Detect byte-swapped AC-3 and support decoding it directly.
[ffmpeg.git] / libavcodec / h264_loopfilter.c
index c390a4f..619ab19 100644 (file)
@@ -2,29 +2,30 @@
  * H.26L/H.264/AVC/JVT/14496-10/... loop filter
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * 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 <michaelni@gmx.at>
  */
 
+#include "libavutil/intreadwrite.h"
 #include "internal.h"
 #include "dsputil.h"
 #include "avcodec.h"
 #include "mathops.h"
 #include "rectangle.h"
 
-#if ARCH_X86
-#include "x86/h264_i386.h"
-#endif
-
 //#undef NDEBUG
 #include <assert.h>
 
@@ -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);
@@ -439,137 +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.
-        //
-        static const int nnz_idx[4] = {4,5,6,3};
-        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[nnz_idx[i]] != 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;
+                AV_ZERO64(bS);
                 mv_done = 1;
             }
-            else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {
-                bS[0] = bS[1] = bS[2] = bS[3] = 1;
-                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
@@ -587,28 +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);
                 }
             }
 
@@ -619,25 +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;
-        //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, s->current_picture.qscale_table[mbn_xy]);
+        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);
             }
         }
     }
@@ -651,68 +661,15 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint
     int first_vertical_edge_done = 0;
     av_unused int dir;
 
-    //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 & 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 & 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 & 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 & 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];
@@ -720,28 +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][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2]))
-                    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;
@@ -756,9 +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