Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ cad4368a

History | View | Annotate | Download (289 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
}H264Context;
352

    
353
static VLC coeff_token_vlc[4];
354
static VLC chroma_dc_coeff_token_vlc;
355

    
356
static VLC total_zeros_vlc[15];
357
static VLC chroma_dc_total_zeros_vlc[3];
358

    
359
static VLC run_vlc[6];
360
static VLC run7_vlc;
361

    
362
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
363
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
364
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);
365

    
366
static inline uint32_t pack16to32(int a, int b){
367
#ifdef WORDS_BIGENDIAN
368
   return (b&0xFFFF) + (a<<16);
369
#else
370
   return (a&0xFFFF) + (b<<16);
371
#endif
372
}
373

    
374
/**
375
 * fill a rectangle.
376
 * @param h height of the rectangle, should be a constant
377
 * @param w width of the rectangle, should be a constant
378
 * @param size the size of val (1 or 4), should be a constant
379
 */
380
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
381
    uint8_t *p= (uint8_t*)vp;
382
    assert(size==1 || size==4);
383
    
384
    w      *= size;
385
    stride *= size;
386
    
387
    assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
388
    assert((stride&(w-1))==0);
389
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
390
    if(w==2 && h==2){
391
        *(uint16_t*)(p + 0)=
392
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
393
    }else if(w==2 && h==4){
394
        *(uint16_t*)(p + 0*stride)=
395
        *(uint16_t*)(p + 1*stride)=
396
        *(uint16_t*)(p + 2*stride)=
397
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
398
    }else if(w==4 && h==1){
399
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
400
    }else if(w==4 && h==2){
401
        *(uint32_t*)(p + 0*stride)=
402
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
403
    }else if(w==4 && h==4){
404
        *(uint32_t*)(p + 0*stride)=
405
        *(uint32_t*)(p + 1*stride)=
406
        *(uint32_t*)(p + 2*stride)=
407
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
408
    }else if(w==8 && h==1){
409
        *(uint32_t*)(p + 0)=
410
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
411
    }else if(w==8 && h==2){
412
        *(uint32_t*)(p + 0 + 0*stride)=
413
        *(uint32_t*)(p + 4 + 0*stride)=
414
        *(uint32_t*)(p + 0 + 1*stride)=
415
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
416
    }else if(w==8 && h==4){
417
        *(uint64_t*)(p + 0*stride)=
418
        *(uint64_t*)(p + 1*stride)=
419
        *(uint64_t*)(p + 2*stride)=
420
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
421
    }else if(w==16 && h==2){
422
        *(uint64_t*)(p + 0+0*stride)=
423
        *(uint64_t*)(p + 8+0*stride)=
424
        *(uint64_t*)(p + 0+1*stride)=
425
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
426
    }else if(w==16 && h==4){
427
        *(uint64_t*)(p + 0+0*stride)=
428
        *(uint64_t*)(p + 8+0*stride)=
429
        *(uint64_t*)(p + 0+1*stride)=
430
        *(uint64_t*)(p + 8+1*stride)=
431
        *(uint64_t*)(p + 0+2*stride)=
432
        *(uint64_t*)(p + 8+2*stride)=
433
        *(uint64_t*)(p + 0+3*stride)=
434
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
435
    }else
436
        assert(0);
437
}
438

    
439
static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
440
    MpegEncContext * const s = &h->s;
441
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
442
    int topleft_xy, top_xy, topright_xy, left_xy[2];
443
    int topleft_type, top_type, topright_type, left_type[2];
444
    int left_block[8];
445
    int i;
446

    
447
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
448
    // the actual condition is whether we're on the edge of a slice,
449
    // and even then the intra and nnz parts are unnecessary.
450
    if(for_deblock && h->slice_num == 1)
451
        return;
452

    
453
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it 
454
    
455
    top_xy     = mb_xy  - s->mb_stride;
456
    topleft_xy = top_xy - 1;
457
    topright_xy= top_xy + 1;
458
    left_xy[1] = left_xy[0] = mb_xy-1;
459
    left_block[0]= 0;
460
    left_block[1]= 1;
461
    left_block[2]= 2;
462
    left_block[3]= 3;
463
    left_block[4]= 7;
464
    left_block[5]= 10;
465
    left_block[6]= 8;
466
    left_block[7]= 11;
467
    if(h->mb_aff_frame){
468
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
469
        const int top_pair_xy      = pair_xy     - s->mb_stride;
470
        const int topleft_pair_xy  = top_pair_xy - 1;
471
        const int topright_pair_xy = top_pair_xy + 1;
472
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
473
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
474
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
475
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
476
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
477
        const int bottom = (s->mb_y & 1);
478
        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);
479
        if (bottom
480
                ? !curr_mb_frame_flag // bottom macroblock
481
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
482
                ) {
483
            top_xy -= s->mb_stride;
484
        }
485
        if (bottom
486
                ? !curr_mb_frame_flag // bottom macroblock
487
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
488
                ) {
489
            topleft_xy -= s->mb_stride;
490
        }
491
        if (bottom
492
                ? !curr_mb_frame_flag // bottom macroblock
493
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
494
                ) {
495
            topright_xy -= s->mb_stride;
496
        }
497
        if (left_mb_frame_flag != curr_mb_frame_flag) {
498
            left_xy[1] = left_xy[0] = pair_xy - 1;
499
            if (curr_mb_frame_flag) {
500
                if (bottom) {
501
                    left_block[0]= 2;
502
                    left_block[1]= 2;
503
                    left_block[2]= 3;
504
                    left_block[3]= 3;
505
                    left_block[4]= 8;
506
                    left_block[5]= 11;
507
                    left_block[6]= 8;
508
                    left_block[7]= 11;
509
                } else {
510
                    left_block[0]= 0;
511
                    left_block[1]= 0;
512
                    left_block[2]= 1;
513
                    left_block[3]= 1;
514
                    left_block[4]= 7;
515
                    left_block[5]= 10;
516
                    left_block[6]= 7;
517
                    left_block[7]= 10;
518
                }
519
            } else {
520
                left_xy[1] += s->mb_stride;
521
                //left_block[0]= 0;
522
                left_block[1]= 2;
523
                left_block[2]= 0;
524
                left_block[3]= 2;
525
                //left_block[4]= 7;
526
                left_block[5]= 10;
527
                left_block[6]= 7;
528
                left_block[7]= 10;
529
            }
530
        }
531
    }
532

    
533
    h->top_mb_xy = top_xy;
534
    h->left_mb_xy[0] = left_xy[0];
535
    h->left_mb_xy[1] = left_xy[1];
536
    if(for_deblock){
537
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
538
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
539
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
540
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
541
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
542
    }else{
543
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
544
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
545
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
546
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
547
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
548
    }
549

    
550
    if(IS_INTRA(mb_type)){
551
        h->topleft_samples_available= 
552
        h->top_samples_available= 
553
        h->left_samples_available= 0xFFFF;
554
        h->topright_samples_available= 0xEEEA;
555

    
556
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
557
            h->topleft_samples_available= 0xB3FF;
558
            h->top_samples_available= 0x33FF;
559
            h->topright_samples_available= 0x26EA;
560
        }
561
        for(i=0; i<2; i++){
562
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
563
                h->topleft_samples_available&= 0xDF5F;
564
                h->left_samples_available&= 0x5F5F;
565
            }
566
        }
567
        
568
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
569
            h->topleft_samples_available&= 0x7FFF;
570
        
571
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
572
            h->topright_samples_available&= 0xFBFF;
573
    
574
        if(IS_INTRA4x4(mb_type)){
575
            if(IS_INTRA4x4(top_type)){
576
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
577
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
578
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
579
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
580
            }else{
581
                int pred;
582
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
583
                    pred= -1;
584
                else{
585
                    pred= 2;
586
                }
587
                h->intra4x4_pred_mode_cache[4+8*0]=
588
                h->intra4x4_pred_mode_cache[5+8*0]=
589
                h->intra4x4_pred_mode_cache[6+8*0]=
590
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
591
            }
592
            for(i=0; i<2; i++){
593
                if(IS_INTRA4x4(left_type[i])){
594
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
595
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
596
                }else{
597
                    int pred;
598
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
599
                        pred= -1;
600
                    else{
601
                        pred= 2;
602
                    }
603
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
604
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
605
                }
606
            }
607
        }
608
    }
609
    
610
    
611
/*
612
0 . T T. T T T T 
613
1 L . .L . . . . 
614
2 L . .L . . . . 
615
3 . T TL . . . . 
616
4 L . .L . . . . 
617
5 L . .. . . . . 
618
*/
619
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
620
    if(top_type){
621
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
622
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
623
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
624
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
625
    
626
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
627
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
628
    
629
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
630
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
631
        
632
    }else{
633
        h->non_zero_count_cache[4+8*0]=      
634
        h->non_zero_count_cache[5+8*0]=
635
        h->non_zero_count_cache[6+8*0]=
636
        h->non_zero_count_cache[7+8*0]=
637
    
638
        h->non_zero_count_cache[1+8*0]=
639
        h->non_zero_count_cache[2+8*0]=
640
    
641
        h->non_zero_count_cache[1+8*3]=
642
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
643
        
644
    }
645

    
646
    for (i=0; i<2; i++) {
647
        if(left_type[i]){
648
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
649
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
650
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
651
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
652
        }else{
653
            h->non_zero_count_cache[3+8*1 + 2*8*i]= 
654
            h->non_zero_count_cache[3+8*2 + 2*8*i]= 
655
            h->non_zero_count_cache[0+8*1 +   8*i]= 
656
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
657
        }
658
    }
659

    
660
    if( h->pps.cabac ) {
661
        // top_cbp
662
        if(top_type) {
663
            h->top_cbp = h->cbp_table[top_xy];
664
        } else if(IS_INTRA(mb_type)) {
665
            h->top_cbp = 0x1C0;
666
        } else {
667
            h->top_cbp = 0;
668
        }
669
        // left_cbp
670
        if (left_type[0]) {
671
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
672
        } else if(IS_INTRA(mb_type)) {
673
            h->left_cbp = 0x1C0;
674
        } else {
675
            h->left_cbp = 0;
676
        }
677
        if (left_type[0]) {
678
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
679
        }
680
        if (left_type[1]) {
681
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
682
        }
683
    }
684

    
685
#if 1
686
    //FIXME direct mb can skip much of this
687
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
688
        int list;
689
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
690
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
691
                /*if(!h->mv_cache_clean[list]){
692
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
693
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
694
                    h->mv_cache_clean[list]= 1;
695
                }*/
696
                continue;
697
            }
698
            h->mv_cache_clean[list]= 0;
699
            
700
            if(IS_INTER(top_type)){
701
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
702
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
703
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
704
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
705
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
706
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
707
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
708
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
709
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
710
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
711
            }else{
712
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
713
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
714
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
715
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
716
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
717
            }
718

    
719
            //FIXME unify cleanup or sth
720
            if(IS_INTER(left_type[0])){
721
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
722
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
723
                *(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]];
724
                *(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]];
725
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
726
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
727
            }else{
728
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
729
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
730
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
731
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
732
            }
733
            
734
            if(IS_INTER(left_type[1])){
735
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
736
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
737
                *(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]];
738
                *(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]];
739
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
740
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
741
            }else{
742
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
743
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
744
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
745
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
746
                assert((!left_type[0]) == (!left_type[1]));
747
            }
748

    
749
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
750
                continue;
751

    
752
            if(IS_INTER(topleft_type)){
753
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
754
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
755
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
756
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
757
            }else{
758
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
759
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
760
            }
761
            
762
            if(IS_INTER(topright_type)){
763
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
764
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
765
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
766
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
767
            }else{
768
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
769
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
770
            }
771
            
772

    
773
            h->ref_cache[list][scan8[5 ]+1] = 
774
            h->ref_cache[list][scan8[7 ]+1] = 
775
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
776
            h->ref_cache[list][scan8[4 ]] = 
777
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
778
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
779
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
780
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
781
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
782
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
783

    
784
            if( h->pps.cabac ) {
785
                /* XXX beurk, Load mvd */
786
                if(IS_INTER(topleft_type)){
787
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
788
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
789
                }else{
790
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
791
                }
792

    
793
                if(IS_INTER(top_type)){
794
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
795
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
796
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
797
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
798
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
799
                }else{
800
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
801
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
802
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
803
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
804
                }
805
                if(IS_INTER(left_type[0])){
806
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
807
                    *(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]];
808
                    *(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]];
809
                }else{
810
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
811
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
812
                }
813
                if(IS_INTER(left_type[1])){
814
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
815
                    *(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]];
816
                    *(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]];
817
                }else{
818
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
819
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
820
                }
821
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
822
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
823
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
824
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
825
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
826

    
827
                if(h->slice_type == B_TYPE){
828
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
829

    
830
                    if(IS_DIRECT(top_type)){
831
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
832
                    }else if(IS_8X8(top_type)){
833
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
834
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
835
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
836
                    }else{
837
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
838
                    }
839
                    
840
                    //FIXME interlacing
841
                    if(IS_DIRECT(left_type[0])){
842
                        h->direct_cache[scan8[0] - 1 + 0*8]=
843
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
844
                    }else if(IS_8X8(left_type[0])){
845
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
846
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
847
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
848
                    }else{
849
                        h->direct_cache[scan8[0] - 1 + 0*8]=
850
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
851
                    }
852
                }
853
            }
854
        }
855
    }
856
#endif
857

    
858
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
859
}
860

    
861
static inline void write_back_intra_pred_mode(H264Context *h){
862
    MpegEncContext * const s = &h->s;
863
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
864

    
865
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
866
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
867
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
868
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
869
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
870
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
871
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
872
}
873

    
874
/**
875
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
876
 */
877
static inline int check_intra4x4_pred_mode(H264Context *h){
878
    MpegEncContext * const s = &h->s;
879
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
880
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
881
    int i;
882
    
883
    if(!(h->top_samples_available&0x8000)){
884
        for(i=0; i<4; i++){
885
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
886
            if(status<0){
887
                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);
888
                return -1;
889
            } else if(status){
890
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
891
            }
892
        }
893
    }
894
    
895
    if(!(h->left_samples_available&0x8000)){
896
        for(i=0; i<4; i++){
897
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
898
            if(status<0){
899
                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);
900
                return -1;
901
            } else if(status){
902
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
903
            }
904
        }
905
    }
906

    
907
    return 0;
908
} //FIXME cleanup like next
909

    
910
/**
911
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
912
 */
913
static inline int check_intra_pred_mode(H264Context *h, int mode){
914
    MpegEncContext * const s = &h->s;
915
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
916
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
917
    
918
    if(mode < 0 || mode > 6) {
919
        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);
920
        return -1;
921
    }
922
    
923
    if(!(h->top_samples_available&0x8000)){
924
        mode= top[ mode ];
925
        if(mode<0){
926
            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);
927
            return -1;
928
        }
929
    }
930
    
931
    if(!(h->left_samples_available&0x8000)){
932
        mode= left[ mode ];
933
        if(mode<0){
934
            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);
935
            return -1;
936
        } 
937
    }
938

    
939
    return mode;
940
}
941

    
942
/**
943
 * gets the predicted intra4x4 prediction mode.
944
 */
945
static inline int pred_intra_mode(H264Context *h, int n){
946
    const int index8= scan8[n];
947
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
948
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
949
    const int min= FFMIN(left, top);
950

    
951
    tprintf("mode:%d %d min:%d\n", left ,top, min);
952

    
953
    if(min<0) return DC_PRED;
954
    else      return min;
955
}
956

    
957
static inline void write_back_non_zero_count(H264Context *h){
958
    MpegEncContext * const s = &h->s;
959
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
960

    
961
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
962
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
963
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
964
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
965
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
966
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
967
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
968
    
969
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
970
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
971
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
972

    
973
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
974
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
975
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
976
}
977

    
978
/**
979
 * gets the predicted number of non zero coefficients.
980
 * @param n block index
981
 */
982
static inline int pred_non_zero_count(H264Context *h, int n){
983
    const int index8= scan8[n];
984
    const int left= h->non_zero_count_cache[index8 - 1];
985
    const int top = h->non_zero_count_cache[index8 - 8];
986
    int i= left + top;
987
    
988
    if(i<64) i= (i+1)>>1;
989

    
990
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
991

    
992
    return i&31;
993
}
994

    
995
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
996
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
997

    
998
    if(topright_ref != PART_NOT_AVAILABLE){
999
        *C= h->mv_cache[list][ i - 8 + part_width ];
1000
        return topright_ref;
1001
    }else{
1002
        tprintf("topright MV not available\n");
1003

    
1004
        *C= h->mv_cache[list][ i - 8 - 1 ];
1005
        return h->ref_cache[list][ i - 8 - 1 ];
1006
    }
1007
}
1008

    
1009
/**
1010
 * gets the predicted MV.
1011
 * @param n the block index
1012
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1013
 * @param mx the x component of the predicted motion vector
1014
 * @param my the y component of the predicted motion vector
1015
 */
1016
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1017
    const int index8= scan8[n];
1018
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1019
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1020
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1021
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1022
    const int16_t * C;
1023
    int diagonal_ref, match_count;
1024

    
1025
    assert(part_width==1 || part_width==2 || part_width==4);
1026

    
1027
/* mv_cache
1028
  B . . A T T T T 
1029
  U . . L . . , .
1030
  U . . L . . . .
1031
  U . . L . . , .
1032
  . . . L . . . .
1033
*/
1034

    
1035
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1036
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1037
    tprintf("pred_motion match_count=%d\n", match_count);
1038
    if(match_count > 1){ //most common
1039
        *mx= mid_pred(A[0], B[0], C[0]);
1040
        *my= mid_pred(A[1], B[1], C[1]);
1041
    }else if(match_count==1){
1042
        if(left_ref==ref){
1043
            *mx= A[0];
1044
            *my= A[1];        
1045
        }else if(top_ref==ref){
1046
            *mx= B[0];
1047
            *my= B[1];        
1048
        }else{
1049
            *mx= C[0];
1050
            *my= C[1];        
1051
        }
1052
    }else{
1053
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1054
            *mx= A[0];
1055
            *my= A[1];        
1056
        }else{
1057
            *mx= mid_pred(A[0], B[0], C[0]);
1058
            *my= mid_pred(A[1], B[1], C[1]);
1059
        }
1060
    }
1061
        
1062
    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);
1063
}
1064

    
1065
/**
1066
 * gets the directionally predicted 16x8 MV.
1067
 * @param n the block index
1068
 * @param mx the x component of the predicted motion vector
1069
 * @param my the y component of the predicted motion vector
1070
 */
1071
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1072
    if(n==0){
1073
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1074
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1075

    
1076
        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);
1077
        
1078
        if(top_ref == ref){
1079
            *mx= B[0];
1080
            *my= B[1];
1081
            return;
1082
        }
1083
    }else{
1084
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1085
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1086
        
1087
        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);
1088

    
1089
        if(left_ref == ref){
1090
            *mx= A[0];
1091
            *my= A[1];
1092
            return;
1093
        }
1094
    }
1095

    
1096
    //RARE
1097
    pred_motion(h, n, 4, list, ref, mx, my);
1098
}
1099

    
1100
/**
1101
 * gets the directionally predicted 8x16 MV.
1102
 * @param n the block index
1103
 * @param mx the x component of the predicted motion vector
1104
 * @param my the y component of the predicted motion vector
1105
 */
1106
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1107
    if(n==0){
1108
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1109
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1110
        
1111
        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);
1112

    
1113
        if(left_ref == ref){
1114
            *mx= A[0];
1115
            *my= A[1];
1116
            return;
1117
        }
1118
    }else{
1119
        const int16_t * C;
1120
        int diagonal_ref;
1121

    
1122
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1123
        
1124
        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);
1125

    
1126
        if(diagonal_ref == ref){ 
1127
            *mx= C[0];
1128
            *my= C[1];
1129
            return;
1130
        }
1131
    }
1132

    
1133
    //RARE
1134
    pred_motion(h, n, 2, list, ref, mx, my);
1135
}
1136

    
1137
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1138
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1139
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1140

    
1141
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1142

    
1143
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1144
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1145
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1146
       
1147
        *mx = *my = 0;
1148
        return;
1149
    }
1150
        
1151
    pred_motion(h, 0, 4, 0, 0, mx, my);
1152

    
1153
    return;
1154
}
1155

    
1156
static inline void direct_dist_scale_factor(H264Context * const h){
1157
    const int poc = h->s.current_picture_ptr->poc;
1158
    const int poc1 = h->ref_list[1][0].poc;
1159
    int i;
1160
    for(i=0; i<h->ref_count[0]; i++){
1161
        int poc0 = h->ref_list[0][i].poc;
1162
        int td = clip(poc1 - poc0, -128, 127);
1163
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1164
            h->dist_scale_factor[i] = 256;
1165
        }else{
1166
            int tb = clip(poc - poc0, -128, 127);
1167
            int tx = (16384 + (ABS(td) >> 1)) / td;
1168
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1169
        }
1170
    }
1171
}
1172
static inline void direct_ref_list_init(H264Context * const h){
1173
    MpegEncContext * const s = &h->s;
1174
    Picture * const ref1 = &h->ref_list[1][0];
1175
    Picture * const cur = s->current_picture_ptr;
1176
    int list, i, j;
1177
    if(cur->pict_type == I_TYPE)
1178
        cur->ref_count[0] = 0;
1179
    if(cur->pict_type != B_TYPE)
1180
        cur->ref_count[1] = 0;
1181
    for(list=0; list<2; list++){
1182
        cur->ref_count[list] = h->ref_count[list];
1183
        for(j=0; j<h->ref_count[list]; j++)
1184
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1185
    }
1186
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1187
        return;
1188
    for(list=0; list<2; list++){
1189
        for(i=0; i<ref1->ref_count[list]; i++){
1190
            const int poc = ref1->ref_poc[list][i];
1191
            h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1192
            for(j=0; j<h->ref_count[list]; j++)
1193
                if(h->ref_list[list][j].poc == poc){
1194
                    h->map_col_to_list0[list][i] = j;
1195
                    break;
1196
                }
1197
        }
1198
    }
1199
}
1200

    
1201
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1202
    MpegEncContext * const s = &h->s;
1203
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1204
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1205
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1206
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1207
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1208
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1209
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1210
    const int is_b8x8 = IS_8X8(*mb_type);
1211
    int sub_mb_type;
1212
    int i8, i4;
1213

    
1214
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1215
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1216
         * so we know exactly what block size to use */
1217
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1218
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1219
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1220
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1221
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1222
    }else{
1223
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1224
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1225
    }
1226
    if(!is_b8x8)
1227
        *mb_type |= MB_TYPE_DIRECT2;
1228

    
1229
    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);
1230
    
1231
    if(h->direct_spatial_mv_pred){
1232
        int ref[2];
1233
        int mv[2][2];
1234
        int list;
1235

    
1236
        /* ref = min(neighbors) */
1237
        for(list=0; list<2; list++){
1238
            int refa = h->ref_cache[list][scan8[0] - 1];
1239
            int refb = h->ref_cache[list][scan8[0] - 8];
1240
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1241
            if(refc == -2)
1242
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1243
            ref[list] = refa;
1244
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1245
                ref[list] = refb;
1246
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1247
                ref[list] = refc;
1248
            if(ref[list] < 0)
1249
                ref[list] = -1;
1250
        }
1251

    
1252
        if(ref[0] < 0 && ref[1] < 0){
1253
            ref[0] = ref[1] = 0;
1254
            mv[0][0] = mv[0][1] =
1255
            mv[1][0] = mv[1][1] = 0;
1256
        }else{
1257
            for(list=0; list<2; list++){
1258
                if(ref[list] >= 0)
1259
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1260
                else
1261
                    mv[list][0] = mv[list][1] = 0;
1262
            }
1263
        }
1264

    
1265
        if(ref[1] < 0){
1266
            *mb_type &= ~MB_TYPE_P0L1;
1267
            sub_mb_type &= ~MB_TYPE_P0L1;
1268
        }else if(ref[0] < 0){
1269
            *mb_type &= ~MB_TYPE_P0L0;
1270
            sub_mb_type &= ~MB_TYPE_P0L0;
1271
        }
1272

    
1273
        if(IS_16X16(*mb_type)){
1274
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1275
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1276
            if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1277
                ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1278
                if(ref[0] > 0)
1279
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1280
                else
1281
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1282
                if(ref[1] > 0)
1283
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1284
                else
1285
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1286
            }else{
1287
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1288
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1289
            }
1290
        }else{
1291
            for(i8=0; i8<4; i8++){
1292
                const int x8 = i8&1;
1293
                const int y8 = i8>>1;
1294
    
1295
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1296
                    continue;
1297
                h->sub_mb_type[i8] = sub_mb_type;
1298
    
1299
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1300
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1301
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1302
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1303
    
1304
                /* col_zero_flag */
1305
                if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1306
                    for(i4=0; i4<4; i4++){
1307
                        const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1308
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1309
                            if(ref[0] == 0)
1310
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1311
                            if(ref[1] == 0)
1312
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1313
                        }
1314
                    }
1315
                }
1316
            }
1317
        }
1318
    }else{ /* direct temporal mv pred */
1319
        if(IS_16X16(*mb_type)){
1320
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1321
            if(IS_INTRA(mb_type_col)){
1322
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1323
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1324
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1325
            }else{
1326
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1327
                                                : h->map_col_to_list0[1][l1ref1[0]];
1328
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1329
                const int16_t *mv_col = l1mv0[0];
1330
                int mv_l0[2];
1331
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1332
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1333
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1334
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1335
                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);
1336
            }
1337
        }else{
1338
            for(i8=0; i8<4; i8++){
1339
                const int x8 = i8&1;
1340
                const int y8 = i8>>1;
1341
                int ref0, dist_scale_factor;
1342
    
1343
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1344
                    continue;
1345
                h->sub_mb_type[i8] = sub_mb_type;
1346
                if(IS_INTRA(mb_type_col)){
1347
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1348
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1349
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1350
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1351
                    continue;
1352
                }
1353
    
1354
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1355
                if(ref0 >= 0)
1356
                    ref0 = h->map_col_to_list0[0][ref0];
1357
                else
1358
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1359
                dist_scale_factor = h->dist_scale_factor[ref0];
1360
    
1361
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1362
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1363
                for(i4=0; i4<4; i4++){
1364
                    const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1365
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1366
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1367
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1368
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1369
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1370
                }
1371
            }
1372
        }
1373
    }
1374
}
1375

    
1376
static inline void write_back_motion(H264Context *h, int mb_type){
1377
    MpegEncContext * const s = &h->s;
1378
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1379
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1380
    int list;
1381

    
1382
    for(list=0; list<2; list++){
1383
        int y;
1384
        if(!USES_LIST(mb_type, list)){
1385
            if(1){ //FIXME skip or never read if mb_type doesn't use it
1386
                for(y=0; y<4; y++){
1387
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1388
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1389
                }
1390
                if( h->pps.cabac ) {
1391
                    /* FIXME needed ? */
1392
                    for(y=0; y<4; y++){
1393
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1394
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1395
                    }
1396
                }
1397
                for(y=0; y<2; y++){
1398
                    s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1399
                    s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1400
                }
1401
            }
1402
            continue;
1403
        }
1404
        
1405
        for(y=0; y<4; y++){
1406
            *(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];
1407
            *(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];
1408
        }
1409
        if( h->pps.cabac ) {
1410
            for(y=0; y<4; y++){
1411
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1412
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1413
            }
1414
        }
1415
        for(y=0; y<2; y++){
1416
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1417
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1418
        }
1419
    }
1420
    
1421
    if(h->slice_type == B_TYPE && h->pps.cabac){
1422
        if(IS_8X8(mb_type)){
1423
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1424
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1425
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1426
        }
1427
    }
1428
}
1429

    
1430
/**
1431
 * Decodes a network abstraction layer unit.
1432
 * @param consumed is the number of bytes used as input
1433
 * @param length is the length of the array
1434
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1435
 * @returns decoded bytes, might be src+1 if no escapes 
1436
 */
1437
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1438
    int i, si, di;
1439
    uint8_t *dst;
1440

    
1441
//    src[0]&0x80;                //forbidden bit
1442
    h->nal_ref_idc= src[0]>>5;
1443
    h->nal_unit_type= src[0]&0x1F;
1444

    
1445
    src++; length--;
1446
#if 0    
1447
    for(i=0; i<length; i++)
1448
        printf("%2X ", src[i]);
1449
#endif
1450
    for(i=0; i+1<length; i+=2){
1451
        if(src[i]) continue;
1452
        if(i>0 && src[i-1]==0) i--;
1453
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1454
            if(src[i+2]!=3){
1455
                /* startcode, so we must be past the end */
1456
                length=i;
1457
            }
1458
            break;
1459
        }
1460
    }
1461

    
1462
    if(i>=length-1){ //no escaped 0
1463
        *dst_length= length;
1464
        *consumed= length+1; //+1 for the header
1465
        return src; 
1466
    }
1467

    
1468
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1469
    dst= h->rbsp_buffer;
1470

    
1471
//printf("decoding esc\n");
1472
    si=di=0;
1473
    while(si<length){ 
1474
        //remove escapes (very rare 1:2^22)
1475
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1476
            if(src[si+2]==3){ //escape
1477
                dst[di++]= 0;
1478
                dst[di++]= 0;
1479
                si+=3;
1480
                continue;
1481
            }else //next start code
1482
                break;
1483
        }
1484

    
1485
        dst[di++]= src[si++];
1486
    }
1487

    
1488
    *dst_length= di;
1489
    *consumed= si + 1;//+1 for the header
1490
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1491
    return dst;
1492
}
1493

    
1494
#if 0
1495
/**
1496
 * @param src the data which should be escaped
1497
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1498
 * @param length the length of the src data
1499
 * @param dst_length the length of the dst array
1500
 * @returns length of escaped data in bytes or -1 if an error occured
1501
 */
1502
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1503
    int i, escape_count, si, di;
1504
    uint8_t *temp;
1505
    
1506
    assert(length>=0);
1507
    assert(dst_length>0);
1508
    
1509
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1510

1511
    if(length==0) return 1;
1512

1513
    escape_count= 0;
1514
    for(i=0; i<length; i+=2){
1515
        if(src[i]) continue;
1516
        if(i>0 && src[i-1]==0) 
1517
            i--;
1518
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1519
            escape_count++;
1520
            i+=2;
1521
        }
1522
    }
1523
    
1524
    if(escape_count==0){ 
1525
        if(dst+1 != src)
1526
            memcpy(dst+1, src, length);
1527
        return length + 1;
1528
    }
1529
    
1530
    if(length + escape_count + 1> dst_length)
1531
        return -1;
1532

1533
    //this should be damn rare (hopefully)
1534

1535
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1536
    temp= h->rbsp_buffer;
1537
//printf("encoding esc\n");
1538
    
1539
    si= 0;
1540
    di= 0;
1541
    while(si < length){
1542
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1543
            temp[di++]= 0; si++;
1544
            temp[di++]= 0; si++;
1545
            temp[di++]= 3; 
1546
            temp[di++]= src[si++];
1547
        }
1548
        else
1549
            temp[di++]= src[si++];
1550
    }
1551
    memcpy(dst+1, temp, length+escape_count);
1552
    
1553
    assert(di == length+escape_count);
1554
    
1555
    return di + 1;
1556
}
1557

1558
/**
1559
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1560
 */
1561
static void encode_rbsp_trailing(PutBitContext *pb){
1562
    int length;
1563
    put_bits(pb, 1, 1);
1564
    length= (-put_bits_count(pb))&7;
1565
    if(length) put_bits(pb, length, 0);
1566
}
1567
#endif
1568

    
1569
/**
1570
 * identifies the exact end of the bitstream
1571
 * @return the length of the trailing, or 0 if damaged
1572
 */
1573
static int decode_rbsp_trailing(uint8_t *src){
1574
    int v= *src;
1575
    int r;
1576

    
1577
    tprintf("rbsp trailing %X\n", v);
1578

    
1579
    for(r=1; r<9; r++){
1580
        if(v&1) return r;
1581
        v>>=1;
1582
    }
1583
    return 0;
1584
}
1585

    
1586
/**
1587
 * idct tranforms the 16 dc values and dequantize them.
1588
 * @param qp quantization parameter
1589
 */
1590
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1591
    const int qmul= dequant_coeff[qp][0];
1592
#define stride 16
1593
    int i;
1594
    int temp[16]; //FIXME check if this is a good idea
1595
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1596
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1597

    
1598
//memset(block, 64, 2*256);
1599
//return;
1600
    for(i=0; i<4; i++){
1601
        const int offset= y_offset[i];
1602
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1603
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1604
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1605
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1606

    
1607
        temp[4*i+0]= z0+z3;
1608
        temp[4*i+1]= z1+z2;
1609
        temp[4*i+2]= z1-z2;
1610
        temp[4*i+3]= z0-z3;
1611
    }
1612

    
1613
    for(i=0; i<4; i++){
1614
        const int offset= x_offset[i];
1615
        const int z0= temp[4*0+i] + temp[4*2+i];
1616
        const int z1= temp[4*0+i] - temp[4*2+i];
1617
        const int z2= temp[4*1+i] - temp[4*3+i];
1618
        const int z3= temp[4*1+i] + temp[4*3+i];
1619

    
1620
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1621
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1622
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1623
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1624
    }
1625
}
1626

    
1627
#if 0
1628
/**
1629
 * dct tranforms the 16 dc values.
1630
 * @param qp quantization parameter ??? FIXME
1631
 */
1632
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1633
//    const int qmul= dequant_coeff[qp][0];
1634
    int i;
1635
    int temp[16]; //FIXME check if this is a good idea
1636
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1637
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1638

1639
    for(i=0; i<4; i++){
1640
        const int offset= y_offset[i];
1641
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1642
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1643
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1644
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1645

1646
        temp[4*i+0]= z0+z3;
1647
        temp[4*i+1]= z1+z2;
1648
        temp[4*i+2]= z1-z2;
1649
        temp[4*i+3]= z0-z3;
1650
    }
1651

1652
    for(i=0; i<4; i++){
1653
        const int offset= x_offset[i];
1654
        const int z0= temp[4*0+i] + temp[4*2+i];
1655
        const int z1= temp[4*0+i] - temp[4*2+i];
1656
        const int z2= temp[4*1+i] - temp[4*3+i];
1657
        const int z3= temp[4*1+i] + temp[4*3+i];
1658

1659
        block[stride*0 +offset]= (z0 + z3)>>1;
1660
        block[stride*2 +offset]= (z1 + z2)>>1;
1661
        block[stride*8 +offset]= (z1 - z2)>>1;
1662
        block[stride*10+offset]= (z0 - z3)>>1;
1663
    }
1664
}
1665
#endif
1666

    
1667
#undef xStride
1668
#undef stride
1669

    
1670
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1671
    const int qmul= dequant_coeff[qp][0];
1672
    const int stride= 16*2;
1673
    const int xStride= 16;
1674
    int a,b,c,d,e;
1675

    
1676
    a= block[stride*0 + xStride*0];
1677
    b= block[stride*0 + xStride*1];
1678
    c= block[stride*1 + xStride*0];
1679
    d= block[stride*1 + xStride*1];
1680

    
1681
    e= a-b;
1682
    a= a+b;
1683
    b= c-d;
1684
    c= c+d;
1685

    
1686
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1687
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1688
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1689
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1690
}
1691

    
1692
#if 0
1693
static void chroma_dc_dct_c(DCTELEM *block){
1694
    const int stride= 16*2;
1695
    const int xStride= 16;
1696
    int a,b,c,d,e;
1697

1698
    a= block[stride*0 + xStride*0];
1699
    b= block[stride*0 + xStride*1];
1700
    c= block[stride*1 + xStride*0];
1701
    d= block[stride*1 + xStride*1];
1702

1703
    e= a-b;
1704
    a= a+b;
1705
    b= c-d;
1706
    c= c+d;
1707

1708
    block[stride*0 + xStride*0]= (a+c);
1709
    block[stride*0 + xStride*1]= (e+b);
1710
    block[stride*1 + xStride*0]= (a-c);
1711
    block[stride*1 + xStride*1]= (e-b);
1712
}
1713
#endif
1714

    
1715
/**
1716
 * gets the chroma qp.
1717
 */
1718
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1719
    
1720
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1721
}
1722

    
1723

    
1724
#if 0
1725
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1726
    int i;
1727
    //FIXME try int temp instead of block
1728
    
1729
    for(i=0; i<4; i++){
1730
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1731
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1732
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1733
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1734
        const int z0= d0 + d3;
1735
        const int z3= d0 - d3;
1736
        const int z1= d1 + d2;
1737
        const int z2= d1 - d2;
1738
        
1739
        block[0 + 4*i]=   z0 +   z1;
1740
        block[1 + 4*i]= 2*z3 +   z2;
1741
        block[2 + 4*i]=   z0 -   z1;
1742
        block[3 + 4*i]=   z3 - 2*z2;
1743
    }    
1744

1745
    for(i=0; i<4; i++){
1746
        const int z0= block[0*4 + i] + block[3*4 + i];
1747
        const int z3= block[0*4 + i] - block[3*4 + i];
1748
        const int z1= block[1*4 + i] + block[2*4 + i];
1749
        const int z2= block[1*4 + i] - block[2*4 + i];
1750
        
1751
        block[0*4 + i]=   z0 +   z1;
1752
        block[1*4 + i]= 2*z3 +   z2;
1753
        block[2*4 + i]=   z0 -   z1;
1754
        block[3*4 + i]=   z3 - 2*z2;
1755
    }
1756
}
1757
#endif
1758

    
1759
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1760
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1761
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1762
    int i;
1763
    const int * const quant_table= quant_coeff[qscale];
1764
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1765
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1766
    const unsigned int threshold2= (threshold1<<1);
1767
    int last_non_zero;
1768

    
1769
    if(seperate_dc){
1770
        if(qscale<=18){
1771
            //avoid overflows
1772
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1773
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1774
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1775

    
1776
            int level= block[0]*quant_coeff[qscale+18][0];
1777
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1778
                if(level>0){
1779
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1780
                    block[0]= level;
1781
                }else{
1782
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1783
                    block[0]= -level;
1784
                }
1785
//                last_non_zero = i;
1786
            }else{
1787
                block[0]=0;
1788
            }
1789
        }else{
1790
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1791
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1792
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1793

    
1794
            int level= block[0]*quant_table[0];
1795
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1796
                if(level>0){
1797
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1798
                    block[0]= level;
1799
                }else{
1800
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1801
                    block[0]= -level;
1802
                }
1803
//                last_non_zero = i;
1804
            }else{
1805
                block[0]=0;
1806
            }
1807
        }
1808
        last_non_zero= 0;
1809
        i=1;
1810
    }else{
1811
        last_non_zero= -1;
1812
        i=0;
1813
    }
1814

    
1815
    for(; i<16; i++){
1816
        const int j= scantable[i];
1817
        int level= block[j]*quant_table[j];
1818

    
1819
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1820
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1821
        if(((unsigned)(level+threshold1))>threshold2){
1822
            if(level>0){
1823
                level= (bias + level)>>QUANT_SHIFT;
1824
                block[j]= level;
1825
            }else{
1826
                level= (bias - level)>>QUANT_SHIFT;
1827
                block[j]= -level;
1828
            }
1829
            last_non_zero = i;
1830
        }else{
1831
            block[j]=0;
1832
        }
1833
    }
1834

    
1835
    return last_non_zero;
1836
}
1837

    
1838
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1839
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1840
    ((uint32_t*)(src+0*stride))[0]= a;
1841
    ((uint32_t*)(src+1*stride))[0]= a;
1842
    ((uint32_t*)(src+2*stride))[0]= a;
1843
    ((uint32_t*)(src+3*stride))[0]= a;
1844
}
1845

    
1846
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1847
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1848
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1849
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1850
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1851
}
1852

    
1853
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1854
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1855
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1856
    
1857
    ((uint32_t*)(src+0*stride))[0]= 
1858
    ((uint32_t*)(src+1*stride))[0]= 
1859
    ((uint32_t*)(src+2*stride))[0]= 
1860
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1861
}
1862

    
1863
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1864
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1865
    
1866
    ((uint32_t*)(src+0*stride))[0]= 
1867
    ((uint32_t*)(src+1*stride))[0]= 
1868
    ((uint32_t*)(src+2*stride))[0]= 
1869
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1870
}
1871

    
1872
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1873
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1874
    
1875
    ((uint32_t*)(src+0*stride))[0]= 
1876
    ((uint32_t*)(src+1*stride))[0]= 
1877
    ((uint32_t*)(src+2*stride))[0]= 
1878
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1879
}
1880

    
1881
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1882
    ((uint32_t*)(src+0*stride))[0]= 
1883
    ((uint32_t*)(src+1*stride))[0]= 
1884
    ((uint32_t*)(src+2*stride))[0]= 
1885
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1886
}
1887

    
1888

    
1889
#define LOAD_TOP_RIGHT_EDGE\
1890
    const int t4= topright[0];\
1891
    const int t5= topright[1];\
1892
    const int t6= topright[2];\
1893
    const int t7= topright[3];\
1894

    
1895
#define LOAD_LEFT_EDGE\
1896
    const int l0= src[-1+0*stride];\
1897
    const int l1= src[-1+1*stride];\
1898
    const int l2= src[-1+2*stride];\
1899
    const int l3= src[-1+3*stride];\
1900

    
1901
#define LOAD_TOP_EDGE\
1902
    const int t0= src[ 0-1*stride];\
1903
    const int t1= src[ 1-1*stride];\
1904
    const int t2= src[ 2-1*stride];\
1905
    const int t3= src[ 3-1*stride];\
1906

    
1907
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1908
    const int lt= src[-1-1*stride];
1909
    LOAD_TOP_EDGE
1910
    LOAD_LEFT_EDGE
1911

    
1912
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1913
    src[0+2*stride]=
1914
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1915
    src[0+1*stride]=
1916
    src[1+2*stride]=
1917
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1918
    src[0+0*stride]=
1919
    src[1+1*stride]=
1920
    src[2+2*stride]=
1921
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1922
    src[1+0*stride]=
1923
    src[2+1*stride]=
1924
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1925
    src[2+0*stride]=
1926
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1927
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1928
}
1929

    
1930
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1931
    LOAD_TOP_EDGE    
1932
    LOAD_TOP_RIGHT_EDGE    
1933
//    LOAD_LEFT_EDGE    
1934

    
1935
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1936
    src[1+0*stride]=
1937
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1938
    src[2+0*stride]=
1939
    src[1+1*stride]=
1940
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1941
    src[3+0*stride]=
1942
    src[2+1*stride]=
1943
    src[1+2*stride]=
1944
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1945
    src[3+1*stride]=
1946
    src[2+2*stride]=
1947
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1948
    src[3+2*stride]=
1949
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1950
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1951
}
1952

    
1953
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1954
    const int lt= src[-1-1*stride];
1955
    LOAD_TOP_EDGE    
1956
    LOAD_LEFT_EDGE    
1957
    const __attribute__((unused)) int unu= l3;
1958

    
1959
    src[0+0*stride]=
1960
    src[1+2*stride]=(lt + t0 + 1)>>1;
1961
    src[1+0*stride]=
1962
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1963
    src[2+0*stride]=
1964
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1965
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1966
    src[0+1*stride]=
1967
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1968
    src[1+1*stride]=
1969
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1970
    src[2+1*stride]=
1971
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1972
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1973
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1974
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1975
}
1976

    
1977
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1978
    LOAD_TOP_EDGE    
1979
    LOAD_TOP_RIGHT_EDGE    
1980
    const __attribute__((unused)) int unu= t7;
1981

    
1982
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1983
    src[1+0*stride]=
1984
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1985
    src[2+0*stride]=
1986
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1987
    src[3+0*stride]=
1988
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1989
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1990
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1991
    src[1+1*stride]=
1992
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1993
    src[2+1*stride]=
1994
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1995
    src[3+1*stride]=
1996
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1997
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1998
}
1999

    
2000
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2001
    LOAD_LEFT_EDGE    
2002

    
2003
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2004
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2005
    src[2+0*stride]=
2006
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2007
    src[3+0*stride]=
2008
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2009
    src[2+1*stride]=
2010
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2011
    src[3+1*stride]=
2012
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2013
    src[3+2*stride]=
2014
    src[1+3*stride]=
2015
    src[0+3*stride]=
2016
    src[2+2*stride]=
2017
    src[2+3*stride]=
2018
    src[3+3*stride]=l3;
2019
}
2020
    
2021
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2022
    const int lt= src[-1-1*stride];
2023
    LOAD_TOP_EDGE    
2024
    LOAD_LEFT_EDGE    
2025
    const __attribute__((unused)) int unu= t3;
2026

    
2027
    src[0+0*stride]=
2028
    src[2+1*stride]=(lt + l0 + 1)>>1;
2029
    src[1+0*stride]=
2030
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2031
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2032
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2033
    src[0+1*stride]=
2034
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2035
    src[1+1*stride]=
2036
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2037
    src[0+2*stride]=
2038
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2039
    src[1+2*stride]=
2040
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2041
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2042
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2043
}
2044

    
2045
static void pred16x16_vertical_c(uint8_t *src, int stride){
2046
    int i;
2047
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2048
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2049
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2050
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2051
    
2052
    for(i=0; i<16; i++){
2053
        ((uint32_t*)(src+i*stride))[0]= a;
2054
        ((uint32_t*)(src+i*stride))[1]= b;
2055
        ((uint32_t*)(src+i*stride))[2]= c;
2056
        ((uint32_t*)(src+i*stride))[3]= d;
2057
    }
2058
}
2059

    
2060
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2061
    int i;
2062

    
2063
    for(i=0; i<16; i++){
2064
        ((uint32_t*)(src+i*stride))[0]=
2065
        ((uint32_t*)(src+i*stride))[1]=
2066
        ((uint32_t*)(src+i*stride))[2]=
2067
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2068
    }
2069
}
2070

    
2071
static void pred16x16_dc_c(uint8_t *src, int stride){
2072
    int i, dc=0;
2073

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

    
2082
    dc= 0x01010101*((dc + 16)>>5);
2083

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

    
2092
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2093
    int i, dc=0;
2094

    
2095
    for(i=0;i<16; i++){
2096
        dc+= src[-1+i*stride];
2097
    }
2098
    
2099
    dc= 0x01010101*((dc + 8)>>4);
2100

    
2101
    for(i=0; i<16; i++){
2102
        ((uint32_t*)(src+i*stride))[0]=
2103
        ((uint32_t*)(src+i*stride))[1]=
2104
        ((uint32_t*)(src+i*stride))[2]=
2105
        ((uint32_t*)(src+i*stride))[3]= dc;
2106
    }
2107
}
2108

    
2109
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2110
    int i, dc=0;
2111

    
2112
    for(i=0;i<16; i++){
2113
        dc+= src[i-stride];
2114
    }
2115
    dc= 0x01010101*((dc + 8)>>4);
2116

    
2117
    for(i=0; i<16; i++){
2118
        ((uint32_t*)(src+i*stride))[0]=
2119
        ((uint32_t*)(src+i*stride))[1]=
2120
        ((uint32_t*)(src+i*stride))[2]=
2121
        ((uint32_t*)(src+i*stride))[3]= dc;
2122
    }
2123
}
2124

    
2125
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2126
    int i;
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]= 0x01010101U*128U;
2133
    }
2134
}
2135

    
2136
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2137
  int i, j, k;
2138
  int a;
2139
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2140
  const uint8_t * const src0 = src+7-stride;
2141
  const uint8_t *src1 = src+8*stride-1;
2142
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2143
  int H = src0[1] - src0[-1];
2144
  int V = src1[0] - src2[ 0];
2145
  for(k=2; k<=8; ++k) {
2146
    src1 += stride; src2 -= stride;
2147
    H += k*(src0[k] - src0[-k]);
2148
    V += k*(src1[0] - src2[ 0]);
2149
  }
2150
  if(svq3){
2151
    H = ( 5*(H/4) ) / 16;
2152
    V = ( 5*(V/4) ) / 16;
2153

    
2154
    /* required for 100% accuracy */
2155
    i = H; H = V; V = i;
2156
  }else{
2157
    H = ( 5*H+32 ) >> 6;
2158
    V = ( 5*V+32 ) >> 6;
2159
  }
2160

    
2161
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2162
  for(j=16; j>0; --j) {
2163
    int b = a;
2164
    a += V;
2165
    for(i=-16; i<0; i+=4) {
2166
      src[16+i] = cm[ (b    ) >> 5 ];
2167
      src[17+i] = cm[ (b+  H) >> 5 ];
2168
      src[18+i] = cm[ (b+2*H) >> 5 ];
2169
      src[19+i] = cm[ (b+3*H) >> 5 ];
2170
      b += 4*H;
2171
    }
2172
    src += stride;
2173
  }
2174
}
2175

    
2176
static void pred16x16_plane_c(uint8_t *src, int stride){
2177
    pred16x16_plane_compat_c(src, stride, 0);
2178
}
2179

    
2180
static void pred8x8_vertical_c(uint8_t *src, int stride){
2181
    int i;
2182
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2183
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2184
    
2185
    for(i=0; i<8; i++){
2186
        ((uint32_t*)(src+i*stride))[0]= a;
2187
        ((uint32_t*)(src+i*stride))[1]= b;
2188
    }
2189
}
2190

    
2191
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2192
    int i;
2193

    
2194
    for(i=0; i<8; i++){
2195
        ((uint32_t*)(src+i*stride))[0]=
2196
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2197
    }
2198
}
2199

    
2200
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2201
    int i;
2202

    
2203
    for(i=0; i<8; i++){
2204
        ((uint32_t*)(src+i*stride))[0]= 
2205
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2206
    }
2207
}
2208

    
2209
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2210
    int i;
2211
    int dc0, dc2;
2212

    
2213
    dc0=dc2=0;
2214
    for(i=0;i<4; i++){
2215
        dc0+= src[-1+i*stride];
2216
        dc2+= src[-1+(i+4)*stride];
2217
    }
2218
    dc0= 0x01010101*((dc0 + 2)>>2);
2219
    dc2= 0x01010101*((dc2 + 2)>>2);
2220

    
2221
    for(i=0; i<4; i++){
2222
        ((uint32_t*)(src+i*stride))[0]=
2223
        ((uint32_t*)(src+i*stride))[1]= dc0;
2224
    }
2225
    for(i=4; i<8; i++){
2226
        ((uint32_t*)(src+i*stride))[0]=
2227
        ((uint32_t*)(src+i*stride))[1]= dc2;
2228
    }
2229
}
2230

    
2231
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2232
    int i;
2233
    int dc0, dc1;
2234

    
2235
    dc0=dc1=0;
2236
    for(i=0;i<4; i++){
2237
        dc0+= src[i-stride];
2238
        dc1+= src[4+i-stride];
2239
    }
2240
    dc0= 0x01010101*((dc0 + 2)>>2);
2241
    dc1= 0x01010101*((dc1 + 2)>>2);
2242

    
2243
    for(i=0; i<4; i++){
2244
        ((uint32_t*)(src+i*stride))[0]= dc0;
2245
        ((uint32_t*)(src+i*stride))[1]= dc1;
2246
    }
2247
    for(i=4; i<8; i++){
2248
        ((uint32_t*)(src+i*stride))[0]= dc0;
2249
        ((uint32_t*)(src+i*stride))[1]= dc1;
2250
    }
2251
}
2252

    
2253

    
2254
static void pred8x8_dc_c(uint8_t *src, int stride){
2255
    int i;
2256
    int dc0, dc1, dc2, dc3;
2257

    
2258
    dc0=dc1=dc2=0;
2259
    for(i=0;i<4; i++){
2260
        dc0+= src[-1+i*stride] + src[i-stride];
2261
        dc1+= src[4+i-stride];
2262
        dc2+= src[-1+(i+4)*stride];
2263
    }
2264
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2265
    dc0= 0x01010101*((dc0 + 4)>>3);
2266
    dc1= 0x01010101*((dc1 + 2)>>2);
2267
    dc2= 0x01010101*((dc2 + 2)>>2);
2268

    
2269
    for(i=0; i<4; i++){
2270
        ((uint32_t*)(src+i*stride))[0]= dc0;
2271
        ((uint32_t*)(src+i*stride))[1]= dc1;
2272
    }
2273
    for(i=4; i<8; i++){
2274
        ((uint32_t*)(src+i*stride))[0]= dc2;
2275
        ((uint32_t*)(src+i*stride))[1]= dc3;
2276
    }
2277
}
2278

    
2279
static void pred8x8_plane_c(uint8_t *src, int stride){
2280
  int j, k;
2281
  int a;
2282
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2283
  const uint8_t * const src0 = src+3-stride;
2284
  const uint8_t *src1 = src+4*stride-1;
2285
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2286
  int H = src0[1] - src0[-1];
2287
  int V = src1[0] - src2[ 0];
2288
  for(k=2; k<=4; ++k) {
2289
    src1 += stride; src2 -= stride;
2290
    H += k*(src0[k] - src0[-k]);
2291
    V += k*(src1[0] - src2[ 0]);
2292
  }
2293
  H = ( 17*H+16 ) >> 5;
2294
  V = ( 17*V+16 ) >> 5;
2295

    
2296
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2297
  for(j=8; j>0; --j) {
2298
    int b = a;
2299
    a += V;
2300
    src[0] = cm[ (b    ) >> 5 ];
2301
    src[1] = cm[ (b+  H) >> 5 ];
2302
    src[2] = cm[ (b+2*H) >> 5 ];
2303
    src[3] = cm[ (b+3*H) >> 5 ];
2304
    src[4] = cm[ (b+4*H) >> 5 ];
2305
    src[5] = cm[ (b+5*H) >> 5 ];
2306
    src[6] = cm[ (b+6*H) >> 5 ];
2307
    src[7] = cm[ (b+7*H) >> 5 ];
2308
    src += stride;
2309
  }
2310
}
2311

    
2312
#define SRC(x,y) src[(x)+(y)*stride]
2313
#define PL(y) \
2314
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2315
#define PREDICT_8x8_LOAD_LEFT \
2316
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2317
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2318
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2319
    const int l7 = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2320

    
2321
#define PT(x) \
2322
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2323
#define PREDICT_8x8_LOAD_TOP \
2324
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2325
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2326
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2327
    const int t7 = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2328
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2329

    
2330
#define PTR(x) \
2331
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2332
#define PREDICT_8x8_LOAD_TOPRIGHT \
2333
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2334
    if(has_topright) { \
2335
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2336
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2337
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2338

    
2339
#define PREDICT_8x8_LOAD_TOPLEFT \
2340
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2341

    
2342
#define PREDICT_8x8_DC(v) \
2343
    int y; \
2344
    for( y = 0; y < 8; y++ ) { \
2345
        ((uint32_t*)src)[0] = \
2346
        ((uint32_t*)src)[1] = v; \
2347
        src += stride; \
2348
    }
2349

    
2350
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2351
{
2352
    PREDICT_8x8_DC(0x80808080);
2353
}
2354
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2355
{
2356
    PREDICT_8x8_LOAD_LEFT;
2357
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2358
    PREDICT_8x8_DC(dc);
2359
}
2360
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2361
{
2362
    PREDICT_8x8_LOAD_TOP;
2363
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2364
    PREDICT_8x8_DC(dc);
2365
}
2366
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2367
{
2368
    PREDICT_8x8_LOAD_LEFT;
2369
    PREDICT_8x8_LOAD_TOP;
2370
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2371
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2372
    PREDICT_8x8_DC(dc);
2373
}
2374
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2375
{
2376
    PREDICT_8x8_LOAD_LEFT;
2377
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2378
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2379
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2380
#undef ROW
2381
}
2382
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2383
{
2384
    int y;
2385
    PREDICT_8x8_LOAD_TOP;
2386
    src[0] = t0;
2387
    src[1] = t1;
2388
    src[2] = t2;
2389
    src[3] = t3;
2390
    src[4] = t4;
2391
    src[5] = t5;
2392
    src[6] = t6;
2393
    src[7] = t7;
2394
    for( y = 1; y < 8; y++ )
2395
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2396
}
2397
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2398
{
2399
    PREDICT_8x8_LOAD_TOP;
2400
    PREDICT_8x8_LOAD_TOPRIGHT;
2401
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2402
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2403
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2404
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2405
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2406
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2407
    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;
2408
    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;
2409
    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;
2410
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2411
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2412
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2413
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2414
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2415
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2416
}
2417
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2418
{
2419
    PREDICT_8x8_LOAD_TOP;
2420
    PREDICT_8x8_LOAD_LEFT;
2421
    PREDICT_8x8_LOAD_TOPLEFT;
2422
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2423
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2424
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2425
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2426
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2427
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2428
    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;
2429
    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;
2430
    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;
2431
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2432
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2433
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2434
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2435
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2436
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2437
  
2438
}
2439
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2440
{
2441
    PREDICT_8x8_LOAD_TOP;
2442
    PREDICT_8x8_LOAD_LEFT;
2443
    PREDICT_8x8_LOAD_TOPLEFT;
2444
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2445
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2446
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2447
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2448
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2449
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2450
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2451
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2452
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2453
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2454
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2455
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2456
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2457
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2458
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2459
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2460
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2461
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2462
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2463
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2464
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2465
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2466
}
2467
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2468
{
2469
    PREDICT_8x8_LOAD_TOP;
2470
    PREDICT_8x8_LOAD_LEFT;
2471
    PREDICT_8x8_LOAD_TOPLEFT;
2472
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2473
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2474
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2475
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2476
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2477
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2478
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2479
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2480
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2481
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2482
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2483
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2484
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2485
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2486
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2487
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2488
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2489
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2490
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2491
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2492
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2493
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2494
}
2495
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2496
{
2497
    PREDICT_8x8_LOAD_TOP;
2498
    PREDICT_8x8_LOAD_TOPRIGHT;
2499
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2500
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2501
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2502
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2503
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2504
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2505
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2506
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2507
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2508
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2509
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2510
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2511
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2512
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2513
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2514
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2515
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2516
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2517
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2518
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2519
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2520
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2521
}
2522
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2523
{
2524
    PREDICT_8x8_LOAD_LEFT;
2525
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2526
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2527
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2528
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2529
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2530
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2531
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2532
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2533
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2534
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2535
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2536
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2537
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2538
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2539
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2540
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2541
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2542
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2543
}
2544
#undef PREDICT_8x8_LOAD_LEFT
2545
#undef PREDICT_8x8_LOAD_TOP
2546
#undef PREDICT_8x8_LOAD_TOPLEFT
2547
#undef PREDICT_8x8_LOAD_TOPRIGHT
2548
#undef PREDICT_8x8_DC
2549
#undef PTR
2550
#undef PT
2551
#undef PL
2552
#undef SRC
2553

    
2554
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2555
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2556
                           int src_x_offset, int src_y_offset,
2557
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2558
    MpegEncContext * const s = &h->s;
2559
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2560
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2561
    const int luma_xy= (mx&3) + ((my&3)<<2);
2562
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2563
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2564
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2565
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2566
    int extra_height= extra_width;
2567
    int emu=0;
2568
    const int full_mx= mx>>2;
2569
    const int full_my= my>>2;
2570
    
2571
    assert(pic->data[0]);
2572
    
2573
    if(mx&7) extra_width -= 3;
2574
    if(my&7) extra_height -= 3;
2575
    
2576
    if(   full_mx < 0-extra_width 
2577
       || full_my < 0-extra_height 
2578
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
2579
       || full_my + 16/*FIXME*/ > s->height + extra_height){
2580
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
2581
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2582
        emu=1;
2583
    }
2584
    
2585
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2586
    if(!square){
2587
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2588
    }
2589
    
2590
    if(s->flags&CODEC_FLAG_GRAY) return;
2591
    
2592
    if(emu){
2593
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2594
            src_cb= s->edge_emu_buffer;
2595
    }
2596
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2597

    
2598
    if(emu){
2599
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2600
            src_cr= s->edge_emu_buffer;
2601
    }
2602
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2603
}
2604

    
2605
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2606
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2607
                           int x_offset, int y_offset,
2608
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2609
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2610
                           int list0, int list1){
2611
    MpegEncContext * const s = &h->s;
2612
    qpel_mc_func *qpix_op=  qpix_put;
2613
    h264_chroma_mc_func chroma_op= chroma_put;
2614
    
2615
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2616
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2617
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2618
    x_offset += 8*s->mb_x;
2619
    y_offset += 8*s->mb_y;
2620
    
2621
    if(list0){
2622
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2623
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2624
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2625
                           qpix_op, chroma_op);
2626

    
2627
        qpix_op=  qpix_avg;
2628
        chroma_op= chroma_avg;
2629
    }
2630

    
2631
    if(list1){
2632
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2633
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2634
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2635
                           qpix_op, chroma_op);
2636
    }
2637
}
2638

    
2639
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2640
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2641
                           int x_offset, int y_offset,
2642
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2643
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2644
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2645
                           int list0, int list1){
2646
    MpegEncContext * const s = &h->s;
2647

    
2648
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2649
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2650
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2651
    x_offset += 8*s->mb_x;
2652
    y_offset += 8*s->mb_y;
2653
    
2654
    if(list0 && list1){
2655
        /* don't optimize for luma-only case, since B-frames usually
2656
         * use implicit weights => chroma too. */
2657
        uint8_t *tmp_cb = s->obmc_scratchpad;
2658
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2659
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2660
        int refn0 = h->ref_cache[0][ scan8[n] ];
2661
        int refn1 = h->ref_cache[1][ scan8[n] ];
2662

    
2663
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2664
                    dest_y, dest_cb, dest_cr,
2665
                    x_offset, y_offset, qpix_put, chroma_put);
2666
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2667
                    tmp_y, tmp_cb, tmp_cr,
2668
                    x_offset, y_offset, qpix_put, chroma_put);
2669

    
2670
        if(h->use_weight == 2){
2671
            int weight0 = h->implicit_weight[refn0][refn1];
2672
            int weight1 = 64 - weight0;
2673
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2674
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2675
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2676
        }else{
2677
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2678
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2679
                            h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2680
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2681
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2682
                            h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2683
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2684
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2685
                            h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2686
        }
2687
    }else{
2688
        int list = list1 ? 1 : 0;
2689
        int refn = h->ref_cache[list][ scan8[n] ];
2690
        Picture *ref= &h->ref_list[list][refn];
2691
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2692
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2693
                    qpix_put, chroma_put);
2694

    
2695
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2696
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2697
        if(h->use_weight_chroma){
2698
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2699
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2700
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2701
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2702
        }
2703
    }
2704
}
2705

    
2706
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2707
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2708
                           int x_offset, int y_offset,
2709
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2710
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2711
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2712
                           int list0, int list1){
2713
    if((h->use_weight==2 && list0 && list1
2714
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2715
       || h->use_weight==1)
2716
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2717
                         x_offset, y_offset, qpix_put, chroma_put,
2718
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2719
    else
2720
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2721
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2722
}
2723

    
2724
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2725
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2726
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2727
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2728
    MpegEncContext * const s = &h->s;
2729
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2730
    const int mb_type= s->current_picture.mb_type[mb_xy];
2731
    
2732
    assert(IS_INTER(mb_type));
2733
    
2734
    if(IS_16X16(mb_type)){
2735
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2736
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2737
                &weight_op[0], &weight_avg[0],
2738
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2739
    }else if(IS_16X8(mb_type)){
2740
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2741
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2742
                &weight_op[1], &weight_avg[1],
2743
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2744
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2745
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2746
                &weight_op[1], &weight_avg[1],
2747
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2748
    }else if(IS_8X16(mb_type)){
2749
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2750
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2751
                &weight_op[2], &weight_avg[2],
2752
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2753
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2754
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2755
                &weight_op[2], &weight_avg[2],
2756
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2757
    }else{
2758
        int i;
2759
        
2760
        assert(IS_8X8(mb_type));
2761

    
2762
        for(i=0; i<4; i++){
2763
            const int sub_mb_type= h->sub_mb_type[i];
2764
            const int n= 4*i;
2765
            int x_offset= (i&1)<<2;
2766
            int y_offset= (i&2)<<1;
2767

    
2768
            if(IS_SUB_8X8(sub_mb_type)){
2769
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2770
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2771
                    &weight_op[3], &weight_avg[3],
2772
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2773
            }else if(IS_SUB_8X4(sub_mb_type)){
2774
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2775
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2776
                    &weight_op[4], &weight_avg[4],
2777
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2778
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2779
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2780
                    &weight_op[4], &weight_avg[4],
2781
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2782
            }else if(IS_SUB_4X8(sub_mb_type)){
2783
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2784
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2785
                    &weight_op[5], &weight_avg[5],
2786
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2787
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2788
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2789
                    &weight_op[5], &weight_avg[5],
2790
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2791
            }else{
2792
                int j;
2793
                assert(IS_SUB_4X4(sub_mb_type));
2794
                for(j=0; j<4; j++){
2795
                    int sub_x_offset= x_offset + 2*(j&1);
2796
                    int sub_y_offset= y_offset +   (j&2);
2797
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2798
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2799
                        &weight_op[6], &weight_avg[6],
2800
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2801
                }
2802
            }
2803
        }
2804
    }
2805
}
2806

    
2807
static void decode_init_vlc(H264Context *h){
2808
    static int done = 0;
2809

    
2810
    if (!done) {
2811
        int i;
2812
        done = 1;
2813

    
2814
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2815
                 &chroma_dc_coeff_token_len [0], 1, 1,
2816
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2817

    
2818
        for(i=0; i<4; i++){
2819
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2820
                     &coeff_token_len [i][0], 1, 1,
2821
                     &coeff_token_bits[i][0], 1, 1, 1);
2822
        }
2823

    
2824
        for(i=0; i<3; i++){
2825
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2826
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2827
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2828
        }
2829
        for(i=0; i<15; i++){
2830
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2831
                     &total_zeros_len [i][0], 1, 1,
2832
                     &total_zeros_bits[i][0], 1, 1, 1);
2833
        }
2834

    
2835
        for(i=0; i<6; i++){
2836
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2837
                     &run_len [i][0], 1, 1,
2838
                     &run_bits[i][0], 1, 1, 1);
2839
        }
2840
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2841
                 &run_len [6][0], 1, 1,
2842
                 &run_bits[6][0], 1, 1, 1);
2843
    }
2844
}
2845

    
2846
/**
2847
 * Sets the intra prediction function pointers.
2848
 */
2849
static void init_pred_ptrs(H264Context *h){
2850
//    MpegEncContext * const s = &h->s;
2851

    
2852
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2853
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2854
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2855
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2856
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2857
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2858
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2859
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2860
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2861
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2862
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2863
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2864

    
2865
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2866
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2867
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2868
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2869
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2870
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2871
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2872
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2873
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2874
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2875
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2876
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2877

    
2878
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2879
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2880
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2881
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2882
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2883
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2884
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2885

    
2886
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2887
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2888
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2889
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2890
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2891
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2892
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2893
}
2894

    
2895
static void free_tables(H264Context *h){
2896
    av_freep(&h->intra4x4_pred_mode);
2897
    av_freep(&h->chroma_pred_mode_table);
2898
    av_freep(&h->cbp_table);
2899
    av_freep(&h->mvd_table[0]);
2900
    av_freep(&h->mvd_table[1]);
2901
    av_freep(&h->direct_table);
2902
    av_freep(&h->non_zero_count);
2903
    av_freep(&h->slice_table_base);
2904
    av_freep(&h->top_borders[1]);
2905
    av_freep(&h->top_borders[0]);
2906
    h->slice_table= NULL;
2907

    
2908
    av_freep(&h->mb2b_xy);
2909
    av_freep(&h->mb2b8_xy);
2910

    
2911
    av_freep(&h->dequant4_coeff);
2912
    av_freep(&h->dequant8_coeff);
2913

    
2914
    av_freep(&h->s.obmc_scratchpad);
2915
}
2916

    
2917
/**
2918
 * allocates tables.
2919
 * needs width/height
2920
 */
2921
static int alloc_tables(H264Context *h){
2922
    MpegEncContext * const s = &h->s;
2923
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2924
    int x,y,q;
2925

    
2926
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2927

    
2928
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2929
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2930
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2931
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2932
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2933

    
2934
    if( h->pps.cabac ) {
2935
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2936
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2937
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2938
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2939
    }
2940

    
2941
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2942
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2943

    
2944
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2945
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2946
    for(y=0; y<s->mb_height; y++){
2947
        for(x=0; x<s->mb_width; x++){
2948
            const int mb_xy= x + y*s->mb_stride;
2949
            const int b_xy = 4*x + 4*y*h->b_stride;
2950
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2951
        
2952
            h->mb2b_xy [mb_xy]= b_xy;
2953
            h->mb2b8_xy[mb_xy]= b8_xy;
2954
        }
2955
    }
2956

    
2957
    CHECKED_ALLOCZ(h->dequant4_coeff, 52*16 * sizeof(uint16_t));
2958
    CHECKED_ALLOCZ(h->dequant8_coeff, 52*64 * sizeof(uint16_t));
2959
    memcpy(h->dequant4_coeff, dequant_coeff, 52*16 * sizeof(uint16_t));
2960
    for(q=0; q<52; q++){
2961
        int shift = div6[q];
2962
        int idx = rem6[q];
2963
        if(shift >= 2) // qp<12 are shifted during dequant
2964
            shift -= 2;
2965
        for(x=0; x<64; x++)
2966
            h->dequant8_coeff[q][x] = dequant8_coeff_init[idx][
2967
                dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] << shift;
2968
    }
2969
    if(h->sps.transform_bypass){
2970
        for(x=0; x<16; x++)
2971
            h->dequant4_coeff[0][x] = 1;
2972
        for(x=0; x<64; x++)
2973
            h->dequant8_coeff[0][x] = 1<<2;
2974
    }
2975

    
2976
    s->obmc_scratchpad = NULL;
2977

    
2978
    return 0;
2979
fail:
2980
    free_tables(h);
2981
    return -1;
2982
}
2983

    
2984
static void common_init(H264Context *h){
2985
    MpegEncContext * const s = &h->s;
2986

    
2987
    s->width = s->avctx->width;
2988
    s->height = s->avctx->height;
2989
    s->codec_id= s->avctx->codec->id;
2990
    
2991
    init_pred_ptrs(h);
2992

    
2993
    s->unrestricted_mv=1;
2994
    s->decode=1; //FIXME
2995
}
2996

    
2997
static int decode_init(AVCodecContext *avctx){
2998
    H264Context *h= avctx->priv_data;
2999
    MpegEncContext * const s = &h->s;
3000

    
3001
    MPV_decode_defaults(s);
3002
    
3003
    s->avctx = avctx;
3004
    common_init(h);
3005

    
3006
    s->out_format = FMT_H264;
3007
    s->workaround_bugs= avctx->workaround_bugs;
3008

    
3009
    // set defaults
3010
//    s->decode_mb= ff_h263_decode_mb;
3011
    s->low_delay= 1;
3012
    avctx->pix_fmt= PIX_FMT_YUV420P;
3013

    
3014
    decode_init_vlc(h);
3015
    
3016
    if(avctx->extradata_size > 0 && avctx->extradata &&
3017
       *(char *)avctx->extradata == 1){
3018
        h->is_avc = 1;
3019
        h->got_avcC = 0;
3020
    } else {
3021
        h->is_avc = 0;
3022
    }
3023

    
3024
    return 0;
3025
}
3026

    
3027
static void frame_start(H264Context *h){
3028
    MpegEncContext * const s = &h->s;
3029
    int i;
3030

    
3031
    MPV_frame_start(s, s->avctx);
3032
    ff_er_frame_start(s);
3033

    
3034
    assert(s->linesize && s->uvlinesize);
3035

    
3036
    for(i=0; i<16; i++){
3037
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3038
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3039
    }
3040
    for(i=0; i<4; i++){
3041
        h->block_offset[16+i]=
3042
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3043
        h->block_offset[24+16+i]=
3044
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3045
    }
3046

    
3047
    /* can't be in alloc_tables because linesize isn't known there.
3048
     * FIXME: redo bipred weight to not require extra buffer? */
3049
    if(!s->obmc_scratchpad)
3050
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3051

    
3052
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3053
}
3054

    
3055
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3056
    MpegEncContext * const s = &h->s;
3057
    int i;
3058
    
3059
    src_y  -=   linesize;
3060
    src_cb -= uvlinesize;
3061
    src_cr -= uvlinesize;
3062

    
3063
    // There are two lines saved, the line above the the top macroblock of a pair,
3064
    // and the line above the bottom macroblock
3065
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3066
    for(i=1; i<17; i++){
3067
        h->left_border[i]= src_y[15+i*  linesize];
3068
    }
3069
    
3070
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3071
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3072

    
3073
    if(!(s->flags&CODEC_FLAG_GRAY)){
3074
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3075
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3076
        for(i=1; i<9; i++){
3077
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3078
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3079
        }
3080
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3081
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3082
    }
3083
}
3084

    
3085
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){
3086
    MpegEncContext * const s = &h->s;
3087
    int temp8, i;
3088
    uint64_t temp64;
3089
    int deblock_left = (s->mb_x > 0);
3090
    int deblock_top  = (s->mb_y > 0);
3091

    
3092
    src_y  -=   linesize + 1;
3093
    src_cb -= uvlinesize + 1;
3094
    src_cr -= uvlinesize + 1;
3095

    
3096
#define XCHG(a,b,t,xchg)\
3097
t= a;\
3098
if(xchg)\
3099
    a= b;\
3100
b= t;
3101

    
3102
    if(deblock_left){
3103
        for(i = !deblock_top; i<17; i++){
3104
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3105
        }
3106
    }
3107

    
3108
    if(deblock_top){
3109
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3110
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3111
        if(s->mb_x+1 < s->mb_width){
3112
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3113
        }
3114
    }
3115

    
3116
    if(!(s->flags&CODEC_FLAG_GRAY)){
3117
        if(deblock_left){
3118
            for(i = !deblock_top; i<9; i++){
3119
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3120
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3121
            }
3122
        }
3123
        if(deblock_top){
3124
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3125
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3126
        }
3127
    }
3128
}
3129

    
3130
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3131
    MpegEncContext * const s = &h->s;
3132
    int i;
3133
    
3134
    src_y  -= 2 *   linesize;
3135
    src_cb -= 2 * uvlinesize;
3136
    src_cr -= 2 * uvlinesize;
3137

    
3138
    // There are two lines saved, the line above the the top macroblock of a pair,
3139
    // and the line above the bottom macroblock
3140
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3141
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3142
    for(i=2; i<34; i++){
3143
        h->left_border[i]= src_y[15+i*  linesize];
3144
    }
3145
    
3146
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3147
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3148
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3149
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3150

    
3151
    if(!(s->flags&CODEC_FLAG_GRAY)){
3152
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3153
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3154
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3155
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3156
        for(i=2; i<18; i++){
3157
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3158
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3159
        }
3160
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3161
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3162
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3163
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3164
    }
3165
}
3166

    
3167
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){
3168
    MpegEncContext * const s = &h->s;
3169
    int temp8, i;
3170
    uint64_t temp64;
3171
    int deblock_left = (s->mb_x > 0);
3172
    int deblock_top  = (s->mb_y > 0);
3173

    
3174
    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);
3175

    
3176
    src_y  -= 2 *   linesize + 1;
3177
    src_cb -= 2 * uvlinesize + 1;
3178
    src_cr -= 2 * uvlinesize + 1;
3179

    
3180
#define XCHG(a,b,t,xchg)\
3181
t= a;\
3182
if(xchg)\
3183
    a= b;\
3184
b= t;
3185

    
3186
    if(deblock_left){
3187
        for(i = (!deblock_top)<<1; i<34; i++){
3188
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3189
        }
3190
    }
3191

    
3192
    if(deblock_top){
3193
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3194
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3195
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3196
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3197
    }
3198

    
3199
    if(!(s->flags&CODEC_FLAG_GRAY)){
3200
        if(deblock_left){
3201
            for(i = (!deblock_top) << 1; i<18; i++){
3202
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3203
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3204
            }
3205
        }
3206
        if(deblock_top){
3207
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3208
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3209
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3210
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3211
        }
3212
    }
3213
}
3214

    
3215
static void hl_decode_mb(H264Context *h){
3216
    MpegEncContext * const s = &h->s;
3217
    const int mb_x= s->mb_x;
3218
    const int mb_y= s->mb_y;
3219
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3220
    const int mb_type= s->current_picture.mb_type[mb_xy];
3221
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3222
    int linesize, uvlinesize /*dct_offset*/;
3223
    int i;
3224
    int *block_offset = &h->block_offset[0];
3225
    const unsigned int bottom = mb_y & 1;
3226
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3227
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3228

    
3229
    if(!s->decode)
3230
        return;
3231

    
3232
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3233
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3234
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3235

    
3236
    if (h->mb_field_decoding_flag) {
3237
        linesize = s->linesize * 2;
3238
        uvlinesize = s->uvlinesize * 2;
3239
        block_offset = &h->block_offset[24];
3240
        if(mb_y&1){ //FIXME move out of this func?
3241
            dest_y -= s->linesize*15;
3242
            dest_cb-= s->uvlinesize*7;
3243
            dest_cr-= s->uvlinesize*7;
3244
        }
3245
    } else {
3246
        linesize = s->linesize;
3247
        uvlinesize = s->uvlinesize;
3248
//        dct_offset = s->linesize * 16;
3249
    }
3250
    
3251
    idct_add = transform_bypass
3252
             ? IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4
3253
             : IS_8x8DCT(mb_type) ? s->dsp.h264_idct8_add : s->dsp.h264_idct_add;
3254

    
3255
    if (IS_INTRA_PCM(mb_type)) {
3256
        unsigned int x, y;
3257

    
3258
        // The pixels are stored in h->mb array in the same order as levels,
3259
        // copy them in output in the correct order.
3260
        for(i=0; i<16; i++) {
3261
            for (y=0; y<4; y++) {
3262
                for (x=0; x<4; x++) {
3263
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3264
                }
3265
            }
3266
        }
3267
        for(i=16; i<16+4; i++) {
3268
            for (y=0; y<4; y++) {
3269
                for (x=0; x<4; x++) {
3270
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3271
                }
3272
            }
3273
        }
3274
        for(i=20; i<20+4; i++) {
3275
            for (y=0; y<4; y++) {
3276
                for (x=0; x<4; x++) {
3277
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3278
                }
3279
            }
3280
        }
3281
    } else {
3282
        if(IS_INTRA(mb_type)){
3283
            if(h->deblocking_filter) {
3284
                if (h->mb_aff_frame) {
3285
                    if (!bottom)
3286
                        xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3287
                } else {
3288
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3289
                }
3290
            }
3291

    
3292
            if(!(s->flags&CODEC_FLAG_GRAY)){
3293
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3294
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3295
            }
3296

    
3297
            if(IS_INTRA4x4(mb_type)){
3298
                if(!s->encoding){
3299
                    if(IS_8x8DCT(mb_type)){
3300
                        for(i=0; i<16; i+=4){
3301
                            uint8_t * const ptr= dest_y + block_offset[i];
3302
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3303
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3304
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3305
                            if(h->non_zero_count_cache[ scan8[i] ])
3306
                                idct_add(ptr, h->mb + i*16, linesize);
3307
                        }
3308
                    }else
3309
                    for(i=0; i<16; i++){
3310
                        uint8_t * const ptr= dest_y + block_offset[i];
3311
                        uint8_t *topright;
3312
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3313
                        int tr;
3314

    
3315
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3316
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3317
                            assert(mb_y || linesize <= block_offset[i]);
3318
                            if(!topright_avail){
3319
                                tr= ptr[3 - linesize]*0x01010101;
3320
                                topright= (uint8_t*) &tr;
3321
                            }else 
3322
                                topright= ptr + 4 - linesize;
3323
                        }else
3324
                            topright= NULL;
3325

    
3326
                        h->pred4x4[ dir ](ptr, topright, linesize);
3327
                        if(h->non_zero_count_cache[ scan8[i] ]){
3328
                            if(s->codec_id == CODEC_ID_H264)
3329
                                idct_add(ptr, h->mb + i*16, linesize);
3330
                            else
3331
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3332
                        }
3333
                    }
3334
                }
3335
            }else{
3336
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3337
                if(s->codec_id == CODEC_ID_H264){
3338
                    if(!transform_bypass)
3339
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3340
                }else
3341
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3342
            }
3343
            if(h->deblocking_filter) {
3344
                if (h->mb_aff_frame) {
3345
                    if (bottom) {
3346
                        uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
3347
                        uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3348
                        uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3349
                        s->mb_y--;
3350
                        xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3351
                        s->mb_y++;
3352
                    }
3353
                } else {
3354
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3355
                }
3356
            }
3357
        }else if(s->codec_id == CODEC_ID_H264){
3358
            hl_motion(h, dest_y, dest_cb, dest_cr,
3359
                      s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
3360
                      s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3361
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3362
        }
3363

    
3364

    
3365
        if(!IS_INTRA4x4(mb_type)){
3366
            if(s->codec_id == CODEC_ID_H264){
3367
                const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3368
                for(i=0; i<16; i+=di){
3369
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3370
                        uint8_t * const ptr= dest_y + block_offset[i];
3371
                        idct_add(ptr, h->mb + i*16, linesize);
3372
                    }
3373
                }
3374
            }else{
3375
                for(i=0; i<16; i++){
3376
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3377
                        uint8_t * const ptr= dest_y + block_offset[i];
3378
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3379
                    }
3380
                }
3381
            }
3382
        }
3383

    
3384
        if(!(s->flags&CODEC_FLAG_GRAY)){
3385
            idct_add = transform_bypass ? s->dsp.add_pixels4 : s->dsp.h264_idct_add;
3386
            if(!transform_bypass){
3387
                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3388
                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3389
            }
3390
            if(s->codec_id == CODEC_ID_H264){
3391
                for(i=16; i<16+4; i++){
3392
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3393
                        uint8_t * const ptr= dest_cb + block_offset[i];
3394
                        idct_add(ptr, h->mb + i*16, uvlinesize);
3395
                    }
3396
                }
3397
                for(i=20; i<20+4; i++){
3398
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3399
                        uint8_t * const ptr= dest_cr + block_offset[i];
3400
                        idct_add(ptr, h->mb + i*16, uvlinesize);
3401
                    }
3402
                }
3403
            }else{
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
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
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
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3414
                    }
3415
                }
3416
            }
3417
        }
3418
    }
3419
    if(h->deblocking_filter) {
3420
        if (h->mb_aff_frame) {
3421
            const int mb_y = s->mb_y - 1;
3422
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3423
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3424
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3425
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3426
            uint8_t tmp = s->current_picture.data[1][384];
3427
            if (!bottom) return;
3428
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3429
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3430
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3431

    
3432
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3433
            // TODO deblock a pair
3434
            // top 
3435
            s->mb_y--;
3436
            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);
3437
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3438
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3439
            if (tmp != s->current_picture.data[1][384]) {
3440
                tprintf("modified pixel 8,1 (1)\n");
3441
            }
3442
            // bottom
3443
            s->mb_y++;
3444
            tprintf("call mbaff filter_mb\n");
3445
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3446
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3447
            if (tmp != s->current_picture.data[1][384]) {
3448
                tprintf("modified pixel 8,1 (2)\n");
3449
            }
3450
        } else {
3451
            tprintf("call filter_mb\n");
3452
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3453
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3454
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3455
        }
3456
    }
3457
}
3458

    
3459
/**
3460
 * fills the default_ref_list.
3461
 */
3462
static int fill_default_ref_list(H264Context *h){
3463
    MpegEncContext * const s = &h->s;
3464
    int i;
3465
    int smallest_poc_greater_than_current = -1;
3466
    Picture sorted_short_ref[32];
3467
    
3468
    if(h->slice_type==B_TYPE){
3469
        int out_i;
3470
        int limit= INT_MIN;
3471

    
3472
        /* sort frame according to poc in B slice */
3473
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3474
            int best_i=INT_MIN;
3475
            int best_poc=INT_MAX;
3476

    
3477
            for(i=0; i<h->short_ref_count; i++){
3478
                const int poc= h->short_ref[i]->poc;
3479
                if(poc > limit && poc < best_poc){
3480
                    best_poc= poc;
3481
                    best_i= i;
3482
                }
3483
            }
3484
            
3485
            assert(best_i != INT_MIN);
3486
            
3487
            limit= best_poc;
3488
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3489
            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);
3490
            if (-1 == smallest_poc_greater_than_current) {
3491
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3492
                    smallest_poc_greater_than_current = out_i;
3493
                }
3494
            }
3495
        }
3496
    }
3497

    
3498
    if(s->picture_structure == PICT_FRAME){
3499
        if(h->slice_type==B_TYPE){
3500
            int list;
3501
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3502

    
3503
            // find the largest poc
3504
            for(list=0; list<2; list++){
3505
                int index = 0;
3506
                int j= -99;
3507
                int step= list ? -1 : 1;
3508

    
3509
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3510
                    while(j<0 || j>= h->short_ref_count){
3511
                        if(j != -99 && step == (list ? -1 : 1))
3512
                            return -1;
3513
                        step = -step;
3514
                        j= smallest_poc_greater_than_current + (step>>1);
3515
                    }
3516
                    if(sorted_short_ref[j].reference != 3) continue;
3517
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3518
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3519
                }
3520

    
3521
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3522
                    if(h->long_ref[i] == NULL) continue;
3523
                    if(h->long_ref[i]->reference != 3) continue;
3524

    
3525
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3526
                    h->default_ref_list[ list ][index++].pic_id= i;;
3527
                }
3528
                
3529
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3530
                    // swap the two first elements of L1 when
3531
                    // L0 and L1 are identical
3532
                    Picture temp= h->default_ref_list[1][0];
3533
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3534
                    h->default_ref_list[1][1] = temp;
3535
                }
3536

    
3537
                if(index < h->ref_count[ list ])
3538
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3539
            }
3540
        }else{
3541
            int index=0;
3542
            for(i=0; i<h->short_ref_count; i++){
3543
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3544
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3545
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3546
            }
3547
            for(i = 0; i < 16; i++){
3548
                if(h->long_ref[i] == NULL) continue;
3549
                if(h->long_ref[i]->reference != 3) continue;
3550
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3551
                h->default_ref_list[0][index++].pic_id= i;;
3552
            }
3553
            if(index < h->ref_count[0])
3554
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3555
        }
3556
    }else{ //FIELD
3557
        if(h->slice_type==B_TYPE){
3558
        }else{
3559
            //FIXME second field balh
3560
        }
3561
    }
3562
#ifdef TRACE
3563
    for (i=0; i<h->ref_count[0]; i++) {
3564
        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]);
3565
    }
3566
    if(h->slice_type==B_TYPE){
3567
        for (i=0; i<h->ref_count[1]; i++) {
3568
            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]);
3569
        }
3570
    }
3571
#endif
3572
    return 0;
3573
}
3574

    
3575
static void print_short_term(H264Context *h);
3576
static void print_long_term(H264Context *h);
3577

    
3578
static int decode_ref_pic_list_reordering(H264Context *h){
3579
    MpegEncContext * const s = &h->s;
3580
    int list, index;
3581
    
3582
    print_short_term(h);
3583
    print_long_term(h);
3584
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3585
    
3586
    for(list=0; list<2; list++){
3587
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3588

    
3589
        if(get_bits1(&s->gb)){
3590
            int pred= h->curr_pic_num;
3591

    
3592
            for(index=0; ; index++){
3593
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3594
                int pic_id;
3595
                int i;
3596
                Picture *ref = NULL;
3597
                
3598
                if(reordering_of_pic_nums_idc==3) 
3599
                    break;
3600
                
3601
                if(index >= h->ref_count[list]){
3602
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3603
                    return -1;
3604
                }
3605
                
3606
                if(reordering_of_pic_nums_idc<3){
3607
                    if(reordering_of_pic_nums_idc<2){
3608
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3609

    
3610
                        if(abs_diff_pic_num >= h->max_pic_num){
3611
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3612
                            return -1;
3613
                        }
3614

    
3615
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3616
                        else                                pred+= abs_diff_pic_num;
3617
                        pred &= h->max_pic_num - 1;
3618
                    
3619
                        for(i= h->short_ref_count-1; i>=0; i--){
3620
                            ref = h->short_ref[i];
3621
                            assert(ref->reference == 3);
3622
                            assert(!ref->long_ref);
3623
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3624
                                break;
3625
                        }
3626
                        if(i>=0)
3627
                            ref->pic_id= ref->frame_num;
3628
                    }else{
3629
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3630
                        ref = h->long_ref[pic_id];
3631
                        ref->pic_id= pic_id;
3632
                        assert(ref->reference == 3);
3633
                        assert(ref->long_ref);
3634
                        i=0;
3635
                    }
3636

    
3637
                    if (i < 0) {
3638
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3639
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3640
                    } else {
3641
                        for(i=index; i+1<h->ref_count[list]; i++){
3642
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3643
                                break;
3644
                        }
3645
                        for(; i > index; i--){
3646
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3647
                        }
3648
                        h->ref_list[list][index]= *ref;
3649
                    }
3650
                }else{
3651
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3652
                    return -1;
3653
                }
3654
            }
3655
        }
3656

    
3657
        if(h->slice_type!=B_TYPE) break;
3658
    }
3659
    for(list=0; list<2; list++){
3660
        for(index= 0; index < h->ref_count[list]; index++){
3661
            if(!h->ref_list[list][index].data[0])
3662
                h->ref_list[list][index]= s->current_picture;
3663
        }
3664
        if(h->slice_type!=B_TYPE) break;
3665
    }
3666
    
3667
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3668
        direct_dist_scale_factor(h);
3669
    direct_ref_list_init(h);
3670
    return 0;    
3671
}
3672

    
3673
static int pred_weight_table(H264Context *h){
3674
    MpegEncContext * const s = &h->s;
3675
    int list, i;
3676
    int luma_def, chroma_def;
3677
    
3678
    h->use_weight= 0;
3679
    h->use_weight_chroma= 0;
3680
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3681
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3682
    luma_def = 1<<h->luma_log2_weight_denom;
3683
    chroma_def = 1<<h->chroma_log2_weight_denom;
3684

    
3685
    for(list=0; list<2; list++){
3686
        for(i=0; i<h->ref_count[list]; i++){
3687
            int luma_weight_flag, chroma_weight_flag;
3688
            
3689
            luma_weight_flag= get_bits1(&s->gb);
3690
            if(luma_weight_flag){
3691
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3692
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3693
                if(   h->luma_weight[list][i] != luma_def
3694
                   || h->luma_offset[list][i] != 0)
3695
                    h->use_weight= 1;
3696
            }else{
3697
                h->luma_weight[list][i]= luma_def;
3698
                h->luma_offset[list][i]= 0;
3699
            }
3700

    
3701
            chroma_weight_flag= get_bits1(&s->gb);
3702
            if(chroma_weight_flag){
3703
                int j;
3704
                for(j=0; j<2; j++){
3705
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3706
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3707
                    if(   h->chroma_weight[list][i][j] != chroma_def
3708
                       || h->chroma_offset[list][i][j] != 0)
3709
                        h->use_weight_chroma= 1;
3710
                }
3711
            }else{
3712
                int j;
3713
                for(j=0; j<2; j++){
3714
                    h->chroma_weight[list][i][j]= chroma_def;
3715
                    h->chroma_offset[list][i][j]= 0;
3716
                }
3717
            }
3718
        }
3719
        if(h->slice_type != B_TYPE) break;
3720
    }
3721
    h->use_weight= h->use_weight || h->use_weight_chroma;
3722
    return 0;
3723
}
3724

    
3725
static void implicit_weight_table(H264Context *h){
3726
    MpegEncContext * const s = &h->s;
3727
    int ref0, ref1;
3728
    int cur_poc = s->current_picture_ptr->poc;
3729

    
3730
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3731
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3732
        h->use_weight= 0;
3733
        h->use_weight_chroma= 0;
3734
        return;
3735
    }
3736

    
3737
    h->use_weight= 2;
3738
    h->use_weight_chroma= 2;
3739
    h->luma_log2_weight_denom= 5;
3740
    h->chroma_log2_weight_denom= 5;
3741

    
3742
    /* FIXME: MBAFF */
3743
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3744
        int poc0 = h->ref_list[0][ref0].poc;
3745
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3746
            int poc1 = h->ref_list[1][ref1].poc;
3747
            int td = clip(poc1 - poc0, -128, 127);
3748
            if(td){
3749
                int tb = clip(cur_poc - poc0, -128, 127);
3750
                int tx = (16384 + (ABS(td) >> 1)) / td;
3751
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3752
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3753
                    h->implicit_weight[ref0][ref1] = 32;
3754
                else
3755
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3756
            }else
3757
                h->implicit_weight[ref0][ref1] = 32;
3758
        }
3759
    }
3760
}
3761

    
3762
static inline void unreference_pic(H264Context *h, Picture *pic){
3763
    int i;
3764
    pic->reference=0;
3765
    if(pic == h->delayed_output_pic)
3766
        pic->reference=1;
3767
    else{
3768
        for(i = 0; h->delayed_pic[i]; i++)
3769
            if(pic == h->delayed_pic[i]){
3770
                pic->reference=1;
3771
                break;
3772
            }
3773
    }
3774
}
3775

    
3776
/**
3777
 * instantaneous decoder refresh.
3778
 */
3779
static void idr(H264Context *h){
3780
    int i;
3781

    
3782
    for(i=0; i<16; i++){
3783
        if (h->long_ref[i] != NULL) {
3784
            unreference_pic(h, h->long_ref[i]);
3785
            h->long_ref[i]= NULL;
3786
        }
3787
    }
3788
    h->long_ref_count=0;
3789

    
3790
    for(i=0; i<h->short_ref_count; i++){
3791
        unreference_pic(h, h->short_ref[i]);
3792
        h->short_ref[i]= NULL;
3793
    }
3794
    h->short_ref_count=0;
3795
}
3796

    
3797
/* forget old pics after a seek */
3798
static void flush_dpb(AVCodecContext *avctx){
3799
    H264Context *h= avctx->priv_data;
3800
    int i;
3801
    for(i=0; i<16; i++)
3802
        h->delayed_pic[i]= NULL;
3803
    h->delayed_output_pic= NULL;
3804
    idr(h);
3805
    if(h->s.current_picture_ptr)
3806
        h->s.current_picture_ptr->reference= 0;
3807
}
3808

    
3809
/**
3810
 *
3811
 * @return the removed picture or NULL if an error occurs
3812
 */
3813
static Picture * remove_short(H264Context *h, int frame_num){
3814
    MpegEncContext * const s = &h->s;
3815
    int i;
3816
    
3817
    if(s->avctx->debug&FF_DEBUG_MMCO)
3818
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3819
    
3820
    for(i=0; i<h->short_ref_count; i++){
3821
        Picture *pic= h->short_ref[i];
3822
        if(s->avctx->debug&FF_DEBUG_MMCO)
3823
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3824
        if(pic->frame_num == frame_num){
3825
            h->short_ref[i]= NULL;
3826
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3827
            h->short_ref_count--;
3828
            return pic;
3829
        }
3830
    }
3831
    return NULL;
3832
}
3833

    
3834
/**
3835
 *
3836
 * @return the removed picture or NULL if an error occurs
3837
 */
3838
static Picture * remove_long(H264Context *h, int i){
3839
    Picture *pic;
3840

    
3841
    pic= h->long_ref[i];
3842
    h->long_ref[i]= NULL;
3843
    if(pic) h->long_ref_count--;
3844

    
3845
    return pic;
3846
}
3847

    
3848
/**
3849
 * print short term list
3850
 */
3851
static void print_short_term(H264Context *h) {
3852
    uint32_t i;
3853
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3854
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3855
        for(i=0; i<h->short_ref_count; i++){
3856
            Picture *pic= h->short_ref[i];
3857
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3858
        }
3859
    }
3860
}
3861

    
3862
/**
3863
 * print long term list
3864
 */
3865
static void print_long_term(H264Context *h) {
3866
    uint32_t i;
3867
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3868
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3869
        for(i = 0; i < 16; i++){
3870
            Picture *pic= h->long_ref[i];
3871
            if (pic) {
3872
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3873
            }
3874
        }
3875
    }
3876
}
3877

    
3878
/**
3879
 * Executes the reference picture marking (memory management control operations).
3880
 */
3881
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3882
    MpegEncContext * const s = &h->s;
3883
    int i, j;
3884
    int current_is_long=0;
3885
    Picture *pic;
3886
    
3887
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3888
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3889
        
3890
    for(i=0; i<mmco_count; i++){
3891
        if(s->avctx->debug&FF_DEBUG_MMCO)
3892
            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);
3893

    
3894
        switch(mmco[i].opcode){
3895
        case MMCO_SHORT2UNUSED:
3896
            pic= remove_short(h, mmco[i].short_frame_num);
3897
            if(pic)
3898
                unreference_pic(h, pic);
3899
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3900
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
3901
            break;
3902
        case MMCO_SHORT2LONG:
3903
            pic= remove_long(h, mmco[i].long_index);
3904
            if(pic) unreference_pic(h, pic);
3905
            
3906
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3907
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3908
            h->long_ref_count++;
3909
            break;
3910
        case MMCO_LONG2UNUSED:
3911
            pic= remove_long(h, mmco[i].long_index);
3912
            if(pic)
3913
                unreference_pic(h, pic);
3914
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3915
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
3916
            break;
3917
        case MMCO_LONG:
3918
            pic= remove_long(h, mmco[i].long_index);
3919
            if(pic) unreference_pic(h, pic);
3920
            
3921
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3922
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3923
            h->long_ref_count++;
3924
            
3925
            current_is_long=1;
3926
            break;
3927
        case MMCO_SET_MAX_LONG:
3928
            assert(mmco[i].long_index <= 16);
3929
            // just remove the long term which index is greater than new max
3930
            for(j = mmco[i].long_index; j<16; j++){
3931
                pic = remove_long(h, j);
3932
                if (pic) unreference_pic(h, pic);
3933
            }
3934
            break;
3935
        case MMCO_RESET:
3936
            while(h->short_ref_count){
3937
                pic= remove_short(h, h->short_ref[0]->frame_num);
3938
                unreference_pic(h, pic);
3939
            }
3940
            for(j = 0; j < 16; j++) {
3941
                pic= remove_long(h, j);
3942
                if(pic) unreference_pic(h, pic);
3943
            }
3944
            break;
3945
        default: assert(0);
3946
        }
3947
    }
3948
    
3949
    if(!current_is_long){
3950
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3951
        if(pic){
3952
            unreference_pic(h, pic);
3953
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3954
        }
3955
        
3956
        if(h->short_ref_count)
3957
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3958

    
3959
        h->short_ref[0]= s->current_picture_ptr;
3960
        h->short_ref[0]->long_ref=0;
3961
        h->short_ref_count++;
3962
    }
3963
    
3964
    print_short_term(h);
3965
    print_long_term(h);
3966
    return 0; 
3967
}
3968

    
3969
static int decode_ref_pic_marking(H264Context *h){
3970
    MpegEncContext * const s = &h->s;
3971
    int i;
3972
    
3973
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3974
        s->broken_link= get_bits1(&s->gb) -1;
3975
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3976
        if(h->mmco[0].long_index == -1)
3977
            h->mmco_index= 0;
3978
        else{
3979
            h->mmco[0].opcode= MMCO_LONG;
3980
            h->mmco_index= 1;
3981
        } 
3982
    }else{
3983
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3984
            for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3985
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3986

    
3987
                h->mmco[i].opcode= opcode;
3988
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3989
                    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
3990
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3991
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3992
                        return -1;
3993
                    }*/
3994
                }
3995
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3996
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
3997
                    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){
3998
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3999
                        return -1;
4000
                    }
4001
                }
4002
                    
4003
                if(opcode > MMCO_LONG){
4004
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4005
                    return -1;
4006
                }
4007
                if(opcode == MMCO_END)
4008
                    break;
4009
            }
4010
            h->mmco_index= i;
4011
        }else{
4012
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4013

    
4014
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4015
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4016
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4017
                h->mmco_index= 1;
4018
            }else
4019
                h->mmco_index= 0;
4020
        }
4021
    }
4022
    
4023
    return 0; 
4024
}
4025

    
4026
static int init_poc(H264Context *h){
4027
    MpegEncContext * const s = &h->s;
4028
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4029
    int field_poc[2];
4030

    
4031
    if(h->nal_unit_type == NAL_IDR_SLICE){
4032
        h->frame_num_offset= 0;
4033
    }else{
4034
        if(h->frame_num < h->prev_frame_num)
4035
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4036
        else
4037
            h->frame_num_offset= h->prev_frame_num_offset;
4038
    }
4039

    
4040
    if(h->sps.poc_type==0){
4041
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4042

    
4043
        if(h->nal_unit_type == NAL_IDR_SLICE){
4044
             h->prev_poc_msb=
4045
             h->prev_poc_lsb= 0;
4046
        }
4047

    
4048
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4049
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4050
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4051
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4052
        else
4053
            h->poc_msb = h->prev_poc_msb;
4054
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4055
        field_poc[0] = 
4056
        field_poc[1] = h->poc_msb + h->poc_lsb;
4057
        if(s->picture_structure == PICT_FRAME) 
4058
            field_poc[1] += h->delta_poc_bottom;
4059
    }else if(h->sps.poc_type==1){
4060
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4061
        int i;
4062

    
4063
        if(h->sps.poc_cycle_length != 0)
4064
            abs_frame_num = h->frame_num_offset + h->frame_num;
4065
        else
4066
            abs_frame_num = 0;
4067

    
4068
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4069
            abs_frame_num--;
4070
            
4071
        expected_delta_per_poc_cycle = 0;
4072
        for(i=0; i < h->sps.poc_cycle_length; i++)
4073
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4074

    
4075
        if(abs_frame_num > 0){
4076
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4077
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4078

    
4079
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4080
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4081
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4082
        } else
4083
            expectedpoc = 0;
4084

    
4085
        if(h->nal_ref_idc == 0) 
4086
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4087
        
4088
        field_poc[0] = expectedpoc + h->delta_poc[0];
4089
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4090

    
4091
        if(s->picture_structure == PICT_FRAME)
4092
            field_poc[1] += h->delta_poc[1];
4093
    }else{
4094
        int poc;
4095
        if(h->nal_unit_type == NAL_IDR_SLICE){
4096
            poc= 0;
4097
        }else{
4098
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4099
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4100
        }
4101
        field_poc[0]= poc;
4102
        field_poc[1]= poc;
4103
    }
4104
    
4105
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4106
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4107
    if(s->picture_structure != PICT_TOP_FIELD)
4108
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4109
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4110
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4111

    
4112
    return 0;
4113
}
4114

    
4115
/**
4116
 * decodes a slice header.
4117
 * this will allso call MPV_common_init() and frame_start() as needed
4118
 */
4119
static int decode_slice_header(H264Context *h){
4120
    MpegEncContext * const s = &h->s;
4121
    int first_mb_in_slice, pps_id;
4122
    int num_ref_idx_active_override_flag;
4123
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4124
    int slice_type;
4125
    int default_ref_list_done = 0;
4126

    
4127
    s->current_picture.reference= h->nal_ref_idc != 0;
4128
    s->dropable= h->nal_ref_idc == 0;
4129

    
4130
    first_mb_in_slice= get_ue_golomb(&s->gb);
4131

    
4132
    slice_type= get_ue_golomb(&s->gb);
4133
    if(slice_type > 9){
4134
        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);
4135
        return -1;
4136
    }
4137
    if(slice_type > 4){
4138
        slice_type -= 5;
4139
        h->slice_type_fixed=1;
4140
    }else
4141
        h->slice_type_fixed=0;
4142
    
4143
    slice_type= slice_type_map[ slice_type ];
4144
    if (slice_type == I_TYPE
4145
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4146
        default_ref_list_done = 1;
4147
    }
4148
    h->slice_type= slice_type;
4149

    
4150
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4151
        
4152
    pps_id= get_ue_golomb(&s->gb);
4153
    if(pps_id>255){
4154
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4155
        return -1;
4156
    }
4157
    h->pps= h->pps_buffer[pps_id];
4158
    if(h->pps.slice_group_count == 0){
4159
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4160
        return -1;
4161
    }
4162

    
4163
    h->sps= h->sps_buffer[ h->pps.sps_id ];
4164
    if(h->sps.log2_max_frame_num == 0){
4165
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4166
        return -1;
4167
    }
4168
    
4169
    s->mb_width= h->sps.mb_width;
4170
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
4171
    
4172
    h->b_stride=  s->mb_width*4 + 1;
4173
    h->b8_stride= s->mb_width*2 + 1;
4174

    
4175
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
4176
    if(h->sps.frame_mbs_only_flag)
4177
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
4178
    else
4179
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
4180
    
4181
    if (s->context_initialized 
4182
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
4183
        free_tables(h);
4184
        MPV_common_end(s);
4185
    }
4186
    if (!s->context_initialized) {
4187
        if (MPV_common_init(s) < 0)
4188
            return -1;
4189
            
4190
        if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
4191
            memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
4192
            memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
4193
        }else{
4194
            int i;
4195
            for(i=0; i<16; i++){
4196
#define T(x) (x>>2) | ((x<<2) & 0xF)
4197
                h->zigzag_scan[i] = T(zigzag_scan[i]);
4198
                h-> field_scan[i] = T( field_scan[i]);
4199
            }
4200
        }
4201
        if(h->sps.transform_bypass){ //FIXME same ugly
4202
            h->zigzag_scan_q0 = zigzag_scan;
4203
            h->field_scan_q0 = field_scan;
4204
        }else{
4205
            h->zigzag_scan_q0 = h->zigzag_scan;
4206
            h->field_scan_q0 = h->field_scan;
4207
        }
4208

    
4209
        alloc_tables(h);
4210

    
4211
        s->avctx->width = s->width;
4212
        s->avctx->height = s->height;
4213
        s->avctx->sample_aspect_ratio= h->sps.sar;
4214
        if(!s->avctx->sample_aspect_ratio.den)
4215
            s->avctx->sample_aspect_ratio.den = 1;
4216

    
4217
        if(h->sps.timing_info_present_flag){
4218
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
4219
        }
4220
    }
4221

    
4222
    if(h->slice_num == 0){
4223
        frame_start(h);
4224
    }
4225

    
4226
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
4227
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
4228

    
4229
    h->mb_aff_frame = 0;
4230
    if(h->sps.frame_mbs_only_flag){
4231
        s->picture_structure= PICT_FRAME;
4232
    }else{
4233
        if(get_bits1(&s->gb)) { //field_pic_flag
4234
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
4235
        } else {
4236
            s->picture_structure= PICT_FRAME;
4237
            first_mb_in_slice <<= h->sps.mb_aff;
4238
            h->mb_aff_frame = h->sps.mb_aff;
4239
        }
4240
    }
4241

    
4242
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
4243
    s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
4244
    if(s->mb_y >= s->mb_height){
4245
        return -1;
4246
    }
4247
    
4248
    if(s->picture_structure==PICT_FRAME){
4249
        h->curr_pic_num=   h->frame_num;
4250
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
4251
    }else{
4252
        h->curr_pic_num= 2*h->frame_num;
4253
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
4254
    }
4255
        
4256
    if(h->nal_unit_type == NAL_IDR_SLICE){
4257
        get_ue_golomb(&s->gb); /* idr_pic_id */
4258
    }
4259
   
4260
    if(h->sps.poc_type==0){
4261
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
4262
        
4263
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
4264
            h->delta_poc_bottom= get_se_golomb(&s->gb);
4265
        }
4266
    }
4267
    
4268
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
4269
        h->delta_poc[0]= get_se_golomb(&s->gb);
4270
        
4271
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
4272
            h->delta_poc[1]= get_se_golomb(&s->gb);
4273
    }
4274
    
4275
    init_poc(h);
4276
    
4277
    if(h->pps.redundant_pic_cnt_present){
4278
        h->redundant_pic_count= get_ue_golomb(&s->gb);
4279
    }
4280

    
4281
    //set defaults, might be overriden a few line later
4282
    h->ref_count[0]= h->pps.ref_count[0];
4283
    h->ref_count[1]= h->pps.ref_count[1];
4284

    
4285
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
4286
        if(h->slice_type == B_TYPE){
4287
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
4288
        }
4289
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
4290
    
4291
        if(num_ref_idx_active_override_flag){
4292
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
4293
            if(h->slice_type==B_TYPE)
4294
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
4295

    
4296
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
4297
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
4298
                return -1;
4299
            }
4300
        }
4301
    }
4302

    
4303
    if(!default_ref_list_done){
4304
        fill_default_ref_list(h);
4305
    }
4306

    
4307
    if(decode_ref_pic_list_reordering(h) < 0)
4308
        return -1;
4309

    
4310
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
4311
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
4312
        pred_weight_table(h);
4313
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
4314
        implicit_weight_table(h);
4315
    else
4316
        h->use_weight = 0;
4317
    
4318
    if(s->current_picture.reference)
4319
        decode_ref_pic_marking(h);
4320

    
4321
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
4322
        h->cabac_init_idc = get_ue_golomb(&s->gb);
4323

    
4324
    h->last_qscale_diff = 0;
4325
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
4326
    if(s->qscale<0 || s->qscale>51){
4327
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
4328
        return -1;
4329
    }
4330
    h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
4331
    //FIXME qscale / qp ... stuff
4332
    if(h->slice_type == SP_TYPE){
4333
        get_bits1(&s->gb); /* sp_for_switch_flag */
4334
    }
4335
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
4336
        get_se_golomb(&s->gb); /* slice_qs_delta */
4337
    }
4338

    
4339
    h->deblocking_filter = 1;
4340
    h->slice_alpha_c0_offset = 0;
4341