Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ fbd312fd

History | View | Annotate | Download (291 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 transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
65
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
66
    int poc_type;                      ///< pic_order_cnt_type
67
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
68
    int delta_pic_order_always_zero_flag;
69
    int offset_for_non_ref_pic;
70
    int offset_for_top_to_bottom_field;
71
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
72
    int ref_frame_count;               ///< num_ref_frames
73
    int gaps_in_frame_num_allowed_flag;
74
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
75
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
76
    int frame_mbs_only_flag;
77
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
78
    int direct_8x8_inference_flag;
79
    int crop;                   ///< frame_cropping_flag
80
    int crop_left;              ///< frame_cropping_rect_left_offset
81
    int crop_right;             ///< frame_cropping_rect_right_offset
82
    int crop_top;               ///< frame_cropping_rect_top_offset
83
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
84
    int vui_parameters_present_flag;
85
    AVRational sar;
86
    int timing_info_present_flag;
87
    uint32_t num_units_in_tick;
88
    uint32_t time_scale;
89
    int fixed_frame_rate_flag;
90
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
91
    int bitstream_restriction_flag;
92
    int num_reorder_frames;
93
}SPS;
94

    
95
/**
96
 * Picture parameter set
97
 */
98
typedef struct PPS{
99
    int sps_id;
100
    int cabac;                  ///< entropy_coding_mode_flag
101
    int pic_order_present;      ///< pic_order_present_flag
102
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
103
    int mb_slice_group_map_type;
104
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
105
    int weighted_pred;          ///< weighted_pred_flag
106
    int weighted_bipred_idc;
107
    int init_qp;                ///< pic_init_qp_minus26 + 26
108
    int init_qs;                ///< pic_init_qs_minus26 + 26
109
    int chroma_qp_index_offset;
110
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
111
    int constrained_intra_pred; ///< constrained_intra_pred_flag
112
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
114
}PPS;
115

    
116
/**
117
 * Memory management control operation opcode.
118
 */
119
typedef enum MMCOOpcode{
120
    MMCO_END=0,
121
    MMCO_SHORT2UNUSED,
122
    MMCO_LONG2UNUSED,
123
    MMCO_SHORT2LONG,
124
    MMCO_SET_MAX_LONG,
125
    MMCO_RESET, 
126
    MMCO_LONG,
127
} MMCOOpcode;
128

    
129
/**
130
 * Memory management control operation.
131
 */
132
typedef struct MMCO{
133
    MMCOOpcode opcode;
134
    int short_frame_num;
135
    int long_index;
136
} MMCO;
137

    
138
/**
139
 * H264Context
140
 */
141
typedef struct H264Context{
142
    MpegEncContext s;
143
    int nal_ref_idc;        
144
    int nal_unit_type;
145
#define NAL_SLICE                1
146
#define NAL_DPA                        2
147
#define NAL_DPB                        3
148
#define NAL_DPC                        4
149
#define NAL_IDR_SLICE                5
150
#define NAL_SEI                        6
151
#define NAL_SPS                        7
152
#define NAL_PPS                        8
153
#define NAL_PICTURE_DELIMITER        9
154
#define NAL_FILTER_DATA                10
155
    uint8_t *rbsp_buffer;
156
    int rbsp_buffer_size;
157

    
158
    /**
159
      * Used to parse AVC variant of h264
160
      */
161
    int is_avc; ///< this flag is != 0 if codec is avc1
162
    int got_avcC; ///< flag used to parse avcC data only once
163
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
164

    
165
    int chroma_qp; //QPc
166

    
167
    int prev_mb_skipped; //FIXME remove (IMHO not used)
168

    
169
    //prediction stuff
170
    int chroma_pred_mode;
171
    int intra16x16_pred_mode;
172

    
173
    int top_mb_xy;
174
    int left_mb_xy[2];
175
    
176
    int8_t intra4x4_pred_mode_cache[5*8];
177
    int8_t (*intra4x4_pred_mode)[8];
178
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
179
    void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
180
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
181
    void (*pred16x16[4+3])(uint8_t *src, int stride);
182
    unsigned int topleft_samples_available;
183
    unsigned int top_samples_available;
184
    unsigned int topright_samples_available;
185
    unsigned int left_samples_available;
186
    uint8_t (*top_borders[2])[16+2*8];
187
    uint8_t left_border[2*(17+2*9)];
188

    
189
    /**
190
     * non zero coeff count cache.
191
     * is 64 if not available.
192
     */
193
    uint8_t non_zero_count_cache[6*8] __align8;
194
    uint8_t (*non_zero_count)[16];
195

    
196
    /**
197
     * Motion vector cache.
198
     */
199
    int16_t mv_cache[2][5*8][2] __align8;
200
    int8_t ref_cache[2][5*8] __align8;
201
#define LIST_NOT_USED -1 //FIXME rename?
202
#define PART_NOT_AVAILABLE -2
203
    
204
    /**
205
     * is 1 if the specific list MV&references are set to 0,0,-2.
206
     */
207
    int mv_cache_clean[2];
208

    
209
    /**
210
     * number of neighbors (top and/or left) that used 8x8 dct
211
     */
212
    int neighbor_transform_size;
213

    
214
    /**
215
     * block_offset[ 0..23] for frame macroblocks
216
     * block_offset[24..47] for field macroblocks
217
     */
218
    int block_offset[2*(16+8)];
219
    
220
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
221
    uint32_t *mb2b8_xy;
222
    int b_stride; //FIXME use s->b4_stride
223
    int b8_stride;
224

    
225
    int halfpel_flag;
226
    int thirdpel_flag;
227

    
228
    int unknown_svq3_flag;
229
    int next_slice_index;
230

    
231
    SPS sps_buffer[MAX_SPS_COUNT];
232
    SPS sps; ///< current sps
233
    
234
    PPS pps_buffer[MAX_PPS_COUNT];
235
    /**
236
     * current pps
237
     */
238
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
239

    
240
    uint16_t (*dequant4_coeff)[16]; // FIXME quant matrices should be per SPS or PPS
241
    uint16_t (*dequant8_coeff)[64];
242

    
243
    int slice_num;
244
    uint8_t *slice_table_base;
245
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
246
    int slice_type;
247
    int slice_type_fixed;
248
    
249
    //interlacing specific flags
250
    int mb_aff_frame;
251
    int mb_field_decoding_flag;
252
    
253
    int sub_mb_type[4];
254
    
255
    //POC stuff
256
    int poc_lsb;
257
    int poc_msb;
258
    int delta_poc_bottom;
259
    int delta_poc[2];
260
    int frame_num;
261
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
262
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
263
    int frame_num_offset;         ///< for POC type 2
264
    int prev_frame_num_offset;    ///< for POC type 2
265
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
266

    
267
    /**
268
     * frame_num for frames or 2*frame_num for field pics.
269
     */
270
    int curr_pic_num;
271
    
272
    /**
273
     * max_frame_num or 2*max_frame_num for field pics.
274
     */
275
    int max_pic_num;
276

    
277
    //Weighted pred stuff
278
    int use_weight;
279
    int use_weight_chroma;
280
    int luma_log2_weight_denom;
281
    int chroma_log2_weight_denom;
282
    int luma_weight[2][16];
283
    int luma_offset[2][16];
284
    int chroma_weight[2][16][2];
285
    int chroma_offset[2][16][2];
286
    int implicit_weight[16][16];
287
   
288
    //deblock
289
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
290
    int slice_alpha_c0_offset;
291
    int slice_beta_offset;
292
     
293
    int redundant_pic_count;
294
    
295
    int direct_spatial_mv_pred;
296
    int dist_scale_factor[16];
297
    int map_col_to_list0[2][16];
298

    
299
    /**
300
     * num_ref_idx_l0/1_active_minus1 + 1
301
     */
302
    int ref_count[2];// FIXME split for AFF
303
    Picture *short_ref[32];
304
    Picture *long_ref[32];
305
    Picture default_ref_list[2][32];
306
    Picture ref_list[2][32]; //FIXME size?
307
    Picture field_ref_list[2][32]; //FIXME size?
308
    Picture *delayed_pic[16]; //FIXME size?
309
    Picture *delayed_output_pic;
310
    
311
    /**
312
     * memory management control operations buffer.
313
     */
314
    MMCO mmco[MAX_MMCO_COUNT];
315
    int mmco_index;
316
    
317
    int long_ref_count;  ///< number of actual long term references
318
    int short_ref_count; ///< number of actual short term references
319
    
320
    //data partitioning
321
    GetBitContext intra_gb;
322
    GetBitContext inter_gb;
323
    GetBitContext *intra_gb_ptr;
324
    GetBitContext *inter_gb_ptr;
325
    
326
    DCTELEM mb[16*24] __align8;
327

    
328
    /**
329
     * Cabac
330
     */
331
    CABACContext cabac;
332
    uint8_t      cabac_state[460];
333
    int          cabac_init_idc;
334

    
335
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
336
    uint16_t     *cbp_table;
337
    int top_cbp;
338
    int left_cbp;
339
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
340
    uint8_t     *chroma_pred_mode_table;
341
    int         last_qscale_diff;
342
    int16_t     (*mvd_table[2])[2];
343
    int16_t     mvd_cache[2][5*8][2] __align8;
344
    uint8_t     *direct_table;
345
    uint8_t     direct_cache[5*8];
346

    
347
    uint8_t zigzag_scan[16];
348
    uint8_t field_scan[16];
349
    const uint8_t *zigzag_scan_q0;
350
    const uint8_t *field_scan_q0;
351
    
352
    int x264_build;
353
}H264Context;
354

    
355
static VLC coeff_token_vlc[4];
356
static VLC chroma_dc_coeff_token_vlc;
357

    
358
static VLC total_zeros_vlc[15];
359
static VLC chroma_dc_total_zeros_vlc[3];
360

    
361
static VLC run_vlc[6];
362
static VLC run7_vlc;
363

    
364
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
365
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
366
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);
367

    
368
static inline uint32_t pack16to32(int a, int b){
369
#ifdef WORDS_BIGENDIAN
370
   return (b&0xFFFF) + (a<<16);
371
#else
372
   return (a&0xFFFF) + (b<<16);
373
#endif
374
}
375

    
376
/**
377
 * fill a rectangle.
378
 * @param h height of the rectangle, should be a constant
379
 * @param w width of the rectangle, should be a constant
380
 * @param size the size of val (1 or 4), should be a constant
381
 */
382
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
383
    uint8_t *p= (uint8_t*)vp;
384
    assert(size==1 || size==4);
385
    
386
    w      *= size;
387
    stride *= size;
388
    
389
    assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
390
    assert((stride&(w-1))==0);
391
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
392
    if(w==2 && h==2){
393
        *(uint16_t*)(p + 0)=
394
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
395
    }else if(w==2 && h==4){
396
        *(uint16_t*)(p + 0*stride)=
397
        *(uint16_t*)(p + 1*stride)=
398
        *(uint16_t*)(p + 2*stride)=
399
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
400
    }else if(w==4 && h==1){
401
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
402
    }else if(w==4 && h==2){
403
        *(uint32_t*)(p + 0*stride)=
404
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
405
    }else if(w==4 && h==4){
406
        *(uint32_t*)(p + 0*stride)=
407
        *(uint32_t*)(p + 1*stride)=
408
        *(uint32_t*)(p + 2*stride)=
409
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
410
    }else if(w==8 && h==1){
411
        *(uint32_t*)(p + 0)=
412
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
413
    }else if(w==8 && h==2){
414
        *(uint32_t*)(p + 0 + 0*stride)=
415
        *(uint32_t*)(p + 4 + 0*stride)=
416
        *(uint32_t*)(p + 0 + 1*stride)=
417
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
418
    }else if(w==8 && h==4){
419
        *(uint64_t*)(p + 0*stride)=
420
        *(uint64_t*)(p + 1*stride)=
421
        *(uint64_t*)(p + 2*stride)=
422
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
423
    }else if(w==16 && h==2){
424
        *(uint64_t*)(p + 0+0*stride)=
425
        *(uint64_t*)(p + 8+0*stride)=
426
        *(uint64_t*)(p + 0+1*stride)=
427
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
428
    }else if(w==16 && h==4){
429
        *(uint64_t*)(p + 0+0*stride)=
430
        *(uint64_t*)(p + 8+0*stride)=
431
        *(uint64_t*)(p + 0+1*stride)=
432
        *(uint64_t*)(p + 8+1*stride)=
433
        *(uint64_t*)(p + 0+2*stride)=
434
        *(uint64_t*)(p + 8+2*stride)=
435
        *(uint64_t*)(p + 0+3*stride)=
436
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
437
    }else
438
        assert(0);
439
}
440

    
441
static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
442
    MpegEncContext * const s = &h->s;
443
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
444
    int topleft_xy, top_xy, topright_xy, left_xy[2];
445
    int topleft_type, top_type, topright_type, left_type[2];
446
    int left_block[8];
447
    int i;
448

    
449
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
450
    // the actual condition is whether we're on the edge of a slice,
451
    // and even then the intra and nnz parts are unnecessary.
452
    if(for_deblock && h->slice_num == 1)
453
        return;
454

    
455
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it 
456
    
457
    top_xy     = mb_xy  - s->mb_stride;
458
    topleft_xy = top_xy - 1;
459
    topright_xy= top_xy + 1;
460
    left_xy[1] = left_xy[0] = mb_xy-1;
461
    left_block[0]= 0;
462
    left_block[1]= 1;
463
    left_block[2]= 2;
464
    left_block[3]= 3;
465
    left_block[4]= 7;
466
    left_block[5]= 10;
467
    left_block[6]= 8;
468
    left_block[7]= 11;
469
    if(h->mb_aff_frame){
470
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
471
        const int top_pair_xy      = pair_xy     - s->mb_stride;
472
        const int topleft_pair_xy  = top_pair_xy - 1;
473
        const int topright_pair_xy = top_pair_xy + 1;
474
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
475
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
476
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
477
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
478
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
479
        const int bottom = (s->mb_y & 1);
480
        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);
481
        if (bottom
482
                ? !curr_mb_frame_flag // bottom macroblock
483
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
484
                ) {
485
            top_xy -= s->mb_stride;
486
        }
487
        if (bottom
488
                ? !curr_mb_frame_flag // bottom macroblock
489
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
490
                ) {
491
            topleft_xy -= s->mb_stride;
492
        }
493
        if (bottom
494
                ? !curr_mb_frame_flag // bottom macroblock
495
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
496
                ) {
497
            topright_xy -= s->mb_stride;
498
        }
499
        if (left_mb_frame_flag != curr_mb_frame_flag) {
500
            left_xy[1] = left_xy[0] = pair_xy - 1;
501
            if (curr_mb_frame_flag) {
502
                if (bottom) {
503
                    left_block[0]= 2;
504
                    left_block[1]= 2;
505
                    left_block[2]= 3;
506
                    left_block[3]= 3;
507
                    left_block[4]= 8;
508
                    left_block[5]= 11;
509
                    left_block[6]= 8;
510
                    left_block[7]= 11;
511
                } else {
512
                    left_block[0]= 0;
513
                    left_block[1]= 0;
514
                    left_block[2]= 1;
515
                    left_block[3]= 1;
516
                    left_block[4]= 7;
517
                    left_block[5]= 10;
518
                    left_block[6]= 7;
519
                    left_block[7]= 10;
520
                }
521
            } else {
522
                left_xy[1] += s->mb_stride;
523
                //left_block[0]= 0;
524
                left_block[1]= 2;
525
                left_block[2]= 0;
526
                left_block[3]= 2;
527
                //left_block[4]= 7;
528
                left_block[5]= 10;
529
                left_block[6]= 7;
530
                left_block[7]= 10;
531
            }
532
        }
533
    }
534

    
535
    h->top_mb_xy = top_xy;
536
    h->left_mb_xy[0] = left_xy[0];
537
    h->left_mb_xy[1] = left_xy[1];
538
    if(for_deblock){
539
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
540
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
541
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
542
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
543
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
544
    }else{
545
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
546
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
547
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
548
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
549
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
550
    }
551

    
552
    if(IS_INTRA(mb_type)){
553
        h->topleft_samples_available= 
554
        h->top_samples_available= 
555
        h->left_samples_available= 0xFFFF;
556
        h->topright_samples_available= 0xEEEA;
557

    
558
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
559
            h->topleft_samples_available= 0xB3FF;
560
            h->top_samples_available= 0x33FF;
561
            h->topright_samples_available= 0x26EA;
562
        }
563
        for(i=0; i<2; i++){
564
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
565
                h->topleft_samples_available&= 0xDF5F;
566
                h->left_samples_available&= 0x5F5F;
567
            }
568
        }
569
        
570
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
571
            h->topleft_samples_available&= 0x7FFF;
572
        
573
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
574
            h->topright_samples_available&= 0xFBFF;
575
    
576
        if(IS_INTRA4x4(mb_type)){
577
            if(IS_INTRA4x4(top_type)){
578
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
579
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
580
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
581
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
582
            }else{
583
                int pred;
584
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
585
                    pred= -1;
586
                else{
587
                    pred= 2;
588
                }
589
                h->intra4x4_pred_mode_cache[4+8*0]=
590
                h->intra4x4_pred_mode_cache[5+8*0]=
591
                h->intra4x4_pred_mode_cache[6+8*0]=
592
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
593
            }
594
            for(i=0; i<2; i++){
595
                if(IS_INTRA4x4(left_type[i])){
596
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
597
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
598
                }else{
599
                    int pred;
600
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
601
                        pred= -1;
602
                    else{
603
                        pred= 2;
604
                    }
605
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
606
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
607
                }
608
            }
609
        }
610
    }
611
    
612
    
613
/*
614
0 . T T. T T T T 
615
1 L . .L . . . . 
616
2 L . .L . . . . 
617
3 . T TL . . . . 
618
4 L . .L . . . . 
619
5 L . .. . . . . 
620
*/
621
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
622
    if(top_type){
623
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
624
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
625
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
626
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
627
    
628
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
629
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
630
    
631
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
632
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
633
        
634
    }else{
635
        h->non_zero_count_cache[4+8*0]=      
636
        h->non_zero_count_cache[5+8*0]=
637
        h->non_zero_count_cache[6+8*0]=
638
        h->non_zero_count_cache[7+8*0]=
639
    
640
        h->non_zero_count_cache[1+8*0]=
641
        h->non_zero_count_cache[2+8*0]=
642
    
643
        h->non_zero_count_cache[1+8*3]=
644
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
645
        
646
    }
647

    
648
    for (i=0; i<2; i++) {
649
        if(left_type[i]){
650
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
651
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
652
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
653
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
654
        }else{
655
            h->non_zero_count_cache[3+8*1 + 2*8*i]= 
656
            h->non_zero_count_cache[3+8*2 + 2*8*i]= 
657
            h->non_zero_count_cache[0+8*1 +   8*i]= 
658
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
659
        }
660
    }
661

    
662
    if( h->pps.cabac ) {
663
        // top_cbp
664
        if(top_type) {
665
            h->top_cbp = h->cbp_table[top_xy];
666
        } else if(IS_INTRA(mb_type)) {
667
            h->top_cbp = 0x1C0;
668
        } else {
669
            h->top_cbp = 0;
670
        }
671
        // left_cbp
672
        if (left_type[0]) {
673
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
674
        } else if(IS_INTRA(mb_type)) {
675
            h->left_cbp = 0x1C0;
676
        } else {
677
            h->left_cbp = 0;
678
        }
679
        if (left_type[0]) {
680
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
681
        }
682
        if (left_type[1]) {
683
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
684
        }
685
    }
686

    
687
#if 1
688
    //FIXME direct mb can skip much of this
689
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
690
        int list;
691
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
692
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
693
                /*if(!h->mv_cache_clean[list]){
694
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
695
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
696
                    h->mv_cache_clean[list]= 1;
697
                }*/
698
                continue;
699
            }
700
            h->mv_cache_clean[list]= 0;
701
            
702
            if(IS_INTER(top_type)){
703
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
704
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
705
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
706
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
707
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
708
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
709
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
710
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
711
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
712
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
713
            }else{
714
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
715
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
716
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
717
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
718
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
719
            }
720

    
721
            //FIXME unify cleanup or sth
722
            if(IS_INTER(left_type[0])){
723
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
724
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
725
                *(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]];
726
                *(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]];
727
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
728
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
729
            }else{
730
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
731
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
732
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
733
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
734
            }
735
            
736
            if(IS_INTER(left_type[1])){
737
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
738
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
739
                *(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]];
740
                *(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]];
741
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
742
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
743
            }else{
744
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
745
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
746
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
747
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
748
                assert((!left_type[0]) == (!left_type[1]));
749
            }
750

    
751
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
752
                continue;
753

    
754
            if(IS_INTER(topleft_type)){
755
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
756
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
757
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
758
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
759
            }else{
760
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
761
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
762
            }
763
            
764
            if(IS_INTER(topright_type)){
765
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
766
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
767
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
768
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
769
            }else{
770
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
771
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
772
            }
773
            
774

    
775
            h->ref_cache[list][scan8[5 ]+1] = 
776
            h->ref_cache[list][scan8[7 ]+1] = 
777
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
778
            h->ref_cache[list][scan8[4 ]] = 
779
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
780
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
781
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
782
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
783
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
784
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
785

    
786
            if( h->pps.cabac ) {
787
                /* XXX beurk, Load mvd */
788
                if(IS_INTER(topleft_type)){
789
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
790
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
791
                }else{
792
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
793
                }
794

    
795
                if(IS_INTER(top_type)){
796
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
797
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
798
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
799
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
800
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
801
                }else{
802
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
803
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
804
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
805
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
806
                }
807
                if(IS_INTER(left_type[0])){
808
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
809
                    *(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]];
810
                    *(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]];
811
                }else{
812
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
813
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
814
                }
815
                if(IS_INTER(left_type[1])){
816
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
817
                    *(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]];
818
                    *(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]];
819
                }else{
820
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
821
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
822
                }
823
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
824
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
825
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
826
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
827
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
828

    
829
                if(h->slice_type == B_TYPE){
830
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
831

    
832
                    if(IS_DIRECT(top_type)){
833
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
834
                    }else if(IS_8X8(top_type)){
835
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
836
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
837
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
838
                    }else{
839
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
840
                    }
841
                    
842
                    //FIXME interlacing
843
                    if(IS_DIRECT(left_type[0])){
844
                        h->direct_cache[scan8[0] - 1 + 0*8]=
845
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
846
                    }else if(IS_8X8(left_type[0])){
847
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
848
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
849
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
850
                    }else{
851
                        h->direct_cache[scan8[0] - 1 + 0*8]=
852
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
853
                    }
854
                }
855
            }
856
        }
857
    }
858
#endif
859

    
860
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
861
}
862

    
863
static inline void write_back_intra_pred_mode(H264Context *h){
864
    MpegEncContext * const s = &h->s;
865
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
866

    
867
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
868
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
869
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
870
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
871
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
872
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
873
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
874
}
875

    
876
/**
877
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
878
 */
879
static inline int check_intra4x4_pred_mode(H264Context *h){
880
    MpegEncContext * const s = &h->s;
881
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
882
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
883
    int i;
884
    
885
    if(!(h->top_samples_available&0x8000)){
886
        for(i=0; i<4; i++){
887
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
888
            if(status<0){
889
                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);
890
                return -1;
891
            } else if(status){
892
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
893
            }
894
        }
895
    }
896
    
897
    if(!(h->left_samples_available&0x8000)){
898
        for(i=0; i<4; i++){
899
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
900
            if(status<0){
901
                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);
902
                return -1;
903
            } else if(status){
904
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
905
            }
906
        }
907
    }
908

    
909
    return 0;
910
} //FIXME cleanup like next
911

    
912
/**
913
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
914
 */
915
static inline int check_intra_pred_mode(H264Context *h, int mode){
916
    MpegEncContext * const s = &h->s;
917
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
918
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
919
    
920
    if(mode < 0 || mode > 6) {
921
        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);
922
        return -1;
923
    }
924
    
925
    if(!(h->top_samples_available&0x8000)){
926
        mode= top[ mode ];
927
        if(mode<0){
928
            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);
929
            return -1;
930
        }
931
    }
932
    
933
    if(!(h->left_samples_available&0x8000)){
934
        mode= left[ mode ];
935
        if(mode<0){
936
            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);
937
            return -1;
938
        } 
939
    }
940

    
941
    return mode;
942
}
943

    
944
/**
945
 * gets the predicted intra4x4 prediction mode.
946
 */
947
static inline int pred_intra_mode(H264Context *h, int n){
948
    const int index8= scan8[n];
949
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
950
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
951
    const int min= FFMIN(left, top);
952

    
953
    tprintf("mode:%d %d min:%d\n", left ,top, min);
954

    
955
    if(min<0) return DC_PRED;
956
    else      return min;
957
}
958

    
959
static inline void write_back_non_zero_count(H264Context *h){
960
    MpegEncContext * const s = &h->s;
961
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
962

    
963
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
964
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
965
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
966
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
967
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
968
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
969
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
970
    
971
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
972
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
973
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
974

    
975
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
976
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
977
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
978
}
979

    
980
/**
981
 * gets the predicted number of non zero coefficients.
982
 * @param n block index
983
 */
984
static inline int pred_non_zero_count(H264Context *h, int n){
985
    const int index8= scan8[n];
986
    const int left= h->non_zero_count_cache[index8 - 1];
987
    const int top = h->non_zero_count_cache[index8 - 8];
988
    int i= left + top;
989
    
990
    if(i<64) i= (i+1)>>1;
991

    
992
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
993

    
994
    return i&31;
995
}
996

    
997
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
998
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
999

    
1000
    if(topright_ref != PART_NOT_AVAILABLE){
1001
        *C= h->mv_cache[list][ i - 8 + part_width ];
1002
        return topright_ref;
1003
    }else{
1004
        tprintf("topright MV not available\n");
1005

    
1006
        *C= h->mv_cache[list][ i - 8 - 1 ];
1007
        return h->ref_cache[list][ i - 8 - 1 ];
1008
    }
1009
}
1010

    
1011
/**
1012
 * gets the predicted MV.
1013
 * @param n the block index
1014
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1015
 * @param mx the x component of the predicted motion vector
1016
 * @param my the y component of the predicted motion vector
1017
 */
1018
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1019
    const int index8= scan8[n];
1020
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1021
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1022
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1023
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1024
    const int16_t * C;
1025
    int diagonal_ref, match_count;
1026

    
1027
    assert(part_width==1 || part_width==2 || part_width==4);
1028

    
1029
/* mv_cache
1030
  B . . A T T T T 
1031
  U . . L . . , .
1032
  U . . L . . . .
1033
  U . . L . . , .
1034
  . . . L . . . .
1035
*/
1036

    
1037
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1038
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1039
    tprintf("pred_motion match_count=%d\n", match_count);
1040
    if(match_count > 1){ //most common
1041
        *mx= mid_pred(A[0], B[0], C[0]);
1042
        *my= mid_pred(A[1], B[1], C[1]);
1043
    }else if(match_count==1){
1044
        if(left_ref==ref){
1045
            *mx= A[0];
1046
            *my= A[1];        
1047
        }else if(top_ref==ref){
1048
            *mx= B[0];
1049
            *my= B[1];        
1050
        }else{
1051
            *mx= C[0];
1052
            *my= C[1];        
1053
        }
1054
    }else{
1055
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1056
            *mx= A[0];
1057
            *my= A[1];        
1058
        }else{
1059
            *mx= mid_pred(A[0], B[0], C[0]);
1060
            *my= mid_pred(A[1], B[1], C[1]);
1061
        }
1062
    }
1063
        
1064
    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);
1065
}
1066

    
1067
/**
1068
 * gets the directionally predicted 16x8 MV.
1069
 * @param n the block index
1070
 * @param mx the x component of the predicted motion vector
1071
 * @param my the y component of the predicted motion vector
1072
 */
1073
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1074
    if(n==0){
1075
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1076
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1077

    
1078
        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);
1079
        
1080
        if(top_ref == ref){
1081
            *mx= B[0];
1082
            *my= B[1];
1083
            return;
1084
        }
1085
    }else{
1086
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1087
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1088
        
1089
        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);
1090

    
1091
        if(left_ref == ref){
1092
            *mx= A[0];
1093
            *my= A[1];
1094
            return;
1095
        }
1096
    }
1097

    
1098
    //RARE
1099
    pred_motion(h, n, 4, list, ref, mx, my);
1100
}
1101

    
1102
/**
1103
 * gets the directionally predicted 8x16 MV.
1104
 * @param n the block index
1105
 * @param mx the x component of the predicted motion vector
1106
 * @param my the y component of the predicted motion vector
1107
 */
1108
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1109
    if(n==0){
1110
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1111
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1112
        
1113
        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);
1114

    
1115
        if(left_ref == ref){
1116
            *mx= A[0];
1117
            *my= A[1];
1118
            return;
1119
        }
1120
    }else{
1121
        const int16_t * C;
1122
        int diagonal_ref;
1123

    
1124
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1125
        
1126
        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);
1127

    
1128
        if(diagonal_ref == ref){ 
1129
            *mx= C[0];
1130
            *my= C[1];
1131
            return;
1132
        }
1133
    }
1134

    
1135
    //RARE
1136
    pred_motion(h, n, 2, list, ref, mx, my);
1137
}
1138

    
1139
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1140
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1141
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1142

    
1143
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1144

    
1145
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1146
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1147
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1148
       
1149
        *mx = *my = 0;
1150
        return;
1151
    }
1152
        
1153
    pred_motion(h, 0, 4, 0, 0, mx, my);
1154

    
1155
    return;
1156
}
1157

    
1158
static inline void direct_dist_scale_factor(H264Context * const h){
1159
    const int poc = h->s.current_picture_ptr->poc;
1160
    const int poc1 = h->ref_list[1][0].poc;
1161
    int i;
1162
    for(i=0; i<h->ref_count[0]; i++){
1163
        int poc0 = h->ref_list[0][i].poc;
1164
        int td = clip(poc1 - poc0, -128, 127);
1165
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1166
            h->dist_scale_factor[i] = 256;
1167
        }else{
1168
            int tb = clip(poc - poc0, -128, 127);
1169
            int tx = (16384 + (ABS(td) >> 1)) / td;
1170
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1171
        }
1172
    }
1173
}
1174
static inline void direct_ref_list_init(H264Context * const h){
1175
    MpegEncContext * const s = &h->s;
1176
    Picture * const ref1 = &h->ref_list[1][0];
1177
    Picture * const cur = s->current_picture_ptr;
1178
    int list, i, j;
1179
    if(cur->pict_type == I_TYPE)
1180
        cur->ref_count[0] = 0;
1181
    if(cur->pict_type != B_TYPE)
1182
        cur->ref_count[1] = 0;
1183
    for(list=0; list<2; list++){
1184
        cur->ref_count[list] = h->ref_count[list];
1185
        for(j=0; j<h->ref_count[list]; j++)
1186
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1187
    }
1188
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1189
        return;
1190
    for(list=0; list<2; list++){
1191
        for(i=0; i<ref1->ref_count[list]; i++){
1192
            const int poc = ref1->ref_poc[list][i];
1193
            h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1194
            for(j=0; j<h->ref_count[list]; j++)
1195
                if(h->ref_list[list][j].poc == poc){
1196
                    h->map_col_to_list0[list][i] = j;
1197
                    break;
1198
                }
1199
        }
1200
    }
1201
}
1202

    
1203
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1204
    MpegEncContext * const s = &h->s;
1205
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1206
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1207
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1208
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1209
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1210
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1211
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1212
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1213
    const int is_b8x8 = IS_8X8(*mb_type);
1214
    int sub_mb_type;
1215
    int i8, i4;
1216

    
1217
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1218
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1219
         * so we know exactly what block size to use */
1220
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1221
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1222
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1223
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1224
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1225
    }else{
1226
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1227
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1228
    }
1229
    if(!is_b8x8)
1230
        *mb_type |= MB_TYPE_DIRECT2;
1231

    
1232
    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);
1233
    
1234
    if(h->direct_spatial_mv_pred){
1235
        int ref[2];
1236
        int mv[2][2];
1237
        int list;
1238

    
1239
        /* ref = min(neighbors) */
1240
        for(list=0; list<2; list++){
1241
            int refa = h->ref_cache[list][scan8[0] - 1];
1242
            int refb = h->ref_cache[list][scan8[0] - 8];
1243
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1244
            if(refc == -2)
1245
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1246
            ref[list] = refa;
1247
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1248
                ref[list] = refb;
1249
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1250
                ref[list] = refc;
1251
            if(ref[list] < 0)
1252
                ref[list] = -1;
1253
        }
1254

    
1255
        if(ref[0] < 0 && ref[1] < 0){
1256
            ref[0] = ref[1] = 0;
1257
            mv[0][0] = mv[0][1] =
1258
            mv[1][0] = mv[1][1] = 0;
1259
        }else{
1260
            for(list=0; list<2; list++){
1261
                if(ref[list] >= 0)
1262
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1263
                else
1264
                    mv[list][0] = mv[list][1] = 0;
1265
            }
1266
        }
1267

    
1268
        if(ref[1] < 0){
1269
            *mb_type &= ~MB_TYPE_P0L1;
1270
            sub_mb_type &= ~MB_TYPE_P0L1;
1271
        }else if(ref[0] < 0){
1272
            *mb_type &= ~MB_TYPE_P0L0;
1273
            sub_mb_type &= ~MB_TYPE_P0L0;
1274
        }
1275

    
1276
        if(IS_16X16(*mb_type)){
1277
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1278
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1279
            if(!IS_INTRA(mb_type_col) 
1280
               && (   (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1281
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1282
                       && (h->x264_build>33 || !h->x264_build)))){
1283
                if(ref[0] > 0)
1284
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1285
                else
1286
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1287
                if(ref[1] > 0)
1288
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1289
                else
1290
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1291
            }else{
1292
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1293
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1294
            }
1295
        }else{
1296
            for(i8=0; i8<4; i8++){
1297
                const int x8 = i8&1;
1298
                const int y8 = i8>>1;
1299
    
1300
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1301
                    continue;
1302
                h->sub_mb_type[i8] = sub_mb_type;
1303
    
1304
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1305
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1306
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1307
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1308
    
1309
                /* col_zero_flag */
1310
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0 
1311
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0 
1312
                                                  && (h->x264_build>33 || !h->x264_build)))){
1313
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1314
                    for(i4=0; i4<4; i4++){
1315
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1316
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1317
                            if(ref[0] == 0)
1318
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1319
                            if(ref[1] == 0)
1320
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1321
                        }
1322
                    }
1323
                }
1324
            }
1325
        }
1326
    }else{ /* direct temporal mv pred */
1327
        if(IS_16X16(*mb_type)){
1328
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1329
            if(IS_INTRA(mb_type_col)){
1330
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1331
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1332
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1333
            }else{
1334
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1335
                                                : h->map_col_to_list0[1][l1ref1[0]];
1336
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1337
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1338
                int mv_l0[2];
1339
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1340
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1341
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1342
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1343
                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);
1344
            }
1345
        }else{
1346
            for(i8=0; i8<4; i8++){
1347
                const int x8 = i8&1;
1348
                const int y8 = i8>>1;
1349
                int ref0, dist_scale_factor;
1350
                const int16_t (*l1mv)[2]= l1mv0;
1351

    
1352
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1353
                    continue;
1354
                h->sub_mb_type[i8] = sub_mb_type;
1355
                if(IS_INTRA(mb_type_col)){
1356
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1357
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1358
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1359
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1360
                    continue;
1361
                }
1362
    
1363
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1364
                if(ref0 >= 0)
1365
                    ref0 = h->map_col_to_list0[0][ref0];
1366
                else{
1367
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1368
                    l1mv= l1mv1;
1369
                }
1370
                dist_scale_factor = h->dist_scale_factor[ref0];
1371
    
1372
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1373
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1374
                for(i4=0; i4<4; i4++){
1375
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1376
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1377
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1378
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1379
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1380
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1381
                }
1382
            }
1383
        }
1384
    }
1385
}
1386

    
1387
static inline void write_back_motion(H264Context *h, int mb_type){
1388
    MpegEncContext * const s = &h->s;
1389
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1390
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1391
    int list;
1392

    
1393
    for(list=0; list<2; list++){
1394
        int y;
1395
        if(!USES_LIST(mb_type, list)){
1396
            if(1){ //FIXME skip or never read if mb_type doesn't use it
1397
                for(y=0; y<4; y++){
1398
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1399
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1400
                }
1401
                if( h->pps.cabac ) {
1402
                    /* FIXME needed ? */
1403
                    for(y=0; y<4; y++){
1404
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1405
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1406
                    }
1407
                }
1408
                for(y=0; y<2; y++){
1409
                    s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1410
                    s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1411
                }
1412
            }
1413
            continue;
1414
        }
1415
        
1416
        for(y=0; y<4; y++){
1417
            *(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];
1418
            *(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];
1419
        }
1420
        if( h->pps.cabac ) {
1421
            for(y=0; y<4; y++){
1422
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1423
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1424
            }
1425
        }
1426
        for(y=0; y<2; y++){
1427
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1428
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1429
        }
1430
    }
1431
    
1432
    if(h->slice_type == B_TYPE && h->pps.cabac){
1433
        if(IS_8X8(mb_type)){
1434
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1435
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1436
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1437
        }
1438
    }
1439
}
1440

    
1441
/**
1442
 * Decodes a network abstraction layer unit.
1443
 * @param consumed is the number of bytes used as input
1444
 * @param length is the length of the array
1445
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1446
 * @returns decoded bytes, might be src+1 if no escapes 
1447
 */
1448
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1449
    int i, si, di;
1450
    uint8_t *dst;
1451

    
1452
//    src[0]&0x80;                //forbidden bit
1453
    h->nal_ref_idc= src[0]>>5;
1454
    h->nal_unit_type= src[0]&0x1F;
1455

    
1456
    src++; length--;
1457
#if 0    
1458
    for(i=0; i<length; i++)
1459
        printf("%2X ", src[i]);
1460
#endif
1461
    for(i=0; i+1<length; i+=2){
1462
        if(src[i]) continue;
1463
        if(i>0 && src[i-1]==0) i--;
1464
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1465
            if(src[i+2]!=3){
1466
                /* startcode, so we must be past the end */
1467
                length=i;
1468
            }
1469
            break;
1470
        }
1471
    }
1472

    
1473
    if(i>=length-1){ //no escaped 0
1474
        *dst_length= length;
1475
        *consumed= length+1; //+1 for the header
1476
        return src; 
1477
    }
1478

    
1479
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1480
    dst= h->rbsp_buffer;
1481

    
1482
//printf("decoding esc\n");
1483
    si=di=0;
1484
    while(si<length){ 
1485
        //remove escapes (very rare 1:2^22)
1486
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1487
            if(src[si+2]==3){ //escape
1488
                dst[di++]= 0;
1489
                dst[di++]= 0;
1490
                si+=3;
1491
                continue;
1492
            }else //next start code
1493
                break;
1494
        }
1495

    
1496
        dst[di++]= src[si++];
1497
    }
1498

    
1499
    *dst_length= di;
1500
    *consumed= si + 1;//+1 for the header
1501
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1502
    return dst;
1503
}
1504

    
1505
#if 0
1506
/**
1507
 * @param src the data which should be escaped
1508
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1509
 * @param length the length of the src data
1510
 * @param dst_length the length of the dst array
1511
 * @returns length of escaped data in bytes or -1 if an error occured
1512
 */
1513
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1514
    int i, escape_count, si, di;
1515
    uint8_t *temp;
1516
    
1517
    assert(length>=0);
1518
    assert(dst_length>0);
1519
    
1520
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1521

1522
    if(length==0) return 1;
1523

1524
    escape_count= 0;
1525
    for(i=0; i<length; i+=2){
1526
        if(src[i]) continue;
1527
        if(i>0 && src[i-1]==0) 
1528
            i--;
1529
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1530
            escape_count++;
1531
            i+=2;
1532
        }
1533
    }
1534
    
1535
    if(escape_count==0){ 
1536
        if(dst+1 != src)
1537
            memcpy(dst+1, src, length);
1538
        return length + 1;
1539
    }
1540
    
1541
    if(length + escape_count + 1> dst_length)
1542
        return -1;
1543

1544
    //this should be damn rare (hopefully)
1545

1546
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1547
    temp= h->rbsp_buffer;
1548
//printf("encoding esc\n");
1549
    
1550
    si= 0;
1551
    di= 0;
1552
    while(si < length){
1553
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1554
            temp[di++]= 0; si++;
1555
            temp[di++]= 0; si++;
1556
            temp[di++]= 3; 
1557
            temp[di++]= src[si++];
1558
        }
1559
        else
1560
            temp[di++]= src[si++];
1561
    }
1562
    memcpy(dst+1, temp, length+escape_count);
1563
    
1564
    assert(di == length+escape_count);
1565
    
1566
    return di + 1;
1567
}
1568

1569
/**
1570
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1571
 */
1572
static void encode_rbsp_trailing(PutBitContext *pb){
1573
    int length;
1574
    put_bits(pb, 1, 1);
1575
    length= (-put_bits_count(pb))&7;
1576
    if(length) put_bits(pb, length, 0);
1577
}
1578
#endif
1579

    
1580
/**
1581
 * identifies the exact end of the bitstream
1582
 * @return the length of the trailing, or 0 if damaged
1583
 */
1584
static int decode_rbsp_trailing(uint8_t *src){
1585
    int v= *src;
1586
    int r;
1587

    
1588
    tprintf("rbsp trailing %X\n", v);
1589

    
1590
    for(r=1; r<9; r++){
1591
        if(v&1) return r;
1592
        v>>=1;
1593
    }
1594
    return 0;
1595
}
1596

    
1597
/**
1598
 * idct tranforms the 16 dc values and dequantize them.
1599
 * @param qp quantization parameter
1600
 */
1601
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1602
    const int qmul= dequant_coeff[qp][0];
1603
#define stride 16
1604
    int i;
1605
    int temp[16]; //FIXME check if this is a good idea
1606
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1607
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1608

    
1609
//memset(block, 64, 2*256);
1610
//return;
1611
    for(i=0; i<4; i++){
1612
        const int offset= y_offset[i];
1613
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1614
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1615
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1616
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1617

    
1618
        temp[4*i+0]= z0+z3;
1619
        temp[4*i+1]= z1+z2;
1620
        temp[4*i+2]= z1-z2;
1621
        temp[4*i+3]= z0-z3;
1622
    }
1623

    
1624
    for(i=0; i<4; i++){
1625
        const int offset= x_offset[i];
1626
        const int z0= temp[4*0+i] + temp[4*2+i];
1627
        const int z1= temp[4*0+i] - temp[4*2+i];
1628
        const int z2= temp[4*1+i] - temp[4*3+i];
1629
        const int z3= temp[4*1+i] + temp[4*3+i];
1630

    
1631
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1632
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1633
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1634
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1635
    }
1636
}
1637

    
1638
#if 0
1639
/**
1640
 * dct tranforms the 16 dc values.
1641
 * @param qp quantization parameter ??? FIXME
1642
 */
1643
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1644
//    const int qmul= dequant_coeff[qp][0];
1645
    int i;
1646
    int temp[16]; //FIXME check if this is a good idea
1647
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1648
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1649

1650
    for(i=0; i<4; i++){
1651
        const int offset= y_offset[i];
1652
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1653
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1654
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1655
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1656

1657
        temp[4*i+0]= z0+z3;
1658
        temp[4*i+1]= z1+z2;
1659
        temp[4*i+2]= z1-z2;
1660
        temp[4*i+3]= z0-z3;
1661
    }
1662

1663
    for(i=0; i<4; i++){
1664
        const int offset= x_offset[i];
1665
        const int z0= temp[4*0+i] + temp[4*2+i];
1666
        const int z1= temp[4*0+i] - temp[4*2+i];
1667
        const int z2= temp[4*1+i] - temp[4*3+i];
1668
        const int z3= temp[4*1+i] + temp[4*3+i];
1669

1670
        block[stride*0 +offset]= (z0 + z3)>>1;
1671
        block[stride*2 +offset]= (z1 + z2)>>1;
1672
        block[stride*8 +offset]= (z1 - z2)>>1;
1673
        block[stride*10+offset]= (z0 - z3)>>1;
1674
    }
1675
}
1676
#endif
1677

    
1678
#undef xStride
1679
#undef stride
1680

    
1681
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1682
    const int qmul= dequant_coeff[qp][0];
1683
    const int stride= 16*2;
1684
    const int xStride= 16;
1685
    int a,b,c,d,e;
1686

    
1687
    a= block[stride*0 + xStride*0];
1688
    b= block[stride*0 + xStride*1];
1689
    c= block[stride*1 + xStride*0];
1690
    d= block[stride*1 + xStride*1];
1691

    
1692
    e= a-b;
1693
    a= a+b;
1694
    b= c-d;
1695
    c= c+d;
1696

    
1697
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1698
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1699
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1700
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1701
}
1702

    
1703
#if 0
1704
static void chroma_dc_dct_c(DCTELEM *block){
1705
    const int stride= 16*2;
1706
    const int xStride= 16;
1707
    int a,b,c,d,e;
1708

1709
    a= block[stride*0 + xStride*0];
1710
    b= block[stride*0 + xStride*1];
1711
    c= block[stride*1 + xStride*0];
1712
    d= block[stride*1 + xStride*1];
1713

1714
    e= a-b;
1715
    a= a+b;
1716
    b= c-d;
1717
    c= c+d;
1718

1719
    block[stride*0 + xStride*0]= (a+c);
1720
    block[stride*0 + xStride*1]= (e+b);
1721
    block[stride*1 + xStride*0]= (a-c);
1722
    block[stride*1 + xStride*1]= (e-b);
1723
}
1724
#endif
1725

    
1726
/**
1727
 * gets the chroma qp.
1728
 */
1729
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1730
    
1731
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1732
}
1733

    
1734

    
1735
#if 0
1736
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1737
    int i;
1738
    //FIXME try int temp instead of block
1739
    
1740
    for(i=0; i<4; i++){
1741
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1742
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1743
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1744
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1745
        const int z0= d0 + d3;
1746
        const int z3= d0 - d3;
1747
        const int z1= d1 + d2;
1748
        const int z2= d1 - d2;
1749
        
1750
        block[0 + 4*i]=   z0 +   z1;
1751
        block[1 + 4*i]= 2*z3 +   z2;
1752
        block[2 + 4*i]=   z0 -   z1;
1753
        block[3 + 4*i]=   z3 - 2*z2;
1754
    }    
1755

1756
    for(i=0; i<4; i++){
1757
        const int z0= block[0*4 + i] + block[3*4 + i];
1758
        const int z3= block[0*4 + i] - block[3*4 + i];
1759
        const int z1= block[1*4 + i] + block[2*4 + i];
1760
        const int z2= block[1*4 + i] - block[2*4 + i];
1761
        
1762
        block[0*4 + i]=   z0 +   z1;
1763
        block[1*4 + i]= 2*z3 +   z2;
1764
        block[2*4 + i]=   z0 -   z1;
1765
        block[3*4 + i]=   z3 - 2*z2;
1766
    }
1767
}
1768
#endif
1769

    
1770
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1771
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1772
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1773
    int i;
1774
    const int * const quant_table= quant_coeff[qscale];
1775
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1776
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1777
    const unsigned int threshold2= (threshold1<<1);
1778
    int last_non_zero;
1779

    
1780
    if(seperate_dc){
1781
        if(qscale<=18){
1782
            //avoid overflows
1783
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1784
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1785
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1786

    
1787
            int level= block[0]*quant_coeff[qscale+18][0];
1788
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1789
                if(level>0){
1790
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1791
                    block[0]= level;
1792
                }else{
1793
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1794
                    block[0]= -level;
1795
                }
1796
//                last_non_zero = i;
1797
            }else{
1798
                block[0]=0;
1799
            }
1800
        }else{
1801
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1802
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1803
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1804

    
1805
            int level= block[0]*quant_table[0];
1806
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1807
                if(level>0){
1808
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1809
                    block[0]= level;
1810
                }else{
1811
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1812
                    block[0]= -level;
1813
                }
1814
//                last_non_zero = i;
1815
            }else{
1816
                block[0]=0;
1817
            }
1818
        }
1819
        last_non_zero= 0;
1820
        i=1;
1821
    }else{
1822
        last_non_zero= -1;
1823
        i=0;
1824
    }
1825

    
1826
    for(; i<16; i++){
1827
        const int j= scantable[i];
1828
        int level= block[j]*quant_table[j];
1829

    
1830
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1831
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1832
        if(((unsigned)(level+threshold1))>threshold2){
1833
            if(level>0){
1834
                level= (bias + level)>>QUANT_SHIFT;
1835
                block[j]= level;
1836
            }else{
1837
                level= (bias - level)>>QUANT_SHIFT;
1838
                block[j]= -level;
1839
            }
1840
            last_non_zero = i;
1841
        }else{
1842
            block[j]=0;
1843
        }
1844
    }
1845

    
1846
    return last_non_zero;
1847
}
1848

    
1849
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1850
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1851
    ((uint32_t*)(src+0*stride))[0]= a;
1852
    ((uint32_t*)(src+1*stride))[0]= a;
1853
    ((uint32_t*)(src+2*stride))[0]= a;
1854
    ((uint32_t*)(src+3*stride))[0]= a;
1855
}
1856

    
1857
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1858
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1859
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1860
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1861
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1862
}
1863

    
1864
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1865
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1866
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1867
    
1868
    ((uint32_t*)(src+0*stride))[0]= 
1869
    ((uint32_t*)(src+1*stride))[0]= 
1870
    ((uint32_t*)(src+2*stride))[0]= 
1871
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1872
}
1873

    
1874
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1875
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1876
    
1877
    ((uint32_t*)(src+0*stride))[0]= 
1878
    ((uint32_t*)(src+1*stride))[0]= 
1879
    ((uint32_t*)(src+2*stride))[0]= 
1880
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1881
}
1882

    
1883
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1884
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1885
    
1886
    ((uint32_t*)(src+0*stride))[0]= 
1887
    ((uint32_t*)(src+1*stride))[0]= 
1888
    ((uint32_t*)(src+2*stride))[0]= 
1889
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1890
}
1891

    
1892
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1893
    ((uint32_t*)(src+0*stride))[0]= 
1894
    ((uint32_t*)(src+1*stride))[0]= 
1895
    ((uint32_t*)(src+2*stride))[0]= 
1896
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1897
}
1898

    
1899

    
1900
#define LOAD_TOP_RIGHT_EDGE\
1901
    const int t4= topright[0];\
1902
    const int t5= topright[1];\
1903
    const int t6= topright[2];\
1904
    const int t7= topright[3];\
1905

    
1906
#define LOAD_LEFT_EDGE\
1907
    const int l0= src[-1+0*stride];\
1908
    const int l1= src[-1+1*stride];\
1909
    const int l2= src[-1+2*stride];\
1910
    const int l3= src[-1+3*stride];\
1911

    
1912
#define LOAD_TOP_EDGE\
1913
    const int t0= src[ 0-1*stride];\
1914
    const int t1= src[ 1-1*stride];\
1915
    const int t2= src[ 2-1*stride];\
1916
    const int t3= src[ 3-1*stride];\
1917

    
1918
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1919
    const int lt= src[-1-1*stride];
1920
    LOAD_TOP_EDGE
1921
    LOAD_LEFT_EDGE
1922

    
1923
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1924
    src[0+2*stride]=
1925
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1926
    src[0+1*stride]=
1927
    src[1+2*stride]=
1928
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1929
    src[0+0*stride]=
1930
    src[1+1*stride]=
1931
    src[2+2*stride]=
1932
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1933
    src[1+0*stride]=
1934
    src[2+1*stride]=
1935
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1936
    src[2+0*stride]=
1937
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1938
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1939
}
1940

    
1941
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1942
    LOAD_TOP_EDGE    
1943
    LOAD_TOP_RIGHT_EDGE    
1944
//    LOAD_LEFT_EDGE    
1945

    
1946
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1947
    src[1+0*stride]=
1948
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1949
    src[2+0*stride]=
1950
    src[1+1*stride]=
1951
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1952
    src[3+0*stride]=
1953
    src[2+1*stride]=
1954
    src[1+2*stride]=
1955
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1956
    src[3+1*stride]=
1957
    src[2+2*stride]=
1958
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1959
    src[3+2*stride]=
1960
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1961
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1962
}
1963

    
1964
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1965
    const int lt= src[-1-1*stride];
1966
    LOAD_TOP_EDGE    
1967
    LOAD_LEFT_EDGE    
1968
    const __attribute__((unused)) int unu= l3;
1969

    
1970
    src[0+0*stride]=
1971
    src[1+2*stride]=(lt + t0 + 1)>>1;
1972
    src[1+0*stride]=
1973
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1974
    src[2+0*stride]=
1975
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1976
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1977
    src[0+1*stride]=
1978
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1979
    src[1+1*stride]=
1980
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1981
    src[2+1*stride]=
1982
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1983
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1984
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1985
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1986
}
1987

    
1988
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1989
    LOAD_TOP_EDGE    
1990
    LOAD_TOP_RIGHT_EDGE    
1991
    const __attribute__((unused)) int unu= t7;
1992

    
1993
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1994
    src[1+0*stride]=
1995
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1996
    src[2+0*stride]=
1997
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1998
    src[3+0*stride]=
1999
    src[2+2*stride]=(t3 + t4+ 1)>>1;
2000
    src[3+2*stride]=(t4 + t5+ 1)>>1;
2001
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2002
    src[1+1*stride]=
2003
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2004
    src[2+1*stride]=
2005
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2006
    src[3+1*stride]=
2007
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2008
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2009
}
2010

    
2011
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2012
    LOAD_LEFT_EDGE    
2013

    
2014
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2015
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2016
    src[2+0*stride]=
2017
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2018
    src[3+0*stride]=
2019
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2020
    src[2+1*stride]=
2021
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2022
    src[3+1*stride]=
2023
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2024
    src[3+2*stride]=
2025
    src[1+3*stride]=
2026
    src[0+3*stride]=
2027
    src[2+2*stride]=
2028
    src[2+3*stride]=
2029
    src[3+3*stride]=l3;
2030
}
2031
    
2032
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2033
    const int lt= src[-1-1*stride];
2034
    LOAD_TOP_EDGE    
2035
    LOAD_LEFT_EDGE    
2036
    const __attribute__((unused)) int unu= t3;
2037

    
2038
    src[0+0*stride]=
2039
    src[2+1*stride]=(lt + l0 + 1)>>1;
2040
    src[1+0*stride]=
2041
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2042
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2043
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2044
    src[0+1*stride]=
2045
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2046
    src[1+1*stride]=
2047
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2048
    src[0+2*stride]=
2049
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2050
    src[1+2*stride]=
2051
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2052
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2053
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2054
}
2055

    
2056
static void pred16x16_vertical_c(uint8_t *src, int stride){
2057
    int i;
2058
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2059
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2060
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2061
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2062
    
2063
    for(i=0; i<16; i++){
2064
        ((uint32_t*)(src+i*stride))[0]= a;
2065
        ((uint32_t*)(src+i*stride))[1]= b;
2066
        ((uint32_t*)(src+i*stride))[2]= c;
2067
        ((uint32_t*)(src+i*stride))[3]= d;
2068
    }
2069
}
2070

    
2071
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2072
    int i;
2073

    
2074
    for(i=0; i<16; i++){
2075
        ((uint32_t*)(src+i*stride))[0]=
2076
        ((uint32_t*)(src+i*stride))[1]=
2077
        ((uint32_t*)(src+i*stride))[2]=
2078
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2079
    }
2080
}
2081

    
2082
static void pred16x16_dc_c(uint8_t *src, int stride){
2083
    int i, dc=0;
2084

    
2085
    for(i=0;i<16; i++){
2086
        dc+= src[-1+i*stride];
2087
    }
2088
    
2089
    for(i=0;i<16; i++){
2090
        dc+= src[i-stride];
2091
    }
2092

    
2093
    dc= 0x01010101*((dc + 16)>>5);
2094

    
2095
    for(i=0; i<16; i++){
2096
        ((uint32_t*)(src+i*stride))[0]=
2097
        ((uint32_t*)(src+i*stride))[1]=
2098
        ((uint32_t*)(src+i*stride))[2]=
2099
        ((uint32_t*)(src+i*stride))[3]= dc;
2100
    }
2101
}
2102

    
2103
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2104
    int i, dc=0;
2105

    
2106
    for(i=0;i<16; i++){
2107
        dc+= src[-1+i*stride];
2108
    }
2109
    
2110
    dc= 0x01010101*((dc + 8)>>4);
2111

    
2112
    for(i=0; i<16; i++){
2113
        ((uint32_t*)(src+i*stride))[0]=
2114
        ((uint32_t*)(src+i*stride))[1]=
2115
        ((uint32_t*)(src+i*stride))[2]=
2116
        ((uint32_t*)(src+i*stride))[3]= dc;
2117
    }
2118
}
2119

    
2120
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2121
    int i, dc=0;
2122

    
2123
    for(i=0;i<16; i++){
2124
        dc+= src[i-stride];
2125
    }
2126
    dc= 0x01010101*((dc + 8)>>4);
2127

    
2128
    for(i=0; i<16; i++){
2129
        ((uint32_t*)(src+i*stride))[0]=
2130
        ((uint32_t*)(src+i*stride))[1]=
2131
        ((uint32_t*)(src+i*stride))[2]=
2132
        ((uint32_t*)(src+i*stride))[3]= dc;
2133
    }
2134
}
2135

    
2136
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2137
    int i;
2138

    
2139
    for(i=0; i<16; i++){
2140
        ((uint32_t*)(src+i*stride))[0]=
2141
        ((uint32_t*)(src+i*stride))[1]=
2142
        ((uint32_t*)(src+i*stride))[2]=
2143
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2144
    }
2145
}
2146

    
2147
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2148
  int i, j, k;
2149
  int a;
2150
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2151
  const uint8_t * const src0 = src+7-stride;
2152
  const uint8_t *src1 = src+8*stride-1;
2153
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2154
  int H = src0[1] - src0[-1];
2155
  int V = src1[0] - src2[ 0];
2156
  for(k=2; k<=8; ++k) {
2157
    src1 += stride; src2 -= stride;
2158
    H += k*(src0[k] - src0[-k]);
2159
    V += k*(src1[0] - src2[ 0]);
2160
  }
2161
  if(svq3){
2162
    H = ( 5*(H/4) ) / 16;
2163
    V = ( 5*(V/4) ) / 16;
2164

    
2165
    /* required for 100% accuracy */
2166
    i = H; H = V; V = i;
2167
  }else{
2168
    H = ( 5*H+32 ) >> 6;
2169
    V = ( 5*V+32 ) >> 6;
2170
  }
2171

    
2172
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2173
  for(j=16; j>0; --j) {
2174
    int b = a;
2175
    a += V;
2176
    for(i=-16; i<0; i+=4) {
2177
      src[16+i] = cm[ (b    ) >> 5 ];
2178
      src[17+i] = cm[ (b+  H) >> 5 ];
2179
      src[18+i] = cm[ (b+2*H) >> 5 ];
2180
      src[19+i] = cm[ (b+3*H) >> 5 ];
2181
      b += 4*H;
2182
    }
2183
    src += stride;
2184
  }
2185
}
2186

    
2187
static void pred16x16_plane_c(uint8_t *src, int stride){
2188
    pred16x16_plane_compat_c(src, stride, 0);
2189
}
2190

    
2191
static void pred8x8_vertical_c(uint8_t *src, int stride){
2192
    int i;
2193
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2194
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2195
    
2196
    for(i=0; i<8; i++){
2197
        ((uint32_t*)(src+i*stride))[0]= a;
2198
        ((uint32_t*)(src+i*stride))[1]= b;
2199
    }
2200
}
2201

    
2202
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2203
    int i;
2204

    
2205
    for(i=0; i<8; i++){
2206
        ((uint32_t*)(src+i*stride))[0]=
2207
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2208
    }
2209
}
2210

    
2211
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2212
    int i;
2213

    
2214
    for(i=0; i<8; i++){
2215
        ((uint32_t*)(src+i*stride))[0]= 
2216
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2217
    }
2218
}
2219

    
2220
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2221
    int i;
2222
    int dc0, dc2;
2223

    
2224
    dc0=dc2=0;
2225
    for(i=0;i<4; i++){
2226
        dc0+= src[-1+i*stride];
2227
        dc2+= src[-1+(i+4)*stride];
2228
    }
2229
    dc0= 0x01010101*((dc0 + 2)>>2);
2230
    dc2= 0x01010101*((dc2 + 2)>>2);
2231

    
2232
    for(i=0; i<4; i++){
2233
        ((uint32_t*)(src+i*stride))[0]=
2234
        ((uint32_t*)(src+i*stride))[1]= dc0;
2235
    }
2236
    for(i=4; i<8; i++){
2237
        ((uint32_t*)(src+i*stride))[0]=
2238
        ((uint32_t*)(src+i*stride))[1]= dc2;
2239
    }
2240
}
2241

    
2242
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2243
    int i;
2244
    int dc0, dc1;
2245

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

    
2254
    for(i=0; i<4; i++){
2255
        ((uint32_t*)(src+i*stride))[0]= dc0;
2256
        ((uint32_t*)(src+i*stride))[1]= dc1;
2257
    }
2258
    for(i=4; i<8; i++){
2259
        ((uint32_t*)(src+i*stride))[0]= dc0;
2260
        ((uint32_t*)(src+i*stride))[1]= dc1;
2261
    }
2262
}
2263

    
2264

    
2265
static void pred8x8_dc_c(uint8_t *src, int stride){
2266
    int i;
2267
    int dc0, dc1, dc2, dc3;
2268

    
2269
    dc0=dc1=dc2=0;
2270
    for(i=0;i<4; i++){
2271
        dc0+= src[-1+i*stride] + src[i-stride];
2272
        dc1+= src[4+i-stride];
2273
        dc2+= src[-1+(i+4)*stride];
2274
    }
2275
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2276
    dc0= 0x01010101*((dc0 + 4)>>3);
2277
    dc1= 0x01010101*((dc1 + 2)>>2);
2278
    dc2= 0x01010101*((dc2 + 2)>>2);
2279

    
2280
    for(i=0; i<4; i++){
2281
        ((uint32_t*)(src+i*stride))[0]= dc0;
2282
        ((uint32_t*)(src+i*stride))[1]= dc1;
2283
    }
2284
    for(i=4; i<8; i++){
2285
        ((uint32_t*)(src+i*stride))[0]= dc2;
2286
        ((uint32_t*)(src+i*stride))[1]= dc3;
2287
    }
2288
}
2289

    
2290
static void pred8x8_plane_c(uint8_t *src, int stride){
2291
  int j, k;
2292
  int a;
2293
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2294
  const uint8_t * const src0 = src+3-stride;
2295
  const uint8_t *src1 = src+4*stride-1;
2296
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2297
  int H = src0[1] - src0[-1];
2298
  int V = src1[0] - src2[ 0];
2299
  for(k=2; k<=4; ++k) {
2300
    src1 += stride; src2 -= stride;
2301
    H += k*(src0[k] - src0[-k]);
2302
    V += k*(src1[0] - src2[ 0]);
2303
  }
2304
  H = ( 17*H+16 ) >> 5;
2305
  V = ( 17*V+16 ) >> 5;
2306

    
2307
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2308
  for(j=8; j>0; --j) {
2309
    int b = a;
2310
    a += V;
2311
    src[0] = cm[ (b    ) >> 5 ];
2312
    src[1] = cm[ (b+  H) >> 5 ];
2313
    src[2] = cm[ (b+2*H) >> 5 ];
2314
    src[3] = cm[ (b+3*H) >> 5 ];
2315
    src[4] = cm[ (b+4*H) >> 5 ];
2316
    src[5] = cm[ (b+5*H) >> 5 ];
2317
    src[6] = cm[ (b+6*H) >> 5 ];
2318
    src[7] = cm[ (b+7*H) >> 5 ];
2319
    src += stride;
2320
  }
2321
}
2322

    
2323
#define SRC(x,y) src[(x)+(y)*stride]
2324
#define PL(y) \
2325
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2326
#define PREDICT_8x8_LOAD_LEFT \
2327
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2328
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2329
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2330
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2331

    
2332
#define PT(x) \
2333
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2334
#define PREDICT_8x8_LOAD_TOP \
2335
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2336
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2337
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2338
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2339
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2340

    
2341
#define PTR(x) \
2342
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2343
#define PREDICT_8x8_LOAD_TOPRIGHT \
2344
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2345
    if(has_topright) { \
2346
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2347
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2348
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2349

    
2350
#define PREDICT_8x8_LOAD_TOPLEFT \
2351
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2352

    
2353
#define PREDICT_8x8_DC(v) \
2354
    int y; \
2355
    for( y = 0; y < 8; y++ ) { \
2356
        ((uint32_t*)src)[0] = \
2357
        ((uint32_t*)src)[1] = v; \
2358
        src += stride; \
2359
    }
2360

    
2361
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2362
{
2363
    PREDICT_8x8_DC(0x80808080);
2364
}
2365
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2366
{
2367
    PREDICT_8x8_LOAD_LEFT;
2368
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2369
    PREDICT_8x8_DC(dc);
2370
}
2371
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2372
{
2373
    PREDICT_8x8_LOAD_TOP;
2374
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2375
    PREDICT_8x8_DC(dc);
2376
}
2377
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2378
{
2379
    PREDICT_8x8_LOAD_LEFT;
2380
    PREDICT_8x8_LOAD_TOP;
2381
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2382
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2383
    PREDICT_8x8_DC(dc);
2384
}
2385
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2386
{
2387
    PREDICT_8x8_LOAD_LEFT;
2388
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2389
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2390
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2391
#undef ROW
2392
}
2393
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2394
{
2395
    int y;
2396
    PREDICT_8x8_LOAD_TOP;
2397
    src[0] = t0;
2398
    src[1] = t1;
2399
    src[2] = t2;
2400
    src[3] = t3;
2401
    src[4] = t4;
2402
    src[5] = t5;
2403
    src[6] = t6;
2404
    src[7] = t7;
2405
    for( y = 1; y < 8; y++ )
2406
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2407
}
2408
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2409
{
2410
    PREDICT_8x8_LOAD_TOP;
2411
    PREDICT_8x8_LOAD_TOPRIGHT;
2412
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2413
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2414
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2415
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2416
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2417
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2418
    SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;
2419
    SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;
2420
    SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;
2421
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2422
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2423
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2424
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2425
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2426
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2427
}
2428
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2429
{
2430
    PREDICT_8x8_LOAD_TOP;
2431
    PREDICT_8x8_LOAD_LEFT;
2432
    PREDICT_8x8_LOAD_TOPLEFT;
2433
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2434
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2435
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2436
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2437
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2438
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2439
    SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;
2440
    SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;
2441
    SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;
2442
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2443
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2444
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2445
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2446
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2447
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2448
  
2449
}
2450
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2451
{
2452
    PREDICT_8x8_LOAD_TOP;
2453
    PREDICT_8x8_LOAD_LEFT;
2454
    PREDICT_8x8_LOAD_TOPLEFT;
2455
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2456
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2457
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2458
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2459
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2460
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2461
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2462
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2463
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2464
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2465
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2466
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2467
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2468
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2469
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2470
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2471
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2472
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2473
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2474
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2475
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2476
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2477
}
2478
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2479
{
2480
    PREDICT_8x8_LOAD_TOP;
2481
    PREDICT_8x8_LOAD_LEFT;
2482
    PREDICT_8x8_LOAD_TOPLEFT;
2483
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2484
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2485
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2486
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2487
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2488
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2489
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2490
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2491
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2492
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2493
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2494
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2495
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2496
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2497
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2498
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2499
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2500
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2501
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2502
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2503
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2504
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2505
}
2506
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2507
{
2508
    PREDICT_8x8_LOAD_TOP;
2509
    PREDICT_8x8_LOAD_TOPRIGHT;
2510
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2511
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2512
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2513
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2514
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2515
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2516
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2517
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2518
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2519
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2520
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2521
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2522
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2523
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2524
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2525
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2526
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2527
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2528
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2529
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2530
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2531
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2532
}
2533
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2534
{
2535
    PREDICT_8x8_LOAD_LEFT;
2536
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2537
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2538
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2539
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2540
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2541
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2542
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2543
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2544
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2545
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2546
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2547
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2548
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2549
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2550
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2551
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2552
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2553
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2554
}
2555
#undef PREDICT_8x8_LOAD_LEFT
2556
#undef PREDICT_8x8_LOAD_TOP
2557
#undef PREDICT_8x8_LOAD_TOPLEFT
2558
#undef PREDICT_8x8_LOAD_TOPRIGHT
2559
#undef PREDICT_8x8_DC
2560
#undef PTR
2561
#undef PT
2562
#undef PL
2563
#undef SRC
2564

    
2565
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2566
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2567
                           int src_x_offset, int src_y_offset,
2568
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2569
    MpegEncContext * const s = &h->s;
2570
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2571
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2572
    const int luma_xy= (mx&3) + ((my&3)<<2);
2573
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2574
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2575
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2576
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2577
    int extra_height= extra_width;
2578
    int emu=0;
2579
    const int full_mx= mx>>2;
2580
    const int full_my= my>>2;
2581
    const int pic_width  = 16*s->mb_width;
2582
    const int pic_height = 16*s->mb_height;
2583
    
2584
    assert(pic->data[0]);
2585
    
2586
    if(mx&7) extra_width -= 3;
2587
    if(my&7) extra_height -= 3;
2588
    
2589
    if(   full_mx < 0-extra_width 
2590
       || full_my < 0-extra_height 
2591
       || full_mx + 16/*FIXME*/ > pic_width + extra_width 
2592
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2593
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
2594
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2595
        emu=1;
2596
    }
2597
    
2598
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2599
    if(!square){
2600
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2601
    }
2602
    
2603
    if(s->flags&CODEC_FLAG_GRAY) return;
2604
    
2605
    if(emu){
2606
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2607
            src_cb= s->edge_emu_buffer;
2608
    }
2609
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2610

    
2611
    if(emu){
2612
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2613
            src_cr= s->edge_emu_buffer;
2614
    }
2615
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2616
}
2617

    
2618
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2619
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2620
                           int x_offset, int y_offset,
2621
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2622
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2623
                           int list0, int list1){
2624
    MpegEncContext * const s = &h->s;
2625
    qpel_mc_func *qpix_op=  qpix_put;
2626
    h264_chroma_mc_func chroma_op= chroma_put;
2627
    
2628
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2629
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2630
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2631
    x_offset += 8*s->mb_x;
2632
    y_offset += 8*s->mb_y;
2633
    
2634
    if(list0){
2635
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2636
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2637
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2638
                           qpix_op, chroma_op);
2639

    
2640
        qpix_op=  qpix_avg;
2641
        chroma_op= chroma_avg;
2642
    }
2643

    
2644
    if(list1){
2645
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2646
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2647
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2648
                           qpix_op, chroma_op);
2649
    }
2650
}
2651

    
2652
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2653
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2654
                           int x_offset, int y_offset,
2655
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2656
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2657
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2658
                           int list0, int list1){
2659
    MpegEncContext * const s = &h->s;
2660

    
2661
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2662
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2663
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2664
    x_offset += 8*s->mb_x;
2665
    y_offset += 8*s->mb_y;
2666
    
2667
    if(list0 && list1){
2668
        /* don't optimize for luma-only case, since B-frames usually
2669
         * use implicit weights => chroma too. */
2670
        uint8_t *tmp_cb = s->obmc_scratchpad;
2671
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2672
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2673
        int refn0 = h->ref_cache[0][ scan8[n] ];
2674
        int refn1 = h->ref_cache[1][ scan8[n] ];
2675

    
2676
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2677
                    dest_y, dest_cb, dest_cr,
2678
                    x_offset, y_offset, qpix_put, chroma_put);
2679
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2680
                    tmp_y, tmp_cb, tmp_cr,
2681
                    x_offset, y_offset, qpix_put, chroma_put);
2682

    
2683
        if(h->use_weight == 2){
2684
            int weight0 = h->implicit_weight[refn0][refn1];
2685
            int weight1 = 64 - weight0;
2686
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2687
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2688
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2689
        }else{
2690
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2691
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2692
                            h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2693
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2694
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2695
                            h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2696
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2697
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2698
                            h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2699
        }
2700
    }else{
2701
        int list = list1 ? 1 : 0;
2702
        int refn = h->ref_cache[list][ scan8[n] ];
2703
        Picture *ref= &h->ref_list[list][refn];
2704
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2705
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2706
                    qpix_put, chroma_put);
2707

    
2708
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2709
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2710
        if(h->use_weight_chroma){
2711
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2712
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2713
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2714
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2715
        }
2716
    }
2717
}
2718

    
2719
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2720
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2721
                           int x_offset, int y_offset,
2722
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2723
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2724
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2725
                           int list0, int list1){
2726
    if((h->use_weight==2 && list0 && list1
2727
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2728
       || h->use_weight==1)
2729
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2730
                         x_offset, y_offset, qpix_put, chroma_put,
2731
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2732
    else
2733
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2734
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2735
}
2736

    
2737
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2738
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2739
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2740
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2741
    MpegEncContext * const s = &h->s;
2742
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2743
    const int mb_type= s->current_picture.mb_type[mb_xy];
2744
    
2745
    assert(IS_INTER(mb_type));
2746
    
2747
    if(IS_16X16(mb_type)){
2748
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2749
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2750
                &weight_op[0], &weight_avg[0],
2751
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2752
    }else if(IS_16X8(mb_type)){
2753
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2754
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2755
                &weight_op[1], &weight_avg[1],
2756
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2757
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2758
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2759
                &weight_op[1], &weight_avg[1],
2760
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2761
    }else if(IS_8X16(mb_type)){
2762
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2763
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2764
                &weight_op[2], &weight_avg[2],
2765
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2766
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2767
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2768
                &weight_op[2], &weight_avg[2],
2769
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2770
    }else{
2771
        int i;
2772
        
2773
        assert(IS_8X8(mb_type));
2774

    
2775
        for(i=0; i<4; i++){
2776
            const int sub_mb_type= h->sub_mb_type[i];
2777
            const int n= 4*i;
2778
            int x_offset= (i&1)<<2;
2779
            int y_offset= (i&2)<<1;
2780

    
2781
            if(IS_SUB_8X8(sub_mb_type)){
2782
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2783
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2784
                    &weight_op[3], &weight_avg[3],
2785
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2786
            }else if(IS_SUB_8X4(sub_mb_type)){
2787
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2788
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2789
                    &weight_op[4], &weight_avg[4],
2790
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2791
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2792
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2793
                    &weight_op[4], &weight_avg[4],
2794
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2795
            }else if(IS_SUB_4X8(sub_mb_type)){
2796
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2797
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2798
                    &weight_op[5], &weight_avg[5],
2799
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2800
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2801
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2802
                    &weight_op[5], &weight_avg[5],
2803
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2804
            }else{
2805
                int j;
2806
                assert(IS_SUB_4X4(sub_mb_type));
2807
                for(j=0; j<4; j++){
2808
                    int sub_x_offset= x_offset + 2*(j&1);
2809
                    int sub_y_offset= y_offset +   (j&2);
2810
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2811
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2812
                        &weight_op[6], &weight_avg[6],
2813
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2814
                }
2815
            }
2816
        }
2817
    }
2818
}
2819

    
2820
static void decode_init_vlc(H264Context *h){
2821
    static int done = 0;
2822

    
2823
    if (!done) {
2824
        int i;
2825
        done = 1;
2826

    
2827
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2828
                 &chroma_dc_coeff_token_len [0], 1, 1,
2829
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2830

    
2831
        for(i=0; i<4; i++){
2832
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2833
                     &coeff_token_len [i][0], 1, 1,
2834
                     &coeff_token_bits[i][0], 1, 1, 1);
2835
        }
2836

    
2837
        for(i=0; i<3; i++){
2838
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2839
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2840
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2841
        }
2842
        for(i=0; i<15; i++){
2843
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2844
                     &total_zeros_len [i][0], 1, 1,
2845
                     &total_zeros_bits[i][0], 1, 1, 1);
2846
        }
2847

    
2848
        for(i=0; i<6; i++){
2849
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2850
                     &run_len [i][0], 1, 1,
2851
                     &run_bits[i][0], 1, 1, 1);
2852
        }
2853
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2854
                 &run_len [6][0], 1, 1,
2855
                 &run_bits[6][0], 1, 1, 1);
2856
    }
2857
}
2858

    
2859
/**
2860
 * Sets the intra prediction function pointers.
2861
 */
2862
static void init_pred_ptrs(H264Context *h){
2863
//    MpegEncContext * const s = &h->s;
2864

    
2865
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2866
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2867
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2868
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2869
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2870
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2871
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2872
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2873
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2874
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2875
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2876
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2877

    
2878
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2879
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2880
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2881
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2882
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2883
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2884
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2885
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2886
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2887
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2888
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2889
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2890

    
2891
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2892
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2893
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2894
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2895
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2896
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2897
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2898

    
2899
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2900
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2901
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2902
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2903
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2904
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2905
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2906
}
2907

    
2908
static void free_tables(H264Context *h){
2909
    av_freep(&h->intra4x4_pred_mode);
2910
    av_freep(&h->chroma_pred_mode_table);
2911
    av_freep(&h->cbp_table);
2912
    av_freep(&h->mvd_table[0]);
2913
    av_freep(&h->mvd_table[1]);
2914
    av_freep(&h->direct_table);
2915
    av_freep(&h->non_zero_count);
2916
    av_freep(&h->slice_table_base);
2917
    av_freep(&h->top_borders[1]);
2918
    av_freep(&h->top_borders[0]);
2919
    h->slice_table= NULL;
2920

    
2921
    av_freep(&h->mb2b_xy);
2922
    av_freep(&h->mb2b8_xy);
2923

    
2924
    av_freep(&h->dequant4_coeff);
2925
    av_freep(&h->dequant8_coeff);
2926

    
2927
    av_freep(&h->s.obmc_scratchpad);
2928
}
2929

    
2930
/**
2931
 * allocates tables.
2932
 * needs width/height
2933
 */
2934
static int alloc_tables(H264Context *h){
2935
    MpegEncContext * const s = &h->s;
2936
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2937
    int x,y,q;
2938

    
2939
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2940

    
2941
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2942
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2943
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2944
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2945
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2946

    
2947
    if( h->pps.cabac ) {
2948
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2949
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2950
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2951
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2952
    }
2953

    
2954
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2955
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2956

    
2957
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2958
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2959
    for(y=0; y<s->mb_height; y++){
2960
        for(x=0; x<s->mb_width; x++){
2961
            const int mb_xy= x + y*s->mb_stride;
2962
            const int b_xy = 4*x + 4*y*h->b_stride;
2963
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2964
        
2965
            h->mb2b_xy [mb_xy]= b_xy;
2966
            h->mb2b8_xy[mb_xy]= b8_xy;
2967
        }
2968
    }
2969

    
2970
    CHECKED_ALLOCZ(h->dequant4_coeff, 52*16 * sizeof(uint16_t));
2971
    CHECKED_ALLOCZ(h->dequant8_coeff, 52*64 * sizeof(uint16_t));
2972
    memcpy(h->dequant4_coeff, dequant_coeff, 52*16 * sizeof(uint16_t));
2973
    for(q=0; q<52; q++){
2974
        int shift = div6[q];
2975
        int idx = rem6[q];
2976
        if(shift >= 2) // qp<12 are shifted during dequant
2977
            shift -= 2;
2978
        for(x=0; x<64; x++)
2979
            h->dequant8_coeff[q][x] = dequant8_coeff_init[idx][
2980
                dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] << shift;
2981
    }
2982
    if(h->sps.transform_bypass){
2983
        for(x=0; x<16; x++)
2984
            h->dequant4_coeff[0][x] = 1;
2985
        for(x=0; x<64; x++)
2986
            h->dequant8_coeff[0][x] = 1<<2;
2987
    }
2988

    
2989
    s->obmc_scratchpad = NULL;
2990

    
2991
    return 0;
2992
fail:
2993
    free_tables(h);
2994
    return -1;
2995
}
2996

    
2997
static void common_init(H264Context *h){
2998
    MpegEncContext * const s = &h->s;
2999

    
3000
    s->width = s->avctx->width;
3001
    s->height = s->avctx->height;
3002
    s->codec_id= s->avctx->codec->id;
3003
    
3004
    init_pred_ptrs(h);
3005

    
3006
    s->unrestricted_mv=1;
3007
    s->decode=1; //FIXME
3008
}
3009

    
3010
static int decode_init(AVCodecContext *avctx){
3011
    H264Context *h= avctx->priv_data;
3012
    MpegEncContext * const s = &h->s;
3013

    
3014
    MPV_decode_defaults(s);
3015
    
3016
    s->avctx = avctx;
3017
    common_init(h);
3018

    
3019
    s->out_format = FMT_H264;
3020
    s->workaround_bugs= avctx->workaround_bugs;
3021

    
3022
    // set defaults
3023
//    s->decode_mb= ff_h263_decode_mb;
3024
    s->low_delay= 1;
3025
    avctx->pix_fmt= PIX_FMT_YUV420P;
3026

    
3027
    decode_init_vlc(h);
3028
    
3029
    if(avctx->extradata_size > 0 && avctx->extradata &&
3030
       *(char *)avctx->extradata == 1){
3031
        h->is_avc = 1;
3032
        h->got_avcC = 0;
3033
    } else {
3034
        h->is_avc = 0;
3035
    }
3036

    
3037
    return 0;
3038
}
3039

    
3040
static void frame_start(H264Context *h){
3041
    MpegEncContext * const s = &h->s;
3042
    int i;
3043

    
3044
    MPV_frame_start(s, s->avctx);
3045
    ff_er_frame_start(s);
3046

    
3047
    assert(s->linesize && s->uvlinesize);
3048

    
3049
    for(i=0; i<16; i++){
3050
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3051
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3052
    }
3053
    for(i=0; i<4; i++){
3054
        h->block_offset[16+i]=
3055
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3056
        h->block_offset[24+16+i]=
3057
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3058
    }
3059

    
3060
    /* can't be in alloc_tables because linesize isn't known there.
3061
     * FIXME: redo bipred weight to not require extra buffer? */
3062
    if(!s->obmc_scratchpad)
3063
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3064

    
3065
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3066
}
3067

    
3068
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3069
    MpegEncContext * const s = &h->s;
3070
    int i;
3071
    
3072
    src_y  -=   linesize;
3073
    src_cb -= uvlinesize;
3074
    src_cr -= uvlinesize;
3075

    
3076
    // There are two lines saved, the line above the the top macroblock of a pair,
3077
    // and the line above the bottom macroblock
3078
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3079
    for(i=1; i<17; i++){
3080
        h->left_border[i]= src_y[15+i*  linesize];
3081
    }
3082
    
3083
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3084
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3085

    
3086
    if(!(s->flags&CODEC_FLAG_GRAY)){
3087
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3088
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3089
        for(i=1; i<9; i++){
3090
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3091
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3092
        }
3093
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3094
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3095
    }
3096
}
3097

    
3098
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){
3099
    MpegEncContext * const s = &h->s;
3100
    int temp8, i;
3101
    uint64_t temp64;
3102
    int deblock_left = (s->mb_x > 0);
3103
    int deblock_top  = (s->mb_y > 0);
3104

    
3105
    src_y  -=   linesize + 1;
3106
    src_cb -= uvlinesize + 1;
3107
    src_cr -= uvlinesize + 1;
3108

    
3109
#define XCHG(a,b,t,xchg)\
3110
t= a;\
3111
if(xchg)\
3112
    a= b;\
3113
b= t;
3114

    
3115
    if(deblock_left){
3116
        for(i = !deblock_top; i<17; i++){
3117
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3118
        }
3119
    }
3120

    
3121
    if(deblock_top){
3122
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3123
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3124
        if(s->mb_x+1 < s->mb_width){
3125
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3126
        }
3127
    }
3128

    
3129
    if(!(s->flags&CODEC_FLAG_GRAY)){
3130
        if(deblock_left){
3131
            for(i = !deblock_top; i<9; i++){
3132
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3133
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3134
            }
3135
        }
3136
        if(deblock_top){
3137
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3138
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3139
        }
3140
    }
3141
}
3142

    
3143
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3144
    MpegEncContext * const s = &h->s;
3145
    int i;
3146
    
3147
    src_y  -= 2 *   linesize;
3148
    src_cb -= 2 * uvlinesize;
3149
    src_cr -= 2 * uvlinesize;
3150

    
3151
    // There are two lines saved, the line above the the top macroblock of a pair,
3152
    // and the line above the bottom macroblock
3153
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3154
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3155
    for(i=2; i<34; i++){
3156
        h->left_border[i]= src_y[15+i*  linesize];
3157
    }
3158
    
3159
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3160
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3161
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3162
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3163

    
3164
    if(!(s->flags&CODEC_FLAG_GRAY)){
3165
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3166
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3167
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3168
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3169
        for(i=2; i<18; i++){
3170
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3171
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3172
        }
3173
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3174
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3175
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3176
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3177
    }
3178
}
3179

    
3180
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){
3181
    MpegEncContext * const s = &h->s;
3182
    int temp8, i;
3183
    uint64_t temp64;
3184
    int deblock_left = (s->mb_x > 0);
3185
    int deblock_top  = (s->mb_y > 0);
3186

    
3187
    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);
3188

    
3189
    src_y  -= 2 *   linesize + 1;
3190
    src_cb -= 2 * uvlinesize + 1;
3191
    src_cr -= 2 * uvlinesize + 1;
3192

    
3193
#define XCHG(a,b,t,xchg)\
3194
t= a;\
3195
if(xchg)\
3196
    a= b;\
3197
b= t;
3198

    
3199
    if(deblock_left){
3200
        for(i = (!deblock_top)<<1; i<34; i++){
3201
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3202
        }
3203
    }
3204

    
3205
    if(deblock_top){
3206
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3207
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3208
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3209
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3210
    }
3211

    
3212
    if(!(s->flags&CODEC_FLAG_GRAY)){
3213
        if(deblock_left){
3214
            for(i = (!deblock_top) << 1; i<18; i++){
3215
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3216
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3217
            }
3218
        }
3219
        if(deblock_top){
3220
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3221
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3222
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3223
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3224
        }
3225
    }
3226
}
3227

    
3228
static void hl_decode_mb(H264Context *h){
3229
    MpegEncContext * const s = &h->s;
3230
    const int mb_x= s->mb_x;
3231
    const int mb_y= s->mb_y;
3232
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3233
    const int mb_type= s->current_picture.mb_type[mb_xy];
3234
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3235
    int linesize, uvlinesize /*dct_offset*/;
3236
    int i;
3237
    int *block_offset = &h->block_offset[0];
3238
    const unsigned int bottom = mb_y & 1;
3239
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3240
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3241

    
3242
    if(!s->decode)
3243
        return;
3244

    
3245
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3246
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3247
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3248

    
3249
    if (h->mb_field_decoding_flag) {
3250
        linesize = s->linesize * 2;
3251
        uvlinesize = s->uvlinesize * 2;
3252
        block_offset = &h->block_offset[24];
3253
        if(mb_y&1){ //FIXME move out of this func?
3254
            dest_y -= s->linesize*15;
3255
            dest_cb-= s->uvlinesize*7;
3256
            dest_cr-= s->uvlinesize*7;
3257
        }
3258
    } else {
3259
        linesize = s->linesize;
3260
        uvlinesize = s->uvlinesize;
3261
//        dct_offset = s->linesize * 16;
3262
    }
3263
    
3264
    idct_add = transform_bypass
3265
             ? IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4
3266
             : IS_8x8DCT(mb_type) ? s->dsp.h264_idct8_add : s->dsp.h264_idct_add;
3267

    
3268
    if (IS_INTRA_PCM(mb_type)) {
3269
        unsigned int x, y;
3270

    
3271
        // The pixels are stored in h->mb array in the same order as levels,
3272
        // copy them in output in the correct order.
3273
        for(i=0; i<16; i++) {
3274
            for (y=0; y<4; y++) {
3275
                for (x=0; x<4; x++) {
3276
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3277
                }
3278
            }
3279
        }
3280
        for(i=16; i<16+4; i++) {
3281
            for (y=0; y<4; y++) {
3282
                for (x=0; x<4; x++) {
3283
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3284
                }
3285
            }
3286
        }
3287
        for(i=20; i<20+4; i++) {
3288
            for (y=0; y<4; y++) {
3289
                for (x=0; x<4; x++) {
3290
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3291
                }
3292
            }
3293
        }
3294
    } else {
3295
        if(IS_INTRA(mb_type)){
3296
            if(h->deblocking_filter) {
3297
                if (h->mb_aff_frame) {
3298
                    if (!bottom)
3299
                        xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3300
                } else {
3301
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3302
                }
3303
            }
3304

    
3305
            if(!(s->flags&CODEC_FLAG_GRAY)){
3306
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3307
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3308
            }
3309

    
3310
            if(IS_INTRA4x4(mb_type)){
3311
                if(!s->encoding){
3312
                    if(IS_8x8DCT(mb_type)){
3313
                        for(i=0; i<16; i+=4){
3314
                            uint8_t * const ptr= dest_y + block_offset[i];
3315
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3316
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3317
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3318
                            if(h->non_zero_count_cache[ scan8[i] ])
3319
                                idct_add(ptr, h->mb + i*16, linesize);
3320
                        }
3321
                    }else
3322
                    for(i=0; i<16; i++){
3323
                        uint8_t * const ptr= dest_y + block_offset[i];
3324
                        uint8_t *topright;
3325
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3326
                        int tr;
3327

    
3328
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3329
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3330
                            assert(mb_y || linesize <= block_offset[i]);
3331
                            if(!topright_avail){
3332
                                tr= ptr[3 - linesize]*0x01010101;
3333
                                topright= (uint8_t*) &tr;
3334
                            }else 
3335
                                topright= ptr + 4 - linesize;
3336
                        }else
3337
                            topright= NULL;
3338

    
3339
                        h->pred4x4[ dir ](ptr, topright, linesize);
3340
                        if(h->non_zero_count_cache[ scan8[i] ]){
3341
                            if(s->codec_id == CODEC_ID_H264)
3342
                                idct_add(ptr, h->mb + i*16, linesize);
3343
                            else
3344
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3345
                        }
3346
                    }
3347
                }
3348
            }else{
3349
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3350
                if(s->codec_id == CODEC_ID_H264){
3351
                    if(!transform_bypass)
3352
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3353
                }else
3354
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3355
            }
3356
            if(h->deblocking_filter) {
3357
                if (h->mb_aff_frame) {
3358
                    if (bottom) {
3359
                        uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
3360
                        uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3361
                        uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3362
                        s->mb_y--;
3363
                        xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3364
                        s->mb_y++;
3365
                    }
3366
                } else {
3367
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3368
                }
3369
            }
3370
        }else if(s->codec_id == CODEC_ID_H264){
3371
            hl_motion(h, dest_y, dest_cb, dest_cr,
3372
                      s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
3373
                      s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3374
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3375
        }
3376

    
3377

    
3378
        if(!IS_INTRA4x4(mb_type)){
3379
            if(s->codec_id == CODEC_ID_H264){
3380
                const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3381
                for(i=0; i<16; i+=di){
3382
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3383
                        uint8_t * const ptr= dest_y + block_offset[i];
3384
                        idct_add(ptr, h->mb + i*16, linesize);
3385
                    }
3386
                }
3387
            }else{
3388
                for(i=0; i<16; i++){
3389
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3390
                        uint8_t * const ptr= dest_y + block_offset[i];
3391
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3392
                    }
3393
                }
3394
            }
3395
        }
3396

    
3397
        if(!(s->flags&CODEC_FLAG_GRAY)){
3398
            idct_add = transform_bypass ? s->dsp.add_pixels4 : s->dsp.h264_idct_add;
3399
            if(!transform_bypass){
3400
                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3401
                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3402
            }
3403
            if(s->codec_id == CODEC_ID_H264){
3404
                for(i=16; i<16+4; i++){
3405
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3406
                        uint8_t * const ptr= dest_cb + block_offset[i];
3407
                        idct_add(ptr, h->mb + i*16, uvlinesize);
3408
                    }
3409
                }
3410
                for(i=20; i<20+4; i++){
3411
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3412
                        uint8_t * const ptr= dest_cr + block_offset[i];
3413
                        idct_add(ptr, h->mb + i*16, uvlinesize);
3414
                    }
3415
                }
3416
            }else{
3417
                for(i=16; i<16+4; i++){
3418
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3419
                        uint8_t * const ptr= dest_cb + block_offset[i];
3420
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3421
                    }
3422
                }
3423
                for(i=20; i<20+4; i++){
3424
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3425
                        uint8_t * const ptr= dest_cr + block_offset[i];
3426
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3427
                    }
3428
                }
3429
            }
3430
        }
3431
    }
3432
    if(h->deblocking_filter) {
3433
        if (h->mb_aff_frame) {
3434
            const int mb_y = s->mb_y - 1;
3435
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3436
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3437
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3438
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3439
            uint8_t tmp = s->current_picture.data[1][384];
3440
            if (!bottom) return;
3441
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3442
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3443
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3444

    
3445
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3446
            // TODO deblock a pair
3447
            // top 
3448
            s->mb_y--;
3449
            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);
3450
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3451
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3452
            if (tmp != s->current_picture.data[1][384]) {
3453
                tprintf("modified pixel 8,1 (1)\n");
3454
            }
3455
            // bottom
3456
            s->mb_y++;
3457
            tprintf("call mbaff filter_mb\n");
3458
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3459
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3460
            if (tmp != s->current_picture.data[1][384]) {
3461
                tprintf("modified pixel 8,1 (2)\n");
3462
            }
3463
        } else {
3464
            tprintf("call filter_mb\n");
3465
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3466
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3467
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3468
        }
3469
    }
3470
}
3471

    
3472
/**
3473
 * fills the default_ref_list.
3474
 */
3475
static int fill_default_ref_list(H264Context *h){
3476
    MpegEncContext * const s = &h->s;
3477
    int i;
3478
    int smallest_poc_greater_than_current = -1;
3479
    Picture sorted_short_ref[32];
3480
    
3481
    if(h->slice_type==B_TYPE){
3482
        int out_i;
3483
        int limit= INT_MIN;
3484

    
3485
        /* sort frame according to poc in B slice */
3486
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3487
            int best_i=INT_MIN;
3488
            int best_poc=INT_MAX;
3489

    
3490
            for(i=0; i<h->short_ref_count; i++){
3491
                const int poc= h->short_ref[i]->poc;
3492
                if(poc > limit && poc < best_poc){
3493
                    best_poc= poc;
3494
                    best_i= i;
3495
                }
3496
            }
3497
            
3498
            assert(best_i != INT_MIN);
3499
            
3500
            limit= best_poc;
3501
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3502
            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);
3503
            if (-1 == smallest_poc_greater_than_current) {
3504
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3505
                    smallest_poc_greater_than_current = out_i;
3506
                }
3507
            }
3508
        }
3509
    }
3510

    
3511
    if(s->picture_structure == PICT_FRAME){
3512
        if(h->slice_type==B_TYPE){
3513
            int list;
3514
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3515

    
3516
            // find the largest poc
3517
            for(list=0; list<2; list++){
3518
                int index = 0;
3519
                int j= -99;
3520
                int step= list ? -1 : 1;
3521

    
3522
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3523
                    while(j<0 || j>= h->short_ref_count){
3524
                        if(j != -99 && step == (list ? -1 : 1))
3525
                            return -1;
3526
                        step = -step;
3527
                        j= smallest_poc_greater_than_current + (step>>1);
3528
                    }
3529
                    if(sorted_short_ref[j].reference != 3) continue;
3530
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3531
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3532
                }
3533

    
3534
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3535
                    if(h->long_ref[i] == NULL) continue;
3536
                    if(h->long_ref[i]->reference != 3) continue;
3537

    
3538
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3539
                    h->default_ref_list[ list ][index++].pic_id= i;;
3540
                }
3541
                
3542
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3543
                    // swap the two first elements of L1 when
3544
                    // L0 and L1 are identical
3545
                    Picture temp= h->default_ref_list[1][0];
3546
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3547
                    h->default_ref_list[1][1] = temp;
3548
                }
3549

    
3550
                if(index < h->ref_count[ list ])
3551
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3552
            }
3553
        }else{
3554
            int index=0;
3555
            for(i=0; i<h->short_ref_count; i++){
3556
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3557
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3558
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3559
            }
3560
            for(i = 0; i < 16; i++){
3561
                if(h->long_ref[i] == NULL) continue;
3562
                if(h->long_ref[i]->reference != 3) continue;
3563
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3564
                h->default_ref_list[0][index++].pic_id= i;;
3565
            }
3566
            if(index < h->ref_count[0])
3567
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3568
        }
3569
    }else{ //FIELD
3570
        if(h->slice_type==B_TYPE){
3571
        }else{
3572
            //FIXME second field balh
3573
        }
3574
    }
3575
#ifdef TRACE
3576
    for (i=0; i<h->ref_count[0]; i++) {
3577
        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]);
3578
    }
3579
    if(h->slice_type==B_TYPE){
3580
        for (i=0; i<h->ref_count[1]; i++) {
3581
            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]);
3582
        }
3583
    }
3584
#endif
3585
    return 0;
3586
}
3587

    
3588
static void print_short_term(H264Context *h);
3589
static void print_long_term(H264Context *h);
3590

    
3591
static int decode_ref_pic_list_reordering(H264Context *h){
3592
    MpegEncContext * const s = &h->s;
3593
    int list, index;
3594
    
3595
    print_short_term(h);
3596
    print_long_term(h);
3597
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3598
    
3599
    for(list=0; list<2; list++){
3600
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3601

    
3602
        if(get_bits1(&s->gb)){
3603
            int pred= h->curr_pic_num;
3604

    
3605
            for(index=0; ; index++){
3606
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3607
                int pic_id;
3608
                int i;
3609
                Picture *ref = NULL;
3610
                
3611
                if(reordering_of_pic_nums_idc==3) 
3612
                    break;
3613
                
3614
                if(index >= h->ref_count[list]){
3615
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3616
                    return -1;
3617
                }
3618
                
3619
                if(reordering_of_pic_nums_idc<3){
3620
                    if(reordering_of_pic_nums_idc<2){
3621
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3622

    
3623
                        if(abs_diff_pic_num >= h->max_pic_num){
3624
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3625
                            return -1;
3626
                        }
3627

    
3628
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3629
                        else                                pred+= abs_diff_pic_num;
3630
                        pred &= h->max_pic_num - 1;
3631
                    
3632
                        for(i= h->short_ref_count-1; i>=0; i--){
3633
                            ref = h->short_ref[i];
3634
                            assert(ref->reference == 3);
3635
                            assert(!ref->long_ref);
3636
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3637
                                break;
3638
                        }
3639
                        if(i>=0)
3640
                            ref->pic_id= ref->frame_num;
3641
                    }else{
3642
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3643
                        ref = h->long_ref[pic_id];
3644
                        ref->pic_id= pic_id;
3645
                        assert(ref->reference == 3);
3646
                        assert(ref->long_ref);
3647
                        i=0;
3648
                    }
3649

    
3650
                    if (i < 0) {
3651
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3652
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3653
                    } else {
3654
                        for(i=index; i+1<h->ref_count[list]; i++){
3655
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3656
                                break;
3657
                        }
3658
                        for(; i > index; i--){
3659
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3660
                        }
3661
                        h->ref_list[list][index]= *ref;
3662
                    }
3663
                }else{
3664
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3665
                    return -1;
3666
                }
3667
            }
3668
        }
3669

    
3670
        if(h->slice_type!=B_TYPE) break;
3671
    }
3672
    for(list=0; list<2; list++){
3673
        for(index= 0; index < h->ref_count[list]; index++){
3674
            if(!h->ref_list[list][index].data[0])
3675
                h->ref_list[list][index]= s->current_picture;
3676
        }
3677
        if(h->slice_type!=B_TYPE) break;
3678
    }
3679
    
3680
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3681
        direct_dist_scale_factor(h);
3682
    direct_ref_list_init(h);
3683
    return 0;    
3684
}
3685

    
3686
static int pred_weight_table(H264Context *h){
3687
    MpegEncContext * const s = &h->s;
3688
    int list, i;
3689
    int luma_def, chroma_def;
3690
    
3691
    h->use_weight= 0;
3692
    h->use_weight_chroma= 0;
3693
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3694
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3695
    luma_def = 1<<h->luma_log2_weight_denom;
3696
    chroma_def = 1<<h->chroma_log2_weight_denom;
3697

    
3698
    for(list=0; list<2; list++){
3699
        for(i=0; i<h->ref_count[list]; i++){
3700
            int luma_weight_flag, chroma_weight_flag;
3701
            
3702
            luma_weight_flag= get_bits1(&s->gb);
3703
            if(luma_weight_flag){
3704
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3705
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3706
                if(   h->luma_weight[list][i] != luma_def
3707
                   || h->luma_offset[list][i] != 0)
3708
                    h->use_weight= 1;
3709
            }else{
3710
                h->luma_weight[list][i]= luma_def;
3711
                h->luma_offset[list][i]= 0;
3712
            }
3713

    
3714
            chroma_weight_flag= get_bits1(&s->gb);
3715
            if(chroma_weight_flag){
3716
                int j;
3717
                for(j=0; j<2; j++){
3718
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3719
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3720
                    if(   h->chroma_weight[list][i][j] != chroma_def
3721
                       || h->chroma_offset[list][i][j] != 0)
3722
                        h->use_weight_chroma= 1;
3723
                }
3724
            }else{
3725
                int j;
3726
                for(j=0; j<2; j++){
3727
                    h->chroma_weight[list][i][j]= chroma_def;
3728
                    h->chroma_offset[list][i][j]= 0;
3729
                }
3730
            }
3731
        }
3732
        if(h->slice_type != B_TYPE) break;
3733
    }
3734
    h->use_weight= h->use_weight || h->use_weight_chroma;
3735
    return 0;
3736
}
3737

    
3738
static void implicit_weight_table(H264Context *h){
3739
    MpegEncContext * const s = &h->s;
3740
    int ref0, ref1;
3741
    int cur_poc = s->current_picture_ptr->poc;
3742

    
3743
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3744
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3745
        h->use_weight= 0;
3746
        h->use_weight_chroma= 0;
3747
        return;
3748
    }
3749

    
3750
    h->use_weight= 2;
3751
    h->use_weight_chroma= 2;
3752
    h->luma_log2_weight_denom= 5;
3753
    h->chroma_log2_weight_denom= 5;
3754

    
3755
    /* FIXME: MBAFF */
3756
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3757
        int poc0 = h->ref_list[0][ref0].poc;
3758
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3759
            int poc1 = h->ref_list[1][ref1].poc;
3760
            int td = clip(poc1 - poc0, -128, 127);
3761
            if(td){
3762
                int tb = clip(cur_poc - poc0, -128, 127);
3763
                int tx = (16384 + (ABS(td) >> 1)) / td;
3764
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3765
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3766
                    h->implicit_weight[ref0][ref1] = 32;
3767
                else
3768
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3769
            }else
3770
                h->implicit_weight[ref0][ref1] = 32;
3771
        }
3772
    }
3773
}
3774

    
3775
static inline void unreference_pic(H264Context *h, Picture *pic){
3776
    int i;
3777
    pic->reference=0;
3778
    if(pic == h->delayed_output_pic)
3779
        pic->reference=1;
3780
    else{
3781
        for(i = 0; h->delayed_pic[i]; i++)
3782
            if(pic == h->delayed_pic[i]){
3783
                pic->reference=1;
3784
                break;
3785
            }
3786
    }
3787
}
3788

    
3789
/**
3790
 * instantaneous decoder refresh.
3791
 */
3792
static void idr(H264Context *h){
3793
    int i;
3794

    
3795
    for(i=0; i<16; i++){
3796
        if (h->long_ref[i] != NULL) {
3797
            unreference_pic(h, h->long_ref[i]);
3798
            h->long_ref[i]= NULL;
3799
        }
3800
    }
3801
    h->long_ref_count=0;
3802

    
3803
    for(i=0; i<h->short_ref_count; i++){
3804
        unreference_pic(h, h->short_ref[i]);
3805
        h->short_ref[i]= NULL;
3806
    }
3807
    h->short_ref_count=0;
3808
}
3809

    
3810
/* forget old pics after a seek */
3811
static void flush_dpb(AVCodecContext *avctx){
3812
    H264Context *h= avctx->priv_data;
3813
    int i;
3814
    for(i=0; i<16; i++)
3815
        h->delayed_pic[i]= NULL;
3816
    h->delayed_output_pic= NULL;
3817
    idr(h);
3818
    if(h->s.current_picture_ptr)
3819
        h->s.current_picture_ptr->reference= 0;
3820
}
3821

    
3822
/**
3823
 *
3824
 * @return the removed picture or NULL if an error occurs
3825
 */
3826
static Picture * remove_short(H264Context *h, int frame_num){
3827
    MpegEncContext * const s = &h->s;
3828
    int i;
3829
    
3830
    if(s->avctx->debug&FF_DEBUG_MMCO)
3831
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3832
    
3833
    for(i=0; i<h->short_ref_count; i++){
3834
        Picture *pic= h->short_ref[i];
3835
        if(s->avctx->debug&FF_DEBUG_MMCO)
3836
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3837
        if(pic->frame_num == frame_num){
3838
            h->short_ref[i]= NULL;
3839
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3840
            h->short_ref_count--;
3841
            return pic;
3842
        }
3843
    }
3844
    return NULL;
3845
}
3846

    
3847
/**
3848
 *
3849
 * @return the removed picture or NULL if an error occurs
3850
 */
3851
static Picture * remove_long(H264Context *h, int i){
3852
    Picture *pic;
3853

    
3854
    pic= h->long_ref[i];
3855
    h->long_ref[i]= NULL;
3856
    if(pic) h->long_ref_count--;
3857

    
3858
    return pic;
3859
}
3860

    
3861
/**
3862
 * print short term list
3863
 */
3864
static void print_short_term(H264Context *h) {
3865
    uint32_t i;
3866
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3867
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3868
        for(i=0; i<h->short_ref_count; i++){
3869
            Picture *pic= h->short_ref[i];
3870
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3871
        }
3872
    }
3873
}
3874

    
3875
/**
3876
 * print long term list
3877
 */
3878
static void print_long_term(H264Context *h) {
3879
    uint32_t i;
3880
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3881
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3882
        for(i = 0; i < 16; i++){
3883
            Picture *pic= h->long_ref[i];
3884
            if (pic) {
3885
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3886
            }
3887
        }
3888
    }
3889
}
3890

    
3891
/**
3892
 * Executes the reference picture marking (memory management control operations).
3893
 */
3894
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3895
    MpegEncContext * const s = &h->s;
3896
    int i, j;
3897
    int current_is_long=0;
3898
    Picture *pic;
3899
    
3900
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3901
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3902
        
3903
    for(i=0; i<mmco_count; i++){
3904
        if(s->avctx->debug&FF_DEBUG_MMCO)
3905
            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);
3906

    
3907
        switch(mmco[i].opcode){
3908
        case MMCO_SHORT2UNUSED:
3909
            pic= remove_short(h, mmco[i].short_frame_num);
3910
            if(pic)
3911
                unreference_pic(h, pic);
3912
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3913
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
3914
            break;
3915
        case MMCO_SHORT2LONG:
3916
            pic= remove_long(h, mmco[i].long_index);
3917
            if(pic) unreference_pic(h, pic);
3918
            
3919
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3920
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3921
            h->long_ref_count++;
3922
            break;
3923
        case MMCO_LONG2UNUSED:
3924
            pic= remove_long(h, mmco[i].long_index);
3925
            if(pic)
3926
                unreference_pic(h, pic);
3927
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3928
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
3929
            break;
3930
        case MMCO_LONG:
3931
            pic= remove_long(h, mmco[i].long_index);
3932
            if(pic) unreference_pic(h, pic);
3933
            
3934
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3935
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3936
            h->long_ref_count++;
3937
            
3938
            current_is_long=1;
3939
            break;
3940
        case MMCO_SET_MAX_LONG:
3941
            assert(mmco[i].long_index <= 16);
3942
            // just remove the long term which index is greater than new max
3943
            for(j = mmco[i].long_index; j<16; j++){
3944
                pic = remove_long(h, j);
3945
                if (pic) unreference_pic(h, pic);
3946
            }
3947
            break;
3948
        case MMCO_RESET:
3949
            while(h->short_ref_count){
3950
                pic= remove_short(h, h->short_ref[0]->frame_num);
3951
                unreference_pic(h, pic);
3952
            }
3953
            for(j = 0; j < 16; j++) {
3954
                pic= remove_long(h, j);
3955
                if(pic) unreference_pic(h, pic);
3956
            }
3957
            break;
3958
        default: assert(0);
3959
        }
3960
    }
3961
    
3962
    if(!current_is_long){
3963
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3964
        if(pic){
3965
            unreference_pic(h, pic);
3966
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3967
        }
3968
        
3969
        if(h->short_ref_count)
3970
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3971

    
3972
        h->short_ref[0]= s->current_picture_ptr;
3973
        h->short_ref[0]->long_ref=0;
3974
        h->short_ref_count++;
3975
    }
3976
    
3977
    print_short_term(h);
3978
    print_long_term(h);
3979
    return 0; 
3980
}
3981

    
3982
static int decode_ref_pic_marking(H264Context *h){
3983
    MpegEncContext * const s = &h->s;
3984
    int i;
3985
    
3986
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3987
        s->broken_link= get_bits1(&s->gb) -1;
3988
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3989
        if(h->mmco[0].long_index == -1)
3990
            h->mmco_index= 0;
3991
        else{
3992
            h->mmco[0].opcode= MMCO_LONG;
3993
            h->mmco_index= 1;
3994
        } 
3995
    }else{
3996
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3997
            for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3998
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3999

    
4000
                h->mmco[i].opcode= opcode;
4001
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
4002
                    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
4003
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
4004
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
4005
                        return -1;
4006
                    }*/
4007
                }
4008
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
4009
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
4010
                    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){
4011
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
4012
                        return -1;
4013
                    }
4014
                }
4015
                    
4016
                if(opcode > MMCO_LONG){
4017
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4018
                    return -1;
4019
                }
4020
                if(opcode == MMCO_END)
4021
                    break;
4022
            }
4023
            h->mmco_index= i;
4024
        }else{
4025
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4026

    
4027
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4028
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4029
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4030
                h->mmco_index= 1;
4031
            }else
4032
                h->mmco_index= 0;
4033
        }
4034
    }
4035
    
4036
    return 0; 
4037
}
4038

    
4039
static int init_poc(H264Context *h){
4040
    MpegEncContext * const s = &h->s;
4041
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4042
    int field_poc[2];
4043

    
4044
    if(h->nal_unit_type == NAL_IDR_SLICE){
4045
        h->frame_num_offset= 0;
4046
    }else{
4047
        if(h->frame_num < h->prev_frame_num)
4048
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4049
        else
4050
            h->frame_num_offset= h->prev_frame_num_offset;
4051
    }
4052

    
4053
    if(h->sps.poc_type==0){
4054
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4055

    
4056
        if(h->nal_unit_type == NAL_IDR_SLICE){
4057
             h->prev_poc_msb=
4058
             h->prev_poc_lsb= 0;
4059
        }
4060

    
4061
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4062
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4063
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4064
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4065
        else
4066
            h->poc_msb = h->prev_poc_msb;
4067
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4068
        field_poc[0] = 
4069
        field_poc[1] = h->poc_msb + h->poc_lsb;
4070
        if(s->picture_structure == PICT_FRAME) 
4071
            field_poc[1] += h->delta_poc_bottom;
4072
    }else if(h->sps.poc_type==1){
4073
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4074
        int i;
4075

    
4076
        if(h->sps.poc_cycle_length != 0)
4077
            abs_frame_num = h->frame_num_offset + h->frame_num;
4078
        else
4079
            abs_frame_num = 0;
4080

    
4081
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4082
            abs_frame_num--;
4083
            
4084
        expected_delta_per_poc_cycle = 0;
4085
        for(i=0; i < h->sps.poc_cycle_length; i++)
4086
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4087

    
4088
        if(abs_frame_num > 0){
4089
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4090
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4091

    
4092
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4093
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4094
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4095
        } else
4096
            expectedpoc = 0;
4097

    
4098
        if(h->nal_ref_idc == 0) 
4099
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4100
        
4101
        field_poc[0] = expectedpoc + h->delta_poc[0];
4102
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4103

    
4104
        if(s->picture_structure == PICT_FRAME)
4105
            field_poc[1] += h->delta_poc[1];
4106
    }else{
4107
        int poc;
4108
        if(h->nal_unit_type == NAL_IDR_SLICE){
4109
            poc= 0;
4110
        }else{
4111
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4112
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4113
        }
4114
        field_poc[0]= poc;
4115
        field_poc[1]= poc;
4116
    }
4117
    
4118
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4119
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4120
    if(s->picture_structure != PICT_TOP_FIELD)
4121
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4122
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4123
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4124

    
4125
    return 0;
4126
}
4127

    
4128
/**
4129
 * decodes a slice header.
4130
 * this will allso call MPV_common_init() and frame_start() as needed
4131
 */
4132
static int decode_slice_header(H264Context *h){
4133
    MpegEncContext * const s = &h->s;
4134
    int first_mb_in_slice, pps_id;
4135
    int num_ref_idx_active_override_flag;
4136
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4137
    int slice_type;
4138
    int default_ref_list_done = 0;
4139

    
4140
    s->current_picture.reference= h->nal_ref_idc != 0;
4141
    s->dropable= h->nal_ref_idc == 0;
4142

    
4143
    first_mb_in_slice= get_ue_golomb(&s->gb);
4144

    
4145
    slice_type= get_ue_golomb(&s->gb);
4146
    if(slice_type > 9){
4147
        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);
4148
        return -1;
4149
    }
4150
    if(slice_type > 4){
4151
        slice_type -= 5;
4152
        h->slice_type_fixed=1;
4153
    }else
4154
        h->slice_type_fixed=0;
4155
    
4156
    slice_type= slice_type_map[ slice_type ];
4157
    if (slice_type == I_TYPE
4158
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4159
        default_ref_list_done = 1;
4160
    }
4161
    h->slice_type= slice_type;
4162

    
4163
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4164
        
4165
    pps_id= get_ue_golomb(&s->gb);
4166
    if(pps_id>255){
4167
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4168
        return -1;
4169
    }
4170
    h->pps= h->pps_buffer[pps_id];
4171
    if(h->pps.slice_group_count == 0){
4172
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4173
        return -1;
4174
    }
4175

    
4176
    h->sps= h->sps_buffer[ h->pps.sps_id ];
4177
    if(h->sps.log2_max_frame_num == 0){
4178
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4179
        return -1;
4180
    }
4181
    
4182
    s->mb_width= h->sps.mb_width;
4183
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
4184
    
4185
    h->b_stride=  s->mb_width*4 + 1;
4186
    h->b8_stride= s->mb_width*2 + 1;
4187

    
4188
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
4189
    if(h->sps.frame_mbs_only_flag)
4190
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
4191
    else
4192
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
4193
    
4194
    if (s->context_initialized 
4195
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
4196
        free_tables(h);
4197
        MPV_common_end(s);
4198
    }
4199
    if (!s->context_initialized) {
4200
        if (MPV_common_init(s) < 0)
4201
            return -1;
4202
            
4203
        if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
4204
            memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
4205
            memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
4206
        }else{
4207
            int i;
4208
            for(i=0; i<16; i++){
4209
#define T(x) (x>>2) | ((x<<2) & 0xF)
4210
                h->zigzag_scan[i] = T(zigzag_scan[i]);
4211
                h-> field_scan[i] = T( field_scan[i]);
4212
            }
4213
        }
4214
        if(h->sps.transform_bypass){ //FIXME same ugly
4215
            h->zigzag_scan_q0 = zigzag_scan;
4216
            h->field_scan_q0 = field_scan;
4217
        }else{
4218
            h->zigzag_scan_q0 = h->zigzag_scan;
4219
            h->field_scan_q0 = h->field_scan;
4220
        }
4221

    
4222
        alloc_tables(h);
4223

    
4224
        s->avctx->width = s->width;
4225
        s->avctx->height = s->height;
4226
        s->avctx->sample_aspect_ratio= h->sps.sar;
4227
        if(!s->avctx->sample_aspect_ratio.den)
4228
            s->avctx->sample_aspect_ratio.den = 1;
4229

    
4230
        if(h->sps.timing_info_present_flag){
4231
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
4232
        }
4233
    }
4234

    
4235
    if(h->slice_num == 0){
4236
        frame_start(h);
4237
    }
4238

    
4239
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
4240
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
4241

    
4242
    h->mb_aff_frame = 0;
4243
    if(h->sps.frame_mbs_only_flag){
4244
        s->picture_structure= PICT_FRAME;
4245
    }else{
4246
        if(get_bits1(&s->gb)) { //field_pic_flag
4247
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
4248
        } else {
4249
            s->picture_structure= PICT_FRAME;
4250
            first_mb_in_slice <<= h->sps.mb_aff;
4251
            h->mb_aff_frame = h->sps.mb_aff;
4252
        }
4253
    }
4254

    
4255
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
4256
    s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
4257
    if(s->mb_y >= s->mb_height){
4258
        return -1;
4259
    }
4260
    
4261
    if(s->picture_structure==PICT_FRAME){
4262
        h->curr_pic_num=   h->frame_num;
4263
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
4264
    }else{
4265
        h->curr_pic_num= 2*h->frame_num;
4266
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
4267
    }
4268
        
4269
    if(h->nal_unit_type == NAL_IDR_SLICE){
4270
        get_ue_golomb(&s->gb); /* idr_pic_id */
4271
    }
4272
   
4273
    if(h->sps.poc_type==0){
4274
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
4275
        
4276
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
4277
            h->delta_poc_bottom= get_se_golomb(&s->gb);
4278
        }
4279
    }
4280
    
4281
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
4282
        h->delta_poc[0]= get_se_golomb(&s->gb);
4283
        
4284
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
4285
            h->delta_poc[1]= get_se_golomb(&s->gb);
4286
    }
4287
    
4288
    init_poc(h);
4289
    
4290
    if(h->pps.redundant_pic_cnt_present){
4291
        h->redundant_pic_count= get_ue_golomb(&s->gb);
4292
    }
4293

    
4294
    //set defaults, might be overriden a few line later
4295
    h->ref_count[0]= h->pps.ref_count[0];
4296
    h->ref_count[1]= h->pps.ref_count[1];
4297

    
4298
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
4299
        if(h->slice_type == B_TYPE){
4300
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
4301
        }
4302
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
4303
    
4304
        if(num_ref_idx_active_override_flag){
4305
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
4306
            if(h->slice_type==B_TYPE)
4307
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
4308

    
4309
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
4310
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
4311
                return -1;
4312
            }
4313
        }
4314
    }
4315

    
4316
    if(!default_ref_list_done){
4317
        fill_default_ref_list(h);
4318
    }
4319

    
4320
    if(decode_ref_pic_list_reordering(h) < 0)
4321
        return -1;
4322

    
4323
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
4324
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
4325
        pred_weight_table(h);
4326
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
4327
        implicit_weight_table(h);
4328
    else
4329
        h->use_weight = 0;
4330
    
4331
    if(s->current_picture.reference)
4332
        decode_ref_pic_marking(h);
4333

    
4334
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
4335
        h->cabac_init_idc = get_ue_golomb(&s->gb);