Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 5d18eaad

History | View | Annotate | Download (317 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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
/* Compiling in interlaced support reduces the speed
58
 * of progressive decoding by about 2%. */
59
#define ALLOW_INTERLACE
60

    
61
#ifdef ALLOW_INTERLACE
62
#define MB_MBAFF h->mb_mbaff
63
#define MB_FIELD h->mb_field_decoding_flag
64
#define FRAME_MBAFF h->mb_aff_frame
65
#else
66
#define MB_MBAFF 0
67
#define MB_FIELD 0
68
#define FRAME_MBAFF 0
69
#undef  IS_INTERLACED
70
#define IS_INTERLACED(mb_type) 0
71
#endif
72

    
73
/**
74
 * Sequence parameter set
75
 */
76
typedef struct SPS{
77

    
78
    int profile_idc;
79
    int level_idc;
80
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
81
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
82
    int poc_type;                      ///< pic_order_cnt_type
83
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
84
    int delta_pic_order_always_zero_flag;
85
    int offset_for_non_ref_pic;
86
    int offset_for_top_to_bottom_field;
87
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
88
    int ref_frame_count;               ///< num_ref_frames
89
    int gaps_in_frame_num_allowed_flag;
90
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
91
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
92
    int frame_mbs_only_flag;
93
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
94
    int direct_8x8_inference_flag;
95
    int crop;                   ///< frame_cropping_flag
96
    int crop_left;              ///< frame_cropping_rect_left_offset
97
    int crop_right;             ///< frame_cropping_rect_right_offset
98
    int crop_top;               ///< frame_cropping_rect_top_offset
99
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
100
    int vui_parameters_present_flag;
101
    AVRational sar;
102
    int timing_info_present_flag;
103
    uint32_t num_units_in_tick;
104
    uint32_t time_scale;
105
    int fixed_frame_rate_flag;
106
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
107
    int bitstream_restriction_flag;
108
    int num_reorder_frames;
109
    int scaling_matrix_present;
110
    uint8_t scaling_matrix4[6][16];
111
    uint8_t scaling_matrix8[2][64];
112
}SPS;
113

    
114
/**
115
 * Picture parameter set
116
 */
117
typedef struct PPS{
118
    int sps_id;
119
    int cabac;                  ///< entropy_coding_mode_flag
120
    int pic_order_present;      ///< pic_order_present_flag
121
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
122
    int mb_slice_group_map_type;
123
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
124
    int weighted_pred;          ///< weighted_pred_flag
125
    int weighted_bipred_idc;
126
    int init_qp;                ///< pic_init_qp_minus26 + 26
127
    int init_qs;                ///< pic_init_qs_minus26 + 26
128
    int chroma_qp_index_offset;
129
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
130
    int constrained_intra_pred; ///< constrained_intra_pred_flag
131
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
132
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
133
    uint8_t scaling_matrix4[6][16];
134
    uint8_t scaling_matrix8[2][64];
135
}PPS;
136

    
137
/**
138
 * Memory management control operation opcode.
139
 */
140
typedef enum MMCOOpcode{
141
    MMCO_END=0,
142
    MMCO_SHORT2UNUSED,
143
    MMCO_LONG2UNUSED,
144
    MMCO_SHORT2LONG,
145
    MMCO_SET_MAX_LONG,
146
    MMCO_RESET,
147
    MMCO_LONG,
148
} MMCOOpcode;
149

    
150
/**
151
 * Memory management control operation.
152
 */
153
typedef struct MMCO{
154
    MMCOOpcode opcode;
155
    int short_frame_num;
156
    int long_index;
157
} MMCO;
158

    
159
/**
160
 * H264Context
161
 */
162
typedef struct H264Context{
163
    MpegEncContext s;
164
    int nal_ref_idc;
165
    int nal_unit_type;
166
#define NAL_SLICE                1
167
#define NAL_DPA                  2
168
#define NAL_DPB                  3
169
#define NAL_DPC                  4
170
#define NAL_IDR_SLICE            5
171
#define NAL_SEI                  6
172
#define NAL_SPS                  7
173
#define NAL_PPS                  8
174
#define NAL_AUD                  9
175
#define NAL_END_SEQUENCE        10
176
#define NAL_END_STREAM          11
177
#define NAL_FILLER_DATA         12
178
#define NAL_SPS_EXT             13
179
#define NAL_AUXILIARY_SLICE     19
180
    uint8_t *rbsp_buffer;
181
    unsigned int rbsp_buffer_size;
182

    
183
    /**
184
      * Used to parse AVC variant of h264
185
      */
186
    int is_avc; ///< this flag is != 0 if codec is avc1
187
    int got_avcC; ///< flag used to parse avcC data only once
188
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
189

    
190
    int chroma_qp; //QPc
191

    
192
    int prev_mb_skipped;
193
    int next_mb_skipped;
194

    
195
    //prediction stuff
196
    int chroma_pred_mode;
197
    int intra16x16_pred_mode;
198

    
199
    int top_mb_xy;
200
    int left_mb_xy[2];
201

    
202
    int8_t intra4x4_pred_mode_cache[5*8];
203
    int8_t (*intra4x4_pred_mode)[8];
204
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
205
    void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
206
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
207
    void (*pred16x16[4+3])(uint8_t *src, int stride);
208
    unsigned int topleft_samples_available;
209
    unsigned int top_samples_available;
210
    unsigned int topright_samples_available;
211
    unsigned int left_samples_available;
212
    uint8_t (*top_borders[2])[16+2*8];
213
    uint8_t left_border[2*(17+2*9)];
214

    
215
    /**
216
     * non zero coeff count cache.
217
     * is 64 if not available.
218
     */
219
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
220
    uint8_t (*non_zero_count)[16];
221

    
222
    /**
223
     * Motion vector cache.
224
     */
225
    DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
226
    DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
227
#define LIST_NOT_USED -1 //FIXME rename?
228
#define PART_NOT_AVAILABLE -2
229

    
230
    /**
231
     * is 1 if the specific list MV&references are set to 0,0,-2.
232
     */
233
    int mv_cache_clean[2];
234

    
235
    /**
236
     * number of neighbors (top and/or left) that used 8x8 dct
237
     */
238
    int neighbor_transform_size;
239

    
240
    /**
241
     * block_offset[ 0..23] for frame macroblocks
242
     * block_offset[24..47] for field macroblocks
243
     */
244
    int block_offset[2*(16+8)];
245

    
246
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
247
    uint32_t *mb2b8_xy;
248
    int b_stride; //FIXME use s->b4_stride
249
    int b8_stride;
250

    
251
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
252
    int mb_uvlinesize;
253

    
254
    int emu_edge_width;
255
    int emu_edge_height;
256

    
257
    int halfpel_flag;
258
    int thirdpel_flag;
259

    
260
    int unknown_svq3_flag;
261
    int next_slice_index;
262

    
263
    SPS sps_buffer[MAX_SPS_COUNT];
264
    SPS sps; ///< current sps
265

    
266
    PPS pps_buffer[MAX_PPS_COUNT];
267
    /**
268
     * current pps
269
     */
270
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
271

    
272
    uint32_t dequant4_buffer[6][52][16];
273
    uint32_t dequant8_buffer[2][52][64];
274
    uint32_t (*dequant4_coeff[6])[16];
275
    uint32_t (*dequant8_coeff[2])[64];
276
    int dequant_coeff_pps;     ///< reinit tables when pps changes
277

    
278
    int slice_num;
279
    uint8_t *slice_table_base;
280
    uint8_t *slice_table;      ///< slice_table_base + 2*mb_stride + 1
281
    int slice_type;
282
    int slice_type_fixed;
283

    
284
    //interlacing specific flags
285
    int mb_aff_frame;
286
    int mb_field_decoding_flag;
287
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
288

    
289
    int sub_mb_type[4];
290

    
291
    //POC stuff
292
    int poc_lsb;
293
    int poc_msb;
294
    int delta_poc_bottom;
295
    int delta_poc[2];
296
    int frame_num;
297
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
298
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
299
    int frame_num_offset;         ///< for POC type 2
300
    int prev_frame_num_offset;    ///< for POC type 2
301
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
302

    
303
    /**
304
     * frame_num for frames or 2*frame_num for field pics.
305
     */
306
    int curr_pic_num;
307

    
308
    /**
309
     * max_frame_num or 2*max_frame_num for field pics.
310
     */
311
    int max_pic_num;
312

    
313
    //Weighted pred stuff
314
    int use_weight;
315
    int use_weight_chroma;
316
    int luma_log2_weight_denom;
317
    int chroma_log2_weight_denom;
318
    int luma_weight[2][48];
319
    int luma_offset[2][48];
320
    int chroma_weight[2][48][2];
321
    int chroma_offset[2][48][2];
322
    int implicit_weight[48][48];
323

    
324
    //deblock
325
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
326
    int slice_alpha_c0_offset;
327
    int slice_beta_offset;
328

    
329
    int redundant_pic_count;
330

    
331
    int direct_spatial_mv_pred;
332
    int dist_scale_factor[16];
333
    int dist_scale_factor_field[32];
334
    int map_col_to_list0[2][16];
335
    int map_col_to_list0_field[2][32];
336

    
337
    /**
338
     * num_ref_idx_l0/1_active_minus1 + 1
339
     */
340
    int ref_count[2];            ///< counts frames or fields, depending on current mb mode
341
    Picture *short_ref[32];
342
    Picture *long_ref[32];
343
    Picture default_ref_list[2][32];
344
    Picture ref_list[2][48];     ///< 0..15: frame refs, 16..47: mbaff field refs
345
    Picture *delayed_pic[16]; //FIXME size?
346
    Picture *delayed_output_pic;
347

    
348
    /**
349
     * memory management control operations buffer.
350
     */
351
    MMCO mmco[MAX_MMCO_COUNT];
352
    int mmco_index;
353

    
354
    int long_ref_count;  ///< number of actual long term references
355
    int short_ref_count; ///< number of actual short term references
356

    
357
    //data partitioning
358
    GetBitContext intra_gb;
359
    GetBitContext inter_gb;
360
    GetBitContext *intra_gb_ptr;
361
    GetBitContext *inter_gb_ptr;
362

    
363
    DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);
364

    
365
    /**
366
     * Cabac
367
     */
368
    CABACContext cabac;
369
    uint8_t      cabac_state[460];
370
    int          cabac_init_idc;
371

    
372
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
373
    uint16_t     *cbp_table;
374
    int top_cbp;
375
    int left_cbp;
376
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
377
    uint8_t     *chroma_pred_mode_table;
378
    int         last_qscale_diff;
379
    int16_t     (*mvd_table[2])[2];
380
    DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
381
    uint8_t     *direct_table;
382
    uint8_t     direct_cache[5*8];
383

    
384
    uint8_t zigzag_scan[16];
385
    uint8_t zigzag_scan8x8[64];
386
    uint8_t zigzag_scan8x8_cavlc[64];
387
    uint8_t field_scan[16];
388
    uint8_t field_scan8x8[64];
389
    uint8_t field_scan8x8_cavlc[64];
390
    const uint8_t *zigzag_scan_q0;
391
    const uint8_t *zigzag_scan8x8_q0;
392
    const uint8_t *zigzag_scan8x8_cavlc_q0;
393
    const uint8_t *field_scan_q0;
394
    const uint8_t *field_scan8x8_q0;
395
    const uint8_t *field_scan8x8_cavlc_q0;
396

    
397
    int x264_build;
398
}H264Context;
399

    
400
static VLC coeff_token_vlc[4];
401
static VLC chroma_dc_coeff_token_vlc;
402

    
403
static VLC total_zeros_vlc[15];
404
static VLC chroma_dc_total_zeros_vlc[3];
405

    
406
static VLC run_vlc[6];
407
static VLC run7_vlc;
408

    
409
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
410
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
411
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);
412

    
413
static always_inline uint32_t pack16to32(int a, int b){
414
#ifdef WORDS_BIGENDIAN
415
   return (b&0xFFFF) + (a<<16);
416
#else
417
   return (a&0xFFFF) + (b<<16);
418
#endif
419
}
420

    
421
/**
422
 * fill a rectangle.
423
 * @param h height of the rectangle, should be a constant
424
 * @param w width of the rectangle, should be a constant
425
 * @param size the size of val (1 or 4), should be a constant
426
 */
427
static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
428
    uint8_t *p= (uint8_t*)vp;
429
    assert(size==1 || size==4);
430
    assert(w<=4);
431

    
432
    w      *= size;
433
    stride *= size;
434

    
435
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
436
    assert((stride&(w-1))==0);
437
    if(w==2){
438
        const uint16_t v= size==4 ? val : val*0x0101;
439
        *(uint16_t*)(p + 0*stride)= v;
440
        if(h==1) return;
441
        *(uint16_t*)(p + 1*stride)= v;
442
        if(h==2) return;
443
        *(uint16_t*)(p + 2*stride)=
444
        *(uint16_t*)(p + 3*stride)= v;
445
    }else if(w==4){
446
        const uint32_t v= size==4 ? val : val*0x01010101;
447
        *(uint32_t*)(p + 0*stride)= v;
448
        if(h==1) return;
449
        *(uint32_t*)(p + 1*stride)= v;
450
        if(h==2) return;
451
        *(uint32_t*)(p + 2*stride)=
452
        *(uint32_t*)(p + 3*stride)= v;
453
    }else if(w==8){
454
    //gcc can't optimize 64bit math on x86_32
455
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
456
        const uint64_t v= val*0x0100000001ULL;
457
        *(uint64_t*)(p + 0*stride)= v;
458
        if(h==1) return;
459
        *(uint64_t*)(p + 1*stride)= v;
460
        if(h==2) return;
461
        *(uint64_t*)(p + 2*stride)=
462
        *(uint64_t*)(p + 3*stride)= v;
463
    }else if(w==16){
464
        const uint64_t v= val*0x0100000001ULL;
465
        *(uint64_t*)(p + 0+0*stride)=
466
        *(uint64_t*)(p + 8+0*stride)=
467
        *(uint64_t*)(p + 0+1*stride)=
468
        *(uint64_t*)(p + 8+1*stride)= v;
469
        if(h==2) return;
470
        *(uint64_t*)(p + 0+2*stride)=
471
        *(uint64_t*)(p + 8+2*stride)=
472
        *(uint64_t*)(p + 0+3*stride)=
473
        *(uint64_t*)(p + 8+3*stride)= v;
474
#else
475
        *(uint32_t*)(p + 0+0*stride)=
476
        *(uint32_t*)(p + 4+0*stride)= val;
477
        if(h==1) return;
478
        *(uint32_t*)(p + 0+1*stride)=
479
        *(uint32_t*)(p + 4+1*stride)= val;
480
        if(h==2) return;
481
        *(uint32_t*)(p + 0+2*stride)=
482
        *(uint32_t*)(p + 4+2*stride)=
483
        *(uint32_t*)(p + 0+3*stride)=
484
        *(uint32_t*)(p + 4+3*stride)= val;
485
    }else if(w==16){
486
        *(uint32_t*)(p + 0+0*stride)=
487
        *(uint32_t*)(p + 4+0*stride)=
488
        *(uint32_t*)(p + 8+0*stride)=
489
        *(uint32_t*)(p +12+0*stride)=
490
        *(uint32_t*)(p + 0+1*stride)=
491
        *(uint32_t*)(p + 4+1*stride)=
492
        *(uint32_t*)(p + 8+1*stride)=
493
        *(uint32_t*)(p +12+1*stride)= val;
494
        if(h==2) return;
495
        *(uint32_t*)(p + 0+2*stride)=
496
        *(uint32_t*)(p + 4+2*stride)=
497
        *(uint32_t*)(p + 8+2*stride)=
498
        *(uint32_t*)(p +12+2*stride)=
499
        *(uint32_t*)(p + 0+3*stride)=
500
        *(uint32_t*)(p + 4+3*stride)=
501
        *(uint32_t*)(p + 8+3*stride)=
502
        *(uint32_t*)(p +12+3*stride)= val;
503
#endif
504
    }else
505
        assert(0);
506
    assert(h==4);
507
}
508

    
509
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
510
    MpegEncContext * const s = &h->s;
511
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
512
    int topleft_xy, top_xy, topright_xy, left_xy[2];
513
    int topleft_type, top_type, topright_type, left_type[2];
514
    int left_block[8];
515
    int i;
516

    
517
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
518
    // the actual condition is whether we're on the edge of a slice,
519
    // and even then the intra and nnz parts are unnecessary.
520
    if(for_deblock && h->slice_num == 1 && !FRAME_MBAFF)
521
        return;
522

    
523
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
524

    
525
    top_xy     = mb_xy  - s->mb_stride;
526
    topleft_xy = top_xy - 1;
527
    topright_xy= top_xy + 1;
528
    left_xy[1] = left_xy[0] = mb_xy-1;
529
    left_block[0]= 0;
530
    left_block[1]= 1;
531
    left_block[2]= 2;
532
    left_block[3]= 3;
533
    left_block[4]= 7;
534
    left_block[5]= 10;
535
    left_block[6]= 8;
536
    left_block[7]= 11;
537
    if(FRAME_MBAFF){
538
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
539
        const int top_pair_xy      = pair_xy     - s->mb_stride;
540
        const int topleft_pair_xy  = top_pair_xy - 1;
541
        const int topright_pair_xy = top_pair_xy + 1;
542
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
543
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
544
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
545
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
546
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
547
        const int bottom = (s->mb_y & 1);
548
        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);
549
        if (bottom
550
                ? !curr_mb_frame_flag // bottom macroblock
551
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
552
                ) {
553
            top_xy -= s->mb_stride;
554
        }
555
        if (bottom
556
                ? !curr_mb_frame_flag // bottom macroblock
557
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
558
                ) {
559
            topleft_xy -= s->mb_stride;
560
        }
561
        if (bottom
562
                ? !curr_mb_frame_flag // bottom macroblock
563
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
564
                ) {
565
            topright_xy -= s->mb_stride;
566
        }
567
        if (left_mb_frame_flag != curr_mb_frame_flag) {
568
            left_xy[1] = left_xy[0] = pair_xy - 1;
569
            if (curr_mb_frame_flag) {
570
                if (bottom) {
571
                    left_block[0]= 2;
572
                    left_block[1]= 2;
573
                    left_block[2]= 3;
574
                    left_block[3]= 3;
575
                    left_block[4]= 8;
576
                    left_block[5]= 11;
577
                    left_block[6]= 8;
578
                    left_block[7]= 11;
579
                } else {
580
                    left_block[0]= 0;
581
                    left_block[1]= 0;
582
                    left_block[2]= 1;
583
                    left_block[3]= 1;
584
                    left_block[4]= 7;
585
                    left_block[5]= 10;
586
                    left_block[6]= 7;
587
                    left_block[7]= 10;
588
                }
589
            } else {
590
                left_xy[1] += s->mb_stride;
591
                //left_block[0]= 0;
592
                left_block[1]= 2;
593
                left_block[2]= 0;
594
                left_block[3]= 2;
595
                //left_block[4]= 7;
596
                left_block[5]= 10;
597
                left_block[6]= 7;
598
                left_block[7]= 10;
599
            }
600
        }
601
    }
602

    
603
    h->top_mb_xy = top_xy;
604
    h->left_mb_xy[0] = left_xy[0];
605
    h->left_mb_xy[1] = left_xy[1];
606
    if(for_deblock){
607
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
608
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
609
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
610
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
611
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
612

    
613
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
614
            int list;
615
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
616
            for(i=0; i<16; i++)
617
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
618
            for(list=0; list<1+(h->slice_type==B_TYPE); list++){
619
                if(USES_LIST(mb_type,list)){
620
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
621
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
622
                    uint8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
623
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
624
                        dst[0] = src[0];
625
                        dst[1] = src[1];
626
                        dst[2] = src[2];
627
                        dst[3] = src[3];
628
                    }
629
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
630
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
631
                    ref += h->b8_stride;
632
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
633
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
634
                }else{
635
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
636
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
637
                }
638
            }
639
        }
640
    }else{
641
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
642
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
643
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
644
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
645
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
646
    }
647

    
648
    if(IS_INTRA(mb_type)){
649
        h->topleft_samples_available=
650
        h->top_samples_available=
651
        h->left_samples_available= 0xFFFF;
652
        h->topright_samples_available= 0xEEEA;
653

    
654
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
655
            h->topleft_samples_available= 0xB3FF;
656
            h->top_samples_available= 0x33FF;
657
            h->topright_samples_available= 0x26EA;
658
        }
659
        for(i=0; i<2; i++){
660
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
661
                h->topleft_samples_available&= 0xDF5F;
662
                h->left_samples_available&= 0x5F5F;
663
            }
664
        }
665

    
666
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
667
            h->topleft_samples_available&= 0x7FFF;
668

    
669
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
670
            h->topright_samples_available&= 0xFBFF;
671

    
672
        if(IS_INTRA4x4(mb_type)){
673
            if(IS_INTRA4x4(top_type)){
674
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
675
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
676
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
677
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
678
            }else{
679
                int pred;
680
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
681
                    pred= -1;
682
                else{
683
                    pred= 2;
684
                }
685
                h->intra4x4_pred_mode_cache[4+8*0]=
686
                h->intra4x4_pred_mode_cache[5+8*0]=
687
                h->intra4x4_pred_mode_cache[6+8*0]=
688
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
689
            }
690
            for(i=0; i<2; i++){
691
                if(IS_INTRA4x4(left_type[i])){
692
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
693
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
694
                }else{
695
                    int pred;
696
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
697
                        pred= -1;
698
                    else{
699
                        pred= 2;
700
                    }
701
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
702
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
703
                }
704
            }
705
        }
706
    }
707

    
708

    
709
/*
710
0 . T T. T T T T
711
1 L . .L . . . .
712
2 L . .L . . . .
713
3 . T TL . . . .
714
4 L . .L . . . .
715
5 L . .. . . . .
716
*/
717
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
718
    if(top_type){
719
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
720
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
721
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
722
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
723

    
724
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
725
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
726

    
727
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
728
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
729

    
730
    }else{
731
        h->non_zero_count_cache[4+8*0]=
732
        h->non_zero_count_cache[5+8*0]=
733
        h->non_zero_count_cache[6+8*0]=
734
        h->non_zero_count_cache[7+8*0]=
735

    
736
        h->non_zero_count_cache[1+8*0]=
737
        h->non_zero_count_cache[2+8*0]=
738

    
739
        h->non_zero_count_cache[1+8*3]=
740
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
741

    
742
    }
743

    
744
    for (i=0; i<2; i++) {
745
        if(left_type[i]){
746
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
747
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
748
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
749
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
750
        }else{
751
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
752
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
753
            h->non_zero_count_cache[0+8*1 +   8*i]=
754
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
755
        }
756
    }
757

    
758
    if( h->pps.cabac ) {
759
        // top_cbp
760
        if(top_type) {
761
            h->top_cbp = h->cbp_table[top_xy];
762
        } else if(IS_INTRA(mb_type)) {
763
            h->top_cbp = 0x1C0;
764
        } else {
765
            h->top_cbp = 0;
766
        }
767
        // left_cbp
768
        if (left_type[0]) {
769
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
770
        } else if(IS_INTRA(mb_type)) {
771
            h->left_cbp = 0x1C0;
772
        } else {
773
            h->left_cbp = 0;
774
        }
775
        if (left_type[0]) {
776
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
777
        }
778
        if (left_type[1]) {
779
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
780
        }
781
    }
782

    
783
#if 1
784
    //FIXME direct mb can skip much of this
785
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
786
        int list;
787
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
788
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
789
                /*if(!h->mv_cache_clean[list]){
790
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
791
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
792
                    h->mv_cache_clean[list]= 1;
793
                }*/
794
                continue;
795
            }
796
            h->mv_cache_clean[list]= 0;
797

    
798
            if(USES_LIST(top_type, list)){
799
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
800
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
801
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
802
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
803
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
804
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
805
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
806
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
807
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
808
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
809
            }else{
810
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
811
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
812
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
813
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
814
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
815
            }
816

    
817
            //FIXME unify cleanup or sth
818
            if(USES_LIST(left_type[0], list)){
819
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
820
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
821
                *(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]];
822
                *(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]];
823
                h->ref_cache[list][scan8[0] - 1 + 0*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
824
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1]>>1)];
825
            }else{
826
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
827
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
828
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
829
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
830
            }
831

    
832
            if(USES_LIST(left_type[1], list)){
833
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
834
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
835
                *(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]];
836
                *(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]];
837
                h->ref_cache[list][scan8[0] - 1 + 2*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
838
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[3]>>1)];
839
            }else{
840
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
841
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
842
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
843
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
844
                assert((!left_type[0]) == (!left_type[1]));
845
            }
846

    
847
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
848
                continue;
849

    
850
            if(USES_LIST(topleft_type, list)){
851
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
852
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
853
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
854
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
855
            }else{
856
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
857
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
858
            }
859

    
860
            if(USES_LIST(topright_type, list)){
861
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
862
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
863
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
864
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
865
            }else{
866
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
867
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
868
            }
869

    
870

    
871
            h->ref_cache[list][scan8[5 ]+1] =
872
            h->ref_cache[list][scan8[7 ]+1] =
873
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
874
            h->ref_cache[list][scan8[4 ]] =
875
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
876
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
877
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
878
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
879
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
880
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
881

    
882
            if( h->pps.cabac ) {
883
                /* XXX beurk, Load mvd */
884
                if(USES_LIST(top_type, list)){
885
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
886
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
887
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
888
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
889
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
890
                }else{
891
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
892
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
893
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
894
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
895
                }
896
                if(USES_LIST(left_type[0], list)){
897
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
898
                    *(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]];
899
                    *(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]];
900
                }else{
901
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
902
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
903
                }
904
                if(USES_LIST(left_type[1], list)){
905
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
906
                    *(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]];
907
                    *(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]];
908
                }else{
909
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
910
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
911
                }
912
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
913
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
914
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
915
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
916
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
917

    
918
                if(h->slice_type == B_TYPE){
919
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
920

    
921
                    if(IS_DIRECT(top_type)){
922
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
923
                    }else if(IS_8X8(top_type)){
924
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
925
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
926
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
927
                    }else{
928
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
929
                    }
930

    
931
                    if(IS_DIRECT(left_type[0]))
932
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
933
                    else if(IS_8X8(left_type[0]))
934
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
935
                    else
936
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
937

    
938
                    if(IS_DIRECT(left_type[1]))
939
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
940
                    else if(IS_8X8(left_type[1]))
941
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
942
                    else
943
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
944
                }
945
            }
946

    
947
            if(FRAME_MBAFF){
948
#define MAP_MVS\
949
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
950
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
951
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
952
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
953
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
954
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
955
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
956
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
957
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
958
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
959
                if(MB_FIELD){
960
#define MAP_F2F(idx, mb_type)\
961
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
962
                        h->ref_cache[list][idx] <<= 1;\
963
                        h->mv_cache[list][idx][1] /= 2;\
964
                        h->mvd_cache[list][idx][1] /= 2;\
965
                    }
966
                    MAP_MVS
967
#undef MAP_F2F
968
                }else{
969
#define MAP_F2F(idx, mb_type)\
970
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
971
                        h->ref_cache[list][idx] >>= 1;\
972
                        h->mv_cache[list][idx][1] <<= 1;\
973
                        h->mvd_cache[list][idx][1] <<= 1;\
974
                    }
975
                    MAP_MVS
976
#undef MAP_F2F
977
                }
978
            }
979
        }
980
    }
981
#endif
982

    
983
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
984
}
985

    
986
static inline void write_back_intra_pred_mode(H264Context *h){
987
    MpegEncContext * const s = &h->s;
988
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
989

    
990
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
991
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
992
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
993
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
994
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
995
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
996
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
997
}
998

    
999
/**
1000
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
1001
 */
1002
static inline int check_intra4x4_pred_mode(H264Context *h){
1003
    MpegEncContext * const s = &h->s;
1004
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
1005
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
1006
    int i;
1007

    
1008
    if(!(h->top_samples_available&0x8000)){
1009
        for(i=0; i<4; i++){
1010
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
1011
            if(status<0){
1012
                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);
1013
                return -1;
1014
            } else if(status){
1015
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
1016
            }
1017
        }
1018
    }
1019

    
1020
    if(!(h->left_samples_available&0x8000)){
1021
        for(i=0; i<4; i++){
1022
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
1023
            if(status<0){
1024
                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);
1025
                return -1;
1026
            } else if(status){
1027
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
1028
            }
1029
        }
1030
    }
1031

    
1032
    return 0;
1033
} //FIXME cleanup like next
1034

    
1035
/**
1036
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
1037
 */
1038
static inline int check_intra_pred_mode(H264Context *h, int mode){
1039
    MpegEncContext * const s = &h->s;
1040
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
1041
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
1042

    
1043
    if(mode < 0 || mode > 6) {
1044
        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);
1045
        return -1;
1046
    }
1047

    
1048
    if(!(h->top_samples_available&0x8000)){
1049
        mode= top[ mode ];
1050
        if(mode<0){
1051
            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);
1052
            return -1;
1053
        }
1054
    }
1055

    
1056
    if(!(h->left_samples_available&0x8000)){
1057
        mode= left[ mode ];
1058
        if(mode<0){
1059
            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);
1060
            return -1;
1061
        }
1062
    }
1063

    
1064
    return mode;
1065
}
1066

    
1067
/**
1068
 * gets the predicted intra4x4 prediction mode.
1069
 */
1070
static inline int pred_intra_mode(H264Context *h, int n){
1071
    const int index8= scan8[n];
1072
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1073
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1074
    const int min= FFMIN(left, top);
1075

    
1076
    tprintf("mode:%d %d min:%d\n", left ,top, min);
1077

    
1078
    if(min<0) return DC_PRED;
1079
    else      return min;
1080
}
1081

    
1082
static inline void write_back_non_zero_count(H264Context *h){
1083
    MpegEncContext * const s = &h->s;
1084
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1085

    
1086
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1087
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1088
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
1089
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
1090
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1091
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1092
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
1093

    
1094
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
1095
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
1096
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
1097

    
1098
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
1099
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
1100
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
1101

    
1102
    if(FRAME_MBAFF){
1103
        // store all luma nnzs, for deblocking
1104
        int v = 0, i;
1105
        for(i=0; i<16; i++)
1106
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
1107
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
1108
    }
1109
}
1110

    
1111
/**
1112
 * gets the predicted number of non zero coefficients.
1113
 * @param n block index
1114
 */
1115
static inline int pred_non_zero_count(H264Context *h, int n){
1116
    const int index8= scan8[n];
1117
    const int left= h->non_zero_count_cache[index8 - 1];
1118
    const int top = h->non_zero_count_cache[index8 - 8];
1119
    int i= left + top;
1120

    
1121
    if(i<64) i= (i+1)>>1;
1122

    
1123
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1124

    
1125
    return i&31;
1126
}
1127

    
1128
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1129
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1130

    
1131
    /* there is no consistent mapping of mvs to neighboring locations that will
1132
     * make mbaff happy, so we can't move all this logic to fill_caches */
1133
    if(FRAME_MBAFF){
1134
        MpegEncContext *s = &h->s;
1135
        const int *mb_types = s->current_picture_ptr->mb_type;
1136
        const int16_t *mv;
1137
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
1138
        *C = h->mv_cache[list][scan8[0]-2];
1139

    
1140
        if(!MB_FIELD
1141
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
1142
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
1143
            if(IS_INTERLACED(mb_types[topright_xy])){
1144
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
1145
                const int x4 = X4, y4 = Y4;\
1146
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
1147
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
1148
                    return LIST_NOT_USED;\
1149
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
1150
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
1151
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
1152
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
1153

    
1154
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
1155
            }
1156
        }
1157
        if(topright_ref == PART_NOT_AVAILABLE
1158
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
1159
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
1160
            if(!MB_FIELD
1161
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
1162
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
1163
            }
1164
            if(MB_FIELD
1165
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
1166
               && i >= scan8[0]+8){
1167
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
1168
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
1169
            }
1170
        }
1171
#undef SET_DIAG_MV
1172
    }
1173

    
1174
    if(topright_ref != PART_NOT_AVAILABLE){
1175
        *C= h->mv_cache[list][ i - 8 + part_width ];
1176
        return topright_ref;
1177
    }else{
1178
        tprintf("topright MV not available\n");
1179

    
1180
        *C= h->mv_cache[list][ i - 8 - 1 ];
1181
        return h->ref_cache[list][ i - 8 - 1 ];
1182
    }
1183
}
1184

    
1185
/**
1186
 * gets the predicted MV.
1187
 * @param n the block index
1188
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1189
 * @param mx the x component of the predicted motion vector
1190
 * @param my the y component of the predicted motion vector
1191
 */
1192
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1193
    const int index8= scan8[n];
1194
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1195
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1196
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1197
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1198
    const int16_t * C;
1199
    int diagonal_ref, match_count;
1200

    
1201
    assert(part_width==1 || part_width==2 || part_width==4);
1202

    
1203
/* mv_cache
1204
  B . . A T T T T
1205
  U . . L . . , .
1206
  U . . L . . . .
1207
  U . . L . . , .
1208
  . . . L . . . .
1209
*/
1210

    
1211
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1212
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1213
    tprintf("pred_motion match_count=%d\n", match_count);
1214
    if(match_count > 1){ //most common
1215
        *mx= mid_pred(A[0], B[0], C[0]);
1216
        *my= mid_pred(A[1], B[1], C[1]);
1217
    }else if(match_count==1){
1218
        if(left_ref==ref){
1219
            *mx= A[0];
1220
            *my= A[1];
1221
        }else if(top_ref==ref){
1222
            *mx= B[0];
1223
            *my= B[1];
1224
        }else{
1225
            *mx= C[0];
1226
            *my= C[1];
1227
        }
1228
    }else{
1229
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1230
            *mx= A[0];
1231
            *my= A[1];
1232
        }else{
1233
            *mx= mid_pred(A[0], B[0], C[0]);
1234
            *my= mid_pred(A[1], B[1], C[1]);
1235
        }
1236
    }
1237

    
1238
    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);
1239
}
1240

    
1241
/**
1242
 * gets the directionally predicted 16x8 MV.
1243
 * @param n the block index
1244
 * @param mx the x component of the predicted motion vector
1245
 * @param my the y component of the predicted motion vector
1246
 */
1247
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1248
    if(n==0){
1249
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1250
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1251

    
1252
        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);
1253

    
1254
        if(top_ref == ref){
1255
            *mx= B[0];
1256
            *my= B[1];
1257
            return;
1258
        }
1259
    }else{
1260
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1261
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1262

    
1263
        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);
1264

    
1265
        if(left_ref == ref){
1266
            *mx= A[0];
1267
            *my= A[1];
1268
            return;
1269
        }
1270
    }
1271

    
1272
    //RARE
1273
    pred_motion(h, n, 4, list, ref, mx, my);
1274
}
1275

    
1276
/**
1277
 * gets the directionally predicted 8x16 MV.
1278
 * @param n the block index
1279
 * @param mx the x component of the predicted motion vector
1280
 * @param my the y component of the predicted motion vector
1281
 */
1282
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1283
    if(n==0){
1284
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1285
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1286

    
1287
        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);
1288

    
1289
        if(left_ref == ref){
1290
            *mx= A[0];
1291
            *my= A[1];
1292
            return;
1293
        }
1294
    }else{
1295
        const int16_t * C;
1296
        int diagonal_ref;
1297

    
1298
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1299

    
1300
        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);
1301

    
1302
        if(diagonal_ref == ref){
1303
            *mx= C[0];
1304
            *my= C[1];
1305
            return;
1306
        }
1307
    }
1308

    
1309
    //RARE
1310
    pred_motion(h, n, 2, list, ref, mx, my);
1311
}
1312

    
1313
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1314
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1315
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1316

    
1317
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1318

    
1319
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1320
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1321
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1322

    
1323
        *mx = *my = 0;
1324
        return;
1325
    }
1326

    
1327
    pred_motion(h, 0, 4, 0, 0, mx, my);
1328

    
1329
    return;
1330
}
1331

    
1332
static inline void direct_dist_scale_factor(H264Context * const h){
1333
    const int poc = h->s.current_picture_ptr->poc;
1334
    const int poc1 = h->ref_list[1][0].poc;
1335
    int i;
1336
    for(i=0; i<h->ref_count[0]; i++){
1337
        int poc0 = h->ref_list[0][i].poc;
1338
        int td = clip(poc1 - poc0, -128, 127);
1339
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1340
            h->dist_scale_factor[i] = 256;
1341
        }else{
1342
            int tb = clip(poc - poc0, -128, 127);
1343
            int tx = (16384 + (ABS(td) >> 1)) / td;
1344
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1345
        }
1346
    }
1347
    if(FRAME_MBAFF){
1348
        for(i=0; i<h->ref_count[0]; i++){
1349
            h->dist_scale_factor_field[2*i] =
1350
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
1351
        }
1352
    }
1353
}
1354
static inline void direct_ref_list_init(H264Context * const h){
1355
    MpegEncContext * const s = &h->s;
1356
    Picture * const ref1 = &h->ref_list[1][0];
1357
    Picture * const cur = s->current_picture_ptr;
1358
    int list, i, j;
1359
    if(cur->pict_type == I_TYPE)
1360
        cur->ref_count[0] = 0;
1361
    if(cur->pict_type != B_TYPE)
1362
        cur->ref_count[1] = 0;
1363
    for(list=0; list<2; list++){
1364
        cur->ref_count[list] = h->ref_count[list];
1365
        for(j=0; j<h->ref_count[list]; j++)
1366
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1367
    }
1368
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1369
        return;
1370
    for(list=0; list<2; list++){
1371
        for(i=0; i<ref1->ref_count[list]; i++){
1372
            const int poc = ref1->ref_poc[list][i];
1373
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1374
            for(j=0; j<h->ref_count[list]; j++)
1375
                if(h->ref_list[list][j].poc == poc){
1376
                    h->map_col_to_list0[list][i] = j;
1377
                    break;
1378
                }
1379
        }
1380
    }
1381
    if(FRAME_MBAFF){
1382
        for(list=0; list<2; list++){
1383
            for(i=0; i<ref1->ref_count[list]; i++){
1384
                j = h->map_col_to_list0[list][i];
1385
                h->map_col_to_list0_field[list][2*i] = 2*j;
1386
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1387
            }
1388
        }
1389
    }
1390
}
1391

    
1392
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1393
    MpegEncContext * const s = &h->s;
1394
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1395
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1396
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1397
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1398
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1399
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1400
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1401
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1402
    const int is_b8x8 = IS_8X8(*mb_type);
1403
    int sub_mb_type;
1404
    int i8, i4;
1405

    
1406
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1407
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1408
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1409
         * so we know exactly what block size to use */
1410
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1411
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1412
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1413
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1414
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1415
    }else{
1416
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1417
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1418
    }
1419
    if(!is_b8x8)
1420
        *mb_type |= MB_TYPE_DIRECT2;
1421
    if(MB_FIELD)
1422
        *mb_type |= MB_TYPE_INTERLACED;
1423

    
1424
    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);
1425

    
1426
    if(h->direct_spatial_mv_pred){
1427
        int ref[2];
1428
        int mv[2][2];
1429
        int list;
1430

    
1431
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1432

    
1433
        /* ref = min(neighbors) */
1434
        for(list=0; list<2; list++){
1435
            int refa = h->ref_cache[list][scan8[0] - 1];
1436
            int refb = h->ref_cache[list][scan8[0] - 8];
1437
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1438
            if(refc == -2)
1439
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1440
            ref[list] = refa;
1441
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1442
                ref[list] = refb;
1443
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1444
                ref[list] = refc;
1445
            if(ref[list] < 0)
1446
                ref[list] = -1;
1447
        }
1448

    
1449
        if(ref[0] < 0 && ref[1] < 0){
1450
            ref[0] = ref[1] = 0;
1451
            mv[0][0] = mv[0][1] =
1452
            mv[1][0] = mv[1][1] = 0;
1453
        }else{
1454
            for(list=0; list<2; list++){
1455
                if(ref[list] >= 0)
1456
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1457
                else
1458
                    mv[list][0] = mv[list][1] = 0;
1459
            }
1460
        }
1461

    
1462
        if(ref[1] < 0){
1463
            *mb_type &= ~MB_TYPE_P0L1;
1464
            sub_mb_type &= ~MB_TYPE_P0L1;
1465
        }else if(ref[0] < 0){
1466
            *mb_type &= ~MB_TYPE_P0L0;
1467
            sub_mb_type &= ~MB_TYPE_P0L0;
1468
        }
1469

    
1470
        if(IS_16X16(*mb_type)){
1471
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1472
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1473
            if(!IS_INTRA(mb_type_col)
1474
               && (   (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1475
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1476
                       && (h->x264_build>33 || !h->x264_build)))){
1477
                if(ref[0] > 0)
1478
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1479
                else
1480
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1481
                if(ref[1] > 0)
1482
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1483
                else
1484
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1485
            }else{
1486
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1487
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1488
            }
1489
        }else{
1490
            for(i8=0; i8<4; i8++){
1491
                const int x8 = i8&1;
1492
                const int y8 = i8>>1;
1493

    
1494
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1495
                    continue;
1496
                h->sub_mb_type[i8] = sub_mb_type;
1497

    
1498
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1499
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1500
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1501
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1502

    
1503
                /* col_zero_flag */
1504
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1505
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1506
                                                  && (h->x264_build>33 || !h->x264_build)))){
1507
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1508
                    if(IS_SUB_8X8(sub_mb_type)){
1509
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1510
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1511
                            if(ref[0] == 0)
1512
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1513
                            if(ref[1] == 0)
1514
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1515
                        }
1516
                    }else
1517
                    for(i4=0; i4<4; i4++){
1518
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1519
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1520
                            if(ref[0] == 0)
1521
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1522
                            if(ref[1] == 0)
1523
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1524
                        }
1525
                    }
1526
                }
1527
            }
1528
        }
1529
    }else{ /* direct temporal mv pred */
1530
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1531
        const int *dist_scale_factor = h->dist_scale_factor;
1532

    
1533
        if(FRAME_MBAFF){
1534
            if(IS_INTERLACED(*mb_type)){
1535
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1536
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1537
                dist_scale_factor = h->dist_scale_factor_field;
1538
            }
1539
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1540
                /* FIXME assumes direct_8x8_inference == 1 */
1541
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1542
                int mb_types_col[2];
1543
                int y_shift;
1544

    
1545
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1546
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1547
                         | (*mb_type & MB_TYPE_INTERLACED);
1548
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1549

    
1550
                if(IS_INTERLACED(*mb_type)){
1551
                    /* frame to field scaling */
1552
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1553
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1554
                    if(s->mb_y&1){
1555
                        l1ref0 -= 2*h->b8_stride;
1556
                        l1ref1 -= 2*h->b8_stride;
1557
                        l1mv0 -= 4*h->b_stride;
1558
                        l1mv1 -= 4*h->b_stride;
1559
                    }
1560
                    y_shift = 0;
1561

    
1562
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1563
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1564
                       && !is_b8x8)
1565
                        *mb_type |= MB_TYPE_16x8;
1566
                    else
1567
                        *mb_type |= MB_TYPE_8x8;
1568
                }else{
1569
                    /* field to frame scaling */
1570
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1571
                     * but in MBAFF, top and bottom POC are equal */
1572
                    int dy = (s->mb_y&1) ? 1 : 2;
1573
                    mb_types_col[0] =
1574
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1575
                    l1ref0 += dy*h->b8_stride;
1576
                    l1ref1 += dy*h->b8_stride;
1577
                    l1mv0 += 2*dy*h->b_stride;
1578
                    l1mv1 += 2*dy*h->b_stride;
1579
                    y_shift = 2;
1580

    
1581
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1582
                       && !is_b8x8)
1583
                        *mb_type |= MB_TYPE_16x16;
1584
                    else
1585
                        *mb_type |= MB_TYPE_8x8;
1586
                }
1587

    
1588
                for(i8=0; i8<4; i8++){
1589
                    const int x8 = i8&1;
1590
                    const int y8 = i8>>1;
1591
                    int ref0, scale;
1592
                    const int16_t (*l1mv)[2]= l1mv0;
1593

    
1594
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1595
                        continue;
1596
                    h->sub_mb_type[i8] = sub_mb_type;
1597

    
1598
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1599
                    if(IS_INTRA(mb_types_col[y8])){
1600
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1601
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1602
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1603
                        continue;
1604
                    }
1605

    
1606
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1607
                    if(ref0 >= 0)
1608
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1609
                    else{
1610
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1611
                        l1mv= l1mv1;
1612
                    }
1613
                    scale = dist_scale_factor[ref0];
1614
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1615

    
1616
                    {
1617
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1618
                        int my_col = (mv_col[1]<<y_shift)/2;
1619
                        int mx = (scale * mv_col[0] + 128) >> 8;
1620
                        int my = (scale * my_col + 128) >> 8;
1621
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1622
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1623
                    }
1624
                }
1625
                return;
1626
            }
1627
        }
1628

    
1629
        /* one-to-one mv scaling */
1630

    
1631
        if(IS_16X16(*mb_type)){
1632
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1633
            if(IS_INTRA(mb_type_col)){
1634
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1635
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1636
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1637
            }else{
1638
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1639
                                                : map_col_to_list0[1][l1ref1[0]];
1640
                const int scale = dist_scale_factor[ref0];
1641
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1642
                int mv_l0[2];
1643
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1644
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1645
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1646
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1647
                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);
1648
            }
1649
        }else{
1650
            for(i8=0; i8<4; i8++){
1651
                const int x8 = i8&1;
1652
                const int y8 = i8>>1;
1653
                int ref0, scale;
1654
                const int16_t (*l1mv)[2]= l1mv0;
1655

    
1656
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1657
                    continue;
1658
                h->sub_mb_type[i8] = sub_mb_type;
1659
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1660
                if(IS_INTRA(mb_type_col)){
1661
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1662
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1663
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1664
                    continue;
1665
                }
1666

    
1667
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1668
                if(ref0 >= 0)
1669
                    ref0 = map_col_to_list0[0][ref0];
1670
                else{
1671
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1672
                    l1mv= l1mv1;
1673
                }
1674
                scale = dist_scale_factor[ref0];
1675

    
1676
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1677
                if(IS_SUB_8X8(sub_mb_type)){
1678
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1679
                    int mx = (scale * mv_col[0] + 128) >> 8;
1680
                    int my = (scale * mv_col[1] + 128) >> 8;
1681
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1682
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1683
                }else
1684
                for(i4=0; i4<4; i4++){
1685
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1686
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1687
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1688
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1689
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1690
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1691
                }
1692
            }
1693
        }
1694
    }
1695
}
1696

    
1697
static inline void write_back_motion(H264Context *h, int mb_type){
1698
    MpegEncContext * const s = &h->s;
1699
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1700
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1701
    int list;
1702

    
1703
    if(!USES_LIST(mb_type, 0))
1704
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1705

    
1706
    for(list=0; list<2; list++){
1707
        int y;
1708
        if(!USES_LIST(mb_type, list))
1709
            continue;
1710

    
1711
        for(y=0; y<4; y++){
1712
            *(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];
1713
            *(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];
1714
        }
1715
        if( h->pps.cabac ) {
1716
            for(y=0; y<4; y++){
1717
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1718
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1719
            }
1720
        }
1721

    
1722
        {
1723
            uint8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1724
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1725
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1726
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1727
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1728
        }
1729
    }
1730

    
1731
    if(h->slice_type == B_TYPE && h->pps.cabac){
1732
        if(IS_8X8(mb_type)){
1733
            uint8_t *direct_table = &h->direct_table[b8_xy];
1734
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1735
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1736
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1737
        }
1738
    }
1739
}
1740

    
1741
/**
1742
 * Decodes a network abstraction layer unit.
1743
 * @param consumed is the number of bytes used as input
1744
 * @param length is the length of the array
1745
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1746
 * @returns decoded bytes, might be src+1 if no escapes
1747
 */
1748
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1749
    int i, si, di;
1750
    uint8_t *dst;
1751

    
1752
//    src[0]&0x80;                //forbidden bit
1753
    h->nal_ref_idc= src[0]>>5;
1754
    h->nal_unit_type= src[0]&0x1F;
1755

    
1756
    src++; length--;
1757
#if 0
1758
    for(i=0; i<length; i++)
1759
        printf("%2X ", src[i]);
1760
#endif
1761
    for(i=0; i+1<length; i+=2){
1762
        if(src[i]) continue;
1763
        if(i>0 && src[i-1]==0) i--;
1764
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1765
            if(src[i+2]!=3){
1766
                /* startcode, so we must be past the end */
1767
                length=i;
1768
            }
1769
            break;
1770
        }
1771
    }
1772

    
1773
    if(i>=length-1){ //no escaped 0
1774
        *dst_length= length;
1775
        *consumed= length+1; //+1 for the header
1776
        return src;
1777
    }
1778

    
1779
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1780
    dst= h->rbsp_buffer;
1781

    
1782
//printf("decoding esc\n");
1783
    si=di=0;
1784
    while(si<length){
1785
        //remove escapes (very rare 1:2^22)
1786
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1787
            if(src[si+2]==3){ //escape
1788
                dst[di++]= 0;
1789
                dst[di++]= 0;
1790
                si+=3;
1791
                continue;
1792
            }else //next start code
1793
                break;
1794
        }
1795

    
1796
        dst[di++]= src[si++];
1797
    }
1798

    
1799
    *dst_length= di;
1800
    *consumed= si + 1;//+1 for the header
1801
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1802
    return dst;
1803
}
1804

    
1805
#if 0
1806
/**
1807
 * @param src the data which should be escaped
1808
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1809
 * @param length the length of the src data
1810
 * @param dst_length the length of the dst array
1811
 * @returns length of escaped data in bytes or -1 if an error occured
1812
 */
1813
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1814
    int i, escape_count, si, di;
1815
    uint8_t *temp;
1816

1817
    assert(length>=0);
1818
    assert(dst_length>0);
1819

1820
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1821

1822
    if(length==0) return 1;
1823

1824
    escape_count= 0;
1825
    for(i=0; i<length; i+=2){
1826
        if(src[i]) continue;
1827
        if(i>0 && src[i-1]==0)
1828
            i--;
1829
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1830
            escape_count++;
1831
            i+=2;
1832
        }
1833
    }
1834

1835
    if(escape_count==0){
1836
        if(dst+1 != src)
1837
            memcpy(dst+1, src, length);
1838
        return length + 1;
1839
    }
1840

1841
    if(length + escape_count + 1> dst_length)
1842
        return -1;
1843

1844
    //this should be damn rare (hopefully)
1845

1846
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1847
    temp= h->rbsp_buffer;
1848
//printf("encoding esc\n");
1849

1850
    si= 0;
1851
    di= 0;
1852
    while(si < length){
1853
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1854
            temp[di++]= 0; si++;
1855
            temp[di++]= 0; si++;
1856
            temp[di++]= 3;
1857
            temp[di++]= src[si++];
1858
        }
1859
        else
1860
            temp[di++]= src[si++];
1861
    }
1862
    memcpy(dst+1, temp, length+escape_count);
1863

1864
    assert(di == length+escape_count);
1865

1866
    return di + 1;
1867
}
1868

1869
/**
1870
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1871
 */
1872
static void encode_rbsp_trailing(PutBitContext *pb){
1873
    int length;
1874
    put_bits(pb, 1, 1);
1875
    length= (-put_bits_count(pb))&7;
1876
    if(length) put_bits(pb, length, 0);
1877
}
1878
#endif
1879

    
1880
/**
1881
 * identifies the exact end of the bitstream
1882
 * @return the length of the trailing, or 0 if damaged
1883
 */
1884
static int decode_rbsp_trailing(uint8_t *src){
1885
    int v= *src;
1886
    int r;
1887

    
1888
    tprintf("rbsp trailing %X\n", v);
1889

    
1890
    for(r=1; r<9; r++){
1891
        if(v&1) return r;
1892
        v>>=1;
1893
    }
1894
    return 0;
1895
}
1896

    
1897
/**
1898
 * idct tranforms the 16 dc values and dequantize them.
1899
 * @param qp quantization parameter
1900
 */
1901
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1902
#define stride 16
1903
    int i;
1904
    int temp[16]; //FIXME check if this is a good idea
1905
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1906
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1907

    
1908
//memset(block, 64, 2*256);
1909
//return;
1910
    for(i=0; i<4; i++){
1911
        const int offset= y_offset[i];
1912
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1913
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1914
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1915
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1916

    
1917
        temp[4*i+0]= z0+z3;
1918
        temp[4*i+1]= z1+z2;
1919
        temp[4*i+2]= z1-z2;
1920
        temp[4*i+3]= z0-z3;
1921
    }
1922

    
1923
    for(i=0; i<4; i++){
1924
        const int offset= x_offset[i];
1925
        const int z0= temp[4*0+i] + temp[4*2+i];
1926
        const int z1= temp[4*0+i] - temp[4*2+i];
1927
        const int z2= temp[4*1+i] - temp[4*3+i];
1928
        const int z3= temp[4*1+i] + temp[4*3+i];
1929

    
1930
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1931
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1932
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1933
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1934
    }
1935
}
1936

    
1937
#if 0
1938
/**
1939
 * dct tranforms the 16 dc values.
1940
 * @param qp quantization parameter ??? FIXME
1941
 */
1942
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1943
//    const int qmul= dequant_coeff[qp][0];
1944
    int i;
1945
    int temp[16]; //FIXME check if this is a good idea
1946
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1947
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1948

1949
    for(i=0; i<4; i++){
1950
        const int offset= y_offset[i];
1951
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1952
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1953
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1954
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1955

1956
        temp[4*i+0]= z0+z3;
1957
        temp[4*i+1]= z1+z2;
1958
        temp[4*i+2]= z1-z2;
1959
        temp[4*i+3]= z0-z3;
1960
    }
1961

1962
    for(i=0; i<4; i++){
1963
        const int offset= x_offset[i];
1964
        const int z0= temp[4*0+i] + temp[4*2+i];
1965
        const int z1= temp[4*0+i] - temp[4*2+i];
1966
        const int z2= temp[4*1+i] - temp[4*3+i];
1967
        const int z3= temp[4*1+i] + temp[4*3+i];
1968

1969
        block[stride*0 +offset]= (z0 + z3)>>1;
1970
        block[stride*2 +offset]= (z1 + z2)>>1;
1971
        block[stride*8 +offset]= (z1 - z2)>>1;
1972
        block[stride*10+offset]= (z0 - z3)>>1;
1973
    }
1974
}
1975
#endif
1976

    
1977
#undef xStride
1978
#undef stride
1979

    
1980
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1981
    const int stride= 16*2;
1982
    const int xStride= 16;
1983
    int a,b,c,d,e;
1984

    
1985
    a= block[stride*0 + xStride*0];
1986
    b= block[stride*0 + xStride*1];
1987
    c= block[stride*1 + xStride*0];
1988
    d= block[stride*1 + xStride*1];
1989

    
1990
    e= a-b;
1991
    a= a+b;
1992
    b= c-d;
1993
    c= c+d;
1994

    
1995
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1996
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1997
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1998
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1999
}
2000

    
2001
#if 0
2002
static void chroma_dc_dct_c(DCTELEM *block){
2003
    const int stride= 16*2;
2004
    const int xStride= 16;
2005
    int a,b,c,d,e;
2006

2007
    a= block[stride*0 + xStride*0];
2008
    b= block[stride*0 + xStride*1];
2009
    c= block[stride*1 + xStride*0];
2010
    d= block[stride*1 + xStride*1];
2011

2012
    e= a-b;
2013
    a= a+b;
2014
    b= c-d;
2015
    c= c+d;
2016

2017
    block[stride*0 + xStride*0]= (a+c);
2018
    block[stride*0 + xStride*1]= (e+b);
2019
    block[stride*1 + xStride*0]= (a-c);
2020
    block[stride*1 + xStride*1]= (e-b);
2021
}
2022
#endif
2023

    
2024
/**
2025
 * gets the chroma qp.
2026
 */
2027
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
2028

    
2029
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
2030
}
2031

    
2032

    
2033
#if 0
2034
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
2035
    int i;
2036
    //FIXME try int temp instead of block
2037

2038
    for(i=0; i<4; i++){
2039
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
2040
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
2041
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
2042
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
2043
        const int z0= d0 + d3;
2044
        const int z3= d0 - d3;
2045
        const int z1= d1 + d2;
2046
        const int z2= d1 - d2;
2047

2048
        block[0 + 4*i]=   z0 +   z1;
2049
        block[1 + 4*i]= 2*z3 +   z2;
2050
        block[2 + 4*i]=   z0 -   z1;
2051
        block[3 + 4*i]=   z3 - 2*z2;
2052
    }
2053

2054
    for(i=0; i<4; i++){
2055
        const int z0= block[0*4 + i] + block[3*4 + i];
2056
        const int z3= block[0*4 + i] - block[3*4 + i];
2057
        const int z1= block[1*4 + i] + block[2*4 + i];
2058
        const int z2= block[1*4 + i] - block[2*4 + i];
2059

2060
        block[0*4 + i]=   z0 +   z1;
2061
        block[1*4 + i]= 2*z3 +   z2;
2062
        block[2*4 + i]=   z0 -   z1;
2063
        block[3*4 + i]=   z3 - 2*z2;
2064
    }
2065
}
2066
#endif
2067

    
2068
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
2069
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
2070
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
2071
    int i;
2072
    const int * const quant_table= quant_coeff[qscale];
2073
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
2074
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
2075
    const unsigned int threshold2= (threshold1<<1);
2076
    int last_non_zero;
2077

    
2078
    if(seperate_dc){
2079
        if(qscale<=18){
2080
            //avoid overflows
2081
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
2082
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
2083
            const unsigned int dc_threshold2= (dc_threshold1<<1);
2084

    
2085
            int level= block[0]*quant_coeff[qscale+18][0];
2086
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
2087
                if(level>0){
2088
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
2089
                    block[0]= level;
2090
                }else{
2091
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
2092
                    block[0]= -level;
2093
                }
2094
//                last_non_zero = i;
2095
            }else{
2096
                block[0]=0;
2097
            }
2098
        }else{
2099
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
2100
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
2101
            const unsigned int dc_threshold2= (dc_threshold1<<1);
2102

    
2103
            int level= block[0]*quant_table[0];
2104
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
2105
                if(level>0){
2106
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
2107
                    block[0]= level;
2108
                }else{
2109
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
2110
                    block[0]= -level;
2111
                }
2112
//                last_non_zero = i;
2113
            }else{
2114
                block[0]=0;
2115
            }
2116
        }
2117
        last_non_zero= 0;
2118
        i=1;
2119
    }else{
2120
        last_non_zero= -1;
2121
        i=0;
2122
    }
2123

    
2124
    for(; i<16; i++){
2125
        const int j= scantable[i];
2126
        int level= block[j]*quant_table[j];
2127

    
2128
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
2129
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
2130
        if(((unsigned)(level+threshold1))>threshold2){
2131
            if(level>0){
2132
                level= (bias + level)>>QUANT_SHIFT;
2133
                block[j]= level;
2134
            }else{
2135
                level= (bias - level)>>QUANT_SHIFT;
2136
                block[j]= -level;
2137
            }
2138
            last_non_zero = i;
2139
        }else{
2140
            block[j]=0;
2141
        }
2142
    }
2143

    
2144
    return last_non_zero;
2145
}
2146

    
2147
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
2148
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2149
    ((uint32_t*)(src+0*stride))[0]= a;
2150
    ((uint32_t*)(src+1*stride))[0]= a;
2151
    ((uint32_t*)(src+2*stride))[0]= a;
2152
    ((uint32_t*)(src+3*stride))[0]= a;
2153
}
2154

    
2155
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
2156
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
2157
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
2158
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
2159
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
2160
}
2161

    
2162
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
2163
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
2164
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
2165

    
2166
    ((uint32_t*)(src+0*stride))[0]=
2167
    ((uint32_t*)(src+1*stride))[0]=
2168
    ((uint32_t*)(src+2*stride))[0]=
2169
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2170
}
2171

    
2172
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
2173
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
2174

    
2175
    ((uint32_t*)(src+0*stride))[0]=
2176
    ((uint32_t*)(src+1*stride))[0]=
2177
    ((uint32_t*)(src+2*stride))[0]=
2178
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2179
}
2180

    
2181
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
2182
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
2183

    
2184
    ((uint32_t*)(src+0*stride))[0]=
2185
    ((uint32_t*)(src+1*stride))[0]=
2186
    ((uint32_t*)(src+2*stride))[0]=
2187
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2188
}
2189

    
2190
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
2191
    ((uint32_t*)(src+0*stride))[0]=
2192
    ((uint32_t*)(src+1*stride))[0]=
2193
    ((uint32_t*)(src+2*stride))[0]=
2194
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
2195
}
2196

    
2197

    
2198
#define LOAD_TOP_RIGHT_EDGE\
2199
    const int t4= topright[0];\
2200
    const int t5= topright[1];\
2201
    const int t6= topright[2];\
2202
    const int t7= topright[3];\
2203

    
2204
#define LOAD_LEFT_EDGE\
2205
    const int l0= src[-1+0*stride];\
2206
    const int l1= src[-1+1*stride];\
2207
    const int l2= src[-1+2*stride];\
2208
    const int l3= src[-1+3*stride];\
2209

    
2210
#define LOAD_TOP_EDGE\
2211
    const int t0= src[ 0-1*stride];\
2212
    const int t1= src[ 1-1*stride];\
2213
    const int t2= src[ 2-1*stride];\
2214
    const int t3= src[ 3-1*stride];\
2215

    
2216
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
2217
    const int lt= src[-1-1*stride];
2218
    LOAD_TOP_EDGE
2219
    LOAD_LEFT_EDGE
2220

    
2221
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
2222
    src[0+2*stride]=
2223
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
2224
    src[0+1*stride]=
2225
    src[1+2*stride]=
2226
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
2227
    src[0+0*stride]=
2228
    src[1+1*stride]=
2229
    src[2+2*stride]=
2230
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2231
    src[1+0*stride]=
2232
    src[2+1*stride]=
2233
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
2234
    src[2+0*stride]=
2235
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2236
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2237
}
2238

    
2239
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
2240
    LOAD_TOP_EDGE
2241
    LOAD_TOP_RIGHT_EDGE
2242
//    LOAD_LEFT_EDGE
2243

    
2244
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
2245
    src[1+0*stride]=
2246
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
2247
    src[2+0*stride]=
2248
    src[1+1*stride]=
2249
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
2250
    src[3+0*stride]=
2251
    src[2+1*stride]=
2252
    src[1+2*stride]=
2253
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
2254
    src[3+1*stride]=
2255
    src[2+2*stride]=
2256
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
2257
    src[3+2*stride]=
2258
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
2259
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
2260
}
2261

    
2262
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
2263
    const int lt= src[-1-1*stride];
2264
    LOAD_TOP_EDGE
2265
    LOAD_LEFT_EDGE
2266
    const __attribute__((unused)) int unu= l3;
2267

    
2268
    src[0+0*stride]=
2269
    src[1+2*stride]=(lt + t0 + 1)>>1;
2270
    src[1+0*stride]=
2271
    src[2+2*stride]=(t0 + t1 + 1)>>1;
2272
    src[2+0*stride]=
2273
    src[3+2*stride]=(t1 + t2 + 1)>>1;
2274
    src[3+0*stride]=(t2 + t3 + 1)>>1;
2275
    src[0+1*stride]=
2276
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2277
    src[1+1*stride]=
2278
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2279
    src[2+1*stride]=
2280
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2281
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2282
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2283
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2284
}
2285

    
2286
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
2287
    LOAD_TOP_EDGE
2288
    LOAD_TOP_RIGHT_EDGE
2289
    const __attribute__((unused)) int unu= t7;
2290

    
2291
    src[0+0*stride]=(t0 + t1 + 1)>>1;
2292
    src[1+0*stride]=
2293
    src[0+2*stride]=(t1 + t2 + 1)>>1;
2294
    src[2+0*stride]=
2295
    src[1+2*stride]=(t2 + t3 + 1)>>1;
2296
    src[3+0*stride]=
2297
    src[2+2*stride]=(t3 + t4+ 1)>>1;
2298
    src[3+2*stride]=(t4 + t5+ 1)>>1;
2299
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2300
    src[1+1*stride]=
2301
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2302
    src[2+1*stride]=
2303
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2304
    src[3+1*stride]=
2305
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2306
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2307
}
2308

    
2309
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2310
    LOAD_LEFT_EDGE
2311

    
2312
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2313
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2314
    src[2+0*stride]=
2315
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2316
    src[3+0*stride]=
2317
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2318
    src[2+1*stride]=
2319
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2320
    src[3+1*stride]=
2321
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2322
    src[3+2*stride]=
2323
    src[1+3*stride]=
2324
    src[0+3*stride]=
2325
    src[2+2*stride]=
2326
    src[2+3*stride]=
2327
    src[3+3*stride]=l3;
2328
}
2329

    
2330
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2331
    const int lt= src[-1-1*stride];
2332
    LOAD_TOP_EDGE
2333
    LOAD_LEFT_EDGE
2334
    const __attribute__((unused)) int unu= t3;
2335

    
2336
    src[0+0*stride]=
2337
    src[2+1*stride]=(lt + l0 + 1)>>1;
2338
    src[1+0*stride]=
2339
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2340
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2341
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2342
    src[0+1*stride]=
2343
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2344
    src[1+1*stride]=
2345
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2346
    src[0+2*stride]=
2347
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2348
    src[1+2*stride]=
2349
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2350
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2351
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2352
}
2353

    
2354
static void pred16x16_vertical_c(uint8_t *src, int stride){
2355
    int i;
2356
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2357
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2358
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2359
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2360

    
2361
    for(i=0; i<16; i++){
2362
        ((uint32_t*)(src+i*stride))[0]= a;
2363
        ((uint32_t*)(src+i*stride))[1]= b;
2364
        ((uint32_t*)(src+i*stride))[2]= c;
2365
        ((uint32_t*)(src+i*stride))[3]= d;
2366
    }
2367
}
2368

    
2369
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2370
    int i;
2371

    
2372
    for(i=0; i<16; i++){
2373
        ((uint32_t*)(src+i*stride))[0]=
2374
        ((uint32_t*)(src+i*stride))[1]=
2375
        ((uint32_t*)(src+i*stride))[2]=
2376
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2377
    }
2378
}
2379

    
2380
static void pred16x16_dc_c(uint8_t *src, int stride){
2381
    int i, dc=0;
2382

    
2383
    for(i=0;i<16; i++){
2384
        dc+= src[-1+i*stride];
2385
    }
2386

    
2387
    for(i=0;i<16; i++){
2388
        dc+= src[i-stride];
2389
    }
2390

    
2391
    dc= 0x01010101*((dc + 16)>>5);
2392

    
2393
    for(i=0; i<16; i++){
2394
        ((uint32_t*)(src+i*stride))[0]=
2395
        ((uint32_t*)(src+i*stride))[1]=
2396
        ((uint32_t*)(src+i*stride))[2]=
2397
        ((uint32_t*)(src+i*stride))[3]= dc;
2398
    }
2399
}
2400

    
2401
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2402
    int i, dc=0;
2403

    
2404
    for(i=0;i<16; i++){
2405
        dc+= src[-1+i*stride];
2406
    }
2407

    
2408
    dc= 0x01010101*((dc + 8)>>4);
2409

    
2410
    for(i=0; i<16; i++){
2411
        ((uint32_t*)(src+i*stride))[0]=
2412
        ((uint32_t*)(src+i*stride))[1]=
2413
        ((uint32_t*)(src+i*stride))[2]=
2414
        ((uint32_t*)(src+i*stride))[3]= dc;
2415
    }
2416
}
2417

    
2418
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2419
    int i, dc=0;
2420

    
2421
    for(i=0;i<16; i++){
2422
        dc+= src[i-stride];
2423
    }
2424
    dc= 0x01010101*((dc + 8)>>4);
2425

    
2426
    for(i=0; i<16; i++){
2427
        ((uint32_t*)(src+i*stride))[0]=
2428
        ((uint32_t*)(src+i*stride))[1]=
2429
        ((uint32_t*)(src+i*stride))[2]=
2430
        ((uint32_t*)(src+i*stride))[3]= dc;
2431
    }
2432
}
2433

    
2434
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2435
    int i;
2436

    
2437
    for(i=0; i<16; i++){
2438
        ((uint32_t*)(src+i*stride))[0]=
2439
        ((uint32_t*)(src+i*stride))[1]=
2440
        ((uint32_t*)(src+i*stride))[2]=
2441
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2442
    }
2443
}
2444

    
2445
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2446
  int i, j, k;
2447
  int a;
2448
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2449
  const uint8_t * const src0 = src+7-stride;
2450
  const uint8_t *src1 = src+8*stride-1;
2451
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2452
  int H = src0[1] - src0[-1];
2453
  int V = src1[0] - src2[ 0];
2454
  for(k=2; k<=8; ++k) {
2455
    src1 += stride; src2 -= stride;
2456
    H += k*(src0[k] - src0[-k]);
2457
    V += k*(src1[0] - src2[ 0]);
2458
  }
2459
  if(svq3){
2460
    H = ( 5*(H/4) ) / 16;
2461
    V = ( 5*(V/4) ) / 16;
2462

    
2463
    /* required for 100% accuracy */
2464
    i = H; H = V; V = i;
2465
  }else{
2466
    H = ( 5*H+32 ) >> 6;
2467
    V = ( 5*V+32 ) >> 6;
2468
  }
2469

    
2470
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2471
  for(j=16; j>0; --j) {
2472
    int b = a;
2473
    a += V;
2474
    for(i=-16; i<0; i+=4) {
2475
      src[16+i] = cm[ (b    ) >> 5 ];
2476
      src[17+i] = cm[ (b+  H) >> 5 ];
2477
      src[18+i] = cm[ (b+2*H) >> 5 ];
2478
      src[19+i] = cm[ (b+3*H) >> 5 ];
2479
      b += 4*H;
2480
    }
2481
    src += stride;
2482
  }
2483
}
2484

    
2485
static void pred16x16_plane_c(uint8_t *src, int stride){
2486
    pred16x16_plane_compat_c(src, stride, 0);
2487
}
2488

    
2489
static void pred8x8_vertical_c(uint8_t *src, int stride){
2490
    int i;
2491
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2492
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2493

    
2494
    for(i=0; i<8; i++){
2495
        ((uint32_t*)(src+i*stride))[0]= a;
2496
        ((uint32_t*)(src+i*stride))[1]= b;
2497
    }
2498
}
2499

    
2500
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2501
    int i;
2502

    
2503
    for(i=0; i<8; i++){
2504
        ((uint32_t*)(src+i*stride))[0]=
2505
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2506
    }
2507
}
2508

    
2509
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2510
    int i;
2511

    
2512
    for(i=0; i<8; i++){
2513
        ((uint32_t*)(src+i*stride))[0]=
2514
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2515
    }
2516
}
2517

    
2518
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2519
    int i;
2520
    int dc0, dc2;
2521

    
2522
    dc0=dc2=0;
2523
    for(i=0;i<4; i++){
2524
        dc0+= src[-1+i*stride];
2525
        dc2+= src[-1+(i+4)*stride];
2526
    }
2527
    dc0= 0x01010101*((dc0 + 2)>>2);
2528
    dc2= 0x01010101*((dc2 + 2)>>2);
2529

    
2530
    for(i=0; i<4; i++){
2531
        ((uint32_t*)(src+i*stride))[0]=
2532
        ((uint32_t*)(src+i*stride))[1]= dc0;
2533
    }
2534
    for(i=4; i<8; i++){
2535
        ((uint32_t*)(src+i*stride))[0]=
2536
        ((uint32_t*)(src+i*stride))[1]= dc2;
2537
    }
2538
}
2539

    
2540
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2541
    int i;
2542
    int dc0, dc1;
2543

    
2544
    dc0=dc1=0;
2545
    for(i=0;i<4; i++){
2546
        dc0+= src[i-stride];
2547
        dc1+= src[4+i-stride];
2548
    }
2549
    dc0= 0x01010101*((dc0 + 2)>>2);
2550
    dc1= 0x01010101*((dc1 + 2)>>2);
2551

    
2552
    for(i=0; i<4; i++){
2553
        ((uint32_t*)(src+i*stride))[0]= dc0;
2554
        ((uint32_t*)(src+i*stride))[1]= dc1;
2555
    }
2556
    for(i=4; i<8; i++){
2557
        ((uint32_t*)(src+i*stride))[0]= dc0;
2558
        ((uint32_t*)(src+i*stride))[1]= dc1;
2559
    }
2560
}
2561

    
2562

    
2563
static void pred8x8_dc_c(uint8_t *src, int stride){
2564
    int i;
2565
    int dc0, dc1, dc2, dc3;
2566

    
2567
    dc0=dc1=dc2=0;
2568
    for(i=0;i<4; i++){
2569
        dc0+= src[-1+i*stride] + src[i-stride];
2570
        dc1+= src[4+i-stride];
2571
        dc2+= src[-1+(i+4)*stride];
2572
    }
2573
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2574
    dc0= 0x01010101*((dc0 + 4)>>3);
2575
    dc1= 0x01010101*((dc1 + 2)>>2);
2576
    dc2= 0x01010101*((dc2 + 2)>>2);
2577

    
2578
    for(i=0; i<4; i++){
2579
        ((uint32_t*)(src+i*stride))[0]= dc0;
2580
        ((uint32_t*)(src+i*stride))[1]= dc1;
2581
    }
2582
    for(i=4; i<8; i++){
2583
        ((uint32_t*)(src+i*stride))[0]= dc2;
2584
        ((uint32_t*)(src+i*stride))[1]= dc3;
2585
    }
2586
}
2587

    
2588
static void pred8x8_plane_c(uint8_t *src, int stride){
2589
  int j, k;
2590
  int a;
2591
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2592
  const uint8_t * const src0 = src+3-stride;
2593
  const uint8_t *src1 = src+4*stride-1;
2594
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2595
  int H = src0[1] - src0[-1];
2596
  int V = src1[0] - src2[ 0];
2597
  for(k=2; k<=4; ++k) {
2598
    src1 += stride; src2 -= stride;
2599
    H += k*(src0[k] - src0[-k]);
2600
    V += k*(src1[0] - src2[ 0]);
2601
  }
2602
  H = ( 17*H+16 ) >> 5;
2603
  V = ( 17*V+16 ) >> 5;
2604

    
2605
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2606
  for(j=8; j>0; --j) {
2607
    int b = a;
2608
    a += V;
2609
    src[0] = cm[ (b    ) >> 5 ];
2610
    src[1] = cm[ (b+  H) >> 5 ];
2611
    src[2] = cm[ (b+2*H) >> 5 ];
2612
    src[3] = cm[ (b+3*H) >> 5 ];
2613
    src[4] = cm[ (b+4*H) >> 5 ];
2614
    src[5] = cm[ (b+5*H) >> 5 ];
2615
    src[6] = cm[ (b+6*H) >> 5 ];
2616
    src[7] = cm[ (b+7*H) >> 5 ];
2617
    src += stride;
2618
  }
2619
}
2620

    
2621
#define SRC(x,y) src[(x)+(y)*stride]
2622
#define PL(y) \
2623
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2624
#define PREDICT_8x8_LOAD_LEFT \
2625
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2626
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2627
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2628
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2629

    
2630
#define PT(x) \
2631
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2632
#define PREDICT_8x8_LOAD_TOP \
2633
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2634
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2635
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2636
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2637
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2638

    
2639
#define PTR(x) \
2640
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2641
#define PREDICT_8x8_LOAD_TOPRIGHT \
2642
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2643
    if(has_topright) { \
2644
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2645
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2646
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2647

    
2648
#define PREDICT_8x8_LOAD_TOPLEFT \
2649
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2650

    
2651
#define PREDICT_8x8_DC(v) \
2652
    int y; \
2653
    for( y = 0; y < 8; y++ ) { \
2654
        ((uint32_t*)src)[0] = \
2655
        ((uint32_t*)src)[1] = v; \
2656
        src += stride; \
2657
    }
2658

    
2659
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2660
{
2661
    PREDICT_8x8_DC(0x80808080);
2662
}
2663
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2664
{
2665
    PREDICT_8x8_LOAD_LEFT;
2666
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2667
    PREDICT_8x8_DC(dc);
2668
}
2669
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2670
{
2671
    PREDICT_8x8_LOAD_TOP;
2672
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2673
    PREDICT_8x8_DC(dc);
2674
}
2675
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2676
{
2677
    PREDICT_8x8_LOAD_LEFT;
2678
    PREDICT_8x8_LOAD_TOP;
2679
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2680
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2681
    PREDICT_8x8_DC(dc);
2682
}
2683
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2684
{
2685
    PREDICT_8x8_LOAD_LEFT;
2686
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2687
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2688
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2689
#undef ROW
2690
}
2691
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2692
{
2693
    int y;
2694
    PREDICT_8x8_LOAD_TOP;
2695
    src[0] = t0;
2696
    src[1] = t1;
2697
    src[2] = t2;
2698
    src[3] = t3;
2699
    src[4] = t4;
2700
    src[5] = t5;
2701
    src[6] = t6;
2702
    src[7] = t7;
2703
    for( y = 1; y < 8; y++ )
2704
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2705
}
2706
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2707
{
2708
    PREDICT_8x8_LOAD_TOP;
2709
    PREDICT_8x8_LOAD_TOPRIGHT;
2710
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2711
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2712
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2713
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2714
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2715
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2716
    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;
2717
    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;
2718
    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;
2719
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2720
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2721
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2722
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2723
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2724
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2725
}
2726
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2727
{
2728
    PREDICT_8x8_LOAD_TOP;
2729
    PREDICT_8x8_LOAD_LEFT;
2730
    PREDICT_8x8_LOAD_TOPLEFT;
2731
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2732
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2733
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2734
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2735
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2736
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2737
    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;
2738
    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;
2739
    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;
2740
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2741
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2742
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2743
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2744
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2745
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2746

    
2747
}
2748
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2749
{
2750
    PREDICT_8x8_LOAD_TOP;
2751
    PREDICT_8x8_LOAD_LEFT;
2752
    PREDICT_8x8_LOAD_TOPLEFT;
2753
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2754
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2755
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2756
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2757
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2758
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2759
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2760
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2761
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2762
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2763
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2764
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2765
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2766
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2767
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2768
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2769
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2770
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2771
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2772
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2773
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2774
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2775
}
2776
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2777
{
2778
    PREDICT_8x8_LOAD_TOP;
2779
    PREDICT_8x8_LOAD_LEFT;
2780
    PREDICT_8x8_LOAD_TOPLEFT;
2781
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2782
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2783
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2784
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2785
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2786
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2787
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2788
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2789
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2790
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2791
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2792
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2793
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2794
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2795
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2796
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2797
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2798
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2799
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2800
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2801
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2802
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2803
}
2804
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2805
{
2806
    PREDICT_8x8_LOAD_TOP;
2807
    PREDICT_8x8_LOAD_TOPRIGHT;
2808
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2809
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2810
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2811
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2812
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2813
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2814
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2815
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2816
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2817
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2818
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2819
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2820
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2821
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2822
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2823
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2824
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2825
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2826
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2827
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2828
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2829
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2830
}
2831
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2832
{
2833
    PREDICT_8x8_LOAD_LEFT;
2834
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2835
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2836
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2837
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2838
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2839
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2840
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2841
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2842
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2843
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2844
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2845
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2846
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2847
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2848
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2849
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2850
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2851
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2852
}
2853
#undef PREDICT_8x8_LOAD_LEFT
2854
#undef PREDICT_8x8_LOAD_TOP
2855
#undef PREDICT_8x8_LOAD_TOPLEFT
2856
#undef PREDICT_8x8_LOAD_TOPRIGHT
2857
#undef PREDICT_8x8_DC
2858
#undef PTR
2859
#undef PT
2860
#undef PL
2861
#undef SRC
2862

    
2863
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2864
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2865
                           int src_x_offset, int src_y_offset,
2866
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2867
    MpegEncContext * const s = &h->s;
2868
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2869
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2870
    const int luma_xy= (mx&3) + ((my&3)<<2);
2871
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2872
    uint8_t * src_cb, * src_cr;
2873
    int extra_width= h->emu_edge_width;
2874
    int extra_height= h->emu_edge_height;
2875
    int emu=0;
2876
    const int full_mx= mx>>2;
2877
    const int full_my= my>>2;
2878
    const int pic_width  = 16*s->mb_width;
2879
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
2880

    
2881
    if(!pic->data[0])
2882
        return;
2883

    
2884
    if(mx&7) extra_width -= 3;
2885
    if(my&7) extra_height -= 3;
2886

    
2887
    if(   full_mx < 0-extra_width
2888
       || full_my < 0-extra_height
2889
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2890
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2891
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
2892
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2893
        emu=1;
2894
    }
2895

    
2896
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2897
    if(!square){
2898
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2899
    }
2900

    
2901
    if(s->flags&CODEC_FLAG_GRAY) return;
2902

    
2903
    if(MB_MBAFF){
2904
        // chroma offset when predicting from a field of opposite parity
2905
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2906
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2907
    }
2908
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2909
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2910

    
2911
    if(emu){
2912
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2913
            src_cb= s->edge_emu_buffer;
2914
    }
2915
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2916

    
2917
    if(emu){
2918
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2919
            src_cr= s->edge_emu_buffer;
2920
    }
2921
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2922
}
2923

    
2924
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2925
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2926
                           int x_offset, int y_offset,
2927
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2928
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2929
                           int list0, int list1){
2930
    MpegEncContext * const s = &h->s;
2931
    qpel_mc_func *qpix_op=  qpix_put;
2932
    h264_chroma_mc_func chroma_op= chroma_put;
2933

    
2934
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2935
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2936
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2937
    x_offset += 8*s->mb_x;
2938
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2939

    
2940
    if(list0){
2941
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2942
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2943
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2944
                           qpix_op, chroma_op);
2945

    
2946
        qpix_op=  qpix_avg;
2947
        chroma_op= chroma_avg;
2948
    }
2949

    
2950
    if(list1){
2951
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2952
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2953
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2954
                           qpix_op, chroma_op);
2955
    }
2956
}
2957

    
2958
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2959
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2960
                           int x_offset, int y_offset,
2961
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2962
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2963
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2964
                           int list0, int list1){
2965
    MpegEncContext * const s = &h->s;
2966

    
2967
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2968
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2969
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2970
    x_offset += 8*s->mb_x;
2971
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2972

    
2973
    if(list0 && list1){
2974
        /* don't optimize for luma-only case, since B-frames usually
2975
         * use implicit weights => chroma too. */
2976
        uint8_t *tmp_cb = s->obmc_scratchpad;
2977
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2978
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2979
        int refn0 = h->ref_cache[0][ scan8[n] ];
2980
        int refn1 = h->ref_cache[1][ scan8[n] ];
2981

    
2982
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2983
                    dest_y, dest_cb, dest_cr,
2984
                    x_offset, y_offset, qpix_put, chroma_put);
2985
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2986
                    tmp_y, tmp_cb, tmp_cr,
2987
                    x_offset, y_offset, qpix_put, chroma_put);
2988

    
2989
        if(h->use_weight == 2){
2990
            int weight0 = h->implicit_weight[refn0][refn1];
2991
            int weight1 = 64 - weight0;
2992
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
2993
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2994
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2995
        }else{
2996
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2997
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2998
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2999
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3000
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
3001
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
3002
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3003
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
3004
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
3005
        }
3006
    }else{
3007
        int list = list1 ? 1 : 0;
3008
        int refn = h->ref_cache[list][ scan8[n] ];
3009
        Picture *ref= &h->ref_list[list][refn];
3010
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
3011
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
3012
                    qpix_put, chroma_put);
3013

    
3014
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
3015
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
3016
        if(h->use_weight_chroma){
3017
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3018
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
3019
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3020
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
3021
        }
3022
    }
3023
}
3024

    
3025
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
3026
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3027
                           int x_offset, int y_offset,
3028
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
3029
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
3030
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
3031
                           int list0, int list1){
3032
    if((h->use_weight==2 && list0 && list1
3033
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
3034
       || h->use_weight==1)
3035
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
3036
                         x_offset, y_offset, qpix_put, chroma_put,
3037
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
3038
    else
3039
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
3040
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
3041
}
3042

    
3043
static inline void prefetch_motion(H264Context *h, int list){
3044
    /* fetch pixels for estimated mv 4 macroblocks ahead
3045
     * optimized for 64byte cache lines */
3046
    MpegEncContext * const s = &h->s;
3047
    const int refn = h->ref_cache[list][scan8[0]];
3048
    if(refn >= 0){
3049
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
3050
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
3051
        uint8_t **src= h->ref_list[list][refn].data;
3052
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
3053
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
3054
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
3055
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
3056
    }
3057
}
3058

    
3059
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3060
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
3061
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
3062
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
3063
    MpegEncContext * const s = &h->s;
3064
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3065
    const int mb_type= s->current_picture.mb_type[mb_xy];
3066

    
3067
    assert(IS_INTER(mb_type));
3068

    
3069
    prefetch_motion(h, 0);
3070

    
3071
    if(IS_16X16(mb_type)){
3072
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
3073
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
3074
                &weight_op[0], &weight_avg[0],
3075
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3076
    }else if(IS_16X8(mb_type)){
3077
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
3078
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
3079
                &weight_op[1], &weight_avg[1],
3080
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3081
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
3082
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
3083
                &weight_op[1], &weight_avg[1],
3084
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
3085
    }else if(IS_8X16(mb_type)){
3086
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
3087
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3088
                &weight_op[2], &weight_avg[2],
3089
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3090
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
3091
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3092
                &weight_op[2], &weight_avg[2],
3093
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
3094
    }else{
3095
        int i;
3096

    
3097
        assert(IS_8X8(mb_type));
3098

    
3099
        for(i=0; i<4; i++){
3100
            const int sub_mb_type= h->sub_mb_type[i];
3101
            const int n= 4*i;
3102
            int x_offset= (i&1)<<2;
3103
            int y_offset= (i&2)<<1;
3104

    
3105
            if(IS_SUB_8X8(sub_mb_type)){
3106
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3107
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3108
                    &weight_op[3], &weight_avg[3],
3109
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3110
            }else if(IS_SUB_8X4(sub_mb_type)){
3111
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3112
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
3113
                    &weight_op[4], &weight_avg[4],
3114
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3115
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
3116
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
3117
                    &weight_op[4], &weight_avg[4],
3118
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3119
            }else if(IS_SUB_4X8(sub_mb_type)){
3120
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3121
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3122
                    &weight_op[5], &weight_avg[5],
3123
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3124
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
3125
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3126
                    &weight_op[5], &weight_avg[5],
3127
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3128
            }else{
3129
                int j;
3130
                assert(IS_SUB_4X4(sub_mb_type));
3131
                for(j=0; j<4; j++){
3132
                    int sub_x_offset= x_offset + 2*(j&1);
3133
                    int sub_y_offset= y_offset +   (j&2);
3134
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
3135
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3136
                        &weight_op[6], &weight_avg[6],
3137
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3138
                }
3139
            }
3140
        }
3141
    }
3142

    
3143
    prefetch_motion(h, 1);
3144
}
3145

    
3146
static void decode_init_vlc(H264Context *h){
3147
    static int done = 0;
3148

    
3149
    if (!done) {
3150
        int i;
3151
        done = 1;
3152

    
3153
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
3154
                 &chroma_dc_coeff_token_len [0], 1, 1,
3155
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
3156

    
3157
        for(i=0; i<4; i++){
3158
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
3159
                     &coeff_token_len [i][0], 1, 1,
3160
                     &coeff_token_bits[i][0], 1, 1, 1);
3161
        }
3162

    
3163
        for(i=0; i<3; i++){
3164
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
3165
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
3166
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
3167
        }
3168
        for(i=0; i<15; i++){
3169
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
3170
                     &total_zeros_len [i][0], 1, 1,
3171
                     &total_zeros_bits[i][0], 1, 1, 1);
3172
        }
3173

    
3174
        for(i=0; i<6; i++){
3175
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
3176
                     &run_len [i][0], 1, 1,
3177
                     &run_bits[i][0], 1, 1, 1);
3178
        }
3179
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
3180
                 &run_len [6][0], 1, 1,
3181
                 &run_bits[6][0], 1, 1, 1);
3182
    }
3183
}
3184

    
3185
/**
3186
 * Sets the intra prediction function pointers.
3187
 */
3188
static void init_pred_ptrs(H264Context *h){
3189
//    MpegEncContext * const s = &h->s;
3190

    
3191
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
3192
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
3193
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
3194
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
3195
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
3196
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
3197
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
3198
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
3199
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
3200
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
3201
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
3202
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
3203

    
3204
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
3205
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
3206
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
3207
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
3208
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
3209
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
3210
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
3211
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
3212
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
3213
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
3214
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
3215
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
3216

    
3217
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
3218
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
3219
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
3220
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
3221
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
3222
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
3223
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
3224

    
3225
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
3226
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
3227
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
3228
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
3229
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
3230
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
3231
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
3232
}
3233

    
3234
static void free_tables(H264Context *h){
3235
    av_freep(&h->intra4x4_pred_mode);
3236
    av_freep(&h->chroma_pred_mode_table);
3237
    av_freep(&h->cbp_table);
3238
    av_freep(&h->mvd_table[0]);
3239
    av_freep(&h->mvd_table[1]);
3240
    av_freep(&h->direct_table);
3241
    av_freep(&h->non_zero_count);
3242
    av_freep(&h->slice_table_base);
3243
    av_freep(&h->top_borders[1]);
3244
    av_freep(&h->top_borders[0]);
3245
    h->slice_table= NULL;
3246

    
3247
    av_freep(&h->mb2b_xy);
3248
    av_freep(&h->mb2b8_xy);
3249

    
3250
    av_freep(&h->s.obmc_scratchpad);
3251
}
3252

    
3253
static void init_dequant8_coeff_table(H264Context *h){
3254
    int i,q,x;
3255
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
3256
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
3257
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
3258

    
3259
    for(i=0; i<2; i++ ){
3260
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
3261
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
3262
            break;
3263
        }
3264

    
3265
        for(q=0; q<52; q++){
3266
            int shift = div6[q];
3267
            int idx = rem6[q];
3268
            for(x=0; x<64; x++)
3269
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
3270
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
3271
                    h->pps.scaling_matrix8[i][x]) << shift;
3272
        }
3273
    }
3274
}
3275

    
3276
static void init_dequant4_coeff_table(H264Context *h){
3277
    int i,j,q,x;
3278
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
3279
    for(i=0; i<6; i++ ){
3280
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
3281
        for(j=0; j<i; j++){
3282
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
3283
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
3284
                break;
3285
            }
3286
        }
3287
        if(j<i)
3288
            continue;
3289

    
3290
        for(q=0; q<52; q++){
3291
            int shift = div6[q] + 2;
3292
            int idx = rem6[q];
3293
            for(x=0; x<16; x++)
3294
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
3295
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
3296
                    h->pps.scaling_matrix4[i][x]) << shift;
3297
        }
3298
    }
3299
}
3300

    
3301
static void init_dequant_tables(H264Context *h){
3302
    int i,x;
3303
    init_dequant4_coeff_table(h);
3304
    if(h->pps.transform_8x8_mode)
3305
        init_dequant8_coeff_table(h);
3306
    if(h->sps.transform_bypass){
3307
        for(i=0; i<6; i++)
3308
            for(x=0; x<16; x++)
3309
                h->dequant4_coeff[i][0][x] = 1<<6;
3310
        if(h->pps.transform_8x8_mode)
3311
            for(i=0; i<2; i++)
3312
                for(x=0; x<64; x++)
3313
                    h->dequant8_coeff[i][0][x] = 1<<6;
3314
    }
3315
}
3316

    
3317

    
3318
/**
3319
 * allocates tables.
3320
 * needs width/height
3321
 */
3322
static int alloc_tables(H264Context *h){
3323
    MpegEncContext * const s = &h->s;
3324
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
3325
    int x,y;
3326

    
3327
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
3328

    
3329
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
3330
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
3331
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3332
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3333
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
3334

    
3335
    if( h->pps.cabac ) {
3336
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
3337
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3338
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
3339
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
3340
    }
3341

    
3342
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
3343
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
3344

    
3345
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
3346
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
3347
    for(y=0; y<s->mb_height; y++){
3348
        for(x=0; x<s->mb_width; x++){
3349
            const int mb_xy= x + y*s->mb_stride;
3350
            const int b_xy = 4*x + 4*y*h->b_stride;
3351
            const int b8_xy= 2*x + 2*y*h->b8_stride;
3352

    
3353
            h->mb2b_xy [mb_xy]= b_xy;
3354
            h->mb2b8_xy[mb_xy]= b8_xy;
3355
        }
3356
    }
3357

    
3358
    s->obmc_scratchpad = NULL;
3359

    
3360
    if(!h->dequant4_coeff[0])
3361
        init_dequant_tables(h);
3362

    
3363
    return 0;
3364
fail:
3365
    free_tables(h);
3366
    return -1;
3367
}
3368

    
3369
static void common_init(H264Context *h){
3370
    MpegEncContext * const s = &h->s;
3371

    
3372
    s->width = s->avctx->width;
3373
    s->height = s->avctx->height;
3374
    s->codec_id= s->avctx->codec->id;
3375

    
3376
    init_pred_ptrs(h);
3377

    
3378
    h->dequant_coeff_pps= -1;
3379
    s->unrestricted_mv=1;
3380
    s->decode=1; //FIXME
3381

    
3382
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3383
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
3384
}
3385

    
3386
static int decode_init(AVCodecContext *avctx){
3387
    H264Context *h= avctx->priv_data;
3388
    MpegEncContext * const s = &h->s;
3389

    
3390
    MPV_decode_defaults(s);
3391

    
3392
    s->avctx = avctx;
3393
    common_init(h);
3394

    
3395
    s->out_format = FMT_H264;
3396
    s->workaround_bugs= avctx->workaround_bugs;
3397

    
3398
    // set defaults
3399
//    s->decode_mb= ff_h263_decode_mb;
3400
    s->low_delay= 1;
3401
    avctx->pix_fmt= PIX_FMT_YUV420P;
3402

    
3403
    decode_init_vlc(h);
3404

    
3405
    if(avctx->extradata_size > 0 && avctx->extradata &&
3406
       *(char *)avctx->extradata == 1){
3407
        h->is_avc = 1;
3408
        h->got_avcC = 0;
3409
    } else {
3410
        h->is_avc = 0;
3411
    }
3412

    
3413
    return 0;
3414
}
3415

    
3416
static int frame_start(H264Context *h){
3417
    MpegEncContext * const s = &h->s;
3418
    int i;
3419

    
3420
    if(MPV_frame_start(s, s->avctx) < 0)
3421
        return -1;
3422
    ff_er_frame_start(s);
3423

    
3424
    assert(s->linesize && s->uvlinesize);
3425

    
3426
    for(i=0; i<16; i++){
3427
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3428
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3429
    }
3430
    for(i=0; i<4; i++){
3431
        h->block_offset[16+i]=
3432
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3433
        h->block_offset[24+16+i]=
3434
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3435
    }
3436

    
3437
    /* can't be in alloc_tables because linesize isn't known there.
3438
     * FIXME: redo bipred weight to not require extra buffer? */
3439
    if(!s->obmc_scratchpad)
3440
        s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
3441

    
3442
    /* some macroblocks will be accessed before they're available */
3443
    if(FRAME_MBAFF)
3444
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
3445

    
3446
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3447
    return 0;
3448
}
3449

    
3450
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3451
    MpegEncContext * const s = &h->s;
3452
    int i;
3453

    
3454
    src_y  -=   linesize;
3455
    src_cb -= uvlinesize;
3456
    src_cr -= uvlinesize;
3457

    
3458
    // There are two lines saved, the line above the the top macroblock of a pair,
3459
    // and the line above the bottom macroblock
3460
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3461
    for(i=1; i<17; i++){
3462
        h->left_border[i]= src_y[15+i*  linesize];
3463
    }
3464

    
3465
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3466
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3467

    
3468
    if(!(s->flags&CODEC_FLAG_GRAY)){
3469
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3470
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3471
        for(i=1; i<9; i++){
3472
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3473
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3474
        }
3475
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3476
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3477
    }
3478
}
3479

    
3480
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){
3481
    MpegEncContext * const s = &h->s;
3482
    int temp8, i;
3483
    uint64_t temp64;
3484
    int deblock_left = (s->mb_x > 0);
3485
    int deblock_top  = (s->mb_y > 0);
3486

    
3487
    src_y  -=   linesize + 1;
3488
    src_cb -= uvlinesize + 1;
3489
    src_cr -= uvlinesize + 1;
3490

    
3491
#define XCHG(a,b,t,xchg)\
3492
t= a;\
3493
if(xchg)\
3494
    a= b;\
3495
b= t;
3496

    
3497
    if(deblock_left){
3498
        for(i = !deblock_top; i<17; i++){
3499
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3500
        }
3501
    }
3502

    
3503
    if(deblock_top){
3504
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3505
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3506
        if(s->mb_x+1 < s->mb_width){
3507
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3508
        }
3509
    }
3510

    
3511
    if(!(s->flags&CODEC_FLAG_GRAY)){
3512
        if(deblock_left){
3513
            for(i = !deblock_top; i<9; i++){
3514
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3515
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3516
            }
3517
        }
3518
        if(deblock_top){
3519
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3520
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3521
        }
3522
    }
3523
}
3524

    
3525
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3526
    MpegEncContext * const s = &h->s;
3527
    int i;
3528

    
3529
    src_y  -= 2 *   linesize;
3530
    src_cb -= 2 * uvlinesize;
3531
    src_cr -= 2 * uvlinesize;
3532

    
3533
    // There are two lines saved, the line above the the top macroblock of a pair,
3534
    // and the line above the bottom macroblock
3535
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3536
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3537
    for(i=2; i<34; i++){
3538
        h->left_border[i]= src_y[15+i*  linesize];
3539
    }
3540

    
3541
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3542
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3543
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3544
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3545

    
3546
    if(!(s->flags&CODEC_FLAG_GRAY)){
3547
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3548
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3549
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3550
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3551
        for(i=2; i<18; i++){
3552
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3553
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3554
        }
3555
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3556
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3557
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3558
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3559
    }
3560
}
3561

    
3562
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){
3563
    MpegEncContext * const s = &h->s;
3564
    int temp8, i;
3565
    uint64_t temp64;
3566
    int deblock_left = (s->mb_x > 0);
3567
    int deblock_top  = (s->mb_y > 1);
3568

    
3569
    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);
3570

    
3571
    src_y  -= 2 *   linesize + 1;
3572
    src_cb -= 2 * uvlinesize + 1;
3573
    src_cr -= 2 * uvlinesize + 1;
3574

    
3575
#define XCHG(a,b,t,xchg)\
3576
t= a;\
3577
if(xchg)\
3578
    a= b;\
3579
b= t;
3580

    
3581
    if(deblock_left){
3582
        for(i = (!deblock_top)<<1; i<34; i++){
3583
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3584
        }
3585
    }
3586

    
3587
    if(deblock_top){
3588
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3589
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3590
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3591
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3592
        if(s->mb_x+1 < s->mb_width){
3593
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3594
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3595
        }
3596
    }
3597

    
3598
    if(!(s->flags&CODEC_FLAG_GRAY)){
3599
        if(deblock_left){
3600
            for(i = (!deblock_top) << 1; i<18; i++){
3601
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3602
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3603
            }
3604
        }
3605
        if(deblock_top){
3606
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3607
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3608
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3609
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3610
        }
3611
    }
3612
}
3613

    
3614
static void hl_decode_mb(H264Context *h){
3615
    MpegEncContext * const s = &h->s;
3616
    const int mb_x= s->mb_x;
3617
    const int mb_y= s->mb_y;
3618
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3619
    const int mb_type= s->current_picture.mb_type[mb_xy];
3620
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3621
    int linesize, uvlinesize /*dct_offset*/;
3622
    int i;
3623
    int *block_offset = &h->block_offset[0];
3624
    const unsigned int bottom = mb_y & 1;
3625
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3626
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3627
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3628

    
3629
    if(!s->decode)
3630
        return;
3631

    
3632
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3633
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3634
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3635

    
3636
    if (MB_FIELD) {
3637
        linesize   = h->mb_linesize   = s->linesize * 2;
3638
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3639
        block_offset = &h->block_offset[24];
3640
        if(mb_y&1){ //FIXME move out of this func?
3641
            dest_y -= s->linesize*15;
3642
            dest_cb-= s->uvlinesize*7;
3643
            dest_cr-= s->uvlinesize*7;
3644
        }
3645
        if(FRAME_MBAFF) {
3646
            int list;
3647
            for(list=0; list<2; list++){
3648
                if(!USES_LIST(mb_type, list))
3649
                    continue;
3650
                if(IS_16X16(mb_type)){
3651
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
3652
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3653
                }else{
3654
                    for(i=0; i<16; i+=4){
3655
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3656
                        int ref = h->ref_cache[list][scan8[i]];
3657
                        if(ref >= 0)
3658
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3659
                    }
3660
                }
3661
            }
3662
        }
3663
    } else {
3664
        linesize   = h->mb_linesize   = s->linesize;
3665
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3666
//        dct_offset = s->linesize * 16;
3667
    }
3668

    
3669
    if(transform_bypass){
3670
        idct_dc_add =
3671
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3672
    }else if(IS_8x8DCT(mb_type)){
3673
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3674
        idct_add = s->dsp.h264_idct8_add;
3675
    }else{
3676
        idct_dc_add = s->dsp.h264_idct_dc_add;
3677
        idct_add = s->dsp.h264_idct_add;
3678
    }
3679

    
3680
    if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3681
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3682
        int mbt_y = mb_y&~1;
3683
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
3684
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3685
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3686
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3687
    }
3688

    
3689
    if (IS_INTRA_PCM(mb_type)) {
3690
        unsigned int x, y;
3691

    
3692
        // The pixels are stored in h->mb array in the same order as levels,
3693
        // copy them in output in the correct order.
3694
        for(i=0; i<16; i++) {
3695
            for (y=0; y<4; y++) {
3696
                for (x=0; x<4; x++) {
3697
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3698
                }
3699
            }
3700
        }
3701
        for(i=16; i<16+4; i++) {
3702
            for (y=0; y<4; y++) {
3703
                for (x=0; x<4; x++) {
3704
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3705
                }
3706
            }
3707
        }
3708
        for(i=20; i<20+4; i++) {
3709
            for (y=0; y<4; y++) {
3710
                for (x=0; x<4; x++) {
3711
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3712
                }
3713
            }
3714
        }
3715
    } else {
3716
        if(IS_INTRA(mb_type)){
3717
            if(h->deblocking_filter && !FRAME_MBAFF)
3718
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3719

    
3720
            if(!(s->flags&CODEC_FLAG_GRAY)){
3721
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3722
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3723
            }
3724

    
3725
            if(IS_INTRA4x4(mb_type)){
3726
                if(!s->encoding){
3727
                    if(IS_8x8DCT(mb_type)){
3728
                        for(i=0; i<16; i+=4){
3729
                            uint8_t * const ptr= dest_y + block_offset[i];
3730
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3731
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3732
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3733
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3734
                            if(nnz){
3735
                                if(nnz == 1 && h->mb[i*16])
3736
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3737
                                else
3738
                                    idct_add(ptr, h->mb + i*16, linesize);
3739
                            }
3740
                        }
3741
                    }else
3742
                    for(i=0; i<16; i++){
3743
                        uint8_t * const ptr= dest_y + block_offset[i];
3744
                        uint8_t *topright;
3745
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3746
                        int nnz, tr;
3747

    
3748
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3749
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3750
                            assert(mb_y || linesize <= block_offset[i]);
3751
                            if(!topright_avail){
3752
                                tr= ptr[3 - linesize]*0x01010101;
3753
                                topright= (uint8_t*) &tr;
3754
                            }else
3755
                                topright= ptr + 4 - linesize;
3756
                        }else
3757
                            topright= NULL;
3758

    
3759
                        h->pred4x4[ dir ](ptr, topright, linesize);
3760
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3761
                        if(nnz){
3762
                            if(s->codec_id == CODEC_ID_H264){
3763
                                if(nnz == 1 && h->mb[i*16])
3764
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3765
                                else
3766
                                    idct_add(ptr, h->mb + i*16, linesize);
3767
                            }else
3768
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3769
                        }
3770
                    }
3771
                }
3772
            }else{
3773
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3774
                if(s->codec_id == CODEC_ID_H264){
3775
                    if(!transform_bypass)
3776
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3777
                }else
3778
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3779
            }
3780
            if(h->deblocking_filter && !FRAME_MBAFF)
3781
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3782
        }else if(s->codec_id == CODEC_ID_H264){
3783
            hl_motion(h, dest_y, dest_cb, dest_cr,
3784
                      s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3785
                      s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3786
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3787
        }
3788

    
3789

    
3790
        if(!IS_INTRA4x4(mb_type)){
3791
            if(s->codec_id == CODEC_ID_H264){
3792
                if(IS_INTRA16x16(mb_type)){
3793
                    for(i=0; i<16; i++){
3794
                        if(h->non_zero_count_cache[ scan8[i] ])
3795
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3796
                        else if(h->mb[i*16])
3797
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3798
                    }
3799
                }else{
3800
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3801
                    for(i=0; i<16; i+=di){
3802
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3803
                        if(nnz){
3804
                            if(nnz==1 && h->mb[i*16])
3805
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3806
                            else
3807
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3808
                        }
3809
                    }
3810
                }
3811
            }else{
3812
                for(i=0; i<16; i++){
3813
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3814
                        uint8_t * const ptr= dest_y + block_offset[i];
3815
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3816
                    }
3817
                }
3818
            }
3819
        }
3820

    
3821
        if(!(s->flags&CODEC_FLAG_GRAY)){
3822
            uint8_t *dest[2] = {dest_cb, dest_cr};
3823
            if(transform_bypass){
3824
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3825
            }else{
3826
                idct_add = s->dsp.h264_idct_add;
3827
                idct_dc_add = s->dsp.h264_idct_dc_add;
3828
                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp][0]);
3829
                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp][0]);
3830
            }
3831
            if(s->codec_id == CODEC_ID_H264){
3832
                for(i=16; i<16+8; i++){
3833
                    if(h->non_zero_count_cache[ scan8[i] ])
3834
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3835
                    else if(h->mb[i*16])
3836
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3837
                }
3838
            }else{
3839
                for(i=16; i<16+8; i++){
3840
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3841
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3842
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3843
                    }
3844
                }
3845
            }
3846
        }
3847
    }
3848
    if(h->deblocking_filter) {
3849
        if (FRAME_MBAFF) {
3850
            //FIXME try deblocking one mb at a time?
3851
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3852
            const int mb_y = s->mb_y - 1;
3853
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3854
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3855
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3856
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3857
            if (!bottom) return;
3858
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3859
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3860
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3861

    
3862
            if(IS_INTRA(mb_type_top | mb_type_bottom))
3863
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3864

    
3865
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3866
            // deblock a pair
3867
            // top
3868
            s->mb_y--;
3869
            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);
3870
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3871
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
3872
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3873
            // bottom
3874
            s->mb_y++;
3875
            tprintf("call mbaff filter_mb\n");
3876
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3877
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3878
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3879
        } else {
3880
            tprintf("call filter_mb\n");
3881
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3882
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3883
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3884
        }
3885
    }
3886
}
3887

    
3888
/**
3889
 * fills the default_ref_list.
3890
 */
3891
static int fill_default_ref_list(H264Context *h){
3892
    MpegEncContext * const s = &h->s;
3893
    int i;
3894
    int smallest_poc_greater_than_current = -1;
3895
    Picture sorted_short_ref[32];
3896

    
3897
    if(h->slice_type==B_TYPE){
3898
        int out_i;
3899
        int limit= INT_MIN;
3900

    
3901
        /* sort frame according to poc in B slice */
3902
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3903
            int best_i=INT_MIN;
3904
            int best_poc=INT_MAX;
3905

    
3906
            for(i=0; i<h->short_ref_count; i++){
3907
                const int poc= h->short_ref[i]->poc;
3908
                if(poc > limit && poc < best_poc){
3909
                    best_poc= poc;
3910
                    best_i= i;
3911
                }
3912
            }
3913

    
3914
            assert(best_i != INT_MIN);
3915

    
3916
            limit= best_poc;
3917
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3918
            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);
3919
            if (-1 == smallest_poc_greater_than_current) {
3920
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3921
                    smallest_poc_greater_than_current = out_i;
3922
                }
3923
            }
3924
        }
3925
    }
3926

    
3927
    if(s->picture_structure == PICT_FRAME){
3928
        if(h->slice_type==B_TYPE){
3929
            int list;
3930
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3931

    
3932
            // find the largest poc
3933
            for(list=0; list<2; list++){
3934
                int index = 0;
3935
                int j= -99;
3936
                int step= list ? -1 : 1;
3937

    
3938
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3939
                    while(j<0 || j>= h->short_ref_count){
3940
                        if(j != -99 && step == (list ? -1 : 1))
3941
                            return -1;
3942
                        step = -step;
3943
                        j= smallest_poc_greater_than_current + (step>>1);
3944
                    }
3945
                    if(sorted_short_ref[j].reference != 3) continue;
3946
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3947
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3948
                }
3949

    
3950
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3951
                    if(h->long_ref[i] == NULL) continue;
3952
                    if(h->long_ref[i]->reference != 3) continue;
3953

    
3954
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3955
                    h->default_ref_list[ list ][index++].pic_id= i;;
3956
                }
3957

    
3958
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3959
                    // swap the two first elements of L1 when
3960
                    // L0 and L1 are identical
3961
                    Picture temp= h->default_ref_list[1][0];
3962
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3963
                    h->default_ref_list[1][1] = temp;
3964
                }
3965

    
3966
                if(index < h->ref_count[ list ])
3967
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3968
            }
3969
        }else{
3970
            int index=0;
3971
            for(i=0; i<h->short_ref_count; i++){
3972
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3973
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3974
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3975
            }
3976
            for(i = 0; i < 16; i++){
3977
                if(h->long_ref[i] == NULL) continue;
3978
                if(h->long_ref[i]->reference != 3) continue;
3979
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3980
                h->default_ref_list[0][index++].pic_id= i;;
3981
            }
3982
            if(index < h->ref_count[0])
3983
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3984
        }
3985
    }else{ //FIELD
3986
        if(h->slice_type==B_TYPE){
3987
        }else{
3988
            //FIXME second field balh
3989
        }
3990
    }
3991
#ifdef TRACE
3992
    for (i=0; i<h->ref_count[0]; i++) {
3993
        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]);
3994
    }
3995
    if(h->slice_type==B_TYPE){
3996
        for (i=0; i<h->ref_count[1]; i++) {
3997
            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]);
3998
        }
3999
    }
4000
#endif
4001
    return 0;
4002
}
4003

    
4004
static void print_short_term(H264Context *h);
4005
static void print_long_term(H264Context *h);
4006

    
4007
static int decode_ref_pic_list_reordering(H264Context *h){
4008
    MpegEncContext * const s = &h->s;
4009
    int list, index;
4010

    
4011
    print_short_term(h);
4012
    print_long_term(h);
4013
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
4014

    
4015
    for(list=0; list<2; list++){
4016
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
4017

    
4018
        if(get_bits1(&s->gb)){
4019
            int pred= h->curr_pic_num;
4020

    
4021
            for(index=0; ; index++){
4022
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
4023
                int pic_id;
4024
                int i;
4025
                Picture *ref = NULL;
4026

    
4027
                if(reordering_of_pic_nums_idc==3)
4028
                    break;
4029

    
4030
                if(index >= h->ref_count[list]){
4031
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
4032
                    return -1;
4033
                }
4034

    
4035
                if(reordering_of_pic_nums_idc<3){
4036
                    if(reordering_of_pic_nums_idc<2){
4037
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
4038

    
4039
                        if(abs_diff_pic_num >= h->max_pic_num){
4040
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
4041
                            return -1;
4042
                        }
4043

    
4044
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
4045
                        else                                pred+= abs_diff_pic_num;
4046
                        pred &= h->max_pic_num - 1;
4047

    
4048
                        for(i= h->short_ref_count-1; i>=0; i--){
4049
                            ref = h->short_ref[i];
4050
                            assert(ref->reference == 3);
4051
                            assert(!ref->long_ref);
4052
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
4053
                                break;
4054
                        }
4055
                        if(i>=0)
4056
                            ref->pic_id= ref->frame_num;
4057
                    }else{
4058
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
4059
                        ref = h->long_ref[pic_id];
4060
                        ref->pic_id= pic_id;
4061
                        assert(ref->reference == 3);
4062
                        assert(ref->long_ref);
4063
                        i=0;
4064
                    }
4065

    
4066
                    if (i < 0) {
4067
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
4068
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
4069
                    } else {
4070
                        for(i=index; i+1<h->ref_count[list]; i++){
4071
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
4072
                                break;
4073
                        }
4074
                        for(; i > index; i--){
4075
                            h->ref_list[list][i]= h->ref_list[list][i-1];
4076
                        }
4077
                        h->ref_list[list][index]= *ref;
4078
                    }
4079
                }else{
4080
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
4081
                    return -1;
4082
                }
4083
            }
4084
        }
4085

    
4086
        if(h->slice_type!=B_TYPE) break;
4087
    }
4088
    for(list=0; list<2; list++){
4089
        for(index= 0; index < h->ref_count[list]; index++){
4090
            if(!h->ref_list[list][index].data[0])
4091
                h->ref_list[list][index]= s->current_picture;
4092
        }
4093
        if(h->slice_type!=B_TYPE) break;
4094
    }
4095

    
4096
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
4097
        direct_dist_scale_factor(h);
4098
    direct_ref_list_init(h);
4099
    return 0;
4100
}
4101

    
4102
static int fill_mbaff_ref_list(H264Context *h){
4103
    int list, i, j;
4104
    for(list=0; list<2; list++){
4105
        for(i=0; i<h->ref_count[list]; i++){
4106
            Picture *frame = &h->ref_list[list][i];
4107
            Picture *field = &h->ref_list[list][16+2*i];
4108
            field[0] = *frame;
4109
            for(j=0; j<3; j++)
4110
                field[0].linesize[j] <<= 1;
4111
            field[1] = field[0];
4112
            for(j=0; j<3; j++)
4113
                field[1].data[j] += frame->linesize[j];
4114

    
4115
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
4116
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
4117
            for(j=0; j<2; j++){
4118
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
4119
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
4120
            }
4121
        }
4122
    }
4123
    for(j=0; j<h->ref_count[1]; j++){
4124
        for(i=0; i<h->ref_count[0]; i++)
4125
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
4126
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
4127
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
4128
    }
4129
}
4130

    
4131
static int pred_weight_table(H264Context *h){
4132
    MpegEncContext * const s = &h->s;
4133
    int list, i;
4134
    int luma_def, chroma_def;
4135

    
4136
    h->use_weight= 0;
4137
    h->use_weight_chroma= 0;
4138
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
4139
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
4140
    luma_def = 1<<h->luma_log2_weight_denom;
4141
    chroma_def = 1<<h->chroma_log2_weight_denom;
4142

    
4143
    for(list=0; list<2; list++){
4144
        for(i=0; i<h->ref_count[list]; i++){
4145
            int luma_weight_flag, chroma_weight_flag;
4146

    
4147
            luma_weight_flag= get_bits1(&s->gb);
4148
            if(luma_weight_flag){
4149
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
4150
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
4151
                if(   h->luma_weight[list][i] != luma_def
4152
                   || h->luma_offset[list][i] != 0)
4153
                    h->use_weight= 1;
4154
            }else{
4155
                h->luma_weight[list][i]= luma_def;
4156
                h->luma_offset[list][i]= 0;
4157
            }
4158

    
4159
            chroma_weight_flag= get_bits1(&s->gb);
4160
            if(chroma_weight_flag){
4161
                int j;
4162
                for(j=0; j<2; j++){
4163
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
4164
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
4165
                    if(   h->chroma_weight[list][i][j] != chroma_def
4166
                       || h->chroma_offset[list][i][j] != 0)
4167
                        h->use_weight_chroma= 1;
4168
                }
4169
            }else{
4170
                int j;
4171
                for(j=0; j<2; j++){
4172
                    h->chroma_weight[list][i][j]= chroma_def;
4173
                    h->chroma_offset[list][i][j]= 0;
4174
                }
4175
            }
4176
        }
4177
        if(h->slice_type != B_TYPE) break;
4178
    }
4179
    h->use_weight= h->use_weight || h->use_weight_chroma;
4180
    return 0;
4181
}
4182

    
4183
static void implicit_weight_table(H264Context *h){
4184
    MpegEncContext * const s = &h->s;
4185
    int ref0, ref1;
4186
    int cur_poc = s->current_picture_ptr->poc;
4187

    
4188
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
4189
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
4190
        h->use_weight= 0;
4191
        h->use_weight_chroma= 0;
4192
        return;
4193
    }
4194

    
4195
    h->use_weight= 2;
4196
    h->use_weight_chroma= 2;
4197
    h->luma_log2_weight_denom= 5;
4198
    h->chroma_log2_weight_denom= 5;
4199

    
4200
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
4201
        int poc0 = h->ref_list[0][ref0].poc;
4202
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
4203
            int poc1 = h->ref_list[1][ref1].poc;
4204
            int td = clip(poc1 - poc0, -128, 127);
4205
            if(td){
4206
                int tb = clip(cur_poc - poc0, -128, 127);
4207
                int tx = (16384 + (ABS(td) >> 1)) / td;
4208
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
4209
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
4210
                    h->implicit_weight[ref0][ref1] = 32;
4211
                else
4212
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
4213
            }else
4214
                h->implicit_weight[ref0][ref1] = 32;
4215
        }
4216
    }
4217
}
4218

    
4219
static inline void unreference_pic(H264Context *h, Picture *pic){
4220
    int i;
4221
    pic->reference=0;
4222
    if(pic == h->delayed_output_pic)
4223
        pic->reference=1;
4224
    else{
4225
        for(i = 0; h->delayed_pic[i]; i++)
4226
            if(pic == h->delayed_pic[i]){
4227
                pic->reference=1;
4228
                break;
4229
            }
4230
    }
4231
}
4232

    
4233
/**
4234
 * instantaneous decoder refresh.
4235
 */
4236
static void idr(H264Context *h){
4237
    int i;
4238

    
4239
    for(i=0; i<16; i++){
4240
        if (h->long_ref[i] != NULL) {
4241
            unreference_pic(h, h->long_ref[i]);
4242
            h->long_ref[i]= NULL;
4243
        }
4244
    }
4245
    h->long_ref_count=0;
4246

    
4247
    for(i=0; i<h->short_ref_count; i++){
4248
        unreference_pic(h, h->short_ref[i]);
4249
        h->short_ref[i]= NULL;
4250
    }
4251
    h->short_ref_count=0;
4252
}
4253

    
4254
/* forget old pics after a seek */
4255
static void flush_dpb(AVCodecContext *avctx){
4256
    H264Context *h= avctx->priv_data;
4257
    int i;
4258
    for(i=0; i<16; i++) {
4259
        if(h->delayed_pic[i])
4260
            h->delayed_pic[i]->reference= 0;
4261
        h->delayed_pic[i]= NULL;
4262
    }
4263
    if(h->delayed_output_pic)
4264
        h->delayed_output_pic->reference= 0;
4265
    h->delayed_output_pic= NULL;
4266
    idr(h);
4267
    if(h->s.current_picture_ptr)
4268
        h->s.current_picture_ptr->reference= 0;
4269
}
4270

    
4271
/**
4272
 *
4273
 * @return the removed picture or NULL if an error occurs
4274
 */
4275
static Picture * remove_short(H264Context *h, int frame_num){
4276
    MpegEncContext * const s = &h->s;
4277
    int i;
4278

    
4279
    if(s->avctx->debug&FF_DEBUG_MMCO)
4280
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
4281

    
4282