Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 550e66cb

History | View | Annotate | Download (323 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 file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 *
21
 */
22

    
23
/**
24
 * @file h264.c
25
 * H.264 / AVC / MPEG4 part10 codec.
26
 * @author Michael Niedermayer <michaelni@gmx.at>
27
 */
28

    
29
#include "common.h"
30
#include "dsputil.h"
31
#include "avcodec.h"
32
#include "mpegvideo.h"
33
#include "h264data.h"
34
#include "golomb.h"
35

    
36
#include "cabac.h"
37

    
38
//#undef NDEBUG
39
#include <assert.h>
40

    
41
#define interlaced_dct interlaced_dct_is_a_bad_name
42
#define mb_intra mb_intra_isnt_initalized_see_mb_type
43

    
44
#define LUMA_DC_BLOCK_INDEX   25
45
#define CHROMA_DC_BLOCK_INDEX 26
46

    
47
#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
48
#define COEFF_TOKEN_VLC_BITS           8
49
#define TOTAL_ZEROS_VLC_BITS           9
50
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
51
#define RUN_VLC_BITS                   3
52
#define RUN7_VLC_BITS                  6
53

    
54
#define MAX_SPS_COUNT 32
55
#define MAX_PPS_COUNT 256
56

    
57
#define MAX_MMCO_COUNT 66
58

    
59
/* Compiling in interlaced support reduces the speed
60
 * of progressive decoding by about 2%. */
61
#define ALLOW_INTERLACE
62

    
63
#ifdef ALLOW_INTERLACE
64
#define MB_MBAFF h->mb_mbaff
65
#define MB_FIELD h->mb_field_decoding_flag
66
#define FRAME_MBAFF h->mb_aff_frame
67
#else
68
#define MB_MBAFF 0
69
#define MB_FIELD 0
70
#define FRAME_MBAFF 0
71
#undef  IS_INTERLACED
72
#define IS_INTERLACED(mb_type) 0
73
#endif
74

    
75
/**
76
 * Sequence parameter set
77
 */
78
typedef struct SPS{
79

    
80
    int profile_idc;
81
    int level_idc;
82
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
83
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
84
    int poc_type;                      ///< pic_order_cnt_type
85
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
86
    int delta_pic_order_always_zero_flag;
87
    int offset_for_non_ref_pic;
88
    int offset_for_top_to_bottom_field;
89
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
90
    int ref_frame_count;               ///< num_ref_frames
91
    int gaps_in_frame_num_allowed_flag;
92
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
93
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
94
    int frame_mbs_only_flag;
95
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
96
    int direct_8x8_inference_flag;
97
    int crop;                   ///< frame_cropping_flag
98
    int crop_left;              ///< frame_cropping_rect_left_offset
99
    int crop_right;             ///< frame_cropping_rect_right_offset
100
    int crop_top;               ///< frame_cropping_rect_top_offset
101
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
102
    int vui_parameters_present_flag;
103
    AVRational sar;
104
    int timing_info_present_flag;
105
    uint32_t num_units_in_tick;
106
    uint32_t time_scale;
107
    int fixed_frame_rate_flag;
108
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
109
    int bitstream_restriction_flag;
110
    int num_reorder_frames;
111
    int scaling_matrix_present;
112
    uint8_t scaling_matrix4[6][16];
113
    uint8_t scaling_matrix8[2][64];
114
}SPS;
115

    
116
/**
117
 * Picture parameter set
118
 */
119
typedef struct PPS{
120
    int sps_id;
121
    int cabac;                  ///< entropy_coding_mode_flag
122
    int pic_order_present;      ///< pic_order_present_flag
123
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
124
    int mb_slice_group_map_type;
125
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
126
    int weighted_pred;          ///< weighted_pred_flag
127
    int weighted_bipred_idc;
128
    int init_qp;                ///< pic_init_qp_minus26 + 26
129
    int init_qs;                ///< pic_init_qs_minus26 + 26
130
    int chroma_qp_index_offset;
131
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
132
    int constrained_intra_pred; ///< constrained_intra_pred_flag
133
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
134
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
135
    uint8_t scaling_matrix4[6][16];
136
    uint8_t scaling_matrix8[2][64];
137
}PPS;
138

    
139
/**
140
 * Memory management control operation opcode.
141
 */
142
typedef enum MMCOOpcode{
143
    MMCO_END=0,
144
    MMCO_SHORT2UNUSED,
145
    MMCO_LONG2UNUSED,
146
    MMCO_SHORT2LONG,
147
    MMCO_SET_MAX_LONG,
148
    MMCO_RESET,
149
    MMCO_LONG,
150
} MMCOOpcode;
151

    
152
/**
153
 * Memory management control operation.
154
 */
155
typedef struct MMCO{
156
    MMCOOpcode opcode;
157
    int short_frame_num;
158
    int long_index;
159
} MMCO;
160

    
161
/**
162
 * H264Context
163
 */
164
typedef struct H264Context{
165
    MpegEncContext s;
166
    int nal_ref_idc;
167
    int nal_unit_type;
168
    uint8_t *rbsp_buffer;
169
    unsigned int rbsp_buffer_size;
170

    
171
    /**
172
      * Used to parse AVC variant of h264
173
      */
174
    int is_avc; ///< this flag is != 0 if codec is avc1
175
    int got_avcC; ///< flag used to parse avcC data only once
176
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
177

    
178
    int chroma_qp; //QPc
179

    
180
    int prev_mb_skipped;
181
    int next_mb_skipped;
182

    
183
    //prediction stuff
184
    int chroma_pred_mode;
185
    int intra16x16_pred_mode;
186

    
187
    int top_mb_xy;
188
    int left_mb_xy[2];
189

    
190
    int8_t intra4x4_pred_mode_cache[5*8];
191
    int8_t (*intra4x4_pred_mode)[8];
192
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
193
    void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
194
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
195
    void (*pred16x16[4+3])(uint8_t *src, int stride);
196
    unsigned int topleft_samples_available;
197
    unsigned int top_samples_available;
198
    unsigned int topright_samples_available;
199
    unsigned int left_samples_available;
200
    uint8_t (*top_borders[2])[16+2*8];
201
    uint8_t left_border[2*(17+2*9)];
202

    
203
    /**
204
     * non zero coeff count cache.
205
     * is 64 if not available.
206
     */
207
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
208
    uint8_t (*non_zero_count)[16];
209

    
210
    /**
211
     * Motion vector cache.
212
     */
213
    DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
214
    DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
215
#define LIST_NOT_USED -1 //FIXME rename?
216
#define PART_NOT_AVAILABLE -2
217

    
218
    /**
219
     * is 1 if the specific list MV&references are set to 0,0,-2.
220
     */
221
    int mv_cache_clean[2];
222

    
223
    /**
224
     * number of neighbors (top and/or left) that used 8x8 dct
225
     */
226
    int neighbor_transform_size;
227

    
228
    /**
229
     * block_offset[ 0..23] for frame macroblocks
230
     * block_offset[24..47] for field macroblocks
231
     */
232
    int block_offset[2*(16+8)];
233

    
234
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
235
    uint32_t *mb2b8_xy;
236
    int b_stride; //FIXME use s->b4_stride
237
    int b8_stride;
238

    
239
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
240
    int mb_uvlinesize;
241

    
242
    int emu_edge_width;
243
    int emu_edge_height;
244

    
245
    int halfpel_flag;
246
    int thirdpel_flag;
247

    
248
    int unknown_svq3_flag;
249
    int next_slice_index;
250

    
251
    SPS sps_buffer[MAX_SPS_COUNT];
252
    SPS sps; ///< current sps
253

    
254
    PPS pps_buffer[MAX_PPS_COUNT];
255
    /**
256
     * current pps
257
     */
258
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
259

    
260
    uint32_t dequant4_buffer[6][52][16];
261
    uint32_t dequant8_buffer[2][52][64];
262
    uint32_t (*dequant4_coeff[6])[16];
263
    uint32_t (*dequant8_coeff[2])[64];
264
    int dequant_coeff_pps;     ///< reinit tables when pps changes
265

    
266
    int slice_num;
267
    uint8_t *slice_table_base;
268
    uint8_t *slice_table;      ///< slice_table_base + 2*mb_stride + 1
269
    int slice_type;
270
    int slice_type_fixed;
271

    
272
    //interlacing specific flags
273
    int mb_aff_frame;
274
    int mb_field_decoding_flag;
275
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
276

    
277
    int sub_mb_type[4];
278

    
279
    //POC stuff
280
    int poc_lsb;
281
    int poc_msb;
282
    int delta_poc_bottom;
283
    int delta_poc[2];
284
    int frame_num;
285
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
286
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
287
    int frame_num_offset;         ///< for POC type 2
288
    int prev_frame_num_offset;    ///< for POC type 2
289
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
290

    
291
    /**
292
     * frame_num for frames or 2*frame_num for field pics.
293
     */
294
    int curr_pic_num;
295

    
296
    /**
297
     * max_frame_num or 2*max_frame_num for field pics.
298
     */
299
    int max_pic_num;
300

    
301
    //Weighted pred stuff
302
    int use_weight;
303
    int use_weight_chroma;
304
    int luma_log2_weight_denom;
305
    int chroma_log2_weight_denom;
306
    int luma_weight[2][48];
307
    int luma_offset[2][48];
308
    int chroma_weight[2][48][2];
309
    int chroma_offset[2][48][2];
310
    int implicit_weight[48][48];
311

    
312
    //deblock
313
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
314
    int slice_alpha_c0_offset;
315
    int slice_beta_offset;
316

    
317
    int redundant_pic_count;
318

    
319
    int direct_spatial_mv_pred;
320
    int dist_scale_factor[16];
321
    int dist_scale_factor_field[32];
322
    int map_col_to_list0[2][16];
323
    int map_col_to_list0_field[2][32];
324

    
325
    /**
326
     * num_ref_idx_l0/1_active_minus1 + 1
327
     */
328
    int ref_count[2];            ///< counts frames or fields, depending on current mb mode
329
    Picture *short_ref[32];
330
    Picture *long_ref[32];
331
    Picture default_ref_list[2][32];
332
    Picture ref_list[2][48];     ///< 0..15: frame refs, 16..47: mbaff field refs
333
    Picture *delayed_pic[16]; //FIXME size?
334
    Picture *delayed_output_pic;
335

    
336
    /**
337
     * memory management control operations buffer.
338
     */
339
    MMCO mmco[MAX_MMCO_COUNT];
340
    int mmco_index;
341

    
342
    int long_ref_count;  ///< number of actual long term references
343
    int short_ref_count; ///< number of actual short term references
344

    
345
    //data partitioning
346
    GetBitContext intra_gb;
347
    GetBitContext inter_gb;
348
    GetBitContext *intra_gb_ptr;
349
    GetBitContext *inter_gb_ptr;
350

    
351
    DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);
352

    
353
    /**
354
     * Cabac
355
     */
356
    CABACContext cabac;
357
    uint8_t      cabac_state[460];
358
    int          cabac_init_idc;
359

    
360
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
361
    uint16_t     *cbp_table;
362
    int cbp;
363
    int top_cbp;
364
    int left_cbp;
365
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
366
    uint8_t     *chroma_pred_mode_table;
367
    int         last_qscale_diff;
368
    int16_t     (*mvd_table[2])[2];
369
    DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
370
    uint8_t     *direct_table;
371
    uint8_t     direct_cache[5*8];
372

    
373
    uint8_t zigzag_scan[16];
374
    uint8_t zigzag_scan8x8[64];
375
    uint8_t zigzag_scan8x8_cavlc[64];
376
    uint8_t field_scan[16];
377
    uint8_t field_scan8x8[64];
378
    uint8_t field_scan8x8_cavlc[64];
379
    const uint8_t *zigzag_scan_q0;
380
    const uint8_t *zigzag_scan8x8_q0;
381
    const uint8_t *zigzag_scan8x8_cavlc_q0;
382
    const uint8_t *field_scan_q0;
383
    const uint8_t *field_scan8x8_q0;
384
    const uint8_t *field_scan8x8_cavlc_q0;
385

    
386
    int x264_build;
387
}H264Context;
388

    
389
static VLC coeff_token_vlc[4];
390
static VLC chroma_dc_coeff_token_vlc;
391

    
392
static VLC total_zeros_vlc[15];
393
static VLC chroma_dc_total_zeros_vlc[3];
394

    
395
static VLC run_vlc[6];
396
static VLC run7_vlc;
397

    
398
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
399
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
400
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);
401
static void filter_mb_fast( 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);
402

    
403
static always_inline uint32_t pack16to32(int a, int b){
404
#ifdef WORDS_BIGENDIAN
405
   return (b&0xFFFF) + (a<<16);
406
#else
407
   return (a&0xFFFF) + (b<<16);
408
#endif
409
}
410

    
411
/**
412
 * fill a rectangle.
413
 * @param h height of the rectangle, should be a constant
414
 * @param w width of the rectangle, should be a constant
415
 * @param size the size of val (1 or 4), should be a constant
416
 */
417
static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
418
    uint8_t *p= (uint8_t*)vp;
419
    assert(size==1 || size==4);
420
    assert(w<=4);
421

    
422
    w      *= size;
423
    stride *= size;
424

    
425
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
426
    assert((stride&(w-1))==0);
427
    if(w==2){
428
        const uint16_t v= size==4 ? val : val*0x0101;
429
        *(uint16_t*)(p + 0*stride)= v;
430
        if(h==1) return;
431
        *(uint16_t*)(p + 1*stride)= v;
432
        if(h==2) return;
433
        *(uint16_t*)(p + 2*stride)=
434
        *(uint16_t*)(p + 3*stride)= v;
435
    }else if(w==4){
436
        const uint32_t v= size==4 ? val : val*0x01010101;
437
        *(uint32_t*)(p + 0*stride)= v;
438
        if(h==1) return;
439
        *(uint32_t*)(p + 1*stride)= v;
440
        if(h==2) return;
441
        *(uint32_t*)(p + 2*stride)=
442
        *(uint32_t*)(p + 3*stride)= v;
443
    }else if(w==8){
444
    //gcc can't optimize 64bit math on x86_32
445
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
446
        const uint64_t v= val*0x0100000001ULL;
447
        *(uint64_t*)(p + 0*stride)= v;
448
        if(h==1) return;
449
        *(uint64_t*)(p + 1*stride)= v;
450
        if(h==2) return;
451
        *(uint64_t*)(p + 2*stride)=
452
        *(uint64_t*)(p + 3*stride)= v;
453
    }else if(w==16){
454
        const uint64_t v= val*0x0100000001ULL;
455
        *(uint64_t*)(p + 0+0*stride)=
456
        *(uint64_t*)(p + 8+0*stride)=
457
        *(uint64_t*)(p + 0+1*stride)=
458
        *(uint64_t*)(p + 8+1*stride)= v;
459
        if(h==2) return;
460
        *(uint64_t*)(p + 0+2*stride)=
461
        *(uint64_t*)(p + 8+2*stride)=
462
        *(uint64_t*)(p + 0+3*stride)=
463
        *(uint64_t*)(p + 8+3*stride)= v;
464
#else
465
        *(uint32_t*)(p + 0+0*stride)=
466
        *(uint32_t*)(p + 4+0*stride)= val;
467
        if(h==1) return;
468
        *(uint32_t*)(p + 0+1*stride)=
469
        *(uint32_t*)(p + 4+1*stride)= val;
470
        if(h==2) return;
471
        *(uint32_t*)(p + 0+2*stride)=
472
        *(uint32_t*)(p + 4+2*stride)=
473
        *(uint32_t*)(p + 0+3*stride)=
474
        *(uint32_t*)(p + 4+3*stride)= val;
475
    }else if(w==16){
476
        *(uint32_t*)(p + 0+0*stride)=
477
        *(uint32_t*)(p + 4+0*stride)=
478
        *(uint32_t*)(p + 8+0*stride)=
479
        *(uint32_t*)(p +12+0*stride)=
480
        *(uint32_t*)(p + 0+1*stride)=
481
        *(uint32_t*)(p + 4+1*stride)=
482
        *(uint32_t*)(p + 8+1*stride)=
483
        *(uint32_t*)(p +12+1*stride)= val;
484
        if(h==2) return;
485
        *(uint32_t*)(p + 0+2*stride)=
486
        *(uint32_t*)(p + 4+2*stride)=
487
        *(uint32_t*)(p + 8+2*stride)=
488
        *(uint32_t*)(p +12+2*stride)=
489
        *(uint32_t*)(p + 0+3*stride)=
490
        *(uint32_t*)(p + 4+3*stride)=
491
        *(uint32_t*)(p + 8+3*stride)=
492
        *(uint32_t*)(p +12+3*stride)= val;
493
#endif
494
    }else
495
        assert(0);
496
    assert(h==4);
497
}
498

    
499
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
500
    MpegEncContext * const s = &h->s;
501
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
502
    int topleft_xy, top_xy, topright_xy, left_xy[2];
503
    int topleft_type, top_type, topright_type, left_type[2];
504
    int left_block[8];
505
    int i;
506

    
507
    //FIXME deblocking could skip the intra and nnz parts.
508
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
509
        return;
510

    
511
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
512

    
513
    top_xy     = mb_xy  - s->mb_stride;
514
    topleft_xy = top_xy - 1;
515
    topright_xy= top_xy + 1;
516
    left_xy[1] = left_xy[0] = mb_xy-1;
517
    left_block[0]= 0;
518
    left_block[1]= 1;
519
    left_block[2]= 2;
520
    left_block[3]= 3;
521
    left_block[4]= 7;
522
    left_block[5]= 10;
523
    left_block[6]= 8;
524
    left_block[7]= 11;
525
    if(FRAME_MBAFF){
526
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
527
        const int top_pair_xy      = pair_xy     - s->mb_stride;
528
        const int topleft_pair_xy  = top_pair_xy - 1;
529
        const int topright_pair_xy = top_pair_xy + 1;
530
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
531
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
532
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
533
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
534
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
535
        const int bottom = (s->mb_y & 1);
536
        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);
537
        if (bottom
538
                ? !curr_mb_frame_flag // bottom macroblock
539
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
540
                ) {
541
            top_xy -= s->mb_stride;
542
        }
543
        if (bottom
544
                ? !curr_mb_frame_flag // bottom macroblock
545
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
546
                ) {
547
            topleft_xy -= s->mb_stride;
548
        }
549
        if (bottom
550
                ? !curr_mb_frame_flag // bottom macroblock
551
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
552
                ) {
553
            topright_xy -= s->mb_stride;
554
        }
555
        if (left_mb_frame_flag != curr_mb_frame_flag) {
556
            left_xy[1] = left_xy[0] = pair_xy - 1;
557
            if (curr_mb_frame_flag) {
558
                if (bottom) {
559
                    left_block[0]= 2;
560
                    left_block[1]= 2;
561
                    left_block[2]= 3;
562
                    left_block[3]= 3;
563
                    left_block[4]= 8;
564
                    left_block[5]= 11;
565
                    left_block[6]= 8;
566
                    left_block[7]= 11;
567
                } else {
568
                    left_block[0]= 0;
569
                    left_block[1]= 0;
570
                    left_block[2]= 1;
571
                    left_block[3]= 1;
572
                    left_block[4]= 7;
573
                    left_block[5]= 10;
574
                    left_block[6]= 7;
575
                    left_block[7]= 10;
576
                }
577
            } else {
578
                left_xy[1] += s->mb_stride;
579
                //left_block[0]= 0;
580
                left_block[1]= 2;
581
                left_block[2]= 0;
582
                left_block[3]= 2;
583
                //left_block[4]= 7;
584
                left_block[5]= 10;
585
                left_block[6]= 7;
586
                left_block[7]= 10;
587
            }
588
        }
589
    }
590

    
591
    h->top_mb_xy = top_xy;
592
    h->left_mb_xy[0] = left_xy[0];
593
    h->left_mb_xy[1] = left_xy[1];
594
    if(for_deblock){
595
        topleft_type = 0;
596
        topright_type = 0;
597
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
598
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
599
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
600

    
601
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
602
            int list;
603
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
604
            for(i=0; i<16; i++)
605
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
606
            for(list=0; list<1+(h->slice_type==B_TYPE); list++){
607
                if(USES_LIST(mb_type,list)){
608
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
609
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
610
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
611
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
612
                        dst[0] = src[0];
613
                        dst[1] = src[1];
614
                        dst[2] = src[2];
615
                        dst[3] = src[3];
616
                    }
617
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
618
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
619
                    ref += h->b8_stride;
620
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
621
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
622
                }else{
623
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
624
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
625
                }
626
            }
627
        }
628
    }else{
629
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
630
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
631
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
632
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
633
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
634
    }
635

    
636
    if(IS_INTRA(mb_type)){
637
        h->topleft_samples_available=
638
        h->top_samples_available=
639
        h->left_samples_available= 0xFFFF;
640
        h->topright_samples_available= 0xEEEA;
641

    
642
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
643
            h->topleft_samples_available= 0xB3FF;
644
            h->top_samples_available= 0x33FF;
645
            h->topright_samples_available= 0x26EA;
646
        }
647
        for(i=0; i<2; i++){
648
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
649
                h->topleft_samples_available&= 0xDF5F;
650
                h->left_samples_available&= 0x5F5F;
651
            }
652
        }
653

    
654
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
655
            h->topleft_samples_available&= 0x7FFF;
656

    
657
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
658
            h->topright_samples_available&= 0xFBFF;
659

    
660
        if(IS_INTRA4x4(mb_type)){
661
            if(IS_INTRA4x4(top_type)){
662
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
663
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
664
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
665
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
666
            }else{
667
                int pred;
668
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
669
                    pred= -1;
670
                else{
671
                    pred= 2;
672
                }
673
                h->intra4x4_pred_mode_cache[4+8*0]=
674
                h->intra4x4_pred_mode_cache[5+8*0]=
675
                h->intra4x4_pred_mode_cache[6+8*0]=
676
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
677
            }
678
            for(i=0; i<2; i++){
679
                if(IS_INTRA4x4(left_type[i])){
680
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
681
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
682
                }else{
683
                    int pred;
684
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
685
                        pred= -1;
686
                    else{
687
                        pred= 2;
688
                    }
689
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
690
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
691
                }
692
            }
693
        }
694
    }
695

    
696

    
697
/*
698
0 . T T. T T T T
699
1 L . .L . . . .
700
2 L . .L . . . .
701
3 . T TL . . . .
702
4 L . .L . . . .
703
5 L . .. . . . .
704
*/
705
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
706
    if(top_type){
707
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
708
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
709
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
710
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
711

    
712
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
713
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
714

    
715
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
716
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
717

    
718
    }else{
719
        h->non_zero_count_cache[4+8*0]=
720
        h->non_zero_count_cache[5+8*0]=
721
        h->non_zero_count_cache[6+8*0]=
722
        h->non_zero_count_cache[7+8*0]=
723

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

    
727
        h->non_zero_count_cache[1+8*3]=
728
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
729

    
730
    }
731

    
732
    for (i=0; i<2; i++) {
733
        if(left_type[i]){
734
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
735
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
736
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
737
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
738
        }else{
739
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
740
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
741
            h->non_zero_count_cache[0+8*1 +   8*i]=
742
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
743
        }
744
    }
745

    
746
    if( h->pps.cabac ) {
747
        // top_cbp
748
        if(top_type) {
749
            h->top_cbp = h->cbp_table[top_xy];
750
        } else if(IS_INTRA(mb_type)) {
751
            h->top_cbp = 0x1C0;
752
        } else {
753
            h->top_cbp = 0;
754
        }
755
        // left_cbp
756
        if (left_type[0]) {
757
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
758
        } else if(IS_INTRA(mb_type)) {
759
            h->left_cbp = 0x1C0;
760
        } else {
761
            h->left_cbp = 0;
762
        }
763
        if (left_type[0]) {
764
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
765
        }
766
        if (left_type[1]) {
767
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
768
        }
769
    }
770

    
771
#if 1
772
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
773
        int list;
774
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
775
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
776
                /*if(!h->mv_cache_clean[list]){
777
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
778
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
779
                    h->mv_cache_clean[list]= 1;
780
                }*/
781
                continue;
782
            }
783
            h->mv_cache_clean[list]= 0;
784

    
785
            if(USES_LIST(top_type, list)){
786
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
787
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
788
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
789
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
790
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
791
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
792
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
793
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
794
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
795
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
796
            }else{
797
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
798
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
799
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
800
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
801
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
802
            }
803

    
804
            //FIXME unify cleanup or sth
805
            if(USES_LIST(left_type[0], list)){
806
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
807
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
808
                *(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]];
809
                *(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]];
810
                h->ref_cache[list][scan8[0] - 1 + 0*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
811
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1]>>1)];
812
            }else{
813
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
814
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
815
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
816
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
817
            }
818

    
819
            if(USES_LIST(left_type[1], list)){
820
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
821
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
822
                *(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]];
823
                *(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]];
824
                h->ref_cache[list][scan8[0] - 1 + 2*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
825
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[3]>>1)];
826
            }else{
827
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
828
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
829
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
830
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
831
                assert((!left_type[0]) == (!left_type[1]));
832
            }
833

    
834
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
835
                continue;
836

    
837
            if(USES_LIST(topleft_type, list)){
838
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
839
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
840
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
841
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
842
            }else{
843
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
844
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
845
            }
846

    
847
            if(USES_LIST(topright_type, list)){
848
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
849
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
850
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
851
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
852
            }else{
853
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
854
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
855
            }
856

    
857
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
858
                continue;
859

    
860
            h->ref_cache[list][scan8[5 ]+1] =
861
            h->ref_cache[list][scan8[7 ]+1] =
862
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
863
            h->ref_cache[list][scan8[4 ]] =
864
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
865
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
866
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
867
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
868
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
869
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
870

    
871
            if( h->pps.cabac ) {
872
                /* XXX beurk, Load mvd */
873
                if(USES_LIST(top_type, list)){
874
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
875
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
876
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
877
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
878
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
879
                }else{
880
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
881
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
882
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
883
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
884
                }
885
                if(USES_LIST(left_type[0], list)){
886
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
887
                    *(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]];
888
                    *(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]];
889
                }else{
890
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
891
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
892
                }
893
                if(USES_LIST(left_type[1], list)){
894
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
895
                    *(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]];
896
                    *(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]];
897
                }else{
898
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
899
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
900
                }
901
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
902
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
903
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
904
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
905
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
906

    
907
                if(h->slice_type == B_TYPE){
908
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
909

    
910
                    if(IS_DIRECT(top_type)){
911
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
912
                    }else if(IS_8X8(top_type)){
913
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
914
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
915
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
916
                    }else{
917
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
918
                    }
919

    
920
                    if(IS_DIRECT(left_type[0]))
921
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
922
                    else if(IS_8X8(left_type[0]))
923
                        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)];
924
                    else
925
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
926

    
927
                    if(IS_DIRECT(left_type[1]))
928
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
929
                    else if(IS_8X8(left_type[1]))
930
                        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)];
931
                    else
932
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
933
                }
934
            }
935

    
936
            if(FRAME_MBAFF){
937
#define MAP_MVS\
938
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
939
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
940
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
941
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
942
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
943
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
944
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
945
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
946
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
947
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
948
                if(MB_FIELD){
949
#define MAP_F2F(idx, mb_type)\
950
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
951
                        h->ref_cache[list][idx] <<= 1;\
952
                        h->mv_cache[list][idx][1] /= 2;\
953
                        h->mvd_cache[list][idx][1] /= 2;\
954
                    }
955
                    MAP_MVS
956
#undef MAP_F2F
957
                }else{
958
#define MAP_F2F(idx, mb_type)\
959
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
960
                        h->ref_cache[list][idx] >>= 1;\
961
                        h->mv_cache[list][idx][1] <<= 1;\
962
                        h->mvd_cache[list][idx][1] <<= 1;\
963
                    }
964
                    MAP_MVS
965
#undef MAP_F2F
966
                }
967
            }
968
        }
969
    }
970
#endif
971

    
972
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
973
}
974

    
975
static inline void write_back_intra_pred_mode(H264Context *h){
976
    MpegEncContext * const s = &h->s;
977
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
978

    
979
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
980
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
981
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
982
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
983
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
984
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
985
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
986
}
987

    
988
/**
989
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
990
 */
991
static inline int check_intra4x4_pred_mode(H264Context *h){
992
    MpegEncContext * const s = &h->s;
993
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
994
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
995
    int i;
996

    
997
    if(!(h->top_samples_available&0x8000)){
998
        for(i=0; i<4; i++){
999
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
1000
            if(status<0){
1001
                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);
1002
                return -1;
1003
            } else if(status){
1004
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
1005
            }
1006
        }
1007
    }
1008

    
1009
    if(!(h->left_samples_available&0x8000)){
1010
        for(i=0; i<4; i++){
1011
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
1012
            if(status<0){
1013
                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);
1014
                return -1;
1015
            } else if(status){
1016
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
1017
            }
1018
        }
1019
    }
1020

    
1021
    return 0;
1022
} //FIXME cleanup like next
1023

    
1024
/**
1025
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
1026
 */
1027
static inline int check_intra_pred_mode(H264Context *h, int mode){
1028
    MpegEncContext * const s = &h->s;
1029
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
1030
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
1031

    
1032
    if(mode < 0 || mode > 6) {
1033
        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);
1034
        return -1;
1035
    }
1036

    
1037
    if(!(h->top_samples_available&0x8000)){
1038
        mode= top[ mode ];
1039
        if(mode<0){
1040
            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);
1041
            return -1;
1042
        }
1043
    }
1044

    
1045
    if(!(h->left_samples_available&0x8000)){
1046
        mode= left[ mode ];
1047
        if(mode<0){
1048
            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);
1049
            return -1;
1050
        }
1051
    }
1052

    
1053
    return mode;
1054
}
1055

    
1056
/**
1057
 * gets the predicted intra4x4 prediction mode.
1058
 */
1059
static inline int pred_intra_mode(H264Context *h, int n){
1060
    const int index8= scan8[n];
1061
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1062
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1063
    const int min= FFMIN(left, top);
1064

    
1065
    tprintf("mode:%d %d min:%d\n", left ,top, min);
1066

    
1067
    if(min<0) return DC_PRED;
1068
    else      return min;
1069
}
1070

    
1071
static inline void write_back_non_zero_count(H264Context *h){
1072
    MpegEncContext * const s = &h->s;
1073
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1074

    
1075
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1076
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1077
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
1078
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
1079
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1080
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1081
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
1082

    
1083
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
1084
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
1085
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
1086

    
1087
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
1088
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
1089
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
1090

    
1091
    if(FRAME_MBAFF){
1092
        // store all luma nnzs, for deblocking
1093
        int v = 0, i;
1094
        for(i=0; i<16; i++)
1095
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
1096
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
1097
    }
1098
}
1099

    
1100
/**
1101
 * gets the predicted number of non zero coefficients.
1102
 * @param n block index
1103
 */
1104
static inline int pred_non_zero_count(H264Context *h, int n){
1105
    const int index8= scan8[n];
1106
    const int left= h->non_zero_count_cache[index8 - 1];
1107
    const int top = h->non_zero_count_cache[index8 - 8];
1108
    int i= left + top;
1109

    
1110
    if(i<64) i= (i+1)>>1;
1111

    
1112
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1113

    
1114
    return i&31;
1115
}
1116

    
1117
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1118
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1119

    
1120
    /* there is no consistent mapping of mvs to neighboring locations that will
1121
     * make mbaff happy, so we can't move all this logic to fill_caches */
1122
    if(FRAME_MBAFF){
1123
        MpegEncContext *s = &h->s;
1124
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
1125
        const int16_t *mv;
1126
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
1127
        *C = h->mv_cache[list][scan8[0]-2];
1128

    
1129
        if(!MB_FIELD
1130
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
1131
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
1132
            if(IS_INTERLACED(mb_types[topright_xy])){
1133
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
1134
                const int x4 = X4, y4 = Y4;\
1135
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
1136
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
1137
                    return LIST_NOT_USED;\
1138
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
1139
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
1140
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
1141
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
1142

    
1143
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
1144
            }
1145
        }
1146
        if(topright_ref == PART_NOT_AVAILABLE
1147
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
1148
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
1149
            if(!MB_FIELD
1150
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
1151
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
1152
            }
1153
            if(MB_FIELD
1154
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
1155
               && i >= scan8[0]+8){
1156
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
1157
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
1158
            }
1159
        }
1160
#undef SET_DIAG_MV
1161
    }
1162

    
1163
    if(topright_ref != PART_NOT_AVAILABLE){
1164
        *C= h->mv_cache[list][ i - 8 + part_width ];
1165
        return topright_ref;
1166
    }else{
1167
        tprintf("topright MV not available\n");
1168

    
1169
        *C= h->mv_cache[list][ i - 8 - 1 ];
1170
        return h->ref_cache[list][ i - 8 - 1 ];
1171
    }
1172
}
1173

    
1174
/**
1175
 * gets the predicted MV.
1176
 * @param n the block index
1177
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1178
 * @param mx the x component of the predicted motion vector
1179
 * @param my the y component of the predicted motion vector
1180
 */
1181
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1182
    const int index8= scan8[n];
1183
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1184
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1185
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1186
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1187
    const int16_t * C;
1188
    int diagonal_ref, match_count;
1189

    
1190
    assert(part_width==1 || part_width==2 || part_width==4);
1191

    
1192
/* mv_cache
1193
  B . . A T T T T
1194
  U . . L . . , .
1195
  U . . L . . . .
1196
  U . . L . . , .
1197
  . . . L . . . .
1198
*/
1199

    
1200
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1201
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1202
    tprintf("pred_motion match_count=%d\n", match_count);
1203
    if(match_count > 1){ //most common
1204
        *mx= mid_pred(A[0], B[0], C[0]);
1205
        *my= mid_pred(A[1], B[1], C[1]);
1206
    }else if(match_count==1){
1207
        if(left_ref==ref){
1208
            *mx= A[0];
1209
            *my= A[1];
1210
        }else if(top_ref==ref){
1211
            *mx= B[0];
1212
            *my= B[1];
1213
        }else{
1214
            *mx= C[0];
1215
            *my= C[1];
1216
        }
1217
    }else{
1218
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1219
            *mx= A[0];
1220
            *my= A[1];
1221
        }else{
1222
            *mx= mid_pred(A[0], B[0], C[0]);
1223
            *my= mid_pred(A[1], B[1], C[1]);
1224
        }
1225
    }
1226

    
1227
    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);
1228
}
1229

    
1230
/**
1231
 * gets the directionally predicted 16x8 MV.
1232
 * @param n the block index
1233
 * @param mx the x component of the predicted motion vector
1234
 * @param my the y component of the predicted motion vector
1235
 */
1236
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1237
    if(n==0){
1238
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1239
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1240

    
1241
        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);
1242

    
1243
        if(top_ref == ref){
1244
            *mx= B[0];
1245
            *my= B[1];
1246
            return;
1247
        }
1248
    }else{
1249
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1250
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1251

    
1252
        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);
1253

    
1254
        if(left_ref == ref){
1255
            *mx= A[0];
1256
            *my= A[1];
1257
            return;
1258
        }
1259
    }
1260

    
1261
    //RARE
1262
    pred_motion(h, n, 4, list, ref, mx, my);
1263
}
1264

    
1265
/**
1266
 * gets the directionally predicted 8x16 MV.
1267
 * @param n the block index
1268
 * @param mx the x component of the predicted motion vector
1269
 * @param my the y component of the predicted motion vector
1270
 */
1271
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1272
    if(n==0){
1273
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1274
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1275

    
1276
        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);
1277

    
1278
        if(left_ref == ref){
1279
            *mx= A[0];
1280
            *my= A[1];
1281
            return;
1282
        }
1283
    }else{
1284
        const int16_t * C;
1285
        int diagonal_ref;
1286

    
1287
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1288

    
1289
        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);
1290

    
1291
        if(diagonal_ref == ref){
1292
            *mx= C[0];
1293
            *my= C[1];
1294
            return;
1295
        }
1296
    }
1297

    
1298
    //RARE
1299
    pred_motion(h, n, 2, list, ref, mx, my);
1300
}
1301

    
1302
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1303
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1304
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1305

    
1306
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1307

    
1308
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1309
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1310
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1311

    
1312
        *mx = *my = 0;
1313
        return;
1314
    }
1315

    
1316
    pred_motion(h, 0, 4, 0, 0, mx, my);
1317

    
1318
    return;
1319
}
1320

    
1321
static inline void direct_dist_scale_factor(H264Context * const h){
1322
    const int poc = h->s.current_picture_ptr->poc;
1323
    const int poc1 = h->ref_list[1][0].poc;
1324
    int i;
1325
    for(i=0; i<h->ref_count[0]; i++){
1326
        int poc0 = h->ref_list[0][i].poc;
1327
        int td = clip(poc1 - poc0, -128, 127);
1328
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1329
            h->dist_scale_factor[i] = 256;
1330
        }else{
1331
            int tb = clip(poc - poc0, -128, 127);
1332
            int tx = (16384 + (FFABS(td) >> 1)) / td;
1333
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1334
        }
1335
    }
1336
    if(FRAME_MBAFF){
1337
        for(i=0; i<h->ref_count[0]; i++){
1338
            h->dist_scale_factor_field[2*i] =
1339
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
1340
        }
1341
    }
1342
}
1343
static inline void direct_ref_list_init(H264Context * const h){
1344
    MpegEncContext * const s = &h->s;
1345
    Picture * const ref1 = &h->ref_list[1][0];
1346
    Picture * const cur = s->current_picture_ptr;
1347
    int list, i, j;
1348
    if(cur->pict_type == I_TYPE)
1349
        cur->ref_count[0] = 0;
1350
    if(cur->pict_type != B_TYPE)
1351
        cur->ref_count[1] = 0;
1352
    for(list=0; list<2; list++){
1353
        cur->ref_count[list] = h->ref_count[list];
1354
        for(j=0; j<h->ref_count[list]; j++)
1355
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1356
    }
1357
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1358
        return;
1359
    for(list=0; list<2; list++){
1360
        for(i=0; i<ref1->ref_count[list]; i++){
1361
            const int poc = ref1->ref_poc[list][i];
1362
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1363
            for(j=0; j<h->ref_count[list]; j++)
1364
                if(h->ref_list[list][j].poc == poc){
1365
                    h->map_col_to_list0[list][i] = j;
1366
                    break;
1367
                }
1368
        }
1369
    }
1370
    if(FRAME_MBAFF){
1371
        for(list=0; list<2; list++){
1372
            for(i=0; i<ref1->ref_count[list]; i++){
1373
                j = h->map_col_to_list0[list][i];
1374
                h->map_col_to_list0_field[list][2*i] = 2*j;
1375
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1376
            }
1377
        }
1378
    }
1379
}
1380

    
1381
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1382
    MpegEncContext * const s = &h->s;
1383
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1384
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1385
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1386
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1387
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1388
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1389
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1390
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1391
    const int is_b8x8 = IS_8X8(*mb_type);
1392
    int sub_mb_type;
1393
    int i8, i4;
1394

    
1395
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1396
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1397
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1398
         * so we know exactly what block size to use */
1399
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1400
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1401
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1402
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1403
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1404
    }else{
1405
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1406
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1407
    }
1408
    if(!is_b8x8)
1409
        *mb_type |= MB_TYPE_DIRECT2;
1410
    if(MB_FIELD)
1411
        *mb_type |= MB_TYPE_INTERLACED;
1412

    
1413
    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);
1414

    
1415
    if(h->direct_spatial_mv_pred){
1416
        int ref[2];
1417
        int mv[2][2];
1418
        int list;
1419

    
1420
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1421

    
1422
        /* ref = min(neighbors) */
1423
        for(list=0; list<2; list++){
1424
            int refa = h->ref_cache[list][scan8[0] - 1];
1425
            int refb = h->ref_cache[list][scan8[0] - 8];
1426
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1427
            if(refc == -2)
1428
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1429
            ref[list] = refa;
1430
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1431
                ref[list] = refb;
1432
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1433
                ref[list] = refc;
1434
            if(ref[list] < 0)
1435
                ref[list] = -1;
1436
        }
1437

    
1438
        if(ref[0] < 0 && ref[1] < 0){
1439
            ref[0] = ref[1] = 0;
1440
            mv[0][0] = mv[0][1] =
1441
            mv[1][0] = mv[1][1] = 0;
1442
        }else{
1443
            for(list=0; list<2; list++){
1444
                if(ref[list] >= 0)
1445
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1446
                else
1447
                    mv[list][0] = mv[list][1] = 0;
1448
            }
1449
        }
1450

    
1451
        if(ref[1] < 0){
1452
            *mb_type &= ~MB_TYPE_P0L1;
1453
            sub_mb_type &= ~MB_TYPE_P0L1;
1454
        }else if(ref[0] < 0){
1455
            *mb_type &= ~MB_TYPE_P0L0;
1456
            sub_mb_type &= ~MB_TYPE_P0L0;
1457
        }
1458

    
1459
        if(IS_16X16(*mb_type)){
1460
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1461
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1462
            if(!IS_INTRA(mb_type_col)
1463
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1464
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1465
                       && (h->x264_build>33 || !h->x264_build)))){
1466
                if(ref[0] > 0)
1467
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1468
                else
1469
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1470
                if(ref[1] > 0)
1471
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1472
                else
1473
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1474
            }else{
1475
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1476
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1477
            }
1478
        }else{
1479
            for(i8=0; i8<4; i8++){
1480
                const int x8 = i8&1;
1481
                const int y8 = i8>>1;
1482

    
1483
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1484
                    continue;
1485
                h->sub_mb_type[i8] = sub_mb_type;
1486

    
1487
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1488
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1489
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1490
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1491

    
1492
                /* col_zero_flag */
1493
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1494
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1495
                                                  && (h->x264_build>33 || !h->x264_build)))){
1496
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1497
                    if(IS_SUB_8X8(sub_mb_type)){
1498
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1499
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1500
                            if(ref[0] == 0)
1501
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1502
                            if(ref[1] == 0)
1503
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1504
                        }
1505
                    }else
1506
                    for(i4=0; i4<4; i4++){
1507
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1508
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1509
                            if(ref[0] == 0)
1510
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1511
                            if(ref[1] == 0)
1512
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1513
                        }
1514
                    }
1515
                }
1516
            }
1517
        }
1518
    }else{ /* direct temporal mv pred */
1519
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1520
        const int *dist_scale_factor = h->dist_scale_factor;
1521

    
1522
        if(FRAME_MBAFF){
1523
            if(IS_INTERLACED(*mb_type)){
1524
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1525
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1526
                dist_scale_factor = h->dist_scale_factor_field;
1527
            }
1528
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1529
                /* FIXME assumes direct_8x8_inference == 1 */
1530
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1531
                int mb_types_col[2];
1532
                int y_shift;
1533

    
1534
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1535
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1536
                         | (*mb_type & MB_TYPE_INTERLACED);
1537
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1538

    
1539
                if(IS_INTERLACED(*mb_type)){
1540
                    /* frame to field scaling */
1541
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1542
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1543
                    if(s->mb_y&1){
1544
                        l1ref0 -= 2*h->b8_stride;
1545
                        l1ref1 -= 2*h->b8_stride;
1546
                        l1mv0 -= 4*h->b_stride;
1547
                        l1mv1 -= 4*h->b_stride;
1548
                    }
1549
                    y_shift = 0;
1550

    
1551
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1552
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1553
                       && !is_b8x8)
1554
                        *mb_type |= MB_TYPE_16x8;
1555
                    else
1556
                        *mb_type |= MB_TYPE_8x8;
1557
                }else{
1558
                    /* field to frame scaling */
1559
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1560
                     * but in MBAFF, top and bottom POC are equal */
1561
                    int dy = (s->mb_y&1) ? 1 : 2;
1562
                    mb_types_col[0] =
1563
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1564
                    l1ref0 += dy*h->b8_stride;
1565
                    l1ref1 += dy*h->b8_stride;
1566
                    l1mv0 += 2*dy*h->b_stride;
1567
                    l1mv1 += 2*dy*h->b_stride;
1568
                    y_shift = 2;
1569

    
1570
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1571
                       && !is_b8x8)
1572
                        *mb_type |= MB_TYPE_16x16;
1573
                    else
1574
                        *mb_type |= MB_TYPE_8x8;
1575
                }
1576

    
1577
                for(i8=0; i8<4; i8++){
1578
                    const int x8 = i8&1;
1579
                    const int y8 = i8>>1;
1580
                    int ref0, scale;
1581
                    const int16_t (*l1mv)[2]= l1mv0;
1582

    
1583
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1584
                        continue;
1585
                    h->sub_mb_type[i8] = sub_mb_type;
1586

    
1587
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1588
                    if(IS_INTRA(mb_types_col[y8])){
1589
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1590
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1591
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1592
                        continue;
1593
                    }
1594

    
1595
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1596
                    if(ref0 >= 0)
1597
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1598
                    else{
1599
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1600
                        l1mv= l1mv1;
1601
                    }
1602
                    scale = dist_scale_factor[ref0];
1603
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1604

    
1605
                    {
1606
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1607
                        int my_col = (mv_col[1]<<y_shift)/2;
1608
                        int mx = (scale * mv_col[0] + 128) >> 8;
1609
                        int my = (scale * my_col + 128) >> 8;
1610
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1611
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1612
                    }
1613
                }
1614
                return;
1615
            }
1616
        }
1617

    
1618
        /* one-to-one mv scaling */
1619

    
1620
        if(IS_16X16(*mb_type)){
1621
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1622
            if(IS_INTRA(mb_type_col)){
1623
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1624
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1625
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1626
            }else{
1627
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1628
                                                : map_col_to_list0[1][l1ref1[0]];
1629
                const int scale = dist_scale_factor[ref0];
1630
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1631
                int mv_l0[2];
1632
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1633
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1634
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1635
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1636
                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);
1637
            }
1638
        }else{
1639
            for(i8=0; i8<4; i8++){
1640
                const int x8 = i8&1;
1641
                const int y8 = i8>>1;
1642
                int ref0, scale;
1643
                const int16_t (*l1mv)[2]= l1mv0;
1644

    
1645
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1646
                    continue;
1647
                h->sub_mb_type[i8] = sub_mb_type;
1648
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1649
                if(IS_INTRA(mb_type_col)){
1650
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1651
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1652
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1653
                    continue;
1654
                }
1655

    
1656
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1657
                if(ref0 >= 0)
1658
                    ref0 = map_col_to_list0[0][ref0];
1659
                else{
1660
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1661
                    l1mv= l1mv1;
1662
                }
1663
                scale = dist_scale_factor[ref0];
1664

    
1665
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1666
                if(IS_SUB_8X8(sub_mb_type)){
1667
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1668
                    int mx = (scale * mv_col[0] + 128) >> 8;
1669
                    int my = (scale * mv_col[1] + 128) >> 8;
1670
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1671
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1672
                }else
1673
                for(i4=0; i4<4; i4++){
1674
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1675
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1676
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1677
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1678
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1679
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1680
                }
1681
            }
1682
        }
1683
    }
1684
}
1685

    
1686
static inline void write_back_motion(H264Context *h, int mb_type){
1687
    MpegEncContext * const s = &h->s;
1688
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1689
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1690
    int list;
1691

    
1692
    if(!USES_LIST(mb_type, 0))
1693
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1694

    
1695
    for(list=0; list<2; list++){
1696
        int y;
1697
        if(!USES_LIST(mb_type, list))
1698
            continue;
1699

    
1700
        for(y=0; y<4; y++){
1701
            *(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];
1702
            *(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];
1703
        }
1704
        if( h->pps.cabac ) {
1705
            if(IS_SKIP(mb_type))
1706
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1707
            else
1708
            for(y=0; y<4; y++){
1709
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1710
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1711
            }
1712
        }
1713

    
1714
        {
1715
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1716
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1717
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1718
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1719
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1720
        }
1721
    }
1722

    
1723
    if(h->slice_type == B_TYPE && h->pps.cabac){
1724
        if(IS_8X8(mb_type)){
1725
            uint8_t *direct_table = &h->direct_table[b8_xy];
1726
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1727
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1728
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1729
        }
1730
    }
1731
}
1732

    
1733
/**
1734
 * Decodes a network abstraction layer unit.
1735
 * @param consumed is the number of bytes used as input
1736
 * @param length is the length of the array
1737
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1738
 * @returns decoded bytes, might be src+1 if no escapes
1739
 */
1740
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1741
    int i, si, di;
1742
    uint8_t *dst;
1743

    
1744
//    src[0]&0x80;                //forbidden bit
1745
    h->nal_ref_idc= src[0]>>5;
1746
    h->nal_unit_type= src[0]&0x1F;
1747

    
1748
    src++; length--;
1749
#if 0
1750
    for(i=0; i<length; i++)
1751
        printf("%2X ", src[i]);
1752
#endif
1753
    for(i=0; i+1<length; i+=2){
1754
        if(src[i]) continue;
1755
        if(i>0 && src[i-1]==0) i--;
1756
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1757
            if(src[i+2]!=3){
1758
                /* startcode, so we must be past the end */
1759
                length=i;
1760
            }
1761
            break;
1762
        }
1763
    }
1764

    
1765
    if(i>=length-1){ //no escaped 0
1766
        *dst_length= length;
1767
        *consumed= length+1; //+1 for the header
1768
        return src;
1769
    }
1770

    
1771
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1772
    dst= h->rbsp_buffer;
1773

    
1774
//printf("decoding esc\n");
1775
    si=di=0;
1776
    while(si<length){
1777
        //remove escapes (very rare 1:2^22)
1778
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1779
            if(src[si+2]==3){ //escape
1780
                dst[di++]= 0;
1781
                dst[di++]= 0;
1782
                si+=3;
1783
                continue;
1784
            }else //next start code
1785
                break;
1786
        }
1787

    
1788
        dst[di++]= src[si++];
1789
    }
1790

    
1791
    *dst_length= di;
1792
    *consumed= si + 1;//+1 for the header
1793
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1794
    return dst;
1795
}
1796

    
1797
#if 0
1798
/**
1799
 * @param src the data which should be escaped
1800
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1801
 * @param length the length of the src data
1802
 * @param dst_length the length of the dst array
1803
 * @returns length of escaped data in bytes or -1 if an error occured
1804
 */
1805
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1806
    int i, escape_count, si, di;
1807
    uint8_t *temp;
1808

1809
    assert(length>=0);
1810
    assert(dst_length>0);
1811

1812
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1813

1814
    if(length==0) return 1;
1815

1816
    escape_count= 0;
1817
    for(i=0; i<length; i+=2){
1818
        if(src[i]) continue;
1819
        if(i>0 && src[i-1]==0)
1820
            i--;
1821
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1822
            escape_count++;
1823
            i+=2;
1824
        }
1825
    }
1826

1827
    if(escape_count==0){
1828
        if(dst+1 != src)
1829
            memcpy(dst+1, src, length);
1830
        return length + 1;
1831
    }
1832

1833
    if(length + escape_count + 1> dst_length)
1834
        return -1;
1835

1836
    //this should be damn rare (hopefully)
1837

1838
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1839
    temp= h->rbsp_buffer;
1840
//printf("encoding esc\n");
1841

1842
    si= 0;
1843
    di= 0;
1844
    while(si < length){
1845
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1846
            temp[di++]= 0; si++;
1847
            temp[di++]= 0; si++;
1848
            temp[di++]= 3;
1849
            temp[di++]= src[si++];
1850
        }
1851
        else
1852
            temp[di++]= src[si++];
1853
    }
1854
    memcpy(dst+1, temp, length+escape_count);
1855

1856
    assert(di == length+escape_count);
1857

1858
    return di + 1;
1859
}
1860

1861
/**
1862
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1863
 */
1864
static void encode_rbsp_trailing(PutBitContext *pb){
1865
    int length;
1866
    put_bits(pb, 1, 1);
1867
    length= (-put_bits_count(pb))&7;
1868
    if(length) put_bits(pb, length, 0);
1869
}
1870
#endif
1871

    
1872
/**
1873
 * identifies the exact end of the bitstream
1874
 * @return the length of the trailing, or 0 if damaged
1875
 */
1876
static int decode_rbsp_trailing(uint8_t *src){
1877
    int v= *src;
1878
    int r;
1879

    
1880
    tprintf("rbsp trailing %X\n", v);
1881

    
1882
    for(r=1; r<9; r++){
1883
        if(v&1) return r;
1884
        v>>=1;
1885
    }
1886
    return 0;
1887
}
1888

    
1889
/**
1890
 * idct tranforms the 16 dc values and dequantize them.
1891
 * @param qp quantization parameter
1892
 */
1893
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1894
#define stride 16
1895
    int i;
1896
    int temp[16]; //FIXME check if this is a good idea
1897
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1898
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1899

    
1900
//memset(block, 64, 2*256);
1901
//return;
1902
    for(i=0; i<4; i++){
1903
        const int offset= y_offset[i];
1904
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1905
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1906
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1907
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1908

    
1909
        temp[4*i+0]= z0+z3;
1910
        temp[4*i+1]= z1+z2;
1911
        temp[4*i+2]= z1-z2;
1912
        temp[4*i+3]= z0-z3;
1913
    }
1914

    
1915
    for(i=0; i<4; i++){
1916
        const int offset= x_offset[i];
1917
        const int z0= temp[4*0+i] + temp[4*2+i];
1918
        const int z1= temp[4*0+i] - temp[4*2+i];
1919
        const int z2= temp[4*1+i] - temp[4*3+i];
1920
        const int z3= temp[4*1+i] + temp[4*3+i];
1921

    
1922
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1923
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1924
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1925
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1926
    }
1927
}
1928

    
1929
#if 0
1930
/**
1931
 * dct tranforms the 16 dc values.
1932
 * @param qp quantization parameter ??? FIXME
1933
 */
1934
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1935
//    const int qmul= dequant_coeff[qp][0];
1936
    int i;
1937
    int temp[16]; //FIXME check if this is a good idea
1938
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1939
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1940

1941
    for(i=0; i<4; i++){
1942
        const int offset= y_offset[i];
1943
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1944
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1945
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1946
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1947

1948
        temp[4*i+0]= z0+z3;
1949
        temp[4*i+1]= z1+z2;
1950
        temp[4*i+2]= z1-z2;
1951
        temp[4*i+3]= z0-z3;
1952
    }
1953

1954
    for(i=0; i<4; i++){
1955
        const int offset= x_offset[i];
1956
        const int z0= temp[4*0+i] + temp[4*2+i];
1957
        const int z1= temp[4*0+i] - temp[4*2+i];
1958
        const int z2= temp[4*1+i] - temp[4*3+i];
1959
        const int z3= temp[4*1+i] + temp[4*3+i];
1960

1961
        block[stride*0 +offset]= (z0 + z3)>>1;
1962
        block[stride*2 +offset]= (z1 + z2)>>1;
1963
        block[stride*8 +offset]= (z1 - z2)>>1;
1964
        block[stride*10+offset]= (z0 - z3)>>1;
1965
    }
1966
}
1967
#endif
1968

    
1969
#undef xStride
1970
#undef stride
1971

    
1972
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1973
    const int stride= 16*2;
1974
    const int xStride= 16;
1975
    int a,b,c,d,e;
1976

    
1977
    a= block[stride*0 + xStride*0];
1978
    b= block[stride*0 + xStride*1];
1979
    c= block[stride*1 + xStride*0];
1980
    d= block[stride*1 + xStride*1];
1981

    
1982
    e= a-b;
1983
    a= a+b;
1984
    b= c-d;
1985
    c= c+d;
1986

    
1987
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1988
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1989
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1990
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1991
}
1992

    
1993
#if 0
1994
static void chroma_dc_dct_c(DCTELEM *block){
1995
    const int stride= 16*2;
1996
    const int xStride= 16;
1997
    int a,b,c,d,e;
1998

1999
    a= block[stride*0 + xStride*0];
2000
    b= block[stride*0 + xStride*1];
2001
    c= block[stride*1 + xStride*0];
2002
    d= block[stride*1 + xStride*1];
2003

2004
    e= a-b;
2005
    a= a+b;
2006
    b= c-d;
2007
    c= c+d;
2008

2009
    block[stride*0 + xStride*0]= (a+c);
2010
    block[stride*0 + xStride*1]= (e+b);
2011
    block[stride*1 + xStride*0]= (a-c);
2012
    block[stride*1 + xStride*1]= (e-b);
2013
}
2014
#endif
2015

    
2016
/**
2017
 * gets the chroma qp.
2018
 */
2019
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
2020

    
2021
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
2022
}
2023

    
2024

    
2025
#if 0
2026
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
2027
    int i;
2028
    //FIXME try int temp instead of block
2029

2030
    for(i=0; i<4; i++){
2031
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
2032
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
2033
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
2034
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
2035
        const int z0= d0 + d3;
2036
        const int z3= d0 - d3;
2037
        const int z1= d1 + d2;
2038
        const int z2= d1 - d2;
2039

2040
        block[0 + 4*i]=   z0 +   z1;
2041
        block[1 + 4*i]= 2*z3 +   z2;
2042
        block[2 + 4*i]=   z0 -   z1;
2043
        block[3 + 4*i]=   z3 - 2*z2;
2044
    }
2045

2046
    for(i=0; i<4; i++){
2047
        const int z0= block[0*4 + i] + block[3*4 + i];
2048
        const int z3= block[0*4 + i] - block[3*4 + i];
2049
        const int z1= block[1*4 + i] + block[2*4 + i];
2050
        const int z2= block[1*4 + i] - block[2*4 + i];
2051

2052
        block[0*4 + i]=   z0 +   z1;
2053
        block[1*4 + i]= 2*z3 +   z2;
2054
        block[2*4 + i]=   z0 -   z1;
2055
        block[3*4 + i]=   z3 - 2*z2;
2056
    }
2057
}
2058
#endif
2059

    
2060
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
2061
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
2062
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
2063
    int i;
2064
    const int * const quant_table= quant_coeff[qscale];
2065
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
2066
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
2067
    const unsigned int threshold2= (threshold1<<1);
2068
    int last_non_zero;
2069

    
2070
    if(seperate_dc){
2071
        if(qscale<=18){
2072
            //avoid overflows
2073
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
2074
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
2075
            const unsigned int dc_threshold2= (dc_threshold1<<1);
2076

    
2077
            int level= block[0]*quant_coeff[qscale+18][0];
2078
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
2079
                if(level>0){
2080
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
2081
                    block[0]= level;
2082
                }else{
2083
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
2084
                    block[0]= -level;
2085
                }
2086
//                last_non_zero = i;
2087
            }else{
2088
                block[0]=0;
2089
            }
2090
        }else{
2091
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
2092
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
2093
            const unsigned int dc_threshold2= (dc_threshold1<<1);
2094

    
2095
            int level= block[0]*quant_table[0];
2096
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
2097
                if(level>0){
2098
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
2099
                    block[0]= level;
2100
                }else{
2101
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
2102
                    block[0]= -level;
2103
                }
2104
//                last_non_zero = i;
2105
            }else{
2106
                block[0]=0;
2107
            }
2108
        }
2109
        last_non_zero= 0;
2110
        i=1;
2111
    }else{
2112
        last_non_zero= -1;
2113
        i=0;
2114
    }
2115

    
2116
    for(; i<16; i++){
2117
        const int j= scantable[i];
2118
        int level= block[j]*quant_table[j];
2119

    
2120
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
2121
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
2122
        if(((unsigned)(level+threshold1))>threshold2){
2123
            if(level>0){
2124
                level= (bias + level)>>QUANT_SHIFT;
2125
                block[j]= level;
2126
            }else{
2127
                level= (bias - level)>>QUANT_SHIFT;
2128
                block[j]= -level;
2129
            }
2130
            last_non_zero = i;
2131
        }else{
2132
            block[j]=0;
2133
        }
2134
    }
2135

    
2136
    return last_non_zero;
2137
}
2138

    
2139
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
2140
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2141
    ((uint32_t*)(src+0*stride))[0]= a;
2142
    ((uint32_t*)(src+1*stride))[0]= a;
2143
    ((uint32_t*)(src+2*stride))[0]= a;
2144
    ((uint32_t*)(src+3*stride))[0]= a;
2145
}
2146

    
2147
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
2148
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
2149
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
2150
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
2151
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
2152
}
2153

    
2154
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
2155
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
2156
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
2157

    
2158
    ((uint32_t*)(src+0*stride))[0]=
2159
    ((uint32_t*)(src+1*stride))[0]=
2160
    ((uint32_t*)(src+2*stride))[0]=
2161
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2162
}
2163

    
2164
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
2165
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
2166

    
2167
    ((uint32_t*)(src+0*stride))[0]=
2168
    ((uint32_t*)(src+1*stride))[0]=
2169
    ((uint32_t*)(src+2*stride))[0]=
2170
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2171
}
2172

    
2173
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
2174
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
2175

    
2176
    ((uint32_t*)(src+0*stride))[0]=
2177
    ((uint32_t*)(src+1*stride))[0]=
2178
    ((uint32_t*)(src+2*stride))[0]=
2179
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2180
}
2181

    
2182
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
2183
    ((uint32_t*)(src+0*stride))[0]=
2184
    ((uint32_t*)(src+1*stride))[0]=
2185
    ((uint32_t*)(src+2*stride))[0]=
2186
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
2187
}
2188

    
2189

    
2190
#define LOAD_TOP_RIGHT_EDGE\
2191
    const int t4= topright[0];\
2192
    const int t5= topright[1];\
2193
    const int t6= topright[2];\
2194
    const int t7= topright[3];\
2195

    
2196
#define LOAD_LEFT_EDGE\
2197
    const int l0= src[-1+0*stride];\
2198
    const int l1= src[-1+1*stride];\
2199
    const int l2= src[-1+2*stride];\
2200
    const int l3= src[-1+3*stride];\
2201

    
2202
#define LOAD_TOP_EDGE\
2203
    const int t0= src[ 0-1*stride];\
2204
    const int t1= src[ 1-1*stride];\
2205
    const int t2= src[ 2-1*stride];\
2206
    const int t3= src[ 3-1*stride];\
2207

    
2208
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
2209
    const int lt= src[-1-1*stride];
2210
    LOAD_TOP_EDGE
2211
    LOAD_LEFT_EDGE
2212

    
2213
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
2214
    src[0+2*stride]=
2215
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
2216
    src[0+1*stride]=
2217
    src[1+2*stride]=
2218
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
2219
    src[0+0*stride]=
2220
    src[1+1*stride]=
2221
    src[2+2*stride]=
2222
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2223
    src[1+0*stride]=
2224
    src[2+1*stride]=
2225
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
2226
    src[2+0*stride]=
2227
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2228
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2229
}
2230

    
2231
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
2232
    LOAD_TOP_EDGE
2233
    LOAD_TOP_RIGHT_EDGE
2234
//    LOAD_LEFT_EDGE
2235

    
2236
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
2237
    src[1+0*stride]=
2238
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
2239
    src[2+0*stride]=
2240
    src[1+1*stride]=
2241
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
2242
    src[3+0*stride]=
2243
    src[2+1*stride]=
2244
    src[1+2*stride]=
2245
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
2246
    src[3+1*stride]=
2247
    src[2+2*stride]=
2248
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
2249
    src[3+2*stride]=
2250
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
2251
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
2252
}
2253

    
2254
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
2255
    const int lt= src[-1-1*stride];
2256
    LOAD_TOP_EDGE
2257
    LOAD_LEFT_EDGE
2258
    const __attribute__((unused)) int unu= l3;
2259

    
2260
    src[0+0*stride]=
2261
    src[1+2*stride]=(lt + t0 + 1)>>1;
2262
    src[1+0*stride]=
2263
    src[2+2*stride]=(t0 + t1 + 1)>>1;
2264
    src[2+0*stride]=
2265
    src[3+2*stride]=(t1 + t2 + 1)>>1;
2266
    src[3+0*stride]=(t2 + t3 + 1)>>1;
2267
    src[0+1*stride]=
2268
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2269
    src[1+1*stride]=
2270
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2271
    src[2+1*stride]=
2272
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2273
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2274
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2275
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2276
}
2277

    
2278
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
2279
    LOAD_TOP_EDGE
2280
    LOAD_TOP_RIGHT_EDGE
2281
    const __attribute__((unused)) int unu= t7;
2282

    
2283
    src[0+0*stride]=(t0 + t1 + 1)>>1;
2284
    src[1+0*stride]=
2285
    src[0+2*stride]=(t1 + t2 + 1)>>1;
2286
    src[2+0*stride]=
2287
    src[1+2*stride]=(t2 + t3 + 1)>>1;
2288
    src[3+0*stride]=
2289
    src[2+2*stride]=(t3 + t4+ 1)>>1;
2290
    src[3+2*stride]=(t4 + t5+ 1)>>1;
2291
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2292
    src[1+1*stride]=
2293
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2294
    src[2+1*stride]=
2295
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2296
    src[3+1*stride]=
2297
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2298
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2299
}
2300

    
2301
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2302
    LOAD_LEFT_EDGE
2303

    
2304
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2305
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2306
    src[2+0*stride]=
2307
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2308
    src[3+0*stride]=
2309
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2310
    src[2+1*stride]=
2311
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2312
    src[3+1*stride]=
2313
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2314
    src[3+2*stride]=
2315
    src[1+3*stride]=
2316
    src[0+3*stride]=
2317
    src[2+2*stride]=
2318
    src[2+3*stride]=
2319
    src[3+3*stride]=l3;
2320
}
2321

    
2322
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2323
    const int lt= src[-1-1*stride];
2324
    LOAD_TOP_EDGE
2325
    LOAD_LEFT_EDGE
2326
    const __attribute__((unused)) int unu= t3;
2327

    
2328
    src[0+0*stride]=
2329
    src[2+1*stride]=(lt + l0 + 1)>>1;
2330
    src[1+0*stride]=
2331
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2332
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2333
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2334
    src[0+1*stride]=
2335
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2336
    src[1+1*stride]=
2337
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2338
    src[0+2*stride]=
2339
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2340
    src[1+2*stride]=
2341
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2342
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2343
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2344
}
2345

    
2346
static void pred16x16_vertical_c(uint8_t *src, int stride){
2347
    int i;
2348
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2349
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2350
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2351
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2352

    
2353
    for(i=0; i<16; i++){
2354
        ((uint32_t*)(src+i*stride))[0]= a;
2355
        ((uint32_t*)(src+i*stride))[1]= b;
2356
        ((uint32_t*)(src+i*stride))[2]= c;
2357
        ((uint32_t*)(src+i*stride))[3]= d;
2358
    }
2359
}
2360

    
2361
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2362
    int i;
2363

    
2364
    for(i=0; i<16; i++){
2365
        ((uint32_t*)(src+i*stride))[0]=
2366
        ((uint32_t*)(src+i*stride))[1]=
2367
        ((uint32_t*)(src+i*stride))[2]=
2368
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2369
    }
2370
}
2371

    
2372
static void pred16x16_dc_c(uint8_t *src, int stride){
2373
    int i, dc=0;
2374

    
2375
    for(i=0;i<16; i++){
2376
        dc+= src[-1+i*stride];
2377
    }
2378

    
2379
    for(i=0;i<16; i++){
2380
        dc+= src[i-stride];
2381
    }
2382

    
2383
    dc= 0x01010101*((dc + 16)>>5);
2384

    
2385
    for(i=0; i<16; i++){
2386
        ((uint32_t*)(src+i*stride))[0]=
2387
        ((uint32_t*)(src+i*stride))[1]=
2388
        ((uint32_t*)(src+i*stride))[2]=
2389
        ((uint32_t*)(src+i*stride))[3]= dc;
2390
    }
2391
}
2392

    
2393
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2394
    int i, dc=0;
2395

    
2396
    for(i=0;i<16; i++){
2397
        dc+= src[-1+i*stride];
2398
    }
2399

    
2400
    dc= 0x01010101*((dc + 8)>>4);
2401

    
2402
    for(i=0; i<16; i++){
2403
        ((uint32_t*)(src+i*stride))[0]=
2404
        ((uint32_t*)(src+i*stride))[1]=
2405
        ((uint32_t*)(src+i*stride))[2]=
2406
        ((uint32_t*)(src+i*stride))[3]= dc;
2407
    }
2408
}
2409

    
2410
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2411
    int i, dc=0;
2412

    
2413
    for(i=0;i<16; i++){
2414
        dc+= src[i-stride];
2415
    }
2416
    dc= 0x01010101*((dc + 8)>>4);
2417

    
2418
    for(i=0; i<16; i++){
2419
        ((uint32_t*)(src+i*stride))[0]=
2420
        ((uint32_t*)(src+i*stride))[1]=
2421
        ((uint32_t*)(src+i*stride))[2]=
2422
        ((uint32_t*)(src+i*stride))[3]= dc;
2423
    }
2424
}
2425

    
2426
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2427
    int i;
2428

    
2429
    for(i=0; i<16; i++){
2430
        ((uint32_t*)(src+i*stride))[0]=
2431
        ((uint32_t*)(src+i*stride))[1]=
2432
        ((uint32_t*)(src+i*stride))[2]=
2433
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2434
    }
2435
}
2436

    
2437
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2438
  int i, j, k;
2439
  int a;
2440
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2441
  const uint8_t * const src0 = src+7-stride;
2442
  const uint8_t *src1 = src+8*stride-1;
2443
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2444
  int H = src0[1] - src0[-1];
2445
  int V = src1[0] - src2[ 0];
2446
  for(k=2; k<=8; ++k) {
2447
    src1 += stride; src2 -= stride;
2448
    H += k*(src0[k] - src0[-k]);
2449
    V += k*(src1[0] - src2[ 0]);
2450
  }
2451
  if(svq3){
2452
    H = ( 5*(H/4) ) / 16;
2453
    V = ( 5*(V/4) ) / 16;
2454

    
2455
    /* required for 100% accuracy */
2456
    i = H; H = V; V = i;
2457
  }else{
2458
    H = ( 5*H+32 ) >> 6;
2459
    V = ( 5*V+32 ) >> 6;
2460
  }
2461

    
2462
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2463
  for(j=16; j>0; --j) {
2464
    int b = a;
2465
    a += V;
2466
    for(i=-16; i<0; i+=4) {
2467
      src[16+i] = cm[ (b    ) >> 5 ];
2468
      src[17+i] = cm[ (b+  H) >> 5 ];
2469
      src[18+i] = cm[ (b+2*H) >> 5 ];
2470
      src[19+i] = cm[ (b+3*H) >> 5 ];
2471
      b += 4*H;
2472
    }
2473
    src += stride;
2474
  }
2475
}
2476

    
2477
static void pred16x16_plane_c(uint8_t *src, int stride){
2478
    pred16x16_plane_compat_c(src, stride, 0);
2479
}
2480

    
2481
static void pred8x8_vertical_c(uint8_t *src, int stride){
2482
    int i;
2483
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2484
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2485

    
2486
    for(i=0; i<8; i++){
2487
        ((uint32_t*)(src+i*stride))[0]= a;
2488
        ((uint32_t*)(src+i*stride))[1]= b;
2489
    }
2490
}
2491

    
2492
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2493
    int i;
2494

    
2495
    for(i=0; i<8; i++){
2496
        ((uint32_t*)(src+i*stride))[0]=
2497
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2498
    }
2499
}
2500

    
2501
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2502
    int i;
2503

    
2504
    for(i=0; i<8; i++){
2505
        ((uint32_t*)(src+i*stride))[0]=
2506
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2507
    }
2508
}
2509

    
2510
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2511
    int i;
2512
    int dc0, dc2;
2513

    
2514
    dc0=dc2=0;
2515
    for(i=0;i<4; i++){
2516
        dc0+= src[-1+i*stride];
2517
        dc2+= src[-1+(i+4)*stride];
2518
    }
2519
    dc0= 0x01010101*((dc0 + 2)>>2);
2520
    dc2= 0x01010101*((dc2 + 2)>>2);
2521

    
2522
    for(i=0; i<4; i++){
2523
        ((uint32_t*)(src+i*stride))[0]=
2524
        ((uint32_t*)(src+i*stride))[1]= dc0;
2525
    }
2526
    for(i=4; i<8; i++){
2527
        ((uint32_t*)(src+i*stride))[0]=
2528
        ((uint32_t*)(src+i*stride))[1]= dc2;
2529
    }
2530
}
2531

    
2532
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2533
    int i;
2534
    int dc0, dc1;
2535

    
2536
    dc0=dc1=0;
2537
    for(i=0;i<4; i++){
2538
        dc0+= src[i-stride];
2539
        dc1+= src[4+i-stride];
2540
    }
2541
    dc0= 0x01010101*((dc0 + 2)>>2);
2542
    dc1= 0x01010101*((dc1 + 2)>>2);
2543

    
2544
    for(i=0; i<4; i++){
2545
        ((uint32_t*)(src+i*stride))[0]= dc0;
2546
        ((uint32_t*)(src+i*stride))[1]= dc1;
2547
    }
2548
    for(i=4; i<8; i++){
2549
        ((uint32_t*)(src+i*stride))[0]= dc0;
2550
        ((uint32_t*)(src+i*stride))[1]= dc1;
2551
    }
2552
}
2553

    
2554

    
2555
static void pred8x8_dc_c(uint8_t *src, int stride){
2556
    int i;
2557
    int dc0, dc1, dc2, dc3;
2558

    
2559
    dc0=dc1=dc2=0;
2560
    for(i=0;i<4; i++){
2561
        dc0+= src[-1+i*stride] + src[i-stride];
2562
        dc1+= src[4+i-stride];
2563
        dc2+= src[-1+(i+4)*stride];
2564
    }
2565
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2566
    dc0= 0x01010101*((dc0 + 4)>>3);
2567
    dc1= 0x01010101*((dc1 + 2)>>2);
2568
    dc2= 0x01010101*((dc2 + 2)>>2);
2569

    
2570
    for(i=0; i<4; i++){
2571
        ((uint32_t*)(src+i*stride))[0]= dc0;
2572
        ((uint32_t*)(src+i*stride))[1]= dc1;
2573
    }
2574
    for(i=4; i<8; i++){
2575
        ((uint32_t*)(src+i*stride))[0]= dc2;
2576
        ((uint32_t*)(src+i*stride))[1]= dc3;
2577
    }
2578
}
2579

    
2580
static void pred8x8_plane_c(uint8_t *src, int stride){
2581
  int j, k;
2582
  int a;
2583
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2584
  const uint8_t * const src0 = src+3-stride;
2585
  const uint8_t *src1 = src+4*stride-1;
2586
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2587
  int H = src0[1] - src0[-1];
2588
  int V = src1[0] - src2[ 0];
2589
  for(k=2; k<=4; ++k) {
2590
    src1 += stride; src2 -= stride;
2591
    H += k*(src0[k] - src0[-k]);
2592
    V += k*(src1[0] - src2[ 0]);
2593
  }
2594
  H = ( 17*H+16 ) >> 5;
2595
  V = ( 17*V+16 ) >> 5;
2596

    
2597
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2598
  for(j=8; j>0; --j) {
2599
    int b = a;
2600
    a += V;
2601
    src[0] = cm[ (b    ) >> 5 ];
2602
    src[1] = cm[ (b+  H) >> 5 ];
2603
    src[2] = cm[ (b+2*H) >> 5 ];
2604
    src[3] = cm[ (b+3*H) >> 5 ];
2605
    src[4] = cm[ (b+4*H) >> 5 ];
2606
    src[5] = cm[ (b+5*H) >> 5 ];
2607
    src[6] = cm[ (b+6*H) >> 5 ];
2608
    src[7] = cm[ (b+7*H) >> 5 ];
2609
    src += stride;
2610
  }
2611
}
2612

    
2613
#define SRC(x,y) src[(x)+(y)*stride]
2614
#define PL(y) \
2615
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2616
#define PREDICT_8x8_LOAD_LEFT \
2617
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2618
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2619
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2620
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2621

    
2622
#define PT(x) \
2623
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2624
#define PREDICT_8x8_LOAD_TOP \
2625
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2626
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2627
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2628
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2629
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2630

    
2631
#define PTR(x) \
2632
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2633
#define PREDICT_8x8_LOAD_TOPRIGHT \
2634
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2635
    if(has_topright) { \
2636
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2637
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2638
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2639

    
2640
#define PREDICT_8x8_LOAD_TOPLEFT \
2641
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2642

    
2643
#define PREDICT_8x8_DC(v) \
2644
    int y; \
2645
    for( y = 0; y < 8; y++ ) { \
2646
        ((uint32_t*)src)[0] = \
2647
        ((uint32_t*)src)[1] = v; \
2648
        src += stride; \
2649
    }
2650

    
2651
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2652
{
2653
    PREDICT_8x8_DC(0x80808080);
2654
}
2655
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2656
{
2657
    PREDICT_8x8_LOAD_LEFT;
2658
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2659
    PREDICT_8x8_DC(dc);
2660
}
2661
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2662
{
2663
    PREDICT_8x8_LOAD_TOP;
2664
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2665
    PREDICT_8x8_DC(dc);
2666
}
2667
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2668
{
2669
    PREDICT_8x8_LOAD_LEFT;
2670
    PREDICT_8x8_LOAD_TOP;
2671
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2672
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2673
    PREDICT_8x8_DC(dc);
2674
}
2675
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2676
{
2677
    PREDICT_8x8_LOAD_LEFT;
2678
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2679
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2680
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2681
#undef ROW
2682
}
2683
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2684
{
2685
    int y;
2686
    PREDICT_8x8_LOAD_TOP;
2687
    src[0] = t0;
2688
    src[1] = t1;
2689
    src[2] = t2;
2690
    src[3] = t3;
2691
    src[4] = t4;
2692
    src[5] = t5;
2693
    src[6] = t6;
2694
    src[7] = t7;
2695
    for( y = 1; y < 8; y++ )
2696
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2697
}
2698
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2699
{
2700
    PREDICT_8x8_LOAD_TOP;
2701
    PREDICT_8x8_LOAD_TOPRIGHT;
2702
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2703
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2704
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2705
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2706
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2707
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2708
    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;
2709
    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;
2710
    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;
2711
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2712
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2713
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2714
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2715
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2716
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2717
}
2718
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2719
{
2720
    PREDICT_8x8_LOAD_TOP;
2721
    PREDICT_8x8_LOAD_LEFT;
2722
    PREDICT_8x8_LOAD_TOPLEFT;
2723
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2724
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2725
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2726
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2727
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2728
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2729
    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;
2730
    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;
2731
    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;
2732
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2733
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2734
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2735
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2736
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2737
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2738

    
2739
}
2740
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2741
{
2742
    PREDICT_8x8_LOAD_TOP;
2743
    PREDICT_8x8_LOAD_LEFT;
2744
    PREDICT_8x8_LOAD_TOPLEFT;
2745
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2746
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2747
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2748
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2749
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2750
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2751
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2752
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2753
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2754
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2755
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2756
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2757
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2758
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2759
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2760
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2761
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2762
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2763
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2764
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2765
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2766
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2767
}
2768
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2769
{
2770
    PREDICT_8x8_LOAD_TOP;
2771
    PREDICT_8x8_LOAD_LEFT;
2772
    PREDICT_8x8_LOAD_TOPLEFT;
2773
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2774
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2775
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2776
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2777
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2778
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2779
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2780
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2781
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2782
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2783
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2784
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2785
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2786
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2787
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2788
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2789
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2790
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2791
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2792
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2793
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2794
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2795
}
2796
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2797
{
2798
    PREDICT_8x8_LOAD_TOP;
2799
    PREDICT_8x8_LOAD_TOPRIGHT;
2800
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2801
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2802
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2803
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2804
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2805
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2806
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2807
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2808
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2809
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2810
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2811
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2812
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2813
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2814
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2815
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2816
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2817
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2818
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2819
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2820
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2821
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2822
}
2823
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2824
{
2825
    PREDICT_8x8_LOAD_LEFT;
2826
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2827
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2828
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2829
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2830
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2831
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2832
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2833
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2834
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2835
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2836
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2837
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2838
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2839
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2840
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2841
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2842
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2843
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2844
}
2845
#undef PREDICT_8x8_LOAD_LEFT
2846
#undef PREDICT_8x8_LOAD_TOP
2847
#undef PREDICT_8x8_LOAD_TOPLEFT
2848
#undef PREDICT_8x8_LOAD_TOPRIGHT
2849
#undef PREDICT_8x8_DC
2850
#undef PTR
2851
#undef PT
2852
#undef PL
2853
#undef SRC
2854

    
2855
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2856
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2857
                           int src_x_offset, int src_y_offset,
2858
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2859
    MpegEncContext * const s = &h->s;
2860
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2861
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2862
    const int luma_xy= (mx&3) + ((my&3)<<2);
2863
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2864
    uint8_t * src_cb, * src_cr;
2865
    int extra_width= h->emu_edge_width;
2866
    int extra_height= h->emu_edge_height;
2867
    int emu=0;
2868
    const int full_mx= mx>>2;
2869
    const int full_my= my>>2;
2870
    const int pic_width  = 16*s->mb_width;
2871
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
2872

    
2873
    if(!pic->data[0])
2874
        return;
2875

    
2876
    if(mx&7) extra_width -= 3;
2877
    if(my&7) extra_height -= 3;
2878

    
2879
    if(   full_mx < 0-extra_width
2880
       || full_my < 0-extra_height
2881
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2882
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2883
        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);
2884
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2885
        emu=1;
2886
    }
2887

    
2888
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2889
    if(!square){
2890
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2891
    }
2892

    
2893
    if(s->flags&CODEC_FLAG_GRAY) return;
2894

    
2895
    if(MB_MBAFF){
2896
        // chroma offset when predicting from a field of opposite parity
2897
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2898
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2899
    }
2900
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2901
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2902

    
2903
    if(emu){
2904
        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);
2905
            src_cb= s->edge_emu_buffer;
2906
    }
2907
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2908

    
2909
    if(emu){
2910
        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);
2911
            src_cr= s->edge_emu_buffer;
2912
    }
2913
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2914
}
2915

    
2916
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2917
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2918
                           int x_offset, int y_offset,
2919
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2920
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2921
                           int list0, int list1){
2922
    MpegEncContext * const s = &h->s;
2923
    qpel_mc_func *qpix_op=  qpix_put;
2924
    h264_chroma_mc_func chroma_op= chroma_put;
2925

    
2926
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2927
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2928
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2929
    x_offset += 8*s->mb_x;
2930
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2931

    
2932
    if(list0){
2933
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2934
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2935
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2936
                           qpix_op, chroma_op);
2937

    
2938
        qpix_op=  qpix_avg;
2939
        chroma_op= chroma_avg;
2940
    }
2941

    
2942
    if(list1){
2943
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2944
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2945
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2946
                           qpix_op, chroma_op);
2947
    }
2948
}
2949

    
2950
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2951
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2952
                           int x_offset, int y_offset,
2953
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2954
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2955
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2956
                           int list0, int list1){
2957
    MpegEncContext * const s = &h->s;
2958

    
2959
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2960
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2961
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2962
    x_offset += 8*s->mb_x;
2963
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2964

    
2965
    if(list0 && list1){
2966
        /* don't optimize for luma-only case, since B-frames usually
2967
         * use implicit weights => chroma too. */
2968
        uint8_t *tmp_cb = s->obmc_scratchpad;
2969
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2970
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2971
        int refn0 = h->ref_cache[0][ scan8[n] ];
2972
        int refn1 = h->ref_cache[1][ scan8[n] ];
2973

    
2974
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2975
                    dest_y, dest_cb, dest_cr,
2976
                    x_offset, y_offset, qpix_put, chroma_put);
2977
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2978
                    tmp_y, tmp_cb, tmp_cr,
2979
                    x_offset, y_offset, qpix_put, chroma_put);
2980

    
2981
        if(h->use_weight == 2){
2982
            int weight0 = h->implicit_weight[refn0][refn1];
2983
            int weight1 = 64 - weight0;
2984
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
2985
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2986
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2987
        }else{
2988
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2989
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2990
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2991
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2992
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2993
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2994
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2995
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2996
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2997
        }
2998
    }else{
2999
        int list = list1 ? 1 : 0;
3000
        int refn = h->ref_cache[list][ scan8[n] ];
3001
        Picture *ref= &h->ref_list[list][refn];
3002
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
3003
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
3004
                    qpix_put, chroma_put);
3005

    
3006
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
3007
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
3008
        if(h->use_weight_chroma){
3009
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3010
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
3011
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3012
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
3013
        }
3014
    }
3015
}
3016

    
3017
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
3018
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3019
                           int x_offset, int y_offset,
3020
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
3021
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
3022
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
3023
                           int list0, int list1){
3024
    if((h->use_weight==2 && list0 && list1
3025
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
3026
       || h->use_weight==1)
3027
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
3028
                         x_offset, y_offset, qpix_put, chroma_put,
3029
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
3030
    else
3031
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
3032
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
3033
}
3034

    
3035
static inline void prefetch_motion(H264Context *h, int list){
3036
    /* fetch pixels for estimated mv 4 macroblocks ahead
3037
     * optimized for 64byte cache lines */
3038
    MpegEncContext * const s = &h->s;
3039
    const int refn = h->ref_cache[list][scan8[0]];
3040
    if(refn >= 0){
3041
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
3042
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
3043
        uint8_t **src= h->ref_list[list][refn].data;
3044
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
3045
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
3046
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
3047
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
3048
    }
3049
}
3050

    
3051
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3052
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
3053
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
3054
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
3055
    MpegEncContext * const s = &h->s;
3056
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3057
    const int mb_type= s->current_picture.mb_type[mb_xy];
3058

    
3059
    assert(IS_INTER(mb_type));
3060

    
3061
    prefetch_motion(h, 0);
3062

    
3063
    if(IS_16X16(mb_type)){
3064
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
3065
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
3066
                &weight_op[0], &weight_avg[0],
3067
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3068
    }else if(IS_16X8(mb_type)){
3069
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
3070
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
3071
                &weight_op[1], &weight_avg[1],
3072
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3073
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
3074
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
3075
                &weight_op[1], &weight_avg[1],
3076
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
3077
    }else if(IS_8X16(mb_type)){
3078
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
3079
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3080
                &weight_op[2], &weight_avg[2],
3081
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3082
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
3083
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3084
                &weight_op[2], &weight_avg[2],
3085
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
3086
    }else{
3087
        int i;
3088

    
3089
        assert(IS_8X8(mb_type));
3090

    
3091
        for(i=0; i<4; i++){
3092
            const int sub_mb_type= h->sub_mb_type[i];
3093
            const int n= 4*i;
3094
            int x_offset= (i&1)<<2;
3095
            int y_offset= (i&2)<<1;
3096

    
3097
            if(IS_SUB_8X8(sub_mb_type)){
3098
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3099
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3100
                    &weight_op[3], &weight_avg[3],
3101
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3102
            }else if(IS_SUB_8X4(sub_mb_type)){
3103
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3104
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
3105
                    &weight_op[4], &weight_avg[4],
3106
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3107
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
3108
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
3109
                    &weight_op[4], &weight_avg[4],
3110
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3111
            }else if(IS_SUB_4X8(sub_mb_type)){
3112
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3113
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3114
                    &weight_op[5], &weight_avg[5],
3115
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3116
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
3117
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3118
                    &weight_op[5], &weight_avg[5],
3119
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3120
            }else{
3121
                int j;
3122
                assert(IS_SUB_4X4(sub_mb_type));
3123
                for(j=0; j<4; j++){
3124
                    int sub_x_offset= x_offset + 2*(j&1);
3125
                    int sub_y_offset= y_offset +   (j&2);
3126
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
3127
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3128
                        &weight_op[6], &weight_avg[6],
3129
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3130
                }
3131
            }
3132
        }
3133
    }
3134

    
3135
    prefetch_motion(h, 1);
3136
}
3137

    
3138
static void decode_init_vlc(){
3139
    static int done = 0;
3140

    
3141
    if (!done) {
3142
        int i;
3143
        done = 1;
3144

    
3145
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
3146
                 &chroma_dc_coeff_token_len [0], 1, 1,
3147
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
3148

    
3149
        for(i=0; i<4; i++){
3150
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
3151
                     &coeff_token_len [i][0], 1, 1,
3152
                     &coeff_token_bits[i][0], 1, 1, 1);
3153
        }
3154

    
3155
        for(i=0; i<3; i++){
3156
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
3157
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
3158
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
3159
        }
3160
        for(i=0; i<15; i++){
3161
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
3162
                     &total_zeros_len [i][0], 1, 1,
3163
                     &total_zeros_bits[i][0], 1, 1, 1);
3164
        }
3165

    
3166
        for(i=0; i<6; i++){
3167
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
3168
                     &run_len [i][0], 1, 1,
3169
                     &run_bits[i][0], 1, 1, 1);
3170
        }
3171
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
3172
                 &run_len [6][0], 1, 1,
3173
                 &run_bits[6][0], 1, 1, 1);
3174
    }
3175
}
3176

    
3177
/**
3178
 * Sets the intra prediction function pointers.
3179
 */
3180
static void init_pred_ptrs(H264Context *h){
3181
//    MpegEncContext * const s = &h->s;
3182

    
3183
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
3184
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
3185
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
3186
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
3187
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
3188
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
3189
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
3190
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
3191
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
3192
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
3193
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
3194
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
3195

    
3196
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
3197
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
3198
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
3199
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
3200
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
3201
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
3202
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
3203
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
3204
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
3205
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
3206
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
3207
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
3208

    
3209
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
3210
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
3211
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
3212
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
3213
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
3214
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
3215
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
3216

    
3217
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
3218
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
3219
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
3220
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
3221
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
3222
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
3223
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
3224
}
3225

    
3226
static void free_tables(H264Context *h){
3227
    av_freep(&h->intra4x4_pred_mode);
3228
    av_freep(&h->chroma_pred_mode_table);
3229
    av_freep(&h->cbp_table);
3230
    av_freep(&h->mvd_table[0]);
3231
    av_freep(&h->mvd_table[1]);
3232
    av_freep(&h->direct_table);
3233
    av_freep(&h->non_zero_count);
3234
    av_freep(&h->slice_table_base);
3235
    av_freep(&h->top_borders[1]);
3236
    av_freep(&h->top_borders[0]);
3237
    h->slice_table= NULL;
3238

    
3239
    av_freep(&h->mb2b_xy);
3240
    av_freep(&h->mb2b8_xy);
3241

    
3242
    av_freep(&h->s.obmc_scratchpad);
3243
}
3244

    
3245
static void init_dequant8_coeff_table(H264Context *h){
3246
    int i,q,x;
3247
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
3248
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
3249
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
3250

    
3251
    for(i=0; i<2; i++ ){
3252
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
3253
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
3254
            break;
3255
        }
3256

    
3257
        for(q=0; q<52; q++){
3258
            int shift = div6[q];
3259
            int idx = rem6[q];
3260
            for(x=0; x<64; x++)
3261
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
3262
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
3263
                    h->pps.scaling_matrix8[i][x]) << shift;
3264
        }
3265
    }
3266
}
3267

    
3268
static void init_dequant4_coeff_table(H264Context *h){
3269
    int i,j,q,x;
3270
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
3271
    for(i=0; i<6; i++ ){
3272
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
3273
        for(j=0; j<i; j++){
3274
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
3275
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
3276
                break;
3277
            }
3278
        }
3279
        if(j<i)
3280
            continue;
3281

    
3282
        for(q=0; q<52; q++){
3283
            int shift = div6[q] + 2;
3284
            int idx = rem6[q];
3285
            for(x=0; x<16; x++)
3286
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
3287
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
3288
                    h->pps.scaling_matrix4[i][x]) << shift;
3289
        }
3290
    }
3291
}
3292

    
3293
static void init_dequant_tables(H264Context *h){
3294
    int i,x;
3295
    init_dequant4_coeff_table(h);
3296
    if(h->pps.transform_8x8_mode)
3297
        init_dequant8_coeff_table(h);
3298
    if(h->sps.transform_bypass){
3299
        for(i=0; i<6; i++)
3300
            for(x=0; x<16; x++)
3301
                h->dequant4_coeff[i][0][x] = 1<<6;
3302
        if(h->pps.transform_8x8_mode)
3303
            for(i=0; i<2; i++)
3304
                for(x=0; x<64; x++)
3305
                    h->dequant8_coeff[i][0][x] = 1<<6;
3306
    }
3307
}
3308

    
3309

    
3310
/**
3311
 * allocates tables.
3312
 * needs width/height
3313
 */
3314
static int alloc_tables(H264Context *h){
3315
    MpegEncContext * const s = &h->s;
3316
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
3317
    int x,y;
3318

    
3319
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
3320

    
3321
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
3322
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
3323
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3324
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3325
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
3326

    
3327
    if( h->pps.cabac ) {
3328
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
3329
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3330
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
3331
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
3332
    }
3333

    
3334
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
3335
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
3336

    
3337
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
3338
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
3339
    for(y=0; y<s->mb_height; y++){
3340
        for(x=0; x<s->mb_width; x++){
3341
            const int mb_xy= x + y*s->mb_stride;
3342
            const int b_xy = 4*x + 4*y*h->b_stride;
3343
            const int b8_xy= 2*x + 2*y*h->b8_stride;
3344

    
3345
            h->mb2b_xy [mb_xy]= b_xy;
3346
            h->mb2b8_xy[mb_xy]= b8_xy;
3347
        }
3348
    }
3349

    
3350
    s->obmc_scratchpad = NULL;
3351

    
3352
    if(!h->dequant4_coeff[0])
3353
        init_dequant_tables(h);
3354

    
3355
    return 0;
3356
fail:
3357
    free_tables(h);
3358
    return -1;
3359
}
3360

    
3361
static void common_init(H264Context *h){
3362
    MpegEncContext * const s = &h->s;
3363

    
3364
    s->width = s->avctx->width;
3365
    s->height = s->avctx->height;
3366
    s->codec_id= s->avctx->codec->id;
3367

    
3368
    init_pred_ptrs(h);
3369

    
3370
    h->dequant_coeff_pps= -1;
3371
    s->unrestricted_mv=1;
3372
    s->decode=1; //FIXME
3373

    
3374
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3375
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
3376
}
3377

    
3378
static int decode_init(AVCodecContext *avctx){
3379
    H264Context *h= avctx->priv_data;
3380
    MpegEncContext * const s = &h->s;
3381

    
3382
    MPV_decode_defaults(s);
3383

    
3384
    s->avctx = avctx;
3385
    common_init(h);
3386

    
3387
    s->out_format = FMT_H264;
3388
    s->workaround_bugs= avctx->workaround_bugs;
3389

    
3390
    // set defaults
3391
//    s->decode_mb= ff_h263_decode_mb;
3392
    s->low_delay= 1;
3393
    avctx->pix_fmt= PIX_FMT_YUV420P;
3394

    
3395
    decode_init_vlc();
3396

    
3397
    if(avctx->extradata_size > 0 && avctx->extradata &&
3398
       *(char *)avctx->extradata == 1){
3399
        h->is_avc = 1;
3400
        h->got_avcC = 0;
3401
    } else {
3402
        h->is_avc = 0;
3403
    }
3404

    
3405
    return 0;
3406
}
3407

    
3408
static int frame_start(H264Context *h){
3409
    MpegEncContext * const s = &h->s;
3410
    int i;
3411

    
3412
    if(MPV_frame_start(s, s->avctx) < 0)
3413
        return -1;
3414
    ff_er_frame_start(s);
3415

    
3416
    assert(s->linesize && s->uvlinesize);
3417

    
3418
    for(i=0; i<16; i++){
3419
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3420
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3421
    }
3422
    for(i=0; i<4; i++){
3423
        h->block_offset[16+i]=
3424
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3425
        h->block_offset[24+16+i]=
3426
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3427
    }
3428

    
3429
    /* can't be in alloc_tables because linesize isn't known there.
3430
     * FIXME: redo bipred weight to not require extra buffer? */
3431
    if(!s->obmc_scratchpad)
3432
        s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
3433

    
3434
    /* some macroblocks will be accessed before they're available */
3435
    if(FRAME_MBAFF)
3436
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
3437

    
3438
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3439
    return 0;
3440
}
3441

    
3442
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3443
    MpegEncContext * const s = &h->s;
3444
    int i;
3445

    
3446
    src_y  -=   linesize;
3447
    src_cb -= uvlinesize;
3448
    src_cr -= uvlinesize;
3449

    
3450
    // There are two lines saved, the line above the the top macroblock of a pair,
3451
    // and the line above the bottom macroblock
3452
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3453
    for(i=1; i<17; i++){
3454
        h->left_border[i]= src_y[15+i*  linesize];
3455
    }
3456

    
3457
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3458
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3459

    
3460
    if(!(s->flags&CODEC_FLAG_GRAY)){
3461
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3462
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3463
        for(i=1; i<9; i++){
3464
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3465
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3466
        }
3467
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3468
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3469
    }
3470
}
3471

    
3472
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){
3473
    MpegEncContext * const s = &h->s;
3474
    int temp8, i;
3475
    uint64_t temp64;
3476
    int deblock_left = (s->mb_x > 0);
3477
    int deblock_top  = (s->mb_y > 0);
3478

    
3479
    src_y  -=   linesize + 1;
3480
    src_cb -= uvlinesize + 1;
3481
    src_cr -= uvlinesize + 1;
3482

    
3483
#define XCHG(a,b,t,xchg)\
3484
t= a;\
3485
if(xchg)\
3486
    a= b;\
3487
b= t;
3488

    
3489
    if(deblock_left){
3490
        for(i = !deblock_top; i<17; i++){
3491
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3492
        }
3493
    }
3494

    
3495
    if(deblock_top){
3496
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3497
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3498
        if(s->mb_x+1 < s->mb_width){
3499
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3500
        }
3501
    }
3502

    
3503
    if(!(s->flags&CODEC_FLAG_GRAY)){
3504
        if(deblock_left){
3505
            for(i = !deblock_top; i<9; i++){
3506
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3507
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3508
            }
3509
        }
3510
        if(deblock_top){
3511
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3512
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3513
        }
3514
    }
3515
}
3516

    
3517
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3518
    MpegEncContext * const s = &h->s;
3519
    int i;
3520

    
3521
    src_y  -= 2 *   linesize;
3522
    src_cb -= 2 * uvlinesize;
3523
    src_cr -= 2 * uvlinesize;
3524

    
3525
    // There are two lines saved, the line above the the top macroblock of a pair,
3526
    // and the line above the bottom macroblock
3527
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3528
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3529
    for(i=2; i<34; i++){
3530
        h->left_border[i]= src_y[15+i*  linesize];
3531
    }
3532

    
3533
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3534
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3535
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3536
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3537

    
3538
    if(!(s->flags&CODEC_FLAG_GRAY)){
3539
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3540
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3541
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3542
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3543
        for(i=2; i<18; i++){
3544
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3545
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3546
        }
3547
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3548
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3549
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3550
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3551
    }
3552
}
3553

    
3554
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){
3555
    MpegEncContext * const s = &h->s;
3556
    int temp8, i;
3557
    uint64_t temp64;
3558
    int deblock_left = (s->mb_x > 0);
3559
    int deblock_top  = (s->mb_y > 1);
3560

    
3561
    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);
3562

    
3563
    src_y  -= 2 *   linesize + 1;
3564
    src_cb -= 2 * uvlinesize + 1;
3565
    src_cr -= 2 * uvlinesize + 1;
3566

    
3567
#define XCHG(a,b,t,xchg)\
3568
t= a;\
3569
if(xchg)\
3570
    a= b;\
3571
b= t;
3572

    
3573
    if(deblock_left){
3574
        for(i = (!deblock_top)<<1; i<34; i++){
3575
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3576
        }
3577
    }
3578

    
3579
    if(deblock_top){
3580
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3581
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3582
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3583
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3584
        if(s->mb_x+1 < s->mb_width){
3585
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3586
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3587
        }
3588
    }
3589

    
3590
    if(!(s->flags&CODEC_FLAG_GRAY)){
3591
        if(deblock_left){
3592
            for(i = (!deblock_top) << 1; i<18; i++){
3593
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3594
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3595
            }
3596
        }
3597
        if(deblock_top){
3598
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3599
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3600
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3601
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3602
        }
3603
    }
3604
}
3605

    
3606
static void hl_decode_mb(H264Context *h){
3607
    MpegEncContext * const s = &h->s;
3608
    const int mb_x= s->mb_x;
3609
    const int mb_y= s->mb_y;
3610
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3611
    const int mb_type= s->current_picture.mb_type[mb_xy];
3612
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3613
    int linesize, uvlinesize /*dct_offset*/;
3614
    int i;
3615
    int *block_offset = &h->block_offset[0];
3616
    const unsigned int bottom = mb_y & 1;
3617
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3618
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3619
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3620

    
3621
    if(!s->decode)
3622
        return;
3623

    
3624
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3625
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3626
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3627

    
3628
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
3629
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
3630

    
3631
    if (MB_FIELD) {
3632
        linesize   = h->mb_linesize   = s->linesize * 2;
3633
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3634
        block_offset = &h->block_offset[24];
3635
        if(mb_y&1){ //FIXME move out of this func?
3636
            dest_y -= s->linesize*15;
3637
            dest_cb-= s->uvlinesize*7;
3638
            dest_cr-= s->uvlinesize*7;
3639
        }
3640
        if(FRAME_MBAFF) {
3641
            int list;
3642
            for(list=0; list<2; list++){
3643
                if(!USES_LIST(mb_type, list))
3644
                    continue;
3645
                if(IS_16X16(mb_type)){
3646
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
3647
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3648
                }else{
3649
                    for(i=0; i<16; i+=4){
3650
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3651
                        int ref = h->ref_cache[list][scan8[i]];
3652
                        if(ref >= 0)
3653
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3654
                    }
3655
                }
3656
            }
3657
        }
3658
    } else {
3659
        linesize   = h->mb_linesize   = s->linesize;
3660
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3661
//        dct_offset = s->linesize * 16;
3662
    }
3663

    
3664
    if(transform_bypass){
3665
        idct_dc_add =
3666
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3667
    }else if(IS_8x8DCT(mb_type)){
3668
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3669
        idct_add = s->dsp.h264_idct8_add;
3670
    }else{
3671
        idct_dc_add = s->dsp.h264_idct_dc_add;
3672
        idct_add = s->dsp.h264_idct_add;
3673
    }
3674

    
3675
    if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3676
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3677
        int mbt_y = mb_y&~1;
3678
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
3679
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3680
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3681
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3682
    }
3683

    
3684
    if (IS_INTRA_PCM(mb_type)) {
3685
        unsigned int x, y;
3686

    
3687
        // The pixels are stored in h->mb array in the same order as levels,
3688
        // copy them in output in the correct order.
3689
        for(i=0; i<16; i++) {
3690
            for (y=0; y<4; y++) {
3691
                for (x=0; x<4; x++) {
3692
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3693
                }
3694
            }
3695
        }
3696
        for(i=16; i<16+4; i++) {
3697
            for (y=0; y<4; y++) {
3698
                for (x=0; x<4; x++) {
3699
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3700
                }
3701
            }
3702
        }
3703
        for(i=20; i<20+4; i++) {
3704
            for (y=0; y<4; y++) {
3705
                for (x=0; x<4; x++) {
3706
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3707
                }
3708
            }
3709
        }
3710
    } else {
3711
        if(IS_INTRA(mb_type)){
3712
            if(h->deblocking_filter && !FRAME_MBAFF)
3713
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3714

    
3715
            if(!(s->flags&CODEC_FLAG_GRAY)){
3716
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3717
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3718
            }
3719

    
3720
            if(IS_INTRA4x4(mb_type)){
3721
                if(!s->encoding){
3722
                    if(IS_8x8DCT(mb_type)){
3723
                        for(i=0; i<16; i+=4){
3724
                            uint8_t * const ptr= dest_y + block_offset[i];
3725
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3726
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3727
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3728
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3729
                            if(nnz){
3730
                                if(nnz == 1 && h->mb[i*16])
3731
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3732
                                else
3733
                                    idct_add(ptr, h->mb + i*16, linesize);
3734
                            }
3735
                        }
3736
                    }else
3737
                    for(i=0; i<16; i++){
3738
                        uint8_t * const ptr= dest_y + block_offset[i];
3739
                        uint8_t *topright;
3740
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3741
                        int nnz, tr;
3742

    
3743
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3744
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3745
                            assert(mb_y || linesize <= block_offset[i]);
3746
                            if(!topright_avail){
3747
                                tr= ptr[3 - linesize]*0x01010101;
3748
                                topright= (uint8_t*) &tr;
3749
                            }else
3750
                                topright= ptr + 4 - linesize;
3751
                        }else
3752
                            topright= NULL;
3753

    
3754
                        h->pred4x4[ dir ](ptr, topright, linesize);
3755
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3756
                        if(nnz){
3757
                            if(s->codec_id == CODEC_ID_H264){
3758
                                if(nnz == 1 && h->mb[i*16])
3759
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3760
                                else
3761
                                    idct_add(ptr, h->mb + i*16, linesize);
3762
                            }else
3763
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3764
                        }
3765
                    }
3766
                }
3767
            }else{
3768
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3769
                if(s->codec_id == CODEC_ID_H264){
3770
                    if(!transform_bypass)
3771
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3772
                }else
3773
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3774
            }
3775
            if(h->deblocking_filter && !FRAME_MBAFF)
3776
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3777
        }else if(s->codec_id == CODEC_ID_H264){
3778
            hl_motion(h, dest_y, dest_cb, dest_cr,
3779
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
3780
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
3781
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3782
        }
3783

    
3784

    
3785
        if(!IS_INTRA4x4(mb_type)){
3786
            if(s->codec_id == CODEC_ID_H264){
3787
                if(IS_INTRA16x16(mb_type)){
3788
                    for(i=0; i<16; i++){
3789
                        if(h->non_zero_count_cache[ scan8[i] ])
3790
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3791
                        else if(h->mb[i*16])
3792
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3793
                    }
3794
                }else{
3795
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3796
                    for(i=0; i<16; i+=di){
3797
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3798
                        if(nnz){
3799
                            if(nnz==1 && h->mb[i*16])
3800
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3801
                            else
3802
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3803
                        }
3804
                    }
3805
                }
3806
            }else{
3807
                for(i=0; i<16; i++){
3808
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3809
                        uint8_t * const ptr= dest_y + block_offset[i];
3810
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3811
                    }
3812
                }
3813
            }
3814
        }
3815

    
3816
        if(!(s->flags&CODEC_FLAG_GRAY)){
3817
            uint8_t *dest[2] = {dest_cb, dest_cr};
3818
            if(transform_bypass){
3819
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3820
            }else{
3821
                idct_add = s->dsp.h264_idct_add;
3822
                idct_dc_add = s->dsp.h264_idct_dc_add;
3823
                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]);
3824
                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]);
3825
            }
3826
            if(s->codec_id == CODEC_ID_H264){
3827
                for(i=16; i<16+8; i++){
3828
                    if(h->non_zero_count_cache[ scan8[i] ])
3829
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3830
                    else if(h->mb[i*16])
3831
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3832
                }
3833
            }else{
3834
                for(i=16; i<16+8; i++){
3835
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3836
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3837
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3838
                    }
3839
                }
3840
            }
3841
        }
3842
    }
3843
    if(h->deblocking_filter) {
3844
        if (FRAME_MBAFF) {
3845
            //FIXME try deblocking one mb at a time?
3846
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3847
            const int mb_y = s->mb_y - 1;
3848
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3849
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3850
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3851
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3852
            if (!bottom) return;
3853
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3854
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3855
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3856

    
3857
            if(IS_INTRA(mb_type_top | mb_type_bottom))
3858
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3859

    
3860
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3861
            // deblock a pair
3862
            // top
3863
            s->mb_y--;
3864
            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);
3865
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3866
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
3867
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3868
            // bottom
3869
            s->mb_y++;
3870
            tprintf("call mbaff filter_mb\n");
3871
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3872
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3873
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3874
        } else {
3875
            tprintf("call filter_mb\n");
3876
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3877
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3878
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3879
        }
3880
    }
3881
}
3882

    
3883
/**
3884
 * fills the default_ref_list.
3885
 */
3886
static int fill_default_ref_list(H264Context *h){
3887
    MpegEncContext * const s = &h->s;
3888
    int i;
3889
    int smallest_poc_greater_than_current = -1;
3890
    Picture sorted_short_ref[32];
3891

    
3892
    if(h->slice_type==B_TYPE){
3893
        int out_i;
3894
        int limit= INT_MIN;
3895

    
3896
        /* sort frame according to poc in B slice */
3897
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3898
            int best_i=INT_MIN;
3899
            int best_poc=INT_MAX;
3900

    
3901
            for(i=0; i<h->short_ref_count; i++){
3902
                const int poc= h->short_ref[i]->poc;
3903
                if(poc > limit && poc < best_poc){
3904
                    best_poc= poc;
3905
                    best_i= i;
3906
                }
3907
            }
3908

    
3909
            assert(best_i != INT_MIN);
3910

    
3911
            limit= best_poc;
3912
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3913
            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);
3914
            if (-1 == smallest_poc_greater_than_current) {
3915
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3916
                    smallest_poc_greater_than_current = out_i;
3917
                }
3918
            }
3919
        }
3920
    }
3921

    
3922
    if(s->picture_structure == PICT_FRAME){
3923
        if(h->slice_type==B_TYPE){
3924
            int list;
3925
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3926

    
3927
            // find the largest poc
3928
            for(list=0; list<2; list++){
3929
                int index = 0;
3930
                int j= -99;
3931
                int step= list ? -1 : 1;
3932

    
3933
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3934
                    while(j<0 || j>= h->short_ref_count){
3935
                        if(j != -99 && step == (list ? -1 : 1))
3936
                            return -1;
3937
                        step = -step;
3938
                        j= smallest_poc_greater_than_current + (step>>1);
3939
                    }
3940
                    if(sorted_short_ref[j].reference != 3) continue;
3941
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3942
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3943
                }
3944

    
3945
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3946
                    if(h->long_ref[i] == NULL) continue;
3947
                    if(h->long_ref[i]->reference != 3) continue;
3948

    
3949
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3950
                    h->default_ref_list[ list ][index++].pic_id= i;;
3951
                }
3952

    
3953
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3954
                    // swap the two first elements of L1 when
3955
                    // L0 and L1 are identical
3956
                    Picture temp= h->default_ref_list[1][0];
3957
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3958
                    h->default_ref_list[1][1] = temp;
3959
                }
3960

    
3961
                if(index < h->ref_count[ list ])
3962
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3963
            }
3964
        }else{
3965
            int index=0;
3966
            for(i=0; i<h->short_ref_count; i++){
3967
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3968
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3969
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3970
            }
3971
            for(i = 0; i < 16; i++){
3972
                if(h->long_ref[i] == NULL) continue;
3973
                if(h->long_ref[i]->reference != 3) continue;
3974
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3975
                h->default_ref_list[0][index++].pic_id= i;;
3976
            }
3977
            if(index < h->ref_count[0])
3978
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3979
        }
3980
    }else{ //FIELD
3981
        if(h->slice_type==B_TYPE){
3982
        }else{
3983
            //FIXME second field balh
3984
        }
3985
    }
3986
#ifdef TRACE
3987
    for (i=0; i<h->ref_count[0]; i++) {
3988
        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]);
3989
    }
3990
    if(h->slice_type==B_TYPE){
3991
        for (i=0; i<h->ref_count[1]; i++) {
3992
            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]);
3993
        }
3994
    }
3995
#endif
3996
    return 0;
3997
}
3998

    
3999
static void print_short_term(H264Context *h);
4000
static void print_long_term(H264Context *h);
4001

    
4002
static int decode_ref_pic_list_reordering(H264Context *h){
4003
    MpegEncContext * const s = &h->s;
4004
    int list, index;
4005

    
4006
    print_short_term(h);
4007
    print_long_term(h);
4008
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
4009

    
4010
    for(list=0; list<2; list++){
4011
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
4012

    
4013
        if(get_bits1(&s->gb)){
4014
            int pred= h->curr_pic_num;
4015

    
4016
            for(index=0; ; index++){
4017
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
4018
                int pic_id;
4019
                int i;
4020
                Picture *ref = NULL;
4021

    
4022
                if(reordering_of_pic_nums_idc==3)
4023
                    break;
4024

    
4025
                if(index >= h->ref_count[list]){
4026
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
4027
                    return -1;
4028
                }
4029

    
4030
                if(reordering_of_pic_nums_idc<3){
4031
                    if(reordering_of_pic_nums_idc<2){
4032
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
4033

    
4034
                        if(abs_diff_pic_num >= h->max_pic_num){
4035
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
4036
                            return -1;
4037
                        }
4038

    
4039
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
4040
                        else                                pred+= abs_diff_pic_num;
4041
                        pred &= h->max_pic_num - 1;
4042

    
4043
                        for(i= h->short_ref_count-1; i>=0; i--){
4044
                            ref = h->short_ref[i];
4045
                            assert(ref->reference == 3);
4046
                            assert(!ref->long_ref);
4047
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
4048
                                break;
4049
                        }
4050
                        if(i>=0)
4051
                            ref->pic_id= ref->frame_num;
4052
                    }else{
4053
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
4054
                        ref = h->long_ref[pic_id];
4055
                        ref->pic_id= pic_id;
4056
                        assert(ref->reference == 3);
4057
                        assert(ref->long_ref);
4058
                        i=0;
4059
                    }
4060

    
4061
                    if (i < 0) {
4062
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
4063
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
4064
                    } else {
4065
                        for(i=index; i+1<h->ref_count[list]; i++){
4066
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
4067
                                break;
4068
                        }
4069
                        for(; i > index; i--){
4070
                            h->ref_list[list][i]= h->ref_list[list][i-1];
4071
                        }
4072
                        h->ref_list[list][index]= *ref;
4073
                    }
4074
                }else{
4075
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
4076
                    return -1;
4077
                }
4078
            }
4079
        }
4080

    
4081
        if(h->slice_type!=B_TYPE) break;
4082
    }
4083
    for(list=0; list<2; list++){
4084
        for(index= 0; index < h->ref_count[list]; index++){
4085
            if(!h->ref_list[list][index].data[0])
4086
                h->ref_list[list][index]= s->current_picture;
4087
        }
4088
        if(h->slice_type!=B_TYPE) break;
4089
    }
4090

    
4091
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
4092
        direct_dist_scale_factor(h);
4093
    direct_ref_list_init(h);
4094
    return 0;
4095
}
4096

    
4097
static void fill_mbaff_ref_list(H264Context *h){
4098
    int list, i, j;
4099
    for(list=0; list<2; list++){
4100
        for(i=0; i<h->ref_count[list]; i++){
4101
            Picture *frame = &h->ref_list[list][i];
4102
            Picture *field = &h->ref_list[list][16+2*i];
4103
            field[0] = *frame;
4104
            for(j=0; j<3; j++)
4105
                field[0].linesize[j] <<= 1;
4106
            field[1] = field[0];
4107
            for(j=0; j<3; j++)
4108
                field[1].data[j] += frame->linesize[j];
4109

    
4110
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
4111
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
4112
            for(j=0; j<2; j++){
4113
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
4114
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
4115
            }
4116
        }
4117
    }
4118
    for(j=0; j<h->ref_count[1]; j++){
4119
        for(i=0; i<h->ref_count[0]; i++)
4120
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
4121
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
4122
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
4123
    }
4124
}
4125

    
4126
static int pred_weight_table(H264Context *h){
4127
    MpegEncContext * const s = &h->s;
4128
    int list, i;
4129
    int luma_def, chroma_def;
4130

    
4131
    h->use_weight= 0;
4132
    h->use_weight_chroma= 0;
4133
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
4134
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
4135
    luma_def = 1<<h->luma_log2_weight_denom;
4136
    chroma_def = 1<<h->chroma_log2_weight_denom;
4137

    
4138
    for(list=0; list<2; list++){
4139
        for(i=0; i<h->ref_count[list]; i++){
4140
            int luma_weight_flag, chroma_weight_flag;
4141

    
4142
            luma_weight_flag= get_bits1(&s->gb);
4143
            if(luma_weight_flag){
4144
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
4145
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
4146
                if(   h->luma_weight[list][i] != luma_def
4147
                   || h->luma_offset[list][i] != 0)
4148
                    h->use_weight= 1;
4149
            }else{
4150
                h->luma_weight[list][i]= luma_def;
4151
                h->luma_offset[list][i]= 0;
4152
            }
4153

    
4154
            chroma_weight_flag= get_bits1(&s->gb);
4155
            if(chroma_weight_flag){
4156
                int j;
4157
                for(j=0; j<2; j++){
4158
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
4159
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
4160
                    if(   h->chroma_weight[list][i][j] != chroma_def
4161
                       || h->chroma_offset[list][i][j] != 0)
4162
                        h->use_weight_chroma= 1;
4163
                }
4164
            }else{
4165
                int j;
4166
                for(j=0; j<2; j++){
4167
                    h->chroma_weight[list][i][j]= chroma_def;
4168
                    h->chroma_offset[list][i][j]= 0;
4169
                }
4170
            }
4171
        }
4172
        if(h->slice_type != B_TYPE) break;
4173
    }
4174
    h->use_weight= h->use_weight || h->use_weight_chroma;
4175
    return 0;
4176
}
4177

    
4178
static void implicit_weight_table(H264Context *h){
4179
    MpegEncContext * const s = &h->s;
4180
    int ref0, ref1;
4181
    int cur_poc = s->current_picture_ptr->poc;
4182

    
4183
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
4184
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
4185
        h->use_weight= 0;
4186
        h->use_weight_chroma= 0;
4187
        return;
4188
    }
4189

    
4190
    h->use_weight= 2;
4191
    h->use_weight_chroma= 2;
4192
    h->luma_log2_weight_denom= 5;
4193
    h->chroma_log2_weight_denom= 5;
4194

    
4195
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
4196
        int poc0 = h->ref_list[0][ref0].poc;
4197
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
4198
            int poc1 = h->ref_list[1][ref1].poc;
4199
            int td = clip(poc1 - poc0, -128, 127);
4200
            if(td){
4201
                int tb = clip(cur_poc - poc0, -128, 127);
4202
                int tx = (16384 + (FFABS(td) >> 1)) / td;
4203
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
4204
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
4205
                    h->implicit_weight[ref0][ref1] = 32;
4206
                else
4207
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
4208
            }else
4209
                h->implicit_weight[ref0][ref1] = 32;
4210
        }
4211
    }
4212
}
4213

    
4214
static inline void unreference_pic(H264Context *h, Picture *pic){
4215
    int i;
4216
    pic->reference=0;
4217
    if(pic == h->delayed_output_pic)
4218
        pic->reference=1;
4219
    else{
4220
        for(i = 0; h->delayed_pic[i]; i++)
4221
            if(pic == h->delayed_pic[i]){
4222
                pic->reference=1;
4223
                break;
4224
            }
4225
    }
4226
}
4227

    
4228
/**
4229
 * instantaneous decoder refresh.
4230
 */
4231
static void idr(H264Context *h){
4232
    int i;
4233

    
4234
    for(i=0; i<16; i++){
4235
        if (h->long_ref[i] != NULL) {
4236
            unreference_pic(h, h->long_ref[i]);
4237
            h->long_ref[i]= NULL;
4238
        }
4239
    }
4240
    h->long_ref_count=0;
4241

    
4242
    for(i=0; i<h->short_ref_count; i++){
4243
        unreference_pic(h, h->short_ref[i]);
4244
        h->short_ref[i]= NULL;
4245
    }
4246
    h->short_ref_count=0;
4247
}
4248

    
4249
/* forget old pics after a seek */
4250
static void flush_dpb(AVCodecContext *avctx){
4251
    H264Context *h= avctx->priv_data;
4252
    int i;
4253
    for(i=0; i<16; i++) {
4254
        if(h->delayed_pic[i])
4255
            h->delayed_pic[i]->reference= 0;
4256
        h->delayed_pic[i]= NULL;
4257
    }
4258
    if(h->delayed_output_pic)
4259
        h->delayed_output_pic->reference= 0;
4260
    h->delayed_output_pic= NULL;
4261
    idr(h);
4262
    if(h->s.current_picture_ptr)
4263
        h->s.current_picture_ptr->reference= 0;
4264
}
4265

    
4266
/**
4267
 *
4268
 * @return the removed picture or NULL if an error occurs
4269
 */
4270
static Picture * remove_short(H264Context *h, int frame_num){
4271
    MpegEncContext * const s = &h->s;
4272
    int i;
4273

    
4274
    if(s->avctx->debug&FF_DEBUG_MMCO)
4275
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
4276

    
4277
    for(i=0; i<h->short_ref_count; i++){
4278
        Picture *pic= h->short_ref[i];
4279
        if(s->avctx->debug&FF_DEBUG_MMCO)
4280
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
4281
        if(pic->frame_num == frame_num){