X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fh264_loopfilter.c;h=710e03706e8d8e43490405a63fd91e5b15671d55;hb=81a646140f80c3e15c1e68883a5ecfb9b2bc8a6a;hp=f95af66d02c95ea7ad2cbf74277139a7aea297bd;hpb=b6303e6d2a267a06500c58bd4b6a02683ad6388b;p=ffmpeg diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c index f95af66d02c..710e03706e8 100644 --- a/libavcodec/h264_loopfilter.c +++ b/libavcodec/h264_loopfilter.c @@ -20,11 +20,12 @@ */ /** - * @file libavcodec/h264_loopfilter.c + * @file * H.264 / AVC / MPEG4 part10 loop filter. * @author Michael Niedermayer */ +#include "libavutil/intreadwrite.h" #include "internal.h" #include "dsputil.h" #include "avcodec.h" @@ -33,10 +34,6 @@ #include "mathops.h" #include "rectangle.h" -#if ARCH_X86 -#include "x86/h264_i386.h" -#endif - //#undef NDEBUG #include @@ -103,66 +100,55 @@ static const uint8_t tc0_table[52*3][4] = { {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, }; -static void av_noinline filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { - const int index_a = qp + h->slice_alpha_c0_offset; - const int alpha = (alpha_table+52)[index_a]; - const int beta = (beta_table+52)[qp + h->slice_beta_offset]; +static void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) { + const unsigned int index_a = qp + h->slice_alpha_c0_offset; + const int alpha = alpha_table[index_a]; + const int beta = beta_table[qp + h->slice_beta_offset]; if (alpha ==0 || beta == 0) return; if( bS[0] < 4 ) { int8_t tc[4]; - tc[0] = (tc0_table+52)[index_a][bS[0]]; - tc[1] = (tc0_table+52)[index_a][bS[1]]; - tc[2] = (tc0_table+52)[index_a][bS[2]]; - tc[3] = (tc0_table+52)[index_a][bS[3]]; - h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc); + tc[0] = tc0_table[index_a][bS[0]]; + tc[1] = tc0_table[index_a][bS[1]]; + tc[2] = tc0_table[index_a][bS[2]]; + tc[3] = tc0_table[index_a][bS[3]]; + h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc); } else { - h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta); + h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta); } } -static void av_noinline filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { - const int index_a = qp + h->slice_alpha_c0_offset; - const int alpha = (alpha_table+52)[index_a]; - const int beta = (beta_table+52)[qp + h->slice_beta_offset]; +static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) { + const unsigned int index_a = qp + h->slice_alpha_c0_offset; + const int alpha = alpha_table[index_a]; + const int beta = beta_table[qp + h->slice_beta_offset]; if (alpha ==0 || beta == 0) return; if( bS[0] < 4 ) { int8_t tc[4]; - tc[0] = (tc0_table+52)[index_a][bS[0]]+1; - tc[1] = (tc0_table+52)[index_a][bS[1]]+1; - tc[2] = (tc0_table+52)[index_a][bS[2]]+1; - tc[3] = (tc0_table+52)[index_a][bS[3]]+1; - h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc); + tc[0] = tc0_table[index_a][bS[0]]+1; + tc[1] = tc0_table[index_a][bS[1]]+1; + tc[2] = tc0_table[index_a][bS[2]]+1; + tc[3] = tc0_table[index_a][bS[3]]+1; + h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc); } else { - h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta); + h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta); } } -static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) { +static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) { int i; - for( i = 0; i < 16; i++, pix += stride) { - int index_a; - int alpha; - int beta; - - int qp_index; - int bS_index = (i >> 1); - if (!MB_FIELD) { - bS_index &= ~1; - bS_index |= (i & 1); - } + int index_a = qp + h->slice_alpha_c0_offset; + int alpha = alpha_table[index_a]; + int beta = beta_table[qp + h->slice_beta_offset]; + for( i = 0; i < 8; i++, pix += stride) { + const int bS_index = (i >> 1) * bsi; if( bS[bS_index] == 0 ) { continue; } - qp_index = MB_FIELD ? (i >> 3) : (i & 1); - index_a = qp[qp_index] + h->slice_alpha_c0_offset; - alpha = (alpha_table+52)[index_a]; - beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset]; - if( bS[bS_index] < 4 ) { - const int tc0 = (tc0_table+52)[index_a][bS[bS_index]]; + const int tc0 = tc0_table[index_a][bS[bS_index]]; const int p0 = pix[-1]; const int p1 = pix[-2]; const int p2 = pix[-3]; @@ -177,10 +163,12 @@ static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int int i_delta; if( FFABS( p2 - p0 ) < beta ) { + if(tc0) pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); tc++; } if( FFABS( q2 - q0 ) < beta ) { + if(tc0) pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); tc++; } @@ -236,27 +224,20 @@ static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int } } } -static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) { +static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) { int i; - for( i = 0; i < 8; i++, pix += stride) { - int index_a; - int alpha; - int beta; - - int qp_index; - int bS_index = i; + int index_a = qp + h->slice_alpha_c0_offset; + int alpha = alpha_table[index_a]; + int beta = beta_table[qp + h->slice_beta_offset]; + for( i = 0; i < 4; i++, pix += stride) { + const int bS_index = i*bsi; if( bS[bS_index] == 0 ) { continue; } - qp_index = MB_FIELD ? (i >> 2) : (i & 1); - index_a = qp[qp_index] + h->slice_alpha_c0_offset; - alpha = (alpha_table+52)[index_a]; - beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset]; - if( bS[bS_index] < 4 ) { - const int tc = (tc0_table+52)[index_a][bS[bS_index]] + 1; + const int tc = tc0_table[index_a][bS[bS_index]] + 1; const int p0 = pix[-1]; const int p1 = pix[-2]; const int q0 = pix[0]; @@ -289,58 +270,56 @@ static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, in } } -static void av_noinline filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { - const int index_a = qp + h->slice_alpha_c0_offset; - const int alpha = (alpha_table+52)[index_a]; - const int beta = (beta_table+52)[qp + h->slice_beta_offset]; +static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) { + const unsigned int index_a = qp + h->slice_alpha_c0_offset; + const int alpha = alpha_table[index_a]; + const int beta = beta_table[qp + h->slice_beta_offset]; if (alpha ==0 || beta == 0) return; if( bS[0] < 4 ) { int8_t tc[4]; - tc[0] = (tc0_table+52)[index_a][bS[0]]; - tc[1] = (tc0_table+52)[index_a][bS[1]]; - tc[2] = (tc0_table+52)[index_a][bS[2]]; - tc[3] = (tc0_table+52)[index_a][bS[3]]; - h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc); + tc[0] = tc0_table[index_a][bS[0]]; + tc[1] = tc0_table[index_a][bS[1]]; + tc[2] = tc0_table[index_a][bS[2]]; + tc[3] = tc0_table[index_a][bS[3]]; + h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc); } else { - h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta); + h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta); } } -static void av_noinline filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { - const int index_a = qp + h->slice_alpha_c0_offset; - const int alpha = (alpha_table+52)[index_a]; - const int beta = (beta_table+52)[qp + h->slice_beta_offset]; +static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) { + const unsigned int index_a = qp + h->slice_alpha_c0_offset; + const int alpha = alpha_table[index_a]; + const int beta = beta_table[qp + h->slice_beta_offset]; if (alpha ==0 || beta == 0) return; if( bS[0] < 4 ) { int8_t tc[4]; - tc[0] = (tc0_table+52)[index_a][bS[0]]+1; - tc[1] = (tc0_table+52)[index_a][bS[1]]+1; - tc[2] = (tc0_table+52)[index_a][bS[2]]+1; - tc[3] = (tc0_table+52)[index_a][bS[3]]+1; - h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc); + tc[0] = tc0_table[index_a][bS[0]]+1; + tc[1] = tc0_table[index_a][bS[1]]+1; + tc[2] = tc0_table[index_a][bS[2]]+1; + tc[3] = tc0_table[index_a][bS[3]]+1; + h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc); } else { - h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta); + h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta); } } void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) { MpegEncContext * const s = &h->s; - int mb_y_firstrow = s->picture_structure == PICT_BOTTOM_FIELD; - int mb_xy, mb_type; + int mb_xy; + int mb_type, left_type; int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh; mb_xy = h->mb_xy; - if(mb_x==0 || mb_y==mb_y_firstrow || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff || - !(s->flags2 & CODEC_FLAG2_FAST) || //FIXME filter_mb_fast is broken, thus hasto be, but should not under CODEC_FLAG2_FAST - (h->deblocking_filter == 2 && (h->slice_table[mb_xy] != h->slice_table[h->top_mb_xy] || - h->slice_table[mb_xy] != h->slice_table[mb_xy - 1]))) { + if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) { ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize); return; } assert(!FRAME_MBAFF); + left_type= h->left_type[0]; mb_type = s->current_picture.mb_type[mb_xy]; qp = s->current_picture.qscale_table[mb_xy]; @@ -353,7 +332,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, qp1 = (qp + qp1 + 1) >> 1; qpc0 = (qpc + qpc0 + 1) >> 1; qpc1 = (qpc + qpc1 + 1) >> 1; - qp_thresh = 15 - h->slice_alpha_c0_offset; + qp_thresh = 15+52 - h->slice_alpha_c0_offset; if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh && qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh) return; @@ -362,71 +341,71 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, int16_t bS4[4] = {4,4,4,4}; int16_t bS3[4] = {3,3,3,3}; int16_t *bSH = FIELD_PICTURE ? bS3 : bS4; + if(left_type) + filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h); if( IS_8x8DCT(mb_type) ) { - filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 ); - filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp ); - filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 ); - filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp ); + filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h); + filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h); + filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h); } else { - filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 ); - filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp ); - filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp ); - filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp ); - filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 ); - filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp ); - filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp ); - filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp ); + filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h); + filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h); + filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h); + filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h); + filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h); + filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h); + filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h); + } + if(left_type){ + filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h); + filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h); } - filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 ); - filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc ); - filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 ); - filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc ); - filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1 ); - filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc ); - filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1 ); - filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc ); + filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h); + filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h); + filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h); + filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h); + filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h); + filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h); return; } else { - DECLARE_ALIGNED_8(int16_t, bS[2][4][4]); - uint64_t (*bSv)[4] = (uint64_t(*)[4])bS; + LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]); int edges; if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) { edges = 4; - bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL; + AV_WN64A(bS[0][0], 0x0002000200020002ULL); + AV_WN64A(bS[0][2], 0x0002000200020002ULL); + AV_WN64A(bS[1][0], 0x0002000200020002ULL); + AV_WN64A(bS[1][2], 0x0002000200020002ULL); } else { - int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : - (mb_type & MB_TYPE_16x8) ? 1 : 0; - int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) - && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16)) - ? 3 : 0; - int step = IS_8x8DCT(mb_type) ? 2 : 1; - edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4; - s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache, - (h->slice_type_nos == FF_B_TYPE), edges, step, mask_edge0, mask_edge1, FIELD_PICTURE); + 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; + 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; + int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1; + edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4; + h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache, + h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE); } - if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) ) - bSv[0][0] = 0x0004000400040004ULL; - if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) ) - bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL; + if( IS_INTRA(left_type) ) + AV_WN64A(bS[0][0], 0x0004000400040004ULL); + if( IS_INTRA(h->top_type) ) + AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL); #define FILTER(hv,dir,edge)\ - if(bSv[dir][edge]) {\ - filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\ + if(AV_RN64A(bS[dir][edge])) { \ + filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\ if(!(edge&1)) {\ - filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\ - filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\ + filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ + filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\ }\ } - if( edges == 1 ) { + if(left_type) FILTER(v,0,0); + if( edges == 1 ) { FILTER(h,1,0); } else if( IS_8x8DCT(mb_type) ) { - FILTER(v,0,0); FILTER(v,0,2); FILTER(h,1,0); FILTER(h,1,2); } else { - FILTER(v,0,0); FILTER(v,0,1); FILTER(v,0,2); FILTER(v,0,3); @@ -439,136 +418,193 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, } } +static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){ + int v; + + v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx]; + if(!v && h->ref_cache[0][b_idx]!=-1) + v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U | + FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit; + + if(h->list_count==2){ + if(!v) + v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] | + h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U | + FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit; + + if(v){ + if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] | + h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx]) + return 1; + return + h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U | + FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit | + h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U | + FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit; + } + } + + return v; +} static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int dir) { MpegEncContext * const s = &h->s; int edge; const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; - const int mbm_type = s->current_picture.mb_type[mbm_xy]; - int (*ref2frm) [64] = h->ref2frm[ h->slice_num &(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2); - int (*ref2frmm)[64] = h->ref2frm[ h->slice_table[mbm_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2); - int start = h->slice_table[mbm_xy] == 0xFFFF ? 1 : 0; + const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type; - const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP)) - == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4; // how often to recheck mv-based bS when iterating between edges - const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 : - (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0; + static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1}, + {0,3,1,1,3,3,3,3}}; + const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7]; + const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4; + // how often to recheck mv-based bS when iterating along each edge const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)); - if (first_vertical_edge_done) { - start = 1; - } - - if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy]) - start = 1; - - if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0 - && !IS_INTERLACED(mb_type) - && IS_INTERLACED(mbm_type) - ) { - // This is a special case in the norm where the filtering must - // be done twice (one each of the field) even if we are in a - // frame macroblock. - // - unsigned int tmp_linesize = 2 * linesize; - unsigned int tmp_uvlinesize = 2 * uvlinesize; - int mbn_xy = mb_xy - 2 * s->mb_stride; - int qp; - int i, j; - int16_t bS[4]; + if(mbm_type && !first_vertical_edge_done){ + + if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) + && IS_INTERLACED(mbm_type&~mb_type) + ) { + // This is a special case in the norm where the filtering must + // be done twice (one each of the field) even if we are in a + // frame macroblock. + // + unsigned int tmp_linesize = 2 * linesize; + unsigned int tmp_uvlinesize = 2 * uvlinesize; + int mbn_xy = mb_xy - 2 * s->mb_stride; + int j; + + for(j=0; j<2; j++, mbn_xy += s->mb_stride){ + DECLARE_ALIGNED(8, int16_t, bS)[4]; + int qp; + if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) { + AV_WN64A(bS, 0x0003000300030003ULL); + } else { + if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){ + bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]); + bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]); + bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]); + bS[3]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+3]); + }else{ + const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8; + int i; + for( i = 0; i < 4; i++ ) { + bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]); + } + } + } + // Do not use s->qscale as luma quantizer because it has not the same + // value in IPCM macroblocks. + qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; + tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize); + { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } + filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h ); + filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS, + ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h); + filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS, + ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h); + } + }else{ + DECLARE_ALIGNED(8, int16_t, bS)[4]; + int qp; - for(j=0; j<2; j++, mbn_xy += s->mb_stride){ - if( IS_INTRA(mb_type) || - IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) { - bS[0] = bS[1] = bS[2] = bS[3] = 3; + if( IS_INTRA(mb_type|mbm_type)) { + AV_WN64A(bS, 0x0003000300030003ULL); + if ( (!IS_INTERLACED(mb_type|mbm_type)) + || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) + ) + AV_WN64A(bS, 0x0004000400040004ULL); } else { - const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy]; + int i; + int mv_done; + + if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) { + AV_WN64A(bS, 0x0001000100010001ULL); + mv_done = 1; + } + else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { + int b_idx= 8 + 4; + int bn_idx= b_idx - (dir ? 8:1); + + bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit); + mv_done = 1; + } + else + mv_done = 0; + for( i = 0; i < 4; i++ ) { - if( h->non_zero_count_cache[scan8[0]+i] != 0 || - mbn_nnz[i+4+3*8] != 0 ) + int x = dir == 0 ? 0 : i; + int y = dir == 0 ? i : 0; + int b_idx= 8 + 4 + x + 8*y; + int bn_idx= b_idx - (dir ? 8:1); + + if( h->non_zero_count_cache[b_idx] | + h->non_zero_count_cache[bn_idx] ) { bS[i] = 2; - else - bS[i] = 1; + } + else if(!mv_done) + { + bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit); + } } } + + /* Filter edge */ // Do not use s->qscale as luma quantizer because it has not the same // value in IPCM macroblocks. - qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; - tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize); - { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } - filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp ); - filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS, - ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1); - filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS, - ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1); + if(bS[0]+bS[1]+bS[2]+bS[3]){ + qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1; + //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], s->current_picture.qscale_table[mbn_xy]); + tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize); + //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } + if( dir == 0 ) { + filter_mb_edgev( &img_y[0], linesize, bS, qp, h ); + { + int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1; + filter_mb_edgecv( &img_cb[0], uvlinesize, bS, qp, h); + if(h->pps.chroma_qp_diff) + qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1; + filter_mb_edgecv( &img_cr[0], uvlinesize, bS, qp, h); + } + } else { + filter_mb_edgeh( &img_y[0], linesize, bS, qp, h ); + { + int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1; + filter_mb_edgech( &img_cb[0], uvlinesize, bS, qp, h); + if(h->pps.chroma_qp_diff) + qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1; + filter_mb_edgech( &img_cr[0], uvlinesize, bS, qp, h); + } + } + } } - - start = 1; } /* Calculate bS */ - for( edge = start; edge < edges; edge++ ) { - /* mbn_xy: neighbor macroblock */ - const int mbn_xy = edge > 0 ? mb_xy : mbm_xy; - const int mbn_type = s->current_picture.mb_type[mbn_xy]; - int (*ref2frmn)[64] = edge > 0 ? ref2frm : ref2frmm; - int16_t bS[4]; + for( edge = 1; edge < edges; edge++ ) { + DECLARE_ALIGNED(8, int16_t, bS)[4]; int qp; - if( (edge&1) && IS_8x8DCT(mb_type) ) + if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type) continue; - if( IS_INTRA(mb_type) || - IS_INTRA(mbn_type) ) { - int value; - if (edge == 0) { - if ( (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type)) - || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) - ) { - value = 4; - } else { - value = 3; - } - } else { - value = 3; - } - bS[0] = bS[1] = bS[2] = bS[3] = value; + if( IS_INTRA(mb_type)) { + AV_WN64A(bS, 0x0003000300030003ULL); } else { - int i, l; + int i; int mv_done; if( edge & mask_edge ) { - bS[0] = bS[1] = bS[2] = bS[3] = 0; - mv_done = 1; - } - else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) { - bS[0] = bS[1] = bS[2] = bS[3] = 1; + AV_ZERO64(bS); mv_done = 1; } - else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { + else if( mask_par0 ) { int b_idx= 8 + 4 + edge * (dir ? 8:1); int bn_idx= b_idx - (dir ? 8:1); - int v = 0; - - for( l = 0; !v && l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) { - v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] | - h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U | - FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit; - } - if(h->slice_type_nos == FF_B_TYPE && v){ - v=0; - for( l = 0; !v && l < 2; l++ ) { - int ln= 1-l; - v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] | - h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U | - FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit; - } - } - - bS[0] = bS[1] = bS[2] = bS[3] = v; + bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit); mv_done = 1; } else @@ -586,28 +622,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u } else if(!mv_done) { - bS[i] = 0; - for( l = 0; l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) { - if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] | - h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U | - FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) { - bS[i] = 1; - break; - } - } - - if(h->slice_type_nos == FF_B_TYPE && bS[i]){ - bS[i] = 0; - for( l = 0; l < 2; l++ ) { - int ln= 1-l; - if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] | - h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U | - FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) { - bS[i] = 1; - break; - } - } - } + bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit); } } @@ -618,25 +633,21 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u /* Filter edge */ // Do not use s->qscale as luma quantizer because it has not the same // value in IPCM macroblocks. - qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; + qp = s->current_picture.qscale_table[mb_xy]; //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], s->current_picture.qscale_table[mbn_xy]); tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize); //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } if( dir == 0 ) { - filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp ); + filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h ); if( (edge&1) == 0 ) { - filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, - ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1); - filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, - ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1); + filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, h->chroma_qp[0], h); + filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, h->chroma_qp[1], h); } } else { - filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp ); + filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h ); if( (edge&1) == 0 ) { - filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, - ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1); - filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, - ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1); + filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h); + filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h); } } } @@ -649,69 +660,16 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4; int first_vertical_edge_done = 0; av_unused int dir; - int list; - - //for sufficiently low qp, filtering wouldn't do anything - //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp - if(!FRAME_MBAFF){ - int qp_thresh = h->qp_thresh; - int qp = s->current_picture.qscale_table[mb_xy]; - if(qp <= qp_thresh - && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh) - && (h->top_mb_xy < 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){ - return; - } - } - // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs - if(!h->pps.cabac && h->pps.transform_8x8_mode){ - int top_type, left_type[2]; - top_type = s->current_picture.mb_type[h->top_mb_xy] ; - left_type[0] = s->current_picture.mb_type[h->left_mb_xy[0]]; - left_type[1] = s->current_picture.mb_type[h->left_mb_xy[1]]; - - if(IS_8x8DCT(top_type)){ - h->non_zero_count_cache[4+8*0]= - h->non_zero_count_cache[5+8*0]= h->cbp_table[h->top_mb_xy] & 4; - h->non_zero_count_cache[6+8*0]= - h->non_zero_count_cache[7+8*0]= h->cbp_table[h->top_mb_xy] & 8; - } - if(IS_8x8DCT(left_type[0])){ - h->non_zero_count_cache[3+8*1]= - h->non_zero_count_cache[3+8*2]= h->cbp_table[h->left_mb_xy[0]]&2; //FIXME check MBAFF - } - if(IS_8x8DCT(left_type[1])){ - h->non_zero_count_cache[3+8*3]= - h->non_zero_count_cache[3+8*4]= h->cbp_table[h->left_mb_xy[1]]&8; //FIXME check MBAFF - } - - if(IS_8x8DCT(mb_type)){ - h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]= - h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp_table[mb_xy] & 1; - - h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]= - h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp_table[mb_xy] & 2; - - h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]= - h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp_table[mb_xy] & 4; - - h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]= - h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp_table[mb_xy] & 8; - } - } if (FRAME_MBAFF - // left mb is in picture - && h->slice_table[mb_xy-1] != 0xFFFF // and current and left pair do not have the same interlaced type - && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1])) - // and left mb is in the same slice if deblocking_filter == 2 - && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) { + && IS_INTERLACED(mb_type^h->left_type[0]) + // and left mb is in available to us + && h->left_type[0]) { /* First vertical edge is different in MBAFF frames * There are 8 different bS to compute and 2 different Qp */ - const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride; - const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride }; - int16_t bS[8]; + DECLARE_ALIGNED(8, int16_t, bS)[8]; int qp[2]; int bqp[2]; int rqp[2]; @@ -719,28 +677,40 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint int i; first_vertical_edge_done = 1; - if( IS_INTRA(mb_type) ) - bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4; - else { + if( IS_INTRA(mb_type) ) { + AV_WN64A(&bS[0], 0x0004000400040004ULL); + AV_WN64A(&bS[4], 0x0004000400040004ULL); + } else { + static const uint8_t offset[2][2][8]={ + { + {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1}, + {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3}, + },{ + {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3}, + {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3}, + } + }; + const uint8_t *off= offset[MB_FIELD][mb_y&1]; for( i = 0; i < 8; i++ ) { - int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1]; + int j= MB_FIELD ? i>>2 : i&1; + int mbn_xy = h->left_mb_xy[j]; + int mbn_type= h->left_type[j]; - if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) + if( IS_INTRA( mbn_type ) ) bS[i] = 4; - else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 || - ((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ? + else{ + bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] | + ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ? (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2)) : - h->non_zero_count[mbn_xy][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8])) - bS[i] = 2; - else - bS[i] = 1; + h->non_zero_count[mbn_xy][ off[i] ])); + } } } mb_qp = s->current_picture.qscale_table[mb_xy]; - mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]]; - mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]]; + mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]]; + mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]]; qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1; bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) + get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1; @@ -755,9 +725,21 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint /* Filter edge */ tprintf(s->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize); { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } - filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp ); - filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, bqp ); - filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, rqp ); + if(MB_FIELD){ + filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] ); + filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] ); + filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] ); + filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] ); + filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] ); + filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] ); + }else{ + filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] ); + filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] ); + filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] ); + filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] ); + filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] ); + filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] ); + } } #if CONFIG_SMALL