Mercurial > libavcodec.hg
comparison h264_loopfilter.c @ 11203:10c06a9bd3d9 libavcodec
H264: use alias-safe macros
This eliminates all aliasing violation warnings in h264 code.
No measurable speed difference with gcc-4.4.3 on i7.
author | mru |
---|---|
date | Thu, 18 Feb 2010 16:24:31 +0000 |
parents | d464f498e19f |
children | 98970e51365a |
comparison
equal
deleted
inserted
replaced
11202:778139a5e058 | 11203:10c06a9bd3d9 |
---|---|
23 * @file libavcodec/h264_loopfilter.c | 23 * @file libavcodec/h264_loopfilter.c |
24 * H.264 / AVC / MPEG4 part10 loop filter. | 24 * H.264 / AVC / MPEG4 part10 loop filter. |
25 * @author Michael Niedermayer <michaelni@gmx.at> | 25 * @author Michael Niedermayer <michaelni@gmx.at> |
26 */ | 26 */ |
27 | 27 |
28 #include "libavutil/intreadwrite.h" | |
28 #include "internal.h" | 29 #include "internal.h" |
29 #include "dsputil.h" | 30 #include "dsputil.h" |
30 #include "avcodec.h" | 31 #include "avcodec.h" |
31 #include "mpegvideo.h" | 32 #include "mpegvideo.h" |
32 #include "h264.h" | 33 #include "h264.h" |
366 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h); | 367 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h); |
367 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h); | 368 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h); |
368 return; | 369 return; |
369 } else { | 370 } else { |
370 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]); | 371 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]); |
371 uint64_t (*bSv)[4] = (uint64_t(*)[4])bS; | |
372 int edges; | 372 int edges; |
373 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) { | 373 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) { |
374 edges = 4; | 374 edges = 4; |
375 bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL; | 375 AV_WN64A(bS[0][0], 0x0002000200020002ULL); |
376 AV_WN64A(bS[0][2], 0x0002000200020002ULL); | |
377 AV_WN64A(bS[1][0], 0x0002000200020002ULL); | |
378 AV_WN64A(bS[1][2], 0x0002000200020002ULL); | |
376 } else { | 379 } else { |
377 int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0; | 380 int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0; |
378 int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0; | 381 int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0; |
379 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1; | 382 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1; |
380 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4; | 383 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4; |
381 s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache, | 384 s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache, |
382 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE); | 385 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE); |
383 } | 386 } |
384 if( IS_INTRA(left_type) ) | 387 if( IS_INTRA(left_type) ) |
385 bSv[0][0] = 0x0004000400040004ULL; | 388 AV_WN64A(bS[0][0], 0x0004000400040004ULL); |
386 if( IS_INTRA(h->top_type) ) | 389 if( IS_INTRA(h->top_type) ) |
387 bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL; | 390 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL); |
388 | 391 |
389 #define FILTER(hv,dir,edge)\ | 392 #define FILTER(hv,dir,edge)\ |
390 if(bSv[dir][edge]) {\ | 393 if(AV_RN64A(bS[dir][edge])) { \ |
391 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\ | 394 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\ |
392 if(!(edge&1)) {\ | 395 if(!(edge&1)) {\ |
393 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ | 396 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ |
394 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ | 397 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ |
395 }\ | 398 }\ |
475 | 478 |
476 for(j=0; j<2; j++, mbn_xy += s->mb_stride){ | 479 for(j=0; j<2; j++, mbn_xy += s->mb_stride){ |
477 DECLARE_ALIGNED_8(int16_t, bS)[4]; | 480 DECLARE_ALIGNED_8(int16_t, bS)[4]; |
478 int qp; | 481 int qp; |
479 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) { | 482 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) { |
480 *(uint64_t*)bS= 0x0003000300030003ULL; | 483 AV_WN64A(bS, 0x0003000300030003ULL); |
481 } else { | 484 } else { |
482 if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){ | 485 if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){ |
483 bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]); | 486 bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]); |
484 bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]); | 487 bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]); |
485 bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]); | 488 bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]); |
506 }else{ | 509 }else{ |
507 DECLARE_ALIGNED_8(int16_t, bS)[4]; | 510 DECLARE_ALIGNED_8(int16_t, bS)[4]; |
508 int qp; | 511 int qp; |
509 | 512 |
510 if( IS_INTRA(mb_type|mbm_type)) { | 513 if( IS_INTRA(mb_type|mbm_type)) { |
511 *(uint64_t*)bS= 0x0003000300030003ULL; | 514 AV_WN64A(bS, 0x0003000300030003ULL); |
512 if ( (!IS_INTERLACED(mb_type|mbm_type)) | 515 if ( (!IS_INTERLACED(mb_type|mbm_type)) |
513 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) | 516 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) |
514 ) | 517 ) |
515 *(uint64_t*)bS= 0x0004000400040004ULL; | 518 AV_WN64A(bS, 0x0004000400040004ULL); |
516 } else { | 519 } else { |
517 int i; | 520 int i; |
518 int mv_done; | 521 int mv_done; |
519 | 522 |
520 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) { | 523 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) { |
521 *(uint64_t*)bS= 0x0001000100010001ULL; | 524 AV_WN64A(bS, 0x0001000100010001ULL); |
522 mv_done = 1; | 525 mv_done = 1; |
523 } | 526 } |
524 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { | 527 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { |
525 int b_idx= 8 + 4; | 528 int b_idx= 8 + 4; |
526 int bn_idx= b_idx - (dir ? 8:1); | 529 int bn_idx= b_idx - (dir ? 8:1); |
586 | 589 |
587 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type) | 590 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type) |
588 continue; | 591 continue; |
589 | 592 |
590 if( IS_INTRA(mb_type)) { | 593 if( IS_INTRA(mb_type)) { |
591 *(uint64_t*)bS= 0x0003000300030003ULL; | 594 AV_WN64A(bS, 0x0003000300030003ULL); |
592 } else { | 595 } else { |
593 int i; | 596 int i; |
594 int mv_done; | 597 int mv_done; |
595 | 598 |
596 if( edge & mask_edge ) { | 599 if( edge & mask_edge ) { |
597 *(uint64_t*)bS= 0; | 600 AV_ZERO64(bS); |
598 mv_done = 1; | 601 mv_done = 1; |
599 } | 602 } |
600 else if( mask_par0 ) { | 603 else if( mask_par0 ) { |
601 int b_idx= 8 + 4 + edge * (dir ? 8:1); | 604 int b_idx= 8 + 4 + edge * (dir ? 8:1); |
602 int bn_idx= b_idx - (dir ? 8:1); | 605 int bn_idx= b_idx - (dir ? 8:1); |
672 int rqp[2]; | 675 int rqp[2]; |
673 int mb_qp, mbn0_qp, mbn1_qp; | 676 int mb_qp, mbn0_qp, mbn1_qp; |
674 int i; | 677 int i; |
675 first_vertical_edge_done = 1; | 678 first_vertical_edge_done = 1; |
676 | 679 |
677 if( IS_INTRA(mb_type) ) | 680 if( IS_INTRA(mb_type) ) { |
678 *(uint64_t*)&bS[0]= | 681 AV_WN64A(&bS[0], 0x0004000400040004ULL); |
679 *(uint64_t*)&bS[4]= 0x0004000400040004ULL; | 682 AV_WN64A(&bS[4], 0x0004000400040004ULL); |
680 else { | 683 } else { |
681 static const uint8_t offset[2][2][8]={ | 684 static const uint8_t offset[2][2][8]={ |
682 { | 685 { |
683 {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1}, | 686 {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1}, |
684 {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3}, | 687 {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3}, |
685 },{ | 688 },{ |