X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fh264_loopfilter.c;h=b045d2388a74524fd572eb4786ac47ee5a34d5d9;hb=9a07c1332cfe092b57b5758f22b686ca58806c60;hp=28557531e4b0d946877659696fd1eda17f8c7259;hpb=16e5e39ab418e5450550da7a62fce29c782cd2c8;p=ffmpeg diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c index 28557531e4b..b045d2388a7 100644 --- a/libavcodec/h264_loopfilter.c +++ b/libavcodec/h264_loopfilter.c @@ -2,29 +2,30 @@ * H.26L/H.264/AVC/JVT/14496-10/... loop filter * Copyright (c) 2003 Michael Niedermayer * - * This file is part of FFmpeg. + * This file is part of Libav. * - * FFmpeg is free software; you can redistribute it and/or + * Libav is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * - * FFmpeg is distributed in the hope that it will be useful, + * Libav is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software + * License along with Libav; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ /** - * @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" @@ -99,329 +100,378 @@ 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( 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; +/* 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 + 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]]; tc[2] = tc0_table[index_a][bS[2]]; tc[3] = tc0_table[index_a][bS[3]]; - h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc); + 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( 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; + +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 + 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; tc[2] = tc0_table[index_a][bS[2]]+1; tc[3] = tc0_table[index_a][bS[3]]+1; - h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc); + 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[4], int bsi, int qp ) { - int 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 < 8; i++, pix += stride) { - const int bS_index = (i >> 1) * bsi; - - if( bS[bS_index] == 0 ) { - continue; - } - - if( bS[bS_index] < 4 ) { - 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]; - const int q0 = pix[0]; - const int q1 = pix[1]; - const int q2 = pix[2]; - - if( FFABS( p0 - q0 ) < alpha && - FFABS( p1 - p0 ) < beta && - FFABS( q1 - q0 ) < beta ) { - int tc = tc0; - 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++; - } - - i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); - pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */ - pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ - tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1); - } - }else{ - const int p0 = pix[-1]; - const int p1 = pix[-2]; - const int p2 = pix[-3]; - - const int q0 = pix[0]; - const int q1 = pix[1]; - const int q2 = pix[2]; - - if( FFABS( p0 - q0 ) < alpha && - FFABS( p1 - p0 ) < beta && - FFABS( q1 - q0 ) < beta ) { +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(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ - if( FFABS( p2 - p0 ) < beta) - { - const int p3 = pix[-4]; - /* p0', p1', p2' */ - pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; - pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; - pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; - } else { - /* p0' */ - pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; - } - if( FFABS( q2 - q0 ) < beta) - { - const int q3 = pix[3]; - /* q0', q1', q2' */ - pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; - pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; - pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; - } else { - /* q0' */ - pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; - } - }else{ - /* p0', q0' */ - pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; - pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; - } - tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]); - } - } + 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]]; + tc[2] = tc0_table[index_a][bS[2*bsi]]; + tc[3] = tc0_table[index_a][bS[3*bsi]]; + h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc); + } else { + 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[4], int bsi, int qp ) { - int 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; - } - if( bS[bS_index] < 4 ) { - 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]; - const int q1 = pix[1]; - - if( FFABS( p0 - q0 ) < alpha && - FFABS( p1 - p0 ) < beta && - FFABS( q1 - q0 ) < beta ) { - const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); - - pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */ - pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */ - tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1); - } - }else{ - const int p0 = pix[-1]; - const int p1 = pix[-2]; - const int q0 = pix[0]; - const int q1 = pix[1]; - - if( FFABS( p0 - q0 ) < alpha && - FFABS( p1 - p0 ) < beta && - FFABS( q1 - q0 ) < beta ) { - - pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ - pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ - tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]); - } - } +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 || !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; + tc[2] = tc0_table[index_a][bS[2*bsi]] + 1; + tc[3] = tc0_table[index_a][bS[3*bsi]] + 1; + h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc); + } else { + h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta); } } -static void av_noinline 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; +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 + 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]]; tc[2] = tc0_table[index_a][bS[2]]; tc[3] = tc0_table[index_a][bS[3]]; - h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc); + 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( 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; +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 + 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; tc[2] = tc0_table[index_a][bS[2]]+1; tc[3] = tc0_table[index_a][bS[3]]+1; - h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc); + 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) { +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; - int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh; - - mb_xy = h->mb_xy; - - if(!h->top_type || !h->left_type[0] || !s->dsp.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); - - 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 chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY)); + 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<cbp&7) == 7 ) { + if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) { 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)) - && (h->left_type[0] & (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, + 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[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, h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE); } - if( IS_INTRA(h->left_type[0]) ) - bSv[0][0] = 0x0004000400040004ULL; - if( IS_INTRA(h->top_type) ) - bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL; - -#define FILTER(hv,dir,edge)\ - if(bSv[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( &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( IS_INTRA(left_type) ) + AV_WN64A(bS[0][0], 0x0004000400040004ULL); + if( IS_INTRA(top_type) ) + AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL); + +#define FILTER(hv,dir,edge,intra)\ + if(AV_RN64A(bS[dir][edge])) { \ + 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; + } -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) { +#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; + + 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 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}, @@ -431,103 +481,162 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u // 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)); - int start = h->slice_table[mbm_xy] == 0xFFFF - || first_vertical_edge_done - || (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_num); - - - if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 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]; + + 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.f.mb_type[mbn_xy])) { + AV_WN64A(bS, 0x0003000300030003ULL); + } else { + 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; + 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.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, 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], 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], a, b, h, 0); + filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0); + } + } + } + }else{ + DECLARE_ALIGNED(8, int16_t, bS)[4]; int qp; - if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) { - *(uint64_t*)bS= 0x0003000300030003ULL; + + 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] + 4+3*8; 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++ ) { - bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]); + 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 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( &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); + if(bS[0]+bS[1]+bS[2]+bS[3]){ + 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); + 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, a, b, h, 1 ); + if (chroma) { + if (chroma444) { + 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], 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, a, b, h, 1 ); + if (chroma) { + if (chroma444) { + 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], a, b, h, 1); + filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1); + } + } + } + } } - - 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]; - DECLARE_ALIGNED_8(int16_t, bS)[4]; + 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|mbn_type)) { - *(uint64_t*)bS= 0x0003000300030003ULL; - if (edge == 0) { - if ( (!IS_INTERLACED(mb_type|mbm_type)) - || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) - ) - *(uint64_t*)bS= 0x0004000400040004ULL; - } + if( IS_INTRA(mb_type)) { + AV_WN64A(bS, 0x0003000300030003ULL); } else { - int i, l; + int i; int mv_done; if( edge & mask_edge ) { - *(uint64_t*)bS= 0; - mv_done = 1; - } - else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) { - *(uint64_t*)bS= 0x0001000100010001ULL; + 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 < h->list_count; l++ ) { - v |= h->ref_cache[l][b_idx] != 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->list_count==2 && v){ - v=0; - for( l = 0; !v && l < 2; l++ ) { - int ln= 1-l; - v |= h->ref_cache[l][b_idx] != 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 @@ -545,28 +654,7 @@ 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 < h->list_count; l++ ) { - if( h->ref_cache[l][b_idx] != 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->list_count == 2 && bS[i]){ - bS[i] = 0; - for( l = 0; l < 2; l++ ) { - int ln= 1-l; - if( h->ref_cache[l][b_idx] != 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); } } @@ -577,27 +665,38 @@ 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; - //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], linesize, bS, qp, h ); - if( (edge&1) == 0 ) { - int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; - filter_mb_edgecv( &img_cb[2*edge], 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[mbn_xy] ) + 1 ) >> 1; - filter_mb_edgecv( &img_cr[2*edge], uvlinesize, 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], 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], 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( (edge&1) == 0 ) { - int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; - filter_mb_edgech( &img_cb[2*edge*uvlinesize], 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[mbn_xy] ) + 1 ) >> 1; - filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, qp, 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); + } + } } } } @@ -606,21 +705,24 @@ 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 list; + 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]) - // and left mb is in the same slice if deblocking_filter == 2 - && h->left_type[0]) { + && IS_INTERLACED(mb_type^h->left_type[LTOP]) + // and left mb is in available to us + && h->left_type[LTOP]) { /* First vertical edge is different in MBAFF frames * There are 8 different bS to compute and 2 different Qp */ - DECLARE_ALIGNED_8(int16_t, bS)[8]; + DECLARE_ALIGNED(8, int16_t, bS)[8]; int qp[2]; int bqp[2]; int rqp[2]; @@ -628,30 +730,40 @@ 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) ) - *(uint64_t*)&bS[0]= - *(uint64_t*)&bS[4]= 0x0004000400040004ULL; - 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]={ + { + {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1}, + {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3}, + },{ + {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3}, + {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3}, + } + }; + 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][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8])); + 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; @@ -667,27 +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_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] ); + 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){ + 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_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] ); + 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){ + 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, 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, 0); - filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 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 }