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#include "avcodec.h" 29#include "dsputil.h" 30#include "fft.h" 31#include "vorbis.h" 32#include "vorbis_enc_data.h" 33 34#define BITSTREAM_WRITER_LE 35#include "put_bits.h" 36 37#undef NDEBUG 38#include <assert.h> 39 40typedef struct { 41 int nentries; 42 uint8_t *lens; 43 uint32_t *codewords; 44 int ndimentions; 45 float min; 46 float delta; 47 int seq_p; 48 int lookup; 49 int *quantlist; 50 float *dimentions; 51 float *pow2; 52} vorbis_enc_codebook; 53 54typedef struct { 55 int dim; 56 int subclass; 57 int masterbook; 58 int *books; 59} vorbis_enc_floor_class; 60 61typedef struct { 62 int partitions; 63 int *partition_to_class; 64 int nclasses; 65 vorbis_enc_floor_class *classes; 66 int multiplier; 67 int rangebits; 68 int values; 69 vorbis_floor1_entry *list; 70} vorbis_enc_floor; 71 72typedef struct { 73 int type; 74 int begin; 75 int end; 76 int partition_size; 77 int classifications; 78 int classbook; 79 int8_t (*books)[8]; 80 float (*maxes)[2]; 81} vorbis_enc_residue; 82 83typedef struct { 84 int submaps; 85 int *mux; 86 int *floor; 87 int *residue; 88 int coupling_steps; 89 int *magnitude; 90 int *angle; 91} vorbis_enc_mapping; 92 93typedef struct { 94 int blockflag; 95 int mapping; 96} vorbis_enc_mode; 97 98typedef struct { 99 int channels; 100 int sample_rate; 101 int log2_blocksize[2]; 102 FFTContext mdct[2]; 103 const float *win[2]; 104 int have_saved; 105 float *saved; 106 float *samples; 107 float *floor; // also used for tmp values for mdct 108 float *coeffs; // also used for residue after floor 109 float quality; 110 111 int ncodebooks; 112 vorbis_enc_codebook *codebooks; 113 114 int nfloors; 115 vorbis_enc_floor *floors; 116 117 int nresidues; 118 vorbis_enc_residue *residues; 119 120 int nmappings; 121 vorbis_enc_mapping *mappings; 122 123 int nmodes; 124 vorbis_enc_mode *modes; 125 126 int64_t sample_count; 127} vorbis_enc_context; 128 129static inline void put_codeword(PutBitContext *pb, vorbis_enc_codebook *cb, 130 int entry) 131{ 132 assert(entry >= 0); 133 assert(entry < cb->nentries); 134 assert(cb->lens[entry]); 135 put_bits(pb, cb->lens[entry], cb->codewords[entry]); 136} 137 138static int cb_lookup_vals(int lookup, int dimentions, int entries) 139{ 140 if (lookup == 1) 141 return ff_vorbis_nth_root(entries, dimentions); 142 else if (lookup == 2) 143 return dimentions *entries; 144 return 0; 145} 146 147static void ready_codebook(vorbis_enc_codebook *cb) 148{ 149 int i; 150 151 ff_vorbis_len2vlc(cb->lens, cb->codewords, cb->nentries); 152 153 if (!cb->lookup) { 154 cb->pow2 = cb->dimentions = NULL; 155 } else { 156 int vals = cb_lookup_vals(cb->lookup, cb->ndimentions, cb->nentries); 157 cb->dimentions = av_malloc(sizeof(float) * cb->nentries * cb->ndimentions); 158 cb->pow2 = av_mallocz(sizeof(float) * cb->nentries); 159 for (i = 0; i < cb->nentries; i++) { 160 float last = 0; 161 int j; 162 int div = 1; 163 for (j = 0; j < cb->ndimentions; j++) { 164 int off; 165 if (cb->lookup == 1) 166 off = (i / div) % vals; // lookup type 1 167 else 168 off = i * cb->ndimentions + j; // lookup type 2 169 170 cb->dimentions[i * cb->ndimentions + j] = last + cb->min + cb->quantlist[off] * cb->delta; 171 if (cb->seq_p) 172 last = cb->dimentions[i * cb->ndimentions + j]; 173 cb->pow2[i] += cb->dimentions[i * cb->ndimentions + j] * cb->dimentions[i * cb->ndimentions + j]; 174 div *= vals; 175 } 176 cb->pow2[i] /= 2.; 177 } 178 } 179} 180 181static void ready_residue(vorbis_enc_residue *rc, vorbis_enc_context *venc) 182{ 183 int i; 184 assert(rc->type == 2); 185 rc->maxes = av_mallocz(sizeof(float[2]) * rc->classifications); 186 for (i = 0; i < rc->classifications; i++) { 187 int j; 188 vorbis_enc_codebook * cb; 189 for (j = 0; j < 8; j++) 190 if (rc->books[i][j] != -1) 191 break; 192 if (j == 8) // zero 193 continue; 194 cb = &venc->codebooks[rc->books[i][j]]; 195 assert(cb->ndimentions >= 2); 196 assert(cb->lookup); 197 198 for (j = 0; j < cb->nentries; j++) { 199 float a; 200 if (!cb->lens[j]) 201 continue; 202 a = fabs(cb->dimentions[j * cb->ndimentions]); 203 if (a > rc->maxes[i][0]) 204 rc->maxes[i][0] = a; 205 a = fabs(cb->dimentions[j * cb->ndimentions + 1]); 206 if (a > rc->maxes[i][1]) 207 rc->maxes[i][1] = a; 208 } 209 } 210 // small bias 211 for (i = 0; i < rc->classifications; i++) { 212 rc->maxes[i][0] += 0.8; 213 rc->maxes[i][1] += 0.8; 214 } 215} 216 217static void create_vorbis_context(vorbis_enc_context *venc, 218 AVCodecContext *avccontext) 219{ 220 vorbis_enc_floor *fc; 221 vorbis_enc_residue *rc; 222 vorbis_enc_mapping *mc; 223 int i, book; 224 225 venc->channels = avccontext->channels; 226 venc->sample_rate = avccontext->sample_rate; 227 venc->log2_blocksize[0] = venc->log2_blocksize[1] = 11; 228 229 venc->ncodebooks = FF_ARRAY_ELEMS(cvectors); 230 venc->codebooks = av_malloc(sizeof(vorbis_enc_codebook) * venc->ncodebooks); 231 232 // codebook 0..14 - floor1 book, values 0..255 233 // codebook 15 residue masterbook 234 // codebook 16..29 residue 235 for (book = 0; book < venc->ncodebooks; book++) { 236 vorbis_enc_codebook *cb = &venc->codebooks[book]; 237 int vals; 238 cb->ndimentions = cvectors[book].dim; 239 cb->nentries = cvectors[book].real_len; 240 cb->min = cvectors[book].min; 241 cb->delta = cvectors[book].delta; 242 cb->lookup = cvectors[book].lookup; 243 cb->seq_p = 0; 244 245 cb->lens = av_malloc(sizeof(uint8_t) * cb->nentries); 246 cb->codewords = av_malloc(sizeof(uint32_t) * cb->nentries); 247 memcpy(cb->lens, cvectors[book].clens, cvectors[book].len); 248 memset(cb->lens + cvectors[book].len, 0, cb->nentries - cvectors[book].len); 249 250 if (cb->lookup) { 251 vals = cb_lookup_vals(cb->lookup, cb->ndimentions, cb->nentries); 252 cb->quantlist = av_malloc(sizeof(int) * vals); 253 for (i = 0; i < vals; i++) 254 cb->quantlist[i] = cvectors[book].quant[i]; 255 } else { 256 cb->quantlist = NULL; 257 } 258 ready_codebook(cb); 259 } 260 261 venc->nfloors = 1; 262 venc->floors = av_malloc(sizeof(vorbis_enc_floor) * venc->nfloors); 263 264 // just 1 floor 265 fc = &venc->floors[0]; 266 fc->partitions = 8; 267 fc->partition_to_class = av_malloc(sizeof(int) * fc->partitions); 268 fc->nclasses = 0; 269 for (i = 0; i < fc->partitions; i++) { 270 static const int a[] = {0, 1, 2, 2, 3, 3, 4, 4}; 271 fc->partition_to_class[i] = a[i]; 272 fc->nclasses = FFMAX(fc->nclasses, fc->partition_to_class[i]); 273 } 274 fc->nclasses++; 275 fc->classes = av_malloc(sizeof(vorbis_enc_floor_class) * fc->nclasses); 276 for (i = 0; i < fc->nclasses; i++) { 277 vorbis_enc_floor_class * c = &fc->classes[i]; 278 int j, books; 279 c->dim = floor_classes[i].dim; 280 c->subclass = floor_classes[i].subclass; 281 c->masterbook = floor_classes[i].masterbook; 282 books = (1 << c->subclass); 283 c->books = av_malloc(sizeof(int) * books); 284 for (j = 0; j < books; j++) 285 c->books[j] = floor_classes[i].nbooks[j]; 286 } 287 fc->multiplier = 2; 288 fc->rangebits = venc->log2_blocksize[0] - 1; 289 290 fc->values = 2; 291 for (i = 0; i < fc->partitions; i++) 292 fc->values += fc->classes[fc->partition_to_class[i]].dim; 293 294 fc->list = av_malloc(sizeof(vorbis_floor1_entry) * fc->values); 295 fc->list[0].x = 0; 296 fc->list[1].x = 1 << fc->rangebits; 297 for (i = 2; i < fc->values; i++) { 298 static const int a[] = { 299 93, 23,372, 6, 46,186,750, 14, 33, 65, 300 130,260,556, 3, 10, 18, 28, 39, 55, 79, 301 111,158,220,312,464,650,850 302 }; 303 fc->list[i].x = a[i - 2]; 304 } 305 ff_vorbis_ready_floor1_list(fc->list, fc->values); 306 307 venc->nresidues = 1; 308 venc->residues = av_malloc(sizeof(vorbis_enc_residue) * venc->nresidues); 309 310 // single residue 311 rc = &venc->residues[0]; 312 rc->type = 2; 313 rc->begin = 0; 314 rc->end = 1600; 315 rc->partition_size = 32; 316 rc->classifications = 10; 317 rc->classbook = 15; 318 rc->books = av_malloc(sizeof(*rc->books) * rc->classifications); 319 { 320 static const int8_t a[10][8] = { 321 { -1, -1, -1, -1, -1, -1, -1, -1, }, 322 { -1, -1, 16, -1, -1, -1, -1, -1, }, 323 { -1, -1, 17, -1, -1, -1, -1, -1, }, 324 { -1, -1, 18, -1, -1, -1, -1, -1, }, 325 { -1, -1, 19, -1, -1, -1, -1, -1, }, 326 { -1, -1, 20, -1, -1, -1, -1, -1, }, 327 { -1, -1, 21, -1, -1, -1, -1, -1, }, 328 { 22, 23, -1, -1, -1, -1, -1, -1, }, 329 { 24, 25, -1, -1, -1, -1, -1, -1, }, 330 { 26, 27, 28, -1, -1, -1, -1, -1, }, 331 }; 332 memcpy(rc->books, a, sizeof a); 333 } 334 ready_residue(rc, venc); 335 336 venc->nmappings = 1; 337 venc->mappings = av_malloc(sizeof(vorbis_enc_mapping) * venc->nmappings); 338 339 // single mapping 340 mc = &venc->mappings[0]; 341 mc->submaps = 1; 342 mc->mux = av_malloc(sizeof(int) * venc->channels); 343 for (i = 0; i < venc->channels; i++) 344 mc->mux[i] = 0; 345 mc->floor = av_malloc(sizeof(int) * mc->submaps); 346 mc->residue = av_malloc(sizeof(int) * mc->submaps); 347 for (i = 0; i < mc->submaps; i++) { 348 mc->floor[i] = 0; 349 mc->residue[i] = 0; 350 } 351 mc->coupling_steps = venc->channels == 2 ? 1 : 0; 352 mc->magnitude = av_malloc(sizeof(int) * mc->coupling_steps); 353 mc->angle = av_malloc(sizeof(int) * mc->coupling_steps); 354 if (mc->coupling_steps) { 355 mc->magnitude[0] = 0; 356 mc->angle[0] = 1; 357 } 358 359 venc->nmodes = 1; 360 venc->modes = av_malloc(sizeof(vorbis_enc_mode) * venc->nmodes); 361 362 // single mode 363 venc->modes[0].blockflag = 0; 364 venc->modes[0].mapping = 0; 365 366 venc->have_saved = 0; 367 venc->saved = av_malloc(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1]) / 2); 368 venc->samples = av_malloc(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1])); 369 venc->floor = av_malloc(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1]) / 2); 370 venc->coeffs = av_malloc(sizeof(float) * venc->channels * (1 << venc->log2_blocksize[1]) / 2); 371 372 venc->win[0] = ff_vorbis_vwin[venc->log2_blocksize[0] - 6]; 373 venc->win[1] = ff_vorbis_vwin[venc->log2_blocksize[1] - 6]; 374 375 ff_mdct_init(&venc->mdct[0], venc->log2_blocksize[0], 0, 1.0); 376 ff_mdct_init(&venc->mdct[1], venc->log2_blocksize[1], 0, 1.0); 377} 378 379static void put_float(PutBitContext *pb, float f) 380{ 381 int exp, mant; 382 uint32_t res = 0; 383 mant = (int)ldexp(frexp(f, &exp), 20); 384 exp += 788 - 20; 385 if (mant < 0) { 386 res |= (1 << 31); 387 mant = -mant; 388 } 389 res |= mant | (exp << 21); 390 put_bits32(pb, res); 391} 392 393static void put_codebook_header(PutBitContext *pb, vorbis_enc_codebook *cb) 394{ 395 int i; 396 int ordered = 0; 397 398 put_bits(pb, 24, 0x564342); //magic 399 put_bits(pb, 16, cb->ndimentions); 400 put_bits(pb, 24, cb->nentries); 401 402 for (i = 1; i < cb->nentries; i++) 403 if (cb->lens[i] < cb->lens[i-1]) 404 break; 405 if (i == cb->nentries) 406 ordered = 1; 407 408 put_bits(pb, 1, ordered); 409 if (ordered) { 410 int len = cb->lens[0]; 411 put_bits(pb, 5, len - 1); 412 i = 0; 413 while (i < cb->nentries) { 414 int j; 415 for (j = 0; j+i < cb->nentries; j++) 416 if (cb->lens[j+i] != len) 417 break; 418 put_bits(pb, ilog(cb->nentries - i), j); 419 i += j; 420 len++; 421 } 422 } else { 423 int sparse = 0; 424 for (i = 0; i < cb->nentries; i++) 425 if (!cb->lens[i]) 426 break; 427 if (i != cb->nentries) 428 sparse = 1; 429 put_bits(pb, 1, sparse); 430 431 for (i = 0; i < cb->nentries; i++) { 432 if (sparse) 433 put_bits(pb, 1, !!cb->lens[i]); 434 if (cb->lens[i]) 435 put_bits(pb, 5, cb->lens[i] - 1); 436 } 437 } 438 439 put_bits(pb, 4, cb->lookup); 440 if (cb->lookup) { 441 int tmp = cb_lookup_vals(cb->lookup, cb->ndimentions, cb->nentries); 442 int bits = ilog(cb->quantlist[0]); 443 444 for (i = 1; i < tmp; i++) 445 bits = FFMAX(bits, ilog(cb->quantlist[i])); 446 447 put_float(pb, cb->min); 448 put_float(pb, cb->delta); 449 450 put_bits(pb, 4, bits - 1); 451 put_bits(pb, 1, cb->seq_p); 452 453 for (i = 0; i < tmp; i++) 454 put_bits(pb, bits, cb->quantlist[i]); 455 } 456} 457 458static void put_floor_header(PutBitContext *pb, vorbis_enc_floor *fc) 459{ 460 int i; 461 462 put_bits(pb, 16, 1); // type, only floor1 is supported 463 464 put_bits(pb, 5, fc->partitions); 465 466 for (i = 0; i < fc->partitions; i++) 467 put_bits(pb, 4, fc->partition_to_class[i]); 468 469 for (i = 0; i < fc->nclasses; i++) { 470 int j, books; 471 472 put_bits(pb, 3, fc->classes[i].dim - 1); 473 put_bits(pb, 2, fc->classes[i].subclass); 474 475 if (fc->classes[i].subclass) 476 put_bits(pb, 8, fc->classes[i].masterbook); 477 478 books = (1 << fc->classes[i].subclass); 479 480 for (j = 0; j < books; j++) 481 put_bits(pb, 8, fc->classes[i].books[j] + 1); 482 } 483 484 put_bits(pb, 2, fc->multiplier - 1); 485 put_bits(pb, 4, fc->rangebits); 486 487 for (i = 2; i < fc->values; i++) 488 put_bits(pb, fc->rangebits, fc->list[i].x); 489} 490 491static void put_residue_header(PutBitContext *pb, vorbis_enc_residue *rc) 492{ 493 int i; 494 495 put_bits(pb, 16, rc->type); 496 497 put_bits(pb, 24, rc->begin); 498 put_bits(pb, 24, rc->end); 499 put_bits(pb, 24, rc->partition_size - 1); 500 put_bits(pb, 6, rc->classifications - 1); 501 put_bits(pb, 8, rc->classbook); 502 503 for (i = 0; i < rc->classifications; i++) { 504 int j, tmp = 0; 505 for (j = 0; j < 8; j++) 506 tmp |= (rc->books[i][j] != -1) << j; 507 508 put_bits(pb, 3, tmp & 7); 509 put_bits(pb, 1, tmp > 7); 510 511 if (tmp > 7) 512 put_bits(pb, 5, tmp >> 3); 513 } 514 515 for (i = 0; i < rc->classifications; i++) { 516 int j; 517 for (j = 0; j < 8; j++) 518 if (rc->books[i][j] != -1) 519 put_bits(pb, 8, rc->books[i][j]); 520 } 521} 522 523static int put_main_header(vorbis_enc_context *venc, uint8_t **out) 524{ 525 int i; 526 PutBitContext pb; 527 uint8_t buffer[50000] = {0}, *p = buffer; 528 int buffer_len = sizeof buffer; 529 int len, hlens[3]; 530 531 // identification header 532 init_put_bits(&pb, p, buffer_len); 533 put_bits(&pb, 8, 1); //magic 534 for (i = 0; "vorbis"[i]; i++) 535 put_bits(&pb, 8, "vorbis"[i]); 536 put_bits32(&pb, 0); // version 537 put_bits(&pb, 8, venc->channels); 538 put_bits32(&pb, venc->sample_rate); 539 put_bits32(&pb, 0); // bitrate 540 put_bits32(&pb, 0); // bitrate 541 put_bits32(&pb, 0); // bitrate 542 put_bits(&pb, 4, venc->log2_blocksize[0]); 543 put_bits(&pb, 4, venc->log2_blocksize[1]); 544 put_bits(&pb, 1, 1); // framing 545 546 flush_put_bits(&pb); 547 hlens[0] = put_bits_count(&pb) >> 3; 548 buffer_len -= hlens[0]; 549 p += hlens[0]; 550 551 // comment header 552 init_put_bits(&pb, p, buffer_len); 553 put_bits(&pb, 8, 3); //magic 554 for (i = 0; "vorbis"[i]; i++) 555 put_bits(&pb, 8, "vorbis"[i]); 556 put_bits32(&pb, 0); // vendor length TODO 557 put_bits32(&pb, 0); // amount of comments 558 put_bits(&pb, 1, 1); // framing 559 560 flush_put_bits(&pb); 561 hlens[1] = put_bits_count(&pb) >> 3; 562 buffer_len -= hlens[1]; 563 p += hlens[1]; 564 565 // setup header 566 init_put_bits(&pb, p, buffer_len); 567 put_bits(&pb, 8, 5); //magic 568 for (i = 0; "vorbis"[i]; i++) 569 put_bits(&pb, 8, "vorbis"[i]); 570 571 // codebooks 572 put_bits(&pb, 8, venc->ncodebooks - 1); 573 for (i = 0; i < venc->ncodebooks; i++) 574 put_codebook_header(&pb, &venc->codebooks[i]); 575 576 // time domain, reserved, zero 577 put_bits(&pb, 6, 0); 578 put_bits(&pb, 16, 0); 579 580 // floors 581 put_bits(&pb, 6, venc->nfloors - 1); 582 for (i = 0; i < venc->nfloors; i++) 583 put_floor_header(&pb, &venc->floors[i]); 584 585 // residues 586 put_bits(&pb, 6, venc->nresidues - 1); 587 for (i = 0; i < venc->nresidues; i++) 588 put_residue_header(&pb, &venc->residues[i]); 589 590 // mappings 591 put_bits(&pb, 6, venc->nmappings - 1); 592 for (i = 0; i < venc->nmappings; i++) { 593 vorbis_enc_mapping *mc = &venc->mappings[i]; 594 int j; 595 put_bits(&pb, 16, 0); // mapping type 596 597 put_bits(&pb, 1, mc->submaps > 1); 598 if (mc->submaps > 1) 599 put_bits(&pb, 4, mc->submaps - 1); 600 601 put_bits(&pb, 1, !!mc->coupling_steps); 602 if (mc->coupling_steps) { 603 put_bits(&pb, 8, mc->coupling_steps - 1); 604 for (j = 0; j < mc->coupling_steps; j++) { 605 put_bits(&pb, ilog(venc->channels - 1), mc->magnitude[j]); 606 put_bits(&pb, ilog(venc->channels - 1), mc->angle[j]); 607 } 608 } 609 610 put_bits(&pb, 2, 0); // reserved 611 612 if (mc->submaps > 1) 613 for (j = 0; j < venc->channels; j++) 614 put_bits(&pb, 4, mc->mux[j]); 615 616 for (j = 0; j < mc->submaps; j++) { 617 put_bits(&pb, 8, 0); // reserved time configuration 618 put_bits(&pb, 8, mc->floor[j]); 619 put_bits(&pb, 8, mc->residue[j]); 620 } 621 } 622 623 // modes 624 put_bits(&pb, 6, venc->nmodes - 1); 625 for (i = 0; i < venc->nmodes; i++) { 626 put_bits(&pb, 1, venc->modes[i].blockflag); 627 put_bits(&pb, 16, 0); // reserved window type 628 put_bits(&pb, 16, 0); // reserved transform type 629 put_bits(&pb, 8, venc->modes[i].mapping); 630 } 631 632 put_bits(&pb, 1, 1); // framing 633 634 flush_put_bits(&pb); 635 hlens[2] = put_bits_count(&pb) >> 3; 636 637 len = hlens[0] + hlens[1] + hlens[2]; 638 p = *out = av_mallocz(64 + len + len/255); 639 640 *p++ = 2; 641 p += av_xiphlacing(p, hlens[0]); 642 p += av_xiphlacing(p, hlens[1]); 643 buffer_len = 0; 644 for (i = 0; i < 3; i++) { 645 memcpy(p, buffer + buffer_len, hlens[i]); 646 p += hlens[i]; 647 buffer_len += hlens[i]; 648 } 649 650 return p - *out; 651} 652 653static float get_floor_average(vorbis_enc_floor * fc, float *coeffs, int i) 654{ 655 int begin = fc->list[fc->list[FFMAX(i-1, 0)].sort].x; 656 int end = fc->list[fc->list[FFMIN(i+1, fc->values - 1)].sort].x; 657 int j; 658 float average = 0; 659 660 for (j = begin; j < end; j++) 661 average += fabs(coeffs[j]); 662 return average / (end - begin); 663} 664 665static void floor_fit(vorbis_enc_context *venc, vorbis_enc_floor *fc, 666 float *coeffs, uint_fast16_t *posts, int samples) 667{ 668 int range = 255 / fc->multiplier + 1; 669 int i; 670 float tot_average = 0.; 671 float averages[fc->values]; 672 for (i = 0; i < fc->values; i++) { 673 averages[i] = get_floor_average(fc, coeffs, i); 674 tot_average += averages[i]; 675 } 676 tot_average /= fc->values; 677 tot_average /= venc->quality; 678 679 for (i = 0; i < fc->values; i++) { 680 int position = fc->list[fc->list[i].sort].x; 681 float average = averages[i]; 682 int j; 683 684 average *= pow(tot_average / average, 0.5) * pow(1.25, position/200.); // MAGIC! 685 for (j = 0; j < range - 1; j++) 686 if (ff_vorbis_floor1_inverse_db_table[j * fc->multiplier] > average) 687 break; 688 posts[fc->list[i].sort] = j; 689 } 690} 691 692static int render_point(int x0, int y0, int x1, int y1, int x) 693{ 694 return y0 + (x - x0) * (y1 - y0) / (x1 - x0); 695} 696 697static void floor_encode(vorbis_enc_context *venc, vorbis_enc_floor *fc, 698 PutBitContext *pb, uint_fast16_t *posts, 699 float *floor, int samples) 700{ 701 int range = 255 / fc->multiplier + 1; 702 int coded[fc->values]; // first 2 values are unused 703 int i, counter; 704 705 put_bits(pb, 1, 1); // non zero 706 put_bits(pb, ilog(range - 1), posts[0]); 707 put_bits(pb, ilog(range - 1), posts[1]); 708 coded[0] = coded[1] = 1; 709 710 for (i = 2; i < fc->values; i++) { 711 int predicted = render_point(fc->list[fc->list[i].low].x, 712 posts[fc->list[i].low], 713 fc->list[fc->list[i].high].x, 714 posts[fc->list[i].high], 715 fc->list[i].x); 716 int highroom = range - predicted; 717 int lowroom = predicted; 718 int room = FFMIN(highroom, lowroom); 719 if (predicted == posts[i]) { 720 coded[i] = 0; // must be used later as flag! 721 continue; 722 } else { 723 if (!coded[fc->list[i].low ]) 724 coded[fc->list[i].low ] = -1; 725 if (!coded[fc->list[i].high]) 726 coded[fc->list[i].high] = -1; 727 } 728 if (posts[i] > predicted) { 729 if (posts[i] - predicted > room) 730 coded[i] = posts[i] - predicted + lowroom; 731 else 732 coded[i] = (posts[i] - predicted) << 1; 733 } else { 734 if (predicted - posts[i] > room) 735 coded[i] = predicted - posts[i] + highroom - 1; 736 else 737 coded[i] = ((predicted - posts[i]) << 1) - 1; 738 } 739 } 740 741 counter = 2; 742 for (i = 0; i < fc->partitions; i++) { 743 vorbis_enc_floor_class * c = &fc->classes[fc->partition_to_class[i]]; 744 int k, cval = 0, csub = 1<<c->subclass; 745 if (c->subclass) { 746 vorbis_enc_codebook * book = &venc->codebooks[c->masterbook]; 747 int cshift = 0; 748 for (k = 0; k < c->dim; k++) { 749 int l; 750 for (l = 0; l < csub; l++) { 751 int maxval = 1; 752 if (c->books[l] != -1) 753 maxval = venc->codebooks[c->books[l]].nentries; 754 // coded could be -1, but this still works, cause that is 0 755 if (coded[counter + k] < maxval) 756 break; 757 } 758 assert(l != csub); 759 cval |= l << cshift; 760 cshift += c->subclass; 761 } 762 put_codeword(pb, book, cval); 763 } 764 for (k = 0; k < c->dim; k++) { 765 int book = c->books[cval & (csub-1)]; 766 int entry = coded[counter++]; 767 cval >>= c->subclass; 768 if (book == -1) 769 continue; 770 if (entry == -1) 771 entry = 0; 772 put_codeword(pb, &venc->codebooks[book], entry); 773 } 774 } 775 776 ff_vorbis_floor1_render_list(fc->list, fc->values, posts, coded, 777 fc->multiplier, floor, samples); 778} 779 780static float *put_vector(vorbis_enc_codebook *book, PutBitContext *pb, 781 float *num) 782{ 783 int i, entry = -1; 784 float distance = FLT_MAX; 785 assert(book->dimentions); 786 for (i = 0; i < book->nentries; i++) { 787 float * vec = book->dimentions + i * book->ndimentions, d = book->pow2[i]; 788 int j; 789 if (!book->lens[i]) 790 continue; 791 for (j = 0; j < book->ndimentions; j++) 792 d -= vec[j] * num[j]; 793 if (distance > d) { 794 entry = i; 795 distance = d; 796 } 797 } 798 put_codeword(pb, book, entry); 799 return &book->dimentions[entry * book->ndimentions]; 800} 801 802static void residue_encode(vorbis_enc_context *venc, vorbis_enc_residue *rc, 803 PutBitContext *pb, float *coeffs, int samples, 804 int real_ch) 805{ 806 int pass, i, j, p, k; 807 int psize = rc->partition_size; 808 int partitions = (rc->end - rc->begin) / psize; 809 int channels = (rc->type == 2) ? 1 : real_ch; 810 int classes[channels][partitions]; 811 int classwords = venc->codebooks[rc->classbook].ndimentions; 812 813 assert(rc->type == 2); 814 assert(real_ch == 2); 815 for (p = 0; p < partitions; p++) { 816 float max1 = 0., max2 = 0.; 817 int s = rc->begin + p * psize; 818 for (k = s; k < s + psize; k += 2) { 819 max1 = FFMAX(max1, fabs(coeffs[ k / real_ch])); 820 max2 = FFMAX(max2, fabs(coeffs[samples + k / real_ch])); 821 } 822 823 for (i = 0; i < rc->classifications - 1; i++) 824 if (max1 < rc->maxes[i][0] && max2 < rc->maxes[i][1]) 825 break; 826 classes[0][p] = i; 827 } 828 829 for (pass = 0; pass < 8; pass++) { 830 p = 0; 831 while (p < partitions) { 832 if (pass == 0) 833 for (j = 0; j < channels; j++) { 834 vorbis_enc_codebook * book = &venc->codebooks[rc->classbook]; 835 int entry = 0; 836 for (i = 0; i < classwords; i++) { 837 entry *= rc->classifications; 838 entry += classes[j][p + i]; 839 } 840 put_codeword(pb, book, entry); 841 } 842 for (i = 0; i < classwords && p < partitions; i++, p++) { 843 for (j = 0; j < channels; j++) { 844 int nbook = rc->books[classes[j][p]][pass]; 845 vorbis_enc_codebook * book = &venc->codebooks[nbook]; 846 float *buf = coeffs + samples*j + rc->begin + p*psize; 847 if (nbook == -1) 848 continue; 849 850 assert(rc->type == 0 || rc->type == 2); 851 assert(!(psize % book->ndimentions)); 852 853 if (rc->type == 0) { 854 for (k = 0; k < psize; k += book->ndimentions) { 855 float *a = put_vector(book, pb, &buf[k]); 856 int l; 857 for (l = 0; l < book->ndimentions; l++) 858 buf[k + l] -= a[l]; 859 } 860 } else { 861 int s = rc->begin + p * psize, a1, b1; 862 a1 = (s % real_ch) * samples; 863 b1 = s / real_ch; 864 s = real_ch * samples; 865 for (k = 0; k < psize; k += book->ndimentions) { 866 int dim, a2 = a1, b2 = b1; 867 float vec[book->ndimentions], *pv = vec; 868 for (dim = book->ndimentions; dim--; ) { 869 *pv++ = coeffs[a2 + b2]; 870 if ((a2 += samples) == s) { 871 a2 = 0; 872 b2++; 873 } 874 } 875 pv = put_vector(book, pb, vec); 876 for (dim = book->ndimentions; dim--; ) { 877 coeffs[a1 + b1] -= *pv++; 878 if ((a1 += samples) == s) { 879 a1 = 0; 880 b1++; 881 } 882 } 883 } 884 } 885 } 886 } 887 } 888 } 889} 890 891static int apply_window_and_mdct(vorbis_enc_context *venc, signed short *audio, 892 int samples) 893{ 894 int i, j, channel; 895 const float * win = venc->win[0]; 896 int window_len = 1 << (venc->log2_blocksize[0] - 1); 897 float n = (float)(1 << venc->log2_blocksize[0]) / 4.; 898 // FIXME use dsp 899 900 if (!venc->have_saved && !samples) 901 return 0; 902 903 if (venc->have_saved) { 904 for (channel = 0; channel < venc->channels; channel++) 905 memcpy(venc->samples + channel * window_len * 2, 906 venc->saved + channel * window_len, sizeof(float) * window_len); 907 } else { 908 for (channel = 0; channel < venc->channels; channel++) 909 memset(venc->samples + channel * window_len * 2, 0, 910 sizeof(float) * window_len); 911 } 912 913 if (samples) { 914 for (channel = 0; channel < venc->channels; channel++) { 915 float * offset = venc->samples + channel*window_len*2 + window_len; 916 j = channel; 917 for (i = 0; i < samples; i++, j += venc->channels) 918 offset[i] = -audio[j] / 32768. / n * win[window_len - i - 1]; //FIXME find out why the sign has to be fliped 919 } 920 } else { 921 for (channel = 0; channel < venc->channels; channel++) 922 memset(venc->samples + channel * window_len * 2 + window_len, 923 0, sizeof(float) * window_len); 924 } 925 926 for (channel = 0; channel < venc->channels; channel++) 927 ff_mdct_calc(&venc->mdct[0], venc->coeffs + channel * window_len, 928 venc->samples + channel * window_len * 2); 929 930 if (samples) { 931 for (channel = 0; channel < venc->channels; channel++) { 932 float *offset = venc->saved + channel * window_len; 933 j = channel; 934 for (i = 0; i < samples; i++, j += venc->channels) 935 offset[i] = -audio[j] / 32768. / n * win[i]; //FIXME find out why the sign has to be fliped 936 } 937 venc->have_saved = 1; 938 } else { 939 venc->have_saved = 0; 940 } 941 return 1; 942} 943 944static av_cold int vorbis_encode_init(AVCodecContext *avccontext) 945{ 946 vorbis_enc_context *venc = avccontext->priv_data; 947 948 if (avccontext->channels != 2) { 949 av_log(avccontext, AV_LOG_ERROR, "Current FFmpeg Vorbis encoder only supports 2 channels.\n"); 950 return -1; 951 } 952 953 create_vorbis_context(venc, avccontext); 954 955 if (avccontext->flags & CODEC_FLAG_QSCALE) 956 venc->quality = avccontext->global_quality / (float)FF_QP2LAMBDA / 10.; 957 else 958 venc->quality = 1.; 959 venc->quality *= venc->quality; 960 961 avccontext->extradata_size = put_main_header(venc, (uint8_t**)&avccontext->extradata); 962 963 avccontext->frame_size = 1 << (venc->log2_blocksize[0] - 1); 964 965 avccontext->coded_frame = avcodec_alloc_frame(); 966 avccontext->coded_frame->key_frame = 1; 967 968 return 0; 969} 970 971static int vorbis_encode_frame(AVCodecContext *avccontext, 972 unsigned char *packets, 973 int buf_size, void *data) 974{ 975 vorbis_enc_context *venc = avccontext->priv_data; 976 signed short *audio = data; 977 int samples = data ? avccontext->frame_size : 0; 978 vorbis_enc_mode *mode; 979 vorbis_enc_mapping *mapping; 980 PutBitContext pb; 981 int i; 982 983 if (!apply_window_and_mdct(venc, audio, samples)) 984 return 0; 985 samples = 1 << (venc->log2_blocksize[0] - 1); 986 987 init_put_bits(&pb, packets, buf_size); 988 989 put_bits(&pb, 1, 0); // magic bit 990 991 put_bits(&pb, ilog(venc->nmodes - 1), 0); // 0 bits, the mode 992 993 mode = &venc->modes[0]; 994 mapping = &venc->mappings[mode->mapping]; 995 if (mode->blockflag) { 996 put_bits(&pb, 1, 0); 997 put_bits(&pb, 1, 0); 998 } 999 1000 for (i = 0; i < venc->channels; i++) { 1001 vorbis_enc_floor *fc = &venc->floors[mapping->floor[mapping->mux[i]]]; 1002 uint_fast16_t posts[fc->values]; 1003 floor_fit(venc, fc, &venc->coeffs[i * samples], posts, samples); 1004 floor_encode(venc, fc, &pb, posts, &venc->floor[i * samples], samples); 1005 } 1006 1007 for (i = 0; i < venc->channels * samples; i++) 1008 venc->coeffs[i] /= venc->floor[i]; 1009 1010 for (i = 0; i < mapping->coupling_steps; i++) { 1011 float *mag = venc->coeffs + mapping->magnitude[i] * samples; 1012 float *ang = venc->coeffs + mapping->angle[i] * samples; 1013 int j; 1014 for (j = 0; j < samples; j++) { 1015 float a = ang[j]; 1016 ang[j] -= mag[j]; 1017 if (mag[j] > 0) 1018 ang[j] = -ang[j]; 1019 if (ang[j] < 0) 1020 mag[j] = a; 1021 } 1022 } 1023 1024 residue_encode(venc, &venc->residues[mapping->residue[mapping->mux[0]]], 1025 &pb, venc->coeffs, samples, venc->channels); 1026 1027 avccontext->coded_frame->pts = venc->sample_count; 1028 venc->sample_count += avccontext->frame_size; 1029 flush_put_bits(&pb); 1030 return put_bits_count(&pb) >> 3; 1031} 1032 1033 1034static av_cold int vorbis_encode_close(AVCodecContext *avccontext) 1035{ 1036 vorbis_enc_context *venc = avccontext->priv_data; 1037 int i; 1038 1039 if (venc->codebooks) 1040 for (i = 0; i < venc->ncodebooks; i++) { 1041 av_freep(&venc->codebooks[i].lens); 1042 av_freep(&venc->codebooks[i].codewords); 1043 av_freep(&venc->codebooks[i].quantlist); 1044 av_freep(&venc->codebooks[i].dimentions); 1045 av_freep(&venc->codebooks[i].pow2); 1046 } 1047 av_freep(&venc->codebooks); 1048 1049 if (venc->floors) 1050 for (i = 0; i < venc->nfloors; i++) { 1051 int j; 1052 if (venc->floors[i].classes) 1053 for (j = 0; j < venc->floors[i].nclasses; j++) 1054 av_freep(&venc->floors[i].classes[j].books); 1055 av_freep(&venc->floors[i].classes); 1056 av_freep(&venc->floors[i].partition_to_class); 1057 av_freep(&venc->floors[i].list); 1058 } 1059 av_freep(&venc->floors); 1060 1061 if (venc->residues) 1062 for (i = 0; i < venc->nresidues; i++) { 1063 av_freep(&venc->residues[i].books); 1064 av_freep(&venc->residues[i].maxes); 1065 } 1066 av_freep(&venc->residues); 1067 1068 if (venc->mappings) 1069 for (i = 0; i < venc->nmappings; i++) { 1070 av_freep(&venc->mappings[i].mux); 1071 av_freep(&venc->mappings[i].floor); 1072 av_freep(&venc->mappings[i].residue); 1073 av_freep(&venc->mappings[i].magnitude); 1074 av_freep(&venc->mappings[i].angle); 1075 } 1076 av_freep(&venc->mappings); 1077 1078 av_freep(&venc->modes); 1079 1080 av_freep(&venc->saved); 1081 av_freep(&venc->samples); 1082 av_freep(&venc->floor); 1083 av_freep(&venc->coeffs); 1084 1085 ff_mdct_end(&venc->mdct[0]); 1086 ff_mdct_end(&venc->mdct[1]); 1087 1088 av_freep(&avccontext->coded_frame); 1089 av_freep(&avccontext->extradata); 1090 1091 return 0 ; 1092} 1093 1094AVCodec vorbis_encoder = { 1095 "vorbis", 1096 AVMEDIA_TYPE_AUDIO, 1097 CODEC_ID_VORBIS, 1098 sizeof(vorbis_enc_context), 1099 vorbis_encode_init, 1100 vorbis_encode_frame, 1101 vorbis_encode_close, 1102 .capabilities= CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL, 1103 .sample_fmts = (const enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE}, 1104 .long_name = NULL_IF_CONFIG_SMALL("Vorbis"), 1105}; 1106