Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 806bb93f

History | View | Annotate | Download (286 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 36940eca Loren Merritt
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
65 0da71265 Michael Niedermayer
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
66
    int poc_type;                      ///< pic_order_cnt_type
67
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
68
    int delta_pic_order_always_zero_flag;
69
    int offset_for_non_ref_pic;
70
    int offset_for_top_to_bottom_field;
71
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
72
    int ref_frame_count;               ///< num_ref_frames
73 a15e68de Michael Niedermayer
    int gaps_in_frame_num_allowed_flag;
74 0da71265 Michael Niedermayer
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
75
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
76
    int frame_mbs_only_flag;
77
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
78
    int direct_8x8_inference_flag;
79 a15e68de Michael Niedermayer
    int crop;                   ///< frame_cropping_flag
80
    int crop_left;              ///< frame_cropping_rect_left_offset
81
    int crop_right;             ///< frame_cropping_rect_right_offset
82
    int crop_top;               ///< frame_cropping_rect_top_offset
83
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
84 0da71265 Michael Niedermayer
    int vui_parameters_present_flag;
85 5ff85f1d Michael Niedermayer
    AVRational sar;
86 ec587a50 Måns Rullgård
    int timing_info_present_flag;
87
    uint32_t num_units_in_tick;
88
    uint32_t time_scale;
89
    int fixed_frame_rate_flag;
90 0da71265 Michael Niedermayer
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
91 4e4d983e Loren Merritt
    int bitstream_restriction_flag;
92
    int num_reorder_frames;
93 0da71265 Michael Niedermayer
}SPS;
94
95
/**
96
 * Picture parameter set
97
 */
98
typedef struct PPS{
99
    int sps_id;
100
    int cabac;                  ///< entropy_coding_mode_flag
101
    int pic_order_present;      ///< pic_order_present_flag
102
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
103
    int mb_slice_group_map_type;
104
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
105
    int weighted_pred;          ///< weighted_pred_flag
106
    int weighted_bipred_idc;
107
    int init_qp;                ///< pic_init_qp_minus26 + 26
108
    int init_qs;                ///< pic_init_qs_minus26 + 26
109
    int chroma_qp_index_offset;
110
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
111
    int constrained_intra_pred; ///< constrained_intra_pred_flag
112
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 43efd19a Loren Merritt
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
114 0da71265 Michael Niedermayer
}PPS;
115
116
/**
117
 * Memory management control operation opcode.
118
 */
119
typedef enum MMCOOpcode{
120
    MMCO_END=0,
121
    MMCO_SHORT2UNUSED,
122
    MMCO_LONG2UNUSED,
123
    MMCO_SHORT2LONG,
124
    MMCO_SET_MAX_LONG,
125
    MMCO_RESET, 
126
    MMCO_LONG,
127
} MMCOOpcode;
128
129
/**
130
 * Memory management control operation.
131
 */
132
typedef struct MMCO{
133
    MMCOOpcode opcode;
134
    int short_frame_num;
135
    int long_index;
136
} MMCO;
137
138
/**
139
 * H264Context
140
 */
141
typedef struct H264Context{
142
    MpegEncContext s;
143
    int nal_ref_idc;        
144
    int nal_unit_type;
145
#define NAL_SLICE                1
146
#define NAL_DPA                        2
147
#define NAL_DPB                        3
148
#define NAL_DPC                        4
149
#define NAL_IDR_SLICE                5
150
#define NAL_SEI                        6
151
#define NAL_SPS                        7
152
#define NAL_PPS                        8
153
#define NAL_PICTURE_DELIMITER        9
154
#define NAL_FILTER_DATA                10
155
    uint8_t *rbsp_buffer;
156
    int rbsp_buffer_size;
157
158 4770b1b4 Roberto Togni
    /**
159
      * Used to parse AVC variant of h264
160
      */
161
    int is_avc; ///< this flag is != 0 if codec is avc1
162
    int got_avcC; ///< flag used to parse avcC data only once
163
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
164
165 0da71265 Michael Niedermayer
    int chroma_qp; //QPc
166
167 160d679c Mike Melanson
    int prev_mb_skipped; //FIXME remove (IMHO not used)
168 0da71265 Michael Niedermayer
169
    //prediction stuff
170
    int chroma_pred_mode;
171
    int intra16x16_pred_mode;
172 6ba71fc4 Loïc Le Loarer
173
    int top_mb_xy;
174
    int left_mb_xy[2];
175 0da71265 Michael Niedermayer
    
176
    int8_t intra4x4_pred_mode_cache[5*8];
177
    int8_t (*intra4x4_pred_mode)[8];
178
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
179 43efd19a Loren Merritt
    void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
180 0da71265 Michael Niedermayer
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
181
    void (*pred16x16[4+3])(uint8_t *src, int stride);
182
    unsigned int topleft_samples_available;
183
    unsigned int top_samples_available;
184
    unsigned int topright_samples_available;
185
    unsigned int left_samples_available;
186 6ba71fc4 Loïc Le Loarer
    uint8_t (*top_borders[2])[16+2*8];
187
    uint8_t left_border[2*(17+2*9)];
188 0da71265 Michael Niedermayer
189
    /**
190
     * non zero coeff count cache.
191
     * is 64 if not available.
192
     */
193 bd9ea6a4 Michael Niedermayer
    uint8_t non_zero_count_cache[6*8] __align8;
194 53c05b1e Michael Niedermayer
    uint8_t (*non_zero_count)[16];
195 0da71265 Michael Niedermayer
196
    /**
197
     * Motion vector cache.
198
     */
199 bd9ea6a4 Michael Niedermayer
    int16_t mv_cache[2][5*8][2] __align8;
200
    int8_t ref_cache[2][5*8] __align8;
201 0da71265 Michael Niedermayer
#define LIST_NOT_USED -1 //FIXME rename?
202
#define PART_NOT_AVAILABLE -2
203
    
204
    /**
205
     * is 1 if the specific list MV&references are set to 0,0,-2.
206
     */
207
    int mv_cache_clean[2];
208
209 6867a90b Loic Le Loarer
    /**
210 43efd19a Loren Merritt
     * number of neighbors (top and/or left) that used 8x8 dct
211
     */
212
    int neighbor_transform_size;
213
214
    /**
215 6867a90b Loic Le Loarer
     * block_offset[ 0..23] for frame macroblocks
216
     * block_offset[24..47] for field macroblocks
217
     */
218
    int block_offset[2*(16+8)];
219 0da71265 Michael Niedermayer
    
220 a55f20bd Loren Merritt
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
221
    uint32_t *mb2b8_xy;
222 650cec0c Loren Merritt
    int b_stride; //FIXME use s->b4_stride
223 0da71265 Michael Niedermayer
    int b8_stride;
224
225 8b82a956 Michael Niedermayer
    int halfpel_flag;
226
    int thirdpel_flag;
227
228 da3b9756 Mike Melanson
    int unknown_svq3_flag;
229
    int next_slice_index;
230
231 0da71265 Michael Niedermayer
    SPS sps_buffer[MAX_SPS_COUNT];
232
    SPS sps; ///< current sps
233
    
234
    PPS pps_buffer[MAX_PPS_COUNT];
235
    /**
236
     * current pps
237
     */
238 3b66c4c5 Kevin Baragona
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
239 0da71265 Michael Niedermayer
240 36940eca Loren Merritt
    uint16_t (*dequant4_coeff)[16]; // FIXME quant matrices should be per SPS or PPS
241 43efd19a Loren Merritt
    uint16_t (*dequant8_coeff)[64];
242
243 0da71265 Michael Niedermayer
    int slice_num;
244
    uint8_t *slice_table_base;
245
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
246
    int slice_type;
247
    int slice_type_fixed;
248
    
249
    //interlacing specific flags
250 6ba71fc4 Loïc Le Loarer
    int mb_aff_frame;
251 0da71265 Michael Niedermayer
    int mb_field_decoding_flag;
252
    
253
    int sub_mb_type[4];
254
    
255
    //POC stuff
256
    int poc_lsb;
257
    int poc_msb;
258
    int delta_poc_bottom;
259
    int delta_poc[2];
260
    int frame_num;
261
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
262
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
263
    int frame_num_offset;         ///< for POC type 2
264
    int prev_frame_num_offset;    ///< for POC type 2
265
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
266
267
    /**
268
     * frame_num for frames or 2*frame_num for field pics.
269
     */
270
    int curr_pic_num;
271
    
272
    /**
273
     * max_frame_num or 2*max_frame_num for field pics.
274
     */
275
    int max_pic_num;
276
277
    //Weighted pred stuff
278 9f2d1b4f Loren Merritt
    int use_weight;
279
    int use_weight_chroma;
280 0da71265 Michael Niedermayer
    int luma_log2_weight_denom;
281
    int chroma_log2_weight_denom;
282
    int luma_weight[2][16];
283
    int luma_offset[2][16];
284
    int chroma_weight[2][16][2];
285
    int chroma_offset[2][16][2];
286 9f2d1b4f Loren Merritt
    int implicit_weight[16][16];
287 0da71265 Michael Niedermayer
   
288
    //deblock
289 53c05b1e Michael Niedermayer
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
290 980a82b7 Michael Niedermayer
    int slice_alpha_c0_offset;
291
    int slice_beta_offset;
292 0da71265 Michael Niedermayer
     
293
    int redundant_pic_count;
294
    
295
    int direct_spatial_mv_pred;
296 5ad984c9 Loren Merritt
    int dist_scale_factor[16];
297 2f944356 Loren Merritt
    int map_col_to_list0[2][16];
298 0da71265 Michael Niedermayer
299
    /**
300
     * num_ref_idx_l0/1_active_minus1 + 1
301
     */
302
    int ref_count[2];// FIXME split for AFF
303 17107065 Loren Merritt
    Picture *short_ref[32];
304
    Picture *long_ref[32];
305 0da71265 Michael Niedermayer
    Picture default_ref_list[2][32];
306
    Picture ref_list[2][32]; //FIXME size?
307
    Picture field_ref_list[2][32]; //FIXME size?
308 a4dae92b Loren Merritt
    Picture *delayed_pic[16]; //FIXME size?
309 4e4d983e Loren Merritt
    Picture *delayed_output_pic;
310 0da71265 Michael Niedermayer
    
311
    /**
312
     * memory management control operations buffer.
313
     */
314
    MMCO mmco[MAX_MMCO_COUNT];
315
    int mmco_index;
316
    
317
    int long_ref_count;  ///< number of actual long term references
318
    int short_ref_count; ///< number of actual short term references
319
    
320
    //data partitioning
321
    GetBitContext intra_gb;
322
    GetBitContext inter_gb;
323
    GetBitContext *intra_gb_ptr;
324
    GetBitContext *inter_gb_ptr;
325
    
326
    DCTELEM mb[16*24] __align8;
327 e5017ab8 Laurent Aimar
328
    /**
329
     * Cabac
330
     */
331
    CABACContext cabac;
332 43efd19a Loren Merritt
    uint8_t      cabac_state[460];
333 e5017ab8 Laurent Aimar
    int          cabac_init_idc;
334
335
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
336
    uint16_t     *cbp_table;
337 3981c385 Michael Niedermayer
    int top_cbp;
338
    int left_cbp;
339 693399ba Laurent Aimar
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
340 e5017ab8 Laurent Aimar
    uint8_t     *chroma_pred_mode_table;
341
    int         last_qscale_diff;
342 9e528114 Laurent Aimar
    int16_t     (*mvd_table[2])[2];
343 bd9ea6a4 Michael Niedermayer
    int16_t     mvd_cache[2][5*8][2] __align8;
344 5ad984c9 Loren Merritt
    uint8_t     *direct_table;
345
    uint8_t     direct_cache[5*8];
346 e5017ab8 Laurent Aimar
347 e4b36d44 Michael Niedermayer
    uint8_t zigzag_scan[16];
348
    uint8_t field_scan[16];
349 36940eca Loren Merritt
    const uint8_t *zigzag_scan_q0;
350
    const uint8_t *field_scan_q0;
351 0da71265 Michael Niedermayer
}H264Context;
352
353
static VLC coeff_token_vlc[4];
354
static VLC chroma_dc_coeff_token_vlc;
355
356
static VLC total_zeros_vlc[15];
357
static VLC chroma_dc_total_zeros_vlc[3];
358
359
static VLC run_vlc[6];
360
static VLC run7_vlc;
361
362 8b82a956 Michael Niedermayer
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
363
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
364 6ba71fc4 Loïc Le Loarer
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
365 8b82a956 Michael Niedermayer
366 377ec888 Michael Niedermayer
static inline uint32_t pack16to32(int a, int b){
367
#ifdef WORDS_BIGENDIAN
368
   return (b&0xFFFF) + (a<<16);
369
#else
370
   return (a&0xFFFF) + (b<<16);
371
#endif
372
}
373
374 0da71265 Michael Niedermayer
/**
375
 * fill a rectangle.
376 5175b937 Loic Le Loarer
 * @param h height of the rectangle, should be a constant
377
 * @param w width of the rectangle, should be a constant
378 0da71265 Michael Niedermayer
 * @param size the size of val (1 or 4), should be a constant
379
 */
380 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
381
    uint8_t *p= (uint8_t*)vp;
382 0da71265 Michael Niedermayer
    assert(size==1 || size==4);
383
    
384
    w      *= size;
385
    stride *= size;
386
    
387 bd9ea6a4 Michael Niedermayer
    assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
388 d9c780a8 Michael Niedermayer
    assert((stride&(w-1))==0);
389 3b66c4c5 Kevin Baragona
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
390 0da71265 Michael Niedermayer
    if(w==2 && h==2){
391
        *(uint16_t*)(p + 0)=
392
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
393
    }else if(w==2 && h==4){
394
        *(uint16_t*)(p + 0*stride)=
395
        *(uint16_t*)(p + 1*stride)=
396
        *(uint16_t*)(p + 2*stride)=
397
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
398 f7a8c179 Michael Niedermayer
    }else if(w==4 && h==1){
399
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
400 0da71265 Michael Niedermayer
    }else if(w==4 && h==2){
401
        *(uint32_t*)(p + 0*stride)=
402
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
403
    }else if(w==4 && h==4){
404
        *(uint32_t*)(p + 0*stride)=
405
        *(uint32_t*)(p + 1*stride)=
406
        *(uint32_t*)(p + 2*stride)=
407
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
408
    }else if(w==8 && h==1){
409
        *(uint32_t*)(p + 0)=
410
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
411
    }else if(w==8 && h==2){
412
        *(uint32_t*)(p + 0 + 0*stride)=
413
        *(uint32_t*)(p + 4 + 0*stride)=
414
        *(uint32_t*)(p + 0 + 1*stride)=
415
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
416
    }else if(w==8 && h==4){
417
        *(uint64_t*)(p + 0*stride)=
418
        *(uint64_t*)(p + 1*stride)=
419
        *(uint64_t*)(p + 2*stride)=
420
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
421
    }else if(w==16 && h==2){
422
        *(uint64_t*)(p + 0+0*stride)=
423
        *(uint64_t*)(p + 8+0*stride)=
424
        *(uint64_t*)(p + 0+1*stride)=
425
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
426
    }else if(w==16 && h==4){
427
        *(uint64_t*)(p + 0+0*stride)=
428
        *(uint64_t*)(p + 8+0*stride)=
429
        *(uint64_t*)(p + 0+1*stride)=
430
        *(uint64_t*)(p + 8+1*stride)=
431
        *(uint64_t*)(p + 0+2*stride)=
432
        *(uint64_t*)(p + 8+2*stride)=
433
        *(uint64_t*)(p + 0+3*stride)=
434
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
435
    }else
436
        assert(0);
437
}
438
439 46f2f05f Michael Niedermayer
static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
440 0da71265 Michael Niedermayer
    MpegEncContext * const s = &h->s;
441 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
442 0da71265 Michael Niedermayer
    int topleft_xy, top_xy, topright_xy, left_xy[2];
443
    int topleft_type, top_type, topright_type, left_type[2];
444 6867a90b Loic Le Loarer
    int left_block[8];
445 0da71265 Michael Niedermayer
    int i;
446
447 e2e5894a Loren Merritt
    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
448
    // the actual condition is whether we're on the edge of a slice,
449
    // and even then the intra and nnz parts are unnecessary.
450
    if(for_deblock && h->slice_num == 1)
451
        return;
452
453 3b66c4c5 Kevin Baragona
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it 
454 0da71265 Michael Niedermayer
    
455 6867a90b Loic Le Loarer
    top_xy     = mb_xy  - s->mb_stride;
456
    topleft_xy = top_xy - 1;
457
    topright_xy= top_xy + 1;
458
    left_xy[1] = left_xy[0] = mb_xy-1;
459
    left_block[0]= 0;
460
    left_block[1]= 1;
461
    left_block[2]= 2;
462
    left_block[3]= 3;
463
    left_block[4]= 7;
464
    left_block[5]= 10;
465
    left_block[6]= 8;
466
    left_block[7]= 11;
467 6ba71fc4 Loïc Le Loarer
    if(h->mb_aff_frame){
468 6867a90b Loic Le Loarer
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
469
        const int top_pair_xy      = pair_xy     - s->mb_stride;
470
        const int topleft_pair_xy  = top_pair_xy - 1;
471
        const int topright_pair_xy = top_pair_xy + 1;
472
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
473
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
474
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
475
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
476
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
477
        const int bottom = (s->mb_y & 1);
478
        tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
479
        if (bottom
480
                ? !curr_mb_frame_flag // bottom macroblock
481
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
482
                ) {
483
            top_xy -= s->mb_stride;
484
        }
485
        if (bottom
486
                ? !curr_mb_frame_flag // bottom macroblock
487
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
488
                ) {
489
            topleft_xy -= s->mb_stride;
490
        }
491
        if (bottom
492
                ? !curr_mb_frame_flag // bottom macroblock
493
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
494
                ) {
495
            topright_xy -= s->mb_stride;
496
        }
497
        if (left_mb_frame_flag != curr_mb_frame_flag) {
498
            left_xy[1] = left_xy[0] = pair_xy - 1;
499
            if (curr_mb_frame_flag) {
500
                if (bottom) {
501
                    left_block[0]= 2;
502
                    left_block[1]= 2;
503
                    left_block[2]= 3;
504
                    left_block[3]= 3;
505
                    left_block[4]= 8;
506
                    left_block[5]= 11;
507
                    left_block[6]= 8;
508
                    left_block[7]= 11;
509
                } else {
510
                    left_block[0]= 0;
511
                    left_block[1]= 0;
512
                    left_block[2]= 1;
513
                    left_block[3]= 1;
514
                    left_block[4]= 7;
515
                    left_block[5]= 10;
516
                    left_block[6]= 7;
517
                    left_block[7]= 10;
518
                }
519
            } else {
520
                left_xy[1] += s->mb_stride;
521
                //left_block[0]= 0;
522
                left_block[1]= 2;
523
                left_block[2]= 0;
524
                left_block[3]= 2;
525
                //left_block[4]= 7;
526
                left_block[5]= 10;
527
                left_block[6]= 7;
528
                left_block[7]= 10;
529
            }
530
        }
531 0da71265 Michael Niedermayer
    }
532
533 826de46e Loïc Le Loarer
    h->top_mb_xy = top_xy;
534
    h->left_mb_xy[0] = left_xy[0];
535
    h->left_mb_xy[1] = left_xy[1];
536 6ba71fc4 Loïc Le Loarer
    if(for_deblock){
537 46f2f05f Michael Niedermayer
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
538
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
539
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
540
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
541
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
542
    }else{
543
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
544
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
545
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
546
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
547
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
548
    }
549 0da71265 Michael Niedermayer
550
    if(IS_INTRA(mb_type)){
551
        h->topleft_samples_available= 
552
        h->top_samples_available= 
553
        h->left_samples_available= 0xFFFF;
554
        h->topright_samples_available= 0xEEEA;
555
556
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
557
            h->topleft_samples_available= 0xB3FF;
558
            h->top_samples_available= 0x33FF;
559
            h->topright_samples_available= 0x26EA;
560
        }
561
        for(i=0; i<2; i++){
562
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
563
                h->topleft_samples_available&= 0xDF5F;
564
                h->left_samples_available&= 0x5F5F;
565
            }
566
        }
567
        
568
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
569
            h->topleft_samples_available&= 0x7FFF;
570
        
571
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
572
            h->topright_samples_available&= 0xFBFF;
573
    
574
        if(IS_INTRA4x4(mb_type)){
575
            if(IS_INTRA4x4(top_type)){
576
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
577
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
578
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
579
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
580
            }else{
581
                int pred;
582 6fbcaaa0 Loic Le Loarer
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
583 0da71265 Michael Niedermayer
                    pred= -1;
584 6fbcaaa0 Loic Le Loarer
                else{
585
                    pred= 2;
586 0da71265 Michael Niedermayer
                }
587
                h->intra4x4_pred_mode_cache[4+8*0]=
588
                h->intra4x4_pred_mode_cache[5+8*0]=
589
                h->intra4x4_pred_mode_cache[6+8*0]=
590
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
591
            }
592
            for(i=0; i<2; i++){
593
                if(IS_INTRA4x4(left_type[i])){
594
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
595
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
596
                }else{
597
                    int pred;
598 6fbcaaa0 Loic Le Loarer
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
599 0da71265 Michael Niedermayer
                        pred= -1;
600 6fbcaaa0 Loic Le Loarer
                    else{
601
                        pred= 2;
602 0da71265 Michael Niedermayer
                    }
603
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
604
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
605
                }
606
            }
607
        }
608
    }
609
    
610
    
611
/*
612
0 . T T. T T T T 
613
1 L . .L . . . . 
614
2 L . .L . . . . 
615
3 . T TL . . . . 
616
4 L . .L . . . . 
617
5 L . .. . . . . 
618
*/
619
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
620
    if(top_type){
621 6867a90b Loic Le Loarer
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
622
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
623
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
624 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
625 0da71265 Michael Niedermayer
    
626 6867a90b Loic Le Loarer
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
627 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
628 0da71265 Michael Niedermayer
    
629 6867a90b Loic Le Loarer
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
630 53c05b1e Michael Niedermayer
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
631 3981c385 Michael Niedermayer
        
632 0da71265 Michael Niedermayer
    }else{
633
        h->non_zero_count_cache[4+8*0]=      
634
        h->non_zero_count_cache[5+8*0]=
635
        h->non_zero_count_cache[6+8*0]=
636
        h->non_zero_count_cache[7+8*0]=
637
    
638
        h->non_zero_count_cache[1+8*0]=
639
        h->non_zero_count_cache[2+8*0]=
640
    
641
        h->non_zero_count_cache[1+8*3]=
642 3981c385 Michael Niedermayer
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
643
        
644 0da71265 Michael Niedermayer
    }
645 826de46e Loïc Le Loarer
646 6867a90b Loic Le Loarer
    for (i=0; i<2; i++) {
647
        if(left_type[i]){
648
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
649
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
650
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
651
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
652
        }else{
653
            h->non_zero_count_cache[3+8*1 + 2*8*i]= 
654
            h->non_zero_count_cache[3+8*2 + 2*8*i]= 
655
            h->non_zero_count_cache[0+8*1 +   8*i]= 
656
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
657 826de46e Loïc Le Loarer
        }
658
    }
659
660
    if( h->pps.cabac ) {
661
        // top_cbp
662
        if(top_type) {
663
            h->top_cbp = h->cbp_table[top_xy];
664
        } else if(IS_INTRA(mb_type)) {
665
            h->top_cbp = 0x1C0;
666
        } else {
667
            h->top_cbp = 0;
668
        }
669
        // left_cbp
670
        if (left_type[0]) {
671
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
672
        } else if(IS_INTRA(mb_type)) {
673
            h->left_cbp = 0x1C0;
674
        } else {
675
            h->left_cbp = 0;
676
        }
677
        if (left_type[0]) {
678
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
679
        }
680
        if (left_type[1]) {
681
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
682 6867a90b Loic Le Loarer
        }
683 0da71265 Michael Niedermayer
    }
684 6867a90b Loic Le Loarer
685 0da71265 Michael Niedermayer
#if 1
686 5ad984c9 Loren Merritt
    //FIXME direct mb can skip much of this
687 e2e5894a Loren Merritt
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
688 0da71265 Michael Niedermayer
        int list;
689 e2e5894a Loren Merritt
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
690
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
691 0da71265 Michael Niedermayer
                /*if(!h->mv_cache_clean[list]){
692
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
693
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
694
                    h->mv_cache_clean[list]= 1;
695
                }*/
696 5ad984c9 Loren Merritt
                continue;
697 0da71265 Michael Niedermayer
            }
698
            h->mv_cache_clean[list]= 0;
699
            
700
            if(IS_INTER(top_type)){
701
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
702
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
703
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
704
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
705
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
706
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
707
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
708
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
709
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
710
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
711
            }else{
712
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
713
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
714
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
715
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
716
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
717
            }
718
719
            //FIXME unify cleanup or sth
720
            if(IS_INTER(left_type[0])){
721
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
722
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
723
                *(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]];
724
                *(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]];
725
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
726
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
727
            }else{
728
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
729
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
730
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
731
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
732
            }
733
            
734
            if(IS_INTER(left_type[1])){
735
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
736
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
737
                *(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]];
738
                *(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]];
739
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
740
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
741
            }else{
742
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
743
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
744
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
745
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
746 d9c780a8 Michael Niedermayer
                assert((!left_type[0]) == (!left_type[1]));
747 0da71265 Michael Niedermayer
            }
748
749 e2e5894a Loren Merritt
            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
750 46f2f05f Michael Niedermayer
                continue;
751
752 e2e5894a Loren Merritt
            if(IS_INTER(topleft_type)){
753
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
754
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
755
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
756
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
757
            }else{
758
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
759
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
760
            }
761
            
762
            if(IS_INTER(topright_type)){
763
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
764
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
765
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
766
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
767
            }else{
768
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
769
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
770
            }
771
            
772
773 0da71265 Michael Niedermayer
            h->ref_cache[list][scan8[5 ]+1] = 
774
            h->ref_cache[list][scan8[7 ]+1] = 
775 3b66c4c5 Kevin Baragona
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
776 0da71265 Michael Niedermayer
            h->ref_cache[list][scan8[4 ]] = 
777
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
778
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
779
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
780 3b66c4c5 Kevin Baragona
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
781 0da71265 Michael Niedermayer
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
782
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
783 9e528114 Laurent Aimar
784
            if( h->pps.cabac ) {
785
                /* XXX beurk, Load mvd */
786
                if(IS_INTER(topleft_type)){
787
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
788
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
789
                }else{
790
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
791
                }
792
793
                if(IS_INTER(top_type)){
794
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
795
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
796
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
797
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
798
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
799
                }else{
800
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
801
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
802
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
803
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
804
                }
805
                if(IS_INTER(left_type[0])){
806
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
807
                    *(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]];
808
                    *(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]];
809
                }else{
810
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
811
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
812
                }
813
                if(IS_INTER(left_type[1])){
814
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
815
                    *(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]];
816
                    *(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]];
817
                }else{
818
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
819
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
820
                }
821
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
822
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
823 3b66c4c5 Kevin Baragona
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
824 9e528114 Laurent Aimar
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
825
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
826 5ad984c9 Loren Merritt
827
                if(h->slice_type == B_TYPE){
828
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
829
830
                    if(IS_DIRECT(top_type)){
831
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
832
                    }else if(IS_8X8(top_type)){
833
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
834
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
835
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
836
                    }else{
837
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
838
                    }
839
                    
840
                    //FIXME interlacing
841
                    if(IS_DIRECT(left_type[0])){
842
                        h->direct_cache[scan8[0] - 1 + 0*8]=
843
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
844
                    }else if(IS_8X8(left_type[0])){
845
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
846
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
847
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
848
                    }else{
849
                        h->direct_cache[scan8[0] - 1 + 0*8]=
850
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
851
                    }
852
                }
853 9e528114 Laurent Aimar
            }
854 0da71265 Michael Niedermayer
        }
855
    }
856
#endif
857 43efd19a Loren Merritt
858
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
859 0da71265 Michael Niedermayer
}
860
861
static inline void write_back_intra_pred_mode(H264Context *h){
862
    MpegEncContext * const s = &h->s;
863 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
864 0da71265 Michael Niedermayer
865
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
866
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
867
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
868
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
869
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
870
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
871
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
872
}
873
874
/**
875
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
876
 */
877
static inline int check_intra4x4_pred_mode(H264Context *h){
878
    MpegEncContext * const s = &h->s;
879
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
880
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
881
    int i;
882
    
883
    if(!(h->top_samples_available&0x8000)){
884
        for(i=0; i<4; i++){
885
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
886
            if(status<0){
887 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);
888 0da71265 Michael Niedermayer
                return -1;
889
            } else if(status){
890
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
891
            }
892
        }
893
    }
894
    
895
    if(!(h->left_samples_available&0x8000)){
896
        for(i=0; i<4; i++){
897
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
898
            if(status<0){
899 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);
900 0da71265 Michael Niedermayer
                return -1;
901
            } else if(status){
902
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
903
            }
904
        }
905
    }
906
907
    return 0;
908
} //FIXME cleanup like next
909
910
/**
911
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
912
 */
913
static inline int check_intra_pred_mode(H264Context *h, int mode){
914
    MpegEncContext * const s = &h->s;
915
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
916
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
917
    
918 5175b937 Loic Le Loarer
    if(mode < 0 || mode > 6) {
919
        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);
920 7440fe83 Michael Niedermayer
        return -1;
921 5175b937 Loic Le Loarer
    }
922 7440fe83 Michael Niedermayer
    
923 0da71265 Michael Niedermayer
    if(!(h->top_samples_available&0x8000)){
924
        mode= top[ mode ];
925
        if(mode<0){
926 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);
927 0da71265 Michael Niedermayer
            return -1;
928
        }
929
    }
930
    
931
    if(!(h->left_samples_available&0x8000)){
932
        mode= left[ mode ];
933
        if(mode<0){
934 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);
935 0da71265 Michael Niedermayer
            return -1;
936
        } 
937
    }
938
939
    return mode;
940
}
941
942
/**
943
 * gets the predicted intra4x4 prediction mode.
944
 */
945
static inline int pred_intra_mode(H264Context *h, int n){
946
    const int index8= scan8[n];
947
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
948
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
949
    const int min= FFMIN(left, top);
950
951 95c26348 Michael Niedermayer
    tprintf("mode:%d %d min:%d\n", left ,top, min);
952 0da71265 Michael Niedermayer
953
    if(min<0) return DC_PRED;
954
    else      return min;
955
}
956
957
static inline void write_back_non_zero_count(H264Context *h){
958
    MpegEncContext * const s = &h->s;
959 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
960 0da71265 Michael Niedermayer
961 6867a90b Loic Le Loarer
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
962
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
963
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
964 53c05b1e Michael Niedermayer
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
965 6867a90b Loic Le Loarer
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
966
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
967
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
968 53c05b1e Michael Niedermayer
    
969 6867a90b Loic Le Loarer
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
970 53c05b1e Michael Niedermayer
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
971 6867a90b Loic Le Loarer
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
972 53c05b1e Michael Niedermayer
973 6867a90b Loic Le Loarer
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
974 53c05b1e Michael Niedermayer
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
975 6867a90b Loic Le Loarer
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
976 0da71265 Michael Niedermayer
}
977
978
/**
979
 * gets the predicted number of non zero coefficients.
980
 * @param n block index
981
 */
982
static inline int pred_non_zero_count(H264Context *h, int n){
983
    const int index8= scan8[n];
984
    const int left= h->non_zero_count_cache[index8 - 1];
985
    const int top = h->non_zero_count_cache[index8 - 8];
986
    int i= left + top;
987
    
988
    if(i<64) i= (i+1)>>1;
989
990 95c26348 Michael Niedermayer
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
991 0da71265 Michael Niedermayer
992
    return i&31;
993
}
994
995 1924f3ce Michael Niedermayer
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
996
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
997
998
    if(topright_ref != PART_NOT_AVAILABLE){
999
        *C= h->mv_cache[list][ i - 8 + part_width ];
1000
        return topright_ref;
1001
    }else{
1002 95c26348 Michael Niedermayer
        tprintf("topright MV not available\n");
1003
1004 1924f3ce Michael Niedermayer
        *C= h->mv_cache[list][ i - 8 - 1 ];
1005
        return h->ref_cache[list][ i - 8 - 1 ];
1006
    }
1007
}
1008
1009 0da71265 Michael Niedermayer
/**
1010
 * gets the predicted MV.
1011
 * @param n the block index
1012
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1013
 * @param mx the x component of the predicted motion vector
1014
 * @param my the y component of the predicted motion vector
1015
 */
1016
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1017
    const int index8= scan8[n];
1018
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
1019
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
1020
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1021
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1022 1924f3ce Michael Niedermayer
    const int16_t * C;
1023
    int diagonal_ref, match_count;
1024
1025 0da71265 Michael Niedermayer
    assert(part_width==1 || part_width==2 || part_width==4);
1026 1924f3ce Michael Niedermayer
1027 0da71265 Michael Niedermayer
/* mv_cache
1028
  B . . A T T T T 
1029
  U . . L . . , .
1030
  U . . L . . . .
1031
  U . . L . . , .
1032
  . . . L . . . .
1033
*/
1034 1924f3ce Michael Niedermayer
1035
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1036
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1037 827c91bf Loic Le Loarer
    tprintf("pred_motion match_count=%d\n", match_count);
1038 1924f3ce Michael Niedermayer
    if(match_count > 1){ //most common
1039
        *mx= mid_pred(A[0], B[0], C[0]);
1040
        *my= mid_pred(A[1], B[1], C[1]);
1041
    }else if(match_count==1){
1042
        if(left_ref==ref){
1043
            *mx= A[0];
1044
            *my= A[1];        
1045
        }else if(top_ref==ref){
1046
            *mx= B[0];
1047
            *my= B[1];        
1048 0da71265 Michael Niedermayer
        }else{
1049 1924f3ce Michael Niedermayer
            *mx= C[0];
1050
            *my= C[1];        
1051 0da71265 Michael Niedermayer
        }
1052
    }else{
1053 1924f3ce Michael Niedermayer
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1054 0da71265 Michael Niedermayer
            *mx= A[0];
1055 1924f3ce Michael Niedermayer
            *my= A[1];        
1056 0da71265 Michael Niedermayer
        }else{
1057 1924f3ce Michael Niedermayer
            *mx= mid_pred(A[0], B[0], C[0]);
1058
            *my= mid_pred(A[1], B[1], C[1]);
1059 0da71265 Michael Niedermayer
        }
1060
    }
1061 1924f3ce Michael Niedermayer
        
1062 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);
1063 0da71265 Michael Niedermayer
}
1064
1065
/**
1066
 * gets the directionally predicted 16x8 MV.
1067
 * @param n the block index
1068
 * @param mx the x component of the predicted motion vector
1069
 * @param my the y component of the predicted motion vector
1070
 */
1071
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1072
    if(n==0){
1073
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
1074
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1075
1076 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);
1077 0da71265 Michael Niedermayer
        
1078
        if(top_ref == ref){
1079
            *mx= B[0];
1080
            *my= B[1];
1081
            return;
1082
        }
1083
    }else{
1084
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
1085
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1086
        
1087 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);
1088 0da71265 Michael Niedermayer
1089
        if(left_ref == ref){
1090
            *mx= A[0];
1091
            *my= A[1];
1092
            return;
1093
        }
1094
    }
1095
1096
    //RARE
1097
    pred_motion(h, n, 4, list, ref, mx, my);
1098
}
1099
1100
/**
1101
 * gets the directionally predicted 8x16 MV.
1102
 * @param n the block index
1103
 * @param mx the x component of the predicted motion vector
1104
 * @param my the y component of the predicted motion vector
1105
 */
1106
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1107
    if(n==0){
1108
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
1109
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1110
        
1111 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);
1112 0da71265 Michael Niedermayer
1113
        if(left_ref == ref){
1114
            *mx= A[0];
1115
            *my= A[1];
1116
            return;
1117
        }
1118
    }else{
1119 1924f3ce Michael Niedermayer
        const int16_t * C;
1120
        int diagonal_ref;
1121
1122
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1123 0da71265 Michael Niedermayer
        
1124 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);
1125 0da71265 Michael Niedermayer
1126 1924f3ce Michael Niedermayer
        if(diagonal_ref == ref){ 
1127 0da71265 Michael Niedermayer
            *mx= C[0];
1128
            *my= C[1];
1129
            return;
1130
        }
1131
    }
1132
1133
    //RARE
1134
    pred_motion(h, n, 2, list, ref, mx, my);
1135
}
1136
1137
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1138
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1139
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1140
1141 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);
1142 0da71265 Michael Niedermayer
1143
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1144
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1145
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1146
       
1147
        *mx = *my = 0;
1148
        return;
1149
    }
1150
        
1151
    pred_motion(h, 0, 4, 0, 0, mx, my);
1152
1153
    return;
1154
}
1155
1156 5ad984c9 Loren Merritt
static inline void direct_dist_scale_factor(H264Context * const h){
1157
    const int poc = h->s.current_picture_ptr->poc;
1158
    const int poc1 = h->ref_list[1][0].poc;
1159
    int i;
1160
    for(i=0; i<h->ref_count[0]; i++){
1161
        int poc0 = h->ref_list[0][i].poc;
1162
        int td = clip(poc1 - poc0, -128, 127);
1163
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1164
            h->dist_scale_factor[i] = 256;
1165
        }else{
1166
            int tb = clip(poc - poc0, -128, 127);
1167
            int tx = (16384 + (ABS(td) >> 1)) / td;
1168
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1169
        }
1170
    }
1171
}
1172 2f944356 Loren Merritt
static inline void direct_ref_list_init(H264Context * const h){
1173
    MpegEncContext * const s = &h->s;
1174
    Picture * const ref1 = &h->ref_list[1][0];
1175
    Picture * const cur = s->current_picture_ptr;
1176
    int list, i, j;
1177
    if(cur->pict_type == I_TYPE)
1178
        cur->ref_count[0] = 0;
1179
    if(cur->pict_type != B_TYPE)
1180
        cur->ref_count[1] = 0;
1181
    for(list=0; list<2; list++){
1182
        cur->ref_count[list] = h->ref_count[list];
1183
        for(j=0; j<h->ref_count[list]; j++)
1184
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1185
    }
1186
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1187
        return;
1188
    for(list=0; list<2; list++){
1189
        for(i=0; i<ref1->ref_count[list]; i++){
1190
            const int poc = ref1->ref_poc[list][i];
1191
            h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1192
            for(j=0; j<h->ref_count[list]; j++)
1193
                if(h->ref_list[list][j].poc == poc){
1194
                    h->map_col_to_list0[list][i] = j;
1195
                    break;
1196
                }
1197
        }
1198
    }
1199
}
1200 5ad984c9 Loren Merritt
1201
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1202
    MpegEncContext * const s = &h->s;
1203
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1204
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1205
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1206
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1207
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1208
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1209 2f944356 Loren Merritt
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1210 5ad984c9 Loren Merritt
    const int is_b8x8 = IS_8X8(*mb_type);
1211
    int sub_mb_type;
1212
    int i8, i4;
1213
1214
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1215
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1216
         * so we know exactly what block size to use */
1217
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1218 3622988f Loren Merritt
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1219 5ad984c9 Loren Merritt
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1220
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1221
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1222
    }else{
1223
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1224 3622988f Loren Merritt
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1225 5ad984c9 Loren Merritt
    }
1226
    if(!is_b8x8)
1227
        *mb_type |= MB_TYPE_DIRECT2;
1228
1229 827c91bf Loic Le Loarer
    tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1230
    
1231 5ad984c9 Loren Merritt
    if(h->direct_spatial_mv_pred){
1232
        int ref[2];
1233
        int mv[2][2];
1234
        int list;
1235
1236
        /* ref = min(neighbors) */
1237
        for(list=0; list<2; list++){
1238
            int refa = h->ref_cache[list][scan8[0] - 1];
1239
            int refb = h->ref_cache[list][scan8[0] - 8];
1240
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1241
            if(refc == -2)
1242
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1243
            ref[list] = refa;
1244
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1245
                ref[list] = refb;
1246
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1247
                ref[list] = refc;
1248
            if(ref[list] < 0)
1249
                ref[list] = -1;
1250
        }
1251
1252
        if(ref[0] < 0 && ref[1] < 0){
1253
            ref[0] = ref[1] = 0;
1254
            mv[0][0] = mv[0][1] =
1255
            mv[1][0] = mv[1][1] = 0;
1256
        }else{
1257
            for(list=0; list<2; list++){
1258
                if(ref[list] >= 0)
1259
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1260
                else
1261
                    mv[list][0] = mv[list][1] = 0;
1262
            }
1263
        }
1264
1265
        if(ref[1] < 0){
1266
            *mb_type &= ~MB_TYPE_P0L1;
1267
            sub_mb_type &= ~MB_TYPE_P0L1;
1268
        }else if(ref[0] < 0){
1269
            *mb_type &= ~MB_TYPE_P0L0;
1270
            sub_mb_type &= ~MB_TYPE_P0L0;
1271
        }
1272
1273
        if(IS_16X16(*mb_type)){
1274
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1275
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1276
            if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1277
                ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1278
                if(ref[0] > 0)
1279
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1280
                else
1281
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1282
                if(ref[1] > 0)
1283
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1284
                else
1285
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1286
            }else{
1287
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1288
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1289
            }
1290
        }else{
1291
            for(i8=0; i8<4; i8++){
1292
                const int x8 = i8&1;
1293
                const int y8 = i8>>1;
1294
    
1295
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1296
                    continue;
1297
                h->sub_mb_type[i8] = sub_mb_type;
1298
    
1299
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1300
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1301
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1302
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1303
    
1304
                /* col_zero_flag */
1305
                if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1306
                    for(i4=0; i4<4; i4++){
1307
                        const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1308
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1309
                            if(ref[0] == 0)
1310
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1311
                            if(ref[1] == 0)
1312
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1313
                        }
1314
                    }
1315
                }
1316
            }
1317
        }
1318
    }else{ /* direct temporal mv pred */
1319
        if(IS_16X16(*mb_type)){
1320
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1321
            if(IS_INTRA(mb_type_col)){
1322
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1323
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1324
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1325
            }else{
1326 2f944356 Loren Merritt
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1327
                                                : h->map_col_to_list0[1][l1ref1[0]];
1328 5ad984c9 Loren Merritt
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1329
                const int16_t *mv_col = l1mv0[0];
1330
                int mv_l0[2];
1331
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1332
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1333
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1334
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1335
                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);
1336
            }
1337
        }else{
1338
            for(i8=0; i8<4; i8++){
1339
                const int x8 = i8&1;
1340
                const int y8 = i8>>1;
1341
                int ref0, dist_scale_factor;
1342
    
1343
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1344
                    continue;
1345
                h->sub_mb_type[i8] = sub_mb_type;
1346
                if(IS_INTRA(mb_type_col)){
1347
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1348
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1349
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1350
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1351
                    continue;
1352
                }
1353
    
1354
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1355 2f944356 Loren Merritt
                if(ref0 >= 0)
1356
                    ref0 = h->map_col_to_list0[0][ref0];
1357
                else
1358
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1359 5ad984c9 Loren Merritt
                dist_scale_factor = h->dist_scale_factor[ref0];
1360
    
1361
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1362
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1363
                for(i4=0; i4<4; i4++){
1364
                    const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1365
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1366
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1367
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1368
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1369
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1370
                }
1371
            }
1372
        }
1373
    }
1374
}
1375
1376 0da71265 Michael Niedermayer
static inline void write_back_motion(H264Context *h, int mb_type){
1377
    MpegEncContext * const s = &h->s;
1378
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1379
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1380
    int list;
1381
1382
    for(list=0; list<2; list++){
1383
        int y;
1384 b75e10d5 Loren Merritt
        if(!USES_LIST(mb_type, list)){
1385 3b66c4c5 Kevin Baragona
            if(1){ //FIXME skip or never read if mb_type doesn't use it
1386 0da71265 Michael Niedermayer
                for(y=0; y<4; y++){
1387
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1388
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1389
                }
1390 9e528114 Laurent Aimar
                if( h->pps.cabac ) {
1391
                    /* FIXME needed ? */
1392
                    for(y=0; y<4; y++){
1393
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1394
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1395
                    }
1396
                }
1397 0da71265 Michael Niedermayer
                for(y=0; y<2; y++){
1398 0444b023 Michael Niedermayer
                    s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1399
                    s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1400 0da71265 Michael Niedermayer
                }
1401
            }
1402 5ad984c9 Loren Merritt
            continue;
1403 0da71265 Michael Niedermayer
        }
1404
        
1405
        for(y=0; y<4; y++){
1406
            *(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];
1407
            *(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];
1408
        }
1409 9e528114 Laurent Aimar
        if( h->pps.cabac ) {
1410
            for(y=0; y<4; y++){
1411
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1412
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1413
            }
1414
        }
1415 0da71265 Michael Niedermayer
        for(y=0; y<2; y++){
1416
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1417
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1418
        }
1419
    }
1420 5ad984c9 Loren Merritt
    
1421
    if(h->slice_type == B_TYPE && h->pps.cabac){
1422
        if(IS_8X8(mb_type)){
1423
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1424
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1425
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1426
        }
1427
    }
1428 0da71265 Michael Niedermayer
}
1429
1430
/**
1431
 * Decodes a network abstraction layer unit.
1432
 * @param consumed is the number of bytes used as input
1433
 * @param length is the length of the array
1434 3b66c4c5 Kevin Baragona
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1435 0da71265 Michael Niedermayer
 * @returns decoded bytes, might be src+1 if no escapes 
1436
 */
1437
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1438
    int i, si, di;
1439
    uint8_t *dst;
1440
1441
//    src[0]&0x80;                //forbidden bit
1442
    h->nal_ref_idc= src[0]>>5;
1443
    h->nal_unit_type= src[0]&0x1F;
1444
1445
    src++; length--;
1446
#if 0    
1447
    for(i=0; i<length; i++)
1448
        printf("%2X ", src[i]);
1449
#endif
1450
    for(i=0; i+1<length; i+=2){
1451
        if(src[i]) continue;
1452
        if(i>0 && src[i-1]==0) i--;
1453
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1454
            if(src[i+2]!=3){
1455
                /* startcode, so we must be past the end */
1456
                length=i;
1457
            }
1458
            break;
1459
        }
1460
    }
1461
1462
    if(i>=length-1){ //no escaped 0
1463
        *dst_length= length;
1464
        *consumed= length+1; //+1 for the header
1465
        return src; 
1466
    }
1467
1468
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1469
    dst= h->rbsp_buffer;
1470
1471 3b66c4c5 Kevin Baragona
//printf("decoding esc\n");
1472 0da71265 Michael Niedermayer
    si=di=0;
1473
    while(si<length){ 
1474
        //remove escapes (very rare 1:2^22)
1475
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1476
            if(src[si+2]==3){ //escape
1477
                dst[di++]= 0;
1478
                dst[di++]= 0;
1479
                si+=3;
1480 c8470cc1 Michael Niedermayer
                continue;
1481 0da71265 Michael Niedermayer
            }else //next start code
1482
                break;
1483
        }
1484
1485
        dst[di++]= src[si++];
1486
    }
1487
1488
    *dst_length= di;
1489
    *consumed= si + 1;//+1 for the header
1490
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1491
    return dst;
1492
}
1493
1494 e5017ab8 Laurent Aimar
#if 0
1495 0da71265 Michael Niedermayer
/**
1496
 * @param src the data which should be escaped
1497
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1498
 * @param length the length of the src data
1499
 * @param dst_length the length of the dst array
1500
 * @returns length of escaped data in bytes or -1 if an error occured
1501
 */
1502
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1503
    int i, escape_count, si, di;
1504
    uint8_t *temp;
1505
    
1506
    assert(length>=0);
1507
    assert(dst_length>0);
1508
    
1509
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1510

1511
    if(length==0) return 1;
1512

1513
    escape_count= 0;
1514
    for(i=0; i<length; i+=2){
1515
        if(src[i]) continue;
1516
        if(i>0 && src[i-1]==0) 
1517
            i--;
1518
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1519
            escape_count++;
1520
            i+=2;
1521
        }
1522
    }
1523
    
1524
    if(escape_count==0){ 
1525
        if(dst+1 != src)
1526
            memcpy(dst+1, src, length);
1527
        return length + 1;
1528
    }
1529
    
1530
    if(length + escape_count + 1> dst_length)
1531
        return -1;
1532

1533
    //this should be damn rare (hopefully)
1534

1535
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1536
    temp= h->rbsp_buffer;
1537
//printf("encoding esc\n");
1538
    
1539
    si= 0;
1540
    di= 0;
1541
    while(si < length){
1542
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1543
            temp[di++]= 0; si++;
1544
            temp[di++]= 0; si++;
1545
            temp[di++]= 3; 
1546
            temp[di++]= src[si++];
1547
        }
1548
        else
1549
            temp[di++]= src[si++];
1550
    }
1551
    memcpy(dst+1, temp, length+escape_count);
1552
    
1553
    assert(di == length+escape_count);
1554
    
1555
    return di + 1;
1556
}
1557

1558
/**
1559
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1560
 */
1561
static void encode_rbsp_trailing(PutBitContext *pb){
1562
    int length;
1563
    put_bits(pb, 1, 1);
1564 fe455f33 Alex Beregszaszi
    length= (-put_bits_count(pb))&7;
1565 0da71265 Michael Niedermayer
    if(length) put_bits(pb, length, 0);
1566
}
1567 e5017ab8 Laurent Aimar
#endif
1568 0da71265 Michael Niedermayer
1569
/**
1570
 * identifies the exact end of the bitstream
1571
 * @return the length of the trailing, or 0 if damaged
1572
 */
1573
static int decode_rbsp_trailing(uint8_t *src){
1574
    int v= *src;
1575
    int r;
1576
1577 95c26348 Michael Niedermayer
    tprintf("rbsp trailing %X\n", v);
1578 0da71265 Michael Niedermayer
1579
    for(r=1; r<9; r++){
1580
        if(v&1) return r;
1581
        v>>=1;
1582
    }
1583
    return 0;
1584
}
1585
1586
/**
1587
 * idct tranforms the 16 dc values and dequantize them.
1588
 * @param qp quantization parameter
1589
 */
1590
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1591
    const int qmul= dequant_coeff[qp][0];
1592
#define stride 16
1593
    int i;
1594
    int temp[16]; //FIXME check if this is a good idea
1595
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1596
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1597
1598
//memset(block, 64, 2*256);
1599
//return;
1600
    for(i=0; i<4; i++){
1601
        const int offset= y_offset[i];
1602
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1603
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1604
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1605
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1606
1607
        temp[4*i+0]= z0+z3;
1608
        temp[4*i+1]= z1+z2;
1609
        temp[4*i+2]= z1-z2;
1610
        temp[4*i+3]= z0-z3;
1611
    }
1612
1613
    for(i=0; i<4; i++){
1614
        const int offset= x_offset[i];
1615
        const int z0= temp[4*0+i] + temp[4*2+i];
1616
        const int z1= temp[4*0+i] - temp[4*2+i];
1617
        const int z2= temp[4*1+i] - temp[4*3+i];
1618
        const int z3= temp[4*1+i] + temp[4*3+i];
1619
1620
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1621
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1622
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1623
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1624
    }
1625
}
1626
1627 e5017ab8 Laurent Aimar
#if 0
1628 0da71265 Michael Niedermayer
/**
1629
 * dct tranforms the 16 dc values.
1630
 * @param qp quantization parameter ??? FIXME
1631
 */
1632
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1633
//    const int qmul= dequant_coeff[qp][0];
1634
    int i;
1635
    int temp[16]; //FIXME check if this is a good idea
1636
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1637
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1638

1639
    for(i=0; i<4; i++){
1640
        const int offset= y_offset[i];
1641
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1642
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1643
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1644
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1645

1646
        temp[4*i+0]= z0+z3;
1647
        temp[4*i+1]= z1+z2;
1648
        temp[4*i+2]= z1-z2;
1649
        temp[4*i+3]= z0-z3;
1650
    }
1651

1652
    for(i=0; i<4; i++){
1653
        const int offset= x_offset[i];
1654
        const int z0= temp[4*0+i] + temp[4*2+i];
1655
        const int z1= temp[4*0+i] - temp[4*2+i];
1656
        const int z2= temp[4*1+i] - temp[4*3+i];
1657
        const int z3= temp[4*1+i] + temp[4*3+i];
1658

1659
        block[stride*0 +offset]= (z0 + z3)>>1;
1660
        block[stride*2 +offset]= (z1 + z2)>>1;
1661
        block[stride*8 +offset]= (z1 - z2)>>1;
1662
        block[stride*10+offset]= (z0 - z3)>>1;
1663
    }
1664
}
1665 e5017ab8 Laurent Aimar
#endif
1666
1667 0da71265 Michael Niedermayer
#undef xStride
1668
#undef stride
1669
1670
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1671
    const int qmul= dequant_coeff[qp][0];
1672
    const int stride= 16*2;
1673
    const int xStride= 16;
1674
    int a,b,c,d,e;
1675
1676
    a= block[stride*0 + xStride*0];
1677
    b= block[stride*0 + xStride*1];
1678
    c= block[stride*1 + xStride*0];
1679
    d= block[stride*1 + xStride*1];
1680
1681
    e= a-b;
1682
    a= a+b;
1683
    b= c-d;
1684
    c= c+d;
1685
1686
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1687
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1688
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1689
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1690
}
1691
1692 e5017ab8 Laurent Aimar
#if 0
1693 0da71265 Michael Niedermayer
static void chroma_dc_dct_c(DCTELEM *block){
1694
    const int stride= 16*2;
1695
    const int xStride= 16;
1696
    int a,b,c,d,e;
1697

1698
    a= block[stride*0 + xStride*0];
1699
    b= block[stride*0 + xStride*1];
1700
    c= block[stride*1 + xStride*0];
1701
    d= block[stride*1 + xStride*1];
1702

1703
    e= a-b;
1704
    a= a+b;
1705
    b= c-d;
1706
    c= c+d;
1707

1708
    block[stride*0 + xStride*0]= (a+c);
1709
    block[stride*0 + xStride*1]= (e+b);
1710
    block[stride*1 + xStride*0]= (a-c);
1711
    block[stride*1 + xStride*1]= (e-b);
1712
}
1713 e5017ab8 Laurent Aimar
#endif
1714 0da71265 Michael Niedermayer
1715
/**
1716
 * gets the chroma qp.
1717
 */
1718 6ba71fc4 Loïc Le Loarer
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1719 0da71265 Michael Niedermayer
    
1720 6ba71fc4 Loïc Le Loarer
    return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1721 0da71265 Michael Niedermayer
}
1722
1723
1724 e5017ab8 Laurent Aimar
#if 0
1725 0da71265 Michael Niedermayer
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1726
    int i;
1727
    //FIXME try int temp instead of block
1728
    
1729
    for(i=0; i<4; i++){
1730
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1731
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1732
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1733
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1734
        const int z0= d0 + d3;
1735
        const int z3= d0 - d3;
1736
        const int z1= d1 + d2;
1737
        const int z2= d1 - d2;
1738
        
1739
        block[0 + 4*i]=   z0 +   z1;
1740
        block[1 + 4*i]= 2*z3 +   z2;
1741
        block[2 + 4*i]=   z0 -   z1;
1742
        block[3 + 4*i]=   z3 - 2*z2;
1743
    }    
1744

1745
    for(i=0; i<4; i++){
1746
        const int z0= block[0*4 + i] + block[3*4 + i];
1747
        const int z3= block[0*4 + i] - block[3*4 + i];
1748
        const int z1= block[1*4 + i] + block[2*4 + i];
1749
        const int z2= block[1*4 + i] - block[2*4 + i];
1750
        
1751
        block[0*4 + i]=   z0 +   z1;
1752
        block[1*4 + i]= 2*z3 +   z2;
1753
        block[2*4 + i]=   z0 -   z1;
1754
        block[3*4 + i]=   z3 - 2*z2;
1755
    }
1756
}
1757 e5017ab8 Laurent Aimar
#endif
1758 0da71265 Michael Niedermayer
1759 3b66c4c5 Kevin Baragona
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1760 0da71265 Michael Niedermayer
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1761
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1762
    int i;
1763
    const int * const quant_table= quant_coeff[qscale];
1764
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1765
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1766
    const unsigned int threshold2= (threshold1<<1);
1767
    int last_non_zero;
1768
1769
    if(seperate_dc){
1770
        if(qscale<=18){
1771
            //avoid overflows
1772
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1773
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1774
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1775
1776
            int level= block[0]*quant_coeff[qscale+18][0];
1777
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1778
                if(level>0){
1779
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1780
                    block[0]= level;
1781
                }else{
1782
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1783
                    block[0]= -level;
1784
                }
1785
//                last_non_zero = i;
1786
            }else{
1787
                block[0]=0;
1788
            }
1789
        }else{
1790
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1791
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1792
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1793
1794
            int level= block[0]*quant_table[0];
1795
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1796
                if(level>0){
1797
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1798
                    block[0]= level;
1799
                }else{
1800
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1801
                    block[0]= -level;
1802
                }
1803
//                last_non_zero = i;
1804
            }else{
1805
                block[0]=0;
1806
            }
1807
        }
1808
        last_non_zero= 0;
1809
        i=1;
1810
    }else{
1811
        last_non_zero= -1;
1812
        i=0;
1813
    }
1814
1815
    for(; i<16; i++){
1816
        const int j= scantable[i];
1817
        int level= block[j]*quant_table[j];
1818
1819
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1820
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1821
        if(((unsigned)(level+threshold1))>threshold2){
1822
            if(level>0){
1823
                level= (bias + level)>>QUANT_SHIFT;
1824
                block[j]= level;
1825
            }else{
1826
                level= (bias - level)>>QUANT_SHIFT;
1827
                block[j]= -level;
1828
            }
1829
            last_non_zero = i;
1830
        }else{
1831
            block[j]=0;
1832
        }
1833
    }
1834
1835
    return last_non_zero;
1836
}
1837
1838
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1839
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1840
    ((uint32_t*)(src+0*stride))[0]= a;
1841
    ((uint32_t*)(src+1*stride))[0]= a;
1842
    ((uint32_t*)(src+2*stride))[0]= a;
1843
    ((uint32_t*)(src+3*stride))[0]= a;
1844
}
1845
1846
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1847
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1848
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1849
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1850
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1851
}
1852
1853
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1854
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1855
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1856
    
1857
    ((uint32_t*)(src+0*stride))[0]= 
1858
    ((uint32_t*)(src+1*stride))[0]= 
1859
    ((uint32_t*)(src+2*stride))[0]= 
1860
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1861
}
1862
1863
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1864
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1865
    
1866
    ((uint32_t*)(src+0*stride))[0]= 
1867
    ((uint32_t*)(src+1*stride))[0]= 
1868
    ((uint32_t*)(src+2*stride))[0]= 
1869
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1870
}
1871
1872
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1873
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1874
    
1875
    ((uint32_t*)(src+0*stride))[0]= 
1876
    ((uint32_t*)(src+1*stride))[0]= 
1877
    ((uint32_t*)(src+2*stride))[0]= 
1878
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1879
}
1880
1881
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1882
    ((uint32_t*)(src+0*stride))[0]= 
1883
    ((uint32_t*)(src+1*stride))[0]= 
1884
    ((uint32_t*)(src+2*stride))[0]= 
1885
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1886
}
1887
1888
1889
#define LOAD_TOP_RIGHT_EDGE\
1890
    const int t4= topright[0];\
1891
    const int t5= topright[1];\
1892
    const int t6= topright[2];\
1893
    const int t7= topright[3];\
1894
1895
#define LOAD_LEFT_EDGE\
1896
    const int l0= src[-1+0*stride];\
1897
    const int l1= src[-1+1*stride];\
1898
    const int l2= src[-1+2*stride];\
1899
    const int l3= src[-1+3*stride];\
1900
1901
#define LOAD_TOP_EDGE\
1902
    const int t0= src[ 0-1*stride];\
1903
    const int t1= src[ 1-1*stride];\
1904
    const int t2= src[ 2-1*stride];\
1905
    const int t3= src[ 3-1*stride];\
1906
1907
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1908
    const int lt= src[-1-1*stride];
1909
    LOAD_TOP_EDGE
1910
    LOAD_LEFT_EDGE
1911
1912
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1913
    src[0+2*stride]=
1914
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1915
    src[0+1*stride]=
1916
    src[1+2*stride]=
1917
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1918
    src[0+0*stride]=
1919
    src[1+1*stride]=
1920
    src[2+2*stride]=
1921
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1922
    src[1+0*stride]=
1923
    src[2+1*stride]=
1924
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1925
    src[2+0*stride]=
1926
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1927
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1928 4cfbf61b Falk Hüffner
}
1929 0da71265 Michael Niedermayer
1930
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1931
    LOAD_TOP_EDGE    
1932
    LOAD_TOP_RIGHT_EDGE    
1933
//    LOAD_LEFT_EDGE    
1934
1935
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1936
    src[1+0*stride]=
1937
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1938
    src[2+0*stride]=
1939
    src[1+1*stride]=
1940
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1941
    src[3+0*stride]=
1942
    src[2+1*stride]=
1943
    src[1+2*stride]=
1944
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1945
    src[3+1*stride]=
1946
    src[2+2*stride]=
1947
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1948
    src[3+2*stride]=
1949
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1950
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1951 4cfbf61b Falk Hüffner
}
1952 0da71265 Michael Niedermayer
1953
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1954
    const int lt= src[-1-1*stride];
1955
    LOAD_TOP_EDGE    
1956
    LOAD_LEFT_EDGE    
1957
    const __attribute__((unused)) int unu= l3;
1958
1959
    src[0+0*stride]=
1960
    src[1+2*stride]=(lt + t0 + 1)>>1;
1961
    src[1+0*stride]=
1962
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1963
    src[2+0*stride]=
1964
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1965
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1966
    src[0+1*stride]=
1967
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1968
    src[1+1*stride]=
1969
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1970
    src[2+1*stride]=
1971
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1972
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1973
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1974
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1975 4cfbf61b Falk Hüffner
}
1976 0da71265 Michael Niedermayer
1977
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1978
    LOAD_TOP_EDGE    
1979
    LOAD_TOP_RIGHT_EDGE    
1980
    const __attribute__((unused)) int unu= t7;
1981
1982
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1983
    src[1+0*stride]=
1984
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1985
    src[2+0*stride]=
1986
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1987
    src[3+0*stride]=
1988
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1989
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1990
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1991
    src[1+1*stride]=
1992
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1993
    src[2+1*stride]=
1994
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1995
    src[3+1*stride]=
1996
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1997
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1998 4cfbf61b Falk Hüffner
}
1999 0da71265 Michael Niedermayer
2000
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2001
    LOAD_LEFT_EDGE    
2002
2003
    src[0+0*stride]=(l0 + l1 + 1)>>1;
2004
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2005
    src[2+0*stride]=
2006
    src[0+1*stride]=(l1 + l2 + 1)>>1;
2007
    src[3+0*stride]=
2008
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2009
    src[2+1*stride]=
2010
    src[0+2*stride]=(l2 + l3 + 1)>>1;
2011
    src[3+1*stride]=
2012
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2013
    src[3+2*stride]=
2014
    src[1+3*stride]=
2015
    src[0+3*stride]=
2016
    src[2+2*stride]=
2017
    src[2+3*stride]=
2018
    src[3+3*stride]=l3;
2019 4cfbf61b Falk Hüffner
}
2020 0da71265 Michael Niedermayer
    
2021
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2022
    const int lt= src[-1-1*stride];
2023
    LOAD_TOP_EDGE    
2024
    LOAD_LEFT_EDGE    
2025
    const __attribute__((unused)) int unu= t3;
2026
2027
    src[0+0*stride]=
2028
    src[2+1*stride]=(lt + l0 + 1)>>1;
2029
    src[1+0*stride]=
2030
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2031
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2032
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2033
    src[0+1*stride]=
2034
    src[2+2*stride]=(l0 + l1 + 1)>>1;
2035
    src[1+1*stride]=
2036
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2037
    src[0+2*stride]=
2038
    src[2+3*stride]=(l1 + l2+ 1)>>1;
2039
    src[1+2*stride]=
2040
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2041
    src[0+3*stride]=(l2 + l3 + 1)>>1;
2042
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2043 4cfbf61b Falk Hüffner
}
2044 0da71265 Michael Niedermayer
2045
static void pred16x16_vertical_c(uint8_t *src, int stride){
2046
    int i;
2047
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2048
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2049
    const uint32_t c= ((uint32_t*)(src-stride))[2];
2050
    const uint32_t d= ((uint32_t*)(src-stride))[3];
2051
    
2052
    for(i=0; i<16; i++){
2053
        ((uint32_t*)(src+i*stride))[0]= a;
2054
        ((uint32_t*)(src+i*stride))[1]= b;
2055
        ((uint32_t*)(src+i*stride))[2]= c;
2056
        ((uint32_t*)(src+i*stride))[3]= d;
2057
    }
2058
}
2059
2060
static void pred16x16_horizontal_c(uint8_t *src, int stride){
2061
    int i;
2062
2063
    for(i=0; i<16; i++){
2064
        ((uint32_t*)(src+i*stride))[0]=
2065
        ((uint32_t*)(src+i*stride))[1]=
2066
        ((uint32_t*)(src+i*stride))[2]=
2067
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2068
    }
2069
}
2070
2071
static void pred16x16_dc_c(uint8_t *src, int stride){
2072
    int i, dc=0;
2073
2074
    for(i=0;i<16; i++){
2075
        dc+= src[-1+i*stride];
2076
    }
2077
    
2078
    for(i=0;i<16; i++){
2079
        dc+= src[i-stride];
2080
    }
2081
2082
    dc= 0x01010101*((dc + 16)>>5);
2083
2084
    for(i=0; i<16; i++){
2085
        ((uint32_t*)(src+i*stride))[0]=
2086
        ((uint32_t*)(src+i*stride))[1]=
2087
        ((uint32_t*)(src+i*stride))[2]=
2088
        ((uint32_t*)(src+i*stride))[3]= dc;
2089
    }
2090
}
2091
2092
static void pred16x16_left_dc_c(uint8_t *src, int stride){
2093
    int i, dc=0;
2094
2095
    for(i=0;i<16; i++){
2096
        dc+= src[-1+i*stride];
2097
    }
2098
    
2099
    dc= 0x01010101*((dc + 8)>>4);
2100
2101
    for(i=0; i<16; i++){
2102
        ((uint32_t*)(src+i*stride))[0]=
2103
        ((uint32_t*)(src+i*stride))[1]=
2104
        ((uint32_t*)(src+i*stride))[2]=
2105
        ((uint32_t*)(src+i*stride))[3]= dc;
2106
    }
2107
}
2108
2109
static void pred16x16_top_dc_c(uint8_t *src, int stride){
2110
    int i, dc=0;
2111
2112
    for(i=0;i<16; i++){
2113
        dc+= src[i-stride];
2114
    }
2115
    dc= 0x01010101*((dc + 8)>>4);
2116
2117
    for(i=0; i<16; i++){
2118
        ((uint32_t*)(src+i*stride))[0]=
2119
        ((uint32_t*)(src+i*stride))[1]=
2120
        ((uint32_t*)(src+i*stride))[2]=
2121
        ((uint32_t*)(src+i*stride))[3]= dc;
2122
    }
2123
}
2124
2125
static void pred16x16_128_dc_c(uint8_t *src, int stride){
2126
    int i;
2127
2128
    for(i=0; i<16; i++){
2129
        ((uint32_t*)(src+i*stride))[0]=
2130
        ((uint32_t*)(src+i*stride))[1]=
2131
        ((uint32_t*)(src+i*stride))[2]=
2132
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2133
    }
2134
}
2135
2136 8b82a956 Michael Niedermayer
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2137 30f73fc7 Michael Niedermayer
  int i, j, k;
2138
  int a;
2139
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2140
  const uint8_t * const src0 = src+7-stride;
2141
  const uint8_t *src1 = src+8*stride-1;
2142
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
2143
  int H = src0[1] - src0[-1];
2144
  int V = src1[0] - src2[ 0];
2145
  for(k=2; k<=8; ++k) {
2146
    src1 += stride; src2 -= stride;
2147
    H += k*(src0[k] - src0[-k]);
2148
    V += k*(src1[0] - src2[ 0]);
2149
  }
2150 8b82a956 Michael Niedermayer
  if(svq3){
2151
    H = ( 5*(H/4) ) / 16;
2152
    V = ( 5*(V/4) ) / 16;
2153 2e26c8d2 Mike Melanson
2154
    /* required for 100% accuracy */
2155
    i = H; H = V; V = i;
2156 8b82a956 Michael Niedermayer
  }else{
2157
    H = ( 5*H+32 ) >> 6;
2158
    V = ( 5*V+32 ) >> 6;
2159
  }
2160 30f73fc7 Michael Niedermayer
2161
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2162
  for(j=16; j>0; --j) {
2163
    int b = a;
2164
    a += V;
2165
    for(i=-16; i<0; i+=4) {
2166
      src[16+i] = cm[ (b    ) >> 5 ];
2167
      src[17+i] = cm[ (b+  H) >> 5 ];
2168
      src[18+i] = cm[ (b+2*H) >> 5 ];
2169
      src[19+i] = cm[ (b+3*H) >> 5 ];
2170
      b += 4*H;
2171 0da71265 Michael Niedermayer
    }
2172 30f73fc7 Michael Niedermayer
    src += stride;
2173
  }
2174 0da71265 Michael Niedermayer
}
2175
2176 8b82a956 Michael Niedermayer
static void pred16x16_plane_c(uint8_t *src, int stride){
2177
    pred16x16_plane_compat_c(src, stride, 0);
2178
}
2179
2180 0da71265 Michael Niedermayer
static void pred8x8_vertical_c(uint8_t *src, int stride){
2181
    int i;
2182
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2183
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2184
    
2185
    for(i=0; i<8; i++){
2186
        ((uint32_t*)(src+i*stride))[0]= a;
2187
        ((uint32_t*)(src+i*stride))[1]= b;
2188
    }
2189
}
2190
2191
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2192
    int i;
2193
2194
    for(i=0; i<8; i++){
2195
        ((uint32_t*)(src+i*stride))[0]=
2196
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2197
    }
2198
}
2199
2200
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2201
    int i;
2202
2203 43efd19a Loren Merritt
    for(i=0; i<8; i++){
2204 0da71265 Michael Niedermayer
        ((uint32_t*)(src+i*stride))[0]= 
2205
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2206
    }
2207
}
2208
2209
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2210
    int i;
2211
    int dc0, dc2;
2212
2213
    dc0=dc2=0;
2214
    for(i=0;i<4; i++){
2215
        dc0+= src[-1+i*stride];
2216
        dc2+= src[-1+(i+4)*stride];
2217
    }
2218
    dc0= 0x01010101*((dc0 + 2)>>2);
2219
    dc2= 0x01010101*((dc2 + 2)>>2);
2220
2221
    for(i=0; i<4; i++){
2222
        ((uint32_t*)(src+i*stride))[0]=
2223
        ((uint32_t*)(src+i*stride))[1]= dc0;
2224
    }
2225
    for(i=4; i<8; i++){
2226
        ((uint32_t*)(src+i*stride))[0]=
2227
        ((uint32_t*)(src+i*stride))[1]= dc2;
2228
    }
2229
}
2230
2231
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2232
    int i;
2233
    int dc0, dc1;
2234
2235
    dc0=dc1=0;
2236
    for(i=0;i<4; i++){
2237
        dc0+= src[i-stride];
2238
        dc1+= src[4+i-stride];
2239
    }
2240
    dc0= 0x01010101*((dc0 + 2)>>2);
2241
    dc1= 0x01010101*((dc1 + 2)>>2);
2242
2243
    for(i=0; i<4; i++){
2244
        ((uint32_t*)(src+i*stride))[0]= dc0;
2245
        ((uint32_t*)(src+i*stride))[1]= dc1;
2246
    }
2247
    for(i=4; i<8; i++){
2248
        ((uint32_t*)(src+i*stride))[0]= dc0;
2249
        ((uint32_t*)(src+i*stride))[1]= dc1;
2250
    }
2251
}
2252
2253
2254
static void pred8x8_dc_c(uint8_t *src, int stride){
2255
    int i;
2256
    int dc0, dc1, dc2, dc3;
2257
2258
    dc0=dc1=dc2=0;
2259
    for(i=0;i<4; i++){
2260
        dc0+= src[-1+i*stride] + src[i-stride];
2261
        dc1+= src[4+i-stride];
2262
        dc2+= src[-1+(i+4)*stride];
2263
    }
2264
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2265
    dc0= 0x01010101*((dc0 + 4)>>3);
2266
    dc1= 0x01010101*((dc1 + 2)>>2);
2267
    dc2= 0x01010101*((dc2 + 2)>>2);
2268
2269
    for(i=0; i<4; i++){
2270
        ((uint32_t*)(src+i*stride))[0]= dc0;
2271
        ((uint32_t*)(src+i*stride))[1]= dc1;
2272
    }
2273
    for(i=4; i<8; i++){
2274
        ((uint32_t*)(src+i*stride))[0]= dc2;
2275
        ((uint32_t*)(src+i*stride))[1]= dc3;
2276
    }
2277
}
2278
2279
static void pred8x8_plane_c(uint8_t *src, int stride){
2280 30f73fc7 Michael Niedermayer
  int j, k;
2281
  int a;
2282
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2283
  const uint8_t * const src0 = src+3-stride;
2284
  const uint8_t *src1 = src+4*stride-1;
2285
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2286
  int H = src0[1] - src0[-1];
2287
  int V = src1[0] - src2[ 0];
2288
  for(k=2; k<=4; ++k) {
2289
    src1 += stride; src2 -= stride;
2290
    H += k*(src0[k] - src0[-k]);
2291
    V += k*(src1[0] - src2[ 0]);
2292
  }
2293
  H = ( 17*H+16 ) >> 5;
2294
  V = ( 17*V+16 ) >> 5;
2295
2296
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2297
  for(j=8; j>0; --j) {
2298
    int b = a;
2299
    a += V;
2300
    src[0] = cm[ (b    ) >> 5 ];
2301
    src[1] = cm[ (b+  H) >> 5 ];
2302
    src[2] = cm[ (b+2*H) >> 5 ];
2303
    src[3] = cm[ (b+3*H) >> 5 ];
2304
    src[4] = cm[ (b+4*H) >> 5 ];
2305
    src[5] = cm[ (b+5*H) >> 5 ];
2306
    src[6] = cm[ (b+6*H) >> 5 ];
2307
    src[7] = cm[ (b+7*H) >> 5 ];
2308
    src += stride;
2309
  }
2310 0da71265 Michael Niedermayer
}
2311
2312 43efd19a Loren Merritt
#define SRC(x,y) src[(x)+(y)*stride]
2313
#define PL(y) \
2314
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2315
#define PREDICT_8x8_LOAD_LEFT \
2316
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2317
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2318
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2319 fb3ca6ba Måns Rullgård
    const int l7 = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2320 43efd19a Loren Merritt
2321
#define PT(x) \
2322
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2323
#define PREDICT_8x8_LOAD_TOP \
2324
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2325
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2326
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2327
    const int t7 = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2328 fb3ca6ba Måns Rullgård
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2329 43efd19a Loren Merritt
2330
#define PTR(x) \
2331
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2332
#define PREDICT_8x8_LOAD_TOPRIGHT \
2333
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2334
    if(has_topright) { \
2335
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2336
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2337
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2338
2339
#define PREDICT_8x8_LOAD_TOPLEFT \
2340 fb3ca6ba Måns Rullgård
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2341 43efd19a Loren Merritt
2342
#define PREDICT_8x8_DC(v) \
2343
    int y; \
2344
    for( y = 0; y < 8; y++ ) { \
2345
        ((uint32_t*)src)[0] = \
2346
        ((uint32_t*)src)[1] = v; \
2347
        src += stride; \
2348
    }
2349
2350
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2351
{
2352
    PREDICT_8x8_DC(0x80808080);
2353
}
2354
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2355
{
2356
    PREDICT_8x8_LOAD_LEFT;
2357
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2358
    PREDICT_8x8_DC(dc);
2359
}
2360
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2361
{
2362
    PREDICT_8x8_LOAD_TOP;
2363
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2364
    PREDICT_8x8_DC(dc);
2365
}
2366
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2367
{
2368
    PREDICT_8x8_LOAD_LEFT;
2369
    PREDICT_8x8_LOAD_TOP;
2370
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2371
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2372
    PREDICT_8x8_DC(dc);
2373
}
2374
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2375
{
2376
    PREDICT_8x8_LOAD_LEFT;
2377
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2378
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2379
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2380
#undef ROW
2381
}
2382
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2383
{
2384
    int y;
2385
    PREDICT_8x8_LOAD_TOP;
2386
    src[0] = t0;
2387
    src[1] = t1;
2388
    src[2] = t2;
2389
    src[3] = t3;
2390
    src[4] = t4;
2391
    src[5] = t5;
2392
    src[6] = t6;
2393
    src[7] = t7;
2394
    for( y = 1; y < 8; y++ )
2395
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2396
}
2397
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2398
{
2399
    PREDICT_8x8_LOAD_TOP;
2400
    PREDICT_8x8_LOAD_TOPRIGHT;
2401
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2402
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2403
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2404
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2405
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2406
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2407
    SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;
2408
    SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;
2409
    SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;
2410
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2411
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2412
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2413
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2414
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2415
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2416
}
2417
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2418
{
2419
    PREDICT_8x8_LOAD_TOP;
2420
    PREDICT_8x8_LOAD_LEFT;
2421
    PREDICT_8x8_LOAD_TOPLEFT;
2422
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2423
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2424
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2425
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2426
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2427
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2428
    SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;
2429
    SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;
2430
    SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;
2431
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2432
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2433
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2434
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2435
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2436
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2437
  
2438
}
2439
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2440
{
2441
    PREDICT_8x8_LOAD_TOP;
2442
    PREDICT_8x8_LOAD_LEFT;
2443
    PREDICT_8x8_LOAD_TOPLEFT;
2444
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2445
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2446
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2447
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2448
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2449
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2450
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2451
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2452
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2453
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2454
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2455
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2456
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2457
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2458
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2459
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2460
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2461
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2462
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2463
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2464
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2465
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2466
}
2467
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2468
{
2469
    PREDICT_8x8_LOAD_TOP;
2470
    PREDICT_8x8_LOAD_LEFT;
2471
    PREDICT_8x8_LOAD_TOPLEFT;
2472
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2473
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2474
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2475
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2476
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2477
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2478
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2479
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2480
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2481
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2482
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2483
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2484
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2485
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2486
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2487
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2488
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2489
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2490
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2491
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2492
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2493
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2494
}
2495
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2496
{
2497
    PREDICT_8x8_LOAD_TOP;
2498
    PREDICT_8x8_LOAD_TOPRIGHT;
2499
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2500
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2501
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2502
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2503
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2504
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2505
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2506
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2507
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2508
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2509
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2510
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2511
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2512
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2513
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2514
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2515
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2516
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2517
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2518
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2519
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2520
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2521
}
2522
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2523
{
2524
    PREDICT_8x8_LOAD_LEFT;
2525
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2526
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2527
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2528
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2529
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2530
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2531
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2532
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2533
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2534
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2535
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2536
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2537
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2538
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2539
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2540
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2541
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2542
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2543
}
2544
#undef PREDICT_8x8_LOAD_LEFT
2545
#undef PREDICT_8x8_LOAD_TOP
2546
#undef PREDICT_8x8_LOAD_TOPLEFT
2547
#undef PREDICT_8x8_LOAD_TOPRIGHT
2548
#undef PREDICT_8x8_DC
2549
#undef PTR
2550
#undef PT
2551
#undef PL
2552
#undef SRC
2553
2554 0da71265 Michael Niedermayer
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2555
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2556
                           int src_x_offset, int src_y_offset,
2557
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2558
    MpegEncContext * const s = &h->s;
2559
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2560
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2561
    const int luma_xy= (mx&3) + ((my&3)<<2);
2562
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2563
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2564
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2565
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2566
    int extra_height= extra_width;
2567
    int emu=0;
2568
    const int full_mx= mx>>2;
2569
    const int full_my= my>>2;
2570
    
2571
    assert(pic->data[0]);
2572
    
2573
    if(mx&7) extra_width -= 3;
2574
    if(my&7) extra_height -= 3;
2575
    
2576
    if(   full_mx < 0-extra_width 
2577
       || full_my < 0-extra_height 
2578
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
2579
       || full_my + 16/*FIXME*/ > s->height + extra_height){
2580 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);
2581 0da71265 Michael Niedermayer
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2582
        emu=1;
2583
    }
2584
    
2585
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2586
    if(!square){
2587
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2588
    }
2589
    
2590
    if(s->flags&CODEC_FLAG_GRAY) return;
2591
    
2592
    if(emu){
2593 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);
2594 0da71265 Michael Niedermayer
            src_cb= s->edge_emu_buffer;
2595
    }
2596
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2597
2598
    if(emu){
2599 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);
2600 0da71265 Michael Niedermayer
            src_cr= s->edge_emu_buffer;
2601
    }
2602
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2603
}
2604
2605 9f2d1b4f Loren Merritt
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2606 0da71265 Michael Niedermayer
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2607
                           int x_offset, int y_offset,
2608
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2609
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2610
                           int list0, int list1){
2611
    MpegEncContext * const s = &h->s;
2612
    qpel_mc_func *qpix_op=  qpix_put;
2613
    h264_chroma_mc_func chroma_op= chroma_put;
2614
    
2615
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2616
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2617
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2618
    x_offset += 8*s->mb_x;
2619
    y_offset += 8*s->mb_y;
2620
    
2621
    if(list0){
2622 1924f3ce Michael Niedermayer
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2623 0da71265 Michael Niedermayer
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2624
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2625
                           qpix_op, chroma_op);
2626
2627
        qpix_op=  qpix_avg;
2628
        chroma_op= chroma_avg;
2629
    }
2630
2631
    if(list1){
2632 1924f3ce Michael Niedermayer
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2633 0da71265 Michael Niedermayer
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2634
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2635
                           qpix_op, chroma_op);
2636
    }
2637
}
2638
2639 9f2d1b4f Loren Merritt
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2640
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2641
                           int x_offset, int y_offset,
2642
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2643
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2644
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2645
                           int list0, int list1){
2646
    MpegEncContext * const s = &h->s;
2647
2648
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2649
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2650
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2651
    x_offset += 8*s->mb_x;
2652
    y_offset += 8*s->mb_y;
2653
    
2654
    if(list0 && list1){
2655
        /* don't optimize for luma-only case, since B-frames usually
2656
         * use implicit weights => chroma too. */
2657
        uint8_t *tmp_cb = s->obmc_scratchpad;
2658
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2659
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2660
        int refn0 = h->ref_cache[0][ scan8[n] ];
2661
        int refn1 = h->ref_cache[1][ scan8[n] ];
2662
2663
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2664
                    dest_y, dest_cb, dest_cr,
2665
                    x_offset, y_offset, qpix_put, chroma_put);
2666
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2667
                    tmp_y, tmp_cb, tmp_cr,
2668
                    x_offset, y_offset, qpix_put, chroma_put);
2669
2670
        if(h->use_weight == 2){
2671
            int weight0 = h->implicit_weight[refn0][refn1];
2672
            int weight1 = 64 - weight0;
2673
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2674
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2675
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2676
        }else{
2677
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2678
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2679
                            h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2680
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2681
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2682
                            h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2683
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2684
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2685
                            h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2686
        }
2687
    }else{
2688
        int list = list1 ? 1 : 0;
2689
        int refn = h->ref_cache[list][ scan8[n] ];
2690
        Picture *ref= &h->ref_list[list][refn];
2691
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2692
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2693
                    qpix_put, chroma_put);
2694
2695
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2696
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2697
        if(h->use_weight_chroma){
2698
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2699
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2700
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2701
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2702
        }
2703
    }
2704
}
2705
2706
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2707
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2708
                           int x_offset, int y_offset,
2709
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2710
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2711
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2712
                           int list0, int list1){
2713
    if((h->use_weight==2 && list0 && list1
2714
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2715
       || h->use_weight==1)
2716
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2717
                         x_offset, y_offset, qpix_put, chroma_put,
2718
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2719
    else
2720
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2721
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2722
}
2723
2724 0da71265 Michael Niedermayer
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2725
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2726 9f2d1b4f Loren Merritt
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2727
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2728 0da71265 Michael Niedermayer
    MpegEncContext * const s = &h->s;
2729 7bc9090a Michael Niedermayer
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2730 0da71265 Michael Niedermayer
    const int mb_type= s->current_picture.mb_type[mb_xy];
2731
    
2732
    assert(IS_INTER(mb_type));
2733
    
2734
    if(IS_16X16(mb_type)){
2735
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2736
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2737 9f2d1b4f Loren Merritt
                &weight_op[0], &weight_avg[0],
2738 0da71265 Michael Niedermayer
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2739
    }else if(IS_16X8(mb_type)){
2740
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2741
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2742 9f2d1b4f Loren Merritt
                &weight_op[1], &weight_avg[1],
2743 0da71265 Michael Niedermayer
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2744
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2745
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2746 9f2d1b4f Loren Merritt
                &weight_op[1], &weight_avg[1],
2747 0da71265 Michael Niedermayer
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2748
    }else if(IS_8X16(mb_type)){
2749
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2750
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2751 9f2d1b4f Loren Merritt
                &weight_op[2], &weight_avg[2],
2752 0da71265 Michael Niedermayer
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2753
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2754
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2755 9f2d1b4f Loren Merritt
                &weight_op[2], &weight_avg[2],
2756 0da71265 Michael Niedermayer
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2757
    }else{
2758
        int i;
2759
        
2760
        assert(IS_8X8(mb_type));
2761
2762
        for(i=0; i<4; i++){
2763
            const int sub_mb_type= h->sub_mb_type[i];
2764
            const int n= 4*i;
2765
            int x_offset= (i&1)<<2;
2766
            int y_offset= (i&2)<<1;
2767
2768
            if(IS_SUB_8X8(sub_mb_type)){
2769
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2770
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2771 9f2d1b4f Loren Merritt
                    &weight_op[3], &weight_avg[3],
2772 0da71265 Michael Niedermayer
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2773
            }else if(IS_SUB_8X4(sub_mb_type)){
2774
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2775
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2776 9f2d1b4f Loren Merritt
                    &weight_op[4], &weight_avg[4],
2777 0da71265 Michael Niedermayer
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2778
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2779
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2780 9f2d1b4f Loren Merritt
                    &weight_op[4], &weight_avg[4],
2781 0da71265 Michael Niedermayer
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2782
            }else if(IS_SUB_4X8(sub_mb_type)){
2783
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2784
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2785 9f2d1b4f Loren Merritt
                    &weight_op[5], &weight_avg[5],
2786 0da71265 Michael Niedermayer
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2787
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2788
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2789 9f2d1b4f Loren Merritt
                    &weight_op[5], &weight_avg[5],
2790 0da71265 Michael Niedermayer
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2791
            }else{
2792
                int j;
2793
                assert(IS_SUB_4X4(sub_mb_type));
2794
                for(j=0; j<4; j++){
2795
                    int sub_x_offset= x_offset + 2*(j&1);
2796
                    int sub_y_offset= y_offset +   (j&2);
2797
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2798
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2799 9f2d1b4f Loren Merritt
                        &weight_op[6], &weight_avg[6],
2800 0da71265 Michael Niedermayer
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2801
                }
2802
            }
2803
        }
2804
    }
2805
}
2806
2807
static void decode_init_vlc(H264Context *h){
2808
    static int done = 0;
2809
2810
    if (!done) {
2811
        int i;
2812
        done = 1;
2813
2814
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2815
                 &chroma_dc_coeff_token_len [0], 1, 1,
2816 073c2593 Burkhard Plaum
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2817 0da71265 Michael Niedermayer
2818
        for(i=0; i<4; i++){
2819
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2820
                     &coeff_token_len [i][0], 1, 1,
2821 073c2593 Burkhard Plaum
                     &coeff_token_bits[i][0], 1, 1, 1);
2822 0da71265 Michael Niedermayer
        }
2823
2824
        for(i=0; i<3; i++){
2825
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2826
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2827 073c2593 Burkhard Plaum
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2828 0da71265 Michael Niedermayer
        }
2829
        for(i=0; i<15; i++){
2830
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2831
                     &total_zeros_len [i][0], 1, 1,
2832 073c2593 Burkhard Plaum
                     &total_zeros_bits[i][0], 1, 1, 1);
2833 0da71265 Michael Niedermayer
        }
2834
2835
        for(i=0; i<6; i++){
2836
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7