libavcodec.hg: h264.c comparison

comparison h264.c @ 3316:7278f730af27 libavcodec

h264: MBAFF interlaced decoding

author	lorenm
date	Sun, 28 May 2006 22:44:14 +0000
parents	cfd452a6560b
children	8dc886b9568a

comparison

equal deleted inserted replaced

-:cfd452a6560b
+:7278f730af27
 #define MAX_SPS_COUNT 32
 #define MAX_PPS_COUNT 256
 #define MAX_MMCO_COUNT 66
+/* Compiling in interlaced support reduces the speed
+* of progressive decoding by about 2%. */
+#define ALLOW_INTERLACE
+#ifdef ALLOW_INTERLACE
+#define MB_MBAFF h->mb_mbaff
+#define MB_FIELD h->mb_field_decoding_flag
+#define FRAME_MBAFF h->mb_aff_frame
+#else
+#define MB_MBAFF 0
+#define MB_FIELD 0
+#define FRAME_MBAFF 0
+#undef  IS_INTERLACED
+#define IS_INTERLACED(mb_type) 0
+#endif
 /**
 * Sequence parameter set
 */
 typedef struct SPS{
 int got_avcC; ///< flag used to parse avcC data only once
 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
 int chroma_qp; //QPc
-int prev_mb_skipped; //FIXME remove (IMHO not used)
+int prev_mb_skipped;
+int next_mb_skipped;
 //prediction stuff
 int chroma_pred_mode;
 int intra16x16_pred_mode;
 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
 uint32_t *mb2b8_xy;
 int b_stride; //FIXME use s->b4_stride
 int b8_stride;
+int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
+int mb_uvlinesize;
+int emu_edge_width;
+int emu_edge_height;
 int halfpel_flag;
 int thirdpel_flag;
 int unknown_svq3_flag;
 int next_slice_index;
 uint32_t (*dequant8_coeff[2])[64];
 int dequant_coeff_pps;     ///< reinit tables when pps changes
 int slice_num;
 uint8_t *slice_table_base;
-uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
+uint8_t *slice_table;      ///< slice_table_base + 2*mb_stride + 1
 int slice_type;
 int slice_type_fixed;
 //interlacing specific flags
 int mb_aff_frame;
 int mb_field_decoding_flag;
+int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
 int sub_mb_type[4];
 //POC stuff
 int poc_lsb;
 //Weighted pred stuff
 int use_weight;
 int use_weight_chroma;
 int luma_log2_weight_denom;
 int chroma_log2_weight_denom;
-int luma_weight[2][16];
+int luma_weight[2][48];
-int luma_offset[2][16];
+int luma_offset[2][48];
-int chroma_weight[2][16][2];
+int chroma_weight[2][48][2];
-int chroma_offset[2][16][2];
+int chroma_offset[2][48][2];
-int implicit_weight[16][16];
+int implicit_weight[48][48];
 //deblock
 int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
 int slice_alpha_c0_offset;
 int slice_beta_offset;
 int redundant_pic_count;
 int direct_spatial_mv_pred;
 int dist_scale_factor[16];
+int dist_scale_factor_field[32];
 int map_col_to_list0[2][16];
+int map_col_to_list0_field[2][32];
 /**
 * num_ref_idx_l0/1_active_minus1 + 1
 */
-int ref_count[2];// FIXME split for AFF
+int ref_count[2];            ///< counts frames or fields, depending on current mb mode
 Picture *short_ref[32];
 Picture *long_ref[32];
 Picture default_ref_list[2][32];
-Picture ref_list[2][32]; //FIXME size?
+Picture ref_list[2][48];     ///< 0..15: frame refs, 16..47: mbaff field refs
-Picture field_ref_list[2][32]; //FIXME size?
 Picture *delayed_pic[16]; //FIXME size?
 Picture *delayed_output_pic;
 /**
 * memory management control operations buffer.
 DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
 uint8_t     *direct_table;
 uint8_t     direct_cache[5*8];
 uint8_t zigzag_scan[16];
-uint8_t field_scan[16];
 uint8_t zigzag_scan8x8[64];
 uint8_t zigzag_scan8x8_cavlc[64];
+uint8_t field_scan[16];
+uint8_t field_scan8x8[64];
+uint8_t field_scan8x8_cavlc[64];
 const uint8_t *zigzag_scan_q0;
-const uint8_t *field_scan_q0;
 const uint8_t *zigzag_scan8x8_q0;
 const uint8_t *zigzag_scan8x8_cavlc_q0;
+const uint8_t *field_scan_q0;
+const uint8_t *field_scan8x8_q0;
+const uint8_t *field_scan8x8_cavlc_q0;
 int x264_build;
 }H264Context;
 static VLC coeff_token_vlc[4];
 int i;
 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
 // the actual condition is whether we're on the edge of a slice,
 // and even then the intra and nnz parts are unnecessary.
-if(for_deblock && h->slice_num == 1)
+if(for_deblock && h->slice_num == 1 && !FRAME_MBAFF)
 return;
 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
 top_xy     = mb_xy  - s->mb_stride;
 left_block[3]= 3;
 left_block[4]= 7;
 left_block[5]= 10;
 left_block[6]= 8;
 left_block[7]= 11;
-if(h->mb_aff_frame){
+if(FRAME_MBAFF){
 const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
 const int top_pair_xy      = pair_xy     - s->mb_stride;
 const int topleft_pair_xy  = top_pair_xy - 1;
 const int topright_pair_xy = top_pair_xy + 1;
 const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
 top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
+if(FRAME_MBAFF && !IS_INTRA(mb_type)){
+int list;
+int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
+for(i=0; i<16; i++)
+h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
+for(list=0; list<1+(h->slice_type==B_TYPE); list++){
+if(USES_LIST(mb_type,list)){
+uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
+uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
+uint8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
+for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
+dst[0] = src[0];
+dst[1] = src[1];
+dst[2] = src[2];
+dst[3] = src[3];
+}
+*(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
+*(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
+ref += h->b8_stride;
+*(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
+*(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
+}else{
+fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
+fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
+}
+}
+}
 }else{
 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
 top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
 if(USES_LIST(left_type[0], list)){
 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
-h->ref_cache[list][scan8[0] - 1 + 0*8]=
+h->ref_cache[list][scan8[0] - 1 + 0*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
-h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
+h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1]>>1)];
 }else{
 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
 h->ref_cache[list][scan8[0] - 1 + 0*8]=
 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 if(USES_LIST(left_type[1], list)){
 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
-h->ref_cache[list][scan8[0] - 1 + 2*8]=
+h->ref_cache[list][scan8[0] - 1 + 2*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
-h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
+h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[3]>>1)];
 }else{
 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
 h->ref_cache[list][scan8[0] - 1 + 2*8]=
 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
 if( h->pps.cabac ) {
 /* XXX beurk, Load mvd */
-if(USES_LIST(topleft_type, list)){
-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(USES_LIST(top_type, list)){
 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
 }else{
 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
 }
-//FIXME interlacing
+if(IS_DIRECT(left_type[0]))
-if(IS_DIRECT(left_type[0])){
+h->direct_cache[scan8[0] - 1 + 0*8]= 1;
-h->direct_cache[scan8[0] - 1 + 0*8]=
+else if(IS_8X8(left_type[0]))
+h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
+else
+h->direct_cache[scan8[0] - 1 + 0*8]= 0;
+if(IS_DIRECT(left_type[1]))
 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
-}else if(IS_8X8(left_type[0])){
+else if(IS_8X8(left_type[1]))
-int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
+h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
-h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
+else
-h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
-}else{
-h->direct_cache[scan8[0] - 1 + 0*8]=
 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
+}
+}
+if(FRAME_MBAFF){
+#define MAP_MVS\
+MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
+MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
+MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
+MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
+MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
+MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
+MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
+MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
+MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
+MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
+if(MB_FIELD){
+#define MAP_F2F(idx, mb_type)\
+if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
+h->ref_cache[list][idx] <<= 1;\
+h->mv_cache[list][idx][1] /= 2;\
+h->mvd_cache[list][idx][1] /= 2;\
 }
+MAP_MVS
+#undef MAP_F2F
+}else{
+#define MAP_F2F(idx, mb_type)\
+if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
+h->ref_cache[list][idx] >>= 1;\
+h->mv_cache[list][idx][1] <<= 1;\
+h->mvd_cache[list][idx][1] <<= 1;\
+}
+MAP_MVS
+#undef MAP_F2F
 }
 }
 }
 }
 #endif
 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
+if(FRAME_MBAFF){
+// store all luma nnzs, for deblocking
+int v = 0, i;
+for(i=0; i<16; i++)
+v += (!!h->non_zero_count_cache[scan8[i]]) << i;
+*(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
+}
 }
 /**
 * gets the predicted number of non zero coefficients.
 * @param n block index
 return i&31;
 }
 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
+/* there is no consistent mapping of mvs to neighboring locations that will
+* make mbaff happy, so we can't move all this logic to fill_caches */
+if(FRAME_MBAFF){
+MpegEncContext *s = &h->s;
+const int *mb_types = s->current_picture_ptr->mb_type;
+const int16_t *mv;
+*(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
+*C = h->mv_cache[list][scan8[0]-2];
+if(!MB_FIELD
+&& (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
+int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
+if(IS_INTERLACED(mb_types[topright_xy])){
+#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
+const int x4 = X4, y4 = Y4;\
+const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
+if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
+return LIST_NOT_USED;\
+mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
+h->mv_cache[list][scan8[0]-2][0] = mv[0];\
+h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
+return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
+SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
+}
+}
+if(topright_ref == PART_NOT_AVAILABLE
+&& ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
+&& h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
+if(!MB_FIELD
+&& IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
+SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
+}
+if(MB_FIELD
+&& !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
+&& i >= scan8[0]+8){
+// leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
+SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
+}
+}
+#undef SET_DIAG_MV
+}
 if(topright_ref != PART_NOT_AVAILABLE){
 *C= h->mv_cache[list][ i - 8 + part_width ];
 return topright_ref;
 }else{
 int tb = clip(poc - poc0, -128, 127);
 int tx = (16384 + (ABS(td) >> 1)) / td;
 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
 }
 }
+if(FRAME_MBAFF){
+for(i=0; i<h->ref_count[0]; i++){
+h->dist_scale_factor_field[2*i] =
+h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
+}
+}
 }
 static inline void direct_ref_list_init(H264Context * const h){
 MpegEncContext * const s = &h->s;
 Picture * const ref1 = &h->ref_list[1][0];
 Picture * const cur = s->current_picture_ptr;
 h->map_col_to_list0[list][i] = j;
 break;
 }
 }
 }
+if(FRAME_MBAFF){
+for(list=0; list<2; list++){
+for(i=0; i<ref1->ref_count[list]; i++){
+j = h->map_col_to_list0[list][i];
+h->map_col_to_list0_field[list][2*i] = 2*j;
+h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
+}
+}
+}
 }
 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
 MpegEncContext * const s = &h->s;
 const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
 const int is_b8x8 = IS_8X8(*mb_type);
 int sub_mb_type;
 int i8, i4;
+#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
 /* FIXME save sub mb types from previous frames (or derive from MVs)
 * so we know exactly what block size to use */
 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
 *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
-}else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
+}else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
 *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
 }else{
 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
 *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
 }
 if(!is_b8x8)
 *mb_type |= MB_TYPE_DIRECT2;
+if(MB_FIELD)
+*mb_type |= MB_TYPE_INTERLACED;
 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
 if(h->direct_spatial_mv_pred){
 int ref[2];
 int mv[2][2];
 int list;
+/* FIXME interlacing + spatial direct uses wrong colocated block positions */
 /* ref = min(neighbors) */
 for(list=0; list<2; list++){
 int refa = h->ref_cache[list][scan8[0] - 1];
 int refb = h->ref_cache[list][scan8[0] - 8];
 }
 }
 }
 }
 }else{ /* direct temporal mv pred */
+const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
+const int *dist_scale_factor = h->dist_scale_factor;
+if(FRAME_MBAFF){
+if(IS_INTERLACED(*mb_type)){
+map_col_to_list0[0] = h->map_col_to_list0_field[0];
+map_col_to_list0[1] = h->map_col_to_list0_field[1];
+dist_scale_factor = h->dist_scale_factor_field;
+}
+if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
+/* FIXME assumes direct_8x8_inference == 1 */
+const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
+int mb_types_col[2];
+int y_shift;
+*mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
+| (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
+| (*mb_type & MB_TYPE_INTERLACED);
+sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
+if(IS_INTERLACED(*mb_type)){
+/* frame to field scaling */
+mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
+mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
+if(s->mb_y&1){
+l1ref0 -= 2*h->b8_stride;
+l1ref1 -= 2*h->b8_stride;
+l1mv0 -= 4*h->b_stride;
+l1mv1 -= 4*h->b_stride;
+}
+y_shift = 0;
+if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
+&& (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
+&& !is_b8x8)
+*mb_type |= MB_TYPE_16x8;
+else
+*mb_type |= MB_TYPE_8x8;
+}else{
+/* field to frame scaling */
+/* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
+* but in MBAFF, top and bottom POC are equal */
+int dy = (s->mb_y&1) ? 1 : 2;
+mb_types_col[0] =
+mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
+l1ref0 += dy*h->b8_stride;
+l1ref1 += dy*h->b8_stride;
+l1mv0 += 2*dy*h->b_stride;
+l1mv1 += 2*dy*h->b_stride;
+y_shift = 2;
+if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
+&& !is_b8x8)
+*mb_type |= MB_TYPE_16x16;
+else
+*mb_type |= MB_TYPE_8x8;
+}
+for(i8=0; i8<4; i8++){
+const int x8 = i8&1;
+const int y8 = i8>>1;
+int ref0, scale;
+const int16_t (*l1mv)[2]= l1mv0;
+if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
+continue;
+h->sub_mb_type[i8] = sub_mb_type;
+fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
+if(IS_INTRA(mb_types_col[y8])){
+fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
+fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
+fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
+continue;
+}
+ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
+if(ref0 >= 0)
+ref0 = map_col_to_list0[0][ref0*2>>y_shift];
+else{
+ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
+l1mv= l1mv1;
+}
+scale = dist_scale_factor[ref0];
+fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
+{
+const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
+int my_col = (mv_col[1]<<y_shift)/2;
+int mx = (scale * mv_col[0] + 128) >> 8;
+int my = (scale * my_col + 128) >> 8;
+fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
+fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
+}
+}
+return;
+}
+}
+/* one-to-one mv scaling */
 if(IS_16X16(*mb_type)){
 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
 if(IS_INTRA(mb_type_col)){
 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
 }else{
-const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
+const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
-: h->map_col_to_list0[1][l1ref1[0]];
+: map_col_to_list0[1][l1ref1[0]];
-const int dist_scale_factor = h->dist_scale_factor[ref0];
+const int scale = dist_scale_factor[ref0];
 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
 int mv_l0[2];
-mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
+mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
-mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
+mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
 }
 }else{
 for(i8=0; i8<4; i8++){
 const int x8 = i8&1;
 const int y8 = i8>>1;
-int ref0, dist_scale_factor;
+int ref0, scale;
 const int16_t (*l1mv)[2]= l1mv0;
 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
 continue;
 h->sub_mb_type[i8] = sub_mb_type;
+fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
 if(IS_INTRA(mb_type_col)){
 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
-fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
 continue;
 }
 ref0 = l1ref0[x8 + y8*h->b8_stride];
 if(ref0 >= 0)
-ref0 = h->map_col_to_list0[0][ref0];
+ref0 = map_col_to_list0[0][ref0];
 else{
-ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
+ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
 l1mv= l1mv1;
 }
-dist_scale_factor = h->dist_scale_factor[ref0];
+scale = dist_scale_factor[ref0];
 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
-fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
 if(IS_SUB_8X8(sub_mb_type)){
 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
-int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
+int mx = (scale * mv_col[0] + 128) >> 8;
-int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
+int my = (scale * mv_col[1] + 128) >> 8;
 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
 }else
 for(i4=0; i4<4; i4++){
 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
-mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
+mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
-mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
+mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
 }
 }
 }
 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 int src_x_offset, int src_y_offset,
 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
 MpegEncContext * const s = &h->s;
 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
-const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
+int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
 const int luma_xy= (mx&3) + ((my&3)<<2);
-uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
+uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
-uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
+uint8_t * src_cb, * src_cr;
-uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
+int extra_width= h->emu_edge_width;
-int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
+int extra_height= h->emu_edge_height;
-int extra_height= extra_width;
 int emu=0;
 const int full_mx= mx>>2;
 const int full_my= my>>2;
 const int pic_width  = 16*s->mb_width;
-const int pic_height = 16*s->mb_height;
+const int pic_height = 16*s->mb_height >> MB_MBAFF;
 if(!pic->data[0])
 return;
 if(mx&7) extra_width -= 3;
 if(   full_mx < 0-extra_width
 || full_my < 0-extra_height
 || full_mx + 16/*FIXME*/ > pic_width + extra_width
 || full_my + 16/*FIXME*/ > pic_height + extra_height){
-ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
-src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
+src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
 emu=1;
 }
-qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
+qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
 if(!square){
-qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
+qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 }
 if(s->flags&CODEC_FLAG_GRAY) return;
+if(MB_MBAFF){
+// chroma offset when predicting from a field of opposite parity
+my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
+emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
+}
+src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
+src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
 if(emu){
-ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 src_cb= s->edge_emu_buffer;
 }
-chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
+chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 if(emu){
-ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
 src_cr= s->edge_emu_buffer;
 }
-chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
+chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
 }
 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 int x_offset, int y_offset,
 int list0, int list1){
 MpegEncContext * const s = &h->s;
 qpel_mc_func *qpix_op=  qpix_put;
 h264_chroma_mc_func chroma_op= chroma_put;
-dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
+dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
-dest_cb +=   x_offset +   y_offset*s->uvlinesize;
+dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
-dest_cr +=   x_offset +   y_offset*s->uvlinesize;
+dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 x_offset += 8*s->mb_x;
-y_offset += 8*s->mb_y;
+y_offset += 8*(s->mb_y >> MB_MBAFF);
 if(list0){
 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
 dest_y, dest_cb, dest_cr, x_offset, y_offset,
 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
 int list0, int list1){
 MpegEncContext * const s = &h->s;
-dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
+dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
-dest_cb +=   x_offset +   y_offset*s->uvlinesize;
+dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
-dest_cr +=   x_offset +   y_offset*s->uvlinesize;
+dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
 x_offset += 8*s->mb_x;
-y_offset += 8*s->mb_y;
+y_offset += 8*(s->mb_y >> MB_MBAFF);
 if(list0 && list1){
 /* don't optimize for luma-only case, since B-frames usually
 * use implicit weights => chroma too. */
 uint8_t *tmp_cb = s->obmc_scratchpad;
-uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
+uint8_t *tmp_cr = s->obmc_scratchpad + 8;
-uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
+uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
 int refn0 = h->ref_cache[0][ scan8[n] ];
 int refn1 = h->ref_cache[1][ scan8[n] ];
 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
 dest_y, dest_cb, dest_cr,
 x_offset, y_offset, qpix_put, chroma_put);
 if(h->use_weight == 2){
 int weight0 = h->implicit_weight[refn0][refn1];
 int weight1 = 64 - weight0;
-luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0);
+luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
-chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
+chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
-chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
+chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
 }else{
-luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
+luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
-chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
+chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
-chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
+chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
 }
 }else{
 int list = list1 ? 1 : 0;
 Picture *ref= &h->ref_list[list][refn];
 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
 dest_y, dest_cb, dest_cr, x_offset, y_offset,
 qpix_put, chroma_put);
-luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
+luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
 h->luma_weight[list][refn], h->luma_offset[list][refn]);
 if(h->use_weight_chroma){
-chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
+chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
-chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
+chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
 }
 }
 }
 const int refn = h->ref_cache[list][scan8[0]];
 if(refn >= 0){
 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
 uint8_t **src= h->ref_list[list][refn].data;
-int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
+int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
 s->dsp.prefetch(src[0]+off, s->linesize, 4);
 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
 }
 }
 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
 &weight_op[1], &weight_avg[1],
 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 }else if(IS_8X16(mb_type)){
-mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
+mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 &weight_op[2], &weight_avg[2],
 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
-mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
+mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
 &weight_op[2], &weight_avg[2],
 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
 }else{
 int i;
 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
 &weight_op[4], &weight_avg[4],
 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 }else if(IS_SUB_4X8(sub_mb_type)){
-mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 &weight_op[5], &weight_avg[5],
 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
-mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
+mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
 &weight_op[5], &weight_avg[5],
 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
 }else{
 int j;
 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->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
 CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
 CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
 if( h->pps.cabac ) {
 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));
 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
 }
-memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
+memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
-h->slice_table= h->slice_table_base + s->mb_stride + 1;
+h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
 CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
 for(y=0; y<s->mb_height; y++){
 for(x=0; x<s->mb_width; x++){
 }
 /* can't be in alloc_tables because linesize isn't known there.
 * FIXME: redo bipred weight to not require extra buffer? */
 if(!s->obmc_scratchpad)
-s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
+s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
+/* some macroblocks will be accessed before they're available */
+if(FRAME_MBAFF)
+memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
 return 0;
 }
 static inline void xchg_pair_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;
 int deblock_left = (s->mb_x > 0);
-int deblock_top  = (s->mb_y > 0);
+int deblock_top  = (s->mb_y > 1);
 tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
 src_y  -= 2 *   linesize + 1;
 src_cb -= 2 * uvlinesize + 1;
 if(deblock_top){
 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
+if(s->mb_x+1 < s->mb_width){
+XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
+XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
+}
 }
 if(!(s->flags&CODEC_FLAG_GRAY)){
 if(deblock_left){
 for(i = (!deblock_top) << 1; i<18; i++){
 dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
-if (h->mb_field_decoding_flag) {
+if (MB_FIELD) {
-linesize = s->linesize * 2;
+linesize   = h->mb_linesize   = s->linesize * 2;
-uvlinesize = s->uvlinesize * 2;
+uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
 block_offset = &h->block_offset[24];
 if(mb_y&1){ //FIXME move out of this func?
 dest_y -= s->linesize*15;
 dest_cb-= s->uvlinesize*7;
 dest_cr-= s->uvlinesize*7;
 }
+if(FRAME_MBAFF) {
+int list;
+for(list=0; list<2; list++){
+if(!USES_LIST(mb_type, list))
+continue;
+if(IS_16X16(mb_type)){
+int8_t *ref = &h->ref_cache[list][scan8[0]];
+fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
+}else{
+for(i=0; i<16; i+=4){
+//FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
+int ref = h->ref_cache[list][scan8[i]];
+if(ref >= 0)
+fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
+}
+}
+}
+}
 } else {
-linesize = s->linesize;
+linesize   = h->mb_linesize   = s->linesize;
-uvlinesize = s->uvlinesize;
+uvlinesize = h->mb_uvlinesize = s->uvlinesize;
 //        dct_offset = s->linesize * 16;
 }
 if(transform_bypass){
 idct_dc_add =
 }else{
 idct_dc_add = s->dsp.h264_idct_dc_add;
 idct_add = s->dsp.h264_idct_add;
 }
+if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
+&& (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
+int mbt_y = mb_y&~1;
+uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
+uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
+uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
+xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
+}
 if (IS_INTRA_PCM(mb_type)) {
 unsigned int x, y;
 // The pixels are stored in h->mb array in the same order as levels,
 // copy them in output in the correct order.
 }
 }
 }
 } else {
 if(IS_INTRA(mb_type)){
-if(h->deblocking_filter) {
+if(h->deblocking_filter && !FRAME_MBAFF)
-if (h->mb_aff_frame) {
+xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
-if (!bottom)
-xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
-} else {
-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);
 }
 if(!transform_bypass)
 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
 }else
 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
 }
-if(h->deblocking_filter) {
+if(h->deblocking_filter && !FRAME_MBAFF)
-if (h->mb_aff_frame) {
+xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
-if (bottom) {
-uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
-uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
-uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
-s->mb_y--;
-xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
-s->mb_y++;
-}
-} else {
-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,
 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
 }
 }
 }
 }
 if(h->deblocking_filter) {
-if (h->mb_aff_frame) {
+if (FRAME_MBAFF) {
+//FIXME try deblocking one mb at a time?
+// the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
 const int mb_y = s->mb_y - 1;
 uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
 const int mb_xy= mb_x + mb_y*s->mb_stride;
 const int mb_type_top   = s->current_picture.mb_type[mb_xy];
 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
-uint8_t tmp = s->current_picture.data[1][384];
 if (!bottom) return;
 pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
+if(IS_INTRA(mb_type_top | mb_type_bottom))
+xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
-// TODO deblock a pair
+// deblock a pair
 // top
 s->mb_y--;
 tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
+h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
-if (tmp != s->current_picture.data[1][384]) {
-tprintf("modified pixel 8,1 (1)\n");
-}
 // bottom
 s->mb_y++;
 tprintf("call mbaff filter_mb\n");
 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
+h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
-if (tmp != s->current_picture.data[1][384]) {
-tprintf("modified pixel 8,1 (2)\n");
-}
 } else {
 tprintf("call filter_mb\n");
 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
 direct_dist_scale_factor(h);
 direct_ref_list_init(h);
 return 0;
 }
+static int fill_mbaff_ref_list(H264Context *h){
+int list, i, j;
+for(list=0; list<2; list++){
+for(i=0; i<h->ref_count[list]; i++){
+Picture *frame = &h->ref_list[list][i];
+Picture *field = &h->ref_list[list][16+2*i];
+field[0] = *frame;
+for(j=0; j<3; j++)
+field[0].linesize[j] <<= 1;
+field[1] = field[0];
+for(j=0; j<3; j++)
+field[1].data[j] += frame->linesize[j];
+h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
+h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
+for(j=0; j<2; j++){
+h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
+h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
+}
+}
+}
+for(j=0; j<h->ref_count[1]; j++){
+for(i=0; i<h->ref_count[0]; i++)
+h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
+memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
+memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
+}
+}
 static int pred_weight_table(H264Context *h){
 MpegEncContext * const s = &h->s;
 int list, i;
 int luma_def, chroma_def;
 h->use_weight= 2;
 h->use_weight_chroma= 2;
 h->luma_log2_weight_denom= 5;
 h->chroma_log2_weight_denom= 5;
-/* FIXME: MBAFF */
 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
 int poc0 = h->ref_list[0][ref0].poc;
 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
 int poc1 = h->ref_list[1][ref1].poc;
 int td = clip(poc1 - poc0, -128, 127);
 h-> field_scan[i] = T( field_scan[i]);
 #undef T
 }
 }
 if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
-memcpy(h->zigzag_scan8x8, zigzag_scan8x8, 64*sizeof(uint8_t));
+memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
+memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
+memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
 }else{
 int i;
 for(i=0; i<64; i++){
 #define T(x) (x>>3) | ((x&7)<<3)
-h->zigzag_scan8x8[i] = T(zigzag_scan8x8[i]);
+h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
+h->field_scan8x8[i]        = T(field_scan8x8[i]);
+h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
 #undef T
 }
 }
 if(h->sps.transform_bypass){ //FIXME same ugly
-h->zigzag_scan_q0 = zigzag_scan;
+h->zigzag_scan_q0          = zigzag_scan;
-h->field_scan_q0 = field_scan;
+h->zigzag_scan8x8_q0       = zigzag_scan8x8;
-h->zigzag_scan8x8_q0 = zigzag_scan8x8;
 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
+h->field_scan_q0           = field_scan;
+h->field_scan8x8_q0        = field_scan8x8;
+h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
 }else{
-h->zigzag_scan_q0 = h->zigzag_scan;
+h->zigzag_scan_q0          = h->zigzag_scan;
-h->field_scan_q0 = h->field_scan;
+h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
-h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
+h->field_scan_q0           = h->field_scan;
+h->field_scan8x8_q0        = h->field_scan8x8;
+h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
 }
 alloc_tables(h);
 s->avctx->width = s->width;
 }
 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
+h->mb_mbaff = 0;
 h->mb_aff_frame = 0;
 if(h->sps.frame_mbs_only_flag){
 s->picture_structure= PICT_FRAME;
 }else{
 if(get_bits1(&s->gb)) { //field_pic_flag
 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
+av_log(h->s.avctx, AV_LOG_ERROR, "PAFF interlacing is not implemented\n");
 } else {
 s->picture_structure= PICT_FRAME;
-first_mb_in_slice <<= h->sps.mb_aff;
 h->mb_aff_frame = h->sps.mb_aff;
 }
 }
 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
-s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
+s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << h->mb_aff_frame;
 if(s->mb_y >= s->mb_height){
 return -1;
 }
 if(s->picture_structure==PICT_FRAME){
 h->ref_count[1]= h->pps.ref_count[1];
 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
 if(h->slice_type == B_TYPE){
 h->direct_spatial_mv_pred= get_bits1(&s->gb);
+if(h->sps.mb_aff && h->direct_spatial_mv_pred)
+av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + spatial direct mode is not implemented\n");
 }
 num_ref_idx_active_override_flag= get_bits1(&s->gb);
 if(num_ref_idx_active_override_flag){
 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
 h->use_weight = 0;
 if(s->current_picture.reference)
 decode_ref_pic_marking(h);
+if(FRAME_MBAFF)
+fill_mbaff_ref_list(h);
 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( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
 slice_group_change_cycle= get_bits(&s->gb, ?);
 #endif
 h->slice_num++;
+h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
+h->emu_edge_height= FRAME_MBAFF ? 0 : h->emu_edge_width;
 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n",
 h->slice_num,
 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
 }
 return 0;
 }
+static void predict_field_decoding_flag(H264Context *h){
+MpegEncContext * const s = &h->s;
+const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
+int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
+? s->current_picture.mb_type[mb_xy-1]
+: (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
+? s->current_picture.mb_type[mb_xy-s->mb_stride]
+: 0;
+h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
+}
 /**
 * decodes a P_SKIP or B_SKIP macroblock
 */
 static void decode_mb_skip(H264Context *h){
 MpegEncContext * const s = &h->s;
 int mb_type=0;
 memset(h->non_zero_count[mb_xy], 0, 16);
 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
-if(h->mb_aff_frame && s->mb_skip_run==0 && (s->mb_y&1)==0){
+if(MB_FIELD)
-h->mb_field_decoding_flag= get_bits1(&s->gb);
-}
-if(h->mb_field_decoding_flag)
 mb_type|= MB_TYPE_INTERLACED;
 if( h->slice_type == B_TYPE )
 {
 // just for fill_caches. pred_direct_motion will set the real mb_type
 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);
 if (s->mb_skip_run--) {
+if(FRAME_MBAFF && (s->mb_y&1) == 0){
+if(s->mb_skip_run==0)
+h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
+else
+predict_field_decoding_flag(h);
+}
 decode_mb_skip(h);
 return 0;
 }
 }
-if(h->mb_aff_frame){
+if(FRAME_MBAFF){
-if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
+if( (s->mb_y&1) == 0 )
-h->mb_field_decoding_flag = get_bits1(&s->gb);
+h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
 }else
 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
 h->prev_mb_skipped= 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(h->mb_field_decoding_flag)
+if(MB_FIELD)
 mb_type |= MB_TYPE_INTERLACED;
 h->slice_table[ mb_xy ]= h->slice_num;
 if(IS_INTRA_PCM(mb_type)){
 // All coeffs are present
 memset(h->non_zero_count[mb_xy], 16, 16);
 s->current_picture.mb_type[mb_xy]= mb_type;
 return 0;
+}
+if(MB_MBAFF){
+h->ref_count[0] <<= 1;
+h->ref_count[1] <<= 1;
 }
 fill_caches(h, mb_type, 0);
 //mb_pred
 }
 for(list=0; list<2; list++){
 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
 if(ref_count == 0) continue;
-if (h->mb_aff_frame && h->mb_field_decoding_flag) {
-ref_count <<= 1;
-}
 for(i=0; i<4; i++){
 if(IS_DIRECT(h->sub_mb_type[i])) continue;
 if(IS_DIR(h->sub_mb_type[i], 0, list)){
 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
 }else{
 const uint8_t *scan, *scan8x8, *dc_scan;
 //        fill_non_zero_count_cache(h);
 if(IS_INTERLACED(mb_type)){
+scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
 scan= s->qscale ? h->field_scan : h->field_scan_q0;
 dc_scan= luma_dc_field_scan;
 }else{
+scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
 dc_scan= luma_dc_zigzag_scan;
 }
-scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;
 dquant= get_se_golomb(&s->gb);
 if( dquant > 25 || dquant < -26 ){
 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
 }
 s->current_picture.qscale_table[mb_xy]= s->qscale;
 write_back_non_zero_count(h);
+if(MB_MBAFF){
+h->ref_count[0] >>= 1;
+h->ref_count[1] >>= 1;
+}
 return 0;
 }
 static int decode_cabac_field_decoding_flag(H264Context *h) {
 MpegEncContext * const s = &h->s;
 /* TODO SI/SP frames? */
 return -1;
 }
 }
-static int decode_cabac_mb_skip( H264Context *h) {
+static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) {
 MpegEncContext * const s = &h->s;
-const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
+int mba_xy, mbb_xy;
-const int mba_xy = mb_xy - 1;
-const int mbb_xy = mb_xy - s->mb_stride;
 int ctx = 0;
+if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches?
+int mb_xy = mb_x + (mb_y&~1)*s->mb_stride;
+mba_xy = mb_xy - 1;
+if( (mb_y&1)
+&& h->slice_table[mba_xy] == h->slice_num
+&& MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) )
+mba_xy += s->mb_stride;
+if( MB_FIELD ){
+mbb_xy = mb_xy - s->mb_stride;
+if( !(mb_y&1)
+&& h->slice_table[mbb_xy] == h->slice_num
+&& IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) )
+mbb_xy -= s->mb_stride;
+}else
+mbb_xy = mb_x + (mb_y-1)*s->mb_stride;
+}else{
+int mb_xy = mb_x + mb_y*s->mb_stride;
+mba_xy = mb_xy - 1;
+mbb_xy = mb_xy - s->mb_stride;
+}
 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
 ctx++;
 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
 ctx++;
 if( 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( h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
+if( h->last_qscale_diff != 0 )
 ctx++;
 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
 if( ctx < 2 )
 ctx = 2;
 return ctx + 4 * cat;
 }
 static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff) {
 const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
-static const int significant_coeff_flag_field_offset[2] = { 105, 277 };
+static const int significant_coeff_flag_offset[2][6] = {
-static const int last_significant_coeff_flag_field_offset[2] = { 166, 338 };
+{ 105+0, 105+15, 105+29, 105+44, 105+47, 402 },
-static const int significant_coeff_flag_offset[6] = { 0, 15, 29, 44, 47, 297 };
+{ 277+0, 277+15, 277+29, 277+44, 277+47, 436 }
-static const int last_significant_coeff_flag_offset[6] = { 0, 15, 29, 44, 47, 251 };
+};
-static const int coeff_abs_level_m1_offset[6] = { 227+0, 227+10, 227+20, 227+30, 227+39, 426 };
+static const int last_coeff_flag_offset[2][6] = {
-static const int significant_coeff_flag_offset_8x8[63] = {
+{ 166+0, 166+15, 166+29, 166+44, 166+47, 417 },
-0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,
+{ 338+0, 338+15, 338+29, 338+44, 338+47, 451 }
+};
+static const int coeff_abs_level_m1_offset[6] = {
+227+0, 227+10, 227+20, 227+30, 227+39, 426
+};
+static const int significant_coeff_flag_offset_8x8[2][63] = {
+{ 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,
 4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7,
 7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11,
-12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12
+12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 },
+{ 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5,
+6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11,
+9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9,
+9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 }
 };
 static const int last_coeff_flag_offset_8x8[63] = {
 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
 return 0;
 }
 }
 significant_coeff_ctx_base = h->cabac_state
-+ significant_coeff_flag_offset[cat]
++ significant_coeff_flag_offset[MB_FIELD][cat];
-+ significant_coeff_flag_field_offset[h->mb_field_decoding_flag];
 last_coeff_ctx_base = h->cabac_state
-+ last_significant_coeff_flag_offset[cat]
++ last_coeff_flag_offset[MB_FIELD][cat];
-+ last_significant_coeff_flag_field_offset[h->mb_field_decoding_flag];
 abs_level_m1_ctx_base = h->cabac_state
 + coeff_abs_level_m1_offset[cat];
 if( cat == 5 ) {
 #define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \
 last= max_coeff; \
 break; \
 } \
 } \
 }
-DECODE_SIGNIFICANCE( 63, significant_coeff_flag_offset_8x8[last],
+const int *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD];
-last_coeff_flag_offset_8x8[last] );
+DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] );
 } else {
 DECODE_SIGNIFICANCE( max_coeff - 1, last, last );
 }
 if( last == max_coeff -1 ) {
 index[coeff_count++] = last;
 {
 MpegEncContext * const s = &h->s;
 const int mb_xy  = s->mb_x + s->mb_y*s->mb_stride;
 h->top_mb_xy     = mb_xy - s->mb_stride;
 h->left_mb_xy[0] = mb_xy - 1;
-if(h->mb_aff_frame){
+if(FRAME_MBAFF){
 const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
 const int top_pair_xy      = pair_xy     - s->mb_stride;
 const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
-const int curr_mb_frame_flag = !h->mb_field_decoding_flag;
+const int curr_mb_frame_flag = !MB_FIELD;
 const int bottom = (s->mb_y & 1);
 if (bottom
 ? !curr_mb_frame_flag // bottom macroblock
 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
 ) {
 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?)
 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
+int skip;
+/* a skipped mb needs the aff flag from the following mb */
+if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 )
+predict_field_decoding_flag(h);
+if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped )
+skip = h->next_mb_skipped;
+else
+skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y );
 /* read skip flags */
-if( decode_cabac_mb_skip( h ) ) {
+if( skip ) {
+if( FRAME_MBAFF && (s->mb_y&1)==0 ){
+s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP;
+h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 );
+if(h->next_mb_skipped)
+predict_field_decoding_flag(h);
+else
+h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
+}
 decode_mb_skip(h);
 h->cbp_table[mb_xy] = 0;
 h->chroma_pred_mode_table[mb_xy] = 0;
 h->last_qscale_diff = 0;
 return 0;
 }
 }
-if(h->mb_aff_frame){
+if(FRAME_MBAFF){
-if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
+if( (s->mb_y&1) == 0 )
+h->mb_mbaff =
 h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
 }else
 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
 h->prev_mb_skipped = 0;
 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(h->mb_field_decoding_flag)
+if(MB_FIELD)
 mb_type |= MB_TYPE_INTERLACED;
 h->slice_table[ mb_xy ]= h->slice_num;
 if(IS_INTRA_PCM(mb_type)) {
 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
 // All coeffs are present
 memset(h->non_zero_count[mb_xy], 16, 16);
 s->current_picture.mb_type[mb_xy]= mb_type;
 return 0;
+}
+if(MB_MBAFF){
+h->ref_count[0] <<= 1;
+h->ref_count[1] <<= 1;
 }
 fill_caches(h, mb_type, 0);
 if( IS_INTRA( mb_type ) ) {
 if( cbp || IS_INTRA16x16( mb_type ) ) {
 const uint8_t *scan, *scan8x8, *dc_scan;
 int dqp;
 if(IS_INTERLACED(mb_type)){
+scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0;
 scan= s->qscale ? h->field_scan : h->field_scan_q0;
 dc_scan= luma_dc_field_scan;
 }else{
+scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;
 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;
 dc_scan= luma_dc_zigzag_scan;
 }
-scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;
 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
 if( dqp == INT_MIN ){
 av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y);
 return -1;
 } else {
 uint8_t * const nnz= &h->non_zero_count_cache[0];
 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
+h->last_qscale_diff = 0;
 }
 s->current_picture.qscale_table[mb_xy]= s->qscale;
 write_back_non_zero_count(h);
+if(MB_MBAFF){
+h->ref_count[0] >>= 1;
+h->ref_count[1] >>= 1;
+}
 return 0;
 }
 int alpha;
 int beta;
 int qp_index;
 int bS_index = (i >> 1);
-if (h->mb_field_decoding_flag) {
+if (!MB_FIELD) {
 bS_index &= ~1;
 bS_index |= (i & 1);
 }
 if( bS[bS_index] == 0 ) {
 continue;
 }
-qp_index = h->mb_field_decoding_flag ? (i & 1) : (i >> 3);
+qp_index = MB_FIELD ? (i >> 3) : (i & 1);
 index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );
 alpha = alpha_table[index_a];
 beta  = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];
 if( bS[bS_index] < 4 ) {
 const int tc0 = tc0_table[index_a][bS[bS_index] - 1];
-/* 4px edge length */
 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];
 pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */
 pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */
 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
 }
 }else{
-/* 4px edge length */
 const int p0 = pix[-1];
 const int p1 = pix[-2];
 const int p2 = pix[-3];
 const int q0 = pix[0];
 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
 }
 }
 }
 }
-static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp[2] ) {
+static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[8], int qp[2] ) {
 int i;
 for( i = 0; i < 8; i++, pix += stride) {
 int index_a;
 int alpha;
 int beta;
 if( bS[bS_index] == 0 ) {
 continue;
 }
-qp_index = h->mb_field_decoding_flag ? (i & 1) : (i >> 3);
+qp_index = MB_FIELD ? (i >> 2) : (i & 1);
 index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );
 alpha = alpha_table[index_a];
 beta  = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];
 if( bS[bS_index] < 4 ) {
 const int tc = tc0_table[index_a][bS[bS_index] - 1] + 1;
-/* 2px edge length (because we use same bS than the one for luma) */
 const int p0 = pix[-1];
 const int p1 = pix[-2];
 const int q0 = pix[0];
 const int q1 = pix[1];
 }
 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
 MpegEncContext * const s = &h->s;
 const int mb_xy= mb_x + mb_y*s->mb_stride;
+const int mb_type = s->current_picture.mb_type[mb_xy];
+const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
 int first_vertical_edge_done = 0;
 int dir;
 /* FIXME: A given frame may occupy more than one position in
 * the reference list. So ref2frm should be populated with
 * frame numbers, not indices. */
-static const int ref2frm[18] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
+static const int ref2frm[34] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,
+16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};
 //for sufficiently low qp, filtering wouldn't do anything
 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
-if(!h->mb_aff_frame){
+if(!FRAME_MBAFF){
 int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX(0, h->pps.chroma_qp_index_offset);
 int qp = s->current_picture.qscale_table[mb_xy];
 if(qp <= qp_thresh
 && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)
 && (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){
 return;
 }
 }
-if (h->mb_aff_frame
+if (FRAME_MBAFF
 // left mb is in picture
 && h->slice_table[mb_xy-1] != 255
 // and current and left pair do not have the same interlaced type
-&& (IS_INTERLACED(s->current_picture.mb_type[mb_xy]) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
+&& (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
 // and left mb is in the same slice if deblocking_filter == 2
 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
 /* First vertical edge is different in MBAFF frames
 * There are 8 different bS to compute and 2 different Qp
 */
+const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
+const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
 int bS[8];
 int qp[2];
 int chroma_qp[2];
+int mb_qp, mbn0_qp, mbn1_qp;
 int i;
 first_vertical_edge_done = 1;
-for( i = 0; i < 8; i++ ) {
-int y = i>>1;
+if( IS_INTRA(mb_type) )
-int b_idx= 8 + 4 + 8*y;
+bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;
-int bn_idx= b_idx - 1;
+else {
+for( i = 0; i < 8; i++ ) {
-int mbn_xy = h->mb_field_decoding_flag ? h->left_mb_xy[i>>2] : h->left_mb_xy[i&1];
+int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];
-if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
+if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
-IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
+bS[i] = 4;
-bS[i] = 4;
+else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
-} else if( h->non_zero_count_cache[b_idx] != 0 ||
+/* FIXME: with 8x8dct + cavlc, should check cbp instead of nnz */
-/* FIXME: with 8x8dct + cavlc, should check cbp instead of nnz */
+h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2] )
-h->non_zero_count_cache[bn_idx] != 0 ) {
+bS[i] = 2;
-bS[i] = 2;
+else
-} else {
+bS[i] = 1;
-int l;
+}
-bS[i] = 0;
+}
-for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
-if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
+mb_qp = s->current_picture.qscale_table[mb_xy];
-ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
+mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
-ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
+mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
-bS[i] = 1;
+qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
-break;
+chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) +
-}
+get_chroma_qp( h->pps.chroma_qp_index_offset, mbn0_qp ) + 1 ) >> 1;
-}
+qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
-}
+chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) +
-}
+get_chroma_qp( h->pps.chroma_qp_index_offset, mbn1_qp ) + 1 ) >> 1;
-if(bS[0]+bS[1]+bS[2]+bS[3] != 0) {
-// Do not use s->qscale as luma quantizer because it has not the same
+/* Filter edge */
-// value in IPCM macroblocks.
+tprintf("filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize);
-qp[0] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[0]] + 1 ) >> 1;
+{ int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
-chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
+filter_mb_mbaff_edgev ( h, &img_y [0], linesize,   bS, qp );
-get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[0]] ) + 1 ) >> 1;
+filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );
-qp[1] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[1]] + 1 ) >> 1;
+filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );
-chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
-get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[1]] ) + 1 ) >> 1;
-/* Filter edge */
-tprintf("filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize);
-{ int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
-filter_mb_mbaff_edgev ( h, &img_y [0], linesize,   bS, qp );
-filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );
-filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );
-}
 }
 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
 for( dir = 0; dir < 2; dir++ )
 {
 int edge;
 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
-const int mb_type = s->current_picture.mb_type[mb_xy];
 const int mbm_type = s->current_picture.mb_type[mbm_xy];
 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
 const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))
 == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4;
 }
 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
 start = 1;
+if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0
+&& !IS_INTERLACED(mb_type)
+&& IS_INTERLACED(mbm_type)
+) {
+// This is a special case in the norm where the filtering must
+// be done twice (one each of the field) even if we are in a
+// frame macroblock.
+//
+static const int nnz_idx[4] = {4,5,6,3};
+unsigned int tmp_linesize   = 2 *   linesize;
+unsigned int tmp_uvlinesize = 2 * uvlinesize;
+int mbn_xy = mb_xy - 2 * s->mb_stride;
+int qp, chroma_qp;
+int i, j;
+int bS[4];
+for(j=0; j<2; j++, mbn_xy += s->mb_stride){
+if( IS_INTRA(mb_type) ||
+IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {
+bS[0] = bS[1] = bS[2] = bS[3] = 3;
+} else {
+const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy];
+for( i = 0; i < 4; i++ ) {
+if( h->non_zero_count_cache[scan8[0]+i] != 0 ||
+mbn_nnz[nnz_idx[i]] != 0 )
+bS[i] = 2;
+else
+bS[i] = 1;
+}
+}
+// Do not use s->qscale as luma quantizer because it has not the same
+// value in IPCM macroblocks.
+qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
+tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
+{ int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
+filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );
+chroma_qp = ( h->chroma_qp +
+get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
+filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp );
+filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp );
+}
+start = 1;
+}
 /* Calculate bS */
 for( edge = start; edge < edges; edge++ ) {
 /* mbn_xy: neighbor macroblock */
 const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
 const int mbn_type = s->current_picture.mb_type[mbn_xy];
 int qp;
 if( (edge&1) && IS_8x8DCT(mb_type) )
 continue;
-if (h->mb_aff_frame && (dir == 1) && (edge == 0) && ((mb_y & 1) == 0)
-&& !IS_INTERLACED(mb_type)
-&& IS_INTERLACED(mbn_type)
-) {
-// This is a special case in the norm where the filtering must
-// be done twice (one each of the field) even if we are in a
-// frame macroblock.
-//
-unsigned int tmp_linesize   = 2 *   linesize;
-unsigned int tmp_uvlinesize = 2 * uvlinesize;
-int mbn_xy = mb_xy - 2 * s->mb_stride;
-int qp, chroma_qp;
-// first filtering
-if( IS_INTRA(mb_type) ||
-IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {
-bS[0] = bS[1] = bS[2] = bS[3] = 3;
-} else {
-// TODO
-av_log(h->s.avctx, AV_LOG_ERROR, "both non intra (TODO)\n");
-}
-/* Filter edge */
-// Do not use s->qscale as luma quantizer because it has not the same
-// value in IPCM macroblocks.
-qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
-tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
-{ int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
-filter_mb_edgeh( h, &img_y[0], tmp_linesize, bS, qp );
-chroma_qp = ( h->chroma_qp +
-get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
-filter_mb_edgech( h, &img_cb[0], tmp_uvlinesize, bS, chroma_qp );
-filter_mb_edgech( h, &img_cr[0], tmp_uvlinesize, bS, chroma_qp );
-// second filtering
-mbn_xy += s->mb_stride;
-if( IS_INTRA(mb_type) ||
-IS_INTRA(mbn_type) ) {
-bS[0] = bS[1] = bS[2] = bS[3] = 3;
-} else {
-// TODO
-av_log(h->s.avctx, AV_LOG_ERROR, "both non intra (TODO)\n");
-}
-/* Filter edge */
-// Do not use s->qscale as luma quantizer because it has not the same
-// value in IPCM macroblocks.
-qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
-tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
-{ int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
-filter_mb_edgeh( h, &img_y[linesize], tmp_linesize, bS, qp );
-chroma_qp = ( h->chroma_qp +
-get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
-filter_mb_edgech( h, &img_cb[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
-filter_mb_edgech( h, &img_cr[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
-continue;
-}
 if( IS_INTRA(mb_type) ||
 IS_INTRA(mbn_type) ) {
 int value;
 if (edge == 0) {
 if (   (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type))
-|| ((h->mb_aff_frame || (s->picture_structure != PICT_FRAME)) && (dir == 0))
+|| ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
 ) {
 value = 4;
 } else {
 value = 3;
 }
 if( edge & mask_edge ) {
 bS[0] = bS[1] = bS[2] = bS[3] = 0;
 mv_done = 1;
 }
+else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {
+bS[0] = bS[1] = bS[2] = bS[3] = 1;
+mv_done = 1;
+}
 else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
 int b_idx= 8 + 4 + edge * (dir ? 8:1);
 int bn_idx= b_idx - (dir ? 8:1);
 int v = 0;
 for( l = 0; !v && l < 1 + (h->slice_type == B_TYPE); l++ ) {
 v |= ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
-ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4;
+ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
 }
 bS[0] = bS[1] = bS[2] = bS[3] = v;
 mv_done = 1;
 }
 else
 {
 bS[i] = 0;
 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
-ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
+ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
 bS[i] = 1;
 break;
 }
 }
 }
 int ret = decode_mb_cabac(h);
 int eos;
 if(ret>=0) hl_decode_mb(h);
-/* XXX: useless as decode_mb_cabac it doesn't support that ... */
+if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
-if( ret >= 0 && h->mb_aff_frame ) { //FIXME optimal? or let mb_decode decode 16x32 ?
 s->mb_y++;
 if(ret>=0) ret = decode_mb_cabac(h);
 if(ret>=0) hl_decode_mb(h);
 if( ++s->mb_x >= s->mb_width ) {
 s->mb_x = 0;
 ff_draw_horiz_band(s, 16*s->mb_y, 16);
 ++s->mb_y;
-if(h->mb_aff_frame) {
+if(FRAME_MBAFF) {
 ++s->mb_y;
 }
 }
 if( eos || s->mb_y >= s->mb_height ) {
 for(;;){
 int ret = decode_mb_cavlc(h);
 if(ret>=0) hl_decode_mb(h);
-if(ret>=0 && h->mb_aff_frame){ //FIXME optimal? or let mb_decode decode 16x32 ?
+if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
 s->mb_y++;
 ret = decode_mb_cavlc(h);
 if(ret>=0) hl_decode_mb(h);
 s->mb_y--;
 if(++s->mb_x >= s->mb_width){
 s->mb_x=0;
 ff_draw_horiz_band(s, 16*s->mb_y, 16);
 ++s->mb_y;
-if(h->mb_aff_frame) {
+if(FRAME_MBAFF) {
 ++s->mb_y;
 }
 if(s->mb_y >= s->mb_height){
 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))
 return -1;
 sps->frame_mbs_only_flag= get_bits1(&s->gb);
 if(!sps->frame_mbs_only_flag)
-av_log(h->s.avctx, AV_LOG_ERROR, "interlacing is not supported, picture will probably be garbage\n");
-if(!sps->frame_mbs_only_flag)
 sps->mb_aff= get_bits1(&s->gb);
 else
 sps->mb_aff= 0;
 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
+#ifndef ALLOW_INTERLACE
+if(sps->mb_aff)
+av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF support not included; enable it compilation time\n");
+#endif
+if(!sps->direct_8x8_inference_flag && sps->mb_aff)
+av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + !direct_8x8_inference is not implemented\n");
 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);

Mercurial > libavcodec.hg

comparison h264.c @ 3316:7278f730af27 libavcodec