]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/h264.c
data_size = 0 cleanup
[ffmpeg] / libavcodec / h264.c
index 63d1af17460b58b981983dc0a66949e5666bac32..594a4b2642b08b2e7599434ca05f82ed320ec523 100644 (file)
@@ -31,6 +31,8 @@
 #include "h264data.h"
 #include "golomb.h"
 
+#include "cabac.h"
+
 #undef NDEBUG
 #include <assert.h>
 
@@ -59,9 +61,6 @@ typedef struct SPS{
     
     int profile_idc;
     int level_idc;
-    int multiple_slice_groups;         ///< more_than_one_slice_group_allowed_flag
-    int arbitrary_slice_order;         ///< arbitrary_slice_order_allowed_flag
-    int redundant_slices;              ///< redundant_slices_allowed_flag
     int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
     int poc_type;                      ///< pic_order_cnt_type
     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
@@ -70,15 +69,19 @@ typedef struct SPS{
     int offset_for_top_to_bottom_field;
     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
     int ref_frame_count;               ///< num_ref_frames
-    int required_frame_num_update_behaviour_flag;
+    int gaps_in_frame_num_allowed_flag;
     int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
     int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
     int frame_mbs_only_flag;
     int mb_aff;                        ///<mb_adaptive_frame_field_flag
     int direct_8x8_inference_flag;
+    int crop;                   ///< frame_cropping_flag
+    int crop_left;              ///< frame_cropping_rect_left_offset
+    int crop_right;             ///< frame_cropping_rect_right_offset
+    int crop_top;               ///< frame_cropping_rect_top_offset
+    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
     int vui_parameters_present_flag;
-    int sar_width;
-    int sar_height;
+    AVRational sar;
     short offset_for_ref_frame[256]; //FIXME dyn aloc?
 }SPS;
 
@@ -100,11 +103,6 @@ typedef struct PPS{
     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
     int constrained_intra_pred; ///< constrained_intra_pred_flag
     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
-    int crop;                   ///< frame_cropping_flag
-    int crop_left;              ///< frame_cropping_rect_left_offset
-    int crop_right;             ///< frame_cropping_rect_right_offset
-    int crop_top;               ///< frame_cropping_rect_top_offset
-    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
 }PPS;
 
 /**
@@ -166,6 +164,8 @@ typedef struct H264Context{
     unsigned int top_samples_available;
     unsigned int topright_samples_available;
     unsigned int left_samples_available;
+    uint8_t (*top_border)[16+2*8];
+    uint8_t left_border[17+2*9];
 
     /**
      * non zero coeff count cache.
@@ -198,6 +198,9 @@ typedef struct H264Context{
     int halfpel_flag;
     int thirdpel_flag;
 
+    int unknown_svq3_flag;
+    int next_slice_index;
+
     SPS sps_buffer[MAX_SPS_COUNT];
     SPS sps; ///< current sps
     
@@ -249,9 +252,9 @@ typedef struct H264Context{
     int chroma_offset[2][16][2];
    
     //deblock
-    int disable_deblocking_filter_idc;
-    int slice_alpha_c0_offset_div2;
-    int slice_beta_offset_div2;
+    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
+    int slice_alpha_c0_offset;
+    int slice_beta_offset;
      
     int redundant_pic_count;
     
@@ -283,6 +286,22 @@ typedef struct H264Context{
     GetBitContext *inter_gb_ptr;
     
     DCTELEM mb[16*24] __align8;
+
+    /**
+     * Cabac
+     */
+    CABACContext cabac;
+    uint8_t      cabac_state[399];
+    int          cabac_init_idc;
+
+    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
+    uint16_t     *cbp_table;
+    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
+    uint8_t     *chroma_pred_mode_table;
+    int         last_qscale_diff;
+    int16_t     (*mvd_table[2])[2];
+    int16_t     mvd_cache[2][5*8][2];
+
 }H264Context;
 
 static VLC coeff_token_vlc[4];
@@ -296,6 +315,15 @@ static VLC run7_vlc;
 
 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 void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
+
+static inline uint32_t pack16to32(int a, int b){
+#ifdef WORDS_BIGENDIAN
+   return (b&0xFFFF) + (a<<16);
+#else
+   return (a&0xFFFF) + (b<<16);
+#endif
+}
 
 /**
  * fill a rectangle.
@@ -319,6 +347,8 @@ static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t v
         *(uint16_t*)(p + 1*stride)=
         *(uint16_t*)(p + 2*stride)=
         *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
+    }else if(w==4 && h==1){
+        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
     }else if(w==4 && h==2){
         *(uint32_t*)(p + 0*stride)=
         *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
@@ -370,6 +400,9 @@ static inline void fill_caches(H264Context *h, int mb_type){
     
     if(h->sps.mb_aff){
     //FIXME
+        topleft_xy = 0; /* avoid warning */
+        top_xy = 0; /* avoid warning */
+        topright_xy = 0; /* avoid warning */
     }else{
         topleft_xy = mb_xy-1 - s->mb_stride;
         top_xy     = mb_xy   - s->mb_stride;
@@ -598,9 +631,52 @@ static inline void fill_caches(H264Context *h, int mb_type){
             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher 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(IS_INTER(topleft_type)){
+                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
+                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
+                }else{
+                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
+                }
+
+                if(IS_INTER(top_type)){
+                    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(IS_INTER(left_type[0])){
+                    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(IS_INTER(left_type[1])){
+                    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 somewher else)
+                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
+                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
+            }
         }
 //FIXME
-
     }
 #endif
 }
@@ -631,7 +707,7 @@ static inline int check_intra4x4_pred_mode(H264Context *h){
         for(i=0; i<4; i++){
             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
             if(status<0){
-                fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
+                av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
                 return -1;
             } else if(status){
                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
@@ -643,7 +719,7 @@ static inline int check_intra4x4_pred_mode(H264Context *h){
         for(i=0; i<4; i++){
             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
             if(status<0){
-                fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
+                av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
                 return -1;
             } else if(status){
                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
@@ -665,7 +741,7 @@ static inline int check_intra_pred_mode(H264Context *h, int mode){
     if(!(h->top_samples_available&0x8000)){
         mode= top[ mode ];
         if(mode<0){
-            fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
+            av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
             return -1;
         }
     }
@@ -673,7 +749,7 @@ static inline int check_intra_pred_mode(H264Context *h, int mode){
     if(!(h->left_samples_available&0x8000)){
         mode= left[ mode ];
         if(mode<0){
-            fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
+            av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
             return -1;
         } 
     }
@@ -776,7 +852,6 @@ static inline void pred_motion(H264Context * const h, int n, int part_width, int
 
     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
-    
     if(match_count > 1){ //most common
         *mx= mid_pred(A[0], B[0], C[0]);
         *my= mid_pred(A[1], B[1], C[1]);
@@ -909,6 +984,13 @@ static inline void write_back_motion(H264Context *h, int mb_type){
                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
                 }
+                if( h->pps.cabac ) {
+                    /* FIXME needed ? */
+                    for(y=0; y<4; y++){
+                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
+                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
+                    }
+                }
                 for(y=0; y<2; y++){
                     *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
                 }
@@ -920,6 +1002,12 @@ static inline void write_back_motion(H264Context *h, int mb_type){
             *(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 ) {
+            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];
+            }
+        }
         for(y=0; y<2; y++){
             s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
             s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
@@ -977,6 +1065,7 @@ static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *c
                 dst[di++]= 0;
                 dst[di++]= 0;
                 si+=3;
+                continue;
             }else //next start code
                 break;
         }
@@ -990,6 +1079,7 @@ static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *c
     return dst;
 }
 
+#if 0
 /**
  * @param src the data which should be escaped
  * @param dst the target buffer, dst+1 == src is allowed as a special case
@@ -1059,9 +1149,10 @@ static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, in
 static void encode_rbsp_trailing(PutBitContext *pb){
     int length;
     put_bits(pb, 1, 1);
-    length= (-get_bit_count(pb))&7;
+    length= (-put_bits_count(pb))&7;
     if(length) put_bits(pb, length, 0);
 }
+#endif
 
 /**
  * identifies the exact end of the bitstream
@@ -1121,6 +1212,7 @@ static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
     }
 }
 
+#if 0
 /**
  * dct tranforms the 16 dc values.
  * @param qp quantization parameter ??? FIXME
@@ -1158,6 +1250,8 @@ static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
         block[stride*10+offset]= (z0 - z3)>>1;
     }
 }
+#endif
+
 #undef xStride
 #undef stride
 
@@ -1183,6 +1277,7 @@ static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
     block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
 }
 
+#if 0
 static void chroma_dc_dct_c(DCTELEM *block){
     const int stride= 16*2;
     const int xStride= 16;
@@ -1203,6 +1298,7 @@ static void chroma_dc_dct_c(DCTELEM *block){
     block[stride*1 + xStride*0]= (a-c);
     block[stride*1 + xStride*1]= (e-b);
 }
+#endif
 
 /**
  * gets the chroma qp.
@@ -1221,31 +1317,7 @@ static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
     uint8_t *cm = cropTbl + MAX_NEG_CROP;
 
     block[0] += 32;
-#if 1
-    for(i=0; i<4; i++){
-        const int z0=  block[i + 4*0]     +  block[i + 4*2];
-        const int z1=  block[i + 4*0]     -  block[i + 4*2];
-        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
-        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
-
-        block[i + 4*0]= z0 + z3;
-        block[i + 4*1]= z1 + z2;
-        block[i + 4*2]= z1 - z2;
-        block[i + 4*3]= z0 - z3;
-    }
-
-    for(i=0; i<4; i++){
-        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
-        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
-        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
-        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
 
-        dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
-        dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
-        dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
-        dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
-    }
-#else
     for(i=0; i<4; i++){
         const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
         const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
@@ -1269,9 +1341,9 @@ static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
         dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
         dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
     }
-#endif
 }
 
+#if 0
 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
     int i;
     //FIXME try int temp instead of block
@@ -1304,6 +1376,7 @@ static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int st
         block[3*4 + i]=   z3 - 2*z2;
     }
 }
+#endif
 
 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
@@ -1474,7 +1547,7 @@ static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
     src[2+0*stride]=
     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
-};
+}
 
 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
     LOAD_TOP_EDGE    
@@ -1497,7 +1570,7 @@ static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
     src[3+2*stride]=
     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
-};
+}
 
 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
     const int lt= src[-1-1*stride];
@@ -1521,7 +1594,7 @@ static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride
     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
-};
+}
 
 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
     LOAD_TOP_EDGE    
@@ -1544,7 +1617,7 @@ static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride)
     src[3+1*stride]=
     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
-};
+}
 
 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
     LOAD_LEFT_EDGE    
@@ -1565,7 +1638,7 @@ static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride)
     src[2+2*stride]=
     src[2+3*stride]=
     src[3+3*stride]=l3;
-};
+}
     
 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
     const int lt= src[-1-1*stride];
@@ -1589,7 +1662,7 @@ static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int strid
     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
     src[0+3*stride]=(l2 + l3 + 1)>>1;
     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
-};
+}
 
 static void pred16x16_vertical_c(uint8_t *src, int stride){
     int i;
@@ -1699,6 +1772,9 @@ static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int
   if(svq3){
     H = ( 5*(H/4) ) / 16;
     V = ( 5*(V/4) ) / 16;
+
+    /* required for 100% accuracy */
+    i = H; H = V; V = i;
   }else{
     H = ( 5*H+32 ) >> 6;
     V = ( 5*V+32 ) >> 6;
@@ -1885,7 +1961,7 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
        || full_my < 0-extra_height 
        || full_mx + 16/*FIXME*/ > s->width + extra_width 
        || full_my + 16/*FIXME*/ > s->height + extra_height){
-        ff_emulated_edge_mc(s, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
+        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
             src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
         emu=1;
     }
@@ -1898,13 +1974,13 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
     if(s->flags&CODEC_FLAG_GRAY) return;
     
     if(emu){
-        ff_emulated_edge_mc(s, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
+        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
             src_cb= s->edge_emu_buffer;
     }
     chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
 
     if(emu){
-        ff_emulated_edge_mc(s, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
+        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
             src_cr= s->edge_emu_buffer;
     }
     chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
@@ -2090,22 +2166,17 @@ static void init_pred_ptrs(H264Context *h){
     h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
 }
 
-//FIXME factorize
-#define CHECKED_ALLOCZ(p, size)\
-{\
-    p= av_mallocz(size);\
-    if(p==NULL){\
-        perror("malloc");\
-        goto fail;\
-    }\
-}
-
 static void free_tables(H264Context *h){
     av_freep(&h->intra4x4_pred_mode);
+    av_freep(&h->chroma_pred_mode_table);
+    av_freep(&h->cbp_table);
+    av_freep(&h->mvd_table[0]);
+    av_freep(&h->mvd_table[1]);
     av_freep(&h->non_zero_count);
     av_freep(&h->slice_table_base);
+    av_freep(&h->top_border);
     h->slice_table= NULL;
-    
+
     av_freep(&h->mb2b_xy);
     av_freep(&h->mb2b8_xy);
 }
@@ -2120,8 +2191,17 @@ static int alloc_tables(H264Context *h){
     int x,y;
 
     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
+
     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
     CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
+    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
+
+    if( h->pps.cabac ) {
+        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
+        CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
+        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
+        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
+    }
 
     memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
     h->slice_table= h->slice_table_base + s->mb_stride + 1;
@@ -2154,6 +2234,7 @@ static void common_init(H264Context *h){
     
     init_pred_ptrs(h);
 
+    s->unrestricted_mv=1;
     s->decode=1; //FIXME
 }
 
@@ -2161,6 +2242,8 @@ static int decode_init(AVCodecContext *avctx){
     H264Context *h= avctx->priv_data;
     MpegEncContext * const s = &h->s;
 
+    MPV_decode_defaults(s);
+    
     s->avctx = avctx;
     common_init(h);
 
@@ -2168,7 +2251,6 @@ static int decode_init(AVCodecContext *avctx){
     s->workaround_bugs= avctx->workaround_bugs;
 
     // set defaults
-    s->progressive_sequence=1;
 //    s->decode_mb= ff_h263_decode_mb;
     s->low_delay= 1;
     avctx->pix_fmt= PIX_FMT_YUV420P;
@@ -2200,6 +2282,66 @@ static void frame_start(H264Context *h){
 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 }
 
+static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
+    MpegEncContext * const s = &h->s;
+    int i;
+    
+    src_y  -=   linesize;
+    src_cb -= uvlinesize;
+    src_cr -= uvlinesize;
+
+    h->left_border[0]= h->top_border[s->mb_x][15];
+    for(i=1; i<17; i++){
+        h->left_border[i]= src_y[15+i*  linesize];
+    }
+    
+    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
+    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
+
+    if(!(s->flags&CODEC_FLAG_GRAY)){
+        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
+        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
+        for(i=1; i<9; i++){
+            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
+            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
+        }
+        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
+        *(uint64_t*)(h->top_border[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){
+    MpegEncContext * const s = &h->s;
+    int temp8, i;
+    uint64_t temp64;
+
+    src_y  -=   linesize + 1;
+    src_cb -= uvlinesize + 1;
+    src_cr -= uvlinesize + 1;
+
+#define XCHG(a,b,t,xchg)\
+t= a;\
+if(xchg)\
+    a= b;\
+b= t;
+    
+    for(i=0; i<17; i++){
+        XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
+    }
+    
+    XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
+    XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
+
+    if(!(s->flags&CODEC_FLAG_GRAY)){
+        for(i=0; i<9; i++){
+            XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
+            XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
+        }
+        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
+        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
+    }
+}
+
 static void hl_decode_mb(H264Context *h){
     MpegEncContext * const s = &h->s;
     const int mb_x= s->mb_x;
@@ -2235,6 +2377,9 @@ static void hl_decode_mb(H264Context *h){
     }
 
     if(IS_INTRA(mb_type)){
+        if(h->deblocking_filter)
+            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
+
         if(!(s->flags&CODEC_FLAG_GRAY)){
             h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
             h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
@@ -2252,6 +2397,9 @@ static void hl_decode_mb(H264Context *h){
                     if(!topright_avail){
                         tr= ptr[3 - linesize]*0x01010101;
                         topright= (uint8_t*) &tr;
+                    }else if(i==5 && h->deblocking_filter){
+                        tr= *(uint32_t*)h->top_border[mb_x+1];
+                        topright= (uint8_t*) &tr;
                     }
 
                     h->pred4x4[ dir ](ptr, topright, linesize);
@@ -2270,6 +2418,8 @@ static void hl_decode_mb(H264Context *h){
             else
                 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
         }
+        if(h->deblocking_filter)
+            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
     }else if(s->codec_id == CODEC_ID_H264){
         hl_motion(h, dest_y, dest_cb, dest_cr,
                   s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
@@ -2326,10 +2476,10 @@ static void hl_decode_mb(H264Context *h){
             }
         }
     }
-}
-
-static void decode_mb_cabac(H264Context *h){
-//    MpegEncContext * const s = &h->s;
+    if(h->deblocking_filter) {
+        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
+        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
+    }
 }
 
 /**
@@ -2443,7 +2593,7 @@ static int decode_ref_pic_list_reordering(H264Context *h){
                 
                 
                 if(index >= h->ref_count[list]){
-                    fprintf(stderr, "reference count overflow\n");
+                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
                     return -1;
                 }
                 
@@ -2452,7 +2602,7 @@ static int decode_ref_pic_list_reordering(H264Context *h){
                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
 
                         if(abs_diff_pic_num >= h->max_pic_num){
-                            fprintf(stderr, "abs_diff_pic_num overflow\n");
+                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
                             return -1;
                         }
 
@@ -2474,7 +2624,7 @@ static int decode_ref_pic_list_reordering(H264Context *h){
                     }
 
                     if(i < index){
-                        fprintf(stderr, "reference picture missing during reorder\n");
+                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
                     }else if(i > index){
                         Picture tmp= h->ref_list[list][i];
@@ -2486,7 +2636,7 @@ static int decode_ref_pic_list_reordering(H264Context *h){
                 }else if(reordering_of_pic_nums_idc==3) 
                     break;
                 else{
-                    fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
+                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
                     return -1;
                 }
             }
@@ -2556,12 +2706,12 @@ static Picture * remove_short(H264Context *h, int frame_num){
     int i;
     
     if(s->avctx->debug&FF_DEBUG_MMCO)
-        printf("remove short %d count %d\n", frame_num, h->short_ref_count);
+        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
     
     for(i=0; i<h->short_ref_count; i++){
         Picture *pic= h->short_ref[i];
         if(s->avctx->debug&FF_DEBUG_MMCO)
-            printf("%d %d %X\n", i, pic->frame_num, (int)pic);
+            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
         if(pic->frame_num == frame_num){
             h->short_ref[i]= NULL;
             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
@@ -2600,11 +2750,11 @@ static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
     Picture *pic;
     
     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
-        printf("no mmco here\n");
+        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
         
     for(i=0; i<mmco_count; i++){
         if(s->avctx->debug&FF_DEBUG_MMCO)
-            printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
+            av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
 
         switch(mmco[i].opcode){
         case MMCO_SHORT2UNUSED:
@@ -2662,7 +2812,7 @@ static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
         pic= remove_short(h, s->current_picture_ptr->frame_num);
         if(pic){
             pic->reference=0;
-            fprintf(stderr, "illegal short term buffer state detected\n");
+            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
         }
         
         if(h->short_ref_count)
@@ -2705,13 +2855,13 @@ static int decode_ref_pic_marking(H264Context *h){
                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
                     h->mmco[i].long_index= get_ue_golomb(&s->gb);
                     if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){
-                        fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
+                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
                         return -1;
                     }
                 }
                     
                 if(opcode > MMCO_LONG){
-                    fprintf(stderr, "illegal memory management control operation %d\n", opcode);
+                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
                     return -1;
                 }
             }
@@ -2824,7 +2974,6 @@ static int decode_slice_header(H264Context *h){
     int first_mb_in_slice, pps_id;
     int num_ref_idx_active_override_flag;
     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
-    float new_aspect;
 
     s->current_picture.reference= h->nal_ref_idc != 0;
 
@@ -2832,7 +2981,7 @@ static int decode_slice_header(H264Context *h){
 
     h->slice_type= get_ue_golomb(&s->gb);
     if(h->slice_type > 9){
-        fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
+        av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
     }
     if(h->slice_type > 4){
         h->slice_type -= 5;
@@ -2846,18 +2995,18 @@ static int decode_slice_header(H264Context *h){
         
     pps_id= get_ue_golomb(&s->gb);
     if(pps_id>255){
-        fprintf(stderr, "pps_id out of range\n");
+        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
         return -1;
     }
     h->pps= h->pps_buffer[pps_id];
     if(h->pps.slice_group_count == 0){
-        fprintf(stderr, "non existing PPS referenced\n");
+        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
         return -1;
     }
 
     h->sps= h->sps_buffer[ h->pps.sps_id ];
     if(h->sps.log2_max_frame_num == 0){
-        fprintf(stderr, "non existing SPS referenced\n");
+        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
         return -1;
     }
     
@@ -2870,24 +3019,14 @@ static int decode_slice_header(H264Context *h){
     s->mb_x = first_mb_in_slice % s->mb_width;
     s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
     
-    s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
+    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
     if(h->sps.frame_mbs_only_flag)
-        s->height= 16*s->mb_height - 2*(h->pps.crop_top  + h->pps.crop_bottom);
+        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
     else
-        s->height= 16*s->mb_height - 4*(h->pps.crop_top  + h->pps.crop_bottom); //FIXME recheck
+        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
     
-    if(h->pps.crop_left || h->pps.crop_top){
-        fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
-    }
-
-    if(s->aspected_height) //FIXME emms at end of slice ?
-        new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
-    else
-        new_aspect=0;
-
     if (s->context_initialized 
-        && (   s->width != s->avctx->width || s->height != s->avctx->height 
-            || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
+        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
         free_tables(h);
         MPV_common_end(s);
     }
@@ -2899,7 +3038,7 @@ static int decode_slice_header(H264Context *h){
 
         s->avctx->width = s->width;
         s->avctx->height = s->height;
-        s->avctx->aspect_ratio= new_aspect;
+        s->avctx->sample_aspect_ratio= h->sps.sar;
     }
 
     if(first_mb_in_slice == 0){
@@ -2927,7 +3066,7 @@ static int decode_slice_header(H264Context *h){
     }
         
     if(h->nal_unit_type == NAL_IDR_SLICE){
-        int idr_pic_id= get_ue_golomb(&s->gb);
+        get_ue_golomb(&s->gb); /* idr_pic_id */
     }
    
     if(h->sps.poc_type==0){
@@ -2967,7 +3106,7 @@ static int decode_slice_header(H264Context *h){
                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
 
             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
-                fprintf(stderr, "reference overflow\n");
+                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
                 return -1;
             }
         }
@@ -2985,25 +3124,37 @@ static int decode_slice_header(H264Context *h){
     
     if(s->current_picture.reference)
         decode_ref_pic_marking(h);
-    //FIXME CABAC stuff
 
-    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
+    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
+        h->cabac_init_idc = get_ue_golomb(&s->gb);
+
+    h->last_qscale_diff = 0;
+    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
+    if(s->qscale<0 || s->qscale>51){
+        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
+        return -1;
+    }
     //FIXME qscale / qp ... stuff
     if(h->slice_type == SP_TYPE){
-        int sp_for_switch_flag= get_bits1(&s->gb);
+        get_bits1(&s->gb); /* sp_for_switch_flag */
     }
     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
-        int slice_qs_delta= get_se_golomb(&s->gb);
+        get_se_golomb(&s->gb); /* slice_qs_delta */
     }
 
+    h->deblocking_filter = 1;
+    h->slice_alpha_c0_offset = 0;
+    h->slice_beta_offset = 0;
     if( h->pps.deblocking_filter_parameters_present ) {
-        h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
-        if( h->disable_deblocking_filter_idc  !=  1 ) {
-            h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
-            h->slice_beta_offset_div2= get_se_golomb(&s->gb);
+        h->deblocking_filter= get_ue_golomb(&s->gb);
+        if(h->deblocking_filter < 2) 
+            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;
         }
-    }else
-        h->disable_deblocking_filter_idc= 0;
+    }
 
 #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)
@@ -3011,14 +3162,14 @@ static int decode_slice_header(H264Context *h){
 #endif
 
     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
-        printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
+        av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
                first_mb_in_slice, 
-               ff_get_pict_type_char(h->slice_type),
+               av_get_pict_type_char(h->slice_type),
                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->disable_deblocking_filter_idc
+               h->deblocking_filter
                );
     }
 
@@ -3113,7 +3264,7 @@ static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, in
             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
             if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
         }else{
-            fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
+            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
             return -1;
         }
 
@@ -3128,7 +3279,7 @@ static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, in
         if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
 #else        
         if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
-        ? == prefix > 2 or sth
+        /* ? == prefix > 2 or sth */
 #endif
         tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
     }
@@ -3154,7 +3305,7 @@ static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, in
     }
 
     if(zeros_left<0){
-        fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
+        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;
     }
     
@@ -3192,15 +3343,16 @@ static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, in
  * decodes a macroblock
  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
  */
-static int decode_mb(H264Context *h){
+static int decode_mb_cavlc(H264Context *h){
     MpegEncContext * const s = &h->s;
     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
     int mb_type, partition_count, cbp;
 
-    memset(h->mb, 0, sizeof(int16_t)*24*16); //FIXME avoid if allready clear (move after skip handlong?
+    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
 
     tprintf("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 != I_TYPE && h->slice_type != SI_TYPE){
         if(s->mb_skip_run==-1)
             s->mb_skip_run= get_ue_golomb(&s->gb);
@@ -3224,10 +3376,11 @@ static int decode_mb(H264Context *h){
             fill_caches(h, mb_type); //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, (mx&0xFFFF)+(my<<16), 4);
+            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; //FIXME SKIP type
+            s->current_picture.qscale_table[mb_xy]= s->qscale;
             h->slice_table[ mb_xy ]= h->slice_num;
 
             h->prev_mb_skiped= 1;
@@ -3263,7 +3416,7 @@ static int decode_mb(H264Context *h){
        assert(h->slice_type == I_TYPE);
 decode_intra_mb:
         if(mb_type > 25){
-            fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, ff_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
+            av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice to 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;
@@ -3308,7 +3461,9 @@ decode_intra_mb:
     
         skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
         
+        //FIXME deblock filter, non_zero_count_cache init ...
         memset(h->non_zero_count[mb_xy], 16, 16);
+        s->current_picture.qscale_table[mb_xy]= s->qscale;
         
         return 0;
     }
@@ -3359,7 +3514,7 @@ decode_intra_mb:
             for(i=0; i<4; i++){
                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
                 if(h->sub_mb_type[i] >=13){
-                    fprintf(stderr, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
+                    av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d 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;
@@ -3370,7 +3525,7 @@ decode_intra_mb:
             for(i=0; i<4; i++){
                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
                 if(h->sub_mb_type[i] >=4){
-                    fprintf(stderr, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
+                    av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d 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;
@@ -3454,7 +3609,7 @@ decode_intra_mb:
                     my += get_se_golomb(&s->gb);
                     tprintf("final mv:%d %d\n", mx, my);
 
-                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, (mx&0xFFFF) + (my<<16), 4);
+                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
                 }
             }
         }
@@ -3477,7 +3632,7 @@ decode_intra_mb:
                         my += get_se_golomb(&s->gb);
                         tprintf("final mv:%d %d\n", mx, my);
 
-                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (mx&0xFFFF) + (my<<16), 4);
+                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
                     }
                 }
             }
@@ -3501,7 +3656,7 @@ decode_intra_mb:
                         my += get_se_golomb(&s->gb);
                         tprintf("final mv:%d %d\n", mx, my);
 
-                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (mx&0xFFFF) + (my<<16), 4);
+                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
                     }
                 }
             }
@@ -3514,7 +3669,7 @@ decode_intra_mb:
     if(!IS_INTRA16x16(mb_type)){
         cbp= get_ue_golomb(&s->gb);
         if(cbp > 47){
-            fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
+            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
             return -1;
         }
         
@@ -3543,7 +3698,7 @@ decode_intra_mb:
         dquant= get_se_golomb(&s->gb);
 
         if( dquant > 25 || dquant < -26 ){
-            fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
+            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;
         }
         
@@ -3571,7 +3726,7 @@ decode_intra_mb:
                     }
                 }
             }else{
-                memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
+                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
             }
         }else{
             for(i8x8=0; i8x8<4; i8x8++){
@@ -3612,148 +3767,1546 @@ decode_intra_mb:
             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
         }
     }else{
-        memset(&h->non_zero_count_cache[8], 0, 8*5);
+        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);
 
     return 0;
 }
 
-static int decode_slice(H264Context *h){
+static int decode_cabac_mb_type( H264Context *h ) {
     MpegEncContext * const s = &h->s;
-    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
 
-    s->mb_skip_run= -1;
-    
-#if 1
-    for(;;){
-        int ret= decode_mb(h);
-            
-        hl_decode_mb(h);
-        
-        if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
-            s->mb_y++;
-            ret= decode_mb(h);
-            
-            hl_decode_mb(h);
-            s->mb_y--;
-        }
+    if( h->slice_type == I_TYPE ) {
+        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
+        int ctx = 0;
+        int mb_type;
 
-        if(ret<0){
-            fprintf(stderr, "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);
+        if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
+            ctx++;
+        if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
+            ctx++;
 
-            return -1;
-        }
-        
-        if(++s->mb_x >= s->mb_width){
-            s->mb_x=0;
-            ff_draw_horiz_band(s, 16*s->mb_y, 16);
-            if(++s->mb_y >= s->mb_height){
-                tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
+        if( get_cabac( &h->cabac, &h->cabac_state[3+ctx] ) == 0 )
+            return 0;   /* I4x4 */
 
-                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( get_cabac_terminate( &h->cabac ) )
+            return 25;  /* PCM */
 
-                    return 0;
-                }else{
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+        mb_type = 1;    /* I16x16 */
+        if( get_cabac( &h->cabac, &h->cabac_state[3+3] ) )
+            mb_type += 12;  /* cbp_luma != 0 */
 
-                    return -1;
-                }
-            }
+        if( get_cabac( &h->cabac, &h->cabac_state[3+4] ) ) {
+            if( get_cabac( &h->cabac, &h->cabac_state[3+5] ) )
+                mb_type += 4 * 2;   /* cbp_chroma == 2 */
+            else
+                mb_type += 4 * 1;   /* cbp_chroma == 1 */
         }
-        
-        if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
-            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);
-
-                return 0;
-            }else{
-                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( get_cabac( &h->cabac, &h->cabac_state[3+6] ) )
+            mb_type += 2;
+        if( get_cabac( &h->cabac, &h->cabac_state[3+7] ) )
+            mb_type += 1;
+        return mb_type;
+
+    } else if( h->slice_type == P_TYPE ) {
+        if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
+            /* P-type */
+            if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
+                if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
+                    return 0; /* P_L0_D16x16; */
+                else
+                    return 3; /* P_8x8; */
+            } else {
+                if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
+                    return 2; /* P_L0_D8x16; */
+                else
+                    return 1; /* P_L0_D16x8; */
             }
+        } else {
+            int mb_type;
+            /* I-type */
+            if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
+                return 5+0; /* I_4x4 */
+            if( get_cabac_terminate( &h->cabac ) )
+                return 5+25; /*I_PCM */
+            mb_type = 5+1;    /* I16x16 */
+            if( get_cabac( &h->cabac, &h->cabac_state[17+1] ) )
+                mb_type += 12;  /* cbp_luma != 0 */
+
+            if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) ) {
+                if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) )
+                    mb_type += 4 * 2;   /* cbp_chroma == 2 */
+                else
+                    mb_type += 4 * 1;   /* cbp_chroma == 1 */
+            }
+            if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
+                mb_type += 2;
+            if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
+                mb_type += 1;
+
+            return mb_type;
         }
+    } else {
+        /* TODO do others frames types */
+        return -1;
     }
-#endif
-#if 0
-    for(;s->mb_y < s->mb_height; s->mb_y++){
-        for(;s->mb_x < s->mb_width; s->mb_x++){
-            int ret= decode_mb(h);
-            
-            hl_decode_mb(h);
+}
 
-            if(ret<0){
-                fprintf(stderr, "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);
+static int decode_cabac_mb_skip( H264Context *h) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
+    const int mba_xy = mb_xy - 1;
+    const int mbb_xy = mb_xy - s->mb_stride;
+    int ctx = 0;
+
+    if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
+        ctx++;
+    if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
+        ctx++;
+
+    if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
+        return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
+    else /* B-frame */
+        return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
+}
 
-                return -1;
-            }
-        
-            if(++s->mb_x >= s->mb_width){
-                s->mb_x=0;
-                if(++s->mb_y >= s->mb_height){
-                    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);
+static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
+    int mode = 0;
 
-                        return 0;
-                    }else{
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
+        return pred_mode;
 
-                        return -1;
-                    }
-                }
-            }
-        
-            if(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( get_cabac( &h->cabac, &h->cabac_state[69] ) )
+        mode += 1;
+    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
+        mode += 2;
+    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
+        mode += 4;
+    if( mode >= pred_mode )
+        return mode + 1;
+    else
+        return mode;
+}
 
-                    return 0;
-                }else{
-                    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);
+static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
+    const int mba_xy = mb_xy - 1;
+    const int mbb_xy = mb_xy - s->mb_stride;
 
-                    return -1;
-                }
-            }
+    int ctx = 0;
+
+    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
+    if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
+        ctx++;
+
+    if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
+        ctx++;
+
+    if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
+        return 0;
+
+    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
+        return 1;
+    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
+        return 2;
+    else
+        return 3;
+}
+
+static const uint8_t block_idx_x[16] = {
+    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
+};
+static const uint8_t block_idx_y[16] = {
+    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
+};
+static const uint8_t block_idx_xy[4][4] = {
+    { 0, 2, 8,  10},
+    { 1, 3, 9,  11},
+    { 4, 6, 12, 14},
+    { 5, 7, 13, 15}
+};
+
+static int decode_cabac_mb_cbp_luma( H264Context *h) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
+
+    int cbp = 0;
+    int i8x8;
+
+    h->cbp_table[mb_xy] = 0;  /* FIXME aaahahahah beurk */
+
+    for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
+        int mba_xy = -1;
+        int mbb_xy = -1;
+        int x, y;
+        int ctx = 0;
+
+        x = block_idx_x[4*i8x8];
+        y = block_idx_y[4*i8x8];
+
+        if( x > 0 )
+            mba_xy = mb_xy;
+        else if( s->mb_x > 0 )
+            mba_xy = mb_xy - 1;
+
+        if( y > 0 )
+            mbb_xy = mb_xy;
+        else if( s->mb_y > 0 )
+            mbb_xy = mb_xy - s->mb_stride;
+
+        /* No need to test for skip as we put 0 for skip block */
+        if( mba_xy >= 0 ) {
+            int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
+            if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
+                ctx++;
+        }
+
+        if( mbb_xy >= 0 ) {
+            int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
+            if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
+                ctx += 2;
+        }
+
+        if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
+            cbp |= 1 << i8x8;
+            h->cbp_table[mb_xy] = cbp;  /* FIXME aaahahahah beurk */
         }
-        s->mb_x=0;
-        ff_draw_horiz_band(s, 16*s->mb_y, 16);
     }
-#endif
-    return -1; //not reached
+    return cbp;
 }
+static int decode_cabac_mb_cbp_chroma( H264Context *h) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
+    int ctx;
+    int cbp_a, cbp_b;
 
-static inline int decode_vui_parameters(H264Context *h, SPS *sps){
+    /* No need to test for skip */
+    if( s->mb_x > 0 )
+        cbp_a = (h->cbp_table[mb_xy-1]>>4)&0x03;
+    else
+        cbp_a = -1;
+
+    if( s->mb_y > 0 )
+        cbp_b = (h->cbp_table[mb_xy-s->mb_stride]>>4)&0x03;
+    else
+        cbp_b = -1;
+
+    ctx = 0;
+    if( cbp_a > 0 ) ctx++;
+    if( cbp_b > 0 ) ctx += 2;
+    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
+        return 0;
+
+    ctx = 4;
+    if( cbp_a == 2 ) ctx++;
+    if( cbp_b == 2 ) ctx += 2;
+    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) )
+        return 2;
+    else
+        return 1;
+}
+static int decode_cabac_mb_dqp( H264Context *h) {
     MpegEncContext * const s = &h->s;
-    int aspect_ratio_info_present_flag, aspect_ratio_idc;
+    int mbn_xy;
+    int   ctx = 0;
+    int   val = 0;
 
-    aspect_ratio_info_present_flag= get_bits1(&s->gb);
-    
-    if( aspect_ratio_info_present_flag ) {
-        aspect_ratio_idc= get_bits(&s->gb, 8);
-        if( aspect_ratio_idc == EXTENDED_SAR ) {
-            sps->sar_width= get_bits(&s->gb, 16);
-            sps->sar_height= get_bits(&s->gb, 16);
-        }else if(aspect_ratio_idc < 16){
-            sps->sar_width=  pixel_aspect[aspect_ratio_idc][0];
-            sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
-        }else{
-            fprintf(stderr, "illegal aspect ratio\n");
-            return -1;
-        }
-    }else{
-        sps->sar_width= 
-        sps->sar_height= 0;
+    if( s->mb_x > 0 )
+        mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
+    else
+        mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
+
+    if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
+        ctx++;
+
+    while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
+        if( ctx < 2 )
+            ctx = 2;
+        else
+            ctx = 3;
+        val++;
     }
-//            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
-#if 0
-| overscan_info_present_flag                        |0  |u(1)    |
-| if( overscan_info_present_flag )                  |   |        |
-|  overscan_appropriate_flag                        |0  |u(1)    |
-| video_signal_type_present_flag                    |0  |u(1)    |
-| if( video_signal_type_present_flag ) {            |   |        |
+
+    if( val&0x01 )
+        return (val + 1)/2;
+    else
+        return -(val + 1)/2;
+}
+static int decode_cabac_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_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( refa > 0 )
+        ctx++;
+    if( refb > 0 )
+        ctx += 2;
+
+    while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
+        ref++;
+        if( ctx < 4 )
+            ctx = 4;
+        else
+            ctx = 5;
+    }
+    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 ctx;
+    int mvd = 0;
+
+    if( amvd < 3 )
+        ctx = 0;
+    else if( amvd > 32 )
+        ctx = 2;
+    else
+        ctx = 1;
+
+    while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
+        mvd++;
+        if( ctx < 3 )
+            ctx = 3;
+        else if( ctx < 6 )
+            ctx++;
+    }
+
+    if( mvd >= 9 ) {
+        int k = 3;
+        while( get_cabac_bypass( &h->cabac ) ) {
+            mvd += 1 << k;
+            k++;
+        }
+        while( k-- ) {
+            if( get_cabac_bypass( &h->cabac ) )
+                mvd += 1 << k;
+        }
+    }
+    if( mvd != 0 && get_cabac_bypass( &h->cabac ) )
+        return -mvd;
+    return mvd;
+}
+
+
+static int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy  = s->mb_x + s->mb_y*s->mb_stride;
+    int mba_xy = -1;
+    int mbb_xy = -1;
+
+    int nza = -1;
+    int nzb = -1;
+    int ctx = 0;
+
+    if( cat == 0 ) {
+        if( s->mb_x > 0 ) {
+            mba_xy = mb_xy - 1;
+            if( IS_INTRA16x16(s->current_picture.mb_type[mba_xy] ) )
+                    nza = h->cbp_table[mba_xy]&0x100;
+        }
+        if( s->mb_y > 0 ) {
+            mbb_xy = mb_xy - s->mb_stride;
+            if( IS_INTRA16x16(s->current_picture.mb_type[mbb_xy] ) )
+                    nzb = h->cbp_table[mbb_xy]&0x100;
+        }
+    } else if( cat == 1 || cat == 2 ) {
+        int i8x8a, i8x8b;
+        int x, y;
+
+        x = block_idx_x[idx];
+        y = block_idx_y[idx];
+
+        if( x > 0 )
+            mba_xy = mb_xy;
+        else if( s->mb_x > 0 )
+            mba_xy = mb_xy - 1;
+
+        if( y > 0 )
+            mbb_xy = mb_xy;
+        else if( s->mb_y > 0 )
+            mbb_xy = mb_xy - s->mb_stride;
+
+        /* No need to test for skip */
+        if( mba_xy >= 0 ) {
+            i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
+
+            if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
+                ((h->cbp_table[mba_xy]&0x0f)>>i8x8a))
+                nza = h->non_zero_count_cache[scan8[idx] - 1];
+        }
+
+        if( mbb_xy >= 0 ) {
+            i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
+
+            if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
+                ((h->cbp_table[mbb_xy]&0x0f)>>i8x8b))
+                nzb = h->non_zero_count_cache[scan8[idx] - 8];
+        }
+    } else if( cat == 3 ) {
+        if( s->mb_x > 0 ) {
+            mba_xy = mb_xy - 1;
+
+            if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
+                (h->cbp_table[mba_xy]&0x30) )
+                nza = (h->cbp_table[mba_xy]>>(6+idx))&0x01;
+        }
+        if( s->mb_y > 0 ) {
+            mbb_xy = mb_xy - s->mb_stride;
+
+            if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
+                (h->cbp_table[mbb_xy]&0x30) )
+                nzb = (h->cbp_table[mbb_xy]>>(6+idx))&0x01;
+        }
+    } else if( cat == 4 ) {
+        int idxc = idx % 4 ;
+        if( idxc == 1 || idxc == 3 )
+            mba_xy = mb_xy;
+        else if( s->mb_x > 0 )
+            mba_xy = mb_xy -1;
+
+        if( idxc == 2 || idxc == 3 )
+            mbb_xy = mb_xy;
+        else if( s->mb_y > 0 )
+            mbb_xy = mb_xy - s->mb_stride;
+
+        if( mba_xy >= 0 &&
+            !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
+            (h->cbp_table[mba_xy]&0x30) == 0x20 )
+            nza = h->non_zero_count_cache[scan8[16+idx] - 1];
+
+        if( mbb_xy >= 0 &&
+            !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
+            (h->cbp_table[mbb_xy]&0x30) == 0x20 )
+            nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
+    }
+
+    if( ( mba_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
+        ( mba_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) ) ||
+          nza > 0 )
+        ctx++;
+
+    if( ( mbb_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
+        ( mbb_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) ) ||
+          nzb > 0 )
+        ctx += 2;
+
+    return ctx + 4 * cat;
+}
+
+static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
+    const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
+    const uint16_t *qmul= dequant_coeff[qp];
+    static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
+    static const int last_significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
+    static const int coeff_abs_level_m1_offset[5] = { 0, 10, 20, 30, 39 };
+
+    int coeff[16];
+
+    int last = 0;
+    int coeff_count = 0;
+    int nz[16] = {0};
+    int i;
+
+    int abslevel1 = 0;
+    int abslevelgt1 = 0;
+
+    /* cat: 0-> DC 16x16  n = 0
+     *      1-> AC 16x16  n = luma4x4idx
+     *      2-> Luma4x4   n = luma4x4idx
+     *      3-> DC Chroma n = iCbCr
+     *      4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
+     */
+
+    /* read coded block flag */
+    if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
+        if( cat == 1 || cat == 2 )
+            h->non_zero_count_cache[scan8[n]] = 0;
+        else if( cat == 4 )
+            h->non_zero_count_cache[scan8[16+n]] = 0;
+
+        return 0;
+    }
+
+    while( last < max_coeff - 1 ) {
+        int ctx = FFMIN( last, max_coeff - 2 );
+
+        if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+ctx] ) == 0 ) {
+            nz[last++] = 0;
+        }
+        else {
+            nz[last++] = 1;
+            coeff_count++;
+            if( get_cabac( &h->cabac, &h->cabac_state[166+last_significant_coeff_flag_offset[cat]+ctx] ) ) {
+                while( last < max_coeff ) {
+                    nz[last++] = 0;
+                }
+                break;
+            }
+        }
+    }
+    if( last == max_coeff -1 ) {
+        nz[last++] = 1;
+        coeff_count++;
+    }
+
+    if( cat == 0 && coeff_count > 0 )
+        h->cbp_table[mb_xy] |= 0x100;
+    else if( cat == 1 || cat == 2 )
+        h->non_zero_count_cache[scan8[n]] = coeff_count;
+    else if( cat == 3 && coeff_count > 0 )
+        h->cbp_table[mb_xy] |= 0x40 << n;
+    else if( cat == 4 )
+        h->non_zero_count_cache[scan8[16+n]] = coeff_count;
+
+    for( i = coeff_count - 1; i >= 0; i-- ) {
+        int coeff_abs_m1;
+
+        int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 + 1 )) + coeff_abs_level_m1_offset[cat];
+
+        if( get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) == 0 ) {
+            coeff_abs_m1 = 0;
+        } else {
+            coeff_abs_m1 = 1;
+            ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
+            while( coeff_abs_m1 < 14 && get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) ) {
+                coeff_abs_m1++;
+            }
+        }
+
+        if( coeff_abs_m1 >= 14 ) {
+            int j = 0;
+            while( get_cabac_bypass( &h->cabac ) ) {
+                coeff_abs_m1 += 1 << j;
+                j++;
+            }
+
+            while( j-- ) {
+                if( get_cabac_bypass( &h->cabac ) )
+                    coeff_abs_m1 += 1 << j ;
+            }
+        }
+        if( get_cabac_bypass( &h->cabac ) )
+            coeff[i] = -1 *( coeff_abs_m1 + 1 );
+        else
+            coeff[i] = coeff_abs_m1 + 1;
+
+        if( coeff_abs_m1 == 0 )
+            abslevel1++;
+        else
+            abslevelgt1++;
+    }
+
+    if( cat == 0 || cat == 3 ) { /* DC */
+        int j;
+        for( i = 0, j = 0; j < coeff_count; i++ ) {
+            if( nz[i] ) {
+                block[scantable[i]] = coeff[j];
+
+                j++;
+            }
+        }
+
+    } else { /* AC */
+        int j;
+        for( i = 0, j = 0; j < coeff_count; i++ ) {
+            if( nz[i] ) {
+                block[scantable[i]] = coeff[j] * qmul[scantable[i]];
+
+                j++;
+            }
+        }
+    }
+    return 0;
+}
+
+/**
+ * decodes a macroblock
+ * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
+ */
+static int decode_mb_cabac(H264Context *h) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
+    int mb_type, partition_count, cbp = 0;
+
+    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
+
+    if( h->slice_type == B_TYPE ) {
+        av_log( h->s.avctx, AV_LOG_ERROR, "B-frame not supported with CABAC\n" );
+        return -1;
+    }
+    if( h->sps.mb_aff ) {
+        av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
+        return -1;
+    }
+
+    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
+        /* read skip flags */
+        if( decode_cabac_mb_skip( h ) ) {
+            int mx, my;
+
+            /* skip mb */
+            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
+
+            memset(h->non_zero_count[mb_xy], 0, 16);
+            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
+#if 0
+            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
+                h->mb_field_decoding_flag= get_bits1(&s->gb);
+            }
+            if(h->mb_field_decoding_flag)
+                mb_type|= MB_TYPE_INTERLACED;
+#endif
+
+            fill_caches(h, mb_type); //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->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
+            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; //FIXME SKIP type
+            s->current_picture.qscale_table[mb_xy]= s->qscale;
+            h->slice_table[ mb_xy ]= h->slice_num;
+            h->cbp_table[mb_xy] = 0;
+            h->chroma_pred_mode_table[mb_xy] = 0;
+            h->last_qscale_diff = 0;
+
+            h->prev_mb_skiped= 1;
+
+            return 0;
+
+        }
+    }
+    h->prev_mb_skiped = 0;
+
+    if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
+        av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
+        return -1;
+    }
+
+    if( h->slice_type == 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 == 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 0
+    if(h->mb_field_decoding_flag)
+        mb_type |= MB_TYPE_INTERLACED;
+#endif
+
+    s->current_picture.mb_type[mb_xy]= mb_type;
+    h->slice_table[ mb_xy ]= h->slice_num;
+
+    if(IS_INTRA_PCM(mb_type)) {
+        /* TODO */
+        h->cbp_table[mb_xy] = 0xf +4*2;
+        h->chroma_pred_mode_table[mb_xy] = 0;
+        s->current_picture.qscale_table[mb_xy]= s->qscale;
+        return -1;
+    }
+
+    fill_caches(h, mb_type);
+
+    if( IS_INTRA( mb_type ) ) {
+        if( IS_INTRA4x4( mb_type ) ) {
+            int i;
+            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] ] );
+            }
+            write_back_intra_pred_mode(h);
+            if( check_intra4x4_pred_mode(h) < 0 ) return -1;
+        } else {
+            h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
+            if( h->intra16x16_pred_mode < 0 ) return -1;
+        }
+        h->chroma_pred_mode_table[mb_xy] =
+            h->chroma_pred_mode          = decode_cabac_mb_chroma_pre_mode( h );
+
+        h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
+        if( h->chroma_pred_mode < 0 ) return -1;
+    } else if( partition_count == 4 ) {
+        int i, j, sub_partition_count[4], list, ref[2][4];
+
+        /* Only P-frame */
+        for( i = 0; i < 4; i++ ) {
+            h->sub_mb_type[i] = decode_cabac_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;
+        }
+
+        for( list = 0; list < 2; list++ ) {
+            if( h->ref_count[list] > 0 ) {
+                for( i = 0; i < 4; i++ ) {
+                    if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
+                        if( h->ref_count[list] > 1 )
+                            ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
+                        else
+                            ref[list][i] = 0;
+                    } else {
+                        ref[list][i] = -1;
+                    }
+                    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];
+                }
+            }
+        }
+
+        for(list=0; list<2; list++){
+
+            for(i=0; i<4; i++){
+                //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) && !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("final mv:%d %d\n", mx, my);
+
+                        if(IS_SUB_8X8(sub_mb_type)){
+                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
+                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
+                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
+                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
+
+                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
+                            mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
+                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
+                            mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
+                        }else if(IS_SUB_8X4(sub_mb_type)){
+                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
+                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
+
+                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
+                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
+                        }else if(IS_SUB_4X8(sub_mb_type)){
+                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
+                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
+
+                            mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
+                            mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
+                        }else{
+                            assert(IS_SUB_4X4(sub_mb_type));
+                            mv_cache[ 0 ][0]= mx;
+                            mv_cache[ 0 ][1]= my;
+
+                            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;
+                }
+            }
+        }
+    } else if( !IS_DIRECT(mb_type) ) {
+        int list, mx, my, i, mpx, mpy;
+        if(IS_16X16(mb_type)){
+            for(list=0; list<2; list++){
+                if(IS_DIR(mb_type, 0, list)){
+                    if(h->ref_count[list] > 0 ){
+                        const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
+                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
+                    }
+                }
+            }
+            for(list=0; list<2; 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("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 if(IS_16X8(mb_type)){
+            for(list=0; list<2; list++){
+                if(h->ref_count[list]>0){
+                    for(i=0; i<2; i++){
+                        if(IS_DIR(mb_type, i, list)){
+                            const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
+                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
+                        }
+                    }
+                }
+            }
+            for(list=0; list<2; 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("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{
+            assert(IS_8X16(mb_type));
+            for(list=0; list<2; list++){
+                if(h->ref_count[list]>0){
+                    for(i=0; i<2; i++){
+                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
+                            const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
+                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
+                        }
+                    }
+                }
+            }
+            for(list=0; list<2; 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("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);
+                    }
+                }
+            }
+        }
+    }
+
+   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 );
+        cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
+    }
+
+    h->cbp_table[mb_xy] = cbp;
+
+    if( cbp || IS_INTRA16x16( mb_type ) ) {
+        const uint8_t *scan, *dc_scan;
+        int dqp;
+
+        if(IS_INTERLACED(mb_type)){
+            scan= field_scan;
+            dc_scan= luma_dc_field_scan;
+        }else{
+            scan= zigzag_scan;
+            dc_scan= luma_dc_zigzag_scan;
+        }
+
+        h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
+        s->qscale += dqp;
+        if(((unsigned)s->qscale) > 51){
+            if(s->qscale<0) s->qscale+= 52;
+            else            s->qscale-= 52;
+        }
+        h->chroma_qp = get_chroma_qp(h, s->qscale);
+
+        if( IS_INTRA16x16( mb_type ) ) {
+            int i;
+            //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
+            if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
+                return -1;
+            if( cbp&15 ) {
+                for( i = 0; i < 16; i++ ) {
+                    //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
+                    if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
+                        return -1;
+                }
+            } 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) ) {
+                    for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
+                        const int index = 4*i8x8 + i4x4;
+                        //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
+                        if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, 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 ){
+            int c;
+            for( c = 0; c < 2; c++ ) {
+                //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
+                if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
+                    return -1;
+            }
+        }
+
+        if( cbp&0x20 ) {
+            int c, i;
+            for( c = 0; c < 2; 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 );
+                    if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 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 {
+        memset( &h->non_zero_count_cache[8], 0, 8*5 );
+    }
+
+    s->current_picture.qscale_table[mb_xy]= s->qscale;
+    write_back_non_zero_count(h);
+
+    return 0;
+}
+
+
+static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
+    int i, d;
+    const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
+    const int alpha = alpha_table[index_a];
+    const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
+
+    for( i = 0; i < 4; i++ ) {
+        if( bS[i] == 0 ) {
+            pix += 4 * stride;
+            continue;
+        }
+
+        if( bS[i] < 4 ) {
+            const int tc0 = tc0_table[index_a][bS[i] - 1];
+            /* 4px edge length */
+            for( d = 0; d < 4; d++ ) {
+                const int p0 = pix[-1];
+                const int p1 = pix[-2];
+                const int p2 = pix[-3];
+                const int q0 = pix[0];
+                const int q1 = pix[1];
+                const int q2 = pix[2];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+                    int tc = tc0;
+                    int i_delta;
+
+                    if( ABS( p2 - p0 ) < beta ) {
+                        pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
+                        tc++;
+                    }
+                    if( ABS( q2 - q0 ) < beta ) {
+                        pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
+                        tc++;
+                    }
+
+                    i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+                    pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */
+                    pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */
+                }
+                pix += stride;
+            }
+        }else{
+            /* 4px edge length */
+            for( d = 0; d < 4; d++ ) {
+                const int p0 = pix[-1];
+                const int p1 = pix[-2];
+                const int p2 = pix[-3];
+
+                const int q0 = pix[0];
+                const int q1 = pix[1];
+                const int q2 = pix[2];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+
+                    if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
+                        if( ABS( p2 - p0 ) < beta)
+                        {
+                            const int p3 = pix[-4];
+                            /* p0', p1', p2' */
+                            pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
+                            pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
+                            pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
+                        } else {
+                            /* p0' */
+                            pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
+                        }
+                        if( ABS( q2 - q0 ) < beta)
+                        {
+                            const int q3 = pix[3];
+                            /* q0', q1', q2' */
+                            pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
+                            pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
+                            pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
+                        } else {
+                            /* q0' */
+                            pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
+                        }
+                    }else{
+                        /* p0', q0' */
+                        pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
+                        pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
+                    }
+                }
+                pix += stride;
+            }
+        }
+    }
+}
+static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
+    int i, d;
+    const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
+    const int alpha = alpha_table[index_a];
+    const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
+
+    for( i = 0; i < 4; i++ ) {
+        if( bS[i] == 0 ) {
+            pix += 2 * stride;
+            continue;
+        }
+
+        if( bS[i] < 4 ) {
+            const int tc = tc0_table[index_a][bS[i] - 1] + 1;
+            /* 2px edge length (because we use same bS than the one for luma) */
+            for( d = 0; d < 2; d++ ){
+                const int p0 = pix[-1];
+                const int p1 = pix[-2];
+                const int q0 = pix[0];
+                const int q1 = pix[1];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+                    const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+
+                    pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */
+                    pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */
+                }
+                pix += stride;
+            }
+        }else{
+            /* 2px edge length (because we use same bS than the one for luma) */
+            for( d = 0; d < 2; d++ ){
+                const int p0 = pix[-1];
+                const int p1 = pix[-2];
+                const int q0 = pix[0];
+                const int q1 = pix[1];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+
+                    pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
+                    pix[0]  = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
+                }
+                pix += stride;
+            }
+        }
+    }
+}
+
+static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
+    int i, d;
+    const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
+    const int alpha = alpha_table[index_a];
+    const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
+    const int pix_next  = stride;
+
+    for( i = 0; i < 4; i++ ) {
+        if( bS[i] == 0 ) {
+            pix += 4;
+            continue;
+        }
+
+        if( bS[i] < 4 ) {
+            const int tc0 = tc0_table[index_a][bS[i] - 1];
+            /* 4px edge length */
+            for( d = 0; d < 4; d++ ) {
+                const int p0 = pix[-1*pix_next];
+                const int p1 = pix[-2*pix_next];
+                const int p2 = pix[-3*pix_next];
+                const int q0 = pix[0];
+                const int q1 = pix[1*pix_next];
+                const int q2 = pix[2*pix_next];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+
+                    int tc = tc0;
+                    int i_delta;
+
+                    if( ABS( p2 - p0 ) < beta ) {
+                        pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
+                        tc++;
+                    }
+                    if( ABS( q2 - q0 ) < beta ) {
+                        pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
+                        tc++;
+                    }
+
+                    i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+                    pix[-pix_next] = clip_uint8( p0 + i_delta );    /* p0' */
+                    pix[0]         = clip_uint8( q0 - i_delta );    /* q0' */
+                }
+                pix++;
+            }
+        }else{
+            /* 4px edge length */
+            for( d = 0; d < 4; d++ ) {
+                const int p0 = pix[-1*pix_next];
+                const int p1 = pix[-2*pix_next];
+                const int p2 = pix[-3*pix_next];
+                const int q0 = pix[0];
+                const int q1 = pix[1*pix_next];
+                const int q2 = pix[2*pix_next];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+
+                    const int p3 = pix[-4*pix_next];
+                    const int q3 = pix[ 3*pix_next];
+
+                    if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
+                        if( ABS( p2 - p0 ) < beta) {
+                            /* p0', p1', p2' */
+                            pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
+                            pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
+                            pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
+                        } else {
+                            /* p0' */
+                            pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
+                        }
+                        if( ABS( q2 - q0 ) < beta) {
+                            /* q0', q1', q2' */
+                            pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
+                            pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
+                            pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
+                        } else {
+                            /* q0' */
+                            pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
+                        }
+                    }else{
+                        /* p0', q0' */
+                        pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
+                        pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
+                    }
+                }
+                pix++;
+            }
+        }
+    }
+}
+
+static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
+    int i, d;
+    const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
+    const int alpha = alpha_table[index_a];
+    const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
+    const int pix_next  = stride;
+
+    for( i = 0; i < 4; i++ )
+    {
+        if( bS[i] == 0 ) {
+            pix += 2;
+            continue;
+        }
+
+        if( bS[i] < 4 ) {
+            int tc = tc0_table[index_a][bS[i] - 1] + 1;
+            /* 2px edge length (see deblocking_filter_edgecv) */
+            for( d = 0; d < 2; d++ ) {
+                const int p0 = pix[-1*pix_next];
+                const int p1 = pix[-2*pix_next];
+                const int q0 = pix[0];
+                const int q1 = pix[1*pix_next];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+
+                    int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+
+                    pix[-pix_next] = clip_uint8( p0 + i_delta );    /* p0' */
+                    pix[0]         = clip_uint8( q0 - i_delta );    /* q0' */
+                }
+                pix++;
+            }
+        }else{
+            /* 2px edge length (see deblocking_filter_edgecv) */
+            for( d = 0; d < 2; d++ ) {
+                const int p0 = pix[-1*pix_next];
+                const int p1 = pix[-2*pix_next];
+                const int q0 = pix[0];
+                const int q1 = pix[1*pix_next];
+
+                if( ABS( p0 - q0 ) < alpha &&
+                    ABS( p1 - p0 ) < beta &&
+                    ABS( q1 - q0 ) < beta ) {
+
+                    pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
+                    pix[0]         = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
+                }
+                pix++;
+            }
+        }
+    }
+}
+
+static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
+    MpegEncContext * const s = &h->s;
+    const int mb_xy= mb_x + mb_y*s->mb_stride;
+    int linesize, uvlinesize;
+    int dir;
+
+    /* FIXME Implement deblocking filter for field MB */
+    if( h->sps.mb_aff ) {
+        return;
+    }
+    linesize = s->linesize;
+    uvlinesize = s->uvlinesize;
+
+    /* dir : 0 -> vertical edge, 1 -> horizontal edge */
+    for( dir = 0; dir < 2; dir++ )
+    {
+        int start = 0;
+        int edge;
+
+        /* test picture boundary */
+        if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
+            start = 1;
+        }
+        /* FIXME test slice boundary */
+        if( h->deblocking_filter == 2 ) {
+        }
+
+        /* Calculate bS */
+        for( edge = start; edge < 4; edge++ ) {
+            /* mbn_xy: neighbour macroblock (how that works for field ?) */
+            int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
+            int bS[4];
+            int qp;
+
+            if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
+                IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
+                bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
+            } else {
+                int i;
+                for( i = 0; i < 4; i++ ) {
+                    int x = dir == 0 ? edge : i;
+                    int y = dir == 0 ? i    : edge;
+                    int b_idx= 8 + 4 + x + 8*y;
+                    int bn_idx= b_idx - (dir ? 8:1);
+
+                    if( h->non_zero_count_cache[b_idx] != 0 ||
+                        h->non_zero_count_cache[bn_idx] != 0 ) {
+                        bS[i] = 2;
+                    }
+                    else if( h->slice_type == P_TYPE ) {
+                        if( h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] ||
+                            ABS( h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] ) >= 4 ||
+                            ABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= 4 )
+                            bS[i] = 1;
+                        else
+                            bS[i] = 0;
+                    }
+                    else {
+                        /* FIXME Add support for B frame */
+                        return;
+                    }
+                }
+
+                if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
+                    continue;
+            }
+
+            /* Filter edge */
+            qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
+            if( dir == 0 ) {
+                filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
+                if( (edge&1) == 0 ) {
+                    int chroma_qp = ( h->chroma_qp +
+                                      get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
+                    filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
+                    filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
+                }
+            } else {
+                filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
+                if( (edge&1) == 0 ) {
+                    int chroma_qp = ( h->chroma_qp +
+                                      get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
+                    filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
+                    filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
+                }
+            }
+        }
+    }
+}
+
+static int decode_slice(H264Context *h){
+    MpegEncContext * const s = &h->s;
+    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
+
+    s->mb_skip_run= -1;
+
+    if( h->pps.cabac ) {
+        int i;
+
+        /* realign */
+        align_get_bits( &s->gb );
+
+        /* init cabac */
+        ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
+        ff_init_cabac_decoder( &h->cabac,
+                               s->gb.buffer + get_bits_count(&s->gb)/8,
+                               ( s->gb.size_in_bits - get_bits_count(&s->gb) ) );
+        /* calculate pre-state */
+        for( i= 0; i < 399; i++ ) {
+            int pre;
+            if( h->slice_type == I_TYPE )
+                pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
+            else
+                pre = 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;
+        }
+
+        for(;;){
+            int ret = decode_mb_cabac(h);
+            int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
+
+            hl_decode_mb(h);
+
+            /* XXX: useless as decode_mb_cabac it doesn't support that ... */
+            if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
+                s->mb_y++;
+
+                ret = decode_mb_cabac(h);
+                eos = get_cabac_terminate( &h->cabac );
+
+                hl_decode_mb(h);
+                s->mb_y--;
+            }
+
+            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;
+                ff_draw_horiz_band(s, 16*s->mb_y, 16);
+                if( ++s->mb_y >= s->mb_height ) {
+                    tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
+                }
+            }
+
+            if( eos || s->mb_y >= s->mb_height ) {
+                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);
+                return 0;
+            }
+#if 0
+            /* TODO test over-reading in cabac code */
+            else if( read too much in h->cabac ) {
+                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;
+            }
+#endif
+        }
+
+    } else {
+        for(;;){
+            int ret = decode_mb_cavlc(h);
+
+            hl_decode_mb(h);
+
+            if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
+                s->mb_y++;
+                ret = decode_mb_cavlc(h);
+
+                hl_decode_mb(h);
+                s->mb_y--;
+            }
+
+            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;
+                ff_draw_horiz_band(s, 16*s->mb_y, 16);
+                if(++s->mb_y >= s->mb_height){
+                    tprintf("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);
+
+                        return 0;
+                    }else{
+                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+
+                        return -1;
+                    }
+                }
+            }
+
+            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
+                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);
+
+                    return 0;
+                }else{
+                    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 0
+    for(;s->mb_y < s->mb_height; s->mb_y++){
+        for(;s->mb_x < s->mb_width; s->mb_x++){
+            int ret= decode_mb(h);
+            
+            hl_decode_mb(h);
+
+            if(ret<0){
+                fprintf(stderr, "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;
+                if(++s->mb_y >= s->mb_height){
+                    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);
+
+                        return 0;
+                    }else{
+                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+
+                        return -1;
+                    }
+                }
+            }
+        
+            if(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);
+
+                    return 0;
+                }else{
+                    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;
+                }
+            }
+        }
+        s->mb_x=0;
+        ff_draw_horiz_band(s, 16*s->mb_y, 16);
+    }
+#endif
+    return -1; //not reached
+}
+
+static inline int decode_vui_parameters(H264Context *h, SPS *sps){
+    MpegEncContext * const s = &h->s;
+    int aspect_ratio_info_present_flag, aspect_ratio_idc;
+
+    aspect_ratio_info_present_flag= get_bits1(&s->gb);
+    
+    if( aspect_ratio_info_present_flag ) {
+        aspect_ratio_idc= get_bits(&s->gb, 8);
+        if( aspect_ratio_idc == EXTENDED_SAR ) {
+            sps->sar.num= get_bits(&s->gb, 16);
+            sps->sar.den= get_bits(&s->gb, 16);
+        }else if(aspect_ratio_idc < 16){
+            sps->sar=  pixel_aspect[aspect_ratio_idc];
+        }else{
+            av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
+            return -1;
+        }
+    }else{
+        sps->sar.num= 
+        sps->sar.den= 0;
+    }
+//            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
+#if 0
+| overscan_info_present_flag                        |0  |u(1)    |
+| if( overscan_info_present_flag )                  |   |        |
+|  overscan_appropriate_flag                        |0  |u(1)    |
+| video_signal_type_present_flag                    |0  |u(1)    |
+| if( video_signal_type_present_flag ) {            |   |        |
 |  video_format                                     |0  |u(3)    |
 |  video_full_range_flag                            |0  |u(1)    |
 |  colour_description_present_flag                  |0  |u(1)    |
@@ -3801,28 +5354,23 @@ static inline int decode_vui_parameters(H264Context *h, SPS *sps){
 
 static inline int decode_seq_parameter_set(H264Context *h){
     MpegEncContext * const s = &h->s;
-    int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices;
+    int profile_idc, level_idc;
     int sps_id, i;
     SPS *sps;
     
     profile_idc= get_bits(&s->gb, 8);
+    get_bits1(&s->gb);   //constraint_set0_flag
+    get_bits1(&s->gb);   //constraint_set1_flag
+    get_bits1(&s->gb);   //constraint_set2_flag
+    get_bits(&s->gb, 5); // reserved
     level_idc= get_bits(&s->gb, 8);
-    multiple_slice_groups= get_bits1(&s->gb);
-    arbitrary_slice_order= get_bits1(&s->gb);
-    redundant_slices= get_bits1(&s->gb);
-    
     sps_id= get_ue_golomb(&s->gb);
     
     sps= &h->sps_buffer[ sps_id ];
-    
     sps->profile_idc= profile_idc;
     sps->level_idc= level_idc;
-    sps->multiple_slice_groups= multiple_slice_groups;
-    sps->arbitrary_slice_order= arbitrary_slice_order;
-    sps->redundant_slices= redundant_slices;
     
     sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
-
     sps->poc_type= get_ue_golomb(&s->gb);
     
     if(sps->poc_type == 0){ //FIXME #define
@@ -3837,12 +5385,12 @@ static inline int decode_seq_parameter_set(H264Context *h){
             sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
     }
     if(sps->poc_type > 2){
-        fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
+        av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
         return -1;
     }
 
     sps->ref_frame_count= get_ue_golomb(&s->gb);
-    sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb);
+    sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
     sps->mb_width= get_ue_golomb(&s->gb) + 1;
     sps->mb_height= get_ue_golomb(&s->gb) + 1;
     sps->frame_mbs_only_flag= get_bits1(&s->gb);
@@ -3853,18 +5401,36 @@ static inline int decode_seq_parameter_set(H264Context *h){
 
     sps->direct_8x8_inference_flag= get_bits1(&s->gb);
 
+    sps->crop= get_bits1(&s->gb);
+    if(sps->crop){
+        sps->crop_left  = get_ue_golomb(&s->gb);
+        sps->crop_right = get_ue_golomb(&s->gb);
+        sps->crop_top   = get_ue_golomb(&s->gb);
+        sps->crop_bottom= get_ue_golomb(&s->gb);
+        if(sps->crop_left || sps->crop_top){
+            av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
+        }
+    }else{
+        sps->crop_left  = 
+        sps->crop_right = 
+        sps->crop_top   = 
+        sps->crop_bottom= 0;
+    }
+
     sps->vui_parameters_present_flag= get_bits1(&s->gb);
     if( sps->vui_parameters_present_flag )
         decode_vui_parameters(h, sps);
     
     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
-        printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n", 
+        av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n", 
                sps_id, sps->profile_idc, sps->level_idc,
                sps->poc_type,
                sps->ref_frame_count,
                sps->mb_width, sps->mb_height,
                sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
                sps->direct_8x8_inference_flag ? "8B8" : "",
+               sps->crop_left, sps->crop_right, 
+               sps->crop_top, sps->crop_bottom, 
                sps->vui_parameters_present_flag ? "VUI" : ""
                );
     }
@@ -3882,7 +5448,7 @@ static inline int decode_picture_parameter_set(H264Context *h){
     pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
     if(pps->slice_group_count > 1 ){
         pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
-fprintf(stderr, "FMO not supported\n");
+        av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
         switch(pps->mb_slice_group_map_type){
         case 0:
 #if 0
@@ -3920,7 +5486,7 @@ fprintf(stderr, "FMO not supported\n");
     pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
     pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
     if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
-        fprintf(stderr, "reference overflow (pps)\n");
+        av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
         return -1;
     }
     
@@ -3932,21 +5498,9 @@ fprintf(stderr, "FMO not supported\n");
     pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
     pps->constrained_intra_pred= get_bits1(&s->gb);
     pps->redundant_pic_cnt_present = get_bits1(&s->gb);
-    pps->crop= get_bits1(&s->gb);
-    if(pps->crop){
-        pps->crop_left  = get_ue_golomb(&s->gb);
-        pps->crop_right = get_ue_golomb(&s->gb);
-        pps->crop_top   = get_ue_golomb(&s->gb);
-        pps->crop_bottom= get_ue_golomb(&s->gb);
-    }else{
-        pps->crop_left  = 
-        pps->crop_right = 
-        pps->crop_top   = 
-        pps->crop_bottom= 0;
-    }
     
     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
-        printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s crop:%d/%d/%d/%d\n", 
+        av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s\n", 
                pps_id, pps->sps_id,
                pps->cabac ? "CABAC" : "CAVLC",
                pps->slice_group_count,
@@ -3955,9 +5509,7 @@ fprintf(stderr, "FMO not supported\n");
                pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
                pps->deblocking_filter_parameters_present ? "LPAR" : "",
                pps->constrained_intra_pred ? "CONSTR" : "",
-               pps->redundant_pic_cnt_present ? "REDU" : "",
-               pps->crop_left, pps->crop_right, 
-               pps->crop_top, pps->crop_bottom
+               pps->redundant_pic_cnt_present ? "REDU" : ""
                );
     }
     
@@ -3968,8 +5520,7 @@ fprintf(stderr, "FMO not supported\n");
  * finds the end of the current frame in the bitstream.
  * @return the position of the first byte of the next frame, or -1
  */
-static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
-    ParseContext *pc= &s->parse_context;
+static int find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size){
     int i;
     uint32_t state;
 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
@@ -3992,12 +5543,33 @@ static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
     return END_NOT_FOUND;
 }
 
+static int h264_parse(AVCodecParserContext *s,
+                      AVCodecContext *avctx,
+                      uint8_t **poutbuf, int *poutbuf_size, 
+                      const uint8_t *buf, int buf_size)
+{
+    ParseContext *pc = s->priv_data;
+    int next;
+    
+    next= find_frame_end(pc, buf, buf_size);
+
+    if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
+        *poutbuf = NULL;
+        *poutbuf_size = 0;
+        return buf_size;
+    }
+
+    *poutbuf = (uint8_t *)buf;
+    *poutbuf_size = buf_size;
+    return next;
+}
+
 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
     MpegEncContext * const s = &h->s;
     AVCodecContext * const avctx= s->avctx;
     int buf_index=0;
+#if 0
     int i;
-#if 0    
     for(i=0; i<32; i++){
         printf("%X ", buf[i]);
     }
@@ -4024,12 +5596,12 @@ static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
         bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
 
         if(s->avctx->debug&FF_DEBUG_STARTCODE){
-            printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
+            av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
         }
         
         buf_index += consumed;
 
-        if(h->nal_ref_idc < s->hurry_up)
+        if( s->hurry_up == 1 && h->nal_ref_idc  == 0 )
             continue;
         
         switch(h->nal_unit_type){
@@ -4042,7 +5614,7 @@ static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
             s->data_partitioning = 0;
             
             if(decode_slice_header(h) < 0) return -1;
-            if(h->redundant_pic_count==0)
+            if(h->redundant_pic_count==0 && s->hurry_up < 5 )
                 decode_slice(h);
             break;
         case NAL_DPA:
@@ -4061,7 +5633,7 @@ static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
             init_get_bits(&h->inter_gb, ptr, bit_length);
             h->inter_gb_ptr= &h->inter_gb;
 
-            if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
+            if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
                 decode_slice(h);
             break;
         case NAL_SEI:
@@ -4106,6 +5678,7 @@ static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
         assert(h->mmco_index==0);
 
     ff_er_frame_end(s);
+
     MPV_frame_end(s);
 
     return buf_index;
@@ -4138,18 +5711,17 @@ static int decode_frame(AVCodecContext *avctx,
     int buf_index;
     
     s->flags= avctx->flags;
+    s->flags2= avctx->flags2;
 
-    *data_size = 0;
-   
    /* no supplementary picture */
     if (buf_size == 0) {
         return 0;
     }
     
     if(s->flags&CODEC_FLAG_TRUNCATED){
-        int next= find_frame_end(s, buf, buf_size);
+        int next= find_frame_end(&s->parse_context, buf, buf_size);
         
-        if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
+        if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
             return buf_size;
 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
     }
@@ -4174,12 +5746,12 @@ static int decode_frame(AVCodecContext *avctx,
     }
 #endif
     if(!s->current_picture_ptr){
-        fprintf(stderr, "error, NO frame\n");
+        av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
         return -1;
     }
 
     *pict= *(AVFrame*)&s->current_picture; //FIXME 
-    ff_print_debug_info(s, s->current_picture_ptr);
+    ff_print_debug_info(s, pict);
     assert(pict->data[0]);
 //printf("out %d\n", (int)pict->data[0]);
 #if 0 //?
@@ -4230,7 +5802,7 @@ int main(){
     
     dsputil_init(&dsp, &avctx);
 
-    init_put_bits(&pb, temp, SIZE, NULL, NULL);
+    init_put_bits(&pb, temp, SIZE);
     printf("testing unsigned exp golomb\n");
     for(i=0; i<COUNT; i++){
         START_TIMER
@@ -4255,7 +5827,7 @@ int main(){
     }
     
     
-    init_put_bits(&pb, temp, SIZE, NULL, NULL);
+    init_put_bits(&pb, temp, SIZE);
     printf("testing signed exp golomb\n");
     for(i=0; i<COUNT; i++){
         START_TIMER
@@ -4416,4 +5988,12 @@ AVCodec h264_decoder = {
     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
 };
 
+AVCodecParser h264_parser = {
+    { CODEC_ID_H264 },
+    sizeof(ParseContext),
+    NULL,
+    h264_parse,
+    ff_parse_close,
+};
+
 #include "svq3.c"