X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fh264_loopfilter.c;h=b045d2388a74524fd572eb4786ac47ee5a34d5d9;hb=033a86f9bb6fd59ca71d4951b8e2e27cdc1b29d9;hp=1ae534ec963c965746995cbb5e9e42419457cd43;hpb=7b442ad918bfbd1597f94b0a8e00c41468402236;p=ffmpeg diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c index 1ae534ec963..b045d2388a7 100644 --- a/libavcodec/h264_loopfilter.c +++ b/libavcodec/h264_loopfilter.c @@ -100,14 +100,18 @@ 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_always_inline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) { - const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); - const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset; +/* intra: 0 if this loopfilter call is guaranteed to be inter (bS < 4), 1 if it might be intra (bS == 4) */ +static av_always_inline void filter_mb_edgev(uint8_t *pix, int stride, + const int16_t bS[4], + unsigned int qp, int a, int b, + H264Context *h, int intra) +{ + const unsigned int index_a = qp + a; const int alpha = alpha_table[index_a]; - const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset]; + const int beta = beta_table[qp + b]; if (alpha ==0 || beta == 0) return; - if( bS[0] < 4 ) { + if( bS[0] < 4 || !intra ) { int8_t tc[4]; tc[0] = tc0_table[index_a][bS[0]]; tc[1] = tc0_table[index_a][bS[1]]; @@ -118,14 +122,18 @@ static void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, int16_t h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta); } } -static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) { - const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); - const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset; + +static av_always_inline void filter_mb_edgecv(uint8_t *pix, int stride, + const int16_t bS[4], + unsigned int qp, int a, int b, + H264Context *h, int intra) +{ + const unsigned int index_a = qp + a; const int alpha = alpha_table[index_a]; - const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset]; + const int beta = beta_table[qp + b]; if (alpha ==0 || beta == 0) return; - if( bS[0] < 4 ) { + if( bS[0] < 4 || !intra ) { int8_t tc[4]; tc[0] = tc0_table[index_a][bS[0]]+1; tc[1] = tc0_table[index_a][bS[1]]+1; @@ -137,14 +145,18 @@ static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t } } -static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) { - const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); - int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset; - int alpha = alpha_table[index_a]; - int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset]; +static av_always_inline void filter_mb_mbaff_edgev(H264Context *h, uint8_t *pix, + int stride, + const int16_t bS[7], int bsi, + int qp, int a, int b, + int intra) +{ + const unsigned int index_a = qp + a; + const int alpha = alpha_table[index_a]; + const int beta = beta_table[qp + b]; if (alpha ==0 || beta == 0) return; - if( bS[0] < 4 ) { + if( bS[0] < 4 || !intra ) { int8_t tc[4]; tc[0] = tc0_table[index_a][bS[0*bsi]]; tc[1] = tc0_table[index_a][bS[1*bsi]]; @@ -155,14 +167,19 @@ static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta); } } -static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) { - const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); - int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset; - int alpha = alpha_table[index_a]; - int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset]; + +static av_always_inline void filter_mb_mbaff_edgecv(H264Context *h, + uint8_t *pix, int stride, + const int16_t bS[7], + int bsi, int qp, int a, + int b, int intra) +{ + const unsigned int index_a = qp + a; + const int alpha = alpha_table[index_a]; + const int beta = beta_table[qp + b]; if (alpha ==0 || beta == 0) return; - if( bS[0] < 4 ) { + if( bS[0] < 4 || !intra ) { int8_t tc[4]; tc[0] = tc0_table[index_a][bS[0*bsi]] + 1; tc[1] = tc0_table[index_a][bS[1*bsi]] + 1; @@ -174,14 +191,17 @@ static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, in } } -static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) { - const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); - const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset; +static av_always_inline void filter_mb_edgeh(uint8_t *pix, int stride, + const int16_t bS[4], + unsigned int qp, int a, int b, + H264Context *h, int intra) +{ + const unsigned int index_a = qp + a; const int alpha = alpha_table[index_a]; - const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset]; + const int beta = beta_table[qp + b]; if (alpha ==0 || beta == 0) return; - if( bS[0] < 4 ) { + if( bS[0] < 4 || !intra ) { int8_t tc[4]; tc[0] = tc0_table[index_a][bS[0]]; tc[1] = tc0_table[index_a][bS[1]]; @@ -193,14 +213,17 @@ static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t } } -static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) { - const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); - const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset; +static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride, + const int16_t bS[4], + unsigned int qp, int a, int b, + H264Context *h, int intra) +{ + const unsigned int index_a = qp + a; const int alpha = alpha_table[index_a]; - const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset]; + const int beta = beta_table[qp + b]; if (alpha ==0 || beta == 0) return; - if( bS[0] < 4 ) { + if( bS[0] < 4 || !intra ) { int8_t tc[4]; tc[0] = tc0_table[index_a][bS[0]]+1; tc[1] = tc0_table[index_a][bS[1]]+1; @@ -212,74 +235,133 @@ static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t } } -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) { +static av_always_inline void h264_filter_mb_fast_internal(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 pixel_shift) +{ MpegEncContext * const s = &h->s; - int mb_xy; - int mb_type, left_type; - int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh; int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY)); - - mb_xy = h->mb_xy; - - if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff || CHROMA444) { - 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]; - qp0 = s->current_picture.qscale_table[mb_xy-1]; - qp1 = s->current_picture.qscale_table[h->top_mb_xy]; - qpc = get_chroma_qp( h, 0, qp ); - qpc0 = get_chroma_qp( h, 0, qp0 ); - qpc1 = get_chroma_qp( h, 0, qp1 ); + int chroma444 = CHROMA444; + int chroma422 = CHROMA422; + + int mb_xy = h->mb_xy; + int left_type= h->left_type[LTOP]; + int top_type= h->top_type; + + int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); + int a = h->slice_alpha_c0_offset - qp_bd_offset; + int b = h->slice_beta_offset - qp_bd_offset; + + int mb_type = s->current_picture.f.mb_type[mb_xy]; + int qp = s->current_picture.f.qscale_table[mb_xy]; + int qp0 = s->current_picture.f.qscale_table[mb_xy - 1]; + int qp1 = s->current_picture.f.qscale_table[h->top_mb_xy]; + int qpc = get_chroma_qp( h, 0, qp ); + int qpc0 = get_chroma_qp( h, 0, qp0 ); + int qpc1 = get_chroma_qp( h, 0, qp1 ); qp0 = (qp + qp0 + 1) >> 1; qp1 = (qp + qp1 + 1) >> 1; qpc0 = (qpc + qpc0 + 1) >> 1; qpc1 = (qpc + qpc1 + 1) >> 1; - 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; if( IS_INTRA(mb_type) ) { - int16_t bS4[4] = {4,4,4,4}; - int16_t bS3[4] = {3,3,3,3}; - int16_t *bSH = FIELD_PICTURE ? bS3 : bS4; + static const int16_t bS4[4] = {4,4,4,4}; + static const int16_t bS3[4] = {3,3,3,3}; + const int16_t *bSH = FIELD_PICTURE ? bS3 : bS4; if(left_type) - filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h); + filter_mb_edgev( &img_y[4*0<cbp&7) == 7 ) { + if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) { edges = 4; AV_WN64A(bS[0][0], 0x0002000200020002ULL); AV_WN64A(bS[0][2], 0x0002000200020002ULL); @@ -287,7 +369,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, AV_WN64A(bS[1][2], 0x0002000200020002ULL); } else { 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 mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[LTOP] & (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, @@ -295,38 +377,64 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, } if( IS_INTRA(left_type) ) AV_WN64A(bS[0][0], 0x0004000400040004ULL); - if( IS_INTRA(h->top_type) ) + if( IS_INTRA(top_type) ) AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL); -#define FILTER(hv,dir,edge)\ +#define FILTER(hv,dir,edge,intra)\ 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(chroma && !(edge&1)) {\ - 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 );\ + filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1<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; + } + +#if CONFIG_SMALL + h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift); +#else + if(h->pixel_shift){ + h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1); + }else{ + h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0); + } +#endif +} + static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){ int v; @@ -356,12 +464,14 @@ static int check_mv(H264Context *h, long b_idx, long bn_idx, int 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 chroma, int chroma444, int dir) { +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 a, int b, int chroma, int dir) { MpegEncContext * const s = &h->s; int edge; int chroma_qp_avg[2]; + int chroma444 = CHROMA444; + int chroma422 = CHROMA422; const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; - const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type; + const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type; // how often to recheck mv-based bS when iterating between edges static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1}, @@ -389,14 +499,14 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u 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]) ) { + if (IS_INTRA(mb_type | s->current_picture.f.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]); + if (!CABAC && IS_8x8DCT(s->current_picture.f.mb_type[mbn_xy])) { + bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]); + bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]); + bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]); + bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]); }else{ const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4; int i; @@ -407,19 +517,19 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u } // 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.f.qscale_table[mb_xy] + s->current_picture.f.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 ); - chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; - chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; + filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, a, b, h, 0 ); + chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1; + chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1; if (chroma) { if (chroma444) { - filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h); - filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h); + filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0); + filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0); } else { - filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h); - filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h); + filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0); + filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0); } } } @@ -472,32 +582,30 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u // Do not use s->qscale as luma quantizer because it has not the same // value in IPCM macroblocks. 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]); + qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbm_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, linesize, uvlinesize); - //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); } - chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1; - chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1; + chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1; + chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1; if( dir == 0 ) { - filter_mb_edgev( &img_y[0], linesize, bS, qp, h ); + filter_mb_edgev( &img_y[0], linesize, bS, qp, a, b, h, 1 ); if (chroma) { if (chroma444) { - filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h); - filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h); + filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1); + filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1); } else { - filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h); - filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h); + filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1); + filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1); } } } else { - filter_mb_edgeh( &img_y[0], linesize, bS, qp, h ); + filter_mb_edgeh( &img_y[0], linesize, bS, qp, a, b, h, 1 ); if (chroma) { if (chroma444) { - filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h); - filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h); + filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1); + filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1); } else { - filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h); - filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h); + filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1); + filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1); } } } @@ -509,8 +617,9 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u for( edge = 1; edge < edges; edge++ ) { DECLARE_ALIGNED(8, int16_t, bS)[4]; int qp; + const int deblock_edge = !IS_8x8DCT(mb_type & (edge<<24)); // (edge&1) && IS_8x8DCT(mb_type) - if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type) + if (!deblock_edge && (!chroma422 || dir == 0)) continue; if( IS_INTRA(mb_type)) { @@ -556,30 +665,37 @@ 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]; - //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]); + qp = s->current_picture.f.qscale_table[mb_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[4*edge << h->pixel_shift], linesize, bS, qp, h ); + filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, a, b, h, 0 ); if (chroma) { if (chroma444) { - filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h); - filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h); + filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0); + filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0); } else if( (edge&1) == 0 ) { - filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h); - filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h); + filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0); + filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0); } } } else { - filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h ); - if (chroma) { - if (chroma444) { - filter_mb_edgeh ( &img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h); - filter_mb_edgeh ( &img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h); - } else if( (edge&1) == 0 ) { - 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); + if (chroma422) { + if (deblock_edge) + filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0); + if (chroma) { + filter_mb_edgech(&img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0); + filter_mb_edgech(&img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0); + } + } else { + filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0); + if (chroma) { + if (chroma444) { + filter_mb_edgeh (&img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0); + filter_mb_edgeh (&img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0); + } else if ((edge&1) == 0) { + filter_mb_edgech(&img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0); + filter_mb_edgech(&img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0); + } } } } @@ -589,17 +705,20 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u void ff_h264_filter_mb( 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; const int mb_xy= mb_x + mb_y*s->mb_stride; - const int mb_type = s->current_picture.mb_type[mb_xy]; + const int mb_type = s->current_picture.f.mb_type[mb_xy]; const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4; int first_vertical_edge_done = 0; av_unused int dir; int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY)); + int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); + int a = h->slice_alpha_c0_offset - qp_bd_offset; + int b = h->slice_beta_offset - qp_bd_offset; if (FRAME_MBAFF // and current and left pair do not have the same interlaced type - && IS_INTERLACED(mb_type^h->left_type[0]) + && IS_INTERLACED(mb_type^h->left_type[LTOP]) // and left mb is in available to us - && h->left_type[0]) { + && h->left_type[LTOP]) { /* First vertical edge is different in MBAFF frames * There are 8 different bS to compute and 2 different Qp */ @@ -627,24 +746,24 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint const uint8_t *off= offset[MB_FIELD][mb_y&1]; for( i = 0; i < 8; i++ ) { int j= MB_FIELD ? i>>2 : i&1; - int mbn_xy = h->left_mb_xy[j]; - int mbn_type= h->left_type[j]; + int mbn_xy = h->left_mb_xy[LEFT(j)]; + int mbn_type= h->left_type[LEFT(j)]; if( IS_INTRA( mbn_type ) ) bS[i] = 4; 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->cbp_table[mbn_xy] & (((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2) << 12)) : h->non_zero_count[mbn_xy][ off[i] ])); } } } - mb_qp = s->current_picture.qscale_table[mb_xy]; - mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]]; - mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]]; + mb_qp = s->current_picture.f.qscale_table[mb_xy]; + mbn0_qp = s->current_picture.f.qscale_table[h->left_mb_xy[0]]; + mbn1_qp = s->current_picture.f.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; @@ -660,31 +779,50 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint 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"); } 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_edgev ( h, img_y , linesize, bS , 1, qp [0], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1], a, b, 1 ); if (chroma){ - 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] ); + if (CHROMA444) { + filter_mb_mbaff_edgev ( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 ); + } else if (CHROMA422) { + filter_mb_mbaff_edgecv(h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1); + filter_mb_mbaff_edgecv(h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1); + filter_mb_mbaff_edgecv(h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1); + filter_mb_mbaff_edgecv(h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1); + }else{ + filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 ); + filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 ); + filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 ); + filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 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_edgev ( h, img_y , 2* linesize, bS , 2, qp [0], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1], a, b, 1 ); if (chroma){ - 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 (CHROMA444) { + filter_mb_mbaff_edgev ( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 ); + filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 ); + }else{ + filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 ); + filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 ); + filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 ); + filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 ); + } } } } #if CONFIG_SMALL for( dir = 0; dir < 2; dir++ ) - filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, chroma, CHROMA444, dir); + filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, a, b, chroma, dir); #else - filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, chroma, CHROMA444, 0); - filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, chroma, CHROMA444, 1); + filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, a, b, chroma, 0); + filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, a, b, chroma, 1); #endif }