Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 1bd86246

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
    const uint8_t *zigzag_scan_q0;
362
    const uint8_t *field_scan_q0;
363

    
364
    int x264_build;
365
}H264Context;
366

    
367
static VLC coeff_token_vlc[4];
368
static VLC chroma_dc_coeff_token_vlc;
369

    
370
static VLC total_zeros_vlc[15];
371
static VLC chroma_dc_total_zeros_vlc[3];
372

    
373
static VLC run_vlc[6];
374
static VLC run7_vlc;
375

    
376
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
377
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
378
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);
379

    
380
static always_inline uint32_t pack16to32(int a, int b){
381
#ifdef WORDS_BIGENDIAN
382
   return (b&0xFFFF) + (a<<16);
383
#else
384
   return (a&0xFFFF) + (b<<16);
385
#endif
386
}
387

    
388
/**
389
 * fill a rectangle.
390
 * @param h height of the rectangle, should be a constant
391
 * @param w width of the rectangle, should be a constant
392
 * @param size the size of val (1 or 4), should be a constant
393
 */
394
static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
395
    uint8_t *p= (uint8_t*)vp;
396
    assert(size==1 || size==4);
397

    
398
    w      *= size;
399
    stride *= size;
400

    
401
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
402
    assert((stride&(w-1))==0);
403
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
404
    if(w==2 && h==2){
405
        *(uint16_t*)(p + 0)=
406
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
407
    }else if(w==2 && h==4){
408
        *(uint16_t*)(p + 0*stride)=
409
        *(uint16_t*)(p + 1*stride)=
410
        *(uint16_t*)(p + 2*stride)=
411
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
412
    }else if(w==4 && h==1){
413
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
414
    }else if(w==4 && h==2){
415
        *(uint32_t*)(p + 0*stride)=
416
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
417
    }else if(w==4 && h==4){
418
        *(uint32_t*)(p + 0*stride)=
419
        *(uint32_t*)(p + 1*stride)=
420
        *(uint32_t*)(p + 2*stride)=
421
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
422
    }else if(w==8 && h==1){
423
        *(uint32_t*)(p + 0)=
424
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
425
    }else if(w==8 && h==2){
426
        *(uint32_t*)(p + 0 + 0*stride)=
427
        *(uint32_t*)(p + 4 + 0*stride)=
428
        *(uint32_t*)(p + 0 + 1*stride)=
429
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
430
    }else if(w==8 && h==4){
431
        *(uint64_t*)(p + 0*stride)=
432
        *(uint64_t*)(p + 1*stride)=
433
        *(uint64_t*)(p + 2*stride)=
434
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
435
    }else if(w==16 && h==2){
436
        *(uint64_t*)(p + 0+0*stride)=
437
        *(uint64_t*)(p + 8+0*stride)=
438
        *(uint64_t*)(p + 0+1*stride)=
439
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
440
    }else if(w==16 && h==4){
441
        *(uint64_t*)(p + 0+0*stride)=
442
        *(uint64_t*)(p + 8+0*stride)=
443
        *(uint64_t*)(p + 0+1*stride)=
444
        *(uint64_t*)(p + 8+1*stride)=
445
        *(uint64_t*)(p + 0+2*stride)=
446
        *(uint64_t*)(p + 8+2*stride)=
447
        *(uint64_t*)(p + 0+3*stride)=
448
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
449
    }else
450
        assert(0);
451
}
452

    
453
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
454
    MpegEncContext * const s = &h->s;
455
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
456
    int topleft_xy, top_xy, topright_xy, left_xy[2];
457
    int topleft_type, top_type, topright_type, left_type[2];
458
    int left_block[8];
459
    int i;
460

    
461
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
462
    // the actual condition is whether we're on the edge of a slice,
463
    // and even then the intra and nnz parts are unnecessary.
464
    if(for_deblock && h->slice_num == 1)
465
        return;
466

    
467
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
468

    
469
    top_xy     = mb_xy  - s->mb_stride;
470
    topleft_xy = top_xy - 1;
471
    topright_xy= top_xy + 1;
472
    left_xy[1] = left_xy[0] = mb_xy-1;
473
    left_block[0]= 0;
474
    left_block[1]= 1;
475
    left_block[2]= 2;
476
    left_block[3]= 3;
477
    left_block[4]= 7;
478
    left_block[5]= 10;
479
    left_block[6]= 8;
480
    left_block[7]= 11;
481
    if(h->mb_aff_frame){
482
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
483
        const int top_pair_xy      = pair_xy     - s->mb_stride;
484
        const int topleft_pair_xy  = top_pair_xy - 1;
485
        const int topright_pair_xy = top_pair_xy + 1;
486
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
487
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
488
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
489
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
490
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
491
        const int bottom = (s->mb_y & 1);
492
        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);
493
        if (bottom
494
                ? !curr_mb_frame_flag // bottom macroblock
495
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
496
                ) {
497
            top_xy -= s->mb_stride;
498
        }
499
        if (bottom
500
                ? !curr_mb_frame_flag // bottom macroblock
501
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
502
                ) {
503
            topleft_xy -= s->mb_stride;
504
        }
505
        if (bottom
506
                ? !curr_mb_frame_flag // bottom macroblock
507
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
508
                ) {
509
            topright_xy -= s->mb_stride;
510
        }
511
        if (left_mb_frame_flag != curr_mb_frame_flag) {
512
            left_xy[1] = left_xy[0] = pair_xy - 1;
513
            if (curr_mb_frame_flag) {
514
                if (bottom) {
515
                    left_block[0]= 2;
516
                    left_block[1]= 2;
517
                    left_block[2]= 3;
518
                    left_block[3]= 3;
519
                    left_block[4]= 8;
520
                    left_block[5]= 11;
521
                    left_block[6]= 8;
522
                    left_block[7]= 11;
523
                } else {
524
                    left_block[0]= 0;
525
                    left_block[1]= 0;
526
                    left_block[2]= 1;
527
                    left_block[3]= 1;
528
                    left_block[4]= 7;
529
                    left_block[5]= 10;
530
                    left_block[6]= 7;
531
                    left_block[7]= 10;
532
                }
533
            } else {
534
                left_xy[1] += s->mb_stride;
535
                //left_block[0]= 0;
536
                left_block[1]= 2;
537
                left_block[2]= 0;
538
                left_block[3]= 2;
539
                //left_block[4]= 7;
540
                left_block[5]= 10;
541
                left_block[6]= 7;
542
                left_block[7]= 10;
543
            }
544
        }
545
    }
546

    
547
    h->top_mb_xy = top_xy;
548
    h->left_mb_xy[0] = left_xy[0];
549
    h->left_mb_xy[1] = left_xy[1];
550
    if(for_deblock){
551
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
552
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
553
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
554
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
555
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
556
    }else{
557
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
558
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
559
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
560
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
561
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
562
    }
563

    
564
    if(IS_INTRA(mb_type)){
565
        h->topleft_samples_available=
566
        h->top_samples_available=
567
        h->left_samples_available= 0xFFFF;
568
        h->topright_samples_available= 0xEEEA;
569

    
570
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
571
            h->topleft_samples_available= 0xB3FF;
572
            h->top_samples_available= 0x33FF;
573
            h->topright_samples_available= 0x26EA;
574
        }
575
        for(i=0; i<2; i++){
576
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
577
                h->topleft_samples_available&= 0xDF5F;
578
                h->left_samples_available&= 0x5F5F;
579
            }
580
        }
581

    
582
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
583
            h->topleft_samples_available&= 0x7FFF;
584

    
585
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
586
            h->topright_samples_available&= 0xFBFF;
587

    
588
        if(IS_INTRA4x4(mb_type)){
589
            if(IS_INTRA4x4(top_type)){
590
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
591
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
592
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
593
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
594
            }else{
595
                int pred;
596
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
597
                    pred= -1;
598
                else{
599
                    pred= 2;
600
                }
601
                h->intra4x4_pred_mode_cache[4+8*0]=
602
                h->intra4x4_pred_mode_cache[5+8*0]=
603
                h->intra4x4_pred_mode_cache[6+8*0]=
604
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
605
            }
606
            for(i=0; i<2; i++){
607
                if(IS_INTRA4x4(left_type[i])){
608
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
609
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
610
                }else{
611
                    int pred;
612
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
613
                        pred= -1;
614
                    else{
615
                        pred= 2;
616
                    }
617
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
618
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
619
                }
620
            }
621
        }
622
    }
623

    
624

    
625
/*
626
0 . T T. T T T T
627
1 L . .L . . . .
628
2 L . .L . . . .
629
3 . T TL . . . .
630
4 L . .L . . . .
631
5 L . .. . . . .
632
*/
633
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
634
    if(top_type){
635
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
636
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
637
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
638
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
639

    
640
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
641
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
642

    
643
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
644
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
645

    
646
    }else{
647
        h->non_zero_count_cache[4+8*0]=
648
        h->non_zero_count_cache[5+8*0]=
649
        h->non_zero_count_cache[6+8*0]=
650
        h->non_zero_count_cache[7+8*0]=
651

    
652
        h->non_zero_count_cache[1+8*0]=
653
        h->non_zero_count_cache[2+8*0]=
654

    
655
        h->non_zero_count_cache[1+8*3]=
656
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
657

    
658
    }
659

    
660
    for (i=0; i<2; i++) {
661
        if(left_type[i]){
662
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
663
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
664
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
665
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
666
        }else{
667
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
668
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
669
            h->non_zero_count_cache[0+8*1 +   8*i]=
670
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
671
        }
672
    }
673

    
674
    if( h->pps.cabac ) {
675
        // top_cbp
676
        if(top_type) {
677
            h->top_cbp = h->cbp_table[top_xy];
678
        } else if(IS_INTRA(mb_type)) {
679
            h->top_cbp = 0x1C0;
680
        } else {
681
            h->top_cbp = 0;
682
        }
683
        // left_cbp
684
        if (left_type[0]) {
685
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
686
        } else if(IS_INTRA(mb_type)) {
687
            h->left_cbp = 0x1C0;
688
        } else {
689
            h->left_cbp = 0;
690
        }
691
        if (left_type[0]) {
692
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
693
        }
694
        if (left_type[1]) {
695
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
696
        }
697
    }
698

    
699
#if 1
700
    //FIXME direct mb can skip much of this
701
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
702
        int list;
703
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
704
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
705
                /*if(!h->mv_cache_clean[list]){
706
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
707
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
708
                    h->mv_cache_clean[list]= 1;
709
                }*/
710
                continue;
711
            }
712
            h->mv_cache_clean[list]= 0;
713

    
714
            if(IS_INTER(top_type)){
715
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
716
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
717
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
718
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
719
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
720
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
721
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
722
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
723
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
724
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
725
            }else{
726
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
727
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
728
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
729
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
730
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
731
            }
732

    
733
            //FIXME unify cleanup or sth
734
            if(IS_INTER(left_type[0])){
735
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
736
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
737
                *(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]];
738
                *(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]];
739
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
740
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
741
            }else{
742
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
743
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
744
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
745
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
746
            }
747

    
748
            if(IS_INTER(left_type[1])){
749
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
750
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
751
                *(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]];
752
                *(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]];
753
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
754
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
755
            }else{
756
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
757
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
758
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
759
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
760
                assert((!left_type[0]) == (!left_type[1]));
761
            }
762

    
763
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
764
                continue;
765

    
766
            if(IS_INTER(topleft_type)){
767
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
768
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
769
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
770
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
771
            }else{
772
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
773
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
774
            }
775

    
776
            if(IS_INTER(topright_type)){
777
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
778
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
779
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
780
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
781
            }else{
782
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
783
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
784
            }
785

    
786

    
787
            h->ref_cache[list][scan8[5 ]+1] =
788
            h->ref_cache[list][scan8[7 ]+1] =
789
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
790
            h->ref_cache[list][scan8[4 ]] =
791
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
792
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
793
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
794
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
795
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
796
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
797

    
798
            if( h->pps.cabac ) {
799
                /* XXX beurk, Load mvd */
800
                if(IS_INTER(topleft_type)){
801
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
802
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
803
                }else{
804
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
805
                }
806

    
807
                if(IS_INTER(top_type)){
808
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
809
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
810
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
811
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
812
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
813
                }else{
814
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
815
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
816
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
817
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
818
                }
819
                if(IS_INTER(left_type[0])){
820
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
821
                    *(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]];
822
                    *(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]];
823
                }else{
824
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
825
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
826
                }
827
                if(IS_INTER(left_type[1])){
828
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
829
                    *(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]];
830
                    *(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]];
831
                }else{
832
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
833
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
834
                }
835
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
836
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
837
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
838
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
839
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
840

    
841
                if(h->slice_type == B_TYPE){
842
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
843

    
844
                    if(IS_DIRECT(top_type)){
845
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
846
                    }else if(IS_8X8(top_type)){
847
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
848
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
849
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
850
                    }else{
851
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
852
                    }
853

    
854
                    //FIXME interlacing
855
                    if(IS_DIRECT(left_type[0])){
856
                        h->direct_cache[scan8[0] - 1 + 0*8]=
857
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
858
                    }else if(IS_8X8(left_type[0])){
859
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
860
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
861
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
862
                    }else{
863
                        h->direct_cache[scan8[0] - 1 + 0*8]=
864
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
865
                    }
866
                }
867
            }
868
        }
869
    }
870
#endif
871

    
872
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
873
}
874

    
875
static inline void write_back_intra_pred_mode(H264Context *h){
876
    MpegEncContext * const s = &h->s;
877
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
878

    
879
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
880
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
881
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
882
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
883
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
884
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
885
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
886
}
887

    
888
/**
889
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
890
 */
891
static inline int check_intra4x4_pred_mode(H264Context *h){
892
    MpegEncContext * const s = &h->s;
893
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
894
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
895
    int i;
896

    
897
    if(!(h->top_samples_available&0x8000)){
898
        for(i=0; i<4; i++){
899
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
900
            if(status<0){
901
                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);
902
                return -1;
903
            } else if(status){
904
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
905
            }
906
        }
907
    }
908

    
909
    if(!(h->left_samples_available&0x8000)){
910
        for(i=0; i<4; i++){
911
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
912
            if(status<0){
913
                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);
914
                return -1;
915
            } else if(status){
916
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
917
            }
918
        }
919
    }
920

    
921
    return 0;
922
} //FIXME cleanup like next
923

    
924
/**
925
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
926
 */
927
static inline int check_intra_pred_mode(H264Context *h, int mode){
928
    MpegEncContext * const s = &h->s;
929
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
930
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
931

    
932
    if(mode < 0 || mode > 6) {
933
        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);
934
        return -1;
935
    }
936

    
937
    if(!(h->top_samples_available&0x8000)){
938
        mode= top[ mode ];
939
        if(mode<0){
940
            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);
941
            return -1;
942
        }
943
    }
944

    
945
    if(!(h->left_samples_available&0x8000)){
946
        mode= left[ mode ];
947
        if(mode<0){
948
            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);
949
            return -1;
950
        }
951
    }
952

    
953
    return mode;
954
}
955

    
956
/**
957
 * gets the predicted intra4x4 prediction mode.
958
 */
959
static inline int pred_intra_mode(H264Context *h, int n){
960
    const int index8= scan8[n];
961
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
962
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
963
    const int min= FFMIN(left, top);
964

    
965
    tprintf("mode:%d %d min:%d\n", left ,top, min);
966

    
967
    if(min<0) return DC_PRED;
968
    else      return min;
969
}
970

    
971
static inline void write_back_non_zero_count(H264Context *h){
972
    MpegEncContext * const s = &h->s;
973
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
974

    
975
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
976
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
977
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
978
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
979
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
980
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
981
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
982

    
983
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
984
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
985
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
986

    
987
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
988
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
989
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
990
}
991

    
992
/**
993
 * gets the predicted number of non zero coefficients.
994
 * @param n block index
995
 */
996
static inline int pred_non_zero_count(H264Context *h, int n){
997
    const int index8= scan8[n];
998
    const int left= h->non_zero_count_cache[index8 - 1];
999
    const int top = h->non_zero_count_cache[index8 - 8];
1000
    int i= left + top;
1001

    
1002
    if(i<64) i= (i+1)>>1;
1003

    
1004
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1005

    
1006
    return i&31;
1007
}
1008

    
1009
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1010
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1011

    
1012
    if(topright_ref != PART_NOT_AVAILABLE){
1013
        *C= h->mv_cache[list][ i - 8 + part_width ];
1014
        return topright_ref;
1015
    }else{
1016
        tprintf("topright MV not available\n");
1017

    
1018
        *C= h->mv_cache[list][ i - 8 - 1 ];
1019
        return h->ref_cache[list][ i - 8 - 1 ];
1020
    }
1021
}
1022

    
1023
/**
1024
 * gets the predicted MV.
1025
 * @param n the block index
1026
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1027
 * @param mx the x component of the predicted motion vector
1028
 * @param my the y component of the predicted motion vector
1029
 */
1030
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1031
    const int index8= scan8[n];
1032
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1033
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1034
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1035
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1036
    const int16_t * C;
1037
    int diagonal_ref, match_count;
1038

    
1039
    assert(part_width==1 || part_width==2 || part_width==4);
1040

    
1041
/* mv_cache
1042
  B . . A T T T T
1043
  U . . L . . , .
1044
  U . . L . . . .
1045
  U . . L . . , .
1046
  . . . L . . . .
1047
*/
1048

    
1049
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1050
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1051
    tprintf("pred_motion match_count=%d\n", match_count);
1052
    if(match_count > 1){ //most common
1053
        *mx= mid_pred(A[0], B[0], C[0]);
1054
        *my= mid_pred(A[1], B[1], C[1]);
1055
    }else if(match_count==1){
1056
        if(left_ref==ref){
1057
            *mx= A[0];
1058
            *my= A[1];
1059
        }else if(top_ref==ref){
1060
            *mx= B[0];
1061
            *my= B[1];
1062
        }else{
1063
            *mx= C[0];
1064
            *my= C[1];
1065
        }
1066
    }else{
1067
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1068
            *mx= A[0];
1069
            *my= A[1];
1070
        }else{
1071
            *mx= mid_pred(A[0], B[0], C[0]);
1072
            *my= mid_pred(A[1], B[1], C[1]);
1073
        }
1074
    }
1075

    
1076
    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);
1077
}
1078

    
1079
/**
1080
 * gets the directionally predicted 16x8 MV.
1081
 * @param n the block index
1082
 * @param mx the x component of the predicted motion vector
1083
 * @param my the y component of the predicted motion vector
1084
 */
1085
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1086
    if(n==0){
1087
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1088
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1089

    
1090
        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);
1091

    
1092
        if(top_ref == ref){
1093
            *mx= B[0];
1094
            *my= B[1];
1095
            return;
1096
        }
1097
    }else{
1098
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1099
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1100

    
1101
        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);
1102

    
1103
        if(left_ref == ref){
1104
            *mx= A[0];
1105
            *my= A[1];
1106
            return;
1107
        }
1108
    }
1109

    
1110
    //RARE
1111
    pred_motion(h, n, 4, list, ref, mx, my);
1112
}
1113

    
1114
/**
1115
 * gets the directionally predicted 8x16 MV.
1116
 * @param n the block index
1117
 * @param mx the x component of the predicted motion vector
1118
 * @param my the y component of the predicted motion vector
1119
 */
1120
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1121
    if(n==0){
1122
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1123
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1124

    
1125
        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);
1126

    
1127
        if(left_ref == ref){
1128
            *mx= A[0];
1129
            *my= A[1];
1130
            return;
1131
        }
1132
    }else{
1133
        const int16_t * C;
1134
        int diagonal_ref;
1135

    
1136
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1137

    
1138
        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);
1139

    
1140
        if(diagonal_ref == ref){
1141
            *mx= C[0];
1142
            *my= C[1];
1143
            return;
1144
        }
1145
    }
1146

    
1147
    //RARE
1148
    pred_motion(h, n, 2, list, ref, mx, my);
1149
}
1150

    
1151
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1152
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1153
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1154

    
1155
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1156

    
1157
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1158
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1159
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1160

    
1161
        *mx = *my = 0;
1162
        return;
1163
    }
1164

    
1165
    pred_motion(h, 0, 4, 0, 0, mx, my);
1166

    
1167
    return;
1168
}
1169

    
1170
static inline void direct_dist_scale_factor(H264Context * const h){
1171
    const int poc = h->s.current_picture_ptr->poc;
1172
    const int poc1 = h->ref_list[1][0].poc;
1173
    int i;
1174
    for(i=0; i<h->ref_count[0]; i++){
1175
        int poc0 = h->ref_list[0][i].poc;
1176
        int td = clip(poc1 - poc0, -128, 127);
1177
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1178
            h->dist_scale_factor[i] = 256;
1179
        }else{
1180
            int tb = clip(poc - poc0, -128, 127);
1181
            int tx = (16384 + (ABS(td) >> 1)) / td;
1182
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1183
        }
1184
    }
1185
}
1186
static inline void direct_ref_list_init(H264Context * const h){
1187
    MpegEncContext * const s = &h->s;
1188
    Picture * const ref1 = &h->ref_list[1][0];
1189
    Picture * const cur = s->current_picture_ptr;
1190
    int list, i, j;
1191
    if(cur->pict_type == I_TYPE)
1192
        cur->ref_count[0] = 0;
1193
    if(cur->pict_type != B_TYPE)
1194
        cur->ref_count[1] = 0;
1195
    for(list=0; list<2; list++){
1196
        cur->ref_count[list] = h->ref_count[list];
1197
        for(j=0; j<h->ref_count[list]; j++)
1198
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1199
    }
1200
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1201
        return;
1202
    for(list=0; list<2; list++){
1203
        for(i=0; i<ref1->ref_count[list]; i++){
1204
            const int poc = ref1->ref_poc[list][i];
1205
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1206
            for(j=0; j<h->ref_count[list]; j++)
1207
                if(h->ref_list[list][j].poc == poc){
1208
                    h->map_col_to_list0[list][i] = j;
1209
                    break;
1210
                }
1211
        }
1212
    }
1213
}
1214

    
1215
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1216
    MpegEncContext * const s = &h->s;
1217
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1218
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1219
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1220
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1221
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1222
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1223
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1224
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1225
    const int is_b8x8 = IS_8X8(*mb_type);
1226
    int sub_mb_type;
1227
    int i8, i4;
1228

    
1229
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1230
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1231
         * so we know exactly what block size to use */
1232
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1233
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1234
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1235
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1236
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1237
    }else{
1238
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1239
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1240
    }
1241
    if(!is_b8x8)
1242
        *mb_type |= MB_TYPE_DIRECT2;
1243

    
1244
    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);
1245

    
1246
    if(h->direct_spatial_mv_pred){
1247
        int ref[2];
1248
        int mv[2][2];
1249
        int list;
1250

    
1251
        /* ref = min(neighbors) */
1252
        for(list=0; list<2; list++){
1253
            int refa = h->ref_cache[list][scan8[0] - 1];
1254
            int refb = h->ref_cache[list][scan8[0] - 8];
1255
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1256
            if(refc == -2)
1257
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1258
            ref[list] = refa;
1259
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1260
                ref[list] = refb;
1261
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1262
                ref[list] = refc;
1263
            if(ref[list] < 0)
1264
                ref[list] = -1;
1265
        }
1266

    
1267
        if(ref[0] < 0 && ref[1] < 0){
1268
            ref[0] = ref[1] = 0;
1269
            mv[0][0] = mv[0][1] =
1270
            mv[1][0] = mv[1][1] = 0;
1271
        }else{
1272
            for(list=0; list<2; list++){
1273
                if(ref[list] >= 0)
1274
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1275
                else
1276
                    mv[list][0] = mv[list][1] = 0;
1277
            }
1278
        }
1279

    
1280
        if(ref[1] < 0){
1281
            *mb_type &= ~MB_TYPE_P0L1;
1282
            sub_mb_type &= ~MB_TYPE_P0L1;
1283
        }else if(ref[0] < 0){
1284
            *mb_type &= ~MB_TYPE_P0L0;
1285
            sub_mb_type &= ~MB_TYPE_P0L0;
1286
        }
1287

    
1288
        if(IS_16X16(*mb_type)){
1289
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1290
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1291
            if(!IS_INTRA(mb_type_col)
1292
               && (   (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1293
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1294
                       && (h->x264_build>33 || !h->x264_build)))){
1295
                if(ref[0] > 0)
1296
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1297
                else
1298
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1299
                if(ref[1] > 0)
1300
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1301
                else
1302
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1303
            }else{
1304
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1305
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1306
            }
1307
        }else{
1308
            for(i8=0; i8<4; i8++){
1309
                const int x8 = i8&1;
1310
                const int y8 = i8>>1;
1311

    
1312
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1313
                    continue;
1314
                h->sub_mb_type[i8] = sub_mb_type;
1315

    
1316
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1317
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1318
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1319
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1320

    
1321
                /* col_zero_flag */
1322
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1323
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1324
                                                  && (h->x264_build>33 || !h->x264_build)))){
1325
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1326
                    if(IS_SUB_8X8(sub_mb_type)){
1327
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1328
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1329
                            if(ref[0] == 0)
1330
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1331
                            if(ref[1] == 0)
1332
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1333
                        }
1334
                    }else
1335
                    for(i4=0; i4<4; i4++){
1336
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1337
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1338
                            if(ref[0] == 0)
1339
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1340
                            if(ref[1] == 0)
1341
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1342
                        }
1343
                    }
1344
                }
1345
            }
1346
        }
1347
    }else{ /* direct temporal mv pred */
1348
        if(IS_16X16(*mb_type)){
1349
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1350
            if(IS_INTRA(mb_type_col)){
1351
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1352
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1353
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1354
            }else{
1355
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1356
                                                : h->map_col_to_list0[1][l1ref1[0]];
1357
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1358
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1359
                int mv_l0[2];
1360
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1361
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1362
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1363
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1364
                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);
1365
            }
1366
        }else{
1367
            for(i8=0; i8<4; i8++){
1368
                const int x8 = i8&1;
1369
                const int y8 = i8>>1;
1370
                int ref0, dist_scale_factor;
1371
                const int16_t (*l1mv)[2]= l1mv0;
1372

    
1373
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1374
                    continue;
1375
                h->sub_mb_type[i8] = sub_mb_type;
1376
                if(IS_INTRA(mb_type_col)){
1377
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1378
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1379
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1380
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1381
                    continue;
1382
                }
1383

    
1384
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1385
                if(ref0 >= 0)
1386
                    ref0 = h->map_col_to_list0[0][ref0];
1387
                else{
1388
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1389
                    l1mv= l1mv1;
1390
                }
1391
                dist_scale_factor = h->dist_scale_factor[ref0];
1392

    
1393
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1394
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1395
                if(IS_SUB_8X8(sub_mb_type)){
1396
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1397
                    int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
1398
                    int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
1399
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1400
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1401
                }else
1402
                for(i4=0; i4<4; i4++){
1403
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1404
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1405
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1406
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1407
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1408
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1409
                }
1410
            }
1411
        }
1412
    }
1413
}
1414

    
1415
static inline void write_back_motion(H264Context *h, int mb_type){
1416
    MpegEncContext * const s = &h->s;
1417
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1418
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1419
    int list;
1420

    
1421
    for(list=0; list<2; list++){
1422
        int y;
1423
        if(!USES_LIST(mb_type, list)){
1424
            if(1){ //FIXME skip or never read if mb_type doesn't use it
1425
                for(y=0; y<4; y++){
1426
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1427
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1428
                }
1429
                if( h->pps.cabac ) {
1430
                    /* FIXME needed ? */
1431
                    for(y=0; y<4; y++){
1432
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1433
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1434
                    }
1435
                }
1436
                for(y=0; y<2; y++){
1437
                    s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1438
                    s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1439
                }
1440
            }
1441
            continue;
1442
        }
1443

    
1444
        for(y=0; y<4; y++){
1445
            *(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];
1446
            *(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];
1447
        }
1448
        if( h->pps.cabac ) {
1449
            for(y=0; y<4; y++){
1450
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1451
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1452
            }
1453
        }
1454
        for(y=0; y<2; y++){
1455
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1456
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1457
        }
1458
    }
1459

    
1460
    if(h->slice_type == B_TYPE && h->pps.cabac){
1461
        if(IS_8X8(mb_type)){
1462
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1463
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1464
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1465
        }
1466
    }
1467
}
1468

    
1469
/**
1470
 * Decodes a network abstraction layer unit.
1471
 * @param consumed is the number of bytes used as input
1472
 * @param length is the length of the array
1473
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1474
 * @returns decoded bytes, might be src+1 if no escapes
1475
 */
1476
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1477
    int i, si, di;
1478
    uint8_t *dst;
1479

    
1480
//    src[0]&0x80;                //forbidden bit
1481
    h->nal_ref_idc= src[0]>>5;
1482
    h->nal_unit_type= src[0]&0x1F;
1483

    
1484
    src++; length--;
1485
#if 0
1486
    for(i=0; i<length; i++)
1487
        printf("%2X ", src[i]);
1488
#endif
1489
    for(i=0; i+1<length; i+=2){
1490
        if(src[i]) continue;
1491
        if(i>0 && src[i-1]==0) i--;
1492
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1493
            if(src[i+2]!=3){
1494
                /* startcode, so we must be past the end */
1495
                length=i;
1496
            }
1497
            break;
1498
        }
1499
    }
1500

    
1501
    if(i>=length-1){ //no escaped 0
1502
        *dst_length= length;
1503
        *consumed= length+1; //+1 for the header
1504
        return src;
1505
    }
1506

    
1507
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1508
    dst= h->rbsp_buffer;
1509

    
1510
//printf("decoding esc\n");
1511
    si=di=0;
1512
    while(si<length){
1513
        //remove escapes (very rare 1:2^22)
1514
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1515
            if(src[si+2]==3){ //escape
1516
                dst[di++]= 0;
1517
                dst[di++]= 0;
1518
                si+=3;
1519
                continue;
1520
            }else //next start code
1521
                break;
1522
        }
1523

    
1524
        dst[di++]= src[si++];
1525
    }
1526

    
1527
    *dst_length= di;
1528
    *consumed= si + 1;//+1 for the header
1529
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1530
    return dst;
1531
}
1532

    
1533
#if 0
1534
/**
1535
 * @param src the data which should be escaped
1536
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1537
 * @param length the length of the src data
1538
 * @param dst_length the length of the dst array
1539
 * @returns length of escaped data in bytes or -1 if an error occured
1540
 */
1541
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1542
    int i, escape_count, si, di;
1543
    uint8_t *temp;
1544

1545
    assert(length>=0);
1546
    assert(dst_length>0);
1547

1548
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1549

1550
    if(length==0) return 1;
1551

1552
    escape_count= 0;
1553
    for(i=0; i<length; i+=2){
1554
        if(src[i]) continue;
1555
        if(i>0 && src[i-1]==0)
1556
            i--;
1557
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1558
            escape_count++;
1559
            i+=2;
1560
        }
1561
    }
1562

1563
    if(escape_count==0){
1564
        if(dst+1 != src)
1565
            memcpy(dst+1, src, length);
1566
        return length + 1;
1567
    }
1568

1569
    if(length + escape_count + 1> dst_length)
1570
        return -1;
1571

1572
    //this should be damn rare (hopefully)
1573

1574
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1575
    temp= h->rbsp_buffer;
1576
//printf("encoding esc\n");
1577

1578
    si= 0;
1579
    di= 0;
1580
    while(si < length){
1581
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1582
            temp[di++]= 0; si++;
1583
            temp[di++]= 0; si++;
1584
            temp[di++]= 3;
1585
            temp[di++]= src[si++];
1586
        }
1587
        else
1588
            temp[di++]= src[si++];
1589
    }
1590
    memcpy(dst+1, temp, length+escape_count);
1591

1592
    assert(di == length+escape_count);
1593

1594
    return di + 1;
1595
}
1596

1597
/**
1598
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1599
 */
1600
static void encode_rbsp_trailing(PutBitContext *pb){
1601
    int length;
1602
    put_bits(pb, 1, 1);
1603
    length= (-put_bits_count(pb))&7;
1604
    if(length) put_bits(pb, length, 0);
1605
}
1606
#endif
1607

    
1608
/**
1609
 * identifies the exact end of the bitstream
1610
 * @return the length of the trailing, or 0 if damaged
1611
 */
1612
static int decode_rbsp_trailing(uint8_t *src){
1613
    int v= *src;
1614
    int r;
1615

    
1616
    tprintf("rbsp trailing %X\n", v);
1617

    
1618
    for(r=1; r<9; r++){
1619
        if(v&1) return r;
1620
        v>>=1;
1621
    }
1622
    return 0;
1623
}
1624

    
1625
/**
1626
 * idct tranforms the 16 dc values and dequantize them.
1627
 * @param qp quantization parameter
1628
 */
1629
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1630
#define stride 16
1631
    int i;
1632
    int temp[16]; //FIXME check if this is a good idea
1633
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1634
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1635

    
1636
//memset(block, 64, 2*256);
1637
//return;
1638
    for(i=0; i<4; i++){
1639
        const int offset= y_offset[i];
1640
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1641
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1642
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1643
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1644

    
1645
        temp[4*i+0]= z0+z3;
1646
        temp[4*i+1]= z1+z2;
1647
        temp[4*i+2]= z1-z2;
1648
        temp[4*i+3]= z0-z3;
1649
    }
1650

    
1651
    for(i=0; i<4; i++){
1652
        const int offset= x_offset[i];
1653
        const int z0= temp[4*0+i] + temp[4*2+i];
1654
        const int z1= temp[4*0+i] - temp[4*2+i];
1655
        const int z2= temp[4*1+i] - temp[4*3+i];
1656
        const int z3= temp[4*1+i] + temp[4*3+i];
1657

    
1658
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1659
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1660
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1661
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1662
    }
1663
}
1664

    
1665
#if 0
1666
/**
1667
 * dct tranforms the 16 dc values.
1668
 * @param qp quantization parameter ??? FIXME
1669
 */
1670
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1671
//    const int qmul= dequant_coeff[qp][0];
1672
    int i;
1673
    int temp[16]; //FIXME check if this is a good idea
1674
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1675
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1676

1677
    for(i=0; i<4; i++){
1678
        const int offset= y_offset[i];
1679
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1680
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1681
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1682
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1683

1684
        temp[4*i+0]= z0+z3;
1685
        temp[4*i+1]= z1+z2;
1686
        temp[4*i+2]= z1-z2;
1687
        temp[4*i+3]= z0-z3;
1688
    }
1689

1690
    for(i=0; i<4; i++){
1691
        const int offset= x_offset[i];
1692
        const int z0= temp[4*0+i] + temp[4*2+i];
1693
        const int z1= temp[4*0+i] - temp[4*2+i];
1694
        const int z2= temp[4*1+i] - temp[4*3+i];
1695
        const int z3= temp[4*1+i] + temp[4*3+i];
1696

1697
        block[stride*0 +offset]= (z0 + z3)>>1;
1698
        block[stride*2 +offset]= (z1 + z2)>>1;
1699
        block[stride*8 +offset]= (z1 - z2)>>1;
1700
        block[stride*10+offset]= (z0 - z3)>>1;
1701
    }
1702
}
1703
#endif
1704

    
1705
#undef xStride
1706
#undef stride
1707

    
1708
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1709
    const int stride= 16*2;
1710
    const int xStride= 16;
1711
    int a,b,c,d,e;
1712

    
1713
    a= block[stride*0 + xStride*0];
1714
    b= block[stride*0 + xStride*1];
1715
    c= block[stride*1 + xStride*0];
1716
    d= block[stride*1 + xStride*1];
1717

    
1718
    e= a-b;
1719
    a= a+b;
1720
    b= c-d;
1721
    c= c+d;
1722

    
1723
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1724
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1725
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1726
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1727
}
1728

    
1729
#if 0
1730
static void chroma_dc_dct_c(DCTELEM *block){
1731
    const int stride= 16*2;
1732
    const int xStride= 16;
1733
    int a,b,c,d,e;
1734

1735
    a= block[stride*0 + xStride*0];
1736
    b= block[stride*0 + xStride*1];
1737
    c= block[stride*1 + xStride*0];
1738
    d= block[stride*1 + xStride*1];
1739

1740
    e= a-b;
1741
    a= a+b;
1742
    b= c-d;
1743
    c= c+d;
1744

1745
    block[stride*0 + xStride*0]= (a+c);
1746
    block[stride*0 + xStride*1]= (e+b);
1747
    block[stride*1 + xStride*0]= (a-c);
1748
    block[stride*1 + xStride*1]= (e-b);
1749
}
1750
#endif
1751

    
1752
/**
1753
 * gets the chroma qp.
1754
 */
1755
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1756

    
1757
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1758
}
1759

    
1760

    
1761
#if 0
1762
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1763
    int i;
1764
    //FIXME try int temp instead of block
1765

1766
    for(i=0; i<4; i++){
1767
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1768
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1769
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1770
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1771
        const int z0= d0 + d3;
1772
        const int z3= d0 - d3;
1773
        const int z1= d1 + d2;
1774
        const int z2= d1 - d2;
1775

1776
        block[0 + 4*i]=   z0 +   z1;
1777
        block[1 + 4*i]= 2*z3 +   z2;
1778
        block[2 + 4*i]=   z0 -   z1;
1779
        block[3 + 4*i]=   z3 - 2*z2;
1780
    }
1781

1782
    for(i=0; i<4; i++){
1783
        const int z0= block[0*4 + i] + block[3*4 + i];
1784
        const int z3= block[0*4 + i] - block[3*4 + i];
1785
        const int z1= block[1*4 + i] + block[2*4 + i];
1786
        const int z2= block[1*4 + i] - block[2*4 + i];
1787

1788
        block[0*4 + i]=   z0 +   z1;
1789
        block[1*4 + i]= 2*z3 +   z2;
1790
        block[2*4 + i]=   z0 -   z1;
1791
        block[3*4 + i]=   z3 - 2*z2;
1792
    }
1793
}
1794
#endif
1795

    
1796
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1797
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1798
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1799
    int i;
1800
    const int * const quant_table= quant_coeff[qscale];
1801
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1802
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1803
    const unsigned int threshold2= (threshold1<<1);
1804
    int last_non_zero;
1805

    
1806
    if(seperate_dc){
1807
        if(qscale<=18){
1808
            //avoid overflows
1809
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1810
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1811
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1812

    
1813
            int level= block[0]*quant_coeff[qscale+18][0];
1814
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1815
                if(level>0){
1816
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1817
                    block[0]= level;
1818
                }else{
1819
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1820
                    block[0]= -level;
1821
                }
1822
//                last_non_zero = i;
1823
            }else{
1824
                block[0]=0;
1825
            }
1826
        }else{
1827
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1828
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1829
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1830

    
1831
            int level= block[0]*quant_table[0];
1832
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1833
                if(level>0){
1834
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1835
                    block[0]= level;
1836
                }else{
1837
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1838
                    block[0]= -level;
1839
                }
1840
//                last_non_zero = i;
1841
            }else{
1842
                block[0]=0;
1843
            }
1844
        }
1845
        last_non_zero= 0;
1846
        i=1;
1847
    }else{
1848
        last_non_zero= -1;
1849
        i=0;
1850
    }
1851

    
1852
    for(; i<16; i++){
1853
        const int j= scantable[i];
1854
        int level= block[j]*quant_table[j];
1855

    
1856
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1857
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1858
        if(((unsigned)(level+threshold1))>threshold2){
1859
            if(level>0){
1860
                level= (bias + level)>>QUANT_SHIFT;
1861
                block[j]= level;
1862
            }else{
1863
                level= (bias - level)>>QUANT_SHIFT;
1864
                block[j]= -level;
1865
            }
1866
            last_non_zero = i;
1867
        }else{
1868
            block[j]=0;
1869
        }
1870
    }
1871

    
1872
    return last_non_zero;
1873
}
1874

    
1875
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1876
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1877
    ((uint32_t*)(src+0*stride))[0]= a;
1878
    ((uint32_t*)(src+1*stride))[0]= a;
1879
    ((uint32_t*)(src+2*stride))[0]= a;
1880
    ((uint32_t*)(src+3*stride))[0]= a;
1881
}
1882

    
1883
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1884
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1885
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1886
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1887
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1888
}
1889

    
1890
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1891
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1892
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
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_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1901
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+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_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1910
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1911

    
1912
    ((uint32_t*)(src+0*stride))[0]=
1913
    ((uint32_t*)(src+1*stride))[0]=
1914
    ((uint32_t*)(src+2*stride))[0]=
1915
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1916
}
1917

    
1918
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1919
    ((uint32_t*)(src+0*stride))[0]=
1920
    ((uint32_t*)(src+1*stride))[0]=
1921
    ((uint32_t*)(src+2*stride))[0]=
1922
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1923
}
1924

    
1925

    
1926
#define LOAD_TOP_RIGHT_EDGE\
1927
    const int t4= topright[0];\
1928
    const int t5= topright[1];\
1929
    const int t6= topright[2];\
1930
    const int t7= topright[3];\
1931

    
1932
#define LOAD_LEFT_EDGE\
1933
    const int l0= src[-1+0*stride];\
1934
    const int l1= src[-1+1*stride];\
1935
    const int l2= src[-1+2*stride];\
1936
    const int l3= src[-1+3*stride];\
1937

    
1938
#define LOAD_TOP_EDGE\
1939
    const int t0= src[ 0-1*stride];\
1940
    const int t1= src[ 1-1*stride];\
1941
    const int t2= src[ 2-1*stride];\
1942
    const int t3= src[ 3-1*stride];\
1943

    
1944
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1945
    const int lt= src[-1-1*stride];
1946
    LOAD_TOP_EDGE
1947
    LOAD_LEFT_EDGE
1948

    
1949
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1950
    src[0+2*stride]=
1951
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1952
    src[0+1*stride]=
1953
    src[1+2*stride]=
1954
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1955
    src[0+0*stride]=
1956
    src[1+1*stride]=
1957
    src[2+2*stride]=
1958
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1959
    src[1+0*stride]=
1960
    src[2+1*stride]=
1961
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1962
    src[2+0*stride]=
1963
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1964
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1965
}
1966

    
1967
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1968
    LOAD_TOP_EDGE
1969
    LOAD_TOP_RIGHT_EDGE
1970
//    LOAD_LEFT_EDGE
1971

    
1972
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1973
    src[1+0*stride]=
1974
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1975
    src[2+0*stride]=
1976
    src[1+1*stride]=
1977
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1978
    src[3+0*stride]=
1979
    src[2+1*stride]=
1980
    src[1+2*stride]=
1981
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1982
    src[3+1*stride]=
1983
    src[2+2*stride]=
1984
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1985
    src[3+2*stride]=
1986
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1987
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1988
}
1989

    
1990
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1991
    const int lt= src[-1-1*stride];
1992
    LOAD_TOP_EDGE
1993
    LOAD_LEFT_EDGE
1994
    const __attribute__((unused)) int unu= l3;
1995

    
1996
    src[0+0*stride]=
1997
    src[1+2*stride]=(lt + t0 + 1)>>1;
1998
    src[1+0*stride]=
1999
    src[2+2*stride]=(t0 + t1 + 1)>>1;
2000
    src[2+0*stride]=
2001
    src[3+2*stride]=(t1 + t2 + 1)>>1;
2002
    src[3+0*stride]=(t2 + t3 + 1)>>1;
2003
    src[0+1*stride]=
2004
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2005
    src[1+1*stride]=
2006
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2007
    src[2+1*stride]=
2008
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2009
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2010
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2011
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2012
}
2013

    
2014
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
2015
    LOAD_TOP_EDGE
2016
    LOAD_TOP_RIGHT_EDGE
2017
    const __attribute__((unused)) int unu= t7;
2018

    
2019
    src[0+0*stride]=(t0 + t1 + 1)>>1;
2020
    src[1+0*stride]=
2021
    src[0+2*stride]=(t1 + t2 + 1)>>1;
2022
    src[2+0*stride]=
2023
    src[1+2*stride]=(t2 + t3 + 1)>>1;
2024
    src[3+0*stride]=
2025
    src[2+2*stride]=(t3 + t4+ 1)>>1;
2026
    src[3+2*stride]=(t4 + t5+ 1)>>1;
2027
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2028
    src[1+1*stride]=
2029
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2030
    src[2+1*stride]=
2031
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2032
    src[3+1*stride]=
2033
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2034
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2035
}
2036

    
2037
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2038
    LOAD_LEFT_EDGE
2039

    
2040
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2041
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2042
    src[2+0*stride]=
2043
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2044
    src[3+0*stride]=
2045
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2046
    src[2+1*stride]=
2047
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2048
    src[3+1*stride]=
2049
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2050
    src[3+2*stride]=
2051
    src[1+3*stride]=
2052
    src[0+3*stride]=
2053
    src[2+2*stride]=
2054
    src[2+3*stride]=
2055
    src[3+3*stride]=l3;
2056
}
2057

    
2058
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2059
    const int lt= src[-1-1*stride];
2060
    LOAD_TOP_EDGE
2061
    LOAD_LEFT_EDGE
2062
    const __attribute__((unused)) int unu= t3;
2063

    
2064
    src[0+0*stride]=
2065
    src[2+1*stride]=(lt + l0 + 1)>>1;
2066
    src[1+0*stride]=
2067
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2068
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2069
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2070
    src[0+1*stride]=
2071
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2072
    src[1+1*stride]=
2073
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2074
    src[0+2*stride]=
2075
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2076
    src[1+2*stride]=
2077
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2078
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2079
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2080
}
2081

    
2082
static void pred16x16_vertical_c(uint8_t *src, int stride){
2083
    int i;
2084
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2085
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2086
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2087
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2088

    
2089
    for(i=0; i<16; i++){
2090
        ((uint32_t*)(src+i*stride))[0]= a;
2091
        ((uint32_t*)(src+i*stride))[1]= b;
2092
        ((uint32_t*)(src+i*stride))[2]= c;
2093
        ((uint32_t*)(src+i*stride))[3]= d;
2094
    }
2095
}
2096

    
2097
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2098
    int i;
2099

    
2100
    for(i=0; i<16; i++){
2101
        ((uint32_t*)(src+i*stride))[0]=
2102
        ((uint32_t*)(src+i*stride))[1]=
2103
        ((uint32_t*)(src+i*stride))[2]=
2104
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2105
    }
2106
}
2107

    
2108
static void pred16x16_dc_c(uint8_t *src, int stride){
2109
    int i, dc=0;
2110

    
2111
    for(i=0;i<16; i++){
2112
        dc+= src[-1+i*stride];
2113
    }
2114

    
2115
    for(i=0;i<16; i++){
2116
        dc+= src[i-stride];
2117
    }
2118

    
2119
    dc= 0x01010101*((dc + 16)>>5);
2120

    
2121
    for(i=0; i<16; i++){
2122
        ((uint32_t*)(src+i*stride))[0]=
2123
        ((uint32_t*)(src+i*stride))[1]=
2124
        ((uint32_t*)(src+i*stride))[2]=
2125
        ((uint32_t*)(src+i*stride))[3]= dc;
2126
    }
2127
}
2128

    
2129
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2130
    int i, dc=0;
2131

    
2132
    for(i=0;i<16; i++){
2133
        dc+= src[-1+i*stride];
2134
    }
2135

    
2136
    dc= 0x01010101*((dc + 8)>>4);
2137

    
2138
    for(i=0; i<16; i++){
2139
        ((uint32_t*)(src+i*stride))[0]=
2140
        ((uint32_t*)(src+i*stride))[1]=
2141
        ((uint32_t*)(src+i*stride))[2]=
2142
        ((uint32_t*)(src+i*stride))[3]= dc;
2143
    }
2144
}
2145

    
2146
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2147
    int i, dc=0;
2148

    
2149
    for(i=0;i<16; i++){
2150
        dc+= src[i-stride];
2151
    }
2152
    dc= 0x01010101*((dc + 8)>>4);
2153

    
2154
    for(i=0; i<16; i++){
2155
        ((uint32_t*)(src+i*stride))[0]=
2156
        ((uint32_t*)(src+i*stride))[1]=
2157
        ((uint32_t*)(src+i*stride))[2]=
2158
        ((uint32_t*)(src+i*stride))[3]= dc;
2159
    }
2160
}
2161

    
2162
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2163
    int i;
2164

    
2165
    for(i=0; i<16; i++){
2166
        ((uint32_t*)(src+i*stride))[0]=
2167
        ((uint32_t*)(src+i*stride))[1]=
2168
        ((uint32_t*)(src+i*stride))[2]=
2169
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2170
    }
2171
}
2172

    
2173
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2174
  int i, j, k;
2175
  int a;
2176
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2177
  const uint8_t * const src0 = src+7-stride;
2178
  const uint8_t *src1 = src+8*stride-1;
2179
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2180
  int H = src0[1] - src0[-1];
2181
  int V = src1[0] - src2[ 0];
2182
  for(k=2; k<=8; ++k) {
2183
    src1 += stride; src2 -= stride;
2184
    H += k*(src0[k] - src0[-k]);
2185
    V += k*(src1[0] - src2[ 0]);
2186
  }
2187
  if(svq3){
2188
    H = ( 5*(H/4) ) / 16;
2189
    V = ( 5*(V/4) ) / 16;
2190

    
2191
    /* required for 100% accuracy */
2192
    i = H; H = V; V = i;
2193
  }else{
2194
    H = ( 5*H+32 ) >> 6;
2195
    V = ( 5*V+32 ) >> 6;
2196
  }
2197

    
2198
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2199
  for(j=16; j>0; --j) {
2200
    int b = a;
2201
    a += V;
2202
    for(i=-16; i<0; i+=4) {
2203
      src[16+i] = cm[ (b    ) >> 5 ];
2204
      src[17+i] = cm[ (b+  H) >> 5 ];
2205
      src[18+i] = cm[ (b+2*H) >> 5 ];
2206
      src[19+i] = cm[ (b+3*H) >> 5 ];
2207
      b += 4*H;
2208
    }
2209
    src += stride;
2210
  }
2211
}
2212

    
2213
static void pred16x16_plane_c(uint8_t *src, int stride){
2214
    pred16x16_plane_compat_c(src, stride, 0);
2215
}
2216

    
2217
static void pred8x8_vertical_c(uint8_t *src, int stride){
2218
    int i;
2219
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2220
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2221

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

    
2228
static void pred8x8_horizontal_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]= src[-1+i*stride]*0x01010101;
2234
    }
2235
}
2236

    
2237
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2238
    int i;
2239

    
2240
    for(i=0; i<8; i++){
2241
        ((uint32_t*)(src+i*stride))[0]=
2242
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2243
    }
2244
}
2245

    
2246
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2247
    int i;
2248
    int dc0, dc2;
2249

    
2250
    dc0=dc2=0;
2251
    for(i=0;i<4; i++){
2252
        dc0+= src[-1+i*stride];
2253
        dc2+= src[-1+(i+4)*stride];
2254
    }
2255
    dc0= 0x01010101*((dc0 + 2)>>2);
2256
    dc2= 0x01010101*((dc2 + 2)>>2);
2257

    
2258
    for(i=0; i<4; i++){
2259
        ((uint32_t*)(src+i*stride))[0]=
2260
        ((uint32_t*)(src+i*stride))[1]= dc0;
2261
    }
2262
    for(i=4; i<8; i++){
2263
        ((uint32_t*)(src+i*stride))[0]=
2264
        ((uint32_t*)(src+i*stride))[1]= dc2;
2265
    }
2266
}
2267

    
2268
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2269
    int i;
2270
    int dc0, dc1;
2271

    
2272
    dc0=dc1=0;
2273
    for(i=0;i<4; i++){
2274
        dc0+= src[i-stride];
2275
        dc1+= src[4+i-stride];
2276
    }
2277
    dc0= 0x01010101*((dc0 + 2)>>2);
2278
    dc1= 0x01010101*((dc1 + 2)>>2);
2279

    
2280
    for(i=0; i<4; i++){
2281
        ((uint32_t*)(src+i*stride))[0]= dc0;
2282
        ((uint32_t*)(src+i*stride))[1]= dc1;
2283
    }
2284
    for(i=4; i<8; i++){
2285
        ((uint32_t*)(src+i*stride))[0]= dc0;
2286
        ((uint32_t*)(src+i*stride))[1]= dc1;
2287
    }
2288
}
2289

    
2290

    
2291
static void pred8x8_dc_c(uint8_t *src, int stride){
2292
    int i;
2293
    int dc0, dc1, dc2, dc3;
2294

    
2295
    dc0=dc1=dc2=0;
2296
    for(i=0;i<4; i++){
2297
        dc0+= src[-1+i*stride] + src[i-stride];
2298
        dc1+= src[4+i-stride];
2299
        dc2+= src[-1+(i+4)*stride];
2300
    }
2301
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2302
    dc0= 0x01010101*((dc0 + 4)>>3);
2303
    dc1= 0x01010101*((dc1 + 2)>>2);
2304
    dc2= 0x01010101*((dc2 + 2)>>2);
2305

    
2306
    for(i=0; i<4; i++){
2307
        ((uint32_t*)(src+i*stride))[0]= dc0;
2308
        ((uint32_t*)(src+i*stride))[1]= dc1;
2309
    }
2310
    for(i=4; i<8; i++){
2311
        ((uint32_t*)(src+i*stride))[0]= dc2;
2312
        ((uint32_t*)(src+i*stride))[1]= dc3;
2313
    }
2314
}
2315

    
2316
static void pred8x8_plane_c(uint8_t *src, int stride){
2317
  int j, k;
2318
  int a;
2319
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2320
  const uint8_t * const src0 = src+3-stride;
2321
  const uint8_t *src1 = src+4*stride-1;
2322
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2323
  int H = src0[1] - src0[-1];
2324
  int V = src1[0] - src2[ 0];
2325
  for(k=2; k<=4; ++k) {
2326
    src1 += stride; src2 -= stride;
2327
    H += k*(src0[k] - src0[-k]);
2328
    V += k*(src1[0] - src2[ 0]);
2329
  }
2330
  H = ( 17*H+16 ) >> 5;
2331
  V = ( 17*V+16 ) >> 5;
2332

    
2333
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2334
  for(j=8; j>0; --j) {
2335
    int b = a;
2336
    a += V;
2337
    src[0] = cm[ (b    ) >> 5 ];
2338
    src[1] = cm[ (b+  H) >> 5 ];
2339
    src[2] = cm[ (b+2*H) >> 5 ];
2340
    src[3] = cm[ (b+3*H) >> 5 ];
2341
    src[4] = cm[ (b+4*H) >> 5 ];
2342
    src[5] = cm[ (b+5*H) >> 5 ];
2343
    src[6] = cm[ (b+6*H) >> 5 ];
2344
    src[7] = cm[ (b+7*H) >> 5 ];
2345
    src += stride;
2346
  }
2347
}
2348

    
2349
#define SRC(x,y) src[(x)+(y)*stride]
2350
#define PL(y) \
2351
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2352
#define PREDICT_8x8_LOAD_LEFT \
2353
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2354
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2355
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2356
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2357

    
2358
#define PT(x) \
2359
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2360
#define PREDICT_8x8_LOAD_TOP \
2361
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2362
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2363
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2364
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2365
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2366

    
2367
#define PTR(x) \
2368
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2369
#define PREDICT_8x8_LOAD_TOPRIGHT \
2370
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2371
    if(has_topright) { \
2372
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2373
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2374
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2375

    
2376
#define PREDICT_8x8_LOAD_TOPLEFT \
2377
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2378

    
2379
#define PREDICT_8x8_DC(v) \
2380
    int y; \
2381
    for( y = 0; y < 8; y++ ) { \
2382
        ((uint32_t*)src)[0] = \
2383
        ((uint32_t*)src)[1] = v; \
2384
        src += stride; \
2385
    }
2386

    
2387
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2388
{
2389
    PREDICT_8x8_DC(0x80808080);
2390
}
2391
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2392
{
2393
    PREDICT_8x8_LOAD_LEFT;
2394
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2395
    PREDICT_8x8_DC(dc);
2396
}
2397
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2398
{
2399
    PREDICT_8x8_LOAD_TOP;
2400
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2401
    PREDICT_8x8_DC(dc);
2402
}
2403
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2404
{
2405
    PREDICT_8x8_LOAD_LEFT;
2406
    PREDICT_8x8_LOAD_TOP;
2407
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2408
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2409
    PREDICT_8x8_DC(dc);
2410
}
2411
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2412
{
2413
    PREDICT_8x8_LOAD_LEFT;
2414
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2415
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2416
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2417
#undef ROW
2418
}
2419
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2420
{
2421
    int y;
2422
    PREDICT_8x8_LOAD_TOP;
2423
    src[0] = t0;
2424
    src[1] = t1;
2425
    src[2] = t2;
2426
    src[3] = t3;
2427
    src[4] = t4;
2428
    src[5] = t5;
2429
    src[6] = t6;
2430
    src[7] = t7;
2431
    for( y = 1; y < 8; y++ )
2432
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2433
}
2434
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2435
{
2436
    PREDICT_8x8_LOAD_TOP;
2437
    PREDICT_8x8_LOAD_TOPRIGHT;
2438
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2439
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2440
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2441
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2442
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2443
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2444
    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;
2445
    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;
2446
    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;
2447
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2448
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2449
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2450
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2451
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2452
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2453
}
2454
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2455
{
2456
    PREDICT_8x8_LOAD_TOP;
2457
    PREDICT_8x8_LOAD_LEFT;
2458
    PREDICT_8x8_LOAD_TOPLEFT;
2459
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2460
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2461
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2462
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2463
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2464
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2465
    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;
2466
    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;
2467
    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;
2468
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2469
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2470
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2471
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2472
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2473
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2474

    
2475
}
2476
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2477
{
2478
    PREDICT_8x8_LOAD_TOP;
2479
    PREDICT_8x8_LOAD_LEFT;
2480
    PREDICT_8x8_LOAD_TOPLEFT;
2481
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2482
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2483
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2484
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2485
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2486
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2487
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2488
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2489
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2490
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2491
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2492
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2493
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2494
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2495
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2496
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2497
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2498
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2499
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2500
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2501
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2502
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2503
}
2504
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2505
{
2506
    PREDICT_8x8_LOAD_TOP;
2507
    PREDICT_8x8_LOAD_LEFT;
2508
    PREDICT_8x8_LOAD_TOPLEFT;
2509
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2510
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2511
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2512
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2513
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2514
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2515
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2516
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2517
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2518
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2519
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2520
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2521
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2522
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2523
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2524
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2525
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2526
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2527
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2528
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2529
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2530
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2531
}
2532
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2533
{
2534
    PREDICT_8x8_LOAD_TOP;
2535
    PREDICT_8x8_LOAD_TOPRIGHT;
2536
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2537
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2538
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2539
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2540
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2541
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2542
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2543
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2544
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2545
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2546
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2547
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2548
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2549
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2550
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2551
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2552
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2553
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2554
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2555
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2556
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2557
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2558
}
2559
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2560
{
2561
    PREDICT_8x8_LOAD_LEFT;
2562
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2563
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2564
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2565
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2566
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2567
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2568
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2569
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2570
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2571
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2572
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2573
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2574
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2575
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2576
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2577
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2578
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2579
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2580
}
2581
#undef PREDICT_8x8_LOAD_LEFT
2582
#undef PREDICT_8x8_LOAD_TOP
2583
#undef PREDICT_8x8_LOAD_TOPLEFT
2584
#undef PREDICT_8x8_LOAD_TOPRIGHT
2585
#undef PREDICT_8x8_DC
2586
#undef PTR
2587
#undef PT
2588
#undef PL
2589
#undef SRC
2590

    
2591
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2592
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2593
                           int src_x_offset, int src_y_offset,
2594
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2595
    MpegEncContext * const s = &h->s;
2596
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2597
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2598
    const int luma_xy= (mx&3) + ((my&3)<<2);
2599
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2600
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2601
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2602
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2603
    int extra_height= extra_width;
2604
    int emu=0;
2605
    const int full_mx= mx>>2;
2606
    const int full_my= my>>2;
2607
    const int pic_width  = 16*s->mb_width;
2608
    const int pic_height = 16*s->mb_height;
2609

    
2610
    if(!pic->data[0])
2611
        return;
2612

    
2613
    if(mx&7) extra_width -= 3;
2614
    if(my&7) extra_height -= 3;
2615

    
2616
    if(   full_mx < 0-extra_width
2617
       || full_my < 0-extra_height
2618
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2619
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2620
        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);
2621
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2622
        emu=1;
2623
    }
2624

    
2625
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2626
    if(!square){
2627
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2628
    }
2629

    
2630
    if(s->flags&CODEC_FLAG_GRAY) return;
2631

    
2632
    if(emu){
2633
        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);
2634
            src_cb= s->edge_emu_buffer;
2635
    }
2636
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2637

    
2638
    if(emu){
2639
        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);
2640
            src_cr= s->edge_emu_buffer;
2641
    }
2642
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2643
}
2644

    
2645
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2646
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2647
                           int x_offset, int y_offset,
2648
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2649
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2650
                           int list0, int list1){
2651
    MpegEncContext * const s = &h->s;
2652
    qpel_mc_func *qpix_op=  qpix_put;
2653
    h264_chroma_mc_func chroma_op= chroma_put;
2654

    
2655
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2656
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2657
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2658
    x_offset += 8*s->mb_x;
2659
    y_offset += 8*s->mb_y;
2660

    
2661
    if(list0){
2662
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2663
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2664
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2665
                           qpix_op, chroma_op);
2666

    
2667
        qpix_op=  qpix_avg;
2668
        chroma_op= chroma_avg;
2669
    }
2670

    
2671
    if(list1){
2672
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2673
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2674
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2675
                           qpix_op, chroma_op);
2676
    }
2677
}
2678

    
2679
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2680
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2681
                           int x_offset, int y_offset,
2682
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2683
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2684
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2685
                           int list0, int list1){
2686
    MpegEncContext * const s = &h->s;
2687

    
2688
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2689
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2690
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2691
    x_offset += 8*s->mb_x;
2692
    y_offset += 8*s->mb_y;
2693

    
2694
    if(list0 && list1){
2695
        /* don't optimize for luma-only case, since B-frames usually
2696
         * use implicit weights => chroma too. */
2697
        uint8_t *tmp_cb = s->obmc_scratchpad;
2698
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2699
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2700
        int refn0 = h->ref_cache[0][ scan8[n] ];
2701
        int refn1 = h->ref_cache[1][ scan8[n] ];
2702

    
2703
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2704
                    dest_y, dest_cb, dest_cr,
2705
                    x_offset, y_offset, qpix_put, chroma_put);
2706
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2707
                    tmp_y, tmp_cb, tmp_cr,
2708
                    x_offset, y_offset, qpix_put, chroma_put);
2709

    
2710
        if(h->use_weight == 2){
2711
            int weight0 = h->implicit_weight[refn0][refn1];
2712
            int weight1 = 64 - weight0;
2713
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0);
2714
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
2715
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
2716
        }else{
2717
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2718
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2719
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2720
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2721
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2722
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2723
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2724
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2725
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2726
        }
2727
    }else{
2728
        int list = list1 ? 1 : 0;
2729
        int refn = h->ref_cache[list][ scan8[n] ];
2730
        Picture *ref= &h->ref_list[list][refn];
2731
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2732
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2733
                    qpix_put, chroma_put);
2734

    
2735
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2736
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2737
        if(h->use_weight_chroma){
2738
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2739
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2740
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2741
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2742
        }
2743
    }
2744
}
2745

    
2746
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2747
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2748
                           int x_offset, int y_offset,
2749
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2750
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2751
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2752
                           int list0, int list1){
2753
    if((h->use_weight==2 && list0 && list1
2754
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2755
       || h->use_weight==1)
2756
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2757
                         x_offset, y_offset, qpix_put, chroma_put,
2758
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2759
    else
2760
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2761
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2762
}
2763

    
2764
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2765
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2766
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2767
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2768
    MpegEncContext * const s = &h->s;
2769
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2770
    const int mb_type= s->current_picture.mb_type[mb_xy];
2771

    
2772
    assert(IS_INTER(mb_type));
2773

    
2774
    if(IS_16X16(mb_type)){
2775
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2776
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2777
                &weight_op[0], &weight_avg[0],
2778
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2779
    }else if(IS_16X8(mb_type)){
2780
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2781
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2782
                &weight_op[1], &weight_avg[1],
2783
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2784
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2785
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2786
                &weight_op[1], &weight_avg[1],
2787
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2788
    }else if(IS_8X16(mb_type)){
2789
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2790
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2791
                &weight_op[2], &weight_avg[2],
2792
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2793
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2794
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2795
                &weight_op[2], &weight_avg[2],
2796
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2797
    }else{
2798
        int i;
2799

    
2800
        assert(IS_8X8(mb_type));
2801

    
2802
        for(i=0; i<4; i++){
2803
            const int sub_mb_type= h->sub_mb_type[i];
2804
            const int n= 4*i;
2805
            int x_offset= (i&1)<<2;
2806
            int y_offset= (i&2)<<1;
2807

    
2808
            if(IS_SUB_8X8(sub_mb_type)){
2809
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2810
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2811
                    &weight_op[3], &weight_avg[3],
2812
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2813
            }else if(IS_SUB_8X4(sub_mb_type)){
2814
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2815
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2816
                    &weight_op[4], &weight_avg[4],
2817
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2818
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2819
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2820
                    &weight_op[4], &weight_avg[4],
2821
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2822
            }else if(IS_SUB_4X8(sub_mb_type)){
2823
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2824
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2825
                    &weight_op[5], &weight_avg[5],
2826
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2827
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2828
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2829
                    &weight_op[5], &weight_avg[5],
2830
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2831
            }else{
2832
                int j;
2833
                assert(IS_SUB_4X4(sub_mb_type));
2834
                for(j=0; j<4; j++){
2835
                    int sub_x_offset= x_offset + 2*(j&1);
2836
                    int sub_y_offset= y_offset +   (j&2);
2837
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2838
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2839
                        &weight_op[6], &weight_avg[6],
2840
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2841
                }
2842
            }
2843
        }
2844
    }
2845
}
2846

    
2847
static void decode_init_vlc(H264Context *h){
2848
    static int done = 0;
2849

    
2850
    if (!done) {
2851
        int i;
2852
        done = 1;
2853

    
2854
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2855
                 &chroma_dc_coeff_token_len [0], 1, 1,
2856
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2857

    
2858
        for(i=0; i<4; i++){
2859
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2860
                     &coeff_token_len [i][0], 1, 1,
2861
                     &coeff_token_bits[i][0], 1, 1, 1);
2862
        }
2863

    
2864
        for(i=0; i<3; i++){
2865
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2866
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2867
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2868
        }
2869
        for(i=0; i<15; i++){
2870
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2871
                     &total_zeros_len [i][0], 1, 1,
2872
                     &total_zeros_bits[i][0], 1, 1, 1);
2873
        }
2874

    
2875
        for(i=0; i<6; i++){
2876
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2877
                     &run_len [i][0], 1, 1,
2878
                     &run_bits[i][0], 1, 1, 1);
2879
        }
2880
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2881
                 &run_len [6][0], 1, 1,
2882
                 &run_bits[6][0], 1, 1, 1);
2883
    }
2884
}
2885

    
2886
/**
2887
 * Sets the intra prediction function pointers.
2888
 */
2889
static void init_pred_ptrs(H264Context *h){
2890
//    MpegEncContext * const s = &h->s;
2891

    
2892
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2893
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2894
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2895
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2896
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2897
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2898
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2899
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2900
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2901
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2902
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2903
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2904

    
2905
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2906
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2907
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2908
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2909
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2910
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2911
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2912
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2913
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2914
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2915
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2916
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2917

    
2918
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2919
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2920
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2921
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2922
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2923
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2924
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2925

    
2926
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2927
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2928
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2929
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2930
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2931
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2932
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2933
}
2934

    
2935
static void free_tables(H264Context *h){
2936
    av_freep(&h->intra4x4_pred_mode);
2937
    av_freep(&h->chroma_pred_mode_table);
2938
    av_freep(&h->cbp_table);
2939
    av_freep(&h->mvd_table[0]);
2940
    av_freep(&h->mvd_table[1]);
2941
    av_freep(&h->direct_table);
2942
    av_freep(&h->non_zero_count);
2943
    av_freep(&h->slice_table_base);
2944
    av_freep(&h->top_borders[1]);
2945
    av_freep(&h->top_borders[0]);
2946
    h->slice_table= NULL;
2947

    
2948
    av_freep(&h->mb2b_xy);
2949
    av_freep(&h->mb2b8_xy);
2950

    
2951
    av_freep(&h->s.obmc_scratchpad);
2952
}
2953

    
2954
static void init_dequant8_coeff_table(H264Context *h){
2955
    int i,q,x;
2956
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2957
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2958

    
2959
    for(i=0; i<2; i++ ){
2960
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2961
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2962
            break;
2963
        }
2964

    
2965
        for(q=0; q<52; q++){
2966
            int shift = div6[q];
2967
            int idx = rem6[q];
2968
            for(x=0; x<64; x++)
2969
                h->dequant8_coeff[i][q][x] = ((uint32_t)dequant8_coeff_init[idx][
2970
                    dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] * h->pps.scaling_matrix8[i][x]) << shift;
2971
        }
2972
    }
2973
}
2974

    
2975
static void init_dequant4_coeff_table(H264Context *h){
2976
    int i,j,q,x;
2977
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2978
    for(i=0; i<6; i++ ){
2979
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2980
        for(j=0; j<i; j++){
2981
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2982
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2983
                break;
2984
            }
2985
        }
2986
        if(j<i)
2987
            continue;
2988

    
2989
        for(q=0; q<52; q++){
2990
            int shift = div6[q] + 2;
2991
            int idx = rem6[q];
2992
            for(x=0; x<16; x++)
2993
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2994
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2995
                    h->pps.scaling_matrix4[i][x]) << shift;
2996
        }
2997
    }
2998
}
2999

    
3000
static void init_dequant_tables(H264Context *h){
3001
    int i,x;
3002
    init_dequant4_coeff_table(h);
3003
    if(h->pps.transform_8x8_mode)
3004
        init_dequant8_coeff_table(h);
3005
    if(h->sps.transform_bypass){
3006
        for(i=0; i<6; i++)
3007
            for(x=0; x<16; x++)
3008
                h->dequant4_coeff[i][0][x] = 1<<6;
3009
        if(h->pps.transform_8x8_mode)
3010
            for(i=0; i<2; i++)
3011
                for(x=0; x<64; x++)
3012
                    h->dequant8_coeff[i][0][x] = 1<<6;
3013
    }
3014
}
3015

    
3016

    
3017
/**
3018
 * allocates tables.
3019
 * needs width/height
3020
 */
3021
static int alloc_tables(H264Context *h){
3022
    MpegEncContext * const s = &h->s;
3023
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
3024
    int x,y;
3025

    
3026
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
3027

    
3028
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
3029
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
3030
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3031
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
3032
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
3033

    
3034
    if( h->pps.cabac ) {
3035
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
3036
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3037
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
3038
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
3039
    }
3040

    
3041
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
3042
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
3043

    
3044
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
3045
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
3046
    for(y=0; y<s->mb_height; y++){
3047
        for(x=0; x<s->mb_width; x++){
3048
            const int mb_xy= x + y*s->mb_stride;
3049
            const int b_xy = 4*x + 4*y*h->b_stride;
3050
            const int b8_xy= 2*x + 2*y*h->b8_stride;
3051

    
3052
            h->mb2b_xy [mb_xy]= b_xy;
3053
            h->mb2b8_xy[mb_xy]= b8_xy;
3054
        }
3055
    }
3056

    
3057
    s->obmc_scratchpad = NULL;
3058

    
3059
    if(!h->dequant4_coeff[0])
3060
        init_dequant_tables(h);
3061

    
3062
    return 0;
3063
fail:
3064
    free_tables(h);
3065
    return -1;
3066
}
3067

    
3068
static void common_init(H264Context *h){
3069
    MpegEncContext * const s = &h->s;
3070

    
3071
    s->width = s->avctx->width;
3072
    s->height = s->avctx->height;
3073
    s->codec_id= s->avctx->codec->id;
3074

    
3075
    init_pred_ptrs(h);
3076

    
3077
    h->dequant_coeff_pps= -1;
3078
    s->unrestricted_mv=1;
3079
    s->decode=1; //FIXME
3080

    
3081
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3082
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
3083
}
3084

    
3085
static int decode_init(AVCodecContext *avctx){
3086
    H264Context *h= avctx->priv_data;
3087
    MpegEncContext * const s = &h->s;
3088

    
3089
    MPV_decode_defaults(s);
3090

    
3091
    s->avctx = avctx;
3092
    common_init(h);
3093

    
3094
    s->out_format = FMT_H264;
3095
    s->workaround_bugs= avctx->workaround_bugs;
3096

    
3097
    // set defaults
3098
//    s->decode_mb= ff_h263_decode_mb;
3099
    s->low_delay= 1;
3100
    avctx->pix_fmt= PIX_FMT_YUV420P;
3101

    
3102
    decode_init_vlc(h);
3103

    
3104
    if(avctx->extradata_size > 0 && avctx->extradata &&
3105
       *(char *)avctx->extradata == 1){
3106
        h->is_avc = 1;
3107
        h->got_avcC = 0;
3108
    } else {
3109
        h->is_avc = 0;
3110
    }
3111

    
3112
    return 0;
3113
}
3114

    
3115
static int frame_start(H264Context *h){
3116
    MpegEncContext * const s = &h->s;
3117
    int i;
3118

    
3119
    if(MPV_frame_start(s, s->avctx) < 0)
3120
        return -1;
3121
    ff_er_frame_start(s);
3122

    
3123
    assert(s->linesize && s->uvlinesize);
3124

    
3125
    for(i=0; i<16; i++){
3126
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3127
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3128
    }
3129
    for(i=0; i<4; i++){
3130
        h->block_offset[16+i]=
3131
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3132
        h->block_offset[24+16+i]=
3133
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3134
    }
3135

    
3136
    /* can't be in alloc_tables because linesize isn't known there.
3137
     * FIXME: redo bipred weight to not require extra buffer? */
3138
    if(!s->obmc_scratchpad)
3139
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3140

    
3141
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3142
    return 0;
3143
}
3144

    
3145
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3146
    MpegEncContext * const s = &h->s;
3147
    int i;
3148

    
3149
    src_y  -=   linesize;
3150
    src_cb -= uvlinesize;
3151
    src_cr -= uvlinesize;
3152

    
3153
    // There are two lines saved, the line above the the top macroblock of a pair,
3154
    // and the line above the bottom macroblock
3155
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3156
    for(i=1; i<17; i++){
3157
        h->left_border[i]= src_y[15+i*  linesize];
3158
    }
3159

    
3160
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3161
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3162

    
3163
    if(!(s->flags&CODEC_FLAG_GRAY)){
3164
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3165
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3166
        for(i=1; i<9; i++){
3167
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3168
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3169
        }
3170
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3171
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3172
    }
3173
}
3174

    
3175
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){
3176
    MpegEncContext * const s = &h->s;
3177
    int temp8, i;
3178
    uint64_t temp64;
3179
    int deblock_left = (s->mb_x > 0);
3180
    int deblock_top  = (s->mb_y > 0);
3181

    
3182
    src_y  -=   linesize + 1;
3183
    src_cb -= uvlinesize + 1;
3184
    src_cr -= uvlinesize + 1;
3185

    
3186
#define XCHG(a,b,t,xchg)\
3187
t= a;\
3188
if(xchg)\
3189
    a= b;\
3190
b= t;
3191

    
3192
    if(deblock_left){
3193
        for(i = !deblock_top; i<17; i++){
3194
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3195
        }
3196
    }
3197

    
3198
    if(deblock_top){
3199
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3200
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3201
        if(s->mb_x+1 < s->mb_width){
3202
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3203
        }
3204
    }
3205

    
3206
    if(!(s->flags&CODEC_FLAG_GRAY)){
3207
        if(deblock_left){
3208
            for(i = !deblock_top; i<9; i++){
3209
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3210
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3211
            }
3212
        }
3213
        if(deblock_top){
3214
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3215
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3216
        }
3217
    }
3218
}
3219

    
3220
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3221
    MpegEncContext * const s = &h->s;
3222
    int i;
3223

    
3224
    src_y  -= 2 *   linesize;
3225
    src_cb -= 2 * uvlinesize;
3226
    src_cr -= 2 * uvlinesize;
3227

    
3228
    // There are two lines saved, the line above the the top macroblock of a pair,
3229
    // and the line above the bottom macroblock
3230
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3231
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3232
    for(i=2; i<34; i++){
3233
        h->left_border[i]= src_y[15+i*  linesize];
3234
    }
3235

    
3236
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3237
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3238
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3239
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3240

    
3241
    if(!(s->flags&CODEC_FLAG_GRAY)){
3242
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3243
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3244
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3245
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3246
        for(i=2; i<18; i++){
3247
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3248
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3249
        }
3250
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3251
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3252
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3253
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3254
    }
3255
}
3256

    
3257
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){
3258
    MpegEncContext * const s = &h->s;
3259
    int temp8, i;
3260
    uint64_t temp64;
3261
    int deblock_left = (s->mb_x > 0);
3262
    int deblock_top  = (s->mb_y > 0);
3263

    
3264
    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);
3265

    
3266
    src_y  -= 2 *   linesize + 1;
3267
    src_cb -= 2 * uvlinesize + 1;
3268
    src_cr -= 2 * uvlinesize + 1;
3269

    
3270
#define XCHG(a,b,t,xchg)\
3271
t= a;\
3272
if(xchg)\
3273
    a= b;\
3274
b= t;
3275

    
3276
    if(deblock_left){
3277
        for(i = (!deblock_top)<<1; i<34; i++){
3278
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3279
        }
3280
    }
3281

    
3282
    if(deblock_top){
3283
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3284
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3285
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3286
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3287
    }
3288

    
3289
    if(!(s->flags&CODEC_FLAG_GRAY)){
3290
        if(deblock_left){
3291
            for(i = (!deblock_top) << 1; i<18; i++){
3292
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3293
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3294
            }
3295
        }
3296
        if(deblock_top){
3297
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3298
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3299
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3300
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3301
        }
3302
    }
3303
}
3304

    
3305
static void hl_decode_mb(H264Context *h){
3306
    MpegEncContext * const s = &h->s;
3307
    const int mb_x= s->mb_x;
3308
    const int mb_y= s->mb_y;
3309
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3310
    const int mb_type= s->current_picture.mb_type[mb_xy];
3311
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3312
    int linesize, uvlinesize /*dct_offset*/;
3313
    int i;
3314
    int *block_offset = &h->block_offset[0];
3315
    const unsigned int bottom = mb_y & 1;
3316
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3317
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3318
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3319

    
3320
    if(!s->decode)
3321
        return;
3322

    
3323
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3324
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3325
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3326

    
3327
    if (h->mb_field_decoding_flag) {
3328
        linesize = s->linesize * 2;
3329
        uvlinesize = s->uvlinesize * 2;
3330
        block_offset = &h->block_offset[24];
3331
        if(mb_y&1){ //FIXME move out of this func?
3332
            dest_y -= s->linesize*15;
3333
            dest_cb-= s->uvlinesize*7;
3334
            dest_cr-= s->uvlinesize*7;
3335
        }
3336
    } else {
3337
        linesize = s->linesize;
3338
        uvlinesize = s->uvlinesize;
3339
//        dct_offset = s->linesize * 16;
3340
    }
3341

    
3342
    if(transform_bypass){
3343
        idct_dc_add =
3344
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3345
    }else if(IS_8x8DCT(mb_type)){
3346
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3347
        idct_add = s->dsp.h264_idct8_add;
3348
    }else{
3349
        idct_dc_add = s->dsp.h264_idct_dc_add;
3350
        idct_add = s->dsp.h264_idct_add;
3351
    }
3352

    
3353
    if (IS_INTRA_PCM(mb_type)) {
3354
        unsigned int x, y;
3355

    
3356
        // The pixels are stored in h->mb array in the same order as levels,
3357
        // copy them in output in the correct order.
3358
        for(i=0; i<16; i++) {
3359
            for (y=0; y<4; y++) {
3360
                for (x=0; x<4; x++) {
3361
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3362
                }
3363
            }
3364
        }
3365
        for(i=16; i<16+4; i++) {
3366
            for (y=0; y<4; y++) {
3367
                for (x=0; x<4; x++) {
3368
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3369
                }
3370
            }
3371
        }
3372
        for(i=20; i<20+4; i++) {
3373
            for (y=0; y<4; y++) {
3374
                for (x=0; x<4; x++) {
3375
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3376
                }
3377
            }
3378
        }
3379
    } else {
3380
        if(IS_INTRA(mb_type)){
3381
            if(h->deblocking_filter) {
3382
                if (h->mb_aff_frame) {
3383
                    if (!bottom)
3384
                        xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3385
                } else {
3386
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3387
                }
3388
            }
3389

    
3390
            if(!(s->flags&CODEC_FLAG_GRAY)){
3391
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3392
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3393
            }
3394

    
3395
            if(IS_INTRA4x4(mb_type)){
3396
                if(!s->encoding){
3397
                    if(IS_8x8DCT(mb_type)){
3398
                        for(i=0; i<16; i+=4){
3399
                            uint8_t * const ptr= dest_y + block_offset[i];
3400
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3401
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3402
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3403
                                                   (h->topright_samples_available<<(i+1))&0x8000, linesize);
3404
                            if(nnz){
3405
                                if(nnz == 1 && h->mb[i*16])
3406
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3407
                                else
3408
                                    idct_add(ptr, h->mb + i*16, linesize);
3409
                            }
3410
                        }
3411
                    }else
3412
                    for(i=0; i<16; i++){
3413
                        uint8_t * const ptr= dest_y + block_offset[i];
3414
                        uint8_t *topright;
3415
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3416
                        int nnz, tr;
3417

    
3418
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3419
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3420
                            assert(mb_y || linesize <= block_offset[i]);
3421
                            if(!topright_avail){
3422
                                tr= ptr[3 - linesize]*0x01010101;
3423
                                topright= (uint8_t*) &tr;
3424
                            }else
3425
                                topright= ptr + 4 - linesize;
3426
                        }else
3427
                            topright= NULL;
3428

    
3429
                        h->pred4x4[ dir ](ptr, topright, linesize);
3430
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3431
                        if(nnz){
3432
                            if(s->codec_id == CODEC_ID_H264){
3433
                                if(nnz == 1 && h->mb[i*16])
3434
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3435
                                else
3436
                                    idct_add(ptr, h->mb + i*16, linesize);
3437
                            }else
3438
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3439
                        }
3440
                    }
3441
                }
3442
            }else{
3443
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3444
                if(s->codec_id == CODEC_ID_H264){
3445
                    if(!transform_bypass)
3446
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3447
                }else
3448
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3449
            }
3450
            if(h->deblocking_filter) {
3451
                if (h->mb_aff_frame) {
3452
                    if (bottom) {
3453
                        uint8_t *pair_dest_y  = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize  ) + mb_x * 16;
3454
                        uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3455
                        uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3456
                        s->mb_y--;
3457
                        xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3458
                        s->mb_y++;
3459
                    }
3460
                } else {
3461
                    xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3462
                }
3463
            }
3464
        }else if(s->codec_id == CODEC_ID_H264){
3465
            hl_motion(h, dest_y, dest_cb, dest_cr,
3466
                      s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3467
                      s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3468
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3469
        }
3470

    
3471

    
3472
        if(!IS_INTRA4x4(mb_type)){
3473
            if(s->codec_id == CODEC_ID_H264){
3474
                if(IS_INTRA16x16(mb_type)){
3475
                    for(i=0; i<16; i++){
3476
                        if(h->non_zero_count_cache[ scan8[i] ])
3477
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3478
                        else if(h->mb[i*16])
3479
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3480
                    }
3481
                }else{
3482
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3483
                    for(i=0; i<16; i+=di){
3484
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3485
                        if(nnz){
3486
                            if(nnz==1 && h->mb[i*16])
3487
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3488
                            else
3489
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3490
                        }
3491
                    }
3492
                }
3493
            }else{
3494
                for(i=0; i<16; i++){
3495
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3496
                        uint8_t * const ptr= dest_y + block_offset[i];
3497
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3498
                    }
3499
                }
3500
            }
3501
        }
3502

    
3503
        if(!(s->flags&CODEC_FLAG_GRAY)){
3504
            uint8_t *dest[2] = {dest_cb, dest_cr};
3505
            if(transform_bypass){
3506
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3507
            }else{
3508
                idct_add = s->dsp.h264_idct_add;
3509
                idct_dc_add = s->dsp.h264_idct_dc_add;
3510
                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]);
3511
                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]);
3512
            }
3513
            if(s->codec_id == CODEC_ID_H264){
3514
                for(i=16; i<16+8; i++){
3515
                    if(h->non_zero_count_cache[ scan8[i] ])
3516
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3517
                    else if(h->mb[i*16])
3518
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3519
                }
3520
            }else{
3521
                for(i=16; i<16+8; i++){
3522
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3523
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3524
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3525
                    }
3526
                }
3527
            }
3528
        }
3529
    }
3530
    if(h->deblocking_filter) {
3531
        if (h->mb_aff_frame) {
3532
            const int mb_y = s->mb_y - 1;
3533
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3534
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3535
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3536
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3537
            uint8_t tmp = s->current_picture.data[1][384];
3538
            if (!bottom) return;
3539
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3540
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3541
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3542

    
3543
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3544
            // TODO deblock a pair
3545
            // top
3546
            s->mb_y--;
3547
            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);
3548
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3549
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3550
            if (tmp != s->current_picture.data[1][384]) {
3551
                tprintf("modified pixel 8,1 (1)\n");
3552
            }
3553
            // bottom
3554
            s->mb_y++;
3555
            tprintf("call mbaff filter_mb\n");
3556
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3557
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3558
            if (tmp != s->current_picture.data[1][384]) {
3559
                tprintf("modified pixel 8,1 (2)\n");
3560
            }
3561
        } else {
3562
            tprintf("call filter_mb\n");
3563
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3564
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3565
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3566
        }
3567
    }
3568
}
3569

    
3570
/**
3571
 * fills the default_ref_list.
3572
 */
3573
static int fill_default_ref_list(H264Context *h){
3574
    MpegEncContext * const s = &h->s;
3575
    int i;
3576
    int smallest_poc_greater_than_current = -1;
3577
    Picture sorted_short_ref[32];
3578

    
3579
    if(h->slice_type==B_TYPE){
3580
        int out_i;
3581
        int limit= INT_MIN;
3582

    
3583
        /* sort frame according to poc in B slice */
3584
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3585
            int best_i=INT_MIN;
3586
            int best_poc=INT_MAX;
3587

    
3588
            for(i=0; i<h->short_ref_count; i++){
3589
                const int poc= h->short_ref[i]->poc;
3590
                if(poc > limit && poc < best_poc){
3591
                    best_poc= poc;
3592
                    best_i= i;
3593
                }
3594
            }
3595

    
3596
            assert(best_i != INT_MIN);
3597

    
3598
            limit= best_poc;
3599
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3600
            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);
3601
            if (-1 == smallest_poc_greater_than_current) {
3602
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3603
                    smallest_poc_greater_than_current = out_i;
3604
                }
3605
            }
3606
        }
3607
    }
3608

    
3609
    if(s->picture_structure == PICT_FRAME){
3610
        if(h->slice_type==B_TYPE){
3611
            int list;
3612
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3613

    
3614
            // find the largest poc
3615
            for(list=0; list<2; list++){
3616
                int index = 0;
3617
                int j= -99;
3618
                int step= list ? -1 : 1;
3619

    
3620
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3621
                    while(j<0 || j>= h->short_ref_count){
3622
                        if(j != -99 && step == (list ? -1 : 1))
3623
                            return -1;
3624
                        step = -step;
3625
                        j= smallest_poc_greater_than_current + (step>>1);
3626
                    }
3627
                    if(sorted_short_ref[j].reference != 3) continue;
3628
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3629
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3630
                }
3631

    
3632
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3633
                    if(h->long_ref[i] == NULL) continue;
3634
                    if(h->long_ref[i]->reference != 3) continue;
3635

    
3636
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3637
                    h->default_ref_list[ list ][index++].pic_id= i;;
3638
                }
3639

    
3640
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3641
                    // swap the two first elements of L1 when
3642
                    // L0 and L1 are identical
3643
                    Picture temp= h->default_ref_list[1][0];
3644
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3645
                    h->default_ref_list[1][1] = temp;
3646
                }
3647

    
3648
                if(index < h->ref_count[ list ])
3649
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3650
            }
3651
        }else{
3652
            int index=0;
3653
            for(i=0; i<h->short_ref_count; i++){
3654
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3655
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3656
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3657
            }
3658
            for(i = 0; i < 16; i++){
3659
                if(h->long_ref[i] == NULL) continue;
3660
                if(h->long_ref[i]->reference != 3) continue;
3661
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3662
                h->default_ref_list[0][index++].pic_id= i;;
3663
            }
3664
            if(index < h->ref_count[0])
3665
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3666
        }
3667
    }else{ //FIELD
3668
        if(h->slice_type==B_TYPE){
3669
        }else{
3670
            //FIXME second field balh
3671
        }
3672
    }
3673
#ifdef TRACE
3674
    for (i=0; i<h->ref_count[0]; i++) {
3675
        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]);
3676
    }
3677
    if(h->slice_type==B_TYPE){
3678
        for (i=0; i<h->ref_count[1]; i++) {
3679
            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]);
3680
        }
3681
    }
3682
#endif
3683
    return 0;
3684
}
3685

    
3686
static void print_short_term(H264Context *h);
3687
static void print_long_term(H264Context *h);
3688

    
3689
static int decode_ref_pic_list_reordering(H264Context *h){
3690
    MpegEncContext * const s = &h->s;
3691
    int list, index;
3692

    
3693
    print_short_term(h);
3694
    print_long_term(h);
3695
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3696

    
3697
    for(list=0; list<2; list++){
3698
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3699

    
3700
        if(get_bits1(&s->gb)){
3701
            int pred= h->curr_pic_num;
3702

    
3703
            for(index=0; ; index++){
3704
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3705
                int pic_id;
3706
                int i;
3707
                Picture *ref = NULL;
3708

    
3709
                if(reordering_of_pic_nums_idc==3)
3710
                    break;
3711

    
3712
                if(index >= h->ref_count[list]){
3713
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3714
                    return -1;
3715
                }
3716

    
3717
                if(reordering_of_pic_nums_idc<3){
3718
                    if(reordering_of_pic_nums_idc<2){
3719
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3720

    
3721
                        if(abs_diff_pic_num >= h->max_pic_num){
3722
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3723
                            return -1;
3724
                        }
3725

    
3726
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3727
                        else                                pred+= abs_diff_pic_num;
3728
                        pred &= h->max_pic_num - 1;
3729

    
3730
                        for(i= h->short_ref_count-1; i>=0; i--){
3731
                            ref = h->short_ref[i];
3732
                            assert(ref->reference == 3);
3733
                            assert(!ref->long_ref);
3734
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3735
                                break;
3736
                        }
3737
                        if(i>=0)
3738
                            ref->pic_id= ref->frame_num;
3739
                    }else{
3740
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3741
                        ref = h->long_ref[pic_id];
3742
                        ref->pic_id= pic_id;
3743
                        assert(ref->reference == 3);
3744
                        assert(ref->long_ref);
3745
                        i=0;
3746
                    }
3747

    
3748
                    if (i < 0) {
3749
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3750
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3751
                    } else {
3752
                        for(i=index; i+1<h->ref_count[list]; i++){
3753
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3754
                                break;
3755
                        }
3756
                        for(; i > index; i--){
3757
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3758
                        }
3759
                        h->ref_list[list][index]= *ref;
3760
                    }
3761
                }else{
3762
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3763
                    return -1;
3764
                }
3765
            }
3766
        }
3767

    
3768
        if(h->slice_type!=B_TYPE) break;
3769
    }
3770
    for(list=0; list<2; list++){
3771
        for(index= 0; index < h->ref_count[list]; index++){
3772
            if(!h->ref_list[list][index].data[0])
3773
                h->ref_list[list][index]= s->current_picture;
3774
        }
3775
        if(h->slice_type!=B_TYPE) break;
3776
    }
3777

    
3778
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3779
        direct_dist_scale_factor(h);
3780
    direct_ref_list_init(h);
3781
    return 0;
3782
}
3783

    
3784
static int pred_weight_table(H264Context *h){
3785
    MpegEncContext * const s = &h->s;
3786
    int list, i;
3787
    int luma_def, chroma_def;
3788

    
3789
    h->use_weight= 0;
3790
    h->use_weight_chroma= 0;
3791
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3792
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3793
    luma_def = 1<<h->luma_log2_weight_denom;
3794
    chroma_def = 1<<h->chroma_log2_weight_denom;
3795

    
3796
    for(list=0; list<2; list++){
3797
        for(i=0; i<h->ref_count[list]; i++){
3798
            int luma_weight_flag, chroma_weight_flag;
3799

    
3800
            luma_weight_flag= get_bits1(&s->gb);
3801
            if(luma_weight_flag){
3802
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3803
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3804
                if(   h->luma_weight[list][i] != luma_def
3805
                   || h->luma_offset[list][i] != 0)
3806
                    h->use_weight= 1;
3807
            }else{
3808
                h->luma_weight[list][i]= luma_def;
3809
                h->luma_offset[list][i]= 0;
3810
            }
3811

    
3812
            chroma_weight_flag= get_bits1(&s->gb);
3813
            if(chroma_weight_flag){
3814
                int j;
3815
                for(j=0; j<2; j++){
3816
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3817
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3818
                    if(   h->chroma_weight[list][i][j] != chroma_def
3819
                       || h->chroma_offset[list][i][j] != 0)
3820
                        h->use_weight_chroma= 1;
3821
                }
3822
            }else{
3823
                int j;
3824
                for(j=0; j<2; j++){
3825
                    h->chroma_weight[list][i][j]= chroma_def;
3826
                    h->chroma_offset[list][i][j]= 0;
3827
                }
3828
            }
3829
        }
3830
        if(h->slice_type != B_TYPE) break;
3831
    }
3832
    h->use_weight= h->use_weight || h->use_weight_chroma;
3833
    return 0;
3834
}
3835

    
3836
static void implicit_weight_table(H264Context *h){
3837
    MpegEncContext * const s = &h->s;
3838
    int ref0, ref1;
3839
    int cur_poc = s->current_picture_ptr->poc;
3840

    
3841
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3842
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3843
        h->use_weight= 0;
3844
        h->use_weight_chroma= 0;
3845
        return;
3846
    }
3847

    
3848
    h->use_weight= 2;
3849
    h->use_weight_chroma= 2;
3850
    h->luma_log2_weight_denom= 5;
3851
    h->chroma_log2_weight_denom= 5;
3852

    
3853
    /* FIXME: MBAFF */
3854
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3855
        int poc0 = h->ref_list[0][ref0].poc;
3856
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3857
            int poc1 = h->ref_list[1][ref1].poc;
3858
            int td = clip(poc1 - poc0, -128, 127);
3859
            if(td){
3860
                int tb = clip(cur_poc - poc0, -128, 127);
3861
                int tx = (16384 + (ABS(td) >> 1)) / td;
3862
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3863
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3864
                    h->implicit_weight[ref0][ref1] = 32;
3865
                else
3866
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3867
            }else
3868
                h->implicit_weight[ref0][ref1] = 32;
3869
        }
3870
    }
3871
}
3872

    
3873
static inline void unreference_pic(H264Context *h, Picture *pic){
3874
    int i;
3875
    pic->reference=0;
3876
    if(pic == h->delayed_output_pic)
3877
        pic->reference=1;
3878
    else{
3879
        for(i = 0; h->delayed_pic[i]; i++)
3880
            if(pic == h->delayed_pic[i]){
3881
                pic->reference=1;
3882
                break;
3883
            }
3884
    }
3885
}
3886

    
3887
/**
3888
 * instantaneous decoder refresh.
3889
 */
3890
static void idr(H264Context *h){
3891
    int i;
3892

    
3893
    for(i=0; i<16; i++){
3894
        if (h->long_ref[i] != NULL) {
3895
            unreference_pic(h, h->long_ref[i]);
3896
            h->long_ref[i]= NULL;
3897
        }
3898
    }
3899
    h->long_ref_count=0;
3900

    
3901
    for(i=0; i<h->short_ref_count; i++){
3902
        unreference_pic(h, h->short_ref[i]);
3903
        h->short_ref[i]= NULL;
3904
    }
3905
    h->short_ref_count=0;
3906
}
3907

    
3908
/* forget old pics after a seek */
3909
static void flush_dpb(AVCodecContext *avctx){
3910
    H264Context *h= avctx->priv_data;
3911
    int i;
3912
    for(i=0; i<16; i++) {
3913
        if(h->delayed_pic[i])
3914
            h->delayed_pic[i]->reference= 0;
3915
        h->delayed_pic[i]= NULL;
3916
    }
3917
    if(h->delayed_output_pic)
3918
        h->delayed_output_pic->reference= 0;
3919
    h->delayed_output_pic= NULL;
3920
    idr(h);
3921
    if(h->s.current_picture_ptr)
3922
        h->s.current_picture_ptr->reference= 0;
3923
}
3924

    
3925
/**
3926
 *
3927
 * @return the removed picture or NULL if an error occurs
3928
 */
3929
static Picture * remove_short(H264Context *h, int frame_num){
3930
    MpegEncContext * const s = &h->s;
3931
    int i;
3932

    
3933
    if(s->avctx->debug&FF_DEBUG_MMCO)
3934
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3935

    
3936
    for(i=0; i<h->short_ref_count; i++){
3937
        Picture *pic= h->short_ref[i];
3938
        if(s->avctx->debug&FF_DEBUG_MMCO)
3939
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3940
        if(pic->frame_num == frame_num){
3941
            h->short_ref[i]= NULL;
3942
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3943
            h->short_ref_count--;
3944
            return pic;
3945
        }
3946
    }
3947
    return NULL;
3948
}
3949

    
3950
/**
3951
 *
3952
 * @return the removed picture or NULL if an error occurs
3953
 */
3954
static Picture * remove_long(H264Context *h, int i){
3955
    Picture *pic;
3956

    
3957
    pic= h->long_ref[i];
3958
    h->long_ref[i]= NULL;
3959
    if(pic) h->long_ref_count--;
3960

    
3961
    return pic;
3962
}
3963

    
3964
/**
3965
 * print short term list
3966
 */
3967
static void print_short_term(H264Context *h) {
3968
    uint32_t i;
3969
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3970
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3971
        for(i=0; i<h->short_ref_count; i++){
3972
            Picture *pic= h->short_ref[i];
3973
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3974
        }
3975
    }
3976
}
3977

    
3978
/**
3979
 * print long term list
3980
 */
3981
static void print_long_term(H264Context *h) {
3982
    uint32_t i;
3983
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3984
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3985
        for(i = 0; i < 16; i++){
3986
            Picture *pic= h->long_ref[i];
3987
            if (pic) {
3988
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3989
            }
3990
        }
3991
    }
3992
}
3993

    
3994
/**
3995
 * Executes the reference picture marking (memory management control operations).
3996
 */
3997
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3998
    MpegEncContext * const s = &h->s;
3999
    int i, j;
4000
    int current_is_long=0;
4001
    Picture *pic;
4002

    
4003
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
4004
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
4005

    
4006
    for(i=0; i<mmco_count; i++){
4007
        if(s->avctx->debug&FF_DEBUG_MMCO)
4008
            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);
4009

    
4010
        switch(mmco[i].opcode){
4011
        case MMCO_SHORT2UNUSED:
4012
            pic= remove_short(h, mmco[i].short_frame_num);
4013
            if(pic)
4014
                unreference_pic(h, pic);
4015
            else if(s->avctx->debug&FF_DEBUG_MMCO)
4016
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
4017
            break;
4018
        case MMCO_SHORT2LONG:
4019
            pic= remove_long(h, mmco[i].long_index);
4020
            if(pic) unreference_pic(h, pic);
4021

    
4022
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
4023
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
4024
            h->long_ref_count++;
4025
            break;
4026
        case MMCO_LONG2UNUSED:
4027
            pic= remove_long(h, mmco[i].long_index);
4028
            if(pic)
4029
                unreference_pic(h, pic);
4030
            else if(s->avctx->debug&FF_DEBUG_MMCO)
4031
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
4032
            break;
4033
        case MMCO_LONG:
4034
            pic= remove_long(h, mmco[i].long_index);
4035
            if(pic) unreference_pic(h, pic);
4036

    
4037
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
4038
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
4039
            h->long_ref_count++;
4040

    
4041
            current_is_long=1;
4042
            break;
4043
        case MMCO_SET_MAX_LONG:
4044
            assert(mmco[i].long_index <= 16);
4045
            // just remove the long term which index is greater than new max
4046
            for(j = mmco[i].long_index; j<16; j++){
4047
                pic = remove_long(h, j);
4048
                if (pic) unreference_pic(h, pic);
4049
            }
4050
            break;
4051
        case MMCO_RESET:
4052
            while(h->short_ref_count){
4053
                pic= remove_short(h, h->short_ref[0]->frame_num);
4054
                unreference_pic(h, pic);
4055
            }
4056
            for(j = 0; j < 16; j++) {
4057
                pic= remove_long(h, j);
4058
                if(pic) unreference_pic(h, pic);
4059
            }
4060
            break;
4061
        default: assert(0);
4062
        }
4063
    }
4064

    
4065
    if(!current_is_long){
4066
        pic= remove_short(h, s->current_picture_ptr->frame_num);
4067
        if(pic){
4068
            unreference_pic(h, pic);
4069
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
4070
        }
4071

    
4072
        if(h->short_ref_count)
4073
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
4074

    
4075
        h->short_ref[0]= s->current_picture_ptr;
4076
        h->short_ref[0]->long_ref=0;
4077
        h->short_ref_count++;
4078
    }
4079

    
4080
    print_short_term(h);
4081
    print_long_term(h);
4082
    return 0;
4083
}
4084

    
4085
static int decode_ref_pic_marking(H264Context *h){
4086
    MpegEncContext * const s = &h->s;
4087
    int i;
4088

    
4089
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
4090
        s->broken_link= get_bits1(&s->gb) -1;
4091
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
4092
        if(h->mmco[0].long_index == -1)
4093
            h->mmco_index= 0;
4094
        else{
4095
            h->mmco[0].opcode= MMCO_LONG;
4096
            h->mmco_index= 1;
4097
        }
4098
    }else{
4099
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
4100
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
4101
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
4102

    
4103
                h->mmco[i].opcode= opcode;
4104
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
4105
                    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
4106
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
4107
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
4108
                        return -1;
4109
                    }*/
4110
                }
4111
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
4112
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
4113
                    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){
4114
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
4115
                        return -1;
4116
                    }
4117
                }
4118

    
4119
                if(opcode > MMCO_LONG){
4120
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4121
                    return -1;
4122
                }
4123
                if(opcode == MMCO_END)
4124
                    break;
4125
            }
4126
            h->mmco_index= i;
4127
        }else{
4128
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4129

    
4130
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4131
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4132
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4133
                h->mmco_index= 1;
4134
            }else
4135
                h->mmco_index= 0;
4136
        }
4137
    }
4138

    
4139
    return 0;
4140
}
4141

    
4142
static int init_poc(H264Context *h){
4143
    MpegEncContext * const s = &h->s;
4144
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4145
    int field_poc[2];
4146

    
4147
    if(h->nal_unit_type == NAL_IDR_SLICE){
4148
        h->frame_num_offset= 0;
4149
    }else{
4150
        if(h->frame_num < h->prev_frame_num)
4151
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4152
        else
4153
            h->frame_num_offset= h->prev_frame_num_offset;
4154
    }
4155

    
4156
    if(h->sps.poc_type==0){
4157
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4158

    
4159
        if(h->nal_unit_type == NAL_IDR_SLICE){
4160
             h->prev_poc_msb=
4161
             h->prev_poc_lsb= 0;
4162
        }
4163

    
4164
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4165
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4166
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4167
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4168
        else
4169
            h->poc_msb = h->prev_poc_msb;
4170
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4171
        field_poc[0] =
4172
        field_poc[1] = h->poc_msb + h->poc_lsb;
4173
        if(s->picture_structure == PICT_FRAME)
4174
            field_poc[1] += h->delta_poc_bottom;
4175
    }else if(h->sps.poc_type==1){
4176
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4177
        int i;
4178

    
4179
        if(h->sps.poc_cycle_length != 0)
4180
            abs_frame_num = h->frame_num_offset + h->frame_num;
4181
        else
4182
            abs_frame_num = 0;
4183

    
4184
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4185
            abs_frame_num--;
4186

    
4187
        expected_delta_per_poc_cycle = 0;
4188
        for(i=0; i < h->sps.poc_cycle_length; i++)
4189
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4190

    
4191
        if(abs_frame_num > 0){
4192
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4193
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4194

    
4195
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4196
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4197
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4198
        } else
4199
            expectedpoc = 0;
4200

    
4201
        if(h->nal_ref_idc == 0)
4202
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4203

    
4204
        field_poc[0] = expectedpoc + h->delta_poc[0];
4205
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4206

    
4207
        if(s->picture_structure == PICT_FRAME)
4208
            field_poc[1] += h->delta_poc[1];
4209
    }else{
4210
        int poc;
4211
        if(h->nal_unit_type == NAL_IDR_SLICE){
4212
            poc= 0;
4213
        }else{
4214
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4215
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4216
        }
4217
        field_poc[0]= poc;
4218
        field_poc[1]= poc;
4219
    }
4220

    
4221
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4222
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4223
    if(s->picture_structure != PICT_TOP_FIELD)
4224
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4225
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4226
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4227

    
4228
    return 0;
4229
}
4230

    
4231
/**
4232
 * decodes a slice header.
4233
 * this will allso call MPV_common_init() and frame_start() as needed
4234
 */
4235
static int decode_slice_header(H264Context *h){
4236
    MpegEncContext * const s = &h->s;
4237
    int first_mb_in_slice, pps_id;
4238
    int num_ref_idx_active_override_flag;
4239
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4240
    int slice_type;
4241
    int default_ref_list_done = 0;
4242

    
4243
    s->current_picture.reference= h->nal_ref_idc != 0;
4244
    s->dropable= h->nal_ref_idc == 0;
4245

    
4246
    first_mb_in_slice= get_ue_golomb(&s->gb);
4247

    
4248
    slice_type= get_ue_golomb(&s->gb);
4249
    if(slice_type > 9){
4250
        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);
4251
        return -1;
4252
    }
4253
    if(slice_type > 4){
4254
        slice_type -= 5;
4255
        h->slice_type_fixed=1;
4256
    }else
4257
        h->slice_type_fixed=0;
4258

    
4259
    slice_type= slice_type_map[ slice_type ];
4260
    if (slice_type == I_TYPE
4261
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4262
        default_ref_list_done = 1;
4263
    }
4264
    h->slice_type= slice_type;
4265

    
4266
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4267

    
4268
    pps_id= get_ue_golomb(&s->gb);
4269
    if(pps_id>255){
4270
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4271
        return -1;
4272
    }
4273
    h->pps= h->pps_buffer[pps_id];
4274
    if(h->pps.slice_group_count == 0){
4275
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4276
        return -1;
4277
    }
4278

    
4279
    h->sps= h->sps_buffer[ h->pps.sps_id ];
4280
    if(h->sps.log2_max_frame_num == 0){
4281
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4282
        return -1;
4283
    }
4284

    
4285
    if(h->dequant_coeff_pps != pps_id){
4286
        h->dequant_coeff_pps = pps_id;
4287
        init_dequant_tables(h);
4288
    }
4289

    
4290
    s->mb_width= h->sps.mb_width;
4291
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
4292

    
4293
    h->b_stride=  s->mb_width*4 + 1;
4294
    h->b8_stride= s->mb_width*2 + 1;
4295

    
4296
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
4297
    if(h->sps.frame_mbs_only_flag)
4298
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
4299
    else
4300
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
4301

    
4302
    if (s->context_initialized
4303
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
4304
        free_tables(h);
4305
        MPV_common_end(s);
4306
    }
4307
    if (!s->context_initialized) {
4308
        if (MPV_common_init(s) < 0)
4309
            return -1;
4310

    
4311
        if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
4312
            memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
4313
            memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
4314
        }else{
4315
            int i;
4316
            for(i=0; i<16; i++){
4317
#define T(x) (x>>2) | ((x<<2) & 0xF)
4318
                h->zigzag_scan[i] = T(zigzag_scan[i]);
4319
                h-> field_scan[i] = T( field_scan[i]);
4320
            }