comparison rv40.c @ 8225:d133e597db2d libavcodec

RV40 loop filter
author kostya
date Sat, 29 Nov 2008 06:24:47 +0000
parents de344498875e
children 45c3780b1a96
comparison
equal deleted inserted replaced
8224:6771b3544991 8225:d133e597db2d
387 int alpha, int beta, int beta2, int chroma, int edge){ 387 int alpha, int beta, int beta2, int chroma, int edge){
388 rv40_adaptive_loop_filter(src, stride, 1, dmode, lim_q1, lim_p1, 388 rv40_adaptive_loop_filter(src, stride, 1, dmode, lim_q1, lim_p1,
389 alpha, beta, beta2, chroma, edge); 389 alpha, beta, beta2, chroma, edge);
390 } 390 }
391 391
392 enum RV40BlockPos{
393 POS_CUR,
394 POS_TOP,
395 POS_LEFT,
396 POS_BOTTOM,
397 };
398
399 #define MASK_CUR 0x0001
400 #define MASK_RIGHT 0x0008
401 #define MASK_BOTTOM 0x0010
402 #define MASK_TOP 0x1000
403 #define MASK_Y_TOP_ROW 0x000F
404 #define MASK_Y_LAST_ROW 0xF000
405 #define MASK_Y_LEFT_COL 0x1111
406 #define MASK_Y_RIGHT_COL 0x8888
407 #define MASK_C_TOP_ROW 0x0003
408 #define MASK_C_LAST_ROW 0x000C
409 #define MASK_C_LEFT_COL 0x0005
410 #define MASK_C_RIGHT_COL 0x000A
411
412 static const int neighbour_offs_x[4] = { 0, 0, -1, 0 };
413 static const int neighbour_offs_y[4] = { 0, -1, 0, 1 };
414
415 /**
416 * RV40 loop filtering function
417 */
418 static void rv40_loop_filter(RV34DecContext *r, int row)
419 {
420 MpegEncContext *s = &r->s;
421 int mb_pos, mb_x;
422 int i, j, k;
423 uint8_t *Y, *C;
424 int alpha, beta, betaY, betaC;
425 int q;
426 int mbtype[4]; ///< current macroblock and its neighbours types
427 /**
428 * flags indicating that macroblock can be filtered with strong filter
429 * it is set only for intra coded MB and MB with DCs coded separately
430 */
431 int mb_strong[4];
432 int clip[4]; ///< MB filter clipping value calculated from filtering strength
433 /**
434 * coded block patterns for luma part of current macroblock and its neighbours
435 * Format:
436 * LSB corresponds to the top left block,
437 * each nibble represents one row of subblocks.
438 */
439 int cbp[4];
440 /**
441 * coded block patterns for chroma part of current macroblock and its neighbours
442 * Format is the same as for luma with two subblocks in a row.
443 */
444 int uvcbp[4][2];
445 /**
446 * This mask represents the pattern of luma subblocks that should be filtered
447 * in addition to the coded ones because because they lie at the edge of
448 * 8x8 block with different enough motion vectors
449 */
450 int mvmasks[4];
451
452 mb_pos = row * s->mb_stride;
453 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
454 int mbtype = s->current_picture_ptr->mb_type[mb_pos];
455 if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
456 r->cbp_luma [mb_pos] = 0xFFFF;
457 if(IS_INTRA(mbtype))
458 r->cbp_chroma[mb_pos] = 0xFF;
459 }
460 mb_pos = row * s->mb_stride;
461 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
462 int y_h_deblock, y_v_deblock;
463 int c_v_deblock[2], c_h_deblock[2];
464 int clip_left;
465 int avail[4];
466 int y_to_deblock, c_to_deblock[2];
467
468 q = s->current_picture_ptr->qscale_table[mb_pos];
469 alpha = rv40_alpha_tab[q];
470 beta = rv40_beta_tab [q];
471 betaY = betaC = beta * 3;
472 if(s->width * s->height <= 176*144)
473 betaY += beta;
474
475 avail[0] = 1;
476 avail[1] = row;
477 avail[2] = mb_x;
478 avail[3] = row < s->mb_height - 1;
479 for(i = 0; i < 4; i++){
480 if(avail[i]){
481 int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
482 mvmasks[i] = r->deblock_coefs[pos];
483 mbtype [i] = s->current_picture_ptr->mb_type[pos];
484 cbp [i] = r->cbp_luma[pos];
485 uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
486 uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
487 }else{
488 mvmasks[i] = 0;
489 mbtype [i] = mbtype[0];
490 cbp [i] = 0;
491 uvcbp[i][0] = uvcbp[i][1] = 0;
492 }
493 mb_strong[i] = IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i]);
494 clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
495 }
496 y_to_deblock = cbp[POS_CUR]
497 | (cbp[POS_BOTTOM] << 16)
498 | mvmasks[POS_CUR]
499 | (mvmasks[POS_BOTTOM] << 16);
500 /* This pattern contains bits signalling that horizontal edges of
501 * the current block can be filtered.
502 * That happens when either of adjacent subblocks is coded or lies on
503 * the edge of 8x8 blocks with motion vectors differing by more than
504 * 3/4 pel in any component (any edge orientation for some reason).
505 */
506 y_h_deblock = y_to_deblock
507 | ((cbp[POS_CUR] << 4) & ~MASK_Y_TOP_ROW)
508 | ((cbp[POS_TOP] & MASK_Y_LAST_ROW) >> 12);
509 /* This pattern contains bits signalling that vertical edges of
510 * the current block can be filtered.
511 * That happens when either of adjacent subblocks is coded or lies on
512 * the edge of 8x8 blocks with motion vectors differing by more than
513 * 3/4 pel in any component (any edge orientation for some reason).
514 */
515 y_v_deblock = y_to_deblock
516 | ((cbp[POS_CUR] << 1) & ~MASK_Y_LEFT_COL)
517 | ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
518 if(!mb_x)
519 y_v_deblock &= ~MASK_Y_LEFT_COL;
520 if(!row)
521 y_h_deblock &= ~MASK_Y_TOP_ROW;
522 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] || mb_strong[POS_BOTTOM]))
523 y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
524 /* Calculating chroma patterns is similar and easier since there is
525 * no motion vector pattern for them.
526 */
527 for(i = 0; i < 2; i++){
528 c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
529 c_v_deblock[i] = c_to_deblock[i]
530 | ((uvcbp[POS_CUR] [i] << 1) & ~MASK_C_LEFT_COL)
531 | ((uvcbp[POS_LEFT][i] & MASK_C_RIGHT_COL) >> 1);
532 c_h_deblock[i] = c_to_deblock[i]
533 | ((uvcbp[POS_TOP][i] & MASK_C_LAST_ROW) >> 2)
534 | (uvcbp[POS_CUR][i] << 2);
535 if(!mb_x)
536 c_v_deblock[i] &= ~MASK_C_LEFT_COL;
537 if(!row)
538 c_h_deblock[i] &= ~MASK_C_TOP_ROW;
539 if(row == s->mb_height - 1 || mb_strong[POS_CUR] || mb_strong[POS_BOTTOM])
540 c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
541 }
542
543 for(j = 0; j < 16; j += 4){
544 Y = s->current_picture_ptr->data[0] + mb_x*16 + (row*16 + j) * s->linesize;
545 for(i = 0; i < 4; i++, Y += 4){
546 int ij = i + j;
547 int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
548 int dither = j ? ij : i*4;
549
550 // if bottom block is coded then we can filter its top edge
551 // (or bottom edge of this block, which is the same)
552 if(y_h_deblock & (MASK_BOTTOM << ij)){
553 rv40_h_loop_filter(Y+4*s->linesize, s->linesize, dither,
554 y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
555 clip_cur,
556 alpha, beta, betaY, 0, 0);
557 }
558 // filter left block edge in ordinary mode (with low filtering strength)
559 if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){
560 if(!i)
561 clip_left = (cbp[POS_LEFT] | mvmasks[POS_LEFT]) & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
562 else
563 clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
564 rv40_v_loop_filter(Y, s->linesize, dither,
565 clip_cur,
566 clip_left,
567 alpha, beta, betaY, 0, 0);
568 }
569 // filter top edge of the current macroblock when filtering strength is high
570 if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){
571 rv40_h_loop_filter(Y, s->linesize, dither,
572 clip_cur,
573 (cbp[POS_TOP] | mvmasks[POS_TOP]) & (MASK_TOP << i) ? clip[POS_TOP] : 0,
574 alpha, beta, betaY, 0, 1);
575 }
576 // filter left block edge in edge mode (with high filtering strength)
577 if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){
578 clip_left = (cbp[POS_LEFT] | mvmasks[POS_LEFT]) & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
579 rv40_v_loop_filter(Y, s->linesize, dither,
580 clip_cur,
581 clip_left,
582 alpha, beta, betaY, 0, 1);
583 }
584 }
585 }
586 for(k = 0; k < 2; k++){
587 for(j = 0; j < 2; j++){
588 C = s->current_picture_ptr->data[k+1] + mb_x*8 + (row*8 + j*4) * s->uvlinesize;
589 for(i = 0; i < 2; i++, C += 4){
590 int ij = i + j*2;
591 int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
592 if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
593 int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
594 rv40_h_loop_filter(C+4*s->uvlinesize, s->uvlinesize, i*8,
595 clip_bot,
596 clip_cur,
597 alpha, beta, betaC, 1, 0);
598 }
599 if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){
600 if(!i)
601 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
602 else
603 clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
604 rv40_v_loop_filter(C, s->uvlinesize, j*8,
605 clip_cur,
606 clip_left,
607 alpha, beta, betaC, 1, 0);
608 }
609 if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){
610 int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
611 rv40_h_loop_filter(C, s->uvlinesize, i*8,
612 clip_cur,
613 clip_top,
614 alpha, beta, betaC, 1, 1);
615 }
616 if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){
617 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
618 rv40_v_loop_filter(C, s->uvlinesize, j*8,
619 clip_cur,
620 clip_left,
621 alpha, beta, betaC, 1, 1);
622 }
623 }
624 }
625 }
626 }
627 }
628
392 /** 629 /**
393 * Initialize decoder. 630 * Initialize decoder.
394 */ 631 */
395 static av_cold int rv40_decode_init(AVCodecContext *avctx) 632 static av_cold int rv40_decode_init(AVCodecContext *avctx)
396 { 633 {
401 if(!aic_top_vlc.bits) 638 if(!aic_top_vlc.bits)
402 rv40_init_tables(); 639 rv40_init_tables();
403 r->parse_slice_header = rv40_parse_slice_header; 640 r->parse_slice_header = rv40_parse_slice_header;
404 r->decode_intra_types = rv40_decode_intra_types; 641 r->decode_intra_types = rv40_decode_intra_types;
405 r->decode_mb_info = rv40_decode_mb_info; 642 r->decode_mb_info = rv40_decode_mb_info;
643 r->loop_filter = rv40_loop_filter;
406 r->luma_dc_quant_i = rv40_luma_dc_quant[0]; 644 r->luma_dc_quant_i = rv40_luma_dc_quant[0];
407 r->luma_dc_quant_p = rv40_luma_dc_quant[1]; 645 r->luma_dc_quant_p = rv40_luma_dc_quant[1];
408 return 0; 646 return 0;
409 } 647 }
410 648