Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 042ef4b7

History | View | Annotate | Download (313 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file h264.c
24
 * H.264 / AVC / MPEG4 part10 codec.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#include "dsputil.h"
29
#include "avcodec.h"
30
#include "mpegvideo.h"
31
#include "h264.h"
32
#include "h264data.h"
33
#include "h264_parser.h"
34
#include "golomb.h"
35

    
36
#include "cabac.h"
37

    
38
//#undef NDEBUG
39
#include <assert.h>
40

    
41
static VLC coeff_token_vlc[4];
42
static VLC chroma_dc_coeff_token_vlc;
43

    
44
static VLC total_zeros_vlc[15];
45
static VLC chroma_dc_total_zeros_vlc[3];
46

    
47
static VLC run_vlc[6];
48
static VLC run7_vlc;
49

    
50
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
51
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
52
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);
53
static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
54

    
55
static av_always_inline uint32_t pack16to32(int a, int b){
56
#ifdef WORDS_BIGENDIAN
57
   return (b&0xFFFF) + (a<<16);
58
#else
59
   return (a&0xFFFF) + (b<<16);
60
#endif
61
}
62

    
63
const uint8_t ff_rem6[52]={
64
0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
65
};
66

    
67
const uint8_t ff_div6[52]={
68
0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
69
};
70

    
71

    
72
/**
73
 * fill a rectangle.
74
 * @param h height of the rectangle, should be a constant
75
 * @param w width of the rectangle, should be a constant
76
 * @param size the size of val (1 or 4), should be a constant
77
 */
78
static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
79
    uint8_t *p= (uint8_t*)vp;
80
    assert(size==1 || size==4);
81
    assert(w<=4);
82

    
83
    w      *= size;
84
    stride *= size;
85

    
86
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
87
    assert((stride&(w-1))==0);
88
    if(w==2){
89
        const uint16_t v= size==4 ? val : val*0x0101;
90
        *(uint16_t*)(p + 0*stride)= v;
91
        if(h==1) return;
92
        *(uint16_t*)(p + 1*stride)= v;
93
        if(h==2) return;
94
        *(uint16_t*)(p + 2*stride)=
95
        *(uint16_t*)(p + 3*stride)= v;
96
    }else if(w==4){
97
        const uint32_t v= size==4 ? val : val*0x01010101;
98
        *(uint32_t*)(p + 0*stride)= v;
99
        if(h==1) return;
100
        *(uint32_t*)(p + 1*stride)= v;
101
        if(h==2) return;
102
        *(uint32_t*)(p + 2*stride)=
103
        *(uint32_t*)(p + 3*stride)= v;
104
    }else if(w==8){
105
    //gcc can't optimize 64bit math on x86_32
106
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
107
        const uint64_t v= val*0x0100000001ULL;
108
        *(uint64_t*)(p + 0*stride)= v;
109
        if(h==1) return;
110
        *(uint64_t*)(p + 1*stride)= v;
111
        if(h==2) return;
112
        *(uint64_t*)(p + 2*stride)=
113
        *(uint64_t*)(p + 3*stride)= v;
114
    }else if(w==16){
115
        const uint64_t v= val*0x0100000001ULL;
116
        *(uint64_t*)(p + 0+0*stride)=
117
        *(uint64_t*)(p + 8+0*stride)=
118
        *(uint64_t*)(p + 0+1*stride)=
119
        *(uint64_t*)(p + 8+1*stride)= v;
120
        if(h==2) return;
121
        *(uint64_t*)(p + 0+2*stride)=
122
        *(uint64_t*)(p + 8+2*stride)=
123
        *(uint64_t*)(p + 0+3*stride)=
124
        *(uint64_t*)(p + 8+3*stride)= v;
125
#else
126
        *(uint32_t*)(p + 0+0*stride)=
127
        *(uint32_t*)(p + 4+0*stride)= val;
128
        if(h==1) return;
129
        *(uint32_t*)(p + 0+1*stride)=
130
        *(uint32_t*)(p + 4+1*stride)= val;
131
        if(h==2) return;
132
        *(uint32_t*)(p + 0+2*stride)=
133
        *(uint32_t*)(p + 4+2*stride)=
134
        *(uint32_t*)(p + 0+3*stride)=
135
        *(uint32_t*)(p + 4+3*stride)= val;
136
    }else if(w==16){
137
        *(uint32_t*)(p + 0+0*stride)=
138
        *(uint32_t*)(p + 4+0*stride)=
139
        *(uint32_t*)(p + 8+0*stride)=
140
        *(uint32_t*)(p +12+0*stride)=
141
        *(uint32_t*)(p + 0+1*stride)=
142
        *(uint32_t*)(p + 4+1*stride)=
143
        *(uint32_t*)(p + 8+1*stride)=
144
        *(uint32_t*)(p +12+1*stride)= val;
145
        if(h==2) return;
146
        *(uint32_t*)(p + 0+2*stride)=
147
        *(uint32_t*)(p + 4+2*stride)=
148
        *(uint32_t*)(p + 8+2*stride)=
149
        *(uint32_t*)(p +12+2*stride)=
150
        *(uint32_t*)(p + 0+3*stride)=
151
        *(uint32_t*)(p + 4+3*stride)=
152
        *(uint32_t*)(p + 8+3*stride)=
153
        *(uint32_t*)(p +12+3*stride)= val;
154
#endif
155
    }else
156
        assert(0);
157
    assert(h==4);
158
}
159

    
160
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
161
    MpegEncContext * const s = &h->s;
162
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
163
    int topleft_xy, top_xy, topright_xy, left_xy[2];
164
    int topleft_type, top_type, topright_type, left_type[2];
165
    int left_block[8];
166
    int i;
167

    
168
    //FIXME deblocking could skip the intra and nnz parts.
169
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
170
        return;
171

    
172
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
173

    
174
    top_xy     = mb_xy  - s->mb_stride;
175
    topleft_xy = top_xy - 1;
176
    topright_xy= top_xy + 1;
177
    left_xy[1] = left_xy[0] = mb_xy-1;
178
    left_block[0]= 0;
179
    left_block[1]= 1;
180
    left_block[2]= 2;
181
    left_block[3]= 3;
182
    left_block[4]= 7;
183
    left_block[5]= 10;
184
    left_block[6]= 8;
185
    left_block[7]= 11;
186
    if(FRAME_MBAFF){
187
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
188
        const int top_pair_xy      = pair_xy     - s->mb_stride;
189
        const int topleft_pair_xy  = top_pair_xy - 1;
190
        const int topright_pair_xy = top_pair_xy + 1;
191
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
192
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
193
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
194
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
195
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
196
        const int bottom = (s->mb_y & 1);
197
        tprintf(s->avctx, "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);
198
        if (bottom
199
                ? !curr_mb_frame_flag // bottom macroblock
200
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
201
                ) {
202
            top_xy -= s->mb_stride;
203
        }
204
        if (bottom
205
                ? !curr_mb_frame_flag // bottom macroblock
206
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
207
                ) {
208
            topleft_xy -= s->mb_stride;
209
        }
210
        if (bottom
211
                ? !curr_mb_frame_flag // bottom macroblock
212
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
213
                ) {
214
            topright_xy -= s->mb_stride;
215
        }
216
        if (left_mb_frame_flag != curr_mb_frame_flag) {
217
            left_xy[1] = left_xy[0] = pair_xy - 1;
218
            if (curr_mb_frame_flag) {
219
                if (bottom) {
220
                    left_block[0]= 2;
221
                    left_block[1]= 2;
222
                    left_block[2]= 3;
223
                    left_block[3]= 3;
224
                    left_block[4]= 8;
225
                    left_block[5]= 11;
226
                    left_block[6]= 8;
227
                    left_block[7]= 11;
228
                } else {
229
                    left_block[0]= 0;
230
                    left_block[1]= 0;
231
                    left_block[2]= 1;
232
                    left_block[3]= 1;
233
                    left_block[4]= 7;
234
                    left_block[5]= 10;
235
                    left_block[6]= 7;
236
                    left_block[7]= 10;
237
                }
238
            } else {
239
                left_xy[1] += s->mb_stride;
240
                //left_block[0]= 0;
241
                left_block[1]= 2;
242
                left_block[2]= 0;
243
                left_block[3]= 2;
244
                //left_block[4]= 7;
245
                left_block[5]= 10;
246
                left_block[6]= 7;
247
                left_block[7]= 10;
248
            }
249
        }
250
    }
251

    
252
    h->top_mb_xy = top_xy;
253
    h->left_mb_xy[0] = left_xy[0];
254
    h->left_mb_xy[1] = left_xy[1];
255
    if(for_deblock){
256
        topleft_type = 0;
257
        topright_type = 0;
258
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
259
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
260
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
261

    
262
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
263
            int list;
264
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
265
            for(i=0; i<16; i++)
266
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
267
            for(list=0; list<h->list_count; list++){
268
                if(USES_LIST(mb_type,list)){
269
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
270
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
271
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
272
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
273
                        dst[0] = src[0];
274
                        dst[1] = src[1];
275
                        dst[2] = src[2];
276
                        dst[3] = src[3];
277
                    }
278
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
279
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
280
                    ref += h->b8_stride;
281
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
282
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
283
                }else{
284
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
285
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
286
                }
287
            }
288
        }
289
    }else{
290
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
291
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
292
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
293
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
294
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
295
    }
296

    
297
    if(IS_INTRA(mb_type)){
298
        h->topleft_samples_available=
299
        h->top_samples_available=
300
        h->left_samples_available= 0xFFFF;
301
        h->topright_samples_available= 0xEEEA;
302

    
303
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
304
            h->topleft_samples_available= 0xB3FF;
305
            h->top_samples_available= 0x33FF;
306
            h->topright_samples_available= 0x26EA;
307
        }
308
        for(i=0; i<2; i++){
309
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
310
                h->topleft_samples_available&= 0xDF5F;
311
                h->left_samples_available&= 0x5F5F;
312
            }
313
        }
314

    
315
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
316
            h->topleft_samples_available&= 0x7FFF;
317

    
318
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
319
            h->topright_samples_available&= 0xFBFF;
320

    
321
        if(IS_INTRA4x4(mb_type)){
322
            if(IS_INTRA4x4(top_type)){
323
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
324
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
325
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
326
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
327
            }else{
328
                int pred;
329
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
330
                    pred= -1;
331
                else{
332
                    pred= 2;
333
                }
334
                h->intra4x4_pred_mode_cache[4+8*0]=
335
                h->intra4x4_pred_mode_cache[5+8*0]=
336
                h->intra4x4_pred_mode_cache[6+8*0]=
337
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
338
            }
339
            for(i=0; i<2; i++){
340
                if(IS_INTRA4x4(left_type[i])){
341
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
342
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
343
                }else{
344
                    int pred;
345
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
346
                        pred= -1;
347
                    else{
348
                        pred= 2;
349
                    }
350
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
351
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
352
                }
353
            }
354
        }
355
    }
356

    
357

    
358
/*
359
0 . T T. T T T T
360
1 L . .L . . . .
361
2 L . .L . . . .
362
3 . T TL . . . .
363
4 L . .L . . . .
364
5 L . .. . . . .
365
*/
366
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
367
    if(top_type){
368
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
369
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
370
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
371
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
372

    
373
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
374
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
375

    
376
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
377
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
378

    
379
    }else{
380
        h->non_zero_count_cache[4+8*0]=
381
        h->non_zero_count_cache[5+8*0]=
382
        h->non_zero_count_cache[6+8*0]=
383
        h->non_zero_count_cache[7+8*0]=
384

    
385
        h->non_zero_count_cache[1+8*0]=
386
        h->non_zero_count_cache[2+8*0]=
387

    
388
        h->non_zero_count_cache[1+8*3]=
389
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
390

    
391
    }
392

    
393
    for (i=0; i<2; i++) {
394
        if(left_type[i]){
395
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
396
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
397
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
398
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
399
        }else{
400
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
401
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
402
            h->non_zero_count_cache[0+8*1 +   8*i]=
403
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
404
        }
405
    }
406

    
407
    if( h->pps.cabac ) {
408
        // top_cbp
409
        if(top_type) {
410
            h->top_cbp = h->cbp_table[top_xy];
411
        } else if(IS_INTRA(mb_type)) {
412
            h->top_cbp = 0x1C0;
413
        } else {
414
            h->top_cbp = 0;
415
        }
416
        // left_cbp
417
        if (left_type[0]) {
418
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
419
        } else if(IS_INTRA(mb_type)) {
420
            h->left_cbp = 0x1C0;
421
        } else {
422
            h->left_cbp = 0;
423
        }
424
        if (left_type[0]) {
425
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
426
        }
427
        if (left_type[1]) {
428
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
429
        }
430
    }
431

    
432
#if 1
433
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
434
        int list;
435
        for(list=0; list<h->list_count; list++){
436
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
437
                /*if(!h->mv_cache_clean[list]){
438
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
439
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
440
                    h->mv_cache_clean[list]= 1;
441
                }*/
442
                continue;
443
            }
444
            h->mv_cache_clean[list]= 0;
445

    
446
            if(USES_LIST(top_type, list)){
447
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
448
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
449
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
450
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
451
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
452
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
453
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
454
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
455
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
456
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
457
            }else{
458
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
459
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
460
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
461
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
462
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
463
            }
464

    
465
            for(i=0; i<2; i++){
466
                int cache_idx = scan8[0] - 1 + i*2*8;
467
                if(USES_LIST(left_type[i], list)){
468
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
469
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
470
                    *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
471
                    *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
472
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
473
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
474
                }else{
475
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
476
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
477
                    h->ref_cache[list][cache_idx  ]=
478
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
479
                }
480
            }
481

    
482
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
483
                continue;
484

    
485
            if(USES_LIST(topleft_type, list)){
486
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
487
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
488
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
489
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
490
            }else{
491
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
492
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
493
            }
494

    
495
            if(USES_LIST(topright_type, list)){
496
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
497
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
498
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
499
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
500
            }else{
501
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
502
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
503
            }
504

    
505
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
506
                continue;
507

    
508
            h->ref_cache[list][scan8[5 ]+1] =
509
            h->ref_cache[list][scan8[7 ]+1] =
510
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
511
            h->ref_cache[list][scan8[4 ]] =
512
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
513
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
514
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
515
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
516
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
517
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
518

    
519
            if( h->pps.cabac ) {
520
                /* XXX beurk, Load mvd */
521
                if(USES_LIST(top_type, list)){
522
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
523
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
524
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
525
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
526
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
527
                }else{
528
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
529
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
530
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
531
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
532
                }
533
                if(USES_LIST(left_type[0], list)){
534
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
535
                    *(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]];
536
                    *(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]];
537
                }else{
538
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
539
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
540
                }
541
                if(USES_LIST(left_type[1], list)){
542
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
543
                    *(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]];
544
                    *(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]];
545
                }else{
546
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
547
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
548
                }
549
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
550
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
551
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
552
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
553
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
554

    
555
                if(h->slice_type == B_TYPE){
556
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
557

    
558
                    if(IS_DIRECT(top_type)){
559
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
560
                    }else if(IS_8X8(top_type)){
561
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
562
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
563
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
564
                    }else{
565
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
566
                    }
567

    
568
                    if(IS_DIRECT(left_type[0]))
569
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
570
                    else if(IS_8X8(left_type[0]))
571
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
572
                    else
573
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
574

    
575
                    if(IS_DIRECT(left_type[1]))
576
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
577
                    else if(IS_8X8(left_type[1]))
578
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
579
                    else
580
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
581
                }
582
            }
583

    
584
            if(FRAME_MBAFF){
585
#define MAP_MVS\
586
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
587
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
588
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
589
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
590
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
591
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
592
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
593
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
594
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
595
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
596
                if(MB_FIELD){
597
#define MAP_F2F(idx, mb_type)\
598
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
599
                        h->ref_cache[list][idx] <<= 1;\
600
                        h->mv_cache[list][idx][1] /= 2;\
601
                        h->mvd_cache[list][idx][1] /= 2;\
602
                    }
603
                    MAP_MVS
604
#undef MAP_F2F
605
                }else{
606
#define MAP_F2F(idx, mb_type)\
607
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
608
                        h->ref_cache[list][idx] >>= 1;\
609
                        h->mv_cache[list][idx][1] <<= 1;\
610
                        h->mvd_cache[list][idx][1] <<= 1;\
611
                    }
612
                    MAP_MVS
613
#undef MAP_F2F
614
                }
615
            }
616
        }
617
    }
618
#endif
619

    
620
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
621
}
622

    
623
static inline void write_back_intra_pred_mode(H264Context *h){
624
    MpegEncContext * const s = &h->s;
625
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
626

    
627
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
628
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
629
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
630
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
631
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
632
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
633
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
634
}
635

    
636
/**
637
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
638
 */
639
static inline int check_intra4x4_pred_mode(H264Context *h){
640
    MpegEncContext * const s = &h->s;
641
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
642
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
643
    int i;
644

    
645
    if(!(h->top_samples_available&0x8000)){
646
        for(i=0; i<4; i++){
647
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
648
            if(status<0){
649
                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);
650
                return -1;
651
            } else if(status){
652
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
653
            }
654
        }
655
    }
656

    
657
    if(!(h->left_samples_available&0x8000)){
658
        for(i=0; i<4; i++){
659
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
660
            if(status<0){
661
                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);
662
                return -1;
663
            } else if(status){
664
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
665
            }
666
        }
667
    }
668

    
669
    return 0;
670
} //FIXME cleanup like next
671

    
672
/**
673
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
674
 */
675
static inline int check_intra_pred_mode(H264Context *h, int mode){
676
    MpegEncContext * const s = &h->s;
677
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
678
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
679

    
680
    if(mode > 6U) {
681
        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);
682
        return -1;
683
    }
684

    
685
    if(!(h->top_samples_available&0x8000)){
686
        mode= top[ mode ];
687
        if(mode<0){
688
            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);
689
            return -1;
690
        }
691
    }
692

    
693
    if(!(h->left_samples_available&0x8000)){
694
        mode= left[ mode ];
695
        if(mode<0){
696
            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);
697
            return -1;
698
        }
699
    }
700

    
701
    return mode;
702
}
703

    
704
/**
705
 * gets the predicted intra4x4 prediction mode.
706
 */
707
static inline int pred_intra_mode(H264Context *h, int n){
708
    const int index8= scan8[n];
709
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
710
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
711
    const int min= FFMIN(left, top);
712

    
713
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
714

    
715
    if(min<0) return DC_PRED;
716
    else      return min;
717
}
718

    
719
static inline void write_back_non_zero_count(H264Context *h){
720
    MpegEncContext * const s = &h->s;
721
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
722

    
723
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
724
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
725
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
726
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
727
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
728
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
729
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
730

    
731
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
732
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
733
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
734

    
735
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
736
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
737
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
738

    
739
    if(FRAME_MBAFF){
740
        // store all luma nnzs, for deblocking
741
        int v = 0, i;
742
        for(i=0; i<16; i++)
743
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
744
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
745
    }
746
}
747

    
748
/**
749
 * gets the predicted number of non zero coefficients.
750
 * @param n block index
751
 */
752
static inline int pred_non_zero_count(H264Context *h, int n){
753
    const int index8= scan8[n];
754
    const int left= h->non_zero_count_cache[index8 - 1];
755
    const int top = h->non_zero_count_cache[index8 - 8];
756
    int i= left + top;
757

    
758
    if(i<64) i= (i+1)>>1;
759

    
760
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
761

    
762
    return i&31;
763
}
764

    
765
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
766
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
767
    MpegEncContext *s = &h->s;
768

    
769
    /* there is no consistent mapping of mvs to neighboring locations that will
770
     * make mbaff happy, so we can't move all this logic to fill_caches */
771
    if(FRAME_MBAFF){
772
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
773
        const int16_t *mv;
774
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
775
        *C = h->mv_cache[list][scan8[0]-2];
776

    
777
        if(!MB_FIELD
778
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
779
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
780
            if(IS_INTERLACED(mb_types[topright_xy])){
781
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
782
                const int x4 = X4, y4 = Y4;\
783
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
784
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
785
                    return LIST_NOT_USED;\
786
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
787
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
788
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
789
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
790

    
791
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
792
            }
793
        }
794
        if(topright_ref == PART_NOT_AVAILABLE
795
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
796
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
797
            if(!MB_FIELD
798
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
799
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
800
            }
801
            if(MB_FIELD
802
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
803
               && i >= scan8[0]+8){
804
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
805
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
806
            }
807
        }
808
#undef SET_DIAG_MV
809
    }
810

    
811
    if(topright_ref != PART_NOT_AVAILABLE){
812
        *C= h->mv_cache[list][ i - 8 + part_width ];
813
        return topright_ref;
814
    }else{
815
        tprintf(s->avctx, "topright MV not available\n");
816

    
817
        *C= h->mv_cache[list][ i - 8 - 1 ];
818
        return h->ref_cache[list][ i - 8 - 1 ];
819
    }
820
}
821

    
822
/**
823
 * gets the predicted MV.
824
 * @param n the block index
825
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
826
 * @param mx the x component of the predicted motion vector
827
 * @param my the y component of the predicted motion vector
828
 */
829
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
830
    const int index8= scan8[n];
831
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
832
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
833
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
834
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
835
    const int16_t * C;
836
    int diagonal_ref, match_count;
837

    
838
    assert(part_width==1 || part_width==2 || part_width==4);
839

    
840
/* mv_cache
841
  B . . A T T T T
842
  U . . L . . , .
843
  U . . L . . . .
844
  U . . L . . , .
845
  . . . L . . . .
846
*/
847

    
848
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
849
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
850
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
851
    if(match_count > 1){ //most common
852
        *mx= mid_pred(A[0], B[0], C[0]);
853
        *my= mid_pred(A[1], B[1], C[1]);
854
    }else if(match_count==1){
855
        if(left_ref==ref){
856
            *mx= A[0];
857
            *my= A[1];
858
        }else if(top_ref==ref){
859
            *mx= B[0];
860
            *my= B[1];
861
        }else{
862
            *mx= C[0];
863
            *my= C[1];
864
        }
865
    }else{
866
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
867
            *mx= A[0];
868
            *my= A[1];
869
        }else{
870
            *mx= mid_pred(A[0], B[0], C[0]);
871
            *my= mid_pred(A[1], B[1], C[1]);
872
        }
873
    }
874

    
875
    tprintf(h->s.avctx, "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);
876
}
877

    
878
/**
879
 * gets the directionally predicted 16x8 MV.
880
 * @param n the block index
881
 * @param mx the x component of the predicted motion vector
882
 * @param my the y component of the predicted motion vector
883
 */
884
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
885
    if(n==0){
886
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
887
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
888

    
889
        tprintf(h->s.avctx, "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);
890

    
891
        if(top_ref == ref){
892
            *mx= B[0];
893
            *my= B[1];
894
            return;
895
        }
896
    }else{
897
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
898
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
899

    
900
        tprintf(h->s.avctx, "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);
901

    
902
        if(left_ref == ref){
903
            *mx= A[0];
904
            *my= A[1];
905
            return;
906
        }
907
    }
908

    
909
    //RARE
910
    pred_motion(h, n, 4, list, ref, mx, my);
911
}
912

    
913
/**
914
 * gets the directionally predicted 8x16 MV.
915
 * @param n the block index
916
 * @param mx the x component of the predicted motion vector
917
 * @param my the y component of the predicted motion vector
918
 */
919
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
920
    if(n==0){
921
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
922
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
923

    
924
        tprintf(h->s.avctx, "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);
925

    
926
        if(left_ref == ref){
927
            *mx= A[0];
928
            *my= A[1];
929
            return;
930
        }
931
    }else{
932
        const int16_t * C;
933
        int diagonal_ref;
934

    
935
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
936

    
937
        tprintf(h->s.avctx, "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);
938

    
939
        if(diagonal_ref == ref){
940
            *mx= C[0];
941
            *my= C[1];
942
            return;
943
        }
944
    }
945

    
946
    //RARE
947
    pred_motion(h, n, 2, list, ref, mx, my);
948
}
949

    
950
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
951
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
952
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
953

    
954
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
955

    
956
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
957
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
958
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
959

    
960
        *mx = *my = 0;
961
        return;
962
    }
963

    
964
    pred_motion(h, 0, 4, 0, 0, mx, my);
965

    
966
    return;
967
}
968

    
969
static inline void direct_dist_scale_factor(H264Context * const h){
970
    const int poc = h->s.current_picture_ptr->poc;
971
    const int poc1 = h->ref_list[1][0].poc;
972
    int i;
973
    for(i=0; i<h->ref_count[0]; i++){
974
        int poc0 = h->ref_list[0][i].poc;
975
        int td = av_clip(poc1 - poc0, -128, 127);
976
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
977
            h->dist_scale_factor[i] = 256;
978
        }else{
979
            int tb = av_clip(poc - poc0, -128, 127);
980
            int tx = (16384 + (FFABS(td) >> 1)) / td;
981
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
982
        }
983
    }
984
    if(FRAME_MBAFF){
985
        for(i=0; i<h->ref_count[0]; i++){
986
            h->dist_scale_factor_field[2*i] =
987
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
988
        }
989
    }
990
}
991
static inline void direct_ref_list_init(H264Context * const h){
992
    MpegEncContext * const s = &h->s;
993
    Picture * const ref1 = &h->ref_list[1][0];
994
    Picture * const cur = s->current_picture_ptr;
995
    int list, i, j;
996
    if(cur->pict_type == I_TYPE)
997
        cur->ref_count[0] = 0;
998
    if(cur->pict_type != B_TYPE)
999
        cur->ref_count[1] = 0;
1000
    for(list=0; list<2; list++){
1001
        cur->ref_count[list] = h->ref_count[list];
1002
        for(j=0; j<h->ref_count[list]; j++)
1003
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1004
    }
1005
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1006
        return;
1007
    for(list=0; list<2; list++){
1008
        for(i=0; i<ref1->ref_count[list]; i++){
1009
            const int poc = ref1->ref_poc[list][i];
1010
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1011
            for(j=0; j<h->ref_count[list]; j++)
1012
                if(h->ref_list[list][j].poc == poc){
1013
                    h->map_col_to_list0[list][i] = j;
1014
                    break;
1015
                }
1016
        }
1017
    }
1018
    if(FRAME_MBAFF){
1019
        for(list=0; list<2; list++){
1020
            for(i=0; i<ref1->ref_count[list]; i++){
1021
                j = h->map_col_to_list0[list][i];
1022
                h->map_col_to_list0_field[list][2*i] = 2*j;
1023
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1024
            }
1025
        }
1026
    }
1027
}
1028

    
1029
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1030
    MpegEncContext * const s = &h->s;
1031
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1032
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1033
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1034
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1035
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1036
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1037
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1038
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1039
    const int is_b8x8 = IS_8X8(*mb_type);
1040
    unsigned int sub_mb_type;
1041
    int i8, i4;
1042

    
1043
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1044
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1045
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1046
         * so we know exactly what block size to use */
1047
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1048
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1049
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1050
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1051
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1052
    }else{
1053
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1054
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1055
    }
1056
    if(!is_b8x8)
1057
        *mb_type |= MB_TYPE_DIRECT2;
1058
    if(MB_FIELD)
1059
        *mb_type |= MB_TYPE_INTERLACED;
1060

    
1061
    tprintf(s->avctx, "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);
1062

    
1063
    if(h->direct_spatial_mv_pred){
1064
        int ref[2];
1065
        int mv[2][2];
1066
        int list;
1067

    
1068
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1069

    
1070
        /* ref = min(neighbors) */
1071
        for(list=0; list<2; list++){
1072
            int refa = h->ref_cache[list][scan8[0] - 1];
1073
            int refb = h->ref_cache[list][scan8[0] - 8];
1074
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1075
            if(refc == -2)
1076
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1077
            ref[list] = refa;
1078
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1079
                ref[list] = refb;
1080
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1081
                ref[list] = refc;
1082
            if(ref[list] < 0)
1083
                ref[list] = -1;
1084
        }
1085

    
1086
        if(ref[0] < 0 && ref[1] < 0){
1087
            ref[0] = ref[1] = 0;
1088
            mv[0][0] = mv[0][1] =
1089
            mv[1][0] = mv[1][1] = 0;
1090
        }else{
1091
            for(list=0; list<2; list++){
1092
                if(ref[list] >= 0)
1093
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1094
                else
1095
                    mv[list][0] = mv[list][1] = 0;
1096
            }
1097
        }
1098

    
1099
        if(ref[1] < 0){
1100
            *mb_type &= ~MB_TYPE_P0L1;
1101
            sub_mb_type &= ~MB_TYPE_P0L1;
1102
        }else if(ref[0] < 0){
1103
            *mb_type &= ~MB_TYPE_P0L0;
1104
            sub_mb_type &= ~MB_TYPE_P0L0;
1105
        }
1106

    
1107
        if(IS_16X16(*mb_type)){
1108
            int a=0, b=0;
1109

    
1110
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1111
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1112
            if(!IS_INTRA(mb_type_col)
1113
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1114
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1115
                       && (h->x264_build>33 || !h->x264_build)))){
1116
                if(ref[0] > 0)
1117
                    a= pack16to32(mv[0][0],mv[0][1]);
1118
                if(ref[1] > 0)
1119
                    b= pack16to32(mv[1][0],mv[1][1]);
1120
            }else{
1121
                a= pack16to32(mv[0][0],mv[0][1]);
1122
                b= pack16to32(mv[1][0],mv[1][1]);
1123
            }
1124
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1125
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1126
        }else{
1127
            for(i8=0; i8<4; i8++){
1128
                const int x8 = i8&1;
1129
                const int y8 = i8>>1;
1130

    
1131
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1132
                    continue;
1133
                h->sub_mb_type[i8] = sub_mb_type;
1134

    
1135
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1136
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1137
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1138
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1139

    
1140
                /* col_zero_flag */
1141
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1142
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1143
                                                  && (h->x264_build>33 || !h->x264_build)))){
1144
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1145
                    if(IS_SUB_8X8(sub_mb_type)){
1146
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1147
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1148
                            if(ref[0] == 0)
1149
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1150
                            if(ref[1] == 0)
1151
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1152
                        }
1153
                    }else
1154
                    for(i4=0; i4<4; i4++){
1155
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1156
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1157
                            if(ref[0] == 0)
1158
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1159
                            if(ref[1] == 0)
1160
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1161
                        }
1162
                    }
1163
                }
1164
            }
1165
        }
1166
    }else{ /* direct temporal mv pred */
1167
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1168
        const int *dist_scale_factor = h->dist_scale_factor;
1169

    
1170
        if(FRAME_MBAFF){
1171
            if(IS_INTERLACED(*mb_type)){
1172
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1173
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1174
                dist_scale_factor = h->dist_scale_factor_field;
1175
            }
1176
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1177
                /* FIXME assumes direct_8x8_inference == 1 */
1178
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1179
                int mb_types_col[2];
1180
                int y_shift;
1181

    
1182
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1183
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1184
                         | (*mb_type & MB_TYPE_INTERLACED);
1185
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1186

    
1187
                if(IS_INTERLACED(*mb_type)){
1188
                    /* frame to field scaling */
1189
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1190
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1191
                    if(s->mb_y&1){
1192
                        l1ref0 -= 2*h->b8_stride;
1193
                        l1ref1 -= 2*h->b8_stride;
1194
                        l1mv0 -= 4*h->b_stride;
1195
                        l1mv1 -= 4*h->b_stride;
1196
                    }
1197
                    y_shift = 0;
1198

    
1199
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1200
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1201
                       && !is_b8x8)
1202
                        *mb_type |= MB_TYPE_16x8;
1203
                    else
1204
                        *mb_type |= MB_TYPE_8x8;
1205
                }else{
1206
                    /* field to frame scaling */
1207
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1208
                     * but in MBAFF, top and bottom POC are equal */
1209
                    int dy = (s->mb_y&1) ? 1 : 2;
1210
                    mb_types_col[0] =
1211
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1212
                    l1ref0 += dy*h->b8_stride;
1213
                    l1ref1 += dy*h->b8_stride;
1214
                    l1mv0 += 2*dy*h->b_stride;
1215
                    l1mv1 += 2*dy*h->b_stride;
1216
                    y_shift = 2;
1217

    
1218
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1219
                       && !is_b8x8)
1220
                        *mb_type |= MB_TYPE_16x16;
1221
                    else
1222
                        *mb_type |= MB_TYPE_8x8;
1223
                }
1224

    
1225
                for(i8=0; i8<4; i8++){
1226
                    const int x8 = i8&1;
1227
                    const int y8 = i8>>1;
1228
                    int ref0, scale;
1229
                    const int16_t (*l1mv)[2]= l1mv0;
1230

    
1231
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1232
                        continue;
1233
                    h->sub_mb_type[i8] = sub_mb_type;
1234

    
1235
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1236
                    if(IS_INTRA(mb_types_col[y8])){
1237
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1238
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1239
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1240
                        continue;
1241
                    }
1242

    
1243
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1244
                    if(ref0 >= 0)
1245
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1246
                    else{
1247
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1248
                        l1mv= l1mv1;
1249
                    }
1250
                    scale = dist_scale_factor[ref0];
1251
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1252

    
1253
                    {
1254
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1255
                        int my_col = (mv_col[1]<<y_shift)/2;
1256
                        int mx = (scale * mv_col[0] + 128) >> 8;
1257
                        int my = (scale * my_col + 128) >> 8;
1258
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1259
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1260
                    }
1261
                }
1262
                return;
1263
            }
1264
        }
1265

    
1266
        /* one-to-one mv scaling */
1267

    
1268
        if(IS_16X16(*mb_type)){
1269
            int ref, mv0, mv1;
1270

    
1271
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1272
            if(IS_INTRA(mb_type_col)){
1273
                ref=mv0=mv1=0;
1274
            }else{
1275
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1276
                                                : map_col_to_list0[1][l1ref1[0]];
1277
                const int scale = dist_scale_factor[ref0];
1278
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1279
                int mv_l0[2];
1280
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1281
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1282
                ref= ref0;
1283
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1284
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1285
            }
1286
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1287
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1288
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1289
        }else{
1290
            for(i8=0; i8<4; i8++){
1291
                const int x8 = i8&1;
1292
                const int y8 = i8>>1;
1293
                int ref0, scale;
1294
                const int16_t (*l1mv)[2]= l1mv0;
1295

    
1296
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1297
                    continue;
1298
                h->sub_mb_type[i8] = sub_mb_type;
1299
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1300
                if(IS_INTRA(mb_type_col)){
1301
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1302
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1303
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1304
                    continue;
1305
                }
1306

    
1307
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1308
                if(ref0 >= 0)
1309
                    ref0 = map_col_to_list0[0][ref0];
1310
                else{
1311
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1312
                    l1mv= l1mv1;
1313
                }
1314
                scale = dist_scale_factor[ref0];
1315

    
1316
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1317
                if(IS_SUB_8X8(sub_mb_type)){
1318
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1319
                    int mx = (scale * mv_col[0] + 128) >> 8;
1320
                    int my = (scale * mv_col[1] + 128) >> 8;
1321
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1322
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1323
                }else
1324
                for(i4=0; i4<4; i4++){
1325
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1326
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1327
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1328
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1329
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1330
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1331
                }
1332
            }
1333
        }
1334
    }
1335
}
1336

    
1337
static inline void write_back_motion(H264Context *h, int mb_type){
1338
    MpegEncContext * const s = &h->s;
1339
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1340
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1341
    int list;
1342

    
1343
    if(!USES_LIST(mb_type, 0))
1344
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1345

    
1346
    for(list=0; list<h->list_count; list++){
1347
        int y;
1348
        if(!USES_LIST(mb_type, list))
1349
            continue;
1350

    
1351
        for(y=0; y<4; y++){
1352
            *(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];
1353
            *(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];
1354
        }
1355
        if( h->pps.cabac ) {
1356
            if(IS_SKIP(mb_type))
1357
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1358
            else
1359
            for(y=0; y<4; y++){
1360
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1361
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1362
            }
1363
        }
1364

    
1365
        {
1366
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1367
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1368
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1369
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1370
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1371
        }
1372
    }
1373

    
1374
    if(h->slice_type == B_TYPE && h->pps.cabac){
1375
        if(IS_8X8(mb_type)){
1376
            uint8_t *direct_table = &h->direct_table[b8_xy];
1377
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1378
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1379
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1380
        }
1381
    }
1382
}
1383

    
1384
/**
1385
 * Decodes a network abstraction layer unit.
1386
 * @param consumed is the number of bytes used as input
1387
 * @param length is the length of the array
1388
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1389
 * @returns decoded bytes, might be src+1 if no escapes
1390
 */
1391
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1392
    int i, si, di;
1393
    uint8_t *dst;
1394
    int bufidx;
1395

    
1396
//    src[0]&0x80;                //forbidden bit
1397
    h->nal_ref_idc= src[0]>>5;
1398
    h->nal_unit_type= src[0]&0x1F;
1399

    
1400
    src++; length--;
1401
#if 0
1402
    for(i=0; i<length; i++)
1403
        printf("%2X ", src[i]);
1404
#endif
1405
    for(i=0; i+1<length; i+=2){
1406
        if(src[i]) continue;
1407
        if(i>0 && src[i-1]==0) i--;
1408
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1409
            if(src[i+2]!=3){
1410
                /* startcode, so we must be past the end */
1411
                length=i;
1412
            }
1413
            break;
1414
        }
1415
    }
1416

    
1417
    if(i>=length-1){ //no escaped 0
1418
        *dst_length= length;
1419
        *consumed= length+1; //+1 for the header
1420
        return src;
1421
    }
1422

    
1423
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1424
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1425
    dst= h->rbsp_buffer[bufidx];
1426

    
1427
    if (dst == NULL){
1428
        return NULL;
1429
    }
1430

    
1431
//printf("decoding esc\n");
1432
    si=di=0;
1433
    while(si<length){
1434
        //remove escapes (very rare 1:2^22)
1435
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1436
            if(src[si+2]==3){ //escape
1437
                dst[di++]= 0;
1438
                dst[di++]= 0;
1439
                si+=3;
1440
                continue;
1441
            }else //next start code
1442
                break;
1443
        }
1444

    
1445
        dst[di++]= src[si++];
1446
    }
1447

    
1448
    *dst_length= di;
1449
    *consumed= si + 1;//+1 for the header
1450
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1451
    return dst;
1452
}
1453

    
1454
/**
1455
 * identifies the exact end of the bitstream
1456
 * @return the length of the trailing, or 0 if damaged
1457
 */
1458
static int decode_rbsp_trailing(H264Context *h, uint8_t *src){
1459
    int v= *src;
1460
    int r;
1461

    
1462
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1463

    
1464
    for(r=1; r<9; r++){
1465
        if(v&1) return r;
1466
        v>>=1;
1467
    }
1468
    return 0;
1469
}
1470

    
1471
/**
1472
 * idct tranforms the 16 dc values and dequantize them.
1473
 * @param qp quantization parameter
1474
 */
1475
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1476
#define stride 16
1477
    int i;
1478
    int temp[16]; //FIXME check if this is a good idea
1479
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1480
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1481

    
1482
//memset(block, 64, 2*256);
1483
//return;
1484
    for(i=0; i<4; i++){
1485
        const int offset= y_offset[i];
1486
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1487
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1488
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1489
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1490

    
1491
        temp[4*i+0]= z0+z3;
1492
        temp[4*i+1]= z1+z2;
1493
        temp[4*i+2]= z1-z2;
1494
        temp[4*i+3]= z0-z3;
1495
    }
1496

    
1497
    for(i=0; i<4; i++){
1498
        const int offset= x_offset[i];
1499
        const int z0= temp[4*0+i] + temp[4*2+i];
1500
        const int z1= temp[4*0+i] - temp[4*2+i];
1501
        const int z2= temp[4*1+i] - temp[4*3+i];
1502
        const int z3= temp[4*1+i] + temp[4*3+i];
1503

    
1504
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1505
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1506
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1507
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1508
    }
1509
}
1510

    
1511
#if 0
1512
/**
1513
 * dct tranforms the 16 dc values.
1514
 * @param qp quantization parameter ??? FIXME
1515
 */
1516
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1517
//    const int qmul= dequant_coeff[qp][0];
1518
    int i;
1519
    int temp[16]; //FIXME check if this is a good idea
1520
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1521
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1522

1523
    for(i=0; i<4; i++){
1524
        const int offset= y_offset[i];
1525
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1526
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1527
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1528
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1529

1530
        temp[4*i+0]= z0+z3;
1531
        temp[4*i+1]= z1+z2;
1532
        temp[4*i+2]= z1-z2;
1533
        temp[4*i+3]= z0-z3;
1534
    }
1535

1536
    for(i=0; i<4; i++){
1537
        const int offset= x_offset[i];
1538
        const int z0= temp[4*0+i] + temp[4*2+i];
1539
        const int z1= temp[4*0+i] - temp[4*2+i];
1540
        const int z2= temp[4*1+i] - temp[4*3+i];
1541
        const int z3= temp[4*1+i] + temp[4*3+i];
1542

1543
        block[stride*0 +offset]= (z0 + z3)>>1;
1544
        block[stride*2 +offset]= (z1 + z2)>>1;
1545
        block[stride*8 +offset]= (z1 - z2)>>1;
1546
        block[stride*10+offset]= (z0 - z3)>>1;
1547
    }
1548
}
1549
#endif
1550

    
1551
#undef xStride
1552
#undef stride
1553

    
1554
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1555
    const int stride= 16*2;
1556
    const int xStride= 16;
1557
    int a,b,c,d,e;
1558

    
1559
    a= block[stride*0 + xStride*0];
1560
    b= block[stride*0 + xStride*1];
1561
    c= block[stride*1 + xStride*0];
1562
    d= block[stride*1 + xStride*1];
1563

    
1564
    e= a-b;
1565
    a= a+b;
1566
    b= c-d;
1567
    c= c+d;
1568

    
1569
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1570
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1571
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1572
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1573
}
1574

    
1575
#if 0
1576
static void chroma_dc_dct_c(DCTELEM *block){
1577
    const int stride= 16*2;
1578
    const int xStride= 16;
1579
    int a,b,c,d,e;
1580

1581
    a= block[stride*0 + xStride*0];
1582
    b= block[stride*0 + xStride*1];
1583
    c= block[stride*1 + xStride*0];
1584
    d= block[stride*1 + xStride*1];
1585

1586
    e= a-b;
1587
    a= a+b;
1588
    b= c-d;
1589
    c= c+d;
1590

1591
    block[stride*0 + xStride*0]= (a+c);
1592
    block[stride*0 + xStride*1]= (e+b);
1593
    block[stride*1 + xStride*0]= (a-c);
1594
    block[stride*1 + xStride*1]= (e-b);
1595
}
1596
#endif
1597

    
1598
/**
1599
 * gets the chroma qp.
1600
 */
1601
static inline int get_chroma_qp(H264Context *h, int qscale){
1602
    return h->pps.chroma_qp_table[qscale & 0xff];
1603
}
1604

    
1605
//FIXME need to check that this does not overflow signed 32 bit for low qp, i am not sure, it's very close
1606
//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1607
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1608
    int i;
1609
    const int * const quant_table= quant_coeff[qscale];
1610
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1611
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1612
    const unsigned int threshold2= (threshold1<<1);
1613
    int last_non_zero;
1614

    
1615
    if(separate_dc){
1616
        if(qscale<=18){
1617
            //avoid overflows
1618
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1619
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1620
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1621

    
1622
            int level= block[0]*quant_coeff[qscale+18][0];
1623
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1624
                if(level>0){
1625
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1626
                    block[0]= level;
1627
                }else{
1628
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1629
                    block[0]= -level;
1630
                }
1631
//                last_non_zero = i;
1632
            }else{
1633
                block[0]=0;
1634
            }
1635
        }else{
1636
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1637
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1638
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1639

    
1640
            int level= block[0]*quant_table[0];
1641
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1642
                if(level>0){
1643
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1644
                    block[0]= level;
1645
                }else{
1646
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1647
                    block[0]= -level;
1648
                }
1649
//                last_non_zero = i;
1650
            }else{
1651
                block[0]=0;
1652
            }
1653
        }
1654
        last_non_zero= 0;
1655
        i=1;
1656
    }else{
1657
        last_non_zero= -1;
1658
        i=0;
1659
    }
1660

    
1661
    for(; i<16; i++){
1662
        const int j= scantable[i];
1663
        int level= block[j]*quant_table[j];
1664

    
1665
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1666
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1667
        if(((unsigned)(level+threshold1))>threshold2){
1668
            if(level>0){
1669
                level= (bias + level)>>QUANT_SHIFT;
1670
                block[j]= level;
1671
            }else{
1672
                level= (bias - level)>>QUANT_SHIFT;
1673
                block[j]= -level;
1674
            }
1675
            last_non_zero = i;
1676
        }else{
1677
            block[j]=0;
1678
        }
1679
    }
1680

    
1681
    return last_non_zero;
1682
}
1683

    
1684
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1685
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1686
    ((uint32_t*)(src+0*stride))[0]= a;
1687
    ((uint32_t*)(src+1*stride))[0]= a;
1688
    ((uint32_t*)(src+2*stride))[0]= a;
1689
    ((uint32_t*)(src+3*stride))[0]= a;
1690
}
1691

    
1692
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1693
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1694
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1695
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1696
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1697
}
1698

    
1699
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1700
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1701
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1702

    
1703
    ((uint32_t*)(src+0*stride))[0]=
1704
    ((uint32_t*)(src+1*stride))[0]=
1705
    ((uint32_t*)(src+2*stride))[0]=
1706
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1707
}
1708

    
1709
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1710
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1711

    
1712
    ((uint32_t*)(src+0*stride))[0]=
1713
    ((uint32_t*)(src+1*stride))[0]=
1714
    ((uint32_t*)(src+2*stride))[0]=
1715
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1716
}
1717

    
1718
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1719
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1720

    
1721
    ((uint32_t*)(src+0*stride))[0]=
1722
    ((uint32_t*)(src+1*stride))[0]=
1723
    ((uint32_t*)(src+2*stride))[0]=
1724
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1725
}
1726

    
1727
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1728
    ((uint32_t*)(src+0*stride))[0]=
1729
    ((uint32_t*)(src+1*stride))[0]=
1730
    ((uint32_t*)(src+2*stride))[0]=
1731
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1732
}
1733

    
1734

    
1735
#define LOAD_TOP_RIGHT_EDGE\
1736
    const int av_unused t4= topright[0];\
1737
    const int av_unused t5= topright[1];\
1738
    const int av_unused t6= topright[2];\
1739
    const int av_unused t7= topright[3];\
1740

    
1741
#define LOAD_LEFT_EDGE\
1742
    const int av_unused l0= src[-1+0*stride];\
1743
    const int av_unused l1= src[-1+1*stride];\
1744
    const int av_unused l2= src[-1+2*stride];\
1745
    const int av_unused l3= src[-1+3*stride];\
1746

    
1747
#define LOAD_TOP_EDGE\
1748
    const int av_unused t0= src[ 0-1*stride];\
1749
    const int av_unused t1= src[ 1-1*stride];\
1750
    const int av_unused t2= src[ 2-1*stride];\
1751
    const int av_unused t3= src[ 3-1*stride];\
1752

    
1753
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1754
    const int lt= src[-1-1*stride];
1755
    LOAD_TOP_EDGE
1756
    LOAD_LEFT_EDGE
1757

    
1758
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1759
    src[0+2*stride]=
1760
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1761
    src[0+1*stride]=
1762
    src[1+2*stride]=
1763
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1764
    src[0+0*stride]=
1765
    src[1+1*stride]=
1766
    src[2+2*stride]=
1767
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1768
    src[1+0*stride]=
1769
    src[2+1*stride]=
1770
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1771
    src[2+0*stride]=
1772
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1773
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1774
}
1775

    
1776
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1777
    LOAD_TOP_EDGE
1778
    LOAD_TOP_RIGHT_EDGE
1779
//    LOAD_LEFT_EDGE
1780

    
1781
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1782
    src[1+0*stride]=
1783
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1784
    src[2+0*stride]=
1785
    src[1+1*stride]=
1786
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1787
    src[3+0*stride]=
1788
    src[2+1*stride]=
1789
    src[1+2*stride]=
1790
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1791
    src[3+1*stride]=
1792
    src[2+2*stride]=
1793
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1794
    src[3+2*stride]=
1795
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1796
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1797
}
1798

    
1799
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1800
    const int lt= src[-1-1*stride];
1801
    LOAD_TOP_EDGE
1802
    LOAD_LEFT_EDGE
1803

    
1804
    src[0+0*stride]=
1805
    src[1+2*stride]=(lt + t0 + 1)>>1;
1806
    src[1+0*stride]=
1807
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1808
    src[2+0*stride]=
1809
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1810
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1811
    src[0+1*stride]=
1812
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1813
    src[1+1*stride]=
1814
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1815
    src[2+1*stride]=
1816
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1817
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1818
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1819
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1820
}
1821

    
1822
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1823
    LOAD_TOP_EDGE
1824
    LOAD_TOP_RIGHT_EDGE
1825

    
1826
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1827
    src[1+0*stride]=
1828
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1829
    src[2+0*stride]=
1830
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1831
    src[3+0*stride]=
1832
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1833
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1834
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1835
    src[1+1*stride]=
1836
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1837
    src[2+1*stride]=
1838
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1839
    src[3+1*stride]=
1840
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1841
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1842
}
1843

    
1844
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1845
    LOAD_LEFT_EDGE
1846

    
1847
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1848
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1849
    src[2+0*stride]=
1850
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1851
    src[3+0*stride]=
1852
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1853
    src[2+1*stride]=
1854
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1855
    src[3+1*stride]=
1856
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1857
    src[3+2*stride]=
1858
    src[1+3*stride]=
1859
    src[0+3*stride]=
1860
    src[2+2*stride]=
1861
    src[2+3*stride]=
1862
    src[3+3*stride]=l3;
1863
}
1864

    
1865
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1866
    const int lt= src[-1-1*stride];
1867
    LOAD_TOP_EDGE
1868
    LOAD_LEFT_EDGE
1869

    
1870
    src[0+0*stride]=
1871
    src[2+1*stride]=(lt + l0 + 1)>>1;
1872
    src[1+0*stride]=
1873
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1874
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1875
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1876
    src[0+1*stride]=
1877
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1878
    src[1+1*stride]=
1879
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1880
    src[0+2*stride]=
1881
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1882
    src[1+2*stride]=
1883
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1884
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1885
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1886
}
1887

    
1888
void ff_pred16x16_vertical_c(uint8_t *src, int stride){
1889
    int i;
1890
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1891
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1892
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1893
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1894

    
1895
    for(i=0; i<16; i++){
1896
        ((uint32_t*)(src+i*stride))[0]= a;
1897
        ((uint32_t*)(src+i*stride))[1]= b;
1898
        ((uint32_t*)(src+i*stride))[2]= c;
1899
        ((uint32_t*)(src+i*stride))[3]= d;
1900
    }
1901
}
1902

    
1903
void ff_pred16x16_horizontal_c(uint8_t *src, int stride){
1904
    int i;
1905

    
1906
    for(i=0; i<16; i++){
1907
        ((uint32_t*)(src+i*stride))[0]=
1908
        ((uint32_t*)(src+i*stride))[1]=
1909
        ((uint32_t*)(src+i*stride))[2]=
1910
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1911
    }
1912
}
1913

    
1914
void ff_pred16x16_dc_c(uint8_t *src, int stride){
1915
    int i, dc=0;
1916

    
1917
    for(i=0;i<16; i++){
1918
        dc+= src[-1+i*stride];
1919
    }
1920

    
1921
    for(i=0;i<16; i++){
1922
        dc+= src[i-stride];
1923
    }
1924

    
1925
    dc= 0x01010101*((dc + 16)>>5);
1926

    
1927
    for(i=0; i<16; i++){
1928
        ((uint32_t*)(src+i*stride))[0]=
1929
        ((uint32_t*)(src+i*stride))[1]=
1930
        ((uint32_t*)(src+i*stride))[2]=
1931
        ((uint32_t*)(src+i*stride))[3]= dc;
1932
    }
1933
}
1934

    
1935
void ff_pred16x16_left_dc_c(uint8_t *src, int stride){
1936
    int i, dc=0;
1937

    
1938
    for(i=0;i<16; i++){
1939
        dc+= src[-1+i*stride];
1940
    }
1941

    
1942
    dc= 0x01010101*((dc + 8)>>4);
1943

    
1944
    for(i=0; i<16; i++){
1945
        ((uint32_t*)(src+i*stride))[0]=
1946
        ((uint32_t*)(src+i*stride))[1]=
1947
        ((uint32_t*)(src+i*stride))[2]=
1948
        ((uint32_t*)(src+i*stride))[3]= dc;
1949
    }
1950
}
1951

    
1952
void ff_pred16x16_top_dc_c(uint8_t *src, int stride){
1953
    int i, dc=0;
1954

    
1955
    for(i=0;i<16; i++){
1956
        dc+= src[i-stride];
1957
    }
1958
    dc= 0x01010101*((dc + 8)>>4);
1959

    
1960
    for(i=0; i<16; i++){
1961
        ((uint32_t*)(src+i*stride))[0]=
1962
        ((uint32_t*)(src+i*stride))[1]=
1963
        ((uint32_t*)(src+i*stride))[2]=
1964
        ((uint32_t*)(src+i*stride))[3]= dc;
1965
    }
1966
}
1967

    
1968
void ff_pred16x16_128_dc_c(uint8_t *src, int stride){
1969
    int i;
1970

    
1971
    for(i=0; i<16; i++){
1972
        ((uint32_t*)(src+i*stride))[0]=
1973
        ((uint32_t*)(src+i*stride))[1]=
1974
        ((uint32_t*)(src+i*stride))[2]=
1975
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1976
    }
1977
}
1978

    
1979
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1980
  int i, j, k;
1981
  int a;
1982
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
1983
  const uint8_t * const src0 = src+7-stride;
1984
  const uint8_t *src1 = src+8*stride-1;
1985
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1986
  int H = src0[1] - src0[-1];
1987
  int V = src1[0] - src2[ 0];
1988
  for(k=2; k<=8; ++k) {
1989
    src1 += stride; src2 -= stride;
1990
    H += k*(src0[k] - src0[-k]);
1991
    V += k*(src1[0] - src2[ 0]);
1992
  }
1993
  if(svq3){
1994
    H = ( 5*(H/4) ) / 16;
1995
    V = ( 5*(V/4) ) / 16;
1996

    
1997
    /* required for 100% accuracy */
1998
    i = H; H = V; V = i;
1999
  }else{
2000
    H = ( 5*H+32 ) >> 6;
2001
    V = ( 5*V+32 ) >> 6;
2002
  }
2003

    
2004
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2005
  for(j=16; j>0; --j) {
2006
    int b = a;
2007
    a += V;
2008
    for(i=-16; i<0; i+=4) {
2009
      src[16+i] = cm[ (b    ) >> 5 ];
2010
      src[17+i] = cm[ (b+  H) >> 5 ];
2011
      src[18+i] = cm[ (b+2*H) >> 5 ];
2012
      src[19+i] = cm[ (b+3*H) >> 5 ];
2013
      b += 4*H;
2014
    }
2015
    src += stride;
2016
  }
2017
}
2018

    
2019
void ff_pred16x16_plane_c(uint8_t *src, int stride){
2020
    pred16x16_plane_compat_c(src, stride, 0);
2021
}
2022

    
2023
void ff_pred8x8_vertical_c(uint8_t *src, int stride){
2024
    int i;
2025
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2026
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2027

    
2028
    for(i=0; i<8; i++){
2029
        ((uint32_t*)(src+i*stride))[0]= a;
2030
        ((uint32_t*)(src+i*stride))[1]= b;
2031
    }
2032
}
2033

    
2034
void ff_pred8x8_horizontal_c(uint8_t *src, int stride){
2035
    int i;
2036

    
2037
    for(i=0; i<8; i++){
2038
        ((uint32_t*)(src+i*stride))[0]=
2039
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2040
    }
2041
}
2042

    
2043
void ff_pred8x8_128_dc_c(uint8_t *src, int stride){
2044
    int i;
2045

    
2046
    for(i=0; i<8; i++){
2047
        ((uint32_t*)(src+i*stride))[0]=
2048
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2049
    }
2050
}
2051

    
2052
void ff_pred8x8_left_dc_c(uint8_t *src, int stride){
2053
    int i;
2054
    int dc0, dc2;
2055

    
2056
    dc0=dc2=0;
2057
    for(i=0;i<4; i++){
2058
        dc0+= src[-1+i*stride];
2059
        dc2+= src[-1+(i+4)*stride];
2060
    }
2061
    dc0= 0x01010101*((dc0 + 2)>>2);
2062
    dc2= 0x01010101*((dc2 + 2)>>2);
2063

    
2064
    for(i=0; i<4; i++){
2065
        ((uint32_t*)(src+i*stride))[0]=
2066
        ((uint32_t*)(src+i*stride))[1]= dc0;
2067
    }
2068
    for(i=4; i<8; i++){
2069
        ((uint32_t*)(src+i*stride))[0]=
2070
        ((uint32_t*)(src+i*stride))[1]= dc2;
2071
    }
2072
}
2073

    
2074
void ff_pred8x8_top_dc_c(uint8_t *src, int stride){
2075
    int i;
2076
    int dc0, dc1;
2077

    
2078
    dc0=dc1=0;
2079
    for(i=0;i<4; i++){
2080
        dc0+= src[i-stride];
2081
        dc1+= src[4+i-stride];
2082
    }
2083
    dc0= 0x01010101*((dc0 + 2)>>2);
2084
    dc1= 0x01010101*((dc1 + 2)>>2);
2085

    
2086
    for(i=0; i<4; i++){
2087
        ((uint32_t*)(src+i*stride))[0]= dc0;
2088
        ((uint32_t*)(src+i*stride))[1]= dc1;
2089
    }
2090
    for(i=4; i<8; i++){
2091
        ((uint32_t*)(src+i*stride))[0]= dc0;
2092
        ((uint32_t*)(src+i*stride))[1]= dc1;
2093
    }
2094
}
2095

    
2096

    
2097
void ff_pred8x8_dc_c(uint8_t *src, int stride){
2098
    int i;
2099
    int dc0, dc1, dc2, dc3;
2100

    
2101
    dc0=dc1=dc2=0;
2102
    for(i=0;i<4; i++){
2103
        dc0+= src[-1+i*stride] + src[i-stride];
2104
        dc1+= src[4+i-stride];
2105
        dc2+= src[-1+(i+4)*stride];
2106
    }
2107
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2108
    dc0= 0x01010101*((dc0 + 4)>>3);
2109
    dc1= 0x01010101*((dc1 + 2)>>2);
2110
    dc2= 0x01010101*((dc2 + 2)>>2);
2111

    
2112
    for(i=0; i<4; i++){
2113
        ((uint32_t*)(src+i*stride))[0]= dc0;
2114
        ((uint32_t*)(src+i*stride))[1]= dc1;
2115
    }
2116
    for(i=4; i<8; i++){
2117
        ((uint32_t*)(src+i*stride))[0]= dc2;
2118
        ((uint32_t*)(src+i*stride))[1]= dc3;
2119
    }
2120
}
2121

    
2122
void ff_pred8x8_plane_c(uint8_t *src, int stride){
2123
  int j, k;
2124
  int a;
2125
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2126
  const uint8_t * const src0 = src+3-stride;
2127
  const uint8_t *src1 = src+4*stride-1;
2128
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2129
  int H = src0[1] - src0[-1];
2130
  int V = src1[0] - src2[ 0];
2131
  for(k=2; k<=4; ++k) {
2132
    src1 += stride; src2 -= stride;
2133
    H += k*(src0[k] - src0[-k]);
2134
    V += k*(src1[0] - src2[ 0]);
2135
  }
2136
  H = ( 17*H+16 ) >> 5;
2137
  V = ( 17*V+16 ) >> 5;
2138

    
2139
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2140
  for(j=8; j>0; --j) {
2141
    int b = a;
2142
    a += V;
2143
    src[0] = cm[ (b    ) >> 5 ];
2144
    src[1] = cm[ (b+  H) >> 5 ];
2145
    src[2] = cm[ (b+2*H) >> 5 ];
2146
    src[3] = cm[ (b+3*H) >> 5 ];
2147
    src[4] = cm[ (b+4*H) >> 5 ];
2148
    src[5] = cm[ (b+5*H) >> 5 ];
2149
    src[6] = cm[ (b+6*H) >> 5 ];
2150
    src[7] = cm[ (b+7*H) >> 5 ];
2151
    src += stride;
2152
  }
2153
}
2154

    
2155
#define SRC(x,y) src[(x)+(y)*stride]
2156
#define PL(y) \
2157
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2158
#define PREDICT_8x8_LOAD_LEFT \
2159
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2160
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2161
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2162
    const int l7 av_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2163

    
2164
#define PT(x) \
2165
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2166
#define PREDICT_8x8_LOAD_TOP \
2167
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2168
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2169
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2170
    const int t7 av_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2171
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2172

    
2173
#define PTR(x) \
2174
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2175
#define PREDICT_8x8_LOAD_TOPRIGHT \
2176
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2177
    if(has_topright) { \
2178
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2179
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2180
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2181

    
2182
#define PREDICT_8x8_LOAD_TOPLEFT \
2183
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2184

    
2185
#define PREDICT_8x8_DC(v) \
2186
    int y; \
2187
    for( y = 0; y < 8; y++ ) { \
2188
        ((uint32_t*)src)[0] = \
2189
        ((uint32_t*)src)[1] = v; \
2190
        src += stride; \
2191
    }
2192

    
2193
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2194
{
2195
    PREDICT_8x8_DC(0x80808080);
2196
}
2197
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2198
{
2199
    PREDICT_8x8_LOAD_LEFT;
2200
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2201
    PREDICT_8x8_DC(dc);
2202
}
2203
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2204
{
2205
    PREDICT_8x8_LOAD_TOP;
2206
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2207
    PREDICT_8x8_DC(dc);
2208
}
2209
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2210
{
2211
    PREDICT_8x8_LOAD_LEFT;
2212
    PREDICT_8x8_LOAD_TOP;
2213
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2214
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2215
    PREDICT_8x8_DC(dc);
2216
}
2217
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2218
{
2219
    PREDICT_8x8_LOAD_LEFT;
2220
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2221
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2222
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2223
#undef ROW
2224
}
2225
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2226
{
2227
    int y;
2228
    PREDICT_8x8_LOAD_TOP;
2229
    src[0] = t0;
2230
    src[1] = t1;
2231
    src[2] = t2;
2232
    src[3] = t3;
2233
    src[4] = t4;
2234
    src[5] = t5;
2235
    src[6] = t6;
2236
    src[7] = t7;
2237
    for( y = 1; y < 8; y++ )
2238
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2239
}
2240
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2241
{
2242
    PREDICT_8x8_LOAD_TOP;
2243
    PREDICT_8x8_LOAD_TOPRIGHT;
2244
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2245
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2246
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2247
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2248
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2249
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2250
    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;
2251
    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;
2252
    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;
2253
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2254
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2255
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2256
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2257
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2258
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2259
}
2260
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2261
{
2262
    PREDICT_8x8_LOAD_TOP;
2263
    PREDICT_8x8_LOAD_LEFT;
2264
    PREDICT_8x8_LOAD_TOPLEFT;
2265
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2266
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2267
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2268
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2269
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2270
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2271
    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;
2272
    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;
2273
    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;
2274
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2275
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2276
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2277
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2278
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2279
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2280

    
2281
}
2282
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2283
{
2284
    PREDICT_8x8_LOAD_TOP;
2285
    PREDICT_8x8_LOAD_LEFT;
2286
    PREDICT_8x8_LOAD_TOPLEFT;
2287
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2288
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2289
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2290
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2291
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2292
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2293
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2294
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2295
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2296
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2297
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2298
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2299
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2300
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2301
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2302
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2303
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2304
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2305
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2306
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2307
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2308
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2309
}
2310
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2311
{
2312
    PREDICT_8x8_LOAD_TOP;
2313
    PREDICT_8x8_LOAD_LEFT;
2314
    PREDICT_8x8_LOAD_TOPLEFT;
2315
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2316
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2317
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2318
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2319
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2320
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2321
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2322
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2323
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2324
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2325
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2326
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2327
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2328
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2329
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2330
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2331
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2332
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2333
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2334
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2335
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2336
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2337
}
2338
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2339
{
2340
    PREDICT_8x8_LOAD_TOP;
2341
    PREDICT_8x8_LOAD_TOPRIGHT;
2342
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2343
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2344
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2345
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2346
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2347
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2348
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2349
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2350
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2351
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2352
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2353
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2354
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2355
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2356
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2357
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2358
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2359
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2360
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2361
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2362
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2363
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2364
}
2365
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2366
{
2367
    PREDICT_8x8_LOAD_LEFT;
2368
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2369
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2370
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2371
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2372
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2373
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2374
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2375
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2376
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2377
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2378
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2379
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2380
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2381
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2382
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2383
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2384
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2385
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2386
}
2387
#undef PREDICT_8x8_LOAD_LEFT
2388
#undef PREDICT_8x8_LOAD_TOP
2389
#undef PREDICT_8x8_LOAD_TOPLEFT
2390
#undef PREDICT_8x8_LOAD_TOPRIGHT
2391
#undef PREDICT_8x8_DC
2392
#undef PTR
2393
#undef PT
2394
#undef PL
2395
#undef SRC
2396

    
2397
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2398
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2399
                           int src_x_offset, int src_y_offset,
2400
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2401
    MpegEncContext * const s = &h->s;
2402
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2403
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2404
    const int luma_xy= (mx&3) + ((my&3)<<2);
2405
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2406
    uint8_t * src_cb, * src_cr;
2407
    int extra_width= h->emu_edge_width;
2408
    int extra_height= h->emu_edge_height;
2409
    int emu=0;
2410
    const int full_mx= mx>>2;
2411
    const int full_my= my>>2;
2412
    const int pic_width  = 16*s->mb_width;
2413
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
2414

    
2415
    if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
2416
        return;
2417

    
2418
    if(mx&7) extra_width -= 3;
2419
    if(my&7) extra_height -= 3;
2420

    
2421
    if(   full_mx < 0-extra_width
2422
       || full_my < 0-extra_height
2423
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2424
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2425
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
2426
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2427
        emu=1;
2428
    }
2429

    
2430
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2431
    if(!square){
2432
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2433
    }
2434

    
2435
    if(s->flags&CODEC_FLAG_GRAY) return;
2436

    
2437
    if(MB_MBAFF){
2438
        // chroma offset when predicting from a field of opposite parity
2439
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2440
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2441
    }
2442
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2443
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2444

    
2445
    if(emu){
2446
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2447
            src_cb= s->edge_emu_buffer;
2448
    }
2449
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2450

    
2451
    if(emu){
2452
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2453
            src_cr= s->edge_emu_buffer;
2454
    }
2455
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2456
}
2457

    
2458
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2459
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2460
                           int x_offset, int y_offset,
2461
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2462
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2463
                           int list0, int list1){
2464
    MpegEncContext * const s = &h->s;
2465
    qpel_mc_func *qpix_op=  qpix_put;
2466
    h264_chroma_mc_func chroma_op= chroma_put;
2467

    
2468
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2469
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2470
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2471
    x_offset += 8*s->mb_x;
2472
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2473

    
2474
    if(list0){
2475
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2476
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2477
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2478
                           qpix_op, chroma_op);
2479

    
2480
        qpix_op=  qpix_avg;
2481
        chroma_op= chroma_avg;
2482
    }
2483

    
2484
    if(list1){
2485
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2486
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2487
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2488
                           qpix_op, chroma_op);
2489
    }
2490
}
2491

    
2492
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2493
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2494
                           int x_offset, int y_offset,
2495
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2496
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2497
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2498
                           int list0, int list1){
2499
    MpegEncContext * const s = &h->s;
2500

    
2501
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2502
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2503
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2504
    x_offset += 8*s->mb_x;
2505
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2506

    
2507
    if(list0 && list1){
2508
        /* don't optimize for luma-only case, since B-frames usually
2509
         * use implicit weights => chroma too. */
2510
        uint8_t *tmp_cb = s->obmc_scratchpad;
2511
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2512
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2513
        int refn0 = h->ref_cache[0][ scan8[n] ];
2514
        int refn1 = h->ref_cache[1][ scan8[n] ];
2515

    
2516
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2517
                    dest_y, dest_cb, dest_cr,
2518
                    x_offset, y_offset, qpix_put, chroma_put);
2519
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2520
                    tmp_y, tmp_cb, tmp_cr,
2521
                    x_offset, y_offset, qpix_put, chroma_put);
2522

    
2523
        if(h->use_weight == 2){
2524
            int weight0 = h->implicit_weight[refn0][refn1];
2525
            int weight1 = 64 - weight0;
2526
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
2527
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2528
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2529
        }else{
2530
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2531
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2532
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2533
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2534
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2535
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2536
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2537
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2538
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2539
        }
2540
    }else{
2541
        int list = list1 ? 1 : 0;
2542
        int refn = h->ref_cache[list][ scan8[n] ];
2543
        Picture *ref= &h->ref_list[list][refn];
2544
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2545
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2546
                    qpix_put, chroma_put);
2547

    
2548
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
2549
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2550
        if(h->use_weight_chroma){
2551
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2552
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2553
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2554
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2555
        }
2556
    }
2557
}
2558

    
2559
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2560
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2561
                           int x_offset, int y_offset,
2562
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2563
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2564
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2565
                           int list0, int list1){
2566
    if((h->use_weight==2 && list0 && list1
2567
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2568
       || h->use_weight==1)
2569
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2570
                         x_offset, y_offset, qpix_put, chroma_put,
2571
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2572
    else
2573
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2574
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2575
}
2576

    
2577
static inline void prefetch_motion(H264Context *h, int list){
2578
    /* fetch pixels for estimated mv 4 macroblocks ahead
2579
     * optimized for 64byte cache lines */
2580
    MpegEncContext * const s = &h->s;
2581
    const int refn = h->ref_cache[list][scan8[0]];
2582
    if(refn >= 0){
2583
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2584
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2585
        uint8_t **src= h->ref_list[list][refn].data;
2586
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
2587
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
2588
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2589
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2590
    }
2591
}
2592

    
2593
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2594
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2595
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2596
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2597
    MpegEncContext * const s = &h->s;
2598
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2599
    const int mb_type= s->current_picture.mb_type[mb_xy];
2600

    
2601
    assert(IS_INTER(mb_type));
2602

    
2603
    prefetch_motion(h, 0);
2604

    
2605
    if(IS_16X16(mb_type)){
2606
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2607
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2608
                &weight_op[0], &weight_avg[0],
2609
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2610
    }else if(IS_16X8(mb_type)){
2611
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2612
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2613
                &weight_op[1], &weight_avg[1],
2614
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2615
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2616
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2617
                &weight_op[1], &weight_avg[1],
2618
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2619
    }else if(IS_8X16(mb_type)){
2620
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
2621
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2622
                &weight_op[2], &weight_avg[2],
2623
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2624
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
2625
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2626
                &weight_op[2], &weight_avg[2],
2627
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2628
    }else{
2629
        int i;
2630

    
2631
        assert(IS_8X8(mb_type));
2632

    
2633
        for(i=0; i<4; i++){
2634
            const int sub_mb_type= h->sub_mb_type[i];
2635
            const int n= 4*i;
2636
            int x_offset= (i&1)<<2;
2637
            int y_offset= (i&2)<<1;
2638

    
2639
            if(IS_SUB_8X8(sub_mb_type)){
2640
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2641
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2642
                    &weight_op[3], &weight_avg[3],
2643
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2644
            }else if(IS_SUB_8X4(sub_mb_type)){
2645
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2646
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2647
                    &weight_op[4], &weight_avg[4],
2648
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2649
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2650
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2651
                    &weight_op[4], &weight_avg[4],
2652
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2653
            }else if(IS_SUB_4X8(sub_mb_type)){
2654
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2655
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2656
                    &weight_op[5], &weight_avg[5],
2657
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2658
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2659
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2660
                    &weight_op[5], &weight_avg[5],
2661
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2662
            }else{
2663
                int j;
2664
                assert(IS_SUB_4X4(sub_mb_type));
2665
                for(j=0; j<4; j++){
2666
                    int sub_x_offset= x_offset + 2*(j&1);
2667
                    int sub_y_offset= y_offset +   (j&2);
2668
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2669
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2670
                        &weight_op[6], &weight_avg[6],
2671
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2672
                }
2673
            }
2674
        }
2675
    }
2676

    
2677
    prefetch_motion(h, 1);
2678
}
2679

    
2680
static void decode_init_vlc(void){
2681
    static int done = 0;
2682

    
2683
    if (!done) {
2684
        int i;
2685
        done = 1;
2686

    
2687
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2688
                 &chroma_dc_coeff_token_len [0], 1, 1,
2689
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2690

    
2691
        for(i=0; i<4; i++){
2692
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2693
                     &coeff_token_len [i][0], 1, 1,
2694
                     &coeff_token_bits[i][0], 1, 1, 1);
2695
        }
2696

    
2697
        for(i=0; i<3; i++){
2698
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2699
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2700
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2701
        }
2702
        for(i=0; i<15; i++){
2703
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2704
                     &total_zeros_len [i][0], 1, 1,
2705
                     &total_zeros_bits[i][0], 1, 1, 1);
2706
        }
2707

    
2708
        for(i=0; i<6; i++){
2709
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2710
                     &run_len [i][0], 1, 1,
2711
                     &run_bits[i][0], 1, 1, 1);
2712
        }
2713
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2714
                 &run_len [6][0], 1, 1,
2715
                 &run_bits[6][0], 1, 1, 1);
2716
    }
2717
}
2718

    
2719
/**
2720
 * Sets the intra prediction function pointers.
2721
 */
2722
static void init_pred_ptrs(H264Context *h){
2723
//    MpegEncContext * const s = &h->s;
2724

    
2725
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2726
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2727
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2728
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2729
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2730
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2731
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2732
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2733
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2734
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2735
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2736
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2737

    
2738
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2739
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2740
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2741
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2742
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2743
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2744
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2745
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2746
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2747
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2748
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2749
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2750

    
2751
    h->pred8x8[DC_PRED8x8     ]= ff_pred8x8_dc_c;
2752
    h->pred8x8[VERT_PRED8x8   ]= ff_pred8x8_vertical_c;
2753
    h->pred8x8[HOR_PRED8x8    ]= ff_pred8x8_horizontal_c;
2754
    h->pred8x8[PLANE_PRED8x8  ]= ff_pred8x8_plane_c;
2755
    h->pred8x8[LEFT_DC_PRED8x8]= ff_pred8x8_left_dc_c;
2756
    h->pred8x8[TOP_DC_PRED8x8 ]= ff_pred8x8_top_dc_c;
2757
    h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c;
2758

    
2759
    h->pred16x16[DC_PRED8x8     ]= ff_pred16x16_dc_c;
2760
    h->pred16x16[VERT_PRED8x8   ]= ff_pred16x16_vertical_c;
2761
    h->pred16x16[HOR_PRED8x8    ]= ff_pred16x16_horizontal_c;
2762
    h->pred16x16[PLANE_PRED8x8  ]= ff_pred16x16_plane_c;
2763
    h->pred16x16[LEFT_DC_PRED8x8]= ff_pred16x16_left_dc_c;
2764
    h->pred16x16[TOP_DC_PRED8x8 ]= ff_pred16x16_top_dc_c;
2765
    h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c;
2766
}
2767

    
2768
static void free_tables(H264Context *h){
2769
    int i;
2770
    av_freep(&h->intra4x4_pred_mode);
2771
    av_freep(&h->chroma_pred_mode_table);
2772
    av_freep(&h->cbp_table);
2773
    av_freep(&h->mvd_table[0]);
2774
    av_freep(&h->mvd_table[1]);
2775
    av_freep(&h->direct_table);
2776
    av_freep(&h->non_zero_count);
2777
    av_freep(&h->slice_table_base);
2778
    av_freep(&h->top_borders[1]);
2779
    av_freep(&h->top_borders[0]);
2780
    h->slice_table= NULL;
2781

    
2782
    av_freep(&h->mb2b_xy);
2783
    av_freep(&h->mb2b8_xy);
2784

    
2785
    av_freep(&h->s.obmc_scratchpad);
2786

    
2787
    for(i = 0; i < MAX_SPS_COUNT; i++)
2788
        av_freep(h->sps_buffers + i);
2789

    
2790
    for(i = 0; i < MAX_PPS_COUNT; i++)
2791
        av_freep(h->pps_buffers + i);
2792
}
2793

    
2794
static void init_dequant8_coeff_table(H264Context *h){
2795
    int i,q,x;
2796
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2797
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2798
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2799

    
2800
    for(i=0; i<2; i++ ){
2801
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2802
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2803
            break;
2804
        }
2805

    
2806
        for(q=0; q<52; q++){
2807
            int shift = ff_div6[q];
2808
            int idx = ff_rem6[q];
2809
            for(x=0; x<64; x++)
2810
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2811
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2812
                    h->pps.scaling_matrix8[i][x]) << shift;
2813
        }
2814
    }
2815
}
2816

    
2817
static void init_dequant4_coeff_table(H264Context *h){
2818
    int i,j,q,x;
2819
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2820
    for(i=0; i<6; i++ ){
2821
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2822
        for(j=0; j<i; j++){
2823
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2824
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2825
                break;
2826
            }
2827
        }
2828
        if(j<i)
2829
            continue;
2830

    
2831
        for(q=0; q<52; q++){
2832
            int shift = ff_div6[q] + 2;
2833
            int idx = ff_rem6[q];
2834
            for(x=0; x<16; x++)
2835
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2836
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2837
                    h->pps.scaling_matrix4[i][x]) << shift;
2838
        }
2839
    }
2840
}
2841

    
2842
static void init_dequant_tables(H264Context *h){
2843
    int i,x;
2844
    init_dequant4_coeff_table(h);
2845
    if(h->pps.transform_8x8_mode)
2846
        init_dequant8_coeff_table(h);
2847
    if(h->sps.transform_bypass){
2848
        for(i=0; i<6; i++)
2849
            for(x=0; x<16; x++)
2850
                h->dequant4_coeff[i][0][x] = 1<<6;
2851
        if(h->pps.transform_8x8_mode)
2852
            for(i=0; i<2; i++)
2853
                for(x=0; x<64; x++)
2854
                    h->dequant8_coeff[i][0][x] = 1<<6;
2855
    }
2856
}
2857

    
2858

    
2859
/**
2860
 * allocates tables.
2861
 * needs width/height
2862
 */
2863
static int alloc_tables(H264Context *h){
2864
    MpegEncContext * const s = &h->s;
2865
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2866
    int x,y;
2867

    
2868
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2869

    
2870
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2871
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2872
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2873
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2874
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2875

    
2876
    if( h->pps.cabac ) {
2877
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2878
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2879
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2880
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2881
    }
2882

    
2883
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2884
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2885

    
2886
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2887
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2888
    for(y=0; y<s->mb_height; y++){
2889
        for(x=0; x<s->mb_width; x++){
2890
            const int mb_xy= x + y*s->mb_stride;
2891
            const int b_xy = 4*x + 4*y*h->b_stride;
2892
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2893

    
2894
            h->mb2b_xy [mb_xy]= b_xy;
2895
            h->mb2b8_xy[mb_xy]= b8_xy;
2896
        }
2897
    }
2898

    
2899
    s->obmc_scratchpad = NULL;
2900

    
2901
    if(!h->dequant4_coeff[0])
2902
        init_dequant_tables(h);
2903

    
2904
    return 0;
2905
fail:
2906
    free_tables(h);
2907
    return -1;
2908
}
2909

    
2910
static void common_init(H264Context *h){
2911
    MpegEncContext * const s = &h->s;
2912

    
2913
    s->width = s->avctx->width;
2914
    s->height = s->avctx->height;
2915
    s->codec_id= s->avctx->codec->id;
2916

    
2917
    init_pred_ptrs(h);
2918

    
2919
    h->dequant_coeff_pps= -1;
2920
    s->unrestricted_mv=1;
2921
    s->decode=1; //FIXME
2922

    
2923
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2924
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2925
}
2926

    
2927
static int decode_init(AVCodecContext *avctx){
2928
    H264Context *h= avctx->priv_data;
2929
    MpegEncContext * const s = &h->s;
2930

    
2931
    MPV_decode_defaults(s);
2932

    
2933
    s->avctx = avctx;
2934
    common_init(h);
2935

    
2936
    s->out_format = FMT_H264;
2937
    s->workaround_bugs= avctx->workaround_bugs;
2938

    
2939
    // set defaults
2940
//    s->decode_mb= ff_h263_decode_mb;
2941
    s->low_delay= 1;
2942
    avctx->pix_fmt= PIX_FMT_YUV420P;
2943

    
2944
    decode_init_vlc();
2945

    
2946
    if(avctx->extradata_size > 0 && avctx->extradata &&
2947
       *(char *)avctx->extradata == 1){
2948
        h->is_avc = 1;
2949
        h->got_avcC = 0;
2950
    } else {
2951
        h->is_avc = 0;
2952
    }
2953

    
2954
    return 0;
2955
}
2956

    
2957
static int frame_start(H264Context *h){
2958
    MpegEncContext * const s = &h->s;
2959
    int i;
2960

    
2961
    if(MPV_frame_start(s, s->avctx) < 0)
2962
        return -1;
2963
    ff_er_frame_start(s);
2964

    
2965
    assert(s->linesize && s->uvlinesize);
2966

    
2967
    for(i=0; i<16; i++){
2968
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2969
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2970
    }
2971
    for(i=0; i<4; i++){
2972
        h->block_offset[16+i]=
2973
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2974
        h->block_offset[24+16+i]=
2975
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2976
    }
2977

    
2978
    /* can't be in alloc_tables because linesize isn't known there.
2979
     * FIXME: redo bipred weight to not require extra buffer? */
2980
    if(!s->obmc_scratchpad)
2981
        s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2982

    
2983
    /* some macroblocks will be accessed before they're available */
2984
    if(FRAME_MBAFF)
2985
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2986

    
2987
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2988
    return 0;
2989
}
2990

    
2991
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
2992
    MpegEncContext * const s = &h->s;
2993
    int i;
2994

    
2995
    src_y  -=   linesize;
2996
    src_cb -= uvlinesize;
2997
    src_cr -= uvlinesize;
2998

    
2999
    // There are two lines saved, the line above the the top macroblock of a pair,
3000
    // and the line above the bottom macroblock
3001
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3002
    for(i=1; i<17; i++){
3003
        h->left_border[i]= src_y[15+i*  linesize];
3004
    }
3005

    
3006
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3007
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3008

    
3009
    if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3010
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3011
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3012
        for(i=1; i<9; i++){
3013
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3014
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3015
        }
3016
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3017
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3018
    }
3019
}
3020

    
3021
static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
3022
    MpegEncContext * const s = &h->s;
3023
    int temp8, i;
3024
    uint64_t temp64;
3025
    int deblock_left;
3026
    int deblock_top;
3027
    int mb_xy;
3028

    
3029
    if(h->deblocking_filter == 2) {
3030
        mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3031
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
3032
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
3033
    } else {
3034
        deblock_left = (s->mb_x > 0);
3035
        deblock_top =  (s->mb_y > 0);
3036
    }
3037

    
3038
    src_y  -=   linesize + 1;
3039
    src_cb -= uvlinesize + 1;
3040
    src_cr -= uvlinesize + 1;
3041

    
3042
#define XCHG(a,b,t,xchg)\
3043
t= a;\
3044
if(xchg)\
3045
    a= b;\
3046
b= t;
3047

    
3048
    if(deblock_left){
3049
        for(i = !deblock_top; i<17; i++){
3050
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3051
        }
3052
    }
3053

    
3054
    if(deblock_top){
3055
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3056
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3057
        if(s->mb_x+1 < s->mb_width){
3058
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3059
        }
3060
    }
3061

    
3062
    if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3063
        if(deblock_left){
3064
            for(i = !deblock_top; i<9; i++){
3065
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3066
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3067
            }
3068
        }
3069
        if(deblock_top){
3070
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3071
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3072
        }
3073
    }
3074
}
3075

    
3076
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3077
    MpegEncContext * const s = &h->s;
3078
    int i;
3079

    
3080
    src_y  -= 2 *   linesize;
3081
    src_cb -= 2 * uvlinesize;
3082
    src_cr -= 2 * uvlinesize;
3083

    
3084
    // There are two lines saved, the line above the the top macroblock of a pair,
3085
    // and the line above the bottom macroblock
3086
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3087
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3088
    for(i=2; i<34; i++){
3089
        h->left_border[i]= src_y[15+i*  linesize];
3090
    }
3091

    
3092
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3093
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3094
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3095
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3096

    
3097
    if(!(s->flags&CODEC_FLAG_GRAY)){
3098
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3099
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3100
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3101
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3102
        for(i=2; i<18; i++){
3103
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3104
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3105
        }
3106
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3107
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3108
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3109
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3110
    }
3111
}
3112

    
3113
static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
3114
    MpegEncContext * const s = &h->s;
3115
    int temp8, i;
3116
    uint64_t temp64;
3117
    int deblock_left = (s->mb_x > 0);
3118
    int deblock_top  = (s->mb_y > 1);
3119

    
3120
    tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
3121

    
3122
    src_y  -= 2 *   linesize + 1;
3123
    src_cb -= 2 * uvlinesize + 1;
3124
    src_cr -= 2 * uvlinesize + 1;
3125

    
3126
#define XCHG(a,b,t,xchg)\
3127
t= a;\
3128
if(xchg)\
3129
    a= b;\
3130
b= t;
3131

    
3132
    if(deblock_left){
3133
        for(i = (!deblock_top)<<1; i<34; i++){
3134
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3135
        }
3136
    }
3137

    
3138
    if(deblock_top){
3139
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3140
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3141
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3142
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3143
        if(s->mb_x+1 < s->mb_width){
3144
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3145
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3146
        }
3147
    }
3148

    
3149
    if(!(s->flags&CODEC_FLAG_GRAY)){
3150
        if(deblock_left){
3151
            for(i = (!deblock_top) << 1; i<18; i++){
3152
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3153
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3154
            }
3155
        }
3156
        if(deblock_top){
3157
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3158
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3159
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3160
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3161
        }
3162
    }
3163
}
3164

    
3165
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
3166
    MpegEncContext * const s = &h->s;
3167
    const int mb_x= s->mb_x;
3168
    const int mb_y= s->mb_y;
3169
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3170
    const int mb_type= s->current_picture.mb_type[mb_xy];
3171
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3172
    int linesize, uvlinesize /*dct_offset*/;
3173
    int i;
3174
    int *block_offset = &h->block_offset[0];
3175
    const unsigned int bottom = mb_y & 1;
3176
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
3177
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3178
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3179

    
3180
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3181
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3182
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3183

    
3184
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
3185
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
3186

    
3187
    if (!simple && MB_FIELD) {
3188
        linesize   = h->mb_linesize   = s->linesize * 2;
3189
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3190
        block_offset = &h->block_offset[24];
3191
        if(mb_y&1){ //FIXME move out of this func?
3192
            dest_y -= s->linesize*15;
3193
            dest_cb-= s->uvlinesize*7;
3194
            dest_cr-= s->uvlinesize*7;
3195
        }
3196
        if(FRAME_MBAFF) {
3197
            int list;
3198
            for(list=0; list<h->list_count; list++){
3199
                if(!USES_LIST(mb_type, list))
3200
                    continue;
3201
                if(IS_16X16(mb_type)){
3202
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
3203
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3204
                }else{
3205
                    for(i=0; i<16; i+=4){
3206
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3207
                        int ref = h->ref_cache[list][scan8[i]];
3208
                        if(ref >= 0)
3209
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3210
                    }
3211
                }
3212
            }
3213
        }
3214
    } else {
3215
        linesize   = h->mb_linesize   = s->linesize;
3216
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3217
//        dct_offset = s->linesize * 16;
3218
    }
3219

    
3220
    if(transform_bypass){
3221
        idct_dc_add =
3222
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3223
    }else if(IS_8x8DCT(mb_type)){
3224
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3225
        idct_add = s->dsp.h264_idct8_add;
3226
    }else{
3227
        idct_dc_add = s->dsp.h264_idct_dc_add;
3228
        idct_add = s->dsp.h264_idct_add;
3229
    }
3230

    
3231
    if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3232
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3233
        int mbt_y = mb_y&~1;
3234
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
3235
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3236
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3237
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3238
    }
3239

    
3240
    if (!simple && IS_INTRA_PCM(mb_type)) {
3241
        unsigned int x, y;
3242

    
3243
        // The pixels are stored in h->mb array in the same order as levels,
3244
        // copy them in output in the correct order.
3245
        for(i=0; i<16; i++) {
3246
            for (y=0; y<4; y++) {
3247
                for (x=0; x<4; x++) {
3248
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3249
                }
3250
            }
3251
        }
3252
        for(i=16; i<16+4; i++) {
3253
            for (y=0; y<4; y++) {
3254
                for (x=0; x<4; x++) {
3255
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3256
                }
3257
            }
3258
        }
3259
        for(i=20; i<20+4; i++) {
3260
            for (y=0; y<4; y++) {
3261
                for (x=0; x<4; x++) {
3262
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3263
                }
3264
            }
3265
        }
3266
    } else {
3267
        if(IS_INTRA(mb_type)){
3268
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3269
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
3270

    
3271
            if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3272
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3273
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3274
            }
3275

    
3276
            if(IS_INTRA4x4(mb_type)){
3277
                if(simple || !s->encoding){
3278
                    if(IS_8x8DCT(mb_type)){
3279
                        for(i=0; i<16; i+=4){
3280
                            uint8_t * const ptr= dest_y + block_offset[i];
3281
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3282
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3283
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3284
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
3285
                            if(nnz){
3286
                                if(nnz == 1 && h->mb[i*16])
3287
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3288
                                else
3289
                                    idct_add(ptr, h->mb + i*16, linesize);
3290
                            }
3291
                        }
3292
                    }else
3293
                    for(i=0; i<16; i++){
3294
                        uint8_t * const ptr= dest_y + block_offset[i];
3295
                        uint8_t *topright;
3296
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3297
                        int nnz, tr;
3298

    
3299
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3300
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3301
                            assert(mb_y || linesize <= block_offset[i]);
3302
                            if(!topright_avail){
3303
                                tr= ptr[3 - linesize]*0x01010101;
3304
                                topright= (uint8_t*) &tr;
3305
                            }else
3306
                                topright= ptr + 4 - linesize;
3307
                        }else
3308
                            topright= NULL;
3309

    
3310
                        h->pred4x4[ dir ](ptr, topright, linesize);
3311
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3312
                        if(nnz){
3313
                            if(is_h264){
3314
                                if(nnz == 1 && h->mb[i*16])
3315
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3316
                                else
3317
                                    idct_add(ptr, h->mb + i*16, linesize);
3318
                            }else
3319
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3320
                        }
3321
                    }
3322
                }
3323
            }else{
3324
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3325
                if(is_h264){
3326
                    if(!transform_bypass)
3327
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3328
                }else
3329
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3330
            }
3331
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3332
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
3333
        }else if(is_h264){
3334
            hl_motion(h, dest_y, dest_cb, dest_cr,
3335
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
3336
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
3337
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3338
        }
3339

    
3340

    
3341
        if(!IS_INTRA4x4(mb_type)){
3342
            if(is_h264){
3343
                if(IS_INTRA16x16(mb_type)){
3344
                    for(i=0; i<16; i++){
3345
                        if(h->non_zero_count_cache[ scan8[i] ])
3346
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3347
                        else if(h->mb[i*16])
3348
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3349
                    }
3350
                }else{
3351
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3352
                    for(i=0; i<16; i+=di){
3353
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3354
                        if(nnz){
3355
                            if(nnz==1 && h->mb[i*16])
3356
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3357
                            else
3358
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3359
                        }
3360
                    }
3361
                }
3362
            }else{
3363
                for(i=0; i<16; i++){
3364
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3365
                        uint8_t * const ptr= dest_y + block_offset[i];
3366
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3367
                    }
3368
                }
3369
            }
3370
        }
3371

    
3372
        if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3373
            uint8_t *dest[2] = {dest_cb, dest_cr};
3374
            if(transform_bypass){
3375
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3376
            }else{
3377
                idct_add = s->dsp.h264_idct_add;
3378
                idct_dc_add = s->dsp.h264_idct_dc_add;
3379
                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp][0]);
3380
                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp][0]);
3381
            }
3382
            if(is_h264){
3383
                for(i=16; i<16+8; i++){
3384
                    if(h->non_zero_count_cache[ scan8[i] ])
3385
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3386
                    else if(h->mb[i*16])
3387
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3388
                }
3389
            }else{
3390
                for(i=16; i<16+8; i++){
3391
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3392
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3393
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3394
                    }
3395
                }
3396
            }
3397
        }
3398
    }
3399
    if(h->deblocking_filter) {
3400
        if (!simple && FRAME_MBAFF) {
3401
            //FIXME try deblocking one mb at a time?
3402
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3403
            const int mb_y = s->mb_y - 1;
3404
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3405
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3406
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3407
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3408
            if (!bottom) return;
3409
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3410
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3411
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3412

    
3413
            if(IS_INTRA(mb_type_top | mb_type_bottom))
3414
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3415

    
3416
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3417
            // deblock a pair
3418
            // top
3419
            s->mb_y--;
3420
            tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3421
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3422
            h->chroma_qp = get_chroma_qp(h, s->current_picture.qscale_table[mb_xy]);
3423
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3424
            // bottom
3425
            s->mb_y++;
3426
            tprintf(h->s.avctx, "call mbaff filter_mb\n");
3427
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3428
            h->chroma_qp = get_chroma_qp(h, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3429
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3430
        } else {
3431
            tprintf(h->s.avctx, "call filter_mb\n");
3432
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
3433
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3434
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3435
        }
3436
    }
3437
}
3438

    
3439
/**
3440
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
3441
 */
3442
static void hl_decode_mb_simple(H264Context *h){
3443
    hl_decode_mb_internal(h, 1);
3444
}
3445

    
3446
/**
3447
 * Process a macroblock; this handles edge cases, such as interlacing.
3448
 */
3449
static void av_noinline hl_decode_mb_complex(H264Context *h){
3450
    hl_decode_mb_internal(h, 0);
3451
}
3452

    
3453
static void hl_decode_mb(H264Context *h){
3454
    MpegEncContext * const s = &h->s;
3455
    const int mb_x= s->mb_x;
3456
    const int mb_y= s->mb_y;
3457
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3458
    const int mb_type= s->current_picture.mb_type[mb_xy];
3459
    int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 || (s->flags&CODEC_FLAG_GRAY) || s->encoding;
3460

    
3461
    if(!s->decode)
3462
        return;
3463

    
3464
    if (is_complex)
3465
        hl_decode_mb_complex(h);
3466
    else hl_decode_mb_simple(h);
3467
}
3468

    
3469
/**
3470
 * fills the default_ref_list.
3471
 */
3472
static int fill_default_ref_list(H264Context *h){
3473
    MpegEncContext * const s = &h->s;
3474
    int i;
3475
    int smallest_poc_greater_than_current = -1;
3476
    Picture sorted_short_ref[32];
3477

    
3478
    if(h->slice_type==B_TYPE){
3479
        int out_i;
3480
        int limit= INT_MIN;
3481

    
3482
        /* sort frame according to poc in B slice */
3483
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3484
            int best_i=INT_MIN;
3485
            int best_poc=INT_MAX;
3486

    
3487
            for(i=0; i<h->short_ref_count; i++){
3488
                const int poc= h->short_ref[i]->poc;
3489
                if(poc > limit && poc < best_poc){
3490
                    best_poc= poc;
3491
                    best_i= i;
3492
                }
3493
            }
3494

    
3495
            assert(best_i != INT_MIN);
3496

    
3497
            limit= best_poc;
3498
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3499
            tprintf(h->s.avctx, "sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
3500
            if (-1 == smallest_poc_greater_than_current) {
3501
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3502
                    smallest_poc_greater_than_current = out_i;
3503
                }
3504
            }
3505
        }
3506
    }
3507

    
3508
    if(s->picture_structure == PICT_FRAME){
3509
        if(h->slice_type==B_TYPE){
3510
            int list;
3511
            tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3512

    
3513
            // find the largest poc
3514
            for(list=0; list<2; list++){
3515
                int index = 0;
3516
                int j= -99;
3517
                int step= list ? -1 : 1;
3518

    
3519
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3520
                    while(j<0 || j>= h->short_ref_count){
3521
                        if(j != -99 && step == (list ? -1 : 1))
3522
                            return -1;
3523
                        step = -step;
3524
                        j= smallest_poc_greater_than_current + (step>>1);
3525
                    }
3526
                    if(sorted_short_ref[j].reference != 3) continue;
3527
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3528
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3529
                }
3530

    
3531
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3532
                    if(h->long_ref[i] == NULL) continue;
3533
                    if(h->long_ref[i]->reference != 3) continue;
3534

    
3535
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3536
                    h->default_ref_list[ list ][index++].pic_id= i;;
3537
                }
3538

    
3539
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3540
                    // swap the two first elements of L1 when
3541
                    // L0 and L1 are identical
3542
                    Picture temp= h->default_ref_list[1][0];
3543
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3544
                    h->default_ref_list[1][1] = temp;
3545
                }
3546

    
3547
                if(index < h->ref_count[ list ])
3548
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3549
            }
3550
        }else{
3551
            int index=0;
3552
            for(i=0; i<h->short_ref_count; i++){
3553
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3554
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3555
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3556
            }
3557
            for(i = 0; i < 16; i++){
3558
                if(h->long_ref[i] == NULL) continue;
3559
                if(h->long_ref[i]->reference != 3) continue;
3560
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3561
                h->default_ref_list[0][index++].pic_id= i;;
3562
            }
3563
            if(index < h->ref_count[0])
3564
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3565
        }
3566
    }else{ //FIELD
3567
        if(h->slice_type==B_TYPE){
3568
        }else{
3569
            //FIXME second field balh
3570
        }
3571
    }
3572
#ifdef TRACE
3573
    for (i=0; i<h->ref_count[0]; i++) {
3574
        tprintf(h->s.avctx, "List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
3575
    }
3576
    if(h->slice_type==B_TYPE){
3577
        for (i=0; i<h->ref_count[1]; i++) {
3578
            tprintf(h->s.avctx, "List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
3579
        }
3580
    }
3581
#endif
3582
    return 0;
3583
}
3584

    
3585
static void print_short_term(H264Context *h);
3586
static void print_long_term(H264Context *h);
3587

    
3588
static int decode_ref_pic_list_reordering(H264Context *h){
3589
    MpegEncContext * const s = &h->s;
3590
    int list, index;
3591

    
3592
    print_short_term(h);
3593
    print_long_term(h);
3594
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3595

    
3596
    for(list=0; list<h->list_count; list++){
3597
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3598

    
3599
        if(get_bits1(&s->gb)){
3600
            int pred= h->curr_pic_num;
3601

    
3602
            for(index=0; ; index++){
3603
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3604
                unsigned int pic_id;
3605
                int i;
3606
                Picture *ref = NULL;
3607

    
3608
                if(reordering_of_pic_nums_idc==3)
3609
                    break;
3610

    
3611
                if(index >= h->ref_count[list]){
3612
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3613
                    return -1;
3614
                }
3615

    
3616
                if(reordering_of_pic_nums_idc<3){
3617
                    if(reordering_of_pic_nums_idc<2){
3618
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3619

    
3620
                        if(abs_diff_pic_num >= h->max_pic_num){
3621
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3622
                            return -1;
3623
                        }
3624

    
3625
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3626
                        else                                pred+= abs_diff_pic_num;
3627
                        pred &= h->max_pic_num - 1;
3628

    
3629
                        for(i= h->short_ref_count-1; i>=0; i--){
3630
                            ref = h->short_ref[i];
3631
                            assert(ref->reference == 3);
3632
                            assert(!ref->long_ref);
3633
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3634
                                break;
3635
                        }
3636
                        if(i>=0)
3637
                            ref->pic_id= ref->frame_num;
3638
                    }else{
3639
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3640
                        if(pic_id>31){
3641
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
3642
                            return -1;
3643
                        }
3644
                        ref = h->long_ref[pic_id];
3645
                        if(ref){
3646
                            ref->pic_id= pic_id;
3647
                            assert(ref->reference == 3);
3648
                            assert(ref->long_ref);
3649
                            i=0;
3650
                        }else{
3651
                            i=-1;
3652
                        }
3653
                    }
3654

    
3655
                    if (i < 0) {
3656
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3657
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3658
                    } else {
3659
                        for(i=index; i+1<h->ref_count[list]; i++){
3660
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3661
                                break;
3662
                        }
3663
                        for(; i > index; i--){
3664
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3665
                        }
3666
                        h->ref_list[list][index]= *ref;
3667
                    }
3668
                }else{
3669
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3670
                    return -1;
3671
                }
3672
            }
3673
        }
3674
    }
3675
    for(list=0; list<h->list_count; list++){
3676
        for(index= 0; index < h->ref_count[list]; index++){
3677
            if(!h->ref_list[list][index].data[0])
3678
                h->ref_list[list][index]= s->current_picture;
3679
        }
3680
    }
3681

    
3682
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3683
        direct_dist_scale_factor(h);
3684
    direct_ref_list_init(h);
3685
    return 0;
3686
}
3687

    
3688
static void fill_mbaff_ref_list(H264Context *h){
3689
    int list, i, j;
3690
    for(list=0; list<2; list++){ //FIXME try list_count
3691
        for(i=0; i<h->ref_count[list]; i++){
3692
            Picture *frame = &h->ref_list[list][i];
3693
            Picture *field = &h->ref_list[list][16+2*i];
3694
            field[0] = *frame;
3695
            for(j=0; j<3; j++)
3696
                field[0].linesize[j] <<= 1;
3697
            field[1] = field[0];
3698
            for(j=0; j<3; j++)
3699
                field[1].data[j] += frame->linesize[j];
3700

    
3701
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
3702
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
3703
            for(j=0; j<2; j++){
3704
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
3705
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
3706
            }
3707
        }
3708
    }
3709
    for(j=0; j<h->ref_count[1]; j++){
3710
        for(i=0; i<h->ref_count[0]; i++)
3711
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
3712
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
3713
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
3714
    }
3715
}
3716

    
3717
static int pred_weight_table(H264Context *h){
3718
    MpegEncContext * const s = &h->s;
3719
    int list, i;
3720
    int luma_def, chroma_def;
3721

    
3722
    h->use_weight= 0;
3723
    h->use_weight_chroma= 0;
3724
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3725
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3726
    luma_def = 1<<h->luma_log2_weight_denom;
3727
    chroma_def = 1<<h->chroma_log2_weight_denom;
3728

    
3729
    for(list=0; list<2; list++){
3730
        for(i=0; i<h->ref_count[list]; i++){
3731
            int luma_weight_flag, chroma_weight_flag;
3732

    
3733
            luma_weight_flag= get_bits1(&s->gb);
3734
            if(luma_weight_flag){
3735
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3736
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3737
                if(   h->luma_weight[list][i] != luma_def
3738
                   || h->luma_offset[list][i] != 0)
3739
                    h->use_weight= 1;
3740
            }else{
3741
                h->luma_weight[list][i]= luma_def;
3742
                h->luma_offset[list][i]= 0;
3743
            }
3744

    
3745
            chroma_weight_flag= get_bits1(&s->gb);
3746
            if(chroma_weight_flag){
3747
                int j;
3748
                for(j=0; j<2; j++){
3749
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3750
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3751
                    if(   h->chroma_weight[list][i][j] != chroma_def
3752
                       || h->chroma_offset[list][i][j] != 0)
3753
                        h->use_weight_chroma= 1;
3754
                }
3755
            }else{
3756
                int j;
3757
                for(j=0; j<2; j++){
3758
                    h->chroma_weight[list][i][j]= chroma_def;
3759
                    h->chroma_offset[list][i][j]= 0;
3760
                }
3761
            }
3762
        }
3763
        if(h->slice_type != B_TYPE) break;
3764
    }
3765
    h->use_weight= h->use_weight || h->use_weight_chroma;
3766
    return 0;
3767
}
3768

    
3769
static void implicit_weight_table(H264Context *h){
3770
    MpegEncContext * const s = &h->s;
3771
    int ref0, ref1;
3772
    int cur_poc = s->current_picture_ptr->poc;
3773

    
3774
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3775
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3776
        h->use_weight= 0;
3777
        h->use_weight_chroma= 0;
3778
        return;
3779
    }
3780

    
3781
    h->use_weight= 2;
3782
    h->use_weight_chroma= 2;
3783
    h->luma_log2_weight_denom= 5;
3784
    h->chroma_log2_weight_denom= 5;
3785

    
3786
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3787
        int poc0 = h->ref_list[0][ref0].poc;
3788
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3789
            int poc1 = h->ref_list[1][ref1].poc;
3790
            int td = av_clip(poc1 - poc0, -128, 127);
3791
            if(td){
3792
                int tb = av_clip(cur_poc - poc0, -128, 127);
3793
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3794
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3795
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3796
                    h->implicit_weight[ref0][ref1] = 32;
3797
                else
3798
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3799
            }else
3800
                h->implicit_weight[ref0][ref1] = 32;
3801
        }
3802
    }
3803
}
3804

    
3805
static inline void unreference_pic(H264Context *h, Picture *pic){
3806
    int i;
3807
    pic->reference=0;
3808
    if(pic == h->delayed_output_pic)
3809
        pic->reference=1;
3810
    else{
3811
        for(i = 0; h->delayed_pic[i]; i++)
3812
            if(pic == h->delayed_pic[i]){
3813
                pic->reference=1;
3814
                break;
3815
            }
3816
    }
3817
}
3818

    
3819
/**
3820
 * instantaneous decoder refresh.
3821
 */
3822
static void idr(H264Context *h){
3823
    int i;
3824

    
3825
    for(i=0; i<16; i++){
3826
        if (h->long_ref[i] != NULL) {
3827
            unreference_pic(h, h->long_ref[i]);
3828
            h->long_ref[i]= NULL;
3829
        }
3830
    }
3831
    h->long_ref_count=0;
3832

    
3833
    for(i=0; i<h->short_ref_count; i++){
3834
        unreference_pic(h, h->short_ref[i]);
3835
        h->short_ref[i]= NULL;
3836
    }
3837
    h->short_ref_count=0;
3838
}
3839

    
3840
/* forget old pics after a seek */
3841
static void flush_dpb(AVCodecContext *avctx){
3842
    H264Context *h= avctx->priv_data;
3843
    int i;
3844
    for(i=0; i<16; i++) {
3845
        if(h->delayed_pic[i])
3846
            h->delayed_pic[i]->reference= 0;
3847
        h->delayed_pic[i]= NULL;
3848
    }
3849
    if(h->delayed_output_pic)
3850
        h->delayed_output_pic->reference= 0;
3851
    h->delayed_output_pic= NULL;
3852
    idr(h);
3853
    if(h->s.current_picture_ptr)
3854
        h->s.current_picture_ptr->reference= 0;
3855
}
3856

    
3857
/**
3858
 *
3859
 * @return the removed picture or NULL if an error occurs
3860
 */
3861
static Picture * remove_short(H264Context *h, int frame_num){
3862
    MpegEncContext * const s = &h->s;
3863
    int i;
3864

    
3865
    if(s->avctx->debug&FF_DEBUG_MMCO)
3866
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3867

    
3868
    for(i=0; i<h->short_ref_count; i++){
3869
        Picture *pic= h->short_ref[i];
3870
        if(s->avctx->debug&FF_DEBUG_MMCO)
3871
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3872
        if(pic->frame_num == frame_num){
3873
            h->short_ref[i]= NULL;
3874
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3875
            h->short_ref_count--;
3876
            return pic;
3877
        }
3878
    }
3879
    return NULL;
3880
}
3881

    
3882
/**
3883
 *
3884
 * @return the removed picture or NULL if an error occurs
3885
 */
3886
static Picture * remove_long(H264Context *h, int i){
3887
    Picture *pic;
3888

    
3889
    pic= h->long_ref[i];
3890
    h->long_ref[i]= NULL;
3891
    if(pic) h->long_ref_count--;
3892

    
3893
    return pic;
3894
}
3895

    
3896
/**
3897
 * print short term list
3898
 */
3899
static void print_short_term(H264Context *h) {
3900
    uint32_t i;
3901
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3902
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3903
        for(i=0; i<h->short_ref_count; i++){
3904
            Picture *pic= h->short_ref[i];
3905
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3906
        }
3907
    }
3908
}
3909

    
3910
/**
3911
 * print long term list
3912
 */
3913
static void print_long_term(H264Context *h) {
3914
    uint32_t i;
3915
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3916
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3917
        for(i = 0; i < 16; i++){
3918
            Picture *pic= h->long_ref[i];
3919
            if (pic) {
3920
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3921
            }
3922
        }
3923
    }
3924
}
3925

    
3926
/**
3927
 * Executes the reference picture marking (memory management control operations).
3928
 */
3929
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3930
    MpegEncContext * const s = &h->s;
3931
    int i, j;
3932
    int current_is_long=0;
3933
    Picture *pic;
3934

    
3935
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3936
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3937

    
3938
    for(i=0; i<mmco_count; i++){
3939
        if(s->avctx->debug&FF_DEBUG_MMCO)
3940
            av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
3941

    
3942
        switch(mmco[i].opcode){
3943
        case MMCO_SHORT2UNUSED:
3944
            pic= remove_short(h, mmco[i].short_frame_num);
3945
            if(pic)
3946
                unreference_pic(h, pic);
3947
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3948
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
3949
            break;
3950
        case MMCO_SHORT2LONG:
3951
            pic= remove_long(h, mmco[i].long_index);
3952
            if(pic) unreference_pic(h, pic);
3953

    
3954
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3955
            if (h->long_ref[ mmco[i].long_index ]){
3956
                h->long_ref[ mmco[i].long_index ]->long_ref=1;
3957
                h->long_ref_count++;
3958
            }
3959
            break;
3960
        case MMCO_LONG2UNUSED:
3961
            pic= remove_long(h, mmco[i].long_index);
3962
            if(pic)
3963
                unreference_pic(h, pic);
3964
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3965
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
3966
            break;
3967
        case MMCO_LONG:
3968
            pic= remove_long(h, mmco[i].long_index);
3969
            if(pic) unreference_pic(h, pic);
3970

    
3971
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3972
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3973
            h->long_ref_count++;
3974

    
3975
            current_is_long=1;
3976
            break;
3977
        case MMCO_SET_MAX_LONG:
3978
            assert(mmco[i].long_index <= 16);
3979
            // just remove the long term which index is greater than new max
3980
            for(j = mmco[i].long_index; j<16; j++){
3981
                pic = remove_long(h, j);
3982
                if (pic) unreference_pic(h, pic);
3983
            }
3984
            break;
3985
        case MMCO_RESET:
3986
            while(h->short_ref_count){
3987
                pic= remove_short(h, h->short_ref[0]->frame_num);
3988
                if(pic) unreference_pic(h, pic);
3989
            }
3990
            for(j = 0; j < 16; j++) {
3991
                pic= remove_long(h, j);
3992
                if(pic) unreference_pic(h, pic);
3993
            }
3994
            break;
3995
        default: assert(0);
3996
        }
3997
    }
3998

    
3999
    if(!current_is_long){
4000
        pic= remove_short(h, s->current_picture_ptr->frame_num);
4001
        if(pic){
4002
            unreference_pic(h, pic);
4003
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
4004
        }
4005

    
4006
        if(h->short_ref_count)
4007
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
4008

    
4009
        h->short_ref[0]= s->current_picture_ptr;
4010
        h->short_ref[0]->long_ref=0;
4011
        h->short_ref_count++;
4012
    }
4013

    
4014
    print_short_term(h);
4015
    print_long_term(h);
4016
    return 0;
4017
}
4018

    
4019
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
4020
    MpegEncContext * const s = &h->s;
4021
    int i;
4022

    
4023
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
4024
        s->broken_link= get_bits1(gb) -1;
4025
        h->mmco[0].long_index= get_bits1(gb) - 1; // current_long_term_idx
4026
        if(h->mmco[0].long_index == -1)
4027
            h->mmco_index= 0;
4028
        else{
4029
            h->mmco[0].opcode= MMCO_LONG;
4030
            h->mmco_index= 1;
4031
        }
4032
    }else{
4033
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
4034
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
4035
                MMCOOpcode opcode= get_ue_golomb(gb);
4036

    
4037
                h->mmco[i].opcode= opcode;
4038
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
4039
                    h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
4040
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
4041
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
4042
                        return -1;
4043
                    }*/
4044
                }
4045
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
4046
                    unsigned int long_index= get_ue_golomb(gb);
4047
                    if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ long_index >= 16){
4048
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
4049
                        return -1;
4050
                    }
4051
                    h->mmco[i].long_index= long_index;
4052
                }
4053

    
4054
                if(opcode > (unsigned)MMCO_LONG){
4055
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4056
                    return -1;
4057
                }
4058
                if(opcode == MMCO_END)
4059
                    break;
4060
            }
4061
            h->mmco_index= i;
4062
        }else{
4063
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4064

    
4065
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4066
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4067
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4068
                h->mmco_index= 1;
4069
            }else
4070
                h->mmco_index= 0;
4071
        }
4072
    }
4073

    
4074
    return 0;
4075
}
4076

    
4077
static int init_poc(H264Context *h){
4078
    MpegEncContext * const s = &h->s;
4079
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4080
    int field_poc[2];
4081

    
4082
    if(h->nal_unit_type == NAL_IDR_SLICE){
4083
        h->frame_num_offset= 0;
4084
    }else{
4085
        if(h->frame_num < h->prev_frame_num)
4086
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4087
        else
4088
            h->frame_num_offset= h->prev_frame_num_offset;
4089
    }
4090

    
4091
    if(h->sps.poc_type==0){
4092
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4093

    
4094
        if(h->nal_unit_type == NAL_IDR_SLICE){
4095
             h->prev_poc_msb=
4096
             h->prev_poc_lsb= 0;
4097
        }
4098

    
4099
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4100
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4101
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4102
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4103
        else
4104
            h->poc_msb = h->prev_poc_msb;
4105
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4106
        field_poc[0] =
4107
        field_poc[1] = h->poc_msb + h->poc_lsb;
4108
        if(s->picture_structure == PICT_FRAME)
4109
            field_poc[1] += h->delta_poc_bottom;
4110
    }else if(h->sps.poc_type==1){
4111
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4112
        int i;
4113

    
4114
        if(h->sps.poc_cycle_length != 0)
4115
            abs_frame_num = h->frame_num_offset + h->frame_num;
4116
        else
4117
            abs_frame_num = 0;
4118

    
4119
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4120
            abs_frame_num--;
4121

    
4122
        expected_delta_per_poc_cycle = 0;
4123
        for(i=0; i < h->sps.poc_cycle_length; i++)
4124
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4125

    
4126
        if(abs_frame_num > 0){
4127
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4128
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4129

    
4130
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4131
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4132
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4133
        } else
4134
            expectedpoc = 0;
4135

    
4136
        if(h->nal_ref_idc == 0)
4137
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4138

    
4139
        field_poc[0] = expectedpoc + h->delta_poc[0];
4140
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4141

    
4142
        if(s->picture_structure == PICT_FRAME)
4143
            field_poc[1] += h->delta_poc[1];
4144
    }else{
4145
        int poc;
4146
        if(h->nal_unit_type == NAL_IDR_SLICE){
4147
            poc= 0;
4148
        }else{
4149
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4150
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4151
        }
4152
        field_poc[0]= poc;
4153
        field_poc[1]= poc;
4154
    }
4155

    
4156
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4157
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4158
    if(s->picture_structure != PICT_TOP_FIELD)
4159
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4160
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4161
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4162

    
4163
    return 0;
4164
}
4165

    
4166

    
4167
/**
4168
 * initialize scan tables
4169
 */
4170
static void init_scan_tables(H264Context *h){
4171
    MpegEncContext * const s = &h->s;
4172
    int i;
4173
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
4174
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
4175
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
4176
    }else{
4177
        for(i=0; i<16; i++){
4178
#define T(x) (x>>2) | ((x<<2) & 0xF)
4179
            h->zigzag_scan[i] = T(zigzag_scan[i]);
4180
            h-> field_scan[i] = T( field_scan[i]);
4181
#undef T
4182
        }
4183
    }
4184
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
4185
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
4186
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
4187
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
4188
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
4189
    }else{
4190
        for(i=0; i<64; i++){
4191
#define T(x) (x>>3) | ((x&7)<<3)
4192
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
4193
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
4194
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
4195
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
4196
#undef T
4197
        }
4198
    }
4199
    if(h->sps.transform_bypass){ //FIXME same ugly
4200
        h->zigzag_scan_q0          = zigzag_scan;
4201
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
4202
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
4203
        h->field_scan_q0           = field_scan;
4204
        h->field_scan8x8_q0        = field_scan8x8;
4205
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
4206
    }else{
4207
        h->zigzag_scan_q0          = h->zigzag_scan;
4208
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
4209
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
4210
        h->field_scan_q0           = h->field_scan;
4211
        h->field_scan8x8_q0        = h->field_scan8x8;
4212
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
4213
    }
4214
}
4215
/**
4216
 * decodes a slice header.
4217
 * this will allso call MPV_common_init() and frame_start() as needed
4218
 */
4219
static int decode_slice_header(H264Context *h){
4220
    MpegEncContext * const s = &h->s;
4221
    unsigned int first_mb_in_slice;
4222
    unsigned int pps_id;
4223
    int num_ref_idx_active_override_flag;
4224
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4225
    unsigned int slice_type, tmp;
4226
    int default_ref_list_done = 0;
4227

    
4228
    s->current_picture.reference= h->nal_ref_idc != 0;
4229
    s->dropable= h->nal_ref_idc == 0;
4230

    
4231
    first_mb_in_slice= get_ue_golomb(&s->gb);
4232

    
4233
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
4234
        h->slice_num = 0;
4235
        s->current_picture_ptr= NULL;
4236
    }
4237

    
4238
    slice_type= get_ue_golomb(&s->gb);
4239
    if(slice_type > 9){
4240
        av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
4241
        return -1;
4242
    }
4243
    if(slice_type > 4){
4244
        slice_type -= 5;
4245
        h->slice_type_fixed=1;
4246
    }else
4247
        h->slice_type_fixed=0;
4248

    
4249
    slice_type= slice_type_map[ slice_type ];
4250
    if (slice_type == I_TYPE
4251
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4252
        default_ref_list_done = 1;
4253
    }
4254
    h->slice_type= slice_type;
4255

    
4256
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4257

    
4258
    pps_id= get_ue_golomb(&s->gb);
4259
    if(pps_id>=MAX_PPS_COUNT){
4260
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4261
        return -1;
4262
    }
4263
    if(!h->pps_buffers[pps_id]) {
4264
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4265
        return -1;
4266
    }
4267
    h->pps= *h->pps_buffers[pps_id];
4268

    
4269
    if(!h->sps_buffers[h->pps.sps_id]) {
4270
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4271
        return -1;
4272
    }
4273
    h->sps = *h->sps_buffers[h->pps.sps_id];
4274

    
4275
    if(h->dequant_coeff_pps != pps_id){
4276
        h->dequant_coeff_pps = pps_id;
4277
        init_dequant_tables(h);
4278
    }
4279

    
4280
    s->mb_width= h->sps.mb_width;