1/* 2 * Copyright (C) 2003-2004 the ffmpeg project 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 libavcodec/vp3.c 23 * On2 VP3 Video Decoder 24 * 25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx) 26 * For more information about the VP3 coding process, visit: 27 * http://wiki.multimedia.cx/index.php?title=On2_VP3 28 * 29 * Theora decoder by Alex Beregszaszi 30 */ 31 32#include <stdio.h> 33#include <stdlib.h> 34#include <string.h> 35#include <unistd.h> 36 37#include "avcodec.h" 38#include "dsputil.h" 39#include "bitstream.h" 40 41#include "vp3data.h" 42#include "xiph.h" 43 44#define FRAGMENT_PIXELS 8 45 46static av_cold int vp3_decode_end(AVCodecContext *avctx); 47 48typedef struct Coeff { 49 struct Coeff *next; 50 DCTELEM coeff; 51 uint8_t index; 52} Coeff; 53 54//FIXME split things out into their own arrays 55typedef struct Vp3Fragment { 56 Coeff *next_coeff; 57 /* address of first pixel taking into account which plane the fragment 58 * lives on as well as the plane stride */ 59 int first_pixel; 60 /* this is the macroblock that the fragment belongs to */ 61 uint16_t macroblock; 62 uint8_t coding_method; 63 int8_t motion_x; 64 int8_t motion_y; 65} Vp3Fragment; 66 67#define SB_NOT_CODED 0 68#define SB_PARTIALLY_CODED 1 69#define SB_FULLY_CODED 2 70 71#define MODE_INTER_NO_MV 0 72#define MODE_INTRA 1 73#define MODE_INTER_PLUS_MV 2 74#define MODE_INTER_LAST_MV 3 75#define MODE_INTER_PRIOR_LAST 4 76#define MODE_USING_GOLDEN 5 77#define MODE_GOLDEN_MV 6 78#define MODE_INTER_FOURMV 7 79#define CODING_MODE_COUNT 8 80 81/* special internal mode */ 82#define MODE_COPY 8 83 84/* There are 6 preset schemes, plus a free-form scheme */ 85static const int ModeAlphabet[6][CODING_MODE_COUNT] = 86{ 87 /* scheme 1: Last motion vector dominates */ 88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, 89 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV, 90 MODE_INTRA, MODE_USING_GOLDEN, 91 MODE_GOLDEN_MV, MODE_INTER_FOURMV }, 92 93 /* scheme 2 */ 94 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, 95 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV, 96 MODE_INTRA, MODE_USING_GOLDEN, 97 MODE_GOLDEN_MV, MODE_INTER_FOURMV }, 98 99 /* scheme 3 */ 100 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV, 101 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV, 102 MODE_INTRA, MODE_USING_GOLDEN, 103 MODE_GOLDEN_MV, MODE_INTER_FOURMV }, 104 105 /* scheme 4 */ 106 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV, 107 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST, 108 MODE_INTRA, MODE_USING_GOLDEN, 109 MODE_GOLDEN_MV, MODE_INTER_FOURMV }, 110 111 /* scheme 5: No motion vector dominates */ 112 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV, 113 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV, 114 MODE_INTRA, MODE_USING_GOLDEN, 115 MODE_GOLDEN_MV, MODE_INTER_FOURMV }, 116 117 /* scheme 6 */ 118 { MODE_INTER_NO_MV, MODE_USING_GOLDEN, 119 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST, 120 MODE_INTER_PLUS_MV, MODE_INTRA, 121 MODE_GOLDEN_MV, MODE_INTER_FOURMV }, 122 123}; 124 125#define MIN_DEQUANT_VAL 2 126 127typedef struct Vp3DecodeContext { 128 AVCodecContext *avctx; 129 int theora, theora_tables; 130 int version; 131 int width, height; 132 AVFrame golden_frame; 133 AVFrame last_frame; 134 AVFrame current_frame; 135 int keyframe; 136 DSPContext dsp; 137 int flipped_image; 138 139 int qis[3]; 140 int nqis; 141 int quality_index; 142 int last_quality_index; 143 144 int superblock_count; 145 int y_superblock_width; 146 int y_superblock_height; 147 int c_superblock_width; 148 int c_superblock_height; 149 int u_superblock_start; 150 int v_superblock_start; 151 unsigned char *superblock_coding; 152 153 int macroblock_count; 154 int macroblock_width; 155 int macroblock_height; 156 157 int fragment_count; 158 int fragment_width; 159 int fragment_height; 160 161 Vp3Fragment *all_fragments; 162 uint8_t *coeff_counts; 163 Coeff *coeffs; 164 Coeff *next_coeff; 165 int fragment_start[3]; 166 167 ScanTable scantable; 168 169 /* tables */ 170 uint16_t coded_dc_scale_factor[64]; 171 uint32_t coded_ac_scale_factor[64]; 172 uint8_t base_matrix[384][64]; 173 uint8_t qr_count[2][3]; 174 uint8_t qr_size [2][3][64]; 175 uint16_t qr_base[2][3][64]; 176 177 /* this is a list of indexes into the all_fragments array indicating 178 * which of the fragments are coded */ 179 int *coded_fragment_list; 180 int coded_fragment_list_index; 181 int pixel_addresses_initialized; 182 183 VLC dc_vlc[16]; 184 VLC ac_vlc_1[16]; 185 VLC ac_vlc_2[16]; 186 VLC ac_vlc_3[16]; 187 VLC ac_vlc_4[16]; 188 189 VLC superblock_run_length_vlc; 190 VLC fragment_run_length_vlc; 191 VLC mode_code_vlc; 192 VLC motion_vector_vlc; 193 194 /* these arrays need to be on 16-byte boundaries since SSE2 operations 195 * index into them */ 196 DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]); //<qmat[is_inter][plane] 197 198 /* This table contains superblock_count * 16 entries. Each set of 16 199 * numbers corresponds to the fragment indexes 0..15 of the superblock. 200 * An entry will be -1 to indicate that no entry corresponds to that 201 * index. */ 202 int *superblock_fragments; 203 204 /* This table contains superblock_count * 4 entries. Each set of 4 205 * numbers corresponds to the macroblock indexes 0..3 of the superblock. 206 * An entry will be -1 to indicate that no entry corresponds to that 207 * index. */ 208 int *superblock_macroblocks; 209 210 /* This table contains macroblock_count * 6 entries. Each set of 6 211 * numbers corresponds to the fragment indexes 0..5 which comprise 212 * the macroblock (4 Y fragments and 2 C fragments). */ 213 int *macroblock_fragments; 214 /* This is an array that indicates how a particular macroblock 215 * is coded. */ 216 unsigned char *macroblock_coding; 217 218 int first_coded_y_fragment; 219 int first_coded_c_fragment; 220 int last_coded_y_fragment; 221 int last_coded_c_fragment; 222 223 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc 224 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16 225 226 /* Huffman decode */ 227 int hti; 228 unsigned int hbits; 229 int entries; 230 int huff_code_size; 231 uint16_t huffman_table[80][32][2]; 232 233 uint8_t filter_limit_values[64]; 234 DECLARE_ALIGNED_8(int, bounding_values_array[256+2]); 235} Vp3DecodeContext; 236 237/************************************************************************ 238 * VP3 specific functions 239 ************************************************************************/ 240 241/* 242 * This function sets up all of the various blocks mappings: 243 * superblocks <-> fragments, macroblocks <-> fragments, 244 * superblocks <-> macroblocks 245 * 246 * Returns 0 is successful; returns 1 if *anything* went wrong. 247 */ 248static int init_block_mapping(Vp3DecodeContext *s) 249{ 250 int i, j; 251 signed int hilbert_walk_mb[4]; 252 253 int current_fragment = 0; 254 int current_width = 0; 255 int current_height = 0; 256 int right_edge = 0; 257 int bottom_edge = 0; 258 int superblock_row_inc = 0; 259 int *hilbert = NULL; 260 int mapping_index = 0; 261 262 int current_macroblock; 263 int c_fragment; 264 265 signed char travel_width[16] = { 266 1, 1, 0, -1, 267 0, 0, 1, 0, 268 1, 0, 1, 0, 269 0, -1, 0, 1 270 }; 271 272 signed char travel_height[16] = { 273 0, 0, 1, 0, 274 1, 1, 0, -1, 275 0, 1, 0, -1, 276 -1, 0, -1, 0 277 }; 278 279 signed char travel_width_mb[4] = { 280 1, 0, 1, 0 281 }; 282 283 signed char travel_height_mb[4] = { 284 0, 1, 0, -1 285 }; 286 287 hilbert_walk_mb[0] = 1; 288 hilbert_walk_mb[1] = s->macroblock_width; 289 hilbert_walk_mb[2] = 1; 290 hilbert_walk_mb[3] = -s->macroblock_width; 291 292 /* iterate through each superblock (all planes) and map the fragments */ 293 for (i = 0; i < s->superblock_count; i++) { 294 /* time to re-assign the limits? */ 295 if (i == 0) { 296 297 /* start of Y superblocks */ 298 right_edge = s->fragment_width; 299 bottom_edge = s->fragment_height; 300 current_width = -1; 301 current_height = 0; 302 superblock_row_inc = 3 * s->fragment_width - 303 (s->y_superblock_width * 4 - s->fragment_width); 304 305 /* the first operation for this variable is to advance by 1 */ 306 current_fragment = -1; 307 308 } else if (i == s->u_superblock_start) { 309 310 /* start of U superblocks */ 311 right_edge = s->fragment_width / 2; 312 bottom_edge = s->fragment_height / 2; 313 current_width = -1; 314 current_height = 0; 315 superblock_row_inc = 3 * (s->fragment_width / 2) - 316 (s->c_superblock_width * 4 - s->fragment_width / 2); 317 318 /* the first operation for this variable is to advance by 1 */ 319 current_fragment = s->fragment_start[1] - 1; 320 321 } else if (i == s->v_superblock_start) { 322 323 /* start of V superblocks */ 324 right_edge = s->fragment_width / 2; 325 bottom_edge = s->fragment_height / 2; 326 current_width = -1; 327 current_height = 0; 328 superblock_row_inc = 3 * (s->fragment_width / 2) - 329 (s->c_superblock_width * 4 - s->fragment_width / 2); 330 331 /* the first operation for this variable is to advance by 1 */ 332 current_fragment = s->fragment_start[2] - 1; 333 334 } 335 336 if (current_width >= right_edge - 1) { 337 /* reset width and move to next superblock row */ 338 current_width = -1; 339 current_height += 4; 340 341 /* fragment is now at the start of a new superblock row */ 342 current_fragment += superblock_row_inc; 343 } 344 345 /* iterate through all 16 fragments in a superblock */ 346 for (j = 0; j < 16; j++) { 347 current_fragment += travel_width[j] + right_edge * travel_height[j]; 348 current_width += travel_width[j]; 349 current_height += travel_height[j]; 350 351 /* check if the fragment is in bounds */ 352 if ((current_width < right_edge) && 353 (current_height < bottom_edge)) { 354 s->superblock_fragments[mapping_index] = current_fragment; 355 } else { 356 s->superblock_fragments[mapping_index] = -1; 357 } 358 359 mapping_index++; 360 } 361 } 362 363 /* initialize the superblock <-> macroblock mapping; iterate through 364 * all of the Y plane superblocks to build this mapping */ 365 right_edge = s->macroblock_width; 366 bottom_edge = s->macroblock_height; 367 current_width = -1; 368 current_height = 0; 369 superblock_row_inc = s->macroblock_width - 370 (s->y_superblock_width * 2 - s->macroblock_width); 371 hilbert = hilbert_walk_mb; 372 mapping_index = 0; 373 current_macroblock = -1; 374 for (i = 0; i < s->u_superblock_start; i++) { 375 376 if (current_width >= right_edge - 1) { 377 /* reset width and move to next superblock row */ 378 current_width = -1; 379 current_height += 2; 380 381 /* macroblock is now at the start of a new superblock row */ 382 current_macroblock += superblock_row_inc; 383 } 384 385 /* iterate through each potential macroblock in the superblock */ 386 for (j = 0; j < 4; j++) { 387 current_macroblock += hilbert_walk_mb[j]; 388 current_width += travel_width_mb[j]; 389 current_height += travel_height_mb[j]; 390 391 /* check if the macroblock is in bounds */ 392 if ((current_width < right_edge) && 393 (current_height < bottom_edge)) { 394 s->superblock_macroblocks[mapping_index] = current_macroblock; 395 } else { 396 s->superblock_macroblocks[mapping_index] = -1; 397 } 398 399 mapping_index++; 400 } 401 } 402 403 /* initialize the macroblock <-> fragment mapping */ 404 current_fragment = 0; 405 current_macroblock = 0; 406 mapping_index = 0; 407 for (i = 0; i < s->fragment_height; i += 2) { 408 409 for (j = 0; j < s->fragment_width; j += 2) { 410 411 s->all_fragments[current_fragment].macroblock = current_macroblock; 412 s->macroblock_fragments[mapping_index++] = current_fragment; 413 414 if (j + 1 < s->fragment_width) { 415 s->all_fragments[current_fragment + 1].macroblock = current_macroblock; 416 s->macroblock_fragments[mapping_index++] = current_fragment + 1; 417 } else 418 s->macroblock_fragments[mapping_index++] = -1; 419 420 if (i + 1 < s->fragment_height) { 421 s->all_fragments[current_fragment + s->fragment_width].macroblock = 422 current_macroblock; 423 s->macroblock_fragments[mapping_index++] = 424 current_fragment + s->fragment_width; 425 } else 426 s->macroblock_fragments[mapping_index++] = -1; 427 428 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) { 429 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock = 430 current_macroblock; 431 s->macroblock_fragments[mapping_index++] = 432 current_fragment + s->fragment_width + 1; 433 } else 434 s->macroblock_fragments[mapping_index++] = -1; 435 436 /* C planes */ 437 c_fragment = s->fragment_start[1] + 438 (i * s->fragment_width / 4) + (j / 2); 439 s->all_fragments[c_fragment].macroblock = s->macroblock_count; 440 s->macroblock_fragments[mapping_index++] = c_fragment; 441 442 c_fragment = s->fragment_start[2] + 443 (i * s->fragment_width / 4) + (j / 2); 444 s->all_fragments[c_fragment].macroblock = s->macroblock_count; 445 s->macroblock_fragments[mapping_index++] = c_fragment; 446 447 if (j + 2 <= s->fragment_width) 448 current_fragment += 2; 449 else 450 current_fragment++; 451 current_macroblock++; 452 } 453 454 current_fragment += s->fragment_width; 455 } 456 457 return 0; /* successful path out */ 458} 459 460/* 461 * This function wipes out all of the fragment data. 462 */ 463static void init_frame(Vp3DecodeContext *s, GetBitContext *gb) 464{ 465 int i; 466 467 /* zero out all of the fragment information */ 468 s->coded_fragment_list_index = 0; 469 for (i = 0; i < s->fragment_count; i++) { 470 s->coeff_counts[i] = 0; 471 s->all_fragments[i].motion_x = 127; 472 s->all_fragments[i].motion_y = 127; 473 s->all_fragments[i].next_coeff= NULL; 474 s->coeffs[i].index= 475 s->coeffs[i].coeff=0; 476 s->coeffs[i].next= NULL; 477 } 478} 479 480/* 481 * This function sets up the dequantization tables used for a particular 482 * frame. 483 */ 484static void init_dequantizer(Vp3DecodeContext *s) 485{ 486 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index]; 487 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index]; 488 int i, plane, inter, qri, bmi, bmj, qistart; 489 490 for(inter=0; inter<2; inter++){ 491 for(plane=0; plane<3; plane++){ 492 int sum=0; 493 for(qri=0; qri<s->qr_count[inter][plane]; qri++){ 494 sum+= s->qr_size[inter][plane][qri]; 495 if(s->quality_index <= sum) 496 break; 497 } 498 qistart= sum - s->qr_size[inter][plane][qri]; 499 bmi= s->qr_base[inter][plane][qri ]; 500 bmj= s->qr_base[inter][plane][qri+1]; 501 for(i=0; i<64; i++){ 502 int coeff= ( 2*(sum -s->quality_index)*s->base_matrix[bmi][i] 503 - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i] 504 + s->qr_size[inter][plane][qri]) 505 / (2*s->qr_size[inter][plane][qri]); 506 507 int qmin= 8<<(inter + !i); 508 int qscale= i ? ac_scale_factor : dc_scale_factor; 509 510 s->qmat[inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096); 511 } 512 } 513 } 514 515 memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune 516} 517 518/* 519 * This function initializes the loop filter boundary limits if the frame's 520 * quality index is different from the previous frame's. 521 */ 522static void init_loop_filter(Vp3DecodeContext *s) 523{ 524 int *bounding_values= s->bounding_values_array+127; 525 int filter_limit; 526 int x; 527 528 filter_limit = s->filter_limit_values[s->quality_index]; 529 530 /* set up the bounding values */ 531 memset(s->bounding_values_array, 0, 256 * sizeof(int)); 532 for (x = 0; x < filter_limit; x++) { 533 bounding_values[-x - filter_limit] = -filter_limit + x; 534 bounding_values[-x] = -x; 535 bounding_values[x] = x; 536 bounding_values[x + filter_limit] = filter_limit - x; 537 } 538 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202; 539} 540 541/* 542 * This function unpacks all of the superblock/macroblock/fragment coding 543 * information from the bitstream. 544 */ 545static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb) 546{ 547 int bit = 0; 548 int current_superblock = 0; 549 int current_run = 0; 550 int decode_fully_flags = 0; 551 int decode_partial_blocks = 0; 552 int first_c_fragment_seen; 553 554 int i, j; 555 int current_fragment; 556 557 if (s->keyframe) { 558 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count); 559 560 } else { 561 562 /* unpack the list of partially-coded superblocks */ 563 bit = get_bits1(gb); 564 /* toggle the bit because as soon as the first run length is 565 * fetched the bit will be toggled again */ 566 bit ^= 1; 567 while (current_superblock < s->superblock_count) { 568 if (current_run-- == 0) { 569 bit ^= 1; 570 current_run = get_vlc2(gb, 571 s->superblock_run_length_vlc.table, 6, 2); 572 if (current_run == 33) 573 current_run += get_bits(gb, 12); 574 575 /* if any of the superblocks are not partially coded, flag 576 * a boolean to decode the list of fully-coded superblocks */ 577 if (bit == 0) { 578 decode_fully_flags = 1; 579 } else { 580 581 /* make a note of the fact that there are partially coded 582 * superblocks */ 583 decode_partial_blocks = 1; 584 } 585 } 586 s->superblock_coding[current_superblock++] = bit; 587 } 588 589 /* unpack the list of fully coded superblocks if any of the blocks were 590 * not marked as partially coded in the previous step */ 591 if (decode_fully_flags) { 592 593 current_superblock = 0; 594 current_run = 0; 595 bit = get_bits1(gb); 596 /* toggle the bit because as soon as the first run length is 597 * fetched the bit will be toggled again */ 598 bit ^= 1; 599 while (current_superblock < s->superblock_count) { 600 601 /* skip any superblocks already marked as partially coded */ 602 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) { 603 604 if (current_run-- == 0) { 605 bit ^= 1; 606 current_run = get_vlc2(gb, 607 s->superblock_run_length_vlc.table, 6, 2); 608 if (current_run == 33) 609 current_run += get_bits(gb, 12); 610 } 611 s->superblock_coding[current_superblock] = 2*bit; 612 } 613 current_superblock++; 614 } 615 } 616 617 /* if there were partial blocks, initialize bitstream for 618 * unpacking fragment codings */ 619 if (decode_partial_blocks) { 620 621 current_run = 0; 622 bit = get_bits1(gb); 623 /* toggle the bit because as soon as the first run length is 624 * fetched the bit will be toggled again */ 625 bit ^= 1; 626 } 627 } 628 629 /* figure out which fragments are coded; iterate through each 630 * superblock (all planes) */ 631 s->coded_fragment_list_index = 0; 632 s->next_coeff= s->coeffs + s->fragment_count; 633 s->first_coded_y_fragment = s->first_coded_c_fragment = 0; 634 s->last_coded_y_fragment = s->last_coded_c_fragment = -1; 635 first_c_fragment_seen = 0; 636 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count); 637 for (i = 0; i < s->superblock_count; i++) { 638 639 /* iterate through all 16 fragments in a superblock */ 640 for (j = 0; j < 16; j++) { 641 642 /* if the fragment is in bounds, check its coding status */ 643 current_fragment = s->superblock_fragments[i * 16 + j]; 644 if (current_fragment >= s->fragment_count) { 645 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n", 646 current_fragment, s->fragment_count); 647 return 1; 648 } 649 if (current_fragment != -1) { 650 if (s->superblock_coding[i] == SB_NOT_CODED) { 651 652 /* copy all the fragments from the prior frame */ 653 s->all_fragments[current_fragment].coding_method = 654 MODE_COPY; 655 656 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) { 657 658 /* fragment may or may not be coded; this is the case 659 * that cares about the fragment coding runs */ 660 if (current_run-- == 0) { 661 bit ^= 1; 662 current_run = get_vlc2(gb, 663 s->fragment_run_length_vlc.table, 5, 2); 664 } 665 666 if (bit) { 667 /* default mode; actual mode will be decoded in 668 * the next phase */ 669 s->all_fragments[current_fragment].coding_method = 670 MODE_INTER_NO_MV; 671 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment; 672 s->coded_fragment_list[s->coded_fragment_list_index] = 673 current_fragment; 674 if ((current_fragment >= s->fragment_start[1]) && 675 (s->last_coded_y_fragment == -1) && 676 (!first_c_fragment_seen)) { 677 s->first_coded_c_fragment = s->coded_fragment_list_index; 678 s->last_coded_y_fragment = s->first_coded_c_fragment - 1; 679 first_c_fragment_seen = 1; 680 } 681 s->coded_fragment_list_index++; 682 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV; 683 } else { 684 /* not coded; copy this fragment from the prior frame */ 685 s->all_fragments[current_fragment].coding_method = 686 MODE_COPY; 687 } 688 689 } else { 690 691 /* fragments are fully coded in this superblock; actual 692 * coding will be determined in next step */ 693 s->all_fragments[current_fragment].coding_method = 694 MODE_INTER_NO_MV; 695 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment; 696 s->coded_fragment_list[s->coded_fragment_list_index] = 697 current_fragment; 698 if ((current_fragment >= s->fragment_start[1]) && 699 (s->last_coded_y_fragment == -1) && 700 (!first_c_fragment_seen)) { 701 s->first_coded_c_fragment = s->coded_fragment_list_index; 702 s->last_coded_y_fragment = s->first_coded_c_fragment - 1; 703 first_c_fragment_seen = 1; 704 } 705 s->coded_fragment_list_index++; 706 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV; 707 } 708 } 709 } 710 } 711 712 if (!first_c_fragment_seen) 713 /* only Y fragments coded in this frame */ 714 s->last_coded_y_fragment = s->coded_fragment_list_index - 1; 715 else 716 /* end the list of coded C fragments */ 717 s->last_coded_c_fragment = s->coded_fragment_list_index - 1; 718 719 return 0; 720} 721 722/* 723 * This function unpacks all the coding mode data for individual macroblocks 724 * from the bitstream. 725 */ 726static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb) 727{ 728 int i, j, k; 729 int scheme; 730 int current_macroblock; 731 int current_fragment; 732 int coding_mode; 733 int custom_mode_alphabet[CODING_MODE_COUNT]; 734 735 if (s->keyframe) { 736 for (i = 0; i < s->fragment_count; i++) 737 s->all_fragments[i].coding_method = MODE_INTRA; 738 739 } else { 740 741 /* fetch the mode coding scheme for this frame */ 742 scheme = get_bits(gb, 3); 743 744 /* is it a custom coding scheme? */ 745 if (scheme == 0) { 746 for (i = 0; i < 8; i++) 747 custom_mode_alphabet[i] = MODE_INTER_NO_MV; 748 for (i = 0; i < 8; i++) 749 custom_mode_alphabet[get_bits(gb, 3)] = i; 750 } 751 752 /* iterate through all of the macroblocks that contain 1 or more 753 * coded fragments */ 754 for (i = 0; i < s->u_superblock_start; i++) { 755 756 for (j = 0; j < 4; j++) { 757 current_macroblock = s->superblock_macroblocks[i * 4 + j]; 758 if ((current_macroblock == -1) || 759 (s->macroblock_coding[current_macroblock] == MODE_COPY)) 760 continue; 761 if (current_macroblock >= s->macroblock_count) { 762 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n", 763 current_macroblock, s->macroblock_count); 764 return 1; 765 } 766 767 /* mode 7 means get 3 bits for each coding mode */ 768 if (scheme == 7) 769 coding_mode = get_bits(gb, 3); 770 else if(scheme == 0) 771 coding_mode = custom_mode_alphabet 772 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)]; 773 else 774 coding_mode = ModeAlphabet[scheme-1] 775 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)]; 776 777 s->macroblock_coding[current_macroblock] = coding_mode; 778 for (k = 0; k < 6; k++) { 779 current_fragment = 780 s->macroblock_fragments[current_macroblock * 6 + k]; 781 if (current_fragment == -1) 782 continue; 783 if (current_fragment >= s->fragment_count) { 784 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n", 785 current_fragment, s->fragment_count); 786 return 1; 787 } 788 if (s->all_fragments[current_fragment].coding_method != 789 MODE_COPY) 790 s->all_fragments[current_fragment].coding_method = 791 coding_mode; 792 } 793 } 794 } 795 } 796 797 return 0; 798} 799 800/* 801 * This function unpacks all the motion vectors for the individual 802 * macroblocks from the bitstream. 803 */ 804static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb) 805{ 806 int i, j, k, l; 807 int coding_mode; 808 int motion_x[6]; 809 int motion_y[6]; 810 int last_motion_x = 0; 811 int last_motion_y = 0; 812 int prior_last_motion_x = 0; 813 int prior_last_motion_y = 0; 814 int current_macroblock; 815 int current_fragment; 816 817 if (s->keyframe) 818 return 0; 819 820 memset(motion_x, 0, 6 * sizeof(int)); 821 memset(motion_y, 0, 6 * sizeof(int)); 822 823 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */ 824 coding_mode = get_bits1(gb); 825 826 /* iterate through all of the macroblocks that contain 1 or more 827 * coded fragments */ 828 for (i = 0; i < s->u_superblock_start; i++) { 829 830 for (j = 0; j < 4; j++) { 831 current_macroblock = s->superblock_macroblocks[i * 4 + j]; 832 if ((current_macroblock == -1) || 833 (s->macroblock_coding[current_macroblock] == MODE_COPY)) 834 continue; 835 if (current_macroblock >= s->macroblock_count) { 836 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n", 837 current_macroblock, s->macroblock_count); 838 return 1; 839 } 840 841 current_fragment = s->macroblock_fragments[current_macroblock * 6]; 842 if (current_fragment >= s->fragment_count) { 843 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n", 844 current_fragment, s->fragment_count); 845 return 1; 846 } 847 switch (s->macroblock_coding[current_macroblock]) { 848 849 case MODE_INTER_PLUS_MV: 850 case MODE_GOLDEN_MV: 851 /* all 6 fragments use the same motion vector */ 852 if (coding_mode == 0) { 853 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; 854 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; 855 } else { 856 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)]; 857 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)]; 858 } 859 860 for (k = 1; k < 6; k++) { 861 motion_x[k] = motion_x[0]; 862 motion_y[k] = motion_y[0]; 863 } 864 865 /* vector maintenance, only on MODE_INTER_PLUS_MV */ 866 if (s->macroblock_coding[current_macroblock] == 867 MODE_INTER_PLUS_MV) { 868 prior_last_motion_x = last_motion_x; 869 prior_last_motion_y = last_motion_y; 870 last_motion_x = motion_x[0]; 871 last_motion_y = motion_y[0]; 872 } 873 break; 874 875 case MODE_INTER_FOURMV: 876 /* vector maintenance */ 877 prior_last_motion_x = last_motion_x; 878 prior_last_motion_y = last_motion_y; 879 880 /* fetch 4 vectors from the bitstream, one for each 881 * Y fragment, then average for the C fragment vectors */ 882 motion_x[4] = motion_y[4] = 0; 883 for (k = 0; k < 4; k++) { 884 for (l = 0; l < s->coded_fragment_list_index; l++) 885 if (s->coded_fragment_list[l] == s->macroblock_fragments[6*current_macroblock + k]) 886 break; 887 if (l < s->coded_fragment_list_index) { 888 if (coding_mode == 0) { 889 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; 890 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)]; 891 } else { 892 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)]; 893 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)]; 894 } 895 last_motion_x = motion_x[k]; 896 last_motion_y = motion_y[k]; 897 } else { 898 motion_x[k] = 0; 899 motion_y[k] = 0; 900 } 901 motion_x[4] += motion_x[k]; 902 motion_y[4] += motion_y[k]; 903 } 904 905 motion_x[5]= 906 motion_x[4]= RSHIFT(motion_x[4], 2); 907 motion_y[5]= 908 motion_y[4]= RSHIFT(motion_y[4], 2); 909 break; 910 911 case MODE_INTER_LAST_MV: 912 /* all 6 fragments use the last motion vector */ 913 motion_x[0] = last_motion_x; 914 motion_y[0] = last_motion_y; 915 for (k = 1; k < 6; k++) { 916 motion_x[k] = motion_x[0]; 917 motion_y[k] = motion_y[0]; 918 } 919 920 /* no vector maintenance (last vector remains the 921 * last vector) */ 922 break; 923 924 case MODE_INTER_PRIOR_LAST: 925 /* all 6 fragments use the motion vector prior to the 926 * last motion vector */ 927 motion_x[0] = prior_last_motion_x; 928 motion_y[0] = prior_last_motion_y; 929 for (k = 1; k < 6; k++) { 930 motion_x[k] = motion_x[0]; 931 motion_y[k] = motion_y[0]; 932 } 933 934 /* vector maintenance */ 935 prior_last_motion_x = last_motion_x; 936 prior_last_motion_y = last_motion_y; 937 last_motion_x = motion_x[0]; 938 last_motion_y = motion_y[0]; 939 break; 940 941 default: 942 /* covers intra, inter without MV, golden without MV */ 943 memset(motion_x, 0, 6 * sizeof(int)); 944 memset(motion_y, 0, 6 * sizeof(int)); 945 946 /* no vector maintenance */ 947 break; 948 } 949 950 /* assign the motion vectors to the correct fragments */ 951 for (k = 0; k < 6; k++) { 952 current_fragment = 953 s->macroblock_fragments[current_macroblock * 6 + k]; 954 if (current_fragment == -1) 955 continue; 956 if (current_fragment >= s->fragment_count) { 957 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n", 958 current_fragment, s->fragment_count); 959 return 1; 960 } 961 s->all_fragments[current_fragment].motion_x = motion_x[k]; 962 s->all_fragments[current_fragment].motion_y = motion_y[k]; 963 } 964 } 965 } 966 967 return 0; 968} 969 970/* 971 * This function is called by unpack_dct_coeffs() to extract the VLCs from 972 * the bitstream. The VLCs encode tokens which are used to unpack DCT 973 * data. This function unpacks all the VLCs for either the Y plane or both 974 * C planes, and is called for DC coefficients or different AC coefficient 975 * levels (since different coefficient types require different VLC tables. 976 * 977 * This function returns a residual eob run. E.g, if a particular token gave 978 * instructions to EOB the next 5 fragments and there were only 2 fragments 979 * left in the current fragment range, 3 would be returned so that it could 980 * be passed into the next call to this same function. 981 */ 982static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb, 983 VLC *table, int coeff_index, 984 int first_fragment, int last_fragment, 985 int eob_run) 986{ 987 int i; 988 int token; 989 int zero_run = 0; 990 DCTELEM coeff = 0; 991 Vp3Fragment *fragment; 992 uint8_t *perm= s->scantable.permutated; 993 int bits_to_get; 994 995 if ((first_fragment >= s->fragment_count) || 996 (last_fragment >= s->fragment_count)) { 997 998 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n", 999 first_fragment, last_fragment); 1000 return 0; 1001 } 1002 1003 for (i = first_fragment; i <= last_fragment; i++) { 1004 int fragment_num = s->coded_fragment_list[i]; 1005 1006 if (s->coeff_counts[fragment_num] > coeff_index) 1007 continue; 1008 fragment = &s->all_fragments[fragment_num]; 1009 1010 if (!eob_run) { 1011 /* decode a VLC into a token */ 1012 token = get_vlc2(gb, table->table, 5, 3); 1013 /* use the token to get a zero run, a coefficient, and an eob run */ 1014 if (token <= 6) { 1015 eob_run = eob_run_base[token]; 1016 if (eob_run_get_bits[token]) 1017 eob_run += get_bits(gb, eob_run_get_bits[token]); 1018 coeff = zero_run = 0; 1019 } else { 1020 bits_to_get = coeff_get_bits[token]; 1021 if (!bits_to_get) 1022 coeff = coeff_tables[token][0]; 1023 else 1024 coeff = coeff_tables[token][get_bits(gb, bits_to_get)]; 1025 1026 zero_run = zero_run_base[token]; 1027 if (zero_run_get_bits[token]) 1028 zero_run += get_bits(gb, zero_run_get_bits[token]); 1029 } 1030 } 1031 1032 if (!eob_run) { 1033 s->coeff_counts[fragment_num] += zero_run; 1034 if (s->coeff_counts[fragment_num] < 64){ 1035 fragment->next_coeff->coeff= coeff; 1036 fragment->next_coeff->index= perm[s->coeff_counts[fragment_num]++]; //FIXME perm here already? 1037 fragment->next_coeff->next= s->next_coeff; 1038 s->next_coeff->next=NULL; 1039 fragment->next_coeff= s->next_coeff++; 1040 } 1041 } else { 1042 s->coeff_counts[fragment_num] |= 128; 1043 eob_run--; 1044 } 1045 } 1046 1047 return eob_run; 1048} 1049 1050/* 1051 * This function unpacks all of the DCT coefficient data from the 1052 * bitstream. 1053 */ 1054static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb) 1055{ 1056 int i; 1057 int dc_y_table; 1058 int dc_c_table; 1059 int ac_y_table; 1060 int ac_c_table; 1061 int residual_eob_run = 0; 1062 1063 /* fetch the DC table indexes */ 1064 dc_y_table = get_bits(gb, 4); 1065 dc_c_table = get_bits(gb, 4); 1066 1067 /* unpack the Y plane DC coefficients */ 1068 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0, 1069 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run); 1070 1071 /* unpack the C plane DC coefficients */ 1072 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0, 1073 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run); 1074 1075 /* fetch the AC table indexes */ 1076 ac_y_table = get_bits(gb, 4); 1077 ac_c_table = get_bits(gb, 4); 1078 1079 /* unpack the group 1 AC coefficients (coeffs 1-5) */ 1080 for (i = 1; i <= 5; i++) { 1081 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i, 1082 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run); 1083 1084 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i, 1085 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run); 1086 } 1087 1088 /* unpack the group 2 AC coefficients (coeffs 6-14) */ 1089 for (i = 6; i <= 14; i++) { 1090 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i, 1091 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run); 1092 1093 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i, 1094 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run); 1095 } 1096 1097 /* unpack the group 3 AC coefficients (coeffs 15-27) */ 1098 for (i = 15; i <= 27; i++) { 1099 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i, 1100 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run); 1101 1102 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i, 1103 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run); 1104 } 1105 1106 /* unpack the group 4 AC coefficients (coeffs 28-63) */ 1107 for (i = 28; i <= 63; i++) { 1108 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i, 1109 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run); 1110 1111 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i, 1112 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run); 1113 } 1114 1115 return 0; 1116} 1117 1118/* 1119 * This function reverses the DC prediction for each coded fragment in 1120 * the frame. Much of this function is adapted directly from the original 1121 * VP3 source code. 1122 */ 1123#define COMPATIBLE_FRAME(x) \ 1124 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type) 1125#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY) 1126#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this 1127 1128static void reverse_dc_prediction(Vp3DecodeContext *s, 1129 int first_fragment, 1130 int fragment_width, 1131 int fragment_height) 1132{ 1133 1134#define PUL 8 1135#define PU 4 1136#define PUR 2 1137#define PL 1 1138 1139 int x, y; 1140 int i = first_fragment; 1141 1142 int predicted_dc; 1143 1144 /* DC values for the left, up-left, up, and up-right fragments */ 1145 int vl, vul, vu, vur; 1146 1147 /* indexes for the left, up-left, up, and up-right fragments */ 1148 int l, ul, u, ur; 1149 1150 /* 1151 * The 6 fields mean: 1152 * 0: up-left multiplier 1153 * 1: up multiplier 1154 * 2: up-right multiplier 1155 * 3: left multiplier 1156 */ 1157 int predictor_transform[16][4] = { 1158 { 0, 0, 0, 0}, 1159 { 0, 0, 0,128}, // PL 1160 { 0, 0,128, 0}, // PUR 1161 { 0, 0, 53, 75}, // PUR|PL 1162 { 0,128, 0, 0}, // PU 1163 { 0, 64, 0, 64}, // PU|PL 1164 { 0,128, 0, 0}, // PU|PUR 1165 { 0, 0, 53, 75}, // PU|PUR|PL 1166 {128, 0, 0, 0}, // PUL 1167 { 0, 0, 0,128}, // PUL|PL 1168 { 64, 0, 64, 0}, // PUL|PUR 1169 { 0, 0, 53, 75}, // PUL|PUR|PL 1170 { 0,128, 0, 0}, // PUL|PU 1171 {-104,116, 0,116}, // PUL|PU|PL 1172 { 24, 80, 24, 0}, // PUL|PU|PUR 1173 {-104,116, 0,116} // PUL|PU|PUR|PL 1174 }; 1175 1176 /* This table shows which types of blocks can use other blocks for 1177 * prediction. For example, INTRA is the only mode in this table to 1178 * have a frame number of 0. That means INTRA blocks can only predict 1179 * from other INTRA blocks. There are 2 golden frame coding types; 1180 * blocks encoding in these modes can only predict from other blocks 1181 * that were encoded with these 1 of these 2 modes. */ 1182 unsigned char compatible_frame[8] = { 1183 1, /* MODE_INTER_NO_MV */ 1184 0, /* MODE_INTRA */ 1185 1, /* MODE_INTER_PLUS_MV */ 1186 1, /* MODE_INTER_LAST_MV */ 1187 1, /* MODE_INTER_PRIOR_MV */ 1188 2, /* MODE_USING_GOLDEN */ 1189 2, /* MODE_GOLDEN_MV */ 1190 1 /* MODE_INTER_FOUR_MV */ 1191 }; 1192 int current_frame_type; 1193 1194 /* there is a last DC predictor for each of the 3 frame types */ 1195 short last_dc[3]; 1196 1197 int transform = 0; 1198 1199 vul = vu = vur = vl = 0; 1200 last_dc[0] = last_dc[1] = last_dc[2] = 0; 1201 1202 /* for each fragment row... */ 1203 for (y = 0; y < fragment_height; y++) { 1204 1205 /* for each fragment in a row... */ 1206 for (x = 0; x < fragment_width; x++, i++) { 1207 1208 /* reverse prediction if this block was coded */ 1209 if (s->all_fragments[i].coding_method != MODE_COPY) { 1210 1211 current_frame_type = 1212 compatible_frame[s->all_fragments[i].coding_method]; 1213 1214 transform= 0; 1215 if(x){ 1216 l= i-1; 1217 vl = DC_COEFF(l); 1218 if(FRAME_CODED(l) && COMPATIBLE_FRAME(l)) 1219 transform |= PL; 1220 } 1221 if(y){ 1222 u= i-fragment_width; 1223 vu = DC_COEFF(u); 1224 if(FRAME_CODED(u) && COMPATIBLE_FRAME(u)) 1225 transform |= PU; 1226 if(x){ 1227 ul= i-fragment_width-1; 1228 vul = DC_COEFF(ul); 1229 if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul)) 1230 transform |= PUL; 1231 } 1232 if(x + 1 < fragment_width){ 1233 ur= i-fragment_width+1; 1234 vur = DC_COEFF(ur); 1235 if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur)) 1236 transform |= PUR; 1237 } 1238 } 1239 1240 if (transform == 0) { 1241 1242 /* if there were no fragments to predict from, use last 1243 * DC saved */ 1244 predicted_dc = last_dc[current_frame_type]; 1245 } else { 1246 1247 /* apply the appropriate predictor transform */ 1248 predicted_dc = 1249 (predictor_transform[transform][0] * vul) + 1250 (predictor_transform[transform][1] * vu) + 1251 (predictor_transform[transform][2] * vur) + 1252 (predictor_transform[transform][3] * vl); 1253 1254 predicted_dc /= 128; 1255 1256 /* check for outranging on the [ul u l] and 1257 * [ul u ur l] predictors */ 1258 if ((transform == 13) || (transform == 15)) { 1259 if (FFABS(predicted_dc - vu) > 128) 1260 predicted_dc = vu; 1261 else if (FFABS(predicted_dc - vl) > 128) 1262 predicted_dc = vl; 1263 else if (FFABS(predicted_dc - vul) > 128) 1264 predicted_dc = vul; 1265 } 1266 } 1267 1268 /* at long last, apply the predictor */ 1269 if(s->coeffs[i].index){ 1270 *s->next_coeff= s->coeffs[i]; 1271 s->coeffs[i].index=0; 1272 s->coeffs[i].coeff=0; 1273 s->coeffs[i].next= s->next_coeff++; 1274 } 1275 s->coeffs[i].coeff += predicted_dc; 1276 /* save the DC */ 1277 last_dc[current_frame_type] = DC_COEFF(i); 1278 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){ 1279 s->coeff_counts[i]= 129; 1280// s->all_fragments[i].next_coeff= s->next_coeff; 1281 s->coeffs[i].next= s->next_coeff; 1282 (s->next_coeff++)->next=NULL; 1283 } 1284 } 1285 } 1286 } 1287} 1288 1289/* 1290 * Perform the final rendering for a particular slice of data. 1291 * The slice number ranges from 0..(macroblock_height - 1). 1292 */ 1293static void render_slice(Vp3DecodeContext *s, int slice) 1294{ 1295 int x; 1296 int16_t *dequantizer; 1297 DECLARE_ALIGNED_16(DCTELEM, block[64]); 1298 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef; 1299 int motion_halfpel_index; 1300 uint8_t *motion_source; 1301 int plane; 1302 int current_macroblock_entry = slice * s->macroblock_width * 6; 1303 1304 if (slice >= s->macroblock_height) 1305 return; 1306 1307 for (plane = 0; plane < 3; plane++) { 1308 uint8_t *output_plane = s->current_frame.data [plane]; 1309 uint8_t * last_plane = s-> last_frame.data [plane]; 1310 uint8_t *golden_plane = s-> golden_frame.data [plane]; 1311 int stride = s->current_frame.linesize[plane]; 1312 int plane_width = s->width >> !!plane; 1313 int plane_height = s->height >> !!plane; 1314 int y = slice * FRAGMENT_PIXELS << !plane ; 1315 int slice_height = y + (FRAGMENT_PIXELS << !plane); 1316 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane]; 1317 1318 if (!s->flipped_image) stride = -stride; 1319 1320 1321 if(FFABS(stride) > 2048) 1322 return; //various tables are fixed size 1323 1324 /* for each fragment row in the slice (both of them)... */ 1325 for (; y < slice_height; y += 8) { 1326 1327 /* for each fragment in a row... */ 1328 for (x = 0; x < plane_width; x += 8, i++) { 1329 1330 if ((i < 0) || (i >= s->fragment_count)) { 1331 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i); 1332 return; 1333 } 1334 1335 /* transform if this block was coded */ 1336 if ((s->all_fragments[i].coding_method != MODE_COPY) && 1337 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) { 1338 1339 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) || 1340 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV)) 1341 motion_source= golden_plane; 1342 else 1343 motion_source= last_plane; 1344 1345 motion_source += s->all_fragments[i].first_pixel; 1346 motion_halfpel_index = 0; 1347 1348 /* sort out the motion vector if this fragment is coded 1349 * using a motion vector method */ 1350 if ((s->all_fragments[i].coding_method > MODE_INTRA) && 1351 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) { 1352 int src_x, src_y; 1353 motion_x = s->all_fragments[i].motion_x; 1354 motion_y = s->all_fragments[i].motion_y; 1355 if(plane){ 1356 motion_x= (motion_x>>1) | (motion_x&1); 1357 motion_y= (motion_y>>1) | (motion_y&1); 1358 } 1359 1360 src_x= (motion_x>>1) + x; 1361 src_y= (motion_y>>1) + y; 1362 if ((motion_x == 127) || (motion_y == 127)) 1363 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y); 1364 1365 motion_halfpel_index = motion_x & 0x01; 1366 motion_source += (motion_x >> 1); 1367 1368 motion_halfpel_index |= (motion_y & 0x01) << 1; 1369 motion_source += ((motion_y >> 1) * stride); 1370 1371 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){ 1372 uint8_t *temp= s->edge_emu_buffer; 1373 if(stride<0) temp -= 9*stride; 1374 else temp += 9*stride; 1375 1376 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height); 1377 motion_source= temp; 1378 } 1379 } 1380 1381 1382 /* first, take care of copying a block from either the 1383 * previous or the golden frame */ 1384 if (s->all_fragments[i].coding_method != MODE_INTRA) { 1385 /* Note, it is possible to implement all MC cases with 1386 put_no_rnd_pixels_l2 which would look more like the 1387 VP3 source but this would be slower as 1388 put_no_rnd_pixels_tab is better optimzed */ 1389 if(motion_halfpel_index != 3){ 1390 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index]( 1391 output_plane + s->all_fragments[i].first_pixel, 1392 motion_source, stride, 8); 1393 }else{ 1394 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1 1395 s->dsp.put_no_rnd_pixels_l2[1]( 1396 output_plane + s->all_fragments[i].first_pixel, 1397 motion_source - d, 1398 motion_source + stride + 1 + d, 1399 stride, 8); 1400 } 1401 dequantizer = s->qmat[1][plane]; 1402 }else{ 1403 dequantizer = s->qmat[0][plane]; 1404 } 1405 1406 /* dequantize the DCT coefficients */ 1407 if(s->avctx->idct_algo==FF_IDCT_VP3){ 1408 Coeff *coeff= s->coeffs + i; 1409 s->dsp.clear_block(block); 1410 while(coeff->next){ 1411 block[coeff->index]= coeff->coeff * dequantizer[coeff->index]; 1412 coeff= coeff->next; 1413 } 1414 }else{ 1415 Coeff *coeff= s->coeffs + i; 1416 s->dsp.clear_block(block); 1417 while(coeff->next){ 1418 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2; 1419 coeff= coeff->next; 1420 } 1421 } 1422 1423 /* invert DCT and place (or add) in final output */ 1424 1425 if (s->all_fragments[i].coding_method == MODE_INTRA) { 1426 if(s->avctx->idct_algo!=FF_IDCT_VP3) 1427 block[0] += 128<<3; 1428 s->dsp.idct_put( 1429 output_plane + s->all_fragments[i].first_pixel, 1430 stride, 1431 block); 1432 } else { 1433 s->dsp.idct_add( 1434 output_plane + s->all_fragments[i].first_pixel, 1435 stride, 1436 block); 1437 } 1438 } else { 1439 1440 /* copy directly from the previous frame */ 1441 s->dsp.put_pixels_tab[1][0]( 1442 output_plane + s->all_fragments[i].first_pixel, 1443 last_plane + s->all_fragments[i].first_pixel, 1444 stride, 8); 1445 1446 } 1447#if 0 1448 /* perform the left edge filter if: 1449 * - the fragment is not on the left column 1450 * - the fragment is coded in this frame 1451 * - the fragment is not coded in this frame but the left 1452 * fragment is coded in this frame (this is done instead 1453 * of a right edge filter when rendering the left fragment 1454 * since this fragment is not available yet) */ 1455 if ((x > 0) && 1456 ((s->all_fragments[i].coding_method != MODE_COPY) || 1457 ((s->all_fragments[i].coding_method == MODE_COPY) && 1458 (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) { 1459 horizontal_filter( 1460 output_plane + s->all_fragments[i].first_pixel + 7*stride, 1461 -stride, s->bounding_values_array + 127); 1462 } 1463 1464 /* perform the top edge filter if: 1465 * - the fragment is not on the top row 1466 * - the fragment is coded in this frame 1467 * - the fragment is not coded in this frame but the above 1468 * fragment is coded in this frame (this is done instead 1469 * of a bottom edge filter when rendering the above 1470 * fragment since this fragment is not available yet) */ 1471 if ((y > 0) && 1472 ((s->all_fragments[i].coding_method != MODE_COPY) || 1473 ((s->all_fragments[i].coding_method == MODE_COPY) && 1474 (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) { 1475 vertical_filter( 1476 output_plane + s->all_fragments[i].first_pixel - stride, 1477 -stride, s->bounding_values_array + 127); 1478 } 1479#endif 1480 } 1481 } 1482 } 1483 1484 /* this looks like a good place for slice dispatch... */ 1485 /* algorithm: 1486 * if (slice == s->macroblock_height - 1) 1487 * dispatch (both last slice & 2nd-to-last slice); 1488 * else if (slice > 0) 1489 * dispatch (slice - 1); 1490 */ 1491 1492 emms_c(); 1493} 1494 1495static void apply_loop_filter(Vp3DecodeContext *s) 1496{ 1497 int plane; 1498 int x, y; 1499 int *bounding_values= s->bounding_values_array+127; 1500 1501#if 0 1502 int bounding_values_array[256]; 1503 int filter_limit; 1504 1505 /* find the right loop limit value */ 1506 for (x = 63; x >= 0; x--) { 1507 if (vp31_ac_scale_factor[x] >= s->quality_index) 1508 break; 1509 } 1510 filter_limit = vp31_filter_limit_values[s->quality_index]; 1511 1512 /* set up the bounding values */ 1513 memset(bounding_values_array, 0, 256 * sizeof(int)); 1514 for (x = 0; x < filter_limit; x++) { 1515 bounding_values[-x - filter_limit] = -filter_limit + x; 1516 bounding_values[-x] = -x; 1517 bounding_values[x] = x; 1518 bounding_values[x + filter_limit] = filter_limit - x; 1519 } 1520#endif 1521 1522 for (plane = 0; plane < 3; plane++) { 1523 int width = s->fragment_width >> !!plane; 1524 int height = s->fragment_height >> !!plane; 1525 int fragment = s->fragment_start [plane]; 1526 int stride = s->current_frame.linesize[plane]; 1527 uint8_t *plane_data = s->current_frame.data [plane]; 1528 if (!s->flipped_image) stride = -stride; 1529 1530 for (y = 0; y < height; y++) { 1531 1532 for (x = 0; x < width; x++) { 1533 /* do not perform left edge filter for left columns frags */ 1534 if ((x > 0) && 1535 (s->all_fragments[fragment].coding_method != MODE_COPY)) { 1536 s->dsp.vp3_h_loop_filter( 1537 plane_data + s->all_fragments[fragment].first_pixel, 1538 stride, bounding_values); 1539 } 1540 1541 /* do not perform top edge filter for top row fragments */ 1542 if ((y > 0) && 1543 (s->all_fragments[fragment].coding_method != MODE_COPY)) { 1544 s->dsp.vp3_v_loop_filter( 1545 plane_data + s->all_fragments[fragment].first_pixel, 1546 stride, bounding_values); 1547 } 1548 1549 /* do not perform right edge filter for right column 1550 * fragments or if right fragment neighbor is also coded 1551 * in this frame (it will be filtered in next iteration) */ 1552 if ((x < width - 1) && 1553 (s->all_fragments[fragment].coding_method != MODE_COPY) && 1554 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) { 1555 s->dsp.vp3_h_loop_filter( 1556 plane_data + s->all_fragments[fragment + 1].first_pixel, 1557 stride, bounding_values); 1558 } 1559 1560 /* do not perform bottom edge filter for bottom row 1561 * fragments or if bottom fragment neighbor is also coded 1562 * in this frame (it will be filtered in the next row) */ 1563 if ((y < height - 1) && 1564 (s->all_fragments[fragment].coding_method != MODE_COPY) && 1565 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) { 1566 s->dsp.vp3_v_loop_filter( 1567 plane_data + s->all_fragments[fragment + width].first_pixel, 1568 stride, bounding_values); 1569 } 1570 1571 fragment++; 1572 } 1573 } 1574 } 1575} 1576 1577/* 1578 * This function computes the first pixel addresses for each fragment. 1579 * This function needs to be invoked after the first frame is allocated 1580 * so that it has access to the plane strides. 1581 */ 1582static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s) 1583{ 1584#define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift 1585#define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0 1586 1587 int i, x, y; 1588 const int y_inc = s->flipped_image ? 1 : -1; 1589 1590 /* figure out the first pixel addresses for each of the fragments */ 1591 /* Y plane */ 1592 i = 0; 1593 for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) { 1594 for (x = 0; x < s->fragment_width; x++) { 1595 s->all_fragments[i++].first_pixel = 1596 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS - 1597 s->golden_frame.linesize[0] + 1598 x * FRAGMENT_PIXELS; 1599 } 1600 } 1601 1602 /* U plane */ 1603 i = s->fragment_start[1]; 1604 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) { 1605 for (x = 0; x < s->fragment_width / 2; x++) { 1606 s->all_fragments[i++].first_pixel = 1607 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS - 1608 s->golden_frame.linesize[1] + 1609 x * FRAGMENT_PIXELS; 1610 } 1611 } 1612 1613 /* V plane */ 1614 i = s->fragment_start[2]; 1615 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) { 1616 for (x = 0; x < s->fragment_width / 2; x++) { 1617 s->all_fragments[i++].first_pixel = 1618 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS - 1619 s->golden_frame.linesize[2] + 1620 x * FRAGMENT_PIXELS; 1621 } 1622 } 1623} 1624 1625/* 1626 * This is the ffmpeg/libavcodec API init function. 1627 */ 1628static av_cold int vp3_decode_init(AVCodecContext *avctx) 1629{ 1630 Vp3DecodeContext *s = avctx->priv_data; 1631 int i, inter, plane; 1632 int c_width; 1633 int c_height; 1634 int y_superblock_count; 1635 int c_superblock_count; 1636 1637 if (avctx->codec_tag == MKTAG('V','P','3','0')) 1638 s->version = 0; 1639 else 1640 s->version = 1; 1641 1642 s->avctx = avctx; 1643 s->width = (avctx->width + 15) & 0xFFFFFFF0; 1644 s->height = (avctx->height + 15) & 0xFFFFFFF0; 1645 avctx->pix_fmt = PIX_FMT_YUV420P; 1646 if(avctx->idct_algo==FF_IDCT_AUTO) 1647 avctx->idct_algo=FF_IDCT_VP3; 1648 dsputil_init(&s->dsp, avctx); 1649 1650 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct); 1651 1652 /* initialize to an impossible value which will force a recalculation 1653 * in the first frame decode */ 1654 s->quality_index = -1; 1655 1656 s->y_superblock_width = (s->width + 31) / 32; 1657 s->y_superblock_height = (s->height + 31) / 32; 1658 y_superblock_count = s->y_superblock_width * s->y_superblock_height; 1659 1660 /* work out the dimensions for the C planes */ 1661 c_width = s->width / 2; 1662 c_height = s->height / 2; 1663 s->c_superblock_width = (c_width + 31) / 32; 1664 s->c_superblock_height = (c_height + 31) / 32; 1665 c_superblock_count = s->c_superblock_width * s->c_superblock_height; 1666 1667 s->superblock_count = y_superblock_count + (c_superblock_count * 2); 1668 s->u_superblock_start = y_superblock_count; 1669 s->v_superblock_start = s->u_superblock_start + c_superblock_count; 1670 s->superblock_coding = av_malloc(s->superblock_count); 1671 1672 s->macroblock_width = (s->width + 15) / 16; 1673 s->macroblock_height = (s->height + 15) / 16; 1674 s->macroblock_count = s->macroblock_width * s->macroblock_height; 1675 1676 s->fragment_width = s->width / FRAGMENT_PIXELS; 1677 s->fragment_height = s->height / FRAGMENT_PIXELS; 1678 1679 /* fragment count covers all 8x8 blocks for all 3 planes */ 1680 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2; 1681 s->fragment_start[1] = s->fragment_width * s->fragment_height; 1682 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4; 1683 1684 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment)); 1685 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts)); 1686 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65); 1687 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int)); 1688 s->pixel_addresses_initialized = 0; 1689 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts || 1690 !s->coeffs || !s->coded_fragment_list) { 1691 vp3_decode_end(avctx); 1692 return -1; 1693 } 1694 1695 if (!s->theora_tables) 1696 { 1697 for (i = 0; i < 64; i++) { 1698 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i]; 1699 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i]; 1700 s->base_matrix[0][i] = vp31_intra_y_dequant[i]; 1701 s->base_matrix[1][i] = vp31_intra_c_dequant[i]; 1702 s->base_matrix[2][i] = vp31_inter_dequant[i]; 1703 s->filter_limit_values[i] = vp31_filter_limit_values[i]; 1704 } 1705 1706 for(inter=0; inter<2; inter++){ 1707 for(plane=0; plane<3; plane++){ 1708 s->qr_count[inter][plane]= 1; 1709 s->qr_size [inter][plane][0]= 63; 1710 s->qr_base [inter][plane][0]= 1711 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter; 1712 } 1713 } 1714 1715 /* init VLC tables */ 1716 for (i = 0; i < 16; i++) { 1717 1718 /* DC histograms */ 1719 init_vlc(&s->dc_vlc[i], 5, 32, 1720 &dc_bias[i][0][1], 4, 2, 1721 &dc_bias[i][0][0], 4, 2, 0); 1722 1723 /* group 1 AC histograms */ 1724 init_vlc(&s->ac_vlc_1[i], 5, 32, 1725 &ac_bias_0[i][0][1], 4, 2, 1726 &ac_bias_0[i][0][0], 4, 2, 0); 1727 1728 /* group 2 AC histograms */ 1729 init_vlc(&s->ac_vlc_2[i], 5, 32, 1730 &ac_bias_1[i][0][1], 4, 2, 1731 &ac_bias_1[i][0][0], 4, 2, 0); 1732 1733 /* group 3 AC histograms */ 1734 init_vlc(&s->ac_vlc_3[i], 5, 32, 1735 &ac_bias_2[i][0][1], 4, 2, 1736 &ac_bias_2[i][0][0], 4, 2, 0); 1737 1738 /* group 4 AC histograms */ 1739 init_vlc(&s->ac_vlc_4[i], 5, 32, 1740 &ac_bias_3[i][0][1], 4, 2, 1741 &ac_bias_3[i][0][0], 4, 2, 0); 1742 } 1743 } else { 1744 for (i = 0; i < 16; i++) { 1745 1746 /* DC histograms */ 1747 if (init_vlc(&s->dc_vlc[i], 5, 32, 1748 &s->huffman_table[i][0][1], 4, 2, 1749 &s->huffman_table[i][0][0], 4, 2, 0) < 0) 1750 goto vlc_fail; 1751 1752 /* group 1 AC histograms */ 1753 if (init_vlc(&s->ac_vlc_1[i], 5, 32, 1754 &s->huffman_table[i+16][0][1], 4, 2, 1755 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0) 1756 goto vlc_fail; 1757 1758 /* group 2 AC histograms */ 1759 if (init_vlc(&s->ac_vlc_2[i], 5, 32, 1760 &s->huffman_table[i+16*2][0][1], 4, 2, 1761 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0) 1762 goto vlc_fail; 1763 1764 /* group 3 AC histograms */ 1765 if (init_vlc(&s->ac_vlc_3[i], 5, 32, 1766 &s->huffman_table[i+16*3][0][1], 4, 2, 1767 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0) 1768 goto vlc_fail; 1769 1770 /* group 4 AC histograms */ 1771 if (init_vlc(&s->ac_vlc_4[i], 5, 32, 1772 &s->huffman_table[i+16*4][0][1], 4, 2, 1773 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0) 1774 goto vlc_fail; 1775 } 1776 } 1777 1778 init_vlc(&s->superblock_run_length_vlc, 6, 34, 1779 &superblock_run_length_vlc_table[0][1], 4, 2, 1780 &superblock_run_length_vlc_table[0][0], 4, 2, 0); 1781 1782 init_vlc(&s->fragment_run_length_vlc, 5, 30, 1783 &fragment_run_length_vlc_table[0][1], 4, 2, 1784 &fragment_run_length_vlc_table[0][0], 4, 2, 0); 1785 1786 init_vlc(&s->mode_code_vlc, 3, 8, 1787 &mode_code_vlc_table[0][1], 2, 1, 1788 &mode_code_vlc_table[0][0], 2, 1, 0); 1789 1790 init_vlc(&s->motion_vector_vlc, 6, 63, 1791 &motion_vector_vlc_table[0][1], 2, 1, 1792 &motion_vector_vlc_table[0][0], 2, 1, 0); 1793 1794 /* work out the block mapping tables */ 1795 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int)); 1796 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int)); 1797 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int)); 1798 s->macroblock_coding = av_malloc(s->macroblock_count + 1); 1799 if (!s->superblock_fragments || !s->superblock_macroblocks || 1800 !s->macroblock_fragments || !s->macroblock_coding) { 1801 vp3_decode_end(avctx); 1802 return -1; 1803 } 1804 init_block_mapping(s); 1805 1806 for (i = 0; i < 3; i++) { 1807 s->current_frame.data[i] = NULL; 1808 s->last_frame.data[i] = NULL; 1809 s->golden_frame.data[i] = NULL; 1810 } 1811 1812 return 0; 1813 1814vlc_fail: 1815 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n"); 1816 return -1; 1817} 1818 1819/* 1820 * This is the ffmpeg/libavcodec API frame decode function. 1821 */ 1822static int vp3_decode_frame(AVCodecContext *avctx, 1823 void *data, int *data_size, 1824 const uint8_t *buf, int buf_size) 1825{ 1826 Vp3DecodeContext *s = avctx->priv_data; 1827 GetBitContext gb; 1828 static int counter = 0; 1829 int i; 1830 1831 init_get_bits(&gb, buf, buf_size * 8); 1832 1833 if (s->theora && get_bits1(&gb)) 1834 { 1835 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n"); 1836 return -1; 1837 } 1838 1839 s->keyframe = !get_bits1(&gb); 1840 if (!s->theora) 1841 skip_bits(&gb, 1); 1842 s->last_quality_index = s->quality_index; 1843 1844 s->nqis=0; 1845 do{ 1846 s->qis[s->nqis++]= get_bits(&gb, 6); 1847 } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb)); 1848 1849 s->quality_index= s->qis[0]; 1850 1851 if (s->avctx->debug & FF_DEBUG_PICT_INFO) 1852 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n", 1853 s->keyframe?"key":"", counter, s->quality_index); 1854 counter++; 1855 1856 if (s->quality_index != s->last_quality_index) { 1857 init_dequantizer(s); 1858 init_loop_filter(s); 1859 } 1860 1861 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe) 1862 return buf_size; 1863 1864 if (s->keyframe) { 1865 if (!s->theora) 1866 { 1867 skip_bits(&gb, 4); /* width code */ 1868 skip_bits(&gb, 4); /* height code */ 1869 if (s->version) 1870 { 1871 s->version = get_bits(&gb, 5); 1872 if (counter == 1) 1873 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version); 1874 } 1875 } 1876 if (s->version || s->theora) 1877 { 1878 if (get_bits1(&gb)) 1879 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n"); 1880 skip_bits(&gb, 2); /* reserved? */ 1881 } 1882 1883 if (s->last_frame.data[0] == s->golden_frame.data[0]) { 1884 if (s->golden_frame.data[0]) 1885 avctx->release_buffer(avctx, &s->golden_frame); 1886 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */ 1887 } else { 1888 if (s->golden_frame.data[0]) 1889 avctx->release_buffer(avctx, &s->golden_frame); 1890 if (s->last_frame.data[0]) 1891 avctx->release_buffer(avctx, &s->last_frame); 1892 } 1893 1894 s->golden_frame.reference = 3; 1895 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) { 1896 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n"); 1897 return -1; 1898 } 1899 1900 /* golden frame is also the current frame */ 1901 s->current_frame= s->golden_frame; 1902 1903 /* time to figure out pixel addresses? */ 1904 if (!s->pixel_addresses_initialized) 1905 { 1906 vp3_calculate_pixel_addresses(s); 1907 s->pixel_addresses_initialized = 1; 1908 } 1909 } else { 1910 /* allocate a new current frame */ 1911 s->current_frame.reference = 3; 1912 if (!s->pixel_addresses_initialized) { 1913 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n"); 1914 return -1; 1915 } 1916 if(avctx->get_buffer(avctx, &s->current_frame) < 0) { 1917 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n"); 1918 return -1; 1919 } 1920 } 1921 1922 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame 1923 s->current_frame.qstride= 0; 1924 1925 init_frame(s, &gb); 1926 1927 if (unpack_superblocks(s, &gb)){ 1928 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n"); 1929 return -1; 1930 } 1931 if (unpack_modes(s, &gb)){ 1932 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n"); 1933 return -1; 1934 } 1935 if (unpack_vectors(s, &gb)){ 1936 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n"); 1937 return -1; 1938 } 1939 if (unpack_dct_coeffs(s, &gb)){ 1940 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n"); 1941 return -1; 1942 } 1943 1944 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height); 1945 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) { 1946 reverse_dc_prediction(s, s->fragment_start[1], 1947 s->fragment_width / 2, s->fragment_height / 2); 1948 reverse_dc_prediction(s, s->fragment_start[2], 1949 s->fragment_width / 2, s->fragment_height / 2); 1950 } 1951 1952 for (i = 0; i < s->macroblock_height; i++) 1953 render_slice(s, i); 1954 1955 apply_loop_filter(s); 1956 1957 *data_size=sizeof(AVFrame); 1958 *(AVFrame*)data= s->current_frame; 1959 1960 /* release the last frame, if it is allocated and if it is not the 1961 * golden frame */ 1962 if ((s->last_frame.data[0]) && 1963 (s->last_frame.data[0] != s->golden_frame.data[0])) 1964 avctx->release_buffer(avctx, &s->last_frame); 1965 1966 /* shuffle frames (last = current) */ 1967 s->last_frame= s->current_frame; 1968 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */ 1969 1970 return buf_size; 1971} 1972 1973/* 1974 * This is the ffmpeg/libavcodec API module cleanup function. 1975 */ 1976static av_cold int vp3_decode_end(AVCodecContext *avctx) 1977{ 1978 Vp3DecodeContext *s = avctx->priv_data; 1979 int i; 1980 1981 av_free(s->superblock_coding); 1982 av_free(s->all_fragments); 1983 av_free(s->coeff_counts); 1984 av_free(s->coeffs); 1985 av_free(s->coded_fragment_list); 1986 av_free(s->superblock_fragments); 1987 av_free(s->superblock_macroblocks); 1988 av_free(s->macroblock_fragments); 1989 av_free(s->macroblock_coding); 1990 1991 for (i = 0; i < 16; i++) { 1992 free_vlc(&s->dc_vlc[i]); 1993 free_vlc(&s->ac_vlc_1[i]); 1994 free_vlc(&s->ac_vlc_2[i]); 1995 free_vlc(&s->ac_vlc_3[i]); 1996 free_vlc(&s->ac_vlc_4[i]); 1997 } 1998 1999 free_vlc(&s->superblock_run_length_vlc); 2000 free_vlc(&s->fragment_run_length_vlc); 2001 free_vlc(&s->mode_code_vlc); 2002 free_vlc(&s->motion_vector_vlc); 2003 2004 /* release all frames */ 2005 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0]) 2006 avctx->release_buffer(avctx, &s->golden_frame); 2007 if (s->last_frame.data[0]) 2008 avctx->release_buffer(avctx, &s->last_frame); 2009 /* no need to release the current_frame since it will always be pointing 2010 * to the same frame as either the golden or last frame */ 2011 2012 return 0; 2013} 2014 2015static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb) 2016{ 2017 Vp3DecodeContext *s = avctx->priv_data; 2018 2019 if (get_bits1(gb)) { 2020 int token; 2021 if (s->entries >= 32) { /* overflow */ 2022 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); 2023 return -1; 2024 } 2025 token = get_bits(gb, 5); 2026 //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size); 2027 s->huffman_table[s->hti][token][0] = s->hbits; 2028 s->huffman_table[s->hti][token][1] = s->huff_code_size; 2029 s->entries++; 2030 } 2031 else { 2032 if (s->huff_code_size >= 32) {/* overflow */ 2033 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n"); 2034 return -1; 2035 } 2036 s->huff_code_size++; 2037 s->hbits <<= 1; 2038 if (read_huffman_tree(avctx, gb)) 2039 return -1; 2040 s->hbits |= 1; 2041 if (read_huffman_tree(avctx, gb)) 2042 return -1; 2043 s->hbits >>= 1; 2044 s->huff_code_size--; 2045 } 2046 return 0; 2047} 2048 2049#if CONFIG_THEORA_DECODER 2050static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb) 2051{ 2052 Vp3DecodeContext *s = avctx->priv_data; 2053 int visible_width, visible_height; 2054 2055 s->theora = get_bits_long(gb, 24); 2056 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora); 2057 2058 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */ 2059 /* but previous versions have the image flipped relative to vp3 */ 2060 if (s->theora < 0x030200) 2061 { 2062 s->flipped_image = 1; 2063 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n"); 2064 } 2065 2066 visible_width = s->width = get_bits(gb, 16) << 4; 2067 visible_height = s->height = get_bits(gb, 16) << 4; 2068 2069 if(avcodec_check_dimensions(avctx, s->width, s->height)){ 2070 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height); 2071 s->width= s->height= 0; 2072 return -1; 2073 } 2074 2075 if (s->theora >= 0x030400) 2076 { 2077 skip_bits(gb, 32); /* total number of superblocks in a frame */ 2078 // fixme, the next field is 36bits long 2079 skip_bits(gb, 32); /* total number of blocks in a frame */ 2080 skip_bits(gb, 4); /* total number of blocks in a frame */ 2081 skip_bits(gb, 32); /* total number of macroblocks in a frame */ 2082 } 2083 2084 if (s->theora >= 0x030200) { 2085 visible_width = get_bits_long(gb, 24); 2086 visible_height = get_bits_long(gb, 24); 2087 2088 skip_bits(gb, 8); /* offset x */ 2089 skip_bits(gb, 8); /* offset y */ 2090 } 2091 2092 skip_bits(gb, 32); /* fps numerator */ 2093 skip_bits(gb, 32); /* fps denumerator */ 2094 skip_bits(gb, 24); /* aspect numerator */ 2095 skip_bits(gb, 24); /* aspect denumerator */ 2096 2097 if (s->theora < 0x030200) 2098 skip_bits(gb, 5); /* keyframe frequency force */ 2099 skip_bits(gb, 8); /* colorspace */ 2100 if (s->theora >= 0x030400) 2101 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */ 2102 skip_bits(gb, 24); /* bitrate */ 2103 2104 skip_bits(gb, 6); /* quality hint */ 2105 2106 if (s->theora >= 0x030200) 2107 { 2108 skip_bits(gb, 5); /* keyframe frequency force */ 2109 2110 if (s->theora < 0x030400) 2111 skip_bits(gb, 5); /* spare bits */ 2112 } 2113 2114// align_get_bits(gb); 2115 2116 if ( visible_width <= s->width && visible_width > s->width-16 2117 && visible_height <= s->height && visible_height > s->height-16) 2118 avcodec_set_dimensions(avctx, visible_width, visible_height); 2119 else 2120 avcodec_set_dimensions(avctx, s->width, s->height); 2121 2122 return 0; 2123} 2124 2125static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb) 2126{ 2127 Vp3DecodeContext *s = avctx->priv_data; 2128 int i, n, matrices, inter, plane; 2129 2130 if (s->theora >= 0x030200) { 2131 n = get_bits(gb, 3); 2132 /* loop filter limit values table */ 2133 for (i = 0; i < 64; i++) 2134 s->filter_limit_values[i] = get_bits(gb, n); 2135 } 2136 2137 if (s->theora >= 0x030200) 2138 n = get_bits(gb, 4) + 1; 2139 else 2140 n = 16; 2141 /* quality threshold table */ 2142 for (i = 0; i < 64; i++) 2143 s->coded_ac_scale_factor[i] = get_bits(gb, n); 2144 2145 if (s->theora >= 0x030200) 2146 n = get_bits(gb, 4) + 1; 2147 else 2148 n = 16; 2149 /* dc scale factor table */ 2150 for (i = 0; i < 64; i++) 2151 s->coded_dc_scale_factor[i] = get_bits(gb, n); 2152 2153 if (s->theora >= 0x030200) 2154 matrices = get_bits(gb, 9) + 1; 2155 else 2156 matrices = 3; 2157 2158 if(matrices > 384){ 2159 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n"); 2160 return -1; 2161 } 2162 2163 for(n=0; n<matrices; n++){ 2164 for (i = 0; i < 64; i++) 2165 s->base_matrix[n][i]= get_bits(gb, 8); 2166 } 2167 2168 for (inter = 0; inter <= 1; inter++) { 2169 for (plane = 0; plane <= 2; plane++) { 2170 int newqr= 1; 2171 if (inter || plane > 0) 2172 newqr = get_bits1(gb); 2173 if (!newqr) { 2174 int qtj, plj; 2175 if(inter && get_bits1(gb)){ 2176 qtj = 0; 2177 plj = plane; 2178 }else{ 2179 qtj= (3*inter + plane - 1) / 3; 2180 plj= (plane + 2) % 3; 2181 } 2182 s->qr_count[inter][plane]= s->qr_count[qtj][plj]; 2183 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0])); 2184 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0])); 2185 } else { 2186 int qri= 0; 2187 int qi = 0; 2188 2189 for(;;){ 2190 i= get_bits(gb, av_log2(matrices-1)+1); 2191 if(i>= matrices){ 2192 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n"); 2193 return -1; 2194 } 2195 s->qr_base[inter][plane][qri]= i; 2196 if(qi >= 63) 2197 break; 2198 i = get_bits(gb, av_log2(63-qi)+1) + 1; 2199 s->qr_size[inter][plane][qri++]= i; 2200 qi += i; 2201 } 2202 2203 if (qi > 63) { 2204 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi); 2205 return -1; 2206 } 2207 s->qr_count[inter][plane]= qri; 2208 } 2209 } 2210 } 2211 2212 /* Huffman tables */ 2213 for (s->hti = 0; s->hti < 80; s->hti++) { 2214 s->entries = 0; 2215 s->huff_code_size = 1; 2216 if (!get_bits1(gb)) { 2217 s->hbits = 0; 2218 if(read_huffman_tree(avctx, gb)) 2219 return -1; 2220 s->hbits = 1; 2221 if(read_huffman_tree(avctx, gb)) 2222 return -1; 2223 } 2224 } 2225 2226 s->theora_tables = 1; 2227 2228 return 0; 2229} 2230 2231static av_cold int theora_decode_init(AVCodecContext *avctx) 2232{ 2233 Vp3DecodeContext *s = avctx->priv_data; 2234 GetBitContext gb; 2235 int ptype; 2236 uint8_t *header_start[3]; 2237 int header_len[3]; 2238 int i; 2239 2240 s->theora = 1; 2241 2242 if (!avctx->extradata_size) 2243 { 2244 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n"); 2245 return -1; 2246 } 2247 2248 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size, 2249 42, header_start, header_len) < 0) { 2250 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n"); 2251 return -1; 2252 } 2253 2254 for(i=0;i<3;i++) { 2255 init_get_bits(&gb, header_start[i], header_len[i] * 8); 2256 2257 ptype = get_bits(&gb, 8); 2258 2259 if (!(ptype & 0x80)) 2260 { 2261 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n"); 2262// return -1; 2263 } 2264 2265 // FIXME: Check for this as well. 2266 skip_bits(&gb, 6*8); /* "theora" */ 2267 2268 switch(ptype) 2269 { 2270 case 0x80: 2271 theora_decode_header(avctx, &gb); 2272 break; 2273 case 0x81: 2274// FIXME: is this needed? it breaks sometimes 2275// theora_decode_comments(avctx, gb); 2276 break; 2277 case 0x82: 2278 if (theora_decode_tables(avctx, &gb)) 2279 return -1; 2280 break; 2281 default: 2282 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80); 2283 break; 2284 } 2285 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb)) 2286 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype); 2287 if (s->theora < 0x030200) 2288 break; 2289 } 2290 2291 vp3_decode_init(avctx); 2292 return 0; 2293} 2294 2295AVCodec theora_decoder = { 2296 "theora", 2297 CODEC_TYPE_VIDEO, 2298 CODEC_ID_THEORA, 2299 sizeof(Vp3DecodeContext), 2300 theora_decode_init, 2301 NULL, 2302 vp3_decode_end, 2303 vp3_decode_frame, 2304 0, 2305 NULL, 2306 .long_name = NULL_IF_CONFIG_SMALL("Theora"), 2307}; 2308#endif 2309 2310AVCodec vp3_decoder = { 2311 "vp3", 2312 CODEC_TYPE_VIDEO, 2313 CODEC_ID_VP3, 2314 sizeof(Vp3DecodeContext), 2315 vp3_decode_init, 2316 NULL, 2317 vp3_decode_end, 2318 vp3_decode_frame, 2319 0, 2320 NULL, 2321 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"), 2322}; 2323