Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 53b19144

History | View | Annotate | Download (300 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
 *
19
 */
20

    
21
/**
22
 * @file h264.c
23
 * H.264 / AVC / MPEG4 part10 codec.
24
 * @author Michael Niedermayer <michaelni@gmx.at>
25
 */
26

    
27
#include "common.h"
28
#include "dsputil.h"
29
#include "avcodec.h"
30
#include "mpegvideo.h"
31
#include "h264data.h"
32
#include "golomb.h"
33

    
34
#include "cabac.h"
35

    
36
#undef NDEBUG
37
#include <assert.h>
38

    
39
#define interlaced_dct interlaced_dct_is_a_bad_name
40
#define mb_intra mb_intra_isnt_initalized_see_mb_type
41

    
42
#define LUMA_DC_BLOCK_INDEX   25
43
#define CHROMA_DC_BLOCK_INDEX 26
44

    
45
#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46
#define COEFF_TOKEN_VLC_BITS           8
47
#define TOTAL_ZEROS_VLC_BITS           9
48
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49
#define RUN_VLC_BITS                   3
50
#define RUN7_VLC_BITS                  6
51

    
52
#define MAX_SPS_COUNT 32
53
#define MAX_PPS_COUNT 256
54

    
55
#define MAX_MMCO_COUNT 66
56

    
57
/**
58
 * Sequence parameter set
59
 */
60
typedef struct SPS{
61

    
62
    int profile_idc;
63
    int level_idc;
64
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
65
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
66
    int poc_type;                      ///< pic_order_cnt_type
67
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
68
    int delta_pic_order_always_zero_flag;
69
    int offset_for_non_ref_pic;
70
    int offset_for_top_to_bottom_field;
71
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
72
    int ref_frame_count;               ///< num_ref_frames
73
    int gaps_in_frame_num_allowed_flag;
74
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
75
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
76
    int frame_mbs_only_flag;
77
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
78
    int direct_8x8_inference_flag;
79
    int crop;                   ///< frame_cropping_flag
80
    int crop_left;              ///< frame_cropping_rect_left_offset
81
    int crop_right;             ///< frame_cropping_rect_right_offset
82
    int crop_top;               ///< frame_cropping_rect_top_offset
83
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
84
    int vui_parameters_present_flag;
85
    AVRational sar;
86
    int timing_info_present_flag;
87
    uint32_t num_units_in_tick;
88
    uint32_t time_scale;
89
    int fixed_frame_rate_flag;
90
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
91
    int bitstream_restriction_flag;
92
    int num_reorder_frames;
93
    int scaling_matrix_present;
94
    uint8_t scaling_matrix4[6][16];
95
    uint8_t scaling_matrix8[2][64];
96
}SPS;
97

    
98
/**
99
 * Picture parameter set
100
 */
101
typedef struct PPS{
102
    int sps_id;
103
    int cabac;                  ///< entropy_coding_mode_flag
104
    int pic_order_present;      ///< pic_order_present_flag
105
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
106
    int mb_slice_group_map_type;
107
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
108
    int weighted_pred;          ///< weighted_pred_flag
109
    int weighted_bipred_idc;
110
    int init_qp;                ///< pic_init_qp_minus26 + 26
111
    int init_qs;                ///< pic_init_qs_minus26 + 26
112
    int chroma_qp_index_offset;
113
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
114
    int constrained_intra_pred; ///< constrained_intra_pred_flag
115
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
116
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
117
    uint8_t scaling_matrix4[6][16];
118
    uint8_t scaling_matrix8[2][64];
119
}PPS;
120

    
121
/**
122
 * Memory management control operation opcode.
123
 */
124
typedef enum MMCOOpcode{
125
    MMCO_END=0,
126
    MMCO_SHORT2UNUSED,
127
    MMCO_LONG2UNUSED,
128
    MMCO_SHORT2LONG,
129
    MMCO_SET_MAX_LONG,
130
    MMCO_RESET,
131
    MMCO_LONG,
132
} MMCOOpcode;
133

    
134
/**
135
 * Memory management control operation.
136
 */
137
typedef struct MMCO{
138
    MMCOOpcode opcode;
139
    int short_frame_num;
140
    int long_index;
141
} MMCO;
142

    
143
/**
144
 * H264Context
145
 */
146
typedef struct H264Context{
147
    MpegEncContext s;
148
    int nal_ref_idc;
149
    int nal_unit_type;
150
#define NAL_SLICE                1
151
#define NAL_DPA                  2
152
#define NAL_DPB                  3
153
#define NAL_DPC                  4
154
#define NAL_IDR_SLICE            5
155
#define NAL_SEI                  6
156
#define NAL_SPS                  7
157
#define NAL_PPS                  8
158
#define NAL_AUD                  9
159
#define NAL_END_SEQUENCE        10
160
#define NAL_END_STREAM          11
161
#define NAL_FILLER_DATA         12
162
#define NAL_SPS_EXT             13
163
#define NAL_AUXILIARY_SLICE     19
164
    uint8_t *rbsp_buffer;
165
    unsigned int rbsp_buffer_size;
166

    
167
    /**
168
      * Used to parse AVC variant of h264
169
      */
170
    int is_avc; ///< this flag is != 0 if codec is avc1
171
    int got_avcC; ///< flag used to parse avcC data only once
172
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
173

    
174
    int chroma_qp; //QPc
175

    
176
    int prev_mb_skipped; //FIXME remove (IMHO not used)
177

    
178
    //prediction stuff
179
    int chroma_pred_mode;
180
    int intra16x16_pred_mode;
181

    
182
    int top_mb_xy;
183
    int left_mb_xy[2];
184

    
185
    int8_t intra4x4_pred_mode_cache[5*8];
186
    int8_t (*intra4x4_pred_mode)[8];
187
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
188
    void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
189
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
190
    void (*pred16x16[4+3])(uint8_t *src, int stride);
191
    unsigned int topleft_samples_available;
192
    unsigned int top_samples_available;
193
    unsigned int topright_samples_available;
194
    unsigned int left_samples_available;
195
    uint8_t (*top_borders[2])[16+2*8];
196
    uint8_t left_border[2*(17+2*9)];
197

    
198
    /**
199
     * non zero coeff count cache.
200
     * is 64 if not available.
201
     */
202
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
203
    uint8_t (*non_zero_count)[16];
204

    
205
    /**
206
     * Motion vector cache.
207
     */
208
    DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
209
    DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
210
#define LIST_NOT_USED -1 //FIXME rename?
211
#define PART_NOT_AVAILABLE -2
212

    
213
    /**
214
     * is 1 if the specific list MV&references are set to 0,0,-2.
215
     */
216
    int mv_cache_clean[2];
217

    
218
    /**
219
     * number of neighbors (top and/or left) that used 8x8 dct
220
     */
221
    int neighbor_transform_size;
222

    
223
    /**
224
     * block_offset[ 0..23] for frame macroblocks
225
     * block_offset[24..47] for field macroblocks
226
     */
227
    int block_offset[2*(16+8)];
228

    
229
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
230
    uint32_t *mb2b8_xy;
231
    int b_stride; //FIXME use s->b4_stride
232
    int b8_stride;
233

    
234
    int halfpel_flag;
235
    int thirdpel_flag;
236

    
237
    int unknown_svq3_flag;
238
    int next_slice_index;
239

    
240
    SPS sps_buffer[MAX_SPS_COUNT];
241
    SPS sps; ///< current sps
242

    
243
    PPS pps_buffer[MAX_PPS_COUNT];
244
    /**
245
     * current pps
246
     */
247
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
248

    
249
    uint32_t dequant4_buffer[6][52][16];
250
    uint32_t dequant8_buffer[2][52][64];
251
    uint32_t (*dequant4_coeff[6])[16];
252
    uint32_t (*dequant8_coeff[2])[64];
253
    int dequant_coeff_pps;     ///< reinit tables when pps changes
254

    
255
    int slice_num;
256
    uint8_t *slice_table_base;
257
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
258
    int slice_type;
259
    int slice_type_fixed;
260

    
261
    //interlacing specific flags
262
    int mb_aff_frame;
263
    int mb_field_decoding_flag;
264

    
265
    int sub_mb_type[4];
266

    
267
    //POC stuff
268
    int poc_lsb;
269
    int poc_msb;
270
    int delta_poc_bottom;
271
    int delta_poc[2];
272
    int frame_num;
273
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
274
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
275
    int frame_num_offset;         ///< for POC type 2
276
    int prev_frame_num_offset;    ///< for POC type 2
277
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
278

    
279
    /**
280
     * frame_num for frames or 2*frame_num for field pics.
281
     */
282
    int curr_pic_num;
283

    
284
    /**
285
     * max_frame_num or 2*max_frame_num for field pics.
286
     */
287
    int max_pic_num;
288

    
289
    //Weighted pred stuff
290
    int use_weight;
291
    int use_weight_chroma;
292
    int luma_log2_weight_denom;
293
    int chroma_log2_weight_denom;
294
    int luma_weight[2][16];
295
    int luma_offset[2][16];
296
    int chroma_weight[2][16][2];
297
    int chroma_offset[2][16][2];
298
    int implicit_weight[16][16];
299

    
300
    //deblock
301
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
302
    int slice_alpha_c0_offset;
303
    int slice_beta_offset;
304

    
305
    int redundant_pic_count;
306

    
307
    int direct_spatial_mv_pred;
308
    int dist_scale_factor[16];
309
    int map_col_to_list0[2][16];
310

    
311
    /**
312
     * num_ref_idx_l0/1_active_minus1 + 1
313
     */
314
    int ref_count[2];// FIXME split for AFF
315
    Picture *short_ref[32];
316
    Picture *long_ref[32];
317
    Picture default_ref_list[2][32];
318
    Picture ref_list[2][32]; //FIXME size?
319
    Picture field_ref_list[2][32]; //FIXME size?
320
    Picture *delayed_pic[16]; //FIXME size?
321
    Picture *delayed_output_pic;
322

    
323
    /**
324
     * memory management control operations buffer.
325
     */
326
    MMCO mmco[MAX_MMCO_COUNT];
327
    int mmco_index;
328

    
329
    int long_ref_count;  ///< number of actual long term references
330
    int short_ref_count; ///< number of actual short term references
331

    
332
    //data partitioning
333
    GetBitContext intra_gb;
334
    GetBitContext inter_gb;
335
    GetBitContext *intra_gb_ptr;
336
    GetBitContext *inter_gb_ptr;
337

    
338
    DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);
339

    
340
    /**
341
     * Cabac
342
     */
343
    CABACContext cabac;
344
    uint8_t      cabac_state[460];
345
    int          cabac_init_idc;
346

    
347
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
348
    uint16_t     *cbp_table;
349
    int top_cbp;
350
    int left_cbp;
351
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
352
    uint8_t     *chroma_pred_mode_table;
353
    int         last_qscale_diff;
354
    int16_t     (*mvd_table[2])[2];
355
    DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
356
    uint8_t     *direct_table;
357
    uint8_t     direct_cache[5*8];
358

    
359
    uint8_t zigzag_scan[16];
360
    uint8_t field_scan[16];
361
    uint8_t zigzag_scan8x8[64];
362
    uint8_t zigzag_scan8x8_cavlc[64];
363
    const uint8_t *zigzag_scan_q0;
364
    const uint8_t *field_scan_q0;
365
    const uint8_t *zigzag_scan8x8_q0;
366
    const uint8_t *zigzag_scan8x8_cavlc_q0;
367

    
368
    int x264_build;
369
}H264Context;
370

    
371
static VLC coeff_token_vlc[4];
372
static VLC chroma_dc_coeff_token_vlc;
373

    
374
static VLC total_zeros_vlc[15];
375
static VLC chroma_dc_total_zeros_vlc[3];
376

    
377
static VLC run_vlc[6];
378
static VLC run7_vlc;
379

    
380
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
381
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
382
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);
383

    
384
static always_inline uint32_t pack16to32(int a, int b){
385
#ifdef WORDS_BIGENDIAN
386
   return (b&0xFFFF) + (a<<16);
387
#else
388
   return (a&0xFFFF) + (b<<16);
389
#endif
390
}
391

    
392
/**
393
 * fill a rectangle.
394
 * @param h height of the rectangle, should be a constant
395
 * @param w width of the rectangle, should be a constant
396
 * @param size the size of val (1 or 4), should be a constant
397
 */
398
static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
399
    uint8_t *p= (uint8_t*)vp;
400
    assert(size==1 || size==4);
401

    
402
    w      *= size;
403
    stride *= size;
404

    
405
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
406
    assert((stride&(w-1))==0);
407
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
408
    if(w==2 && h==2){
409
        *(uint16_t*)(p + 0)=
410
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
411
    }else if(w==2 && h==4){
412
        *(uint16_t*)(p + 0*stride)=
413
        *(uint16_t*)(p + 1*stride)=
414
        *(uint16_t*)(p + 2*stride)=
415
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
416
    }else if(w==4 && h==1){
417
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
418
    }else if(w==4 && h==2){
419
        *(uint32_t*)(p + 0*stride)=
420
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
421
    }else if(w==4 && h==4){
422
        *(uint32_t*)(p + 0*stride)=
423
        *(uint32_t*)(p + 1*stride)=
424
        *(uint32_t*)(p + 2*stride)=
425
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
426
    }else if(w==8 && h==1){
427
        *(uint32_t*)(p + 0)=
428
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
429
    }else if(w==8 && h==2){
430
        *(uint32_t*)(p + 0 + 0*stride)=
431
        *(uint32_t*)(p + 4 + 0*stride)=
432
        *(uint32_t*)(p + 0 + 1*stride)=
433
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
434
    }else if(w==8 && h==4){
435
        *(uint64_t*)(p + 0*stride)=
436
        *(uint64_t*)(p + 1*stride)=
437
        *(uint64_t*)(p + 2*stride)=
438
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
439
    }else if(w==16 && h==2){
440
        *(uint64_t*)(p + 0+0*stride)=
441
        *(uint64_t*)(p + 8+0*stride)=
442
        *(uint64_t*)(p + 0+1*stride)=
443
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
444
    }else if(w==16 && h==4){
445
        *(uint64_t*)(p + 0+0*stride)=
446
        *(uint64_t*)(p + 8+0*stride)=
447
        *(uint64_t*)(p + 0+1*stride)=
448
        *(uint64_t*)(p + 8+1*stride)=
449
        *(uint64_t*)(p + 0+2*stride)=
450
        *(uint64_t*)(p + 8+2*stride)=
451
        *(uint64_t*)(p + 0+3*stride)=
452
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
453
    }else
454
        assert(0);
455
}
456

    
457
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
458
    MpegEncContext * const s = &h->s;
459
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
460
    int topleft_xy, top_xy, topright_xy, left_xy[2];
461
    int topleft_type, top_type, topright_type, left_type[2];
462
    int left_block[8];
463
    int i;
464

    
465
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
466
    // the actual condition is whether we're on the edge of a slice,
467
    // and even then the intra and nnz parts are unnecessary.
468
    if(for_deblock && h->slice_num == 1)
469
        return;
470

    
471
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
472

    
473
    top_xy     = mb_xy  - s->mb_stride;
474
    topleft_xy = top_xy - 1;
475
    topright_xy= top_xy + 1;
476
    left_xy[1] = left_xy[0] = mb_xy-1;
477
    left_block[0]= 0;
478
    left_block[1]= 1;
479
    left_block[2]= 2;
480
    left_block[3]= 3;
481
    left_block[4]= 7;
482
    left_block[5]= 10;
483
    left_block[6]= 8;
484
    left_block[7]= 11;
485
    if(h->mb_aff_frame){
486
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
487
        const int top_pair_xy      = pair_xy     - s->mb_stride;
488
        const int topleft_pair_xy  = top_pair_xy - 1;
489
        const int topright_pair_xy = top_pair_xy + 1;
490
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
491
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
492
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
493
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
494
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
495
        const int bottom = (s->mb_y & 1);
496
        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);
497
        if (bottom
498
                ? !curr_mb_frame_flag // bottom macroblock
499
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
500
                ) {
501
            top_xy -= s->mb_stride;
502
        }
503
        if (bottom
504
                ? !curr_mb_frame_flag // bottom macroblock
505
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
506
                ) {
507
            topleft_xy -= s->mb_stride;
508
        }
509
        if (bottom
510
                ? !curr_mb_frame_flag // bottom macroblock
511
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
512
                ) {
513
            topright_xy -= s->mb_stride;
514
        }
515
        if (left_mb_frame_flag != curr_mb_frame_flag) {
516
            left_xy[1] = left_xy[0] = pair_xy - 1;
517
            if (curr_mb_frame_flag) {
518
                if (bottom) {
519
                    left_block[0]= 2;
520
                    left_block[1]= 2;
521
                    left_block[2]= 3;
522
                    left_block[3]= 3;
523
                    left_block[4]= 8;
524
                    left_block[5]= 11;
525
                    left_block[6]= 8;
526
                    left_block[7]= 11;
527
                } else {
528
                    left_block[0]= 0;
529
                    left_block[1]= 0;
530
                    left_block[2]= 1;
531
                    left_block[3]= 1;
532
                    left_block[4]= 7;
533
                    left_block[5]= 10;
534
                    left_block[6]= 7;
535
                    left_block[7]= 10;
536
                }
537
            } else {
538
                left_xy[1] += s->mb_stride;
539
                //left_block[0]= 0;
540
                left_block[1]= 2;
541
                left_block[2]= 0;
542
                left_block[3]= 2;
543
                //left_block[4]= 7;
544
                left_block[5]= 10;
545
                left_block[6]= 7;
546
                left_block[7]= 10;
547
            }
548
        }
549
    }
550

    
551
    h->top_mb_xy = top_xy;
552
    h->left_mb_xy[0] = left_xy[0];
553
    h->left_mb_xy[1] = left_xy[1];
554
    if(for_deblock){
555
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
556
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
557
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
558
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
559
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
560
    }else{
561
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
562
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
563
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
564
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
565
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
566
    }
567

    
568
    if(IS_INTRA(mb_type)){
569
        h->topleft_samples_available=
570
        h->top_samples_available=
571
        h->left_samples_available= 0xFFFF;
572
        h->topright_samples_available= 0xEEEA;
573

    
574
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
575
            h->topleft_samples_available= 0xB3FF;
576
            h->top_samples_available= 0x33FF;
577
            h->topright_samples_available= 0x26EA;
578
        }
579
        for(i=0; i<2; i++){
580
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
581
                h->topleft_samples_available&= 0xDF5F;
582
                h->left_samples_available&= 0x5F5F;
583
            }
584
        }
585

    
586
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
587
            h->topleft_samples_available&= 0x7FFF;
588

    
589
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
590
            h->topright_samples_available&= 0xFBFF;
591

    
592
        if(IS_INTRA4x4(mb_type)){
593
            if(IS_INTRA4x4(top_type)){
594
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
595
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
596
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
597
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
598
            }else{
599
                int pred;
600
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
601
                    pred= -1;
602
                else{
603
                    pred= 2;
604
                }
605
                h->intra4x4_pred_mode_cache[4+8*0]=
606
                h->intra4x4_pred_mode_cache[5+8*0]=
607
                h->intra4x4_pred_mode_cache[6+8*0]=
608
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
609
            }
610
            for(i=0; i<2; i++){
611
                if(IS_INTRA4x4(left_type[i])){
612
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
613
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
614
                }else{
615
                    int pred;
616
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
617
                        pred= -1;
618
                    else{
619
                        pred= 2;
620
                    }
621
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
622
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
623
                }
624
            }
625
        }
626
    }
627

    
628

    
629
/*
630
0 . T T. T T T T
631
1 L . .L . . . .
632
2 L . .L . . . .
633
3 . T TL . . . .
634
4 L . .L . . . .
635
5 L . .. . . . .
636
*/
637
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
638
    if(top_type){
639
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
640
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
641
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
642
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
643

    
644
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
645
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
646

    
647
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
648
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
649

    
650
    }else{
651
        h->non_zero_count_cache[4+8*0]=
652
        h->non_zero_count_cache[5+8*0]=
653
        h->non_zero_count_cache[6+8*0]=
654
        h->non_zero_count_cache[7+8*0]=
655

    
656
        h->non_zero_count_cache[1+8*0]=
657
        h->non_zero_count_cache[2+8*0]=
658

    
659
        h->non_zero_count_cache[1+8*3]=
660
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
661

    
662
    }
663

    
664
    for (i=0; i<2; i++) {
665
        if(left_type[i]){
666
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
667
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
668
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
669
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
670
        }else{
671
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
672
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
673
            h->non_zero_count_cache[0+8*1 +   8*i]=
674
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
675
        }
676
    }
677

    
678
    if( h->pps.cabac ) {
679
        // top_cbp
680
        if(top_type) {
681
            h->top_cbp = h->cbp_table[top_xy];
682
        } else if(IS_INTRA(mb_type)) {
683
            h->top_cbp = 0x1C0;
684
        } else {
685
            h->top_cbp = 0;
686
        }
687
        // left_cbp
688
        if (left_type[0]) {
689
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
690
        } else if(IS_INTRA(mb_type)) {
691
            h->left_cbp = 0x1C0;
692
        } else {
693
            h->left_cbp = 0;
694
        }
695
        if (left_type[0]) {
696
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
697
        }
698
        if (left_type[1]) {
699
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
700
        }
701
    }
702

    
703
#if 1
704
    //FIXME direct mb can skip much of this
705
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
706
        int list;
707
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
708
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
709
                /*if(!h->mv_cache_clean[list]){
710
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
711
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
712
                    h->mv_cache_clean[list]= 1;
713
                }*/
714
                continue;
715
            }
716
            h->mv_cache_clean[list]= 0;
717

    
718
            if(USES_LIST(top_type, list)){
719
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
720
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
721
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
722
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
723
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
724
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
725
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
726
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
727
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
728
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
729
            }else{
730
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
731
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
732
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
733
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
734
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
735
            }
736

    
737
            //FIXME unify cleanup or sth
738
            if(USES_LIST(left_type[0], list)){
739
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
740
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
741
                *(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]];
742
                *(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]];
743
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
744
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
745
            }else{
746
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
747
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
748
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
749
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
750
            }
751

    
752
            if(USES_LIST(left_type[1], list)){
753
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
754
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
755
                *(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]];
756
                *(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]];
757
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
758
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
759
            }else{
760
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
761
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
762
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
763
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
764
                assert((!left_type[0]) == (!left_type[1]));
765
            }
766

    
767
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
768
                continue;
769

    
770
            if(USES_LIST(topleft_type, list)){
771
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
772
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
773
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
774
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
775
            }else{
776
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
777
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
778
            }
779

    
780
            if(USES_LIST(topright_type, list)){
781
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
782
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
783
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
784
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
785
            }else{
786
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
787
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
788
            }
789

    
790

    
791
            h->ref_cache[list][scan8[5 ]+1] =
792
            h->ref_cache[list][scan8[7 ]+1] =
793
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
794
            h->ref_cache[list][scan8[4 ]] =
795
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
796
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
797
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
798
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
799
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
800
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
801

    
802
            if( h->pps.cabac ) {
803
                /* XXX beurk, Load mvd */
804
                if(USES_LIST(topleft_type, list)){
805
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
806
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
807
                }else{
808
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
809
                }
810

    
811
                if(USES_LIST(top_type, list)){
812
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
813
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
814
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
815
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
816
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
817
                }else{
818
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
819
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
820
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
821
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
822
                }
823
                if(USES_LIST(left_type[0], list)){
824
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
825
                    *(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]];
826
                    *(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]];
827
                }else{
828
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
829
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
830
                }
831
                if(USES_LIST(left_type[1], list)){
832
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
833
                    *(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]];
834
                    *(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]];
835
                }else{
836
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
837
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
838
                }
839
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
840
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
841
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
842
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
843
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
844

    
845
                if(h->slice_type == B_TYPE){
846
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
847

    
848
                    if(IS_DIRECT(top_type)){
849
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
850
                    }else if(IS_8X8(top_type)){
851
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
852
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
853
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
854
                    }else{
855
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
856
                    }
857

    
858
                    //FIXME interlacing
859
                    if(IS_DIRECT(left_type[0])){
860
                        h->direct_cache[scan8[0] - 1 + 0*8]=
861
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
862
                    }else if(IS_8X8(left_type[0])){
863
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
864
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
865
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
866
                    }else{
867
                        h->direct_cache[scan8[0] - 1 + 0*8]=
868
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
869
                    }
870
                }
871
            }
872
        }
873
    }
874
#endif
875

    
876
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
877
}
878

    
879
static inline void write_back_intra_pred_mode(H264Context *h){
880
    MpegEncContext * const s = &h->s;
881
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
882

    
883
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
884
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
885
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
886
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
887
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
888
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
889
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
890
}
891

    
892
/**
893
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
894
 */
895
static inline int check_intra4x4_pred_mode(H264Context *h){
896
    MpegEncContext * const s = &h->s;
897
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
898
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
899
    int i;
900

    
901
    if(!(h->top_samples_available&0x8000)){
902
        for(i=0; i<4; i++){
903
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
904
            if(status<0){
905
                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);
906
                return -1;
907
            } else if(status){
908
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
909
            }
910
        }
911
    }
912

    
913
    if(!(h->left_samples_available&0x8000)){
914
        for(i=0; i<4; i++){
915
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
916
            if(status<0){
917
                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);
918
                return -1;
919
            } else if(status){
920
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
921
            }
922
        }
923
    }
924

    
925
    return 0;
926
} //FIXME cleanup like next
927

    
928
/**
929
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
930
 */
931
static inline int check_intra_pred_mode(H264Context *h, int mode){
932
    MpegEncContext * const s = &h->s;
933
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
934
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
935

    
936
    if(mode < 0 || mode > 6) {
937
        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);
938
        return -1;
939
    }
940

    
941
    if(!(h->top_samples_available&0x8000)){
942
        mode= top[ mode ];
943
        if(mode<0){
944
            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);
945
            return -1;
946
        }
947
    }
948

    
949
    if(!(h->left_samples_available&0x8000)){
950
        mode= left[ mode ];
951
        if(mode<0){
952
            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);
953
            return -1;
954
        }
955
    }
956

    
957
    return mode;
958
}
959

    
960
/**
961
 * gets the predicted intra4x4 prediction mode.
962
 */
963
static inline int pred_intra_mode(H264Context *h, int n){
964
    const int index8= scan8[n];
965
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
966
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
967
    const int min= FFMIN(left, top);
968

    
969
    tprintf("mode:%d %d min:%d\n", left ,top, min);
970

    
971
    if(min<0) return DC_PRED;
972
    else      return min;
973
}
974

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

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

    
987
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
988
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
989
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
990

    
991
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
992
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
993
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
994
}
995

    
996
/**
997
 * gets the predicted number of non zero coefficients.
998
 * @param n block index
999
 */
1000
static inline int pred_non_zero_count(H264Context *h, int n){
1001
    const int index8= scan8[n];
1002
    const int left= h->non_zero_count_cache[index8 - 1];
1003
    const int top = h->non_zero_count_cache[index8 - 8];
1004
    int i= left + top;
1005

    
1006
    if(i<64) i= (i+1)>>1;
1007

    
1008
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1009

    
1010
    return i&31;
1011
}
1012

    
1013
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1014
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1015

    
1016
    if(topright_ref != PART_NOT_AVAILABLE){
1017
        *C= h->mv_cache[list][ i - 8 + part_width ];
1018
        return topright_ref;
1019
    }else{
1020
        tprintf("topright MV not available\n");
1021

    
1022
        *C= h->mv_cache[list][ i - 8 - 1 ];
1023
        return h->ref_cache[list][ i - 8 - 1 ];
1024
    }
1025
}
1026

    
1027
/**
1028
 * gets the predicted MV.
1029
 * @param n the block index
1030
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1031
 * @param mx the x component of the predicted motion vector
1032
 * @param my the y component of the predicted motion vector
1033
 */
1034
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1035
    const int index8= scan8[n];
1036
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1037
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1038
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1039
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1040
    const int16_t * C;
1041
    int diagonal_ref, match_count;
1042

    
1043
    assert(part_width==1 || part_width==2 || part_width==4);
1044

    
1045
/* mv_cache
1046
  B . . A T T T T
1047
  U . . L . . , .
1048
  U . . L . . . .
1049
  U . . L . . , .
1050
  . . . L . . . .
1051
*/
1052

    
1053
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1054
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1055
    tprintf("pred_motion match_count=%d\n", match_count);
1056
    if(match_count > 1){ //most common
1057
        *mx= mid_pred(A[0], B[0], C[0]);
1058
        *my= mid_pred(A[1], B[1], C[1]);
1059
    }else if(match_count==1){
1060
        if(left_ref==ref){
1061
            *mx= A[0];
1062
            *my= A[1];
1063
        }else if(top_ref==ref){
1064
            *mx= B[0];
1065
            *my= B[1];
1066
        }else{
1067
            *mx= C[0];
1068
            *my= C[1];
1069
        }
1070
    }else{
1071
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1072
            *mx= A[0];
1073
            *my= A[1];
1074
        }else{
1075
            *mx= mid_pred(A[0], B[0], C[0]);
1076
            *my= mid_pred(A[1], B[1], C[1]);
1077
        }
1078
    }
1079

    
1080
    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);
1081
}
1082

    
1083
/**
1084
 * gets the directionally predicted 16x8 MV.
1085
 * @param n the block index
1086
 * @param mx the x component of the predicted motion vector
1087
 * @param my the y component of the predicted motion vector
1088
 */
1089
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1090
    if(n==0){
1091
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1092
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1093

    
1094
        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);
1095

    
1096
        if(top_ref == ref){
1097
            *mx= B[0];
1098
            *my= B[1];
1099
            return;
1100
        }
1101
    }else{
1102
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1103
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1104

    
1105
        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);
1106

    
1107
        if(left_ref == ref){
1108
            *mx= A[0];
1109
            *my= A[1];
1110
            return;
1111
        }
1112
    }
1113

    
1114
    //RARE
1115
    pred_motion(h, n, 4, list, ref, mx, my);
1116
}
1117

    
1118
/**
1119
 * gets the directionally predicted 8x16 MV.
1120
 * @param n the block index
1121
 * @param mx the x component of the predicted motion vector
1122
 * @param my the y component of the predicted motion vector
1123
 */
1124
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1125
    if(n==0){
1126
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1127
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1128

    
1129
        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);
1130

    
1131
        if(left_ref == ref){
1132
            *mx= A[0];
1133
            *my= A[1];
1134
            return;
1135
        }
1136
    }else{
1137
        const int16_t * C;
1138
        int diagonal_ref;
1139

    
1140
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1141

    
1142
        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);
1143

    
1144
        if(diagonal_ref == ref){
1145
            *mx= C[0];
1146
            *my= C[1];
1147
            return;
1148
        }
1149
    }
1150

    
1151
    //RARE
1152
    pred_motion(h, n, 2, list, ref, mx, my);
1153
}
1154

    
1155
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1156
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1157
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1158

    
1159
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1160

    
1161
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1162
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1163
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1164

    
1165
        *mx = *my = 0;
1166
        return;
1167
    }
1168

    
1169
    pred_motion(h, 0, 4, 0, 0, mx, my);
1170

    
1171
    return;
1172
}
1173

    
1174
static inline void direct_dist_scale_factor(H264Context * const h){
1175
    const int poc = h->s.current_picture_ptr->poc;
1176
    const int poc1 = h->ref_list[1][0].poc;
1177
    int i;
1178
    for(i=0; i<h->ref_count[0]; i++){
1179
        int poc0 = h->ref_list[0][i].poc;
1180
        int td = clip(poc1 - poc0, -128, 127);
1181
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1182
            h->dist_scale_factor[i] = 256;
1183
        }else{
1184
            int tb = clip(poc - poc0, -128, 127);
1185
            int tx = (16384 + (ABS(td) >> 1)) / td;
1186
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1187
        }
1188
    }
1189
}
1190
static inline void direct_ref_list_init(H264Context * const h){
1191
    MpegEncContext * const s = &h->s;
1192
    Picture * const ref1 = &h->ref_list[1][0];
1193
    Picture * const cur = s->current_picture_ptr;
1194
    int list, i, j;
1195
    if(cur->pict_type == I_TYPE)
1196
        cur->ref_count[0] = 0;
1197
    if(cur->pict_type != B_TYPE)
1198
        cur->ref_count[1] = 0;
1199
    for(list=0; list<2; list++){
1200
        cur->ref_count[list] = h->ref_count[list];
1201
        for(j=0; j<h->ref_count[list]; j++)
1202
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1203
    }
1204
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1205
        return;
1206
    for(list=0; list<2; list++){
1207
        for(i=0; i<ref1->ref_count[list]; i++){
1208
            const int poc = ref1->ref_poc[list][i];
1209
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1210
            for(j=0; j<h->ref_count[list]; j++)
1211
                if(h->ref_list[list][j].poc == poc){
1212
                    h->map_col_to_list0[list][i] = j;
1213
                    break;
1214
                }
1215
        }
1216
    }
1217
}
1218

    
1219
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1220
    MpegEncContext * const s = &h->s;
1221
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1222
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1223
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1224
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1225
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1226
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1227
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1228
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1229
    const int is_b8x8 = IS_8X8(*mb_type);
1230
    int sub_mb_type;
1231
    int i8, i4;
1232

    
1233
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1234
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1235
         * so we know exactly what block size to use */
1236
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1237
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1238
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1239
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1240
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1241
    }else{
1242
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1243
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1244
    }
1245
    if(!is_b8x8)
1246
        *mb_type |= MB_TYPE_DIRECT2;
1247

    
1248
    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);
1249

    
1250
    if(h->direct_spatial_mv_pred){
1251
        int ref[2];
1252
        int mv[2][2];
1253
        int list;
1254

    
1255
        /* ref = min(neighbors) */
1256
        for(list=0; list<2; list++){
1257
            int refa = h->ref_cache[list][scan8[0] - 1];
1258
            int refb = h->ref_cache[list][scan8[0] - 8];
1259
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1260
            if(refc == -2)
1261
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1262
            ref[list] = refa;
1263
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1264
                ref[list] = refb;
1265
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1266
                ref[list] = refc;
1267
            if(ref[list] < 0)
1268
                ref[list] = -1;
1269
        }
1270

    
1271
        if(ref[0] < 0 && ref[1] < 0){
1272
            ref[0] = ref[1] = 0;
1273
            mv[0][0] = mv[0][1] =
1274
            mv[1][0] = mv[1][1] = 0;
1275
        }else{
1276
            for(list=0; list<2; list++){
1277
                if(ref[list] >= 0)
1278
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1279
                else
1280
                    mv[list][0] = mv[list][1] = 0;
1281
            }
1282
        }
1283

    
1284
        if(ref[1] < 0){
1285
            *mb_type &= ~MB_TYPE_P0L1;
1286
            sub_mb_type &= ~MB_TYPE_P0L1;
1287
        }else if(ref[0] < 0){
1288
            *mb_type &= ~MB_TYPE_P0L0;
1289
            sub_mb_type &= ~MB_TYPE_P0L0;
1290
        }
1291

    
1292
        if(IS_16X16(*mb_type)){
1293
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1294
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1295
            if(!IS_INTRA(mb_type_col)
1296
               && (   (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1297
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1298
                       && (h->x264_build>33 || !h->x264_build)))){
1299
                if(ref[0] > 0)
1300
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1301
                else
1302
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1303
                if(ref[1] > 0)
1304
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1305
                else
1306
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1307
            }else{
1308
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1309
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1310
            }
1311
        }else{
1312
            for(i8=0; i8<4; i8++){
1313
                const int x8 = i8&1;
1314
                const int y8 = i8>>1;
1315

    
1316
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1317
                    continue;
1318
                h->sub_mb_type[i8] = sub_mb_type;
1319

    
1320
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1321
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1322
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1323
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1324

    
1325
                /* col_zero_flag */
1326
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1327
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1328
                                                  && (h->x264_build>33 || !h->x264_build)))){
1329
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1330
                    if(IS_SUB_8X8(sub_mb_type)){
1331
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1332
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1333
                            if(ref[0] == 0)
1334
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1335
                            if(ref[1] == 0)
1336
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1337
                        }
1338
                    }else
1339
                    for(i4=0; i4<4; i4++){
1340
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1341
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1342
                            if(ref[0] == 0)
1343
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1344
                            if(ref[1] == 0)
1345
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1346
                        }
1347
                    }
1348
                }
1349
            }
1350
        }
1351
    }else{ /* direct temporal mv pred */
1352
        if(IS_16X16(*mb_type)){
1353
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1354
            if(IS_INTRA(mb_type_col)){
1355
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1356
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1357
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1358
            }else{
1359
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1360
                                                : h->map_col_to_list0[1][l1ref1[0]];
1361
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1362
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1363
                int mv_l0[2];
1364
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1365
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1366
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1367
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1368
                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);
1369
            }
1370
        }else{
1371
            for(i8=0; i8<4; i8++){
1372
                const int x8 = i8&1;
1373
                const int y8 = i8>>1;
1374
                int ref0, dist_scale_factor;
1375
                const int16_t (*l1mv)[2]= l1mv0;
1376

    
1377
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1378
                    continue;
1379
                h->sub_mb_type[i8] = sub_mb_type;
1380
                if(IS_INTRA(mb_type_col)){
1381
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1382
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1383
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1384
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1385
                    continue;
1386
                }
1387

    
1388
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1389
                if(ref0 >= 0)
1390
                    ref0 = h->map_col_to_list0[0][ref0];
1391
                else{
1392
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1393
                    l1mv= l1mv1;
1394
                }
1395
                dist_scale_factor = h->dist_scale_factor[ref0];
1396

    
1397
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1398
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1399
                if(IS_SUB_8X8(sub_mb_type)){
1400
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1401
                    int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
1402
                    int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
1403
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1404
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1405
                }else
1406
                for(i4=0; i4<4; i4++){
1407
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1408
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1409
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1410
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1411
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1412
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1413
                }
1414
            }
1415
        }
1416
    }
1417
}
1418

    
1419
static inline void write_back_motion(H264Context *h, int mb_type){
1420
    MpegEncContext * const s = &h->s;
1421
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1422
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1423
    int list;
1424

    
1425
    for(list=0; list<2; list++){
1426
        int y;
1427
        if(!USES_LIST(mb_type, list))
1428
            continue;
1429

    
1430
        for(y=0; y<4; y++){
1431
            *(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];
1432
            *(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];
1433
        }
1434
        if( h->pps.cabac ) {
1435
            for(y=0; y<4; y++){
1436
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1437
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1438
            }
1439
        }
1440

    
1441
        {
1442
            uint8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1443
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1444
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1445
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1446
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1447
        }
1448
    }
1449

    
1450
    if(h->slice_type == B_TYPE && h->pps.cabac){
1451
        if(IS_8X8(mb_type)){
1452
            uint8_t *direct_table = &h->direct_table[b8_xy];
1453
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1454
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1455
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1456
        }
1457
    }
1458
}
1459

    
1460
/**
1461
 * Decodes a network abstraction layer unit.
1462
 * @param consumed is the number of bytes used as input
1463
 * @param length is the length of the array
1464
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1465
 * @returns decoded bytes, might be src+1 if no escapes
1466
 */
1467
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1468
    int i, si, di;
1469
    uint8_t *dst;
1470

    
1471
//    src[0]&0x80;                //forbidden bit
1472
    h->nal_ref_idc= src[0]>>5;
1473
    h->nal_unit_type= src[0]&0x1F;
1474

    
1475
    src++; length--;
1476
#if 0
1477
    for(i=0; i<length; i++)
1478
        printf("%2X ", src[i]);
1479
#endif
1480
    for(i=0; i+1<length; i+=2){
1481
        if(src[i]) continue;
1482
        if(i>0 && src[i-1]==0) i--;
1483
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1484
            if(src[i+2]!=3){
1485
                /* startcode, so we must be past the end */
1486
                length=i;
1487
            }
1488
            break;
1489
        }
1490
    }
1491

    
1492
    if(i>=length-1){ //no escaped 0
1493
        *dst_length= length;
1494
        *consumed= length+1; //+1 for the header
1495
        return src;
1496
    }
1497

    
1498
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1499
    dst= h->rbsp_buffer;
1500

    
1501
//printf("decoding esc\n");
1502
    si=di=0;
1503
    while(si<length){
1504
        //remove escapes (very rare 1:2^22)
1505
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1506
            if(src[si+2]==3){ //escape
1507
                dst[di++]= 0;
1508
                dst[di++]= 0;
1509
                si+=3;
1510
                continue;
1511
            }else //next start code
1512
                break;
1513
        }
1514

    
1515
        dst[di++]= src[si++];
1516
    }
1517

    
1518
    *dst_length= di;
1519
    *consumed= si + 1;//+1 for the header
1520
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1521
    return dst;
1522
}
1523

    
1524
#if 0
1525
/**
1526
 * @param src the data which should be escaped
1527
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1528
 * @param length the length of the src data
1529
 * @param dst_length the length of the dst array
1530
 * @returns length of escaped data in bytes or -1 if an error occured
1531
 */
1532
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1533
    int i, escape_count, si, di;
1534
    uint8_t *temp;
1535

1536
    assert(length>=0);
1537
    assert(dst_length>0);
1538

1539
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1540

1541
    if(length==0) return 1;
1542

1543
    escape_count= 0;
1544
    for(i=0; i<length; i+=2){
1545
        if(src[i]) continue;
1546
        if(i>0 && src[i-1]==0)
1547
            i--;
1548
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1549
            escape_count++;
1550
            i+=2;
1551
        }
1552
    }
1553

1554
    if(escape_count==0){
1555
        if(dst+1 != src)
1556
            memcpy(dst+1, src, length);
1557
        return length + 1;
1558
    }
1559

1560
    if(length + escape_count + 1> dst_length)
1561
        return -1;
1562

1563
    //this should be damn rare (hopefully)
1564

1565
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1566
    temp= h->rbsp_buffer;
1567
//printf("encoding esc\n");
1568

1569
    si= 0;
1570
    di= 0;
1571
    while(si < length){
1572
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1573
            temp[di++]= 0; si++;
1574
            temp[di++]= 0; si++;
1575
            temp[di++]= 3;
1576
            temp[di++]= src[si++];
1577
        }
1578
        else
1579
            temp[di++]= src[si++];
1580
    }
1581
    memcpy(dst+1, temp, length+escape_count);
1582

1583
    assert(di == length+escape_count);
1584

1585
    return di + 1;
1586
}
1587

1588
/**
1589
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1590
 */
1591
static void encode_rbsp_trailing(PutBitContext *pb){
1592
    int length;
1593
    put_bits(pb, 1, 1);
1594
    length= (-put_bits_count(pb))&7;
1595
    if(length) put_bits(pb, length, 0);
1596
}
1597
#endif
1598

    
1599
/**
1600
 * identifies the exact end of the bitstream
1601
 * @return the length of the trailing, or 0 if damaged
1602
 */
1603
static int decode_rbsp_trailing(uint8_t *src){
1604
    int v= *src;
1605
    int r;
1606

    
1607
    tprintf("rbsp trailing %X\n", v);
1608

    
1609
    for(r=1; r<9; r++){
1610
        if(v&1) return r;
1611
        v>>=1;
1612
    }
1613
    return 0;
1614
}
1615

    
1616
/**
1617
 * idct tranforms the 16 dc values and dequantize them.
1618
 * @param qp quantization parameter
1619
 */
1620
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1621
#define stride 16
1622
    int i;
1623
    int temp[16]; //FIXME check if this is a good idea
1624
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1625
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1626

    
1627
//memset(block, 64, 2*256);
1628
//return;
1629
    for(i=0; i<4; i++){
1630
        const int offset= y_offset[i];
1631
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1632
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1633
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1634
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1635

    
1636
        temp[4*i+0]= z0+z3;
1637
        temp[4*i+1]= z1+z2;
1638
        temp[4*i+2]= z1-z2;
1639
        temp[4*i+3]= z0-z3;
1640
    }
1641

    
1642
    for(i=0; i<4; i++){
1643
        const int offset= x_offset[i];
1644
        const int z0= temp[4*0+i] + temp[4*2+i];
1645
        const int z1= temp[4*0+i] - temp[4*2+i];
1646
        const int z2= temp[4*1+i] - temp[4*3+i];
1647
        const int z3= temp[4*1+i] + temp[4*3+i];
1648

    
1649
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1650
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1651
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1652
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1653
    }
1654
}
1655

    
1656
#if 0
1657
/**
1658
 * dct tranforms the 16 dc values.
1659
 * @param qp quantization parameter ??? FIXME
1660
 */
1661
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1662
//    const int qmul= dequant_coeff[qp][0];
1663
    int i;
1664
    int temp[16]; //FIXME check if this is a good idea
1665
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1666
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1667

1668
    for(i=0; i<4; i++){
1669
        const int offset= y_offset[i];
1670
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1671
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1672
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1673
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1674

1675
        temp[4*i+0]= z0+z3;
1676
        temp[4*i+1]= z1+z2;
1677
        temp[4*i+2]= z1-z2;
1678
        temp[4*i+3]= z0-z3;
1679
    }
1680

1681
    for(i=0; i<4; i++){
1682
        const int offset= x_offset[i];
1683
        const int z0= temp[4*0+i] + temp[4*2+i];
1684
        const int z1= temp[4*0+i] - temp[4*2+i];
1685
        const int z2= temp[4*1+i] - temp[4*3+i];
1686
        const int z3= temp[4*1+i] + temp[4*3+i];
1687

1688
        block[stride*0 +offset]= (z0 + z3)>>1;
1689
        block[stride*2 +offset]= (z1 + z2)>>1;
1690
        block[stride*8 +offset]= (z1 - z2)>>1;
1691
        block[stride*10+offset]= (z0 - z3)>>1;
1692
    }
1693
}
1694
#endif
1695

    
1696
#undef xStride
1697
#undef stride
1698

    
1699
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1700
    const int stride= 16*2;
1701
    const int xStride= 16;
1702
    int a,b,c,d,e;
1703

    
1704
    a= block[stride*0 + xStride*0];
1705
    b= block[stride*0 + xStride*1];
1706
    c= block[stride*1 + xStride*0];
1707
    d= block[stride*1 + xStride*1];
1708

    
1709
    e= a-b;
1710
    a= a+b;
1711
    b= c-d;
1712
    c= c+d;
1713

    
1714
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1715
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1716
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1717
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1718
}
1719

    
1720
#if 0
1721
static void chroma_dc_dct_c(DCTELEM *block){
1722
    const int stride= 16*2;
1723
    const int xStride= 16;
1724
    int a,b,c,d,e;
1725

1726
    a= block[stride*0 + xStride*0];
1727
    b= block[stride*0 + xStride*1];
1728
    c= block[stride*1 + xStride*0];
1729
    d= block[stride*1 + xStride*1];
1730

1731
    e= a-b;
1732
    a= a+b;
1733
    b= c-d;
1734
    c= c+d;
1735

1736
    block[stride*0 + xStride*0]= (a+c);
1737
    block[stride*0 + xStride*1]= (e+b);
1738
    block[stride*1 + xStride*0]= (a-c);
1739
    block[stride*1 + xStride*1]= (e-b);
1740
}
1741
#endif
1742

    
1743
/**
1744
 * gets the chroma qp.
1745
 */
1746
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1747

    
1748
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1749
}
1750

    
1751

    
1752
#if 0
1753
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1754
    int i;
1755
    //FIXME try int temp instead of block
1756

1757
    for(i=0; i<4; i++){
1758
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1759
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1760
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1761
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1762
        const int z0= d0 + d3;
1763
        const int z3= d0 - d3;
1764
        const int z1= d1 + d2;
1765
        const int z2= d1 - d2;
1766

1767
        block[0 + 4*i]=   z0 +   z1;
1768
        block[1 + 4*i]= 2*z3 +   z2;
1769
        block[2 + 4*i]=   z0 -   z1;
1770
        block[3 + 4*i]=   z3 - 2*z2;
1771
    }
1772

1773
    for(i=0; i<4; i++){
1774
        const int z0= block[0*4 + i] + block[3*4 + i];
1775
        const int z3= block[0*4 + i] - block[3*4 + i];
1776
        const int z1= block[1*4 + i] + block[2*4 + i];
1777
        const int z2= block[1*4 + i] - block[2*4 + i];
1778

1779
        block[0*4 + i]=   z0 +   z1;
1780
        block[1*4 + i]= 2*z3 +   z2;
1781
        block[2*4 + i]=   z0 -   z1;
1782
        block[3*4 + i]=   z3 - 2*z2;
1783
    }
1784
}
1785
#endif
1786

    
1787
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1788
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1789
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1790
    int i;
1791
    const int * const quant_table= quant_coeff[qscale];
1792
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1793
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1794
    const unsigned int threshold2= (threshold1<<1);
1795
    int last_non_zero;
1796

    
1797
    if(seperate_dc){
1798
        if(qscale<=18){
1799
            //avoid overflows
1800
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1801
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1802
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1803

    
1804
            int level= block[0]*quant_coeff[qscale+18][0];
1805
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1806
                if(level>0){
1807
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1808
                    block[0]= level;
1809
                }else{
1810
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1811
                    block[0]= -level;
1812
                }
1813
//                last_non_zero = i;
1814
            }else{
1815
                block[0]=0;
1816
            }
1817
        }else{
1818
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1819
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1820
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1821

    
1822
            int level= block[0]*quant_table[0];
1823
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1824
                if(level>0){
1825
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1826
                    block[0]= level;
1827
                }else{
1828
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1829
                    block[0]= -level;
1830
                }
1831
//                last_non_zero = i;
1832
            }else{
1833
                block[0]=0;
1834
            }
1835
        }
1836
        last_non_zero= 0;
1837
        i=1;
1838
    }else{
1839
        last_non_zero= -1;
1840
        i=0;
1841
    }
1842

    
1843
    for(; i<16; i++){
1844
        const int j= scantable[i];
1845
        int level= block[j]*quant_table[j];
1846

    
1847
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1848
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1849
        if(((unsigned)(level+threshold1))>threshold2){
1850
            if(level>0){
1851
                level= (bias + level)>>QUANT_SHIFT;
1852
                block[j]= level;
1853
            }else{
1854
                level= (bias - level)>>QUANT_SHIFT;
1855
                block[j]= -level;
1856
            }
1857
            last_non_zero = i;
1858
        }else{
1859
            block[j]=0;
1860
        }
1861
    }
1862

    
1863
    return last_non_zero;
1864
}
1865

    
1866
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1867
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1868
    ((uint32_t*)(src+0*stride))[0]= a;
1869
    ((uint32_t*)(src+1*stride))[0]= a;
1870
    ((uint32_t*)(src+2*stride))[0]= a;
1871
    ((uint32_t*)(src+3*stride))[0]= a;
1872
}
1873

    
1874
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1875
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1876
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1877
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1878
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1879
}
1880

    
1881
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1882
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1883
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1884

    
1885
    ((uint32_t*)(src+0*stride))[0]=
1886
    ((uint32_t*)(src+1*stride))[0]=
1887
    ((uint32_t*)(src+2*stride))[0]=
1888
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1889
}
1890

    
1891
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1892
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1893

    
1894
    ((uint32_t*)(src+0*stride))[0]=
1895
    ((uint32_t*)(src+1*stride))[0]=
1896
    ((uint32_t*)(src+2*stride))[0]=
1897
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1898
}
1899

    
1900
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1901
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1902

    
1903
    ((uint32_t*)(src+0*stride))[0]=
1904
    ((uint32_t*)(src+1*stride))[0]=
1905
    ((uint32_t*)(src+2*stride))[0]=
1906
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1907
}
1908

    
1909
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1910
    ((uint32_t*)(src+0*stride))[0]=
1911
    ((uint32_t*)(src+1*stride))[0]=
1912
    ((uint32_t*)(src+2*stride))[0]=
1913
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1914
}
1915

    
1916

    
1917
#define LOAD_TOP_RIGHT_EDGE\
1918
    const int t4= topright[0];\
1919
    const int t5= topright[1];\
1920
    const int t6= topright[2];\
1921
    const int t7= topright[3];\
1922

    
1923
#define LOAD_LEFT_EDGE\
1924
    const int l0= src[-1+0*stride];\
1925
    const int l1= src[-1+1*stride];\
1926
    const int l2= src[-1+2*stride];\
1927
    const int l3= src[-1+3*stride];\
1928

    
1929
#define LOAD_TOP_EDGE\
1930
    const int t0= src[ 0-1*stride];\
1931
    const int t1= src[ 1-1*stride];\
1932
    const int t2= src[ 2-1*stride];\
1933
    const int t3= src[ 3-1*stride];\
1934

    
1935
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1936
    const int lt= src[-1-1*stride];
1937
    LOAD_TOP_EDGE
1938
    LOAD_LEFT_EDGE
1939

    
1940
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1941
    src[0+2*stride]=
1942
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1943
    src[0+1*stride]=
1944
    src[1+2*stride]=
1945
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1946
    src[0+0*stride]=
1947
    src[1+1*stride]=
1948
    src[2+2*stride]=
1949
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1950
    src[1+0*stride]=
1951
    src[2+1*stride]=
1952
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1953
    src[2+0*stride]=
1954
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1955
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1956
}
1957

    
1958
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1959
    LOAD_TOP_EDGE
1960
    LOAD_TOP_RIGHT_EDGE
1961
//    LOAD_LEFT_EDGE
1962

    
1963
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1964
    src[1+0*stride]=
1965
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1966
    src[2+0*stride]=
1967
    src[1+1*stride]=
1968
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1969
    src[3+0*stride]=
1970
    src[2+1*stride]=
1971
    src[1+2*stride]=
1972
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1973
    src[3+1*stride]=
1974
    src[2+2*stride]=
1975
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1976
    src[3+2*stride]=
1977
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1978
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1979
}
1980

    
1981
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1982
    const int lt= src[-1-1*stride];
1983
    LOAD_TOP_EDGE
1984
    LOAD_LEFT_EDGE
1985
    const __attribute__((unused)) int unu= l3;
1986

    
1987
    src[0+0*stride]=
1988
    src[1+2*stride]=(lt + t0 + 1)>>1;
1989
    src[1+0*stride]=
1990
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1991
    src[2+0*stride]=
1992
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1993
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1994
    src[0+1*stride]=
1995
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1996
    src[1+1*stride]=
1997
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1998
    src[2+1*stride]=
1999
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2000
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2001
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2002
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2003
}
2004

    
2005
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
2006
    LOAD_TOP_EDGE
2007
    LOAD_TOP_RIGHT_EDGE
2008
    const __attribute__((unused)) int unu= t7;
2009

    
2010
    src[0+0*stride]=(t0 + t1 + 1)>>1;
2011
    src[1+0*stride]=
2012
    src[0+2*stride]=(t1 + t2 + 1)>>1;
2013
    src[2+0*stride]=
2014
    src[1+2*stride]=(t2 + t3 + 1)>>1;
2015
    src[3+0*stride]=
2016
    src[2+2*stride]=(t3 + t4+ 1)>>1;
2017
    src[3+2*stride]=(t4 + t5+ 1)>>1;
2018
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2019
    src[1+1*stride]=
2020
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2021
    src[2+1*stride]=
2022
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2023
    src[3+1*stride]=
2024
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2025
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2026
}
2027

    
2028
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2029
    LOAD_LEFT_EDGE
2030

    
2031
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2032
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2033
    src[2+0*stride]=
2034
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2035
    src[3+0*stride]=
2036
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2037
    src[2+1*stride]=
2038
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2039
    src[3+1*stride]=
2040
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2041
    src[3+2*stride]=
2042
    src[1+3*stride]=
2043
    src[0+3*stride]=
2044
    src[2+2*stride]=
2045
    src[2+3*stride]=
2046
    src[3+3*stride]=l3;
2047
}
2048

    
2049
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2050
    const int lt= src[-1-1*stride];
2051
    LOAD_TOP_EDGE
2052
    LOAD_LEFT_EDGE
2053
    const __attribute__((unused)) int unu= t3;
2054

    
2055
    src[0+0*stride]=
2056
    src[2+1*stride]=(lt + l0 + 1)>>1;
2057
    src[1+0*stride]=
2058
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2059
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2060
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2061
    src[0+1*stride]=
2062
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2063
    src[1+1*stride]=
2064
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2065
    src[0+2*stride]=
2066
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2067
    src[1+2*stride]=
2068
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2069
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2070
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2071
}
2072

    
2073
static void pred16x16_vertical_c(uint8_t *src, int stride){
2074
    int i;
2075
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2076
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2077
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2078
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2079

    
2080
    for(i=0; i<16; i++){
2081
        ((uint32_t*)(src+i*stride))[0]= a;
2082
        ((uint32_t*)(src+i*stride))[1]= b;
2083
        ((uint32_t*)(src+i*stride))[2]= c;
2084
        ((uint32_t*)(src+i*stride))[3]= d;
2085
    }
2086
}
2087

    
2088
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2089
    int i;
2090

    
2091
    for(i=0; i<16; i++){
2092
        ((uint32_t*)(src+i*stride))[0]=
2093
        ((uint32_t*)(src+i*stride))[1]=
2094
        ((uint32_t*)(src+i*stride))[2]=
2095
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2096
    }
2097
}
2098

    
2099
static void pred16x16_dc_c(uint8_t *src, int stride){
2100
    int i, dc=0;
2101

    
2102
    for(i=0;i<16; i++){
2103
        dc+= src[-1+i*stride];
2104
    }
2105

    
2106
    for(i=0;i<16; i++){
2107
        dc+= src[i-stride];
2108
    }
2109

    
2110
    dc= 0x01010101*((dc + 16)>>5);
2111

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

    
2120
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2121
    int i, dc=0;
2122

    
2123
    for(i=0;i<16; i++){
2124
        dc+= src[-1+i*stride];
2125
    }
2126

    
2127
    dc= 0x01010101*((dc + 8)>>4);
2128

    
2129
    for(i=0; i<16; i++){
2130
        ((uint32_t*)(src+i*stride))[0]=
2131
        ((uint32_t*)(src+i*stride))[1]=
2132
        ((uint32_t*)(src+i*stride))[2]=
2133
        ((uint32_t*)(src+i*stride))[3]= dc;
2134
    }
2135
}
2136

    
2137
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2138
    int i, dc=0;
2139

    
2140
    for(i=0;i<16; i++){
2141
        dc+= src[i-stride];
2142
    }
2143
    dc= 0x01010101*((dc + 8)>>4);
2144

    
2145
    for(i=0; i<16; i++){
2146
        ((uint32_t*)(src+i*stride))[0]=
2147
        ((uint32_t*)(src+i*stride))[1]=
2148
        ((uint32_t*)(src+i*stride))[2]=
2149
        ((uint32_t*)(src+i*stride))[3]= dc;
2150
    }
2151
}
2152

    
2153
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2154
    int i;
2155

    
2156
    for(i=0; i<16; i++){
2157
        ((uint32_t*)(src+i*stride))[0]=
2158
        ((uint32_t*)(src+i*stride))[1]=
2159
        ((uint32_t*)(src+i*stride))[2]=
2160
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2161
    }
2162
}
2163

    
2164
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2165
  int i, j, k;
2166
  int a;
2167
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2168
  const uint8_t * const src0 = src+7-stride;
2169
  const uint8_t *src1 = src+8*stride-1;
2170
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2171
  int H = src0[1] - src0[-1];
2172
  int V = src1[0] - src2[ 0];
2173
  for(k=2; k<=8; ++k) {
2174
    src1 += stride; src2 -= stride;
2175
    H += k*(src0[k] - src0[-k]);
2176
    V += k*(src1[0] - src2[ 0]);
2177
  }
2178
  if(svq3){
2179
    H = ( 5*(H/4) ) / 16;
2180
    V = ( 5*(V/4) ) / 16;
2181

    
2182
    /* required for 100% accuracy */
2183
    i = H; H = V; V = i;
2184
  }else{
2185
    H = ( 5*H+32 ) >> 6;
2186
    V = ( 5*V+32 ) >> 6;
2187
  }
2188

    
2189
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2190
  for(j=16; j>0; --j) {
2191
    int b = a;
2192
    a += V;
2193
    for(i=-16; i<0; i+=4) {
2194
      src[16+i] = cm[ (b    ) >> 5 ];
2195
      src[17+i] = cm[ (b+  H) >> 5 ];
2196
      src[18+i] = cm[ (b+2*H) >> 5 ];
2197
      src[19+i] = cm[ (b+3*H) >> 5 ];
2198
      b += 4*H;
2199
    }
2200
    src += stride;
2201
  }
2202
}
2203

    
2204
static void pred16x16_plane_c(uint8_t *src, int stride){
2205
    pred16x16_plane_compat_c(src, stride, 0);
2206
}
2207

    
2208
static void pred8x8_vertical_c(uint8_t *src, int stride){
2209
    int i;
2210
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2211
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2212

    
2213
    for(i=0; i<8; i++){
2214
        ((uint32_t*)(src+i*stride))[0]= a;
2215
        ((uint32_t*)(src+i*stride))[1]= b;
2216
    }
2217
}
2218

    
2219
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2220
    int i;
2221

    
2222
    for(i=0; i<8; i++){
2223
        ((uint32_t*)(src+i*stride))[0]=
2224
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2225
    }
2226
}
2227

    
2228
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2229
    int i;
2230

    
2231
    for(i=0; i<8; i++){
2232
        ((uint32_t*)(src+i*stride))[0]=
2233
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2234
    }
2235
}
2236

    
2237
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2238
    int i;
2239
    int dc0, dc2;
2240

    
2241
    dc0=dc2=0;
2242
    for(i=0;i<4; i++){
2243
        dc0+= src[-1+i*stride];
2244
        dc2+= src[-1+(i+4)*stride];
2245
    }
2246
    dc0= 0x01010101*((dc0 + 2)>>2);
2247
    dc2= 0x01010101*((dc2 + 2)>>2);
2248

    
2249
    for(i=0; i<4; i++){
2250
        ((uint32_t*)(src+i*stride))[0]=
2251
        ((uint32_t*)(src+i*stride))[1]= dc0;
2252
    }
2253
    for(i=4; i<8; i++){
2254
        ((uint32_t*)(src+i*stride))[0]=
2255
        ((uint32_t*)(src+i*stride))[1]= dc2;
2256
    }
2257
}
2258

    
2259
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2260
    int i;
2261
    int dc0, dc1;
2262

    
2263
    dc0=dc1=0;
2264
    for(i=0;i<4; i++){
2265
        dc0+= src[i-stride];
2266
        dc1+= src[4+i-stride];
2267
    }
2268
    dc0= 0x01010101*((dc0 + 2)>>2);
2269
    dc1= 0x01010101*((dc1 + 2)>>2);
2270

    
2271
    for(i=0; i<4; i++){
2272
        ((uint32_t*)(src+i*stride))[0]= dc0;
2273
        ((uint32_t*)(src+i*stride))[1]= dc1;
2274
    }
2275
    for(i=4; i<8; i++){
2276
        ((uint32_t*)(src+i*stride))[0]= dc0;
2277
        ((uint32_t*)(src+i*stride))[1]= dc1;
2278
    }
2279
}
2280

    
2281

    
2282
static void pred8x8_dc_c(uint8_t *src, int stride){
2283
    int i;
2284
    int dc0, dc1, dc2, dc3;
2285

    
2286
    dc0=dc1=dc2=0;
2287
    for(i=0;i<4; i++){
2288
        dc0+= src[-1+i*stride] + src[i-stride];
2289
        dc1+= src[4+i-stride];
2290
        dc2+= src[-1+(i+4)*stride];
2291
    }
2292
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2293
    dc0= 0x01010101*((dc0 + 4)>>3);
2294
    dc1= 0x01010101*((dc1 + 2)>>2);
2295
    dc2= 0x01010101*((dc2 + 2)>>2);
2296

    
2297
    for(i=0; i<4; i++){
2298
        ((uint32_t*)(src+i*stride))[0]= dc0;
2299
        ((uint32_t*)(src+i*stride))[1]= dc1;
2300
    }
2301
    for(i=4; i<8; i++){
2302
        ((uint32_t*)(src+i*stride))[0]= dc2;
2303
        ((uint32_t*)(src+i*stride))[1]= dc3;
2304
    }
2305
}
2306

    
2307
static void pred8x8_plane_c(uint8_t *src, int stride){
2308
  int j, k;
2309
  int a;
2310
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2311
  const uint8_t * const src0 = src+3-stride;
2312
  const uint8_t *src1 = src+4*stride-1;
2313
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2314
  int H = src0[1] - src0[-1];
2315
  int V = src1[0] - src2[ 0];
2316
  for(k=2; k<=4; ++k) {
2317
    src1 += stride; src2 -= stride;
2318
    H += k*(src0[k] - src0[-k]);
2319
    V += k*(src1[0] - src2[ 0]);
2320
  }
2321
  H = ( 17*H+16 ) >> 5;
2322
  V = ( 17*V+16 ) >> 5;
2323

    
2324
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2325
  for(j=8; j>0; --j) {
2326
    int b = a;
2327
    a += V;
2328
    src[0] = cm[ (b    ) >> 5 ];
2329
    src[1] = cm[ (b+  H) >> 5 ];
2330
    src[2] = cm[ (b+2*H) >> 5 ];
2331
    src[3] = cm[ (b+3*H) >> 5 ];
2332
    src[4] = cm[ (b+4*H) >> 5 ];
2333
    src[5] = cm[ (b+5*H) >> 5 ];
2334
    src[6] = cm[ (b+6*H) >> 5 ];
2335
    src[7] = cm[ (b+7*H) >> 5 ];
2336
    src += stride;
2337
  }
2338
}
2339

    
2340
#define SRC(x,y) src[(x)+(y)*stride]
2341
#define PL(y) \
2342
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2343
#define PREDICT_8x8_LOAD_LEFT \
2344
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2345
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2346
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2347
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2348

    
2349
#define PT(x) \
2350
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2351
#define PREDICT_8x8_LOAD_TOP \
2352
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2353
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2354
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2355
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2356
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2357

    
2358
#define PTR(x) \
2359
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2360
#define PREDICT_8x8_LOAD_TOPRIGHT \
2361
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2362
    if(has_topright) { \
2363
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2364
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2365
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2366

    
2367
#define PREDICT_8x8_LOAD_TOPLEFT \
2368
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2369

    
2370
#define PREDICT_8x8_DC(v) \
2371
    int y; \
2372
    for( y = 0; y < 8; y++ ) { \
2373
        ((uint32_t*)src)[0] = \
2374
        ((uint32_t*)src)[1] = v; \
2375
        src += stride; \
2376
    }
2377

    
2378
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2379
{
2380
    PREDICT_8x8_DC(0x80808080);
2381
}
2382
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2383
{
2384
    PREDICT_8x8_LOAD_LEFT;
2385
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2386
    PREDICT_8x8_DC(dc);
2387
}
2388
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2389
{
2390
    PREDICT_8x8_LOAD_TOP;
2391
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2392
    PREDICT_8x8_DC(dc);
2393
}
2394
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2395
{
2396
    PREDICT_8x8_LOAD_LEFT;
2397
    PREDICT_8x8_LOAD_TOP;
2398
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2399
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2400
    PREDICT_8x8_DC(dc);
2401
}
2402
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2403
{
2404
    PREDICT_8x8_LOAD_LEFT;
2405
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2406
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2407
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2408
#undef ROW
2409
}
2410
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2411
{
2412
    int y;
2413
    PREDICT_8x8_LOAD_TOP;
2414
    src[0] = t0;
2415
    src[1] = t1;
2416
    src[2] = t2;
2417
    src[3] = t3;
2418
    src[4] = t4;
2419
    src[5] = t5;
2420
    src[6] = t6;
2421
    src[7] = t7;
2422
    for( y = 1; y < 8; y++ )
2423
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2424
}
2425
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2426
{
2427
    PREDICT_8x8_LOAD_TOP;
2428
    PREDICT_8x8_LOAD_TOPRIGHT;
2429
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2430
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2431
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2432
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2433
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2434
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2435
    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;
2436
    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;
2437
    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;
2438
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2439
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2440
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2441
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2442
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2443
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2444
}
2445
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2446
{
2447
    PREDICT_8x8_LOAD_TOP;
2448
    PREDICT_8x8_LOAD_LEFT;
2449
    PREDICT_8x8_LOAD_TOPLEFT;
2450
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2451
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2452
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2453
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2454
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2455
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2456
    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;
2457
    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;
2458
    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;
2459
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2460
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2461
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2462
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2463
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2464
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2465

    
2466
}
2467
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2468
{
2469
    PREDICT_8x8_LOAD_TOP;
2470
    PREDICT_8x8_LOAD_LEFT;
2471
    PREDICT_8x8_LOAD_TOPLEFT;
2472
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2473
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2474
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2475
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2476
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2477
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2478
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2479
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2480
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2481
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2482
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2483
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2484
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2485
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2486
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2487
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2488
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2489
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2490
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2491
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2492
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2493
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2494
}
2495
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2496
{
2497
    PREDICT_8x8_LOAD_TOP;
2498
    PREDICT_8x8_LOAD_LEFT;
2499
    PREDICT_8x8_LOAD_TOPLEFT;
2500
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2501
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2502
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2503
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2504
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2505
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2506
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2507
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2508
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2509
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2510
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2511
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2512
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2513
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2514
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2515
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2516
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2517
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2518
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2519
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2520
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2521
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2522
}
2523
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2524
{
2525
    PREDICT_8x8_LOAD_TOP;
2526
    PREDICT_8x8_LOAD_TOPRIGHT;
2527
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2528
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2529
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2530
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2531
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2532
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2533
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2534
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2535
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2536
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2537
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2538
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2539
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2540
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2541
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2542
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2543
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2544
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2545
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2546
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2547
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2548
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2549
}
2550
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2551
{
2552
    PREDICT_8x8_LOAD_LEFT;
2553
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2554
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2555
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2556
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2557
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2558
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2559
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2560
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2561
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2562
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2563
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2564
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2565
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2566
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2567
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2568
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2569
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2570
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2571
}
2572
#undef PREDICT_8x8_LOAD_LEFT
2573
#undef PREDICT_8x8_LOAD_TOP
2574
#undef PREDICT_8x8_LOAD_TOPLEFT
2575
#undef PREDICT_8x8_LOAD_TOPRIGHT
2576
#undef PREDICT_8x8_DC
2577
#undef PTR
2578
#undef PT
2579
#undef PL
2580
#undef SRC
2581

    
2582
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2583
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2584
                           int src_x_offset, int src_y_offset,
2585
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2586
    MpegEncContext * const s = &h->s;
2587
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2588
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2589
    const int luma_xy= (mx&3) + ((my&3)<<2);
2590
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2591
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2592
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2593
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2594
    int extra_height= extra_width;
2595
    int emu=0;
2596
    const int full_mx= mx>>2;
2597
    const int full_my= my>>2;
2598
    const int pic_width  = 16*s->mb_width;
2599
    const int pic_height = 16*s->mb_height;
2600

    
2601
    if(!pic->data[0])
2602
        return;
2603

    
2604
    if(mx&7) extra_width -= 3;
2605
    if(my&7) extra_height -= 3;
2606

    
2607
    if(   full_mx < 0-extra_width
2608
       || full_my < 0-extra_height
2609
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2610
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2611
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
2612
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2613
        emu=1;
2614
    }
2615

    
2616
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2617
    if(!square){
2618
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2619
    }
2620

    
2621
    if(s->flags&CODEC_FLAG_GRAY) return;
2622

    
2623
    if(emu){
2624
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2625
            src_cb= s->edge_emu_buffer;
2626
    }
2627
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2628

    
2629
    if(emu){
2630
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2631
            src_cr= s->edge_emu_buffer;
2632
    }
2633
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2634
}
2635

    
2636
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2637
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2638
                           int x_offset, int y_offset,
2639
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2640
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2641
                           int list0, int list1){
2642
    MpegEncContext * const s = &h->s;
2643
    qpel_mc_func *qpix_op=  qpix_put;
2644
    h264_chroma_mc_func chroma_op= chroma_put;
2645

    
2646
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2647
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2648
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2649
    x_offset += 8*s->mb_x;
2650
    y_offset += 8*s->mb_y;
2651

    
2652
    if(list0){
2653
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2654
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2655
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2656
                           qpix_op, chroma_op);
2657

    
2658
        qpix_op=  qpix_avg;
2659
        chroma_op= chroma_avg;
2660
    }
2661

    
2662
    if(list1){
2663
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2664
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2665
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2666
                           qpix_op, chroma_op);
2667
    }
2668
}
2669

    
2670
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2671
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2672
                           int x_offset, int y_offset,
2673
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2674
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2675
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2676
                           int list0, int list1){
2677
    MpegEncContext * const s = &h->s;
2678

    
2679
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2680
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2681
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2682
    x_offset += 8*s->mb_x;
2683
    y_offset += 8*s->mb_y;
2684

    
2685
    if(list0 && list1){
2686
        /* don't optimize for luma-only case, since B-frames usually
2687
         * use implicit weights => chroma too. */
2688
        uint8_t *tmp_cb = s->obmc_scratchpad;
2689
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2690
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2691
        int refn0 = h->ref_cache[0][ scan8[n] ];
2692
        int refn1 = h->ref_cache[1][ scan8[n] ];
2693

    
2694
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2695
                    dest_y, dest_cb, dest_cr,
2696
                    x_offset, y_offset, qpix_put, chroma_put);
2697
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2698
                    tmp_y, tmp_cb, tmp_cr,
2699
                    x_offset, y_offset, qpix_put, chroma_put);
2700

    
2701
        if(h->use_weight == 2){
2702
            int weight0 = h->implicit_weight[refn0][refn1];
2703
            int weight1 = 64 - weight0;
2704
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0);
2705
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
2706
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
2707
        }else{
2708
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2709
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2710
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2711
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2712
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2713
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2714
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2715
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2716
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2717
        }
2718
    }else{
2719
        int list = list1 ? 1 : 0;
2720
        int refn = h->ref_cache[list][ scan8[n] ];
2721
        Picture *ref= &h->ref_list[list][refn];
2722
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2723
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2724
                    qpix_put, chroma_put);
2725

    
2726
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2727
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2728
        if(h->use_weight_chroma){
2729
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2730
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2731
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2732
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2733
        }
2734
    }
2735
}
2736

    
2737
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2738
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2739
                           int x_offset, int y_offset,
2740
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2741
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2742
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2743
                           int list0, int list1){
2744
    if((h->use_weight==2 && list0 && list1
2745
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2746
       || h->use_weight==1)
2747
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2748
                         x_offset, y_offset, qpix_put, chroma_put,
2749
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2750
    else
2751
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2752
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2753
}
2754

    
2755
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2756
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2757
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2758
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2759
    MpegEncContext * const s = &h->s;
2760
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2761
    const int mb_type= s->current_picture.mb_type[mb_xy];
2762

    
2763
    assert(IS_INTER(mb_type));
2764

    
2765
    if(IS_16X16(mb_type)){
2766
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2767
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2768
                &weight_op[0], &weight_avg[0],
2769
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2770
    }else if(IS_16X8(mb_type)){
2771
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2772
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2773
                &weight_op[1], &weight_avg[1],
2774
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2775
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2776
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2777
                &weight_op[1], &weight_avg[1],
2778
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2779
    }else if(IS_8X16(mb_type)){
2780
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2781
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2782
                &weight_op[2], &weight_avg[2],
2783
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2784
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2785
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2786
                &weight_op[2], &weight_avg[2],
2787
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2788
    }else{
2789
        int i;
2790

    
2791
        assert(IS_8X8(mb_type));
2792

    
2793
        for(i=0; i<4; i++){
2794
            const int sub_mb_type= h->sub_mb_type[i];
2795
            const int n= 4*i;
2796
            int x_offset= (i&1)<<2;
2797
            int y_offset= (i&2)<<1;
2798

    
2799
            if(IS_SUB_8X8(sub_mb_type)){
2800
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2801
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2802
                    &weight_op[3], &weight_avg[3],
2803
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2804
            }else if(IS_SUB_8X4(sub_mb_type)){
2805
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2806
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2807
                    &weight_op[4], &weight_avg[4],
2808
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2809
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2810
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2811
                    &weight_op[4], &weight_avg[4],
2812
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2813
            }else if(IS_SUB_4X8(sub_mb_type)){
2814
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2815
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2816
                    &weight_op[5], &weight_avg[5],
2817
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2818
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2819
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2820
                    &weight_op[5], &weight_avg[5],
2821
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2822
            }else{
2823
                int j;
2824
                assert(IS_SUB_4X4(sub_mb_type));
2825
                for(j=0; j<4; j++){
2826
                    int sub_x_offset= x_offset + 2*(j&1);
2827
                    int sub_y_offset= y_offset +   (j&2);
2828
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2829
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2830
                        &weight_op[6], &weight_avg[6],
2831
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2832
                }
2833
            }
2834
        }
2835
    }
2836
}
2837

    
2838
static void decode_init_vlc(H264Context *h){
2839
    static int done = 0;
2840

    
2841
    if (!done) {
2842
        int i;
2843
        done = 1;
2844

    
2845
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2846
                 &chroma_dc_coeff_token_len [0], 1, 1,
2847
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2848

    
2849
        for(i=0; i<4; i++){
2850
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2851
                     &coeff_token_len [i][0], 1, 1,
2852
                     &coeff_token_bits[i][0], 1, 1, 1);
2853
        }
2854

    
2855
        for(i=0; i<3; i++){
2856
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2857
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2858
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2859
        }
2860
        for(i=0; i<15; i++){
2861
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2862
                     &total_zeros_len [i][0], 1, 1,
2863
                     &total_zeros_bits[i][0], 1, 1, 1);
2864
        }
2865

    
2866
        for(i=0; i<6; i++){
2867
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2868
                     &run_len [i][0], 1, 1,
2869
                     &run_bits[i][0], 1, 1, 1);
2870
        }
2871
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2872
                 &run_len [6][0], 1, 1,
2873
                 &run_bits[6][0], 1, 1, 1);
2874
    }
2875
}
2876

    
2877
/**
2878
 * Sets the intra prediction function pointers.
2879
 */
2880
static void init_pred_ptrs(H264Context *h){
2881
//    MpegEncContext * const s = &h->s;
2882

    
2883
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2884
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2885
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2886
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2887
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2888
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2889
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2890
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2891
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2892
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2893
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2894
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2895

    
2896
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2897
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2898
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2899
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2900
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2901
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2902
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2903
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2904
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2905
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2906
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2907
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2908

    
2909
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2910
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2911
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2912
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2913
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2914
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2915
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2916

    
2917
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2918
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2919
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2920
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2921
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2922
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2923
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2924
}
2925

    
2926
static void free_tables(H264Context *h){
2927
    av_freep(&h->intra4x4_pred_mode);
2928
    av_freep(&h->chroma_pred_mode_table);
2929
    av_freep(&h->cbp_table);
2930
    av_freep(&h->mvd_table[0]);
2931
    av_freep(&h->mvd_table[1]);
2932
    av_freep(&h->direct_table);
2933
    av_freep(&h->non_zero_count);
2934
    av_freep(&h->slice_table_base);
2935
    av_freep(&h->top_borders[1]);
2936
    av_freep(&h->top_borders[0]);
2937
    h->slice_table= NULL;
2938

    
2939
    av_freep(&h->mb2b_xy);
2940
    av_freep(&h->mb2b8_xy);
2941

    
2942
    av_freep(&h->s.obmc_scratchpad);
2943
}
2944

    
2945
static void init_dequant8_coeff_table(H264Context *h){
2946
    int i,q,x;
2947
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2948
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2949
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2950

    
2951
    for(i=0; i<2; i++ ){
2952
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2953
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2954
            break;
2955
        }
2956

    
2957
        for(q=0; q<52; q++){
2958
            int shift = div6[q];
2959
            int idx = rem6[q];
2960
            for(x=0; x<64; x++)
2961
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2962
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2963
                    h->pps.scaling_matrix8[i][x]) << shift;
2964
        }
2965
    }
2966
}
2967

    
2968
static void init_dequant4_coeff_table(H264Context *h){
2969
    int i,j,q,x;
2970
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2971
    for(i=0; i<6; i++ ){
2972
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2973
        for(j=0; j<i; j++){
2974
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2975
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2976
                break;
2977
            }
2978
        }
2979
        if(j<i)
2980
            continue;
2981

    
2982
        for(q=0; q<52; q++){
2983
            int shift = div6[q] + 2;
2984
            int idx = rem6[q];
2985
            for(x=0; x<16; x++)
2986
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2987
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2988
                    h->pps.scaling_matrix4[i][x]) << shift;
2989
        }
2990
    }
2991
}
2992

    
2993
static void init_dequant_tables(H264Context *h){
2994
    int i,x;
2995
    init_dequant4_coeff_table(h);
2996
    if(h->pps.transform_8x8_mode)
2997
        init_dequant8_coeff_table(h);
2998
    if(h->sps.transform_bypass){
2999
        for(i=0; i<6; i++)
3000
            for(x=0; x<16; x++)
3001
                h->dequant4_coeff[i][0][x] = 1<<6;
3002
        if(h->pps.transform_8x8_mode)
3003
            for(i=0; i<2; i++)
3004
                for(x=0; x<64; x++)
3005
                    h->dequant8_coeff[i][0][x] = 1<<6;
3006
    }
3007
}
3008

    
3009

    
3010
/**
3011
 * allocates tables.
3012
 * needs width/height
3013
 */
3014
static int alloc_tables(H264Context *h){
3015
    MpegEncContext * const s = &h->s;
3016
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
3017
    int x,y;
3018

    
3019
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
3020

    
3021
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
3022
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
3023
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3024
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3025
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
3026

    
3027
    if( h->pps.cabac ) {
3028
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
3029
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3030
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
3031
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
3032
    }
3033

    
3034
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
3035
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
3036

    
3037
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
3038
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
3039
    for(y=0; y<s->mb_height; y++){
3040
        for(x=0; x<s->mb_width; x++){
3041
            const int mb_xy= x + y*s->mb_stride;
3042
            const int b_xy = 4*x + 4*y*h->b_stride;
3043
            const int b8_xy= 2*x + 2*y*h->b8_stride;
3044

    
3045
            h->mb2b_xy [mb_xy]= b_xy;
3046
            h->mb2b8_xy[mb_xy]= b8_xy;
3047
        }
3048
    }
3049

    
3050
    s->obmc_scratchpad = NULL;
3051

    
3052
    if(!h->dequant4_coeff[0])
3053
        init_dequant_tables(h);
3054

    
3055
    return 0;
3056
fail:
3057
    free_tables(h);
3058
    return -1;
3059
}
3060

    
3061
static void common_init(H264Context *h){
3062
    MpegEncContext * const s = &h->s;
3063

    
3064
    s->width = s->avctx->width;
3065
    s->height = s->avctx->height;
3066
    s->codec_id= s->avctx->codec->id;
3067

    
3068
    init_pred_ptrs(h);
3069

    
3070
    h->dequant_coeff_pps= -1;
3071
    s->unrestricted_mv=1;
3072
    s->decode=1; //FIXME
3073

    
3074
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3075
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
3076
}
3077

    
3078
static int decode_init(AVCodecContext *avctx){
3079
    H264Context *h= avctx->priv_data;
3080
    MpegEncContext * const s = &h->s;
3081

    
3082
    MPV_decode_defaults(s);
3083

    
3084
    s->avctx = avctx;
3085
    common_init(h);
3086

    
3087
    s->out_format = FMT_H264;
3088
    s->workaround_bugs= avctx->workaround_bugs;
3089

    
3090
    // set defaults
3091
//    s->decode_mb= ff_h263_decode_mb;
3092
    s->low_delay= 1;
3093
    avctx->pix_fmt= PIX_FMT_YUV420P;
3094

    
3095
    decode_init_vlc(h);
3096

    
3097
    if(avctx->extradata_size > 0 && avctx->extradata &&
3098
       *(char *)avctx->extradata == 1){
3099
        h->is_avc = 1;
3100
        h->got_avcC = 0;
3101
    } else {
3102
        h->is_avc = 0;
3103
    }
3104

    
3105
    return 0;
3106
}
3107

    
3108
static int frame_start(H264Context *h){
3109
    MpegEncContext * const s = &h->s;
3110
    int i;
3111

    
3112
    if(MPV_frame_start(s, s->avctx) < 0)
3113
        return -1;
3114
    ff_er_frame_start(s);
3115

    
3116
    assert(s->linesize && s->uvlinesize);
3117

    
3118
    for(i=0; i<16; i++){
3119
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3120
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3121
    }
3122
    for(i=0; i<4; i++){
3123
        h->block_offset[16+i]=
3124
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3125
        h->block_offset[24+16+i]=
3126
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3127
    }
3128

    
3129
    /* can't be in alloc_tables because linesize isn't known there.
3130
     * FIXME: redo bipred weight to not require extra buffer? */
3131
    if(!s->obmc_scratchpad)
3132
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3133

    
3134
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3135
    return 0;
3136
}
3137

    
3138
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3139
    MpegEncContext * const s = &h->s;
3140
    int i;
3141

    
3142
    src_y  -=   linesize;
3143
    src_cb -= uvlinesize;
3144
    src_cr -= uvlinesize;
3145

    
3146
    // There are two lines saved, the line above the the top macroblock of a pair,
3147
    // and the line above the bottom macroblock
3148
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3149
    for(i=1; i<17; i++){
3150
        h->left_border[i]= src_y[15+i*  linesize];
3151
    }
3152

    
3153
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3154
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3155

    
3156
    if(!(s->flags&CODEC_FLAG_GRAY)){
3157
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3158
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3159
        for(i=1; i<9; i++){
3160
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3161
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3162
        }
3163
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3164
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3165
    }
3166
}
3167

    
3168
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){
3169
    MpegEncContext * const s = &h->s;
3170
    int temp8, i;
3171
    uint64_t temp64;
3172
    int deblock_left = (s->mb_x > 0);
3173
    int deblock_top  = (s->mb_y > 0);
3174

    
3175
    src_y  -=   linesize + 1;
3176
    src_cb -= uvlinesize + 1;
3177
    src_cr -= uvlinesize + 1;
3178

    
3179
#define XCHG(a,b,t,xchg)\
3180
t= a;\
3181
if(xchg)\
3182
    a= b;\
3183
b= t;
3184

    
3185
    if(deblock_left){
3186
        for(i = !deblock_top; i<17; i++){
3187
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3188
        }
3189
    }
3190

    
3191
    if(deblock_top){
3192
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3193
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3194
        if(s->mb_x+1 < s->mb_width){
3195
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3196
        }
3197
    }
3198

    
3199
    if(!(s->flags&CODEC_FLAG_GRAY)){
3200
        if(deblock_left){
3201
            for(i = !deblock_top; i<9; i++){
3202
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3203
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3204
            }
3205
        }
3206
        if(deblock_top){
3207
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3208
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3209
        }
3210
    }
3211
}
3212

    
3213
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3214
    MpegEncContext * const s = &h->s;
3215
    int i;
3216

    
3217
    src_y  -= 2 *   linesize;
3218
    src_cb -= 2 * uvlinesize;
3219
    src_cr -= 2 * uvlinesize;
3220

    
3221
    // There are two lines saved, the line above the the top macroblock of a pair,
3222
    // and the line above the bottom macroblock
3223
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3224
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3225
    for(i=2; i<34; i++){
3226
        h->left_border[i]= src_y[15+i*  linesize];
3227
    }
3228

    
3229
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3230
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3231
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3232
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3233

    
3234
    if(!(s->flags&CODEC_FLAG_GRAY)){
3235
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3236
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3237
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3238
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3239
        for(i=2; i<18; i++){
3240
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3241
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3242
        }
3243
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3244
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3245
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3246
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3247
    }
3248
}
3249

    
3250
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){
3251
    MpegEncContext * const s = &h->s;
3252
    int temp8, i;
3253
    uint64_t temp64;
3254
    int deblock_left = (s->mb_x > 0);
3255
    int deblock_top  = (s->mb_y > 0);
3256

    
3257
    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);
3258

    
3259
    src_y  -= 2 *   linesize + 1;
3260
    src_cb -= 2 * uvlinesize + 1;
3261
    src_cr -= 2 * uvlinesize + 1;
3262

    
3263
#define XCHG(a,b,t,xchg)\
3264
t= a;\
3265
if(xchg)\
3266
    a= b;\
3267
b= t;
3268

    
3269
    if(deblock_left){
3270
        for(i = (!deblock_top)<<1; i<34; i++){
3271
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3272
        }
3273
    }
3274

    
3275
    if(deblock_top){
3276
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3277
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3278
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3279
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3280
    }
3281

    
3282
    if(!(s->flags&CODEC_FLAG_GRAY)){
3283
        if(deblock_left){
3284
            for(i = (!deblock_top) << 1; i<18; i++){
3285
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3286
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3287
            }
3288
        }
3289
        if(deblock_top){
3290
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3291
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3292
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3293
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3294
        }
3295
    }
3296
}
3297

    
3298
static void hl_decode_mb(H264Context *h){
3299
    MpegEncContext * const s = &h->s;
3300
    const int mb_x= s->mb_x;
3301
    const int mb_y= s->mb_y;
3302
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3303
    const int mb_type= s->current_picture.mb_type[mb_xy];
3304
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3305
    int linesize, uvlinesize /*dct_offset*/;
3306
    int i;
3307
    int *block_offset = &h->block_offset[0];
3308
    const unsigned int bottom = mb_y & 1;
3309
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3310
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3311
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3312

    
3313
    if(!s->decode)
3314
        return;
3315

    
3316
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3317
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3318
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3319

    
3320
    if (h->mb_field_decoding_flag) {
3321
        linesize = s->linesize * 2;
3322
        uvlinesize = s->uvlinesize * 2;
3323
        block_offset = &h->block_offset[24];
3324
        if(mb_y&1){ //FIXME move out of this func?
3325
            dest_y -= s->linesize*15;
3326
            dest_cb-= s->uvlinesize*7;
3327
            dest_cr-= s->uvlinesize*7;
3328
        }
3329
    } else {
3330
        linesize = s->linesize;
3331
        uvlinesize = s->uvlinesize;
3332
//        dct_offset = s->linesize * 16;
3333
    }
3334

    
3335
    if(transform_bypass){
3336
        idct_dc_add =
3337
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3338
    }else if(IS_8x8DCT(mb_type)){
3339
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3340
        idct_add = s->dsp.h264_idct8_add;
3341
    }else{
3342
        idct_dc_add = s->dsp.h264_idct_dc_add;
3343
        idct_add = s->dsp.h264_idct_add;
3344
    }
3345

    
3346
    if (IS_INTRA_PCM(mb_type)) {
3347
        unsigned int x, y;
3348

    
3349
        // The pixels are stored in h->mb array in the same order as levels,
3350
        // copy them in output in the correct order.
3351
        for(i=0; i<16; i++) {
3352
            for (y=0; y<4; y++) {
3353
                for (x=0; x<4; x++) {
3354
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3355
                }
3356
            }
3357
        }
3358
        for(i=16; i<16+4; i++) {
3359
            for (y=0; y<4; y++) {
3360
                for (x=0; x<4; x++) {
3361
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3362
                }
3363
            }
3364
        }
3365
        for(i=20; i<20+4; i++) {
3366
            for (y=0; y<4; y++) {
3367
                for (x=0; x<4; x++) {
3368
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3369
                }
3370
            }
3371
        }
3372
    } else {
3373
        if(IS_INTRA(mb_type)){
3374
            if(h->deblocking_filter) {
3375
                if (h->mb_aff_frame) {
3376
                    if (!bottom)
3377
                        xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3378
                } else {
3379
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3380
                }
3381
            }
3382

    
3383
            if(!(s->flags&CODEC_FLAG_GRAY)){
3384
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3385
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3386
            }
3387

    
3388
            if(IS_INTRA4x4(mb_type)){
3389
                if(!s->encoding){
3390
                    if(IS_8x8DCT(mb_type)){
3391
                        for(i=0; i<16; i+=4){
3392
                            uint8_t * const ptr= dest_y + block_offset[i];
3393
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3394
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3395
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3396
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3397
                            if(nnz){
3398
                                if(nnz == 1 && h->mb[i*16])
3399
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3400
                                else
3401
                                    idct_add(ptr, h->mb + i*16, linesize);
3402
                            }
3403
                        }
3404
                    }else
3405
                    for(i=0; i<16; i++){
3406
                        uint8_t * const ptr= dest_y + block_offset[i];
3407
                        uint8_t *topright;
3408
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3409
                        int nnz, tr;
3410

    
3411
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3412
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3413
                            assert(mb_y || linesize <= block_offset[i]);
3414
                            if(!topright_avail){
3415
                                tr= ptr[3 - linesize]*0x01010101;
3416
                                topright= (uint8_t*) &tr;
3417
                            }else
3418
                                topright= ptr + 4 - linesize;
3419
                        }else
3420
                            topright= NULL;
3421

    
3422
                        h->pred4x4[ dir ](ptr, topright, linesize);
3423
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3424
                        if(nnz){
3425
                            if(s->codec_id == CODEC_ID_H264){
3426
                                if(nnz == 1 && h->mb[i*16])
3427
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3428
                                else
3429
                                    idct_add(ptr, h->mb + i*16, linesize);
3430
                            }else
3431
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3432
                        }
3433
                    }
3434
                }
3435
            }else{
3436
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3437
                if(s->codec_id == CODEC_ID_H264){
3438
                    if(!transform_bypass)
3439
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3440
                }else
3441
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3442
            }
3443
            if(h->deblocking_filter) {
3444
                if (h->mb_aff_frame) {
3445
                    if (bottom) {
3446
                        uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
3447
                        uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3448
                        uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3449
                        s->mb_y--;
3450
                        xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3451
                        s->mb_y++;
3452
                    }
3453
                } else {
3454
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3455
                }
3456
            }
3457
        }else if(s->codec_id == CODEC_ID_H264){
3458
            hl_motion(h, dest_y, dest_cb, dest_cr,
3459
                      s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3460
                      s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3461
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3462
        }
3463

    
3464

    
3465
        if(!IS_INTRA4x4(mb_type)){
3466
            if(s->codec_id == CODEC_ID_H264){
3467
                if(IS_INTRA16x16(mb_type)){
3468
                    for(i=0; i<16; i++){
3469
                        if(h->non_zero_count_cache[ scan8[i] ])
3470
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3471
                        else if(h->mb[i*16])
3472
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3473
                    }
3474
                }else{
3475
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3476
                    for(i=0; i<16; i+=di){
3477
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3478
                        if(nnz){
3479
                            if(nnz==1 && h->mb[i*16])
3480
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3481
                            else
3482
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3483
                        }
3484
                    }
3485
                }
3486
            }else{
3487
                for(i=0; i<16; i++){
3488
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3489
                        uint8_t * const ptr= dest_y + block_offset[i];
3490
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3491
                    }
3492
                }
3493
            }
3494
        }
3495

    
3496
        if(!(s->flags&CODEC_FLAG_GRAY)){
3497
            uint8_t *dest[2] = {dest_cb, dest_cr};
3498
            if(transform_bypass){
3499
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3500
            }else{
3501
                idct_add = s->dsp.h264_idct_add;
3502
                idct_dc_add = s->dsp.h264_idct_dc_add;
3503
                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]);
3504
                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]);
3505
            }
3506
            if(s->codec_id == CODEC_ID_H264){
3507
                for(i=16; i<16+8; i++){
3508
                    if(h->non_zero_count_cache[ scan8[i] ])
3509
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3510
                    else if(h->mb[i*16])
3511
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3512
                }
3513
            }else{
3514
                for(i=16; i<16+8; i++){
3515
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3516
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3517
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3518
                    }
3519
                }
3520
            }
3521
        }
3522
    }
3523
    if(h->deblocking_filter) {
3524
        if (h->mb_aff_frame) {
3525
            const int mb_y = s->mb_y - 1;
3526
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3527
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3528
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3529
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3530
            uint8_t tmp = s->current_picture.data[1][384];
3531
            if (!bottom) return;
3532
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3533
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3534
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3535

    
3536
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3537
            // TODO deblock a pair
3538
            // top
3539
            s->mb_y--;
3540
            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);
3541
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3542
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3543
            if (tmp != s->current_picture.data[1][384]) {
3544
                tprintf("modified pixel 8,1 (1)\n");
3545
            }
3546
            // bottom
3547
            s->mb_y++;
3548
            tprintf("call mbaff filter_mb\n");
3549
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3550
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3551
            if (tmp != s->current_picture.data[1][384]) {
3552
                tprintf("modified pixel 8,1 (2)\n");
3553
            }
3554
        } else {
3555
            tprintf("call filter_mb\n");
3556
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3557
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3558
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3559
        }
3560
    }
3561
}
3562

    
3563
/**
3564
 * fills the default_ref_list.
3565
 */
3566
static int fill_default_ref_list(H264Context *h){
3567
    MpegEncContext * const s = &h->s;
3568
    int i;
3569
    int smallest_poc_greater_than_current = -1;
3570
    Picture sorted_short_ref[32];
3571

    
3572
    if(h->slice_type==B_TYPE){
3573
        int out_i;
3574
        int limit= INT_MIN;
3575

    
3576
        /* sort frame according to poc in B slice */
3577
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3578
            int best_i=INT_MIN;
3579
            int best_poc=INT_MAX;
3580

    
3581
            for(i=0; i<h->short_ref_count; i++){
3582
                const int poc= h->short_ref[i]->poc;
3583
                if(poc > limit && poc < best_poc){
3584
                    best_poc= poc;
3585
                    best_i= i;
3586
                }
3587
            }
3588

    
3589
            assert(best_i != INT_MIN);
3590

    
3591
            limit= best_poc;
3592
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3593
            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);
3594
            if (-1 == smallest_poc_greater_than_current) {
3595
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3596
                    smallest_poc_greater_than_current = out_i;
3597
                }
3598
            }
3599
        }
3600
    }
3601

    
3602
    if(s->picture_structure == PICT_FRAME){
3603
        if(h->slice_type==B_TYPE){
3604
            int list;
3605
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3606

    
3607
            // find the largest poc
3608
            for(list=0; list<2; list++){
3609
                int index = 0;
3610
                int j= -99;
3611
                int step= list ? -1 : 1;
3612

    
3613
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3614
                    while(j<0 || j>= h->short_ref_count){
3615
                        if(j != -99 && step == (list ? -1 : 1))
3616
                            return -1;
3617
                        step = -step;
3618
                        j= smallest_poc_greater_than_current + (step>>1);
3619
                    }
3620
                    if(sorted_short_ref[j].reference != 3) continue;
3621
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3622
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3623
                }
3624

    
3625
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3626
                    if(h->long_ref[i] == NULL) continue;
3627
                    if(h->long_ref[i]->reference != 3) continue;
3628

    
3629
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3630
                    h->default_ref_list[ list ][index++].pic_id= i;;
3631
                }
3632

    
3633
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3634
                    // swap the two first elements of L1 when
3635
                    // L0 and L1 are identical
3636
                    Picture temp= h->default_ref_list[1][0];
3637
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3638
                    h->default_ref_list[1][1] = temp;
3639
                }
3640

    
3641
                if(index < h->ref_count[ list ])
3642
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3643
            }
3644
        }else{
3645
            int index=0;
3646
            for(i=0; i<h->short_ref_count; i++){
3647
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3648
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3649
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3650
            }
3651
            for(i = 0; i < 16; i++){
3652
                if(h->long_ref[i] == NULL) continue;
3653
                if(h->long_ref[i]->reference != 3) continue;
3654
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3655
                h->default_ref_list[0][index++].pic_id= i;;
3656
            }
3657
            if(index < h->ref_count[0])
3658
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3659
        }
3660
    }else{ //FIELD
3661
        if(h->slice_type==B_TYPE){
3662
        }else{
3663
            //FIXME second field balh
3664
        }
3665
    }
3666
#ifdef TRACE
3667
    for (i=0; i<h->ref_count[0]; i++) {
3668
        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]);
3669
    }
3670
    if(h->slice_type==B_TYPE){
3671
        for (i=0; i<h->ref_count[1]; i++) {
3672
            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]);
3673
        }
3674
    }
3675
#endif
3676
    return 0;
3677
}
3678

    
3679
static void print_short_term(H264Context *h);
3680
static void print_long_term(H264Context *h);
3681

    
3682
static int decode_ref_pic_list_reordering(H264Context *h){
3683
    MpegEncContext * const s = &h->s;
3684
    int list, index;
3685

    
3686
    print_short_term(h);
3687
    print_long_term(h);
3688
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3689

    
3690
    for(list=0; list<2; list++){
3691
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3692

    
3693
        if(get_bits1(&s->gb)){
3694
            int pred= h->curr_pic_num;
3695

    
3696
            for(index=0; ; index++){
3697
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3698
                int pic_id;
3699
                int i;
3700
                Picture *ref = NULL;
3701

    
3702
                if(reordering_of_pic_nums_idc==3)
3703
                    break;
3704

    
3705
                if(index >= h->ref_count[list]){
3706
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3707
                    return -1;
3708
                }
3709

    
3710
                if(reordering_of_pic_nums_idc<3){
3711
                    if(reordering_of_pic_nums_idc<2){
3712
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3713

    
3714
                        if(abs_diff_pic_num >= h->max_pic_num){
3715
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3716
                            return -1;
3717
                        }
3718

    
3719
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3720
                        else                                pred+= abs_diff_pic_num;
3721
                        pred &= h->max_pic_num - 1;
3722

    
3723
                        for(i= h->short_ref_count-1; i>=0; i--){
3724
                            ref = h->short_ref[i];
3725
                            assert(ref->reference == 3);
3726
                            assert(!ref->long_ref);
3727
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3728
                                break;
3729
                        }
3730
                        if(i>=0)
3731
                            ref->pic_id= ref->frame_num;
3732
                    }else{
3733
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3734
                        ref = h->long_ref[pic_id];
3735
                        ref->pic_id= pic_id;
3736
                        assert(ref->reference == 3);
3737
                        assert(ref->long_ref);
3738
                        i=0;
3739
                    }
3740

    
3741
                    if (i < 0) {
3742
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3743
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3744
                    } else {
3745
                        for(i=index; i+1<h->ref_count[list]; i++){
3746
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3747
                                break;
3748
                        }
3749
                        for(; i > index; i--){
3750
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3751
                        }
3752
                        h->ref_list[list][index]= *ref;
3753
                    }
3754
                }else{
3755
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3756
                    return -1;
3757
                }
3758
            }
3759
        }
3760

    
3761
        if(h->slice_type!=B_TYPE) break;
3762
    }
3763
    for(list=0; list<2; list++){
3764
        for(index= 0; index < h->ref_count[list]; index++){
3765
            if(!h->ref_list[list][index].data[0])
3766
                h->ref_list[list][index]= s->current_picture;
3767
        }
3768
        if(h->slice_type!=B_TYPE) break;
3769
    }
3770

    
3771
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3772
        direct_dist_scale_factor(h);
3773
    direct_ref_list_init(h);
3774
    return 0;
3775
}
3776

    
3777
static int pred_weight_table(H264Context *h){
3778
    MpegEncContext * const s = &h->s;
3779
    int list, i;
3780
    int luma_def, chroma_def;
3781

    
3782
    h->use_weight= 0;
3783
    h->use_weight_chroma= 0;
3784
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3785
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3786
    luma_def = 1<<h->luma_log2_weight_denom;
3787
    chroma_def = 1<<h->chroma_log2_weight_denom;
3788

    
3789
    for(list=0; list<2; list++){
3790
        for(i=0; i<h->ref_count[list]; i++){
3791
            int luma_weight_flag, chroma_weight_flag;
3792

    
3793
            luma_weight_flag= get_bits1(&s->gb);
3794
            if(luma_weight_flag){
3795
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3796
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3797
                if(   h->luma_weight[list][i] != luma_def
3798
                   || h->luma_offset[list][i] != 0)
3799
                    h->use_weight= 1;
3800
            }else{
3801
                h->luma_weight[list][i]= luma_def;
3802
                h->luma_offset[list][i]= 0;
3803
            }
3804

    
3805
            chroma_weight_flag= get_bits1(&s->gb);
3806
            if(chroma_weight_flag){
3807
                int j;
3808
                for(j=0; j<2; j++){
3809
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3810
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3811
                    if(   h->chroma_weight[list][i][j] != chroma_def
3812
                       || h->chroma_offset[list][i][j] != 0)
3813
                        h->use_weight_chroma= 1;
3814
                }
3815
            }else{
3816
                int j;
3817
                for(j=0; j<2; j++){
3818
                    h->chroma_weight[list][i][j]= chroma_def;
3819
                    h->chroma_offset[list][i][j]= 0;
3820
                }
3821
            }
3822
        }
3823
        if(h->slice_type != B_TYPE) break;
3824
    }
3825
    h->use_weight= h->use_weight || h->use_weight_chroma;
3826
    return 0;
3827
}
3828

    
3829
static void implicit_weight_table(H264Context *h){
3830
    MpegEncContext * const s = &h->s;
3831
    int ref0, ref1;
3832
    int cur_poc = s->current_picture_ptr->poc;
3833

    
3834
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3835
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3836
        h->use_weight= 0;
3837
        h->use_weight_chroma= 0;
3838
        return;
3839
    }
3840

    
3841
    h->use_weight= 2;
3842
    h->use_weight_chroma= 2;
3843
    h->luma_log2_weight_denom= 5;
3844
    h->chroma_log2_weight_denom= 5;
3845

    
3846
    /* FIXME: MBAFF */
3847
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3848
        int poc0 = h->ref_list[0][ref0].poc;
3849
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3850
            int poc1 = h->ref_list[1][ref1].poc;
3851
            int td = clip(poc1 - poc0, -128, 127);
3852
            if(td){
3853
                int tb = clip(cur_poc - poc0, -128, 127);
3854
                int tx = (16384 + (ABS(td) >> 1)) / td;
3855
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3856
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3857
                    h->implicit_weight[ref0][ref1] = 32;
3858
                else
3859
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3860
            }else
3861
                h->implicit_weight[ref0][ref1] = 32;
3862
        }
3863
    }
3864
}
3865

    
3866
static inline void unreference_pic(H264Context *h, Picture *pic){
3867
    int i;
3868
    pic->reference=0;
3869
    if(pic == h->delayed_output_pic)
3870
        pic->reference=1;
3871
    else{
3872
        for(i = 0; h->delayed_pic[i]; i++)
3873
            if(pic == h->delayed_pic[i]){
3874
                pic->reference=1;
3875
                break;
3876
            }
3877
    }
3878
}
3879

    
3880
/**
3881
 * instantaneous decoder refresh.
3882
 */
3883
static void idr(H264Context *h){
3884
    int i;
3885

    
3886
    for(i=0; i<16; i++){
3887
        if (h->long_ref[i] != NULL) {
3888
            unreference_pic(h, h->long_ref[i]);
3889
            h->long_ref[i]= NULL;
3890
        }
3891
    }
3892
    h->long_ref_count=0;
3893

    
3894
    for(i=0; i<h->short_ref_count; i++){
3895
        unreference_pic(h, h->short_ref[i]);
3896
        h->short_ref[i]= NULL;
3897
    }
3898
    h->short_ref_count=0;
3899
}
3900

    
3901
/* forget old pics after a seek */
3902
static void flush_dpb(AVCodecContext *avctx){
3903
    H264Context *h= avctx->priv_data;
3904
    int i;
3905
    for(i=0; i<16; i++) {
3906
        if(h->delayed_pic[i])
3907
            h->delayed_pic[i]->reference= 0;
3908
        h->delayed_pic[i]= NULL;
3909
    }
3910
    if(h->delayed_output_pic)
3911
        h->delayed_output_pic->reference= 0;
3912
    h->delayed_output_pic= NULL;
3913
    idr(h);
3914
    if(h->s.current_picture_ptr)
3915
        h->s.current_picture_ptr->reference= 0;
3916
}
3917

    
3918
/**
3919
 *
3920
 * @return the removed picture or NULL if an error occurs
3921
 */
3922
static Picture * remove_short(H264Context *h, int frame_num){
3923
    MpegEncContext * const s = &h->s;
3924
    int i;
3925

    
3926
    if(s->avctx->debug&FF_DEBUG_MMCO)
3927
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3928

    
3929
    for(i=0; i<h->short_ref_count; i++){
3930
        Picture *pic= h->short_ref[i];
3931
        if(s->avctx->debug&FF_DEBUG_MMCO)
3932
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3933
        if(pic->frame_num == frame_num){
3934
            h->short_ref[i]= NULL;
3935
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3936
            h->short_ref_count--;
3937
            return pic;
3938
        }
3939
    }
3940
    return NULL;
3941
}
3942

    
3943
/**
3944
 *
3945
 * @return the removed picture or NULL if an error occurs
3946
 */
3947
static Picture * remove_long(H264Context *h, int i){
3948
    Picture *pic;
3949

    
3950
    pic= h->long_ref[i];
3951
    h->long_ref[i]= NULL;
3952
    if(pic) h->long_ref_count--;
3953

    
3954
    return pic;
3955
}
3956

    
3957
/**
3958
 * print short term list
3959
 */
3960
static void print_short_term(H264Context *h) {
3961
    uint32_t i;
3962
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3963
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3964
        for(i=0; i<h->short_ref_count; i++){
3965
            Picture *pic= h->short_ref[i];
3966
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3967
        }
3968
    }
3969
}
3970

    
3971
/**
3972
 * print long term list
3973
 */
3974
static void print_long_term(H264Context *h) {
3975
    uint32_t i;
3976
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3977
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3978
        for(i = 0; i < 16; i++){
3979
            Picture *pic= h->long_ref[i];
3980
            if (pic) {
3981
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3982
            }
3983
        }
3984
    }
3985
}
3986

    
3987
/**
3988
 * Executes the reference picture marking (memory management control operations).
3989
 */
3990
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3991
    MpegEncContext * const s = &h->s;
3992
    int i, j;
3993
    int current_is_long=0;
3994
    Picture *pic;
3995

    
3996
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3997
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3998

    
3999
    for(i=0; i<mmco_count; i++){
4000
        if(s->avctx->debug&FF_DEBUG_MMCO)
4001
            av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
4002

    
4003
        switch(mmco[i].opcode){
4004
        case MMCO_SHORT2UNUSED:
4005
            pic= remove_short(h, mmco[i].short_frame_num);
4006
            if(pic)
4007
                unreference_pic(h, pic);
4008
            else if(s->avctx->debug&FF_DEBUG_MMCO)
4009
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
4010
            break;
4011
        case MMCO_SHORT2LONG:
4012
            pic= remove_long(h, mmco[i].long_index);
4013
            if(pic) unreference_pic(h, pic);
4014

    
4015
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
4016
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
4017
            h->long_ref_count++;
4018
            break;
4019
        case MMCO_LONG2UNUSED:
4020
            pic= remove_long(h, mmco[i].long_index);
4021
            if(pic)
4022
                unreference_pic(h, pic);
4023
            else if(s->avctx->debug&FF_DEBUG_MMCO)
4024
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
4025
            break;
4026
        case MMCO_LONG:
4027
            pic= remove_long(h, mmco[i].long_index);
4028
            if(pic) unreference_pic(h, pic);
4029

    
4030
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
4031
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
4032
            h->long_ref_count++;
4033

    
4034
            current_is_long=1;
4035
            break;
4036
        case MMCO_SET_MAX_LONG:
4037
            assert(mmco[i].long_index <= 16);
4038
            // just remove the long term which index is greater than new max
4039
            for(j = mmco[i].long_index; j<16; j++){
4040
                pic = remove_long(h, j);
4041
                if (pic) unreference_pic(h, pic);
4042
            }
4043
            break;
4044
        case MMCO_RESET:
4045
            while(h->short_ref_count){
4046
                pic= remove_short(h, h->short_ref[0]->frame_num);
4047
                unreference_pic(h, pic);
4048
            }
4049
            for(j = 0; j < 16; j++) {
4050
                pic= remove_long(h, j);
4051
                if(pic) unreference_pic(h, pic);
4052
            }
4053
            break;
4054
        default: assert(0);
4055
        }
4056
    }
4057

    
4058
    if(!current_is_long){
4059
        pic= remove_short(h, s->current_picture_ptr->frame_num);
4060
        if(pic){
4061
            unreference_pic(h, pic);
4062
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
4063
        }
4064

    
4065
        if(h->short_ref_count)
4066
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
4067

    
4068
        h->short_ref[0]= s->current_picture_ptr;
4069
        h->short_ref[0]->long_ref=0;
4070
        h->short_ref_count++;
4071
    }
4072

    
4073
    print_short_term(h);
4074
    print_long_term(h);
4075
    return 0;
4076
}
4077

    
4078
static int decode_ref_pic_marking(H264Context *h){
4079
    MpegEncContext * const s = &h->s;
4080
    int i;
4081

    
4082
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
4083
        s->broken_link= get_bits1(&s->gb) -1;
4084
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
4085
        if(h->mmco[0].long_index == -1)
4086
            h->mmco_index= 0;
4087
        else{
4088
            h->mmco[0].opcode= MMCO_LONG;
4089
            h->mmco_index= 1;
4090
        }
4091
    }else{
4092
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
4093
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
4094
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
4095

    
4096
                h->mmco[i].opcode= opcode;
4097
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
4098
                    h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
4099
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
4100
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
4101
                        return -1;
4102
                    }*/
4103
                }
4104
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
4105
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
4106
                    if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){
4107
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
4108
                        return -1;
4109
                    }
4110
                }
4111

    
4112
                if(opcode > MMCO_LONG){
4113
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4114
                    return -1;
4115
                }
4116
                if(opcode == MMCO_END)
4117
                    break;
4118
            }
4119
            h->mmco_index= i;
4120
        }else{
4121
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4122

    
4123
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4124
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4125
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4126
                h->mmco_index= 1;
4127
            }else
4128
                h->mmco_index= 0;
4129
        }
4130
    }
4131

    
4132
    return 0;
4133
}
4134

    
4135
static int init_poc(H264Context *h){
4136
    MpegEncContext * const s = &h->s;
4137
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4138
    int field_poc[2];
4139

    
4140
    if(h->nal_unit_type == NAL_IDR_SLICE){
4141
        h->frame_num_offset= 0;
4142
    }else{
4143
        if(h->frame_num < h->prev_frame_num)
4144
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4145
        else
4146
            h->frame_num_offset= h->prev_frame_num_offset;
4147
    }
4148

    
4149
    if(h->sps.poc_type==0){
4150
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4151

    
4152
        if(h->nal_unit_type == NAL_IDR_SLICE){
4153
             h->prev_poc_msb=
4154
             h->prev_poc_lsb= 0;
4155
        }
4156

    
4157
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4158
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4159
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4160
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4161
        else
4162
            h->poc_msb = h->prev_poc_msb;
4163
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4164
        field_poc[0] =
4165
        field_poc[1] = h->poc_msb + h->poc_lsb;
4166
        if(s->picture_structure == PICT_FRAME)
4167
            field_poc[1] += h->delta_poc_bottom;
4168
    }else if(h->sps.poc_type==1){
4169
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4170
        int i;
4171

    
4172
        if(h->sps.poc_cycle_length != 0)
4173
            abs_frame_num = h->frame_num_offset + h->frame_num;
4174
        else
4175
            abs_frame_num = 0;
4176

    
4177
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4178
            abs_frame_num--;
4179

    
4180
        expected_delta_per_poc_cycle = 0;
4181
        for(i=0; i < h->sps.poc_cycle_length; i++)
4182
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4183

    
4184
        if(abs_frame_num > 0){
4185
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4186
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4187

    
4188
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4189
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4190
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4191
        } else
4192
            expectedpoc = 0;
4193

    
4194
        if(h->nal_ref_idc == 0)
4195
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4196

    
4197
        field_poc[0] = expectedpoc + h->delta_poc[0];
4198
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4199

    
4200
        if(s->picture_structure == PICT_FRAME)
4201
            field_poc[1] += h->delta_poc[1];
4202
    }else{
4203
        int poc;
4204
        if(h->nal_unit_type == NAL_IDR_SLICE){
4205
            poc= 0;
4206
        }else{
4207
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4208
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4209
        }
4210
        field_poc[0]= poc;
4211
        field_poc[1]= poc;
4212
    }
4213

    
4214
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4215
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4216
    if(s->picture_structure != PICT_TOP_FIELD)
4217
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4218
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4219
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4220

    
4221
    return 0;
4222
}
4223

    
4224
/**
4225
 * decodes a slice header.
4226
 * this will allso call MPV_common_init() and frame_start() as needed
4227
 */
4228
static int decode_slice_header(H264Context *h){
4229
    MpegEncContext * const s = &h->s;
4230
    int first_mb_in_slice, pps_id;
4231
    int num_ref_idx_active_override_flag;
4232
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4233
    int slice_type;
4234
    int default_ref_list_done = 0;
4235

    
4236
    s->current_picture.reference= h->nal_ref_idc != 0;
4237
    s->dropable= h->nal_ref_idc == 0;
4238

    
4239
    first_mb_in_slice= get_ue_golomb(&s->gb);
4240

    
4241
    slice_type= get_ue_golomb(&s->gb);
4242
    if(slice_type > 9){
4243
        av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
4244
        return -1;
4245
    }
4246
    if(slice_type > 4){
4247
        slice_type -= 5;
4248
        h->slice_type_fixed=1;
4249
    }else
4250
        h->slice_type_fixed=0;
4251

    
4252
    slice_type= slice_type_map[ slice_type ];
4253
    if (slice_type == I_TYPE
4254
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4255
        default_ref_list_done = 1;
4256
    }
4257
    h->slice_type= slice_type;
4258

    
4259
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4260

    
4261
    pps_id= get_ue_golomb(&s->gb);
4262
    if(pps_id>255){
4263
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4264
        return -1;
4265
    }
4266
    h->pps= h->pps_buffer[pps_id];
4267
    if(h->pps.slice_group_count == 0){
4268
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4269
        return -1;
4270
    }
4271

    
4272
    h->sps= h->sps_buffer[ h->pps.sps_id ];
4273
    if(h->sps.log2_max_frame_num == 0){
4274
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4275
        return -1;
4276
    }
4277

    
4278
    if(h->dequant_coeff_pps != pps_id){
4279
        h->dequant_coeff_pps = pps_id;
4280
        init_dequant_tables(h);
4281
    }
4282

    
4283
    s->mb_width= h->sps.mb_width;
4284
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
4285

    
4286
    h->b_stride=  s->mb_width*4;
4287
    h->b8_stride= s->mb_width*2;
4288

    
4289
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
4290
    if(h->sps.frame_mbs_only_flag)
4291
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
4292
    else
4293
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
4294

    
4295
    if (s->context_initialized
4296
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
4297
        free_tables(h);
4298
        MPV_common_end(s);
4299
    }
4300
    if (!s->context_initialized) {
4301
        if (MPV_common_init(s) < 0)
4302
            return -1;
4303

    
4304
        if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
4305
            memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
4306
            memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
4307
        }else{
4308
            int i;
4309
            for(i=0; i<16; i++){
4310
#define T(x) (x>>2) | ((x<<2) & 0xF)
4311
                h->zigzag_scan[i] = T(zigzag_scan[i]);
4312
                h-> field_scan[i] = T( field_scan[i]);
4313
#undef T
4314
            }
4315
        }
4316
        if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
4317
            memcpy(h->zigzag_scan8x8, zigzag_scan8x8, 64*sizeof(uint8_t));
4318
            memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
4319
        }else{
4320
            int i;
4321
            for(i=0; i<64; i++){