Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ b566bd65

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
#define NAL_SLICE                1
169
#define NAL_DPA                  2
170
#define NAL_DPB                  3
171
#define NAL_DPC                  4
172
#define NAL_IDR_SLICE            5
173
#define NAL_SEI                  6
174
#define NAL_SPS                  7
175
#define NAL_PPS                  8
176
#define NAL_AUD                  9
177
#define NAL_END_SEQUENCE        10
178
#define NAL_END_STREAM          11
179
#define NAL_FILLER_DATA         12
180
#define NAL_SPS_EXT             13
181
#define NAL_AUXILIARY_SLICE     19
182
    uint8_t *rbsp_buffer;
183
    unsigned int rbsp_buffer_size;
184

    
185
    /**
186
      * Used to parse AVC variant of h264
187
      */
188
    int is_avc; ///< this flag is != 0 if codec is avc1
189
    int got_avcC; ///< flag used to parse avcC data only once
190
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
191

    
192
    int chroma_qp; //QPc
193

    
194
    int prev_mb_skipped;
195
    int next_mb_skipped;
196

    
197
    //prediction stuff
198
    int chroma_pred_mode;
199
    int intra16x16_pred_mode;
200

    
201
    int top_mb_xy;
202
    int left_mb_xy[2];
203

    
204
    int8_t intra4x4_pred_mode_cache[5*8];
205
    int8_t (*intra4x4_pred_mode)[8];
206
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
207
    void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
208
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
209
    void (*pred16x16[4+3])(uint8_t *src, int stride);
210
    unsigned int topleft_samples_available;
211
    unsigned int top_samples_available;
212
    unsigned int topright_samples_available;
213
    unsigned int left_samples_available;
214
    uint8_t (*top_borders[2])[16+2*8];
215
    uint8_t left_border[2*(17+2*9)];
216

    
217
    /**
218
     * non zero coeff count cache.
219
     * is 64 if not available.
220
     */
221
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
222
    uint8_t (*non_zero_count)[16];
223

    
224
    /**
225
     * Motion vector cache.
226
     */
227
    DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
228
    DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
229
#define LIST_NOT_USED -1 //FIXME rename?
230
#define PART_NOT_AVAILABLE -2
231

    
232
    /**
233
     * is 1 if the specific list MV&references are set to 0,0,-2.
234
     */
235
    int mv_cache_clean[2];
236

    
237
    /**
238
     * number of neighbors (top and/or left) that used 8x8 dct
239
     */
240
    int neighbor_transform_size;
241

    
242
    /**
243
     * block_offset[ 0..23] for frame macroblocks
244
     * block_offset[24..47] for field macroblocks
245
     */
246
    int block_offset[2*(16+8)];
247

    
248
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
249
    uint32_t *mb2b8_xy;
250
    int b_stride; //FIXME use s->b4_stride
251
    int b8_stride;
252

    
253
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
254
    int mb_uvlinesize;
255

    
256
    int emu_edge_width;
257
    int emu_edge_height;
258

    
259
    int halfpel_flag;
260
    int thirdpel_flag;
261

    
262
    int unknown_svq3_flag;
263
    int next_slice_index;
264

    
265
    SPS sps_buffer[MAX_SPS_COUNT];
266
    SPS sps; ///< current sps
267

    
268
    PPS pps_buffer[MAX_PPS_COUNT];
269
    /**
270
     * current pps
271
     */
272
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
273

    
274
    uint32_t dequant4_buffer[6][52][16];
275
    uint32_t dequant8_buffer[2][52][64];
276
    uint32_t (*dequant4_coeff[6])[16];
277
    uint32_t (*dequant8_coeff[2])[64];
278
    int dequant_coeff_pps;     ///< reinit tables when pps changes
279

    
280
    int slice_num;
281
    uint8_t *slice_table_base;
282
    uint8_t *slice_table;      ///< slice_table_base + 2*mb_stride + 1
283
    int slice_type;
284
    int slice_type_fixed;
285

    
286
    //interlacing specific flags
287
    int mb_aff_frame;
288
    int mb_field_decoding_flag;
289
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
290

    
291
    int sub_mb_type[4];
292

    
293
    //POC stuff
294
    int poc_lsb;
295
    int poc_msb;
296
    int delta_poc_bottom;
297
    int delta_poc[2];
298
    int frame_num;
299
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
300
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
301
    int frame_num_offset;         ///< for POC type 2
302
    int prev_frame_num_offset;    ///< for POC type 2
303
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
304

    
305
    /**
306
     * frame_num for frames or 2*frame_num for field pics.
307
     */
308
    int curr_pic_num;
309

    
310
    /**
311
     * max_frame_num or 2*max_frame_num for field pics.
312
     */
313
    int max_pic_num;
314

    
315
    //Weighted pred stuff
316
    int use_weight;
317
    int use_weight_chroma;
318
    int luma_log2_weight_denom;
319
    int chroma_log2_weight_denom;
320
    int luma_weight[2][48];
321
    int luma_offset[2][48];
322
    int chroma_weight[2][48][2];
323
    int chroma_offset[2][48][2];
324
    int implicit_weight[48][48];
325

    
326
    //deblock
327
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
328
    int slice_alpha_c0_offset;
329
    int slice_beta_offset;
330

    
331
    int redundant_pic_count;
332

    
333
    int direct_spatial_mv_pred;
334
    int dist_scale_factor[16];
335
    int dist_scale_factor_field[32];
336
    int map_col_to_list0[2][16];
337
    int map_col_to_list0_field[2][32];
338

    
339
    /**
340
     * num_ref_idx_l0/1_active_minus1 + 1
341
     */
342
    int ref_count[2];            ///< counts frames or fields, depending on current mb mode
343
    Picture *short_ref[32];
344
    Picture *long_ref[32];
345
    Picture default_ref_list[2][32];
346
    Picture ref_list[2][48];     ///< 0..15: frame refs, 16..47: mbaff field refs
347
    Picture *delayed_pic[16]; //FIXME size?
348
    Picture *delayed_output_pic;
349

    
350
    /**
351
     * memory management control operations buffer.
352
     */
353
    MMCO mmco[MAX_MMCO_COUNT];
354
    int mmco_index;
355

    
356
    int long_ref_count;  ///< number of actual long term references
357
    int short_ref_count; ///< number of actual short term references
358

    
359
    //data partitioning
360
    GetBitContext intra_gb;
361
    GetBitContext inter_gb;
362
    GetBitContext *intra_gb_ptr;
363
    GetBitContext *inter_gb_ptr;
364

    
365
    DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);
366

    
367
    /**
368
     * Cabac
369
     */
370
    CABACContext cabac;
371
    uint8_t      cabac_state[460];
372
    int          cabac_init_idc;
373

    
374
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
375
    uint16_t     *cbp_table;
376
    int cbp;
377
    int top_cbp;
378
    int left_cbp;
379
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
380
    uint8_t     *chroma_pred_mode_table;
381
    int         last_qscale_diff;
382
    int16_t     (*mvd_table[2])[2];
383
    DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
384
    uint8_t     *direct_table;
385
    uint8_t     direct_cache[5*8];
386

    
387
    uint8_t zigzag_scan[16];
388
    uint8_t zigzag_scan8x8[64];
389
    uint8_t zigzag_scan8x8_cavlc[64];
390
    uint8_t field_scan[16];
391
    uint8_t field_scan8x8[64];
392
    uint8_t field_scan8x8_cavlc[64];
393
    const uint8_t *zigzag_scan_q0;
394
    const uint8_t *zigzag_scan8x8_q0;
395
    const uint8_t *zigzag_scan8x8_cavlc_q0;
396
    const uint8_t *field_scan_q0;
397
    const uint8_t *field_scan8x8_q0;
398
    const uint8_t *field_scan8x8_cavlc_q0;
399

    
400
    int x264_build;
401
}H264Context;
402

    
403
static VLC coeff_token_vlc[4];
404
static VLC chroma_dc_coeff_token_vlc;
405

    
406
static VLC total_zeros_vlc[15];
407
static VLC chroma_dc_total_zeros_vlc[3];
408

    
409
static VLC run_vlc[6];
410
static VLC run7_vlc;
411

    
412
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
413
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
414
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);
415
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);
416

    
417
static always_inline uint32_t pack16to32(int a, int b){
418
#ifdef WORDS_BIGENDIAN
419
   return (b&0xFFFF) + (a<<16);
420
#else
421
   return (a&0xFFFF) + (b<<16);
422
#endif
423
}
424

    
425
/**
426
 * fill a rectangle.
427
 * @param h height of the rectangle, should be a constant
428
 * @param w width of the rectangle, should be a constant
429
 * @param size the size of val (1 or 4), should be a constant
430
 */
431
static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
432
    uint8_t *p= (uint8_t*)vp;
433
    assert(size==1 || size==4);
434
    assert(w<=4);
435

    
436
    w      *= size;
437
    stride *= size;
438

    
439
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
440
    assert((stride&(w-1))==0);
441
    if(w==2){
442
        const uint16_t v= size==4 ? val : val*0x0101;
443
        *(uint16_t*)(p + 0*stride)= v;
444
        if(h==1) return;
445
        *(uint16_t*)(p + 1*stride)= v;
446
        if(h==2) return;
447
        *(uint16_t*)(p + 2*stride)=
448
        *(uint16_t*)(p + 3*stride)= v;
449
    }else if(w==4){
450
        const uint32_t v= size==4 ? val : val*0x01010101;
451
        *(uint32_t*)(p + 0*stride)= v;
452
        if(h==1) return;
453
        *(uint32_t*)(p + 1*stride)= v;
454
        if(h==2) return;
455
        *(uint32_t*)(p + 2*stride)=
456
        *(uint32_t*)(p + 3*stride)= v;
457
    }else if(w==8){
458
    //gcc can't optimize 64bit math on x86_32
459
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
460
        const uint64_t v= val*0x0100000001ULL;
461
        *(uint64_t*)(p + 0*stride)= v;
462
        if(h==1) return;
463
        *(uint64_t*)(p + 1*stride)= v;
464
        if(h==2) return;
465
        *(uint64_t*)(p + 2*stride)=
466
        *(uint64_t*)(p + 3*stride)= v;
467
    }else if(w==16){
468
        const uint64_t v= val*0x0100000001ULL;
469
        *(uint64_t*)(p + 0+0*stride)=
470
        *(uint64_t*)(p + 8+0*stride)=
471
        *(uint64_t*)(p + 0+1*stride)=
472
        *(uint64_t*)(p + 8+1*stride)= v;
473
        if(h==2) return;
474
        *(uint64_t*)(p + 0+2*stride)=
475
        *(uint64_t*)(p + 8+2*stride)=
476
        *(uint64_t*)(p + 0+3*stride)=
477
        *(uint64_t*)(p + 8+3*stride)= v;
478
#else
479
        *(uint32_t*)(p + 0+0*stride)=
480
        *(uint32_t*)(p + 4+0*stride)= val;
481
        if(h==1) return;
482
        *(uint32_t*)(p + 0+1*stride)=
483
        *(uint32_t*)(p + 4+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 + 0+3*stride)=
488
        *(uint32_t*)(p + 4+3*stride)= val;
489
    }else if(w==16){
490
        *(uint32_t*)(p + 0+0*stride)=
491
        *(uint32_t*)(p + 4+0*stride)=
492
        *(uint32_t*)(p + 8+0*stride)=
493
        *(uint32_t*)(p +12+0*stride)=
494
        *(uint32_t*)(p + 0+1*stride)=
495
        *(uint32_t*)(p + 4+1*stride)=
496
        *(uint32_t*)(p + 8+1*stride)=
497
        *(uint32_t*)(p +12+1*stride)= val;
498
        if(h==2) return;
499
        *(uint32_t*)(p + 0+2*stride)=
500
        *(uint32_t*)(p + 4+2*stride)=
501
        *(uint32_t*)(p + 8+2*stride)=
502
        *(uint32_t*)(p +12+2*stride)=
503
        *(uint32_t*)(p + 0+3*stride)=
504
        *(uint32_t*)(p + 4+3*stride)=
505
        *(uint32_t*)(p + 8+3*stride)=
506
        *(uint32_t*)(p +12+3*stride)= val;
507
#endif
508
    }else
509
        assert(0);
510
    assert(h==4);
511
}
512

    
513
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
514
    MpegEncContext * const s = &h->s;
515
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
516
    int topleft_xy, top_xy, topright_xy, left_xy[2];
517
    int topleft_type, top_type, topright_type, left_type[2];
518
    int left_block[8];
519
    int i;
520

    
521
    //FIXME deblocking could skip the intra and nnz parts.
522
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
523
        return;
524

    
525
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
526

    
527
    top_xy     = mb_xy  - s->mb_stride;
528
    topleft_xy = top_xy - 1;
529
    topright_xy= top_xy + 1;
530
    left_xy[1] = left_xy[0] = mb_xy-1;
531
    left_block[0]= 0;
532
    left_block[1]= 1;
533
    left_block[2]= 2;
534
    left_block[3]= 3;
535
    left_block[4]= 7;
536
    left_block[5]= 10;
537
    left_block[6]= 8;
538
    left_block[7]= 11;
539
    if(FRAME_MBAFF){
540
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
541
        const int top_pair_xy      = pair_xy     - s->mb_stride;
542
        const int topleft_pair_xy  = top_pair_xy - 1;
543
        const int topright_pair_xy = top_pair_xy + 1;
544
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
545
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
546
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
547
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
548
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
549
        const int bottom = (s->mb_y & 1);
550
        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);
551
        if (bottom
552
                ? !curr_mb_frame_flag // bottom macroblock
553
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
554
                ) {
555
            top_xy -= s->mb_stride;
556
        }
557
        if (bottom
558
                ? !curr_mb_frame_flag // bottom macroblock
559
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
560
                ) {
561
            topleft_xy -= s->mb_stride;
562
        }
563
        if (bottom
564
                ? !curr_mb_frame_flag // bottom macroblock
565
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
566
                ) {
567
            topright_xy -= s->mb_stride;
568
        }
569
        if (left_mb_frame_flag != curr_mb_frame_flag) {
570
            left_xy[1] = left_xy[0] = pair_xy - 1;
571
            if (curr_mb_frame_flag) {
572
                if (bottom) {
573
                    left_block[0]= 2;
574
                    left_block[1]= 2;
575
                    left_block[2]= 3;
576
                    left_block[3]= 3;
577
                    left_block[4]= 8;
578
                    left_block[5]= 11;
579
                    left_block[6]= 8;
580
                    left_block[7]= 11;
581
                } else {
582
                    left_block[0]= 0;
583
                    left_block[1]= 0;
584
                    left_block[2]= 1;
585
                    left_block[3]= 1;
586
                    left_block[4]= 7;
587
                    left_block[5]= 10;
588
                    left_block[6]= 7;
589
                    left_block[7]= 10;
590
                }
591
            } else {
592
                left_xy[1] += s->mb_stride;
593
                //left_block[0]= 0;
594
                left_block[1]= 2;
595
                left_block[2]= 0;
596
                left_block[3]= 2;
597
                //left_block[4]= 7;
598
                left_block[5]= 10;
599
                left_block[6]= 7;
600
                left_block[7]= 10;
601
            }
602
        }
603
    }
604

    
605
    h->top_mb_xy = top_xy;
606
    h->left_mb_xy[0] = left_xy[0];
607
    h->left_mb_xy[1] = left_xy[1];
608
    if(for_deblock){
609
        topleft_type = 0;
610
        topright_type = 0;
611
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
612
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
613
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
614

    
615
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
616
            int list;
617
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
618
            for(i=0; i<16; i++)
619
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
620
            for(list=0; list<1+(h->slice_type==B_TYPE); list++){
621
                if(USES_LIST(mb_type,list)){
622
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
623
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
624
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
625
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
626
                        dst[0] = src[0];
627
                        dst[1] = src[1];
628
                        dst[2] = src[2];
629
                        dst[3] = src[3];
630
                    }
631
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
632
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
633
                    ref += h->b8_stride;
634
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
635
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
636
                }else{
637
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
638
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
639
                }
640
            }
641
        }
642
    }else{
643
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
644
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
645
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
646
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
647
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
648
    }
649

    
650
    if(IS_INTRA(mb_type)){
651
        h->topleft_samples_available=
652
        h->top_samples_available=
653
        h->left_samples_available= 0xFFFF;
654
        h->topright_samples_available= 0xEEEA;
655

    
656
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
657
            h->topleft_samples_available= 0xB3FF;
658
            h->top_samples_available= 0x33FF;
659
            h->topright_samples_available= 0x26EA;
660
        }
661
        for(i=0; i<2; i++){
662
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
663
                h->topleft_samples_available&= 0xDF5F;
664
                h->left_samples_available&= 0x5F5F;
665
            }
666
        }
667

    
668
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
669
            h->topleft_samples_available&= 0x7FFF;
670

    
671
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
672
            h->topright_samples_available&= 0xFBFF;
673

    
674
        if(IS_INTRA4x4(mb_type)){
675
            if(IS_INTRA4x4(top_type)){
676
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
677
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
678
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
679
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
680
            }else{
681
                int pred;
682
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
683
                    pred= -1;
684
                else{
685
                    pred= 2;
686
                }
687
                h->intra4x4_pred_mode_cache[4+8*0]=
688
                h->intra4x4_pred_mode_cache[5+8*0]=
689
                h->intra4x4_pred_mode_cache[6+8*0]=
690
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
691
            }
692
            for(i=0; i<2; i++){
693
                if(IS_INTRA4x4(left_type[i])){
694
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
695
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
696
                }else{
697
                    int pred;
698
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
699
                        pred= -1;
700
                    else{
701
                        pred= 2;
702
                    }
703
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
704
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
705
                }
706
            }
707
        }
708
    }
709

    
710

    
711
/*
712
0 . T T. T T T T
713
1 L . .L . . . .
714
2 L . .L . . . .
715
3 . T TL . . . .
716
4 L . .L . . . .
717
5 L . .. . . . .
718
*/
719
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
720
    if(top_type){
721
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
722
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
723
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
724
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
725

    
726
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
727
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
728

    
729
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
730
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
731

    
732
    }else{
733
        h->non_zero_count_cache[4+8*0]=
734
        h->non_zero_count_cache[5+8*0]=
735
        h->non_zero_count_cache[6+8*0]=
736
        h->non_zero_count_cache[7+8*0]=
737

    
738
        h->non_zero_count_cache[1+8*0]=
739
        h->non_zero_count_cache[2+8*0]=
740

    
741
        h->non_zero_count_cache[1+8*3]=
742
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
743

    
744
    }
745

    
746
    for (i=0; i<2; i++) {
747
        if(left_type[i]){
748
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
749
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
750
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
751
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
752
        }else{
753
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
754
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
755
            h->non_zero_count_cache[0+8*1 +   8*i]=
756
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
757
        }
758
    }
759

    
760
    if( h->pps.cabac ) {
761
        // top_cbp
762
        if(top_type) {
763
            h->top_cbp = h->cbp_table[top_xy];
764
        } else if(IS_INTRA(mb_type)) {
765
            h->top_cbp = 0x1C0;
766
        } else {
767
            h->top_cbp = 0;
768
        }
769
        // left_cbp
770
        if (left_type[0]) {
771
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
772
        } else if(IS_INTRA(mb_type)) {
773
            h->left_cbp = 0x1C0;
774
        } else {
775
            h->left_cbp = 0;
776
        }
777
        if (left_type[0]) {
778
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
779
        }
780
        if (left_type[1]) {
781
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
782
        }
783
    }
784

    
785
#if 1
786
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
787
        int list;
788
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
789
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
790
                /*if(!h->mv_cache_clean[list]){
791
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
792
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
793
                    h->mv_cache_clean[list]= 1;
794
                }*/
795
                continue;
796
            }
797
            h->mv_cache_clean[list]= 0;
798

    
799
            if(USES_LIST(top_type, list)){
800
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
801
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
802
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
803
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
804
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
805
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
806
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
807
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
808
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
809
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
810
            }else{
811
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
812
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
813
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
814
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
815
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
816
            }
817

    
818
            //FIXME unify cleanup or sth
819
            if(USES_LIST(left_type[0], list)){
820
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
821
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
822
                *(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]];
823
                *(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]];
824
                h->ref_cache[list][scan8[0] - 1 + 0*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
825
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1]>>1)];
826
            }else{
827
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
828
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
829
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
830
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
831
            }
832

    
833
            if(USES_LIST(left_type[1], list)){
834
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
835
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
836
                *(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]];
837
                *(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]];
838
                h->ref_cache[list][scan8[0] - 1 + 2*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
839
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[3]>>1)];
840
            }else{
841
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
842
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
843
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
844
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
845
                assert((!left_type[0]) == (!left_type[1]));
846
            }
847

    
848
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
849
                continue;
850

    
851
            if(USES_LIST(topleft_type, list)){
852
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
853
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
854
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
855
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
856
            }else{
857
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
858
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
859
            }
860

    
861
            if(USES_LIST(topright_type, list)){
862
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
863
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
864
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
865
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
866
            }else{
867
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
868
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
869
            }
870

    
871
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
872
                continue;
873

    
874
            h->ref_cache[list][scan8[5 ]+1] =
875
            h->ref_cache[list][scan8[7 ]+1] =
876
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
877
            h->ref_cache[list][scan8[4 ]] =
878
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
879
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
880
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
881
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
882
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
883
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
884

    
885
            if( h->pps.cabac ) {
886
                /* XXX beurk, Load mvd */
887
                if(USES_LIST(top_type, list)){
888
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
889
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
890
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
891
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
892
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
893
                }else{
894
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
895
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
896
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
897
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
898
                }
899
                if(USES_LIST(left_type[0], list)){
900
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
901
                    *(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]];
902
                    *(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]];
903
                }else{
904
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
905
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
906
                }
907
                if(USES_LIST(left_type[1], list)){
908
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
909
                    *(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]];
910
                    *(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]];
911
                }else{
912
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
913
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
914
                }
915
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
916
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
917
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
918
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
919
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
920

    
921
                if(h->slice_type == B_TYPE){
922
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
923

    
924
                    if(IS_DIRECT(top_type)){
925
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
926
                    }else if(IS_8X8(top_type)){
927
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
928
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
929
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
930
                    }else{
931
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
932
                    }
933

    
934
                    if(IS_DIRECT(left_type[0]))
935
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
936
                    else if(IS_8X8(left_type[0]))
937
                        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)];
938
                    else
939
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
940

    
941
                    if(IS_DIRECT(left_type[1]))
942
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
943
                    else if(IS_8X8(left_type[1]))
944
                        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)];
945
                    else
946
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
947
                }
948
            }
949

    
950
            if(FRAME_MBAFF){
951
#define MAP_MVS\
952
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
953
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
954
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
955
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
956
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
957
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
958
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
959
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
960
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
961
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
962
                if(MB_FIELD){
963
#define MAP_F2F(idx, mb_type)\
964
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
965
                        h->ref_cache[list][idx] <<= 1;\
966
                        h->mv_cache[list][idx][1] /= 2;\
967
                        h->mvd_cache[list][idx][1] /= 2;\
968
                    }
969
                    MAP_MVS
970
#undef MAP_F2F
971
                }else{
972
#define MAP_F2F(idx, mb_type)\
973
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
974
                        h->ref_cache[list][idx] >>= 1;\
975
                        h->mv_cache[list][idx][1] <<= 1;\
976
                        h->mvd_cache[list][idx][1] <<= 1;\
977
                    }
978
                    MAP_MVS
979
#undef MAP_F2F
980
                }
981
            }
982
        }
983
    }
984
#endif
985

    
986
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
987
}
988

    
989
static inline void write_back_intra_pred_mode(H264Context *h){
990
    MpegEncContext * const s = &h->s;
991
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
992

    
993
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
994
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
995
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
996
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
997
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
998
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
999
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
1000
}
1001

    
1002
/**
1003
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
1004
 */
1005
static inline int check_intra4x4_pred_mode(H264Context *h){
1006
    MpegEncContext * const s = &h->s;
1007
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
1008
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
1009
    int i;
1010

    
1011
    if(!(h->top_samples_available&0x8000)){
1012
        for(i=0; i<4; i++){
1013
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
1014
            if(status<0){
1015
                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);
1016
                return -1;
1017
            } else if(status){
1018
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
1019
            }
1020
        }
1021
    }
1022

    
1023
    if(!(h->left_samples_available&0x8000)){
1024
        for(i=0; i<4; i++){
1025
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
1026
            if(status<0){
1027
                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);
1028
                return -1;
1029
            } else if(status){
1030
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
1031
            }
1032
        }
1033
    }
1034

    
1035
    return 0;
1036
} //FIXME cleanup like next
1037

    
1038
/**
1039
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
1040
 */
1041
static inline int check_intra_pred_mode(H264Context *h, int mode){
1042
    MpegEncContext * const s = &h->s;
1043
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
1044
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
1045

    
1046
    if(mode < 0 || mode > 6) {
1047
        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);
1048
        return -1;
1049
    }
1050

    
1051
    if(!(h->top_samples_available&0x8000)){
1052
        mode= top[ mode ];
1053
        if(mode<0){
1054
            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);
1055
            return -1;
1056
        }
1057
    }
1058

    
1059
    if(!(h->left_samples_available&0x8000)){
1060
        mode= left[ mode ];
1061
        if(mode<0){
1062
            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);
1063
            return -1;
1064
        }
1065
    }
1066

    
1067
    return mode;
1068
}
1069

    
1070
/**
1071
 * gets the predicted intra4x4 prediction mode.
1072
 */
1073
static inline int pred_intra_mode(H264Context *h, int n){
1074
    const int index8= scan8[n];
1075
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1076
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1077
    const int min= FFMIN(left, top);
1078

    
1079
    tprintf("mode:%d %d min:%d\n", left ,top, min);
1080

    
1081
    if(min<0) return DC_PRED;
1082
    else      return min;
1083
}
1084

    
1085
static inline void write_back_non_zero_count(H264Context *h){
1086
    MpegEncContext * const s = &h->s;
1087
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1088

    
1089
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1090
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1091
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
1092
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
1093
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1094
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1095
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
1096

    
1097
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
1098
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
1099
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
1100

    
1101
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
1102
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
1103
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
1104

    
1105
    if(FRAME_MBAFF){
1106
        // store all luma nnzs, for deblocking
1107
        int v = 0, i;
1108
        for(i=0; i<16; i++)
1109
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
1110
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
1111
    }
1112
}
1113

    
1114
/**
1115
 * gets the predicted number of non zero coefficients.
1116
 * @param n block index
1117
 */
1118
static inline int pred_non_zero_count(H264Context *h, int n){
1119
    const int index8= scan8[n];
1120
    const int left= h->non_zero_count_cache[index8 - 1];
1121
    const int top = h->non_zero_count_cache[index8 - 8];
1122
    int i= left + top;
1123

    
1124
    if(i<64) i= (i+1)>>1;
1125

    
1126
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1127

    
1128
    return i&31;
1129
}
1130

    
1131
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1132
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1133

    
1134
    /* there is no consistent mapping of mvs to neighboring locations that will
1135
     * make mbaff happy, so we can't move all this logic to fill_caches */
1136
    if(FRAME_MBAFF){
1137
        MpegEncContext *s = &h->s;
1138
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
1139
        const int16_t *mv;
1140
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
1141
        *C = h->mv_cache[list][scan8[0]-2];
1142

    
1143
        if(!MB_FIELD
1144
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
1145
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
1146
            if(IS_INTERLACED(mb_types[topright_xy])){
1147
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
1148
                const int x4 = X4, y4 = Y4;\
1149
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
1150
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
1151
                    return LIST_NOT_USED;\
1152
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
1153
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
1154
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
1155
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
1156

    
1157
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
1158
            }
1159
        }
1160
        if(topright_ref == PART_NOT_AVAILABLE
1161
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
1162
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
1163
            if(!MB_FIELD
1164
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
1165
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
1166
            }
1167
            if(MB_FIELD
1168
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
1169
               && i >= scan8[0]+8){
1170
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
1171
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
1172
            }
1173
        }
1174
#undef SET_DIAG_MV
1175
    }
1176

    
1177
    if(topright_ref != PART_NOT_AVAILABLE){
1178
        *C= h->mv_cache[list][ i - 8 + part_width ];
1179
        return topright_ref;
1180
    }else{
1181
        tprintf("topright MV not available\n");
1182

    
1183
        *C= h->mv_cache[list][ i - 8 - 1 ];
1184
        return h->ref_cache[list][ i - 8 - 1 ];
1185
    }
1186
}
1187

    
1188
/**
1189
 * gets the predicted MV.
1190
 * @param n the block index
1191
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1192
 * @param mx the x component of the predicted motion vector
1193
 * @param my the y component of the predicted motion vector
1194
 */
1195
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1196
    const int index8= scan8[n];
1197
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1198
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1199
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1200
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1201
    const int16_t * C;
1202
    int diagonal_ref, match_count;
1203

    
1204
    assert(part_width==1 || part_width==2 || part_width==4);
1205

    
1206
/* mv_cache
1207
  B . . A T T T T
1208
  U . . L . . , .
1209
  U . . L . . . .
1210
  U . . L . . , .
1211
  . . . L . . . .
1212
*/
1213

    
1214
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1215
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1216
    tprintf("pred_motion match_count=%d\n", match_count);
1217
    if(match_count > 1){ //most common
1218
        *mx= mid_pred(A[0], B[0], C[0]);
1219
        *my= mid_pred(A[1], B[1], C[1]);
1220
    }else if(match_count==1){
1221
        if(left_ref==ref){
1222
            *mx= A[0];
1223
            *my= A[1];
1224
        }else if(top_ref==ref){
1225
            *mx= B[0];
1226
            *my= B[1];
1227
        }else{
1228
            *mx= C[0];
1229
            *my= C[1];
1230
        }
1231
    }else{
1232
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1233
            *mx= A[0];
1234
            *my= A[1];
1235
        }else{
1236
            *mx= mid_pred(A[0], B[0], C[0]);
1237
            *my= mid_pred(A[1], B[1], C[1]);
1238
        }
1239
    }
1240

    
1241
    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);
1242
}
1243

    
1244
/**
1245
 * gets the directionally predicted 16x8 MV.
1246
 * @param n the block index
1247
 * @param mx the x component of the predicted motion vector
1248
 * @param my the y component of the predicted motion vector
1249
 */
1250
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1251
    if(n==0){
1252
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1253
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1254

    
1255
        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);
1256

    
1257
        if(top_ref == ref){
1258
            *mx= B[0];
1259
            *my= B[1];
1260
            return;
1261
        }
1262
    }else{
1263
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1264
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1265

    
1266
        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);
1267

    
1268
        if(left_ref == ref){
1269
            *mx= A[0];
1270
            *my= A[1];
1271
            return;
1272
        }
1273
    }
1274

    
1275
    //RARE
1276
    pred_motion(h, n, 4, list, ref, mx, my);
1277
}
1278

    
1279
/**
1280
 * gets the directionally predicted 8x16 MV.
1281
 * @param n the block index
1282
 * @param mx the x component of the predicted motion vector
1283
 * @param my the y component of the predicted motion vector
1284
 */
1285
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1286
    if(n==0){
1287
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1288
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1289

    
1290
        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);
1291

    
1292
        if(left_ref == ref){
1293
            *mx= A[0];
1294
            *my= A[1];
1295
            return;
1296
        }
1297
    }else{
1298
        const int16_t * C;
1299
        int diagonal_ref;
1300

    
1301
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1302

    
1303
        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);
1304

    
1305
        if(diagonal_ref == ref){
1306
            *mx= C[0];
1307
            *my= C[1];
1308
            return;
1309
        }
1310
    }
1311

    
1312
    //RARE
1313
    pred_motion(h, n, 2, list, ref, mx, my);
1314
}
1315

    
1316
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1317
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1318
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1319

    
1320
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1321

    
1322
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1323
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1324
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1325

    
1326
        *mx = *my = 0;
1327
        return;
1328
    }
1329

    
1330
    pred_motion(h, 0, 4, 0, 0, mx, my);
1331

    
1332
    return;
1333
}
1334

    
1335
static inline void direct_dist_scale_factor(H264Context * const h){
1336
    const int poc = h->s.current_picture_ptr->poc;
1337
    const int poc1 = h->ref_list[1][0].poc;
1338
    int i;
1339
    for(i=0; i<h->ref_count[0]; i++){
1340
        int poc0 = h->ref_list[0][i].poc;
1341
        int td = clip(poc1 - poc0, -128, 127);
1342
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1343
            h->dist_scale_factor[i] = 256;
1344
        }else{
1345
            int tb = clip(poc - poc0, -128, 127);
1346
            int tx = (16384 + (FFABS(td) >> 1)) / td;
1347
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1348
        }
1349
    }
1350
    if(FRAME_MBAFF){
1351
        for(i=0; i<h->ref_count[0]; i++){
1352
            h->dist_scale_factor_field[2*i] =
1353
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
1354
        }
1355
    }
1356
}
1357
static inline void direct_ref_list_init(H264Context * const h){
1358
    MpegEncContext * const s = &h->s;
1359
    Picture * const ref1 = &h->ref_list[1][0];
1360
    Picture * const cur = s->current_picture_ptr;
1361
    int list, i, j;
1362
    if(cur->pict_type == I_TYPE)
1363
        cur->ref_count[0] = 0;
1364
    if(cur->pict_type != B_TYPE)
1365
        cur->ref_count[1] = 0;
1366
    for(list=0; list<2; list++){
1367
        cur->ref_count[list] = h->ref_count[list];
1368
        for(j=0; j<h->ref_count[list]; j++)
1369
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1370
    }
1371
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1372
        return;
1373
    for(list=0; list<2; list++){
1374
        for(i=0; i<ref1->ref_count[list]; i++){
1375
            const int poc = ref1->ref_poc[list][i];
1376
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1377
            for(j=0; j<h->ref_count[list]; j++)
1378
                if(h->ref_list[list][j].poc == poc){
1379
                    h->map_col_to_list0[list][i] = j;
1380
                    break;
1381
                }
1382
        }
1383
    }
1384
    if(FRAME_MBAFF){
1385
        for(list=0; list<2; list++){
1386
            for(i=0; i<ref1->ref_count[list]; i++){
1387
                j = h->map_col_to_list0[list][i];
1388
                h->map_col_to_list0_field[list][2*i] = 2*j;
1389
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1390
            }
1391
        }
1392
    }
1393
}
1394

    
1395
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1396
    MpegEncContext * const s = &h->s;
1397
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1398
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1399
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1400
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1401
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1402
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1403
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1404
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1405
    const int is_b8x8 = IS_8X8(*mb_type);
1406
    int sub_mb_type;
1407
    int i8, i4;
1408

    
1409
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1410
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1411
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1412
         * so we know exactly what block size to use */
1413
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1414
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1415
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1416
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1417
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1418
    }else{
1419
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1420
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1421
    }
1422
    if(!is_b8x8)
1423
        *mb_type |= MB_TYPE_DIRECT2;
1424
    if(MB_FIELD)
1425
        *mb_type |= MB_TYPE_INTERLACED;
1426

    
1427
    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);
1428

    
1429
    if(h->direct_spatial_mv_pred){
1430
        int ref[2];
1431
        int mv[2][2];
1432
        int list;
1433

    
1434
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1435

    
1436
        /* ref = min(neighbors) */
1437
        for(list=0; list<2; list++){
1438
            int refa = h->ref_cache[list][scan8[0] - 1];
1439
            int refb = h->ref_cache[list][scan8[0] - 8];
1440
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1441
            if(refc == -2)
1442
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1443
            ref[list] = refa;
1444
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1445
                ref[list] = refb;
1446
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1447
                ref[list] = refc;
1448
            if(ref[list] < 0)
1449
                ref[list] = -1;
1450
        }
1451

    
1452
        if(ref[0] < 0 && ref[1] < 0){
1453
            ref[0] = ref[1] = 0;
1454
            mv[0][0] = mv[0][1] =
1455
            mv[1][0] = mv[1][1] = 0;
1456
        }else{
1457
            for(list=0; list<2; list++){
1458
                if(ref[list] >= 0)
1459
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1460
                else
1461
                    mv[list][0] = mv[list][1] = 0;
1462
            }
1463
        }
1464

    
1465
        if(ref[1] < 0){
1466
            *mb_type &= ~MB_TYPE_P0L1;
1467
            sub_mb_type &= ~MB_TYPE_P0L1;
1468
        }else if(ref[0] < 0){
1469
            *mb_type &= ~MB_TYPE_P0L0;
1470
            sub_mb_type &= ~MB_TYPE_P0L0;
1471
        }
1472

    
1473
        if(IS_16X16(*mb_type)){
1474
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1475
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1476
            if(!IS_INTRA(mb_type_col)
1477
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1478
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1479
                       && (h->x264_build>33 || !h->x264_build)))){
1480
                if(ref[0] > 0)
1481
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1482
                else
1483
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1484
                if(ref[1] > 0)
1485
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1486
                else
1487
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1488
            }else{
1489
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1490
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1491
            }
1492
        }else{
1493
            for(i8=0; i8<4; i8++){
1494
                const int x8 = i8&1;
1495
                const int y8 = i8>>1;
1496

    
1497
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1498
                    continue;
1499
                h->sub_mb_type[i8] = sub_mb_type;
1500

    
1501
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1502
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1503
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1504
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1505

    
1506
                /* col_zero_flag */
1507
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1508
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1509
                                                  && (h->x264_build>33 || !h->x264_build)))){
1510
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1511
                    if(IS_SUB_8X8(sub_mb_type)){
1512
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1513
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1514
                            if(ref[0] == 0)
1515
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1516
                            if(ref[1] == 0)
1517
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1518
                        }
1519
                    }else
1520
                    for(i4=0; i4<4; i4++){
1521
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1522
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1523
                            if(ref[0] == 0)
1524
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1525
                            if(ref[1] == 0)
1526
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1527
                        }
1528
                    }
1529
                }
1530
            }
1531
        }
1532
    }else{ /* direct temporal mv pred */
1533
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1534
        const int *dist_scale_factor = h->dist_scale_factor;
1535

    
1536
        if(FRAME_MBAFF){
1537
            if(IS_INTERLACED(*mb_type)){
1538
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1539
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1540
                dist_scale_factor = h->dist_scale_factor_field;
1541
            }
1542
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1543
                /* FIXME assumes direct_8x8_inference == 1 */
1544
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1545
                int mb_types_col[2];
1546
                int y_shift;
1547

    
1548
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1549
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1550
                         | (*mb_type & MB_TYPE_INTERLACED);
1551
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1552

    
1553
                if(IS_INTERLACED(*mb_type)){
1554
                    /* frame to field scaling */
1555
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1556
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1557
                    if(s->mb_y&1){
1558
                        l1ref0 -= 2*h->b8_stride;
1559
                        l1ref1 -= 2*h->b8_stride;
1560
                        l1mv0 -= 4*h->b_stride;
1561
                        l1mv1 -= 4*h->b_stride;
1562
                    }
1563
                    y_shift = 0;
1564

    
1565
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1566
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1567
                       && !is_b8x8)
1568
                        *mb_type |= MB_TYPE_16x8;
1569
                    else
1570
                        *mb_type |= MB_TYPE_8x8;
1571
                }else{
1572
                    /* field to frame scaling */
1573
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1574
                     * but in MBAFF, top and bottom POC are equal */
1575
                    int dy = (s->mb_y&1) ? 1 : 2;
1576
                    mb_types_col[0] =
1577
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1578
                    l1ref0 += dy*h->b8_stride;
1579
                    l1ref1 += dy*h->b8_stride;
1580
                    l1mv0 += 2*dy*h->b_stride;
1581
                    l1mv1 += 2*dy*h->b_stride;
1582
                    y_shift = 2;
1583

    
1584
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1585
                       && !is_b8x8)
1586
                        *mb_type |= MB_TYPE_16x16;
1587
                    else
1588
                        *mb_type |= MB_TYPE_8x8;
1589
                }
1590

    
1591
                for(i8=0; i8<4; i8++){
1592
                    const int x8 = i8&1;
1593
                    const int y8 = i8>>1;
1594
                    int ref0, scale;
1595
                    const int16_t (*l1mv)[2]= l1mv0;
1596

    
1597
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1598
                        continue;
1599
                    h->sub_mb_type[i8] = sub_mb_type;
1600

    
1601
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1602
                    if(IS_INTRA(mb_types_col[y8])){
1603
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1604
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1605
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1606
                        continue;
1607
                    }
1608

    
1609
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1610
                    if(ref0 >= 0)
1611
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1612
                    else{
1613
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1614
                        l1mv= l1mv1;
1615
                    }
1616
                    scale = dist_scale_factor[ref0];
1617
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1618

    
1619
                    {
1620
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1621
                        int my_col = (mv_col[1]<<y_shift)/2;
1622
                        int mx = (scale * mv_col[0] + 128) >> 8;
1623
                        int my = (scale * my_col + 128) >> 8;
1624
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1625
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1626
                    }
1627
                }
1628
                return;
1629
            }
1630
        }
1631

    
1632
        /* one-to-one mv scaling */
1633

    
1634
        if(IS_16X16(*mb_type)){
1635
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1636
            if(IS_INTRA(mb_type_col)){
1637
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1638
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1639
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1640
            }else{
1641
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1642
                                                : map_col_to_list0[1][l1ref1[0]];
1643
                const int scale = dist_scale_factor[ref0];
1644
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1645
                int mv_l0[2];
1646
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1647
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1648
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1649
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1650
                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);
1651
            }
1652
        }else{
1653
            for(i8=0; i8<4; i8++){
1654
                const int x8 = i8&1;
1655
                const int y8 = i8>>1;
1656
                int ref0, scale;
1657
                const int16_t (*l1mv)[2]= l1mv0;
1658

    
1659
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1660
                    continue;
1661
                h->sub_mb_type[i8] = sub_mb_type;
1662
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1663
                if(IS_INTRA(mb_type_col)){
1664
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1665
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1666
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1667
                    continue;
1668
                }
1669

    
1670
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1671
                if(ref0 >= 0)
1672
                    ref0 = map_col_to_list0[0][ref0];
1673
                else{
1674
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1675
                    l1mv= l1mv1;
1676
                }
1677
                scale = dist_scale_factor[ref0];
1678

    
1679
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1680
                if(IS_SUB_8X8(sub_mb_type)){
1681
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1682
                    int mx = (scale * mv_col[0] + 128) >> 8;
1683
                    int my = (scale * mv_col[1] + 128) >> 8;
1684
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1685
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1686
                }else
1687
                for(i4=0; i4<4; i4++){
1688
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1689
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1690
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1691
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1692
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1693
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1694
                }
1695
            }
1696
        }
1697
    }
1698
}
1699

    
1700
static inline void write_back_motion(H264Context *h, int mb_type){
1701
    MpegEncContext * const s = &h->s;
1702
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1703
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1704
    int list;
1705

    
1706
    if(!USES_LIST(mb_type, 0))
1707
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1708

    
1709
    for(list=0; list<2; list++){
1710
        int y;
1711
        if(!USES_LIST(mb_type, list))
1712
            continue;
1713

    
1714
        for(y=0; y<4; y++){
1715
            *(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];
1716
            *(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];
1717
        }
1718
        if( h->pps.cabac ) {
1719
            if(IS_SKIP(mb_type))
1720
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1721
            else
1722
            for(y=0; y<4; y++){
1723
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1724
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1725
            }
1726
        }
1727

    
1728
        {
1729
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1730
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1731
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1732
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1733
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1734
        }
1735
    }
1736

    
1737
    if(h->slice_type == B_TYPE && h->pps.cabac){
1738
        if(IS_8X8(mb_type)){
1739
            uint8_t *direct_table = &h->direct_table[b8_xy];
1740
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1741
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1742
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1743
        }
1744
    }
1745
}
1746

    
1747
/**
1748
 * Decodes a network abstraction layer unit.
1749
 * @param consumed is the number of bytes used as input
1750
 * @param length is the length of the array
1751
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1752
 * @returns decoded bytes, might be src+1 if no escapes
1753
 */
1754
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1755
    int i, si, di;
1756
    uint8_t *dst;
1757

    
1758
//    src[0]&0x80;                //forbidden bit
1759
    h->nal_ref_idc= src[0]>>5;
1760
    h->nal_unit_type= src[0]&0x1F;
1761

    
1762
    src++; length--;
1763
#if 0
1764
    for(i=0; i<length; i++)
1765
        printf("%2X ", src[i]);
1766
#endif
1767
    for(i=0; i+1<length; i+=2){
1768
        if(src[i]) continue;
1769
        if(i>0 && src[i-1]==0) i--;
1770
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1771
            if(src[i+2]!=3){
1772
                /* startcode, so we must be past the end */
1773
                length=i;
1774
            }
1775
            break;
1776
        }
1777
    }
1778

    
1779
    if(i>=length-1){ //no escaped 0
1780
        *dst_length= length;
1781
        *consumed= length+1; //+1 for the header
1782
        return src;
1783
    }
1784

    
1785
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1786
    dst= h->rbsp_buffer;
1787

    
1788
//printf("decoding esc\n");
1789
    si=di=0;
1790
    while(si<length){
1791
        //remove escapes (very rare 1:2^22)
1792
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1793
            if(src[si+2]==3){ //escape
1794
                dst[di++]= 0;
1795
                dst[di++]= 0;
1796
                si+=3;
1797
                continue;
1798
            }else //next start code
1799
                break;
1800
        }
1801

    
1802
        dst[di++]= src[si++];
1803
    }
1804

    
1805
    *dst_length= di;
1806
    *consumed= si + 1;//+1 for the header
1807
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1808
    return dst;
1809
}
1810

    
1811
#if 0
1812
/**
1813
 * @param src the data which should be escaped
1814
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1815
 * @param length the length of the src data
1816
 * @param dst_length the length of the dst array
1817
 * @returns length of escaped data in bytes or -1 if an error occured
1818
 */
1819
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1820
    int i, escape_count, si, di;
1821
    uint8_t *temp;
1822

1823
    assert(length>=0);
1824
    assert(dst_length>0);
1825

1826
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1827

1828
    if(length==0) return 1;
1829

1830
    escape_count= 0;
1831
    for(i=0; i<length; i+=2){
1832
        if(src[i]) continue;
1833
        if(i>0 && src[i-1]==0)
1834
            i--;
1835
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1836
            escape_count++;
1837
            i+=2;
1838
        }
1839
    }
1840

1841
    if(escape_count==0){
1842
        if(dst+1 != src)
1843
            memcpy(dst+1, src, length);
1844
        return length + 1;
1845
    }
1846

1847
    if(length + escape_count + 1> dst_length)
1848
        return -1;
1849

1850
    //this should be damn rare (hopefully)
1851

1852
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1853
    temp= h->rbsp_buffer;
1854
//printf("encoding esc\n");
1855

1856
    si= 0;
1857
    di= 0;
1858
    while(si < length){
1859
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1860
            temp[di++]= 0; si++;
1861
            temp[di++]= 0; si++;
1862
            temp[di++]= 3;
1863
            temp[di++]= src[si++];
1864
        }
1865
        else
1866
            temp[di++]= src[si++];
1867
    }
1868
    memcpy(dst+1, temp, length+escape_count);
1869

1870
    assert(di == length+escape_count);
1871

1872
    return di + 1;
1873
}
1874

1875
/**
1876
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1877
 */
1878
static void encode_rbsp_trailing(PutBitContext *pb){
1879
    int length;
1880
    put_bits(pb, 1, 1);
1881
    length= (-put_bits_count(pb))&7;
1882
    if(length) put_bits(pb, length, 0);
1883
}
1884
#endif
1885

    
1886
/**
1887
 * identifies the exact end of the bitstream
1888
 * @return the length of the trailing, or 0 if damaged
1889
 */
1890
static int decode_rbsp_trailing(uint8_t *src){
1891
    int v= *src;
1892
    int r;
1893

    
1894
    tprintf("rbsp trailing %X\n", v);
1895

    
1896
    for(r=1; r<9; r++){
1897
        if(v&1) return r;
1898
        v>>=1;
1899
    }
1900
    return 0;
1901
}
1902

    
1903
/**
1904
 * idct tranforms the 16 dc values and dequantize them.
1905
 * @param qp quantization parameter
1906
 */
1907
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1908
#define stride 16
1909
    int i;
1910
    int temp[16]; //FIXME check if this is a good idea
1911
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1912
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1913

    
1914
//memset(block, 64, 2*256);
1915
//return;
1916
    for(i=0; i<4; i++){
1917
        const int offset= y_offset[i];
1918
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1919
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1920
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1921
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1922

    
1923
        temp[4*i+0]= z0+z3;
1924
        temp[4*i+1]= z1+z2;
1925
        temp[4*i+2]= z1-z2;
1926
        temp[4*i+3]= z0-z3;
1927
    }
1928

    
1929
    for(i=0; i<4; i++){
1930
        const int offset= x_offset[i];
1931
        const int z0= temp[4*0+i] + temp[4*2+i];
1932
        const int z1= temp[4*0+i] - temp[4*2+i];
1933
        const int z2= temp[4*1+i] - temp[4*3+i];
1934
        const int z3= temp[4*1+i] + temp[4*3+i];
1935

    
1936
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1937
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1938
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1939
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1940
    }
1941
}
1942

    
1943
#if 0
1944
/**
1945
 * dct tranforms the 16 dc values.
1946
 * @param qp quantization parameter ??? FIXME
1947
 */
1948
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1949
//    const int qmul= dequant_coeff[qp][0];
1950
    int i;
1951
    int temp[16]; //FIXME check if this is a good idea
1952
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1953
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1954

1955
    for(i=0; i<4; i++){
1956
        const int offset= y_offset[i];
1957
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1958
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1959
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1960
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1961

1962
        temp[4*i+0]= z0+z3;
1963
        temp[4*i+1]= z1+z2;
1964
        temp[4*i+2]= z1-z2;
1965
        temp[4*i+3]= z0-z3;
1966
    }
1967

1968
    for(i=0; i<4; i++){
1969
        const int offset= x_offset[i];
1970
        const int z0= temp[4*0+i] + temp[4*2+i];
1971
        const int z1= temp[4*0+i] - temp[4*2+i];
1972
        const int z2= temp[4*1+i] - temp[4*3+i];
1973
        const int z3= temp[4*1+i] + temp[4*3+i];
1974

1975
        block[stride*0 +offset]= (z0 + z3)>>1;
1976
        block[stride*2 +offset]= (z1 + z2)>>1;
1977
        block[stride*8 +offset]= (z1 - z2)>>1;
1978
        block[stride*10+offset]= (z0 - z3)>>1;
1979
    }
1980
}
1981
#endif
1982

    
1983
#undef xStride
1984
#undef stride
1985

    
1986
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1987
    const int stride= 16*2;
1988
    const int xStride= 16;
1989
    int a,b,c,d,e;
1990

    
1991
    a= block[stride*0 + xStride*0];
1992
    b= block[stride*0 + xStride*1];
1993
    c= block[stride*1 + xStride*0];
1994
    d= block[stride*1 + xStride*1];
1995

    
1996
    e= a-b;
1997
    a= a+b;
1998
    b= c-d;
1999
    c= c+d;
2000

    
2001
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
2002
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
2003
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
2004
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
2005
}
2006

    
2007
#if 0
2008
static void chroma_dc_dct_c(DCTELEM *block){
2009
    const int stride= 16*2;
2010
    const int xStride= 16;
2011
    int a,b,c,d,e;
2012

2013
    a= block[stride*0 + xStride*0];
2014
    b= block[stride*0 + xStride*1];
2015
    c= block[stride*1 + xStride*0];
2016
    d= block[stride*1 + xStride*1];
2017

2018
    e= a-b;
2019
    a= a+b;
2020
    b= c-d;
2021
    c= c+d;
2022

2023
    block[stride*0 + xStride*0]= (a+c);
2024
    block[stride*0 + xStride*1]= (e+b);
2025
    block[stride*1 + xStride*0]= (a-c);
2026
    block[stride*1 + xStride*1]= (e-b);
2027
}
2028
#endif
2029

    
2030
/**
2031
 * gets the chroma qp.
2032
 */
2033
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
2034

    
2035
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
2036
}
2037

    
2038

    
2039
#if 0
2040
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
2041
    int i;
2042
    //FIXME try int temp instead of block
2043

2044
    for(i=0; i<4; i++){
2045
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
2046
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
2047
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
2048
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
2049
        const int z0= d0 + d3;
2050
        const int z3= d0 - d3;
2051
        const int z1= d1 + d2;
2052
        const int z2= d1 - d2;
2053

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

2060
    for(i=0; i<4; i++){
2061
        const int z0= block[0*4 + i] + block[3*4 + i];
2062
        const int z3= block[0*4 + i] - block[3*4 + i];
2063
        const int z1= block[1*4 + i] + block[2*4 + i];
2064
        const int z2= block[1*4 + i] - block[2*4 + i];
2065

2066
        block[0*4 + i]=   z0 +   z1;
2067
        block[1*4 + i]= 2*z3 +   z2;
2068
        block[2*4 + i]=   z0 -   z1;
2069
        block[3*4 + i]=   z3 - 2*z2;
2070
    }
2071
}
2072
#endif
2073

    
2074
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
2075
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
2076
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
2077
    int i;
2078
    const int * const quant_table= quant_coeff[qscale];
2079
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
2080
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
2081
    const unsigned int threshold2= (threshold1<<1);
2082
    int last_non_zero;
2083

    
2084
    if(seperate_dc){
2085
        if(qscale<=18){
2086
            //avoid overflows
2087
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
2088
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
2089
            const unsigned int dc_threshold2= (dc_threshold1<<1);
2090

    
2091
            int level= block[0]*quant_coeff[qscale+18][0];
2092
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
2093
                if(level>0){
2094
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
2095
                    block[0]= level;
2096
                }else{
2097
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
2098
                    block[0]= -level;
2099
                }
2100
//                last_non_zero = i;
2101
            }else{
2102
                block[0]=0;
2103
            }
2104
        }else{
2105
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
2106
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
2107
            const unsigned int dc_threshold2= (dc_threshold1<<1);
2108

    
2109
            int level= block[0]*quant_table[0];
2110
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
2111
                if(level>0){
2112
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
2113
                    block[0]= level;
2114
                }else{
2115
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
2116
                    block[0]= -level;
2117
                }
2118
//                last_non_zero = i;
2119
            }else{
2120
                block[0]=0;
2121
            }
2122
        }
2123
        last_non_zero= 0;
2124
        i=1;
2125
    }else{
2126
        last_non_zero= -1;
2127
        i=0;
2128
    }
2129

    
2130
    for(; i<16; i++){
2131
        const int j= scantable[i];
2132
        int level= block[j]*quant_table[j];
2133

    
2134
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
2135
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
2136
        if(((unsigned)(level+threshold1))>threshold2){
2137
            if(level>0){
2138
                level= (bias + level)>>QUANT_SHIFT;
2139
                block[j]= level;
2140
            }else{
2141
                level= (bias - level)>>QUANT_SHIFT;
2142
                block[j]= -level;
2143
            }
2144
            last_non_zero = i;
2145
        }else{
2146
            block[j]=0;
2147
        }
2148
    }
2149

    
2150
    return last_non_zero;
2151
}
2152

    
2153
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
2154
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2155
    ((uint32_t*)(src+0*stride))[0]= a;
2156
    ((uint32_t*)(src+1*stride))[0]= a;
2157
    ((uint32_t*)(src+2*stride))[0]= a;
2158
    ((uint32_t*)(src+3*stride))[0]= a;
2159
}
2160

    
2161
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
2162
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
2163
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
2164
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
2165
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
2166
}
2167

    
2168
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
2169
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
2170
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
2171

    
2172
    ((uint32_t*)(src+0*stride))[0]=
2173
    ((uint32_t*)(src+1*stride))[0]=
2174
    ((uint32_t*)(src+2*stride))[0]=
2175
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2176
}
2177

    
2178
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
2179
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
2180

    
2181
    ((uint32_t*)(src+0*stride))[0]=
2182
    ((uint32_t*)(src+1*stride))[0]=
2183
    ((uint32_t*)(src+2*stride))[0]=
2184
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2185
}
2186

    
2187
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
2188
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
2189

    
2190
    ((uint32_t*)(src+0*stride))[0]=
2191
    ((uint32_t*)(src+1*stride))[0]=
2192
    ((uint32_t*)(src+2*stride))[0]=
2193
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
2194
}
2195

    
2196
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
2197
    ((uint32_t*)(src+0*stride))[0]=
2198
    ((uint32_t*)(src+1*stride))[0]=
2199
    ((uint32_t*)(src+2*stride))[0]=
2200
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
2201
}
2202

    
2203

    
2204
#define LOAD_TOP_RIGHT_EDGE\
2205
    const int t4= topright[0];\
2206
    const int t5= topright[1];\
2207
    const int t6= topright[2];\
2208
    const int t7= topright[3];\
2209

    
2210
#define LOAD_LEFT_EDGE\
2211
    const int l0= src[-1+0*stride];\
2212
    const int l1= src[-1+1*stride];\
2213
    const int l2= src[-1+2*stride];\
2214
    const int l3= src[-1+3*stride];\
2215

    
2216
#define LOAD_TOP_EDGE\
2217
    const int t0= src[ 0-1*stride];\
2218
    const int t1= src[ 1-1*stride];\
2219
    const int t2= src[ 2-1*stride];\
2220
    const int t3= src[ 3-1*stride];\
2221

    
2222
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
2223
    const int lt= src[-1-1*stride];
2224
    LOAD_TOP_EDGE
2225
    LOAD_LEFT_EDGE
2226

    
2227
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
2228
    src[0+2*stride]=
2229
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
2230
    src[0+1*stride]=
2231
    src[1+2*stride]=
2232
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
2233
    src[0+0*stride]=
2234
    src[1+1*stride]=
2235
    src[2+2*stride]=
2236
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2237
    src[1+0*stride]=
2238
    src[2+1*stride]=
2239
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
2240
    src[2+0*stride]=
2241
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2242
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2243
}
2244

    
2245
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
2246
    LOAD_TOP_EDGE
2247
    LOAD_TOP_RIGHT_EDGE
2248
//    LOAD_LEFT_EDGE
2249

    
2250
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
2251
    src[1+0*stride]=
2252
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
2253
    src[2+0*stride]=
2254
    src[1+1*stride]=
2255
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
2256
    src[3+0*stride]=
2257
    src[2+1*stride]=
2258
    src[1+2*stride]=
2259
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
2260
    src[3+1*stride]=
2261
    src[2+2*stride]=
2262
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
2263
    src[3+2*stride]=
2264
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
2265
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
2266
}
2267

    
2268
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
2269
    const int lt= src[-1-1*stride];
2270
    LOAD_TOP_EDGE
2271
    LOAD_LEFT_EDGE
2272
    const __attribute__((unused)) int unu= l3;
2273

    
2274
    src[0+0*stride]=
2275
    src[1+2*stride]=(lt + t0 + 1)>>1;
2276
    src[1+0*stride]=
2277
    src[2+2*stride]=(t0 + t1 + 1)>>1;
2278
    src[2+0*stride]=
2279
    src[3+2*stride]=(t1 + t2 + 1)>>1;
2280
    src[3+0*stride]=(t2 + t3 + 1)>>1;
2281
    src[0+1*stride]=
2282
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2283
    src[1+1*stride]=
2284
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2285
    src[2+1*stride]=
2286
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2287
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2288
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2289
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2290
}
2291

    
2292
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
2293
    LOAD_TOP_EDGE
2294
    LOAD_TOP_RIGHT_EDGE
2295
    const __attribute__((unused)) int unu= t7;
2296

    
2297
    src[0+0*stride]=(t0 + t1 + 1)>>1;
2298
    src[1+0*stride]=
2299
    src[0+2*stride]=(t1 + t2 + 1)>>1;
2300
    src[2+0*stride]=
2301
    src[1+2*stride]=(t2 + t3 + 1)>>1;
2302
    src[3+0*stride]=
2303
    src[2+2*stride]=(t3 + t4+ 1)>>1;
2304
    src[3+2*stride]=(t4 + t5+ 1)>>1;
2305
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2306
    src[1+1*stride]=
2307
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2308
    src[2+1*stride]=
2309
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2310
    src[3+1*stride]=
2311
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2312
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2313
}
2314

    
2315
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2316
    LOAD_LEFT_EDGE
2317

    
2318
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2319
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2320
    src[2+0*stride]=
2321
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2322
    src[3+0*stride]=
2323
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2324
    src[2+1*stride]=
2325
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2326
    src[3+1*stride]=
2327
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2328
    src[3+2*stride]=
2329
    src[1+3*stride]=
2330
    src[0+3*stride]=
2331
    src[2+2*stride]=
2332
    src[2+3*stride]=
2333
    src[3+3*stride]=l3;
2334
}
2335

    
2336
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2337
    const int lt= src[-1-1*stride];
2338
    LOAD_TOP_EDGE
2339
    LOAD_LEFT_EDGE
2340
    const __attribute__((unused)) int unu= t3;
2341

    
2342
    src[0+0*stride]=
2343
    src[2+1*stride]=(lt + l0 + 1)>>1;
2344
    src[1+0*stride]=
2345
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2346
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2347
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2348
    src[0+1*stride]=
2349
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2350
    src[1+1*stride]=
2351
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2352
    src[0+2*stride]=
2353
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2354
    src[1+2*stride]=
2355
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2356
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2357
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2358
}
2359

    
2360
static void pred16x16_vertical_c(uint8_t *src, int stride){
2361
    int i;
2362
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2363
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2364
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2365
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2366

    
2367
    for(i=0; i<16; i++){
2368
        ((uint32_t*)(src+i*stride))[0]= a;
2369
        ((uint32_t*)(src+i*stride))[1]= b;
2370
        ((uint32_t*)(src+i*stride))[2]= c;
2371
        ((uint32_t*)(src+i*stride))[3]= d;
2372
    }
2373
}
2374

    
2375
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2376
    int i;
2377

    
2378
    for(i=0; i<16; i++){
2379
        ((uint32_t*)(src+i*stride))[0]=
2380
        ((uint32_t*)(src+i*stride))[1]=
2381
        ((uint32_t*)(src+i*stride))[2]=
2382
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2383
    }
2384
}
2385

    
2386
static void pred16x16_dc_c(uint8_t *src, int stride){
2387
    int i, dc=0;
2388

    
2389
    for(i=0;i<16; i++){
2390
        dc+= src[-1+i*stride];
2391
    }
2392

    
2393
    for(i=0;i<16; i++){
2394
        dc+= src[i-stride];
2395
    }
2396

    
2397
    dc= 0x01010101*((dc + 16)>>5);
2398

    
2399
    for(i=0; i<16; i++){
2400
        ((uint32_t*)(src+i*stride))[0]=
2401
        ((uint32_t*)(src+i*stride))[1]=
2402
        ((uint32_t*)(src+i*stride))[2]=
2403
        ((uint32_t*)(src+i*stride))[3]= dc;
2404
    }
2405
}
2406

    
2407
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2408
    int i, dc=0;
2409

    
2410
    for(i=0;i<16; i++){
2411
        dc+= src[-1+i*stride];
2412
    }
2413

    
2414
    dc= 0x01010101*((dc + 8)>>4);
2415

    
2416
    for(i=0; i<16; i++){
2417
        ((uint32_t*)(src+i*stride))[0]=
2418
        ((uint32_t*)(src+i*stride))[1]=
2419
        ((uint32_t*)(src+i*stride))[2]=
2420
        ((uint32_t*)(src+i*stride))[3]= dc;
2421
    }
2422
}
2423

    
2424
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2425
    int i, dc=0;
2426

    
2427
    for(i=0;i<16; i++){
2428
        dc+= src[i-stride];
2429
    }
2430
    dc= 0x01010101*((dc + 8)>>4);
2431

    
2432
    for(i=0; i<16; i++){
2433
        ((uint32_t*)(src+i*stride))[0]=
2434
        ((uint32_t*)(src+i*stride))[1]=
2435
        ((uint32_t*)(src+i*stride))[2]=
2436
        ((uint32_t*)(src+i*stride))[3]= dc;
2437
    }
2438
}
2439

    
2440
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2441
    int i;
2442

    
2443
    for(i=0; i<16; i++){
2444
        ((uint32_t*)(src+i*stride))[0]=
2445
        ((uint32_t*)(src+i*stride))[1]=
2446
        ((uint32_t*)(src+i*stride))[2]=
2447
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2448
    }
2449
}
2450

    
2451
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2452
  int i, j, k;
2453
  int a;
2454
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2455
  const uint8_t * const src0 = src+7-stride;
2456
  const uint8_t *src1 = src+8*stride-1;
2457
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2458
  int H = src0[1] - src0[-1];
2459
  int V = src1[0] - src2[ 0];
2460
  for(k=2; k<=8; ++k) {
2461
    src1 += stride; src2 -= stride;
2462
    H += k*(src0[k] - src0[-k]);
2463
    V += k*(src1[0] - src2[ 0]);
2464
  }
2465
  if(svq3){
2466
    H = ( 5*(H/4) ) / 16;
2467
    V = ( 5*(V/4) ) / 16;
2468

    
2469
    /* required for 100% accuracy */
2470
    i = H; H = V; V = i;
2471
  }else{
2472
    H = ( 5*H+32 ) >> 6;
2473
    V = ( 5*V+32 ) >> 6;
2474
  }
2475

    
2476
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2477
  for(j=16; j>0; --j) {
2478
    int b = a;
2479
    a += V;
2480
    for(i=-16; i<0; i+=4) {
2481
      src[16+i] = cm[ (b    ) >> 5 ];
2482
      src[17+i] = cm[ (b+  H) >> 5 ];
2483
      src[18+i] = cm[ (b+2*H) >> 5 ];
2484
      src[19+i] = cm[ (b+3*H) >> 5 ];
2485
      b += 4*H;
2486
    }
2487
    src += stride;
2488
  }
2489
}
2490

    
2491
static void pred16x16_plane_c(uint8_t *src, int stride){
2492
    pred16x16_plane_compat_c(src, stride, 0);
2493
}
2494

    
2495
static void pred8x8_vertical_c(uint8_t *src, int stride){
2496
    int i;
2497
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2498
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2499

    
2500
    for(i=0; i<8; i++){
2501
        ((uint32_t*)(src+i*stride))[0]= a;
2502
        ((uint32_t*)(src+i*stride))[1]= b;
2503
    }
2504
}
2505

    
2506
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2507
    int i;
2508

    
2509
    for(i=0; i<8; i++){
2510
        ((uint32_t*)(src+i*stride))[0]=
2511
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2512
    }
2513
}
2514

    
2515
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2516
    int i;
2517

    
2518
    for(i=0; i<8; i++){
2519
        ((uint32_t*)(src+i*stride))[0]=
2520
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2521
    }
2522
}
2523

    
2524
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2525
    int i;
2526
    int dc0, dc2;
2527

    
2528
    dc0=dc2=0;
2529
    for(i=0;i<4; i++){
2530
        dc0+= src[-1+i*stride];
2531
        dc2+= src[-1+(i+4)*stride];
2532
    }
2533
    dc0= 0x01010101*((dc0 + 2)>>2);
2534
    dc2= 0x01010101*((dc2 + 2)>>2);
2535

    
2536
    for(i=0; i<4; i++){
2537
        ((uint32_t*)(src+i*stride))[0]=
2538
        ((uint32_t*)(src+i*stride))[1]= dc0;
2539
    }
2540
    for(i=4; i<8; i++){
2541
        ((uint32_t*)(src+i*stride))[0]=
2542
        ((uint32_t*)(src+i*stride))[1]= dc2;
2543
    }
2544
}
2545

    
2546
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2547
    int i;
2548
    int dc0, dc1;
2549

    
2550
    dc0=dc1=0;
2551
    for(i=0;i<4; i++){
2552
        dc0+= src[i-stride];
2553
        dc1+= src[4+i-stride];
2554
    }
2555
    dc0= 0x01010101*((dc0 + 2)>>2);
2556
    dc1= 0x01010101*((dc1 + 2)>>2);
2557

    
2558
    for(i=0; i<4; i++){
2559
        ((uint32_t*)(src+i*stride))[0]= dc0;
2560
        ((uint32_t*)(src+i*stride))[1]= dc1;
2561
    }
2562
    for(i=4; i<8; i++){
2563
        ((uint32_t*)(src+i*stride))[0]= dc0;
2564
        ((uint32_t*)(src+i*stride))[1]= dc1;
2565
    }
2566
}
2567

    
2568

    
2569
static void pred8x8_dc_c(uint8_t *src, int stride){
2570
    int i;
2571
    int dc0, dc1, dc2, dc3;
2572

    
2573
    dc0=dc1=dc2=0;
2574
    for(i=0;i<4; i++){
2575
        dc0+= src[-1+i*stride] + src[i-stride];
2576
        dc1+= src[4+i-stride];
2577
        dc2+= src[-1+(i+4)*stride];
2578
    }
2579
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2580
    dc0= 0x01010101*((dc0 + 4)>>3);
2581
    dc1= 0x01010101*((dc1 + 2)>>2);
2582
    dc2= 0x01010101*((dc2 + 2)>>2);
2583

    
2584
    for(i=0; i<4; i++){
2585
        ((uint32_t*)(src+i*stride))[0]= dc0;
2586
        ((uint32_t*)(src+i*stride))[1]= dc1;
2587
    }
2588
    for(i=4; i<8; i++){
2589
        ((uint32_t*)(src+i*stride))[0]= dc2;
2590
        ((uint32_t*)(src+i*stride))[1]= dc3;
2591
    }
2592
}
2593

    
2594
static void pred8x8_plane_c(uint8_t *src, int stride){
2595
  int j, k;
2596
  int a;
2597
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2598
  const uint8_t * const src0 = src+3-stride;
2599
  const uint8_t *src1 = src+4*stride-1;
2600
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2601
  int H = src0[1] - src0[-1];
2602
  int V = src1[0] - src2[ 0];
2603
  for(k=2; k<=4; ++k) {
2604
    src1 += stride; src2 -= stride;
2605
    H += k*(src0[k] - src0[-k]);
2606
    V += k*(src1[0] - src2[ 0]);
2607
  }
2608
  H = ( 17*H+16 ) >> 5;
2609
  V = ( 17*V+16 ) >> 5;
2610

    
2611
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2612
  for(j=8; j>0; --j) {
2613
    int b = a;
2614
    a += V;
2615
    src[0] = cm[ (b    ) >> 5 ];
2616
    src[1] = cm[ (b+  H) >> 5 ];
2617
    src[2] = cm[ (b+2*H) >> 5 ];
2618
    src[3] = cm[ (b+3*H) >> 5 ];
2619
    src[4] = cm[ (b+4*H) >> 5 ];
2620
    src[5] = cm[ (b+5*H) >> 5 ];
2621
    src[6] = cm[ (b+6*H) >> 5 ];
2622
    src[7] = cm[ (b+7*H) >> 5 ];
2623
    src += stride;
2624
  }
2625
}
2626

    
2627
#define SRC(x,y) src[(x)+(y)*stride]
2628
#define PL(y) \
2629
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2630
#define PREDICT_8x8_LOAD_LEFT \
2631
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2632
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2633
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2634
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2635

    
2636
#define PT(x) \
2637
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2638
#define PREDICT_8x8_LOAD_TOP \
2639
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2640
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2641
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2642
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2643
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2644

    
2645
#define PTR(x) \
2646
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2647
#define PREDICT_8x8_LOAD_TOPRIGHT \
2648
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2649
    if(has_topright) { \
2650
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2651
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2652
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2653

    
2654
#define PREDICT_8x8_LOAD_TOPLEFT \
2655
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2656

    
2657
#define PREDICT_8x8_DC(v) \
2658
    int y; \
2659
    for( y = 0; y < 8; y++ ) { \
2660
        ((uint32_t*)src)[0] = \
2661
        ((uint32_t*)src)[1] = v; \
2662
        src += stride; \
2663
    }
2664

    
2665
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2666
{
2667
    PREDICT_8x8_DC(0x80808080);
2668
}
2669
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2670
{
2671
    PREDICT_8x8_LOAD_LEFT;
2672
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2673
    PREDICT_8x8_DC(dc);
2674
}
2675
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2676
{
2677
    PREDICT_8x8_LOAD_TOP;
2678
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2679
    PREDICT_8x8_DC(dc);
2680
}
2681
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2682
{
2683
    PREDICT_8x8_LOAD_LEFT;
2684
    PREDICT_8x8_LOAD_TOP;
2685
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2686
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2687
    PREDICT_8x8_DC(dc);
2688
}
2689
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2690
{
2691
    PREDICT_8x8_LOAD_LEFT;
2692
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2693
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2694
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2695
#undef ROW
2696
}
2697
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2698
{
2699
    int y;
2700
    PREDICT_8x8_LOAD_TOP;
2701
    src[0] = t0;
2702
    src[1] = t1;
2703
    src[2] = t2;
2704
    src[3] = t3;
2705
    src[4] = t4;
2706
    src[5] = t5;
2707
    src[6] = t6;
2708
    src[7] = t7;
2709
    for( y = 1; y < 8; y++ )
2710
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2711
}
2712
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2713
{
2714
    PREDICT_8x8_LOAD_TOP;
2715
    PREDICT_8x8_LOAD_TOPRIGHT;
2716
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2717
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2718
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2719
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2720
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2721
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2722
    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;
2723
    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;
2724
    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;
2725
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2726
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2727
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2728
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2729
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2730
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2731
}
2732
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2733
{
2734
    PREDICT_8x8_LOAD_TOP;
2735
    PREDICT_8x8_LOAD_LEFT;
2736
    PREDICT_8x8_LOAD_TOPLEFT;
2737
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2738
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2739
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2740
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2741
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2742
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2743
    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;
2744
    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;
2745
    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;
2746
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2747
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2748
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2749
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2750
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2751
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2752

    
2753
}
2754
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2755
{
2756
    PREDICT_8x8_LOAD_TOP;
2757
    PREDICT_8x8_LOAD_LEFT;
2758
    PREDICT_8x8_LOAD_TOPLEFT;
2759
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2760
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2761
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2762
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2763
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2764
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2765
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2766
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2767
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2768
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2769
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2770
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2771
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2772
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2773
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2774
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2775
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2776
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2777
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2778
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2779
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2780
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2781
}
2782
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2783
{
2784
    PREDICT_8x8_LOAD_TOP;
2785
    PREDICT_8x8_LOAD_LEFT;
2786
    PREDICT_8x8_LOAD_TOPLEFT;
2787
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2788
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2789
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2790
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2791
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2792
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2793
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2794
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2795
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2796
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2797
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2798
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2799
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2800
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2801
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2802
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2803
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2804
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2805
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2806
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2807
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2808
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2809
}
2810
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2811
{
2812
    PREDICT_8x8_LOAD_TOP;
2813
    PREDICT_8x8_LOAD_TOPRIGHT;
2814
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2815
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2816
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2817
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2818
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2819
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2820
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2821
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2822
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2823
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2824
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2825
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2826
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2827
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2828
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2829
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2830
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2831
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2832
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2833
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2834
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2835
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2836
}
2837
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2838
{
2839
    PREDICT_8x8_LOAD_LEFT;
2840
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2841
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2842
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2843
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2844
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2845
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2846
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2847
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2848
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2849
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2850
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2851
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2852
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2853
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2854
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2855
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2856
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2857
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2858
}
2859
#undef PREDICT_8x8_LOAD_LEFT
2860
#undef PREDICT_8x8_LOAD_TOP
2861
#undef PREDICT_8x8_LOAD_TOPLEFT
2862
#undef PREDICT_8x8_LOAD_TOPRIGHT
2863
#undef PREDICT_8x8_DC
2864
#undef PTR
2865
#undef PT
2866
#undef PL
2867
#undef SRC
2868

    
2869
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2870
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2871
                           int src_x_offset, int src_y_offset,
2872
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2873
    MpegEncContext * const s = &h->s;
2874
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2875
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2876
    const int luma_xy= (mx&3) + ((my&3)<<2);
2877
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2878
    uint8_t * src_cb, * src_cr;
2879
    int extra_width= h->emu_edge_width;
2880
    int extra_height= h->emu_edge_height;
2881
    int emu=0;
2882
    const int full_mx= mx>>2;
2883
    const int full_my= my>>2;
2884
    const int pic_width  = 16*s->mb_width;
2885
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
2886

    
2887
    if(!pic->data[0])
2888
        return;
2889

    
2890
    if(mx&7) extra_width -= 3;
2891
    if(my&7) extra_height -= 3;
2892

    
2893
    if(   full_mx < 0-extra_width
2894
       || full_my < 0-extra_height
2895
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2896
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2897
        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);
2898
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2899
        emu=1;
2900
    }
2901

    
2902
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2903
    if(!square){
2904
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2905
    }
2906

    
2907
    if(s->flags&CODEC_FLAG_GRAY) return;
2908

    
2909
    if(MB_MBAFF){
2910
        // chroma offset when predicting from a field of opposite parity
2911
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2912
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2913
    }
2914
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2915
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2916

    
2917
    if(emu){
2918
        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);
2919
            src_cb= s->edge_emu_buffer;
2920
    }
2921
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2922

    
2923
    if(emu){
2924
        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);
2925
            src_cr= s->edge_emu_buffer;
2926
    }
2927
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2928
}
2929

    
2930
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2931
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2932
                           int x_offset, int y_offset,
2933
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2934
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2935
                           int list0, int list1){
2936
    MpegEncContext * const s = &h->s;
2937
    qpel_mc_func *qpix_op=  qpix_put;
2938
    h264_chroma_mc_func chroma_op= chroma_put;
2939

    
2940
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2941
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2942
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2943
    x_offset += 8*s->mb_x;
2944
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2945

    
2946
    if(list0){
2947
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2948
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2949
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2950
                           qpix_op, chroma_op);
2951

    
2952
        qpix_op=  qpix_avg;
2953
        chroma_op= chroma_avg;
2954
    }
2955

    
2956
    if(list1){
2957
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2958
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2959
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2960
                           qpix_op, chroma_op);
2961
    }
2962
}
2963

    
2964
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2965
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2966
                           int x_offset, int y_offset,
2967
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2968
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2969
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2970
                           int list0, int list1){
2971
    MpegEncContext * const s = &h->s;
2972

    
2973
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2974
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2975
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2976
    x_offset += 8*s->mb_x;
2977
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2978

    
2979
    if(list0 && list1){
2980
        /* don't optimize for luma-only case, since B-frames usually
2981
         * use implicit weights => chroma too. */
2982
        uint8_t *tmp_cb = s->obmc_scratchpad;
2983
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2984
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2985
        int refn0 = h->ref_cache[0][ scan8[n] ];
2986
        int refn1 = h->ref_cache[1][ scan8[n] ];
2987

    
2988
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2989
                    dest_y, dest_cb, dest_cr,
2990
                    x_offset, y_offset, qpix_put, chroma_put);
2991
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2992
                    tmp_y, tmp_cb, tmp_cr,
2993
                    x_offset, y_offset, qpix_put, chroma_put);
2994

    
2995
        if(h->use_weight == 2){
2996
            int weight0 = h->implicit_weight[refn0][refn1];
2997
            int weight1 = 64 - weight0;
2998
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
2999
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
3000
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
3001
        }else{
3002
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
3003
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
3004
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
3005
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3006
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
3007
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
3008
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3009
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
3010
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
3011
        }
3012
    }else{
3013
        int list = list1 ? 1 : 0;
3014
        int refn = h->ref_cache[list][ scan8[n] ];
3015
        Picture *ref= &h->ref_list[list][refn];
3016
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
3017
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
3018
                    qpix_put, chroma_put);
3019

    
3020
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
3021
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
3022
        if(h->use_weight_chroma){
3023
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3024
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
3025
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
3026
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
3027
        }
3028
    }
3029
}
3030

    
3031
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
3032
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3033
                           int x_offset, int y_offset,
3034
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
3035
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
3036
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
3037
                           int list0, int list1){
3038
    if((h->use_weight==2 && list0 && list1
3039
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
3040
       || h->use_weight==1)
3041
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
3042
                         x_offset, y_offset, qpix_put, chroma_put,
3043
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
3044
    else
3045
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
3046
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
3047
}
3048

    
3049
static inline void prefetch_motion(H264Context *h, int list){
3050
    /* fetch pixels for estimated mv 4 macroblocks ahead
3051
     * optimized for 64byte cache lines */
3052
    MpegEncContext * const s = &h->s;
3053
    const int refn = h->ref_cache[list][scan8[0]];
3054
    if(refn >= 0){
3055
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
3056
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
3057
        uint8_t **src= h->ref_list[list][refn].data;
3058
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
3059
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
3060
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
3061
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
3062
    }
3063
}
3064

    
3065
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3066
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
3067
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
3068
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
3069
    MpegEncContext * const s = &h->s;
3070
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3071
    const int mb_type= s->current_picture.mb_type[mb_xy];
3072

    
3073
    assert(IS_INTER(mb_type));
3074

    
3075
    prefetch_motion(h, 0);
3076

    
3077
    if(IS_16X16(mb_type)){
3078
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
3079
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
3080
                &weight_op[0], &weight_avg[0],
3081
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3082
    }else if(IS_16X8(mb_type)){
3083
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
3084
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
3085
                &weight_op[1], &weight_avg[1],
3086
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3087
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
3088
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
3089
                &weight_op[1], &weight_avg[1],
3090
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
3091
    }else if(IS_8X16(mb_type)){
3092
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
3093
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3094
                &weight_op[2], &weight_avg[2],
3095
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
3096
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
3097
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3098
                &weight_op[2], &weight_avg[2],
3099
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
3100
    }else{
3101
        int i;
3102

    
3103
        assert(IS_8X8(mb_type));
3104

    
3105
        for(i=0; i<4; i++){
3106
            const int sub_mb_type= h->sub_mb_type[i];
3107
            const int n= 4*i;
3108
            int x_offset= (i&1)<<2;
3109
            int y_offset= (i&2)<<1;
3110

    
3111
            if(IS_SUB_8X8(sub_mb_type)){
3112
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3113
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
3114
                    &weight_op[3], &weight_avg[3],
3115
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3116
            }else if(IS_SUB_8X4(sub_mb_type)){
3117
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3118
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
3119
                    &weight_op[4], &weight_avg[4],
3120
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3121
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
3122
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
3123
                    &weight_op[4], &weight_avg[4],
3124
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3125
            }else if(IS_SUB_4X8(sub_mb_type)){
3126
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
3127
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3128
                    &weight_op[5], &weight_avg[5],
3129
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3130
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
3131
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3132
                    &weight_op[5], &weight_avg[5],
3133
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3134
            }else{
3135
                int j;
3136
                assert(IS_SUB_4X4(sub_mb_type));
3137
                for(j=0; j<4; j++){
3138
                    int sub_x_offset= x_offset + 2*(j&1);
3139
                    int sub_y_offset= y_offset +   (j&2);
3140
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
3141
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
3142
                        &weight_op[6], &weight_avg[6],
3143
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3144
                }
3145
            }
3146
        }
3147
    }
3148

    
3149
    prefetch_motion(h, 1);
3150
}
3151

    
3152
static void decode_init_vlc(H264Context *h){
3153
    static int done = 0;
3154

    
3155
    if (!done) {
3156
        int i;
3157
        done = 1;
3158

    
3159
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
3160
                 &chroma_dc_coeff_token_len [0], 1, 1,
3161
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
3162

    
3163
        for(i=0; i<4; i++){
3164
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
3165
                     &coeff_token_len [i][0], 1, 1,
3166
                     &coeff_token_bits[i][0], 1, 1, 1);
3167
        }
3168

    
3169
        for(i=0; i<3; i++){
3170
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
3171
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
3172
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
3173
        }
3174
        for(i=0; i<15; i++){
3175
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
3176
                     &total_zeros_len [i][0], 1, 1,
3177
                     &total_zeros_bits[i][0], 1, 1, 1);
3178
        }
3179

    
3180
        for(i=0; i<6; i++){
3181
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
3182
                     &run_len [i][0], 1, 1,
3183
                     &run_bits[i][0], 1, 1, 1);
3184
        }
3185
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
3186
                 &run_len [6][0], 1, 1,
3187
                 &run_bits[6][0], 1, 1, 1);
3188
    }
3189
}
3190

    
3191
/**
3192
 * Sets the intra prediction function pointers.
3193
 */
3194
static void init_pred_ptrs(H264Context *h){
3195
//    MpegEncContext * const s = &h->s;
3196

    
3197
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
3198
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
3199
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
3200
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
3201
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
3202
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
3203
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
3204
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
3205
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
3206
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
3207
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
3208
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
3209

    
3210
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
3211
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
3212
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
3213
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
3214
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
3215
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
3216
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
3217
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
3218
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
3219
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
3220
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
3221
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
3222

    
3223
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
3224
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
3225
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
3226
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
3227
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
3228
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
3229
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
3230

    
3231
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
3232
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
3233
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
3234
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
3235
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
3236
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
3237
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
3238
}
3239

    
3240
static void free_tables(H264Context *h){
3241
    av_freep(&h->intra4x4_pred_mode);
3242
    av_freep(&h->chroma_pred_mode_table);
3243
    av_freep(&h->cbp_table);
3244
    av_freep(&h->mvd_table[0]);
3245
    av_freep(&h->mvd_table[1]);
3246
    av_freep(&h->direct_table);
3247
    av_freep(&h->non_zero_count);
3248
    av_freep(&h->slice_table_base);
3249
    av_freep(&h->top_borders[1]);
3250
    av_freep(&h->top_borders[0]);
3251
    h->slice_table= NULL;
3252

    
3253
    av_freep(&h->mb2b_xy);
3254
    av_freep(&h->mb2b8_xy);
3255

    
3256
    av_freep(&h->s.obmc_scratchpad);
3257
}
3258

    
3259
static void init_dequant8_coeff_table(H264Context *h){
3260
    int i,q,x;
3261
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
3262
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
3263
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
3264

    
3265
    for(i=0; i<2; i++ ){
3266
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
3267
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
3268
            break;
3269
        }
3270

    
3271
        for(q=0; q<52; q++){
3272
            int shift = div6[q];
3273
            int idx = rem6[q];
3274
            for(x=0; x<64; x++)
3275
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
3276
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
3277
                    h->pps.scaling_matrix8[i][x]) << shift;
3278
        }
3279
    }
3280
}
3281

    
3282
static void init_dequant4_coeff_table(H264Context *h){
3283
    int i,j,q,x;
3284
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
3285
    for(i=0; i<6; i++ ){
3286
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
3287
        for(j=0; j<i; j++){
3288
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
3289
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
3290
                break;
3291
            }
3292
        }
3293
        if(j<i)
3294
            continue;
3295

    
3296
        for(q=0; q<52; q++){
3297
            int shift = div6[q] + 2;
3298
            int idx = rem6[q];
3299
            for(x=0; x<16; x++)
3300
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
3301
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
3302
                    h->pps.scaling_matrix4[i][x]) << shift;
3303
        }
3304
    }
3305
}
3306

    
3307
static void init_dequant_tables(H264Context *h){
3308
    int i,x;
3309
    init_dequant4_coeff_table(h);
3310
    if(h->pps.transform_8x8_mode)
3311
        init_dequant8_coeff_table(h);
3312
    if(h->sps.transform_bypass){
3313
        for(i=0; i<6; i++)
3314
            for(x=0; x<16; x++)
3315
                h->dequant4_coeff[i][0][x] = 1<<6;
3316
        if(h->pps.transform_8x8_mode)
3317
            for(i=0; i<2; i++)
3318
                for(x=0; x<64; x++)
3319
                    h->dequant8_coeff[i][0][x] = 1<<6;
3320
    }
3321
}
3322

    
3323

    
3324
/**
3325
 * allocates tables.
3326
 * needs width/height
3327
 */
3328
static int alloc_tables(H264Context *h){
3329
    MpegEncContext * const s = &h->s;
3330
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
3331
    int x,y;
3332

    
3333
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
3334

    
3335
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
3336
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
3337
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3338
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3339
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
3340

    
3341
    if( h->pps.cabac ) {
3342
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
3343
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3344
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
3345
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
3346
    }
3347

    
3348
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
3349
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
3350

    
3351
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
3352
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
3353
    for(y=0; y<s->mb_height; y++){
3354
        for(x=0; x<s->mb_width; x++){
3355
            const int mb_xy= x + y*s->mb_stride;
3356
            const int b_xy = 4*x + 4*y*h->b_stride;
3357
            const int b8_xy= 2*x + 2*y*h->b8_stride;
3358

    
3359
            h->mb2b_xy [mb_xy]= b_xy;
3360
            h->mb2b8_xy[mb_xy]= b8_xy;
3361
        }
3362
    }
3363

    
3364
    s->obmc_scratchpad = NULL;
3365

    
3366
    if(!h->dequant4_coeff[0])
3367
        init_dequant_tables(h);
3368

    
3369
    return 0;
3370
fail:
3371
    free_tables(h);
3372
    return -1;
3373
}
3374

    
3375
static void common_init(H264Context *h){
3376
    MpegEncContext * const s = &h->s;
3377

    
3378
    s->width = s->avctx->width;
3379
    s->height = s->avctx->height;
3380
    s->codec_id= s->avctx->codec->id;
3381

    
3382
    init_pred_ptrs(h);
3383

    
3384
    h->dequant_coeff_pps= -1;
3385
    s->unrestricted_mv=1;
3386
    s->decode=1; //FIXME
3387

    
3388
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3389
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
3390
}
3391

    
3392
static int decode_init(AVCodecContext *avctx){
3393
    H264Context *h= avctx->priv_data;
3394
    MpegEncContext * const s = &h->s;
3395

    
3396
    MPV_decode_defaults(s);
3397

    
3398
    s->avctx = avctx;
3399
    common_init(h);
3400

    
3401
    s->out_format = FMT_H264;
3402
    s->workaround_bugs= avctx->workaround_bugs;
3403

    
3404
    // set defaults
3405
//    s->decode_mb= ff_h263_decode_mb;
3406
    s->low_delay= 1;
3407
    avctx->pix_fmt= PIX_FMT_YUV420P;
3408

    
3409
    decode_init_vlc(h);
3410

    
3411
    if(avctx->extradata_size > 0 && avctx->extradata &&
3412
       *(char *)avctx->extradata == 1){
3413
        h->is_avc = 1;
3414
        h->got_avcC = 0;
3415
    } else {
3416
        h->is_avc = 0;
3417
    }
3418

    
3419
    return 0;
3420
}
3421

    
3422
static int frame_start(H264Context *h){
3423
    MpegEncContext * const s = &h->s;
3424
    int i;
3425

    
3426
    if(MPV_frame_start(s, s->avctx) < 0)
3427
        return -1;
3428
    ff_er_frame_start(s);
3429

    
3430
    assert(s->linesize && s->uvlinesize);
3431

    
3432
    for(i=0; i<16; i++){
3433
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3434
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3435
    }
3436
    for(i=0; i<4; i++){
3437
        h->block_offset[16+i]=
3438
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3439
        h->block_offset[24+16+i]=
3440
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3441
    }
3442

    
3443
    /* can't be in alloc_tables because linesize isn't known there.
3444
     * FIXME: redo bipred weight to not require extra buffer? */
3445
    if(!s->obmc_scratchpad)
3446
        s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
3447

    
3448
    /* some macroblocks will be accessed before they're available */
3449
    if(FRAME_MBAFF)
3450
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
3451

    
3452
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3453
    return 0;
3454
}
3455

    
3456
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3457
    MpegEncContext * const s = &h->s;
3458
    int i;
3459

    
3460
    src_y  -=   linesize;
3461
    src_cb -= uvlinesize;
3462
    src_cr -= uvlinesize;
3463

    
3464
    // There are two lines saved, the line above the the top macroblock of a pair,
3465
    // and the line above the bottom macroblock
3466
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3467
    for(i=1; i<17; i++){
3468
        h->left_border[i]= src_y[15+i*  linesize];
3469
    }
3470

    
3471
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3472
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3473

    
3474
    if(!(s->flags&CODEC_FLAG_GRAY)){
3475
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3476
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3477
        for(i=1; i<9; i++){
3478
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3479
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3480
        }
3481
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3482
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3483
    }
3484
}
3485

    
3486
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){
3487
    MpegEncContext * const s = &h->s;
3488
    int temp8, i;
3489
    uint64_t temp64;
3490
    int deblock_left = (s->mb_x > 0);
3491
    int deblock_top  = (s->mb_y > 0);
3492

    
3493
    src_y  -=   linesize + 1;
3494
    src_cb -= uvlinesize + 1;
3495
    src_cr -= uvlinesize + 1;
3496

    
3497
#define XCHG(a,b,t,xchg)\
3498
t= a;\
3499
if(xchg)\
3500
    a= b;\
3501
b= t;
3502

    
3503
    if(deblock_left){
3504
        for(i = !deblock_top; i<17; i++){
3505
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3506
        }
3507
    }
3508

    
3509
    if(deblock_top){
3510
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3511
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3512
        if(s->mb_x+1 < s->mb_width){
3513
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3514
        }
3515
    }
3516

    
3517
    if(!(s->flags&CODEC_FLAG_GRAY)){
3518
        if(deblock_left){
3519
            for(i = !deblock_top; i<9; i++){
3520
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3521
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3522
            }
3523
        }
3524
        if(deblock_top){
3525
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3526
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3527
        }
3528
    }
3529
}
3530

    
3531
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3532
    MpegEncContext * const s = &h->s;
3533
    int i;
3534

    
3535
    src_y  -= 2 *   linesize;
3536
    src_cb -= 2 * uvlinesize;
3537
    src_cr -= 2 * uvlinesize;
3538

    
3539
    // There are two lines saved, the line above the the top macroblock of a pair,
3540
    // and the line above the bottom macroblock
3541
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3542
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3543
    for(i=2; i<34; i++){
3544
        h->left_border[i]= src_y[15+i*  linesize];
3545
    }
3546

    
3547
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3548
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3549
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3550
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3551

    
3552
    if(!(s->flags&CODEC_FLAG_GRAY)){
3553
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3554
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3555
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3556
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3557
        for(i=2; i<18; i++){
3558
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3559
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3560
        }
3561
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3562
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3563
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3564
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3565
    }
3566
}
3567

    
3568
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){
3569
    MpegEncContext * const s = &h->s;
3570
    int temp8, i;
3571
    uint64_t temp64;
3572
    int deblock_left = (s->mb_x > 0);
3573
    int deblock_top  = (s->mb_y > 1);
3574

    
3575
    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);
3576

    
3577
    src_y  -= 2 *   linesize + 1;
3578
    src_cb -= 2 * uvlinesize + 1;
3579
    src_cr -= 2 * uvlinesize + 1;
3580

    
3581
#define XCHG(a,b,t,xchg)\
3582
t= a;\
3583
if(xchg)\
3584
    a= b;\
3585
b= t;
3586

    
3587
    if(deblock_left){
3588
        for(i = (!deblock_top)<<1; i<34; i++){
3589
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3590
        }
3591
    }
3592

    
3593
    if(deblock_top){
3594
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3595
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3596
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3597
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3598
        if(s->mb_x+1 < s->mb_width){
3599
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3600
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3601
        }
3602
    }
3603

    
3604
    if(!(s->flags&CODEC_FLAG_GRAY)){
3605
        if(deblock_left){
3606
            for(i = (!deblock_top) << 1; i<18; i++){
3607
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3608
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3609
            }
3610
        }
3611
        if(deblock_top){
3612
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3613
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3614
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3615
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3616
        }
3617
    }
3618
}
3619

    
3620
static void hl_decode_mb(H264Context *h){
3621
    MpegEncContext * const s = &h->s;
3622
    const int mb_x= s->mb_x;
3623
    const int mb_y= s->mb_y;
3624
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3625
    const int mb_type= s->current_picture.mb_type[mb_xy];
3626
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3627
    int linesize, uvlinesize /*dct_offset*/;
3628
    int i;
3629
    int *block_offset = &h->block_offset[0];
3630
    const unsigned int bottom = mb_y & 1;
3631
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3632
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3633
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3634

    
3635
    if(!s->decode)
3636
        return;
3637

    
3638
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3639
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3640
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3641

    
3642
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
3643
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
3644

    
3645
    if (MB_FIELD) {
3646
        linesize   = h->mb_linesize   = s->linesize * 2;
3647
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3648
        block_offset = &h->block_offset[24];
3649
        if(mb_y&1){ //FIXME move out of this func?
3650
            dest_y -= s->linesize*15;
3651
            dest_cb-= s->uvlinesize*7;
3652
            dest_cr-= s->uvlinesize*7;
3653
        }
3654
        if(FRAME_MBAFF) {
3655
            int list;
3656
            for(list=0; list<2; list++){
3657
                if(!USES_LIST(mb_type, list))
3658
                    continue;
3659
                if(IS_16X16(mb_type)){
3660
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
3661
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3662
                }else{
3663
                    for(i=0; i<16; i+=4){
3664
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3665
                        int ref = h->ref_cache[list][scan8[i]];
3666
                        if(ref >= 0)
3667
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3668
                    }
3669
                }
3670
            }
3671
        }
3672
    } else {
3673
        linesize   = h->mb_linesize   = s->linesize;
3674
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3675
//        dct_offset = s->linesize * 16;
3676
    }
3677

    
3678
    if(transform_bypass){
3679
        idct_dc_add =
3680
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3681
    }else if(IS_8x8DCT(mb_type)){
3682
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3683
        idct_add = s->dsp.h264_idct8_add;
3684
    }else{
3685
        idct_dc_add = s->dsp.h264_idct_dc_add;
3686
        idct_add = s->dsp.h264_idct_add;
3687
    }
3688

    
3689
    if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3690
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3691
        int mbt_y = mb_y&~1;
3692
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
3693
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3694
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3695
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3696
    }
3697

    
3698
    if (IS_INTRA_PCM(mb_type)) {
3699
        unsigned int x, y;
3700

    
3701
        // The pixels are stored in h->mb array in the same order as levels,
3702
        // copy them in output in the correct order.
3703
        for(i=0; i<16; i++) {
3704
            for (y=0; y<4; y++) {
3705
                for (x=0; x<4; x++) {
3706
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3707
                }
3708
            }
3709
        }
3710
        for(i=16; i<16+4; i++) {
3711
            for (y=0; y<4; y++) {
3712
                for (x=0; x<4; x++) {
3713
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3714
                }
3715
            }
3716
        }
3717
        for(i=20; i<20+4; i++) {
3718
            for (y=0; y<4; y++) {
3719
                for (x=0; x<4; x++) {
3720
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3721
                }
3722
            }
3723
        }
3724
    } else {
3725
        if(IS_INTRA(mb_type)){
3726
            if(h->deblocking_filter && !FRAME_MBAFF)
3727
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3728

    
3729
            if(!(s->flags&CODEC_FLAG_GRAY)){
3730
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3731
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3732
            }
3733

    
3734
            if(IS_INTRA4x4(mb_type)){
3735
                if(!s->encoding){
3736
                    if(IS_8x8DCT(mb_type)){
3737
                        for(i=0; i<16; i+=4){
3738
                            uint8_t * const ptr= dest_y + block_offset[i];
3739
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3740
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3741
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3742
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3743
                            if(nnz){
3744
                                if(nnz == 1 && h->mb[i*16])
3745
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3746
                                else
3747
                                    idct_add(ptr, h->mb + i*16, linesize);
3748
                            }
3749
                        }
3750
                    }else
3751
                    for(i=0; i<16; i++){
3752
                        uint8_t * const ptr= dest_y + block_offset[i];
3753
                        uint8_t *topright;
3754
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3755
                        int nnz, tr;
3756

    
3757
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3758
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3759
                            assert(mb_y || linesize <= block_offset[i]);
3760
                            if(!topright_avail){
3761
                                tr= ptr[3 - linesize]*0x01010101;
3762
                                topright= (uint8_t*) &tr;
3763
                            }else
3764
                                topright= ptr + 4 - linesize;
3765
                        }else
3766
                            topright= NULL;
3767

    
3768
                        h->pred4x4[ dir ](ptr, topright, linesize);
3769
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3770
                        if(nnz){
3771
                            if(s->codec_id == CODEC_ID_H264){
3772
                                if(nnz == 1 && h->mb[i*16])
3773
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3774
                                else
3775
                                    idct_add(ptr, h->mb + i*16, linesize);
3776
                            }else
3777
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3778
                        }
3779
                    }
3780
                }
3781
            }else{
3782
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3783
                if(s->codec_id == CODEC_ID_H264){
3784
                    if(!transform_bypass)
3785
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3786
                }else
3787
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3788
            }
3789
            if(h->deblocking_filter && !FRAME_MBAFF)
3790
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3791
        }else if(s->codec_id == CODEC_ID_H264){
3792
            hl_motion(h, dest_y, dest_cb, dest_cr,
3793
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
3794
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
3795
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3796
        }
3797

    
3798

    
3799
        if(!IS_INTRA4x4(mb_type)){
3800
            if(s->codec_id == CODEC_ID_H264){
3801
                if(IS_INTRA16x16(mb_type)){
3802
                    for(i=0; i<16; i++){
3803
                        if(h->non_zero_count_cache[ scan8[i] ])
3804
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3805
                        else if(h->mb[i*16])
3806
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3807
                    }
3808
                }else{
3809
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3810
                    for(i=0; i<16; i+=di){
3811
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3812
                        if(nnz){
3813
                            if(nnz==1 && h->mb[i*16])
3814
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3815
                            else
3816
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3817
                        }
3818
                    }
3819
                }
3820
            }else{
3821
                for(i=0; i<16; i++){
3822
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3823
                        uint8_t * const ptr= dest_y + block_offset[i];
3824
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3825
                    }
3826
                }
3827
            }
3828
        }
3829

    
3830
        if(!(s->flags&CODEC_FLAG_GRAY)){
3831
            uint8_t *dest[2] = {dest_cb, dest_cr};
3832
            if(transform_bypass){
3833
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3834
            }else{
3835
                idct_add = s->dsp.h264_idct_add;
3836
                idct_dc_add = s->dsp.h264_idct_dc_add;
3837
                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]);
3838
                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]);
3839
            }
3840
            if(s->codec_id == CODEC_ID_H264){
3841
                for(i=16; i<16+8; i++){
3842
                    if(h->non_zero_count_cache[ scan8[i] ])
3843
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3844
                    else if(h->mb[i*16])
3845
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3846
                }
3847
            }else{
3848
                for(i=16; i<16+8; i++){
3849
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3850
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3851
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3852
                    }
3853
                }
3854
            }
3855
        }
3856
    }
3857
    if(h->deblocking_filter) {
3858
        if (FRAME_MBAFF) {
3859
            //FIXME try deblocking one mb at a time?
3860
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3861
            const int mb_y = s->mb_y - 1;
3862
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3863
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3864
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3865
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3866
            if (!bottom) return;
3867
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3868
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3869
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3870

    
3871
            if(IS_INTRA(mb_type_top | mb_type_bottom))
3872
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3873

    
3874
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3875
            // deblock a pair
3876
            // top
3877
            s->mb_y--;
3878
            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);
3879
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3880
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
3881
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3882
            // bottom
3883
            s->mb_y++;
3884
            tprintf("call mbaff filter_mb\n");
3885
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3886
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3887
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3888
        } else {
3889
            tprintf("call filter_mb\n");
3890
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3891
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3892
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3893
        }
3894
    }
3895
}
3896

    
3897
/**
3898
 * fills the default_ref_list.
3899
 */
3900
static int fill_default_ref_list(H264Context *h){
3901
    MpegEncContext * const s = &h->s;
3902
    int i;
3903
    int smallest_poc_greater_than_current = -1;
3904
    Picture sorted_short_ref[32];
3905

    
3906
    if(h->slice_type==B_TYPE){
3907
        int out_i;
3908
        int limit= INT_MIN;
3909

    
3910
        /* sort frame according to poc in B slice */
3911
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3912
            int best_i=INT_MIN;
3913
            int best_poc=INT_MAX;
3914

    
3915
            for(i=0; i<h->short_ref_count; i++){
3916
                const int poc= h->short_ref[i]->poc;
3917
                if(poc > limit && poc < best_poc){
3918
                    best_poc= poc;
3919
                    best_i= i;
3920
                }
3921
            }
3922

    
3923
            assert(best_i != INT_MIN);
3924

    
3925
            limit= best_poc;
3926
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3927
            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);
3928
            if (-1 == smallest_poc_greater_than_current) {
3929
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3930
                    smallest_poc_greater_than_current = out_i;
3931
                }
3932
            }
3933
        }
3934
    }
3935

    
3936
    if(s->picture_structure == PICT_FRAME){
3937
        if(h->slice_type==B_TYPE){
3938
            int list;
3939
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3940

    
3941
            // find the largest poc
3942
            for(list=0; list<2; list++){
3943
                int index = 0;
3944
                int j= -99;
3945
                int step= list ? -1 : 1;
3946

    
3947
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3948
                    while(j<0 || j>= h->short_ref_count){
3949
                        if(j != -99 && step == (list ? -1 : 1))
3950
                            return -1;
3951
                        step = -step;
3952
                        j= smallest_poc_greater_than_current + (step>>1);
3953
                    }
3954
                    if(sorted_short_ref[j].reference != 3) continue;
3955
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3956
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3957
                }
3958

    
3959
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3960
                    if(h->long_ref[i] == NULL) continue;
3961
                    if(h->long_ref[i]->reference != 3) continue;
3962

    
3963
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3964
                    h->default_ref_list[ list ][index++].pic_id= i;;
3965
                }
3966

    
3967
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3968
                    // swap the two first elements of L1 when
3969
                    // L0 and L1 are identical
3970
                    Picture temp= h->default_ref_list[1][0];
3971
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3972
                    h->default_ref_list[1][1] = temp;
3973
                }
3974

    
3975
                if(index < h->ref_count[ list ])
3976
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3977
            }
3978
        }else{
3979
            int index=0;
3980
            for(i=0; i<h->short_ref_count; i++){
3981
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3982
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3983
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3984
            }
3985
            for(i = 0; i < 16; i++){
3986
                if(h->long_ref[i] == NULL) continue;
3987
                if(h->long_ref[i]->reference != 3) continue;
3988
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3989
                h->default_ref_list[0][index++].pic_id= i;;
3990
            }
3991
            if(index < h->ref_count[0])
3992
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3993
        }
3994
    }else{ //FIELD
3995
        if(h->slice_type==B_TYPE){
3996
        }else{
3997
            //FIXME second field balh
3998
        }
3999
    }
4000
#ifdef TRACE
4001
    for (i=0; i<h->ref_count[0]; i++) {
4002
        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]);
4003
    }
4004
    if(h->slice_type==B_TYPE){
4005
        for (i=0; i<h->ref_count[1]; i++) {
4006
            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]);
4007
        }
4008
    }
4009
#endif
4010
    return 0;
4011
}
4012

    
4013
static void print_short_term(H264Context *h);
4014
static void print_long_term(H264Context *h);
4015

    
4016
static int decode_ref_pic_list_reordering(H264Context *h){
4017
    MpegEncContext * const s = &h->s;
4018
    int list, index;
4019

    
4020
    print_short_term(h);
4021
    print_long_term(h);
4022
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
4023

    
4024
    for(list=0; list<2; list++){
4025
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
4026

    
4027
        if(get_bits1(&s->gb)){
4028
            int pred= h->curr_pic_num;
4029

    
4030
            for(index=0; ; index++){
4031
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
4032
                int pic_id;
4033
                int i;
4034
                Picture *ref = NULL;
4035

    
4036
                if(reordering_of_pic_nums_idc==3)
4037
                    break;
4038

    
4039
                if(index >= h->ref_count[list]){
4040
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
4041
                    return -1;
4042
                }
4043

    
4044
                if(reordering_of_pic_nums_idc<3){
4045
                    if(reordering_of_pic_nums_idc<2){
4046
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
4047

    
4048
                        if(abs_diff_pic_num >= h->max_pic_num){
4049
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
4050
                            return -1;
4051
                        }
4052

    
4053
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
4054
                        else                                pred+= abs_diff_pic_num;
4055
                        pred &= h->max_pic_num - 1;
4056

    
4057
                        for(i= h->short_ref_count-1; i>=0; i--){
4058
                            ref = h->short_ref[i];
4059
                            assert(ref->reference == 3);
4060
                            assert(!ref->long_ref);
4061
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
4062
                                break;
4063
                        }
4064
                        if(i>=0)
4065
                            ref->pic_id= ref->frame_num;
4066
                    }else{
4067
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
4068
                        ref = h->long_ref[pic_id];
4069
                        ref->pic_id= pic_id;
4070
                        assert(ref->reference == 3);
4071
                        assert(ref->long_ref);
4072
                        i=0;
4073
                    }
4074

    
4075
                    if (i < 0) {
4076
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
4077
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
4078
                    } else {
4079
                        for(i=index; i+1<h->ref_count[list]; i++){
4080
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
4081
                                break;
4082
                        }
4083
                        for(; i > index; i--){
4084
                            h->ref_list[list][i]= h->ref_list[list][i-1];
4085
                        }
4086
                        h->ref_list[list][index]= *ref;
4087
                    }
4088
                }else{
4089
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
4090
                    return -1;
4091
                }
4092
            }
4093
        }
4094

    
4095
        if(h->slice_type!=B_TYPE) break;
4096
    }
4097
    for(list=0; list<2; list++){
4098
        for(index= 0; index < h->ref_count[list]; index++){
4099
            if(!h->ref_list[list][index].data[0])
4100
                h->ref_list[list][index]= s->current_picture;
4101
        }
4102
        if(h->slice_type!=B_TYPE) break;
4103
    }
4104

    
4105
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
4106
        direct_dist_scale_factor(h);
4107
    direct_ref_list_init(h);
4108
    return 0;
4109
}
4110

    
4111
static void fill_mbaff_ref_list(H264Context *h){
4112
    int list, i, j;
4113
    for(list=0; list<2; list++){
4114
        for(i=0; i<h->ref_count[list]; i++){
4115
            Picture *frame = &h->ref_list[list][i];
4116
            Picture *field = &h->ref_list[list][16+2*i];
4117
            field[0] = *frame;
4118
            for(j=0; j<3; j++)
4119
                field[0].linesize[j] <<= 1;
4120
            field[1] = field[0];
4121
            for(j=0; j<3; j++)
4122
                field[1].data[j] += frame->linesize[j];
4123

    
4124
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
4125
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
4126
            for(j=0; j<2; j++){
4127
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
4128
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
4129
            }
4130
        }
4131
    }
4132
    for(j=0; j<h->ref_count[1]; j++){
4133
        for(i=0; i<h->ref_count[0]; i++)
4134
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
4135
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
4136
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
4137
    }
4138
}
4139

    
4140
static int pred_weight_table(H264Context *h){
4141
    MpegEncContext * const s = &h->s;
4142
    int list, i;
4143
    int luma_def, chroma_def;
4144

    
4145
    h->use_weight= 0;
4146
    h->use_weight_chroma= 0;
4147
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
4148
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
4149
    luma_def = 1<<h->luma_log2_weight_denom;
4150
    chroma_def = 1<<h->chroma_log2_weight_denom;
4151

    
4152
    for(list=0; list<2; list++){
4153
        for(i=0; i<h->ref_count[list]; i++){
4154
            int luma_weight_flag, chroma_weight_flag;
4155

    
4156
            luma_weight_flag= get_bits1(&s->gb);
4157
            if(luma_weight_flag){
4158
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
4159
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
4160
                if(   h->luma_weight[list][i] != luma_def
4161
                   || h->luma_offset[list][i] != 0)
4162
                    h->use_weight= 1;
4163
            }else{
4164
                h->luma_weight[list][i]= luma_def;
4165
                h->luma_offset[list][i]= 0;
4166
            }
4167

    
4168
            chroma_weight_flag= get_bits1(&s->gb);
4169
            if(chroma_weight_flag){
4170
                int j;
4171
                for(j=0; j<2; j++){
4172
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
4173
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
4174
                    if(   h->chroma_weight[list][i][j] != chroma_def
4175
                       || h->chroma_offset[list][i][j] != 0)
4176
                        h->use_weight_chroma= 1;
4177
                }
4178
            }else{
4179
                int j;
4180
                for(j=0; j<2; j++){
4181
                    h->chroma_weight[list][i][j]= chroma_def;
4182
                    h->chroma_offset[list][i][j]= 0;
4183
                }
4184
            }
4185
        }
4186
        if(h->slice_type != B_TYPE) break;
4187
    }
4188
    h->use_weight= h->use_weight || h->use_weight_chroma;
4189
    return 0;
4190
}
4191

    
4192
static void implicit_weight_table(H264Context *h){
4193
    MpegEncContext * const s = &h->s;
4194
    int ref0, ref1;
4195
    int cur_poc = s->current_picture_ptr->poc;
4196

    
4197
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
4198
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
4199
        h->use_weight= 0;
4200
        h->use_weight_chroma= 0;
4201
        return;
4202
    }
4203

    
4204
    h->use_weight= 2;
4205
    h->use_weight_chroma= 2;
4206
    h->luma_log2_weight_denom= 5;
4207
    h->chroma_log2_weight_denom= 5;
4208

    
4209
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
4210
        int poc0 = h->ref_list[0][ref0].poc;
4211
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
4212
            int poc1 = h->ref_list[1][ref1].poc;
4213
            int td = clip(poc1 - poc0, -128, 127);
4214
            if(td){
4215
                int tb = clip(cur_poc - poc0, -128, 127);
4216
                int tx = (16384 + (FFABS(td) >> 1)) / td;
4217
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
4218
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
4219
                    h->implicit_weight[ref0][ref1] = 32;
4220
                else
4221
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
4222
            }else
4223
                h->implicit_weight[ref0][ref1] = 32;
4224
        }
4225
    }
4226
}
4227

    
4228
static inline void unreference_pic(H264Context *h, Picture *pic){
4229
    int i;
4230
    pic->reference=0;
4231
    if(pic == h->delayed_output_pic)
4232
        pic->reference=1;
4233
    else{
4234
        for(i = 0; h->delayed_pic[i]; i++)
4235
            if(pic == h->delayed_pic[i]){
4236
                pic->reference=1;
4237
                break;
4238
            }
4239
    }
4240
}
4241

    
4242
/**
4243
 * instantaneous decoder refresh.
4244
 */
4245
static void idr(H264Context *h){
4246
    int i;
4247

    
4248
    for(i=0; i<16; i++){
4249
        if (h->long_ref[i] != NULL) {
4250
            unreference_pic(h, h->long_ref[i]);
4251
            h->long_ref[i]= NULL;
4252
        }
4253
    }
4254
    h->long_ref_count=0;
4255

    
4256
    for(i=0; i<h->short_ref_count; i++){
4257
        unreference_pic(h, h->short_ref[i]);
4258
        h->short_ref[i]= NULL;
4259
    }
4260
    h->short_ref_count=0;
4261
}
4262

    
4263
/* forget old pics after a seek */
4264
static void flush_dpb(AVCodecContext *avctx){
4265
    H264Context *h= avctx->priv_data;
4266
    int i;
4267
    for(i=0; i<16; i++) {
4268
        if(h->delayed_pic[i])
4269
            h->delayed_pic[i]->reference= 0;
4270
        h->delayed_pic[i]= NULL;
4271
    }
4272
    if(h->delayed_output_pic)
4273
        h->delayed_output_pic->reference= 0;
4274
    h->delayed_output_pic= NULL;
4275
    idr(h);
4276
    if(h->s.current_picture_ptr)
4277
        h->s.current_picture_ptr->reference= 0;
4278
}
4279

    
4280
/**