39ee4c6a2c3f19d9815601306359d3b0c8b2ba48
[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     int bitstream_restriction_flag;
91     int num_reorder_frames;
92 }SPS;
93
94 /**
95  * Picture parameter set
96  */
97 typedef struct PPS{
98     int sps_id;
99     int cabac;                  ///< entropy_coding_mode_flag
100     int pic_order_present;      ///< pic_order_present_flag
101     int slice_group_count;      ///< num_slice_groups_minus1 + 1
102     int mb_slice_group_map_type;
103     int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
104     int weighted_pred;          ///< weighted_pred_flag
105     int weighted_bipred_idc;
106     int init_qp;                ///< pic_init_qp_minus26 + 26
107     int init_qs;                ///< pic_init_qs_minus26 + 26
108     int chroma_qp_index_offset;
109     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
110     int constrained_intra_pred; ///< constrained_intra_pred_flag
111     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
112 }PPS;
113
114 /**
115  * Memory management control operation opcode.
116  */
117 typedef enum MMCOOpcode{
118     MMCO_END=0,
119     MMCO_SHORT2UNUSED,
120     MMCO_LONG2UNUSED,
121     MMCO_SHORT2LONG,
122     MMCO_SET_MAX_LONG,
123     MMCO_RESET, 
124     MMCO_LONG,
125 } MMCOOpcode;
126
127 /**
128  * Memory management control operation.
129  */
130 typedef struct MMCO{
131     MMCOOpcode opcode;
132     int short_frame_num;
133     int long_index;
134 } MMCO;
135
136 /**
137  * H264Context
138  */
139 typedef struct H264Context{
140     MpegEncContext s;
141     int nal_ref_idc;    
142     int nal_unit_type;
143 #define NAL_SLICE               1
144 #define NAL_DPA                 2
145 #define NAL_DPB                 3
146 #define NAL_DPC                 4
147 #define NAL_IDR_SLICE           5
148 #define NAL_SEI                 6
149 #define NAL_SPS                 7
150 #define NAL_PPS                 8
151 #define NAL_PICTURE_DELIMITER   9
152 #define NAL_FILTER_DATA         10
153     uint8_t *rbsp_buffer;
154     int rbsp_buffer_size;
155
156     /**
157       * Used to parse AVC variant of h264
158       */
159     int is_avc; ///< this flag is != 0 if codec is avc1
160     int got_avcC; ///< flag used to parse avcC data only once
161     int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
162
163     int chroma_qp; //QPc
164
165     int prev_mb_skiped; //FIXME remove (IMHO not used)
166
167     //prediction stuff
168     int chroma_pred_mode;
169     int intra16x16_pred_mode;
170
171     int top_mb_xy;
172     int left_mb_xy[2];
173     
174     int8_t intra4x4_pred_mode_cache[5*8];
175     int8_t (*intra4x4_pred_mode)[8];
176     void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
177     void (*pred8x8  [4+3])(uint8_t *src, int stride);
178     void (*pred16x16[4+3])(uint8_t *src, int stride);
179     unsigned int topleft_samples_available;
180     unsigned int top_samples_available;
181     unsigned int topright_samples_available;
182     unsigned int left_samples_available;
183     uint8_t (*top_borders[2])[16+2*8];
184     uint8_t left_border[2*(17+2*9)];
185
186     /**
187      * non zero coeff count cache.
188      * is 64 if not available.
189      */
190     uint8_t non_zero_count_cache[6*8] __align8;
191     uint8_t (*non_zero_count)[16];
192
193     /**
194      * Motion vector cache.
195      */
196     int16_t mv_cache[2][5*8][2] __align8;
197     int8_t ref_cache[2][5*8] __align8;
198 #define LIST_NOT_USED -1 //FIXME rename?
199 #define PART_NOT_AVAILABLE -2
200     
201     /**
202      * is 1 if the specific list MV&references are set to 0,0,-2.
203      */
204     int mv_cache_clean[2];
205
206     /**
207      * block_offset[ 0..23] for frame macroblocks
208      * block_offset[24..47] for field macroblocks
209      */
210     int block_offset[2*(16+8)];
211     
212     uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
213     uint16_t *mb2b8_xy;
214     int b_stride; //FIXME use s->b4_stride
215     int b8_stride;
216
217     int halfpel_flag;
218     int thirdpel_flag;
219
220     int unknown_svq3_flag;
221     int next_slice_index;
222
223     SPS sps_buffer[MAX_SPS_COUNT];
224     SPS sps; ///< current sps
225     
226     PPS pps_buffer[MAX_PPS_COUNT];
227     /**
228      * current pps
229      */
230     PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
231
232     int slice_num;
233     uint8_t *slice_table_base;
234     uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
235     int slice_type;
236     int slice_type_fixed;
237     
238     //interlacing specific flags
239     int mb_aff_frame;
240     int mb_field_decoding_flag;
241     
242     int sub_mb_type[4];
243     
244     //POC stuff
245     int poc_lsb;
246     int poc_msb;
247     int delta_poc_bottom;
248     int delta_poc[2];
249     int frame_num;
250     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
251     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
252     int frame_num_offset;         ///< for POC type 2
253     int prev_frame_num_offset;    ///< for POC type 2
254     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
255
256     /**
257      * frame_num for frames or 2*frame_num for field pics.
258      */
259     int curr_pic_num;
260     
261     /**
262      * max_frame_num or 2*max_frame_num for field pics.
263      */
264     int max_pic_num;
265
266     //Weighted pred stuff
267     int use_weight;
268     int use_weight_chroma;
269     int luma_log2_weight_denom;
270     int chroma_log2_weight_denom;
271     int luma_weight[2][16];
272     int luma_offset[2][16];
273     int chroma_weight[2][16][2];
274     int chroma_offset[2][16][2];
275     int implicit_weight[16][16];
276    
277     //deblock
278     int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
279     int slice_alpha_c0_offset;
280     int slice_beta_offset;
281      
282     int redundant_pic_count;
283     
284     int direct_spatial_mv_pred;
285     int dist_scale_factor[16];
286     int map_col_to_list0[2][16];
287
288     /**
289      * num_ref_idx_l0/1_active_minus1 + 1
290      */
291     int ref_count[2];// FIXME split for AFF
292     Picture *short_ref[32];
293     Picture *long_ref[32];
294     Picture default_ref_list[2][32];
295     Picture ref_list[2][32]; //FIXME size?
296     Picture field_ref_list[2][32]; //FIXME size?
297     Picture *delayed_pic[16]; //FIXME size?
298     Picture *delayed_output_pic;
299     
300     /**
301      * memory management control operations buffer.
302      */
303     MMCO mmco[MAX_MMCO_COUNT];
304     int mmco_index;
305     
306     int long_ref_count;  ///< number of actual long term references
307     int short_ref_count; ///< number of actual short term references
308     
309     //data partitioning
310     GetBitContext intra_gb;
311     GetBitContext inter_gb;
312     GetBitContext *intra_gb_ptr;
313     GetBitContext *inter_gb_ptr;
314     
315     DCTELEM mb[16*24] __align8;
316
317     /**
318      * Cabac
319      */
320     CABACContext cabac;
321     uint8_t      cabac_state[399];
322     int          cabac_init_idc;
323
324     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
325     uint16_t     *cbp_table;
326     int top_cbp;
327     int left_cbp;
328     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
329     uint8_t     *chroma_pred_mode_table;
330     int         last_qscale_diff;
331     int16_t     (*mvd_table[2])[2];
332     int16_t     mvd_cache[2][5*8][2] __align8;
333     uint8_t     *direct_table;
334     uint8_t     direct_cache[5*8];
335
336 }H264Context;
337
338 static VLC coeff_token_vlc[4];
339 static VLC chroma_dc_coeff_token_vlc;
340
341 static VLC total_zeros_vlc[15];
342 static VLC chroma_dc_total_zeros_vlc[3];
343
344 static VLC run_vlc[6];
345 static VLC run7_vlc;
346
347 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
348 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
349 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
350
351 static inline uint32_t pack16to32(int a, int b){
352 #ifdef WORDS_BIGENDIAN
353    return (b&0xFFFF) + (a<<16);
354 #else
355    return (a&0xFFFF) + (b<<16);
356 #endif
357 }
358
359 /**
360  * fill a rectangle.
361  * @param h height of the rectangle, should be a constant
362  * @param w width of the rectangle, should be a constant
363  * @param size the size of val (1 or 4), should be a constant
364  */
365 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
366     uint8_t *p= (uint8_t*)vp;
367     assert(size==1 || size==4);
368     
369     w      *= size;
370     stride *= size;
371     
372     assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
373 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
374     if(w==2 && h==2){
375         *(uint16_t*)(p + 0)=
376         *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
377     }else if(w==2 && h==4){
378         *(uint16_t*)(p + 0*stride)=
379         *(uint16_t*)(p + 1*stride)=
380         *(uint16_t*)(p + 2*stride)=
381         *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
382     }else if(w==4 && h==1){
383         *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
384     }else if(w==4 && h==2){
385         *(uint32_t*)(p + 0*stride)=
386         *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
387     }else if(w==4 && h==4){
388         *(uint32_t*)(p + 0*stride)=
389         *(uint32_t*)(p + 1*stride)=
390         *(uint32_t*)(p + 2*stride)=
391         *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
392     }else if(w==8 && h==1){
393         *(uint32_t*)(p + 0)=
394         *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
395     }else if(w==8 && h==2){
396         *(uint32_t*)(p + 0 + 0*stride)=
397         *(uint32_t*)(p + 4 + 0*stride)=
398         *(uint32_t*)(p + 0 + 1*stride)=
399         *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
400     }else if(w==8 && h==4){
401         *(uint64_t*)(p + 0*stride)=
402         *(uint64_t*)(p + 1*stride)=
403         *(uint64_t*)(p + 2*stride)=
404         *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
405     }else if(w==16 && h==2){
406         *(uint64_t*)(p + 0+0*stride)=
407         *(uint64_t*)(p + 8+0*stride)=
408         *(uint64_t*)(p + 0+1*stride)=
409         *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
410     }else if(w==16 && h==4){
411         *(uint64_t*)(p + 0+0*stride)=
412         *(uint64_t*)(p + 8+0*stride)=
413         *(uint64_t*)(p + 0+1*stride)=
414         *(uint64_t*)(p + 8+1*stride)=
415         *(uint64_t*)(p + 0+2*stride)=
416         *(uint64_t*)(p + 8+2*stride)=
417         *(uint64_t*)(p + 0+3*stride)=
418         *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
419     }else
420         assert(0);
421 }
422
423 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
424     MpegEncContext * const s = &h->s;
425     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
426     int topleft_xy, top_xy, topright_xy, left_xy[2];
427     int topleft_type, top_type, topright_type, left_type[2];
428     int left_block[8];
429     int i;
430
431     //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
432     
433     top_xy     = mb_xy  - s->mb_stride;
434     topleft_xy = top_xy - 1;
435     topright_xy= top_xy + 1;
436     left_xy[1] = left_xy[0] = mb_xy-1;
437     left_block[0]= 0;
438     left_block[1]= 1;
439     left_block[2]= 2;
440     left_block[3]= 3;
441     left_block[4]= 7;
442     left_block[5]= 10;
443     left_block[6]= 8;
444     left_block[7]= 11;
445     if(h->mb_aff_frame){
446         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
447         const int top_pair_xy      = pair_xy     - s->mb_stride;
448         const int topleft_pair_xy  = top_pair_xy - 1;
449         const int topright_pair_xy = top_pair_xy + 1;
450         const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
451         const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
452         const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
453         const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
454         const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
455         const int bottom = (s->mb_y & 1);
456         tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
457         if (bottom
458                 ? !curr_mb_frame_flag // bottom macroblock
459                 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
460                 ) {
461             top_xy -= s->mb_stride;
462         }
463         if (bottom
464                 ? !curr_mb_frame_flag // bottom macroblock
465                 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
466                 ) {
467             topleft_xy -= s->mb_stride;
468         }
469         if (bottom
470                 ? !curr_mb_frame_flag // bottom macroblock
471                 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
472                 ) {
473             topright_xy -= s->mb_stride;
474         }
475         if (left_mb_frame_flag != curr_mb_frame_flag) {
476             left_xy[1] = left_xy[0] = pair_xy - 1;
477             if (curr_mb_frame_flag) {
478                 if (bottom) {
479                     left_block[0]= 2;
480                     left_block[1]= 2;
481                     left_block[2]= 3;
482                     left_block[3]= 3;
483                     left_block[4]= 8;
484                     left_block[5]= 11;
485                     left_block[6]= 8;
486                     left_block[7]= 11;
487                 } else {
488                     left_block[0]= 0;
489                     left_block[1]= 0;
490                     left_block[2]= 1;
491                     left_block[3]= 1;
492                     left_block[4]= 7;
493                     left_block[5]= 10;
494                     left_block[6]= 7;
495                     left_block[7]= 10;
496                 }
497             } else {
498                 left_xy[1] += s->mb_stride;
499                 //left_block[0]= 0;
500                 left_block[1]= 2;
501                 left_block[2]= 0;
502                 left_block[3]= 2;
503                 //left_block[4]= 7;
504                 left_block[5]= 10;
505                 left_block[6]= 7;
506                 left_block[7]= 10;
507             }
508         }
509     }
510
511     h->top_mb_xy = top_xy;
512     h->left_mb_xy[0] = left_xy[0];
513     h->left_mb_xy[1] = left_xy[1];
514     if(for_deblock){
515         topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
516         top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
517         topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
518         left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
519         left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
520     }else{
521         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
522         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
523         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
524         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
525         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
526     }
527
528     if(IS_INTRA(mb_type)){
529         h->topleft_samples_available= 
530         h->top_samples_available= 
531         h->left_samples_available= 0xFFFF;
532         h->topright_samples_available= 0xEEEA;
533
534         if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
535             h->topleft_samples_available= 0xB3FF;
536             h->top_samples_available= 0x33FF;
537             h->topright_samples_available= 0x26EA;
538         }
539         for(i=0; i<2; i++){
540             if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
541                 h->topleft_samples_available&= 0xDF5F;
542                 h->left_samples_available&= 0x5F5F;
543             }
544         }
545         
546         if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
547             h->topleft_samples_available&= 0x7FFF;
548         
549         if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
550             h->topright_samples_available&= 0xFBFF;
551     
552         if(IS_INTRA4x4(mb_type)){
553             if(IS_INTRA4x4(top_type)){
554                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
555                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
556                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
557                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
558             }else{
559                 int pred;
560                 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
561                     pred= -1;
562                 else{
563                     pred= 2;
564                 }
565                 h->intra4x4_pred_mode_cache[4+8*0]=
566                 h->intra4x4_pred_mode_cache[5+8*0]=
567                 h->intra4x4_pred_mode_cache[6+8*0]=
568                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
569             }
570             for(i=0; i<2; i++){
571                 if(IS_INTRA4x4(left_type[i])){
572                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
573                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
574                 }else{
575                     int pred;
576                     if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
577                         pred= -1;
578                     else{
579                         pred= 2;
580                     }
581                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
582                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
583                 }
584             }
585         }
586     }
587     
588     
589 /*
590 0 . T T. T T T T 
591 1 L . .L . . . . 
592 2 L . .L . . . . 
593 3 . T TL . . . . 
594 4 L . .L . . . . 
595 5 L . .. . . . . 
596 */
597 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
598     if(top_type){
599         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
600         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
601         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
602         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
603     
604         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
605         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
606     
607         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
608         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
609         
610     }else{
611         h->non_zero_count_cache[4+8*0]=      
612         h->non_zero_count_cache[5+8*0]=
613         h->non_zero_count_cache[6+8*0]=
614         h->non_zero_count_cache[7+8*0]=
615     
616         h->non_zero_count_cache[1+8*0]=
617         h->non_zero_count_cache[2+8*0]=
618     
619         h->non_zero_count_cache[1+8*3]=
620         h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
621         
622     }
623
624     for (i=0; i<2; i++) {
625         if(left_type[i]){
626             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
627             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
628             h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
629             h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
630         }else{
631             h->non_zero_count_cache[3+8*1 + 2*8*i]= 
632             h->non_zero_count_cache[3+8*2 + 2*8*i]= 
633             h->non_zero_count_cache[0+8*1 +   8*i]= 
634             h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
635         }
636     }
637
638     if( h->pps.cabac ) {
639         // top_cbp
640         if(top_type) {
641             h->top_cbp = h->cbp_table[top_xy];
642         } else if(IS_INTRA(mb_type)) {
643             h->top_cbp = 0x1C0;
644         } else {
645             h->top_cbp = 0;
646         }
647         // left_cbp
648         if (left_type[0]) {
649             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
650         } else if(IS_INTRA(mb_type)) {
651             h->left_cbp = 0x1C0;
652         } else {
653             h->left_cbp = 0;
654         }
655         if (left_type[0]) {
656             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
657         }
658         if (left_type[1]) {
659             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
660         }
661     }
662
663 #if 1
664     //FIXME direct mb can skip much of this
665     if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
666         int list;
667         for(list=0; list<2; list++){
668             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !for_deblock){
669                 /*if(!h->mv_cache_clean[list]){
670                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
671                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
672                     h->mv_cache_clean[list]= 1;
673                 }*/
674                 continue;
675             }
676             h->mv_cache_clean[list]= 0;
677             
678             if(IS_INTER(topleft_type)){
679                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
680                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
681                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
682                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
683             }else{
684                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
685                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
686             }
687             
688             if(IS_INTER(top_type)){
689                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
690                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
691                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
692                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
693                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
694                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
695                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
696                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
697                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
698                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
699             }else{
700                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
701                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
702                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
703                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
704                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
705             }
706
707             if(IS_INTER(topright_type)){
708                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
709                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
710                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
711                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
712             }else{
713                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
714                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
715             }
716             
717             //FIXME unify cleanup or sth
718             if(IS_INTER(left_type[0])){
719                 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
720                 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
721                 *(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]];
722                 *(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]];
723                 h->ref_cache[list][scan8[0] - 1 + 0*8]= 
724                 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
725             }else{
726                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
727                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
728                 h->ref_cache[list][scan8[0] - 1 + 0*8]=
729                 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
730             }
731             
732             if(IS_INTER(left_type[1])){
733                 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
734                 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
735                 *(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]];
736                 *(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]];
737                 h->ref_cache[list][scan8[0] - 1 + 2*8]= 
738                 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
739             }else{
740                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
741                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
742                 h->ref_cache[list][scan8[0] - 1 + 2*8]=
743                 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
744             }
745
746             if(for_deblock)
747                 continue;
748
749             h->ref_cache[list][scan8[5 ]+1] = 
750             h->ref_cache[list][scan8[7 ]+1] = 
751             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
752             h->ref_cache[list][scan8[4 ]] = 
753             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
754             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
755             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
756             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
757             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
758             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
759
760             if( h->pps.cabac ) {
761                 /* XXX beurk, Load mvd */
762                 if(IS_INTER(topleft_type)){
763                     const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
764                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
765                 }else{
766                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
767                 }
768
769                 if(IS_INTER(top_type)){
770                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
771                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
772                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
773                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
774                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
775                 }else{
776                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
777                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
778                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
779                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
780                 }
781                 if(IS_INTER(left_type[0])){
782                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
783                     *(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]];
784                     *(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]];
785                 }else{
786                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
787                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
788                 }
789                 if(IS_INTER(left_type[1])){
790                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
791                     *(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]];
792                     *(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]];
793                 }else{
794                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
795                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
796                 }
797                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
798                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
799                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
800                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
801                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
802
803                 if(h->slice_type == B_TYPE){
804                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
805
806                     if(IS_DIRECT(top_type)){
807                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
808                     }else if(IS_8X8(top_type)){
809                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
810                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
811                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
812                     }else{
813                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
814                     }
815                     
816                     //FIXME interlacing
817                     if(IS_DIRECT(left_type[0])){
818                         h->direct_cache[scan8[0] - 1 + 0*8]=
819                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
820                     }else if(IS_8X8(left_type[0])){
821                         int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
822                         h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
823                         h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
824                     }else{
825                         h->direct_cache[scan8[0] - 1 + 0*8]=
826                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
827                     }
828                 }
829             }
830         }
831     }
832 #endif
833 }
834
835 static inline void write_back_intra_pred_mode(H264Context *h){
836     MpegEncContext * const s = &h->s;
837     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
838
839     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
840     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
841     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
842     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
843     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
844     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
845     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
846 }
847
848 /**
849  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
850  */
851 static inline int check_intra4x4_pred_mode(H264Context *h){
852     MpegEncContext * const s = &h->s;
853     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
854     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
855     int i;
856     
857     if(!(h->top_samples_available&0x8000)){
858         for(i=0; i<4; i++){
859             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
860             if(status<0){
861                 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);
862                 return -1;
863             } else if(status){
864                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
865             }
866         }
867     }
868     
869     if(!(h->left_samples_available&0x8000)){
870         for(i=0; i<4; i++){
871             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
872             if(status<0){
873                 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);
874                 return -1;
875             } else if(status){
876                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
877             }
878         }
879     }
880
881     return 0;
882 } //FIXME cleanup like next
883
884 /**
885  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
886  */
887 static inline int check_intra_pred_mode(H264Context *h, int mode){
888     MpegEncContext * const s = &h->s;
889     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
890     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
891     
892     if(mode < 0 || mode > 6) {
893         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);
894         return -1;
895     }
896     
897     if(!(h->top_samples_available&0x8000)){
898         mode= top[ mode ];
899         if(mode<0){
900             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);
901             return -1;
902         }
903     }
904     
905     if(!(h->left_samples_available&0x8000)){
906         mode= left[ mode ];
907         if(mode<0){
908             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);
909             return -1;
910         } 
911     }
912
913     return mode;
914 }
915
916 /**
917  * gets the predicted intra4x4 prediction mode.
918  */
919 static inline int pred_intra_mode(H264Context *h, int n){
920     const int index8= scan8[n];
921     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
922     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
923     const int min= FFMIN(left, top);
924
925     tprintf("mode:%d %d min:%d\n", left ,top, min);
926
927     if(min<0) return DC_PRED;
928     else      return min;
929 }
930
931 static inline void write_back_non_zero_count(H264Context *h){
932     MpegEncContext * const s = &h->s;
933     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
934
935     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
936     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
937     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
938     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
939     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
940     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
941     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
942     
943     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
944     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
945     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
946
947     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
948     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
949     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
950 }
951
952 /**
953  * gets the predicted number of non zero coefficients.
954  * @param n block index
955  */
956 static inline int pred_non_zero_count(H264Context *h, int n){
957     const int index8= scan8[n];
958     const int left= h->non_zero_count_cache[index8 - 1];
959     const int top = h->non_zero_count_cache[index8 - 8];
960     int i= left + top;
961     
962     if(i<64) i= (i+1)>>1;
963
964     tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
965
966     return i&31;
967 }
968
969 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
970     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
971
972     if(topright_ref != PART_NOT_AVAILABLE){
973         *C= h->mv_cache[list][ i - 8 + part_width ];
974         return topright_ref;
975     }else{
976         tprintf("topright MV not available\n");
977
978         *C= h->mv_cache[list][ i - 8 - 1 ];
979         return h->ref_cache[list][ i - 8 - 1 ];
980     }
981 }
982
983 /**
984  * gets the predicted MV.
985  * @param n the block index
986  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
987  * @param mx the x component of the predicted motion vector
988  * @param my the y component of the predicted motion vector
989  */
990 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
991     const int index8= scan8[n];
992     const int top_ref=      h->ref_cache[list][ index8 - 8 ];
993     const int left_ref=     h->ref_cache[list][ index8 - 1 ];
994     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
995     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
996     const int16_t * C;
997     int diagonal_ref, match_count;
998
999     assert(part_width==1 || part_width==2 || part_width==4);
1000
1001 /* mv_cache
1002   B . . A T T T T 
1003   U . . L . . , .
1004   U . . L . . . .
1005   U . . L . . , .
1006   . . . L . . . .
1007 */
1008
1009     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1010     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1011     tprintf("pred_motion match_count=%d\n", match_count);
1012     if(match_count > 1){ //most common
1013         *mx= mid_pred(A[0], B[0], C[0]);
1014         *my= mid_pred(A[1], B[1], C[1]);
1015     }else if(match_count==1){
1016         if(left_ref==ref){
1017             *mx= A[0];
1018             *my= A[1];        
1019         }else if(top_ref==ref){
1020             *mx= B[0];
1021             *my= B[1];        
1022         }else{
1023             *mx= C[0];
1024             *my= C[1];        
1025         }
1026     }else{
1027         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1028             *mx= A[0];
1029             *my= A[1];        
1030         }else{
1031             *mx= mid_pred(A[0], B[0], C[0]);
1032             *my= mid_pred(A[1], B[1], C[1]);
1033         }
1034     }
1035         
1036     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);
1037 }
1038
1039 /**
1040  * gets the directionally predicted 16x8 MV.
1041  * @param n the block index
1042  * @param mx the x component of the predicted motion vector
1043  * @param my the y component of the predicted motion vector
1044  */
1045 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1046     if(n==0){
1047         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1048         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1049
1050         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);
1051         
1052         if(top_ref == ref){
1053             *mx= B[0];
1054             *my= B[1];
1055             return;
1056         }
1057     }else{
1058         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1059         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1060         
1061         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);
1062
1063         if(left_ref == ref){
1064             *mx= A[0];
1065             *my= A[1];
1066             return;
1067         }
1068     }
1069
1070     //RARE
1071     pred_motion(h, n, 4, list, ref, mx, my);
1072 }
1073
1074 /**
1075  * gets the directionally predicted 8x16 MV.
1076  * @param n the block index
1077  * @param mx the x component of the predicted motion vector
1078  * @param my the y component of the predicted motion vector
1079  */
1080 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1081     if(n==0){
1082         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1083         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1084         
1085         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);
1086
1087         if(left_ref == ref){
1088             *mx= A[0];
1089             *my= A[1];
1090             return;
1091         }
1092     }else{
1093         const int16_t * C;
1094         int diagonal_ref;
1095
1096         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1097         
1098         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);
1099
1100         if(diagonal_ref == ref){ 
1101             *mx= C[0];
1102             *my= C[1];
1103             return;
1104         }
1105     }
1106
1107     //RARE
1108     pred_motion(h, n, 2, list, ref, mx, my);
1109 }
1110
1111 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1112     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1113     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1114
1115     tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1116
1117     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1118        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1119        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1120        
1121         *mx = *my = 0;
1122         return;
1123     }
1124         
1125     pred_motion(h, 0, 4, 0, 0, mx, my);
1126
1127     return;
1128 }
1129
1130 static inline void direct_dist_scale_factor(H264Context * const h){
1131     const int poc = h->s.current_picture_ptr->poc;
1132     const int poc1 = h->ref_list[1][0].poc;
1133     int i;
1134     for(i=0; i<h->ref_count[0]; i++){
1135         int poc0 = h->ref_list[0][i].poc;
1136         int td = clip(poc1 - poc0, -128, 127);
1137         if(td == 0 /* FIXME || pic0 is a long-term ref */){
1138             h->dist_scale_factor[i] = 256;
1139         }else{
1140             int tb = clip(poc - poc0, -128, 127);
1141             int tx = (16384 + (ABS(td) >> 1)) / td;
1142             h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1143         }
1144     }
1145 }
1146 static inline void direct_ref_list_init(H264Context * const h){
1147     MpegEncContext * const s = &h->s;
1148     Picture * const ref1 = &h->ref_list[1][0];
1149     Picture * const cur = s->current_picture_ptr;
1150     int list, i, j;
1151     if(cur->pict_type == I_TYPE)
1152         cur->ref_count[0] = 0;
1153     if(cur->pict_type != B_TYPE)
1154         cur->ref_count[1] = 0;
1155     for(list=0; list<2; list++){
1156         cur->ref_count[list] = h->ref_count[list];
1157         for(j=0; j<h->ref_count[list]; j++)
1158             cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1159     }
1160     if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1161         return;
1162     for(list=0; list<2; list++){
1163         for(i=0; i<ref1->ref_count[list]; i++){
1164             const int poc = ref1->ref_poc[list][i];
1165             h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1166             for(j=0; j<h->ref_count[list]; j++)
1167                 if(h->ref_list[list][j].poc == poc){
1168                     h->map_col_to_list0[list][i] = j;
1169                     break;
1170                 }
1171         }
1172     }
1173 }
1174
1175 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1176     MpegEncContext * const s = &h->s;
1177     const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1178     const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1179     const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1180     const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1181     const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1182     const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1183     const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1184     const int is_b8x8 = IS_8X8(*mb_type);
1185     int sub_mb_type;
1186     int i8, i4;
1187
1188     if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1189         /* FIXME save sub mb types from previous frames (or derive from MVs)
1190          * so we know exactly what block size to use */
1191         sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1192         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1193     }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1194         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1195         *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1196     }else{
1197         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1198         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1199     }
1200     if(!is_b8x8)
1201         *mb_type |= MB_TYPE_DIRECT2;
1202
1203     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);
1204     
1205     if(h->direct_spatial_mv_pred){
1206         int ref[2];
1207         int mv[2][2];
1208         int list;
1209
1210         /* ref = min(neighbors) */
1211         for(list=0; list<2; list++){
1212             int refa = h->ref_cache[list][scan8[0] - 1];
1213             int refb = h->ref_cache[list][scan8[0] - 8];
1214             int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1215             if(refc == -2)
1216                 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1217             ref[list] = refa;
1218             if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1219                 ref[list] = refb;
1220             if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1221                 ref[list] = refc;
1222             if(ref[list] < 0)
1223                 ref[list] = -1;
1224         }
1225
1226         if(ref[0] < 0 && ref[1] < 0){
1227             ref[0] = ref[1] = 0;
1228             mv[0][0] = mv[0][1] =
1229             mv[1][0] = mv[1][1] = 0;
1230         }else{
1231             for(list=0; list<2; list++){
1232                 if(ref[list] >= 0)
1233                     pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1234                 else
1235                     mv[list][0] = mv[list][1] = 0;
1236             }
1237         }
1238
1239         if(ref[1] < 0){
1240             *mb_type &= ~MB_TYPE_P0L1;
1241             sub_mb_type &= ~MB_TYPE_P0L1;
1242         }else if(ref[0] < 0){
1243             *mb_type &= ~MB_TYPE_P0L0;
1244             sub_mb_type &= ~MB_TYPE_P0L0;
1245         }
1246
1247         if(IS_16X16(*mb_type)){
1248             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1249             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1250             if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1251                 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1252                 if(ref[0] > 0)
1253                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1254                 else
1255                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1256                 if(ref[1] > 0)
1257                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1258                 else
1259                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1260             }else{
1261                 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1262                 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1263             }
1264         }else{
1265             for(i8=0; i8<4; i8++){
1266                 const int x8 = i8&1;
1267                 const int y8 = i8>>1;
1268     
1269                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1270                     continue;
1271                 h->sub_mb_type[i8] = sub_mb_type;
1272     
1273                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1274                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1275                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1276                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1277     
1278                 /* col_zero_flag */
1279                 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1280                     for(i4=0; i4<4; i4++){
1281                         const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1282                         if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1283                             if(ref[0] == 0)
1284                                 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1285                             if(ref[1] == 0)
1286                                 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1287                         }
1288                     }
1289                 }
1290             }
1291         }
1292     }else{ /* direct temporal mv pred */
1293         if(IS_16X16(*mb_type)){
1294             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1295             if(IS_INTRA(mb_type_col)){
1296                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1297                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1298                 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1299             }else{
1300                 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1301                                                 : h->map_col_to_list0[1][l1ref1[0]];
1302                 const int dist_scale_factor = h->dist_scale_factor[ref0];
1303                 const int16_t *mv_col = l1mv0[0];
1304                 int mv_l0[2];
1305                 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1306                 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1307                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1308                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1309                 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);
1310             }
1311         }else{
1312             for(i8=0; i8<4; i8++){
1313                 const int x8 = i8&1;
1314                 const int y8 = i8>>1;
1315                 int ref0, dist_scale_factor;
1316     
1317                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1318                     continue;
1319                 h->sub_mb_type[i8] = sub_mb_type;
1320                 if(IS_INTRA(mb_type_col)){
1321                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1322                     fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1323                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1324                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1325                     continue;
1326                 }
1327     
1328                 ref0 = l1ref0[x8 + y8*h->b8_stride];
1329                 if(ref0 >= 0)
1330                     ref0 = h->map_col_to_list0[0][ref0];
1331                 else
1332                     ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1333                 dist_scale_factor = h->dist_scale_factor[ref0];
1334     
1335                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1336                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1337                 for(i4=0; i4<4; i4++){
1338                     const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1339                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1340                     mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1341                     mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1342                     *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1343                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1344                 }
1345             }
1346         }
1347     }
1348 }
1349
1350 static inline void write_back_motion(H264Context *h, int mb_type){
1351     MpegEncContext * const s = &h->s;
1352     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1353     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1354     int list;
1355
1356     for(list=0; list<2; list++){
1357         int y;
1358         if(!USES_LIST(mb_type, list)){
1359             if(1){ //FIXME skip or never read if mb_type doesnt use it
1360                 for(y=0; y<4; y++){
1361                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1362                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1363                 }
1364                 if( h->pps.cabac ) {
1365                     /* FIXME needed ? */
1366                     for(y=0; y<4; y++){
1367                         *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1368                         *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1369                     }
1370                 }
1371                 for(y=0; y<2; y++){
1372                     *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1373                 }
1374             }
1375             continue;
1376         }
1377         
1378         for(y=0; y<4; y++){
1379             *(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];
1380             *(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];
1381         }
1382         if( h->pps.cabac ) {
1383             for(y=0; y<4; y++){
1384                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1385                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1386             }
1387         }
1388         for(y=0; y<2; y++){
1389             s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1390             s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1391         }
1392     }
1393     
1394     if(h->slice_type == B_TYPE && h->pps.cabac){
1395         if(IS_8X8(mb_type)){
1396             h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1397             h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1398             h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1399         }
1400     }
1401 }
1402
1403 /**
1404  * Decodes a network abstraction layer unit.
1405  * @param consumed is the number of bytes used as input
1406  * @param length is the length of the array
1407  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1408  * @returns decoded bytes, might be src+1 if no escapes 
1409  */
1410 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1411     int i, si, di;
1412     uint8_t *dst;
1413
1414 //    src[0]&0x80;              //forbidden bit
1415     h->nal_ref_idc= src[0]>>5;
1416     h->nal_unit_type= src[0]&0x1F;
1417
1418     src++; length--;
1419 #if 0    
1420     for(i=0; i<length; i++)
1421         printf("%2X ", src[i]);
1422 #endif
1423     for(i=0; i+1<length; i+=2){
1424         if(src[i]) continue;
1425         if(i>0 && src[i-1]==0) i--;
1426         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1427             if(src[i+2]!=3){
1428                 /* startcode, so we must be past the end */
1429                 length=i;
1430             }
1431             break;
1432         }
1433     }
1434
1435     if(i>=length-1){ //no escaped 0
1436         *dst_length= length;
1437         *consumed= length+1; //+1 for the header
1438         return src; 
1439     }
1440
1441     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1442     dst= h->rbsp_buffer;
1443
1444 //printf("deoding esc\n");
1445     si=di=0;
1446     while(si<length){ 
1447         //remove escapes (very rare 1:2^22)
1448         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1449             if(src[si+2]==3){ //escape
1450                 dst[di++]= 0;
1451                 dst[di++]= 0;
1452                 si+=3;
1453                 continue;
1454             }else //next start code
1455                 break;
1456         }
1457
1458         dst[di++]= src[si++];
1459     }
1460
1461     *dst_length= di;
1462     *consumed= si + 1;//+1 for the header
1463 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1464     return dst;
1465 }
1466
1467 #if 0
1468 /**
1469  * @param src the data which should be escaped
1470  * @param dst the target buffer, dst+1 == src is allowed as a special case
1471  * @param length the length of the src data
1472  * @param dst_length the length of the dst array
1473  * @returns length of escaped data in bytes or -1 if an error occured
1474  */
1475 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1476     int i, escape_count, si, di;
1477     uint8_t *temp;
1478     
1479     assert(length>=0);
1480     assert(dst_length>0);
1481     
1482     dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1483
1484     if(length==0) return 1;
1485
1486     escape_count= 0;
1487     for(i=0; i<length; i+=2){
1488         if(src[i]) continue;
1489         if(i>0 && src[i-1]==0) 
1490             i--;
1491         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1492             escape_count++;
1493             i+=2;
1494         }
1495     }
1496     
1497     if(escape_count==0){ 
1498         if(dst+1 != src)
1499             memcpy(dst+1, src, length);
1500         return length + 1;
1501     }
1502     
1503     if(length + escape_count + 1> dst_length)
1504         return -1;
1505
1506     //this should be damn rare (hopefully)
1507
1508     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1509     temp= h->rbsp_buffer;
1510 //printf("encoding esc\n");
1511     
1512     si= 0;
1513     di= 0;
1514     while(si < length){
1515         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1516             temp[di++]= 0; si++;
1517             temp[di++]= 0; si++;
1518             temp[di++]= 3; 
1519             temp[di++]= src[si++];
1520         }
1521         else
1522             temp[di++]= src[si++];
1523     }
1524     memcpy(dst+1, temp, length+escape_count);
1525     
1526     assert(di == length+escape_count);
1527     
1528     return di + 1;
1529 }
1530
1531 /**
1532  * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1533  */
1534 static void encode_rbsp_trailing(PutBitContext *pb){
1535     int length;
1536     put_bits(pb, 1, 1);
1537     length= (-put_bits_count(pb))&7;
1538     if(length) put_bits(pb, length, 0);
1539 }
1540 #endif
1541
1542 /**
1543  * identifies the exact end of the bitstream
1544  * @return the length of the trailing, or 0 if damaged
1545  */
1546 static int decode_rbsp_trailing(uint8_t *src){
1547     int v= *src;
1548     int r;
1549
1550     tprintf("rbsp trailing %X\n", v);
1551
1552     for(r=1; r<9; r++){
1553         if(v&1) return r;
1554         v>>=1;
1555     }
1556     return 0;
1557 }
1558
1559 /**
1560  * idct tranforms the 16 dc values and dequantize them.
1561  * @param qp quantization parameter
1562  */
1563 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1564     const int qmul= dequant_coeff[qp][0];
1565 #define stride 16
1566     int i;
1567     int temp[16]; //FIXME check if this is a good idea
1568     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1569     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1570
1571 //memset(block, 64, 2*256);
1572 //return;
1573     for(i=0; i<4; i++){
1574         const int offset= y_offset[i];
1575         const int z0= block[offset+stride*0] + block[offset+stride*4];
1576         const int z1= block[offset+stride*0] - block[offset+stride*4];
1577         const int z2= block[offset+stride*1] - block[offset+stride*5];
1578         const int z3= block[offset+stride*1] + block[offset+stride*5];
1579
1580         temp[4*i+0]= z0+z3;
1581         temp[4*i+1]= z1+z2;
1582         temp[4*i+2]= z1-z2;
1583         temp[4*i+3]= z0-z3;
1584     }
1585
1586     for(i=0; i<4; i++){
1587         const int offset= x_offset[i];
1588         const int z0= temp[4*0+i] + temp[4*2+i];
1589         const int z1= temp[4*0+i] - temp[4*2+i];
1590         const int z2= temp[4*1+i] - temp[4*3+i];
1591         const int z3= temp[4*1+i] + temp[4*3+i];
1592
1593         block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1594         block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1595         block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1596         block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1597     }
1598 }
1599
1600 #if 0
1601 /**
1602  * dct tranforms the 16 dc values.
1603  * @param qp quantization parameter ??? FIXME
1604  */
1605 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1606 //    const int qmul= dequant_coeff[qp][0];
1607     int i;
1608     int temp[16]; //FIXME check if this is a good idea
1609     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1610     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1611
1612     for(i=0; i<4; i++){
1613         const int offset= y_offset[i];
1614         const int z0= block[offset+stride*0] + block[offset+stride*4];
1615         const int z1= block[offset+stride*0] - block[offset+stride*4];
1616         const int z2= block[offset+stride*1] - block[offset+stride*5];
1617         const int z3= block[offset+stride*1] + block[offset+stride*5];
1618
1619         temp[4*i+0]= z0+z3;
1620         temp[4*i+1]= z1+z2;
1621         temp[4*i+2]= z1-z2;
1622         temp[4*i+3]= z0-z3;
1623     }
1624
1625     for(i=0; i<4; i++){
1626         const int offset= x_offset[i];
1627         const int z0= temp[4*0+i] + temp[4*2+i];
1628         const int z1= temp[4*0+i] - temp[4*2+i];
1629         const int z2= temp[4*1+i] - temp[4*3+i];
1630         const int z3= temp[4*1+i] + temp[4*3+i];
1631
1632         block[stride*0 +offset]= (z0 + z3)>>1;
1633         block[stride*2 +offset]= (z1 + z2)>>1;
1634         block[stride*8 +offset]= (z1 - z2)>>1;
1635         block[stride*10+offset]= (z0 - z3)>>1;
1636     }
1637 }
1638 #endif
1639
1640 #undef xStride
1641 #undef stride
1642
1643 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1644     const int qmul= dequant_coeff[qp][0];
1645     const int stride= 16*2;
1646     const int xStride= 16;
1647     int a,b,c,d,e;
1648
1649     a= block[stride*0 + xStride*0];
1650     b= block[stride*0 + xStride*1];
1651     c= block[stride*1 + xStride*0];
1652     d= block[stride*1 + xStride*1];
1653
1654     e= a-b;
1655     a= a+b;
1656     b= c-d;
1657     c= c+d;
1658
1659     block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1660     block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1661     block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1662     block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1663 }
1664
1665 #if 0
1666 static void chroma_dc_dct_c(DCTELEM *block){
1667     const int stride= 16*2;
1668     const int xStride= 16;
1669     int a,b,c,d,e;
1670
1671     a= block[stride*0 + xStride*0];
1672     b= block[stride*0 + xStride*1];
1673     c= block[stride*1 + xStride*0];
1674     d= block[stride*1 + xStride*1];
1675
1676     e= a-b;
1677     a= a+b;
1678     b= c-d;
1679     c= c+d;
1680
1681     block[stride*0 + xStride*0]= (a+c);
1682     block[stride*0 + xStride*1]= (e+b);
1683     block[stride*1 + xStride*0]= (a-c);
1684     block[stride*1 + xStride*1]= (e-b);
1685 }
1686 #endif
1687
1688 /**
1689  * gets the chroma qp.
1690  */
1691 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1692     
1693     return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1694 }
1695
1696
1697 #if 0
1698 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1699     int i;
1700     //FIXME try int temp instead of block
1701     
1702     for(i=0; i<4; i++){
1703         const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1704         const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1705         const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1706         const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1707         const int z0= d0 + d3;
1708         const int z3= d0 - d3;
1709         const int z1= d1 + d2;
1710         const int z2= d1 - d2;
1711         
1712         block[0 + 4*i]=   z0 +   z1;
1713         block[1 + 4*i]= 2*z3 +   z2;
1714         block[2 + 4*i]=   z0 -   z1;
1715         block[3 + 4*i]=   z3 - 2*z2;
1716     }    
1717
1718     for(i=0; i<4; i++){
1719         const int z0= block[0*4 + i] + block[3*4 + i];
1720         const int z3= block[0*4 + i] - block[3*4 + i];
1721         const int z1= block[1*4 + i] + block[2*4 + i];
1722         const int z2= block[1*4 + i] - block[2*4 + i];
1723         
1724         block[0*4 + i]=   z0 +   z1;
1725         block[1*4 + i]= 2*z3 +   z2;
1726         block[2*4 + i]=   z0 -   z1;
1727         block[3*4 + i]=   z3 - 2*z2;
1728     }
1729 }
1730 #endif
1731
1732 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1733 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1734 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1735     int i;
1736     const int * const quant_table= quant_coeff[qscale];
1737     const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1738     const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1739     const unsigned int threshold2= (threshold1<<1);
1740     int last_non_zero;
1741
1742     if(seperate_dc){
1743         if(qscale<=18){
1744             //avoid overflows
1745             const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1746             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1747             const unsigned int dc_threshold2= (dc_threshold1<<1);
1748
1749             int level= block[0]*quant_coeff[qscale+18][0];
1750             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1751                 if(level>0){
1752                     level= (dc_bias + level)>>(QUANT_SHIFT-2);
1753                     block[0]= level;
1754                 }else{
1755                     level= (dc_bias - level)>>(QUANT_SHIFT-2);
1756                     block[0]= -level;
1757                 }
1758 //                last_non_zero = i;
1759             }else{
1760                 block[0]=0;
1761             }
1762         }else{
1763             const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1764             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1765             const unsigned int dc_threshold2= (dc_threshold1<<1);
1766
1767             int level= block[0]*quant_table[0];
1768             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1769                 if(level>0){
1770                     level= (dc_bias + level)>>(QUANT_SHIFT+1);
1771                     block[0]= level;
1772                 }else{
1773                     level= (dc_bias - level)>>(QUANT_SHIFT+1);
1774                     block[0]= -level;
1775                 }
1776 //                last_non_zero = i;
1777             }else{
1778                 block[0]=0;
1779             }
1780         }
1781         last_non_zero= 0;
1782         i=1;
1783     }else{
1784         last_non_zero= -1;
1785         i=0;
1786     }
1787
1788     for(; i<16; i++){
1789         const int j= scantable[i];
1790         int level= block[j]*quant_table[j];
1791
1792 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1793 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1794         if(((unsigned)(level+threshold1))>threshold2){
1795             if(level>0){
1796                 level= (bias + level)>>QUANT_SHIFT;
1797                 block[j]= level;
1798             }else{
1799                 level= (bias - level)>>QUANT_SHIFT;
1800                 block[j]= -level;
1801             }
1802             last_non_zero = i;
1803         }else{
1804             block[j]=0;
1805         }
1806     }
1807
1808     return last_non_zero;
1809 }
1810
1811 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1812     const uint32_t a= ((uint32_t*)(src-stride))[0];
1813     ((uint32_t*)(src+0*stride))[0]= a;
1814     ((uint32_t*)(src+1*stride))[0]= a;
1815     ((uint32_t*)(src+2*stride))[0]= a;
1816     ((uint32_t*)(src+3*stride))[0]= a;
1817 }
1818
1819 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1820     ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1821     ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1822     ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1823     ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1824 }
1825
1826 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1827     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1828                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1829     
1830     ((uint32_t*)(src+0*stride))[0]= 
1831     ((uint32_t*)(src+1*stride))[0]= 
1832     ((uint32_t*)(src+2*stride))[0]= 
1833     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1834 }
1835
1836 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1837     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1838     
1839     ((uint32_t*)(src+0*stride))[0]= 
1840     ((uint32_t*)(src+1*stride))[0]= 
1841     ((uint32_t*)(src+2*stride))[0]= 
1842     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1843 }
1844
1845 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1846     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1847     
1848     ((uint32_t*)(src+0*stride))[0]= 
1849     ((uint32_t*)(src+1*stride))[0]= 
1850     ((uint32_t*)(src+2*stride))[0]= 
1851     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1852 }
1853
1854 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1855     ((uint32_t*)(src+0*stride))[0]= 
1856     ((uint32_t*)(src+1*stride))[0]= 
1857     ((uint32_t*)(src+2*stride))[0]= 
1858     ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1859 }
1860
1861
1862 #define LOAD_TOP_RIGHT_EDGE\
1863     const int t4= topright[0];\
1864     const int t5= topright[1];\
1865     const int t6= topright[2];\
1866     const int t7= topright[3];\
1867
1868 #define LOAD_LEFT_EDGE\
1869     const int l0= src[-1+0*stride];\
1870     const int l1= src[-1+1*stride];\
1871     const int l2= src[-1+2*stride];\
1872     const int l3= src[-1+3*stride];\
1873
1874 #define LOAD_TOP_EDGE\
1875     const int t0= src[ 0-1*stride];\
1876     const int t1= src[ 1-1*stride];\
1877     const int t2= src[ 2-1*stride];\
1878     const int t3= src[ 3-1*stride];\
1879
1880 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1881     const int lt= src[-1-1*stride];
1882     LOAD_TOP_EDGE
1883     LOAD_LEFT_EDGE
1884
1885     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1886     src[0+2*stride]=
1887     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1888     src[0+1*stride]=
1889     src[1+2*stride]=
1890     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1891     src[0+0*stride]=
1892     src[1+1*stride]=
1893     src[2+2*stride]=
1894     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1895     src[1+0*stride]=
1896     src[2+1*stride]=
1897     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1898     src[2+0*stride]=
1899     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1900     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1901 }
1902
1903 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1904     LOAD_TOP_EDGE    
1905     LOAD_TOP_RIGHT_EDGE    
1906 //    LOAD_LEFT_EDGE    
1907
1908     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1909     src[1+0*stride]=
1910     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1911     src[2+0*stride]=
1912     src[1+1*stride]=
1913     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1914     src[3+0*stride]=
1915     src[2+1*stride]=
1916     src[1+2*stride]=
1917     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1918     src[3+1*stride]=
1919     src[2+2*stride]=
1920     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1921     src[3+2*stride]=
1922     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1923     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1924 }
1925
1926 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1927     const int lt= src[-1-1*stride];
1928     LOAD_TOP_EDGE    
1929     LOAD_LEFT_EDGE    
1930     const __attribute__((unused)) int unu= l3;
1931
1932     src[0+0*stride]=
1933     src[1+2*stride]=(lt + t0 + 1)>>1;
1934     src[1+0*stride]=
1935     src[2+2*stride]=(t0 + t1 + 1)>>1;
1936     src[2+0*stride]=
1937     src[3+2*stride]=(t1 + t2 + 1)>>1;
1938     src[3+0*stride]=(t2 + t3 + 1)>>1;
1939     src[0+1*stride]=
1940     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1941     src[1+1*stride]=
1942     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1943     src[2+1*stride]=
1944     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1945     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1946     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1947     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1948 }
1949
1950 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1951     LOAD_TOP_EDGE    
1952     LOAD_TOP_RIGHT_EDGE    
1953     const __attribute__((unused)) int unu= t7;
1954
1955     src[0+0*stride]=(t0 + t1 + 1)>>1;
1956     src[1+0*stride]=
1957     src[0+2*stride]=(t1 + t2 + 1)>>1;
1958     src[2+0*stride]=
1959     src[1+2*stride]=(t2 + t3 + 1)>>1;
1960     src[3+0*stride]=
1961     src[2+2*stride]=(t3 + t4+ 1)>>1;
1962     src[3+2*stride]=(t4 + t5+ 1)>>1;
1963     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1964     src[1+1*stride]=
1965     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1966     src[2+1*stride]=
1967     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1968     src[3+1*stride]=
1969     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1970     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1971 }
1972
1973 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1974     LOAD_LEFT_EDGE    
1975
1976     src[0+0*stride]=(l0 + l1 + 1)>>1;
1977     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1978     src[2+0*stride]=
1979     src[0+1*stride]=(l1 + l2 + 1)>>1;
1980     src[3+0*stride]=
1981     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1982     src[2+1*stride]=
1983     src[0+2*stride]=(l2 + l3 + 1)>>1;
1984     src[3+1*stride]=
1985     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1986     src[3+2*stride]=
1987     src[1+3*stride]=
1988     src[0+3*stride]=
1989     src[2+2*stride]=
1990     src[2+3*stride]=
1991     src[3+3*stride]=l3;
1992 }
1993     
1994 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1995     const int lt= src[-1-1*stride];
1996     LOAD_TOP_EDGE    
1997     LOAD_LEFT_EDGE    
1998     const __attribute__((unused)) int unu= t3;
1999
2000     src[0+0*stride]=
2001     src[2+1*stride]=(lt + l0 + 1)>>1;
2002     src[1+0*stride]=
2003     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2004     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2005     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2006     src[0+1*stride]=
2007     src[2+2*stride]=(l0 + l1 + 1)>>1;
2008     src[1+1*stride]=
2009     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2010     src[0+2*stride]=
2011     src[2+3*stride]=(l1 + l2+ 1)>>1;
2012     src[1+2*stride]=
2013     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2014     src[0+3*stride]=(l2 + l3 + 1)>>1;
2015     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2016 }
2017
2018 static void pred16x16_vertical_c(uint8_t *src, int stride){
2019     int i;
2020     const uint32_t a= ((uint32_t*)(src-stride))[0];
2021     const uint32_t b= ((uint32_t*)(src-stride))[1];
2022     const uint32_t c= ((uint32_t*)(src-stride))[2];
2023     const uint32_t d= ((uint32_t*)(src-stride))[3];
2024     
2025     for(i=0; i<16; i++){
2026         ((uint32_t*)(src+i*stride))[0]= a;
2027         ((uint32_t*)(src+i*stride))[1]= b;
2028         ((uint32_t*)(src+i*stride))[2]= c;
2029         ((uint32_t*)(src+i*stride))[3]= d;
2030     }
2031 }
2032
2033 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2034     int i;
2035
2036     for(i=0; i<16; i++){
2037         ((uint32_t*)(src+i*stride))[0]=
2038         ((uint32_t*)(src+i*stride))[1]=
2039         ((uint32_t*)(src+i*stride))[2]=
2040         ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2041     }
2042 }
2043
2044 static void pred16x16_dc_c(uint8_t *src, int stride){
2045     int i, dc=0;
2046
2047     for(i=0;i<16; i++){
2048         dc+= src[-1+i*stride];
2049     }
2050     
2051     for(i=0;i<16; i++){
2052         dc+= src[i-stride];
2053     }
2054
2055     dc= 0x01010101*((dc + 16)>>5);
2056
2057     for(i=0; i<16; i++){
2058         ((uint32_t*)(src+i*stride))[0]=
2059         ((uint32_t*)(src+i*stride))[1]=
2060         ((uint32_t*)(src+i*stride))[2]=
2061         ((uint32_t*)(src+i*stride))[3]= dc;
2062     }
2063 }
2064
2065 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2066     int i, dc=0;
2067
2068     for(i=0;i<16; i++){
2069         dc+= src[-1+i*stride];
2070     }
2071     
2072     dc= 0x01010101*((dc + 8)>>4);
2073
2074     for(i=0; i<16; i++){
2075         ((uint32_t*)(src+i*stride))[0]=
2076         ((uint32_t*)(src+i*stride))[1]=
2077         ((uint32_t*)(src+i*stride))[2]=
2078         ((uint32_t*)(src+i*stride))[3]= dc;
2079     }
2080 }
2081
2082 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2083     int i, dc=0;
2084
2085     for(i=0;i<16; i++){
2086         dc+= src[i-stride];
2087     }
2088     dc= 0x01010101*((dc + 8)>>4);
2089
2090     for(i=0; i<16; i++){
2091         ((uint32_t*)(src+i*stride))[0]=
2092         ((uint32_t*)(src+i*stride))[1]=
2093         ((uint32_t*)(src+i*stride))[2]=
2094         ((uint32_t*)(src+i*stride))[3]= dc;
2095     }
2096 }
2097
2098 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2099     int i;
2100
2101     for(i=0; i<16; i++){
2102         ((uint32_t*)(src+i*stride))[0]=
2103         ((uint32_t*)(src+i*stride))[1]=
2104         ((uint32_t*)(src+i*stride))[2]=
2105         ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2106     }
2107 }
2108
2109 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2110   int i, j, k;
2111   int a;
2112   uint8_t *cm = cropTbl + MAX_NEG_CROP;
2113   const uint8_t * const src0 = src+7-stride;
2114   const uint8_t *src1 = src+8*stride-1;
2115   const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2116   int H = src0[1] - src0[-1];
2117   int V = src1[0] - src2[ 0];
2118   for(k=2; k<=8; ++k) {
2119     src1 += stride; src2 -= stride;
2120     H += k*(src0[k] - src0[-k]);
2121     V += k*(src1[0] - src2[ 0]);
2122   }
2123   if(svq3){
2124     H = ( 5*(H/4) ) / 16;
2125     V = ( 5*(V/4) ) / 16;
2126
2127     /* required for 100% accuracy */
2128     i = H; H = V; V = i;
2129   }else{
2130     H = ( 5*H+32 ) >> 6;
2131     V = ( 5*V+32 ) >> 6;
2132   }
2133
2134   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2135   for(j=16; j>0; --j) {
2136     int b = a;
2137     a += V;
2138     for(i=-16; i<0; i+=4) {
2139       src[16+i] = cm[ (b    ) >> 5 ];
2140       src[17+i] = cm[ (b+  H) >> 5 ];
2141       src[18+i] = cm[ (b+2*H) >> 5 ];
2142       src[19+i] = cm[ (b+3*H) >> 5 ];
2143       b += 4*H;
2144     }
2145     src += stride;
2146   }
2147 }
2148
2149 static void pred16x16_plane_c(uint8_t *src, int stride){
2150     pred16x16_plane_compat_c(src, stride, 0);
2151 }
2152
2153 static void pred8x8_vertical_c(uint8_t *src, int stride){
2154     int i;
2155     const uint32_t a= ((uint32_t*)(src-stride))[0];
2156     const uint32_t b= ((uint32_t*)(src-stride))[1];
2157     
2158     for(i=0; i<8; i++){
2159         ((uint32_t*)(src+i*stride))[0]= a;
2160         ((uint32_t*)(src+i*stride))[1]= b;
2161     }
2162 }
2163
2164 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2165     int i;
2166
2167     for(i=0; i<8; i++){
2168         ((uint32_t*)(src+i*stride))[0]=
2169         ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2170     }
2171 }
2172
2173 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2174     int i;
2175
2176     for(i=0; i<4; i++){
2177         ((uint32_t*)(src+i*stride))[0]= 
2178         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2179     }
2180     for(i=4; i<8; i++){
2181         ((uint32_t*)(src+i*stride))[0]= 
2182         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2183     }
2184 }
2185
2186 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2187     int i;
2188     int dc0, dc2;
2189
2190     dc0=dc2=0;
2191     for(i=0;i<4; i++){
2192         dc0+= src[-1+i*stride];
2193         dc2+= src[-1+(i+4)*stride];
2194     }
2195     dc0= 0x01010101*((dc0 + 2)>>2);
2196     dc2= 0x01010101*((dc2 + 2)>>2);
2197
2198     for(i=0; i<4; i++){
2199         ((uint32_t*)(src+i*stride))[0]=
2200         ((uint32_t*)(src+i*stride))[1]= dc0;
2201     }
2202     for(i=4; i<8; i++){
2203         ((uint32_t*)(src+i*stride))[0]=
2204         ((uint32_t*)(src+i*stride))[1]= dc2;
2205     }
2206 }
2207
2208 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2209     int i;
2210     int dc0, dc1;
2211
2212     dc0=dc1=0;
2213     for(i=0;i<4; i++){
2214         dc0+= src[i-stride];
2215         dc1+= src[4+i-stride];
2216     }
2217     dc0= 0x01010101*((dc0 + 2)>>2);
2218     dc1= 0x01010101*((dc1 + 2)>>2);
2219
2220     for(i=0; i<4; i++){
2221         ((uint32_t*)(src+i*stride))[0]= dc0;
2222         ((uint32_t*)(src+i*stride))[1]= dc1;
2223     }
2224     for(i=4; i<8; i++){
2225         ((uint32_t*)(src+i*stride))[0]= dc0;
2226         ((uint32_t*)(src+i*stride))[1]= dc1;
2227     }
2228 }
2229
2230
2231 static void pred8x8_dc_c(uint8_t *src, int stride){
2232     int i;
2233     int dc0, dc1, dc2, dc3;
2234
2235     dc0=dc1=dc2=0;
2236     for(i=0;i<4; i++){
2237         dc0+= src[-1+i*stride] + src[i-stride];
2238         dc1+= src[4+i-stride];
2239         dc2+= src[-1+(i+4)*stride];
2240     }
2241     dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2242     dc0= 0x01010101*((dc0 + 4)>>3);
2243     dc1= 0x01010101*((dc1 + 2)>>2);
2244     dc2= 0x01010101*((dc2 + 2)>>2);
2245
2246     for(i=0; i<4; i++){
2247         ((uint32_t*)(src+i*stride))[0]= dc0;
2248         ((uint32_t*)(src+i*stride))[1]= dc1;
2249     }
2250     for(i=4; i<8; i++){
2251         ((uint32_t*)(src+i*stride))[0]= dc2;
2252         ((uint32_t*)(src+i*stride))[1]= dc3;
2253     }
2254 }
2255
2256 static void pred8x8_plane_c(uint8_t *src, int stride){
2257   int j, k;
2258   int a;
2259   uint8_t *cm = cropTbl + MAX_NEG_CROP;
2260   const uint8_t * const src0 = src+3-stride;
2261   const uint8_t *src1 = src+4*stride-1;
2262   const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2263   int H = src0[1] - src0[-1];
2264   int V = src1[0] - src2[ 0];
2265   for(k=2; k<=4; ++k) {
2266     src1 += stride; src2 -= stride;
2267     H += k*(src0[k] - src0[-k]);
2268     V += k*(src1[0] - src2[ 0]);
2269   }
2270   H = ( 17*H+16 ) >> 5;
2271   V = ( 17*V+16 ) >> 5;
2272
2273   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2274   for(j=8; j>0; --j) {
2275     int b = a;
2276     a += V;
2277     src[0] = cm[ (b    ) >> 5 ];
2278     src[1] = cm[ (b+  H) >> 5 ];
2279     src[2] = cm[ (b+2*H) >> 5 ];
2280     src[3] = cm[ (b+3*H) >> 5 ];
2281     src[4] = cm[ (b+4*H) >> 5 ];
2282     src[5] = cm[ (b+5*H) >> 5 ];
2283     src[6] = cm[ (b+6*H) >> 5 ];
2284     src[7] = cm[ (b+7*H) >> 5 ];
2285     src += stride;
2286   }
2287 }
2288
2289 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2290                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2291                            int src_x_offset, int src_y_offset,
2292                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2293     MpegEncContext * const s = &h->s;
2294     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2295     const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2296     const int luma_xy= (mx&3) + ((my&3)<<2);
2297     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2298     uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2299     uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2300     int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2301     int extra_height= extra_width;
2302     int emu=0;
2303     const int full_mx= mx>>2;
2304     const int full_my= my>>2;
2305     
2306     assert(pic->data[0]);
2307     
2308     if(mx&7) extra_width -= 3;
2309     if(my&7) extra_height -= 3;
2310     
2311     if(   full_mx < 0-extra_width 
2312        || full_my < 0-extra_height 
2313        || full_mx + 16/*FIXME*/ > s->width + extra_width 
2314        || full_my + 16/*FIXME*/ > s->height + extra_height){
2315         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);
2316             src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2317         emu=1;
2318     }
2319     
2320     qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2321     if(!square){
2322         qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2323     }
2324     
2325     if(s->flags&CODEC_FLAG_GRAY) return;
2326     
2327     if(emu){
2328         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);
2329             src_cb= s->edge_emu_buffer;
2330     }
2331     chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2332
2333     if(emu){
2334         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);
2335             src_cr= s->edge_emu_buffer;
2336     }
2337     chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2338 }
2339
2340 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2341                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2342                            int x_offset, int y_offset,
2343                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2344                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2345                            int list0, int list1){
2346     MpegEncContext * const s = &h->s;
2347     qpel_mc_func *qpix_op=  qpix_put;
2348     h264_chroma_mc_func chroma_op= chroma_put;
2349     
2350     dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2351     dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2352     dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2353     x_offset += 8*s->mb_x;
2354     y_offset += 8*s->mb_y;
2355     
2356     if(list0){
2357         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2358         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2359                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
2360                            qpix_op, chroma_op);
2361
2362         qpix_op=  qpix_avg;
2363         chroma_op= chroma_avg;
2364     }
2365
2366     if(list1){
2367         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2368         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2369                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
2370                            qpix_op, chroma_op);
2371     }
2372 }
2373
2374 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2375                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2376                            int x_offset, int y_offset,
2377                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2378                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2379                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2380                            int list0, int list1){
2381     MpegEncContext * const s = &h->s;
2382
2383     dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2384     dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2385     dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2386     x_offset += 8*s->mb_x;
2387     y_offset += 8*s->mb_y;
2388     
2389     if(list0 && list1){
2390         /* don't optimize for luma-only case, since B-frames usually
2391          * use implicit weights => chroma too. */
2392         uint8_t *tmp_cb = s->obmc_scratchpad;
2393         uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2394         uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2395         int refn0 = h->ref_cache[0][ scan8[n] ];
2396         int refn1 = h->ref_cache[1][ scan8[n] ];
2397
2398         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2399                     dest_y, dest_cb, dest_cr,
2400                     x_offset, y_offset, qpix_put, chroma_put);
2401         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2402                     tmp_y, tmp_cb, tmp_cr,
2403                     x_offset, y_offset, qpix_put, chroma_put);
2404
2405         if(h->use_weight == 2){
2406             int weight0 = h->implicit_weight[refn0][refn1];
2407             int weight1 = 64 - weight0;
2408             luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2409             chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2410             chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2411         }else{
2412             luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2413                             h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2414                             h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2415             chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2416                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2417                             h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2418             chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2419                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2420                             h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2421         }
2422     }else{
2423         int list = list1 ? 1 : 0;
2424         int refn = h->ref_cache[list][ scan8[n] ];
2425         Picture *ref= &h->ref_list[list][refn];
2426         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2427                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
2428                     qpix_put, chroma_put);
2429
2430         luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2431                        h->luma_weight[list][refn], h->luma_offset[list][refn]);
2432         if(h->use_weight_chroma){
2433             chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2434                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2435             chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2436                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2437         }
2438     }
2439 }
2440
2441 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2442                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2443                            int x_offset, int y_offset,
2444                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2445                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2446                            h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2447                            int list0, int list1){
2448     if((h->use_weight==2 && list0 && list1
2449         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2450        || h->use_weight==1)
2451         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2452                          x_offset, y_offset, qpix_put, chroma_put,
2453                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2454     else
2455         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2456                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2457 }
2458
2459 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2460                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2461                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2462                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2463     MpegEncContext * const s = &h->s;
2464     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2465     const int mb_type= s->current_picture.mb_type[mb_xy];
2466     
2467     assert(IS_INTER(mb_type));
2468     
2469     if(IS_16X16(mb_type)){
2470         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2471                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2472                 &weight_op[0], &weight_avg[0],
2473                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2474     }else if(IS_16X8(mb_type)){
2475         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2476                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2477                 &weight_op[1], &weight_avg[1],
2478                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2479         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2480                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2481                 &weight_op[1], &weight_avg[1],
2482                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2483     }else if(IS_8X16(mb_type)){
2484         mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2485                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2486                 &weight_op[2], &weight_avg[2],
2487                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2488         mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2489                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2490                 &weight_op[2], &weight_avg[2],
2491                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2492     }else{
2493         int i;
2494         
2495         assert(IS_8X8(mb_type));
2496
2497         for(i=0; i<4; i++){
2498             const int sub_mb_type= h->sub_mb_type[i];
2499             const int n= 4*i;
2500             int x_offset= (i&1)<<2;
2501             int y_offset= (i&2)<<1;
2502
2503             if(IS_SUB_8X8(sub_mb_type)){
2504                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2505                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2506                     &weight_op[3], &weight_avg[3],
2507                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2508             }else if(IS_SUB_8X4(sub_mb_type)){
2509                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2510                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2511                     &weight_op[4], &weight_avg[4],
2512                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2513                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2514                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2515                     &weight_op[4], &weight_avg[4],
2516                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2517             }else if(IS_SUB_4X8(sub_mb_type)){
2518                 mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2519                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2520                     &weight_op[5], &weight_avg[5],
2521                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2522                 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2523                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2524                     &weight_op[5], &weight_avg[5],
2525                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2526             }else{
2527                 int j;
2528                 assert(IS_SUB_4X4(sub_mb_type));
2529                 for(j=0; j<4; j++){
2530                     int sub_x_offset= x_offset + 2*(j&1);
2531                     int sub_y_offset= y_offset +   (j&2);
2532                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2533                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2534                         &weight_op[6], &weight_avg[6],
2535                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2536                 }
2537             }
2538         }
2539     }
2540 }
2541
2542 static void decode_init_vlc(H264Context *h){
2543     static int done = 0;
2544
2545     if (!done) {
2546         int i;
2547         done = 1;
2548
2549         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2550                  &chroma_dc_coeff_token_len [0], 1, 1,
2551                  &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2552
2553         for(i=0; i<4; i++){
2554             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2555                      &coeff_token_len [i][0], 1, 1,
2556                      &coeff_token_bits[i][0], 1, 1, 1);
2557         }
2558
2559         for(i=0; i<3; i++){
2560             init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2561                      &chroma_dc_total_zeros_len [i][0], 1, 1,
2562                      &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2563         }
2564         for(i=0; i<15; i++){
2565             init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2566                      &total_zeros_len [i][0], 1, 1,
2567                      &total_zeros_bits[i][0], 1, 1, 1);
2568         }
2569
2570         for(i=0; i<6; i++){
2571             init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2572                      &run_len [i][0], 1, 1,
2573                      &run_bits[i][0], 1, 1, 1);
2574         }
2575         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2576                  &run_len [6][0], 1, 1,
2577                  &run_bits[6][0], 1, 1, 1);
2578     }
2579 }
2580
2581 /**
2582  * Sets the intra prediction function pointers.
2583  */
2584 static void init_pred_ptrs(H264Context *h){
2585 //    MpegEncContext * const s = &h->s;
2586
2587     h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2588     h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2589     h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2590     h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2591     h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2592     h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2593     h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2594     h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2595     h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2596     h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2597     h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2598     h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2599
2600     h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2601     h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2602     h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2603     h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2604     h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2605     h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2606     h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2607
2608     h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2609     h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2610     h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2611     h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2612     h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2613     h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2614     h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2615 }
2616
2617 static void free_tables(H264Context *h){
2618     av_freep(&h->intra4x4_pred_mode);
2619     av_freep(&h->chroma_pred_mode_table);
2620     av_freep(&h->cbp_table);
2621     av_freep(&h->mvd_table[0]);
2622     av_freep(&h->mvd_table[1]);
2623     av_freep(&h->direct_table);
2624     av_freep(&h->non_zero_count);
2625     av_freep(&h->slice_table_base);
2626     av_freep(&h->top_borders[1]);
2627     av_freep(&h->top_borders[0]);
2628     h->slice_table= NULL;
2629
2630     av_freep(&h->mb2b_xy);
2631     av_freep(&h->mb2b8_xy);
2632
2633     av_freep(&h->s.obmc_scratchpad);
2634 }
2635
2636 /**
2637  * allocates tables.
2638  * needs widzh/height
2639  */
2640 static int alloc_tables(H264Context *h){
2641     MpegEncContext * const s = &h->s;
2642     const int big_mb_num= s->mb_stride * (s->mb_height+1);
2643     int x,y;
2644
2645     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2646
2647     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2648     CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2649     CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2650     CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2651     CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2652
2653     if( h->pps.cabac ) {
2654         CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2655         CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2656         CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2657         CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2658     }
2659
2660     memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2661     h->slice_table= h->slice_table_base + s->mb_stride + 1;
2662
2663     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2664     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2665     for(y=0; y<s->mb_height; y++){
2666         for(x=0; x<s->mb_width; x++){
2667             const int mb_xy= x + y*s->mb_stride;
2668             const int b_xy = 4*x + 4*y*h->b_stride;
2669             const int b8_xy= 2*x + 2*y*h->b8_stride;
2670         
2671             h->mb2b_xy [mb_xy]= b_xy;
2672             h->mb2b8_xy[mb_xy]= b8_xy;
2673         }
2674     }
2675
2676     s->obmc_scratchpad = NULL;
2677
2678     return 0;
2679 fail:
2680     free_tables(h);
2681     return -1;
2682 }
2683
2684 static void common_init(H264Context *h){
2685     MpegEncContext * const s = &h->s;
2686
2687     s->width = s->avctx->width;
2688     s->height = s->avctx->height;
2689     s->codec_id= s->avctx->codec->id;
2690     
2691     init_pred_ptrs(h);
2692
2693     s->unrestricted_mv=1;
2694     s->decode=1; //FIXME
2695 }
2696
2697 static int decode_init(AVCodecContext *avctx){
2698     H264Context *h= avctx->priv_data;
2699     MpegEncContext * const s = &h->s;
2700
2701     MPV_decode_defaults(s);
2702     
2703     s->avctx = avctx;
2704     common_init(h);
2705
2706     s->out_format = FMT_H264;
2707     s->workaround_bugs= avctx->workaround_bugs;
2708
2709     // set defaults
2710 //    s->decode_mb= ff_h263_decode_mb;
2711     s->low_delay= 1;
2712     avctx->pix_fmt= PIX_FMT_YUV420P;
2713
2714     decode_init_vlc(h);
2715     
2716     if(avctx->extradata_size > 0 && avctx->extradata &&
2717        *(char *)avctx->extradata == 1){
2718         h->is_avc = 1;
2719         h->got_avcC = 0;
2720     } else {
2721         h->is_avc = 0;
2722     }
2723
2724     return 0;
2725 }
2726
2727 static void frame_start(H264Context *h){
2728     MpegEncContext * const s = &h->s;
2729     int i;
2730
2731     MPV_frame_start(s, s->avctx);
2732     ff_er_frame_start(s);
2733
2734     assert(s->linesize && s->uvlinesize);
2735
2736     for(i=0; i<16; i++){
2737         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2738         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2739     }
2740     for(i=0; i<4; i++){
2741         h->block_offset[16+i]=
2742         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2743         h->block_offset[24+16+i]=
2744         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2745     }
2746
2747     /* can't be in alloc_tables because linesize isn't known there.
2748      * FIXME: redo bipred weight to not require extra buffer? */
2749     if(!s->obmc_scratchpad)
2750         s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2751
2752 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2753 }
2754
2755 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2756     MpegEncContext * const s = &h->s;
2757     int i;
2758     
2759     src_y  -=   linesize;
2760     src_cb -= uvlinesize;
2761     src_cr -= uvlinesize;
2762
2763     // There is two lines saved, the line above the the top macroblock of a pair,
2764     // and the line above the bottom macroblock
2765     h->left_border[0]= h->top_borders[0][s->mb_x][15];
2766     for(i=1; i<17; i++){
2767         h->left_border[i]= src_y[15+i*  linesize];
2768     }
2769     
2770     *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2771     *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2772
2773     if(!(s->flags&CODEC_FLAG_GRAY)){
2774         h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
2775         h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2776         for(i=1; i<9; i++){
2777             h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2778             h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2779         }
2780         *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2781         *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2782     }
2783 }
2784
2785 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){
2786     MpegEncContext * const s = &h->s;
2787     int temp8, i;
2788     uint64_t temp64;
2789     int deblock_left = (s->mb_x > 0);
2790     int deblock_top  = (s->mb_y > 0);
2791
2792     src_y  -=   linesize + 1;
2793     src_cb -= uvlinesize + 1;
2794     src_cr -= uvlinesize + 1;
2795
2796 #define XCHG(a,b,t,xchg)\
2797 t= a;\
2798 if(xchg)\
2799     a= b;\
2800 b= t;
2801
2802     if(deblock_left){
2803         for(i = !deblock_top; i<17; i++){
2804             XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2805         }
2806     }
2807
2808     if(deblock_top){
2809         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2810         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2811     }
2812
2813     if(!(s->flags&CODEC_FLAG_GRAY)){
2814         if(deblock_left){
2815             for(i = !deblock_top; i<9; i++){
2816                 XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2817                 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2818             }
2819         }
2820         if(deblock_top){
2821             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2822             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2823         }
2824     }
2825 }
2826
2827 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2828     MpegEncContext * const s = &h->s;
2829     int i;
2830     
2831     src_y  -= 2 *   linesize;
2832     src_cb -= 2 * uvlinesize;
2833     src_cr -= 2 * uvlinesize;
2834
2835     // There is two lines saved, the line above the the top macroblock of a pair,
2836     // and the line above the bottom macroblock
2837     h->left_border[0]= h->top_borders[0][s->mb_x][15];
2838     h->left_border[1]= h->top_borders[1][s->mb_x][15];
2839     for(i=2; i<34; i++){
2840         h->left_border[i]= src_y[15+i*  linesize];
2841     }
2842     
2843     *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
2844     *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2845     *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
2846     *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2847
2848     if(!(s->flags&CODEC_FLAG_GRAY)){
2849         h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
2850         h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
2851         h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
2852         h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2853         for(i=2; i<18; i++){
2854             h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
2855             h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2856         }
2857         *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2858         *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2859         *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2860         *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2861     }
2862 }
2863
2864 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2865     MpegEncContext * const s = &h->s;
2866     int temp8, i;
2867     uint64_t temp64;
2868     int deblock_left = (s->mb_x > 0);
2869     int deblock_top  = (s->mb_y > 0);
2870
2871     tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
2872
2873     src_y  -= 2 *   linesize + 1;
2874     src_cb -= 2 * uvlinesize + 1;
2875     src_cr -= 2 * uvlinesize + 1;
2876
2877 #define XCHG(a,b,t,xchg)\
2878 t= a;\
2879 if(xchg)\
2880     a= b;\
2881 b= t;
2882
2883     if(deblock_left){
2884         for(i = (!deblock_top)<<1; i<34; i++){
2885             XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2886         }
2887     }
2888
2889     if(deblock_top){
2890         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2891         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2892         XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2893         XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2894     }
2895
2896     if(!(s->flags&CODEC_FLAG_GRAY)){
2897         if(deblock_left){
2898             for(i = (!deblock_top) << 1; i<18; i++){
2899                 XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
2900                 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2901             }
2902         }
2903         if(deblock_top){
2904             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2905             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2906             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2907             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2908         }
2909     }
2910 }
2911
2912 static void hl_decode_mb(H264Context *h){
2913     MpegEncContext * const s = &h->s;
2914     const int mb_x= s->mb_x;
2915     const int mb_y= s->mb_y;
2916     const int mb_xy= mb_x + mb_y*s->mb_stride;
2917     const int mb_type= s->current_picture.mb_type[mb_xy];
2918     uint8_t  *dest_y, *dest_cb, *dest_cr;
2919     int linesize, uvlinesize /*dct_offset*/;
2920     int i;
2921     int *block_offset = &h->block_offset[0];
2922     const unsigned int bottom = mb_y & 1;
2923
2924     if(!s->decode)
2925         return;
2926
2927     dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2928     dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2929     dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2930
2931     if (h->mb_field_decoding_flag) {
2932         linesize = s->linesize * 2;
2933         uvlinesize = s->uvlinesize * 2;
2934         block_offset = &h->block_offset[24];
2935         if(mb_y&1){ //FIXME move out of this func?
2936             dest_y -= s->linesize*15;
2937             dest_cb-= s->uvlinesize*7;
2938             dest_cr-= s->uvlinesize*7;
2939         }
2940     } else {
2941         linesize = s->linesize;
2942         uvlinesize = s->uvlinesize;
2943 //        dct_offset = s->linesize * 16;
2944     }
2945
2946     if (IS_INTRA_PCM(mb_type)) {
2947         unsigned int x, y;
2948
2949         // The pixels are stored in h->mb array in the same order as levels,
2950         // copy them in output in the correct order.
2951         for(i=0; i<16; i++) {
2952             for (y=0; y<4; y++) {
2953                 for (x=0; x<4; x++) {
2954                     *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2955                 }
2956             }
2957         }
2958         for(i=16; i<16+4; i++) {
2959             for (y=0; y<4; y++) {
2960                 for (x=0; x<4; x++) {
2961                     *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2962                 }
2963             }
2964         }
2965         for(i=20; i<20+4; i++) {
2966             for (y=0; y<4; y++) {
2967                 for (x=0; x<4; x++) {
2968                     *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2969                 }
2970             }
2971         }
2972     } else {
2973         if(IS_INTRA(mb_type)){
2974             if(h->deblocking_filter) {
2975                 if (h->mb_aff_frame) {
2976                     if (!bottom)
2977                         xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
2978                 } else {
2979                     xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2980                 }
2981             }
2982
2983             if(!(s->flags&CODEC_FLAG_GRAY)){
2984                 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2985                 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2986             }
2987
2988             if(IS_INTRA4x4(mb_type)){
2989                 if(!s->encoding){
2990                     for(i=0; i<16; i++){
2991                         uint8_t * const ptr= dest_y + block_offset[i];
2992                         uint8_t *topright;
2993                         const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2994                         int tr;
2995
2996                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2997                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2998                             assert(mb_y || linesize <= block_offset[i]);
2999                             if(!topright_avail){
3000                                 tr= ptr[3 - linesize]*0x01010101;
3001                                 topright= (uint8_t*) &tr;
3002                             }else if(i==5 && h->deblocking_filter){
3003                                 tr= *(uint32_t*)h->top_borders[h->mb_aff_frame ? IS_INTERLACED(mb_type) ? bottom : 1 : 0][mb_x+1];
3004                                 topright= (uint8_t*) &tr;
3005                             }else
3006                                 topright= ptr + 4 - linesize;
3007                         }else
3008                             topright= NULL;
3009
3010                         h->pred4x4[ dir ](ptr, topright, linesize);
3011                         if(h->non_zero_count_cache[ scan8[i] ]){
3012                             if(s->codec_id == CODEC_ID_H264)
3013                                 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3014                             else
3015                                 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3016                         }
3017                     }
3018                 }
3019             }else{
3020                 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3021                 if(s->codec_id == CODEC_ID_H264)
3022                     h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3023                 else
3024                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3025             }
3026             if(h->deblocking_filter) {
3027                 if (h->mb_aff_frame) {
3028                     if (bottom) {
3029                         uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
3030                         uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3031                         uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3032                         s->mb_y--;
3033                         xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3034                         s->mb_y++;
3035                     }
3036                 } else {
3037                     xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3038                 }
3039             }
3040         }else if(s->codec_id == CODEC_ID_H264){
3041             hl_motion(h, dest_y, dest_cb, dest_cr,
3042                       s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
3043                       s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3044                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3045         }
3046
3047
3048         if(!IS_INTRA4x4(mb_type)){
3049             if(s->codec_id == CODEC_ID_H264){
3050                 for(i=0; i<16; i++){
3051                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3052                         uint8_t * const ptr= dest_y + block_offset[i];
3053                         s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3054                     }
3055                 }
3056             }else{
3057                 for(i=0; i<16; i++){
3058                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3059                         uint8_t * const ptr= dest_y + block_offset[i];
3060                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3061                     }
3062                 }
3063             }
3064         }
3065
3066         if(!(s->flags&CODEC_FLAG_GRAY)){
3067             chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3068             chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3069             if(s->codec_id == CODEC_ID_H264){
3070                 for(i=16; i<16+4; i++){
3071                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3072                         uint8_t * const ptr= dest_cb + block_offset[i];
3073                         s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3074                     }
3075                 }
3076                 for(i=20; i<20+4; i++){
3077                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3078                         uint8_t * const ptr= dest_cr + block_offset[i];
3079                         s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3080                     }
3081                 }
3082             }else{
3083                 for(i=16; i<16+4; i++){
3084                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3085                         uint8_t * const ptr= dest_cb + block_offset[i];
3086                         svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3087                     }
3088                 }
3089                 for(i=20; i<20+4; i++){
3090                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3091                         uint8_t * const ptr= dest_cr + block_offset[i];
3092                         svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3093                     }
3094                 }
3095             }
3096         }
3097     }
3098     if(h->deblocking_filter) {
3099         if (h->mb_aff_frame) {
3100             const int mb_y = s->mb_y - 1;
3101             uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3102             const int mb_xy= mb_x + mb_y*s->mb_stride;
3103             const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3104             const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3105             uint8_t tmp = s->current_picture.data[1][384];
3106             if (!bottom) return;
3107             pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3108             pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3109             pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3110
3111             backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3112             // TODO deblock a pair
3113             // top 
3114             s->mb_y--;
3115             tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3116             fill_caches(h, mb_type_top, 1); //FIXME dont fill stuff which isnt used by filter_mb
3117             filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3118             if (tmp != s->current_picture.data[1][384]) {
3119                 tprintf("modified pixel 8,1 (1)\n");
3120             }
3121             // bottom
3122             s->mb_y++;
3123             tprintf("call mbaff filter_mb\n");
3124             fill_caches(h, mb_type_bottom, 1); //FIXME dont fill stuff which isnt used by filter_mb
3125             filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3126             if (tmp != s->current_picture.data[1][384]) {
3127                 tprintf("modified pixel 8,1 (2)\n");
3128             }
3129         } else {
3130             tprintf("call filter_mb\n");
3131             backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3132             fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
3133             filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3134         }
3135     }
3136 }
3137
3138 /**
3139  * fills the default_ref_list.
3140  */
3141 static int fill_default_ref_list(H264Context *h){
3142     MpegEncContext * const s = &h->s;
3143     int i;
3144     int smallest_poc_greater_than_current = -1;
3145     Picture sorted_short_ref[32];
3146     
3147     if(h->slice_type==B_TYPE){
3148         int out_i;
3149         int limit= -1;
3150
3151         /* sort frame according to poc in B slice */
3152         for(out_i=0; out_i<h->short_ref_count; out_i++){
3153             int best_i=-1;
3154             int best_poc=INT_MAX;
3155
3156             for(i=0; i<h->short_ref_count; i++){
3157                 const int poc= h->short_ref[i]->poc;
3158                 if(poc > limit && poc < best_poc){
3159                     best_poc= poc;
3160                     best_i= i;
3161                 }
3162             }
3163             
3164             assert(best_i != -1);
3165             
3166             limit= best_poc;
3167             sorted_short_ref[out_i]= *h->short_ref[best_i];
3168             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);
3169             if (-1 == smallest_poc_greater_than_current) {
3170                 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3171                     smallest_poc_greater_than_current = out_i;
3172                 }
3173             }
3174         }
3175     }
3176
3177     if(s->picture_structure == PICT_FRAME){
3178         if(h->slice_type==B_TYPE){
3179             int list;
3180             tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3181
3182             // find the largest poc
3183             for(list=0; list<2; list++){
3184                 int index = 0;
3185                 int j= -99;
3186                 int step= list ? -1 : 1;
3187
3188                 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3189                     while(j<0 || j>= h->short_ref_count){
3190                         step = -step;
3191                         j= smallest_poc_greater_than_current + (step>>1);
3192                     }
3193                     if(sorted_short_ref[j].reference != 3) continue;
3194                     h->default_ref_list[list][index  ]= sorted_short_ref[j];
3195                     h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3196                 }
3197
3198                 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3199                     if(h->long_ref[i] == NULL) continue;
3200                     if(h->long_ref[i]->reference != 3) continue;
3201
3202                     h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3203                     h->default_ref_list[ list ][index++].pic_id= i;;
3204                 }
3205                 
3206                 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3207                     // swap the two first elements of L1 when
3208                     // L0 and L1 are identical
3209                     Picture temp= h->default_ref_list[1][0];
3210                     h->default_ref_list[1][0] = h->default_ref_list[1][1];
3211                     h->default_ref_list[1][0] = temp;
3212                 }
3213
3214                 if(index < h->ref_count[ list ])
3215                     memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3216             }
3217         }else{
3218             int index=0;
3219             for(i=0; i<h->short_ref_count; i++){
3220                 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3221                 h->default_ref_list[0][index  ]= *h->short_ref[i];
3222                 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3223             }
3224             for(i = 0; i < 16; i++){
3225                 if(h->long_ref[i] == NULL) continue;
3226                 if(h->long_ref[i]->reference != 3) continue;
3227                 h->default_ref_list[0][index  ]= *h->long_ref[i];
3228                 h->default_ref_list[0][index++].pic_id= i;;
3229             }
3230             if(index < h->ref_count[0])
3231                 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3232         }
3233     }else{ //FIELD
3234         if(h->slice_type==B_TYPE){
3235         }else{
3236             //FIXME second field balh
3237         }
3238     }
3239 #ifdef TRACE
3240     for (i=0; i<h->ref_count[0]; i++) {
3241         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]);
3242     }
3243     if(h->slice_type==B_TYPE){
3244         for (i=0; i<h->ref_count[1]; i++) {
3245             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]);
3246         }
3247     }
3248 #endif
3249     return 0;
3250 }
3251
3252 static void print_short_term(H264Context *h);
3253 static void print_long_term(H264Context *h);
3254
3255 static int decode_ref_pic_list_reordering(H264Context *h){
3256     MpegEncContext * const s = &h->s;
3257     int list;
3258     
3259     print_short_term(h);
3260     print_long_term(h);
3261     if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
3262     
3263     for(list=0; list<2; list++){
3264         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3265
3266         if(get_bits1(&s->gb)){
3267             int pred= h->curr_pic_num;
3268             int index;
3269
3270             for(index=0; ; index++){
3271                 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3272                 int pic_id;
3273                 int i;
3274                 Picture *ref = NULL;
3275                 
3276                 if(reordering_of_pic_nums_idc==3) 
3277                     break;
3278                 
3279                 if(index >= h->ref_count[list]){
3280                     av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3281                     return -1;
3282                 }
3283                 
3284                 if(reordering_of_pic_nums_idc<3){
3285                     if(reordering_of_pic_nums_idc<2){
3286                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3287
3288                         if(abs_diff_pic_num >= h->max_pic_num){
3289                             av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3290                             return -1;
3291                         }
3292
3293                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3294                         else                                pred+= abs_diff_pic_num;
3295                         pred &= h->max_pic_num - 1;
3296                     
3297                         for(i= h->short_ref_count-1; i>=0; i--){
3298                             ref = h->short_ref[i];
3299                             if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3300                                 break;
3301                         }
3302                     }else{
3303                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3304                         ref = h->long_ref[pic_id];
3305                     }
3306
3307                     if (i < 0) {
3308                         av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3309                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3310                     } else {
3311                         h->ref_list[list][index]= *ref;
3312                     }
3313                 }else{
3314                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3315                     return -1;
3316                 }
3317             }
3318         }
3319
3320         if(h->slice_type!=B_TYPE) break;
3321     }
3322     
3323     if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3324         direct_dist_scale_factor(h);
3325     direct_ref_list_init(h);
3326     return 0;    
3327 }
3328
3329 static int pred_weight_table(H264Context *h){
3330     MpegEncContext * const s = &h->s;
3331     int list, i;
3332     int luma_def, chroma_def;
3333     
3334     h->use_weight= 0;
3335     h->use_weight_chroma= 0;
3336     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3337     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3338     luma_def = 1<<h->luma_log2_weight_denom;
3339     chroma_def = 1<<h->chroma_log2_weight_denom;
3340
3341     for(list=0; list<2; list++){
3342         for(i=0; i<h->ref_count[list]; i++){
3343             int luma_weight_flag, chroma_weight_flag;
3344             
3345             luma_weight_flag= get_bits1(&s->gb);
3346             if(luma_weight_flag){
3347                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3348                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3349                 if(   h->luma_weight[list][i] != luma_def
3350                    || h->luma_offset[list][i] != 0)
3351                     h->use_weight= 1;
3352             }else{
3353                 h->luma_weight[list][i]= luma_def;
3354                 h->luma_offset[list][i]= 0;
3355             }
3356
3357             chroma_weight_flag= get_bits1(&s->gb);
3358             if(chroma_weight_flag){
3359                 int j;
3360                 for(j=0; j<2; j++){
3361                     h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3362                     h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3363                     if(   h->chroma_weight[list][i][j] != chroma_def
3364                        || h->chroma_offset[list][i][j] != 0)
3365                         h->use_weight_chroma= 1;
3366                 }
3367             }else{
3368                 int j;
3369                 for(j=0; j<2; j++){
3370                     h->chroma_weight[list][i][j]= chroma_def;
3371                     h->chroma_offset[list][i][j]= 0;
3372                 }
3373             }
3374         }
3375         if(h->slice_type != B_TYPE) break;
3376     }
3377     h->use_weight= h->use_weight || h->use_weight_chroma;
3378     return 0;
3379 }
3380
3381 static void implicit_weight_table(H264Context *h){
3382     MpegEncContext * const s = &h->s;
3383     int ref0, ref1;
3384     int cur_poc = s->current_picture_ptr->poc;
3385
3386     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3387        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3388         h->use_weight= 0;
3389         h->use_weight_chroma= 0;
3390         return;
3391     }
3392
3393     h->use_weight= 2;
3394     h->use_weight_chroma= 2;
3395     h->luma_log2_weight_denom= 5;
3396     h->chroma_log2_weight_denom= 5;
3397
3398     /* FIXME: MBAFF */
3399     for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3400         int poc0 = h->ref_list[0][ref0].poc;
3401         for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3402             int poc1 = h->ref_list[1][ref1].poc;
3403             int td = clip(poc1 - poc0, -128, 127);
3404             if(td){
3405                 int tb = clip(cur_poc - poc0, -128, 127);
3406                 int tx = (16384 + (ABS(td) >> 1)) / td;
3407                 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3408                 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3409                     h->implicit_weight[ref0][ref1] = 32;
3410                 else
3411                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3412             }else
3413                 h->implicit_weight[ref0][ref1] = 32;
3414         }
3415     }
3416 }
3417
3418 static inline void unreference_pic(H264Context *h, Picture *pic){
3419     int i;
3420     pic->reference=0;
3421     if(pic == h->delayed_output_pic)
3422         pic->reference=1;
3423     else{
3424         for(i = 0; h->delayed_pic[i]; i++)
3425             if(pic == h->delayed_pic[i]){
3426                 pic->reference=1;
3427                 break;
3428             }
3429     }
3430 }
3431
3432 /**
3433  * instantaneous decoder refresh.
3434  */
3435 static void idr(H264Context *h){
3436     int i;
3437
3438     for(i=0; i<16; i++){
3439         if (h->long_ref[i] != NULL) {
3440             unreference_pic(h, h->long_ref[i]);
3441             h->long_ref[i]= NULL;
3442         }
3443     }
3444     h->long_ref_count=0;
3445
3446     for(i=0; i<h->short_ref_count; i++){
3447         unreference_pic(h, h->short_ref[i]);
3448         h->short_ref[i]= NULL;
3449     }
3450     h->short_ref_count=0;
3451 }
3452
3453 /**
3454  *
3455  * @return the removed picture or NULL if an error occures
3456  */
3457 static Picture * remove_short(H264Context *h, int frame_num){
3458     MpegEncContext * const s = &h->s;
3459     int i;
3460     
3461     if(s->avctx->debug&FF_DEBUG_MMCO)
3462         av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3463     
3464     for(i=0; i<h->short_ref_count; i++){
3465         Picture *pic= h->short_ref[i];
3466         if(s->avctx->debug&FF_DEBUG_MMCO)
3467             av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3468         if(pic->frame_num == frame_num){
3469             h->short_ref[i]= NULL;
3470             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3471             h->short_ref_count--;
3472             return pic;
3473         }
3474     }
3475     return NULL;
3476 }
3477
3478 /**
3479  *
3480  * @return the removed picture or NULL if an error occures
3481  */
3482 static Picture * remove_long(H264Context *h, int i){
3483     Picture *pic;
3484
3485     pic= h->long_ref[i];
3486     h->long_ref[i]= NULL;
3487     if(pic) h->long_ref_count--;
3488
3489     return pic;
3490 }
3491
3492 /**
3493  * print short term list
3494  */
3495 static void print_short_term(H264Context *h) {
3496     uint32_t i;
3497     if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3498         av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3499         for(i=0; i<h->short_ref_count; i++){
3500             Picture *pic= h->short_ref[i];
3501             av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3502         }
3503     }
3504 }
3505
3506 /**
3507  * print long term list
3508  */
3509 static void print_long_term(H264Context *h) {
3510     uint32_t i;
3511     if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3512         av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3513         for(i = 0; i < 16; i++){
3514             Picture *pic= h->long_ref[i];
3515             if (pic) {
3516                 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3517             }
3518         }
3519     }
3520 }
3521
3522 /**
3523  * Executes the reference picture marking (memory management control operations).
3524  */
3525 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3526     MpegEncContext * const s = &h->s;
3527     int i, j;
3528     int current_is_long=0;
3529     Picture *pic;
3530     
3531     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3532         av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3533         
3534     for(i=0; i<mmco_count; i++){
3535         if(s->avctx->debug&FF_DEBUG_MMCO)
3536             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);
3537
3538         switch(mmco[i].opcode){
3539         case MMCO_SHORT2UNUSED:
3540             pic= remove_short(h, mmco[i].short_frame_num);
3541             if(pic==NULL) return -1;
3542             unreference_pic(h, pic);
3543             break;
3544         case MMCO_SHORT2LONG:
3545             pic= remove_long(h, mmco[i].long_index);
3546             if(pic) unreference_pic(h, pic);
3547             
3548             h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3549             h->long_ref[ mmco[i].long_index ]->long_ref=1;
3550             h->long_ref_count++;
3551             break;
3552         case MMCO_LONG2UNUSED:
3553             pic= remove_long(h, mmco[i].long_index);
3554             if(pic==NULL) return -1;
3555             unreference_pic(h, pic);
3556             break;
3557         case MMCO_LONG:
3558             pic= remove_long(h, mmco[i].long_index);
3559             if(pic) unreference_pic(h, pic);
3560             
3561             h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3562             h->long_ref[ mmco[i].long_index ]->long_ref=1;
3563             h->long_ref_count++;
3564             
3565             current_is_long=1;
3566             break;
3567         case MMCO_SET_MAX_LONG:
3568             assert(mmco[i].long_index <= 16);
3569             // just remove the long term which index is greater than new max
3570             for(j = mmco[i].long_index; j<16; j++){
3571                 pic = remove_long(h, j);
3572                 if (pic) unreference_pic(h, pic);
3573             }
3574             break;
3575         case MMCO_RESET:
3576             while(h->short_ref_count){
3577                 pic= remove_short(h, h->short_ref[0]->frame_num);
3578                 unreference_pic(h, pic);
3579             }
3580             for(j = 0; j < 16; j++) {
3581                 pic= remove_long(h, j);
3582                 if(pic) unreference_pic(h, pic);
3583             }
3584             break;
3585         default: assert(0);
3586         }
3587     }
3588     
3589     if(!current_is_long){
3590         pic= remove_short(h, s->current_picture_ptr->frame_num);
3591         if(pic){
3592             unreference_pic(h, pic);
3593             av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3594         }
3595         
3596         if(h->short_ref_count)
3597             memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3598
3599         h->short_ref[0]= s->current_picture_ptr;
3600         h->short_ref[0]->long_ref=0;
3601         h->short_ref_count++;
3602     }
3603     
3604     print_short_term(h);
3605     print_long_term(h);
3606     return 0; 
3607 }
3608
3609 static int decode_ref_pic_marking(H264Context *h){
3610     MpegEncContext * const s = &h->s;
3611     int i;
3612     
3613     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3614         s->broken_link= get_bits1(&s->gb) -1;
3615         h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3616         if(h->mmco[0].long_index == -1)
3617             h->mmco_index= 0;
3618         else{
3619             h->mmco[0].opcode= MMCO_LONG;
3620             h->mmco_index= 1;
3621         } 
3622     }else{
3623         if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3624             for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3625                 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3626
3627                 h->mmco[i].opcode= opcode;
3628                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3629                     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
3630 /*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3631                         fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3632                         return -1;
3633                     }*/
3634                 }
3635                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3636                     h->mmco[i].long_index= get_ue_golomb(&s->gb);
3637                     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){
3638                         av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3639                         return -1;
3640                     }
3641                 }
3642                     
3643                 if(opcode > MMCO_LONG){
3644                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3645                     return -1;
3646                 }
3647                 if(opcode == MMCO_END)
3648                     break;
3649             }
3650             h->mmco_index= i;
3651         }else{
3652             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3653
3654             if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3655                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3656                 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3657                 h->mmco_index= 1;
3658             }else
3659                 h->mmco_index= 0;
3660         }
3661     }
3662     
3663     return 0; 
3664 }
3665
3666 static int init_poc(H264Context *h){
3667     MpegEncContext * const s = &h->s;
3668     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3669     int field_poc[2];
3670
3671     if(h->nal_unit_type == NAL_IDR_SLICE){
3672         h->frame_num_offset= 0;
3673     }else{
3674         if(h->frame_num < h->prev_frame_num)
3675             h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3676         else
3677             h->frame_num_offset= h->prev_frame_num_offset;
3678     }
3679
3680     if(h->sps.poc_type==0){
3681         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3682
3683         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3684             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3685         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3686             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3687         else
3688             h->poc_msb = h->prev_poc_msb;
3689 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3690         field_poc[0] = 
3691         field_poc[1] = h->poc_msb + h->poc_lsb;
3692         if(s->picture_structure == PICT_FRAME) 
3693             field_poc[1] += h->delta_poc_bottom;
3694     }else if(h->sps.poc_type==1){
3695         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3696         int i;
3697
3698         if(h->sps.poc_cycle_length != 0)
3699             abs_frame_num = h->frame_num_offset + h->frame_num;
3700         else
3701             abs_frame_num = 0;
3702
3703         if(h->nal_ref_idc==0 && abs_frame_num > 0)
3704             abs_frame_num--;
3705             
3706         expected_delta_per_poc_cycle = 0;
3707         for(i=0; i < h->sps.poc_cycle_length; i++)
3708             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3709
3710         if(abs_frame_num > 0){
3711             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3712             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3713
3714             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3715             for(i = 0; i <= frame_num_in_poc_cycle; i++)
3716                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3717         } else
3718             expectedpoc = 0;
3719
3720         if(h->nal_ref_idc == 0) 
3721             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3722         
3723         field_poc[0] = expectedpoc + h->delta_poc[0];
3724         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3725
3726         if(s->picture_structure == PICT_FRAME)
3727             field_poc[1] += h->delta_poc[1];
3728     }else{
3729         int poc;
3730         if(h->nal_unit_type == NAL_IDR_SLICE){
3731             poc= 0;
3732         }else{
3733             if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3734             else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3735         }
3736         field_poc[0]= poc;
3737         field_poc[1]= poc;
3738     }
3739     
3740     if(s->picture_structure != PICT_BOTTOM_FIELD)
3741         s->current_picture_ptr->field_poc[0]= field_poc[0];
3742     if(s->picture_structure != PICT_TOP_FIELD)
3743         s->current_picture_ptr->field_poc[1]= field_poc[1];
3744     if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3745         s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3746
3747     return 0;
3748 }
3749
3750 /**
3751  * decodes a slice header.
3752  * this will allso call MPV_common_init() and frame_start() as needed
3753  */
3754 static int decode_slice_header(H264Context *h){
3755     MpegEncContext * const s = &h->s;
3756     int first_mb_in_slice, pps_id;
3757     int num_ref_idx_active_override_flag;
3758     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3759     int slice_type;
3760     int default_ref_list_done = 0;
3761
3762     s->current_picture.reference= h->nal_ref_idc != 0;
3763     s->dropable= h->nal_ref_idc == 0;
3764
3765     first_mb_in_slice= get_ue_golomb(&s->gb);
3766
3767     slice_type= get_ue_golomb(&s->gb);
3768     if(slice_type > 9){
3769         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);
3770         return -1;
3771     }
3772     if(slice_type > 4){
3773         slice_type -= 5;
3774         h->slice_type_fixed=1;
3775     }else
3776         h->slice_type_fixed=0;
3777     
3778     slice_type= slice_type_map[ slice_type ];
3779     if (slice_type == I_TYPE
3780         || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3781         default_ref_list_done = 1;
3782     }
3783     h->slice_type= slice_type;
3784
3785     s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3786         
3787     pps_id= get_ue_golomb(&s->gb);
3788     if(pps_id>255){
3789         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3790         return -1;
3791     }
3792     h->pps= h->pps_buffer[pps_id];
3793     if(h->pps.slice_group_count == 0){
3794         av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3795         return -1;
3796     }
3797
3798     h->sps= h->sps_buffer[ h->pps.sps_id ];
3799     if(h->sps.log2_max_frame_num == 0){
3800         av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3801         return -1;
3802     }
3803     
3804     s->mb_width= h->sps.mb_width;
3805     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3806     
3807     h->b_stride=  s->mb_width*4 + 1;
3808     h->b8_stride= s->mb_width*2 + 1;
3809
3810     s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3811     if(h->sps.frame_mbs_only_flag)
3812         s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3813     else
3814         s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3815     
3816     if (s->context_initialized 
3817         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3818         free_tables(h);
3819         MPV_common_end(s);
3820     }
3821     if (!s->context_initialized) {
3822         if (MPV_common_init(s) < 0)
3823             return -1;
3824
3825         alloc_tables(h);
3826
3827         s->avctx->width = s->width;
3828         s->avctx->height = s->height;
3829         s->avctx->sample_aspect_ratio= h->sps.sar;
3830         if(!s->avctx->sample_aspect_ratio.den)
3831             s->avctx->sample_aspect_ratio.den = 1;
3832
3833         if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3834             s->avctx->frame_rate = h->sps.time_scale;
3835             s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3836         }
3837     }
3838
3839     if(h->slice_num == 0){
3840         frame_start(h);
3841     }
3842
3843     s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3844     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3845
3846     h->mb_aff_frame = 0;
3847     if(h->sps.frame_mbs_only_flag){
3848         s->picture_structure= PICT_FRAME;
3849     }else{
3850         if(get_bits1(&s->gb)) { //field_pic_flag
3851             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3852         } else {
3853             s->picture_structure= PICT_FRAME;
3854             first_mb_in_slice <<= 1;
3855             h->mb_aff_frame = h->sps.mb_aff;
3856         }
3857     }
3858
3859     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3860     s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
3861     
3862     if(s->picture_structure==PICT_FRAME){
3863         h->curr_pic_num=   h->frame_num;
3864         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3865     }else{
3866         h->curr_pic_num= 2*h->frame_num;
3867         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3868     }
3869         
3870     if(h->nal_unit_type == NAL_IDR_SLICE){
3871         get_ue_golomb(&s->gb); /* idr_pic_id */
3872     }
3873    
3874     if(h->sps.poc_type==0){
3875         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3876         
3877         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3878             h->delta_poc_bottom= get_se_golomb(&s->gb);
3879         }
3880     }
3881     
3882     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3883         h->delta_poc[0]= get_se_golomb(&s->gb);
3884         
3885         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3886             h->delta_poc[1]= get_se_golomb(&s->gb);
3887     }
3888     
3889     init_poc(h);
3890     
3891     if(h->pps.redundant_pic_cnt_present){
3892         h->redundant_pic_count= get_ue_golomb(&s->gb);
3893     }
3894
3895     //set defaults, might be overriden a few line later
3896     h->ref_count[0]= h->pps.ref_count[0];
3897     h->ref_count[1]= h->pps.ref_count[1];
3898
3899     if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3900         if(h->slice_type == B_TYPE){
3901             h->direct_spatial_mv_pred= get_bits1(&s->gb);
3902         }
3903         num_ref_idx_active_override_flag= get_bits1(&s->gb);
3904     
3905         if(num_ref_idx_active_override_flag){
3906             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3907             if(h->slice_type==B_TYPE)
3908                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3909
3910             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3911                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3912                 return -1;
3913             }
3914         }
3915     }
3916
3917     if(!default_ref_list_done){
3918         fill_default_ref_list(h);
3919     }
3920
3921     decode_ref_pic_list_reordering(h);
3922
3923     if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3924        || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3925         pred_weight_table(h);
3926     else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3927         implicit_weight_table(h);
3928     else
3929         h->use_weight = 0;
3930     
3931     if(s->current_picture.reference)
3932         decode_ref_pic_marking(h);
3933
3934     if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3935         h->cabac_init_idc = get_ue_golomb(&s->gb);
3936
3937     h->last_qscale_diff = 0;
3938     s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3939     if(s->qscale<0 || s->qscale>51){
3940         av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3941         return -1;
3942     }
3943     h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
3944     //FIXME qscale / qp ... stuff
3945     if(h->slice_type == SP_TYPE){
3946         get_bits1(&s->gb); /* sp_for_switch_flag */
3947     }
3948     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3949         get_se_golomb(&s->gb); /* slice_qs_delta */
3950     }
3951
3952     h->deblocking_filter = 1;
3953     h->slice_alpha_c0_offset = 0;
3954     h->slice_beta_offset = 0;
3955     if( h->pps.deblocking_filter_parameters_present ) {
3956         h->deblocking_filter= get_ue_golomb(&s->gb);
3957         if(h->deblocking_filter < 2) 
3958             h->deblocking_filter^= 1; // 1<->0
3959
3960         if( h->deblocking_filter ) {
3961             h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3962             h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3963         }
3964     }
3965
3966 #if 0 //FMO
3967     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3968         slice_group_change_cycle= get_bits(&s->gb, ?);
3969 #endif
3970
3971     h->slice_num++;
3972
3973     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3974         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n", 
3975                h->slice_num,
3976                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3977                first_mb_in_slice, 
3978                av_get_pict_type_char(h->slice_type),
3979                pps_id, h->frame_num,
3980                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3981                h->ref_count[0], h->ref_count[1],
3982                s->qscale,
3983                h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
3984                h->use_weight,
3985                h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3986                );
3987     }
3988
3989     return 0;
3990 }
3991
3992 /**
3993  *
3994  */
3995 static inline int get_level_prefix(GetBitContext *gb){
3996     unsigned int buf;
3997     int log;
3998     
3999     OPEN_READER(re, gb);
4000     UPDATE_CACHE(re, gb);
4001     buf=GET_CACHE(re, gb);
4002     
4003     log= 32 - av_log2(buf);
4004 #ifdef TRACE
4005     print_bin(buf>>(32-log), log);
4006     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__);
4007 #endif
4008
4009     LAST_SKIP_BITS(re, gb, log);
4010     CLOSE_READER(re, gb);
4011
4012     return log-1;
4013 }
4014
4015 /**
4016  * decodes a residual block.
4017  * @param n block index
4018  * @param scantable scantable
4019  * @param max_coeff number of coefficients in the block
4020  * @return <0 if an error occured
4021  */
4022 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
4023     MpegEncContext * const s = &h->s;
4024     const uint16_t *qmul= dequant_coeff[qp];
4025     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};
4026     int level[16], run[16];
4027     int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
4028
4029     //FIXME put trailing_onex into the context
4030
4031     if(n == CHROMA_DC_BLOCK_INDEX){
4032         coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4033         total_coeff= coeff_token>>2;
4034     }else{    
4035         if(n == LUMA_DC_BLOCK_INDEX){
4036             total_coeff= pred_non_zero_count(h, 0);
4037             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4038             total_coeff= coeff_token>>2;
4039         }else{
4040             total_coeff= pred_non_zero_count(h, n);
4041             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4042             total_coeff= coeff_token>>2;
4043             h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4044         }
4045     }
4046
4047     //FIXME set last_non_zero?
4048
4049     if(total_coeff==0)
4050         return 0;
4051         
4052     trailing_ones= coeff_token&3;
4053     tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
4054     assert(total_coeff<=16);
4055     
4056     for(i=0; i<trailing_ones; i++){
4057         level[i]= 1 - 2*get_bits1(gb);
4058     }
4059
4060     suffix_length= total_coeff > 10 && trailing_ones < 3;
4061
4062     for(; i<total_coeff; i++){
4063         const int prefix= get_level_prefix(gb);
4064         int level_code, mask;
4065
4066         if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4067             if(suffix_length)
4068                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4069             else
4070                 level_code= (prefix<<suffix_length); //part
4071         }else if(prefix==14){
4072             if(suffix_length)
4073                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4074             else
4075                 level_code= prefix + get_bits(gb, 4); //part
4076         }else if(prefix==15){
4077             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
4078             if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
4079         }else{
4080             av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4081             return -1;
4082         }
4083
4084         if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
4085
4086         mask= -(level_code&1);
4087         level[i]= (((2+level_code)>>1) ^ mask) - mask;
4088
4089         if(suffix_length==0) suffix_length=1; //FIXME split first iteration
4090
4091 #if 1
4092         if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4093 #else        
4094         if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4095         /* ? == prefix > 2 or sth */
4096 #endif
4097         tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
4098     }
4099
4100     if(total_coeff == max_coeff)
4101         zeros_left=0;
4102     else{
4103         if(n == CHROMA_DC_BLOCK_INDEX)
4104             zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4105         else
4106             zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4107     }
4108     
4109     for(i=0; i<total_coeff-1; i++){
4110         if(zeros_left <=0)
4111             break;
4112         else if(zeros_left < 7){
4113             run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4114         }else{
4115             run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4116         }
4117         zeros_left -= run[i];
4118     }
4119
4120     if(zeros_left<0){
4121         av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4122         return -1;
4123     }
4124     
4125     for(; i<total_coeff-1; i++){
4126         run[i]= 0;
4127     }
4128
4129     run[i]= zeros_left;
4130
4131     coeff_num=-1;
4132     if(n > 24){
4133         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4134             int j;
4135
4136             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4137             j= scantable[ coeff_num ];
4138
4139             block[j]= level[i];
4140         }
4141     }else{
4142         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
4143             int j;
4144
4145             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4146             j= scantable[ coeff_num ];
4147
4148             block[j]= level[i] * qmul[j];
4149 //            printf("%d %d  ", block[j], qmul[j]);
4150         }
4151     }
4152     return 0;
4153 }
4154
4155 /**
4156  * decodes a P_SKIP or B_SKIP macroblock
4157  */
4158 static void decode_mb_skip(H264Context *h){
4159     MpegEncContext * const s = &h->s;
4160     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4161     int mb_type;
4162     
4163     memset(h->non_zero_count[mb_xy], 0, 16);
4164     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4165
4166     if(h->mb_aff_frame && s->mb_skip_run==0 && (s->mb_y&1)==0){
4167         h->mb_field_decoding_flag= get_bits1(&s->gb);
4168     }
4169     if(h->mb_field_decoding_flag)
4170         mb_type|= MB_TYPE_INTERLACED;
4171         
4172     if( h->slice_type == B_TYPE )
4173     {
4174         // just for fill_caches. pred_direct_motion will set the real mb_type
4175         mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4176
4177         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4178         pred_direct_motion(h, &mb_type);
4179         if(h->pps.cabac){
4180             fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4181             fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4182         }
4183     }
4184     else
4185     {
4186         int mx, my;
4187         mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4188
4189         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4190         pred_pskip_motion(h, &mx, &my);
4191         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4192         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4193         if(h->pps.cabac)
4194             fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4195     }
4196
4197     write_back_motion(h, mb_type);
4198     s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
4199     s->current_picture.qscale_table[mb_xy]= s->qscale;
4200     h->slice_table[ mb_xy ]= h->slice_num;
4201     h->prev_mb_skiped= 1;
4202 }
4203
4204 /**
4205  * decodes a macroblock
4206  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4207  */
4208 static int decode_mb_cavlc(H264Context *h){
4209     MpegEncContext * const s = &h->s;
4210     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4211     int mb_type, partition_count, cbp;
4212
4213     s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
4214
4215     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);