c2a8a0c642896936832f922d2cf15e97ac00f1ef
[ffmpeg.git] / libavcodec / h264.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This library is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU Lesser General Public
7  * License as published by the Free Software Foundation; either
8  * version 2 of the License, or (at your option) any later version.
9  *
10  * This library is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * Lesser General Public License for more details.
14  *
15  * You should have received a copy of the GNU Lesser General Public
16  * License along with this library; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  *
19  */
20  
21 /**
22  * @file h264.c
23  * H.264 / AVC / MPEG4 part10 codec.
24  * @author Michael Niedermayer <michaelni@gmx.at>
25  */
26
27 #include "common.h"
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264data.h"
32 #include "golomb.h"
33
34 #include "cabac.h"
35
36 #undef NDEBUG
37 #include <assert.h>
38
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
41
42 #define LUMA_DC_BLOCK_INDEX   25
43 #define CHROMA_DC_BLOCK_INDEX 26
44
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS           8
47 #define TOTAL_ZEROS_VLC_BITS           9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS                   3
50 #define RUN7_VLC_BITS                  6
51
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
54
55 #define MAX_MMCO_COUNT 66
56
57 /**
58  * Sequence parameter set
59  */
60 typedef struct SPS{
61     
62     int profile_idc;
63     int level_idc;
64     int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
65     int poc_type;                      ///< pic_order_cnt_type
66     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
67     int delta_pic_order_always_zero_flag;
68     int offset_for_non_ref_pic;
69     int offset_for_top_to_bottom_field;
70     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
71     int ref_frame_count;               ///< num_ref_frames
72     int gaps_in_frame_num_allowed_flag;
73     int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
74     int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
75     int frame_mbs_only_flag;
76     int mb_aff;                        ///<mb_adaptive_frame_field_flag
77     int direct_8x8_inference_flag;
78     int crop;                   ///< frame_cropping_flag
79     int crop_left;              ///< frame_cropping_rect_left_offset
80     int crop_right;             ///< frame_cropping_rect_right_offset
81     int crop_top;               ///< frame_cropping_rect_top_offset
82     int crop_bottom;            ///< frame_cropping_rect_bottom_offset
83     int vui_parameters_present_flag;
84     AVRational sar;
85     int timing_info_present_flag;
86     uint32_t num_units_in_tick;
87     uint32_t time_scale;
88     int fixed_frame_rate_flag;
89     short offset_for_ref_frame[256]; //FIXME dyn aloc?
90 }SPS;
91
92 /**
93  * Picture parameter set
94  */
95 typedef struct PPS{
96     int sps_id;
97     int cabac;                  ///< entropy_coding_mode_flag
98     int pic_order_present;      ///< pic_order_present_flag
99     int slice_group_count;      ///< num_slice_groups_minus1 + 1
100     int mb_slice_group_map_type;
101     int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
102     int weighted_pred;          ///< weighted_pred_flag
103     int weighted_bipred_idc;
104     int init_qp;                ///< pic_init_qp_minus26 + 26
105     int init_qs;                ///< pic_init_qs_minus26 + 26
106     int chroma_qp_index_offset;
107     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108     int constrained_intra_pred; ///< constrained_intra_pred_flag
109     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
110 }PPS;
111
112 /**
113  * Memory management control operation opcode.
114  */
115 typedef enum MMCOOpcode{
116     MMCO_END=0,
117     MMCO_SHORT2UNUSED,
118     MMCO_LONG2UNUSED,
119     MMCO_SHORT2LONG,
120     MMCO_SET_MAX_LONG,
121     MMCO_RESET, 
122     MMCO_LONG,
123 } MMCOOpcode;
124
125 /**
126  * Memory management control operation.
127  */
128 typedef struct MMCO{
129     MMCOOpcode opcode;
130     int short_frame_num;
131     int long_index;
132 } MMCO;
133
134 /**
135  * H264Context
136  */
137 typedef struct H264Context{
138     MpegEncContext s;
139     int nal_ref_idc;    
140     int nal_unit_type;
141 #define NAL_SLICE               1
142 #define NAL_DPA                 2
143 #define NAL_DPB                 3
144 #define NAL_DPC                 4
145 #define NAL_IDR_SLICE           5
146 #define NAL_SEI                 6
147 #define NAL_SPS                 7
148 #define NAL_PPS                 8
149 #define NAL_PICTURE_DELIMITER   9
150 #define NAL_FILTER_DATA         10
151     uint8_t *rbsp_buffer;
152     int rbsp_buffer_size;
153
154     /**
155       * Used to parse AVC variant of h264
156       */
157     int is_avc; ///< this flag is != 0 if codec is avc1
158     int got_avcC; ///< flag used to parse avcC data only once
159     int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
160
161     int chroma_qp; //QPc
162
163     int prev_mb_skiped; //FIXME remove (IMHO not used)
164
165     //prediction stuff
166     int chroma_pred_mode;
167     int intra16x16_pred_mode;
168     
169     int8_t intra4x4_pred_mode_cache[5*8];
170     int8_t (*intra4x4_pred_mode)[8];
171     void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172     void (*pred8x8  [4+3])(uint8_t *src, int stride);
173     void (*pred16x16[4+3])(uint8_t *src, int stride);
174     unsigned int topleft_samples_available;
175     unsigned int top_samples_available;
176     unsigned int topright_samples_available;
177     unsigned int left_samples_available;
178     uint8_t (*top_border)[16+2*8];
179     uint8_t left_border[17+2*9];
180
181     /**
182      * non zero coeff count cache.
183      * is 64 if not available.
184      */
185     uint8_t non_zero_count_cache[6*8] __align8;
186     uint8_t (*non_zero_count)[16];
187
188     /**
189      * Motion vector cache.
190      */
191     int16_t mv_cache[2][5*8][2] __align8;
192     int8_t ref_cache[2][5*8] __align8;
193 #define LIST_NOT_USED -1 //FIXME rename?
194 #define PART_NOT_AVAILABLE -2
195     
196     /**
197      * is 1 if the specific list MV&references are set to 0,0,-2.
198      */
199     int mv_cache_clean[2];
200
201     int block_offset[16+8];
202     int chroma_subblock_offset[16]; //FIXME remove
203     
204     uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
205     uint16_t *mb2b8_xy;
206     int b_stride; //FIXME use s->b4_stride
207     int b8_stride;
208
209     int halfpel_flag;
210     int thirdpel_flag;
211
212     int unknown_svq3_flag;
213     int next_slice_index;
214
215     SPS sps_buffer[MAX_SPS_COUNT];
216     SPS sps; ///< current sps
217     
218     PPS pps_buffer[MAX_PPS_COUNT];
219     /**
220      * current pps
221      */
222     PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
223
224     int slice_num;
225     uint8_t *slice_table_base;
226     uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
227     int slice_type;
228     int slice_type_fixed;
229     
230     //interlacing specific flags
231     int mb_field_decoding_flag;
232     
233     int sub_mb_type[4];
234     
235     //POC stuff
236     int poc_lsb;
237     int poc_msb;
238     int delta_poc_bottom;
239     int delta_poc[2];
240     int frame_num;
241     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
242     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
243     int frame_num_offset;         ///< for POC type 2
244     int prev_frame_num_offset;    ///< for POC type 2
245     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
246
247     /**
248      * frame_num for frames or 2*frame_num for field pics.
249      */
250     int curr_pic_num;
251     
252     /**
253      * max_frame_num or 2*max_frame_num for field pics.
254      */
255     int max_pic_num;
256
257     //Weighted pred stuff
258     int use_weight;
259     int use_weight_chroma;
260     int luma_log2_weight_denom;
261     int chroma_log2_weight_denom;
262     int luma_weight[2][16];
263     int luma_offset[2][16];
264     int chroma_weight[2][16][2];
265     int chroma_offset[2][16][2];
266     int implicit_weight[16][16];
267    
268     //deblock
269     int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
270     int slice_alpha_c0_offset;
271     int slice_beta_offset;
272      
273     int redundant_pic_count;
274     
275     int direct_spatial_mv_pred;
276     int dist_scale_factor[16];
277
278     /**
279      * num_ref_idx_l0/1_active_minus1 + 1
280      */
281     int ref_count[2];// FIXME split for AFF
282     Picture *short_ref[16];
283     Picture *long_ref[16];
284     Picture default_ref_list[2][32];
285     Picture ref_list[2][32]; //FIXME size?
286     Picture field_ref_list[2][32]; //FIXME size?
287     Picture *delayed_pic[16]; //FIXME size?
288     
289     /**
290      * memory management control operations buffer.
291      */
292     MMCO mmco[MAX_MMCO_COUNT];
293     int mmco_index;
294     
295     int long_ref_count;  ///< number of actual long term references
296     int short_ref_count; ///< number of actual short term references
297     
298     //data partitioning
299     GetBitContext intra_gb;
300     GetBitContext inter_gb;
301     GetBitContext *intra_gb_ptr;
302     GetBitContext *inter_gb_ptr;
303     
304     DCTELEM mb[16*24] __align8;
305
306     /**
307      * Cabac
308      */
309     CABACContext cabac;
310     uint8_t      cabac_state[399];
311     int          cabac_init_idc;
312
313     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
314     uint16_t     *cbp_table;
315     int top_cbp;
316     int left_cbp;
317     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
318     uint8_t     *chroma_pred_mode_table;
319     int         last_qscale_diff;
320     int16_t     (*mvd_table[2])[2];
321     int16_t     mvd_cache[2][5*8][2] __align8;
322     uint8_t     *direct_table;
323     uint8_t     direct_cache[5*8];
324
325 }H264Context;
326
327 static VLC coeff_token_vlc[4];
328 static VLC chroma_dc_coeff_token_vlc;
329
330 static VLC total_zeros_vlc[15];
331 static VLC chroma_dc_total_zeros_vlc[3];
332
333 static VLC run_vlc[6];
334 static VLC run7_vlc;
335
336 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
337 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
338 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
339
340 static inline uint32_t pack16to32(int a, int b){
341 #ifdef WORDS_BIGENDIAN
342    return (b&0xFFFF) + (a<<16);
343 #else
344    return (a&0xFFFF) + (b<<16);
345 #endif
346 }
347
348 /**
349  * fill a rectangle.
350  * @param h height of the rectangle, should be a constant
351  * @param w width of the rectangle, should be a constant
352  * @param size the size of val (1 or 4), should be a constant
353  */
354 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
355     uint8_t *p= (uint8_t*)vp;
356     assert(size==1 || size==4);
357     
358     w      *= size;
359     stride *= size;
360     
361     assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
362 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
363     if(w==2 && h==2){
364         *(uint16_t*)(p + 0)=
365         *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
366     }else if(w==2 && h==4){
367         *(uint16_t*)(p + 0*stride)=
368         *(uint16_t*)(p + 1*stride)=
369         *(uint16_t*)(p + 2*stride)=
370         *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
371     }else if(w==4 && h==1){
372         *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
373     }else if(w==4 && h==2){
374         *(uint32_t*)(p + 0*stride)=
375         *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
376     }else if(w==4 && h==4){
377         *(uint32_t*)(p + 0*stride)=
378         *(uint32_t*)(p + 1*stride)=
379         *(uint32_t*)(p + 2*stride)=
380         *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
381     }else if(w==8 && h==1){
382         *(uint32_t*)(p + 0)=
383         *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
384     }else if(w==8 && h==2){
385         *(uint32_t*)(p + 0 + 0*stride)=
386         *(uint32_t*)(p + 4 + 0*stride)=
387         *(uint32_t*)(p + 0 + 1*stride)=
388         *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
389     }else if(w==8 && h==4){
390         *(uint64_t*)(p + 0*stride)=
391         *(uint64_t*)(p + 1*stride)=
392         *(uint64_t*)(p + 2*stride)=
393         *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
394     }else if(w==16 && h==2){
395         *(uint64_t*)(p + 0+0*stride)=
396         *(uint64_t*)(p + 8+0*stride)=
397         *(uint64_t*)(p + 0+1*stride)=
398         *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
399     }else if(w==16 && h==4){
400         *(uint64_t*)(p + 0+0*stride)=
401         *(uint64_t*)(p + 8+0*stride)=
402         *(uint64_t*)(p + 0+1*stride)=
403         *(uint64_t*)(p + 8+1*stride)=
404         *(uint64_t*)(p + 0+2*stride)=
405         *(uint64_t*)(p + 8+2*stride)=
406         *(uint64_t*)(p + 0+3*stride)=
407         *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
408     }else
409         assert(0);
410 }
411
412 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
413     MpegEncContext * const s = &h->s;
414     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
415     int topleft_xy, top_xy, topright_xy, left_xy[2];
416     int topleft_type, top_type, topright_type, left_type[2];
417     int left_block[4];
418     int i;
419
420     //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
421     
422     if(h->sps.mb_aff){
423     //FIXME
424         topleft_xy = 0; /* avoid warning */
425         top_xy = 0; /* avoid warning */
426         topright_xy = 0; /* avoid warning */
427     }else{
428         topleft_xy = mb_xy-1 - s->mb_stride;
429         top_xy     = mb_xy   - s->mb_stride;
430         topright_xy= mb_xy+1 - s->mb_stride;
431         left_xy[0]   = mb_xy-1;
432         left_xy[1]   = mb_xy-1;
433         left_block[0]= 0;
434         left_block[1]= 1;
435         left_block[2]= 2;
436         left_block[3]= 3;
437     }
438
439     if(for_deblock){
440         topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
441         top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
442         topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
443         left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
444         left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
445     }else{
446         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
447         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
448         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
449         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
450         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
451     }
452
453     if(IS_INTRA(mb_type)){
454         h->topleft_samples_available= 
455         h->top_samples_available= 
456         h->left_samples_available= 0xFFFF;
457         h->topright_samples_available= 0xEEEA;
458
459         if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
460             h->topleft_samples_available= 0xB3FF;
461             h->top_samples_available= 0x33FF;
462             h->topright_samples_available= 0x26EA;
463         }
464         for(i=0; i<2; i++){
465             if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
466                 h->topleft_samples_available&= 0xDF5F;
467                 h->left_samples_available&= 0x5F5F;
468             }
469         }
470         
471         if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
472             h->topleft_samples_available&= 0x7FFF;
473         
474         if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
475             h->topright_samples_available&= 0xFBFF;
476     
477         if(IS_INTRA4x4(mb_type)){
478             if(IS_INTRA4x4(top_type)){
479                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
480                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
481                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
482                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
483             }else{
484                 int pred;
485                 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
486                     pred= -1;
487                 else{
488                     pred= 2;
489                 }
490                 h->intra4x4_pred_mode_cache[4+8*0]=
491                 h->intra4x4_pred_mode_cache[5+8*0]=
492                 h->intra4x4_pred_mode_cache[6+8*0]=
493                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
494             }
495             for(i=0; i<2; i++){
496                 if(IS_INTRA4x4(left_type[i])){
497                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
498                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
499                 }else{
500                     int pred;
501                     if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
502                         pred= -1;
503                     else{
504                         pred= 2;
505                     }
506                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
507                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
508                 }
509             }
510         }
511     }
512     
513     
514 /*
515 0 . T T. T T T T 
516 1 L . .L . . . . 
517 2 L . .L . . . . 
518 3 . T TL . . . . 
519 4 L . .L . . . . 
520 5 L . .. . . . . 
521 */
522 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
523     if(top_type){
524         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
525         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
526         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
527         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
528     
529         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
530         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
531     
532         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
533         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
534         
535         h->top_cbp= h->cbp_table[top_xy];
536     }else{
537         h->non_zero_count_cache[4+8*0]=      
538         h->non_zero_count_cache[5+8*0]=
539         h->non_zero_count_cache[6+8*0]=
540         h->non_zero_count_cache[7+8*0]=
541     
542         h->non_zero_count_cache[1+8*0]=
543         h->non_zero_count_cache[2+8*0]=
544     
545         h->non_zero_count_cache[1+8*3]=
546         h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
547         
548         if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
549         else                  h->top_cbp= 0;
550     }
551     
552     if(left_type[0]){
553         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
554         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
555         h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
556         h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
557         h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
558     }else{
559         h->non_zero_count_cache[3+8*1]= 
560         h->non_zero_count_cache[3+8*2]= 
561         h->non_zero_count_cache[0+8*1]= 
562         h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
563         
564         if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
565         else                  h->left_cbp= 0;
566     }
567     
568     if(left_type[1]){
569         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
570         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
571         h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
572         h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
573     }else{
574         h->non_zero_count_cache[3+8*3]= 
575         h->non_zero_count_cache[3+8*4]= 
576         h->non_zero_count_cache[0+8*2]= 
577         h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
578     }
579     
580 #if 1
581     //FIXME direct mb can skip much of this
582     if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
583         int list;
584         for(list=0; list<2; list++){
585             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !for_deblock){
586                 /*if(!h->mv_cache_clean[list]){
587                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
588                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
589                     h->mv_cache_clean[list]= 1;
590                 }*/
591                 continue;
592             }
593             h->mv_cache_clean[list]= 0;
594             
595             if(IS_INTER(topleft_type)){
596                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
597                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
598                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
599                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
600             }else{
601                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
602                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
603             }
604             
605             if(IS_INTER(top_type)){
606                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
607                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
608                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
609                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
610                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
611                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
612                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
613                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
614                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
615                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
616             }else{
617                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
618                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
619                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
620                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
621                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
622             }
623
624             if(IS_INTER(topright_type)){
625                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
626                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
627                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
628                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
629             }else{
630                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
631                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
632             }
633             
634             //FIXME unify cleanup or sth
635             if(IS_INTER(left_type[0])){
636                 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
637                 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
638                 *(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]];
639                 *(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]];
640                 h->ref_cache[list][scan8[0] - 1 + 0*8]= 
641                 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
642             }else{
643                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
644                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
645                 h->ref_cache[list][scan8[0] - 1 + 0*8]=
646                 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
647             }
648             
649             if(IS_INTER(left_type[1])){
650                 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
651                 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
652                 *(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]];
653                 *(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]];
654                 h->ref_cache[list][scan8[0] - 1 + 2*8]= 
655                 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
656             }else{
657                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
658                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
659                 h->ref_cache[list][scan8[0] - 1 + 2*8]=
660                 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
661             }
662
663             if(for_deblock)
664                 continue;
665
666             h->ref_cache[list][scan8[5 ]+1] = 
667             h->ref_cache[list][scan8[7 ]+1] = 
668             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
669             h->ref_cache[list][scan8[4 ]] = 
670             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
671             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
672             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
673             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
674             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
675             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
676
677             if( h->pps.cabac ) {
678                 /* XXX beurk, Load mvd */
679                 if(IS_INTER(topleft_type)){
680                     const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
681                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
682                 }else{
683                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
684                 }
685
686                 if(IS_INTER(top_type)){
687                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
688                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
689                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
690                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
691                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
692                 }else{
693                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
694                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
695                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
696                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
697                 }
698                 if(IS_INTER(left_type[0])){
699                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
700                     *(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]];
701                     *(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]];
702                 }else{
703                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
704                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
705                 }
706                 if(IS_INTER(left_type[1])){
707                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
708                     *(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]];
709                     *(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]];
710                 }else{
711                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
712                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
713                 }
714                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
715                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
716                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
717                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
718                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
719
720                 if(h->slice_type == B_TYPE){
721                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
722
723                     if(IS_DIRECT(top_type)){
724                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
725                     }else if(IS_8X8(top_type)){
726                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
727                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
728                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
729                     }else{
730                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
731                     }
732                     
733                     //FIXME interlacing
734                     if(IS_DIRECT(left_type[0])){
735                         h->direct_cache[scan8[0] - 1 + 0*8]=
736                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
737                     }else if(IS_8X8(left_type[0])){
738                         int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
739                         h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
740                         h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
741                     }else{
742                         h->direct_cache[scan8[0] - 1 + 0*8]=
743                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
744                     }
745                 }
746             }
747         }
748     }
749 #endif
750 }
751
752 static inline void write_back_intra_pred_mode(H264Context *h){
753     MpegEncContext * const s = &h->s;
754     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
755
756     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
757     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
758     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
759     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
760     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
761     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
762     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
763 }
764
765 /**
766  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
767  */
768 static inline int check_intra4x4_pred_mode(H264Context *h){
769     MpegEncContext * const s = &h->s;
770     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
771     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
772     int i;
773     
774     if(!(h->top_samples_available&0x8000)){
775         for(i=0; i<4; i++){
776             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
777             if(status<0){
778                 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);
779                 return -1;
780             } else if(status){
781                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
782             }
783         }
784     }
785     
786     if(!(h->left_samples_available&0x8000)){
787         for(i=0; i<4; i++){
788             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
789             if(status<0){
790                 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);
791                 return -1;
792             } else if(status){
793                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
794             }
795         }
796     }
797
798     return 0;
799 } //FIXME cleanup like next
800
801 /**
802  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
803  */
804 static inline int check_intra_pred_mode(H264Context *h, int mode){
805     MpegEncContext * const s = &h->s;
806     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
807     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
808     
809     if(mode < 0 || mode > 6) {
810         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
811         return -1;
812     }
813     
814     if(!(h->top_samples_available&0x8000)){
815         mode= top[ mode ];
816         if(mode<0){
817             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);
818             return -1;
819         }
820     }
821     
822     if(!(h->left_samples_available&0x8000)){
823         mode= left[ mode ];
824         if(mode<0){
825             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);
826             return -1;
827         } 
828     }
829
830     return mode;
831 }
832
833 /**
834  * gets the predicted intra4x4 prediction mode.
835  */
836 static inline int pred_intra_mode(H264Context *h, int n){
837     const int index8= scan8[n];
838     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
839     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
840     const int min= FFMIN(left, top);
841
842     tprintf("mode:%d %d min:%d\n", left ,top, min);
843
844     if(min<0) return DC_PRED;
845     else      return min;
846 }
847
848 static inline void write_back_non_zero_count(H264Context *h){
849     MpegEncContext * const s = &h->s;
850     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
851
852     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
853     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
854     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
855     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
856     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
857     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
858     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
859     
860     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
861     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
862     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
863
864     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
865     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
866     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
867 }
868
869 /**
870  * gets the predicted number of non zero coefficients.
871  * @param n block index
872  */
873 static inline int pred_non_zero_count(H264Context *h, int n){
874     const int index8= scan8[n];
875     const int left= h->non_zero_count_cache[index8 - 1];
876     const int top = h->non_zero_count_cache[index8 - 8];
877     int i= left + top;
878     
879     if(i<64) i= (i+1)>>1;
880
881     tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
882
883     return i&31;
884 }
885
886 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
887     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
888
889     if(topright_ref != PART_NOT_AVAILABLE){
890         *C= h->mv_cache[list][ i - 8 + part_width ];
891         return topright_ref;
892     }else{
893         tprintf("topright MV not available\n");
894
895         *C= h->mv_cache[list][ i - 8 - 1 ];
896         return h->ref_cache[list][ i - 8 - 1 ];
897     }
898 }
899
900 /**
901  * gets the predicted MV.
902  * @param n the block index
903  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
904  * @param mx the x component of the predicted motion vector
905  * @param my the y component of the predicted motion vector
906  */
907 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
908     const int index8= scan8[n];
909     const int top_ref=      h->ref_cache[list][ index8 - 8 ];
910     const int left_ref=     h->ref_cache[list][ index8 - 1 ];
911     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
912     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
913     const int16_t * C;
914     int diagonal_ref, match_count;
915
916     assert(part_width==1 || part_width==2 || part_width==4);
917
918 /* mv_cache
919   B . . A T T T T 
920   U . . L . . , .
921   U . . L . . . .
922   U . . L . . , .
923   . . . L . . . .
924 */
925
926     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
927     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
928     tprintf("pred_motion match_count=%d\n", match_count);
929     if(match_count > 1){ //most common
930         *mx= mid_pred(A[0], B[0], C[0]);
931         *my= mid_pred(A[1], B[1], C[1]);
932     }else if(match_count==1){
933         if(left_ref==ref){
934             *mx= A[0];
935             *my= A[1];        
936         }else if(top_ref==ref){
937             *mx= B[0];
938             *my= B[1];        
939         }else{
940             *mx= C[0];
941             *my= C[1];        
942         }
943     }else{
944         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
945             *mx= A[0];
946             *my= A[1];        
947         }else{
948             *mx= mid_pred(A[0], B[0], C[0]);
949             *my= mid_pred(A[1], B[1], C[1]);
950         }
951     }
952         
953     tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
954 }
955
956 /**
957  * gets the directionally predicted 16x8 MV.
958  * @param n the block index
959  * @param mx the x component of the predicted motion vector
960  * @param my the y component of the predicted motion vector
961  */
962 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
963     if(n==0){
964         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
965         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
966
967         tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
968         
969         if(top_ref == ref){
970             *mx= B[0];
971             *my= B[1];
972             return;
973         }
974     }else{
975         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
976         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
977         
978         tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
979
980         if(left_ref == ref){
981             *mx= A[0];
982             *my= A[1];
983             return;
984         }
985     }
986
987     //RARE
988     pred_motion(h, n, 4, list, ref, mx, my);
989 }
990
991 /**
992  * gets the directionally predicted 8x16 MV.
993  * @param n the block index
994  * @param mx the x component of the predicted motion vector
995  * @param my the y component of the predicted motion vector
996  */
997 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
998     if(n==0){
999         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1000         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1001         
1002         tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1003
1004         if(left_ref == ref){
1005             *mx= A[0];
1006             *my= A[1];
1007             return;
1008         }
1009     }else{
1010         const int16_t * C;
1011         int diagonal_ref;
1012
1013         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1014         
1015         tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1016
1017         if(diagonal_ref == ref){ 
1018             *mx= C[0];
1019             *my= C[1];
1020             return;
1021         }
1022     }
1023
1024     //RARE
1025     pred_motion(h, n, 2, list, ref, mx, my);
1026 }
1027
1028 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1029     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1030     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1031
1032     tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1033
1034     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1035        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1036        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1037        
1038         *mx = *my = 0;
1039         return;
1040     }
1041         
1042     pred_motion(h, 0, 4, 0, 0, mx, my);
1043
1044     return;
1045 }
1046
1047 static inline void direct_dist_scale_factor(H264Context * const h){
1048     const int poc = h->s.current_picture_ptr->poc;
1049     const int poc1 = h->ref_list[1][0].poc;
1050     int i;
1051     for(i=0; i<h->ref_count[0]; i++){
1052         int poc0 = h->ref_list[0][i].poc;
1053         int td = clip(poc1 - poc0, -128, 127);
1054         if(td == 0 /* FIXME || pic0 is a long-term ref */){
1055             h->dist_scale_factor[i] = 256;
1056         }else{
1057             int tb = clip(poc - poc0, -128, 127);
1058             int tx = (16384 + (ABS(td) >> 1)) / td;
1059             h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1060         }
1061     }
1062 }
1063
1064 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1065     MpegEncContext * const s = &h->s;
1066     const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1067     const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1068     const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1069     const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1070     const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1071     const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1072     const int is_b8x8 = IS_8X8(*mb_type);
1073     int sub_mb_type;
1074     int i8, i4;
1075
1076     if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1077         /* FIXME save sub mb types from previous frames (or derive from MVs)
1078          * so we know exactly what block size to use */
1079         sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1080         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1081     }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1082         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1083         *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1084     }else{
1085         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1086         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1087     }
1088     if(!is_b8x8)
1089         *mb_type |= MB_TYPE_DIRECT2;
1090
1091     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);
1092     
1093     if(h->direct_spatial_mv_pred){
1094         int ref[2];
1095         int mv[2][2];
1096         int list;
1097
1098         /* ref = min(neighbors) */
1099         for(list=0; list<2; list++){
1100             int refa = h->ref_cache[list][scan8[0] - 1];
1101             int refb = h->ref_cache[list][scan8[0] - 8];
1102             int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1103             if(refc == -2)
1104                 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1105             ref[list] = refa;
1106             if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1107                 ref[list] = refb;
1108             if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1109                 ref[list] = refc;
1110             if(ref[list] < 0)
1111                 ref[list] = -1;
1112         }
1113
1114         if(ref[0] < 0 && ref[1] < 0){
1115             ref[0] = ref[1] = 0;
1116             mv[0][0] = mv[0][1] =
1117             mv[1][0] = mv[1][1] = 0;
1118         }else{
1119             for(list=0; list<2; list++){
1120                 if(ref[list] >= 0)
1121                     pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1122                 else
1123                     mv[list][0] = mv[list][1] = 0;
1124             }
1125         }
1126
1127         if(ref[1] < 0){
1128             *mb_type &= ~MB_TYPE_P0L1;
1129             sub_mb_type &= ~MB_TYPE_P0L1;
1130         }else if(ref[0] < 0){
1131             *mb_type &= ~MB_TYPE_P0L0;
1132             sub_mb_type &= ~MB_TYPE_P0L0;
1133         }
1134
1135         if(IS_16X16(*mb_type)){
1136             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1137             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1138             if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1139                 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1140                 if(ref[0] > 0)
1141                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1142                 else
1143                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1144                 if(ref[1] > 0)
1145                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1146                 else
1147                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1148             }else{
1149                 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1150                 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1151             }
1152         }else{
1153             for(i8=0; i8<4; i8++){
1154                 const int x8 = i8&1;
1155                 const int y8 = i8>>1;
1156     
1157                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1158                     continue;
1159                 h->sub_mb_type[i8] = sub_mb_type;
1160     
1161                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1162                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1163                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1164                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1165     
1166                 /* col_zero_flag */
1167                 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1168                     for(i4=0; i4<4; i4++){
1169                         const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1170                         if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1171                             if(ref[0] == 0)
1172                                 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1173                             if(ref[1] == 0)
1174                                 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1175                         }
1176                     }
1177                 }
1178             }
1179         }
1180     }else{ /* direct temporal mv pred */
1181         /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1182         if(IS_16X16(*mb_type)){
1183             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1184             if(IS_INTRA(mb_type_col)){
1185                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1186                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1187                 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1188             }else{
1189                 const int ref0 = l1ref0[0];
1190                 const int dist_scale_factor = h->dist_scale_factor[ref0];
1191                 const int16_t *mv_col = l1mv0[0];
1192                 int mv_l0[2];
1193                 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1194                 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1195                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1196                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1197                 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);
1198             }
1199         }else{
1200             for(i8=0; i8<4; i8++){
1201                 const int x8 = i8&1;
1202                 const int y8 = i8>>1;
1203                 int ref0, dist_scale_factor;
1204     
1205                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1206                     continue;
1207                 h->sub_mb_type[i8] = sub_mb_type;
1208                 if(IS_INTRA(mb_type_col)){
1209                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1210                     fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1211                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1212                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1213                     continue;
1214                 }
1215     
1216                 ref0 = l1ref0[x8 + y8*h->b8_stride];
1217                 dist_scale_factor = h->dist_scale_factor[ref0];
1218     
1219                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1220                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1221                 for(i4=0; i4<4; i4++){
1222                     const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1223                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1224                     mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1225                     mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1226                     *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1227                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1228                 }
1229             }
1230         }
1231     }
1232 }
1233
1234 static inline void write_back_motion(H264Context *h, int mb_type){
1235     MpegEncContext * const s = &h->s;
1236     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1237     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1238     int list;
1239
1240     for(list=0; list<2; list++){
1241         int y;
1242         if(!USES_LIST(mb_type, list)){
1243             if(1){ //FIXME skip or never read if mb_type doesnt use it
1244                 for(y=0; y<4; y++){
1245                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1246                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1247                 }
1248                 if( h->pps.cabac ) {
1249                     /* FIXME needed ? */
1250                     for(y=0; y<4; y++){
1251                         *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1252                         *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1253                     }
1254                 }
1255                 for(y=0; y<2; y++){
1256                     *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1257                 }
1258             }
1259             continue;
1260         }
1261         
1262         for(y=0; y<4; y++){
1263             *(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];
1264             *(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];
1265         }
1266         if( h->pps.cabac ) {
1267             for(y=0; y<4; y++){
1268                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1269                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1270             }
1271         }
1272         for(y=0; y<2; y++){
1273             s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1274             s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1275         }
1276     }
1277     
1278     if(h->slice_type == B_TYPE && h->pps.cabac){
1279         if(IS_8X8(mb_type)){
1280             h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1281             h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1282             h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1283         }
1284     }
1285 }
1286
1287 /**
1288  * Decodes a network abstraction layer unit.
1289  * @param consumed is the number of bytes used as input
1290  * @param length is the length of the array
1291  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1292  * @returns decoded bytes, might be src+1 if no escapes 
1293  */
1294 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1295     int i, si, di;
1296     uint8_t *dst;
1297
1298 //    src[0]&0x80;              //forbidden bit
1299     h->nal_ref_idc= src[0]>>5;
1300     h->nal_unit_type= src[0]&0x1F;
1301
1302     src++; length--;
1303 #if 0    
1304     for(i=0; i<length; i++)
1305         printf("%2X ", src[i]);
1306 #endif
1307     for(i=0; i+1<length; i+=2){
1308         if(src[i]) continue;
1309         if(i>0 && src[i-1]==0) i--;
1310         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1311             if(src[i+2]!=3){
1312                 /* startcode, so we must be past the end */
1313                 length=i;
1314             }
1315             break;
1316         }
1317     }
1318
1319     if(i>=length-1){ //no escaped 0
1320         *dst_length= length;
1321         *consumed= length+1; //+1 for the header
1322         return src; 
1323     }
1324
1325     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1326     dst= h->rbsp_buffer;
1327
1328 //printf("deoding esc\n");
1329     si=di=0;
1330     while(si<length){ 
1331         //remove escapes (very rare 1:2^22)
1332         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1333             if(src[si+2]==3){ //escape
1334                 dst[di++]= 0;
1335                 dst[di++]= 0;
1336                 si+=3;
1337                 continue;
1338             }else //next start code
1339                 break;
1340         }
1341
1342         dst[di++]= src[si++];
1343     }
1344
1345     *dst_length= di;
1346     *consumed= si + 1;//+1 for the header
1347 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1348     return dst;
1349 }
1350
1351 #if 0
1352 /**
1353  * @param src the data which should be escaped
1354  * @param dst the target buffer, dst+1 == src is allowed as a special case
1355  * @param length the length of the src data
1356  * @param dst_length the length of the dst array
1357  * @returns length of escaped data in bytes or -1 if an error occured
1358  */
1359 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1360     int i, escape_count, si, di;
1361     uint8_t *temp;
1362     
1363     assert(length>=0);
1364     assert(dst_length>0);
1365     
1366     dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1367
1368     if(length==0) return 1;
1369
1370     escape_count= 0;
1371     for(i=0; i<length; i+=2){
1372         if(src[i]) continue;
1373         if(i>0 && src[i-1]==0) 
1374             i--;
1375         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1376             escape_count++;
1377             i+=2;
1378         }
1379     }
1380     
1381     if(escape_count==0){ 
1382         if(dst+1 != src)
1383             memcpy(dst+1, src, length);
1384         return length + 1;
1385     }
1386     
1387     if(length + escape_count + 1> dst_length)
1388         return -1;
1389
1390     //this should be damn rare (hopefully)
1391
1392     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1393     temp= h->rbsp_buffer;
1394 //printf("encoding esc\n");
1395     
1396     si= 0;
1397     di= 0;
1398     while(si < length){
1399         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1400             temp[di++]= 0; si++;
1401             temp[di++]= 0; si++;
1402             temp[di++]= 3; 
1403             temp[di++]= src[si++];
1404         }
1405         else
1406             temp[di++]= src[si++];
1407     }
1408     memcpy(dst+1, temp, length+escape_count);
1409     
1410     assert(di == length+escape_count);
1411     
1412     return di + 1;
1413 }
1414
1415 /**
1416  * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1417  */
1418 static void encode_rbsp_trailing(PutBitContext *pb){
1419     int length;
1420     put_bits(pb, 1, 1);
1421     length= (-put_bits_count(pb))&7;
1422     if(length) put_bits(pb, length, 0);
1423 }
1424 #endif
1425
1426 /**
1427  * identifies the exact end of the bitstream
1428  * @return the length of the trailing, or 0 if damaged
1429  */
1430 static int decode_rbsp_trailing(uint8_t *src){
1431     int v= *src;
1432     int r;
1433
1434     tprintf("rbsp trailing %X\n", v);
1435
1436     for(r=1; r<9; r++){
1437         if(v&1) return r;
1438         v>>=1;
1439     }
1440     return 0;
1441 }
1442
1443 /**
1444  * idct tranforms the 16 dc values and dequantize them.
1445  * @param qp quantization parameter
1446  */
1447 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1448     const int qmul= dequant_coeff[qp][0];
1449 #define stride 16
1450     int i;
1451     int temp[16]; //FIXME check if this is a good idea
1452     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1453     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1454
1455 //memset(block, 64, 2*256);
1456 //return;
1457     for(i=0; i<4; i++){
1458         const int offset= y_offset[i];
1459         const int z0= block[offset+stride*0] + block[offset+stride*4];
1460         const int z1= block[offset+stride*0] - block[offset+stride*4];
1461         const int z2= block[offset+stride*1] - block[offset+stride*5];
1462         const int z3= block[offset+stride*1] + block[offset+stride*5];
1463
1464         temp[4*i+0]= z0+z3;
1465         temp[4*i+1]= z1+z2;
1466         temp[4*i+2]= z1-z2;
1467         temp[4*i+3]= z0-z3;
1468     }
1469
1470     for(i=0; i<4; i++){
1471         const int offset= x_offset[i];
1472         const int z0= temp[4*0+i] + temp[4*2+i];
1473         const int z1= temp[4*0+i] - temp[4*2+i];
1474         const int z2= temp[4*1+i] - temp[4*3+i];
1475         const int z3= temp[4*1+i] + temp[4*3+i];
1476
1477         block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1478         block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1479         block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1480         block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1481     }
1482 }
1483
1484 #if 0
1485 /**
1486  * dct tranforms the 16 dc values.
1487  * @param qp quantization parameter ??? FIXME
1488  */
1489 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1490 //    const int qmul= dequant_coeff[qp][0];
1491     int i;
1492     int temp[16]; //FIXME check if this is a good idea
1493     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1494     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1495
1496     for(i=0; i<4; i++){
1497         const int offset= y_offset[i];
1498         const int z0= block[offset+stride*0] + block[offset+stride*4];
1499         const int z1= block[offset+stride*0] - block[offset+stride*4];
1500         const int z2= block[offset+stride*1] - block[offset+stride*5];
1501         const int z3= block[offset+stride*1] + block[offset+stride*5];
1502
1503         temp[4*i+0]= z0+z3;
1504         temp[4*i+1]= z1+z2;
1505         temp[4*i+2]= z1-z2;
1506         temp[4*i+3]= z0-z3;
1507     }
1508
1509     for(i=0; i<4; i++){
1510         const int offset= x_offset[i];
1511         const int z0= temp[4*0+i] + temp[4*2+i];
1512         const int z1= temp[4*0+i] - temp[4*2+i];
1513         const int z2= temp[4*1+i] - temp[4*3+i];
1514         const int z3= temp[4*1+i] + temp[4*3+i];
1515
1516         block[stride*0 +offset]= (z0 + z3)>>1;
1517         block[stride*2 +offset]= (z1 + z2)>>1;
1518         block[stride*8 +offset]= (z1 - z2)>>1;
1519         block[stride*10+offset]= (z0 - z3)>>1;
1520     }
1521 }
1522 #endif
1523
1524 #undef xStride
1525 #undef stride
1526
1527 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1528     const int qmul= dequant_coeff[qp][0];
1529     const int stride= 16*2;
1530     const int xStride= 16;
1531     int a,b,c,d,e;
1532
1533     a= block[stride*0 + xStride*0];
1534     b= block[stride*0 + xStride*1];
1535     c= block[stride*1 + xStride*0];
1536     d= block[stride*1 + xStride*1];
1537
1538     e= a-b;
1539     a= a+b;
1540     b= c-d;
1541     c= c+d;
1542
1543     block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1544     block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1545     block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1546     block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1547 }
1548
1549 #if 0
1550 static void chroma_dc_dct_c(DCTELEM *block){
1551     const int stride= 16*2;
1552     const int xStride= 16;
1553     int a,b,c,d,e;
1554
1555     a= block[stride*0 + xStride*0];
1556     b= block[stride*0 + xStride*1];
1557     c= block[stride*1 + xStride*0];
1558     d= block[stride*1 + xStride*1];
1559
1560     e= a-b;
1561     a= a+b;
1562     b= c-d;
1563     c= c+d;
1564
1565     block[stride*0 + xStride*0]= (a+c);
1566     block[stride*0 + xStride*1]= (e+b);
1567     block[stride*1 + xStride*0]= (a-c);
1568     block[stride*1 + xStride*1]= (e-b);
1569 }
1570 #endif
1571
1572 /**
1573  * gets the chroma qp.
1574  */
1575 static inline int get_chroma_qp(H264Context *h, int qscale){
1576     
1577     return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1578 }
1579
1580
1581 #if 0
1582 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1583     int i;
1584     //FIXME try int temp instead of block
1585     
1586     for(i=0; i<4; i++){
1587         const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1588         const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1589         const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1590         const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1591         const int z0= d0 + d3;
1592         const int z3= d0 - d3;
1593         const int z1= d1 + d2;
1594         const int z2= d1 - d2;
1595         
1596         block[0 + 4*i]=   z0 +   z1;
1597         block[1 + 4*i]= 2*z3 +   z2;
1598         block[2 + 4*i]=   z0 -   z1;
1599         block[3 + 4*i]=   z3 - 2*z2;
1600     }    
1601
1602     for(i=0; i<4; i++){
1603         const int z0= block[0*4 + i] + block[3*4 + i];
1604         const int z3= block[0*4 + i] - block[3*4 + i];
1605         const int z1= block[1*4 + i] + block[2*4 + i];
1606         const int z2= block[1*4 + i] - block[2*4 + i];
1607         
1608         block[0*4 + i]=   z0 +   z1;
1609         block[1*4 + i]= 2*z3 +   z2;
1610         block[2*4 + i]=   z0 -   z1;
1611         block[3*4 + i]=   z3 - 2*z2;
1612     }
1613 }
1614 #endif
1615
1616 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1617 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1618 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1619     int i;
1620     const int * const quant_table= quant_coeff[qscale];
1621     const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1622     const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1623     const unsigned int threshold2= (threshold1<<1);
1624     int last_non_zero;
1625
1626     if(seperate_dc){
1627         if(qscale<=18){
1628             //avoid overflows
1629             const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1630             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1631             const unsigned int dc_threshold2= (dc_threshold1<<1);
1632
1633             int level= block[0]*quant_coeff[qscale+18][0];
1634             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1635                 if(level>0){
1636                     level= (dc_bias + level)>>(QUANT_SHIFT-2);
1637                     block[0]= level;
1638                 }else{
1639                     level= (dc_bias - level)>>(QUANT_SHIFT-2);
1640                     block[0]= -level;
1641                 }
1642 //                last_non_zero = i;
1643             }else{
1644                 block[0]=0;
1645             }
1646         }else{
1647             const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1648             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1649             const unsigned int dc_threshold2= (dc_threshold1<<1);
1650
1651             int level= block[0]*quant_table[0];
1652             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1653                 if(level>0){
1654                     level= (dc_bias + level)>>(QUANT_SHIFT+1);
1655                     block[0]= level;
1656                 }else{
1657                     level= (dc_bias - level)>>(QUANT_SHIFT+1);
1658                     block[0]= -level;
1659                 }
1660 //                last_non_zero = i;
1661             }else{
1662                 block[0]=0;
1663             }
1664         }
1665         last_non_zero= 0;
1666         i=1;
1667     }else{
1668         last_non_zero= -1;
1669         i=0;
1670     }
1671
1672     for(; i<16; i++){
1673         const int j= scantable[i];
1674         int level= block[j]*quant_table[j];
1675
1676 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1677 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1678         if(((unsigned)(level+threshold1))>threshold2){
1679             if(level>0){
1680                 level= (bias + level)>>QUANT_SHIFT;
1681                 block[j]= level;
1682             }else{
1683                 level= (bias - level)>>QUANT_SHIFT;
1684                 block[j]= -level;
1685             }
1686             last_non_zero = i;
1687         }else{
1688             block[j]=0;
1689         }
1690     }
1691
1692     return last_non_zero;
1693 }
1694
1695 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1696     const uint32_t a= ((uint32_t*)(src-stride))[0];
1697     ((uint32_t*)(src+0*stride))[0]= a;
1698     ((uint32_t*)(src+1*stride))[0]= a;
1699     ((uint32_t*)(src+2*stride))[0]= a;
1700     ((uint32_t*)(src+3*stride))[0]= a;
1701 }
1702
1703 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1704     ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1705     ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1706     ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1707     ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1708 }
1709
1710 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1711     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1712                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1713     
1714     ((uint32_t*)(src+0*stride))[0]= 
1715     ((uint32_t*)(src+1*stride))[0]= 
1716     ((uint32_t*)(src+2*stride))[0]= 
1717     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1718 }
1719
1720 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1721     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1722     
1723     ((uint32_t*)(src+0*stride))[0]= 
1724     ((uint32_t*)(src+1*stride))[0]= 
1725     ((uint32_t*)(src+2*stride))[0]= 
1726     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1727 }
1728
1729 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1730     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1731     
1732     ((uint32_t*)(src+0*stride))[0]= 
1733     ((uint32_t*)(src+1*stride))[0]= 
1734     ((uint32_t*)(src+2*stride))[0]= 
1735     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1736 }
1737
1738 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1739     ((uint32_t*)(src+0*stride))[0]= 
1740     ((uint32_t*)(src+1*stride))[0]= 
1741     ((uint32_t*)(src+2*stride))[0]= 
1742     ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1743 }
1744
1745
1746 #define LOAD_TOP_RIGHT_EDGE\
1747     const int t4= topright[0];\
1748     const int t5= topright[1];\
1749     const int t6= topright[2];\
1750     const int t7= topright[3];\
1751
1752 #define LOAD_LEFT_EDGE\
1753     const int l0= src[-1+0*stride];\
1754     const int l1= src[-1+1*stride];\
1755     const int l2= src[-1+2*stride];\
1756     const int l3= src[-1+3*stride];\
1757
1758 #define LOAD_TOP_EDGE\
1759     const int t0= src[ 0-1*stride];\
1760     const int t1= src[ 1-1*stride];\
1761     const int t2= src[ 2-1*stride];\
1762     const int t3= src[ 3-1*stride];\
1763
1764 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1765     const int lt= src[-1-1*stride];
1766     LOAD_TOP_EDGE
1767     LOAD_LEFT_EDGE
1768
1769     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1770     src[0+2*stride]=
1771     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1772     src[0+1*stride]=
1773     src[1+2*stride]=
1774     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1775     src[0+0*stride]=
1776     src[1+1*stride]=
1777     src[2+2*stride]=
1778     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1779     src[1+0*stride]=
1780     src[2+1*stride]=
1781     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1782     src[2+0*stride]=
1783     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1784     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1785 }
1786
1787 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1788     LOAD_TOP_EDGE    
1789     LOAD_TOP_RIGHT_EDGE    
1790 //    LOAD_LEFT_EDGE    
1791
1792     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1793     src[1+0*stride]=
1794     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1795     src[2+0*stride]=
1796     src[1+1*stride]=
1797     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1798     src[3+0*stride]=
1799     src[2+1*stride]=
1800     src[1+2*stride]=
1801     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1802     src[3+1*stride]=
1803     src[2+2*stride]=
1804     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1805     src[3+2*stride]=
1806     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1807     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1808 }
1809
1810 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1811     const int lt= src[-1-1*stride];
1812     LOAD_TOP_EDGE    
1813     LOAD_LEFT_EDGE    
1814     const __attribute__((unused)) int unu= l3;
1815
1816     src[0+0*stride]=
1817     src[1+2*stride]=(lt + t0 + 1)>>1;
1818     src[1+0*stride]=
1819     src[2+2*stride]=(t0 + t1 + 1)>>1;
1820     src[2+0*stride]=
1821     src[3+2*stride]=(t1 + t2 + 1)>>1;
1822     src[3+0*stride]=(t2 + t3 + 1)>>1;
1823     src[0+1*stride]=
1824     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1825     src[1+1*stride]=
1826     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1827     src[2+1*stride]=
1828     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1829     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1830     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1831     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1832 }
1833
1834 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1835     LOAD_TOP_EDGE    
1836     LOAD_TOP_RIGHT_EDGE    
1837     const __attribute__((unused)) int unu= t7;
1838
1839     src[0+0*stride]=(t0 + t1 + 1)>>1;
1840     src[1+0*stride]=
1841     src[0+2*stride]=(t1 + t2 + 1)>>1;
1842     src[2+0*stride]=
1843     src[1+2*stride]=(t2 + t3 + 1)>>1;
1844     src[3+0*stride]=
1845     src[2+2*stride]=(t3 + t4+ 1)>>1;
1846     src[3+2*stride]=(t4 + t5+ 1)>>1;
1847     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1848     src[1+1*stride]=
1849     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1850     src[2+1*stride]=
1851     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1852     src[3+1*stride]=
1853     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1854     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1855 }
1856
1857 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1858     LOAD_LEFT_EDGE    
1859
1860     src[0+0*stride]=(l0 + l1 + 1)>>1;
1861     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1862     src[2+0*stride]=
1863     src[0+1*stride]=(l1 + l2 + 1)>>1;
1864     src[3+0*stride]=
1865     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1866     src[2+1*stride]=
1867     src[0+2*stride]=(l2 + l3 + 1)>>1;
1868     src[3+1*stride]=
1869     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1870     src[3+2*stride]=
1871     src[1+3*stride]=
1872     src[0+3*stride]=
1873     src[2+2*stride]=
1874     src[2+3*stride]=
1875     src[3+3*stride]=l3;
1876 }
1877     
1878 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1879     const int lt= src[-1-1*stride];
1880     LOAD_TOP_EDGE    
1881     LOAD_LEFT_EDGE    
1882     const __attribute__((unused)) int unu= t3;
1883
1884     src[0+0*stride]=
1885     src[2+1*stride]=(lt + l0 + 1)>>1;
1886     src[1+0*stride]=
1887     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1888     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1889     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1890     src[0+1*stride]=
1891     src[2+2*stride]=(l0 + l1 + 1)>>1;
1892     src[1+1*stride]=
1893     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1894     src[0+2*stride]=
1895     src[2+3*stride]=(l1 + l2+ 1)>>1;
1896     src[1+2*stride]=
1897     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1898     src[0+3*stride]=(l2 + l3 + 1)>>1;
1899     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1900 }
1901
1902 static void pred16x16_vertical_c(uint8_t *src, int stride){
1903     int i;
1904     const uint32_t a= ((uint32_t*)(src-stride))[0];
1905     const uint32_t b= ((uint32_t*)(src-stride))[1];
1906     const uint32_t c= ((uint32_t*)(src-stride))[2];
1907     const uint32_t d= ((uint32_t*)(src-stride))[3];
1908     
1909     for(i=0; i<16; i++){
1910         ((uint32_t*)(src+i*stride))[0]= a;
1911         ((uint32_t*)(src+i*stride))[1]= b;
1912         ((uint32_t*)(src+i*stride))[2]= c;
1913         ((uint32_t*)(src+i*stride))[3]= d;
1914     }
1915 }
1916
1917 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1918     int i;
1919
1920     for(i=0; i<16; i++){
1921         ((uint32_t*)(src+i*stride))[0]=
1922         ((uint32_t*)(src+i*stride))[1]=
1923         ((uint32_t*)(src+i*stride))[2]=
1924         ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1925     }
1926 }
1927
1928 static void pred16x16_dc_c(uint8_t *src, int stride){
1929     int i, dc=0;
1930
1931     for(i=0;i<16; i++){
1932         dc+= src[-1+i*stride];
1933     }
1934     
1935     for(i=0;i<16; i++){
1936         dc+= src[i-stride];
1937     }
1938
1939     dc= 0x01010101*((dc + 16)>>5);
1940
1941     for(i=0; i<16; i++){
1942         ((uint32_t*)(src+i*stride))[0]=
1943         ((uint32_t*)(src+i*stride))[1]=
1944         ((uint32_t*)(src+i*stride))[2]=
1945         ((uint32_t*)(src+i*stride))[3]= dc;
1946     }
1947 }
1948
1949 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1950     int i, dc=0;
1951
1952     for(i=0;i<16; i++){
1953         dc+= src[-1+i*stride];
1954     }
1955     
1956     dc= 0x01010101*((dc + 8)>>4);
1957
1958     for(i=0; i<16; i++){
1959         ((uint32_t*)(src+i*stride))[0]=
1960         ((uint32_t*)(src+i*stride))[1]=
1961         ((uint32_t*)(src+i*stride))[2]=
1962         ((uint32_t*)(src+i*stride))[3]= dc;
1963     }
1964 }
1965
1966 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1967     int i, dc=0;
1968
1969     for(i=0;i<16; i++){
1970         dc+= src[i-stride];
1971     }
1972     dc= 0x01010101*((dc + 8)>>4);
1973
1974     for(i=0; i<16; i++){
1975         ((uint32_t*)(src+i*stride))[0]=
1976         ((uint32_t*)(src+i*stride))[1]=
1977         ((uint32_t*)(src+i*stride))[2]=
1978         ((uint32_t*)(src+i*stride))[3]= dc;
1979     }
1980 }
1981
1982 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1983     int i;
1984
1985     for(i=0; i<16; i++){
1986         ((uint32_t*)(src+i*stride))[0]=
1987         ((uint32_t*)(src+i*stride))[1]=
1988         ((uint32_t*)(src+i*stride))[2]=
1989         ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1990     }
1991 }
1992
1993 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1994   int i, j, k;
1995   int a;
1996   uint8_t *cm = cropTbl + MAX_NEG_CROP;
1997   const uint8_t * const src0 = src+7-stride;
1998   const uint8_t *src1 = src+8*stride-1;
1999   const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2000   int H = src0[1] - src0[-1];
2001   int V = src1[0] - src2[ 0];
2002   for(k=2; k<=8; ++k) {
2003     src1 += stride; src2 -= stride;
2004     H += k*(src0[k] - src0[-k]);
2005     V += k*(src1[0] - src2[ 0]);
2006   }
2007   if(svq3){
2008     H = ( 5*(H/4) ) / 16;
2009     V = ( 5*(V/4) ) / 16;
2010
2011     /* required for 100% accuracy */
2012     i = H; H = V; V = i;
2013   }else{
2014     H = ( 5*H+32 ) >> 6;
2015     V = ( 5*V+32 ) >> 6;
2016   }
2017
2018   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2019   for(j=16; j>0; --j) {
2020     int b = a;
2021     a += V;
2022     for(i=-16; i<0; i+=4) {
2023       src[16+i] = cm[ (b    ) >> 5 ];
2024       src[17+i] = cm[ (b+  H) >> 5 ];
2025       src[18+i] = cm[ (b+2*H) >> 5 ];
2026       src[19+i] = cm[ (b+3*H) >> 5 ];
2027       b += 4*H;
2028     }
2029     src += stride;
2030   }
2031 }
2032
2033 static void pred16x16_plane_c(uint8_t *src, int stride){
2034     pred16x16_plane_compat_c(src, stride, 0);
2035 }
2036
2037 static void pred8x8_vertical_c(uint8_t *src, int stride){
2038     int i;
2039     const uint32_t a= ((uint32_t*)(src-stride))[0];
2040     const uint32_t b= ((uint32_t*)(src-stride))[1];
2041     
2042     for(i=0; i<8; i++){
2043         ((uint32_t*)(src+i*stride))[0]= a;
2044         ((uint32_t*)(src+i*stride))[1]= b;
2045     }
2046 }
2047
2048 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2049     int i;
2050
2051     for(i=0; i<8; i++){
2052         ((uint32_t*)(src+i*stride))[0]=
2053         ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2054     }
2055 }
2056
2057 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2058     int i;
2059
2060     for(i=0; i<4; i++){
2061         ((uint32_t*)(src+i*stride))[0]= 
2062         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2063     }
2064     for(i=4; i<8; i++){
2065         ((uint32_t*)(src+i*stride))[0]= 
2066         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2067     }
2068 }
2069
2070 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2071     int i;
2072     int dc0, dc2;
2073
2074     dc0=dc2=0;
2075     for(i=0;i<4; i++){
2076         dc0+= src[-1+i*stride];
2077         dc2+= src[-1+(i+4)*stride];
2078     }
2079     dc0= 0x01010101*((dc0 + 2)>>2);
2080     dc2= 0x01010101*((dc2 + 2)>>2);
2081
2082     for(i=0; i<4; i++){
2083         ((uint32_t*)(src+i*stride))[0]=
2084         ((uint32_t*)(src+i*stride))[1]= dc0;
2085     }
2086     for(i=4; i<8; i++){
2087         ((uint32_t*)(src+i*stride))[0]=
2088         ((uint32_t*)(src+i*stride))[1]= dc2;
2089     }
2090 }
2091
2092 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2093     int i;
2094     int dc0, dc1;
2095
2096     dc0=dc1=0;
2097     for(i=0;i<4; i++){
2098         dc0+= src[i-stride];
2099         dc1+= src[4+i-stride];
2100     }
2101     dc0= 0x01010101*((dc0 + 2)>>2);
2102     dc1= 0x01010101*((dc1 + 2)>>2);
2103
2104     for(i=0; i<4; i++){
2105         ((uint32_t*)(src+i*stride))[0]= dc0;
2106         ((uint32_t*)(src+i*stride))[1]= dc1;
2107     }
2108     for(i=4; i<8; i++){
2109         ((uint32_t*)(src+i*stride))[0]= dc0;
2110         ((uint32_t*)(src+i*stride))[1]= dc1;
2111     }
2112 }
2113
2114
2115 static void pred8x8_dc_c(uint8_t *src, int stride){
2116     int i;
2117     int dc0, dc1, dc2, dc3;
2118
2119     dc0=dc1=dc2=0;
2120     for(i=0;i<4; i++){
2121         dc0+= src[-1+i*stride] + src[i-stride];
2122         dc1+= src[4+i-stride];
2123         dc2+= src[-1+(i+4)*stride];
2124     }
2125     dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2126     dc0= 0x01010101*((dc0 + 4)>>3);
2127     dc1= 0x01010101*((dc1 + 2)>>2);
2128     dc2= 0x01010101*((dc2 + 2)>>2);
2129
2130     for(i=0; i<4; i++){
2131         ((uint32_t*)(src+i*stride))[0]= dc0;
2132         ((uint32_t*)(src+i*stride))[1]= dc1;
2133     }
2134     for(i=4; i<8; i++){
2135         ((uint32_t*)(src+i*stride))[0]= dc2;
2136         ((uint32_t*)(src+i*stride))[1]= dc3;
2137     }
2138 }
2139
2140 static void pred8x8_plane_c(uint8_t *src, int stride){
2141   int j, k;
2142   int a;
2143   uint8_t *cm = cropTbl + MAX_NEG_CROP;
2144   const uint8_t * const src0 = src+3-stride;
2145   const uint8_t *src1 = src+4*stride-1;
2146   const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2147   int H = src0[1] - src0[-1];
2148   int V = src1[0] - src2[ 0];
2149   for(k=2; k<=4; ++k) {
2150     src1 += stride; src2 -= stride;
2151     H += k*(src0[k] - src0[-k]);
2152     V += k*(src1[0] - src2[ 0]);
2153   }
2154   H = ( 17*H+16 ) >> 5;
2155   V = ( 17*V+16 ) >> 5;
2156
2157   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2158   for(j=8; j>0; --j) {
2159     int b = a;
2160     a += V;
2161     src[0] = cm[ (b    ) >> 5 ];
2162     src[1] = cm[ (b+  H) >> 5 ];
2163     src[2] = cm[ (b+2*H) >> 5 ];
2164     src[3] = cm[ (b+3*H) >> 5 ];
2165     src[4] = cm[ (b+4*H) >> 5 ];
2166     src[5] = cm[ (b+5*H) >> 5 ];
2167     src[6] = cm[ (b+6*H) >> 5 ];
2168     src[7] = cm[ (b+7*H) >> 5 ];
2169     src += stride;
2170   }
2171 }
2172
2173 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2174                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2175                            int src_x_offset, int src_y_offset,
2176                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2177     MpegEncContext * const s = &h->s;
2178     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2179     const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2180     const int luma_xy= (mx&3) + ((my&3)<<2);
2181     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2182     uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2183     uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2184     int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2185     int extra_height= extra_width;
2186     int emu=0;
2187     const int full_mx= mx>>2;
2188     const int full_my= my>>2;
2189     
2190     assert(pic->data[0]);
2191     
2192     if(mx&7) extra_width -= 3;
2193     if(my&7) extra_height -= 3;
2194     
2195     if(   full_mx < 0-extra_width 
2196        || full_my < 0-extra_height 
2197        || full_mx + 16/*FIXME*/ > s->width + extra_width 
2198        || full_my + 16/*FIXME*/ > s->height + extra_height){
2199         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);
2200             src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2201         emu=1;
2202     }
2203     
2204     qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2205     if(!square){
2206         qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2207     }
2208     
2209     if(s->flags&CODEC_FLAG_GRAY) return;
2210     
2211     if(emu){
2212         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);
2213             src_cb= s->edge_emu_buffer;
2214     }
2215     chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2216
2217     if(emu){
2218         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);
2219             src_cr= s->edge_emu_buffer;
2220     }
2221     chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2222 }
2223
2224 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2225                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2226                            int x_offset, int y_offset,
2227                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2228                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2229                            int list0, int list1){
2230     MpegEncContext * const s = &h->s;
2231     qpel_mc_func *qpix_op=  qpix_put;
2232     h264_chroma_mc_func chroma_op= chroma_put;
2233     
2234     dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2235     dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2236     dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2237     x_offset += 8*s->mb_x;
2238     y_offset += 8*s->mb_y;
2239     
2240     if(list0){
2241         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2242         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2243                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
2244                            qpix_op, chroma_op);
2245
2246         qpix_op=  qpix_avg;
2247         chroma_op= chroma_avg;
2248     }
2249
2250     if(list1){
2251         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2252         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2253                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
2254                            qpix_op, chroma_op);
2255     }
2256 }
2257
2258 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2259                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2260                            int x_offset, int y_offset,
2261                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2262                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2263                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2264                            int list0, int list1){
2265     MpegEncContext * const s = &h->s;
2266
2267     dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2268     dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2269     dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2270     x_offset += 8*s->mb_x;
2271     y_offset += 8*s->mb_y;
2272     
2273     if(list0 && list1){
2274         /* don't optimize for luma-only case, since B-frames usually
2275          * use implicit weights => chroma too. */
2276         uint8_t *tmp_cb = s->obmc_scratchpad;
2277         uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2278         uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2279         int refn0 = h->ref_cache[0][ scan8[n] ];
2280         int refn1 = h->ref_cache[1][ scan8[n] ];
2281
2282         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2283                     dest_y, dest_cb, dest_cr,
2284                     x_offset, y_offset, qpix_put, chroma_put);
2285         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2286                     tmp_y, tmp_cb, tmp_cr,
2287                     x_offset, y_offset, qpix_put, chroma_put);
2288
2289         if(h->use_weight == 2){
2290             int weight0 = h->implicit_weight[refn0][refn1];
2291             int weight1 = 64 - weight0;
2292             luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2293             chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2294             chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2295         }else{
2296             luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2297                             h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2298                             h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2299             chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2300                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2301                             h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2302             chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2303                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2304                             h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2305         }
2306     }else{
2307         int list = list1 ? 1 : 0;
2308         int refn = h->ref_cache[list][ scan8[n] ];
2309         Picture *ref= &h->ref_list[list][refn];
2310         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2311                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
2312                     qpix_put, chroma_put);
2313
2314         luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2315                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
2316         if(h->use_weight_chroma){
2317             chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2318                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2319             chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2320                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2321         }
2322     }
2323 }
2324
2325 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2326                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2327                            int x_offset, int y_offset,
2328                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2329                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2330                            h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2331                            int list0, int list1){
2332     if((h->use_weight==2 && list0 && list1
2333         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2334        || h->use_weight==1)
2335         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2336                          x_offset, y_offset, qpix_put, chroma_put,
2337                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2338     else
2339         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2340                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2341 }
2342
2343 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2344                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2345                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2346                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2347     MpegEncContext * const s = &h->s;
2348     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2349     const int mb_type= s->current_picture.mb_type[mb_xy];
2350     
2351     assert(IS_INTER(mb_type));
2352     
2353     if(IS_16X16(mb_type)){
2354         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2355                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2356                 &weight_op[0], &weight_avg[0],
2357                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2358     }else if(IS_16X8(mb_type)){
2359         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2360                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2361                 &weight_op[1], &weight_avg[1],
2362                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2363         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2364                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2365                 &weight_op[1], &weight_avg[1],
2366                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2367     }else if(IS_8X16(mb_type)){
2368         mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2369                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2370                 &weight_op[2], &weight_avg[2],
2371                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2372         mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2373                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2374                 &weight_op[2], &weight_avg[2],
2375                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2376     }else{
2377         int i;
2378         
2379         assert(IS_8X8(mb_type));
2380
2381         for(i=0; i<4; i++){
2382             const int sub_mb_type= h->sub_mb_type[i];
2383             const int n= 4*i;
2384             int x_offset= (i&1)<<2;
2385             int y_offset= (i&2)<<1;
2386
2387             if(IS_SUB_8X8(sub_mb_type)){
2388                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2389                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2390                     &weight_op[3], &weight_avg[3],
2391                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2392             }else if(IS_SUB_8X4(sub_mb_type)){
2393                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2394                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2395                     &weight_op[4], &weight_avg[4],
2396                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2397                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2398                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2399                     &weight_op[4], &weight_avg[4],
2400                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2401             }else if(IS_SUB_4X8(sub_mb_type)){
2402                 mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2403                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2404                     &weight_op[5], &weight_avg[5],
2405                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2406                 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2407                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2408                     &weight_op[5], &weight_avg[5],
2409                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2410             }else{
2411                 int j;
2412                 assert(IS_SUB_4X4(sub_mb_type));
2413                 for(j=0; j<4; j++){
2414                     int sub_x_offset= x_offset + 2*(j&1);
2415                     int sub_y_offset= y_offset +   (j&2);
2416                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2417                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2418                         &weight_op[6], &weight_avg[6],
2419                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2420                 }
2421             }
2422         }
2423     }
2424 }
2425
2426 static void decode_init_vlc(H264Context *h){
2427     static int done = 0;
2428
2429     if (!done) {
2430         int i;
2431         done = 1;
2432
2433         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2434                  &chroma_dc_coeff_token_len [0], 1, 1,
2435                  &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2436
2437         for(i=0; i<4; i++){
2438             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2439                      &coeff_token_len [i][0], 1, 1,
2440                      &coeff_token_bits[i][0], 1, 1, 1);
2441         }
2442
2443         for(i=0; i<3; i++){
2444             init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2445                      &chroma_dc_total_zeros_len [i][0], 1, 1,
2446                      &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2447         }
2448         for(i=0; i<15; i++){
2449             init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2450                      &total_zeros_len [i][0], 1, 1,
2451                      &total_zeros_bits[i][0], 1, 1, 1);
2452         }
2453
2454         for(i=0; i<6; i++){
2455             init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2456                      &run_len [i][0], 1, 1,
2457                      &run_bits[i][0], 1, 1, 1);
2458         }
2459         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2460                  &run_len [6][0], 1, 1,
2461                  &run_bits[6][0], 1, 1, 1);
2462     }
2463 }
2464
2465 /**
2466  * Sets the intra prediction function pointers.
2467  */
2468 static void init_pred_ptrs(H264Context *h){
2469 //    MpegEncContext * const s = &h->s;
2470
2471     h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2472     h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2473     h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2474     h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2475     h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2476     h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2477     h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2478     h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2479     h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2480     h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2481     h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2482     h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2483
2484     h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2485     h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2486     h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2487     h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2488     h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2489     h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2490     h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2491
2492     h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2493     h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2494     h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2495     h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2496     h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2497     h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2498     h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2499 }
2500
2501 static void free_tables(H264Context *h){
2502     av_freep(&h->intra4x4_pred_mode);
2503     av_freep(&h->chroma_pred_mode_table);
2504     av_freep(&h->cbp_table);
2505     av_freep(&h->mvd_table[0]);
2506     av_freep(&h->mvd_table[1]);
2507     av_freep(&h->direct_table);
2508     av_freep(&h->non_zero_count);
2509     av_freep(&h->slice_table_base);
2510     av_freep(&h->top_border);
2511     h->slice_table= NULL;
2512
2513     av_freep(&h->mb2b_xy);
2514     av_freep(&h->mb2b8_xy);
2515
2516     av_freep(&h->s.obmc_scratchpad);
2517 }
2518
2519 /**
2520  * allocates tables.
2521  * needs widzh/height
2522  */
2523 static int alloc_tables(H264Context *h){
2524     MpegEncContext * const s = &h->s;
2525     const int big_mb_num= s->mb_stride * (s->mb_height+1);
2526     int x,y;
2527
2528     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2529
2530     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2531     CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2532     CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2533     CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2534
2535     if( h->pps.cabac ) {
2536         CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2537         CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2538         CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2539         CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2540     }
2541
2542     memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2543     h->slice_table= h->slice_table_base + s->mb_stride + 1;
2544
2545     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2546     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2547     for(y=0; y<s->mb_height; y++){
2548         for(x=0; x<s->mb_width; x++){
2549             const int mb_xy= x + y*s->mb_stride;
2550             const int b_xy = 4*x + 4*y*h->b_stride;
2551             const int b8_xy= 2*x + 2*y*h->b8_stride;
2552         
2553             h->mb2b_xy [mb_xy]= b_xy;
2554             h->mb2b8_xy[mb_xy]= b8_xy;
2555         }
2556     }
2557
2558     s->obmc_scratchpad = NULL;
2559
2560     return 0;
2561 fail:
2562     free_tables(h);
2563     return -1;
2564 }
2565
2566 static void common_init(H264Context *h){
2567     MpegEncContext * const s = &h->s;
2568
2569     s->width = s->avctx->width;
2570     s->height = s->avctx->height;
2571     s->codec_id= s->avctx->codec->id;
2572     
2573     init_pred_ptrs(h);
2574
2575     s->unrestricted_mv=1;
2576     s->decode=1; //FIXME
2577 }
2578
2579 static int decode_init(AVCodecContext *avctx){
2580     H264Context *h= avctx->priv_data;
2581     MpegEncContext * const s = &h->s;
2582
2583     MPV_decode_defaults(s);
2584     
2585     s->avctx = avctx;
2586     common_init(h);
2587
2588     s->out_format = FMT_H264;
2589     s->workaround_bugs= avctx->workaround_bugs;
2590
2591     // set defaults
2592 //    s->decode_mb= ff_h263_decode_mb;
2593     s->low_delay= 1;
2594     avctx->pix_fmt= PIX_FMT_YUV420P;
2595
2596     decode_init_vlc(h);
2597     
2598     if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2599         h->is_avc = 0;
2600     else {
2601         if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2602             av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2603             return -1;
2604         }
2605         h->is_avc = 1;
2606         h->got_avcC = 0;
2607     }
2608
2609     return 0;
2610 }
2611
2612 static void frame_start(H264Context *h){
2613     MpegEncContext * const s = &h->s;
2614     int i;
2615
2616     MPV_frame_start(s, s->avctx);
2617     ff_er_frame_start(s);
2618
2619     assert(s->linesize && s->uvlinesize);
2620
2621     for(i=0; i<16; i++){
2622         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2623         h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2624     }
2625     for(i=0; i<4; i++){
2626         h->block_offset[16+i]=
2627         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2628     }
2629
2630     /* can't be in alloc_tables because linesize isn't known there.
2631      * FIXME: redo bipred weight to not require extra buffer? */
2632     if(!s->obmc_scratchpad)
2633         s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2634
2635 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2636 }
2637
2638 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2639     MpegEncContext * const s = &h->s;
2640     int i;
2641     
2642     src_y  -=   linesize;
2643     src_cb -= uvlinesize;
2644     src_cr -= uvlinesize;
2645
2646     h->left_border[0]= h->top_border[s->mb_x][15];
2647     for(i=1; i<17; i++){
2648         h->left_border[i]= src_y[15+i*  linesize];
2649     }
2650     
2651     *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2652     *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2653
2654     if(!(s->flags&CODEC_FLAG_GRAY)){
2655         h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2656         h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2657         for(i=1; i<9; i++){
2658             h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2659             h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2660         }
2661         *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2662         *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2663     }
2664 }
2665
2666 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){
2667     MpegEncContext * const s = &h->s;
2668     int temp8, i;
2669     uint64_t temp64;
2670     int deblock_left = (s->mb_x > 0);
2671     int deblock_top  = (s->mb_y > 0);
2672
2673     src_y  -=   linesize + 1;
2674     src_cb -= uvlinesize + 1;
2675     src_cr -= uvlinesize + 1;
2676
2677 #define XCHG(a,b,t,xchg)\
2678 t= a;\
2679 if(xchg)\
2680     a= b;\
2681 b= t;
2682
2683     if(deblock_left){
2684         for(i = !deblock_top; i<17; i++){
2685             XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2686         }
2687     }
2688
2689     if(deblock_top){
2690         XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2691         XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2692     }
2693
2694     if(!(s->flags&CODEC_FLAG_GRAY)){
2695         if(deblock_left){
2696             for(i = !deblock_top; i<9; i++){
2697                 XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2698                 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2699             }
2700         }
2701         if(deblock_top){
2702             XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2703             XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2704         }
2705     }
2706 }
2707
2708 static void hl_decode_mb(H264Context *h){
2709     MpegEncContext * const s = &h->s;
2710     const int mb_x= s->mb_x;
2711     const int mb_y= s->mb_y;
2712     const int mb_xy= mb_x + mb_y*s->mb_stride;
2713     const int mb_type= s->current_picture.mb_type[mb_xy];
2714     uint8_t  *dest_y, *dest_cb, *dest_cr;
2715     int linesize, uvlinesize /*dct_offset*/;
2716     int i;
2717
2718     if(!s->decode)
2719         return;
2720
2721     if(s->mb_skiped){
2722     }
2723
2724     dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2725     dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2726     dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2727
2728     if (h->mb_field_decoding_flag) {
2729         linesize = s->linesize * 2;
2730         uvlinesize = s->uvlinesize * 2;
2731         if(mb_y&1){ //FIXME move out of this func?
2732             dest_y -= s->linesize*15;
2733             dest_cb-= s->linesize*7;
2734             dest_cr-= s->linesize*7;
2735         }
2736     } else {
2737         linesize = s->linesize;
2738         uvlinesize = s->uvlinesize;
2739 //        dct_offset = s->linesize * 16;
2740     }
2741
2742     if (IS_INTRA_PCM(mb_type)) {
2743         unsigned int x, y;
2744
2745         // The pixels are stored in h->mb array in the same order as levels,
2746         // copy them in output in the correct order.
2747         for(i=0; i<16; i++) {
2748             for (y=0; y<4; y++) {
2749                 for (x=0; x<4; x++) {
2750                     *(dest_y + h->block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2751                 }
2752             }
2753         }
2754         for(i=16; i<16+4; i++) {
2755             for (y=0; y<4; y++) {
2756                 for (x=0; x<4; x++) {
2757                     *(dest_cb + h->block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2758                 }
2759             }
2760         }
2761         for(i=20; i<20+4; i++) {
2762             for (y=0; y<4; y++) {
2763                 for (x=0; x<4; x++) {
2764                     *(dest_cr + h->block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2765                 }
2766             }
2767         }
2768     } else {
2769         if(IS_INTRA(mb_type)){
2770             if(h->deblocking_filter)
2771                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2772
2773             if(!(s->flags&CODEC_FLAG_GRAY)){
2774                 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2775                 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2776             }
2777
2778             if(IS_INTRA4x4(mb_type)){
2779                 if(!s->encoding){
2780                     for(i=0; i<16; i++){
2781                         uint8_t * const ptr= dest_y + h->block_offset[i];
2782                         uint8_t *topright;
2783                         const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2784                         int tr;
2785
2786                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2787                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2788                             assert(mb_y || linesize <= h->block_offset[i]);
2789                             if(!topright_avail){
2790                                 tr= ptr[3 - linesize]*0x01010101;
2791                                 topright= (uint8_t*) &tr;
2792                             }else if(i==5 && h->deblocking_filter){
2793                                 tr= *(uint32_t*)h->top_border[mb_x+1];
2794                                 topright= (uint8_t*) &tr;
2795                             }else
2796                                 topright= ptr + 4 - linesize;
2797                         }else
2798                             topright= NULL;
2799
2800                         h->pred4x4[ dir ](ptr, topright, linesize);
2801                         if(h->non_zero_count_cache[ scan8[i] ]){
2802                             if(s->codec_id == CODEC_ID_H264)
2803                                 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2804                             else
2805                                 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2806                         }
2807                     }
2808                 }
2809             }else{
2810                 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2811                 if(s->codec_id == CODEC_ID_H264)
2812                     h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2813                 else
2814                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2815             }
2816             if(h->deblocking_filter)
2817                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2818         }else if(s->codec_id == CODEC_ID_H264){
2819             hl_motion(h, dest_y, dest_cb, dest_cr,
2820                       s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2821                       s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2822                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2823         }
2824
2825
2826         if(!IS_INTRA4x4(mb_type)){
2827             if(s->codec_id == CODEC_ID_H264){
2828                 for(i=0; i<16; i++){
2829                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2830                         uint8_t * const ptr= dest_y + h->block_offset[i];
2831                         s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2832                     }
2833                 }
2834             }else{
2835                 for(i=0; i<16; i++){
2836                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2837                         uint8_t * const ptr= dest_y + h->block_offset[i];
2838                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2839                     }
2840                 }
2841             }
2842         }
2843
2844         if(!(s->flags&CODEC_FLAG_GRAY)){
2845             chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2846             chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2847             if(s->codec_id == CODEC_ID_H264){
2848                 for(i=16; i<16+4; i++){
2849                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2850                         uint8_t * const ptr= dest_cb + h->block_offset[i];
2851                         s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2852                     }
2853                 }
2854                 for(i=20; i<20+4; i++){
2855                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2856                         uint8_t * const ptr= dest_cr + h->block_offset[i];
2857                         s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2858                     }
2859                 }
2860             }else{
2861                 for(i=16; i<16+4; i++){
2862                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2863                         uint8_t * const ptr= dest_cb + h->block_offset[i];
2864                         svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2865                     }
2866                 }
2867                 for(i=20; i<20+4; i++){
2868                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2869                         uint8_t * const ptr= dest_cr + h->block_offset[i];
2870                         svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2871                     }
2872                 }
2873             }
2874         }
2875     }
2876     if(h->deblocking_filter) {
2877         backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2878         fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
2879         filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2880     }
2881 }
2882
2883 /**
2884  * fills the default_ref_list.
2885  */
2886 static int fill_default_ref_list(H264Context *h){
2887     MpegEncContext * const s = &h->s;
2888     int i;
2889     int smallest_poc_greater_than_current = -1;
2890     Picture sorted_short_ref[16];
2891     
2892     if(h->slice_type==B_TYPE){
2893         int out_i;
2894         int limit= -1;
2895
2896         /* sort frame according to poc in B slice */
2897         for(out_i=0; out_i<h->short_ref_count; out_i++){
2898             int best_i=-1;
2899             int best_poc=INT_MAX;
2900
2901             for(i=0; i<h->short_ref_count; i++){
2902                 const int poc= h->short_ref[i]->poc;
2903                 if(poc > limit && poc < best_poc){
2904                     best_poc= poc;
2905                     best_i= i;
2906                 }
2907             }
2908             
2909             assert(best_i != -1);
2910             
2911             limit= best_poc;
2912             sorted_short_ref[out_i]= *h->short_ref[best_i];
2913             tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
2914             if (-1 == smallest_poc_greater_than_current) {
2915                 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
2916                     smallest_poc_greater_than_current = out_i;
2917                 }
2918             }
2919         }
2920     }
2921
2922     if(s->picture_structure == PICT_FRAME){
2923         if(h->slice_type==B_TYPE){
2924             int list;
2925             tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
2926
2927             // find the largest poc
2928             for(list=0; list<2; list++){
2929                 int index = 0;
2930                 int j= -99;
2931                 int step= list ? -1 : 1;
2932
2933                 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
2934                     while(j<0 || j>= h->short_ref_count){
2935                         step = -step;
2936                         j= smallest_poc_greater_than_current + (step>>1);
2937                     }
2938                     if(sorted_short_ref[j].reference != 3) continue;
2939                     h->default_ref_list[list][index  ]= sorted_short_ref[j];
2940                     h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
2941                 }
2942
2943                 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
2944                     if(h->long_ref[i] == NULL) continue;
2945                     if(h->long_ref[i]->reference != 3) continue;
2946
2947                     h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2948                     h->default_ref_list[ list ][index++].pic_id= i;;
2949                 }
2950                 
2951                 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
2952                     // swap the two first elements of L1 when
2953                     // L0 and L1 are identical
2954                     Picture temp= h->default_ref_list[1][0];
2955                     h->default_ref_list[1][0] = h->default_ref_list[1][1];
2956                     h->default_ref_list[1][0] = temp;
2957                 }
2958
2959                 if(index < h->ref_count[ list ])
2960                     memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2961             }
2962         }else{
2963             int index=0;
2964             for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2965                 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2966                 h->default_ref_list[0][index  ]= *h->short_ref[i];
2967                 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2968             }
2969             for(i = 0; i < 16 && index < h->ref_count[0]; i++){
2970                 if(h->long_ref[i] == NULL) continue;
2971                 if(h->long_ref[i]->reference != 3) continue;
2972                 h->default_ref_list[0][index  ]= *h->long_ref[i];
2973                 h->default_ref_list[0][index++].pic_id= i;;
2974             }
2975             if(index < h->ref_count[0])
2976                 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2977         }
2978     }else{ //FIELD
2979         if(h->slice_type==B_TYPE){
2980         }else{
2981             //FIXME second field balh
2982         }
2983     }
2984 #ifdef TRACE
2985     for (i=0; i<h->ref_count[0]; i++) {
2986         tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
2987     }
2988     if(h->slice_type==B_TYPE){
2989         for (i=0; i<h->ref_count[1]; i++) {
2990             tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
2991         }
2992     }
2993 #endif
2994     return 0;
2995 }
2996
2997 static void print_short_term(H264Context *h);
2998 static void print_long_term(H264Context *h);
2999
3000 static int decode_ref_pic_list_reordering(H264Context *h){
3001     MpegEncContext * const s = &h->s;
3002     int list;
3003     
3004     print_short_term(h);
3005     print_long_term(h);
3006     if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
3007     
3008     for(list=0; list<2; list++){
3009         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3010
3011         if(get_bits1(&s->gb)){
3012             int pred= h->curr_pic_num;
3013             int index;
3014
3015             for(index=0; ; index++){
3016                 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3017                 int pic_id;
3018                 int i;
3019                 
3020                 if(reordering_of_pic_nums_idc==3) 
3021                     break;
3022                 
3023                 if(index >= h->ref_count[list]){
3024                     av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3025                     return -1;
3026                 }
3027                 
3028                 if(reordering_of_pic_nums_idc<3){
3029                     if(reordering_of_pic_nums_idc<2){
3030                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3031
3032                         if(abs_diff_pic_num >= h->max_pic_num){
3033                             av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3034                             return -1;
3035                         }
3036
3037                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3038                         else                                pred+= abs_diff_pic_num;
3039                         pred &= h->max_pic_num - 1;
3040                     
3041                         for(i= h->ref_count[list]-1; i>=0; i--){
3042                             if(h->ref_list[list][i].data[0] != NULL && h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0) // ignore non existing pictures by testing data[0] pointer
3043                                 break;
3044                         }
3045                     }else{
3046                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3047
3048                         for(i= h->ref_count[list]-1; i>=0; i--){
3049                             if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1) // no need to ignore non existing pictures as non existing pictures have long_ref==0
3050                                 break;
3051                         }
3052                     }
3053
3054                     if (i < 0) {
3055                         av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3056                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3057                     } else if (i != index) /* this test is not necessary, it is only an optimisation to skip double copy of Picture structure in this case */ {
3058                         Picture tmp= h->ref_list[list][i];
3059                         if (i < index) {
3060                             i = h->ref_count[list];
3061                         }
3062                         for(; i > index; i--){
3063                             h->ref_list[list][i]= h->ref_list[list][i-1];
3064                         }
3065                         h->ref_list[list][index]= tmp;
3066                     }
3067                 }else{
3068                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3069                     return -1;
3070                 }
3071             }
3072         }
3073
3074         if(h->slice_type!=B_TYPE) break;
3075     }
3076     
3077     if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3078         direct_dist_scale_factor(h);
3079     return 0;    
3080 }
3081
3082 static int pred_weight_table(H264Context *h){
3083     MpegEncContext * const s = &h->s;
3084     int list, i;
3085     int luma_def, chroma_def;
3086     
3087     h->use_weight= 0;
3088     h->use_weight_chroma= 0;
3089     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3090     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3091     luma_def = 1<<h->luma_log2_weight_denom;
3092     chroma_def = 1<<h->chroma_log2_weight_denom;
3093
3094     for(list=0; list<2; list++){
3095         for(i=0; i<h->ref_count[list]; i++){
3096             int luma_weight_flag, chroma_weight_flag;
3097             
3098             luma_weight_flag= get_bits1(&s->gb);
3099             if(luma_weight_flag){
3100                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3101                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3102                 if(   h->luma_weight[list][i] != luma_def
3103                    || h->luma_offset[list][i] != 0)
3104                     h->use_weight= 1;
3105             }else{
3106                 h->luma_weight[list][i]= luma_def;
3107                 h->luma_offset[list][i]= 0;
3108             }
3109
3110             chroma_weight_flag= get_bits1(&s->gb);
3111             if(chroma_weight_flag){
3112                 int j;
3113                 for(j=0; j<2; j++){
3114                     h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3115                     h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3116                     if(   h->chroma_weight[list][i][j] != chroma_def
3117                        || h->chroma_offset[list][i][j] != 0)
3118                         h->use_weight_chroma= 1;
3119                 }
3120             }else{
3121                 int j;
3122                 for(j=0; j<2; j++){
3123                     h->chroma_weight[list][i][j]= chroma_def;
3124                     h->chroma_offset[list][i][j]= 0;
3125                 }
3126             }
3127         }
3128         if(h->slice_type != B_TYPE) break;
3129     }
3130     h->use_weight= h->use_weight || h->use_weight_chroma;
3131     return 0;
3132 }
3133
3134 static void implicit_weight_table(H264Context *h){
3135     MpegEncContext * const s = &h->s;
3136     int ref0, ref1;
3137     int cur_poc = s->current_picture_ptr->poc;
3138
3139     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3140        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3141         h->use_weight= 0;
3142         h->use_weight_chroma= 0;
3143         return;
3144     }
3145
3146     h->use_weight= 2;
3147     h->use_weight_chroma= 2;
3148     h->luma_log2_weight_denom= 5;
3149     h->chroma_log2_weight_denom= 5;
3150
3151     /* FIXME: MBAFF */
3152     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3153         int poc0 = h->ref_list[0][ref0].poc;
3154         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3155             int poc1 = h->ref_list[1][ref1].poc;
3156             int td = clip(poc1 - poc0, -128, 127);
3157             if(td){
3158                 int tb = clip(cur_poc - poc0, -128, 127);
3159                 int tx = (16384 + (ABS(td) >> 1)) / td;
3160                 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3161                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3162                     h->implicit_weight[ref0][ref1] = 32;
3163                 else
3164                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3165             }else
3166                 h->implicit_weight[ref0][ref1] = 32;
3167         }
3168     }
3169 }
3170
3171 /**
3172  * instantaneous decoder refresh.
3173  */
3174 static void idr(H264Context *h){
3175     int i,j;
3176
3177 #define CHECK_DELAY(pic) \
3178     for(j = 0; h->delayed_pic[j]; j++) \
3179         if(pic == h->delayed_pic[j]){ \
3180             pic->reference=1; \
3181             break; \
3182         }
3183
3184     for(i=0; i<16; i++){
3185         if (h->long_ref[i] != NULL) {
3186             h->long_ref[i]->reference=0;
3187             CHECK_DELAY(h->long_ref[i]);
3188             h->long_ref[i]= NULL;
3189         }
3190     }
3191     h->long_ref_count=0;
3192
3193     for(i=0; i<h->short_ref_count; i++){
3194         h->short_ref[i]->reference=0;
3195         CHECK_DELAY(h->short_ref[i]);
3196         h->short_ref[i]= NULL;
3197     }
3198     h->short_ref_count=0;
3199 }
3200 #undef CHECK_DELAY
3201
3202 /**
3203  *
3204  * @return the removed picture or NULL if an error occures
3205  */
3206 static Picture * remove_short(H264Context *h, int frame_num){
3207     MpegEncContext * const s = &h->s;
3208     int i;
3209     
3210     if(s->avctx->debug&FF_DEBUG_MMCO)
3211         av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3212     
3213     for(i=0; i<h->short_ref_count; i++){
3214         Picture *pic= h->short_ref[i];
3215         if(s->avctx->debug&FF_DEBUG_MMCO)
3216             av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3217         if(pic->frame_num == frame_num){
3218             h->short_ref[i]= NULL;
3219             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3220             h->short_ref_count--;
3221             return pic;
3222         }
3223     }
3224     return NULL;
3225 }
3226
3227 /**
3228  *
3229  * @return the removed picture or NULL if an error occures
3230  */
3231 static Picture * remove_long(H264Context *h, int i){
3232     Picture *pic;
3233
3234     pic= h->long_ref[i];
3235     h->long_ref[i]= NULL;
3236     if(pic) h->long_ref_count--;
3237
3238     return pic;
3239 }
3240
3241 /**
3242  * print short term list
3243  */
3244 static void print_short_term(H264Context *h) {
3245     uint32_t i;
3246     if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3247         av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3248         for(i=0; i<h->short_ref_count; i++){
3249             Picture *pic= h->short_ref[i];
3250             av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3251         }
3252     }
3253 }
3254
3255 /**
3256  * print long term list
3257  */
3258 static void print_long_term(H264Context *h) {
3259     uint32_t i;
3260     if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3261         av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3262         for(i = 0; i < 16; i++){
3263             Picture *pic= h->long_ref[i];
3264             if (pic) {
3265                 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3266             }
3267         }
3268     }
3269 }
3270
3271 /**
3272  * Executes the reference picture marking (memory management control operations).
3273  */
3274 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3275     MpegEncContext * const s = &h->s;
3276     int i, j;
3277     int current_is_long=0;
3278     Picture *pic;
3279     
3280     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3281         av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3282         
3283     for(i=0; i<mmco_count; i++){
3284         if(s->avctx->debug&FF_DEBUG_MMCO)
3285             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);
3286
3287         switch(mmco[i].opcode){
3288         case MMCO_SHORT2UNUSED:
3289             pic= remove_short(h, mmco[i].short_frame_num);
3290             if(pic==NULL) return -1;
3291             pic->reference= 0;
3292             break;
3293         case MMCO_SHORT2LONG:
3294             pic= remove_long(h, mmco[i].long_index);
3295             if(pic) pic->reference=0;
3296             
3297             h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3298             h->long_ref[ mmco[i].long_index ]->long_ref=1;
3299             h->long_ref_count++;
3300             break;
3301         case MMCO_LONG2UNUSED:
3302             pic= remove_long(h, mmco[i].long_index);
3303             if(pic==NULL) return -1;
3304             pic->reference= 0;
3305             break;
3306         case MMCO_LONG:
3307             pic= remove_long(h, mmco[i].long_index);
3308             if(pic) pic->reference=0;
3309             
3310             h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3311             h->long_ref[ mmco[i].long_index ]->long_ref=1;
3312             h->long_ref_count++;
3313             
3314             current_is_long=1;
3315             break;
3316         case MMCO_SET_MAX_LONG:
3317             assert(mmco[i].long_index <= 16);
3318             // just remove the long term which index is greater than new max
3319             for(j = mmco[i].long_index; j<16; j++){
3320                 pic = remove_long(h, j);
3321                 if (pic) pic->reference=0;
3322             }
3323             break;
3324         case MMCO_RESET:
3325             while(h->short_ref_count){
3326                 pic= remove_short(h, h->short_ref[0]->frame_num);
3327                 pic->reference=0;
3328             }
3329             for(j = 0; j < 16; j++) {
3330                 pic= remove_long(h, j);
3331                 if(pic) pic->reference=0;
3332             }
3333             break;
3334         default: assert(0);
3335         }
3336     }
3337     
3338     if(!current_is_long){
3339         pic= remove_short(h, s->current_picture_ptr->frame_num);
3340         if(pic){
3341             pic->reference=0;
3342             av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3343         }
3344         
3345         if(h->short_ref_count)
3346             memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3347
3348         h->short_ref[0]= s->current_picture_ptr;
3349         h->short_ref[0]->long_ref=0;
3350         h->short_ref_count++;
3351     }
3352     
3353     print_short_term(h);
3354     print_long_term(h);
3355     return 0; 
3356 }
3357
3358 static int decode_ref_pic_marking(H264Context *h){
3359     MpegEncContext * const s = &h->s;
3360     int i;
3361     
3362     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3363         s->broken_link= get_bits1(&s->gb) -1;
3364         h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3365         if(h->mmco[0].long_index == -1)
3366             h->mmco_index= 0;
3367         else{
3368             h->mmco[0].opcode= MMCO_LONG;
3369             h->mmco_index= 1;
3370         } 
3371     }else{
3372         if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3373             for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3374                 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3375
3376                 h->mmco[i].opcode= opcode;
3377                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3378                     h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
3379 /*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3380                         fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3381                         return -1;
3382                     }*/
3383                 }
3384                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3385                     h->mmco[i].long_index= get_ue_golomb(&s->gb);
3386                     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){
3387                         av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3388                         return -1;
3389                     }
3390                 }
3391                     
3392                 if(opcode > MMCO_LONG){
3393                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3394                     return -1;
3395                 }
3396                 if(opcode == MMCO_END)
3397                     break;
3398             }
3399             h->mmco_index= i;
3400         }else{
3401             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3402
3403             if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3404                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3405                 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3406                 h->mmco_index= 1;
3407             }else
3408                 h->mmco_index= 0;
3409         }
3410     }
3411     
3412     return 0; 
3413 }
3414
3415 static int init_poc(H264Context *h){
3416     MpegEncContext * const s = &h->s;
3417     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3418     int field_poc[2];
3419
3420     if(h->nal_unit_type == NAL_IDR_SLICE){
3421         h->frame_num_offset= 0;
3422     }else{
3423         if(h->frame_num < h->prev_frame_num)
3424             h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3425         else
3426             h->frame_num_offset= h->prev_frame_num_offset;
3427     }
3428
3429     if(h->sps.poc_type==0){
3430         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3431
3432         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3433             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3434         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3435             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3436         else
3437             h->poc_msb = h->prev_poc_msb;
3438 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3439         field_poc[0] = 
3440         field_poc[1] = h->poc_msb + h->poc_lsb;
3441         if(s->picture_structure == PICT_FRAME) 
3442             field_poc[1] += h->delta_poc_bottom;
3443     }else if(h->sps.poc_type==1){
3444         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3445         int i;
3446
3447         if(h->sps.poc_cycle_length != 0)
3448             abs_frame_num = h->frame_num_offset + h->frame_num;
3449         else
3450             abs_frame_num = 0;
3451
3452         if(h->nal_ref_idc==0 && abs_frame_num > 0)
3453             abs_frame_num--;
3454             
3455         expected_delta_per_poc_cycle = 0;
3456         for(i=0; i < h->sps.poc_cycle_length; i++)
3457             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3458
3459         if(abs_frame_num > 0){
3460             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3461             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3462
3463             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3464             for(i = 0; i <= frame_num_in_poc_cycle; i++)
3465                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3466         } else
3467             expectedpoc = 0;
3468
3469         if(h->nal_ref_idc == 0) 
3470             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3471         
3472         field_poc[0] = expectedpoc + h->delta_poc[0];
3473         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3474
3475         if(s->picture_structure == PICT_FRAME)
3476             field_poc[1] += h->delta_poc[1];
3477     }else{
3478         int poc;
3479         if(h->nal_unit_type == NAL_IDR_SLICE){
3480             poc= 0;
3481         }else{
3482             if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3483             else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3484         }
3485         field_poc[0]= poc;
3486         field_poc[1]= poc;
3487     }
3488     
3489     if(s->picture_structure != PICT_BOTTOM_FIELD)
3490         s->current_picture_ptr->field_poc[0]= field_poc[0];
3491     if(s->picture_structure != PICT_TOP_FIELD)
3492         s->current_picture_ptr->field_poc[1]= field_poc[1];
3493     if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3494         s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3495
3496     return 0;
3497 }
3498
3499 /**
3500  * decodes a slice header.
3501  * this will allso call MPV_common_init() and frame_start() as needed
3502  */
3503 static int decode_slice_header(H264Context *h){
3504     MpegEncContext * const s = &h->s;
3505     int first_mb_in_slice, pps_id;
3506     int num_ref_idx_active_override_flag;
3507     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3508     int slice_type;
3509     int default_ref_list_done = 0;
3510
3511     s->current_picture.reference= h->nal_ref_idc != 0;
3512
3513     first_mb_in_slice= get_ue_golomb(&s->gb);
3514
3515     slice_type= get_ue_golomb(&s->gb);
3516     if(slice_type > 9){
3517         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);
3518         return -1;
3519     }
3520     if(slice_type > 4){
3521         slice_type -= 5;
3522         h->slice_type_fixed=1;
3523     }else
3524         h->slice_type_fixed=0;
3525     
3526     slice_type= slice_type_map[ slice_type ];
3527     if (slice_type == I_TYPE
3528         || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3529         default_ref_list_done = 1;
3530     }
3531     h->slice_type= slice_type;
3532
3533     s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3534         
3535     pps_id= get_ue_golomb(&s->gb);
3536     if(pps_id>255){
3537         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3538         return -1;
3539     }
3540     h->pps= h->pps_buffer[pps_id];
3541     if(h->pps.slice_group_count == 0){
3542         av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3543         return -1;
3544     }
3545
3546     h->sps= h->sps_buffer[ h->pps.sps_id ];
3547     if(h->sps.log2_max_frame_num == 0){
3548         av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3549         return -1;
3550     }
3551     
3552     s->mb_width= h->sps.mb_width;
3553     s->mb_height= h->sps.mb_height;
3554     
3555     h->b_stride=  s->mb_width*4 + 1;
3556     h->b8_stride= s->mb_width*2 + 1;
3557
3558     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3559     s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3560     
3561     s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3562     if(h->sps.frame_mbs_only_flag)
3563         s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3564     else
3565         s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3566     
3567     if (s->context_initialized 
3568         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3569         free_tables(h);
3570         MPV_common_end(s);
3571     }
3572     if (!s->context_initialized) {
3573         if (MPV_common_init(s) < 0)
3574             return -1;
3575
3576         alloc_tables(h);
3577
3578         s->avctx->width = s->width;
3579         s->avctx->height = s->height;
3580         s->avctx->sample_aspect_ratio= h->sps.sar;
3581         if(!s->avctx->sample_aspect_ratio.den)
3582             s->avctx->sample_aspect_ratio.den = 1;
3583
3584         if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3585             s->avctx->frame_rate = h->sps.time_scale;
3586             s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3587         }
3588     }
3589
3590     if(h->slice_num == 0){
3591         frame_start(h);
3592     }
3593
3594     s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3595     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3596
3597     if(h->sps.frame_mbs_only_flag){
3598         s->picture_structure= PICT_FRAME;
3599     }else{
3600         if(get_bits1(&s->gb)) //field_pic_flag
3601             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3602         else
3603             s->picture_structure= PICT_FRAME;
3604     }
3605
3606     if(s->picture_structure==PICT_FRAME){
3607         h->curr_pic_num=   h->frame_num;
3608         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3609     }else{
3610         h->curr_pic_num= 2*h->frame_num;
3611         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3612     }
3613         
3614     if(h->nal_unit_type == NAL_IDR_SLICE){
3615         get_ue_golomb(&s->gb); /* idr_pic_id */
3616     }
3617    
3618     if(h->sps.poc_type==0){
3619         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3620         
3621         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3622             h->delta_poc_bottom= get_se_golomb(&s->gb);
3623         }
3624     }
3625     
3626     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3627         h->delta_poc[0]= get_se_golomb(&s->gb);
3628         
3629         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3630             h->delta_poc[1]= get_se_golomb(&s->gb);
3631     }
3632     
3633     init_poc(h);
3634     
3635     if(h->pps.redundant_pic_cnt_present){
3636         h->redundant_pic_count= get_ue_golomb(&s->gb);
3637     }
3638
3639     //set defaults, might be overriden a few line later
3640     h->ref_count[0]= h->pps.ref_count[0];
3641     h->ref_count[1]= h->pps.ref_count[1];
3642
3643     if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3644         if(h->slice_type == B_TYPE){
3645             h->direct_spatial_mv_pred= get_bits1(&s->gb);
3646         }
3647         num_ref_idx_active_override_flag= get_bits1(&s->gb);
3648     
3649         if(num_ref_idx_active_override_flag){
3650             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3651             if(h->slice_type==B_TYPE)
3652                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3653
3654             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3655                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3656                 return -1;
3657             }
3658         }
3659     }
3660
3661     if(!default_ref_list_done){
3662         fill_default_ref_list(h);
3663     }
3664
3665     decode_ref_pic_list_reordering(h);
3666
3667     if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3668        || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3669         pred_weight_table(h);
3670     else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3671         implicit_weight_table(h);
3672     else
3673         h->use_weight = 0;
3674     
3675     if(s->current_picture.reference)
3676         decode_ref_pic_marking(h);
3677
3678     if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3679         h->cabac_init_idc = get_ue_golomb(&s->gb);
3680
3681     h->last_qscale_diff = 0;
3682     s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3683     if(s->qscale<0 || s->qscale>51){
3684         av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3685         return -1;
3686     }
3687     h->chroma_qp = get_chroma_qp(h, s->qscale);
3688     //FIXME qscale / qp ... stuff
3689     if(h->slice_type == SP_TYPE){
3690         get_bits1(&s->gb); /* sp_for_switch_flag */
3691     }
3692     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3693         get_se_golomb(&s->gb); /* slice_qs_delta */
3694     }
3695
3696     h->deblocking_filter = 1;
3697     h->slice_alpha_c0_offset = 0;
3698     h->slice_beta_offset = 0;
3699     if( h->pps.deblocking_filter_parameters_present ) {
3700         h->deblocking_filter= get_ue_golomb(&s->gb);
3701         if(h->deblocking_filter < 2) 
3702             h->deblocking_filter^= 1; // 1<->0
3703
3704         if( h->deblocking_filter ) {
3705             h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3706             h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3707         }
3708     }
3709
3710 #if 0 //FMO
3711     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3712         slice_group_change_cycle= get_bits(&s->gb, ?);
3713 #endif
3714
3715     h->slice_num++;
3716
3717     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3718         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d weight:%d%s\n", 
3719                h->slice_num, first_mb_in_slice, 
3720                av_get_pict_type_char(h->slice_type),
3721                pps_id, h->frame_num,
3722                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3723                h->ref_count[0], h->ref_count[1],
3724                s->qscale,
3725                h->deblocking_filter,
3726                h->use_weight,
3727                h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3728                );
3729     }
3730
3731     return 0;
3732 }
3733
3734 /**
3735  *
3736  */
3737 static inline int get_level_prefix(GetBitContext *gb){
3738     unsigned int buf;
3739     int log;
3740     
3741     OPEN_READER(re, gb);
3742     UPDATE_CACHE(re, gb);
3743     buf=GET_CACHE(re, gb);
3744     
3745     log= 32 - av_log2(buf);
3746 #ifdef TRACE
3747     print_bin(buf>>(32-log), log);
3748     av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3749 #endif
3750
3751     LAST_SKIP_BITS(re, gb, log);
3752     CLOSE_READER(re, gb);
3753
3754     return log-1;
3755 }
3756
3757 /**
3758  * decodes a residual block.
3759  * @param n block index
3760  * @param scantable scantable
3761  * @param max_coeff number of coefficients in the block
3762  * @return <0 if an error occured
3763  */
3764 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3765     MpegEncContext * const s = &h->s;
3766     const uint16_t *qmul= dequant_coeff[qp];
3767     static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
3768     int level[16], run[16];
3769     int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3770
3771     //FIXME put trailing_onex into the context
3772
3773     if(n == CHROMA_DC_BLOCK_INDEX){
3774         coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3775         total_coeff= coeff_token>>2;
3776     }else{    
3777         if(n == LUMA_DC_BLOCK_INDEX){
3778             total_coeff= pred_non_zero_count(h, 0);
3779             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3780             total_coeff= coeff_token>>2;
3781         }else{
3782             total_coeff= pred_non_zero_count(h, n);
3783             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3784             total_coeff= coeff_token>>2;
3785             h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3786         }
3787     }
3788
3789     //FIXME set last_non_zero?
3790
3791     if(total_coeff==0)
3792         return 0;
3793         
3794     trailing_ones= coeff_token&3;
3795     tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3796     assert(total_coeff<=16);
3797     
3798     for(i=0; i<trailing_ones; i++){
3799         level[i]= 1 - 2*get_bits1(gb);
3800     }
3801
3802     suffix_length= total_coeff > 10 && trailing_ones < 3;
3803
3804     for(; i<total_coeff; i++){
3805         const int prefix= get_level_prefix(gb);
3806         int level_code, mask;
3807
3808         if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3809             if(suffix_length)
3810                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3811             else
3812                 level_code= (prefix<<suffix_length); //part
3813         }else if(prefix==14){
3814             if(suffix_length)
3815                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3816             else
3817                 level_code= prefix + get_bits(gb, 4); //part
3818         }else if(prefix==15){
3819             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3820             if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3821         }else{
3822             av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3823             return -1;
3824         }
3825
3826         if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3827
3828         mask= -(level_code&1);
3829         level[i]= (((2+level_code)>>1) ^ mask) - mask;
3830
3831         if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3832
3833 #if 1
3834         if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3835 #else        
3836         if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3837         /* ? == prefix > 2 or sth */
3838 #endif
3839         tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3840     }
3841
3842     if(total_coeff == max_coeff)
3843         zeros_left=0;
3844     else{
3845         if(n == CHROMA_DC_BLOCK_INDEX)
3846             zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3847         else
3848             zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3849     }
3850     
3851     for(i=0; i<total_coeff-1; i++){
3852         if(zeros_left <=0)
3853             break;
3854         else if(zeros_left < 7){
3855             run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3856         }else{
3857             run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3858         }
3859         zeros_left -= run[i];
3860     }
3861
3862     if(zeros_left<0){
3863         av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3864         return -1;
3865     }
3866     
3867     for(; i<total_coeff-1; i++){
3868         run[i]= 0;
3869     }
3870
3871     run[i]= zeros_left;
3872
3873     coeff_num=-1;
3874     if(n > 24){
3875         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3876             int j;
3877
3878             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3879             j= scantable[ coeff_num ];
3880
3881             block[j]= level[i];
3882         }
3883     }else{
3884         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3885             int j;
3886
3887             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3888             j= scantable[ coeff_num ];
3889
3890             block[j]= level[i] * qmul[j];
3891 //            printf("%d %d  ", block[j], qmul[j]);
3892         }
3893     }
3894     return 0;
3895 }
3896
3897 /**
3898  * decodes a P_SKIP or B_SKIP macroblock
3899  */
3900 static void decode_mb_skip(H264Context *h){
3901     MpegEncContext * const s = &h->s;
3902     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3903     int mb_type;
3904     
3905     memset(h->non_zero_count[mb_xy], 0, 16);
3906     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3907
3908     if( h->slice_type == B_TYPE )
3909     {
3910         // just for fill_caches. pred_direct_motion will set the real mb_type
3911         mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3912         //FIXME mbaff
3913
3914         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
3915         pred_direct_motion(h, &mb_type);
3916         if(h->pps.cabac){
3917             fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3918             fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
3919         }
3920     }
3921     else
3922     {
3923         int mx, my;
3924         mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
3925
3926         if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3927             h->mb_field_decoding_flag= get_bits1(&s->gb);
3928         }
3929         if(h->mb_field_decoding_flag)
3930             mb_type|= MB_TYPE_INTERLACED;
3931         
3932         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
3933         pred_pskip_motion(h, &mx, &my);
3934         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3935         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3936         if(h->pps.cabac)
3937             fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3938     }
3939
3940     write_back_motion(h, mb_type);
3941     s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
3942     s->current_picture.qscale_table[mb_xy]= s->qscale;
3943     h->slice_table[ mb_xy ]= h->slice_num;
3944     h->prev_mb_skiped= 1;
3945 }
3946
3947 /**
3948  * decodes a macroblock
3949  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3950  */
3951 static int decode_mb_cavlc(H264Context *h){
3952     MpegEncContext * const s = &h->s;
3953     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3954     int mb_type, partition_count, cbp;
3955
3956     s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3957
3958     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3959     cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3960                 down the code */
3961     if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3962         if(s->mb_skip_run==-1)
3963             s->mb_skip_run= get_ue_golomb(&s->gb);
3964         
3965         if (s->mb_skip_run--) {
3966             decode_mb_skip(h);
3967             return 0;
3968         }
3969     }
3970     if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3971         if((s->mb_y&1)==0)
3972             h->mb_field_decoding_flag = get_bits1(&s->gb);
3973     }else
3974         h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3975     
3976     h->prev_mb_skiped= 0;
3977     
3978     mb_type= get_ue_golomb(&s->gb);
3979     if(h->slice_type == B_TYPE){
3980         if(mb_type < 23){
3981             partition_count= b_mb_type_info[mb_type].partition_count;
3982             mb_type=         b_mb_type_info[mb_type].type;
3983         }else{
3984             mb_type -= 23;
3985             goto decode_intra_mb;
3986         }
3987     }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3988         if(mb_type < 5){
3989             partition_count= p_mb_type_info[mb_type].partition_count;
3990             mb_type=         p_mb_type_info[mb_type].type;
3991         }else{
3992             mb_type -= 5;
3993             goto decode_intra_mb;
3994         }
3995     }else{
3996        assert(h->slice_type == I_TYPE);
3997 decode_intra_mb:
3998         if(mb_type > 25){
3999             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);
4000             return -1;
4001         }
4002         partition_count=0;
4003         cbp= i_mb_type_info[mb_type].cbp;
4004         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4005         mb_type= i_mb_type_info[mb_type].type;
4006     }
4007
4008     if(h->mb_field_decoding_flag)
4009         mb_type |= MB_TYPE_INTERLACED;
4010
4011     s->current_picture.mb_type[mb_xy]= mb_type;
4012     h->slice_table[ mb_xy ]= h->slice_num;
4013     
4014     if(IS_INTRA_PCM(mb_type)){
4015         unsigned int x, y;
4016         
4017         // we assume these blocks are very rare so we dont optimize it
4018         align_get_bits(&s->gb);
4019         
4020         // The pixels are stored in the same order as levels in h->mb array.
4021         for(y=0; y<16; y++){
4022             const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4023             for(x=0; x<16; x++){
4024                 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4025                 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4026             }
4027         }
4028         for(y=0; y<8; y++){
4029             const int index= 256 + 4*(y&3) + 32*(y>>2);
4030             for(x=0; x<8; x++){
4031                 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4032                 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4033             }
4034         }
4035         for(y=0; y<8; y++){
4036             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4037             for(x=0; x<8; x++){
4038                 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4039                 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4040             }
4041         }
4042     
4043         // In deblocking, the quantiser is 0
4044         s->current_picture.qscale_table[mb_xy]= 0;
4045         h->chroma_qp = get_chroma_qp(h, 0);
4046         // All coeffs are presents
4047         memset(h->non_zero_count[mb_xy], 16, 16);
4048         
4049         return 0;
4050     }
4051         
4052     fill_caches(h, mb_type, 0);
4053
4054     //mb_pred
4055     if(IS_INTRA(mb_type)){
4056 //            init_top_left_availability(h);
4057             if(IS_INTRA4x4(mb_type)){
4058                 int i;
4059
4060 //                fill_intra4x4_pred_table(h);
4061                 for(i=0; i<16; i++){
4062                     const int mode_coded= !get_bits1(&s->gb);
4063                     const int predicted_mode=  pred_intra_mode(h, i);
4064                     int mode;
4065
4066                     if(mode_coded){
4067                         const int rem_mode= get_bits(&s->gb, 3);
4068                         if(rem_mode<predicted_mode)
4069                             mode= rem_mode;
4070                         else
4071                             mode= rem_mode + 1;
4072                     }else{
4073                         mode= predicted_mode;
4074                     }
4075                     
4076                     h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4077                 }
4078                 write_back_intra_pred_mode(h);
4079                 if( check_intra4x4_pred_mode(h) < 0)
4080                     return -1;
4081             }else{
4082                 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4083                 if(h->intra16x16_pred_mode < 0)
4084                     return -1;
4085             }
4086             h->chroma_pred_mode= get_ue_golomb(&s->gb);
4087
4088             h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4089             if(h->chroma_pred_mode < 0)
4090                 return -1;
4091     }else if(partition_count==4){
4092         int i, j, sub_partition_count[4], list, ref[2][4];
4093         
4094         if(h->slice_type == B_TYPE){
4095             for(i=0; i<4; i++){
4096                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4097                 if(h->sub_mb_type[i] >=13){
4098                     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);
4099                     return -1;
4100                 }
4101                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4102                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4103             }
4104             if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4105                || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4106                 pred_direct_motion(h, &mb_type);
4107         }else{
4108             assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4109             for(i=0; i<4; i++){
4110                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4111                 if(h->sub_mb_type[i] >=4){
4112                     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);
4113                     return -1;
4114                 }
4115                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4116                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4117             }
4118         }
4119         
4120         for(list=0; list<2; list++){
4121             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4122             if(ref_count == 0) continue;
4123             for(i=0; i<4; i++){
4124                 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4125                 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4126                     ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4127                 }else{
4128                  //FIXME
4129                     ref[list][i] = -1;
4130                 }
4131             }
4132         }
4133         
4134         for(list=0; list<2; list++){
4135             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4136             if(ref_count == 0) continue;
4137
4138             for(i=0; i<4; i++){
4139                 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4140                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4141                 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4142
4143                 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4144                     const int sub_mb_type= h->sub_mb_type[i];
4145                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4146                     for(j=0; j<sub_partition_count[i]; j++){
4147                         int mx, my;
4148                         const int index= 4*i + block_width*j;
4149                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4150                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4151                         mx += get_se_golomb(&s->gb);
4152                         my += get_se_golomb(&s->gb);
4153                         tprintf("final mv:%d %d\n", mx, my);
4154
4155                         if(IS_SUB_8X8(sub_mb_type)){
4156                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
4157                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4158                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
4159                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4160                         }else if(IS_SUB_8X4(sub_mb_type)){
4161                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4162                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4163                         }else if(IS_SUB_4X8(sub_mb_type)){
4164                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4165                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4166                         }else{
4167                             assert(IS_SUB_4X4(sub_mb_type));
4168                             mv_cache[ 0 ][0]= mx;
4169                             mv_cache[ 0 ][1]= my;
4170                         }
4171                     }
4172                 }else{
4173                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4174                     p[0] = p[1]=
4175                     p[8] = p[9]= 0;
4176                 }
4177             }
4178         }
4179     }else if(IS_DIRECT(mb_type)){
4180         pred_direct_motion(h, &mb_type);
4181         s->current_picture.mb_type[mb_xy]= mb_type;
4182     }else{
4183         int list, mx, my, i;
4184          //FIXME we should set ref_idx_l? to 0 if we use that later ...
4185         if(IS_16X16(mb_type)){
4186             for(list=0; list<2; list++){
4187                 if(h->ref_count[list]>0){
4188                     if(IS_DIR(mb_type, 0, list)){
4189                         const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4190                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4191                     }else
4192                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);
4193                 }
4194             }
4195             for(list=0; list<2; list++){
4196                 if(IS_DIR(mb_type, 0, list)){
4197                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4198                     mx += get_se_golomb(&s->gb);
4199                     my += get_se_golomb(&s->gb);
4200                     tprintf("final mv:%d %d\n", mx, my);
4201
4202