Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ a4dae92b

History | View | Annotate | Download (226 KB)

1 0da71265 Michael Niedermayer
/*
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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  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 e5017ab8 Laurent Aimar
#include "cabac.h"
35
36 0da71265 Michael Niedermayer
#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 log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
65
    int poc_type;                      ///< pic_order_cnt_type
66
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
67
    int delta_pic_order_always_zero_flag;
68
    int offset_for_non_ref_pic;
69
    int offset_for_top_to_bottom_field;
70
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
71
    int ref_frame_count;               ///< num_ref_frames
72 a15e68de Michael Niedermayer
    int gaps_in_frame_num_allowed_flag;
73 0da71265 Michael Niedermayer
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
74
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
75
    int frame_mbs_only_flag;
76
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
77
    int direct_8x8_inference_flag;
78 a15e68de Michael Niedermayer
    int crop;                   ///< frame_cropping_flag
79
    int crop_left;              ///< frame_cropping_rect_left_offset
80
    int crop_right;             ///< frame_cropping_rect_right_offset
81
    int crop_top;               ///< frame_cropping_rect_top_offset
82
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
83 0da71265 Michael Niedermayer
    int vui_parameters_present_flag;
84 5ff85f1d Michael Niedermayer
    AVRational sar;
85 ec587a50 Måns Rullgård
    int timing_info_present_flag;
86
    uint32_t num_units_in_tick;
87
    uint32_t time_scale;
88
    int fixed_frame_rate_flag;
89 0da71265 Michael Niedermayer
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
90
}SPS;
91
92
/**
93
 * Picture parameter set
94
 */
95
typedef struct PPS{
96
    int sps_id;
97
    int cabac;                  ///< entropy_coding_mode_flag
98
    int pic_order_present;      ///< pic_order_present_flag
99
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
100
    int mb_slice_group_map_type;
101
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
102
    int weighted_pred;          ///< weighted_pred_flag
103
    int weighted_bipred_idc;
104
    int init_qp;                ///< pic_init_qp_minus26 + 26
105
    int init_qs;                ///< pic_init_qs_minus26 + 26
106
    int chroma_qp_index_offset;
107
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108
    int constrained_intra_pred; ///< constrained_intra_pred_flag
109
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
110
}PPS;
111
112
/**
113
 * Memory management control operation opcode.
114
 */
115
typedef enum MMCOOpcode{
116
    MMCO_END=0,
117
    MMCO_SHORT2UNUSED,
118
    MMCO_LONG2UNUSED,
119
    MMCO_SHORT2LONG,
120
    MMCO_SET_MAX_LONG,
121
    MMCO_RESET, 
122
    MMCO_LONG,
123
} MMCOOpcode;
124
125
/**
126
 * Memory management control operation.
127
 */
128
typedef struct MMCO{
129
    MMCOOpcode opcode;
130
    int short_frame_num;
131
    int long_index;
132
} MMCO;
133
134
/**
135
 * H264Context
136
 */
137
typedef struct H264Context{
138
    MpegEncContext s;
139
    int nal_ref_idc;        
140
    int nal_unit_type;
141
#define NAL_SLICE                1
142
#define NAL_DPA                        2
143
#define NAL_DPB                        3
144
#define NAL_DPC                        4
145
#define NAL_IDR_SLICE                5
146
#define NAL_SEI                        6
147
#define NAL_SPS                        7
148
#define NAL_PPS                        8
149
#define NAL_PICTURE_DELIMITER        9
150
#define NAL_FILTER_DATA                10
151
    uint8_t *rbsp_buffer;
152
    int rbsp_buffer_size;
153
154 4770b1b4 Roberto Togni
    /**
155
      * Used to parse AVC variant of h264
156
      */
157
    int is_avc; ///< this flag is != 0 if codec is avc1
158
    int got_avcC; ///< flag used to parse avcC data only once
159
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
160
161 0da71265 Michael Niedermayer
    int chroma_qp; //QPc
162
163
    int prev_mb_skiped; //FIXME remove (IMHO not used)
164
165
    //prediction stuff
166
    int chroma_pred_mode;
167
    int intra16x16_pred_mode;
168
    
169
    int8_t intra4x4_pred_mode_cache[5*8];
170
    int8_t (*intra4x4_pred_mode)[8];
171
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
173
    void (*pred16x16[4+3])(uint8_t *src, int stride);
174
    unsigned int topleft_samples_available;
175
    unsigned int top_samples_available;
176
    unsigned int topright_samples_available;
177
    unsigned int left_samples_available;
178 53c05b1e Michael Niedermayer
    uint8_t (*top_border)[16+2*8];
179
    uint8_t left_border[17+2*9];
180 0da71265 Michael Niedermayer
181
    /**
182
     * non zero coeff count cache.
183
     * is 64 if not available.
184
     */
185
    uint8_t non_zero_count_cache[6*8];
186 53c05b1e Michael Niedermayer
    uint8_t (*non_zero_count)[16];
187 0da71265 Michael Niedermayer
188
    /**
189
     * Motion vector cache.
190
     */
191
    int16_t mv_cache[2][5*8][2];
192
    int8_t ref_cache[2][5*8];
193
#define LIST_NOT_USED -1 //FIXME rename?
194
#define PART_NOT_AVAILABLE -2
195
    
196
    /**
197
     * is 1 if the specific list MV&references are set to 0,0,-2.
198
     */
199
    int mv_cache_clean[2];
200
201
    int block_offset[16+8];
202
    int chroma_subblock_offset[16]; //FIXME remove
203
    
204
    uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
205
    uint16_t *mb2b8_xy;
206 650cec0c Loren Merritt
    int b_stride; //FIXME use s->b4_stride
207 0da71265 Michael Niedermayer
    int b8_stride;
208
209 8b82a956 Michael Niedermayer
    int halfpel_flag;
210
    int thirdpel_flag;
211
212 da3b9756 Mike Melanson
    int unknown_svq3_flag;
213
    int next_slice_index;
214
215 0da71265 Michael Niedermayer
    SPS sps_buffer[MAX_SPS_COUNT];
216
    SPS sps; ///< current sps
217
    
218
    PPS pps_buffer[MAX_PPS_COUNT];
219
    /**
220
     * current pps
221
     */
222
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
223
224
    int slice_num;
225
    uint8_t *slice_table_base;
226
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
227
    int slice_type;
228
    int slice_type_fixed;
229
    
230
    //interlacing specific flags
231
    int mb_field_decoding_flag;
232
    
233
    int sub_mb_type[4];
234
    
235
    //POC stuff
236
    int poc_lsb;
237
    int poc_msb;
238
    int delta_poc_bottom;
239
    int delta_poc[2];
240
    int frame_num;
241
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
242
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
243
    int frame_num_offset;         ///< for POC type 2
244
    int prev_frame_num_offset;    ///< for POC type 2
245
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
246
247
    /**
248
     * frame_num for frames or 2*frame_num for field pics.
249
     */
250
    int curr_pic_num;
251
    
252
    /**
253
     * max_frame_num or 2*max_frame_num for field pics.
254
     */
255
    int max_pic_num;
256
257
    //Weighted pred stuff
258
    int luma_log2_weight_denom;
259
    int chroma_log2_weight_denom;
260
    int luma_weight[2][16];
261
    int luma_offset[2][16];
262
    int chroma_weight[2][16][2];
263
    int chroma_offset[2][16][2];
264
   
265
    //deblock
266 53c05b1e Michael Niedermayer
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
267 980a82b7 Michael Niedermayer
    int slice_alpha_c0_offset;
268
    int slice_beta_offset;
269 0da71265 Michael Niedermayer
     
270
    int redundant_pic_count;
271
    
272
    int direct_spatial_mv_pred;
273 5ad984c9 Loren Merritt
    int dist_scale_factor[16];
274 0da71265 Michael Niedermayer
275
    /**
276
     * num_ref_idx_l0/1_active_minus1 + 1
277
     */
278
    int ref_count[2];// FIXME split for AFF
279
    Picture *short_ref[16];
280
    Picture *long_ref[16];
281
    Picture default_ref_list[2][32];
282
    Picture ref_list[2][32]; //FIXME size?
283
    Picture field_ref_list[2][32]; //FIXME size?
284 a4dae92b Loren Merritt
    Picture *delayed_pic[16]; //FIXME size?
285 0da71265 Michael Niedermayer
    
286
    /**
287
     * memory management control operations buffer.
288
     */
289
    MMCO mmco[MAX_MMCO_COUNT];
290
    int mmco_index;
291
    
292
    int long_ref_count;  ///< number of actual long term references
293
    int short_ref_count; ///< number of actual short term references
294
    
295
    //data partitioning
296
    GetBitContext intra_gb;
297
    GetBitContext inter_gb;
298
    GetBitContext *intra_gb_ptr;
299
    GetBitContext *inter_gb_ptr;
300
    
301
    DCTELEM mb[16*24] __align8;
302 e5017ab8 Laurent Aimar
303
    /**
304
     * Cabac
305
     */
306
    CABACContext cabac;
307
    uint8_t      cabac_state[399];
308
    int          cabac_init_idc;
309
310
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
311
    uint16_t     *cbp_table;
312 3981c385 Michael Niedermayer
    int top_cbp;
313
    int left_cbp;
314 693399ba Laurent Aimar
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
315 e5017ab8 Laurent Aimar
    uint8_t     *chroma_pred_mode_table;
316
    int         last_qscale_diff;
317 9e528114 Laurent Aimar
    int16_t     (*mvd_table[2])[2];
318
    int16_t     mvd_cache[2][5*8][2];
319 5ad984c9 Loren Merritt
    uint8_t     *direct_table;
320
    uint8_t     direct_cache[5*8];
321 e5017ab8 Laurent Aimar
322 0da71265 Michael Niedermayer
}H264Context;
323
324
static VLC coeff_token_vlc[4];
325
static VLC chroma_dc_coeff_token_vlc;
326
327
static VLC total_zeros_vlc[15];
328
static VLC chroma_dc_total_zeros_vlc[3];
329
330
static VLC run_vlc[6];
331
static VLC run7_vlc;
332
333 8b82a956 Michael Niedermayer
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
334
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
335 53c05b1e Michael Niedermayer
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
336 8b82a956 Michael Niedermayer
337 377ec888 Michael Niedermayer
static inline uint32_t pack16to32(int a, int b){
338
#ifdef WORDS_BIGENDIAN
339
   return (b&0xFFFF) + (a<<16);
340
#else
341
   return (a&0xFFFF) + (b<<16);
342
#endif
343
}
344
345 0da71265 Michael Niedermayer
/**
346
 * fill a rectangle.
347 5175b937 Loic Le Loarer
 * @param h height of the rectangle, should be a constant
348
 * @param w width of the rectangle, should be a constant
349 0da71265 Michael Niedermayer
 * @param size the size of val (1 or 4), should be a constant
350
 */
351 af6e2fed Michael Niedermayer
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
352
    uint8_t *p= (uint8_t*)vp;
353 0da71265 Michael Niedermayer
    assert(size==1 || size==4);
354
    
355
    w      *= size;
356
    stride *= size;
357
    
358
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
359
    if(w==2 && h==2){
360
        *(uint16_t*)(p + 0)=
361
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
362
    }else if(w==2 && h==4){
363
        *(uint16_t*)(p + 0*stride)=
364
        *(uint16_t*)(p + 1*stride)=
365
        *(uint16_t*)(p + 2*stride)=
366
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
367 f7a8c179 Michael Niedermayer
    }else if(w==4 && h==1){
368
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
369 0da71265 Michael Niedermayer
    }else if(w==4 && h==2){
370
        *(uint32_t*)(p + 0*stride)=
371
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
372
    }else if(w==4 && h==4){
373
        *(uint32_t*)(p + 0*stride)=
374
        *(uint32_t*)(p + 1*stride)=
375
        *(uint32_t*)(p + 2*stride)=
376
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
377
    }else if(w==8 && h==1){
378
        *(uint32_t*)(p + 0)=
379
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
380
    }else if(w==8 && h==2){
381
        *(uint32_t*)(p + 0 + 0*stride)=
382
        *(uint32_t*)(p + 4 + 0*stride)=
383
        *(uint32_t*)(p + 0 + 1*stride)=
384
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
385
    }else if(w==8 && h==4){
386
        *(uint64_t*)(p + 0*stride)=
387
        *(uint64_t*)(p + 1*stride)=
388
        *(uint64_t*)(p + 2*stride)=
389
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
390
    }else if(w==16 && h==2){
391
        *(uint64_t*)(p + 0+0*stride)=
392
        *(uint64_t*)(p + 8+0*stride)=
393
        *(uint64_t*)(p + 0+1*stride)=
394
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
395
    }else if(w==16 && h==4){
396
        *(uint64_t*)(p + 0+0*stride)=
397
        *(uint64_t*)(p + 8+0*stride)=
398
        *(uint64_t*)(p + 0+1*stride)=
399
        *(uint64_t*)(p + 8+1*stride)=
400
        *(uint64_t*)(p + 0+2*stride)=
401
        *(uint64_t*)(p + 8+2*stride)=
402
        *(uint64_t*)(p + 0+3*stride)=
403
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
404
    }else
405
        assert(0);
406
}
407
408
static inline void fill_caches(H264Context *h, int mb_type){
409
    MpegEncContext * const s = &h->s;
410 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
411 0da71265 Michael Niedermayer
    int topleft_xy, top_xy, topright_xy, left_xy[2];
412
    int topleft_type, top_type, topright_type, left_type[2];
413
    int left_block[4];
414
    int i;
415
416
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
417
    
418
    if(h->sps.mb_aff){
419
    //FIXME
420 1df1df0b Fabrice Bellard
        topleft_xy = 0; /* avoid warning */
421
        top_xy = 0; /* avoid warning */
422
        topright_xy = 0; /* avoid warning */
423 0da71265 Michael Niedermayer
    }else{
424 7bc9090a Michael Niedermayer
        topleft_xy = mb_xy-1 - s->mb_stride;
425
        top_xy     = mb_xy   - s->mb_stride;
426
        topright_xy= mb_xy+1 - s->mb_stride;
427 0da71265 Michael Niedermayer
        left_xy[0]   = mb_xy-1;
428
        left_xy[1]   = mb_xy-1;
429
        left_block[0]= 0;
430
        left_block[1]= 1;
431
        left_block[2]= 2;
432
        left_block[3]= 3;
433
    }
434
435
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
436
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
437
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
438
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
439
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
440
441
    if(IS_INTRA(mb_type)){
442
        h->topleft_samples_available= 
443
        h->top_samples_available= 
444
        h->left_samples_available= 0xFFFF;
445
        h->topright_samples_available= 0xEEEA;
446
447
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
448
            h->topleft_samples_available= 0xB3FF;
449
            h->top_samples_available= 0x33FF;
450
            h->topright_samples_available= 0x26EA;
451
        }
452
        for(i=0; i<2; i++){
453
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
454
                h->topleft_samples_available&= 0xDF5F;
455
                h->left_samples_available&= 0x5F5F;
456
            }
457
        }
458
        
459
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
460
            h->topleft_samples_available&= 0x7FFF;
461
        
462
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
463
            h->topright_samples_available&= 0xFBFF;
464
    
465
        if(IS_INTRA4x4(mb_type)){
466
            if(IS_INTRA4x4(top_type)){
467
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
468
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
469
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
470
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
471
            }else{
472
                int pred;
473
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
474
                    pred= 2;
475
                else{
476
                    pred= -1;
477
                }
478
                h->intra4x4_pred_mode_cache[4+8*0]=
479
                h->intra4x4_pred_mode_cache[5+8*0]=
480
                h->intra4x4_pred_mode_cache[6+8*0]=
481
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
482
            }
483
            for(i=0; i<2; i++){
484
                if(IS_INTRA4x4(left_type[i])){
485
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
486
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
487
                }else{
488
                    int pred;
489
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
490
                        pred= 2;
491
                    else{
492
                        pred= -1;
493
                    }
494
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
495
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
496
                }
497
            }
498
        }
499
    }
500
    
501
    
502
/*
503
0 . T T. T T T T 
504
1 L . .L . . . . 
505
2 L . .L . . . . 
506
3 . T TL . . . . 
507
4 L . .L . . . . 
508
5 L . .. . . . . 
509
*/
510
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
511
    if(top_type){
512 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
513
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
514
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
515
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
516 0da71265 Michael Niedermayer
    
517 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
518
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
519 0da71265 Michael Niedermayer
    
520 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
521
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
522 3981c385 Michael Niedermayer
        
523
        h->top_cbp= h->cbp_table[top_xy];
524 0da71265 Michael Niedermayer
    }else{
525
        h->non_zero_count_cache[4+8*0]=      
526
        h->non_zero_count_cache[5+8*0]=
527
        h->non_zero_count_cache[6+8*0]=
528
        h->non_zero_count_cache[7+8*0]=
529
    
530
        h->non_zero_count_cache[1+8*0]=
531
        h->non_zero_count_cache[2+8*0]=
532
    
533
        h->non_zero_count_cache[1+8*3]=
534 3981c385 Michael Niedermayer
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
535
        
536
        if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
537
        else                  h->top_cbp= 0;
538 0da71265 Michael Niedermayer
    }
539
    
540
    if(left_type[0]){
541 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
542
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
543
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
544
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
545 3981c385 Michael Niedermayer
        h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
546 0da71265 Michael Niedermayer
    }else{
547
        h->non_zero_count_cache[3+8*1]= 
548
        h->non_zero_count_cache[3+8*2]= 
549
        h->non_zero_count_cache[0+8*1]= 
550 3981c385 Michael Niedermayer
        h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
551
        
552
        if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
553
        else                  h->left_cbp= 0;
554 0da71265 Michael Niedermayer
    }
555
    
556
    if(left_type[1]){
557 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
558
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
559
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
560
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
561 0da71265 Michael Niedermayer
    }else{
562
        h->non_zero_count_cache[3+8*3]= 
563
        h->non_zero_count_cache[3+8*4]= 
564
        h->non_zero_count_cache[0+8*2]= 
565 3981c385 Michael Niedermayer
        h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
566 0da71265 Michael Niedermayer
    }
567
    
568
#if 1
569 5ad984c9 Loren Merritt
    //FIXME direct mb can skip much of this
570
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
571 0da71265 Michael Niedermayer
        int list;
572
        for(list=0; list<2; list++){
573 5ad984c9 Loren Merritt
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
574 0da71265 Michael Niedermayer
                /*if(!h->mv_cache_clean[list]){
575
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
576
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
577
                    h->mv_cache_clean[list]= 1;
578
                }*/
579 5ad984c9 Loren Merritt
                continue;
580 0da71265 Michael Niedermayer
            }
581
            h->mv_cache_clean[list]= 0;
582
            
583
            if(IS_INTER(topleft_type)){
584
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
585
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
586
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
587
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
588
            }else{
589
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
590
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
591
            }
592
            
593
            if(IS_INTER(top_type)){
594
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
595
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
596
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
597
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
598
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
599
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
600
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
601
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
602
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
603
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
604
            }else{
605
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
606
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
607
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
608
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
609
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
610
            }
611
612
            if(IS_INTER(topright_type)){
613
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
614
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
615
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
616
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
617
            }else{
618
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
619
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
620
            }
621
            
622
            //FIXME unify cleanup or sth
623
            if(IS_INTER(left_type[0])){
624
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
625
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
626
                *(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]];
627
                *(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]];
628
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
629
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
630
            }else{
631
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
632
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
633
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
634
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
635
            }
636
            
637
            if(IS_INTER(left_type[1])){
638
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
639
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
640
                *(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]];
641
                *(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]];
642
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
643
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
644
            }else{
645
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
646
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
647
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
648
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
649
            }
650
651
            h->ref_cache[list][scan8[5 ]+1] = 
652
            h->ref_cache[list][scan8[7 ]+1] = 
653
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
654
            h->ref_cache[list][scan8[4 ]] = 
655
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
656
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
657
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
658
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
659
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
660
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
661 9e528114 Laurent Aimar
662
            if( h->pps.cabac ) {
663
                /* XXX beurk, Load mvd */
664
                if(IS_INTER(topleft_type)){
665
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
666
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
667
                }else{
668
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
669
                }
670
671
                if(IS_INTER(top_type)){
672
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
673
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
674
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
675
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
676
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
677
                }else{
678
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
679
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
680
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
681
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
682
                }
683
                if(IS_INTER(left_type[0])){
684
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
685
                    *(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]];
686
                    *(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]];
687
                }else{
688
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
689
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
690
                }
691
                if(IS_INTER(left_type[1])){
692
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
693
                    *(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]];
694
                    *(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]];
695
                }else{
696
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
697
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
698
                }
699
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
700
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
701
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
702
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
703
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
704 5ad984c9 Loren Merritt
705
                if(h->slice_type == B_TYPE){
706
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
707
708
                    if(IS_DIRECT(top_type)){
709
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
710
                    }else if(IS_8X8(top_type)){
711
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
712
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
713
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
714
                    }else{
715
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
716
                    }
717
                    
718
                    //FIXME interlacing
719
                    if(IS_DIRECT(left_type[0])){
720
                        h->direct_cache[scan8[0] - 1 + 0*8]=
721
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
722
                    }else if(IS_8X8(left_type[0])){
723
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
724
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
725
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
726
                    }else{
727
                        h->direct_cache[scan8[0] - 1 + 0*8]=
728
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
729
                    }
730
                }
731 9e528114 Laurent Aimar
            }
732 0da71265 Michael Niedermayer
        }
733
    }
734
#endif
735
}
736
737
static inline void write_back_intra_pred_mode(H264Context *h){
738
    MpegEncContext * const s = &h->s;
739 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
740 0da71265 Michael Niedermayer
741
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
742
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
743
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
744
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
745
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
746
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
747
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
748
}
749
750
/**
751
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
752
 */
753
static inline int check_intra4x4_pred_mode(H264Context *h){
754
    MpegEncContext * const s = &h->s;
755
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
756
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
757
    int i;
758
    
759
    if(!(h->top_samples_available&0x8000)){
760
        for(i=0; i<4; i++){
761
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
762
            if(status<0){
763 9b879566 Michel Bardiaux
                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);
764 0da71265 Michael Niedermayer
                return -1;
765
            } else if(status){
766
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
767
            }
768
        }
769
    }
770
    
771
    if(!(h->left_samples_available&0x8000)){
772
        for(i=0; i<4; i++){
773
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
774
            if(status<0){
775 9b879566 Michel Bardiaux
                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);
776 0da71265 Michael Niedermayer
                return -1;
777
            } else if(status){
778
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
779
            }
780
        }
781
    }
782
783
    return 0;
784
} //FIXME cleanup like next
785
786
/**
787
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
788
 */
789
static inline int check_intra_pred_mode(H264Context *h, int mode){
790
    MpegEncContext * const s = &h->s;
791
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
792
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
793
    
794 5175b937 Loic Le Loarer
    if(mode < 0 || mode > 6) {
795
        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);
796 7440fe83 Michael Niedermayer
        return -1;
797 5175b937 Loic Le Loarer
    }
798 7440fe83 Michael Niedermayer
    
799 0da71265 Michael Niedermayer
    if(!(h->top_samples_available&0x8000)){
800
        mode= top[ mode ];
801
        if(mode<0){
802 9b879566 Michel Bardiaux
            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);
803 0da71265 Michael Niedermayer
            return -1;
804
        }
805
    }
806
    
807
    if(!(h->left_samples_available&0x8000)){
808
        mode= left[ mode ];
809
        if(mode<0){
810 9b879566 Michel Bardiaux
            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);
811 0da71265 Michael Niedermayer
            return -1;
812
        } 
813
    }
814
815
    return mode;
816
}
817
818
/**
819
 * gets the predicted intra4x4 prediction mode.
820
 */
821
static inline int pred_intra_mode(H264Context *h, int n){
822
    const int index8= scan8[n];
823
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
824
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
825
    const int min= FFMIN(left, top);
826
827 95c26348 Michael Niedermayer
    tprintf("mode:%d %d min:%d\n", left ,top, min);
828 0da71265 Michael Niedermayer
829
    if(min<0) return DC_PRED;
830
    else      return min;
831
}
832
833
static inline void write_back_non_zero_count(H264Context *h){
834
    MpegEncContext * const s = &h->s;
835 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
836 0da71265 Michael Niedermayer
837 53c05b1e Michael Niedermayer
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
838
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
839
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
840
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
841
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
842
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
843
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
844
    
845
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
846
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
847
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
848
849
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
850
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
851
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
852 0da71265 Michael Niedermayer
}
853
854
/**
855
 * gets the predicted number of non zero coefficients.
856
 * @param n block index
857
 */
858
static inline int pred_non_zero_count(H264Context *h, int n){
859
    const int index8= scan8[n];
860
    const int left= h->non_zero_count_cache[index8 - 1];
861
    const int top = h->non_zero_count_cache[index8 - 8];
862
    int i= left + top;
863
    
864
    if(i<64) i= (i+1)>>1;
865
866 95c26348 Michael Niedermayer
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
867 0da71265 Michael Niedermayer
868
    return i&31;
869
}
870
871 1924f3ce Michael Niedermayer
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
872
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
873
874
    if(topright_ref != PART_NOT_AVAILABLE){
875
        *C= h->mv_cache[list][ i - 8 + part_width ];
876
        return topright_ref;
877
    }else{
878 95c26348 Michael Niedermayer
        tprintf("topright MV not available\n");
879
880 1924f3ce Michael Niedermayer
        *C= h->mv_cache[list][ i - 8 - 1 ];
881
        return h->ref_cache[list][ i - 8 - 1 ];
882
    }
883
}
884
885 0da71265 Michael Niedermayer
/**
886
 * gets the predicted MV.
887
 * @param n the block index
888
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
889
 * @param mx the x component of the predicted motion vector
890
 * @param my the y component of the predicted motion vector
891
 */
892
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
893
    const int index8= scan8[n];
894
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
895
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
896
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
897
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
898 1924f3ce Michael Niedermayer
    const int16_t * C;
899
    int diagonal_ref, match_count;
900
901 0da71265 Michael Niedermayer
    assert(part_width==1 || part_width==2 || part_width==4);
902 1924f3ce Michael Niedermayer
903 0da71265 Michael Niedermayer
/* mv_cache
904
  B . . A T T T T 
905
  U . . L . . , .
906
  U . . L . . . .
907
  U . . L . . , .
908
  . . . L . . . .
909
*/
910 1924f3ce Michael Niedermayer
911
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
912
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
913
    if(match_count > 1){ //most common
914
        *mx= mid_pred(A[0], B[0], C[0]);
915
        *my= mid_pred(A[1], B[1], C[1]);
916
    }else if(match_count==1){
917
        if(left_ref==ref){
918
            *mx= A[0];
919
            *my= A[1];        
920
        }else if(top_ref==ref){
921
            *mx= B[0];
922
            *my= B[1];        
923 0da71265 Michael Niedermayer
        }else{
924 1924f3ce Michael Niedermayer
            *mx= C[0];
925
            *my= C[1];        
926 0da71265 Michael Niedermayer
        }
927
    }else{
928 1924f3ce Michael Niedermayer
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
929 0da71265 Michael Niedermayer
            *mx= A[0];
930 1924f3ce Michael Niedermayer
            *my= A[1];        
931 0da71265 Michael Niedermayer
        }else{
932 1924f3ce Michael Niedermayer
            *mx= mid_pred(A[0], B[0], C[0]);
933
            *my= mid_pred(A[1], B[1], C[1]);
934 0da71265 Michael Niedermayer
        }
935
    }
936 1924f3ce Michael Niedermayer
        
937 af6e2fed Michael Niedermayer
    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);
938 0da71265 Michael Niedermayer
}
939
940
/**
941
 * gets the directionally predicted 16x8 MV.
942
 * @param n the block index
943
 * @param mx the x component of the predicted motion vector
944
 * @param my the y component of the predicted motion vector
945
 */
946
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
947
    if(n==0){
948
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
949
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
950
951 579448a0 Loïc Le Loarer
        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);
952 0da71265 Michael Niedermayer
        
953
        if(top_ref == ref){
954
            *mx= B[0];
955
            *my= B[1];
956
            return;
957
        }
958
    }else{
959
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
960
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
961
        
962 579448a0 Loïc Le Loarer
        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);
963 0da71265 Michael Niedermayer
964
        if(left_ref == ref){
965
            *mx= A[0];
966
            *my= A[1];
967
            return;
968
        }
969
    }
970
971
    //RARE
972
    pred_motion(h, n, 4, list, ref, mx, my);
973
}
974
975
/**
976
 * gets the directionally predicted 8x16 MV.
977
 * @param n the block index
978
 * @param mx the x component of the predicted motion vector
979
 * @param my the y component of the predicted motion vector
980
 */
981
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
982
    if(n==0){
983
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
984
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
985
        
986 579448a0 Loïc Le Loarer
        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);
987 0da71265 Michael Niedermayer
988
        if(left_ref == ref){
989
            *mx= A[0];
990
            *my= A[1];
991
            return;
992
        }
993
    }else{
994 1924f3ce Michael Niedermayer
        const int16_t * C;
995
        int diagonal_ref;
996
997
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
998 0da71265 Michael Niedermayer
        
999 579448a0 Loïc Le Loarer
        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);
1000 0da71265 Michael Niedermayer
1001 1924f3ce Michael Niedermayer
        if(diagonal_ref == ref){ 
1002 0da71265 Michael Niedermayer
            *mx= C[0];
1003
            *my= C[1];
1004
            return;
1005
        }
1006
    }
1007
1008
    //RARE
1009
    pred_motion(h, n, 2, list, ref, mx, my);
1010
}
1011
1012
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1013
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1014
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1015
1016 5175b937 Loic Le Loarer
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1017 0da71265 Michael Niedermayer
1018
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1019
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1020
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1021
       
1022
        *mx = *my = 0;
1023
        return;
1024
    }
1025
        
1026
    pred_motion(h, 0, 4, 0, 0, mx, my);
1027
1028
    return;
1029
}
1030
1031 5ad984c9 Loren Merritt
static inline void direct_dist_scale_factor(H264Context * const h){
1032
    const int poc = h->s.current_picture_ptr->poc;
1033
    const int poc1 = h->ref_list[1][0].poc;
1034
    int i;
1035
    for(i=0; i<h->ref_count[0]; i++){
1036
        int poc0 = h->ref_list[0][i].poc;
1037
        int td = clip(poc1 - poc0, -128, 127);
1038
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1039
            h->dist_scale_factor[i] = 256;
1040
        }else{
1041
            int tb = clip(poc - poc0, -128, 127);
1042
            int tx = (16384 + (ABS(td) >> 1)) / td;
1043
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1044
        }
1045
    }
1046
}
1047
1048
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1049
    MpegEncContext * const s = &h->s;
1050
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1051
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1052
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1053
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1054
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1055
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1056
    const int is_b8x8 = IS_8X8(*mb_type);
1057
    int sub_mb_type;
1058
    int i8, i4;
1059
1060
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1061
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1062
         * so we know exactly what block size to use */
1063
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1064
        *mb_type =    MB_TYPE_8x8;
1065
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1066
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1067
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1068
    }else{
1069
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1070
        *mb_type =    MB_TYPE_8x8;
1071
    }
1072
    if(!is_b8x8)
1073
        *mb_type |= MB_TYPE_DIRECT2;
1074
1075
    if(h->direct_spatial_mv_pred){
1076
        int ref[2];
1077
        int mv[2][2];
1078
        int list;
1079
1080
        /* ref = min(neighbors) */
1081
        for(list=0; list<2; list++){
1082
            int refa = h->ref_cache[list][scan8[0] - 1];
1083
            int refb = h->ref_cache[list][scan8[0] - 8];
1084
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1085
            if(refc == -2)
1086
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1087
            ref[list] = refa;
1088
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1089
                ref[list] = refb;
1090
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1091
                ref[list] = refc;
1092
            if(ref[list] < 0)
1093
                ref[list] = -1;
1094
        }
1095
1096
        if(ref[0] < 0 && ref[1] < 0){
1097
            ref[0] = ref[1] = 0;
1098
            mv[0][0] = mv[0][1] =
1099
            mv[1][0] = mv[1][1] = 0;
1100
        }else{
1101
            for(list=0; list<2; list++){
1102
                if(ref[list] >= 0)
1103
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1104
                else
1105
                    mv[list][0] = mv[list][1] = 0;
1106
            }
1107
        }
1108
1109
        if(ref[1] < 0){
1110
            *mb_type &= ~MB_TYPE_P0L1;
1111
            sub_mb_type &= ~MB_TYPE_P0L1;
1112
        }else if(ref[0] < 0){
1113
            *mb_type &= ~MB_TYPE_P0L0;
1114
            sub_mb_type &= ~MB_TYPE_P0L0;
1115
        }
1116
1117
        if(IS_16X16(*mb_type)){
1118
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1119
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1120
            if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1121
                ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1122
                if(ref[0] > 0)
1123
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1124
                else
1125
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1126
                if(ref[1] > 0)
1127
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1128
                else
1129
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1130
            }else{
1131
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1132
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1133
            }
1134
        }else{
1135
            for(i8=0; i8<4; i8++){
1136
                const int x8 = i8&1;
1137
                const int y8 = i8>>1;
1138
    
1139
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1140
                    continue;
1141
                h->sub_mb_type[i8] = sub_mb_type;
1142
    
1143
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1144
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1145
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1146
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1147
    
1148
                /* col_zero_flag */
1149
                if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1150
                    for(i4=0; i4<4; i4++){
1151
                        const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1152
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1153
                            if(ref[0] == 0)
1154
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1155
                            if(ref[1] == 0)
1156
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1157
                        }
1158
                    }
1159
                }
1160
            }
1161
        }
1162
    }else{ /* direct temporal mv pred */
1163
        /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1164
        if(IS_16X16(*mb_type)){
1165
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1166
            if(IS_INTRA(mb_type_col)){
1167
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1168
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1169
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1170
            }else{
1171
                const int ref0 = l1ref0[0];
1172
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1173
                const int16_t *mv_col = l1mv0[0];
1174
                int mv_l0[2];
1175
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1176
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1177
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1178
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1179
                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);
1180
            }
1181
        }else{
1182
            for(i8=0; i8<4; i8++){
1183
                const int x8 = i8&1;
1184
                const int y8 = i8>>1;
1185
                int ref0, dist_scale_factor;
1186
    
1187
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1188
                    continue;
1189
                h->sub_mb_type[i8] = sub_mb_type;
1190
                if(IS_INTRA(mb_type_col)){
1191
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1192
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1193
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1194
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1195
                    continue;
1196
                }
1197
    
1198
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1199
                dist_scale_factor = h->dist_scale_factor[ref0];
1200
    
1201
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1202
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1203
                for(i4=0; i4<4; i4++){
1204
                    const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1205
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1206
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1207
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1208
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1209
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1210
                }
1211
            }
1212
        }
1213
    }
1214
}
1215
1216 0da71265 Michael Niedermayer
static inline void write_back_motion(H264Context *h, int mb_type){
1217
    MpegEncContext * const s = &h->s;
1218
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1219
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1220
    int list;
1221
1222
    for(list=0; list<2; list++){
1223
        int y;
1224
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1225
            if(1){ //FIXME skip or never read if mb_type doesnt use it
1226
                for(y=0; y<4; y++){
1227
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1228
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1229
                }
1230 9e528114 Laurent Aimar
                if( h->pps.cabac ) {
1231
                    /* FIXME needed ? */
1232
                    for(y=0; y<4; y++){
1233
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1234
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1235
                    }
1236
                }
1237 0da71265 Michael Niedermayer
                for(y=0; y<2; y++){
1238 7d7f635d Loren Merritt
                    *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1239 0da71265 Michael Niedermayer
                }
1240
            }
1241 5ad984c9 Loren Merritt
            continue;
1242 0da71265 Michael Niedermayer
        }
1243
        
1244
        for(y=0; y<4; y++){
1245
            *(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];
1246
            *(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];
1247
        }
1248 9e528114 Laurent Aimar
        if( h->pps.cabac ) {
1249
            for(y=0; y<4; y++){
1250
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1251
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1252
            }
1253
        }
1254 0da71265 Michael Niedermayer
        for(y=0; y<2; y++){
1255
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1256
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1257
        }
1258
    }
1259 5ad984c9 Loren Merritt
    
1260
    if(h->slice_type == B_TYPE && h->pps.cabac){
1261
        if(IS_8X8(mb_type)){
1262
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1263
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1264
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1265
        }
1266
    }
1267 0da71265 Michael Niedermayer
}
1268
1269
/**
1270
 * Decodes a network abstraction layer unit.
1271
 * @param consumed is the number of bytes used as input
1272
 * @param length is the length of the array
1273
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1274
 * @returns decoded bytes, might be src+1 if no escapes 
1275
 */
1276
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1277
    int i, si, di;
1278
    uint8_t *dst;
1279
1280
//    src[0]&0x80;                //forbidden bit
1281
    h->nal_ref_idc= src[0]>>5;
1282
    h->nal_unit_type= src[0]&0x1F;
1283
1284
    src++; length--;
1285
#if 0    
1286
    for(i=0; i<length; i++)
1287
        printf("%2X ", src[i]);
1288
#endif
1289
    for(i=0; i+1<length; i+=2){
1290
        if(src[i]) continue;
1291
        if(i>0 && src[i-1]==0) i--;
1292
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1293
            if(src[i+2]!=3){
1294
                /* startcode, so we must be past the end */
1295
                length=i;
1296
            }
1297
            break;
1298
        }
1299
    }
1300
1301
    if(i>=length-1){ //no escaped 0
1302
        *dst_length= length;
1303
        *consumed= length+1; //+1 for the header
1304
        return src; 
1305
    }
1306
1307
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1308
    dst= h->rbsp_buffer;
1309
1310
//printf("deoding esc\n");
1311
    si=di=0;
1312
    while(si<length){ 
1313
        //remove escapes (very rare 1:2^22)
1314
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1315
            if(src[si+2]==3){ //escape
1316
                dst[di++]= 0;
1317
                dst[di++]= 0;
1318
                si+=3;
1319 c8470cc1 Michael Niedermayer
                continue;
1320 0da71265 Michael Niedermayer
            }else //next start code
1321
                break;
1322
        }
1323
1324
        dst[di++]= src[si++];
1325
    }
1326
1327
    *dst_length= di;
1328
    *consumed= si + 1;//+1 for the header
1329
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1330
    return dst;
1331
}
1332
1333 e5017ab8 Laurent Aimar
#if 0
1334 0da71265 Michael Niedermayer
/**
1335
 * @param src the data which should be escaped
1336
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1337
 * @param length the length of the src data
1338
 * @param dst_length the length of the dst array
1339
 * @returns length of escaped data in bytes or -1 if an error occured
1340
 */
1341
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1342
    int i, escape_count, si, di;
1343
    uint8_t *temp;
1344
    
1345
    assert(length>=0);
1346
    assert(dst_length>0);
1347
    
1348
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1349

1350
    if(length==0) return 1;
1351

1352
    escape_count= 0;
1353
    for(i=0; i<length; i+=2){
1354
        if(src[i]) continue;
1355
        if(i>0 && src[i-1]==0) 
1356
            i--;
1357
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1358
            escape_count++;
1359
            i+=2;
1360
        }
1361
    }
1362
    
1363
    if(escape_count==0){ 
1364
        if(dst+1 != src)
1365
            memcpy(dst+1, src, length);
1366
        return length + 1;
1367
    }
1368
    
1369
    if(length + escape_count + 1> dst_length)
1370
        return -1;
1371

1372
    //this should be damn rare (hopefully)
1373

1374
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1375
    temp= h->rbsp_buffer;
1376
//printf("encoding esc\n");
1377
    
1378
    si= 0;
1379
    di= 0;
1380
    while(si < length){
1381
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1382
            temp[di++]= 0; si++;
1383
            temp[di++]= 0; si++;
1384
            temp[di++]= 3; 
1385
            temp[di++]= src[si++];
1386
        }
1387
        else
1388
            temp[di++]= src[si++];
1389
    }
1390
    memcpy(dst+1, temp, length+escape_count);
1391
    
1392
    assert(di == length+escape_count);
1393
    
1394
    return di + 1;
1395
}
1396

1397
/**
1398
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1399
 */
1400
static void encode_rbsp_trailing(PutBitContext *pb){
1401
    int length;
1402
    put_bits(pb, 1, 1);
1403 fe455f33 Alex Beregszaszi
    length= (-put_bits_count(pb))&7;
1404 0da71265 Michael Niedermayer
    if(length) put_bits(pb, length, 0);
1405
}
1406 e5017ab8 Laurent Aimar
#endif
1407 0da71265 Michael Niedermayer
1408
/**
1409
 * identifies the exact end of the bitstream
1410
 * @return the length of the trailing, or 0 if damaged
1411
 */
1412
static int decode_rbsp_trailing(uint8_t *src){
1413
    int v= *src;
1414
    int r;
1415
1416 95c26348 Michael Niedermayer
    tprintf("rbsp trailing %X\n", v);
1417 0da71265 Michael Niedermayer
1418
    for(r=1; r<9; r++){
1419
        if(v&1) return r;
1420
        v>>=1;
1421
    }
1422
    return 0;
1423
}
1424
1425
/**
1426
 * idct tranforms the 16 dc values and dequantize them.
1427
 * @param qp quantization parameter
1428
 */
1429
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1430
    const int qmul= dequant_coeff[qp][0];
1431
#define stride 16
1432
    int i;
1433
    int temp[16]; //FIXME check if this is a good idea
1434
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1435
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1436
1437
//memset(block, 64, 2*256);
1438
//return;
1439
    for(i=0; i<4; i++){
1440
        const int offset= y_offset[i];
1441
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1442
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1443
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1444
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1445
1446
        temp[4*i+0]= z0+z3;
1447
        temp[4*i+1]= z1+z2;
1448
        temp[4*i+2]= z1-z2;
1449
        temp[4*i+3]= z0-z3;
1450
    }
1451
1452
    for(i=0; i<4; i++){
1453
        const int offset= x_offset[i];
1454
        const int z0= temp[4*0+i] + temp[4*2+i];
1455
        const int z1= temp[4*0+i] - temp[4*2+i];
1456
        const int z2= temp[4*1+i] - temp[4*3+i];
1457
        const int z3= temp[4*1+i] + temp[4*3+i];
1458
1459
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1460
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1461
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1462
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1463
    }
1464
}
1465
1466 e5017ab8 Laurent Aimar
#if 0
1467 0da71265 Michael Niedermayer
/**
1468
 * dct tranforms the 16 dc values.
1469
 * @param qp quantization parameter ??? FIXME
1470
 */
1471
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1472
//    const int qmul= dequant_coeff[qp][0];
1473
    int i;
1474
    int temp[16]; //FIXME check if this is a good idea
1475
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1476
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1477

1478
    for(i=0; i<4; i++){
1479
        const int offset= y_offset[i];
1480
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1481
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1482
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1483
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1484

1485
        temp[4*i+0]= z0+z3;
1486
        temp[4*i+1]= z1+z2;
1487
        temp[4*i+2]= z1-z2;
1488
        temp[4*i+3]= z0-z3;
1489
    }
1490

1491
    for(i=0; i<4; i++){
1492
        const int offset= x_offset[i];
1493
        const int z0= temp[4*0+i] + temp[4*2+i];
1494
        const int z1= temp[4*0+i] - temp[4*2+i];
1495
        const int z2= temp[4*1+i] - temp[4*3+i];
1496
        const int z3= temp[4*1+i] + temp[4*3+i];
1497

1498
        block[stride*0 +offset]= (z0 + z3)>>1;
1499
        block[stride*2 +offset]= (z1 + z2)>>1;
1500
        block[stride*8 +offset]= (z1 - z2)>>1;
1501
        block[stride*10+offset]= (z0 - z3)>>1;
1502
    }
1503
}
1504 e5017ab8 Laurent Aimar
#endif
1505
1506 0da71265 Michael Niedermayer
#undef xStride
1507
#undef stride
1508
1509
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1510
    const int qmul= dequant_coeff[qp][0];
1511
    const int stride= 16*2;
1512
    const int xStride= 16;
1513
    int a,b,c,d,e;
1514
1515
    a= block[stride*0 + xStride*0];
1516
    b= block[stride*0 + xStride*1];
1517
    c= block[stride*1 + xStride*0];
1518
    d= block[stride*1 + xStride*1];
1519
1520
    e= a-b;
1521
    a= a+b;
1522
    b= c-d;
1523
    c= c+d;
1524
1525
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1526
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1527
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1528
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1529
}
1530
1531 e5017ab8 Laurent Aimar
#if 0
1532 0da71265 Michael Niedermayer
static void chroma_dc_dct_c(DCTELEM *block){
1533
    const int stride= 16*2;
1534
    const int xStride= 16;
1535
    int a,b,c,d,e;
1536

1537
    a= block[stride*0 + xStride*0];
1538
    b= block[stride*0 + xStride*1];
1539
    c= block[stride*1 + xStride*0];
1540
    d= block[stride*1 + xStride*1];
1541

1542
    e= a-b;
1543
    a= a+b;
1544
    b= c-d;
1545
    c= c+d;
1546

1547
    block[stride*0 + xStride*0]= (a+c);
1548
    block[stride*0 + xStride*1]= (e+b);
1549
    block[stride*1 + xStride*0]= (a-c);
1550
    block[stride*1 + xStride*1]= (e-b);
1551
}
1552 e5017ab8 Laurent Aimar
#endif
1553 0da71265 Michael Niedermayer
1554
/**
1555
 * gets the chroma qp.
1556
 */
1557
static inline int get_chroma_qp(H264Context *h, int qscale){
1558
    
1559
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1560
}
1561
1562
1563 e5017ab8 Laurent Aimar
#if 0
1564 0da71265 Michael Niedermayer
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1565
    int i;
1566
    //FIXME try int temp instead of block
1567
    
1568
    for(i=0; i<4; i++){
1569
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1570
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1571
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1572
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1573
        const int z0= d0 + d3;
1574
        const int z3= d0 - d3;
1575
        const int z1= d1 + d2;
1576
        const int z2= d1 - d2;
1577
        
1578
        block[0 + 4*i]=   z0 +   z1;
1579
        block[1 + 4*i]= 2*z3 +   z2;
1580
        block[2 + 4*i]=   z0 -   z1;
1581
        block[3 + 4*i]=   z3 - 2*z2;
1582
    }    
1583

1584
    for(i=0; i<4; i++){
1585
        const int z0= block[0*4 + i] + block[3*4 + i];
1586
        const int z3= block[0*4 + i] - block[3*4 + i];
1587
        const int z1= block[1*4 + i] + block[2*4 + i];
1588
        const int z2= block[1*4 + i] - block[2*4 + i];
1589
        
1590
        block[0*4 + i]=   z0 +   z1;
1591
        block[1*4 + i]= 2*z3 +   z2;
1592
        block[2*4 + i]=   z0 -   z1;
1593
        block[3*4 + i]=   z3 - 2*z2;
1594
    }
1595
}
1596 e5017ab8 Laurent Aimar
#endif
1597 0da71265 Michael Niedermayer
1598
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1599
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1600
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1601
    int i;
1602
    const int * const quant_table= quant_coeff[qscale];
1603
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1604
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1605
    const unsigned int threshold2= (threshold1<<1);
1606
    int last_non_zero;
1607
1608
    if(seperate_dc){
1609
        if(qscale<=18){
1610
            //avoid overflows
1611
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1612
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1613
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1614
1615
            int level= block[0]*quant_coeff[qscale+18][0];
1616
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1617
                if(level>0){
1618
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1619
                    block[0]= level;
1620
                }else{
1621
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1622
                    block[0]= -level;
1623
                }
1624
//                last_non_zero = i;
1625
            }else{
1626
                block[0]=0;
1627
            }
1628
        }else{
1629
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1630
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1631
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1632
1633
            int level= block[0]*quant_table[0];
1634
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1635
                if(level>0){
1636
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1637
                    block[0]= level;
1638
                }else{
1639
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1640
                    block[0]= -level;
1641
                }
1642
//                last_non_zero = i;
1643
            }else{
1644
                block[0]=0;
1645
            }
1646
        }
1647
        last_non_zero= 0;
1648
        i=1;
1649
    }else{
1650
        last_non_zero= -1;
1651
        i=0;
1652
    }
1653
1654
    for(; i<16; i++){
1655
        const int j= scantable[i];
1656
        int level= block[j]*quant_table[j];
1657
1658
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1659
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1660
        if(((unsigned)(level+threshold1))>threshold2){
1661
            if(level>0){
1662
                level= (bias + level)>>QUANT_SHIFT;
1663
                block[j]= level;
1664
            }else{
1665
                level= (bias - level)>>QUANT_SHIFT;
1666
                block[j]= -level;
1667
            }
1668
            last_non_zero = i;
1669
        }else{
1670
            block[j]=0;
1671
        }
1672
    }
1673
1674
    return last_non_zero;
1675
}
1676
1677
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1678
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1679
    ((uint32_t*)(src+0*stride))[0]= a;
1680
    ((uint32_t*)(src+1*stride))[0]= a;
1681
    ((uint32_t*)(src+2*stride))[0]= a;
1682
    ((uint32_t*)(src+3*stride))[0]= a;
1683
}
1684
1685
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1686
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1687
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1688
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1689
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1690
}
1691
1692
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1693
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1694
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1695
    
1696
    ((uint32_t*)(src+0*stride))[0]= 
1697
    ((uint32_t*)(src+1*stride))[0]= 
1698
    ((uint32_t*)(src+2*stride))[0]= 
1699
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1700
}
1701
1702
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1703
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1704
    
1705
    ((uint32_t*)(src+0*stride))[0]= 
1706
    ((uint32_t*)(src+1*stride))[0]= 
1707
    ((uint32_t*)(src+2*stride))[0]= 
1708
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1709
}
1710
1711
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1712
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1713
    
1714
    ((uint32_t*)(src+0*stride))[0]= 
1715
    ((uint32_t*)(src+1*stride))[0]= 
1716
    ((uint32_t*)(src+2*stride))[0]= 
1717
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1718
}
1719
1720
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1721
    ((uint32_t*)(src+0*stride))[0]= 
1722
    ((uint32_t*)(src+1*stride))[0]= 
1723
    ((uint32_t*)(src+2*stride))[0]= 
1724
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1725
}
1726
1727
1728
#define LOAD_TOP_RIGHT_EDGE\
1729
    const int t4= topright[0];\
1730
    const int t5= topright[1];\
1731
    const int t6= topright[2];\
1732
    const int t7= topright[3];\
1733
1734
#define LOAD_LEFT_EDGE\
1735
    const int l0= src[-1+0*stride];\
1736
    const int l1= src[-1+1*stride];\
1737
    const int l2= src[-1+2*stride];\
1738
    const int l3= src[-1+3*stride];\
1739
1740
#define LOAD_TOP_EDGE\
1741
    const int t0= src[ 0-1*stride];\
1742
    const int t1= src[ 1-1*stride];\
1743
    const int t2= src[ 2-1*stride];\
1744
    const int t3= src[ 3-1*stride];\
1745
1746
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1747
    const int lt= src[-1-1*stride];
1748
    LOAD_TOP_EDGE
1749
    LOAD_LEFT_EDGE
1750
1751
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1752
    src[0+2*stride]=
1753
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1754
    src[0+1*stride]=
1755
    src[1+2*stride]=
1756
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1757
    src[0+0*stride]=
1758
    src[1+1*stride]=
1759
    src[2+2*stride]=
1760
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1761
    src[1+0*stride]=
1762
    src[2+1*stride]=
1763
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1764
    src[2+0*stride]=
1765
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1766
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1767 4cfbf61b Falk Hüffner
}
1768 0da71265 Michael Niedermayer
1769
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1770
    LOAD_TOP_EDGE    
1771
    LOAD_TOP_RIGHT_EDGE    
1772
//    LOAD_LEFT_EDGE    
1773
1774
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1775
    src[1+0*stride]=
1776
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1777
    src[2+0*stride]=
1778
    src[1+1*stride]=
1779
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1780
    src[3+0*stride]=
1781
    src[2+1*stride]=
1782
    src[1+2*stride]=
1783
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1784
    src[3+1*stride]=
1785
    src[2+2*stride]=
1786
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1787
    src[3+2*stride]=
1788
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1789
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1790 4cfbf61b Falk Hüffner
}
1791 0da71265 Michael Niedermayer
1792
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1793
    const int lt= src[-1-1*stride];
1794
    LOAD_TOP_EDGE    
1795
    LOAD_LEFT_EDGE    
1796
    const __attribute__((unused)) int unu= l3;
1797
1798
    src[0+0*stride]=
1799
    src[1+2*stride]=(lt + t0 + 1)>>1;
1800
    src[1+0*stride]=
1801
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1802
    src[2+0*stride]=
1803
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1804
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1805
    src[0+1*stride]=
1806
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1807
    src[1+1*stride]=
1808
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1809
    src[2+1*stride]=
1810
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1811
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1812
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1813
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1814 4cfbf61b Falk Hüffner
}
1815 0da71265 Michael Niedermayer
1816
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1817
    LOAD_TOP_EDGE    
1818
    LOAD_TOP_RIGHT_EDGE    
1819
    const __attribute__((unused)) int unu= t7;
1820
1821
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1822
    src[1+0*stride]=
1823
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1824
    src[2+0*stride]=
1825
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1826
    src[3+0*stride]=
1827
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1828
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1829
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1830
    src[1+1*stride]=
1831
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1832
    src[2+1*stride]=
1833
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1834
    src[3+1*stride]=
1835
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1836
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1837 4cfbf61b Falk Hüffner
}
1838 0da71265 Michael Niedermayer
1839
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1840
    LOAD_LEFT_EDGE    
1841
1842
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1843
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1844
    src[2+0*stride]=
1845
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1846
    src[3+0*stride]=
1847
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1848
    src[2+1*stride]=
1849
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1850
    src[3+1*stride]=
1851
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1852
    src[3+2*stride]=
1853
    src[1+3*stride]=
1854
    src[0+3*stride]=
1855
    src[2+2*stride]=
1856
    src[2+3*stride]=
1857
    src[3+3*stride]=l3;
1858 4cfbf61b Falk Hüffner
}
1859 0da71265 Michael Niedermayer
    
1860
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1861
    const int lt= src[-1-1*stride];
1862
    LOAD_TOP_EDGE    
1863
    LOAD_LEFT_EDGE    
1864
    const __attribute__((unused)) int unu= t3;
1865
1866
    src[0+0*stride]=
1867
    src[2+1*stride]=(lt + l0 + 1)>>1;
1868
    src[1+0*stride]=
1869
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1870
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1871
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1872
    src[0+1*stride]=
1873
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1874
    src[1+1*stride]=
1875
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1876
    src[0+2*stride]=
1877
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1878
    src[1+2*stride]=
1879
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1880
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1881
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1882 4cfbf61b Falk Hüffner
}
1883 0da71265 Michael Niedermayer
1884
static void pred16x16_vertical_c(uint8_t *src, int stride){
1885
    int i;
1886
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1887
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1888
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1889
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1890
    
1891
    for(i=0; i<16; i++){
1892
        ((uint32_t*)(src+i*stride))[0]= a;
1893
        ((uint32_t*)(src+i*stride))[1]= b;
1894
        ((uint32_t*)(src+i*stride))[2]= c;
1895
        ((uint32_t*)(src+i*stride))[3]= d;
1896
    }
1897
}
1898
1899
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1900
    int i;
1901
1902
    for(i=0; i<16; i++){
1903
        ((uint32_t*)(src+i*stride))[0]=
1904
        ((uint32_t*)(src+i*stride))[1]=
1905
        ((uint32_t*)(src+i*stride))[2]=
1906
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1907
    }
1908
}
1909
1910
static void pred16x16_dc_c(uint8_t *src, int stride){
1911
    int i, dc=0;
1912
1913
    for(i=0;i<16; i++){
1914
        dc+= src[-1+i*stride];
1915
    }
1916
    
1917
    for(i=0;i<16; i++){
1918
        dc+= src[i-stride];
1919
    }
1920
1921
    dc= 0x01010101*((dc + 16)>>5);
1922
1923
    for(i=0; i<16; i++){
1924
        ((uint32_t*)(src+i*stride))[0]=
1925
        ((uint32_t*)(src+i*stride))[1]=
1926
        ((uint32_t*)(src+i*stride))[2]=
1927
        ((uint32_t*)(src+i*stride))[3]= dc;
1928
    }
1929
}
1930
1931
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1932
    int i, dc=0;
1933
1934
    for(i=0;i<16; i++){
1935
        dc+= src[-1+i*stride];
1936
    }
1937
    
1938
    dc= 0x01010101*((dc + 8)>>4);
1939
1940
    for(i=0; i<16; i++){
1941
        ((uint32_t*)(src+i*stride))[0]=
1942
        ((uint32_t*)(src+i*stride))[1]=
1943
        ((uint32_t*)(src+i*stride))[2]=
1944
        ((uint32_t*)(src+i*stride))[3]= dc;
1945
    }
1946
}
1947
1948
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1949
    int i, dc=0;
1950
1951
    for(i=0;i<16; i++){
1952
        dc+= src[i-stride];
1953
    }
1954
    dc= 0x01010101*((dc + 8)>>4);
1955
1956
    for(i=0; i<16; i++){
1957
        ((uint32_t*)(src+i*stride))[0]=
1958
        ((uint32_t*)(src+i*stride))[1]=
1959
        ((uint32_t*)(src+i*stride))[2]=
1960
        ((uint32_t*)(src+i*stride))[3]= dc;
1961
    }
1962
}
1963
1964
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1965
    int i;
1966
1967
    for(i=0; i<16; i++){
1968
        ((uint32_t*)(src+i*stride))[0]=
1969
        ((uint32_t*)(src+i*stride))[1]=
1970
        ((uint32_t*)(src+i*stride))[2]=
1971
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1972
    }
1973
}
1974
1975 8b82a956 Michael Niedermayer
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1976 30f73fc7 Michael Niedermayer
  int i, j, k;
1977
  int a;
1978
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1979
  const uint8_t * const src0 = src+7-stride;
1980
  const uint8_t *src1 = src+8*stride-1;
1981
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1982
  int H = src0[1] - src0[-1];
1983
  int V = src1[0] - src2[ 0];
1984
  for(k=2; k<=8; ++k) {
1985
    src1 += stride; src2 -= stride;
1986
    H += k*(src0[k] - src0[-k]);
1987
    V += k*(src1[0] - src2[ 0]);
1988
  }
1989 8b82a956 Michael Niedermayer
  if(svq3){
1990
    H = ( 5*(H/4) ) / 16;
1991
    V = ( 5*(V/4) ) / 16;
1992 2e26c8d2 Mike Melanson
1993
    /* required for 100% accuracy */
1994
    i = H; H = V; V = i;
1995 8b82a956 Michael Niedermayer
  }else{
1996
    H = ( 5*H+32 ) >> 6;
1997
    V = ( 5*V+32 ) >> 6;
1998
  }
1999 30f73fc7 Michael Niedermayer
2000
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2001
  for(j=16; j>0; --j) {
2002
    int b = a;
2003
    a += V;
2004
    for(i=-16; i<0; i+=4) {
2005
      src[16+i] = cm[ (b    ) >> 5 ];
2006
      src[17+i] = cm[ (b+  H) >> 5 ];
2007
      src[18+i] = cm[ (b+2*H) >> 5 ];
2008
      src[19+i] = cm[ (b+3*H) >> 5 ];
2009
      b += 4*H;
2010 0da71265 Michael Niedermayer
    }
2011 30f73fc7 Michael Niedermayer
    src += stride;
2012
  }
2013 0da71265 Michael Niedermayer
}
2014
2015 8b82a956 Michael Niedermayer
static void pred16x16_plane_c(uint8_t *src, int stride){
2016
    pred16x16_plane_compat_c(src, stride, 0);
2017
}
2018
2019 0da71265 Michael Niedermayer
static void pred8x8_vertical_c(uint8_t *src, int stride){
2020
    int i;
2021
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2022
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2023
    
2024
    for(i=0; i<8; i++){
2025
        ((uint32_t*)(src+i*stride))[0]= a;
2026
        ((uint32_t*)(src+i*stride))[1]= b;
2027
    }
2028
}
2029
2030
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2031
    int i;
2032
2033
    for(i=0; i<8; i++){
2034
        ((uint32_t*)(src+i*stride))[0]=
2035
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2036
    }
2037
}
2038
2039
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2040
    int i;
2041
2042
    for(i=0; i<4; i++){
2043
        ((uint32_t*)(src+i*stride))[0]= 
2044
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2045
    }
2046
    for(i=4; i<8; i++){
2047
        ((uint32_t*)(src+i*stride))[0]= 
2048
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2049
    }
2050
}
2051
2052
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2053
    int i;
2054
    int dc0, dc2;
2055
2056
    dc0=dc2=0;
2057
    for(i=0;i<4; i++){
2058
        dc0+= src[-1+i*stride];
2059
        dc2+= src[-1+(i+4)*stride];
2060
    }
2061
    dc0= 0x01010101*((dc0 + 2)>>2);
2062
    dc2= 0x01010101*((dc2 + 2)>>2);
2063
2064
    for(i=0; i<4; i++){
2065
        ((uint32_t*)(src+i*stride))[0]=
2066
        ((uint32_t*)(src+i*stride))[1]= dc0;
2067
    }
2068
    for(i=4; i<8; i++){
2069
        ((uint32_t*)(src+i*stride))[0]=
2070
        ((uint32_t*)(src+i*stride))[1]= dc2;
2071
    }
2072
}
2073
2074
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2075
    int i;
2076
    int dc0, dc1;
2077
2078
    dc0=dc1=0;
2079
    for(i=0;i<4; i++){
2080
        dc0+= src[i-stride];
2081
        dc1+= src[4+i-stride];
2082
    }
2083
    dc0= 0x01010101*((dc0 + 2)>>2);
2084
    dc1= 0x01010101*((dc1 + 2)>>2);
2085
2086
    for(i=0; i<4; i++){
2087
        ((uint32_t*)(src+i*stride))[0]= dc0;
2088
        ((uint32_t*)(src+i*stride))[1]= dc1;
2089
    }
2090
    for(i=4; i<8; i++){
2091
        ((uint32_t*)(src+i*stride))[0]= dc0;
2092
        ((uint32_t*)(src+i*stride))[1]= dc1;
2093
    }
2094
}
2095
2096
2097
static void pred8x8_dc_c(uint8_t *src, int stride){
2098
    int i;
2099
    int dc0, dc1, dc2, dc3;
2100
2101
    dc0=dc1=dc2=0;
2102
    for(i=0;i<4; i++){
2103
        dc0+= src[-1+i*stride] + src[i-stride];
2104
        dc1+= src[4+i-stride];
2105
        dc2+= src[-1+(i+4)*stride];
2106
    }
2107
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2108
    dc0= 0x01010101*((dc0 + 4)>>3);
2109
    dc1= 0x01010101*((dc1 + 2)>>2);
2110
    dc2= 0x01010101*((dc2 + 2)>>2);
2111
2112
    for(i=0; i<4; i++){
2113
        ((uint32_t*)(src+i*stride))[0]= dc0;
2114
        ((uint32_t*)(src+i*stride))[1]= dc1;
2115
    }
2116
    for(i=4; i<8; i++){
2117
        ((uint32_t*)(src+i*stride))[0]= dc2;
2118
        ((uint32_t*)(src+i*stride))[1]= dc3;
2119
    }
2120
}
2121
2122
static void pred8x8_plane_c(uint8_t *src, int stride){
2123 30f73fc7 Michael Niedermayer
  int j, k;
2124
  int a;
2125
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2126
  const uint8_t * const src0 = src+3-stride;
2127
  const uint8_t *src1 = src+4*stride-1;
2128
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2129
  int H = src0[1] - src0[-1];
2130
  int V = src1[0] - src2[ 0];
2131
  for(k=2; k<=4; ++k) {
2132
    src1 += stride; src2 -= stride;
2133
    H += k*(src0[k] - src0[-k]);
2134
    V += k*(src1[0] - src2[ 0]);
2135
  }
2136
  H = ( 17*H+16 ) >> 5;
2137
  V = ( 17*V+16 ) >> 5;
2138
2139
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2140
  for(j=8; j>0; --j) {
2141
    int b = a;
2142
    a += V;
2143
    src[0] = cm[ (b    ) >> 5 ];
2144
    src[1] = cm[ (b+  H) >> 5 ];
2145
    src[2] = cm[ (b+2*H) >> 5 ];
2146
    src[3] = cm[ (b+3*H) >> 5 ];
2147
    src[4] = cm[ (b+4*H) >> 5 ];
2148
    src[5] = cm[ (b+5*H) >> 5 ];
2149
    src[6] = cm[ (b+6*H) >> 5 ];
2150
    src[7] = cm[ (b+7*H) >> 5 ];
2151
    src += stride;
2152
  }
2153 0da71265 Michael Niedermayer
}
2154
2155
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2156
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2157
                           int src_x_offset, int src_y_offset,
2158
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2159
    MpegEncContext * const s = &h->s;
2160
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2161
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2162
    const int luma_xy= (mx&3) + ((my&3)<<2);
2163
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2164
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2165
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2166
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2167
    int extra_height= extra_width;
2168
    int emu=0;
2169
    const int full_mx= mx>>2;
2170
    const int full_my= my>>2;
2171
    
2172
    assert(pic->data[0]);
2173
    
2174
    if(mx&7) extra_width -= 3;
2175
    if(my&7) extra_height -= 3;
2176
    
2177
    if(   full_mx < 0-extra_width 
2178
       || full_my < 0-extra_height 
2179
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
2180
       || full_my + 16/*FIXME*/ > s->height + extra_height){
2181 c009df3f Michael Niedermayer
        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, s->width, s->height);
2182 0da71265 Michael Niedermayer
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2183
        emu=1;
2184
    }
2185
    
2186
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2187
    if(!square){
2188
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2189
    }
2190
    
2191
    if(s->flags&CODEC_FLAG_GRAY) return;
2192
    
2193
    if(emu){
2194 c009df3f Michael Niedermayer
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2195 0da71265 Michael Niedermayer
            src_cb= s->edge_emu_buffer;
2196
    }
2197
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2198
2199
    if(emu){
2200 c009df3f Michael Niedermayer
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2201 0da71265 Michael Niedermayer
            src_cr= s->edge_emu_buffer;
2202
    }
2203
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2204
}
2205
2206
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2207
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2208
                           int x_offset, int y_offset,
2209
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2210
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2211
                           int list0, int list1){
2212
    MpegEncContext * const s = &h->s;
2213
    qpel_mc_func *qpix_op=  qpix_put;
2214
    h264_chroma_mc_func chroma_op= chroma_put;
2215
    
2216
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2217
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2218
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2219
    x_offset += 8*s->mb_x;
2220
    y_offset += 8*s->mb_y;
2221
    
2222
    if(list0){
2223 1924f3ce Michael Niedermayer
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2224 0da71265 Michael Niedermayer
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2225
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2226
                           qpix_op, chroma_op);
2227
2228
        qpix_op=  qpix_avg;
2229
        chroma_op= chroma_avg;
2230
    }
2231
2232
    if(list1){
2233 1924f3ce Michael Niedermayer
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2234 0da71265 Michael Niedermayer
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2235
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2236
                           qpix_op, chroma_op);
2237
    }
2238
}
2239
2240
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2241
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2242
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2243
    MpegEncContext * const s = &h->s;
2244 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2245 0da71265 Michael Niedermayer
    const int mb_type= s->current_picture.mb_type[mb_xy];
2246
    
2247
    assert(IS_INTER(mb_type));
2248
    
2249
    if(IS_16X16(mb_type)){
2250
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2251
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2252
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2253
    }else if(IS_16X8(mb_type)){
2254
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2255
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2256
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2257
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2258
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2259
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2260
    }else if(IS_8X16(mb_type)){
2261
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2262
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2263
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2264
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2265
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2266
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2267
    }else{
2268
        int i;
2269
        
2270
        assert(IS_8X8(mb_type));
2271
2272
        for(i=0; i<4; i++){
2273
            const int sub_mb_type= h->sub_mb_type[i];
2274
            const int n= 4*i;
2275
            int x_offset= (i&1)<<2;
2276
            int y_offset= (i&2)<<1;
2277
2278
            if(IS_SUB_8X8(sub_mb_type)){
2279
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2280
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2281
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2282
            }else if(IS_SUB_8X4(sub_mb_type)){
2283
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2284
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2285
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2286
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2287
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2288
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2289
            }else if(IS_SUB_4X8(sub_mb_type)){
2290
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2291
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2292
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2293
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2294
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2295
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2296
            }else{
2297
                int j;
2298
                assert(IS_SUB_4X4(sub_mb_type));
2299
                for(j=0; j<4; j++){
2300
                    int sub_x_offset= x_offset + 2*(j&1);
2301
                    int sub_y_offset= y_offset +   (j&2);
2302
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2303
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2304
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2305
                }
2306
            }
2307
        }
2308
    }
2309
}
2310
2311
static void decode_init_vlc(H264Context *h){
2312
    static int done = 0;
2313
2314
    if (!done) {
2315
        int i;
2316
        done = 1;
2317
2318
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2319
                 &chroma_dc_coeff_token_len [0], 1, 1,
2320 073c2593 Burkhard Plaum
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2321 0da71265 Michael Niedermayer
2322
        for(i=0; i<4; i++){
2323
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2324
                     &coeff_token_len [i][0], 1, 1,
2325 073c2593 Burkhard Plaum
                     &coeff_token_bits[i][0], 1, 1, 1);
2326 0da71265 Michael Niedermayer
        }
2327
2328
        for(i=0; i<3; i++){
2329
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2330
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2331 073c2593 Burkhard Plaum
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2332 0da71265 Michael Niedermayer
        }
2333
        for(i=0; i<15; i++){
2334
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2335
                     &total_zeros_len [i][0], 1, 1,
2336 073c2593 Burkhard Plaum
                     &total_zeros_bits[i][0], 1, 1, 1);
2337 0da71265 Michael Niedermayer
        }
2338
2339
        for(i=0; i<6; i++){
2340
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2341
                     &run_len [i][0], 1, 1,
2342 073c2593 Burkhard Plaum
                     &run_bits[i][0], 1, 1, 1);
2343 0da71265 Michael Niedermayer
        }
2344
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2345
                 &run_len [6][0], 1, 1,
2346 073c2593 Burkhard Plaum
                 &run_bits[6][0], 1, 1, 1);
2347 0da71265 Michael Niedermayer
    }
2348
}
2349
2350
/**
2351
 * Sets the intra prediction function pointers.
2352
 */
2353
static void init_pred_ptrs(H264Context *h){
2354
//    MpegEncContext * const s = &h->s;
2355
2356
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2357
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2358
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2359
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2360
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2361
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2362
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2363
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2364
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2365
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2366
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2367
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2368
2369
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2370
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2371
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2372
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2373
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2374
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2375
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2376
2377
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2378
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2379
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2380
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2381
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2382
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2383
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2384
}
2385
2386
static void free_tables(H264Context *h){
2387
    av_freep(&h->intra4x4_pred_mode);
2388 e5017ab8 Laurent Aimar
    av_freep(&h->chroma_pred_mode_table);
2389
    av_freep(&h->cbp_table);
2390 9e528114 Laurent Aimar
    av_freep(&h->mvd_table[0]);
2391
    av_freep(&h->mvd_table[1]);
2392 5ad984c9 Loren Merritt
    av_freep(&h->direct_table);
2393 0da71265 Michael Niedermayer
    av_freep(&h->non_zero_count);
2394
    av_freep(&h->slice_table_base);
2395 53c05b1e Michael Niedermayer
    av_freep(&h->top_border);
2396 0da71265 Michael Niedermayer
    h->slice_table= NULL;
2397 e5017ab8 Laurent Aimar
2398 0da71265 Michael Niedermayer
    av_freep(&h->mb2b_xy);
2399
    av_freep(&h->mb2b8_xy);
2400
}
2401
2402
/**
2403
 * allocates tables.
2404
 * needs widzh/height
2405
 */
2406
static int alloc_tables(H264Context *h){
2407
    MpegEncContext * const s = &h->s;
2408 7bc9090a Michael Niedermayer
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2409 0da71265 Michael Niedermayer
    int x,y;
2410
2411
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2412 e5017ab8 Laurent Aimar
2413 53c05b1e Michael Niedermayer
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2414 0da71265 Michael Niedermayer
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2415 53c05b1e Michael Niedermayer
    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2416 5d0e4cb8 Michael Niedermayer
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2417 0da71265 Michael Niedermayer
2418 e5017ab8 Laurent Aimar
    if( h->pps.cabac ) {
2419
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2420 9e528114 Laurent Aimar
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2421
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2422 5ad984c9 Loren Merritt
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2423 e5017ab8 Laurent Aimar
    }
2424
2425 0da71265 Michael Niedermayer
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2426 7bc9090a Michael Niedermayer
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2427 0da71265 Michael Niedermayer
2428
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2429
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2430
    for(y=0; y<s->mb_height; y++){
2431
        for(x=0; x<s->mb_width; x++){
2432 7bc9090a Michael Niedermayer
            const int mb_xy= x + y*s->mb_stride;
2433 0da71265 Michael Niedermayer
            const int b_xy = 4*x + 4*y*h->b_stride;
2434
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2435
        
2436
            h->mb2b_xy [mb_xy]= b_xy;
2437
            h->mb2b8_xy[mb_xy]= b8_xy;
2438
        }
2439
    }
2440
    
2441
    return 0;
2442
fail:
2443
    free_tables(h);
2444
    return -1;
2445
}
2446
2447
static void common_init(H264Context *h){
2448
    MpegEncContext * const s = &h->s;
2449
2450
    s->width = s->avctx->width;
2451
    s->height = s->avctx->height;
2452
    s->codec_id= s->avctx->codec->id;
2453
    
2454
    init_pred_ptrs(h);
2455
2456 9a41c2c7 Michael Niedermayer
    s->unrestricted_mv=1;
2457 0da71265 Michael Niedermayer
    s->decode=1; //FIXME
2458
}
2459
2460
static int decode_init(AVCodecContext *avctx){
2461
    H264Context *h= avctx->priv_data;
2462
    MpegEncContext * const s = &h->s;
2463
2464 3edcacde Michael Niedermayer
    MPV_decode_defaults(s);
2465
    
2466 0da71265 Michael Niedermayer
    s->avctx = avctx;
2467
    common_init(h);
2468
2469
    s->out_format = FMT_H264;
2470
    s->workaround_bugs= avctx->workaround_bugs;
2471
2472
    // set defaults
2473
//    s->decode_mb= ff_h263_decode_mb;
2474
    s->low_delay= 1;
2475
    avctx->pix_fmt= PIX_FMT_YUV420P;
2476
2477
    decode_init_vlc(h);
2478
    
2479 bf835aee Michael Niedermayer
    if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2480 4770b1b4 Roberto Togni
        h->is_avc = 0;
2481
    else {
2482
        if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2483
            av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2484
            return -1;
2485
        }
2486
        h->is_avc = 1;
2487
        h->got_avcC = 0;
2488
    }
2489
2490 0da71265 Michael Niedermayer
    return 0;
2491
}
2492
2493
static void frame_start(H264Context *h){
2494
    MpegEncContext * const s = &h->s;
2495
    int i;
2496
2497
    MPV_frame_start(s, s->avctx);
2498
    ff_er_frame_start(s);
2499 17bfbd70 Michael Niedermayer
    h->mmco_index=0;
2500 0da71265 Michael Niedermayer
2501
    assert(s->linesize && s->uvlinesize);
2502
2503
    for(i=0; i<16; i++){
2504
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2505
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2506
    }
2507
    for(i=0; i<4; i++){
2508
        h->block_offset[16+i]=
2509
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2510
    }
2511
2512
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2513
}
2514
2515 53c05b1e Michael Niedermayer
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2516
    MpegEncContext * const s = &h->s;
2517
    int i;
2518
    
2519
    src_y  -=   linesize;
2520
    src_cb -= uvlinesize;
2521
    src_cr -= uvlinesize;
2522
2523
    h->left_border[0]= h->top_border[s->mb_x][15];
2524
    for(i=1; i<17; i++){
2525
        h->left_border[i]= src_y[15+i*  linesize];
2526
    }
2527
    
2528
    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2529
    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2530
2531
    if(!(s->flags&CODEC_FLAG_GRAY)){
2532
        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2533
        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2534
        for(i=1; i<9; i++){
2535
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2536
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2537
        }
2538
        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2539
        *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2540
    }
2541
}
2542
2543
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){
2544
    MpegEncContext * const s = &h->s;
2545
    int temp8, i;
2546
    uint64_t temp64;
2547 d89dc06a Loren Merritt
    int deblock_left = (s->mb_x > 0);
2548
    int deblock_top  = (s->mb_y > 0);
2549 53c05b1e Michael Niedermayer
2550
    src_y  -=   linesize + 1;
2551
    src_cb -= uvlinesize + 1;
2552
    src_cr -= uvlinesize + 1;
2553
2554
#define XCHG(a,b,t,xchg)\
2555
t= a;\
2556
if(xchg)\
2557
    a= b;\
2558
b= t;
2559 d89dc06a Loren Merritt
2560
    if(deblock_left){
2561
        for(i = !deblock_top; i<17; i++){
2562
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2563
        }
2564
    }
2565
2566
    if(deblock_top){
2567
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2568
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2569 53c05b1e Michael Niedermayer
    }
2570
2571
    if(!(s->flags&CODEC_FLAG_GRAY)){
2572 d89dc06a Loren Merritt
        if(deblock_left){
2573
            for(i = !deblock_top; i<9; i++){
2574
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2575
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2576
            }
2577
        }
2578
        if(deblock_top){
2579
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2580
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2581 53c05b1e Michael Niedermayer
        }
2582
    }
2583
}
2584
2585 0da71265 Michael Niedermayer
static void hl_decode_mb(H264Context *h){
2586
    MpegEncContext * const s = &h->s;
2587
    const int mb_x= s->mb_x;
2588
    const int mb_y= s->mb_y;
2589 7bc9090a Michael Niedermayer
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2590 0da71265 Michael Niedermayer
    const int mb_type= s->current_picture.mb_type[mb_xy];
2591
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2592
    int linesize, uvlinesize /*dct_offset*/;
2593
    int i;
2594
2595
    if(!s->decode)
2596
        return;
2597
2598
    if(s->mb_skiped){
2599
    }
2600
2601
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2602
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2603
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2604
2605
    if (h->mb_field_decoding_flag) {
2606
        linesize = s->linesize * 2;
2607
        uvlinesize = s->uvlinesize * 2;
2608
        if(mb_y&1){ //FIXME move out of this func?
2609
            dest_y -= s->linesize*15;
2610
            dest_cb-= s->linesize*7;
2611
            dest_cr-= s->linesize*7;
2612
        }
2613
    } else {
2614
        linesize = s->linesize;
2615
        uvlinesize = s->uvlinesize;
2616
//        dct_offset = s->linesize * 16;
2617
    }
2618
2619
    if(IS_INTRA(mb_type)){
2620 53c05b1e Michael Niedermayer
        if(h->deblocking_filter)
2621
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2622
2623 0da71265 Michael Niedermayer
        if(!(s->flags&CODEC_FLAG_GRAY)){
2624
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2625
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2626
        }
2627
2628
        if(IS_INTRA4x4(mb_type)){
2629
            if(!s->encoding){
2630
                for(i=0; i<16; i++){
2631
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2632 a9799653 Michael Niedermayer
                    uint8_t *topright;
2633 0da71265 Michael Niedermayer
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2634
                    int tr;
2635
2636 a9799653 Michael Niedermayer
                    if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2637
                        const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2638
                        assert(mb_y || linesize <= h->block_offset[i]);
2639
                        if(!topright_avail){
2640
                            tr= ptr[3 - linesize]*0x01010101;
2641
                            topright= (uint8_t*) &tr;
2642
                        }else if(i==5 && h->deblocking_filter){
2643
                            tr= *(uint32_t*)h->top_border[mb_x+1];
2644
                            topright= (uint8_t*) &tr;
2645
                        }else
2646
                            topright= ptr + 4 - linesize;
2647
                    }else
2648
                        topright= NULL;
2649 0da71265 Michael Niedermayer
2650
                    h->pred4x4[ dir ](ptr, topright, linesize);
2651 8b82a956 Michael Niedermayer
                    if(h->non_zero_count_cache[ scan8[i] ]){
2652
                        if(s->codec_id == CODEC_ID_H264)
2653 0fa8158d Michael Niedermayer
                            s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2654 8b82a956 Michael Niedermayer
                        else
2655
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2656
                    }
2657 0da71265 Michael Niedermayer
                }
2658
            }
2659
        }else{
2660
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2661 8b82a956 Michael Niedermayer
            if(s->codec_id == CODEC_ID_H264)
2662
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2663
            else
2664
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2665 0da71265 Michael Niedermayer
        }
2666 53c05b1e Michael Niedermayer
        if(h->deblocking_filter)
2667
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2668 8b82a956 Michael Niedermayer
    }else if(s->codec_id == CODEC_ID_H264){
2669 0da71265 Michael Niedermayer
        hl_motion(h, dest_y, dest_cb, dest_cr,
2670
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2671
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2672
    }
2673
2674
2675
    if(!IS_INTRA4x4(mb_type)){
2676 4704097a Michael Niedermayer
        if(s->codec_id == CODEC_ID_H264){
2677
            for(i=0; i<16; i++){
2678
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2679
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2680 0fa8158d Michael Niedermayer
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2681 4704097a Michael Niedermayer
                }
2682
            }
2683
        }else{
2684
            for(i=0; i<16; i++){
2685
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2686
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2687 8b82a956 Michael Niedermayer
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2688 4704097a Michael Niedermayer
                }
2689 0da71265 Michael Niedermayer
            }
2690
        }
2691
    }
2692
2693
    if(!(s->flags&CODEC_FLAG_GRAY)){
2694
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2695
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2696 4704097a Michael Niedermayer
        if(s->codec_id == CODEC_ID_H264){
2697
            for(i=16; i<16+4; i++){
2698
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2699
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2700 0fa8158d Michael Niedermayer
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2701 4704097a Michael Niedermayer
                }
2702 0da71265 Michael Niedermayer
            }
2703 4704097a Michael Niedermayer
            for(i=20; i<20+4; i++){
2704
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2705
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2706 0fa8158d Michael Niedermayer
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2707 4704097a Michael Niedermayer
                }
2708
            }
2709
        }else{
2710
            for(i=16; i<16+4; i++){
2711
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2712
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2713 8b82a956 Michael Niedermayer
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2714 4704097a Michael Niedermayer
                }
2715
            }
2716
            for(i=20; i<20+4; i++){
2717
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2718
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2719
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2720
                }
2721 0da71265 Michael Niedermayer
            }
2722
        }
2723
    }
2724 53c05b1e Michael Niedermayer
    if(h->deblocking_filter) {
2725
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2726
        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2727
    }
2728 0da71265 Michael Niedermayer
}
2729
2730
/**
2731
 * fills the default_ref_list.
2732
 */
2733
static int fill_default_ref_list(H264Context *h){
2734
    MpegEncContext * const s = &h->s;
2735
    int i;
2736
    Picture sorted_short_ref[16];
2737
    
2738
    if(h->slice_type==B_TYPE){
2739
        int out_i;
2740
        int limit= -1;
2741
2742
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2743
            int best_i=-1;
2744 792bb815 Michael Niedermayer
            int best_poc=INT_MAX;
2745 0da71265 Michael Niedermayer
2746
            for(i=0; i<h->short_ref_count; i++){
2747
                const int poc= h->short_ref[i]->poc;
2748
                if(poc > limit && poc < best_poc){
2749
                    best_poc= poc;
2750
                    best_i= i;
2751
                }
2752
            }
2753
            
2754
            assert(best_i != -1);
2755
            
2756
            limit= best_poc;
2757
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2758
        }
2759
    }
2760
2761
    if(s->picture_structure == PICT_FRAME){
2762
        if(h->slice_type==B_TYPE){
2763
            const int current_poc= s->current_picture_ptr->poc;
2764
            int list;
2765
2766
            for(list=0; list<2; list++){
2767
                int index=0;
2768
2769
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2770 5ad984c9 Loren Merritt
                    const int i2= list ? i : h->short_ref_count - i - 1;
2771 0da71265 Michael Niedermayer
                    const int poc= sorted_short_ref[i2].poc;
2772
                    
2773
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2774
2775
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2776
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2777
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2778
                    }
2779
                }
2780
2781
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2782
                    if(h->long_ref[i]->reference != 3) continue;
2783
2784
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2785
                    h->default_ref_list[ list ][index++].pic_id= i;;
2786
                }
2787
                
2788
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2789
                    Picture temp= h->default_ref_list[1][0];
2790
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2791
                    h->default_ref_list[1][0] = temp;
2792
                }
2793
2794
                if(index < h->ref_count[ list ])
2795
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2796
            }
2797
        }else{
2798
            int index=0;
2799
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2800
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2801
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2802
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2803
            }
2804
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2805
                if(h->long_ref[i]->reference != 3) continue;
2806
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2807
                h->default_ref_list[0][index++].pic_id= i;;
2808
            }
2809
            if(index < h->ref_count[0])
2810
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2811
        }
2812
    }else{ //FIELD
2813
        if(h->slice_type==B_TYPE){
2814
        }else{
2815
            //FIXME second field balh
2816
        }
2817
    }
2818
    return 0;
2819
}
2820
2821
static int decode_ref_pic_list_reordering(H264Context *h){
2822
    MpegEncContext * const s = &h->s;
2823
    int list;
2824
    
2825
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2826
    
2827
    for(list=0; list<2; list++){
2828
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2829
2830
        if(get_bits1(&s->gb)){
2831
            int pred= h->curr_pic_num;
2832
            int index;
2833
2834
            for(index=0; ; index++){
2835
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2836
                int pic_id;
2837
                int i;
2838
                
2839 0bc42cad Loren Merritt
                if(reordering_of_pic_nums_idc==3) 
2840
                    break;
2841 0da71265 Michael Niedermayer
                
2842
                if(index >= h->ref_count[list]){
2843 9b879566 Michel Bardiaux
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2844 0da71265 Michael Niedermayer
                    return -1;
2845
                }
2846
                
2847
                if(reordering_of_pic_nums_idc<3){
2848
                    if(reordering_of_pic_nums_idc<2){
2849
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2850
2851
                        if(abs_diff_pic_num >= h->max_pic_num){
2852 9b879566 Michel Bardiaux
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2853 0da71265 Michael Niedermayer
                            return -1;
2854
                        }
2855
2856
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2857
                        else                                pred+= abs_diff_pic_num;
2858
                        pred &= h->max_pic_num - 1;
2859
                    
2860
                        for(i= h->ref_count[list]-1; i>=index; i--){
2861
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2862
                                break;
2863
                        }
2864
                    }else{
2865
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2866
2867
                        for(i= h->ref_count[list]-1; i>=index; i--){
2868
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2869
                                break;
2870
                        }
2871
                    }
2872
2873
                    if(i < index){
2874 9b879566 Michel Bardiaux
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2875 0da71265 Michael Niedermayer
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2876
                    }else if(i > index){
2877
                        Picture tmp= h->ref_list[list][i];
2878
                        for(; i>index; i--){
2879
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2880
                        }
2881
                        h->ref_list[list][index]= tmp;
2882
                    }
2883 0bc42cad Loren Merritt
                }else{
2884 9b879566 Michel Bardiaux
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2885 0da71265 Michael Niedermayer
                    return -1;
2886
                }
2887
            }
2888
        }
2889
2890
        if(h->slice_type!=B_TYPE) break;
2891
    }
2892 5ad984c9 Loren Merritt
    
2893
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
2894
        direct_dist_scale_factor(h);
2895 0da71265 Michael Niedermayer
    return 0;    
2896
}
2897
2898
static int pred_weight_table(H264Context *h){
2899
    MpegEncContext * const s = &h->s;
2900
    int list, i;
2901
    
2902
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2903
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2904
2905
    for(list=0; list<2; list++){
2906
        for(i=0; i<h->ref_count[list]; i++){
2907
            int luma_weight_flag, chroma_weight_flag;
2908
            
2909
            luma_weight_flag= get_bits1(&s->gb);
2910
            if(luma_weight_flag){
2911
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2912
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2913
            }
2914
2915
            chroma_weight_flag= get_bits1(&s->gb);
2916
            if(chroma_weight_flag){
2917
                int j;
2918
                for(j=0; j<2; j++){
2919
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2920
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2921
                }
2922
            }
2923
        }
2924
        if(h->slice_type != B_TYPE) break;
2925
    }
2926
    return 0;
2927
}
2928
2929
/**
2930 5175b937 Loic Le Loarer
 * instantaneous decoder refresh.
2931 0da71265 Michael Niedermayer
 */
2932
static void idr(H264Context *h){
2933 a4dae92b Loren Merritt
    int i,j;
2934
2935
#define CHECK_DELAY(pic) \
2936
    for(j = 0; h->delayed_pic[j]; j++) \
2937
        if(pic == h->delayed_pic[j]){ \
2938
            pic->reference=1; \
2939
            break; \
2940
        }
2941 0da71265 Michael Niedermayer
2942
    for(i=0; i<h->long_ref_count; i++){
2943
        h->long_ref[i]->reference=0;
2944 a4dae92b Loren Merritt
        CHECK_DELAY(h->long_ref[i]);
2945 0da71265 Michael Niedermayer
        h->long_ref[i]= NULL;
2946
    }
2947
    h->long_ref_count=0;
2948
2949
    for(i=0; i<h->short_ref_count; i++){
2950
        h->short_ref[i]->reference=0;
2951 a4dae92b Loren Merritt
        CHECK_DELAY(h->short_ref[i]);
2952 0da71265 Michael Niedermayer
        h->short_ref[i]= NULL;
2953
    }
2954
    h->short_ref_count=0;
2955
}
2956 a4dae92b Loren Merritt
#undef CHECK_DELAY
2957 0da71265 Michael Niedermayer
2958
/**
2959
 *
2960
 * @return the removed picture or NULL if an error occures
2961
 */
2962
static Picture * remove_short(H264Context *h, int frame_num){
2963 1924f3ce Michael Niedermayer
    MpegEncContext * const s = &h->s;
2964 0da71265 Michael Niedermayer
    int i;
2965
    
2966 1924f3ce Michael Niedermayer
    if(s->avctx->debug&FF_DEBUG_MMCO)
2967 9b879566 Michel Bardiaux
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2968 1924f3ce Michael Niedermayer
    
2969 0da71265 Michael Niedermayer
    for(i=0; i<h->short_ref_count; i++){
2970
        Picture *pic= h->short_ref[i];
2971 1924f3ce Michael Niedermayer
        if(s->avctx->debug&FF_DEBUG_MMCO)
2972 9b879566 Michel Bardiaux
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2973 0da71265 Michael Niedermayer
        if(pic->frame_num == frame_num){
2974
            h->short_ref[i]= NULL;
2975
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2976
            h->short_ref_count--;
2977
            return pic;
2978
        }
2979
    }
2980
    return NULL;
2981
}
2982
2983
/**
2984
 *
2985
 * @return the removed picture or NULL if an error occures
2986
 */
2987