = 2; 704 p += av_xiphlacing(p, hlens[0]); 705 p += av_xiphlacing(p, hlens[1]); 706 buffer_len = 0; 707 for (i = 0; i < 3; i++) { 708 memcpy(p, buffer + buffer_len, hlens[i]); 709 p += hlens[i]; 710 buffer_len += hlens[i]; 711 } 712 713 return p - *out; 714} 715 716static float get_floor_average(vorbis_enc_floor * fc, float *coeffs, int i) 717{ 718 int begin = fc->list[fc->list[FFMAX(i-1, 0)].sort].x; 719 int end = fc->list[fc->list[FFMIN(i+1, fc->values - 1)].sort].x; 720 int j; 721 float average = 0; 722 723 for (j = begin; j < end; j++) 724 average += fabs(coeffs[j]); 725 return average / (end - begin); 726} 727 728static void floor_fit(vorbis_enc_context *venc, vorbis_enc_floor *fc, 729 float *coeffs, uint16_t *posts, int samples) 730{ 731 int range = 255 / fc->multiplier + 1; 732 int i; 733 float tot_average = 0.0; 734 float averages[MAX_FLOOR_VALUES]; 735 for (i = 0; i < fc->values; i++) { 736 averages[i] = get_floor_average(fc, coeffs, i); 737 tot_average += averages[i]; 738 } 739 tot_average /= fc->values; 740 tot_average /= venc->quality; 741 742 for (i = 0; i < fc->values; i++) { 743 int position = fc->list[fc->list[i].sort].x; 744 float average = averages[i]; 745 int j; 746 747 average = sqrt(tot_average * average) * pow(1.25f, position*0.005f); // MAGIC! 748 for (j = 0; j < range - 1; j++) 749 if (ff_vorbis_floor1_inverse_db_table[j * fc->multiplier] > average) 750 break; 751 posts[fc->list[i].sort] = j; 752 } 753} 754 755static int render_point(int x0, int y0, int x1, int y1, int x) 756{ 757 return y0 + (x - x0) * (y1 - y0) / (x1 - x0); 758} 759 760static int floor_encode(vorbis_enc_context *venc, vorbis_enc_floor *fc, 761 PutBitContext *pb, uint16_t *posts, 762 float *floor, int samples) 763{ 764 int range = 255 / fc->multiplier + 1; 765 int coded[MAX_FLOOR_VALUES]; // first 2 values are unused 766 int i, counter; 767 768 if (pb->size_in_bits - put_bits_count(pb) < 1 + 2 * ilog(range - 1)) 769 return AVERROR(EINVAL); 770 put_bits(pb, 1, 1); // non zero 771 put_bits(pb, ilog(range - 1), posts[0]); 772 put_bits(pb, ilog(range - 1), posts[1]); 773 coded[0] = coded[1] = 1; 774 775 for (i = 2; i < fc->values; i++) { 776 int predicted = render_point(fc->list[fc->list[i].low].x, 777 posts[fc->list[i].low], 778 fc->list[fc->list[i].high].x, 779 posts[fc->list[i].high], 780 fc->list[i].x); 781 int highroom = range - predicted; 782 int lowroom = predicted; 783 int room = FFMIN(highroom, lowroom); 784 if (predicted == posts[i]) { 785 coded[i] = 0; // must be used later as flag! 786 continue; 787 } else { 788 if (!coded[fc->list[i].low ]) 789 coded[fc->list[i].low ] = -1; 790 if (!coded[fc->list[i].high]) 791 coded[fc->list[i].high] = -1; 792 } 793 if (posts[i] > predicted) { 794 if (posts[i] - predicted > room) 795 coded[i] = posts[i] - predicted + lowroom; 796 else 797 coded[i] = (posts[i] - predicted) << 1; 798 } else { 799 if (predicted - posts[i] > room) 800 coded[i] = predicted - posts[i] + highroom - 1; 801 else 802 coded[i] = ((predicted - posts[i]) << 1) - 1; 803 } 804 } 805 806 counter = 2; 807 for (i = 0; i < fc->partitions; i++) { 808 vorbis_enc_floor_class * c = &fc->classes[fc->partition_to_class[i]]; 809 int k, cval = 0, csub = 1<<c->subclass; 810 if (c->subclass) { 811 vorbis_enc_codebook * book = &venc->codebooks[c->masterbook]; 812 int cshift = 0; 813 for (k = 0; k < c->dim; k++) { 814 int l; 815 for (l = 0; l < csub; l++) { 816 int maxval = 1; 817 if (c->books[l] != -1) 818 maxval = venc->codebooks[c->books[l]].nentries; 819 // coded could be -1, but this still works, cause that is 0 820 if (coded[counter + k] < maxval) 821 break; 822 } 823 assert(l != csub); 824 cval |= l << cshift; 825 cshift += c->subclass; 826 } 827 if (put_codeword(pb, book, cval)) 828 return AVERROR(EINVAL); 829 } 830 for (k = 0; k < c->dim; k++) { 831 int book = c->books[cval & (csub-1)]; 832 int entry = coded[counter++]; 833 cval >>= c->subclass; 834 if (book == -1) 835 continue; 836 if (entry == -1) 837 entry = 0; 838 if (put_codeword(pb, &venc->codebooks[book], entry)) 839 return AVERROR(EINVAL); 840 } 841 } 842 843 ff_vorbis_floor1_render_list(fc->list, fc->values, posts, coded, 844 fc->multiplier, floor, samples); 845 846 return 0; 847} 848 849static float *put_vector(vorbis_enc_codebook *book, PutBitContext *pb, 850 float *num) 851{ 852 int i, entry = -1; 853 float distance = FLT_MAX; 854 assert(book->dimensions); 855 for (i = 0; i < book->nentries; i++) { 856 float * vec = book->dimensions + i * book->ndimensions, d = book->pow2[i]; 857 int j; 858 if (!book->lens[i]) 859 continue; 860 for (j = 0; j < book->ndimensions; j++) 861 d -= vec[j] * num[j]; 862 if (distance > d) { 863 entry = i; 864 distance = d; 865 } 866 } 867 if (put_codeword(pb, book, entry)) 868 return NULL; 869 return &book->dimensions[entry * book->ndimensions]; 870} 871 872static int residue_encode(vorbis_enc_context *venc, vorbis_enc_residue *rc, 873 PutBitContext *pb, float *coeffs, int samples, 874 int real_ch) 875{ 876 int pass, i, j, p, k; 877 int psize = rc->partition_size; 878 int partitions = (rc->end - rc->begin) / psize; 879 int channels = (rc->type == 2) ? 1 : real_ch; 880 int classes[MAX_CHANNELS][NUM_RESIDUE_PARTITIONS]; 881 int classwords = venc->codebooks[rc->classbook].ndimensions; 882 883 av_assert0(rc->type == 2); 884 av_assert0(real_ch == 2); 885 for (p = 0; p < partitions; p++) { 886 float max1 = 0.0, max2 = 0.0; 887 int s = rc->begin + p * psize; 888 for (k = s; k < s + psize; k += 2) { 889 max1 = FFMAX(max1, fabs(coeffs[ k / real_ch])); 890 max2 = FFMAX(max2, fabs(coeffs[samples + k / real_ch])); 891 } 892 893 for (i = 0; i < rc->classifications - 1; i++) 894 if (max1 < rc->maxes[i][0] && max2 < rc->maxes[i][1]) 895 break; 896 classes[0][p] = i; 897 } 898 899 for (pass = 0; pass < 8; pass++) { 900 p = 0; 901 while (p < partitions) { 902 if (pass == 0) 903 for (j = 0; j < channels; j++) { 904 vorbis_enc_codebook * book = &venc->codebooks[rc->classbook]; 905 int entry = 0; 906 for (i = 0; i < classwords; i++) { 907 entry *= rc->classifications; 908 entry += classes[j][p + i]; 909 } 910 if (put_codeword(pb, book, entry)) 911 return AVERROR(EINVAL); 912 } 913 for (i = 0; i < classwords && p < partitions; i++, p++) { 914 for (j = 0; j < channels; j++) { 915 int nbook = rc->books[classes[j][p]][pass]; 916 vorbis_enc_codebook * book = &venc->codebooks[nbook]; 917 float *buf = coeffs + samples*j + rc->begin + p*psize; 918 if (nbook == -1) 919 continue; 920 921 assert(rc->type == 0 || rc->type == 2); 922 assert(!(psize % book->ndimensions)); 923 924 if (rc->type == 0) { 925 for (k = 0; k < psize; k += book->ndimensions) { 926 int l; 927 float *a = put_vector(book, pb, &buf[k]); 928 if (!a) 929 return AVERROR(EINVAL); 930 for (l = 0; l < book->ndimensions; l++) 931 buf[k + l] -= a[l]; 932 } 933 } else { 934 int s = rc->begin + p * psize, a1, b1; 935 a1 = (s % real_ch) * samples; 936 b1 = s / real_ch; 937 s = real_ch * samples; 938 for (k = 0; k < psize; k += book->ndimensions) { 939 int dim, a2 = a1, b2 = b1; 940 float vec[MAX_CODEBOOK_DIM], *pv = vec; 941 for (dim = book->ndimensions; dim--; ) { 942 *pv++ = coeffs[a2 + b2]; 943 if ((a2 += samples) == s) { 944 a2 = 0; 945 b2++; 946 } 947 } 948 pv = put_vector(book, pb, vec); 949 if (!pv) 950 return AVERROR(EINVAL); 951 for (dim = book->ndimensions; dim--; ) { 952 coeffs[a1 + b1] -= *pv++; 953 if ((a1 += samples) == s) { 954 a1 = 0; 955 b1++; 956 } 957 } 958 } 959 } 960 } 961 } 962 } 963 } 964 return 0; 965} 966 967static int apply_window_and_mdct(vorbis_enc_context *venc, 968 float **audio, int samples) 969{ 970 int i, channel; 971 const float * win = venc->win[0]; 972 int window_len = 1 << (venc->log2_blocksize[0] - 1); 973 float n = (float)(1 << venc->log2_blocksize[0]) / 4.0; 974 // FIXME use dsp 975 976 if (!venc->have_saved && !samples) 977 return 0; 978 979 if (venc->have_saved) { 980 for (channel = 0; channel < venc->channels; channel++) 981 memcpy(venc->samples + channel * window_len * 2, 982 venc->saved + channel * window_len, sizeof(float) * window_len); 983 } else { 984 for (channel = 0; channel < venc->channels; channel++) 985 memset(venc->samples + channel * window_len * 2, 0, 986 sizeof(float) * window_len); 987 } 988 989 if (samples) { 990 for (channel = 0; channel < venc->channels; channel++) { 991 float * offset = venc->samples + channel*window_len*2 + window_len; 992 for (i = 0; i < samples; i++) 993 offset[i] = audio[channel][i] / n * win[window_len - i - 1]; 994 } 995 } else { 996 for (channel = 0; channel < venc->channels; channel++) 997 memset(venc->samples + channel * window_len * 2 + window_len, 998 0, sizeof(float) * window_len); 999 } 1000 1001 for (channel = 0; channel < venc->channels; channel++) 1002 venc->mdct[0].mdct_calc(&venc->mdct[0], venc->coeffs + channel * window_len, 1003 venc->samples + channel * window_len * 2); 1004 1005 if (samples) { 1006 for (channel = 0; channel < venc->channels; channel++) { 1007 float *offset = venc->saved + channel * window_len; 1008 for (i = 0; i < samples; i++) 1009 offset[i] = audio[channel][i] / n * win[i]; 1010 } 1011 venc->have_saved = 1; 1012 } else { 1013 venc->have_saved = 0; 1014 } 1015 return 1; 1016} 1017 1018static int vorbis_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, 1019 const AVFrame *frame, int *got_packet_ptr) 1020{ 1021 vorbis_enc_context *venc = avctx->priv_data; 1022 float **audio = frame ? (float **)frame->extended_data : NULL; 1023 int samples = frame ? frame->nb_samples : 0; 1024 vorbis_enc_mode *mode; 1025 vorbis_enc_mapping *mapping; 1026 PutBitContext pb; 1027 int i, ret; 1028 1029 if (!apply_window_and_mdct(venc, audio, samples)) 1030 return 0; 1031 samples = 1 << (venc->log2_blocksize[0] - 1); 1032 1033 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192)) < 0) 1034 return ret; 1035 1036 init_put_bits(&pb, avpkt->data, avpkt->size); 1037 1038 if (pb.size_in_bits - put_bits_count(&pb) < 1 + ilog(venc->nmodes - 1)) { 1039 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n"); 1040 return AVERROR(EINVAL); 1041 } 1042 1043 put_bits(&pb, 1, 0); // magic bit 1044 1045 put_bits(&pb, ilog(venc->nmodes - 1), 0); // 0 bits, the mode 1046 1047 mode = &venc->modes[0]; 1048 mapping = &venc->mappings[mode->mapping]; 1049 if (mode->blockflag) { 1050 put_bits(&pb, 1, 0); 1051 put_bits(&pb, 1, 0); 1052 } 1053 1054 for (i = 0; i < venc->channels; i++) { 1055 vorbis_enc_floor *fc = &venc->floors[mapping->floor[mapping->mux[i]]]; 1056 uint16_t posts[MAX_FLOOR_VALUES]; 1057 floor_fit(venc, fc, &venc->coeffs[i * samples], posts, samples); 1058 if (floor_encode(venc, fc, &pb, posts, &venc->floor[i * samples], samples)) { 1059 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n"); 1060 return AVERROR(EINVAL); 1061 } 1062 } 1063 1064 for (i = 0; i < venc->channels * samples; i++) 1065 venc->coeffs[i] /= venc->floor[i]; 1066 1067 for (i = 0; i < mapping->coupling_steps; i++) { 1068 float *mag = venc->coeffs + mapping->magnitude[i] * samples; 1069 float *ang = venc->coeffs + mapping->angle[i] * samples; 1070 int j; 1071 for (j = 0; j < samples; j++) { 1072 float a = ang[j]; 1073 ang[j] -= mag[j]; 1074 if (mag[j] > 0) 1075 ang[j] = -ang[j]; 1076 if (ang[j] < 0) 1077 mag[j] = a; 1078 } 1079 } 1080 1081 if (residue_encode(venc, &venc->residues[mapping->residue[mapping->mux[0]]], 1082 &pb, venc->coeffs, samples, venc->channels)) { 1083 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n"); 1084 return AVERROR(EINVAL); 1085 } 1086 1087 flush_put_bits(&pb); 1088 avpkt->size = put_bits_count(&pb) >> 3; 1089 1090 avpkt->duration = ff_samples_to_time_base(avctx, avctx->frame_size); 1091 if (frame) { 1092 if (frame->pts != AV_NOPTS_VALUE) 1093 avpkt->pts = ff_samples_to_time_base(avctx, frame->pts); 1094 } else 1095 avpkt->pts = venc->next_pts; 1096 if (avpkt->pts != AV_NOPTS_VALUE) 1097 venc->next_pts = avpkt->pts + avpkt->duration; 1098 1099 *got_packet_ptr = 1; 1100 return 0; 1101} 1102 1103 1104static av_cold int vorbis_encode_close(AVCodecContext *avctx) 1105{ 1106 vorbis_enc_context *venc = avctx->priv_data; 1107 int i; 1108 1109 if (venc->codebooks) 1110 for (i = 0; i < venc->ncodebooks; i++) { 1111 av_freep(&venc->codebooks[i].lens); 1112 av_freep(&venc->codebooks[i].codewords); 1113 av_freep(&venc->codebooks[i].quantlist); 1114 av_freep(&venc->codebooks[i].dimensions); 1115 av_freep(&venc->codebooks[i].pow2); 1116 } 1117 av_freep(&venc->codebooks); 1118 1119 if (venc->floors) 1120 for (i = 0; i < venc->nfloors; i++) { 1121 int j; 1122 if (venc->floors[i].classes) 1123 for (j = 0; j < venc->floors[i].nclasses; j++) 1124 av_freep(&venc->floors[i].classes[j].books); 1125 av_freep(&venc->floors[i].classes); 1126 av_freep(&venc->floors[i].partition_to_class); 1127 av_freep(&venc->floors[i].list); 1128 } 1129 av_freep(&venc->floors); 1130 1131 if (venc->residues) 1132 for (i = 0; i < venc->nresidues; i++) { 1133 av_freep(&venc->residues[i].books); 1134 av_freep(&venc->residues[i].maxes); 1135 } 1136 av_freep(&venc->residues); 1137 1138 if (venc->mappings) 1139 for (i = 0; i < venc->nmappings; i++) { 1140 av_freep(&venc->mappings[i].mux); 1141 av_freep(&venc->mappings[i].floor); 1142 av_freep(&venc->mappings[i].residue); 1143 av_freep(&venc->mappings[i].magnitude); 1144 av_freep(&venc->mappings[i].angle); 1145 } 1146 av_freep(&venc->mappings); 1147 1148 av_freep(&venc->modes); 1149 1150 av_freep(&venc->saved); 1151 av_freep(&venc->samples); 1152 av_freep(&venc->floor); 1153 av_freep(&venc->coeffs); 1154 1155 ff_mdct_end(&venc->mdct[0]); 1156 ff_mdct_end(&venc->mdct[1]); 1157 1158 av_freep(&avctx->extradata); 1159 1160 return 0 ; 1161} 1162 1163static av_cold int vorbis_encode_init(AVCodecContext *avctx) 1164{ 1165 vorbis_enc_context *venc = avctx->priv_data; 1166 int ret; 1167 1168 if (avctx->channels != 2) { 1169 av_log(avctx, AV_LOG_ERROR, "Current FFmpeg Vorbis encoder only supports 2 channels.\n"); 1170 return -1; 1171 } 1172 1173 if ((ret = create_vorbis_context(venc, avctx)) < 0) 1174 goto error; 1175 1176 avctx->bit_rate = 0; 1177 if (avctx->flags & CODEC_FLAG_QSCALE) 1178 venc->quality = avctx->global_quality / (float)FF_QP2LAMBDA; 1179 else 1180 venc->quality = 8; 1181 venc->quality *= venc->quality; 1182 1183 if ((ret = put_main_header(venc, (uint8_t**)&avctx->extradata)) < 0) 1184 goto error; 1185 avctx->extradata_size = ret; 1186 1187 avctx->frame_size = 1 << (venc->log2_blocksize[0] - 1); 1188 1189 return 0; 1190error: 1191 vorbis_encode_close(avctx); 1192 return ret; 1193} 1194 1195AVCodec ff_vorbis_encoder = { 1196 .name = "vorbis", 1197 .long_name = NULL_IF_CONFIG_SMALL("Vorbis"), 1198 .type = AVMEDIA_TYPE_AUDIO, 1199 .id = AV_CODEC_ID_VORBIS, 1200 .priv_data_size = sizeof(vorbis_enc_context), 1201 .init = vorbis_encode_init, 1202 .encode2 = vorbis_encode_frame, 1203 .close = vorbis_encode_close, 1204 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL, 1205 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP, 1206 AV_SAMPLE_FMT_NONE }, 1207}; 1208