h264: fix loopfilter with threading at slice boundaries.
[ffmpeg.git] / libavcodec / h264.c
index 9883d65..5338146 100644 (file)
 /*
- * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
+ * H.26L/H.264/AVC/JVT/14496-10/... decoder
  * 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.c
+ * @file
  * H.264 / AVC / MPEG4 part10 codec.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
+#include "libavutil/imgutils.h"
 #include "internal.h"
 #include "dsputil.h"
 #include "avcodec.h"
 #include "mpegvideo.h"
 #include "h264.h"
 #include "h264data.h"
-#include "h264_parser.h"
+#include "h264_mvpred.h"
 #include "golomb.h"
 #include "mathops.h"
 #include "rectangle.h"
 #include "vdpau_internal.h"
+#include "libavutil/avassert.h"
 
 #include "cabac.h"
-#if ARCH_X86
-#include "x86/h264_i386.h"
-#endif
 
 //#undef NDEBUG
 #include <assert.h>
 
-static VLC coeff_token_vlc[4];
-static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
-static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
-
-static VLC chroma_dc_coeff_token_vlc;
-static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
-static const int chroma_dc_coeff_token_vlc_table_size = 256;
-
-static VLC total_zeros_vlc[15];
-static VLC_TYPE total_zeros_vlc_tables[15][512][2];
-static const int total_zeros_vlc_tables_size = 512;
-
-static VLC chroma_dc_total_zeros_vlc[3];
-static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
-static const int chroma_dc_total_zeros_vlc_tables_size = 8;
-
-static VLC run_vlc[6];
-static VLC_TYPE run_vlc_tables[6][8][2];
-static const int run_vlc_tables_size = 8;
-
-static VLC run7_vlc;
-static VLC_TYPE run7_vlc_table[96][2];
-static const int run7_vlc_table_size = 96;
-
-static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
-static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
-
-static const uint8_t rem6[52]={
-0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
+static const uint8_t rem6[QP_MAX_NUM+1]={
+0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
 };
 
-static const uint8_t div6[52]={
-0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
+static const uint8_t div6[QP_MAX_NUM+1]={
+0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
 };
 
-static const uint8_t left_block_options[4][8]={
-    {0,1,2,3,7,10,8,11},
-    {2,2,3,3,8,11,8,11},
-    {0,0,1,1,7,10,7,10},
-    {0,2,0,2,7,10,7,10}
+static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
+    PIX_FMT_DXVA2_VLD,
+    PIX_FMT_VAAPI_VLD,
+    PIX_FMT_YUVJ420P,
+    PIX_FMT_NONE
 };
 
-#define LEVEL_TAB_BITS 8
-static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
-
-static void fill_caches(H264Context *h, int mb_type, int for_deblock){
-    MpegEncContext * const s = &h->s;
-    const int mb_xy= h->mb_xy;
-    int topleft_xy, top_xy, topright_xy, left_xy[2];
-    int topleft_type, top_type, topright_type, left_type[2];
-    const uint8_t * left_block;
-    int topleft_partition= -1;
-    int i;
-
-    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
-
-    //FIXME deblocking could skip the intra and nnz parts.
-    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
-        return;
-
-    /* Wow, what a mess, why didn't they simplify the interlacing & intra
-     * stuff, I can't imagine that these complex rules are worth it. */
-
-    topleft_xy = top_xy - 1;
-    topright_xy= top_xy + 1;
-    left_xy[1] = left_xy[0] = mb_xy-1;
-    left_block = left_block_options[0];
-    if(FRAME_MBAFF){
-        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
-        const int top_pair_xy      = pair_xy     - s->mb_stride;
-        const int topleft_pair_xy  = top_pair_xy - 1;
-        const int topright_pair_xy = top_pair_xy + 1;
-        const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
-        const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
-        const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
-        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
-        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
-        const int bottom = (s->mb_y & 1);
-        tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
-
-        if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
-            top_xy -= s->mb_stride;
-        }
-        if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
-            topleft_xy -= s->mb_stride;
-        } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
-            topleft_xy += s->mb_stride;
-            // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
-            topleft_partition = 0;
-        }
-        if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
-            topright_xy -= s->mb_stride;
-        }
-        if (left_mb_field_flag != curr_mb_field_flag) {
-            left_xy[1] = left_xy[0] = pair_xy - 1;
-            if (curr_mb_field_flag) {
-                left_xy[1] += s->mb_stride;
-                left_block = left_block_options[3];
-            } else {
-                left_block= left_block_options[2 - bottom];
-            }
-        }
-    }
-
-    h->top_mb_xy = top_xy;
-    h->left_mb_xy[0] = left_xy[0];
-    h->left_mb_xy[1] = left_xy[1];
-    if(for_deblock){
-        topleft_type = 0;
-        topright_type = 0;
-        top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
-        left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
-        left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
-
-        if(MB_MBAFF && !IS_INTRA(mb_type)){
-            int list;
-            for(list=0; list<h->list_count; list++){
-                //These values where changed for ease of performing MC, we need to change them back
-                //FIXME maybe we can make MC and loop filter use the same values or prevent
-                //the MC code from changing ref_cache and rather use a temporary array.
-                if(USES_LIST(mb_type,list)){
-                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
-                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
-                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
-                    ref += h->b8_stride;
-                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
-                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
-                }
-            }
-        }
-    }else{
-        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
-        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
-        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
-        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
-        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
-
-    if(IS_INTRA(mb_type)){
-        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
-        h->topleft_samples_available=
-        h->top_samples_available=
-        h->left_samples_available= 0xFFFF;
-        h->topright_samples_available= 0xEEEA;
-
-        if(!(top_type & type_mask)){
-            h->topleft_samples_available= 0xB3FF;
-            h->top_samples_available= 0x33FF;
-            h->topright_samples_available= 0x26EA;
-        }
-        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
-            if(IS_INTERLACED(mb_type)){
-                if(!(left_type[0] & type_mask)){
-                    h->topleft_samples_available&= 0xDFFF;
-                    h->left_samples_available&= 0x5FFF;
-                }
-                if(!(left_type[1] & type_mask)){
-                    h->topleft_samples_available&= 0xFF5F;
-                    h->left_samples_available&= 0xFF5F;
-                }
-            }else{
-                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
-                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
-                assert(left_xy[0] == left_xy[1]);
-                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
-                    h->topleft_samples_available&= 0xDF5F;
-                    h->left_samples_available&= 0x5F5F;
-                }
-            }
-        }else{
-            if(!(left_type[0] & type_mask)){
-                h->topleft_samples_available&= 0xDF5F;
-                h->left_samples_available&= 0x5F5F;
-            }
-        }
-
-        if(!(topleft_type & type_mask))
-            h->topleft_samples_available&= 0x7FFF;
-
-        if(!(topright_type & type_mask))
-            h->topright_samples_available&= 0xFBFF;
-
-        if(IS_INTRA4x4(mb_type)){
-            if(IS_INTRA4x4(top_type)){
-                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
-                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
-                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
-                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
-            }else{
-                int pred;
-                if(!(top_type & type_mask))
-                    pred= -1;
-                else{
-                    pred= 2;
-                }
-                h->intra4x4_pred_mode_cache[4+8*0]=
-                h->intra4x4_pred_mode_cache[5+8*0]=
-                h->intra4x4_pred_mode_cache[6+8*0]=
-                h->intra4x4_pred_mode_cache[7+8*0]= pred;
-            }
-            for(i=0; i<2; i++){
-                if(IS_INTRA4x4(left_type[i])){
-                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
-                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
-                }else{
-                    int pred;
-                    if(!(left_type[i] & type_mask))
-                        pred= -1;
-                    else{
-                        pred= 2;
-                    }
-                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
-                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
-                }
-            }
-        }
-    }
-    }
-
-
-/*
-0 . T T. T T T T
-1 L . .L . . . .
-2 L . .L . . . .
-3 . T TL . . . .
-4 L . .L . . . .
-5 L . .. . . . .
-*/
-//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
-    if(top_type){
-        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
-        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
-        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
-        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
-
-        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
-        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
-
-        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
-        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
-
-    }else{
-        h->non_zero_count_cache[4+8*0]=
-        h->non_zero_count_cache[5+8*0]=
-        h->non_zero_count_cache[6+8*0]=
-        h->non_zero_count_cache[7+8*0]=
-
-        h->non_zero_count_cache[1+8*0]=
-        h->non_zero_count_cache[2+8*0]=
-
-        h->non_zero_count_cache[1+8*3]=
-        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
-
-    }
-
-    for (i=0; i<2; i++) {
-        if(left_type[i]){
-            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
-            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
-            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
-            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
-        }else{
-            h->non_zero_count_cache[3+8*1 + 2*8*i]=
-            h->non_zero_count_cache[3+8*2 + 2*8*i]=
-            h->non_zero_count_cache[0+8*1 +   8*i]=
-            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
-        }
-    }
-
-    if( h->pps.cabac ) {
-        // top_cbp
-        if(top_type) {
-            h->top_cbp = h->cbp_table[top_xy];
-        } else if(IS_INTRA(mb_type)) {
-            h->top_cbp = 0x1C0;
-        } else {
-            h->top_cbp = 0;
-        }
-        // left_cbp
-        if (left_type[0]) {
-            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
-        } else if(IS_INTRA(mb_type)) {
-            h->left_cbp = 0x1C0;
-        } else {
-            h->left_cbp = 0;
-        }
-        if (left_type[0]) {
-            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
-        }
-        if (left_type[1]) {
-            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
-        }
-    }
-
-#if 1
-    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
-        int list;
-        for(list=0; list<h->list_count; list++){
-            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
-                /*if(!h->mv_cache_clean[list]){
-                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
-                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
-                    h->mv_cache_clean[list]= 1;
-                }*/
-                continue;
-            }
-            h->mv_cache_clean[list]= 0;
-
-            if(USES_LIST(top_type, list)){
-                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
-                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
-                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
-                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
-                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
-                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
-                h->ref_cache[list][scan8[0] + 0 - 1*8]=
-                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
-                h->ref_cache[list][scan8[0] + 2 - 1*8]=
-                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
-            }else{
-                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
-                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
-                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
-                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
-                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
-            }
-
-            for(i=0; i<2; i++){
-                int cache_idx = scan8[0] - 1 + i*2*8;
-                if(USES_LIST(left_type[i], list)){
-                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
-                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
-                    *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
-                    *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
-                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
-                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
-                }else{
-                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
-                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
-                    h->ref_cache[list][cache_idx  ]=
-                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-                }
-            }
-
-            if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
-                continue;
-
-            if(USES_LIST(topleft_type, list)){
-                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
-                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
-                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
-                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
-            }else{
-                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
-                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-            }
-
-            if(USES_LIST(topright_type, list)){
-                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
-                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
-                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
-                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
-            }else{
-                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
-                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-            }
-
-            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
-                continue;
-
-            h->ref_cache[list][scan8[5 ]+1] =
-            h->ref_cache[list][scan8[7 ]+1] =
-            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
-            h->ref_cache[list][scan8[4 ]] =
-            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
-            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
-            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
-            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
-            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
-            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
-
-            if( h->pps.cabac ) {
-                /* XXX beurk, Load mvd */
-                if(USES_LIST(top_type, list)){
-                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
-                }else{
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
-                }
-                if(USES_LIST(left_type[0], list)){
-                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
-                }else{
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
-                }
-                if(USES_LIST(left_type[1], list)){
-                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
-                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
-                }else{
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
-                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
-                }
-                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
-                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
-                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
-                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
-                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
-
-                if(h->slice_type_nos == FF_B_TYPE){
-                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
-
-                    if(IS_DIRECT(top_type)){
-                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
-                    }else if(IS_8X8(top_type)){
-                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
-                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
-                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
-                    }else{
-                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
-                    }
-
-                    if(IS_DIRECT(left_type[0]))
-                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
-                    else if(IS_8X8(left_type[0]))
-                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
-                    else
-                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
-
-                    if(IS_DIRECT(left_type[1]))
-                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
-                    else if(IS_8X8(left_type[1]))
-                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
-                    else
-                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
-                }
-            }
-
-            if(FRAME_MBAFF){
-#define MAP_MVS\
-                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
-                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
-                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
-                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
-                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
-                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
-                if(MB_FIELD){
-#define MAP_F2F(idx, mb_type)\
-                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
-                        h->ref_cache[list][idx] <<= 1;\
-                        h->mv_cache[list][idx][1] /= 2;\
-                        h->mvd_cache[list][idx][1] /= 2;\
-                    }
-                    MAP_MVS
-#undef MAP_F2F
-                }else{
-#define MAP_F2F(idx, mb_type)\
-                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
-                        h->ref_cache[list][idx] >>= 1;\
-                        h->mv_cache[list][idx][1] <<= 1;\
-                        h->mvd_cache[list][idx][1] <<= 1;\
-                    }
-                    MAP_MVS
-#undef MAP_F2F
-                }
-            }
-        }
-    }
-#endif
-
-    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
-}
-
 void ff_h264_write_back_intra_pred_mode(H264Context *h){
-    const int mb_xy= h->mb_xy;
+    int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
 
-    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
-    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
-    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
-    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
-    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
-    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
-    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
+    AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
+    mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
+    mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
+    mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
 }
 
 /**
@@ -615,196 +142,6 @@ int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
     return mode;
 }
 
-/**
- * gets the predicted intra4x4 prediction mode.
- */
-static inline int pred_intra_mode(H264Context *h, int n){
-    const int index8= scan8[n];
-    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
-    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
-    const int min= FFMIN(left, top);
-
-    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
-
-    if(min<0) return DC_PRED;
-    else      return min;
-}
-
-static inline void write_back_non_zero_count(H264Context *h){
-    const int mb_xy= h->mb_xy;
-
-    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
-    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
-    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
-    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
-    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
-    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
-    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
-
-    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
-    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
-    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
-
-    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
-    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
-    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
-}
-
-/**
- * gets the predicted number of non-zero coefficients.
- * @param n block index
- */
-static inline int pred_non_zero_count(H264Context *h, int n){
-    const int index8= scan8[n];
-    const int left= h->non_zero_count_cache[index8 - 1];
-    const int top = h->non_zero_count_cache[index8 - 8];
-    int i= left + top;
-
-    if(i<64) i= (i+1)>>1;
-
-    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
-
-    return i&31;
-}
-
-/**
- * gets the directionally predicted 16x8 MV.
- * @param n the block index
- * @param mx the x component of the predicted motion vector
- * @param my the y component of the predicted motion vector
- */
-static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
-    if(n==0){
-        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
-        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
-
-        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
-
-        if(top_ref == ref){
-            *mx= B[0];
-            *my= B[1];
-            return;
-        }
-    }else{
-        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
-        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
-
-        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
-
-        if(left_ref == ref){
-            *mx= A[0];
-            *my= A[1];
-            return;
-        }
-    }
-
-    //RARE
-    pred_motion(h, n, 4, list, ref, mx, my);
-}
-
-/**
- * gets the directionally predicted 8x16 MV.
- * @param n the block index
- * @param mx the x component of the predicted motion vector
- * @param my the y component of the predicted motion vector
- */
-static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
-    if(n==0){
-        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
-        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
-
-        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
-
-        if(left_ref == ref){
-            *mx= A[0];
-            *my= A[1];
-            return;
-        }
-    }else{
-        const int16_t * C;
-        int diagonal_ref;
-
-        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
-
-        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
-
-        if(diagonal_ref == ref){
-            *mx= C[0];
-            *my= C[1];
-            return;
-        }
-    }
-
-    //RARE
-    pred_motion(h, n, 2, list, ref, mx, my);
-}
-
-static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
-    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
-    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
-
-    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
-
-    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
-       || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
-       || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
-
-        *mx = *my = 0;
-        return;
-    }
-
-    pred_motion(h, 0, 4, 0, 0, mx, my);
-
-    return;
-}
-
-static inline void write_back_motion(H264Context *h, int mb_type){
-    MpegEncContext * const s = &h->s;
-    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
-    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
-    int list;
-
-    if(!USES_LIST(mb_type, 0))
-        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
-
-    for(list=0; list<h->list_count; list++){
-        int y;
-        if(!USES_LIST(mb_type, list))
-            continue;
-
-        for(y=0; y<4; y++){
-            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
-            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
-        }
-        if( h->pps.cabac ) {
-            if(IS_SKIP(mb_type))
-                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
-            else
-            for(y=0; y<4; y++){
-                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
-                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
-            }
-        }
-
-        {
-            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
-            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
-            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
-            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
-            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
-        }
-    }
-
-    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
-        if(IS_8X8(mb_type)){
-            uint8_t *direct_table = &h->direct_table[b8_xy];
-            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
-            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
-            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
-        }
-    }
-}
-
 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
     int i, si, di;
     uint8_t *dst;
@@ -815,20 +152,16 @@ const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_l
     h->nal_unit_type= src[0]&0x1F;
 
     src++; length--;
-#if 0
-    for(i=0; i<length; i++)
-        printf("%2X ", src[i]);
-#endif
 
 #if HAVE_FAST_UNALIGNED
 # if HAVE_FAST_64BIT
 #   define RS 7
     for(i=0; i+1<length; i+=9){
-        if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
+        if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
 # else
 #   define RS 3
     for(i=0; i+1<length; i+=5){
-        if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
+        if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
 # endif
             continue;
         if(i>0 && !src[i]) i--;
@@ -895,7 +228,11 @@ nsc:
     return dst;
 }
 
-int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
+/**
+ * Identify the exact end of the bitstream
+ * @return the length of the trailing, or 0 if damaged
+ */
+static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
     int v= *src;
     int r;
 
@@ -908,19 +245,18 @@ int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
     return 0;
 }
 
+#if 0
 /**
- * IDCT transforms the 16 dc values and dequantizes them.
- * @param qp quantization parameter
+ * DCT transforms the 16 dc values.
+ * @param qp quantization parameter ??? FIXME
  */
-static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
-#define stride 16
+static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
+//    const int qmul= dequant_coeff[qp][0];
     int i;
     int temp[16]; //FIXME check if this is a good idea
     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
 
-//memset(block, 64, 2*256);
-//return;
     for(i=0; i<4; i++){
         const int offset= y_offset[i];
         const int z0= block[offset+stride*0] + block[offset+stride*4];
@@ -941,76 +277,16 @@ static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
         const int z2= temp[4*1+i] - temp[4*3+i];
         const int z3= temp[4*1+i] + temp[4*3+i];
 
-        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
-        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
-        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
-        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
+        block[stride*0 +offset]= (z0 + z3)>>1;
+        block[stride*2 +offset]= (z1 + z2)>>1;
+        block[stride*8 +offset]= (z1 - z2)>>1;
+        block[stride*10+offset]= (z0 - z3)>>1;
     }
 }
+#endif
 
-#if 0
-/**
- * DCT transforms the 16 dc values.
- * @param qp quantization parameter ??? FIXME
- */
-static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
-//    const int qmul= dequant_coeff[qp][0];
-    int i;
-    int temp[16]; //FIXME check if this is a good idea
-    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
-    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
-
-    for(i=0; i<4; i++){
-        const int offset= y_offset[i];
-        const int z0= block[offset+stride*0] + block[offset+stride*4];
-        const int z1= block[offset+stride*0] - block[offset+stride*4];
-        const int z2= block[offset+stride*1] - block[offset+stride*5];
-        const int z3= block[offset+stride*1] + block[offset+stride*5];
-
-        temp[4*i+0]= z0+z3;
-        temp[4*i+1]= z1+z2;
-        temp[4*i+2]= z1-z2;
-        temp[4*i+3]= z0-z3;
-    }
-
-    for(i=0; i<4; i++){
-        const int offset= x_offset[i];
-        const int z0= temp[4*0+i] + temp[4*2+i];
-        const int z1= temp[4*0+i] - temp[4*2+i];
-        const int z2= temp[4*1+i] - temp[4*3+i];
-        const int z3= temp[4*1+i] + temp[4*3+i];
-
-        block[stride*0 +offset]= (z0 + z3)>>1;
-        block[stride*2 +offset]= (z1 + z2)>>1;
-        block[stride*8 +offset]= (z1 - z2)>>1;
-        block[stride*10+offset]= (z0 - z3)>>1;
-    }
-}
-#endif
-
-#undef xStride
-#undef stride
-
-static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
-    const int stride= 16*2;
-    const int xStride= 16;
-    int a,b,c,d,e;
-
-    a= block[stride*0 + xStride*0];
-    b= block[stride*0 + xStride*1];
-    c= block[stride*1 + xStride*0];
-    d= block[stride*1 + xStride*1];
-
-    e= a-b;
-    a= a+b;
-    b= c-d;
-    c= c+d;
-
-    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
-    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
-    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
-    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
-}
+#undef xStride
+#undef stride
 
 #if 0
 static void chroma_dc_dct_c(DCTELEM *block){
@@ -1038,12 +314,13 @@ static void chroma_dc_dct_c(DCTELEM *block){
 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                            int src_x_offset, int src_y_offset,
-                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
+                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
+                           int pixel_shift){
     MpegEncContext * const s = &h->s;
     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
     const int luma_xy= (mx&3) + ((my&3)<<2);
-    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
+    uint8_t * src_y = pic->data[0] + ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
     uint8_t * src_cb, * src_cr;
     int extra_width= h->emu_edge_width;
     int extra_height= h->emu_edge_height;
@@ -1060,8 +337,8 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
        || full_my < 0-extra_height
        || full_mx + 16/*FIXME*/ > pic_width + extra_width
        || full_my + 16/*FIXME*/ > pic_height + extra_height){
-        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
-            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+            src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
         emu=1;
     }
 
@@ -1077,17 +354,17 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
         my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
     }
-    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
-    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
+    src_cb= pic->data[1] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
+    src_cr= pic->data[2] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
 
     if(emu){
-        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
             src_cb= s->edge_emu_buffer;
     }
     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 
     if(emu){
-        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
             src_cr= s->edge_emu_buffer;
     }
     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
@@ -1098,14 +375,14 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
                            int x_offset, int y_offset,
                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
-                           int list0, int list1){
+                           int list0, int list1, int pixel_shift){
     MpegEncContext * const s = &h->s;
     qpel_mc_func *qpix_op=  qpix_put;
     h264_chroma_mc_func chroma_op= chroma_put;
 
-    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
-    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
-    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
+    dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->  mb_linesize;
+    dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
+    dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
     x_offset += 8*s->mb_x;
     y_offset += 8*(s->mb_y >> MB_FIELD);
 
@@ -1113,7 +390,7 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op);
+                           qpix_op, chroma_op, pixel_shift);
 
         qpix_op=  qpix_avg;
         chroma_op= chroma_avg;
@@ -1123,7 +400,7 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op);
+                           qpix_op, chroma_op, pixel_shift);
     }
 }
 
@@ -1133,12 +410,12 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
-                           int list0, int list1){
+                           int list0, int list1, int pixel_shift){
     MpegEncContext * const s = &h->s;
 
-    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
-    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
-    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
+    dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->  mb_linesize;
+    dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
+    dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
     x_offset += 8*s->mb_x;
     y_offset += 8*(s->mb_y >> MB_FIELD);
 
@@ -1146,34 +423,34 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         /* don't optimize for luma-only case, since B-frames usually
          * use implicit weights => chroma too. */
         uint8_t *tmp_cb = s->obmc_scratchpad;
-        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
+        uint8_t *tmp_cr = s->obmc_scratchpad + (8 << pixel_shift);
         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
         int refn0 = h->ref_cache[0][ scan8[n] ];
         int refn1 = h->ref_cache[1][ scan8[n] ];
 
         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
                     dest_y, dest_cb, dest_cr,
-                    x_offset, y_offset, qpix_put, chroma_put);
+                    x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
                     tmp_y, tmp_cb, tmp_cr,
-                    x_offset, y_offset, qpix_put, chroma_put);
+                    x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
 
         if(h->use_weight == 2){
-            int weight0 = h->implicit_weight[refn0][refn1];
+            int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
             int weight1 = 64 - weight0;
             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
         }else{
             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
-                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
-                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
+                            h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
+                            h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
-                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
-                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
+                            h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
+                            h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
-                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
-                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
+                            h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
+                            h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
         }
     }else{
         int list = list1 ? 1 : 0;
@@ -1181,15 +458,15 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         Picture *ref= &h->ref_list[list][refn];
         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                    qpix_put, chroma_put);
+                    qpix_put, chroma_put, pixel_shift);
 
         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
-                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
+                       h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
         if(h->use_weight_chroma){
             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
-                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
+                             h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
-                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
+                             h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
         }
     }
 }
@@ -1200,19 +477,21 @@ static inline void mc_part(H264Context *h, int n, int square, int chroma_height,
                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,
-                           int list0, int list1){
+                           int list0, int list1, int pixel_shift){
     if((h->use_weight==2 && list0 && list1
-        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
+        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
        || h->use_weight==1)
         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
                          x_offset, y_offset, qpix_put, chroma_put,
-                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
+                         weight_op[0], weight_op[3], weight_avg[0],
+                         weight_avg[3], list0, list1, pixel_shift);
     else
         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
-                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
+                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
+                    chroma_avg, list0, list1, pixel_shift);
 }
 
-static inline void prefetch_motion(H264Context *h, int list){
+static inline void prefetch_motion(H264Context *h, int list, int pixel_shift){
     /* fetch pixels for estimated mv 4 macroblocks ahead
      * optimized for 64byte cache lines */
     MpegEncContext * const s = &h->s;
@@ -1221,48 +500,54 @@ static inline void prefetch_motion(H264Context *h, int list){
         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
         uint8_t **src= h->ref_list[list][refn].data;
-        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
+        int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
         s->dsp.prefetch(src[0]+off, s->linesize, 4);
-        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
+        off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
     }
 }
 
-static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
-                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
+                      h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+                      int pixel_shift){
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
     const int mb_type= s->current_picture.mb_type[mb_xy];
 
     assert(IS_INTER(mb_type));
 
-    prefetch_motion(h, 0);
+    prefetch_motion(h, 0, pixel_shift);
 
     if(IS_16X16(mb_type)){
         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
-                &weight_op[0], &weight_avg[0],
-                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
+                weight_op, weight_avg,
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift);
     }else if(IS_16X8(mb_type)){
-        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
                 &weight_op[1], &weight_avg[1],
-                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
-        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift);
+        mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
                 &weight_op[1], &weight_avg[1],
-                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
+                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
+                pixel_shift);
     }else if(IS_8X16(mb_type)){
         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
                 &weight_op[2], &weight_avg[2],
-                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift);
         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
                 &weight_op[2], &weight_avg[2],
-                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
+                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
+                pixel_shift);
     }else{
         int i;
 
@@ -1278,25 +563,30 @@ static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t
                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
                     &weight_op[3], &weight_avg[3],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
             }else if(IS_SUB_8X4(sub_mb_type)){
-                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n  , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
                     &weight_op[4], &weight_avg[4],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
-                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
+                mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
                     &weight_op[4], &weight_avg[4],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
             }else if(IS_SUB_4X8(sub_mb_type)){
                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
                     &weight_op[5], &weight_avg[5],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
                     &weight_op[5], &weight_avg[5],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
             }else{
                 int j;
                 assert(IS_SUB_4X4(sub_mb_type));
@@ -1306,112 +596,34 @@ static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t
                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
                         &weight_op[6], &weight_avg[6],
-                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                        pixel_shift);
                 }
             }
         }
     }
 
-    prefetch_motion(h, 1);
-}
-
-static av_cold void init_cavlc_level_tab(void){
-    int suffix_length, mask;
-    unsigned int i;
-
-    for(suffix_length=0; suffix_length<7; suffix_length++){
-        for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
-            int prefix= LEVEL_TAB_BITS - av_log2(2*i);
-            int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
-
-            mask= -(level_code&1);
-            level_code= (((2+level_code)>>1) ^ mask) - mask;
-            if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
-                cavlc_level_tab[suffix_length][i][0]= level_code;
-                cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
-            }else if(prefix + 1 <= LEVEL_TAB_BITS){
-                cavlc_level_tab[suffix_length][i][0]= prefix+100;
-                cavlc_level_tab[suffix_length][i][1]= prefix + 1;
-            }else{
-                cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
-                cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
-            }
-        }
-    }
+    prefetch_motion(h, 1, pixel_shift);
 }
 
-static av_cold void decode_init_vlc(void){
-    static int done = 0;
-
-    if (!done) {
-        int i;
-        int offset;
-        done = 1;
-
-        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
-        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
-        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
-                 &chroma_dc_coeff_token_len [0], 1, 1,
-                 &chroma_dc_coeff_token_bits[0], 1, 1,
-                 INIT_VLC_USE_NEW_STATIC);
-
-        offset = 0;
-        for(i=0; i<4; i++){
-            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
-            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
-            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
-                     &coeff_token_len [i][0], 1, 1,
-                     &coeff_token_bits[i][0], 1, 1,
-                     INIT_VLC_USE_NEW_STATIC);
-            offset += coeff_token_vlc_tables_size[i];
-        }
-        /*
-         * This is a one time safety check to make sure that
-         * the packed static coeff_token_vlc table sizes
-         * were initialized correctly.
-         */
-        assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
-
-        for(i=0; i<3; i++){
-            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
-            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
-            init_vlc(&chroma_dc_total_zeros_vlc[i],
-                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
-                     &chroma_dc_total_zeros_len [i][0], 1, 1,
-                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
-                     INIT_VLC_USE_NEW_STATIC);
-        }
-        for(i=0; i<15; i++){
-            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
-            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
-            init_vlc(&total_zeros_vlc[i],
-                     TOTAL_ZEROS_VLC_BITS, 16,
-                     &total_zeros_len [i][0], 1, 1,
-                     &total_zeros_bits[i][0], 1, 1,
-                     INIT_VLC_USE_NEW_STATIC);
-        }
-
-        for(i=0; i<6; i++){
-            run_vlc[i].table = run_vlc_tables[i];
-            run_vlc[i].table_allocated = run_vlc_tables_size;
-            init_vlc(&run_vlc[i],
-                     RUN_VLC_BITS, 7,
-                     &run_len [i][0], 1, 1,
-                     &run_bits[i][0], 1, 1,
-                     INIT_VLC_USE_NEW_STATIC);
-        }
-        run7_vlc.table = run7_vlc_table,
-        run7_vlc.table_allocated = run7_vlc_table_size;
-        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
-                 &run_len [6][0], 1, 1,
-                 &run_bits[6][0], 1, 1,
-                 INIT_VLC_USE_NEW_STATIC);
-
-        init_cavlc_level_tab();
-    }
+#define hl_motion_fn(sh, bits) \
+static av_always_inline void hl_motion_ ## bits(H264Context *h, \
+                                       uint8_t *dest_y, \
+                                       uint8_t *dest_cb, uint8_t *dest_cr, \
+                                       qpel_mc_func (*qpix_put)[16], \
+                                       h264_chroma_mc_func (*chroma_put), \
+                                       qpel_mc_func (*qpix_avg)[16], \
+                                       h264_chroma_mc_func (*chroma_avg), \
+                                       h264_weight_func *weight_op, \
+                                       h264_biweight_func *weight_avg) \
+{ \
+    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
+              qpix_avg, chroma_avg, weight_op, weight_avg, sh); \
 }
+hl_motion_fn(0, 8);
+hl_motion_fn(1, 16);
 
-static void free_tables(H264Context *h){
+static void free_tables(H264Context *h, int free_rbsp){
     int i;
     H264Context *hx;
     av_freep(&h->intra4x4_pred_mode);
@@ -1423,9 +635,10 @@ static void free_tables(H264Context *h){
     av_freep(&h->non_zero_count);
     av_freep(&h->slice_table_base);
     h->slice_table= NULL;
+    av_freep(&h->list_counts);
 
     av_freep(&h->mb2b_xy);
-    av_freep(&h->mb2b8_xy);
+    av_freep(&h->mb2br_xy);
 
     for(i = 0; i < MAX_THREADS; i++) {
         hx = h->thread_context[i];
@@ -1433,17 +646,19 @@ static void free_tables(H264Context *h){
         av_freep(&hx->top_borders[1]);
         av_freep(&hx->top_borders[0]);
         av_freep(&hx->s.obmc_scratchpad);
-        av_freep(&hx->rbsp_buffer[1]);
-        av_freep(&hx->rbsp_buffer[0]);
-        hx->rbsp_buffer_size[0] = 0;
-        hx->rbsp_buffer_size[1] = 0;
+        if (free_rbsp){
+            av_freep(&hx->rbsp_buffer[1]);
+            av_freep(&hx->rbsp_buffer[0]);
+            hx->rbsp_buffer_size[0] = 0;
+            hx->rbsp_buffer_size[1] = 0;
+        }
         if (i) av_freep(&h->thread_context[i]);
     }
 }
 
 static void init_dequant8_coeff_table(H264Context *h){
     int i,q,x;
-    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
+    const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
     h->dequant8_coeff[0] = h->dequant8_buffer[0];
     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 
@@ -1453,11 +668,11 @@ static void init_dequant8_coeff_table(H264Context *h){
             break;
         }
 
-        for(q=0; q<52; q++){
+        for(q=0; q<max_qp+1; q++){
             int shift = div6[q];
             int idx = rem6[q];
             for(x=0; x<64; x++)
-                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
+                h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
                     h->pps.scaling_matrix8[i][x]) << shift;
         }
@@ -1466,7 +681,7 @@ static void init_dequant8_coeff_table(H264Context *h){
 
 static void init_dequant4_coeff_table(H264Context *h){
     int i,j,q,x;
-    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
+    const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
     for(i=0; i<6; i++ ){
         h->dequant4_coeff[i] = h->dequant4_buffer[i];
         for(j=0; j<i; j++){
@@ -1478,11 +693,11 @@ static void init_dequant4_coeff_table(H264Context *h){
         if(j<i)
             continue;
 
-        for(q=0; q<52; q++){
+        for(q=0; q<max_qp+1; q++){
             int shift = div6[q] + 2;
             int idx = rem6[q];
             for(x=0; x<16; x++)
-                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
+                h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
                     h->pps.scaling_matrix4[i][x]) << shift;
         }
@@ -1509,32 +724,33 @@ static void init_dequant_tables(H264Context *h){
 int ff_h264_alloc_tables(H264Context *h){
     MpegEncContext * const s = &h->s;
     const int big_mb_num= s->mb_stride * (s->mb_height+1);
+    const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
     int x,y;
 
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t), fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
 
     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
     for(y=0; y<s->mb_height; y++){
         for(x=0; x<s->mb_width; x++){
             const int mb_xy= x + y*s->mb_stride;
             const int b_xy = 4*x + 4*y*h->b_stride;
-            const int b8_xy= 2*x + 2*y*h->b8_stride;
 
             h->mb2b_xy [mb_xy]= b_xy;
-            h->mb2b8_xy[mb_xy]= b8_xy;
+            h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
         }
     }
 
@@ -1545,27 +761,29 @@ int ff_h264_alloc_tables(H264Context *h){
 
     return 0;
 fail:
-    free_tables(h);
+    free_tables(h, 1);
     return -1;
 }
 
 /**
  * Mimic alloc_tables(), but for every context thread.
  */
-static void clone_tables(H264Context *dst, H264Context *src){
-    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
+static void clone_tables(H264Context *dst, H264Context *src, int i){
+    MpegEncContext * const s = &src->s;
+    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
     dst->non_zero_count           = src->non_zero_count;
     dst->slice_table              = src->slice_table;
     dst->cbp_table                = src->cbp_table;
     dst->mb2b_xy                  = src->mb2b_xy;
-    dst->mb2b8_xy                 = src->mb2b8_xy;
+    dst->mb2br_xy                 = src->mb2br_xy;
     dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
-    dst->mvd_table[0]             = src->mvd_table[0];
-    dst->mvd_table[1]             = src->mvd_table[1];
+    dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
+    dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
     dst->direct_table             = src->direct_table;
+    dst->list_counts              = src->list_counts;
 
     dst->s.obmc_scratchpad = NULL;
-    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
+    ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma);
 }
 
 /**
@@ -1573,14 +791,19 @@ static void clone_tables(H264Context *dst, H264Context *src){
  * Allocate buffers which are not shared amongst multiple threads.
  */
 static int context_init(H264Context *h){
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t)*2, fail)
+
+    h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
+    h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
 
     return 0;
 fail:
     return -1; // free_tables will clean up for us
 }
 
+static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
+
 static av_cold void common_init(H264Context *h){
     MpegEncContext * const s = &h->s;
 
@@ -1588,7 +811,8 @@ static av_cold void common_init(H264Context *h){
     s->height = s->avctx->height;
     s->codec_id= s->avctx->codec->id;
 
-    ff_h264_pred_init(&h->hpc, s->codec_id);
+    ff_h264dsp_init(&h->h264dsp, 8);
+    ff_h264_pred_init(&h->hpc, s->codec_id, 8);
 
     h->dequant_coeff_pps= -1;
     s->unrestricted_mv=1;
@@ -1600,6 +824,54 @@ static av_cold void common_init(H264Context *h){
     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
 }
 
+int ff_h264_decode_extradata(H264Context *h)
+{
+    AVCodecContext *avctx = h->s.avctx;
+
+    if(*(char *)avctx->extradata == 1){
+        int i, cnt, nalsize;
+        unsigned char *p = avctx->extradata;
+
+        h->is_avc = 1;
+
+        if(avctx->extradata_size < 7) {
+            av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
+            return -1;
+        }
+        /* sps and pps in the avcC always have length coded with 2 bytes,
+           so put a fake nal_length_size = 2 while parsing them */
+        h->nal_length_size = 2;
+        // Decode sps from avcC
+        cnt = *(p+5) & 0x1f; // Number of sps
+        p += 6;
+        for (i = 0; i < cnt; i++) {
+            nalsize = AV_RB16(p) + 2;
+            if(decode_nal_units(h, p, nalsize) < 0) {
+                av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
+                return -1;
+            }
+            p += nalsize;
+        }
+        // Decode pps from avcC
+        cnt = *(p++); // Number of pps
+        for (i = 0; i < cnt; i++) {
+            nalsize = AV_RB16(p) + 2;
+            if(decode_nal_units(h, p, nalsize)  != nalsize) {
+                av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
+                return -1;
+            }
+            p += nalsize;
+        }
+        // Now store right nal length size, that will be use to parse all other nals
+        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
+    } else {
+        h->is_avc = 0;
+        if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
+            return -1;
+    }
+    return 0;
+}
+
 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
     H264Context *h= avctx->priv_data;
     MpegEncContext * const s = &h->s;
@@ -1620,19 +892,15 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 
     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
 
-    decode_init_vlc();
+    ff_h264_decode_init_vlc();
 
-    if(avctx->extradata_size > 0 && avctx->extradata &&
-       *(char *)avctx->extradata == 1){
-        h->is_avc = 1;
-        h->got_avcC = 0;
-    } else {
-        h->is_avc = 0;
-    }
+    h->pixel_shift = 0;
+    h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
 
     h->thread_context[0] = h;
     h->outputed_poc = INT_MIN;
     h->prev_poc_msb= 1<<16;
+    h->x264_build = -1;
     ff_h264_reset_sei(h);
     if(avctx->codec_id == CODEC_ID_H264){
         if(avctx->ticks_per_frame == 1){
@@ -1640,12 +908,23 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
         }
         avctx->ticks_per_frame = 2;
     }
+
+    if(avctx->extradata_size > 0 && avctx->extradata &&
+        ff_h264_decode_extradata(h))
+        return -1;
+
+    if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
+        s->avctx->has_b_frames = h->sps.num_reorder_frames;
+        s->low_delay = 0;
+    }
+
     return 0;
 }
 
 int ff_h264_frame_start(H264Context *h){
     MpegEncContext * const s = &h->s;
     int i;
+    const int pixel_shift = h->pixel_shift;
 
     if(MPV_frame_start(s, s->avctx) < 0)
         return -1;
@@ -1662,25 +941,24 @@ int ff_h264_frame_start(H264Context *h){
     assert(s->linesize && s->uvlinesize);
 
     for(i=0; i<16; i++){
-        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
-        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[24+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
     }
     for(i=0; i<4; i++){
         h->block_offset[16+i]=
-        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
         h->block_offset[24+16+i]=
-        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[24+20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
     }
 
     /* can't be in alloc_tables because linesize isn't known there.
      * FIXME: redo bipred weight to not require extra buffer? */
     for(i = 0; i < s->avctx->thread_count; i++)
-        if(!h->thread_context[i]->s.obmc_scratchpad)
+        if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
 
-    /* some macroblocks will be accessed before they're available */
-    if(FRAME_MBAFF || s->avctx->thread_count > 1)
-        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
+    /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
+    memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 
 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 
@@ -1702,12 +980,9 @@ int ff_h264_frame_start(H264Context *h){
 
 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
     MpegEncContext * const s = &h->s;
-    int i;
-    int step    = 1;
-    int offset  = 1;
-    int uvoffset= 1;
+    uint8_t *top_border;
     int top_idx = 1;
-    int skiplast= 0;
+    const int pixel_shift = h->pixel_shift;
 
     src_y  -=   linesize;
     src_cb -= uvlinesize;
@@ -1715,129 +990,129 @@ static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src
 
     if(!simple && FRAME_MBAFF){
         if(s->mb_y&1){
-            offset  = MB_MBAFF ? 1 : 17;
-            uvoffset= MB_MBAFF ? 1 : 9;
-            if(!MB_MBAFF){
-                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
-                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
-                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
-                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
-                }
-            }
-        }else{
             if(!MB_MBAFF){
-                h->left_border[0]= h->top_borders[0][s->mb_x][15];
+                top_border = h->top_borders[0][s->mb_x];
+                AV_COPY128(top_border, src_y + 15*linesize);
+                if (pixel_shift)
+                    AV_COPY128(top_border+16, src_y+15*linesize+16);
                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
-                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
+                    if (pixel_shift) {
+                        AV_COPY128(top_border+32, src_cb+7*uvlinesize);
+                        AV_COPY128(top_border+48, src_cr+7*uvlinesize);
+                    } else {
+                    AV_COPY64(top_border+16, src_cb+7*uvlinesize);
+                    AV_COPY64(top_border+24, src_cr+7*uvlinesize);
+                    }
                 }
-                skiplast= 1;
             }
-            offset  =
-            uvoffset=
-            top_idx = MB_MBAFF ? 0 : 1;
-        }
-        step= MB_MBAFF ? 2 : 1;
+        }else if(MB_MBAFF){
+            top_idx = 0;
+        }else
+            return;
     }
 
+    top_border = h->top_borders[top_idx][s->mb_x];
     // There are two lines saved, the line above the the top macroblock of a pair,
     // and the line above the bottom macroblock
-    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
-    for(i=1; i<17 - skiplast; i++){
-        h->left_border[offset+i*step]= src_y[15+i*  linesize];
-    }
-
-    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
-    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
+    AV_COPY128(top_border, src_y + 16*linesize);
+    if (pixel_shift)
+        AV_COPY128(top_border+16, src_y+16*linesize+16);
 
     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
-        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
-        for(i=1; i<9 - skiplast; i++){
-            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
-            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
+        if (pixel_shift) {
+            AV_COPY128(top_border+32, src_cb+8*uvlinesize);
+            AV_COPY128(top_border+48, src_cr+8*uvlinesize);
+        } else {
+        AV_COPY64(top_border+16, src_cb+8*uvlinesize);
+        AV_COPY64(top_border+24, src_cr+8*uvlinesize);
         }
-        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
-        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
     }
 }
 
-static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
+static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
+                                  uint8_t *src_cb, uint8_t *src_cr,
+                                  int linesize, int uvlinesize,
+                                  int xchg, int simple, int pixel_shift){
     MpegEncContext * const s = &h->s;
-    int temp8, i;
-    uint64_t temp64;
     int deblock_left;
     int deblock_top;
-    int mb_xy;
-    int step    = 1;
-    int offset  = 1;
-    int uvoffset= 1;
     int top_idx = 1;
+    uint8_t *top_border_m1;
+    uint8_t *top_border;
 
     if(!simple && FRAME_MBAFF){
         if(s->mb_y&1){
-            offset  = MB_MBAFF ? 1 : 17;
-            uvoffset= MB_MBAFF ? 1 : 9;
+            if(!MB_MBAFF)
+                return;
         }else{
-            offset  =
-            uvoffset=
             top_idx = MB_MBAFF ? 0 : 1;
         }
-        step= MB_MBAFF ? 2 : 1;
     }
 
     if(h->deblocking_filter == 2) {
-        mb_xy = h->mb_xy;
-        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
-        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
+        deblock_left = h->left_type[0];
+        deblock_top  = h->top_type;
     } else {
         deblock_left = (s->mb_x > 0);
         deblock_top =  (s->mb_y > !!MB_FIELD);
     }
 
-    src_y  -=   linesize + 1;
-    src_cb -= uvlinesize + 1;
-    src_cr -= uvlinesize + 1;
+    src_y  -=   linesize + 1 + pixel_shift;
+    src_cb -= uvlinesize + 1 + pixel_shift;
+    src_cr -= uvlinesize + 1 + pixel_shift;
 
-#define XCHG(a,b,t,xchg)\
-t= a;\
-if(xchg)\
-    a= b;\
-b= t;
+    top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
+    top_border    = h->top_borders[top_idx][s->mb_x];
 
-    if(deblock_left){
-        for(i = !deblock_top; i<16; i++){
-            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
-        }
-        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
-    }
+#define XCHG(a,b,xchg)\
+    if (pixel_shift) {\
+        if (xchg) {\
+            AV_SWAP64(b+0,a+0);\
+            AV_SWAP64(b+8,a+8);\
+        } else {\
+            AV_COPY128(b,a); \
+        }\
+    } else \
+if (xchg) AV_SWAP64(b,a);\
+else      AV_COPY64(b,a);
 
     if(deblock_top){
-        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
-        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
+        if(deblock_left){
+            XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
+        }
+        XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
+        XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
         if(s->mb_x+1 < s->mb_width){
-            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
+            XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
         }
     }
-
     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-        if(deblock_left){
-            for(i = !deblock_top; i<8; i++){
-                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
-                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
-            }
-            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
-            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
-        }
         if(deblock_top){
-            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
-            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
+            if(deblock_left){
+                XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
+                XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
+            }
+            XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
+            XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
         }
     }
 }
 
-static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
+static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
+    if (high_bit_depth) {
+        return AV_RN32A(((int32_t*)mb) + index);
+    } else
+        return AV_RN16A(mb + index);
+}
+
+static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
+    if (high_bit_depth) {
+        AV_WN32A(((int32_t*)mb) + index, value);
+    } else
+        AV_WN16A(mb + index, value);
+}
+
+static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
     MpegEncContext * const s = &h->s;
     const int mb_x= s->mb_x;
     const int mb_y= s->mb_y;
@@ -1853,12 +1128,14 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 
-    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
-    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
-    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
+    dest_y  = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
+    dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
+    dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
 
-    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
-    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
+    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
+    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
+
+    h->list_counts[mb_xy]= h->list_count;
 
     if (!simple && MB_FIELD) {
         linesize   = h->mb_linesize   = s->linesize * 2;
@@ -1893,6 +1170,28 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
     }
 
     if (!simple && IS_INTRA_PCM(mb_type)) {
+        if (pixel_shift) {
+            const int bit_depth = h->sps.bit_depth_luma;
+            int j;
+            GetBitContext gb;
+            init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
+
+            for (i = 0; i < 16; i++) {
+                uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
+                for (j = 0; j < 16; j++)
+                    tmp_y[j] = get_bits(&gb, bit_depth);
+            }
+            for (i = 0; i < 8; i++) {
+                uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
+                for (j = 0; j < 8; j++)
+                    tmp_cb[j] = get_bits(&gb, bit_depth);
+            }
+            for (i = 0; i < 8; i++) {
+                uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
+                for (j = 0; j < 8; j++)
+                    tmp_cr[j] = get_bits(&gb, bit_depth);
+            }
+        } else {
         for (i=0; i<16; i++) {
             memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
         }
@@ -1900,10 +1199,11 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
             memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
             memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
         }
+        }
     } else {
         if(IS_INTRA(mb_type)){
             if(h->deblocking_filter)
-                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
+                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple, pixel_shift);
 
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
@@ -1917,23 +1217,23 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             idct_dc_add =
                             idct_add    = s->dsp.add_pixels8;
                         }else{
-                            idct_dc_add = s->dsp.h264_idct8_dc_add;
-                            idct_add    = s->dsp.h264_idct8_add;
+                            idct_dc_add = h->h264dsp.h264_idct8_dc_add;
+                            idct_add    = h->h264dsp.h264_idct8_add;
                         }
                         for(i=0; i<16; i+=4){
                             uint8_t * const ptr= dest_y + block_offset[i];
                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
-                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
+                                h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16 << pixel_shift), linesize);
                             }else{
                                 const int nnz = h->non_zero_count_cache[ scan8[i] ];
                                 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
                                                             (h->topright_samples_available<<i)&0x4000, linesize);
                                 if(nnz){
-                                    if(nnz == 1 && h->mb[i*16])
-                                        idct_dc_add(ptr, h->mb + i*16, linesize);
+                                    if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
+                                        idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
                                     else
-                                        idct_add   (ptr, h->mb + i*16, linesize);
+                                        idct_add   (ptr, h->mb + (i*16 << pixel_shift), linesize);
                                 }
                             }
                         }
@@ -1942,26 +1242,32 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             idct_dc_add =
                             idct_add    = s->dsp.add_pixels4;
                         }else{
-                            idct_dc_add = s->dsp.h264_idct_dc_add;
-                            idct_add    = s->dsp.h264_idct_add;
+                            idct_dc_add = h->h264dsp.h264_idct_dc_add;
+                            idct_add    = h->h264dsp.h264_idct_add;
                         }
                         for(i=0; i<16; i++){
                             uint8_t * const ptr= dest_y + block_offset[i];
                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
 
                             if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
-                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
+                                h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16 << pixel_shift), linesize);
                             }else{
                                 uint8_t *topright;
                                 int nnz, tr;
+                                uint64_t tr_high;
                                 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
                                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
                                     assert(mb_y || linesize <= block_offset[i]);
                                     if(!topright_avail){
+                                        if (pixel_shift) {
+                                            tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
+                                            topright= (uint8_t*) &tr_high;
+                                        } else {
                                         tr= ptr[3 - linesize]*0x01010101;
                                         topright= (uint8_t*) &tr;
+                                        }
                                     }else
-                                        topright= ptr + 4 - linesize;
+                                        topright= ptr + (4 << pixel_shift) - linesize;
                                 }else
                                     topright= NULL;
 
@@ -1969,12 +1275,12 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                                 nnz = h->non_zero_count_cache[ scan8[i] ];
                                 if(nnz){
                                     if(is_h264){
-                                        if(nnz == 1 && h->mb[i*16])
-                                            idct_dc_add(ptr, h->mb + i*16, linesize);
+                                        if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
+                                            idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
                                         else
-                                            idct_add   (ptr, h->mb + i*16, linesize);
+                                            idct_add   (ptr, h->mb + (i*16 << pixel_shift), linesize);
                                     }else
-                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
+                                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
                                 }
                             }
                         }
@@ -1983,18 +1289,34 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
             }else{
                 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
                 if(is_h264){
-                    if(!transform_bypass)
-                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
+                    if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
+                        if(!transform_bypass)
+                            h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
+                        else{
+                            static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
+                                                                    8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
+                            for(i = 0; i < 16; i++)
+                                dctcoef_set(h->mb, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc, pixel_shift, i));
+                        }
+                    }
                 }else
-                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
+                    ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
             }
             if(h->deblocking_filter)
-                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
+                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple, pixel_shift);
         }else if(is_h264){
-            hl_motion(h, dest_y, dest_cb, dest_cr,
-                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
+            if (pixel_shift) {
+                hl_motion_16(h, dest_y, dest_cb, dest_cr,
+                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                             h->h264dsp.weight_h264_pixels_tab,
+                             h->h264dsp.biweight_h264_pixels_tab);
+            } else
+                hl_motion_8(h, dest_y, dest_cb, dest_cr,
+                            s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                            s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                            h->h264dsp.weight_h264_pixels_tab,
+                            h->h264dsp.biweight_h264_pixels_tab);
         }
 
 
@@ -2006,12 +1328,12 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
                         }else{
                             for(i=0; i<16; i++){
-                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
-                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
+                                if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
                             }
                         }
                     }else{
-                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
+                         h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
                     }
                 }else if(h->cbp&15){
                     if(transform_bypass){
@@ -2019,14 +1341,14 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                         idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
                         for(i=0; i<16; i+=di){
                             if(h->non_zero_count_cache[ scan8[i] ]){
-                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
+                                idct_add(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
                             }
                         }
                     }else{
                         if(IS_8x8DCT(mb_type)){
-                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
+                            h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
                         }else{
-                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
+                            h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
                         }
                     }
                 }
@@ -2034,7 +1356,7 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                 for(i=0; i<16; i++){
                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
                         uint8_t * const ptr= dest_y + block_offset[i];
-                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
+                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
                     }
                 }
             }
@@ -2044,32 +1366,31 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
             uint8_t *dest[2] = {dest_cb, dest_cr};
             if(transform_bypass){
                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
-                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
-                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
+                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16 << pixel_shift), uvlinesize);
+                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + (20*16 << pixel_shift), uvlinesize);
                 }else{
                     idct_add = s->dsp.add_pixels4;
                     for(i=16; i<16+8; i++){
-                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
-                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
+                        if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
                     }
                 }
             }else{
-                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
-                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
                 if(is_h264){
-                    idct_add = s->dsp.h264_idct_add;
-                    idct_dc_add = s->dsp.h264_idct_dc_add;
-                    for(i=16; i<16+8; i++){
-                        if(h->non_zero_count_cache[ scan8[i] ])
-                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
-                        else if(h->mb[i*16])
-                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
-                    }
+                    if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16 << pixel_shift)       , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
+                    if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + ((16*16+4*16) << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
+                    h->h264dsp.h264_idct_add8(dest, block_offset,
+                                              h->mb, uvlinesize,
+                                              h->non_zero_count_cache);
                 }else{
+                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
+                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
                     for(i=16; i<16+8; i++){
                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
-                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
+                            ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
                         }
                     }
                 }
@@ -2078,32 +1399,23 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
     }
     if(h->cbp || IS_INTRA(mb_type))
         s->dsp.clear_blocks(h->mb);
-
-    if(h->deblocking_filter) {
-        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
-        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
-        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
-        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
-        if (!simple && FRAME_MBAFF) {
-            ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
-        } else {
-            ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
-        }
-    }
 }
 
 /**
  * Process a macroblock; this case avoids checks for expensive uncommon cases.
  */
-static void hl_decode_mb_simple(H264Context *h){
-    hl_decode_mb_internal(h, 1);
+#define hl_decode_mb_simple(sh, bits) \
+static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
+    hl_decode_mb_internal(h, 1, sh); \
 }
+hl_decode_mb_simple(0, 8);
+hl_decode_mb_simple(1, 16);
 
 /**
  * Process a macroblock; this handles edge cases, such as interlacing.
  */
 static void av_noinline hl_decode_mb_complex(H264Context *h){
-    hl_decode_mb_internal(h, 0);
+    hl_decode_mb_internal(h, 0, h->pixel_shift);
 }
 
 void ff_h264_hl_decode_mb(H264Context *h){
@@ -2112,9 +1424,12 @@ void ff_h264_hl_decode_mb(H264Context *h){
     const int mb_type= s->current_picture.mb_type[mb_xy];
     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
 
-    if (is_complex)
+    if (is_complex) {
         hl_decode_mb_complex(h);
-    else hl_decode_mb_simple(h);
+    } else if (h->pixel_shift) {
+        hl_decode_mb_simple_16(h);
+    } else
+        hl_decode_mb_simple_8(h);
 }
 
 static int pred_weight_table(H264Context *h){
@@ -2125,7 +1440,8 @@ static int pred_weight_table(H264Context *h){
     h->use_weight= 0;
     h->use_weight_chroma= 0;
     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
-    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
+    if(CHROMA)
+        h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
     luma_def = 1<<h->luma_log2_weight_denom;
     chroma_def = 1<<h->chroma_log2_weight_denom;
 
@@ -2137,16 +1453,16 @@ static int pred_weight_table(H264Context *h){
 
             luma_weight_flag= get_bits1(&s->gb);
             if(luma_weight_flag){
-                h->luma_weight[list][i]= get_se_golomb(&s->gb);
-                h->luma_offset[list][i]= get_se_golomb(&s->gb);
-                if(   h->luma_weight[list][i] != luma_def
-                   || h->luma_offset[list][i] != 0) {
+                h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
+                h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
+                if(   h->luma_weight[i][list][0] != luma_def
+                   || h->luma_weight[i][list][1] != 0) {
                     h->use_weight= 1;
                     h->luma_weight_flag[list]= 1;
                 }
             }else{
-                h->luma_weight[list][i]= luma_def;
-                h->luma_offset[list][i]= 0;
+                h->luma_weight[i][list][0]= luma_def;
+                h->luma_weight[i][list][1]= 0;
             }
 
             if(CHROMA){
@@ -2154,10 +1470,10 @@ static int pred_weight_table(H264Context *h){
                 if(chroma_weight_flag){
                     int j;
                     for(j=0; j<2; j++){
-                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
-                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
-                        if(   h->chroma_weight[list][i][j] != chroma_def
-                           || h->chroma_offset[list][i][j] != 0) {
+                        h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
+                        h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
+                        if(   h->chroma_weight[i][list][j][0] != chroma_def
+                           || h->chroma_weight[i][list][j][1] != 0) {
                             h->use_weight_chroma= 1;
                             h->chroma_weight_flag[list]= 1;
                         }
@@ -2165,55 +1481,74 @@ static int pred_weight_table(H264Context *h){
                 }else{
                     int j;
                     for(j=0; j<2; j++){
-                        h->chroma_weight[list][i][j]= chroma_def;
-                        h->chroma_offset[list][i][j]= 0;
+                        h->chroma_weight[i][list][j][0]= chroma_def;
+                        h->chroma_weight[i][list][j][1]= 0;
                     }
                 }
             }
         }
-        if(h->slice_type_nos != FF_B_TYPE) break;
+        if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
     }
     h->use_weight= h->use_weight || h->use_weight_chroma;
     return 0;
 }
 
-static void implicit_weight_table(H264Context *h){
+/**
+ * Initialize implicit_weight table.
+ * @param field  0/1 initialize the weight for interlaced MBAFF
+ *                -1 initializes the rest
+ */
+static void implicit_weight_table(H264Context *h, int field){
     MpegEncContext * const s = &h->s;
-    int ref0, ref1, i;
-    int cur_poc = s->current_picture_ptr->poc;
+    int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
 
     for (i = 0; i < 2; i++) {
         h->luma_weight_flag[i]   = 0;
         h->chroma_weight_flag[i] = 0;
     }
 
-    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
+    if(field < 0){
+        cur_poc = s->current_picture_ptr->poc;
+    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
         h->use_weight= 0;
         h->use_weight_chroma= 0;
         return;
     }
+        ref_start= 0;
+        ref_count0= h->ref_count[0];
+        ref_count1= h->ref_count[1];
+    }else{
+        cur_poc = s->current_picture_ptr->field_poc[field];
+        ref_start= 16;
+        ref_count0= 16+2*h->ref_count[0];
+        ref_count1= 16+2*h->ref_count[1];
+    }
 
     h->use_weight= 2;
     h->use_weight_chroma= 2;
     h->luma_log2_weight_denom= 5;
     h->chroma_log2_weight_denom= 5;
 
-    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
+    for(ref0=ref_start; ref0 < ref_count0; ref0++){
         int poc0 = h->ref_list[0][ref0].poc;
-        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
+        for(ref1=ref_start; ref1 < ref_count1; ref1++){
             int poc1 = h->ref_list[1][ref1].poc;
             int td = av_clip(poc1 - poc0, -128, 127);
+            int w= 32;
             if(td){
                 int tb = av_clip(cur_poc - poc0, -128, 127);
                 int tx = (16384 + (FFABS(td) >> 1)) / td;
-                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
-                if(dist_scale_factor < -64 || dist_scale_factor > 128)
-                    h->implicit_weight[ref0][ref1] = 32;
-                else
-                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
-            }else
-                h->implicit_weight[ref0][ref1] = 32;
+                int dist_scale_factor = (tb*tx + 32) >> 8;
+                if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
+                    w = 64 - dist_scale_factor;
+            }
+            if(field<0){
+                h->implicit_weight[ref0][ref1][0]=
+                h->implicit_weight[ref0][ref1][1]= w;
+            }else{
+                h->implicit_weight[ref0][ref1][field]=w;
+            }
         }
     }
 }
@@ -2330,33 +1665,20 @@ static int init_poc(H264Context *h){
  * initialize scan tables
  */
 static void init_scan_tables(H264Context *h){
-    MpegEncContext * const s = &h->s;
     int i;
-    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
-        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
-        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
-    }else{
-        for(i=0; i<16; i++){
+    for(i=0; i<16; i++){
 #define T(x) (x>>2) | ((x<<2) & 0xF)
-            h->zigzag_scan[i] = T(zigzag_scan[i]);
-            h-> field_scan[i] = T( field_scan[i]);
+        h->zigzag_scan[i] = T(zigzag_scan[i]);
+        h-> field_scan[i] = T( field_scan[i]);
 #undef T
-        }
     }
-    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
-        memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
-        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
-        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
-        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
-    }else{
-        for(i=0; i<64; i++){
+    for(i=0; i<64; i++){
 #define T(x) (x>>3) | ((x&7)<<3)
-            h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
-            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
-            h->field_scan8x8[i]        = T(field_scan8x8[i]);
-            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
+        h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
+        h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
+        h->field_scan8x8[i]        = T(field_scan8x8[i]);
+        h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
 #undef T
-        }
     }
     if(h->sps.transform_bypass){ //FIXME same ugly
         h->zigzag_scan_q0          = zigzag_scan;
@@ -2423,7 +1745,7 @@ static void field_end(H264Context *h){
 }
 
 /**
- * Replicates H264 "master" context to thread contexts.
+ * Replicate H264 "master" context to thread contexts.
  */
 static void clone_slice(H264Context *dst, H264Context *src)
 {
@@ -2449,6 +1771,33 @@ static void clone_slice(H264Context *dst, H264Context *src)
     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
 }
 
+/**
+ * computes profile from profile_idc and constraint_set?_flags
+ *
+ * @param sps SPS
+ *
+ * @return profile as defined by FF_PROFILE_H264_*
+ */
+int ff_h264_get_profile(SPS *sps)
+{
+    int profile = sps->profile_idc;
+
+    switch(sps->profile_idc) {
+    case FF_PROFILE_H264_BASELINE:
+        // constraint_set1_flag set to 1
+        profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
+        break;
+    case FF_PROFILE_H264_HIGH_10:
+    case FF_PROFILE_H264_HIGH_422:
+    case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
+        // constraint_set3_flag set to 1
+        profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
+        break;
+    }
+
+    return profile;
+}
+
 /**
  * decodes a slice header.
  * This will also call MPV_common_init() and frame_start() as needed.
@@ -2502,7 +1851,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         h->slice_type_fixed=0;
 
     slice_type= golomb_to_pict_type[ slice_type ];
-    if (slice_type == FF_I_TYPE
+    if (slice_type == AV_PICTURE_TYPE_I
         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
         default_ref_list_done = 1;
     }
@@ -2510,11 +1859,6 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->slice_type_nos= slice_type & 3;
 
     s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
-    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
-        av_log(h->s.avctx, AV_LOG_ERROR,
-               "B picture before any references, skipping\n");
-        return -1;
-    }
 
     pps_id= get_ue_golomb(&s->gb);
     if(pps_id>=MAX_PPS_COUNT){
@@ -2533,6 +1877,10 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     }
     h->sps = *h0->sps_buffers[h->pps.sps_id];
 
+    s->avctx->profile = ff_h264_get_profile(&h->sps);
+    s->avctx->level   = h->sps.level_idc;
+    s->avctx->refs    = h->sps.ref_frame_count;
+
     if(h == h0 && h->dequant_coeff_pps != pps_id){
         h->dequant_coeff_pps = pps_id;
         init_dequant_tables(h);
@@ -2542,19 +1890,19 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
 
     h->b_stride=  s->mb_width*4;
-    h->b8_stride= s->mb_width*2;
 
     s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
     if(h->sps.frame_mbs_only_flag)
         s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
     else
-        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
+        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
 
     if (s->context_initialized
-        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
+        && (   s->width != s->avctx->width || s->height != s->avctx->height
+            || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
         if(h != h0)
             return -1;   // width / height changed during parallelized decoding
-        free_tables(h);
+        free_tables(h, 0);
         flush_dpb(s->avctx);
         MPV_common_end(s);
     }
@@ -2564,8 +1912,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
         avcodec_set_dimensions(s->avctx, s->width, s->height);
         s->avctx->sample_aspect_ratio= h->sps.sar;
-        if(!s->avctx->sample_aspect_ratio.den)
-            s->avctx->sample_aspect_ratio.den = 1;
+        av_assert0(s->avctx->sample_aspect_ratio.den);
 
         if(h->sps.video_signal_type_present_flag){
             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
@@ -2577,13 +1924,29 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         }
 
         if(h->sps.timing_info_present_flag){
-            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
-            if(h->x264_build > 0 && h->x264_build < 44)
-                s->avctx->time_base.den *= 2;
+            int64_t den= h->sps.time_scale;
+            if(h->x264_build < 44U)
+                den *= 2;
             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
-                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
+                      h->sps.num_units_in_tick, den, 1<<30);
+        }
+
+        switch (h->sps.bit_depth_luma) {
+            case 9 :
+                s->avctx->pix_fmt = PIX_FMT_YUV420P9;
+                break;
+            case 10 :
+                s->avctx->pix_fmt = PIX_FMT_YUV420P10;
+                break;
+            default:
+        s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
+                                                 s->avctx->codec->pix_fmts ?
+                                                 s->avctx->codec->pix_fmts :
+                                                 s->avctx->color_range == AVCOL_RANGE_JPEG ?
+                                                 hwaccel_pixfmt_list_h264_jpeg_420 :
+                                                 ff_hwaccel_pixfmt_list_420);
         }
-        s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
+
         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
 
         if (MPV_common_init(s) < 0)
@@ -2599,10 +1962,12 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
+            c->h264dsp = h->h264dsp;
             c->sps = h->sps;
             c->pps = h->pps;
+            c->pixel_shift = h->pixel_shift;
             init_scan_tables(c);
-            clone_tables(c, h);
+            clone_tables(c, h, i);
         }
 
         for(i = 0; i < s->avctx->thread_count; i++)
@@ -2630,13 +1995,30 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     if(h0->current_slice == 0){
         while(h->frame_num !=  h->prev_frame_num &&
               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
-            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
+            Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
+            av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
             if (ff_h264_frame_start(h) < 0)
                 return -1;
             h->prev_frame_num++;
             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
             s->current_picture_ptr->frame_num= h->prev_frame_num;
-            ff_h264_execute_ref_pic_marking(h, NULL, 0);
+            ff_generate_sliding_window_mmcos(h);
+            ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+            /* Error concealment: if a ref is missing, copy the previous ref in its place.
+             * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
+             * about there being no actual duplicates.
+             * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
+             * concealing a lost frame, this probably isn't noticable by comparison, but it should
+             * be fixed. */
+            if (h->short_ref_count) {
+                if (prev) {
+                    av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
+                                  (const uint8_t**)prev->data, prev->linesize,
+                                  s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
+                    h->short_ref[0]->poc = prev->poc+2;
+                }
+                h->short_ref[0]->frame_num = h->prev_frame_num;
+            }
         }
 
         /* See if we have a decoded first field looking for a pair... */
@@ -2738,15 +2120,15 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->ref_count[0]= h->pps.ref_count[0];
     h->ref_count[1]= h->pps.ref_count[1];
 
-    if(h->slice_type_nos != FF_I_TYPE){
-        if(h->slice_type_nos == FF_B_TYPE){
+    if(h->slice_type_nos != AV_PICTURE_TYPE_I){
+        if(h->slice_type_nos == AV_PICTURE_TYPE_B){
             h->direct_spatial_mv_pred= get_bits1(&s->gb);
         }
         num_ref_idx_active_override_flag= get_bits1(&s->gb);
 
         if(num_ref_idx_active_override_flag){
             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
-            if(h->slice_type_nos==FF_B_TYPE)
+            if(h->slice_type_nos==AV_PICTURE_TYPE_B)
                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
 
             if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
@@ -2755,7 +2137,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                 return -1;
             }
         }
-        if(h->slice_type_nos == FF_B_TYPE)
+        if(h->slice_type_nos == AV_PICTURE_TYPE_B)
             h->list_count= 2;
         else
             h->list_count= 1;
@@ -2766,24 +2148,24 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         ff_h264_fill_default_ref_list(h);
     }
 
-    if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
+    if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
         return -1;
 
-    if(h->slice_type_nos!=FF_I_TYPE){
+    if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
         s->last_picture_ptr= &h->ref_list[0][0];
         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
     }
-    if(h->slice_type_nos==FF_B_TYPE){
+    if(h->slice_type_nos==AV_PICTURE_TYPE_B){
         s->next_picture_ptr= &h->ref_list[1][0];
         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
     }
 
-    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
-       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
+    if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
+       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
         pred_weight_table(h);
-    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
-        implicit_weight_table(h);
-    else {
+    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
+        implicit_weight_table(h, -1);
+    }else {
         h->use_weight = 0;
         for (i = 0; i < 2; i++) {
             h->luma_weight_flag[i]   = 0;
@@ -2794,14 +2176,20 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     if(h->nal_ref_idc)
         ff_h264_decode_ref_pic_marking(h0, &s->gb);
 
-    if(FRAME_MBAFF)
+    if(FRAME_MBAFF){
         ff_h264_fill_mbaff_ref_list(h);
 
-    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
+        if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
+            implicit_weight_table(h, 0);
+            implicit_weight_table(h, 1);
+        }
+    }
+
+    if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
         ff_h264_direct_dist_scale_factor(h);
     ff_h264_direct_ref_list_init(h);
 
-    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
+    if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
         tmp = get_ue_golomb_31(&s->gb);
         if(tmp > 2){
             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
@@ -2812,7 +2200,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
     h->last_qscale_diff = 0;
     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
-    if(tmp>51){
+    if(tmp>51+6*(h->sps.bit_depth_luma-8)){
         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
         return -1;
     }
@@ -2820,16 +2208,16 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
     //FIXME qscale / qp ... stuff
-    if(h->slice_type == FF_SP_TYPE){
+    if(h->slice_type == AV_PICTURE_TYPE_SP){
         get_bits1(&s->gb); /* sp_for_switch_flag */
     }
-    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
+    if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
         get_se_golomb(&s->gb); /* slice_qs_delta */
     }
 
     h->deblocking_filter = 1;
-    h->slice_alpha_c0_offset = 0;
-    h->slice_beta_offset = 0;
+    h->slice_alpha_c0_offset = 52;
+    h->slice_beta_offset = 52;
     if( h->pps.deblocking_filter_parameters_present ) {
         tmp= get_ue_golomb_31(&s->gb);
         if(tmp > 2){
@@ -2841,14 +2229,19 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             h->deblocking_filter^= 1; // 1<->0
 
         if( h->deblocking_filter ) {
-            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
-            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
+            h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
+            h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
+            if(   h->slice_alpha_c0_offset > 104U
+               || h->slice_beta_offset     > 104U){
+                av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
+                return -1;
+            }
         }
     }
 
     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
-       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
-       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
+       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
+       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
         h->deblocking_filter= 0;
 
@@ -2867,6 +2260,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                 return 1; // deblocking switched inside frame
         }
     }
+    h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
 
 #if 0 //FMO
     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
@@ -2880,1909 +2274,365 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     }
 
     for(j=0; j<2; j++){
+        int id_list[16];
         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
+        for(i=0; i<16; i++){
+            id_list[i]= 60;
+            if(h->ref_list[j][i].data[0]){
+                int k;
+                uint8_t *base= h->ref_list[j][i].base[0];
+                for(k=0; k<h->short_ref_count; k++)
+                    if(h->short_ref[k]->base[0] == base){
+                        id_list[i]= k;
+                        break;
+                    }
+                for(k=0; k<h->long_ref_count; k++)
+                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
+                        id_list[i]= h->short_ref_count + k;
+                        break;
+                    }
+            }
+        }
+
         ref2frm[0]=
         ref2frm[1]= -1;
         for(i=0; i<16; i++)
-            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
+            ref2frm[i+2]= 4*id_list[i]
                           +(h->ref_list[j][i].reference&3);
         ref2frm[18+0]=
         ref2frm[18+1]= -1;
         for(i=16; i<48; i++)
-            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
+            ref2frm[i+4]= 4*id_list[(i-16)>>1]
                           +(h->ref_list[j][i].reference&3);
     }
 
     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
 
-    s->avctx->refs= h->sps.ref_frame_count;
-
     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
                h->slice_num,
-               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
-               first_mb_in_slice,
-               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
-               pps_id, h->frame_num,
-               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
-               h->ref_count[0], h->ref_count[1],
-               s->qscale,
-               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
-               h->use_weight,
-               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
-               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
-               );
-    }
-
-    return 0;
-}
-
-int ff_h264_get_slice_type(H264Context *h)
-{
-    switch (h->slice_type) {
-    case FF_P_TYPE:  return 0;
-    case FF_B_TYPE:  return 1;
-    case FF_I_TYPE:  return 2;
-    case FF_SP_TYPE: return 3;
-    case FF_SI_TYPE: return 4;
-    default:         return -1;
-    }
-}
-
-/**
- *
- */
-static inline int get_level_prefix(GetBitContext *gb){
-    unsigned int buf;
-    int log;
-
-    OPEN_READER(re, gb);
-    UPDATE_CACHE(re, gb);
-    buf=GET_CACHE(re, gb);
-
-    log= 32 - av_log2(buf);
-#ifdef TRACE
-    print_bin(buf>>(32-log), log);
-    av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
-#endif
-
-    LAST_SKIP_BITS(re, gb, log);
-    CLOSE_READER(re, gb);
-
-    return log-1;
-}
-
-static inline int get_dct8x8_allowed(H264Context *h){
-    if(h->sps.direct_8x8_inference_flag)
-        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
-    else
-        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
-}
-
-/**
- * decodes a residual block.
- * @param n block index
- * @param scantable scantable
- * @param max_coeff number of coefficients in the block
- * @return <0 if an error occurred
- */
-static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
-    MpegEncContext * const s = &h->s;
-    static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
-    int level[16];
-    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
-
-    //FIXME put trailing_onex into the context
-
-    if(n == CHROMA_DC_BLOCK_INDEX){
-        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
-        total_coeff= coeff_token>>2;
-    }else{
-        if(n == LUMA_DC_BLOCK_INDEX){
-            total_coeff= pred_non_zero_count(h, 0);
-            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
-            total_coeff= coeff_token>>2;
-        }else{
-            total_coeff= pred_non_zero_count(h, n);
-            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
-            total_coeff= coeff_token>>2;
-            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
-        }
-    }
-
-    //FIXME set last_non_zero?
-
-    if(total_coeff==0)
-        return 0;
-    if(total_coeff > (unsigned)max_coeff) {
-        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
-        return -1;
-    }
-
-    trailing_ones= coeff_token&3;
-    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
-    assert(total_coeff<=16);
-
-    i = show_bits(gb, 3);
-    skip_bits(gb, trailing_ones);
-    level[0] = 1-((i&4)>>1);
-    level[1] = 1-((i&2)   );
-    level[2] = 1-((i&1)<<1);
-
-    if(trailing_ones<total_coeff) {
-        int mask, prefix;
-        int suffix_length = total_coeff > 10 && trailing_ones < 3;
-        int bitsi= show_bits(gb, LEVEL_TAB_BITS);
-        int level_code= cavlc_level_tab[suffix_length][bitsi][0];
-
-        skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
-        if(level_code >= 100){
-            prefix= level_code - 100;
-            if(prefix == LEVEL_TAB_BITS)
-                prefix += get_level_prefix(gb);
-
-            //first coefficient has suffix_length equal to 0 or 1
-            if(prefix<14){ //FIXME try to build a large unified VLC table for all this
-                if(suffix_length)
-                    level_code= (prefix<<1) + get_bits1(gb); //part
-                else
-                    level_code= prefix; //part
-            }else if(prefix==14){
-                if(suffix_length)
-                    level_code= (prefix<<1) + get_bits1(gb); //part
-                else
-                    level_code= prefix + get_bits(gb, 4); //part
-            }else{
-                level_code= 30 + get_bits(gb, prefix-3); //part
-                if(prefix>=16)
-                    level_code += (1<<(prefix-3))-4096;
-            }
-
-            if(trailing_ones < 3) level_code += 2;
-
-            suffix_length = 2;
-            mask= -(level_code&1);
-            level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
-        }else{
-            if(trailing_ones < 3) level_code += (level_code>>31)|1;
-
-            suffix_length = 1;
-            if(level_code + 3U > 6U)
-                suffix_length++;
-            level[trailing_ones]= level_code;
-        }
-
-        //remaining coefficients have suffix_length > 0
-        for(i=trailing_ones+1;i<total_coeff;i++) {
-            static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
-            int bitsi= show_bits(gb, LEVEL_TAB_BITS);
-            level_code= cavlc_level_tab[suffix_length][bitsi][0];
-
-            skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
-            if(level_code >= 100){
-                prefix= level_code - 100;
-                if(prefix == LEVEL_TAB_BITS){
-                    prefix += get_level_prefix(gb);
-                }
-                if(prefix<15){
-                    level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
-                }else{
-                    level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
-                    if(prefix>=16)
-                        level_code += (1<<(prefix-3))-4096;
-                }
-                mask= -(level_code&1);
-                level_code= (((2+level_code)>>1) ^ mask) - mask;
-            }
-            level[i]= level_code;
-
-            if(suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length])
-                suffix_length++;
-        }
-    }
-
-    if(total_coeff == max_coeff)
-        zeros_left=0;
-    else{
-        if(n == CHROMA_DC_BLOCK_INDEX)
-            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
-        else
-            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
-    }
-
-    coeff_num = zeros_left + total_coeff - 1;
-    j = scantable[coeff_num];
-    if(n > 24){
-        block[j] = level[0];
-        for(i=1;i<total_coeff;i++) {
-            if(zeros_left <= 0)
-                run_before = 0;
-            else if(zeros_left < 7){
-                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
-            }else{
-                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
-            }
-            zeros_left -= run_before;
-            coeff_num -= 1 + run_before;
-            j= scantable[ coeff_num ];
-
-            block[j]= level[i];
-        }
-    }else{
-        block[j] = (level[0] * qmul[j] + 32)>>6;
-        for(i=1;i<total_coeff;i++) {
-            if(zeros_left <= 0)
-                run_before = 0;
-            else if(zeros_left < 7){
-                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
-            }else{
-                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
-            }
-            zeros_left -= run_before;
-            coeff_num -= 1 + run_before;
-            j= scantable[ coeff_num ];
-
-            block[j]= (level[i] * qmul[j] + 32)>>6;
-        }
-    }
-
-    if(zeros_left<0){
-        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
-        return -1;
-    }
-
-    return 0;
-}
-
-static void predict_field_decoding_flag(H264Context *h){
-    MpegEncContext * const s = &h->s;
-    const int mb_xy= h->mb_xy;
-    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
-                ? s->current_picture.mb_type[mb_xy-1]
-                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
-                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
-                : 0;
-    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
-}
-
-/**
- * decodes a P_SKIP or B_SKIP macroblock
- */
-static void decode_mb_skip(H264Context *h){
-    MpegEncContext * const s = &h->s;
-    const int mb_xy= h->mb_xy;
-    int mb_type=0;
-
-    memset(h->non_zero_count[mb_xy], 0, 16);
-    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
-
-    if(MB_FIELD)
-        mb_type|= MB_TYPE_INTERLACED;
-
-    if( h->slice_type_nos == FF_B_TYPE )
-    {
-        // just for fill_caches. pred_direct_motion will set the real mb_type
-        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
-
-        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
-        ff_h264_pred_direct_motion(h, &mb_type);
-        mb_type|= MB_TYPE_SKIP;
-    }
-    else
-    {
-        int mx, my;
-        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
-
-        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
-        pred_pskip_motion(h, &mx, &my);
-        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
-        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
-    }
-
-    write_back_motion(h, mb_type);
-    s->current_picture.mb_type[mb_xy]= mb_type;
-    s->current_picture.qscale_table[mb_xy]= s->qscale;
-    h->slice_table[ mb_xy ]= h->slice_num;
-    h->prev_mb_skipped= 1;
-}
-
-/**
- * decodes a macroblock
- * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
- */
-static int decode_mb_cavlc(H264Context *h){
-    MpegEncContext * const s = &h->s;
-    int mb_xy;
-    int partition_count;
-    unsigned int mb_type, cbp;
-    int dct8x8_allowed= h->pps.transform_8x8_mode;
-
-    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
-
-    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
-    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
-                down the code */
-    if(h->slice_type_nos != FF_I_TYPE){
-        if(s->mb_skip_run==-1)
-            s->mb_skip_run= get_ue_golomb(&s->gb);
-
-        if (s->mb_skip_run--) {
-            if(FRAME_MBAFF && (s->mb_y&1) == 0){
-                if(s->mb_skip_run==0)
-                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
-                else
-                    predict_field_decoding_flag(h);
-            }
-            decode_mb_skip(h);
-            return 0;
-        }
-    }
-    if(FRAME_MBAFF){
-        if( (s->mb_y&1) == 0 )
-            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
-    }
-
-    h->prev_mb_skipped= 0;
-
-    mb_type= get_ue_golomb(&s->gb);
-    if(h->slice_type_nos == FF_B_TYPE){
-        if(mb_type < 23){
-            partition_count= b_mb_type_info[mb_type].partition_count;
-            mb_type=         b_mb_type_info[mb_type].type;
-        }else{
-            mb_type -= 23;
-            goto decode_intra_mb;
-        }
-    }else if(h->slice_type_nos == FF_P_TYPE){
-        if(mb_type < 5){
-            partition_count= p_mb_type_info[mb_type].partition_count;
-            mb_type=         p_mb_type_info[mb_type].type;
-        }else{
-            mb_type -= 5;
-            goto decode_intra_mb;
-        }
-    }else{
-       assert(h->slice_type_nos == FF_I_TYPE);
-        if(h->slice_type == FF_SI_TYPE && mb_type)
-            mb_type--;
-decode_intra_mb:
-        if(mb_type > 25){
-            av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
-            return -1;
-        }
-        partition_count=0;
-        cbp= i_mb_type_info[mb_type].cbp;
-        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
-        mb_type= i_mb_type_info[mb_type].type;
-    }
-
-    if(MB_FIELD)
-        mb_type |= MB_TYPE_INTERLACED;
-
-    h->slice_table[ mb_xy ]= h->slice_num;
-
-    if(IS_INTRA_PCM(mb_type)){
-        unsigned int x;
-
-        // We assume these blocks are very rare so we do not optimize it.
-        align_get_bits(&s->gb);
-
-        // The pixels are stored in the same order as levels in h->mb array.
-        for(x=0; x < (CHROMA ? 384 : 256); x++){
-            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
-        }
-
-        // In deblocking, the quantizer is 0
-        s->current_picture.qscale_table[mb_xy]= 0;
-        // All coeffs are present
-        memset(h->non_zero_count[mb_xy], 16, 16);
-
-        s->current_picture.mb_type[mb_xy]= mb_type;
-        return 0;
-    }
-
-    if(MB_MBAFF){
-        h->ref_count[0] <<= 1;
-        h->ref_count[1] <<= 1;
-    }
-
-    fill_caches(h, mb_type, 0);
-
-    //mb_pred
-    if(IS_INTRA(mb_type)){
-        int pred_mode;
-//            init_top_left_availability(h);
-        if(IS_INTRA4x4(mb_type)){
-            int i;
-            int di = 1;
-            if(dct8x8_allowed && get_bits1(&s->gb)){
-                mb_type |= MB_TYPE_8x8DCT;
-                di = 4;
-            }
-
-//                fill_intra4x4_pred_table(h);
-            for(i=0; i<16; i+=di){
-                int mode= pred_intra_mode(h, i);
-
-                if(!get_bits1(&s->gb)){
-                    const int rem_mode= get_bits(&s->gb, 3);
-                    mode = rem_mode + (rem_mode >= mode);
-                }
-
-                if(di==4)
-                    fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
-                else
-                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
-            }
-            ff_h264_write_back_intra_pred_mode(h);
-            if( ff_h264_check_intra4x4_pred_mode(h) < 0)
-                return -1;
-        }else{
-            h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode(h, h->intra16x16_pred_mode);
-            if(h->intra16x16_pred_mode < 0)
-                return -1;
-        }
-        if(CHROMA){
-            pred_mode= ff_h264_check_intra_pred_mode(h, get_ue_golomb_31(&s->gb));
-            if(pred_mode < 0)
-                return -1;
-            h->chroma_pred_mode= pred_mode;
-        }
-    }else if(partition_count==4){
-        int i, j, sub_partition_count[4], list, ref[2][4];
-
-        if(h->slice_type_nos == FF_B_TYPE){
-            for(i=0; i<4; i++){
-                h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
-                if(h->sub_mb_type[i] >=13){
-                    av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
-                    return -1;
-                }
-                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
-                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
-            }
-            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
-               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
-                ff_h264_pred_direct_motion(h, &mb_type);
-                h->ref_cache[0][scan8[4]] =
-                h->ref_cache[1][scan8[4]] =
-                h->ref_cache[0][scan8[12]] =
-                h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
-            }
-        }else{
-            assert(h->slice_type_nos == FF_P_TYPE); //FIXME SP correct ?
-            for(i=0; i<4; i++){
-                h->sub_mb_type[i]= get_ue_golomb_31(&s->gb);
-                if(h->sub_mb_type[i] >=4){
-                    av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
-                    return -1;
-                }
-                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
-                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
-            }
-        }
-
-        for(list=0; list<h->list_count; list++){
-            int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
-            for(i=0; i<4; i++){
-                if(IS_DIRECT(h->sub_mb_type[i])) continue;
-                if(IS_DIR(h->sub_mb_type[i], 0, list)){
-                    unsigned int tmp;
-                    if(ref_count == 1){
-                        tmp= 0;
-                    }else if(ref_count == 2){
-                        tmp= get_bits1(&s->gb)^1;
-                    }else{
-                        tmp= get_ue_golomb_31(&s->gb);
-                        if(tmp>=ref_count){
-                            av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", tmp);
-                            return -1;
-                        }
-                    }
-                    ref[list][i]= tmp;
-                }else{
-                 //FIXME
-                    ref[list][i] = -1;
-                }
-            }
-        }
-
-        if(dct8x8_allowed)
-            dct8x8_allowed = get_dct8x8_allowed(h);
-
-        for(list=0; list<h->list_count; list++){
-            for(i=0; i<4; i++){
-                if(IS_DIRECT(h->sub_mb_type[i])) {
-                    h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];
-                    continue;
-                }
-                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
-                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
-
-                if(IS_DIR(h->sub_mb_type[i], 0, list)){
-                    const int sub_mb_type= h->sub_mb_type[i];
-                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
-                    for(j=0; j<sub_partition_count[i]; j++){
-                        int mx, my;
-                        const int index= 4*i + block_width*j;
-                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
-                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
-                        mx += get_se_golomb(&s->gb);
-                        my += get_se_golomb(&s->gb);
-                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-
-                        if(IS_SUB_8X8(sub_mb_type)){
-                            mv_cache[ 1 ][0]=
-                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
-                            mv_cache[ 1 ][1]=
-                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
-                        }else if(IS_SUB_8X4(sub_mb_type)){
-                            mv_cache[ 1 ][0]= mx;
-                            mv_cache[ 1 ][1]= my;
-                        }else if(IS_SUB_4X8(sub_mb_type)){
-                            mv_cache[ 8 ][0]= mx;
-                            mv_cache[ 8 ][1]= my;
-                        }
-                        mv_cache[ 0 ][0]= mx;
-                        mv_cache[ 0 ][1]= my;
-                    }
-                }else{
-                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
-                    p[0] = p[1]=
-                    p[8] = p[9]= 0;
-                }
-            }
-        }
-    }else if(IS_DIRECT(mb_type)){
-        ff_h264_pred_direct_motion(h, &mb_type);
-        dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
-    }else{
-        int list, mx, my, i;
-         //FIXME we should set ref_idx_l? to 0 if we use that later ...
-        if(IS_16X16(mb_type)){
-            for(list=0; list<h->list_count; list++){
-                    unsigned int val;
-                    if(IS_DIR(mb_type, 0, list)){
-                        if(h->ref_count[list]==1){
-                            val= 0;
-                        }else if(h->ref_count[list]==2){
-                            val= get_bits1(&s->gb)^1;
-                        }else{
-                            val= get_ue_golomb_31(&s->gb);
-                            if(val >= h->ref_count[list]){
-                                av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
-                                return -1;
-                            }
-                        }
-                    }else
-                        val= LIST_NOT_USED&0xFF;
-                    fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
-            }
-            for(list=0; list<h->list_count; list++){
-                unsigned int val;
-                if(IS_DIR(mb_type, 0, list)){
-                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
-                    mx += get_se_golomb(&s->gb);
-                    my += get_se_golomb(&s->gb);
-                    tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-
-                    val= pack16to32(mx,my);
-                }else
-                    val=0;
-                fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, val, 4);
-            }
-        }
-        else if(IS_16X8(mb_type)){
-            for(list=0; list<h->list_count; list++){
-                    for(i=0; i<2; i++){
-                        unsigned int val;
-                        if(IS_DIR(mb_type, i, list)){
-                            if(h->ref_count[list] == 1){
-                                val= 0;
-                            }else if(h->ref_count[list] == 2){
-                                val= get_bits1(&s->gb)^1;
-                            }else{
-                                val= get_ue_golomb_31(&s->gb);
-                                if(val >= h->ref_count[list]){
-                                    av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
-                                    return -1;
-                                }
-                            }
-                        }else
-                            val= LIST_NOT_USED&0xFF;
-                        fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
-                    }
-            }
-            for(list=0; list<h->list_count; list++){
-                for(i=0; i<2; i++){
-                    unsigned int val;
-                    if(IS_DIR(mb_type, i, list)){
-                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
-                        mx += get_se_golomb(&s->gb);
-                        my += get_se_golomb(&s->gb);
-                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-
-                        val= pack16to32(mx,my);
-                    }else
-                        val=0;
-                    fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 4);
-                }
-            }
-        }else{
-            assert(IS_8X16(mb_type));
-            for(list=0; list<h->list_count; list++){
-                    for(i=0; i<2; i++){
-                        unsigned int val;
-                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
-                            if(h->ref_count[list]==1){
-                                val= 0;
-                            }else if(h->ref_count[list]==2){
-                                val= get_bits1(&s->gb)^1;
-                            }else{
-                                val= get_ue_golomb_31(&s->gb);
-                                if(val >= h->ref_count[list]){
-                                    av_log(h->s.avctx, AV_LOG_ERROR, "ref %u overflow\n", val);
-                                    return -1;
-                                }
-                            }
-                        }else
-                            val= LIST_NOT_USED&0xFF;
-                        fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
-                    }
-            }
-            for(list=0; list<h->list_count; list++){
-                for(i=0; i<2; i++){
-                    unsigned int val;
-                    if(IS_DIR(mb_type, i, list)){
-                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
-                        mx += get_se_golomb(&s->gb);
-                        my += get_se_golomb(&s->gb);
-                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-
-                        val= pack16to32(mx,my);
-                    }else
-                        val=0;
-                    fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 4);
-                }
-            }
-        }
-    }
-
-    if(IS_INTER(mb_type))
-        write_back_motion(h, mb_type);
-
-    if(!IS_INTRA16x16(mb_type)){
-        cbp= get_ue_golomb(&s->gb);
-        if(cbp > 47){
-            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, s->mb_x, s->mb_y);
-            return -1;
-        }
-
-        if(CHROMA){
-            if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
-            else                     cbp= golomb_to_inter_cbp   [cbp];
-        }else{
-            if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
-            else                     cbp= golomb_to_inter_cbp_gray[cbp];
-        }
-    }
-    h->cbp = cbp;
-
-    if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){
-        if(get_bits1(&s->gb)){
-            mb_type |= MB_TYPE_8x8DCT;
-            h->cbp_table[mb_xy]= cbp;
-        }
-    }
-    s->current_picture.mb_type[mb_xy]= mb_type;
-
-    if(cbp || IS_INTRA16x16(mb_type)){
-        int i8x8, i4x4, chroma_idx;
-        int dquant;
-        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
-        const uint8_t *scan, *scan8x8, *dc_scan;
-
-//        fill_non_zero_count_cache(h);
-
-        if(IS_INTERLACED(mb_type)){
-            scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
-            scan= s->qscale ? h->field_scan : h->field_scan_q0;
-            dc_scan= luma_dc_field_scan;
-        }else{
-            scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
-            scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
-            dc_scan= luma_dc_zigzag_scan;
-        }
-
-        dquant= get_se_golomb(&s->gb);
-
-        if( dquant > 25 || dquant < -26 ){
-            av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
-            return -1;
-        }
-
-        s->qscale += dquant;
-        if(((unsigned)s->qscale) > 51){
-            if(s->qscale<0) s->qscale+= 52;
-            else            s->qscale-= 52;
-        }
-
-        h->chroma_qp[0]= get_chroma_qp(h, 0, s->qscale);
-        h->chroma_qp[1]= get_chroma_qp(h, 1, s->qscale);
-        if(IS_INTRA16x16(mb_type)){
-            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){
-                return -1; //FIXME continue if partitioned and other return -1 too
-            }
-
-            assert((cbp&15) == 0 || (cbp&15) == 15);
-
-            if(cbp&15){
-                for(i8x8=0; i8x8<4; i8x8++){
-                    for(i4x4=0; i4x4<4; i4x4++){
-                        const int index= i4x4 + 4*i8x8;
-                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){
-                            return -1;
-                        }
-                    }
-                }
-            }else{
-                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
-            }
-        }else{
-            for(i8x8=0; i8x8<4; i8x8++){
-                if(cbp & (1<<i8x8)){
-                    if(IS_8x8DCT(mb_type)){
-                        DCTELEM *buf = &h->mb[64*i8x8];
-                        uint8_t *nnz;
-                        for(i4x4=0; i4x4<4; i4x4++){
-                            if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,
-                                                h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 )
-                                return -1;
-                        }
-                        nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
-                        nnz[0] += nnz[1] + nnz[8] + nnz[9];
-                    }else{
-                        for(i4x4=0; i4x4<4; i4x4++){
-                            const int index= i4x4 + 4*i8x8;
-
-                            if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){
-                                return -1;
-                            }
-                        }
-                    }
-                }else{
-                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
-                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
-                }
-            }
-        }
-
-        if(cbp&0x30){
-            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
-                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){
-                    return -1;
-                }
-        }
-
-        if(cbp&0x20){
-            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
-                const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[chroma_idx]];
-                for(i4x4=0; i4x4<4; i4x4++){
-                    const int index= 16 + 4*chroma_idx + i4x4;
-                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, qmul, 15) < 0){
-                        return -1;
-                    }
-                }
-            }
-        }else{
-            uint8_t * const nnz= &h->non_zero_count_cache[0];
-            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
-            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
-        }
-    }else{
-        uint8_t * const nnz= &h->non_zero_count_cache[0];
-        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
-        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
-        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
-    }
-    s->current_picture.qscale_table[mb_xy]= s->qscale;
-    write_back_non_zero_count(h);
-
-    if(MB_MBAFF){
-        h->ref_count[0] >>= 1;
-        h->ref_count[1] >>= 1;
-    }
-
-    return 0;
-}
-
-static int decode_cabac_field_decoding_flag(H264Context *h) {
-    MpegEncContext * const s = &h->s;
-    const int mb_x = s->mb_x;
-    const int mb_y = s->mb_y & ~1;
-    const int mba_xy = mb_x - 1 +  mb_y   *s->mb_stride;
-    const int mbb_xy = mb_x     + (mb_y-2)*s->mb_stride;
-
-    unsigned int ctx = 0;
-
-    if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) {
-        ctx += 1;
-    }
-    if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) {
-        ctx += 1;
-    }
-
-    return get_cabac_noinline( &h->cabac, &h->cabac_state[70 + ctx] );
-}
-
-static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
-    uint8_t *state= &h->cabac_state[ctx_base];
-    int mb_type;
-
-    if(intra_slice){
-        MpegEncContext * const s = &h->s;
-        const int mba_xy = h->left_mb_xy[0];
-        const int mbb_xy = h->top_mb_xy;
-        int ctx=0;
-        if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
-            ctx++;
-        if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
-            ctx++;
-        if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 )
-            return 0;   /* I4x4 */
-        state += 2;
-    }else{
-        if( get_cabac_noinline( &h->cabac, &state[0] ) == 0 )
-            return 0;   /* I4x4 */
-    }
-
-    if( get_cabac_terminate( &h->cabac ) )
-        return 25;  /* PCM */
-
-    mb_type = 1; /* I16x16 */
-    mb_type += 12 * get_cabac_noinline( &h->cabac, &state[1] ); /* cbp_luma != 0 */
-    if( get_cabac_noinline( &h->cabac, &state[2] ) ) /* cbp_chroma */
-        mb_type += 4 + 4 * get_cabac_noinline( &h->cabac, &state[2+intra_slice] );
-    mb_type += 2 * get_cabac_noinline( &h->cabac, &state[3+intra_slice] );
-    mb_type += 1 * get_cabac_noinline( &h->cabac, &state[3+2*intra_slice] );
-    return mb_type;
-}
-
-static int decode_cabac_mb_type_b( H264Context *h ) {
-    MpegEncContext * const s = &h->s;
-
-        const int mba_xy = h->left_mb_xy[0];
-        const int mbb_xy = h->top_mb_xy;
-        int ctx = 0;
-        int bits;
-        assert(h->slice_type_nos == FF_B_TYPE);
-
-        if( h->slice_table[mba_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
-            ctx++;
-        if( h->slice_table[mbb_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
-            ctx++;
-
-        if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) )
-            return 0; /* B_Direct_16x16 */
-
-        if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+3] ) ) {
-            return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
-        }
-
-        bits = get_cabac_noinline( &h->cabac, &h->cabac_state[27+4] ) << 3;
-        bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 2;
-        bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 1;
-        bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );
-        if( bits < 8 )
-            return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
-        else if( bits == 13 ) {
-            return decode_cabac_intra_mb_type(h, 32, 0) + 23;
-        } else if( bits == 14 )
-            return 11; /* B_L1_L0_8x16 */
-        else if( bits == 15 )
-            return 22; /* B_8x8 */
-
-        bits= ( bits<<1 ) | get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );
-        return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
-}
-
-static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) {
-    MpegEncContext * const s = &h->s;
-    int mba_xy, mbb_xy;
-    int ctx = 0;
-
-    if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches?
-        int mb_xy = mb_x + (mb_y&~1)*s->mb_stride;
-        mba_xy = mb_xy - 1;
-        if( (mb_y&1)
-            && h->slice_table[mba_xy] == h->slice_num
-            && MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) )
-            mba_xy += s->mb_stride;
-        if( MB_FIELD ){
-            mbb_xy = mb_xy - s->mb_stride;
-            if( !(mb_y&1)
-                && h->slice_table[mbb_xy] == h->slice_num
-                && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) )
-                mbb_xy -= s->mb_stride;
-        }else
-            mbb_xy = mb_x + (mb_y-1)*s->mb_stride;
-    }else{
-        int mb_xy = h->mb_xy;
-        mba_xy = mb_xy - 1;
-        mbb_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
-    }
-
-    if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
-        ctx++;
-    if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
-        ctx++;
-
-    if( h->slice_type_nos == FF_B_TYPE )
-        ctx += 13;
-    return get_cabac_noinline( &h->cabac, &h->cabac_state[11+ctx] );
-}
-
-static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
-    int mode = 0;
-
-    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
-        return pred_mode;
-
-    mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] );
-    mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] );
-    mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] );
-
-    if( mode >= pred_mode )
-        return mode + 1;
-    else
-        return mode;
-}
-
-static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
-    const int mba_xy = h->left_mb_xy[0];
-    const int mbb_xy = h->top_mb_xy;
-
-    int ctx = 0;
-
-    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
-    if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
-        ctx++;
-
-    if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
-        ctx++;
-
-    if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
-        return 0;
-
-    if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )
-        return 1;
-    if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )
-        return 2;
-    else
-        return 3;
-}
-
-static int decode_cabac_mb_cbp_luma( H264Context *h) {
-    int cbp_b, cbp_a, ctx, cbp = 0;
-
-    cbp_a = h->slice_table[h->left_mb_xy[0]] == h->slice_num ? h->left_cbp : -1;
-    cbp_b = h->slice_table[h->top_mb_xy]     == h->slice_num ? h->top_cbp  : -1;
-
-    ctx = !(cbp_a & 0x02) + 2 * !(cbp_b & 0x04);
-    cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]);
-    ctx = !(cbp   & 0x01) + 2 * !(cbp_b & 0x08);
-    cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 1;
-    ctx = !(cbp_a & 0x08) + 2 * !(cbp   & 0x01);
-    cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 2;
-    ctx = !(cbp   & 0x04) + 2 * !(cbp   & 0x02);
-    cbp |= get_cabac_noinline(&h->cabac, &h->cabac_state[73 + ctx]) << 3;
-    return cbp;
-}
-static int decode_cabac_mb_cbp_chroma( H264Context *h) {
-    int ctx;
-    int cbp_a, cbp_b;
-
-    cbp_a = (h->left_cbp>>4)&0x03;
-    cbp_b = (h-> top_cbp>>4)&0x03;
-
-    ctx = 0;
-    if( cbp_a > 0 ) ctx++;
-    if( cbp_b > 0 ) ctx += 2;
-    if( get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
-        return 0;
-
-    ctx = 4;
-    if( cbp_a == 2 ) ctx++;
-    if( cbp_b == 2 ) ctx += 2;
-    return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] );
-}
-static int decode_cabac_mb_dqp( H264Context *h) {
-    int   ctx= h->last_qscale_diff != 0;
-    int   val = 0;
-
-    while( get_cabac_noinline( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
-        ctx= 2+(ctx>>1);
-        val++;
-        if(val > 102) //prevent infinite loop
-            return INT_MIN;
-    }
-
-    if( val&0x01 )
-        return   (val + 1)>>1 ;
-    else
-        return -((val + 1)>>1);
-}
-static int decode_cabac_p_mb_sub_type( H264Context *h ) {
-    if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
-        return 0;   /* 8x8 */
-    if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
-        return 1;   /* 8x4 */
-    if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
-        return 2;   /* 4x8 */
-    return 3;       /* 4x4 */
-}
-static int decode_cabac_b_mb_sub_type( H264Context *h ) {
-    int type;
-    if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
-        return 0;   /* B_Direct_8x8 */
-    if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
-        return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
-    type = 3;
-    if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
-        if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
-            return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
-        type += 4;
-    }
-    type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
-    type +=   get_cabac( &h->cabac, &h->cabac_state[39] );
-    return type;
-}
-
-static inline int decode_cabac_mb_transform_size( H264Context *h ) {
-    return get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] );
-}
-
-static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
-    int refa = h->ref_cache[list][scan8[n] - 1];
-    int refb = h->ref_cache[list][scan8[n] - 8];
-    int ref  = 0;
-    int ctx  = 0;
-
-    if( h->slice_type_nos == FF_B_TYPE) {
-        if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
-            ctx++;
-        if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
-            ctx += 2;
-    } else {
-        if( refa > 0 )
-            ctx++;
-        if( refb > 0 )
-            ctx += 2;
-    }
-
-    while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
-        ref++;
-        ctx = (ctx>>2)+4;
-        if(ref >= 32 /*h->ref_list[list]*/){
-            return -1;
-        }
-    }
-    return ref;
-}
-
-static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
-    int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
-               abs( h->mvd_cache[list][scan8[n] - 8][l] );
-    int ctxbase = (l == 0) ? 40 : 47;
-    int mvd;
-    int ctx = (amvd>2) + (amvd>32);
-
-    if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
-        return 0;
-
-    mvd= 1;
-    ctx= 3;
-    while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
-        mvd++;
-        if( ctx < 6 )
-            ctx++;
-    }
-
-    if( mvd >= 9 ) {
-        int k = 3;
-        while( get_cabac_bypass( &h->cabac ) ) {
-            mvd += 1 << k;
-            k++;
-            if(k>24){
-                av_log(h->s.avctx, AV_LOG_ERROR, "overflow in decode_cabac_mb_mvd\n");
-                return INT_MIN;
-            }
-        }
-        while( k-- ) {
-            if( get_cabac_bypass( &h->cabac ) )
-                mvd += 1 << k;
-        }
-    }
-    return get_cabac_bypass_sign( &h->cabac, -mvd );
-}
-
-static av_always_inline int get_cabac_cbf_ctx( H264Context *h, int cat, int idx, int is_dc ) {
-    int nza, nzb;
-    int ctx = 0;
-
-    if( is_dc ) {
-        if( cat == 0 ) {
-            nza = h->left_cbp&0x100;
-            nzb = h-> top_cbp&0x100;
-        } else {
-            nza = (h->left_cbp>>(6+idx))&0x01;
-            nzb = (h-> top_cbp>>(6+idx))&0x01;
-        }
-    } else {
-        assert(cat == 1 || cat == 2 || cat == 4);
-        nza = h->non_zero_count_cache[scan8[idx] - 1];
-        nzb = h->non_zero_count_cache[scan8[idx] - 8];
-    }
-
-    if( nza > 0 )
-        ctx++;
-
-    if( nzb > 0 )
-        ctx += 2;
-
-    return ctx + 4 * cat;
-}
-
-DECLARE_ASM_CONST(1, uint8_t, last_coeff_flag_offset_8x8[63]) = {
-    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
-    3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
-    5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8
-};
-
-static av_always_inline void decode_cabac_residual_internal( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff, int is_dc ) {
-    static const int significant_coeff_flag_offset[2][6] = {
-      { 105+0, 105+15, 105+29, 105+44, 105+47, 402 },
-      { 277+0, 277+15, 277+29, 277+44, 277+47, 436 }
-    };
-    static const int last_coeff_flag_offset[2][6] = {
-      { 166+0, 166+15, 166+29, 166+44, 166+47, 417 },
-      { 338+0, 338+15, 338+29, 338+44, 338+47, 451 }
-    };
-    static const int coeff_abs_level_m1_offset[6] = {
-        227+0, 227+10, 227+20, 227+30, 227+39, 426
-    };
-    static const uint8_t significant_coeff_flag_offset_8x8[2][63] = {
-      { 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,
-        4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7,
-        7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11,
-       12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 },
-      { 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5,
-        6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11,
-        9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9,
-        9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 }
-    };
-    /* node ctx: 0..3: abslevel1 (with abslevelgt1 == 0).
-     * 4..7: abslevelgt1 + 3 (and abslevel1 doesn't matter).
-     * map node ctx => cabac ctx for level=1 */
-    static const uint8_t coeff_abs_level1_ctx[8] = { 1, 2, 3, 4, 0, 0, 0, 0 };
-    /* map node ctx => cabac ctx for level>1 */
-    static const uint8_t coeff_abs_levelgt1_ctx[8] = { 5, 5, 5, 5, 6, 7, 8, 9 };
-    static const uint8_t coeff_abs_level_transition[2][8] = {
-    /* update node ctx after decoding a level=1 */
-        { 1, 2, 3, 3, 4, 5, 6, 7 },
-    /* update node ctx after decoding a level>1 */
-        { 4, 4, 4, 4, 5, 6, 7, 7 }
-    };
-
-    int index[64];
-
-    int av_unused last;
-    int coeff_count = 0;
-    int node_ctx = 0;
-
-    uint8_t *significant_coeff_ctx_base;
-    uint8_t *last_coeff_ctx_base;
-    uint8_t *abs_level_m1_ctx_base;
-
-#if !ARCH_X86
-#define CABAC_ON_STACK
-#endif
-#ifdef CABAC_ON_STACK
-#define CC &cc
-    CABACContext cc;
-    cc.range     = h->cabac.range;
-    cc.low       = h->cabac.low;
-    cc.bytestream= h->cabac.bytestream;
-#else
-#define CC &h->cabac
-#endif
-
-
-    /* cat: 0-> DC 16x16  n = 0
-     *      1-> AC 16x16  n = luma4x4idx
-     *      2-> Luma4x4   n = luma4x4idx
-     *      3-> DC Chroma n = iCbCr
-     *      4-> AC Chroma n = 16 + 4 * iCbCr + chroma4x4idx
-     *      5-> Luma8x8   n = 4 * luma8x8idx
-     */
-
-    /* read coded block flag */
-    if( is_dc || cat != 5 ) {
-        if( get_cabac( CC, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n, is_dc ) ] ) == 0 ) {
-            if( !is_dc )
-                h->non_zero_count_cache[scan8[n]] = 0;
-
-#ifdef CABAC_ON_STACK
-            h->cabac.range     = cc.range     ;
-            h->cabac.low       = cc.low       ;
-            h->cabac.bytestream= cc.bytestream;
-#endif
-            return;
-        }
-    }
-
-    significant_coeff_ctx_base = h->cabac_state
-        + significant_coeff_flag_offset[MB_FIELD][cat];
-    last_coeff_ctx_base = h->cabac_state
-        + last_coeff_flag_offset[MB_FIELD][cat];
-    abs_level_m1_ctx_base = h->cabac_state
-        + coeff_abs_level_m1_offset[cat];
-
-    if( !is_dc && cat == 5 ) {
-#define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \
-        for(last= 0; last < coefs; last++) { \
-            uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \
-            if( get_cabac( CC, sig_ctx )) { \
-                uint8_t *last_ctx = last_coeff_ctx_base + last_off; \
-                index[coeff_count++] = last; \
-                if( get_cabac( CC, last_ctx ) ) { \
-                    last= max_coeff; \
-                    break; \
-                } \
-            } \
-        }\
-        if( last == max_coeff -1 ) {\
-            index[coeff_count++] = last;\
-        }
-        const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD];
-#if ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS)
-        coeff_count= decode_significance_8x8_x86(CC, significant_coeff_ctx_base, index, sig_off);
-    } else {
-        coeff_count= decode_significance_x86(CC, max_coeff, significant_coeff_ctx_base, index);
-#else
-        DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] );
-    } else {
-        DECODE_SIGNIFICANCE( max_coeff - 1, last, last );
-#endif
-    }
-    assert(coeff_count > 0);
-
-    if( is_dc ) {
-        if( cat == 0 )
-            h->cbp_table[h->mb_xy] |= 0x100;
-        else
-            h->cbp_table[h->mb_xy] |= 0x40 << n;
-    } else {
-        if( cat == 5 )
-            fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1);
-        else {
-            assert( cat == 1 || cat == 2 || cat == 4 );
-            h->non_zero_count_cache[scan8[n]] = coeff_count;
-        }
-    }
-
-    do {
-        uint8_t *ctx = coeff_abs_level1_ctx[node_ctx] + abs_level_m1_ctx_base;
-
-        int j= scantable[index[--coeff_count]];
-
-        if( get_cabac( CC, ctx ) == 0 ) {
-            node_ctx = coeff_abs_level_transition[0][node_ctx];
-            if( is_dc ) {
-                block[j] = get_cabac_bypass_sign( CC, -1);
-            }else{
-                block[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6;
-            }
-        } else {
-            int coeff_abs = 2;
-            ctx = coeff_abs_levelgt1_ctx[node_ctx] + abs_level_m1_ctx_base;
-            node_ctx = coeff_abs_level_transition[1][node_ctx];
-
-            while( coeff_abs < 15 && get_cabac( CC, ctx ) ) {
-                coeff_abs++;
-            }
-
-            if( coeff_abs >= 15 ) {
-                int j = 0;
-                while( get_cabac_bypass( CC ) ) {
-                    j++;
-                }
-
-                coeff_abs=1;
-                while( j-- ) {
-                    coeff_abs += coeff_abs + get_cabac_bypass( CC );
-                }
-                coeff_abs+= 14;
-            }
-
-            if( is_dc ) {
-                block[j] = get_cabac_bypass_sign( CC, -coeff_abs );
-            }else{
-                block[j] = (get_cabac_bypass_sign( CC, -coeff_abs ) * qmul[j] + 32) >> 6;
-            }
-        }
-    } while( coeff_count );
-#ifdef CABAC_ON_STACK
-            h->cabac.range     = cc.range     ;
-            h->cabac.low       = cc.low       ;
-            h->cabac.bytestream= cc.bytestream;
-#endif
-
-}
-
-#if !CONFIG_SMALL
-static void decode_cabac_residual_dc( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff ) {
-    decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, 1);
-}
-
-static void decode_cabac_residual_nondc( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff ) {
-    decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, 0);
-}
-#endif
+               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
+               first_mb_in_slice,
+               av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
+               pps_id, h->frame_num,
+               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
+               h->ref_count[0], h->ref_count[1],
+               s->qscale,
+               h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
+               h->use_weight,
+               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
+               h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
+               );
+    }
 
-static void decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff ) {
-#if CONFIG_SMALL
-    decode_cabac_residual_internal(h, block, cat, n, scantable, qmul, max_coeff, cat == 0 || cat == 3);
-#else
-    if( cat == 0 || cat == 3 ) decode_cabac_residual_dc(h, block, cat, n, scantable, qmul, max_coeff);
-    else decode_cabac_residual_nondc(h, block, cat, n, scantable, qmul, max_coeff);
-#endif
+    return 0;
 }
 
-static inline void compute_mb_neighbors(H264Context *h)
+int ff_h264_get_slice_type(const H264Context *h)
 {
-    MpegEncContext * const s = &h->s;
-    const int mb_xy  = h->mb_xy;
-    h->top_mb_xy     = mb_xy - s->mb_stride;
-    h->left_mb_xy[0] = mb_xy - 1;
-    if(FRAME_MBAFF){
-        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
-        const int top_pair_xy      = pair_xy     - s->mb_stride;
-        const int top_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
-        const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
-        const int curr_mb_field_flag = MB_FIELD;
-        const int bottom = (s->mb_y & 1);
-
-        if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
-            h->top_mb_xy -= s->mb_stride;
-        }
-        if (!left_mb_field_flag == curr_mb_field_flag) {
-            h->left_mb_xy[0] = pair_xy - 1;
-        }
-    } else if (FIELD_PICTURE) {
-        h->top_mb_xy -= s->mb_stride;
+    switch (h->slice_type) {
+    case AV_PICTURE_TYPE_P:  return 0;
+    case AV_PICTURE_TYPE_B:  return 1;
+    case AV_PICTURE_TYPE_I:  return 2;
+    case AV_PICTURE_TYPE_SP: return 3;
+    case AV_PICTURE_TYPE_SI: return 4;
+    default:         return -1;
     }
-    return;
 }
 
 /**
- * decodes a macroblock
- * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
+ *
+ * @return non zero if the loop filter can be skiped
  */
-static int decode_mb_cabac(H264Context *h) {
+static int fill_filter_caches(H264Context *h, int mb_type){
     MpegEncContext * const s = &h->s;
-    int mb_xy;
-    int mb_type, partition_count, cbp = 0;
-    int dct8x8_allowed= h->pps.transform_8x8_mode;
-
-    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
-
-    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
-    if( h->slice_type_nos != FF_I_TYPE ) {
-        int skip;
-        /* a skipped mb needs the aff flag from the following mb */
-        if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 )
-            predict_field_decoding_flag(h);
-        if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped )
-            skip = h->next_mb_skipped;
-        else
-            skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y );
-        /* read skip flags */
-        if( skip ) {
-            if( FRAME_MBAFF && (s->mb_y&1)==0 ){
-                s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP;
-                h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 );
-                if(!h->next_mb_skipped)
-                    h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
-            }
+    const int mb_xy= h->mb_xy;
+    int top_xy, left_xy[2];
+    int top_type, left_type[2];
 
-            decode_mb_skip(h);
+    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
 
-            h->cbp_table[mb_xy] = 0;
-            h->chroma_pred_mode_table[mb_xy] = 0;
-            h->last_qscale_diff = 0;
+    //FIXME deblocking could skip the intra and nnz parts.
 
-            return 0;
+    /* Wow, what a mess, why didn't they simplify the interlacing & intra
+     * stuff, I can't imagine that these complex rules are worth it. */
 
+    left_xy[1] = left_xy[0] = mb_xy-1;
+    if(FRAME_MBAFF){
+        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
+        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
+        if(s->mb_y&1){
+            if (left_mb_field_flag != curr_mb_field_flag) {
+                left_xy[0] -= s->mb_stride;
+            }
+        }else{
+            if(curr_mb_field_flag){
+                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
+            }
+            if (left_mb_field_flag != curr_mb_field_flag) {
+                left_xy[1] += s->mb_stride;
+            }
         }
     }
-    if(FRAME_MBAFF){
-        if( (s->mb_y&1) == 0 )
-            h->mb_mbaff =
-            h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
+
+    h->top_mb_xy = top_xy;
+    h->left_mb_xy[0] = left_xy[0];
+    h->left_mb_xy[1] = left_xy[1];
+    {
+        //for sufficiently low qp, filtering wouldn't do anything
+        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
+        int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
+        int qp = s->current_picture.qscale_table[mb_xy];
+        if(qp <= qp_thresh
+           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
+           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
+            if(!FRAME_MBAFF)
+                return 1;
+            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
+               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
+                return 1;
+        }
+    }
+
+    top_type     = s->current_picture.mb_type[top_xy]    ;
+    left_type[0] = s->current_picture.mb_type[left_xy[0]];
+    left_type[1] = s->current_picture.mb_type[left_xy[1]];
+    if(h->deblocking_filter == 2){
+        if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
+        if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
+    }else{
+        if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
+        if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
     }
+    h->top_type    = top_type    ;
+    h->left_type[0]= left_type[0];
+    h->left_type[1]= left_type[1];
+
+    if(IS_INTRA(mb_type))
+        return 0;
 
-    h->prev_mb_skipped = 0;
+    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
+    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
+    AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
+    AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
+    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
 
-    compute_mb_neighbors(h);
+    h->cbp= h->cbp_table[mb_xy];
 
-    if( h->slice_type_nos == FF_B_TYPE ) {
-        mb_type = decode_cabac_mb_type_b( h );
-        if( mb_type < 23 ){
-            partition_count= b_mb_type_info[mb_type].partition_count;
-            mb_type=         b_mb_type_info[mb_type].type;
-        }else{
-            mb_type -= 23;
-            goto decode_intra_mb;
-        }
-    } else if( h->slice_type_nos == FF_P_TYPE ) {
-        if( get_cabac_noinline( &h->cabac, &h->cabac_state[14] ) == 0 ) {
-            /* P-type */
-            if( get_cabac_noinline( &h->cabac, &h->cabac_state[15] ) == 0 ) {
-                /* P_L0_D16x16, P_8x8 */
-                mb_type= 3 * get_cabac_noinline( &h->cabac, &h->cabac_state[16] );
-            } else {
-                /* P_L0_D8x16, P_L0_D16x8 */
-                mb_type= 2 - get_cabac_noinline( &h->cabac, &h->cabac_state[17] );
+    {
+        int list;
+        for(list=0; list<h->list_count; list++){
+            int8_t *ref;
+            int y, b_stride;
+            int16_t (*mv_dst)[2];
+            int16_t (*mv_src)[2];
+
+            if(!USES_LIST(mb_type, list)){
+                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
+                AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+                AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+                AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+                AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+                continue;
             }
-            partition_count= p_mb_type_info[mb_type].partition_count;
-            mb_type=         p_mb_type_info[mb_type].type;
-        } else {
-            mb_type= decode_cabac_intra_mb_type(h, 17, 0);
-            goto decode_intra_mb;
-        }
-    } else {
-        mb_type= decode_cabac_intra_mb_type(h, 3, 1);
-        if(h->slice_type == FF_SI_TYPE && mb_type)
-            mb_type--;
-        assert(h->slice_type_nos == FF_I_TYPE);
-decode_intra_mb:
-        partition_count = 0;
-        cbp= i_mb_type_info[mb_type].cbp;
-        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
-        mb_type= i_mb_type_info[mb_type].type;
-    }
-    if(MB_FIELD)
-        mb_type |= MB_TYPE_INTERLACED;
 
-    h->slice_table[ mb_xy ]= h->slice_num;
+            ref = &s->current_picture.ref_index[list][4*mb_xy];
+            {
+                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
+                AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
+                ref += 2;
+                AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
+                AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
+            }
 
-    if(IS_INTRA_PCM(mb_type)) {
-        const uint8_t *ptr;
+            b_stride = h->b_stride;
+            mv_dst   = &h->mv_cache[list][scan8[0]];
+            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
+            for(y=0; y<4; y++){
+                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
+            }
 
-        // We assume these blocks are very rare so we do not optimize it.
-        // FIXME The two following lines get the bitstream position in the cabac
-        // decode, I think it should be done by a function in cabac.h (or cabac.c).
-        ptr= h->cabac.bytestream;
-        if(h->cabac.low&0x1) ptr--;
-        if(CABAC_BITS==16){
-            if(h->cabac.low&0x1FF) ptr--;
         }
+    }
 
-        // The pixels are stored in the same order as levels in h->mb array.
-        memcpy(h->mb, ptr, 256); ptr+=256;
-        if(CHROMA){
-            memcpy(h->mb+128, ptr, 128); ptr+=128;
-        }
 
-        ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
-
-        // All blocks are present
-        h->cbp_table[mb_xy] = 0x1ef;
-        h->chroma_pred_mode_table[mb_xy] = 0;
-        // In deblocking, the quantizer is 0
-        s->current_picture.qscale_table[mb_xy]= 0;
-        // All coeffs are present
-        memset(h->non_zero_count[mb_xy], 16, 16);
-        s->current_picture.mb_type[mb_xy]= mb_type;
-        h->last_qscale_diff = 0;
-        return 0;
+/*
+0 . T T. T T T T
+1 L . .L . . . .
+2 L . .L . . . .
+3 . T TL . . . .
+4 L . .L . . . .
+5 L . .. . . . .
+*/
+//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
+    if(top_type){
+        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
     }
 
-    if(MB_MBAFF){
-        h->ref_count[0] <<= 1;
-        h->ref_count[1] <<= 1;
+    if(left_type[0]){
+        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
+        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
+        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
+        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
     }
 
-    fill_caches(h, mb_type, 0);
-
-    if( IS_INTRA( mb_type ) ) {
-        int i, pred_mode;
-        if( IS_INTRA4x4( mb_type ) ) {
-            if( dct8x8_allowed && decode_cabac_mb_transform_size( h ) ) {
-                mb_type |= MB_TYPE_8x8DCT;
-                for( i = 0; i < 16; i+=4 ) {
-                    int pred = pred_intra_mode( h, i );
-                    int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred );
-                    fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
-                }
-            } else {
-                for( i = 0; i < 16; i++ ) {
-                    int pred = pred_intra_mode( h, i );
-                    h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
-
-                //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
-                }
-            }
-            ff_h264_write_back_intra_pred_mode(h);
-            if( ff_h264_check_intra4x4_pred_mode(h) < 0 ) return -1;
-        } else {
-            h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode( h, h->intra16x16_pred_mode );
-            if( h->intra16x16_pred_mode < 0 ) return -1;
-        }
-        if(CHROMA){
-            h->chroma_pred_mode_table[mb_xy] =
-            pred_mode                        = decode_cabac_mb_chroma_pre_mode( h );
-
-            pred_mode= ff_h264_check_intra_pred_mode( h, pred_mode );
-            if( pred_mode < 0 ) return -1;
-            h->chroma_pred_mode= pred_mode;
+    // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
+    if(!CABAC && h->pps.transform_8x8_mode){
+        if(IS_8x8DCT(top_type)){
+            h->non_zero_count_cache[4+8*0]=
+            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
+            h->non_zero_count_cache[6+8*0]=
+            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
         }
-    } else if( partition_count == 4 ) {
-        int i, j, sub_partition_count[4], list, ref[2][4];
-
-        if( h->slice_type_nos == FF_B_TYPE ) {
-            for( i = 0; i < 4; i++ ) {
-                h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
-                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
-                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
-            }
-            if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] |
-                          h->sub_mb_type[2] | h->sub_mb_type[3]) ) {
-                ff_h264_pred_direct_motion(h, &mb_type);
-                h->ref_cache[0][scan8[4]] =
-                h->ref_cache[1][scan8[4]] =
-                h->ref_cache[0][scan8[12]] =
-                h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;
-                if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
-                    for( i = 0; i < 4; i++ )
-                        if( IS_DIRECT(h->sub_mb_type[i]) )
-                            fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
-                }
-            }
-        } else {
-            for( i = 0; i < 4; i++ ) {
-                h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
-                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
-                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
-            }
+        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[left_xy[0]]&2; //FIXME check MBAFF
         }
-
-        for( list = 0; list < h->list_count; list++ ) {
-                for( i = 0; i < 4; i++ ) {
-                    if(IS_DIRECT(h->sub_mb_type[i])) continue;
-                    if(IS_DIR(h->sub_mb_type[i], 0, list)){
-                        if( h->ref_count[list] > 1 ){
-                            ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
-                            if(ref[list][i] >= (unsigned)h->ref_count[list]){
-                                av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref[list][i], h->ref_count[list]);
-                                return -1;
-                            }
-                        }else
-                            ref[list][i] = 0;
-                    } else {
-                        ref[list][i] = -1;
-                    }
-                                                       h->ref_cache[list][ scan8[4*i]+1 ]=
-                    h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
-                }
+        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[left_xy[1]]&8; //FIXME check MBAFF
         }
 
-        if(dct8x8_allowed)
-            dct8x8_allowed = get_dct8x8_allowed(h);
+        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;
 
-        for(list=0; list<h->list_count; list++){
-            for(i=0; i<4; i++){
-                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ];
-                if(IS_DIRECT(h->sub_mb_type[i])){
-                    fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
-                    continue;
-                }
+            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;
 
-                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
-                    const int sub_mb_type= h->sub_mb_type[i];
-                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
-                    for(j=0; j<sub_partition_count[i]; j++){
-                        int mpx, mpy;
-                        int mx, my;
-                        const int index= 4*i + block_width*j;
-                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
-                        int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
-                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
-
-                        mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
-                        my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
-                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-
-                        if(IS_SUB_8X8(sub_mb_type)){
-                            mv_cache[ 1 ][0]=
-                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
-                            mv_cache[ 1 ][1]=
-                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
-
-                            mvd_cache[ 1 ][0]=
-                            mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
-                            mvd_cache[ 1 ][1]=
-                            mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
-                        }else if(IS_SUB_8X4(sub_mb_type)){
-                            mv_cache[ 1 ][0]= mx;
-                            mv_cache[ 1 ][1]= my;
-
-                            mvd_cache[ 1 ][0]= mx - mpx;
-                            mvd_cache[ 1 ][1]= my - mpy;
-                        }else if(IS_SUB_4X8(sub_mb_type)){
-                            mv_cache[ 8 ][0]= mx;
-                            mv_cache[ 8 ][1]= my;
-
-                            mvd_cache[ 8 ][0]= mx - mpx;
-                            mvd_cache[ 8 ][1]= my - mpy;
-                        }
-                        mv_cache[ 0 ][0]= mx;
-                        mv_cache[ 0 ][1]= my;
+            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;
 
-                        mvd_cache[ 0 ][0]= mx - mpx;
-                        mvd_cache[ 0 ][1]= my - mpy;
-                    }
-                }else{
-                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
-                    uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
-                    p[0] = p[1] = p[8] = p[9] = 0;
-                    pd[0]= pd[1]= pd[8]= pd[9]= 0;
-                }
-            }
+            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;
         }
-    } else if( IS_DIRECT(mb_type) ) {
-        ff_h264_pred_direct_motion(h, &mb_type);
-        fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
-        fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
-        dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
-    } else {
-        int list, mx, my, i, mpx, mpy;
-        if(IS_16X16(mb_type)){
-            for(list=0; list<h->list_count; list++){
-                if(IS_DIR(mb_type, 0, list)){
-                    int ref;
-                    if(h->ref_count[list] > 1){
-                        ref= decode_cabac_mb_ref(h, list, 0);
-                        if(ref >= (unsigned)h->ref_count[list]){
-                            av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, h->ref_count[list]);
-                            return -1;
-                        }
-                    }else
-                        ref=0;
-                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
-                }else
-                    fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1); //FIXME factorize and the other fill_rect below too
-            }
-            for(list=0; list<h->list_count; list++){
-                if(IS_DIR(mb_type, 0, list)){
-                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
-
-                    mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
-                    my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
-                    tprintf(s->avctx, "final mv:%d %d\n", mx, my);
+    }
 
-                    fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
-                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
-                }else
-                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
-            }
-        }
-        else if(IS_16X8(mb_type)){
-            for(list=0; list<h->list_count; list++){
-                    for(i=0; i<2; i++){
-                        if(IS_DIR(mb_type, i, list)){
-                            int ref;
-                            if(h->ref_count[list] > 1){
-                                ref= decode_cabac_mb_ref( h, list, 8*i );
-                                if(ref >= (unsigned)h->ref_count[list]){
-                                    av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, h->ref_count[list]);
-                                    return -1;
-                                }
-                            }else
-                                ref=0;
-                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
-                        }else
-                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
-                    }
-            }
-            for(list=0; list<h->list_count; list++){
-                for(i=0; i<2; i++){
-                    if(IS_DIR(mb_type, i, list)){
-                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
-                        mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
-                        my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
-                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-
-                        fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
-                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
-                    }else{
-                        fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
-                        fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
-                    }
-                }
-            }
-        }else{
-            assert(IS_8X16(mb_type));
-            for(list=0; list<h->list_count; list++){
-                    for(i=0; i<2; i++){
-                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
-                            int ref;
-                            if(h->ref_count[list] > 1){
-                                ref= decode_cabac_mb_ref( h, list, 4*i );
-                                if(ref >= (unsigned)h->ref_count[list]){
-                                    av_log(s->avctx, AV_LOG_ERROR, "Reference %d >= %d\n", ref, h->ref_count[list]);
-                                    return -1;
-                                }
-                            }else
-                                ref=0;
-                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
-                        }else
-                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
-                    }
+    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
+        int list;
+        for(list=0; list<h->list_count; list++){
+            if(USES_LIST(top_type, list)){
+                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
+                const int b8_xy= 4*top_xy + 2;
+                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
+                h->ref_cache[list][scan8[0] + 0 - 1*8]=
+                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
+                h->ref_cache[list][scan8[0] + 2 - 1*8]=
+                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
+            }else{
+                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
+                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
             }
-            for(list=0; list<h->list_count; list++){
-                for(i=0; i<2; i++){
-                    if(IS_DIR(mb_type, i, list)){
-                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
-                        mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
-                        my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
-
-                        tprintf(s->avctx, "final mv:%d %d\n", mx, my);
-                        fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
-                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
-                    }else{
-                        fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
-                        fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
-                    }
+
+            if(!IS_INTERLACED(mb_type^left_type[0])){
+                if(USES_LIST(left_type[0], list)){
+                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
+                    const int b8_xy= 4*left_xy[0] + 1;
+                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
+                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
+                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
+                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
+                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
+                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
+                    h->ref_cache[list][scan8[0] - 1 +16 ]=
+                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
+                }else{
+                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
+                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
+                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
+                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
+                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
+                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
+                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
+                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
                 }
             }
         }
     }
 
-   if( IS_INTER( mb_type ) ) {
-        h->chroma_pred_mode_table[mb_xy] = 0;
-        write_back_motion( h, mb_type );
-   }
-
-    if( !IS_INTRA16x16( mb_type ) ) {
-        cbp  = decode_cabac_mb_cbp_luma( h );
-        if(CHROMA)
-            cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
-    }
-
-    h->cbp_table[mb_xy] = h->cbp = cbp;
-
-    if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) {
-        if( decode_cabac_mb_transform_size( h ) )
-            mb_type |= MB_TYPE_8x8DCT;
-    }
-    s->current_picture.mb_type[mb_xy]= mb_type;
-
-    if( cbp || IS_INTRA16x16( mb_type ) ) {
-        const uint8_t *scan, *scan8x8, *dc_scan;
-        const uint32_t *qmul;
-        int dqp;
+    return 0;
+}
 
-        if(IS_INTERLACED(mb_type)){
-            scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0;
-            scan= s->qscale ? h->field_scan : h->field_scan_q0;
-            dc_scan= luma_dc_field_scan;
-        }else{
-            scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;
-            scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
-            dc_scan= luma_dc_zigzag_scan;
-        }
+static void loop_filter(H264Context *h, int start_x, int end_x){
+    MpegEncContext * const s = &h->s;
+    uint8_t  *dest_y, *dest_cb, *dest_cr;
+    int linesize, uvlinesize, mb_x, mb_y;
+    const int end_mb_y= s->mb_y + FRAME_MBAFF;
+    const int old_slice_type= h->slice_type;
+    const int pixel_shift = h->pixel_shift;
 
-        h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
-        if( dqp == INT_MIN ){
-            av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y);
-            return -1;
-        }
-        s->qscale += dqp;
-        if(((unsigned)s->qscale) > 51){
-            if(s->qscale<0) s->qscale+= 52;
-            else            s->qscale-= 52;
-        }
-        h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
-        h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
-
-        if( IS_INTRA16x16( mb_type ) ) {
-            int i;
-            //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
-            decode_cabac_residual( h, h->mb, 0, 0, dc_scan, NULL, 16);
-
-            if( cbp&15 ) {
-                qmul = h->dequant4_coeff[0][s->qscale];
-                for( i = 0; i < 16; i++ ) {
-                    //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
-                    decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, qmul, 15);
-                }
-            } else {
-                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
-            }
-        } else {
-            int i8x8, i4x4;
-            for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
-                if( cbp & (1<<i8x8) ) {
-                    if( IS_8x8DCT(mb_type) ) {
-                        decode_cabac_residual(h, h->mb + 64*i8x8, 5, 4*i8x8,
-                            scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64);
-                    } else {
-                        qmul = h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale];
-                        for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
-                            const int index = 4*i8x8 + i4x4;
-                            //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
-//START_TIMER
-                            decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, qmul, 16);
-//STOP_TIMER("decode_residual")
-                        }
+    if(h->deblocking_filter) {
+        for(mb_x= start_x; mb_x<end_x; mb_x++){
+            for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
+                int mb_xy, mb_type;
+                mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
+                h->slice_num= h->slice_table[mb_xy];
+                mb_type= s->current_picture.mb_type[mb_xy];
+                h->list_count= h->list_counts[mb_xy];
+
+                if(FRAME_MBAFF)
+                    h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
+
+                s->mb_x= mb_x;
+                s->mb_y= mb_y;
+                dest_y  = s->current_picture.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
+                dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
+                dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
+                    //FIXME simplify above
+
+                if (MB_FIELD) {
+                    linesize   = h->mb_linesize   = s->linesize * 2;
+                    uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
+                    if(mb_y&1){ //FIXME move out of this function?
+                        dest_y -= s->linesize*15;
+                        dest_cb-= s->uvlinesize*7;
+                        dest_cr-= s->uvlinesize*7;
                     }
                 } else {
-                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
-                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
+                    linesize   = h->mb_linesize   = s->linesize;
+                    uvlinesize = h->mb_uvlinesize = s->uvlinesize;
                 }
-            }
-        }
-
-        if( cbp&0x30 ){
-            int c;
-            for( c = 0; c < 2; c++ ) {
-                //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
-                decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, NULL, 4);
-            }
-        }
+                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
+                if(fill_filter_caches(h, mb_type))
+                    continue;
+                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
+                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
 
-        if( cbp&0x20 ) {
-            int c, i;
-            for( c = 0; c < 2; c++ ) {
-                qmul = h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[c]];
-                for( i = 0; i < 4; i++ ) {
-                    const int index = 16 + 4 * c + i;
-                    //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
-                    decode_cabac_residual(h, h->mb + 16*index, 4, index, scan + 1, qmul, 15);
+                if (FRAME_MBAFF) {
+                    ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
+                } else {
+                    ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
                 }
             }
-        } else {
-            uint8_t * const nnz= &h->non_zero_count_cache[0];
-            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
-            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
         }
-    } else {
-        uint8_t * const nnz= &h->non_zero_count_cache[0];
-        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
-        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
-        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
-        h->last_qscale_diff = 0;
-    }
-
-    s->current_picture.qscale_table[mb_xy]= s->qscale;
-    write_back_non_zero_count(h);
-
-    if(MB_MBAFF){
-        h->ref_count[0] >>= 1;
-        h->ref_count[1] >>= 1;
     }
+    h->slice_type= old_slice_type;
+    s->mb_x= 0;
+    s->mb_y= end_mb_y - FRAME_MBAFF;
+    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
+    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
+}
 
-    return 0;
+static void predict_field_decoding_flag(H264Context *h){
+    MpegEncContext * const s = &h->s;
+    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
+    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
+                ? s->current_picture.mb_type[mb_xy-1]
+                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
+                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
+                : 0;
+    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
 }
 
 static int decode_slice(struct AVCodecContext *avctx, void *arg){
     H264Context *h = *(void**)arg;
     MpegEncContext * const s = &h->s;
     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
+    int lf_x_start = s->mb_x;
 
     s->mb_skip_run= -1;
 
@@ -4790,8 +2640,6 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
                     (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
 
     if( h->pps.cabac ) {
-        int i;
-
         /* realign */
         align_get_bits( &s->gb );
 
@@ -4800,23 +2648,12 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
         ff_init_cabac_decoder( &h->cabac,
                                s->gb.buffer + get_bits_count(&s->gb)/8,
                                (get_bits_left(&s->gb) + 7)/8);
-        /* calculate pre-state */
-        for( i= 0; i < 460; i++ ) {
-            int pre;
-            if( h->slice_type_nos == FF_I_TYPE )
-                pre = av_clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
-            else
-                pre = av_clip( ((cabac_context_init_PB[h->cabac_init_idc][i][0] * s->qscale) >>4 ) + cabac_context_init_PB[h->cabac_init_idc][i][1], 1, 126 );
-
-            if( pre <= 63 )
-                h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
-            else
-                h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
-        }
+
+        ff_h264_init_cabac_states(h);
 
         for(;;){
 //START_TIMER
-            int ret = decode_mb_cabac(h);
+            int ret = ff_h264_decode_mb_cabac(h);
             int eos;
 //STOP_TIMER("decode_mb_cabac")
 
@@ -4825,13 +2662,18 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
                 s->mb_y++;
 
-                ret = decode_mb_cabac(h);
+                ret = ff_h264_decode_mb_cabac(h);
 
                 if(ret>=0) ff_h264_hl_decode_mb(h);
                 s->mb_y--;
             }
             eos = get_cabac_terminate( &h->cabac );
 
+            if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
+                return 0;
+            }
             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
@@ -4839,30 +2681,34 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             }
 
             if( ++s->mb_x >= s->mb_width ) {
-                s->mb_x = 0;
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
                     ++s->mb_y;
+                    if(FRAME_MBAFF && s->mb_y < s->mb_height)
+                        predict_field_decoding_flag(h);
                 }
             }
 
             if( eos || s->mb_y >= s->mb_height ) {
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
                 return 0;
             }
         }
 
     } else {
         for(;;){
-            int ret = decode_mb_cavlc(h);
+            int ret = ff_h264_decode_mb_cavlc(h);
 
             if(ret>=0) ff_h264_hl_decode_mb(h);
 
             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
                 s->mb_y++;
-                ret = decode_mb_cavlc(h);
+                ret = ff_h264_decode_mb_cavlc(h);
 
                 if(ret>=0) ff_h264_hl_decode_mb(h);
                 s->mb_y--;
@@ -4871,16 +2717,18 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             if(ret<0){
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
                 return -1;
             }
 
             if(++s->mb_x >= s->mb_width){
-                s->mb_x=0;
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
                 ff_draw_horiz_band(s, 16*s->mb_y, 16);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
                     ++s->mb_y;
+                    if(FRAME_MBAFF && s->mb_y < s->mb_height)
+                        predict_field_decoding_flag(h);
                 }
                 if(s->mb_y >= s->mb_height){
                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
@@ -4901,6 +2749,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                    if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
 
                     return 0;
                 }else{
@@ -5009,12 +2858,6 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
     int next_avc= h->is_avc ? 0 : buf_size;
 
     h->max_contexts = avctx->thread_count;
-#if 0
-    int i;
-    for(i=0; i<50; i++){
-        av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
-    }
-#endif
     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
         h->current_slice = 0;
         if (!s->first_field)
@@ -5035,14 +2878,9 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             nalsize = 0;
             for(i = 0; i < h->nal_length_size; i++)
                 nalsize = (nalsize << 8) | buf[buf_index++];
-            if(nalsize <= 1 || nalsize > buf_size - buf_index){
-                if(nalsize == 1){
-                    buf_index++;
-                    continue;
-                }else{
-                    av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
-                    break;
-                }
+            if(nalsize <= 0 || nalsize > buf_size - buf_index){
+                av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
+                break;
             }
             next_avc= buf_index + nalsize;
         } else {
@@ -5065,8 +2903,15 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
         if (ptr==NULL || dst_length < 0){
             return -1;
         }
+        i= buf_index + consumed;
+        if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
+           buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
+            s->workaround_bugs |= FF_BUG_TRUNCATED;
+
+        if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
         while(ptr[dst_length - 1] == 0 && dst_length > 0)
             dst_length--;
+        }
         bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
 
         if(s->avctx->debug&FF_DEBUG_STARTCODE){
@@ -5079,8 +2924,8 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
 
         buf_index += consumed;
 
-        if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
-           ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
+        //FIXME do not discard SEI id
+        if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
             continue;
 
       again:
@@ -5101,18 +2946,20 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             if((err = decode_slice_header(hx, h)))
                break;
 
-            if (s->avctx->hwaccel && h->current_slice == 1) {
-                if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
+            if (h->current_slice == 1) {
+                if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
                     return -1;
+                if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
+                    ff_vdpau_h264_picture_start(s);
             }
 
             s->current_picture_ptr->key_frame |=
                     (hx->nal_unit_type == NAL_IDR_SLICE) ||
                     (h->sei_recovery_frame_cnt >= 0);
-            if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
+            if(hx->redundant_pic_count==0
                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
-               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
-               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
+               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
+               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
                && avctx->skip_frame < AVDISCARD_ALL){
                 if(avctx->hwaccel) {
                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
@@ -5147,10 +2994,9 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
 
             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
                && s->context_initialized
-               && s->hurry_up < 5
                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
-               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
-               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
+               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
+               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
                && avctx->skip_frame < AVDISCARD_ALL)
                 context_count++;
             break;
@@ -5167,6 +3013,20 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
 
             if(avctx->has_b_frames < 2)
                 avctx->has_b_frames= !s->low_delay;
+
+            if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma) {
+                if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
+                    avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
+                    h->pixel_shift = h->sps.bit_depth_luma > 8;
+
+                    ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma);
+                    ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma);
+                    dsputil_init(&s->dsp, s->avctx);
+                } else {
+                    av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
+                    return -1;
+                }
+            }
             break;
         case NAL_PPS:
             init_get_bits(&s->gb, ptr, bit_length);
@@ -5233,6 +3093,7 @@ static int decode_frame(AVCodecContext *avctx,
     s->flags2= avctx->flags2;
 
    /* end of stream, output what is still in the buffers */
+ out:
     if (buf_size == 0) {
         Picture *out;
         int i, out_idx;
@@ -5257,59 +3118,18 @@ static int decode_frame(AVCodecContext *avctx,
         return 0;
     }
 
-    if(h->is_avc && !h->got_avcC) {
-        int i, cnt, nalsize;
-        unsigned char *p = avctx->extradata;
-        if(avctx->extradata_size < 7) {
-            av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
-            return -1;
-        }
-        if(*p != 1) {
-            av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
-            return -1;
-        }
-        /* sps and pps in the avcC always have length coded with 2 bytes,
-           so put a fake nal_length_size = 2 while parsing them */
-        h->nal_length_size = 2;
-        // Decode sps from avcC
-        cnt = *(p+5) & 0x1f; // Number of sps
-        p += 6;
-        for (i = 0; i < cnt; i++) {
-            nalsize = AV_RB16(p) + 2;
-            if(decode_nal_units(h, p, nalsize) < 0) {
-                av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
-                return -1;
-            }
-            p += nalsize;
-        }
-        // Decode pps from avcC
-        cnt = *(p++); // Number of pps
-        for (i = 0; i < cnt; i++) {
-            nalsize = AV_RB16(p) + 2;
-            if(decode_nal_units(h, p, nalsize)  != nalsize) {
-                av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
-                return -1;
-            }
-            p += nalsize;
-        }
-        // Now store right nal length size, that will be use to parse all other nals
-        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
-        // Do not reparse avcC
-        h->got_avcC = 1;
-    }
-
-    if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
-        if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
-            return -1;
-        h->got_avcC = 1;
-    }
-
     buf_index=decode_nal_units(h, buf, buf_size);
     if(buf_index < 0)
         return -1;
 
+    if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
+        buf_size = 0;
+        goto out;
+    }
+
     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
-        if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
+        if (avctx->skip_frame >= AVDISCARD_NONREF)
+            return 0;
         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
         return -1;
     }
@@ -5430,7 +3250,7 @@ static int decode_frame(AVCodecContext *avctx,
             else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
                || (s->low_delay &&
                 ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
-                 || cur->pict_type == FF_B_TYPE)))
+                 || cur->pict_type == AV_PICTURE_TYPE_B)))
             {
                 s->low_delay = 0;
                 s->avctx->has_b_frames++;
@@ -5572,7 +3392,7 @@ int main(void){
         }
 //        printf("\n");
 
-        s->dsp.h264_idct_add(ref, block, 4);
+        h->h264dsp.h264_idct_add(ref, block, 4);
 /*        for(j=0; j<16; j++){
             printf("%d ", ref[j]);
         }
@@ -5661,7 +3481,7 @@ av_cold void ff_h264_free_context(H264Context *h)
 {
     int i;
 
-    free_tables(h); //FIXME cleanup init stuff perhaps
+    free_tables(h, 1); //FIXME cleanup init stuff perhaps
 
     for(i = 0; i < MAX_SPS_COUNT; i++)
         av_freep(h->sps_buffers + i);
@@ -5684,26 +3504,43 @@ av_cold int ff_h264_decode_end(AVCodecContext *avctx)
     return 0;
 }
 
+static const AVProfile profiles[] = {
+    { FF_PROFILE_H264_BASELINE,             "Baseline"              },
+    { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
+    { FF_PROFILE_H264_MAIN,                 "Main"                  },
+    { FF_PROFILE_H264_EXTENDED,             "Extended"              },
+    { FF_PROFILE_H264_HIGH,                 "High"                  },
+    { FF_PROFILE_H264_HIGH_10,              "High 10"               },
+    { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
+    { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
+    { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
+    { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
+    { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
+    { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
+    { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
+    { FF_PROFILE_UNKNOWN },
+};
 
-AVCodec h264_decoder = {
+AVCodec ff_h264_decoder = {
     "h264",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_H264,
     sizeof(H264Context),
     ff_h264_decode_init,
     NULL,
     ff_h264_decode_end,
     decode_frame,
-    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
+    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
+        CODEC_CAP_SLICE_THREADS,
     .flush= flush_dpb,
     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
-    .pix_fmts= ff_hwaccel_pixfmt_list_420,
+    .profiles = NULL_IF_CONFIG_SMALL(profiles),
 };
 
 #if CONFIG_H264_VDPAU_DECODER
-AVCodec h264_vdpau_decoder = {
+AVCodec ff_h264_vdpau_decoder = {
     "h264_vdpau",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_H264,
     sizeof(H264Context),
     ff_h264_decode_init,
@@ -5714,5 +3551,6 @@ AVCodec h264_vdpau_decoder = {
     .flush= flush_dpb,
     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
+    .profiles = NULL_IF_CONFIG_SMALL(profiles),
 };
 #endif