Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 5cf08f23

History | View | Annotate | Download (266 KB)

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_skipped; //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
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
213
    uint32_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 to 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
    assert((stride&(w-1))==0);
374
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
375
    if(w==2 && h==2){
376
        *(uint16_t*)(p + 0)=
377
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
378
    }else if(w==2 && h==4){
379
        *(uint16_t*)(p + 0*stride)=
380
        *(uint16_t*)(p + 1*stride)=
381
        *(uint16_t*)(p + 2*stride)=
382
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
383
    }else if(w==4 && h==1){
384
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
385
    }else if(w==4 && h==2){
386
        *(uint32_t*)(p + 0*stride)=
387
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
388
    }else if(w==4 && h==4){
389
        *(uint32_t*)(p + 0*stride)=
390
        *(uint32_t*)(p + 1*stride)=
391
        *(uint32_t*)(p + 2*stride)=
392
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
393
    }else if(w==8 && h==1){
394
        *(uint32_t*)(p + 0)=
395
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
396
    }else if(w==8 && h==2){
397
        *(uint32_t*)(p + 0 + 0*stride)=
398
        *(uint32_t*)(p + 4 + 0*stride)=
399
        *(uint32_t*)(p + 0 + 1*stride)=
400
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
401
    }else if(w==8 && h==4){
402
        *(uint64_t*)(p + 0*stride)=
403
        *(uint64_t*)(p + 1*stride)=
404
        *(uint64_t*)(p + 2*stride)=
405
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
406
    }else if(w==16 && h==2){
407
        *(uint64_t*)(p + 0+0*stride)=
408
        *(uint64_t*)(p + 8+0*stride)=
409
        *(uint64_t*)(p + 0+1*stride)=
410
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
411
    }else if(w==16 && h==4){
412
        *(uint64_t*)(p + 0+0*stride)=
413
        *(uint64_t*)(p + 8+0*stride)=
414
        *(uint64_t*)(p + 0+1*stride)=
415
        *(uint64_t*)(p + 8+1*stride)=
416
        *(uint64_t*)(p + 0+2*stride)=
417
        *(uint64_t*)(p + 8+2*stride)=
418
        *(uint64_t*)(p + 0+3*stride)=
419
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
420
    }else
421
        assert(0);
422
}
423

    
424
static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
425
    MpegEncContext * const s = &h->s;
426
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
427
    int topleft_xy, top_xy, topright_xy, left_xy[2];
428
    int topleft_type, top_type, topright_type, left_type[2];
429
    int left_block[8];
430
    int i;
431

    
432
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
433
    // the actual condition is whether we're on the edge of a slice,
434
    // and even then the intra and nnz parts are unnecessary.
435
    if(for_deblock && h->slice_num == 1)
436
        return;
437

    
438
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it 
439
    
440
    top_xy     = mb_xy  - s->mb_stride;
441
    topleft_xy = top_xy - 1;
442
    topright_xy= top_xy + 1;
443
    left_xy[1] = left_xy[0] = mb_xy-1;
444
    left_block[0]= 0;
445
    left_block[1]= 1;
446
    left_block[2]= 2;
447
    left_block[3]= 3;
448
    left_block[4]= 7;
449
    left_block[5]= 10;
450
    left_block[6]= 8;
451
    left_block[7]= 11;
452
    if(h->mb_aff_frame){
453
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
454
        const int top_pair_xy      = pair_xy     - s->mb_stride;
455
        const int topleft_pair_xy  = top_pair_xy - 1;
456
        const int topright_pair_xy = top_pair_xy + 1;
457
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
458
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
459
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
460
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
461
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
462
        const int bottom = (s->mb_y & 1);
463
        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);
464
        if (bottom
465
                ? !curr_mb_frame_flag // bottom macroblock
466
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
467
                ) {
468
            top_xy -= s->mb_stride;
469
        }
470
        if (bottom
471
                ? !curr_mb_frame_flag // bottom macroblock
472
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
473
                ) {
474
            topleft_xy -= s->mb_stride;
475
        }
476
        if (bottom
477
                ? !curr_mb_frame_flag // bottom macroblock
478
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
479
                ) {
480
            topright_xy -= s->mb_stride;
481
        }
482
        if (left_mb_frame_flag != curr_mb_frame_flag) {
483
            left_xy[1] = left_xy[0] = pair_xy - 1;
484
            if (curr_mb_frame_flag) {
485
                if (bottom) {
486
                    left_block[0]= 2;
487
                    left_block[1]= 2;
488
                    left_block[2]= 3;
489
                    left_block[3]= 3;
490
                    left_block[4]= 8;
491
                    left_block[5]= 11;
492
                    left_block[6]= 8;
493
                    left_block[7]= 11;
494
                } else {
495
                    left_block[0]= 0;
496
                    left_block[1]= 0;
497
                    left_block[2]= 1;
498
                    left_block[3]= 1;
499
                    left_block[4]= 7;
500
                    left_block[5]= 10;
501
                    left_block[6]= 7;
502
                    left_block[7]= 10;
503
                }
504
            } else {
505
                left_xy[1] += s->mb_stride;
506
                //left_block[0]= 0;
507
                left_block[1]= 2;
508
                left_block[2]= 0;
509
                left_block[3]= 2;
510
                //left_block[4]= 7;
511
                left_block[5]= 10;
512
                left_block[6]= 7;
513
                left_block[7]= 10;
514
            }
515
        }
516
    }
517

    
518
    h->top_mb_xy = top_xy;
519
    h->left_mb_xy[0] = left_xy[0];
520
    h->left_mb_xy[1] = left_xy[1];
521
    if(for_deblock){
522
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
523
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
524
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
525
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
526
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
527
    }else{
528
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
529
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
530
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
531
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
532
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
533
    }
534

    
535
    if(IS_INTRA(mb_type)){
536
        h->topleft_samples_available= 
537
        h->top_samples_available= 
538
        h->left_samples_available= 0xFFFF;
539
        h->topright_samples_available= 0xEEEA;
540

    
541
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
542
            h->topleft_samples_available= 0xB3FF;
543
            h->top_samples_available= 0x33FF;
544
            h->topright_samples_available= 0x26EA;
545
        }
546
        for(i=0; i<2; i++){
547
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
548
                h->topleft_samples_available&= 0xDF5F;
549
                h->left_samples_available&= 0x5F5F;
550
            }
551
        }
552
        
553
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
554
            h->topleft_samples_available&= 0x7FFF;
555
        
556
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
557
            h->topright_samples_available&= 0xFBFF;
558
    
559
        if(IS_INTRA4x4(mb_type)){
560
            if(IS_INTRA4x4(top_type)){
561
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
562
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
563
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
564
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
565
            }else{
566
                int pred;
567
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
568
                    pred= -1;
569
                else{
570
                    pred= 2;
571
                }
572
                h->intra4x4_pred_mode_cache[4+8*0]=
573
                h->intra4x4_pred_mode_cache[5+8*0]=
574
                h->intra4x4_pred_mode_cache[6+8*0]=
575
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
576
            }
577
            for(i=0; i<2; i++){
578
                if(IS_INTRA4x4(left_type[i])){
579
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
580
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
581
                }else{
582
                    int pred;
583
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
584
                        pred= -1;
585
                    else{
586
                        pred= 2;
587
                    }
588
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
589
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
590
                }
591
            }
592
        }
593
    }
594
    
595
    
596
/*
597
0 . T T. T T T T 
598
1 L . .L . . . . 
599
2 L . .L . . . . 
600
3 . T TL . . . . 
601
4 L . .L . . . . 
602
5 L . .. . . . . 
603
*/
604
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
605
    if(top_type){
606
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
607
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
608
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
609
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
610
    
611
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
612
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
613
    
614
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
615
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
616
        
617
    }else{
618
        h->non_zero_count_cache[4+8*0]=      
619
        h->non_zero_count_cache[5+8*0]=
620
        h->non_zero_count_cache[6+8*0]=
621
        h->non_zero_count_cache[7+8*0]=
622
    
623
        h->non_zero_count_cache[1+8*0]=
624
        h->non_zero_count_cache[2+8*0]=
625
    
626
        h->non_zero_count_cache[1+8*3]=
627
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
628
        
629
    }
630

    
631
    for (i=0; i<2; i++) {
632
        if(left_type[i]){
633
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
634
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
635
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
636
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
637
        }else{
638
            h->non_zero_count_cache[3+8*1 + 2*8*i]= 
639
            h->non_zero_count_cache[3+8*2 + 2*8*i]= 
640
            h->non_zero_count_cache[0+8*1 +   8*i]= 
641
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
642
        }
643
    }
644

    
645
    if( h->pps.cabac ) {
646
        // top_cbp
647
        if(top_type) {
648
            h->top_cbp = h->cbp_table[top_xy];
649
        } else if(IS_INTRA(mb_type)) {
650
            h->top_cbp = 0x1C0;
651
        } else {
652
            h->top_cbp = 0;
653
        }
654
        // left_cbp
655
        if (left_type[0]) {
656
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
657
        } else if(IS_INTRA(mb_type)) {
658
            h->left_cbp = 0x1C0;
659
        } else {
660
            h->left_cbp = 0;
661
        }
662
        if (left_type[0]) {
663
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
664
        }
665
        if (left_type[1]) {
666
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
667
        }
668
    }
669

    
670
#if 1
671
    //FIXME direct mb can skip much of this
672
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
673
        int list;
674
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
675
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
676
                /*if(!h->mv_cache_clean[list]){
677
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
678
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
679
                    h->mv_cache_clean[list]= 1;
680
                }*/
681
                continue;
682
            }
683
            h->mv_cache_clean[list]= 0;
684
            
685
            if(IS_INTER(top_type)){
686
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
687
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
688
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
689
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
690
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
691
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
692
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
693
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
694
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
695
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
696
            }else{
697
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
698
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
699
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
700
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
701
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
702
            }
703

    
704
            //FIXME unify cleanup or sth
705
            if(IS_INTER(left_type[0])){
706
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
707
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
708
                *(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]];
709
                *(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]];
710
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
711
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
712
            }else{
713
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
714
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
715
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
716
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
717
            }
718
            
719
            if(IS_INTER(left_type[1])){
720
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
721
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
722
                *(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]];
723
                *(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]];
724
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
725
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
726
            }else{
727
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
728
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
729
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
730
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
731
                assert((!left_type[0]) == (!left_type[1]));
732
            }
733

    
734
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
735
                continue;
736

    
737
            if(IS_INTER(topleft_type)){
738
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
739
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
740
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
741
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
742
            }else{
743
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
744
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
745
            }
746
            
747
            if(IS_INTER(topright_type)){
748
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
749
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
750
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
751
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
752
            }else{
753
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
754
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
755
            }
756
            
757

    
758
            h->ref_cache[list][scan8[5 ]+1] = 
759
            h->ref_cache[list][scan8[7 ]+1] = 
760
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
761
            h->ref_cache[list][scan8[4 ]] = 
762
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
763
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
764
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
765
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
766
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
767
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
768

    
769
            if( h->pps.cabac ) {
770
                /* XXX beurk, Load mvd */
771
                if(IS_INTER(topleft_type)){
772
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
773
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
774
                }else{
775
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
776
                }
777

    
778
                if(IS_INTER(top_type)){
779
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
780
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
781
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
782
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
783
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
784
                }else{
785
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
786
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
787
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
788
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
789
                }
790
                if(IS_INTER(left_type[0])){
791
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
792
                    *(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]];
793
                    *(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]];
794
                }else{
795
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
796
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
797
                }
798
                if(IS_INTER(left_type[1])){
799
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
800
                    *(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]];
801
                    *(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]];
802
                }else{
803
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
804
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
805
                }
806
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
807
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
808
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
809
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
810
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
811

    
812
                if(h->slice_type == B_TYPE){
813
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
814

    
815
                    if(IS_DIRECT(top_type)){
816
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
817
                    }else if(IS_8X8(top_type)){
818
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
819
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
820
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
821
                    }else{
822
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
823
                    }
824
                    
825
                    //FIXME interlacing
826
                    if(IS_DIRECT(left_type[0])){
827
                        h->direct_cache[scan8[0] - 1 + 0*8]=
828
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
829
                    }else if(IS_8X8(left_type[0])){
830
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
831
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
832
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
833
                    }else{
834
                        h->direct_cache[scan8[0] - 1 + 0*8]=
835
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
836
                    }
837
                }
838
            }
839
        }
840
    }
841
#endif
842
}
843

    
844
static inline void write_back_intra_pred_mode(H264Context *h){
845
    MpegEncContext * const s = &h->s;
846
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
847

    
848
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
849
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
850
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
851
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
852
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
853
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
854
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
855
}
856

    
857
/**
858
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
859
 */
860
static inline int check_intra4x4_pred_mode(H264Context *h){
861
    MpegEncContext * const s = &h->s;
862
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
863
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
864
    int i;
865
    
866
    if(!(h->top_samples_available&0x8000)){
867
        for(i=0; i<4; i++){
868
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
869
            if(status<0){
870
                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);
871
                return -1;
872
            } else if(status){
873
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
874
            }
875
        }
876
    }
877
    
878
    if(!(h->left_samples_available&0x8000)){
879
        for(i=0; i<4; i++){
880
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
881
            if(status<0){
882
                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);
883
                return -1;
884
            } else if(status){
885
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
886
            }
887
        }
888
    }
889

    
890
    return 0;
891
} //FIXME cleanup like next
892

    
893
/**
894
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
895
 */
896
static inline int check_intra_pred_mode(H264Context *h, int mode){
897
    MpegEncContext * const s = &h->s;
898
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
899
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
900
    
901
    if(mode < 0 || mode > 6) {
902
        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);
903
        return -1;
904
    }
905
    
906
    if(!(h->top_samples_available&0x8000)){
907
        mode= top[ mode ];
908
        if(mode<0){
909
            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);
910
            return -1;
911
        }
912
    }
913
    
914
    if(!(h->left_samples_available&0x8000)){
915
        mode= left[ mode ];
916
        if(mode<0){
917
            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);
918
            return -1;
919
        } 
920
    }
921

    
922
    return mode;
923
}
924

    
925
/**
926
 * gets the predicted intra4x4 prediction mode.
927
 */
928
static inline int pred_intra_mode(H264Context *h, int n){
929
    const int index8= scan8[n];
930
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
931
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
932
    const int min= FFMIN(left, top);
933

    
934
    tprintf("mode:%d %d min:%d\n", left ,top, min);
935

    
936
    if(min<0) return DC_PRED;
937
    else      return min;
938
}
939

    
940
static inline void write_back_non_zero_count(H264Context *h){
941
    MpegEncContext * const s = &h->s;
942
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
943

    
944
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
945
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
946
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
947
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
948
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
949
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
950
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
951
    
952
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
953
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
954
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
955

    
956
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
957
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
958
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
959
}
960

    
961
/**
962
 * gets the predicted number of non zero coefficients.
963
 * @param n block index
964
 */
965
static inline int pred_non_zero_count(H264Context *h, int n){
966
    const int index8= scan8[n];
967
    const int left= h->non_zero_count_cache[index8 - 1];
968
    const int top = h->non_zero_count_cache[index8 - 8];
969
    int i= left + top;
970
    
971
    if(i<64) i= (i+1)>>1;
972

    
973
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
974

    
975
    return i&31;
976
}
977

    
978
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
979
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
980

    
981
    if(topright_ref != PART_NOT_AVAILABLE){
982
        *C= h->mv_cache[list][ i - 8 + part_width ];
983
        return topright_ref;
984
    }else{
985
        tprintf("topright MV not available\n");
986

    
987
        *C= h->mv_cache[list][ i - 8 - 1 ];
988
        return h->ref_cache[list][ i - 8 - 1 ];
989
    }
990
}
991

    
992
/**
993
 * gets the predicted MV.
994
 * @param n the block index
995
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
996
 * @param mx the x component of the predicted motion vector
997
 * @param my the y component of the predicted motion vector
998
 */
999
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1000
    const int index8= scan8[n];
1001
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1002
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1003
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1004
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1005
    const int16_t * C;
1006
    int diagonal_ref, match_count;
1007

    
1008
    assert(part_width==1 || part_width==2 || part_width==4);
1009

    
1010
/* mv_cache
1011
  B . . A T T T T 
1012
  U . . L . . , .
1013
  U . . L . . . .
1014
  U . . L . . , .
1015
  . . . L . . . .
1016
*/
1017

    
1018
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1019
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1020
    tprintf("pred_motion match_count=%d\n", match_count);
1021
    if(match_count > 1){ //most common
1022
        *mx= mid_pred(A[0], B[0], C[0]);
1023
        *my= mid_pred(A[1], B[1], C[1]);
1024
    }else if(match_count==1){
1025
        if(left_ref==ref){
1026
            *mx= A[0];
1027
            *my= A[1];        
1028
        }else if(top_ref==ref){
1029
            *mx= B[0];
1030
            *my= B[1];        
1031
        }else{
1032
            *mx= C[0];
1033
            *my= C[1];        
1034
        }
1035
    }else{
1036
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1037
            *mx= A[0];
1038
            *my= A[1];        
1039
        }else{
1040
            *mx= mid_pred(A[0], B[0], C[0]);
1041
            *my= mid_pred(A[1], B[1], C[1]);
1042
        }
1043
    }
1044
        
1045
    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);
1046
}
1047

    
1048
/**
1049
 * gets the directionally predicted 16x8 MV.
1050
 * @param n the block index
1051
 * @param mx the x component of the predicted motion vector
1052
 * @param my the y component of the predicted motion vector
1053
 */
1054
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1055
    if(n==0){
1056
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1057
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1058

    
1059
        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);
1060
        
1061
        if(top_ref == ref){
1062
            *mx= B[0];
1063
            *my= B[1];
1064
            return;
1065
        }
1066
    }else{
1067
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1068
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1069
        
1070
        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);
1071

    
1072
        if(left_ref == ref){
1073
            *mx= A[0];
1074
            *my= A[1];
1075
            return;
1076
        }
1077
    }
1078

    
1079
    //RARE
1080
    pred_motion(h, n, 4, list, ref, mx, my);
1081
}
1082

    
1083
/**
1084
 * gets the directionally predicted 8x16 MV.
1085
 * @param n the block index
1086
 * @param mx the x component of the predicted motion vector
1087
 * @param my the y component of the predicted motion vector
1088
 */
1089
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1090
    if(n==0){
1091
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1092
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1093
        
1094
        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);
1095

    
1096
        if(left_ref == ref){
1097
            *mx= A[0];
1098
            *my= A[1];
1099
            return;
1100
        }
1101
    }else{
1102
        const int16_t * C;
1103
        int diagonal_ref;
1104

    
1105
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1106
        
1107
        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);
1108

    
1109
        if(diagonal_ref == ref){ 
1110
            *mx= C[0];
1111
            *my= C[1];
1112
            return;
1113
        }
1114
    }
1115

    
1116
    //RARE
1117
    pred_motion(h, n, 2, list, ref, mx, my);
1118
}
1119

    
1120
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1121
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1122
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1123

    
1124
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1125

    
1126
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1127
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1128
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1129
       
1130
        *mx = *my = 0;
1131
        return;
1132
    }
1133
        
1134
    pred_motion(h, 0, 4, 0, 0, mx, my);
1135

    
1136
    return;
1137
}
1138

    
1139
static inline void direct_dist_scale_factor(H264Context * const h){
1140
    const int poc = h->s.current_picture_ptr->poc;
1141
    const int poc1 = h->ref_list[1][0].poc;
1142
    int i;
1143
    for(i=0; i<h->ref_count[0]; i++){
1144
        int poc0 = h->ref_list[0][i].poc;
1145
        int td = clip(poc1 - poc0, -128, 127);
1146
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1147
            h->dist_scale_factor[i] = 256;
1148
        }else{
1149
            int tb = clip(poc - poc0, -128, 127);
1150
            int tx = (16384 + (ABS(td) >> 1)) / td;
1151
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1152
        }
1153
    }
1154
}
1155
static inline void direct_ref_list_init(H264Context * const h){
1156
    MpegEncContext * const s = &h->s;
1157
    Picture * const ref1 = &h->ref_list[1][0];
1158
    Picture * const cur = s->current_picture_ptr;
1159
    int list, i, j;
1160
    if(cur->pict_type == I_TYPE)
1161
        cur->ref_count[0] = 0;
1162
    if(cur->pict_type != B_TYPE)
1163
        cur->ref_count[1] = 0;
1164
    for(list=0; list<2; list++){
1165
        cur->ref_count[list] = h->ref_count[list];
1166
        for(j=0; j<h->ref_count[list]; j++)
1167
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1168
    }
1169
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1170
        return;
1171
    for(list=0; list<2; list++){
1172
        for(i=0; i<ref1->ref_count[list]; i++){
1173
            const int poc = ref1->ref_poc[list][i];
1174
            h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1175
            for(j=0; j<h->ref_count[list]; j++)
1176
                if(h->ref_list[list][j].poc == poc){
1177
                    h->map_col_to_list0[list][i] = j;
1178
                    break;
1179
                }
1180
        }
1181
    }
1182
}
1183

    
1184
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1185
    MpegEncContext * const s = &h->s;
1186
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1187
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1188
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1189
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1190
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1191
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1192
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1193
    const int is_b8x8 = IS_8X8(*mb_type);
1194
    int sub_mb_type;
1195
    int i8, i4;
1196

    
1197
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1198
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1199
         * so we know exactly what block size to use */
1200
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1201
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1202
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1203
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1204
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1205
    }else{
1206
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1207
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1208
    }
1209
    if(!is_b8x8)
1210
        *mb_type |= MB_TYPE_DIRECT2;
1211

    
1212
    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);
1213
    
1214
    if(h->direct_spatial_mv_pred){
1215
        int ref[2];
1216
        int mv[2][2];
1217
        int list;
1218

    
1219
        /* ref = min(neighbors) */
1220
        for(list=0; list<2; list++){
1221
            int refa = h->ref_cache[list][scan8[0] - 1];
1222
            int refb = h->ref_cache[list][scan8[0] - 8];
1223
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1224
            if(refc == -2)
1225
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1226
            ref[list] = refa;
1227
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1228
                ref[list] = refb;
1229
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1230
                ref[list] = refc;
1231
            if(ref[list] < 0)
1232
                ref[list] = -1;
1233
        }
1234

    
1235
        if(ref[0] < 0 && ref[1] < 0){
1236
            ref[0] = ref[1] = 0;
1237
            mv[0][0] = mv[0][1] =
1238
            mv[1][0] = mv[1][1] = 0;
1239
        }else{
1240
            for(list=0; list<2; list++){
1241
                if(ref[list] >= 0)
1242
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1243
                else
1244
                    mv[list][0] = mv[list][1] = 0;
1245
            }
1246
        }
1247

    
1248
        if(ref[1] < 0){
1249
            *mb_type &= ~MB_TYPE_P0L1;
1250
            sub_mb_type &= ~MB_TYPE_P0L1;
1251
        }else if(ref[0] < 0){
1252
            *mb_type &= ~MB_TYPE_P0L0;
1253
            sub_mb_type &= ~MB_TYPE_P0L0;
1254
        }
1255

    
1256
        if(IS_16X16(*mb_type)){
1257
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1258
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1259
            if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1260
                ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1261
                if(ref[0] > 0)
1262
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1263
                else
1264
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1265
                if(ref[1] > 0)
1266
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1267
                else
1268
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1269
            }else{
1270
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1271
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1272
            }
1273
        }else{
1274
            for(i8=0; i8<4; i8++){
1275
                const int x8 = i8&1;
1276
                const int y8 = i8>>1;
1277
    
1278
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1279
                    continue;
1280
                h->sub_mb_type[i8] = sub_mb_type;
1281
    
1282
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1283
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1284
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1285
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1286
    
1287
                /* col_zero_flag */
1288
                if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1289
                    for(i4=0; i4<4; i4++){
1290
                        const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1291
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1292
                            if(ref[0] == 0)
1293
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1294
                            if(ref[1] == 0)
1295
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1296
                        }
1297
                    }
1298
                }
1299
            }
1300
        }
1301
    }else{ /* direct temporal mv pred */
1302
        if(IS_16X16(*mb_type)){
1303
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1304
            if(IS_INTRA(mb_type_col)){
1305
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1306
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1307
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1308
            }else{
1309
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1310
                                                : h->map_col_to_list0[1][l1ref1[0]];
1311
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1312
                const int16_t *mv_col = l1mv0[0];
1313
                int mv_l0[2];
1314
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1315
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1316
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1317
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1318
                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);
1319
            }
1320
        }else{
1321
            for(i8=0; i8<4; i8++){
1322
                const int x8 = i8&1;
1323
                const int y8 = i8>>1;
1324
                int ref0, dist_scale_factor;
1325
    
1326
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1327
                    continue;
1328
                h->sub_mb_type[i8] = sub_mb_type;
1329
                if(IS_INTRA(mb_type_col)){
1330
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1331
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1332
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1333
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1334
                    continue;
1335
                }
1336
    
1337
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1338
                if(ref0 >= 0)
1339
                    ref0 = h->map_col_to_list0[0][ref0];
1340
                else
1341
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1342
                dist_scale_factor = h->dist_scale_factor[ref0];
1343
    
1344
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1345
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1346
                for(i4=0; i4<4; i4++){
1347
                    const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1348
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1349
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1350
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1351
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1352
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1353
                }
1354
            }
1355
        }
1356
    }
1357
}
1358

    
1359
static inline void write_back_motion(H264Context *h, int mb_type){
1360
    MpegEncContext * const s = &h->s;
1361
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1362
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1363
    int list;
1364

    
1365
    for(list=0; list<2; list++){
1366
        int y;
1367
        if(!USES_LIST(mb_type, list)){
1368
            if(1){ //FIXME skip or never read if mb_type doesn't use it
1369
                for(y=0; y<4; y++){
1370
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1371
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1372
                }
1373
                if( h->pps.cabac ) {
1374
                    /* FIXME needed ? */
1375
                    for(y=0; y<4; y++){
1376
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1377
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1378
                    }
1379
                }
1380
                for(y=0; y<2; y++){
1381
                    s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1382
                    s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1383
                }
1384
            }
1385
            continue;
1386
        }
1387
        
1388
        for(y=0; y<4; y++){
1389
            *(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];
1390
            *(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];
1391
        }
1392
        if( h->pps.cabac ) {
1393
            for(y=0; y<4; y++){
1394
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1395
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1396
            }
1397
        }
1398
        for(y=0; y<2; y++){
1399
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1400
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1401
        }
1402
    }
1403
    
1404
    if(h->slice_type == B_TYPE && h->pps.cabac){
1405
        if(IS_8X8(mb_type)){
1406
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1407
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1408
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1409
        }
1410
    }
1411
}
1412

    
1413
/**
1414
 * Decodes a network abstraction layer unit.
1415
 * @param consumed is the number of bytes used as input
1416
 * @param length is the length of the array
1417
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1418
 * @returns decoded bytes, might be src+1 if no escapes 
1419
 */
1420
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1421
    int i, si, di;
1422
    uint8_t *dst;
1423

    
1424
//    src[0]&0x80;                //forbidden bit
1425
    h->nal_ref_idc= src[0]>>5;
1426
    h->nal_unit_type= src[0]&0x1F;
1427

    
1428
    src++; length--;
1429
#if 0    
1430
    for(i=0; i<length; i++)
1431
        printf("%2X ", src[i]);
1432
#endif
1433
    for(i=0; i+1<length; i+=2){
1434
        if(src[i]) continue;
1435
        if(i>0 && src[i-1]==0) i--;
1436
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1437
            if(src[i+2]!=3){
1438
                /* startcode, so we must be past the end */
1439
                length=i;
1440
            }
1441
            break;
1442
        }
1443
    }
1444

    
1445
    if(i>=length-1){ //no escaped 0
1446
        *dst_length= length;
1447
        *consumed= length+1; //+1 for the header
1448
        return src; 
1449
    }
1450

    
1451
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1452
    dst= h->rbsp_buffer;
1453

    
1454
//printf("decoding esc\n");
1455
    si=di=0;
1456
    while(si<length){ 
1457
        //remove escapes (very rare 1:2^22)
1458
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1459
            if(src[si+2]==3){ //escape
1460
                dst[di++]= 0;
1461
                dst[di++]= 0;
1462
                si+=3;
1463
                continue;
1464
            }else //next start code
1465
                break;
1466
        }
1467

    
1468
        dst[di++]= src[si++];
1469
    }
1470

    
1471
    *dst_length= di;
1472
    *consumed= si + 1;//+1 for the header
1473
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1474
    return dst;
1475
}
1476

    
1477
#if 0
1478
/**
1479
 * @param src the data which should be escaped
1480
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1481
 * @param length the length of the src data
1482
 * @param dst_length the length of the dst array
1483
 * @returns length of escaped data in bytes or -1 if an error occured
1484
 */
1485
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1486
    int i, escape_count, si, di;
1487
    uint8_t *temp;
1488
    
1489
    assert(length>=0);
1490
    assert(dst_length>0);
1491
    
1492
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1493

1494
    if(length==0) return 1;
1495

1496
    escape_count= 0;
1497
    for(i=0; i<length; i+=2){
1498
        if(src[i]) continue;
1499
        if(i>0 && src[i-1]==0) 
1500
            i--;
1501
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1502
            escape_count++;
1503
            i+=2;
1504
        }
1505
    }
1506
    
1507
    if(escape_count==0){ 
1508
        if(dst+1 != src)
1509
            memcpy(dst+1, src, length);
1510
        return length + 1;
1511
    }
1512
    
1513
    if(length + escape_count + 1> dst_length)
1514
        return -1;
1515

1516
    //this should be damn rare (hopefully)
1517

1518
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1519
    temp= h->rbsp_buffer;
1520
//printf("encoding esc\n");
1521
    
1522
    si= 0;
1523
    di= 0;
1524
    while(si < length){
1525
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1526
            temp[di++]= 0; si++;
1527
            temp[di++]= 0; si++;
1528
            temp[di++]= 3; 
1529
            temp[di++]= src[si++];
1530
        }
1531
        else
1532
            temp[di++]= src[si++];
1533
    }
1534
    memcpy(dst+1, temp, length+escape_count);
1535
    
1536
    assert(di == length+escape_count);
1537
    
1538
    return di + 1;
1539
}
1540

1541
/**
1542
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1543
 */
1544
static void encode_rbsp_trailing(PutBitContext *pb){
1545
    int length;
1546
    put_bits(pb, 1, 1);
1547
    length= (-put_bits_count(pb))&7;
1548
    if(length) put_bits(pb, length, 0);
1549
}
1550
#endif
1551

    
1552
/**
1553
 * identifies the exact end of the bitstream
1554
 * @return the length of the trailing, or 0 if damaged
1555
 */
1556
static int decode_rbsp_trailing(uint8_t *src){
1557
    int v= *src;
1558
    int r;
1559

    
1560
    tprintf("rbsp trailing %X\n", v);
1561

    
1562
    for(r=1; r<9; r++){
1563
        if(v&1) return r;
1564
        v>>=1;
1565
    }
1566
    return 0;
1567
}
1568

    
1569
/**
1570
 * idct tranforms the 16 dc values and dequantize them.
1571
 * @param qp quantization parameter
1572
 */
1573
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1574
    const int qmul= dequant_coeff[qp][0];
1575
#define stride 16
1576
    int i;
1577
    int temp[16]; //FIXME check if this is a good idea
1578
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1579
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1580

    
1581
//memset(block, 64, 2*256);
1582
//return;
1583
    for(i=0; i<4; i++){
1584
        const int offset= y_offset[i];
1585
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1586
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1587
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1588
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1589

    
1590
        temp[4*i+0]= z0+z3;
1591
        temp[4*i+1]= z1+z2;
1592
        temp[4*i+2]= z1-z2;
1593
        temp[4*i+3]= z0-z3;
1594
    }
1595

    
1596
    for(i=0; i<4; i++){
1597
        const int offset= x_offset[i];
1598
        const int z0= temp[4*0+i] + temp[4*2+i];
1599
        const int z1= temp[4*0+i] - temp[4*2+i];
1600
        const int z2= temp[4*1+i] - temp[4*3+i];
1601
        const int z3= temp[4*1+i] + temp[4*3+i];
1602

    
1603
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1604
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1605
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1606
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1607
    }
1608
}
1609

    
1610
#if 0
1611
/**
1612
 * dct tranforms the 16 dc values.
1613
 * @param qp quantization parameter ??? FIXME
1614
 */
1615
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1616
//    const int qmul= dequant_coeff[qp][0];
1617
    int i;
1618
    int temp[16]; //FIXME check if this is a good idea
1619
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1620
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1621

1622
    for(i=0; i<4; i++){
1623
        const int offset= y_offset[i];
1624
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1625
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1626
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1627
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1628

1629
        temp[4*i+0]= z0+z3;
1630
        temp[4*i+1]= z1+z2;
1631
        temp[4*i+2]= z1-z2;
1632
        temp[4*i+3]= z0-z3;
1633
    }
1634

1635
    for(i=0; i<4; i++){
1636
        const int offset= x_offset[i];
1637
        const int z0= temp[4*0+i] + temp[4*2+i];
1638
        const int z1= temp[4*0+i] - temp[4*2+i];
1639
        const int z2= temp[4*1+i] - temp[4*3+i];
1640
        const int z3= temp[4*1+i] + temp[4*3+i];
1641

1642
        block[stride*0 +offset]= (z0 + z3)>>1;
1643
        block[stride*2 +offset]= (z1 + z2)>>1;
1644
        block[stride*8 +offset]= (z1 - z2)>>1;
1645
        block[stride*10+offset]= (z0 - z3)>>1;
1646
    }
1647
}
1648
#endif
1649

    
1650
#undef xStride
1651
#undef stride
1652

    
1653
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1654
    const int qmul= dequant_coeff[qp][0];
1655
    const int stride= 16*2;
1656
    const int xStride= 16;
1657
    int a,b,c,d,e;
1658

    
1659
    a= block[stride*0 + xStride*0];
1660
    b= block[stride*0 + xStride*1];
1661
    c= block[stride*1 + xStride*0];
1662
    d= block[stride*1 + xStride*1];
1663

    
1664
    e= a-b;
1665
    a= a+b;
1666
    b= c-d;
1667
    c= c+d;
1668

    
1669
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1670
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1671
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1672
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1673
}
1674

    
1675
#if 0
1676
static void chroma_dc_dct_c(DCTELEM *block){
1677
    const int stride= 16*2;
1678
    const int xStride= 16;
1679
    int a,b,c,d,e;
1680

1681
    a= block[stride*0 + xStride*0];
1682
    b= block[stride*0 + xStride*1];
1683
    c= block[stride*1 + xStride*0];
1684
    d= block[stride*1 + xStride*1];
1685

1686
    e= a-b;
1687
    a= a+b;
1688
    b= c-d;
1689
    c= c+d;
1690

1691
    block[stride*0 + xStride*0]= (a+c);
1692
    block[stride*0 + xStride*1]= (e+b);
1693
    block[stride*1 + xStride*0]= (a-c);
1694
    block[stride*1 + xStride*1]= (e-b);
1695
}
1696
#endif
1697

    
1698
/**
1699
 * gets the chroma qp.
1700
 */
1701
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1702
    
1703
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1704
}
1705

    
1706

    
1707
#if 0
1708
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1709
    int i;
1710
    //FIXME try int temp instead of block
1711
    
1712
    for(i=0; i<4; i++){
1713
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1714
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1715
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1716
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1717
        const int z0= d0 + d3;
1718
        const int z3= d0 - d3;
1719
        const int z1= d1 + d2;
1720
        const int z2= d1 - d2;
1721
        
1722
        block[0 + 4*i]=   z0 +   z1;
1723
        block[1 + 4*i]= 2*z3 +   z2;
1724
        block[2 + 4*i]=   z0 -   z1;
1725
        block[3 + 4*i]=   z3 - 2*z2;
1726
    }    
1727

1728
    for(i=0; i<4; i++){
1729
        const int z0= block[0*4 + i] + block[3*4 + i];
1730
        const int z3= block[0*4 + i] - block[3*4 + i];
1731
        const int z1= block[1*4 + i] + block[2*4 + i];
1732
        const int z2= block[1*4 + i] - block[2*4 + i];
1733
        
1734
        block[0*4 + i]=   z0 +   z1;
1735
        block[1*4 + i]= 2*z3 +   z2;
1736
        block[2*4 + i]=   z0 -   z1;
1737
        block[3*4 + i]=   z3 - 2*z2;
1738
    }
1739
}
1740
#endif
1741

    
1742
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1743
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1744
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1745
    int i;
1746
    const int * const quant_table= quant_coeff[qscale];
1747
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1748
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1749
    const unsigned int threshold2= (threshold1<<1);
1750
    int last_non_zero;
1751

    
1752
    if(seperate_dc){
1753
        if(qscale<=18){
1754
            //avoid overflows
1755
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1756
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1757
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1758

    
1759
            int level= block[0]*quant_coeff[qscale+18][0];
1760
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1761
                if(level>0){
1762
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1763
                    block[0]= level;
1764
                }else{
1765
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1766
                    block[0]= -level;
1767
                }
1768
//                last_non_zero = i;
1769
            }else{
1770
                block[0]=0;
1771
            }
1772
        }else{
1773
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1774
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1775
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1776

    
1777
            int level= block[0]*quant_table[0];
1778
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1779
                if(level>0){
1780
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1781
                    block[0]= level;
1782
                }else{
1783
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1784
                    block[0]= -level;
1785
                }
1786
//                last_non_zero = i;
1787
            }else{
1788
                block[0]=0;
1789
            }
1790
        }
1791
        last_non_zero= 0;
1792
        i=1;
1793
    }else{
1794
        last_non_zero= -1;
1795
        i=0;
1796
    }
1797

    
1798
    for(; i<16; i++){
1799
        const int j= scantable[i];
1800
        int level= block[j]*quant_table[j];
1801

    
1802
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1803
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1804
        if(((unsigned)(level+threshold1))>threshold2){
1805
            if(level>0){
1806
                level= (bias + level)>>QUANT_SHIFT;
1807
                block[j]= level;
1808
            }else{
1809
                level= (bias - level)>>QUANT_SHIFT;
1810
                block[j]= -level;
1811
            }
1812
            last_non_zero = i;
1813
        }else{
1814
            block[j]=0;
1815
        }
1816
    }
1817

    
1818
    return last_non_zero;
1819
}
1820

    
1821
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1822
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1823
    ((uint32_t*)(src+0*stride))[0]= a;
1824
    ((uint32_t*)(src+1*stride))[0]= a;
1825
    ((uint32_t*)(src+2*stride))[0]= a;
1826
    ((uint32_t*)(src+3*stride))[0]= a;
1827
}
1828

    
1829
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1830
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1831
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1832
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1833
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1834
}
1835

    
1836
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1837
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1838
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1839
    
1840
    ((uint32_t*)(src+0*stride))[0]= 
1841
    ((uint32_t*)(src+1*stride))[0]= 
1842
    ((uint32_t*)(src+2*stride))[0]= 
1843
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1844
}
1845

    
1846
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1847
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1848
    
1849
    ((uint32_t*)(src+0*stride))[0]= 
1850
    ((uint32_t*)(src+1*stride))[0]= 
1851
    ((uint32_t*)(src+2*stride))[0]= 
1852
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1853
}
1854

    
1855
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1856
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1857
    
1858
    ((uint32_t*)(src+0*stride))[0]= 
1859
    ((uint32_t*)(src+1*stride))[0]= 
1860
    ((uint32_t*)(src+2*stride))[0]= 
1861
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1862
}
1863

    
1864
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1865
    ((uint32_t*)(src+0*stride))[0]= 
1866
    ((uint32_t*)(src+1*stride))[0]= 
1867
    ((uint32_t*)(src+2*stride))[0]= 
1868
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1869
}
1870

    
1871

    
1872
#define LOAD_TOP_RIGHT_EDGE\
1873
    const int t4= topright[0];\
1874
    const int t5= topright[1];\
1875
    const int t6= topright[2];\
1876
    const int t7= topright[3];\
1877

    
1878
#define LOAD_LEFT_EDGE\
1879
    const int l0= src[-1+0*stride];\
1880
    const int l1= src[-1+1*stride];\
1881
    const int l2= src[-1+2*stride];\
1882
    const int l3= src[-1+3*stride];\
1883

    
1884
#define LOAD_TOP_EDGE\
1885
    const int t0= src[ 0-1*stride];\
1886
    const int t1= src[ 1-1*stride];\
1887
    const int t2= src[ 2-1*stride];\
1888
    const int t3= src[ 3-1*stride];\
1889

    
1890
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1891
    const int lt= src[-1-1*stride];
1892
    LOAD_TOP_EDGE
1893
    LOAD_LEFT_EDGE
1894

    
1895
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1896
    src[0+2*stride]=
1897
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1898
    src[0+1*stride]=
1899
    src[1+2*stride]=
1900
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1901
    src[0+0*stride]=
1902
    src[1+1*stride]=
1903
    src[2+2*stride]=
1904
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1905
    src[1+0*stride]=
1906
    src[2+1*stride]=
1907
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1908
    src[2+0*stride]=
1909
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1910
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1911
}
1912

    
1913
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1914
    LOAD_TOP_EDGE    
1915
    LOAD_TOP_RIGHT_EDGE    
1916
//    LOAD_LEFT_EDGE    
1917

    
1918
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1919
    src[1+0*stride]=
1920
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1921
    src[2+0*stride]=
1922
    src[1+1*stride]=
1923
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1924
    src[3+0*stride]=
1925
    src[2+1*stride]=
1926
    src[1+2*stride]=
1927
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1928
    src[3+1*stride]=
1929
    src[2+2*stride]=
1930
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1931
    src[3+2*stride]=
1932
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1933
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1934
}
1935

    
1936
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1937
    const int lt= src[-1-1*stride];
1938
    LOAD_TOP_EDGE    
1939
    LOAD_LEFT_EDGE    
1940
    const __attribute__((unused)) int unu= l3;
1941

    
1942
    src[0+0*stride]=
1943
    src[1+2*stride]=(lt + t0 + 1)>>1;
1944
    src[1+0*stride]=
1945
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1946
    src[2+0*stride]=
1947
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1948
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1949
    src[0+1*stride]=
1950
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1951
    src[1+1*stride]=
1952
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1953
    src[2+1*stride]=
1954
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1955
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1956
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1957
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1958
}
1959

    
1960
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1961
    LOAD_TOP_EDGE    
1962
    LOAD_TOP_RIGHT_EDGE    
1963
    const __attribute__((unused)) int unu= t7;
1964

    
1965
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1966
    src[1+0*stride]=
1967
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1968
    src[2+0*stride]=
1969
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1970
    src[3+0*stride]=
1971
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1972
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1973
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1974
    src[1+1*stride]=
1975
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1976
    src[2+1*stride]=
1977
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1978
    src[3+1*stride]=
1979
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1980
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1981
}
1982

    
1983
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1984
    LOAD_LEFT_EDGE    
1985

    
1986
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1987
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1988
    src[2+0*stride]=
1989
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1990
    src[3+0*stride]=
1991
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1992
    src[2+1*stride]=
1993
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1994
    src[3+1*stride]=
1995
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1996
    src[3+2*stride]=
1997
    src[1+3*stride]=
1998
    src[0+3*stride]=
1999
    src[2+2*stride]=
2000
    src[2+3*stride]=
2001
    src[3+3*stride]=l3;
2002
}
2003
    
2004
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2005
    const int lt= src[-1-1*stride];
2006
    LOAD_TOP_EDGE    
2007
    LOAD_LEFT_EDGE    
2008
    const __attribute__((unused)) int unu= t3;
2009

    
2010
    src[0+0*stride]=
2011
    src[2+1*stride]=(lt + l0 + 1)>>1;
2012
    src[1+0*stride]=
2013
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2014
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2015
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2016
    src[0+1*stride]=
2017
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2018
    src[1+1*stride]=
2019
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2020
    src[0+2*stride]=
2021
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2022
    src[1+2*stride]=
2023
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2024
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2025
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2026
}
2027

    
2028
static void pred16x16_vertical_c(uint8_t *src, int stride){
2029
    int i;
2030
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2031
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2032
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2033
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2034
    
2035
    for(i=0; i<16; i++){
2036
        ((uint32_t*)(src+i*stride))[0]= a;
2037
        ((uint32_t*)(src+i*stride))[1]= b;
2038
        ((uint32_t*)(src+i*stride))[2]= c;
2039
        ((uint32_t*)(src+i*stride))[3]= d;
2040
    }
2041
}
2042

    
2043
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2044
    int i;
2045

    
2046
    for(i=0; i<16; i++){
2047
        ((uint32_t*)(src+i*stride))[0]=
2048
        ((uint32_t*)(src+i*stride))[1]=
2049
        ((uint32_t*)(src+i*stride))[2]=
2050
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2051
    }
2052
}
2053

    
2054
static void pred16x16_dc_c(uint8_t *src, int stride){
2055
    int i, dc=0;
2056

    
2057
    for(i=0;i<16; i++){
2058
        dc+= src[-1+i*stride];
2059
    }
2060
    
2061
    for(i=0;i<16; i++){
2062
        dc+= src[i-stride];
2063
    }
2064

    
2065
    dc= 0x01010101*((dc + 16)>>5);
2066

    
2067
    for(i=0; i<16; i++){
2068
        ((uint32_t*)(src+i*stride))[0]=
2069
        ((uint32_t*)(src+i*stride))[1]=
2070
        ((uint32_t*)(src+i*stride))[2]=
2071
        ((uint32_t*)(src+i*stride))[3]= dc;
2072
    }
2073
}
2074

    
2075
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2076
    int i, dc=0;
2077

    
2078
    for(i=0;i<16; i++){
2079
        dc+= src[-1+i*stride];
2080
    }
2081
    
2082
    dc= 0x01010101*((dc + 8)>>4);
2083

    
2084
    for(i=0; i<16; i++){
2085
        ((uint32_t*)(src+i*stride))[0]=
2086
        ((uint32_t*)(src+i*stride))[1]=
2087
        ((uint32_t*)(src+i*stride))[2]=
2088
        ((uint32_t*)(src+i*stride))[3]= dc;
2089
    }
2090
}
2091

    
2092
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2093
    int i, dc=0;
2094

    
2095
    for(i=0;i<16; i++){
2096
        dc+= src[i-stride];
2097
    }
2098
    dc= 0x01010101*((dc + 8)>>4);
2099

    
2100
    for(i=0; i<16; i++){
2101
        ((uint32_t*)(src+i*stride))[0]=
2102
        ((uint32_t*)(src+i*stride))[1]=
2103
        ((uint32_t*)(src+i*stride))[2]=
2104
        ((uint32_t*)(src+i*stride))[3]= dc;
2105
    }
2106
}
2107

    
2108
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2109
    int i;
2110

    
2111
    for(i=0; i<16; i++){
2112
        ((uint32_t*)(src+i*stride))[0]=
2113
        ((uint32_t*)(src+i*stride))[1]=
2114
        ((uint32_t*)(src+i*stride))[2]=
2115
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2116
    }
2117
}
2118

    
2119
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2120
  int i, j, k;
2121
  int a;
2122
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2123
  const uint8_t * const src0 = src+7-stride;
2124
  const uint8_t *src1 = src+8*stride-1;
2125
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2126
  int H = src0[1] - src0[-1];
2127
  int V = src1[0] - src2[ 0];
2128
  for(k=2; k<=8; ++k) {
2129
    src1 += stride; src2 -= stride;
2130
    H += k*(src0[k] - src0[-k]);
2131
    V += k*(src1[0] - src2[ 0]);
2132
  }
2133
  if(svq3){
2134
    H = ( 5*(H/4) ) / 16;
2135
    V = ( 5*(V/4) ) / 16;
2136

    
2137
    /* required for 100% accuracy */
2138
    i = H; H = V; V = i;
2139
  }else{
2140
    H = ( 5*H+32 ) >> 6;
2141
    V = ( 5*V+32 ) >> 6;
2142
  }
2143

    
2144
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2145
  for(j=16; j>0; --j) {
2146
    int b = a;
2147
    a += V;
2148
    for(i=-16; i<0; i+=4) {
2149
      src[16+i] = cm[ (b    ) >> 5 ];
2150
      src[17+i] = cm[ (b+  H) >> 5 ];
2151
      src[18+i] = cm[ (b+2*H) >> 5 ];
2152
      src[19+i] = cm[ (b+3*H) >> 5 ];
2153
      b += 4*H;
2154
    }
2155
    src += stride;
2156
  }
2157
}
2158

    
2159
static void pred16x16_plane_c(uint8_t *src, int stride){
2160
    pred16x16_plane_compat_c(src, stride, 0);
2161
}
2162

    
2163
static void pred8x8_vertical_c(uint8_t *src, int stride){
2164
    int i;
2165
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2166
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2167
    
2168
    for(i=0; i<8; i++){
2169
        ((uint32_t*)(src+i*stride))[0]= a;
2170
        ((uint32_t*)(src+i*stride))[1]= b;
2171
    }
2172
}
2173

    
2174
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2175
    int i;
2176

    
2177
    for(i=0; i<8; i++){
2178
        ((uint32_t*)(src+i*stride))[0]=
2179
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2180
    }
2181
}
2182

    
2183
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2184
    int i;
2185

    
2186
    for(i=0; i<4; i++){
2187
        ((uint32_t*)(src+i*stride))[0]= 
2188
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2189
    }
2190
    for(i=4; i<8; i++){
2191
        ((uint32_t*)(src+i*stride))[0]= 
2192
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2193
    }
2194
}
2195

    
2196
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2197
    int i;
2198
    int dc0, dc2;
2199

    
2200
    dc0=dc2=0;
2201
    for(i=0;i<4; i++){
2202
        dc0+= src[-1+i*stride];
2203
        dc2+= src[-1+(i+4)*stride];
2204
    }
2205
    dc0= 0x01010101*((dc0 + 2)>>2);
2206
    dc2= 0x01010101*((dc2 + 2)>>2);
2207

    
2208
    for(i=0; i<4; i++){
2209
        ((uint32_t*)(src+i*stride))[0]=
2210
        ((uint32_t*)(src+i*stride))[1]= dc0;
2211
    }
2212
    for(i=4; i<8; i++){
2213
        ((uint32_t*)(src+i*stride))[0]=
2214
        ((uint32_t*)(src+i*stride))[1]= dc2;
2215
    }
2216
}
2217

    
2218
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2219
    int i;
2220
    int dc0, dc1;
2221

    
2222
    dc0=dc1=0;
2223
    for(i=0;i<4; i++){
2224
        dc0+= src[i-stride];
2225
        dc1+= src[4+i-stride];
2226
    }
2227
    dc0= 0x01010101*((dc0 + 2)>>2);
2228
    dc1= 0x01010101*((dc1 + 2)>>2);
2229

    
2230
    for(i=0; i<4; i++){
2231
        ((uint32_t*)(src+i*stride))[0]= dc0;
2232
        ((uint32_t*)(src+i*stride))[1]= dc1;
2233
    }
2234
    for(i=4; i<8; i++){
2235
        ((uint32_t*)(src+i*stride))[0]= dc0;
2236
        ((uint32_t*)(src+i*stride))[1]= dc1;
2237
    }
2238
}
2239

    
2240

    
2241
static void pred8x8_dc_c(uint8_t *src, int stride){
2242
    int i;
2243
    int dc0, dc1, dc2, dc3;
2244

    
2245
    dc0=dc1=dc2=0;
2246
    for(i=0;i<4; i++){
2247
        dc0+= src[-1+i*stride] + src[i-stride];
2248
        dc1+= src[4+i-stride];
2249
        dc2+= src[-1+(i+4)*stride];
2250
    }
2251
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2252
    dc0= 0x01010101*((dc0 + 4)>>3);
2253
    dc1= 0x01010101*((dc1 + 2)>>2);
2254
    dc2= 0x01010101*((dc2 + 2)>>2);
2255

    
2256
    for(i=0; i<4; i++){
2257
        ((uint32_t*)(src+i*stride))[0]= dc0;
2258
        ((uint32_t*)(src+i*stride))[1]= dc1;
2259
    }
2260
    for(i=4; i<8; i++){
2261
        ((uint32_t*)(src+i*stride))[0]= dc2;
2262
        ((uint32_t*)(src+i*stride))[1]= dc3;
2263
    }
2264
}
2265

    
2266
static void pred8x8_plane_c(uint8_t *src, int stride){
2267
  int j, k;
2268
  int a;
2269
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2270
  const uint8_t * const src0 = src+3-stride;
2271
  const uint8_t *src1 = src+4*stride-1;
2272
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2273
  int H = src0[1] - src0[-1];
2274
  int V = src1[0] - src2[ 0];
2275
  for(k=2; k<=4; ++k) {
2276
    src1 += stride; src2 -= stride;
2277
    H += k*(src0[k] - src0[-k]);
2278
    V += k*(src1[0] - src2[ 0]);
2279
  }
2280
  H = ( 17*H+16 ) >> 5;
2281
  V = ( 17*V+16 ) >> 5;
2282

    
2283
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2284
  for(j=8; j>0; --j) {
2285
    int b = a;
2286
    a += V;
2287
    src[0] = cm[ (b    ) >> 5 ];
2288
    src[1] = cm[ (b+  H) >> 5 ];
2289
    src[2] = cm[ (b+2*H) >> 5 ];
2290
    src[3] = cm[ (b+3*H) >> 5 ];
2291
    src[4] = cm[ (b+4*H) >> 5 ];
2292
    src[5] = cm[ (b+5*H) >> 5 ];
2293
    src[6] = cm[ (b+6*H) >> 5 ];
2294
    src[7] = cm[ (b+7*H) >> 5 ];
2295
    src += stride;
2296
  }
2297
}
2298

    
2299
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2300
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2301
                           int src_x_offset, int src_y_offset,
2302
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2303
    MpegEncContext * const s = &h->s;
2304
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2305
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2306
    const int luma_xy= (mx&3) + ((my&3)<<2);
2307
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2308
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2309
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2310
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2311
    int extra_height= extra_width;
2312
    int emu=0;
2313
    const int full_mx= mx>>2;
2314
    const int full_my= my>>2;
2315
    
2316
    assert(pic->data[0]);
2317
    
2318
    if(mx&7) extra_width -= 3;
2319
    if(my&7) extra_height -= 3;
2320
    
2321
    if(   full_mx < 0-extra_width 
2322
       || full_my < 0-extra_height 
2323
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
2324
       || full_my + 16/*FIXME*/ > s->height + extra_height){
2325
        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);
2326
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2327
        emu=1;
2328
    }
2329
    
2330
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2331
    if(!square){
2332
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2333
    }
2334
    
2335
    if(s->flags&CODEC_FLAG_GRAY) return;
2336
    
2337
    if(emu){
2338
        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);
2339
            src_cb= s->edge_emu_buffer;
2340
    }
2341
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2342

    
2343
    if(emu){
2344
        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);
2345
            src_cr= s->edge_emu_buffer;
2346
    }
2347
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2348
}
2349

    
2350
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2351
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2352
                           int x_offset, int y_offset,
2353
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2354
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2355
                           int list0, int list1){
2356
    MpegEncContext * const s = &h->s;
2357
    qpel_mc_func *qpix_op=  qpix_put;
2358
    h264_chroma_mc_func chroma_op= chroma_put;
2359
    
2360
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2361
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2362
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2363
    x_offset += 8*s->mb_x;
2364
    y_offset += 8*s->mb_y;
2365
    
2366
    if(list0){
2367
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2368
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2369
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2370
                           qpix_op, chroma_op);
2371

    
2372
        qpix_op=  qpix_avg;
2373
        chroma_op= chroma_avg;
2374
    }
2375

    
2376
    if(list1){
2377
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2378
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2379
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2380
                           qpix_op, chroma_op);
2381
    }
2382
}
2383

    
2384
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2385
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2386
                           int x_offset, int y_offset,
2387
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2388
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2389
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2390
                           int list0, int list1){
2391
    MpegEncContext * const s = &h->s;
2392

    
2393
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2394
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2395
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2396
    x_offset += 8*s->mb_x;
2397
    y_offset += 8*s->mb_y;
2398
    
2399
    if(list0 && list1){
2400
        /* don't optimize for luma-only case, since B-frames usually
2401
         * use implicit weights => chroma too. */
2402
        uint8_t *tmp_cb = s->obmc_scratchpad;
2403
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2404
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2405
        int refn0 = h->ref_cache[0][ scan8[n] ];
2406
        int refn1 = h->ref_cache[1][ scan8[n] ];
2407

    
2408
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2409
                    dest_y, dest_cb, dest_cr,
2410
                    x_offset, y_offset, qpix_put, chroma_put);
2411
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2412
                    tmp_y, tmp_cb, tmp_cr,
2413
                    x_offset, y_offset, qpix_put, chroma_put);
2414

    
2415
        if(h->use_weight == 2){
2416
            int weight0 = h->implicit_weight[refn0][refn1];
2417
            int weight1 = 64 - weight0;
2418
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2419
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2420
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2421
        }else{
2422
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2423
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2424
                            h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2425
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2426
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2427
                            h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2428
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2429
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2430
                            h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2431
        }
2432
    }else{
2433
        int list = list1 ? 1 : 0;
2434
        int refn = h->ref_cache[list][ scan8[n] ];
2435
        Picture *ref= &h->ref_list[list][refn];
2436
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2437
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2438
                    qpix_put, chroma_put);
2439

    
2440
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2441
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2442
        if(h->use_weight_chroma){
2443
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2444
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2445
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2446
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2447
        }
2448
    }
2449
}
2450

    
2451
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2452
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2453
                           int x_offset, int y_offset,
2454
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2455
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2456
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2457
                           int list0, int list1){
2458
    if((h->use_weight==2 && list0 && list1
2459
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2460
       || h->use_weight==1)
2461
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2462
                         x_offset, y_offset, qpix_put, chroma_put,
2463
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2464
    else
2465
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2466
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2467
}
2468

    
2469
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2470
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2471
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2472
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2473
    MpegEncContext * const s = &h->s;
2474
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2475
    const int mb_type= s->current_picture.mb_type[mb_xy];
2476
    
2477
    assert(IS_INTER(mb_type));
2478
    
2479
    if(IS_16X16(mb_type)){
2480
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2481
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2482
                &weight_op[0], &weight_avg[0],
2483
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2484
    }else if(IS_16X8(mb_type)){
2485
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2486
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2487
                &weight_op[1], &weight_avg[1],
2488
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2489
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2490
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2491
                &weight_op[1], &weight_avg[1],
2492
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2493
    }else if(IS_8X16(mb_type)){
2494
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2495
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2496
                &weight_op[2], &weight_avg[2],
2497
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2498
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2499
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2500
                &weight_op[2], &weight_avg[2],
2501
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2502
    }else{
2503
        int i;
2504
        
2505
        assert(IS_8X8(mb_type));
2506

    
2507
        for(i=0; i<4; i++){
2508
            const int sub_mb_type= h->sub_mb_type[i];
2509
            const int n= 4*i;
2510
            int x_offset= (i&1)<<2;
2511
            int y_offset= (i&2)<<1;
2512

    
2513
            if(IS_SUB_8X8(sub_mb_type)){
2514
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2515
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2516
                    &weight_op[3], &weight_avg[3],
2517
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2518
            }else if(IS_SUB_8X4(sub_mb_type)){
2519
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2520
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2521
                    &weight_op[4], &weight_avg[4],
2522
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2523
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2524
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2525
                    &weight_op[4], &weight_avg[4],
2526
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2527
            }else if(IS_SUB_4X8(sub_mb_type)){
2528
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2529
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2530
                    &weight_op[5], &weight_avg[5],
2531
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2532
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2533
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2534
                    &weight_op[5], &weight_avg[5],
2535
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2536
            }else{
2537
                int j;
2538
                assert(IS_SUB_4X4(sub_mb_type));
2539
                for(j=0; j<4; j++){
2540
                    int sub_x_offset= x_offset + 2*(j&1);
2541
                    int sub_y_offset= y_offset +   (j&2);
2542
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2543
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2544
                        &weight_op[6], &weight_avg[6],
2545
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2546
                }
2547
            }
2548
        }
2549
    }
2550
}
2551

    
2552
static void decode_init_vlc(H264Context *h){
2553
    static int done = 0;
2554

    
2555
    if (!done) {
2556
        int i;
2557
        done = 1;
2558

    
2559
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2560
                 &chroma_dc_coeff_token_len [0], 1, 1,
2561
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2562

    
2563
        for(i=0; i<4; i++){
2564
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2565
                     &coeff_token_len [i][0], 1, 1,
2566
                     &coeff_token_bits[i][0], 1, 1, 1);
2567
        }
2568

    
2569
        for(i=0; i<3; i++){
2570
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2571
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2572
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2573
        }
2574
        for(i=0; i<15; i++){
2575
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2576
                     &total_zeros_len [i][0], 1, 1,
2577
                     &total_zeros_bits[i][0], 1, 1, 1);
2578
        }
2579

    
2580
        for(i=0; i<6; i++){
2581
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2582
                     &run_len [i][0], 1, 1,
2583
                     &run_bits[i][0], 1, 1, 1);
2584
        }
2585
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2586
                 &run_len [6][0], 1, 1,
2587
                 &run_bits[6][0], 1, 1, 1);
2588
    }
2589
}
2590

    
2591
/**
2592
 * Sets the intra prediction function pointers.
2593
 */
2594
static void init_pred_ptrs(H264Context *h){
2595
//    MpegEncContext * const s = &h->s;
2596

    
2597
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2598
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2599
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2600
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2601
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2602
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2603
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2604
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2605
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2606
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2607
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2608
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2609

    
2610
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2611
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2612
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2613
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2614
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2615
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2616
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2617

    
2618
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2619
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2620
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2621
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2622
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2623
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2624
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2625
}
2626

    
2627
static void free_tables(H264Context *h){
2628
    av_freep(&h->intra4x4_pred_mode);
2629
    av_freep(&h->chroma_pred_mode_table);
2630
    av_freep(&h->cbp_table);
2631
    av_freep(&h->mvd_table[0]);
2632
    av_freep(&h->mvd_table[1]);
2633
    av_freep(&h->direct_table);
2634
    av_freep(&h->non_zero_count);
2635
    av_freep(&h->slice_table_base);
2636
    av_freep(&h->top_borders[1]);
2637
    av_freep(&h->top_borders[0]);
2638
    h->slice_table= NULL;
2639

    
2640
    av_freep(&h->mb2b_xy);
2641
    av_freep(&h->mb2b8_xy);
2642

    
2643
    av_freep(&h->s.obmc_scratchpad);
2644
}
2645

    
2646
/**
2647
 * allocates tables.
2648
 * needs width/height
2649
 */
2650
static int alloc_tables(H264Context *h){
2651
    MpegEncContext * const s = &h->s;
2652
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2653
    int x,y;
2654

    
2655
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2656

    
2657
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2658
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2659
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2660
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2661
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2662

    
2663
    if( h->pps.cabac ) {
2664
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2665
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2666
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2667
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2668
    }
2669

    
2670
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2671
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2672

    
2673
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2674
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2675
    for(y=0; y<s->mb_height; y++){
2676
        for(x=0; x<s->mb_width; x++){
2677
            const int mb_xy= x + y*s->mb_stride;
2678
            const int b_xy = 4*x + 4*y*h->b_stride;
2679
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2680
        
2681
            h->mb2b_xy [mb_xy]= b_xy;
2682
            h->mb2b8_xy[mb_xy]= b8_xy;
2683
        }
2684
    }
2685

    
2686
    s->obmc_scratchpad = NULL;
2687

    
2688
    return 0;
2689
fail:
2690
    free_tables(h);
2691
    return -1;
2692
}
2693

    
2694
static void common_init(H264Context *h){
2695
    MpegEncContext * const s = &h->s;
2696

    
2697
    s->width = s->avctx->width;
2698
    s->height = s->avctx->height;
2699
    s->codec_id= s->avctx->codec->id;
2700
    
2701
    init_pred_ptrs(h);
2702

    
2703
    s->unrestricted_mv=1;
2704
    s->decode=1; //FIXME
2705
}
2706

    
2707
static int decode_init(AVCodecContext *avctx){
2708
    H264Context *h= avctx->priv_data;
2709
    MpegEncContext * const s = &h->s;
2710

    
2711
    MPV_decode_defaults(s);
2712
    
2713
    s->avctx = avctx;
2714
    common_init(h);
2715

    
2716
    s->out_format = FMT_H264;
2717
    s->workaround_bugs= avctx->workaround_bugs;
2718

    
2719
    // set defaults
2720
//    s->decode_mb= ff_h263_decode_mb;
2721
    s->low_delay= 1;
2722
    avctx->pix_fmt= PIX_FMT_YUV420P;
2723

    
2724
    decode_init_vlc(h);
2725
    
2726
    if(avctx->extradata_size > 0 && avctx->extradata &&
2727
       *(char *)avctx->extradata == 1){
2728
        h->is_avc = 1;
2729
        h->got_avcC = 0;
2730
    } else {
2731
        h->is_avc = 0;
2732
    }
2733

    
2734
    return 0;
2735
}
2736

    
2737
static void frame_start(H264Context *h){
2738
    MpegEncContext * const s = &h->s;
2739
    int i;
2740

    
2741
    MPV_frame_start(s, s->avctx);
2742
    ff_er_frame_start(s);
2743

    
2744
    assert(s->linesize && s->uvlinesize);
2745

    
2746
    for(i=0; i<16; i++){
2747
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2748
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2749
    }
2750
    for(i=0; i<4; i++){
2751
        h->block_offset[16+i]=
2752
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2753
        h->block_offset[24+16+i]=
2754
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2755
    }
2756

    
2757
    /* can't be in alloc_tables because linesize isn't known there.
2758
     * FIXME: redo bipred weight to not require extra buffer? */
2759
    if(!s->obmc_scratchpad)
2760
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2761

    
2762
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2763
}
2764

    
2765
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2766
    MpegEncContext * const s = &h->s;
2767
    int i;
2768
    
2769
    src_y  -=   linesize;
2770
    src_cb -= uvlinesize;
2771
    src_cr -= uvlinesize;
2772

    
2773
    // There are two lines saved, the line above the the top macroblock of a pair,
2774
    // and the line above the bottom macroblock
2775
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2776
    for(i=1; i<17; i++){
2777
        h->left_border[i]= src_y[15+i*  linesize];
2778
    }
2779
    
2780
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2781
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2782

    
2783
    if(!(s->flags&CODEC_FLAG_GRAY)){
2784
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
2785
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2786
        for(i=1; i<9; i++){
2787
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2788
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2789
        }
2790
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2791
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2792
    }
2793
}
2794

    
2795
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){
2796
    MpegEncContext * const s = &h->s;
2797
    int temp8, i;
2798
    uint64_t temp64;
2799
    int deblock_left = (s->mb_x > 0);
2800
    int deblock_top  = (s->mb_y > 0);
2801

    
2802
    src_y  -=   linesize + 1;
2803
    src_cb -= uvlinesize + 1;
2804
    src_cr -= uvlinesize + 1;
2805

    
2806
#define XCHG(a,b,t,xchg)\
2807
t= a;\
2808
if(xchg)\
2809
    a= b;\
2810
b= t;
2811

    
2812
    if(deblock_left){
2813
        for(i = !deblock_top; i<17; i++){
2814
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2815
        }
2816
    }
2817

    
2818
    if(deblock_top){
2819
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2820
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2821
    }
2822

    
2823
    if(!(s->flags&CODEC_FLAG_GRAY)){
2824
        if(deblock_left){
2825
            for(i = !deblock_top; i<9; i++){
2826
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2827
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2828
            }
2829
        }
2830
        if(deblock_top){
2831
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2832
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2833
        }
2834
    }
2835
}
2836

    
2837
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2838
    MpegEncContext * const s = &h->s;
2839
    int i;
2840
    
2841
    src_y  -= 2 *   linesize;
2842
    src_cb -= 2 * uvlinesize;
2843
    src_cr -= 2 * uvlinesize;
2844

    
2845
    // There are two lines saved, the line above the the top macroblock of a pair,
2846
    // and the line above the bottom macroblock
2847
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2848
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
2849
    for(i=2; i<34; i++){
2850
        h->left_border[i]= src_y[15+i*  linesize];
2851
    }
2852
    
2853
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
2854
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2855
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
2856
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2857

    
2858
    if(!(s->flags&CODEC_FLAG_GRAY)){
2859
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
2860
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
2861
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
2862
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2863
        for(i=2; i<18; i++){
2864
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
2865
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2866
        }
2867
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2868
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2869
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2870
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2871
    }
2872
}
2873

    
2874
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){
2875
    MpegEncContext * const s = &h->s;
2876
    int temp8, i;
2877
    uint64_t temp64;
2878
    int deblock_left = (s->mb_x > 0);
2879
    int deblock_top  = (s->mb_y > 0);
2880

    
2881
    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);
2882

    
2883
    src_y  -= 2 *   linesize + 1;
2884
    src_cb -= 2 * uvlinesize + 1;
2885
    src_cr -= 2 * uvlinesize + 1;
2886

    
2887
#define XCHG(a,b,t,xchg)\
2888
t= a;\
2889
if(xchg)\
2890
    a= b;\
2891
b= t;
2892

    
2893
    if(deblock_left){
2894
        for(i = (!deblock_top)<<1; i<34; i++){
2895
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2896
        }
2897
    }
2898

    
2899
    if(deblock_top){
2900
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2901
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2902
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2903
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2904
    }
2905

    
2906
    if(!(s->flags&CODEC_FLAG_GRAY)){
2907
        if(deblock_left){
2908
            for(i = (!deblock_top) << 1; i<18; i++){
2909
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
2910
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2911
            }
2912
        }
2913
        if(deblock_top){
2914
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2915
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2916
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2917
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2918
        }
2919
    }
2920
}
2921

    
2922
static void hl_decode_mb(H264Context *h){
2923
    MpegEncContext * const s = &h->s;
2924
    const int mb_x= s->mb_x;
2925
    const int mb_y= s->mb_y;
2926
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2927
    const int mb_type= s->current_picture.mb_type[mb_xy];
2928
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2929
    int linesize, uvlinesize /*dct_offset*/;
2930
    int i;
2931
    int *block_offset = &h->block_offset[0];
2932
    const unsigned int bottom = mb_y & 1;
2933

    
2934
    if(!s->decode)
2935
        return;
2936

    
2937
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2938
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2939
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2940

    
2941
    if (h->mb_field_decoding_flag) {
2942
        linesize = s->linesize * 2;
2943
        uvlinesize = s->uvlinesize * 2;
2944
        block_offset = &h->block_offset[24];
2945
        if(mb_y&1){ //FIXME move out of this func?
2946
            dest_y -= s->linesize*15;
2947
            dest_cb-= s->uvlinesize*7;
2948
            dest_cr-= s->uvlinesize*7;
2949
        }
2950
    } else {
2951
        linesize = s->linesize;
2952
        uvlinesize = s->uvlinesize;
2953
//        dct_offset = s->linesize * 16;
2954
    }
2955

    
2956
    if (IS_INTRA_PCM(mb_type)) {
2957
        unsigned int x, y;
2958

    
2959
        // The pixels are stored in h->mb array in the same order as levels,
2960
        // copy them in output in the correct order.
2961
        for(i=0; i<16; i++) {
2962
            for (y=0; y<4; y++) {
2963
                for (x=0; x<4; x++) {
2964
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2965
                }
2966
            }
2967
        }
2968
        for(i=16; i<16+4; i++) {
2969
            for (y=0; y<4; y++) {
2970
                for (x=0; x<4; x++) {
2971
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2972
                }
2973
            }
2974
        }
2975
        for(i=20; i<20+4; i++) {
2976
            for (y=0; y<4; y++) {
2977
                for (x=0; x<4; x++) {
2978
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2979
                }
2980
            }
2981
        }
2982
    } else {
2983
        if(IS_INTRA(mb_type)){
2984
            if(h->deblocking_filter) {
2985
                if (h->mb_aff_frame) {
2986
                    if (!bottom)
2987
                        xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
2988
                } else {
2989
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2990
                }
2991
            }
2992

    
2993
            if(!(s->flags&CODEC_FLAG_GRAY)){
2994
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2995
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2996
            }
2997

    
2998
            if(IS_INTRA4x4(mb_type)){
2999
                if(!s->encoding){
3000
                    for(i=0; i<16; i++){
3001
                        uint8_t * const ptr= dest_y + block_offset[i];
3002
                        uint8_t *topright;
3003
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3004
                        int tr;
3005

    
3006
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3007
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3008
                            assert(mb_y || linesize <= block_offset[i]);
3009
                            if(!topright_avail){
3010
                                tr= ptr[3 - linesize]*0x01010101;
3011
                                topright= (uint8_t*) &tr;
3012
                            }else if(i==5 && h->deblocking_filter){
3013
                                tr= *(uint32_t*)h->top_borders[h->mb_aff_frame ? IS_INTERLACED(mb_type) ? bottom : 1 : 0][mb_x+1];
3014
                                topright= (uint8_t*) &tr;
3015
                            }else
3016
                                topright= ptr + 4 - linesize;
3017
                        }else
3018
                            topright= NULL;
3019

    
3020
                        h->pred4x4[ dir ](ptr, topright, linesize);
3021
                        if(h->non_zero_count_cache[ scan8[i] ]){
3022
                            if(s->codec_id == CODEC_ID_H264)
3023
                                s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3024
                            else
3025
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3026
                        }
3027
                    }
3028
                }
3029
            }else{
3030
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3031
                if(s->codec_id == CODEC_ID_H264)
3032
                    h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3033
                else
3034
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3035
            }
3036
            if(h->deblocking_filter) {
3037
                if (h->mb_aff_frame) {
3038
                    if (bottom) {
3039
                        uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
3040
                        uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3041
                        uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3042
                        s->mb_y--;
3043
                        xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3044
                        s->mb_y++;
3045
                    }
3046
                } else {
3047
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3048
                }
3049
            }
3050
        }else if(s->codec_id == CODEC_ID_H264){
3051
            hl_motion(h, dest_y, dest_cb, dest_cr,
3052
                      s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
3053
                      s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3054
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3055
        }
3056

    
3057

    
3058
        if(!IS_INTRA4x4(mb_type)){
3059
            if(s->codec_id == CODEC_ID_H264){
3060
                for(i=0; i<16; i++){
3061
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3062
                        uint8_t * const ptr= dest_y + block_offset[i];
3063
                        s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3064
                    }
3065
                }
3066
            }else{
3067
                for(i=0; i<16; i++){
3068
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3069
                        uint8_t * const ptr= dest_y + block_offset[i];
3070
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3071
                    }
3072
                }
3073
            }
3074
        }
3075

    
3076
        if(!(s->flags&CODEC_FLAG_GRAY)){
3077
            chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3078
            chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3079
            if(s->codec_id == CODEC_ID_H264){
3080
                for(i=16; i<16+4; i++){
3081
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3082
                        uint8_t * const ptr= dest_cb + block_offset[i];
3083
                        s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3084
                    }
3085
                }
3086
                for(i=20; i<20+4; i++){
3087
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3088
                        uint8_t * const ptr= dest_cr + block_offset[i];
3089
                        s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3090
                    }
3091
                }
3092
            }else{
3093
                for(i=16; i<16+4; i++){
3094
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3095
                        uint8_t * const ptr= dest_cb + block_offset[i];
3096
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3097
                    }
3098
                }
3099
                for(i=20; i<20+4; i++){
3100
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3101
                        uint8_t * const ptr= dest_cr + block_offset[i];
3102
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3103
                    }
3104
                }
3105
            }
3106
        }
3107
    }
3108
    if(h->deblocking_filter) {
3109
        if (h->mb_aff_frame) {
3110
            const int mb_y = s->mb_y - 1;
3111
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3112
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3113
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3114
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3115
            uint8_t tmp = s->current_picture.data[1][384];
3116
            if (!bottom) return;
3117
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3118
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3119
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3120

    
3121
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3122
            // TODO deblock a pair
3123
            // top 
3124
            s->mb_y--;
3125
            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);
3126
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3127
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3128
            if (tmp != s->current_picture.data[1][384]) {
3129
                tprintf("modified pixel 8,1 (1)\n");
3130
            }
3131
            // bottom
3132
            s->mb_y++;
3133
            tprintf("call mbaff filter_mb\n");
3134
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3135
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3136
            if (tmp != s->current_picture.data[1][384]) {
3137
                tprintf("modified pixel 8,1 (2)\n");
3138
            }
3139
        } else {
3140
            tprintf("call filter_mb\n");
3141
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3142
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3143
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3144
        }
3145
    }
3146
}
3147

    
3148
/**
3149
 * fills the default_ref_list.
3150
 */
3151
static int fill_default_ref_list(H264Context *h){
3152
    MpegEncContext * const s = &h->s;
3153
    int i;
3154
    int smallest_poc_greater_than_current = -1;
3155
    Picture sorted_short_ref[32];
3156
    
3157
    if(h->slice_type==B_TYPE){
3158
        int out_i;
3159
        int limit= -1;
3160

    
3161
        /* sort frame according to poc in B slice */
3162
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3163
            int best_i=-1;
3164
            int best_poc=INT_MAX;
3165

    
3166
            for(i=0; i<h->short_ref_count; i++){
3167
                const int poc= h->short_ref[i]->poc;
3168
                if(poc > limit && poc < best_poc){
3169
                    best_poc= poc;
3170
                    best_i= i;
3171
                }
3172
            }
3173
            
3174
            assert(best_i != -1);
3175
            
3176
            limit= best_poc;
3177
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3178
            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);
3179
            if (-1 == smallest_poc_greater_than_current) {
3180
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3181
                    smallest_poc_greater_than_current = out_i;
3182
                }
3183
            }
3184
        }
3185
    }
3186

    
3187
    if(s->picture_structure == PICT_FRAME){
3188
        if(h->slice_type==B_TYPE){
3189
            int list;
3190
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3191

    
3192
            // find the largest poc
3193
            for(list=0; list<2; list++){
3194
                int index = 0;
3195
                int j= -99;
3196
                int step= list ? -1 : 1;
3197

    
3198
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3199
                    while(j<0 || j>= h->short_ref_count){
3200
                        if(j != -99 && step == (list ? -1 : 1))
3201
                            return -1;
3202
                        step = -step;
3203
                        j= smallest_poc_greater_than_current + (step>>1);
3204
                    }
3205
                    if(sorted_short_ref[j].reference != 3) continue;
3206
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3207
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3208
                }
3209

    
3210
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3211
                    if(h->long_ref[i] == NULL) continue;
3212
                    if(h->long_ref[i]->reference != 3) continue;
3213

    
3214
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3215
                    h->default_ref_list[ list ][index++].pic_id= i;;
3216
                }
3217
                
3218
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3219
                    // swap the two first elements of L1 when
3220
                    // L0 and L1 are identical
3221
                    Picture temp= h->default_ref_list[1][0];
3222
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3223
                    h->default_ref_list[1][1] = temp;
3224
                }
3225

    
3226
                if(index < h->ref_count[ list ])
3227
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3228
            }
3229
        }else{
3230
            int index=0;
3231
            for(i=0; i<h->short_ref_count; i++){
3232
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3233
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3234
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3235
            }
3236
            for(i = 0; i < 16; i++){
3237
                if(h->long_ref[i] == NULL) continue;
3238
                if(h->long_ref[i]->reference != 3) continue;
3239
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3240
                h->default_ref_list[0][index++].pic_id= i;;
3241
            }
3242
            if(index < h->ref_count[0])
3243
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3244
        }
3245
    }else{ //FIELD
3246
        if(h->slice_type==B_TYPE){
3247
        }else{
3248
            //FIXME second field balh
3249
        }
3250
    }
3251
#ifdef TRACE
3252
    for (i=0; i<h->ref_count[0]; i++) {
3253
        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]);
3254
    }
3255
    if(h->slice_type==B_TYPE){
3256
        for (i=0; i<h->ref_count[1]; i++) {
3257
            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]);
3258
        }
3259
    }
3260
#endif
3261
    return 0;
3262
}
3263

    
3264
static void print_short_term(H264Context *h);
3265
static void print_long_term(H264Context *h);
3266

    
3267
static int decode_ref_pic_list_reordering(H264Context *h){
3268
    MpegEncContext * const s = &h->s;
3269
    int list, index;
3270
    
3271
    print_short_term(h);
3272
    print_long_term(h);
3273
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3274
    
3275
    for(list=0; list<2; list++){
3276
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3277

    
3278
        if(get_bits1(&s->gb)){
3279
            int pred= h->curr_pic_num;
3280

    
3281
            for(index=0; ; index++){
3282
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3283
                int pic_id;
3284
                int i;
3285
                Picture *ref = NULL;
3286
                
3287
                if(reordering_of_pic_nums_idc==3) 
3288
                    break;
3289
                
3290
                if(index >= h->ref_count[list]){
3291
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3292
                    return -1;
3293
                }
3294
                
3295
                if(reordering_of_pic_nums_idc<3){
3296
                    if(reordering_of_pic_nums_idc<2){
3297
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3298

    
3299
                        if(abs_diff_pic_num >= h->max_pic_num){
3300
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3301
                            return -1;
3302
                        }
3303

    
3304
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3305
                        else                                pred+= abs_diff_pic_num;
3306
                        pred &= h->max_pic_num - 1;
3307
                    
3308
                        for(i= h->short_ref_count-1; i>=0; i--){
3309
                            ref = h->short_ref[i];
3310
                            assert(ref->reference == 3);
3311
                            assert(!ref->long_ref);
3312
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3313
                                break;
3314
                        }
3315
                        if(i>=0)
3316
                            ref->pic_id= ref->frame_num;
3317
                    }else{
3318
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3319
                        ref = h->long_ref[pic_id];
3320
                        ref->pic_id= pic_id;
3321
                        assert(ref->reference == 3);
3322
                        assert(ref->long_ref);
3323
                        i=0;
3324
                    }
3325

    
3326
                    if (i < 0) {
3327
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3328
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3329
                    } else {
3330
                        for(i=index; i+1<h->ref_count[list]; i++){
3331
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3332
                                break;
3333
                        }
3334
                        for(; i > index; i--){
3335
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3336
                        }
3337
                        h->ref_list[list][index]= *ref;
3338
                    }
3339
                }else{
3340
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3341
                    return -1;
3342
                }
3343
            }
3344
        }
3345

    
3346
        if(h->slice_type!=B_TYPE) break;
3347
    }
3348
    for(list=0; list<2; list++){
3349
        for(index= 0; index < h->ref_count[list]; index++){
3350
            if(!h->ref_list[list][index].data[0])
3351
                h->ref_list[list][index]= s->current_picture;
3352
        }
3353
        if(h->slice_type!=B_TYPE) break;
3354
    }
3355
    
3356
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3357
        direct_dist_scale_factor(h);
3358
    direct_ref_list_init(h);
3359
    return 0;    
3360
}
3361

    
3362
static int pred_weight_table(H264Context *h){
3363
    MpegEncContext * const s = &h->s;
3364
    int list, i;
3365
    int luma_def, chroma_def;
3366
    
3367
    h->use_weight= 0;
3368
    h->use_weight_chroma= 0;
3369
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3370
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3371
    luma_def = 1<<h->luma_log2_weight_denom;
3372
    chroma_def = 1<<h->chroma_log2_weight_denom;
3373

    
3374
    for(list=0; list<2; list++){
3375
        for(i=0; i<h->ref_count[list]; i++){
3376
            int luma_weight_flag, chroma_weight_flag;
3377
            
3378
            luma_weight_flag= get_bits1(&s->gb);
3379
            if(luma_weight_flag){
3380
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3381
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3382
                if(   h->luma_weight[list][i] != luma_def
3383
                   || h->luma_offset[list][i] != 0)
3384
                    h->use_weight= 1;
3385
            }else{
3386
                h->luma_weight[list][i]= luma_def;
3387
                h->luma_offset[list][i]= 0;
3388
            }
3389

    
3390
            chroma_weight_flag= get_bits1(&s->gb);
3391
            if(chroma_weight_flag){
3392
                int j;
3393
                for(j=0; j<2; j++){
3394
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3395
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3396
                    if(   h->chroma_weight[list][i][j] != chroma_def
3397
                       || h->chroma_offset[list][i][j] != 0)
3398
                        h->use_weight_chroma= 1;
3399
                }
3400
            }else{
3401
                int j;
3402
                for(j=0; j<2; j++){
3403
                    h->chroma_weight[list][i][j]= chroma_def;
3404
                    h->chroma_offset[list][i][j]= 0;
3405
                }
3406
            }
3407
        }
3408
        if(h->slice_type != B_TYPE) break;
3409
    }
3410
    h->use_weight= h->use_weight || h->use_weight_chroma;
3411
    return 0;
3412
}
3413

    
3414
static void implicit_weight_table(H264Context *h){
3415
    MpegEncContext * const s = &h->s;
3416
    int ref0, ref1;
3417
    int cur_poc = s->current_picture_ptr->poc;
3418

    
3419
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3420
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3421
        h->use_weight= 0;
3422
        h->use_weight_chroma= 0;
3423
        return;
3424
    }
3425

    
3426
    h->use_weight= 2;
3427
    h->use_weight_chroma= 2;
3428
    h->luma_log2_weight_denom= 5;
3429
    h->chroma_log2_weight_denom= 5;
3430

    
3431
    /* FIXME: MBAFF */
3432
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3433
        int poc0 = h->ref_list[0][ref0].poc;
3434
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3435
            int poc1 = h->ref_list[1][ref1].poc;
3436
            int td = clip(poc1 - poc0, -128, 127);
3437
            if(td){
3438
                int tb = clip(cur_poc - poc0, -128, 127);
3439
                int tx = (16384 + (ABS(td) >> 1)) / td;
3440
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3441
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3442
                    h->implicit_weight[ref0][ref1] = 32;
3443
                else
3444
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3445
            }else
3446
                h->implicit_weight[ref0][ref1] = 32;
3447
        }
3448
    }
3449
}
3450

    
3451
static inline void unreference_pic(H264Context *h, Picture *pic){
3452
    int i;
3453
    pic->reference=0;
3454
    if(pic == h->delayed_output_pic)
3455
        pic->reference=1;
3456
    else{
3457
        for(i = 0; h->delayed_pic[i]; i++)
3458
            if(pic == h->delayed_pic[i]){
3459
                pic->reference=1;
3460
                break;
3461
            }
3462
    }
3463
}
3464

    
3465
/**
3466
 * instantaneous decoder refresh.
3467
 */
3468
static void idr(H264Context *h){
3469
    int i;
3470

    
3471
    for(i=0; i<16; i++){
3472
        if (h->long_ref[i] != NULL) {
3473
            unreference_pic(h, h->long_ref[i]);
3474
            h->long_ref[i]= NULL;
3475
        }
3476
    }
3477
    h->long_ref_count=0;
3478

    
3479
    for(i=0; i<h->short_ref_count; i++){
3480
        unreference_pic(h, h->short_ref[i]);
3481
        h->short_ref[i]= NULL;
3482
    }
3483
    h->short_ref_count=0;
3484
}
3485

    
3486
/* forget old pics after a seek */
3487
static void flush_dpb(AVCodecContext *avctx){
3488
    H264Context *h= avctx->priv_data;
3489
    int i;
3490
    for(i=0; i<16; i++)
3491
        h->delayed_pic[i]= NULL;
3492
    h->delayed_output_pic= NULL;
3493
    idr(h);
3494
}
3495

    
3496
/**
3497
 *
3498
 * @return the removed picture or NULL if an error occurs
3499
 */
3500
static Picture * remove_short(H264Context *h, int frame_num){
3501
    MpegEncContext * const s = &h->s;
3502
    int i;
3503
    
3504
    if(s->avctx->debug&FF_DEBUG_MMCO)
3505
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3506
    
3507
    for(i=0; i<h->short_ref_count; i++){
3508
        Picture *pic= h->short_ref[i];
3509
        if(s->avctx->debug&FF_DEBUG_MMCO)
3510
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3511
        if(pic->frame_num == frame_num){
3512
            h->short_ref[i]= NULL;
3513
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3514
            h->short_ref_count--;
3515
            return pic;
3516
        }
3517
    }
3518
    return NULL;
3519
}
3520

    
3521
/**
3522
 *
3523
 * @return the removed picture or NULL if an error occurs
3524
 */
3525
static Picture * remove_long(H264Context *h, int i){
3526
    Picture *pic;
3527

    
3528
    pic= h->long_ref[i];
3529
    h->long_ref[i]= NULL;
3530
    if(pic) h->long_ref_count--;
3531

    
3532
    return pic;
3533
}
3534

    
3535
/**
3536
 * print short term list
3537
 */
3538
static void print_short_term(H264Context *h) {
3539
    uint32_t i;
3540
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3541
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3542
        for(i=0; i<h->short_ref_count; i++){
3543
            Picture *pic= h->short_ref[i];
3544
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3545
        }
3546
    }
3547
}
3548

    
3549
/**
3550
 * print long term list
3551
 */
3552
static void print_long_term(H264Context *h) {
3553
    uint32_t i;
3554
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3555
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3556
        for(i = 0; i < 16; i++){
3557
            Picture *pic= h->long_ref[i];
3558
            if (pic) {
3559
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3560
            }
3561
        }
3562
    }
3563
}
3564

    
3565
/**
3566
 * Executes the reference picture marking (memory management control operations).
3567
 */
3568
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3569
    MpegEncContext * const s = &h->s;
3570
    int i, j;
3571
    int current_is_long=0;
3572
    Picture *pic;
3573
    
3574
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3575
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3576
        
3577
    for(i=0; i<mmco_count; i++){
3578
        if(s->avctx->debug&FF_DEBUG_MMCO)
3579
            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);
3580

    
3581
        switch(mmco[i].opcode){
3582
        case MMCO_SHORT2UNUSED:
3583
            pic= remove_short(h, mmco[i].short_frame_num);
3584
            if(pic==NULL) return -1;
3585
            unreference_pic(h, pic);
3586
            break;
3587
        case MMCO_SHORT2LONG:
3588
            pic= remove_long(h, mmco[i].long_index);
3589
            if(pic) unreference_pic(h, pic);
3590
            
3591
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3592
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3593
            h->long_ref_count++;
3594
            break;
3595
        case MMCO_LONG2UNUSED:
3596
            pic= remove_long(h, mmco[i].long_index);
3597
            if(pic==NULL) return -1;
3598
            unreference_pic(h, pic);
3599
            break;
3600
        case MMCO_LONG:
3601
            pic= remove_long(h, mmco[i].long_index);
3602
            if(pic) unreference_pic(h, pic);
3603
            
3604
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3605
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3606
            h->long_ref_count++;
3607
            
3608
            current_is_long=1;
3609
            break;
3610
        case MMCO_SET_MAX_LONG:
3611
            assert(mmco[i].long_index <= 16);
3612
            // just remove the long term which index is greater than new max
3613
            for(j = mmco[i].long_index; j<16; j++){
3614
                pic = remove_long(h, j);
3615
                if (pic) unreference_pic(h, pic);
3616
            }
3617
            break;
3618
        case MMCO_RESET:
3619
            while(h->short_ref_count){
3620
                pic= remove_short(h, h->short_ref[0]->frame_num);
3621
                unreference_pic(h, pic);
3622
            }
3623
            for(j = 0; j < 16; j++) {
3624
                pic= remove_long(h, j);
3625
                if(pic) unreference_pic(h, pic);
3626
            }
3627
            break;
3628
        default: assert(0);
3629
        }
3630
    }
3631
    
3632
    if(!current_is_long){
3633
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3634
        if(pic){
3635
            unreference_pic(h, pic);
3636
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3637
        }
3638
        
3639
        if(h->short_ref_count)
3640
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3641

    
3642
        h->short_ref[0]= s->current_picture_ptr;
3643
        h->short_ref[0]->long_ref=0;
3644
        h->short_ref_count++;
3645
    }
3646
    
3647
    print_short_term(h);
3648
    print_long_term(h);
3649
    return 0; 
3650
}
3651

    
3652
static int decode_ref_pic_marking(H264Context *h){
3653
    MpegEncContext * const s = &h->s;
3654
    int i;
3655
    
3656
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3657
        s->broken_link= get_bits1(&s->gb) -1;
3658
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3659
        if(h->mmco[0].long_index == -1)
3660
            h->mmco_index= 0;
3661
        else{
3662
            h->mmco[0].opcode= MMCO_LONG;
3663
            h->mmco_index= 1;
3664
        } 
3665
    }else{
3666
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3667
            for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3668
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3669

    
3670
                h->mmco[i].opcode= opcode;
3671
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3672
                    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
3673
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3674
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3675
                        return -1;
3676
                    }*/
3677
                }
3678
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3679
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
3680
                    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){
3681
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3682
                        return -1;
3683
                    }
3684
                }
3685
                    
3686
                if(opcode > MMCO_LONG){
3687
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3688
                    return -1;
3689
                }
3690
                if(opcode == MMCO_END)
3691
                    break;
3692
            }
3693
            h->mmco_index= i;
3694
        }else{
3695
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3696

    
3697
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3698
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3699
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3700
                h->mmco_index= 1;
3701
            }else
3702
                h->mmco_index= 0;
3703
        }
3704
    }
3705
    
3706
    return 0; 
3707
}
3708

    
3709
static int init_poc(H264Context *h){
3710
    MpegEncContext * const s = &h->s;
3711
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3712
    int field_poc[2];
3713

    
3714
    if(h->nal_unit_type == NAL_IDR_SLICE){
3715
        h->frame_num_offset= 0;
3716
    }else{
3717
        if(h->frame_num < h->prev_frame_num)
3718
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3719
        else
3720
            h->frame_num_offset= h->prev_frame_num_offset;
3721
    }
3722

    
3723
    if(h->sps.poc_type==0){
3724
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3725

    
3726
        if(h->nal_unit_type == NAL_IDR_SLICE){
3727
             h->prev_poc_msb=
3728
             h->prev_poc_lsb= 0;
3729
        }
3730

    
3731
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3732
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3733
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3734
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3735
        else
3736
            h->poc_msb = h->prev_poc_msb;
3737
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3738
        field_poc[0] = 
3739
        field_poc[1] = h->poc_msb + h->poc_lsb;
3740
        if(s->picture_structure == PICT_FRAME) 
3741
            field_poc[1] += h->delta_poc_bottom;
3742
    }else if(h->sps.poc_type==1){
3743
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3744
        int i;
3745

    
3746
        if(h->sps.poc_cycle_length != 0)
3747
            abs_frame_num = h->frame_num_offset + h->frame_num;
3748
        else
3749
            abs_frame_num = 0;
3750

    
3751
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3752
            abs_frame_num--;
3753
            
3754
        expected_delta_per_poc_cycle = 0;
3755
        for(i=0; i < h->sps.poc_cycle_length; i++)
3756
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3757

    
3758
        if(abs_frame_num > 0){
3759
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3760
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3761

    
3762
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3763
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3764
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3765
        } else
3766
            expectedpoc = 0;
3767

    
3768
        if(h->nal_ref_idc == 0) 
3769
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3770
        
3771
        field_poc[0] = expectedpoc + h->delta_poc[0];
3772
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3773

    
3774
        if(s->picture_structure == PICT_FRAME)
3775
            field_poc[1] += h->delta_poc[1];
3776
    }else{
3777
        int poc;
3778
        if(h->nal_unit_type == NAL_IDR_SLICE){
3779
            poc= 0;
3780
        }else{
3781
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3782
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3783
        }
3784
        field_poc[0]= poc;
3785
        field_poc[1]= poc;
3786
    }
3787
    
3788
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3789
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3790
    if(s->picture_structure != PICT_TOP_FIELD)
3791
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3792
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3793
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3794

    
3795
    return 0;
3796
}
3797

    
3798
/**
3799
 * decodes a slice header.
3800
 * this will allso call MPV_common_init() and frame_start() as needed
3801
 */
3802
static int decode_slice_header(H264Context *h){
3803
    MpegEncContext * const s = &h->s;
3804
    int first_mb_in_slice, pps_id;
3805
    int num_ref_idx_active_override_flag;
3806
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3807
    int slice_type;
3808
    int default_ref_list_done = 0;
3809

    
3810
    s->current_picture.reference= h->nal_ref_idc != 0;
3811
    s->dropable= h->nal_ref_idc == 0;
3812

    
3813
    first_mb_in_slice= get_ue_golomb(&s->gb);
3814

    
3815
    slice_type= get_ue_golomb(&s->gb);
3816
    if(slice_type > 9){
3817
        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);
3818
        return -1;
3819
    }
3820
    if(slice_type > 4){
3821
        slice_type -= 5;
3822
        h->slice_type_fixed=1;
3823
    }else
3824
        h->slice_type_fixed=0;
3825
    
3826
    slice_type= slice_type_map[ slice_type ];
3827
    if (slice_type == I_TYPE
3828
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3829
        default_ref_list_done = 1;
3830
    }
3831
    h->slice_type= slice_type;
3832

    
3833
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
3834
        
3835
    pps_id= get_ue_golomb(&s->gb);
3836
    if(pps_id>255){
3837
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3838
        return -1;
3839
    }
3840
    h->pps= h->pps_buffer[pps_id];
3841
    if(h->pps.slice_group_count == 0){
3842
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3843
        return -1;
3844
    }
3845

    
3846
    h->sps= h->sps_buffer[ h->pps.sps_id ];
3847
    if(h->sps.log2_max_frame_num == 0){
3848
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3849
        return -1;
3850
    }
3851
    
3852
    s->mb_width= h->sps.mb_width;
3853
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3854
    
3855
    h->b_stride=  s->mb_width*4 + 1;
3856
    h->b8_stride= s->mb_width*2 + 1;
3857

    
3858
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3859
    if(h->sps.frame_mbs_only_flag)
3860
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3861
    else
3862
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3863
    
3864
    if (s->context_initialized 
3865
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3866
        free_tables(h);
3867
        MPV_common_end(s);
3868
    }
3869
    if (!s->context_initialized) {
3870
        if (MPV_common_init(s) < 0)
3871
            return -1;
3872

    
3873
        alloc_tables(h);
3874

    
3875
        s->avctx->width = s->width;
3876
        s->avctx->height = s->height;
3877
        s->avctx->sample_aspect_ratio= h->sps.sar;
3878
        if(!s->avctx->sample_aspect_ratio.den)
3879
            s->avctx->sample_aspect_ratio.den = 1;
3880

    
3881
        if(h->sps.timing_info_present_flag){
3882
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
3883
        }
3884
    }
3885

    
3886
    if(h->slice_num == 0){
3887
        frame_start(h);
3888
    }
3889

    
3890
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3891
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3892

    
3893
    h->mb_aff_frame = 0;
3894
    if(h->sps.frame_mbs_only_flag){
3895
        s->picture_structure= PICT_FRAME;
3896
    }else{
3897
        if(get_bits1(&s->gb)) { //field_pic_flag
3898
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3899
        } else {
3900
            s->picture_structure= PICT_FRAME;
3901
            first_mb_in_slice <<= 1;
3902
            h->mb_aff_frame = h->sps.mb_aff;
3903
        }
3904
    }
3905

    
3906
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3907
    s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
3908
    
3909
    if(s->picture_structure==PICT_FRAME){
3910
        h->curr_pic_num=   h->frame_num;
3911
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3912
    }else{
3913
        h->curr_pic_num= 2*h->frame_num;
3914
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3915
    }
3916
        
3917
    if(h->nal_unit_type == NAL_IDR_SLICE){
3918
        get_ue_golomb(&s->gb); /* idr_pic_id */
3919
    }
3920
   
3921
    if(h->sps.poc_type==0){
3922
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3923
        
3924
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3925
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3926
        }
3927
    }
3928
    
3929
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3930
        h->delta_poc[0]= get_se_golomb(&s->gb);
3931
        
3932
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3933
            h->delta_poc[1]= get_se_golomb(&s->gb);
3934
    }
3935
    
3936
    init_poc(h);
3937
    
3938
    if(h->pps.redundant_pic_cnt_present){
3939
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3940
    }
3941

    
3942
    //set defaults, might be overriden a few line later
3943
    h->ref_count[0]= h->pps.ref_count[0];
3944
    h->ref_count[1]= h->pps.ref_count[1];
3945

    
3946
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3947
        if(h->slice_type == B_TYPE){
3948
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3949
        }
3950
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3951
    
3952
        if(num_ref_idx_active_override_flag){
3953
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3954
            if(h->slice_type==B_TYPE)
3955
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3956

    
3957
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3958
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3959
                return -1;
3960
            }
3961
        }
3962
    }
3963

    
3964
    if(!default_ref_list_done){
3965
        fill_default_ref_list(h);
3966
    }
3967

    
3968
    decode_ref_pic_list_reordering(h);
3969

    
3970
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3971
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3972
        pred_weight_table(h);
3973
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3974
        implicit_weight_table(h);
3975
    else
3976
        h->use_weight = 0;
3977
    
3978
    if(s->current_picture.reference)
3979
        decode_ref_pic_marking(h);
3980

    
3981
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3982
        h->cabac_init_idc = get_ue_golomb(&s->gb);
3983

    
3984
    h->last_qscale_diff = 0;
3985
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3986
    if(s->qscale<0 || s->qscale>51){
3987
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3988
        return -1;
3989
    }
3990
    h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
3991
    //FIXME qscale / qp ... stuff
3992
    if(h->slice_type == SP_TYPE){
3993
        get_bits1(&s->gb); /* sp_for_switch_flag */
3994
    }
3995
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3996
        get_se_golomb(&s->gb); /* slice_qs_delta */
3997
    }
3998

    
3999
    h->deblocking_filter = 1;
4000
    h->slice_alpha_c0_offset = 0;
4001
    h->slice_beta_offset = 0;
4002
    if( h->pps.deblocking_filter_parameters_present ) {
4003
        h->deblocking_filter= get_ue_golomb(&s->gb);
4004
        if(h->deblocking_filter < 2) 
4005
            h->deblocking_filter^= 1; // 1<->0
4006

    
4007
        if( h->deblocking_filter ) {
4008
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4009
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4010
        }
4011
    }
4012

    
4013
#if 0 //FMO
4014
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4015
        slice_group_change_cycle= get_bits(&s->gb, ?);
4016
#endif
4017

    
4018
    h->slice_num++;
4019

    
4020
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4021
        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", 
4022
               h->slice_num,
4023
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4024
               first_mb_in_slice, 
4025
               av_get_pict_type_char(h->slice_type),
4026
               pps_id, h->frame_num,
4027
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4028
               h->ref_count[0], h->ref_count[1],
4029
               s->qscale,
4030
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4031
               h->use_weight,
4032
               h->use_weight==1 && h->use_weight_chroma ? "c" : ""
4033
               );
4034
    }
4035

    
4036
    return 0;
4037
}
4038

    
4039
/**
4040
 *
4041
 */
4042
static inline int get_level_prefix(GetBitContext *gb){
4043
    unsigned int buf;
4044
    int log;
4045
    
4046
    OPEN_READER(re, gb);
4047
    UPDATE_CACHE(re, gb);
4048
    buf=GET_CACHE(re, gb);
4049
    
4050
    log= 32 - av_log2(buf);
4051
#ifdef TRACE
4052
    print_bin(buf>>(32-log), log);
4053
    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__);
4054
#endif
4055

    
4056
    LAST_SKIP_BITS(re, gb, log);
4057
    CLOSE_READER(re, gb);
4058

    
4059
    return log-1;
4060
}
4061

    
4062
/**
4063
 * decodes a residual block.
4064
 * @param n block index
4065
 * @param scantable scantable
4066
 * @param max_coeff number of coefficients in the block
4067
 * @return <0 if an error occured
4068
 */
4069
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
4070
    MpegEncContext * const s = &h->s;
4071
    const uint16_t *qmul= dequant_coeff[qp];
4072
    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};
4073
    int level[16], run[16];
4074
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
4075

    
4076
    //FIXME put trailing_onex into the context
4077

    
4078
    if(n == CHROMA_DC_BLOCK_INDEX){
4079
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4080
        total_coeff= coeff_token>>2;
4081
    }else{    
4082
        if(n == LUMA_DC_BLOCK_INDEX){
4083
            total_coeff= pred_non_zero_count(h, 0);
4084
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4085
            total_coeff= coeff_token>>2;
4086
        }else{
4087
            total_coeff= pred_non_zero_count(h, n);
4088
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4089
            total_coeff= coeff_token>>2;
4090
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4091
        }
4092
    }
4093

    
4094
    //FIXME set last_non_zero?
4095

    
4096
    if(total_coeff==0)
4097
        return 0;
4098
        
4099
    trailing_ones= coeff_token&3;
4100
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
4101
    assert(total_coeff<=16);
4102
    
4103
    for(i=0; i<trailing_ones; i++){
4104
        level[i]= 1 - 2*get_bits1(gb);
4105
    }
4106

    
4107
    suffix_length= total_coeff > 10 && trailing_ones < 3;
4108

    
4109
    for(; i<total_coeff; i++){
4110
        const int prefix= get_level_prefix(gb);
4111
        int level_code, mask;
4112

    
4113
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4114
            if(suffix_length)
4115
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4116
            else
4117
                level_code= (prefix<<suffix_length); //part
4118
        }else if(prefix==14){
4119
            if(suffix_length)
4120
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4121
            else
4122
                level_code= prefix + get_bits(gb, 4); //part
4123
        }else if(prefix==15){
4124
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
4125
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4126
        }else{
4127
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4128
            return -1;
4129
        }
4130

    
4131
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
4132

    
4133
        mask= -(level_code&1);
4134
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
4135

    
4136
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
4137

    
4138
#if 1
4139
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4140
#else        
4141
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4142
        /* ? == prefix > 2 or sth */
4143
#endif
4144
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
4145
    }
4146

    
4147
    if(total_coeff == max_coeff)
4148
        zeros_left=0;
4149
    else{
4150
        if(n == CHROMA_DC_BLOCK_INDEX)
4151
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4152
        else
4153
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4154
    }
4155
    
4156
    for(i=0; i<total_coeff-1; i++){
4157
        if(zeros_left <=0)
4158
            break;
4159
        else if(zeros_left < 7){
4160
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4161
        }else{
4162
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4163
        }
4164
        zeros_left -= run[i];
4165
    }
4166

    
4167
    if(zeros_left<0){
4168
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4169
        return -1;
4170
    }
4171
    
4172
    for(; i<total_coeff-1; i++){
4173
        run[i]= 0;
4174
    }
4175

    
4176
    run[i]= zeros_left;
4177

    
4178
    coeff_num=-1;
4179
    if(n > 24){
4180
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4181
            int j;
4182

    
4183
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4184
            j= scantable[ coeff_num ];
4185

    
4186
            block[j]= level[i];
4187
        }
4188
    }else{
4189
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
4190
            int j;
4191

    
4192
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4193
            j= scantable[ coeff_num ];
4194

    
4195
            block[j]= level[i] * qmul[j];
4196
//            printf("%d %d  ", block[j], qmul[j]);
4197
        }
4198
    }
4199
    return 0;
4200
}
4201

    
4202
/**
4203
 * decodes a P_SKIP or B_SKIP macroblock
4204
 */
4205
static void decode_mb_skip(H264Context *h){
4206
    MpegEncContext * const s = &h->s;
4207
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4208
    int mb_type=0;
4209
    
4210
    memset(h->non_zero_count[mb_xy], 0, 16);
4211
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4212

    
4213
    if(h->mb_aff_frame && s->mb_skip_run==0 && (s->mb_y&1)==0){
4214
        h->mb_field_decoding_flag= get_bits1(&s->gb);
4215
    }
4216
    if(h->mb_field_decoding_flag)
4217
        mb_type|= MB_TYPE_INTERLACED;
4218

    
4219
    if( h->slice_type == B_TYPE )
4220
    {
4221
        // just for fill_caches. pred_direct_motion will set the real mb_type
4222
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4223

    
4224
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4225
        pred_direct_motion(h, &mb_type);
4226
        if(h->pps.cabac){
4227
            fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4228
            fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4229
        }
4230
    }
4231
    else
4232
    {
4233
        int mx, my;
4234
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4235

    
4236
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4237
        pred_pskip_motion(h, &mx, &my);
4238
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4239
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4240
        if(h->pps.cabac)
4241
            fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4242
    }
4243

    
4244
    write_back_motion(h, mb_type);
4245
    s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
4246
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4247
    h->slice_table[ mb_xy ]= h->slice_num;
4248
    h->prev_mb_skipped= 1;
4249
}
4250

    
4251
/**
4252
 * decodes a macroblock
4253
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4254
 */
4255
static int decode_mb_cavlc(H264Context *h){
4256
    MpegEncContext * const s = &h->s;
4257
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4258
    int mb_type, partition_count, cbp;
4259

    
4260
    s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?    
4261

    
4262
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4263
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4264
                down the code */
4265
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
4266
        if(s->mb_skip_run==-1)
4267
            s->mb_skip_run= get_ue_golomb(&s->gb);
4268
        
4269
        if (s->mb_skip_run--) {
4270
            decode_mb_skip(h);
4271
            return 0;
4272
        }
4273
    }
4274
    if(h->mb_aff_frame){
4275
        if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
4276
            h->mb_field_decoding_flag = get_bits1(&s->gb);
4277
    }else
4278
        h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
4279
    
4280
    h->prev_mb_skipped= 0;
4281
    
4282
    mb_type= get_ue_golomb(&s->gb);
4283
    if(h->slice_type == B_TYPE){
4284
        if(mb_type < 23){
4285
            partition_count= b_mb_type_info[mb_type].partition_count;
4286
            mb_type=         b_mb_type_info[mb_type].type;
4287
        }else{
4288
            mb_type -= 23;
4289
            goto decode_intra_mb;
4290
        }
4291
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
4292
        if(mb_type < 5){
4293
            partition_count= p_mb_type_info[mb_type].partition_count;
4294
            mb_type=         p_mb_type_info[mb_type].type;
4295
        }else{
4296
            mb_type -= 5;
4297
            goto decode_intra_mb;
4298
        }
4299
    }else{
4300
       assert(h->slice_type == I_TYPE);
4301
decode_intra_mb:
4302
        if(mb_type > 25){
4303
            av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice to large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
4304
            return -1;
4305
        }
4306
        partition_count=0;
4307
        cbp= i_mb_type_info[mb_type].cbp;
4308
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4309
        mb_type= i_mb_type_info[mb_type].type;
4310
    }
4311

    
4312
    if(h->mb_field_decoding_flag)
4313
        mb_type |= MB_TYPE_INTERLACED;
4314

    
4315
    s->current_picture.mb_type[mb_xy]= mb_type;
4316
    h->slice_table[ mb_xy ]= h->slice_num;
4317
    
4318
    if(IS_INTRA_PCM(mb_type)){
4319
        unsigned int x, y;
4320
        
4321
        // we assume these blocks are very rare so we dont optimize it
4322
        align_get_bits(&s->gb);
4323
        
4324
        // The pixels are stored in the same order as levels in h->mb array.
4325
        for(y=0; y<16; y++){
4326
            const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4327
            for(x=0; x<16; x++){
4328
                tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4329
                h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4330
            }
4331
        }
4332
        for(y=0; y<8; y++){
4333
            const int index= 256 + 4*(y&3) + 32*(y>>2);
4334
            for(x=0; x<8; x++){
4335
                tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4336
                h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4337
            }
4338
        }
4339
        for(y=0; y<8; y++){
4340
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4341
            for(x=0; x<8; x++){
4342
                tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4343
                h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4344
            }
4345
        }
4346
    
4347
        // In deblocking, the quantizer is 0
4348
        s->current_picture.qscale_table[mb_xy]= 0;
4349
        h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
4350
        // All coeffs are present
4351
        memset(h->non_zero_count[mb_xy], 16, 16);
4352
        
4353
        return 0;
4354
    }
4355
        
4356
    fill_caches(h, mb_type, 0);
4357

    
4358
    //mb_pred
4359
    if(IS_INTRA(mb_type)){
4360
//            init_top_left_availability(h);
4361
            if(IS_INTRA4x4(mb_type)){
4362
                int i;
4363

    
4364
//                fill_intra4x4_pred_table(h);
4365
                for(i=0; i<16; i++){
4366
                    const int mode_coded= !get_bits1(&s->gb);
4367
                    const int predicted_mode=  pred_intra_mode(h, i);
4368
                    int mode;
4369

    
4370
                    if(mode_coded){
4371
                        const int rem_mode= get_bits(&s->gb, 3);
4372
                        if(rem_mode<predicted_mode)
4373
                            mode= rem_mode;
4374
                        else
4375
                            mode= rem_mode + 1;
4376
                    }else{
4377
                        mode= predicted_mode;
4378
                    }
4379
                    
4380
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4381
                }
4382
                write_back_intra_pred_mode(h);
4383
                if( check_intra4x4_pred_mode(h) < 0)
4384
                    return -1;
4385
            }else{
4386
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4387
                if(h->intra16x16_pred_mode < 0)
4388
                    return -1;
4389
            }
4390
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
4391

    
4392
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4393
            if(h->chroma_pred_mode < 0)
4394
                return -1;
4395
    }else if(partition_count==4){
4396
        int i, j, sub_partition_count[4], list, ref[2][4];
4397
        
4398
        if(h->slice_type == B_TYPE){
4399
            for(i=0; i<4; i++){
4400
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4401
                if(h->sub_mb_type[i] >=13){
4402
                    av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4403
                    return -1;
4404
                }
4405
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4406
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4407
            }
4408
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4409
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4410
                pred_direct_motion(h, &mb_type);
4411
        }else{
4412
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4413
            for(i=0; i<4; i++){
4414
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4415
                if(h->sub_mb_type[i] >=4){
4416
                    av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
4417
                    return -1;
4418
                }
4419
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4420
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4421
            }
4422
        }
4423
        
4424
        for(list=0; list<2; list++){
4425
            int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4426
            if(ref_count == 0) continue;
4427
            if (h->mb_aff_frame && h->mb_field_decoding_flag) {
4428
                ref_count <<= 1;
4429
            }
4430
            for(i=0; i<4; i++){
4431
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4432
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4433
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4434
                }else{
4435
                 //FIXME
4436
                    ref[list][i] = -1;
4437
                }
4438
            }
4439
        }
4440
        
4441
        for(list=0; list<2; list++){
4442
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4443
            if(ref_count == 0) continue;
4444

    
4445
            for(i=0; i<4; i++){
4446
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4447
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4448
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4449

    
4450
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4451
                    const int sub_mb_type= h->sub_mb_type[i];
4452
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4453
                    for(j=0; j<sub_partition_count[i]; j++){
4454
                        int mx, my;
4455
                        const int index= 4*i + block_width*j;
4456
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4457
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4458
                        mx += get_se_golomb(&s->gb);
4459
                        my += get_se_golomb(&s->gb);
4460
                        tprintf("final mv:%d %d\n", mx, my);
4461

    
4462
                        if(IS_SUB_8X8(sub_mb_type)){
4463
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
4464
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4465
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
4466
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4467
                        }else if(IS_SUB_8X4(sub_mb_type)){
4468
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4469
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4470
                        }else if(IS_SUB_4X8(sub_mb_type)){
4471
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4472
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4473
                        }