Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ a2e4c0dd

History | View | Annotate | Download (312 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
/**
24
 * @file h264.c
25
 * H.264 / AVC / MPEG4 part10 codec.
26
 * @author Michael Niedermayer <michaelni@gmx.at>
27
 */
28

    
29
#include "dsputil.h"
30
#include "avcodec.h"
31
#include "mpegvideo.h"
32
#include "h264.h"
33
#include "h264data.h"
34
#include "h264_parser.h"
35
#include "golomb.h"
36

    
37
#include "cabac.h"
38

    
39
//#undef NDEBUG
40
#include <assert.h>
41

    
42
static VLC coeff_token_vlc[4];
43
static VLC chroma_dc_coeff_token_vlc;
44

    
45
static VLC total_zeros_vlc[15];
46
static VLC chroma_dc_total_zeros_vlc[3];
47

    
48
static VLC run_vlc[6];
49
static VLC run7_vlc;
50

    
51
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
52
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
53
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);
54
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);
55

    
56
static av_always_inline uint32_t pack16to32(int a, int b){
57
#ifdef WORDS_BIGENDIAN
58
   return (b&0xFFFF) + (a<<16);
59
#else
60
   return (a&0xFFFF) + (b<<16);
61
#endif
62
}
63

    
64
const uint8_t ff_rem6[52]={
65
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,
66
};
67

    
68
const uint8_t ff_div6[52]={
69
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,
70
};
71

    
72

    
73
/**
74
 * fill a rectangle.
75
 * @param h height of the rectangle, should be a constant
76
 * @param w width of the rectangle, should be a constant
77
 * @param size the size of val (1 or 4), should be a constant
78
 */
79
static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
80
    uint8_t *p= (uint8_t*)vp;
81
    assert(size==1 || size==4);
82
    assert(w<=4);
83

    
84
    w      *= size;
85
    stride *= size;
86

    
87
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
88
    assert((stride&(w-1))==0);
89
    if(w==2){
90
        const uint16_t v= size==4 ? val : val*0x0101;
91
        *(uint16_t*)(p + 0*stride)= v;
92
        if(h==1) return;
93
        *(uint16_t*)(p + 1*stride)= v;
94
        if(h==2) return;
95
        *(uint16_t*)(p + 2*stride)=
96
        *(uint16_t*)(p + 3*stride)= v;
97
    }else if(w==4){
98
        const uint32_t v= size==4 ? val : val*0x01010101;
99
        *(uint32_t*)(p + 0*stride)= v;
100
        if(h==1) return;
101
        *(uint32_t*)(p + 1*stride)= v;
102
        if(h==2) return;
103
        *(uint32_t*)(p + 2*stride)=
104
        *(uint32_t*)(p + 3*stride)= v;
105
    }else if(w==8){
106
    //gcc can't optimize 64bit math on x86_32
107
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
108
        const uint64_t v= val*0x0100000001ULL;
109
        *(uint64_t*)(p + 0*stride)= v;
110
        if(h==1) return;
111
        *(uint64_t*)(p + 1*stride)= v;
112
        if(h==2) return;
113
        *(uint64_t*)(p + 2*stride)=
114
        *(uint64_t*)(p + 3*stride)= v;
115
    }else if(w==16){
116
        const uint64_t v= val*0x0100000001ULL;
117
        *(uint64_t*)(p + 0+0*stride)=
118
        *(uint64_t*)(p + 8+0*stride)=
119
        *(uint64_t*)(p + 0+1*stride)=
120
        *(uint64_t*)(p + 8+1*stride)= v;
121
        if(h==2) return;
122
        *(uint64_t*)(p + 0+2*stride)=
123
        *(uint64_t*)(p + 8+2*stride)=
124
        *(uint64_t*)(p + 0+3*stride)=
125
        *(uint64_t*)(p + 8+3*stride)= v;
126
#else
127
        *(uint32_t*)(p + 0+0*stride)=
128
        *(uint32_t*)(p + 4+0*stride)= val;
129
        if(h==1) return;
130
        *(uint32_t*)(p + 0+1*stride)=
131
        *(uint32_t*)(p + 4+1*stride)= val;
132
        if(h==2) return;
133
        *(uint32_t*)(p + 0+2*stride)=
134
        *(uint32_t*)(p + 4+2*stride)=
135
        *(uint32_t*)(p + 0+3*stride)=
136
        *(uint32_t*)(p + 4+3*stride)= val;
137
    }else if(w==16){
138
        *(uint32_t*)(p + 0+0*stride)=
139
        *(uint32_t*)(p + 4+0*stride)=
140
        *(uint32_t*)(p + 8+0*stride)=
141
        *(uint32_t*)(p +12+0*stride)=
142
        *(uint32_t*)(p + 0+1*stride)=
143
        *(uint32_t*)(p + 4+1*stride)=
144
        *(uint32_t*)(p + 8+1*stride)=
145
        *(uint32_t*)(p +12+1*stride)= val;
146
        if(h==2) return;
147
        *(uint32_t*)(p + 0+2*stride)=
148
        *(uint32_t*)(p + 4+2*stride)=
149
        *(uint32_t*)(p + 8+2*stride)=
150
        *(uint32_t*)(p +12+2*stride)=
151
        *(uint32_t*)(p + 0+3*stride)=
152
        *(uint32_t*)(p + 4+3*stride)=
153
        *(uint32_t*)(p + 8+3*stride)=
154
        *(uint32_t*)(p +12+3*stride)= val;
155
#endif
156
    }else
157
        assert(0);
158
    assert(h==4);
159
}
160

    
161
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
162
    MpegEncContext * const s = &h->s;
163
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
164
    int topleft_xy, top_xy, topright_xy, left_xy[2];
165
    int topleft_type, top_type, topright_type, left_type[2];
166
    int left_block[8];
167
    int i;
168

    
169
    //FIXME deblocking could skip the intra and nnz parts.
170
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
171
        return;
172

    
173
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
174

    
175
    top_xy     = mb_xy  - s->mb_stride;
176
    topleft_xy = top_xy - 1;
177
    topright_xy= top_xy + 1;
178
    left_xy[1] = left_xy[0] = mb_xy-1;
179
    left_block[0]= 0;
180
    left_block[1]= 1;
181
    left_block[2]= 2;
182
    left_block[3]= 3;
183
    left_block[4]= 7;
184
    left_block[5]= 10;
185
    left_block[6]= 8;
186
    left_block[7]= 11;
187
    if(FRAME_MBAFF){
188
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
189
        const int top_pair_xy      = pair_xy     - s->mb_stride;
190
        const int topleft_pair_xy  = top_pair_xy - 1;
191
        const int topright_pair_xy = top_pair_xy + 1;
192
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
193
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
194
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
195
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
196
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
197
        const int bottom = (s->mb_y & 1);
198
        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);
199
        if (bottom
200
                ? !curr_mb_frame_flag // bottom macroblock
201
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
202
                ) {
203
            top_xy -= s->mb_stride;
204
        }
205
        if (bottom
206
                ? !curr_mb_frame_flag // bottom macroblock
207
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
208
                ) {
209
            topleft_xy -= s->mb_stride;
210
        }
211
        if (bottom
212
                ? !curr_mb_frame_flag // bottom macroblock
213
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
214
                ) {
215
            topright_xy -= s->mb_stride;
216
        }
217
        if (left_mb_frame_flag != curr_mb_frame_flag) {
218
            left_xy[1] = left_xy[0] = pair_xy - 1;
219
            if (curr_mb_frame_flag) {
220
                if (bottom) {
221
                    left_block[0]= 2;
222
                    left_block[1]= 2;
223
                    left_block[2]= 3;
224
                    left_block[3]= 3;
225
                    left_block[4]= 8;
226
                    left_block[5]= 11;
227
                    left_block[6]= 8;
228
                    left_block[7]= 11;
229
                } else {
230
                    left_block[0]= 0;
231
                    left_block[1]= 0;
232
                    left_block[2]= 1;
233
                    left_block[3]= 1;
234
                    left_block[4]= 7;
235
                    left_block[5]= 10;
236
                    left_block[6]= 7;
237
                    left_block[7]= 10;
238
                }
239
            } else {
240
                left_xy[1] += s->mb_stride;
241
                //left_block[0]= 0;
242
                left_block[1]= 2;
243
                left_block[2]= 0;
244
                left_block[3]= 2;
245
                //left_block[4]= 7;
246
                left_block[5]= 10;
247
                left_block[6]= 7;
248
                left_block[7]= 10;
249
            }
250
        }
251
    }
252

    
253
    h->top_mb_xy = top_xy;
254
    h->left_mb_xy[0] = left_xy[0];
255
    h->left_mb_xy[1] = left_xy[1];
256
    if(for_deblock){
257
        topleft_type = 0;
258
        topright_type = 0;
259
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
260
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
261
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
262

    
263
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
264
            int list;
265
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
266
            for(i=0; i<16; i++)
267
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
268
            for(list=0; list<h->list_count; list++){
269
                if(USES_LIST(mb_type,list)){
270
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
271
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
272
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
273
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
274
                        dst[0] = src[0];
275
                        dst[1] = src[1];
276
                        dst[2] = src[2];
277
                        dst[3] = src[3];
278
                    }
279
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
280
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
281
                    ref += h->b8_stride;
282
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
283
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
284
                }else{
285
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
286
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
287
                }
288
            }
289
        }
290
    }else{
291
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
292
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
293
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
294
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
295
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
296
    }
297

    
298
    if(IS_INTRA(mb_type)){
299
        h->topleft_samples_available=
300
        h->top_samples_available=
301
        h->left_samples_available= 0xFFFF;
302
        h->topright_samples_available= 0xEEEA;
303

    
304
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
305
            h->topleft_samples_available= 0xB3FF;
306
            h->top_samples_available= 0x33FF;
307
            h->topright_samples_available= 0x26EA;
308
        }
309
        for(i=0; i<2; i++){
310
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
311
                h->topleft_samples_available&= 0xDF5F;
312
                h->left_samples_available&= 0x5F5F;
313
            }
314
        }
315

    
316
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
317
            h->topleft_samples_available&= 0x7FFF;
318

    
319
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
320
            h->topright_samples_available&= 0xFBFF;
321

    
322
        if(IS_INTRA4x4(mb_type)){
323
            if(IS_INTRA4x4(top_type)){
324
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
325
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
326
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
327
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
328
            }else{
329
                int pred;
330
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
331
                    pred= -1;
332
                else{
333
                    pred= 2;
334
                }
335
                h->intra4x4_pred_mode_cache[4+8*0]=
336
                h->intra4x4_pred_mode_cache[5+8*0]=
337
                h->intra4x4_pred_mode_cache[6+8*0]=
338
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
339
            }
340
            for(i=0; i<2; i++){
341
                if(IS_INTRA4x4(left_type[i])){
342
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
343
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
344
                }else{
345
                    int pred;
346
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
347
                        pred= -1;
348
                    else{
349
                        pred= 2;
350
                    }
351
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
352
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
353
                }
354
            }
355
        }
356
    }
357

    
358

    
359
/*
360
0 . T T. T T T T
361
1 L . .L . . . .
362
2 L . .L . . . .
363
3 . T TL . . . .
364
4 L . .L . . . .
365
5 L . .. . . . .
366
*/
367
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
368
    if(top_type){
369
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
370
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
371
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
372
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
373

    
374
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
375
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
376

    
377
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
378
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
379

    
380
    }else{
381
        h->non_zero_count_cache[4+8*0]=
382
        h->non_zero_count_cache[5+8*0]=
383
        h->non_zero_count_cache[6+8*0]=
384
        h->non_zero_count_cache[7+8*0]=
385

    
386
        h->non_zero_count_cache[1+8*0]=
387
        h->non_zero_count_cache[2+8*0]=
388

    
389
        h->non_zero_count_cache[1+8*3]=
390
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
391

    
392
    }
393

    
394
    for (i=0; i<2; i++) {
395
        if(left_type[i]){
396
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
397
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
398
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
399
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
400
        }else{
401
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
402
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
403
            h->non_zero_count_cache[0+8*1 +   8*i]=
404
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
405
        }
406
    }
407

    
408
    if( h->pps.cabac ) {
409
        // top_cbp
410
        if(top_type) {
411
            h->top_cbp = h->cbp_table[top_xy];
412
        } else if(IS_INTRA(mb_type)) {
413
            h->top_cbp = 0x1C0;
414
        } else {
415
            h->top_cbp = 0;
416
        }
417
        // left_cbp
418
        if (left_type[0]) {
419
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
420
        } else if(IS_INTRA(mb_type)) {
421
            h->left_cbp = 0x1C0;
422
        } else {
423
            h->left_cbp = 0;
424
        }
425
        if (left_type[0]) {
426
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
427
        }
428
        if (left_type[1]) {
429
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
430
        }
431
    }
432

    
433
#if 1
434
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
435
        int list;
436
        for(list=0; list<h->list_count; list++){
437
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
438
                /*if(!h->mv_cache_clean[list]){
439
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
440
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
441
                    h->mv_cache_clean[list]= 1;
442
                }*/
443
                continue;
444
            }
445
            h->mv_cache_clean[list]= 0;
446

    
447
            if(USES_LIST(top_type, list)){
448
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
449
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
450
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
451
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
452
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
453
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
454
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
455
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
456
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
457
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
458
            }else{
459
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
460
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
461
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
462
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
463
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
464
            }
465

    
466
            for(i=0; i<2; i++){
467
                int cache_idx = scan8[0] - 1 + i*2*8;
468
                if(USES_LIST(left_type[i], list)){
469
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
470
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
471
                    *(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]];
472
                    *(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]];
473
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
474
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
475
                }else{
476
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
477
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
478
                    h->ref_cache[list][cache_idx  ]=
479
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
480
                }
481
            }
482

    
483
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
484
                continue;
485

    
486
            if(USES_LIST(topleft_type, list)){
487
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
488
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
489
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
490
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
491
            }else{
492
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
493
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
494
            }
495

    
496
            if(USES_LIST(topright_type, list)){
497
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
498
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
499
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
500
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
501
            }else{
502
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
503
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
504
            }
505

    
506
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
507
                continue;
508

    
509
            h->ref_cache[list][scan8[5 ]+1] =
510
            h->ref_cache[list][scan8[7 ]+1] =
511
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
512
            h->ref_cache[list][scan8[4 ]] =
513
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
514
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
515
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
516
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
517
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
518
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
519

    
520
            if( h->pps.cabac ) {
521
                /* XXX beurk, Load mvd */
522
                if(USES_LIST(top_type, list)){
523
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
524
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
525
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
526
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
527
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
528
                }else{
529
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
530
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
531
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
532
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
533
                }
534
                if(USES_LIST(left_type[0], list)){
535
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
536
                    *(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]];
537
                    *(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]];
538
                }else{
539
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
540
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
541
                }
542
                if(USES_LIST(left_type[1], list)){
543
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
544
                    *(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]];
545
                    *(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]];
546
                }else{
547
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
548
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
549
                }
550
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
551
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
552
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
553
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
554
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
555

    
556
                if(h->slice_type == B_TYPE){
557
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
558

    
559
                    if(IS_DIRECT(top_type)){
560
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
561
                    }else if(IS_8X8(top_type)){
562
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
563
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
564
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
565
                    }else{
566
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
567
                    }
568

    
569
                    if(IS_DIRECT(left_type[0]))
570
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
571
                    else if(IS_8X8(left_type[0]))
572
                        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)];
573
                    else
574
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
575

    
576
                    if(IS_DIRECT(left_type[1]))
577
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
578
                    else if(IS_8X8(left_type[1]))
579
                        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)];
580
                    else
581
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
582
                }
583
            }
584

    
585
            if(FRAME_MBAFF){
586
#define MAP_MVS\
587
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
588
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
589
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
590
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
591
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
592
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
593
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
594
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
595
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
596
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
597
                if(MB_FIELD){
598
#define MAP_F2F(idx, mb_type)\
599
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
600
                        h->ref_cache[list][idx] <<= 1;\
601
                        h->mv_cache[list][idx][1] /= 2;\
602
                        h->mvd_cache[list][idx][1] /= 2;\
603
                    }
604
                    MAP_MVS
605
#undef MAP_F2F
606
                }else{
607
#define MAP_F2F(idx, mb_type)\
608
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
609
                        h->ref_cache[list][idx] >>= 1;\
610
                        h->mv_cache[list][idx][1] <<= 1;\
611
                        h->mvd_cache[list][idx][1] <<= 1;\
612
                    }
613
                    MAP_MVS
614
#undef MAP_F2F
615
                }
616
            }
617
        }
618
    }
619
#endif
620

    
621
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
622
}
623

    
624
static inline void write_back_intra_pred_mode(H264Context *h){
625
    MpegEncContext * const s = &h->s;
626
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
627

    
628
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
629
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
630
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
631
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
632
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
633
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
634
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
635
}
636

    
637
/**
638
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
639
 */
640
static inline int check_intra4x4_pred_mode(H264Context *h){
641
    MpegEncContext * const s = &h->s;
642
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
643
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
644
    int i;
645

    
646
    if(!(h->top_samples_available&0x8000)){
647
        for(i=0; i<4; i++){
648
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
649
            if(status<0){
650
                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);
651
                return -1;
652
            } else if(status){
653
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
654
            }
655
        }
656
    }
657

    
658
    if(!(h->left_samples_available&0x8000)){
659
        for(i=0; i<4; i++){
660
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
661
            if(status<0){
662
                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);
663
                return -1;
664
            } else if(status){
665
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
666
            }
667
        }
668
    }
669

    
670
    return 0;
671
} //FIXME cleanup like next
672

    
673
/**
674
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
675
 */
676
static inline int check_intra_pred_mode(H264Context *h, int mode){
677
    MpegEncContext * const s = &h->s;
678
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
679
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
680

    
681
    if(mode > 6U) {
682
        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);
683
        return -1;
684
    }
685

    
686
    if(!(h->top_samples_available&0x8000)){
687
        mode= top[ mode ];
688
        if(mode<0){
689
            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);
690
            return -1;
691
        }
692
    }
693

    
694
    if(!(h->left_samples_available&0x8000)){
695
        mode= left[ mode ];
696
        if(mode<0){
697
            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);
698
            return -1;
699
        }
700
    }
701

    
702
    return mode;
703
}
704

    
705
/**
706
 * gets the predicted intra4x4 prediction mode.
707
 */
708
static inline int pred_intra_mode(H264Context *h, int n){
709
    const int index8= scan8[n];
710
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
711
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
712
    const int min= FFMIN(left, top);
713

    
714
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
715

    
716
    if(min<0) return DC_PRED;
717
    else      return min;
718
}
719

    
720
static inline void write_back_non_zero_count(H264Context *h){
721
    MpegEncContext * const s = &h->s;
722
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
723

    
724
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
725
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
726
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
727
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
728
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
729
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
730
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
731

    
732
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
733
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
734
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
735

    
736
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
737
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
738
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
739

    
740
    if(FRAME_MBAFF){
741
        // store all luma nnzs, for deblocking
742
        int v = 0, i;
743
        for(i=0; i<16; i++)
744
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
745
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
746
    }
747
}
748

    
749
/**
750
 * gets the predicted number of non zero coefficients.
751
 * @param n block index
752
 */
753
static inline int pred_non_zero_count(H264Context *h, int n){
754
    const int index8= scan8[n];
755
    const int left= h->non_zero_count_cache[index8 - 1];
756
    const int top = h->non_zero_count_cache[index8 - 8];
757
    int i= left + top;
758

    
759
    if(i<64) i= (i+1)>>1;
760

    
761
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
762

    
763
    return i&31;
764
}
765

    
766
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
767
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
768
    MpegEncContext *s = &h->s;
769

    
770
    /* there is no consistent mapping of mvs to neighboring locations that will
771
     * make mbaff happy, so we can't move all this logic to fill_caches */
772
    if(FRAME_MBAFF){
773
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
774
        const int16_t *mv;
775
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
776
        *C = h->mv_cache[list][scan8[0]-2];
777

    
778
        if(!MB_FIELD
779
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
780
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
781
            if(IS_INTERLACED(mb_types[topright_xy])){
782
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
783
                const int x4 = X4, y4 = Y4;\
784
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
785
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
786
                    return LIST_NOT_USED;\
787
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
788
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
789
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
790
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
791

    
792
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
793
            }
794
        }
795
        if(topright_ref == PART_NOT_AVAILABLE
796
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
797
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
798
            if(!MB_FIELD
799
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
800
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
801
            }
802
            if(MB_FIELD
803
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
804
               && i >= scan8[0]+8){
805
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
806
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
807
            }
808
        }
809
#undef SET_DIAG_MV
810
    }
811

    
812
    if(topright_ref != PART_NOT_AVAILABLE){
813
        *C= h->mv_cache[list][ i - 8 + part_width ];
814
        return topright_ref;
815
    }else{
816
        tprintf(s->avctx, "topright MV not available\n");
817

    
818
        *C= h->mv_cache[list][ i - 8 - 1 ];
819
        return h->ref_cache[list][ i - 8 - 1 ];
820
    }
821
}
822

    
823
/**
824
 * gets the predicted MV.
825
 * @param n the block index
826
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
827
 * @param mx the x component of the predicted motion vector
828
 * @param my the y component of the predicted motion vector
829
 */
830
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
831
    const int index8= scan8[n];
832
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
833
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
834
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
835
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
836
    const int16_t * C;
837
    int diagonal_ref, match_count;
838

    
839
    assert(part_width==1 || part_width==2 || part_width==4);
840

    
841
/* mv_cache
842
  B . . A T T T T
843
  U . . L . . , .
844
  U . . L . . . .
845
  U . . L . . , .
846
  . . . L . . . .
847
*/
848

    
849
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
850
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
851
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
852
    if(match_count > 1){ //most common
853
        *mx= mid_pred(A[0], B[0], C[0]);
854
        *my= mid_pred(A[1], B[1], C[1]);
855
    }else if(match_count==1){
856
        if(left_ref==ref){
857
            *mx= A[0];
858
            *my= A[1];
859
        }else if(top_ref==ref){
860
            *mx= B[0];
861
            *my= B[1];
862
        }else{
863
            *mx= C[0];
864
            *my= C[1];
865
        }
866
    }else{
867
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
868
            *mx= A[0];
869
            *my= A[1];
870
        }else{
871
            *mx= mid_pred(A[0], B[0], C[0]);
872
            *my= mid_pred(A[1], B[1], C[1]);
873
        }
874
    }
875

    
876
    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);
877
}
878

    
879
/**
880
 * gets the directionally predicted 16x8 MV.
881
 * @param n the block index
882
 * @param mx the x component of the predicted motion vector
883
 * @param my the y component of the predicted motion vector
884
 */
885
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
886
    if(n==0){
887
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
888
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
889

    
890
        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);
891

    
892
        if(top_ref == ref){
893
            *mx= B[0];
894
            *my= B[1];
895
            return;
896
        }
897
    }else{
898
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
899
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
900

    
901
        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);
902

    
903
        if(left_ref == ref){
904
            *mx= A[0];
905
            *my= A[1];
906
            return;
907
        }
908
    }
909

    
910
    //RARE
911
    pred_motion(h, n, 4, list, ref, mx, my);
912
}
913

    
914
/**
915
 * gets the directionally predicted 8x16 MV.
916
 * @param n the block index
917
 * @param mx the x component of the predicted motion vector
918
 * @param my the y component of the predicted motion vector
919
 */
920
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
921
    if(n==0){
922
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
923
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
924

    
925
        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);
926

    
927
        if(left_ref == ref){
928
            *mx= A[0];
929
            *my= A[1];
930
            return;
931
        }
932
    }else{
933
        const int16_t * C;
934
        int diagonal_ref;
935

    
936
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
937

    
938
        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);
939

    
940
        if(diagonal_ref == ref){
941
            *mx= C[0];
942
            *my= C[1];
943
            return;
944
        }
945
    }
946

    
947
    //RARE
948
    pred_motion(h, n, 2, list, ref, mx, my);
949
}
950

    
951
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
952
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
953
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
954

    
955
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
956

    
957
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
958
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
959
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
960

    
961
        *mx = *my = 0;
962
        return;
963
    }
964

    
965
    pred_motion(h, 0, 4, 0, 0, mx, my);
966

    
967
    return;
968
}
969

    
970
static inline void direct_dist_scale_factor(H264Context * const h){
971
    const int poc = h->s.current_picture_ptr->poc;
972
    const int poc1 = h->ref_list[1][0].poc;
973
    int i;
974
    for(i=0; i<h->ref_count[0]; i++){
975
        int poc0 = h->ref_list[0][i].poc;
976
        int td = av_clip(poc1 - poc0, -128, 127);
977
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
978
            h->dist_scale_factor[i] = 256;
979
        }else{
980
            int tb = av_clip(poc - poc0, -128, 127);
981
            int tx = (16384 + (FFABS(td) >> 1)) / td;
982
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
983
        }
984
    }
985
    if(FRAME_MBAFF){
986
        for(i=0; i<h->ref_count[0]; i++){
987
            h->dist_scale_factor_field[2*i] =
988
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
989
        }
990
    }
991
}
992
static inline void direct_ref_list_init(H264Context * const h){
993
    MpegEncContext * const s = &h->s;
994
    Picture * const ref1 = &h->ref_list[1][0];
995
    Picture * const cur = s->current_picture_ptr;
996
    int list, i, j;
997
    if(cur->pict_type == I_TYPE)
998
        cur->ref_count[0] = 0;
999
    if(cur->pict_type != B_TYPE)
1000
        cur->ref_count[1] = 0;
1001
    for(list=0; list<2; list++){
1002
        cur->ref_count[list] = h->ref_count[list];
1003
        for(j=0; j<h->ref_count[list]; j++)
1004
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1005
    }
1006
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1007
        return;
1008
    for(list=0; list<2; list++){
1009
        for(i=0; i<ref1->ref_count[list]; i++){
1010
            const int poc = ref1->ref_poc[list][i];
1011
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1012
            for(j=0; j<h->ref_count[list]; j++)
1013
                if(h->ref_list[list][j].poc == poc){
1014
                    h->map_col_to_list0[list][i] = j;
1015
                    break;
1016
                }
1017
        }
1018
    }
1019
    if(FRAME_MBAFF){
1020
        for(list=0; list<2; list++){
1021
            for(i=0; i<ref1->ref_count[list]; i++){
1022
                j = h->map_col_to_list0[list][i];
1023
                h->map_col_to_list0_field[list][2*i] = 2*j;
1024
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1025
            }
1026
        }
1027
    }
1028
}
1029

    
1030
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1031
    MpegEncContext * const s = &h->s;
1032
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1033
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1034
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1035
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1036
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1037
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1038
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1039
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1040
    const int is_b8x8 = IS_8X8(*mb_type);
1041
    unsigned int sub_mb_type;
1042
    int i8, i4;
1043

    
1044
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1045
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1046
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1047
         * so we know exactly what block size to use */
1048
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1049
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1050
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1051
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1052
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1053
    }else{
1054
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1055
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1056
    }
1057
    if(!is_b8x8)
1058
        *mb_type |= MB_TYPE_DIRECT2;
1059
    if(MB_FIELD)
1060
        *mb_type |= MB_TYPE_INTERLACED;
1061

    
1062
    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);
1063

    
1064
    if(h->direct_spatial_mv_pred){
1065
        int ref[2];
1066
        int mv[2][2];
1067
        int list;
1068

    
1069
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1070

    
1071
        /* ref = min(neighbors) */
1072
        for(list=0; list<2; list++){
1073
            int refa = h->ref_cache[list][scan8[0] - 1];
1074
            int refb = h->ref_cache[list][scan8[0] - 8];
1075
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1076
            if(refc == -2)
1077
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1078
            ref[list] = refa;
1079
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1080
                ref[list] = refb;
1081
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1082
                ref[list] = refc;
1083
            if(ref[list] < 0)
1084
                ref[list] = -1;
1085
        }
1086

    
1087
        if(ref[0] < 0 && ref[1] < 0){
1088
            ref[0] = ref[1] = 0;
1089
            mv[0][0] = mv[0][1] =
1090
            mv[1][0] = mv[1][1] = 0;
1091
        }else{
1092
            for(list=0; list<2; list++){
1093
                if(ref[list] >= 0)
1094
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1095
                else
1096
                    mv[list][0] = mv[list][1] = 0;
1097
            }
1098
        }
1099

    
1100
        if(ref[1] < 0){
1101
            *mb_type &= ~MB_TYPE_P0L1;
1102
            sub_mb_type &= ~MB_TYPE_P0L1;
1103
        }else if(ref[0] < 0){
1104
            *mb_type &= ~MB_TYPE_P0L0;
1105
            sub_mb_type &= ~MB_TYPE_P0L0;
1106
        }
1107

    
1108
        if(IS_16X16(*mb_type)){
1109
            int a=0, b=0;
1110

    
1111
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1112
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1113
            if(!IS_INTRA(mb_type_col)
1114
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1115
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1116
                       && (h->x264_build>33 || !h->x264_build)))){
1117
                if(ref[0] > 0)
1118
                    a= pack16to32(mv[0][0],mv[0][1]);
1119
                if(ref[1] > 0)
1120
                    b= pack16to32(mv[1][0],mv[1][1]);
1121
            }else{
1122
                a= pack16to32(mv[0][0],mv[0][1]);
1123
                b= pack16to32(mv[1][0],mv[1][1]);
1124
            }
1125
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1126
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1127
        }else{
1128
            for(i8=0; i8<4; i8++){
1129
                const int x8 = i8&1;
1130
                const int y8 = i8>>1;
1131

    
1132
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1133
                    continue;
1134
                h->sub_mb_type[i8] = sub_mb_type;
1135

    
1136
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1137
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1138
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1139
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1140

    
1141
                /* col_zero_flag */
1142
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1143
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1144
                                                  && (h->x264_build>33 || !h->x264_build)))){
1145
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1146
                    if(IS_SUB_8X8(sub_mb_type)){
1147
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1148
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1149
                            if(ref[0] == 0)
1150
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1151
                            if(ref[1] == 0)
1152
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1153
                        }
1154
                    }else
1155
                    for(i4=0; i4<4; i4++){
1156
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1157
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1158
                            if(ref[0] == 0)
1159
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1160
                            if(ref[1] == 0)
1161
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1162
                        }
1163
                    }
1164
                }
1165
            }
1166
        }
1167
    }else{ /* direct temporal mv pred */
1168
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1169
        const int *dist_scale_factor = h->dist_scale_factor;
1170

    
1171
        if(FRAME_MBAFF){
1172
            if(IS_INTERLACED(*mb_type)){
1173
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1174
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1175
                dist_scale_factor = h->dist_scale_factor_field;
1176
            }
1177
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1178
                /* FIXME assumes direct_8x8_inference == 1 */
1179
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1180
                int mb_types_col[2];
1181
                int y_shift;
1182

    
1183
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1184
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1185
                         | (*mb_type & MB_TYPE_INTERLACED);
1186
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1187

    
1188
                if(IS_INTERLACED(*mb_type)){
1189
                    /* frame to field scaling */
1190
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1191
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1192
                    if(s->mb_y&1){
1193
                        l1ref0 -= 2*h->b8_stride;
1194
                        l1ref1 -= 2*h->b8_stride;
1195
                        l1mv0 -= 4*h->b_stride;
1196
                        l1mv1 -= 4*h->b_stride;
1197
                    }
1198
                    y_shift = 0;
1199

    
1200
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1201
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1202
                       && !is_b8x8)
1203
                        *mb_type |= MB_TYPE_16x8;
1204
                    else
1205
                        *mb_type |= MB_TYPE_8x8;
1206
                }else{
1207
                    /* field to frame scaling */
1208
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1209
                     * but in MBAFF, top and bottom POC are equal */
1210
                    int dy = (s->mb_y&1) ? 1 : 2;
1211
                    mb_types_col[0] =
1212
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1213
                    l1ref0 += dy*h->b8_stride;
1214
                    l1ref1 += dy*h->b8_stride;
1215
                    l1mv0 += 2*dy*h->b_stride;
1216
                    l1mv1 += 2*dy*h->b_stride;
1217
                    y_shift = 2;
1218

    
1219
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1220
                       && !is_b8x8)
1221
                        *mb_type |= MB_TYPE_16x16;
1222
                    else
1223
                        *mb_type |= MB_TYPE_8x8;
1224
                }
1225

    
1226
                for(i8=0; i8<4; i8++){
1227
                    const int x8 = i8&1;
1228
                    const int y8 = i8>>1;
1229
                    int ref0, scale;
1230
                    const int16_t (*l1mv)[2]= l1mv0;
1231

    
1232
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1233
                        continue;
1234
                    h->sub_mb_type[i8] = sub_mb_type;
1235

    
1236
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1237
                    if(IS_INTRA(mb_types_col[y8])){
1238
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1239
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1240
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1241
                        continue;
1242
                    }
1243

    
1244
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1245
                    if(ref0 >= 0)
1246
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1247
                    else{
1248
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1249
                        l1mv= l1mv1;
1250
                    }
1251
                    scale = dist_scale_factor[ref0];
1252
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1253

    
1254
                    {
1255
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1256
                        int my_col = (mv_col[1]<<y_shift)/2;
1257
                        int mx = (scale * mv_col[0] + 128) >> 8;
1258
                        int my = (scale * my_col + 128) >> 8;
1259
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1260
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1261
                    }
1262
                }
1263
                return;
1264
            }
1265
        }
1266

    
1267
        /* one-to-one mv scaling */
1268

    
1269
        if(IS_16X16(*mb_type)){
1270
            int ref, mv0, mv1;
1271

    
1272
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1273
            if(IS_INTRA(mb_type_col)){
1274
                ref=mv0=mv1=0;
1275
            }else{
1276
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1277
                                                : map_col_to_list0[1][l1ref1[0]];
1278
                const int scale = dist_scale_factor[ref0];
1279
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1280
                int mv_l0[2];
1281
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1282
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1283
                ref= ref0;
1284
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1285
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1286
            }
1287
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1288
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1289
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1290
        }else{
1291
            for(i8=0; i8<4; i8++){
1292
                const int x8 = i8&1;
1293
                const int y8 = i8>>1;
1294
                int ref0, scale;
1295
                const int16_t (*l1mv)[2]= l1mv0;
1296

    
1297
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1298
                    continue;
1299
                h->sub_mb_type[i8] = sub_mb_type;
1300
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1301
                if(IS_INTRA(mb_type_col)){
1302
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1303
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1304
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1305
                    continue;
1306
                }
1307

    
1308
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1309
                if(ref0 >= 0)
1310
                    ref0 = map_col_to_list0[0][ref0];
1311
                else{
1312
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1313
                    l1mv= l1mv1;
1314
                }
1315
                scale = dist_scale_factor[ref0];
1316

    
1317
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1318
                if(IS_SUB_8X8(sub_mb_type)){
1319
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1320
                    int mx = (scale * mv_col[0] + 128) >> 8;
1321
                    int my = (scale * mv_col[1] + 128) >> 8;
1322
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1323
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1324
                }else
1325
                for(i4=0; i4<4; i4++){
1326
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1327
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1328
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1329
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1330
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1331
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1332
                }
1333
            }
1334
        }
1335
    }
1336
}
1337

    
1338
static inline void write_back_motion(H264Context *h, int mb_type){
1339
    MpegEncContext * const s = &h->s;
1340
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1341
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1342
    int list;
1343

    
1344
    if(!USES_LIST(mb_type, 0))
1345
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1346

    
1347
    for(list=0; list<h->list_count; list++){
1348
        int y;
1349
        if(!USES_LIST(mb_type, list))
1350
            continue;
1351

    
1352
        for(y=0; y<4; y++){
1353
            *(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];
1354
            *(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];
1355
        }
1356
        if( h->pps.cabac ) {
1357
            if(IS_SKIP(mb_type))
1358
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1359
            else
1360
            for(y=0; y<4; y++){
1361
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1362
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1363
            }
1364
        }
1365

    
1366
        {
1367
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1368
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1369
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1370
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1371
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1372
        }
1373
    }
1374

    
1375
    if(h->slice_type == B_TYPE && h->pps.cabac){
1376
        if(IS_8X8(mb_type)){
1377
            uint8_t *direct_table = &h->direct_table[b8_xy];
1378
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1379
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1380
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1381
        }
1382
    }
1383
}
1384

    
1385
/**
1386
 * Decodes a network abstraction layer unit.
1387
 * @param consumed is the number of bytes used as input
1388
 * @param length is the length of the array
1389
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1390
 * @returns decoded bytes, might be src+1 if no escapes
1391
 */
1392
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1393
    int i, si, di;
1394
    uint8_t *dst;
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
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1424
    dst= h->rbsp_buffer;
1425

    
1426
    if (dst == NULL){
1427
        return NULL;
1428
    }
1429

    
1430
//printf("decoding esc\n");
1431
    si=di=0;
1432
    while(si<length){
1433
        //remove escapes (very rare 1:2^22)
1434
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1435
            if(src[si+2]==3){ //escape
1436
                dst[di++]= 0;
1437
                dst[di++]= 0;
1438
                si+=3;
1439
                continue;
1440
            }else //next start code
1441
                break;
1442
        }
1443

    
1444
        dst[di++]= src[si++];
1445
    }
1446

    
1447
    *dst_length= di;
1448
    *consumed= si + 1;//+1 for the header
1449
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1450
    return dst;
1451
}
1452

    
1453
/**
1454
 * identifies the exact end of the bitstream
1455
 * @return the length of the trailing, or 0 if damaged
1456
 */
1457
static int decode_rbsp_trailing(H264Context *h, uint8_t *src){
1458
    int v= *src;
1459
    int r;
1460

    
1461
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1462

    
1463
    for(r=1; r<9; r++){
1464
        if(v&1) return r;
1465
        v>>=1;
1466
    }
1467
    return 0;
1468
}
1469

    
1470
/**
1471
 * idct tranforms the 16 dc values and dequantize them.
1472
 * @param qp quantization parameter
1473
 */
1474
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1475
#define stride 16
1476
    int i;
1477
    int temp[16]; //FIXME check if this is a good idea
1478
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1479
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1480

    
1481
//memset(block, 64, 2*256);
1482
//return;
1483
    for(i=0; i<4; i++){
1484
        const int offset= y_offset[i];
1485
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1486
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1487
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1488
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1489

    
1490
        temp[4*i+0]= z0+z3;
1491
        temp[4*i+1]= z1+z2;
1492
        temp[4*i+2]= z1-z2;
1493
        temp[4*i+3]= z0-z3;
1494
    }
1495

    
1496
    for(i=0; i<4; i++){
1497
        const int offset= x_offset[i];
1498
        const int z0= temp[4*0+i] + temp[4*2+i];
1499
        const int z1= temp[4*0+i] - temp[4*2+i];
1500
        const int z2= temp[4*1+i] - temp[4*3+i];
1501
        const int z3= temp[4*1+i] + temp[4*3+i];
1502

    
1503
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1504
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1505
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1506
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1507
    }
1508
}
1509

    
1510
#if 0
1511
/**
1512
 * dct tranforms the 16 dc values.
1513
 * @param qp quantization parameter ??? FIXME
1514
 */
1515
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1516
//    const int qmul= dequant_coeff[qp][0];
1517
    int i;
1518
    int temp[16]; //FIXME check if this is a good idea
1519
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1520
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1521

1522
    for(i=0; i<4; i++){
1523
        const int offset= y_offset[i];
1524
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1525
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1526
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1527
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1528

1529
        temp[4*i+0]= z0+z3;
1530
        temp[4*i+1]= z1+z2;
1531
        temp[4*i+2]= z1-z2;
1532
        temp[4*i+3]= z0-z3;
1533
    }
1534

1535
    for(i=0; i<4; i++){
1536
        const int offset= x_offset[i];
1537
        const int z0= temp[4*0+i] + temp[4*2+i];
1538
        const int z1= temp[4*0+i] - temp[4*2+i];
1539
        const int z2= temp[4*1+i] - temp[4*3+i];
1540
        const int z3= temp[4*1+i] + temp[4*3+i];
1541

1542
        block[stride*0 +offset]= (z0 + z3)>>1;
1543
        block[stride*2 +offset]= (z1 + z2)>>1;
1544
        block[stride*8 +offset]= (z1 - z2)>>1;
1545
        block[stride*10+offset]= (z0 - z3)>>1;
1546
    }
1547
}
1548
#endif
1549

    
1550
#undef xStride
1551
#undef stride
1552

    
1553
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1554
    const int stride= 16*2;
1555
    const int xStride= 16;
1556
    int a,b,c,d,e;
1557

    
1558
    a= block[stride*0 + xStride*0];
1559
    b= block[stride*0 + xStride*1];
1560
    c= block[stride*1 + xStride*0];
1561
    d= block[stride*1 + xStride*1];
1562

    
1563
    e= a-b;
1564
    a= a+b;
1565
    b= c-d;
1566
    c= c+d;
1567

    
1568
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1569
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1570
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1571
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1572
}
1573

    
1574
#if 0
1575
static void chroma_dc_dct_c(DCTELEM *block){
1576
    const int stride= 16*2;
1577
    const int xStride= 16;
1578
    int a,b,c,d,e;
1579

1580
    a= block[stride*0 + xStride*0];
1581
    b= block[stride*0 + xStride*1];
1582
    c= block[stride*1 + xStride*0];
1583
    d= block[stride*1 + xStride*1];
1584

1585
    e= a-b;
1586
    a= a+b;
1587
    b= c-d;
1588
    c= c+d;
1589

1590
    block[stride*0 + xStride*0]= (a+c);
1591
    block[stride*0 + xStride*1]= (e+b);
1592
    block[stride*1 + xStride*0]= (a-c);
1593
    block[stride*1 + xStride*1]= (e-b);
1594
}
1595
#endif
1596

    
1597
/**
1598
 * gets the chroma qp.
1599
 */
1600
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1601

    
1602
    return chroma_qp[av_clip(qscale + chroma_qp_index_offset, 0, 51)];
1603
}
1604

    
1605
//FIXME need to check that this doesnt 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 t4= topright[0];\
1737
    const int t5= topright[1];\
1738
    const int t6= topright[2];\
1739
    const int t7= topright[3];\
1740

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

    
1747
#define LOAD_TOP_EDGE\
1748
    const int t0= src[ 0-1*stride];\
1749
    const int t1= src[ 1-1*stride];\
1750
    const int t2= src[ 2-1*stride];\
1751
    const int 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
    const __attribute__((unused)) int unu= l3;
1804

    
1805
    src[0+0*stride]=
1806
    src[1+2*stride]=(lt + t0 + 1)>>1;
1807
    src[1+0*stride]=
1808
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1809
    src[2+0*stride]=
1810
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1811
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1812
    src[0+1*stride]=
1813
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1814
    src[1+1*stride]=
1815
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1816
    src[2+1*stride]=
1817
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1818
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1819
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1820
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1821
}
1822

    
1823
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1824
    LOAD_TOP_EDGE
1825
    LOAD_TOP_RIGHT_EDGE
1826
    const __attribute__((unused)) int unu= t7;
1827

    
1828
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1829
    src[1+0*stride]=
1830
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1831
    src[2+0*stride]=
1832
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1833
    src[3+0*stride]=
1834
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1835
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1836
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1837
    src[1+1*stride]=
1838
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1839
    src[2+1*stride]=
1840
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1841
    src[3+1*stride]=
1842
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1843
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1844
}
1845

    
1846
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1847
    LOAD_LEFT_EDGE
1848

    
1849
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1850
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1851
    src[2+0*stride]=
1852
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1853
    src[3+0*stride]=
1854
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1855
    src[2+1*stride]=
1856
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1857
    src[3+1*stride]=
1858
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1859
    src[3+2*stride]=
1860
    src[1+3*stride]=
1861
    src[0+3*stride]=
1862
    src[2+2*stride]=
1863
    src[2+3*stride]=
1864
    src[3+3*stride]=l3;
1865
}
1866

    
1867
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1868
    const int lt= src[-1-1*stride];
1869
    LOAD_TOP_EDGE
1870
    LOAD_LEFT_EDGE
1871
    const __attribute__((unused)) int unu= t3;
1872

    
1873
    src[0+0*stride]=
1874
    src[2+1*stride]=(lt + l0 + 1)>>1;
1875
    src[1+0*stride]=
1876
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1877
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1878
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1879
    src[0+1*stride]=
1880
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1881
    src[1+1*stride]=
1882
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1883
    src[0+2*stride]=
1884
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1885
    src[1+2*stride]=
1886
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1887
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1888
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1889
}
1890

    
1891
void ff_pred16x16_vertical_c(uint8_t *src, int stride){
1892
    int i;
1893
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1894
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1895
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1896
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1897

    
1898
    for(i=0; i<16; i++){
1899
        ((uint32_t*)(src+i*stride))[0]= a;
1900
        ((uint32_t*)(src+i*stride))[1]= b;
1901
        ((uint32_t*)(src+i*stride))[2]= c;
1902
        ((uint32_t*)(src+i*stride))[3]= d;
1903
    }
1904
}
1905

    
1906
void ff_pred16x16_horizontal_c(uint8_t *src, int stride){
1907
    int i;
1908

    
1909
    for(i=0; i<16; i++){
1910
        ((uint32_t*)(src+i*stride))[0]=
1911
        ((uint32_t*)(src+i*stride))[1]=
1912
        ((uint32_t*)(src+i*stride))[2]=
1913
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1914
    }
1915
}
1916

    
1917
void ff_pred16x16_dc_c(uint8_t *src, int stride){
1918
    int i, dc=0;
1919

    
1920
    for(i=0;i<16; i++){
1921
        dc+= src[-1+i*stride];
1922
    }
1923

    
1924
    for(i=0;i<16; i++){
1925
        dc+= src[i-stride];
1926
    }
1927

    
1928
    dc= 0x01010101*((dc + 16)>>5);
1929

    
1930
    for(i=0; i<16; i++){
1931
        ((uint32_t*)(src+i*stride))[0]=
1932
        ((uint32_t*)(src+i*stride))[1]=
1933
        ((uint32_t*)(src+i*stride))[2]=
1934
        ((uint32_t*)(src+i*stride))[3]= dc;
1935
    }
1936
}
1937

    
1938
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1939
    int i, dc=0;
1940

    
1941
    for(i=0;i<16; i++){
1942
        dc+= src[-1+i*stride];
1943
    }
1944

    
1945
    dc= 0x01010101*((dc + 8)>>4);
1946

    
1947
    for(i=0; i<16; i++){
1948
        ((uint32_t*)(src+i*stride))[0]=
1949
        ((uint32_t*)(src+i*stride))[1]=
1950
        ((uint32_t*)(src+i*stride))[2]=
1951
        ((uint32_t*)(src+i*stride))[3]= dc;
1952
    }
1953
}
1954

    
1955
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1956
    int i, dc=0;
1957

    
1958
    for(i=0;i<16; i++){
1959
        dc+= src[i-stride];
1960
    }
1961
    dc= 0x01010101*((dc + 8)>>4);
1962

    
1963
    for(i=0; i<16; i++){
1964
        ((uint32_t*)(src+i*stride))[0]=
1965
        ((uint32_t*)(src+i*stride))[1]=
1966
        ((uint32_t*)(src+i*stride))[2]=
1967
        ((uint32_t*)(src+i*stride))[3]= dc;
1968
    }
1969
}
1970

    
1971
void ff_pred16x16_128_dc_c(uint8_t *src, int stride){
1972
    int i;
1973

    
1974
    for(i=0; i<16; i++){
1975
        ((uint32_t*)(src+i*stride))[0]=
1976
        ((uint32_t*)(src+i*stride))[1]=
1977
        ((uint32_t*)(src+i*stride))[2]=
1978
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1979
    }
1980
}
1981

    
1982
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1983
  int i, j, k;
1984
  int a;
1985
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
1986
  const uint8_t * const src0 = src+7-stride;
1987
  const uint8_t *src1 = src+8*stride-1;
1988
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1989
  int H = src0[1] - src0[-1];
1990
  int V = src1[0] - src2[ 0];
1991
  for(k=2; k<=8; ++k) {
1992
    src1 += stride; src2 -= stride;
1993
    H += k*(src0[k] - src0[-k]);
1994
    V += k*(src1[0] - src2[ 0]);
1995
  }
1996
  if(svq3){
1997
    H = ( 5*(H/4) ) / 16;
1998
    V = ( 5*(V/4) ) / 16;
1999

    
2000
    /* required for 100% accuracy */
2001
    i = H; H = V; V = i;
2002
  }else{
2003
    H = ( 5*H+32 ) >> 6;
2004
    V = ( 5*V+32 ) >> 6;
2005
  }
2006

    
2007
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2008
  for(j=16; j>0; --j) {
2009
    int b = a;
2010
    a += V;
2011
    for(i=-16; i<0; i+=4) {
2012
      src[16+i] = cm[ (b    ) >> 5 ];
2013
      src[17+i] = cm[ (b+  H) >> 5 ];
2014
      src[18+i] = cm[ (b+2*H) >> 5 ];
2015
      src[19+i] = cm[ (b+3*H) >> 5 ];
2016
      b += 4*H;
2017
    }
2018
    src += stride;
2019
  }
2020
}
2021

    
2022
void ff_pred16x16_plane_c(uint8_t *src, int stride){
2023
    pred16x16_plane_compat_c(src, stride, 0);
2024
}
2025

    
2026
void ff_pred8x8_vertical_c(uint8_t *src, int stride){
2027
    int i;
2028
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2029
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2030

    
2031
    for(i=0; i<8; i++){
2032
        ((uint32_t*)(src+i*stride))[0]= a;
2033
        ((uint32_t*)(src+i*stride))[1]= b;
2034
    }
2035
}
2036

    
2037
void ff_pred8x8_horizontal_c(uint8_t *src, int stride){
2038
    int i;
2039

    
2040
    for(i=0; i<8; i++){
2041
        ((uint32_t*)(src+i*stride))[0]=
2042
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2043
    }
2044
}
2045

    
2046
void ff_pred8x8_128_dc_c(uint8_t *src, int stride){
2047
    int i;
2048

    
2049
    for(i=0; i<8; i++){
2050
        ((uint32_t*)(src+i*stride))[0]=
2051
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2052
    }
2053
}
2054

    
2055
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2056
    int i;
2057
    int dc0, dc2;
2058

    
2059
    dc0=dc2=0;
2060
    for(i=0;i<4; i++){
2061
        dc0+= src[-1+i*stride];
2062
        dc2+= src[-1+(i+4)*stride];
2063
    }
2064
    dc0= 0x01010101*((dc0 + 2)>>2);
2065
    dc2= 0x01010101*((dc2 + 2)>>2);
2066

    
2067
    for(i=0; i<4; i++){
2068
        ((uint32_t*)(src+i*stride))[0]=
2069
        ((uint32_t*)(src+i*stride))[1]= dc0;
2070
    }
2071
    for(i=4; i<8; i++){
2072
        ((uint32_t*)(src+i*stride))[0]=
2073
        ((uint32_t*)(src+i*stride))[1]= dc2;
2074
    }
2075
}
2076

    
2077
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2078
    int i;
2079
    int dc0, dc1;
2080

    
2081
    dc0=dc1=0;
2082
    for(i=0;i<4; i++){
2083
        dc0+= src[i-stride];
2084
        dc1+= src[4+i-stride];
2085
    }
2086
    dc0= 0x01010101*((dc0 + 2)>>2);
2087
    dc1= 0x01010101*((dc1 + 2)>>2);
2088

    
2089
    for(i=0; i<4; i++){
2090
        ((uint32_t*)(src+i*stride))[0]= dc0;
2091
        ((uint32_t*)(src+i*stride))[1]= dc1;
2092
    }
2093
    for(i=4; i<8; i++){
2094
        ((uint32_t*)(src+i*stride))[0]= dc0;
2095
        ((uint32_t*)(src+i*stride))[1]= dc1;
2096
    }
2097
}
2098

    
2099

    
2100
void ff_pred8x8_dc_c(uint8_t *src, int stride){
2101
    int i;
2102
    int dc0, dc1, dc2, dc3;
2103

    
2104
    dc0=dc1=dc2=0;
2105
    for(i=0;i<4; i++){
2106
        dc0+= src[-1+i*stride] + src[i-stride];
2107
        dc1+= src[4+i-stride];
2108
        dc2+= src[-1+(i+4)*stride];
2109
    }
2110
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2111
    dc0= 0x01010101*((dc0 + 4)>>3);
2112
    dc1= 0x01010101*((dc1 + 2)>>2);
2113
    dc2= 0x01010101*((dc2 + 2)>>2);
2114

    
2115
    for(i=0; i<4; i++){
2116
        ((uint32_t*)(src+i*stride))[0]= dc0;
2117
        ((uint32_t*)(src+i*stride))[1]= dc1;
2118
    }
2119
    for(i=4; i<8; i++){
2120
        ((uint32_t*)(src+i*stride))[0]= dc2;
2121
        ((uint32_t*)(src+i*stride))[1]= dc3;
2122
    }
2123
}
2124

    
2125
void ff_pred8x8_plane_c(uint8_t *src, int stride){
2126
  int j, k;
2127
  int a;
2128
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2129
  const uint8_t * const src0 = src+3-stride;
2130
  const uint8_t *src1 = src+4*stride-1;
2131
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2132
  int H = src0[1] - src0[-1];
2133
  int V = src1[0] - src2[ 0];
2134
  for(k=2; k<=4; ++k) {
2135
    src1 += stride; src2 -= stride;
2136
    H += k*(src0[k] - src0[-k]);
2137
    V += k*(src1[0] - src2[ 0]);
2138
  }
2139
  H = ( 17*H+16 ) >> 5;
2140
  V = ( 17*V+16 ) >> 5;
2141

    
2142
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2143
  for(j=8; j>0; --j) {
2144
    int b = a;
2145
    a += V;
2146
    src[0] = cm[ (b    ) >> 5 ];
2147
    src[1] = cm[ (b+  H) >> 5 ];
2148
    src[2] = cm[ (b+2*H) >> 5 ];
2149
    src[3] = cm[ (b+3*H) >> 5 ];
2150
    src[4] = cm[ (b+4*H) >> 5 ];
2151
    src[5] = cm[ (b+5*H) >> 5 ];
2152
    src[6] = cm[ (b+6*H) >> 5 ];
2153
    src[7] = cm[ (b+7*H) >> 5 ];
2154
    src += stride;
2155
  }
2156
}
2157

    
2158
#define SRC(x,y) src[(x)+(y)*stride]
2159
#define PL(y) \
2160
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2161
#define PREDICT_8x8_LOAD_LEFT \
2162
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2163
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2164
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2165
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2166

    
2167
#define PT(x) \
2168
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2169
#define PREDICT_8x8_LOAD_TOP \
2170
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2171
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2172
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2173
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2174
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2175

    
2176
#define PTR(x) \
2177
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2178
#define PREDICT_8x8_LOAD_TOPRIGHT \
2179
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2180
    if(has_topright) { \
2181
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2182
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2183
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2184

    
2185
#define PREDICT_8x8_LOAD_TOPLEFT \
2186
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2187

    
2188
#define PREDICT_8x8_DC(v) \
2189
    int y; \
2190
    for( y = 0; y < 8; y++ ) { \
2191
        ((uint32_t*)src)[0] = \
2192
        ((uint32_t*)src)[1] = v; \
2193
        src += stride; \
2194
    }
2195

    
2196
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2197
{
2198
    PREDICT_8x8_DC(0x80808080);
2199
}
2200
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2201
{
2202
    PREDICT_8x8_LOAD_LEFT;
2203
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2204
    PREDICT_8x8_DC(dc);
2205
}
2206
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2207
{
2208
    PREDICT_8x8_LOAD_TOP;
2209
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2210
    PREDICT_8x8_DC(dc);
2211
}
2212
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2213
{
2214
    PREDICT_8x8_LOAD_LEFT;
2215
    PREDICT_8x8_LOAD_TOP;
2216
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2217
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2218
    PREDICT_8x8_DC(dc);
2219
}
2220
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2221
{
2222
    PREDICT_8x8_LOAD_LEFT;
2223
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2224
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2225
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2226
#undef ROW
2227
}
2228
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2229
{
2230
    int y;
2231
    PREDICT_8x8_LOAD_TOP;
2232
    src[0] = t0;
2233
    src[1] = t1;
2234
    src[2] = t2;
2235
    src[3] = t3;
2236
    src[4] = t4;
2237
    src[5] = t5;
2238
    src[6] = t6;
2239
    src[7] = t7;
2240
    for( y = 1; y < 8; y++ )
2241
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2242
}
2243
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2244
{
2245
    PREDICT_8x8_LOAD_TOP;
2246
    PREDICT_8x8_LOAD_TOPRIGHT;
2247
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2248
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2249
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2250
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2251
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2252
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2253
    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;
2254
    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;
2255
    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;
2256
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2257
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2258
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2259
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2260
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2261
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2262
}
2263
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2264
{
2265
    PREDICT_8x8_LOAD_TOP;
2266
    PREDICT_8x8_LOAD_LEFT;
2267
    PREDICT_8x8_LOAD_TOPLEFT;
2268
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2269
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2270
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2271
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2272
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2273
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2274
    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;
2275
    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;
2276
    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;
2277
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2278
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2279
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2280
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2281
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2282
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2283

    
2284
}
2285
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2286
{
2287
    PREDICT_8x8_LOAD_TOP;
2288
    PREDICT_8x8_LOAD_LEFT;
2289
    PREDICT_8x8_LOAD_TOPLEFT;
2290
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2291
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2292
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2293
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2294
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2295
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2296
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2297
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2298
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2299
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2300
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2301
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2302
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2303
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2304
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2305
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2306
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2307
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2308
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2309
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2310
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2311
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2312
}
2313
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2314
{
2315
    PREDICT_8x8_LOAD_TOP;
2316
    PREDICT_8x8_LOAD_LEFT;
2317
    PREDICT_8x8_LOAD_TOPLEFT;
2318
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2319
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2320
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2321
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2322
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2323
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2324
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2325
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2326
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2327
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2328
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2329
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2330
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2331
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2332
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2333
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2334
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2335
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2336
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2337
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2338
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2339
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2340
}
2341
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2342
{
2343
    PREDICT_8x8_LOAD_TOP;
2344
    PREDICT_8x8_LOAD_TOPRIGHT;
2345
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2346
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2347
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2348
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2349
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2350
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2351
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2352
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2353
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2354
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2355
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2356
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2357
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2358
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2359
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2360
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2361
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2362
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2363
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2364
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2365
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2366
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2367
}
2368
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2369
{
2370
    PREDICT_8x8_LOAD_LEFT;
2371
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2372
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2373
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2374
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2375
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2376
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2377
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2378
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2379
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2380
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2381
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2382
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2383
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2384
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2385
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2386
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2387
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2388
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2389
}
2390
#undef PREDICT_8x8_LOAD_LEFT
2391
#undef PREDICT_8x8_LOAD_TOP
2392
#undef PREDICT_8x8_LOAD_TOPLEFT
2393
#undef PREDICT_8x8_LOAD_TOPRIGHT
2394
#undef PREDICT_8x8_DC
2395
#undef PTR
2396
#undef PT
2397
#undef PL
2398
#undef SRC
2399

    
2400
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2401
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2402
                           int src_x_offset, int src_y_offset,
2403
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2404
    MpegEncContext * const s = &h->s;
2405
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2406
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2407
    const int luma_xy= (mx&3) + ((my&3)<<2);
2408
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2409
    uint8_t * src_cb, * src_cr;
2410
    int extra_width= h->emu_edge_width;
2411
    int extra_height= h->emu_edge_height;
2412
    int emu=0;
2413
    const int full_mx= mx>>2;
2414
    const int full_my= my>>2;
2415
    const int pic_width  = 16*s->mb_width;
2416
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
2417

    
2418
    if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
2419
        return;
2420

    
2421
    if(mx&7) extra_width -= 3;
2422
    if(my&7) extra_height -= 3;
2423

    
2424
    if(   full_mx < 0-extra_width
2425
       || full_my < 0-extra_height
2426
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2427
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2428
        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);
2429
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2430
        emu=1;
2431
    }
2432

    
2433
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2434
    if(!square){
2435
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2436
    }
2437

    
2438
    if(s->flags&CODEC_FLAG_GRAY) return;
2439

    
2440
    if(MB_MBAFF){
2441
        // chroma offset when predicting from a field of opposite parity
2442
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2443
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2444
    }
2445
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2446
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2447

    
2448
    if(emu){
2449
        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);
2450
            src_cb= s->edge_emu_buffer;
2451
    }
2452
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2453

    
2454
    if(emu){
2455
        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);
2456
            src_cr= s->edge_emu_buffer;
2457
    }
2458
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2459
}
2460

    
2461
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2462
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2463
                           int x_offset, int y_offset,
2464
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2465
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2466
                           int list0, int list1){
2467
    MpegEncContext * const s = &h->s;
2468
    qpel_mc_func *qpix_op=  qpix_put;
2469
    h264_chroma_mc_func chroma_op= chroma_put;
2470

    
2471
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2472
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2473
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2474
    x_offset += 8*s->mb_x;
2475
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2476

    
2477
    if(list0){
2478
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2479
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2480
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2481
                           qpix_op, chroma_op);
2482

    
2483
        qpix_op=  qpix_avg;
2484
        chroma_op= chroma_avg;
2485
    }
2486

    
2487
    if(list1){
2488
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2489
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2490
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2491
                           qpix_op, chroma_op);
2492
    }
2493
}
2494

    
2495
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2496
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2497
                           int x_offset, int y_offset,
2498
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2499
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2500
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2501
                           int list0, int list1){
2502
    MpegEncContext * const s = &h->s;
2503

    
2504
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2505
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2506
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2507
    x_offset += 8*s->mb_x;
2508
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2509

    
2510
    if(list0 && list1){
2511
        /* don't optimize for luma-only case, since B-frames usually
2512
         * use implicit weights => chroma too. */
2513
        uint8_t *tmp_cb = s->obmc_scratchpad;
2514
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2515
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2516
        int refn0 = h->ref_cache[0][ scan8[n] ];
2517
        int refn1 = h->ref_cache[1][ scan8[n] ];
2518

    
2519
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2520
                    dest_y, dest_cb, dest_cr,
2521
                    x_offset, y_offset, qpix_put, chroma_put);
2522
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2523
                    tmp_y, tmp_cb, tmp_cr,
2524
                    x_offset, y_offset, qpix_put, chroma_put);
2525

    
2526
        if(h->use_weight == 2){
2527
            int weight0 = h->implicit_weight[refn0][refn1];
2528
            int weight1 = 64 - weight0;
2529
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
2530
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2531
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2532
        }else{
2533
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2534
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2535
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2536
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2537
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2538
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2539
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2540
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2541
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2542
        }
2543
    }else{
2544
        int list = list1 ? 1 : 0;
2545
        int refn = h->ref_cache[list][ scan8[n] ];
2546
        Picture *ref= &h->ref_list[list][refn];
2547
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2548
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2549
                    qpix_put, chroma_put);
2550

    
2551
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
2552
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2553
        if(h->use_weight_chroma){
2554
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2555
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2556
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2557
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2558
        }
2559
    }
2560
}
2561

    
2562
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2563
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2564
                           int x_offset, int y_offset,
2565
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2566
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2567
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2568
                           int list0, int list1){
2569
    if((h->use_weight==2 && list0 && list1
2570
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2571
       || h->use_weight==1)
2572
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2573
                         x_offset, y_offset, qpix_put, chroma_put,
2574
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2575
    else
2576
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2577
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2578
}
2579

    
2580
static inline void prefetch_motion(H264Context *h, int list){
2581
    /* fetch pixels for estimated mv 4 macroblocks ahead
2582
     * optimized for 64byte cache lines */
2583
    MpegEncContext * const s = &h->s;
2584
    const int refn = h->ref_cache[list][scan8[0]];
2585
    if(refn >= 0){
2586
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2587
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2588
        uint8_t **src= h->ref_list[list][refn].data;
2589
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
2590
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
2591
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2592
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2593
    }
2594
}
2595

    
2596
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2597
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2598
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2599
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2600
    MpegEncContext * const s = &h->s;
2601
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2602
    const int mb_type= s->current_picture.mb_type[mb_xy];
2603

    
2604
    assert(IS_INTER(mb_type));
2605

    
2606
    prefetch_motion(h, 0);
2607

    
2608
    if(IS_16X16(mb_type)){
2609
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2610
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2611
                &weight_op[0], &weight_avg[0],
2612
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2613
    }else if(IS_16X8(mb_type)){
2614
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2615
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2616
                &weight_op[1], &weight_avg[1],
2617
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2618
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2619
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2620
                &weight_op[1], &weight_avg[1],
2621
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2622
    }else if(IS_8X16(mb_type)){
2623
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
2624
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2625
                &weight_op[2], &weight_avg[2],
2626
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2627
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
2628
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2629
                &weight_op[2], &weight_avg[2],
2630
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2631
    }else{
2632
        int i;
2633

    
2634
        assert(IS_8X8(mb_type));
2635

    
2636
        for(i=0; i<4; i++){
2637
            const int sub_mb_type= h->sub_mb_type[i];
2638
            const int n= 4*i;
2639
            int x_offset= (i&1)<<2;
2640
            int y_offset= (i&2)<<1;
2641

    
2642
            if(IS_SUB_8X8(sub_mb_type)){
2643
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2644
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2645
                    &weight_op[3], &weight_avg[3],
2646
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2647
            }else if(IS_SUB_8X4(sub_mb_type)){
2648
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2649
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2650
                    &weight_op[4], &weight_avg[4],
2651
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2652
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2653
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2654
                    &weight_op[4], &weight_avg[4],
2655
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2656
            }else if(IS_SUB_4X8(sub_mb_type)){
2657
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2658
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2659
                    &weight_op[5], &weight_avg[5],
2660
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2661
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2662
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2663
                    &weight_op[5], &weight_avg[5],
2664
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2665
            }else{
2666
                int j;
2667
                assert(IS_SUB_4X4(sub_mb_type));
2668
                for(j=0; j<4; j++){
2669
                    int sub_x_offset= x_offset + 2*(j&1);
2670
                    int sub_y_offset= y_offset +   (j&2);
2671
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2672
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2673
                        &weight_op[6], &weight_avg[6],
2674
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2675
                }
2676
            }
2677
        }
2678
    }
2679

    
2680
    prefetch_motion(h, 1);
2681
}
2682

    
2683
static void decode_init_vlc(void){
2684
    static int done = 0;
2685

    
2686
    if (!done) {
2687
        int i;
2688
        done = 1;
2689

    
2690
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2691
                 &chroma_dc_coeff_token_len [0], 1, 1,
2692
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2693

    
2694
        for(i=0; i<4; i++){
2695
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2696
                     &coeff_token_len [i][0], 1, 1,
2697
                     &coeff_token_bits[i][0], 1, 1, 1);
2698
        }
2699

    
2700
        for(i=0; i<3; i++){
2701
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2702
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2703
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2704
        }
2705
        for(i=0; i<15; i++){
2706
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2707
                     &total_zeros_len [i][0], 1, 1,
2708
                     &total_zeros_bits[i][0], 1, 1, 1);
2709
        }
2710

    
2711
        for(i=0; i<6; i++){
2712
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2713
                     &run_len [i][0], 1, 1,
2714
                     &run_bits[i][0], 1, 1, 1);
2715
        }
2716
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2717
                 &run_len [6][0], 1, 1,
2718
                 &run_bits[6][0], 1, 1, 1);
2719
    }
2720
}
2721

    
2722
/**
2723
 * Sets the intra prediction function pointers.
2724
 */
2725
static void init_pred_ptrs(H264Context *h){
2726
//    MpegEncContext * const s = &h->s;
2727

    
2728
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2729
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2730
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2731
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2732
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2733
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2734
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2735
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2736
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2737
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2738
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2739
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2740

    
2741
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2742
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2743
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2744
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2745
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2746
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2747
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2748
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2749
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2750
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2751
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2752
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2753

    
2754
    h->pred8x8[DC_PRED8x8     ]= ff_pred8x8_dc_c;
2755
    h->pred8x8[VERT_PRED8x8   ]= ff_pred8x8_vertical_c;
2756
    h->pred8x8[HOR_PRED8x8    ]= ff_pred8x8_horizontal_c;
2757
    h->pred8x8[PLANE_PRED8x8  ]= ff_pred8x8_plane_c;
2758
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2759
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2760
    h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c;
2761

    
2762
    h->pred16x16[DC_PRED8x8     ]= ff_pred16x16_dc_c;
2763
    h->pred16x16[VERT_PRED8x8   ]= ff_pred16x16_vertical_c;
2764
    h->pred16x16[HOR_PRED8x8    ]= ff_pred16x16_horizontal_c;
2765
    h->pred16x16[PLANE_PRED8x8  ]= ff_pred16x16_plane_c;
2766
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2767
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2768
    h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c;
2769
}
2770

    
2771
static void free_tables(H264Context *h){
2772
    av_freep(&h->intra4x4_pred_mode);
2773
    av_freep(&h->chroma_pred_mode_table);
2774
    av_freep(&h->cbp_table);
2775
    av_freep(&h->mvd_table[0]);
2776
    av_freep(&h->mvd_table[1]);
2777
    av_freep(&h->direct_table);
2778
    av_freep(&h->non_zero_count);
2779
    av_freep(&h->slice_table_base);
2780
    av_freep(&h->top_borders[1]);
2781
    av_freep(&h->top_borders[0]);
2782
    h->slice_table= NULL;
2783

    
2784
    av_freep(&h->mb2b_xy);
2785
    av_freep(&h->mb2b8_xy);
2786

    
2787
    av_freep(&h->s.obmc_scratchpad);
2788
}
2789

    
2790
static void init_dequant8_coeff_table(H264Context *h){
2791
    int i,q,x;
2792
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2793
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2794
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2795

    
2796
    for(i=0; i<2; i++ ){
2797
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2798
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2799
            break;
2800
        }
2801

    
2802
        for(q=0; q<52; q++){
2803
            int shift = ff_div6[q];
2804
            int idx = ff_rem6[q];
2805
            for(x=0; x<64; x++)
2806
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2807
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2808
                    h->pps.scaling_matrix8[i][x]) << shift;
2809
        }
2810
    }
2811
}
2812

    
2813
static void init_dequant4_coeff_table(H264Context *h){
2814
    int i,j,q,x;
2815
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2816
    for(i=0; i<6; i++ ){
2817
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2818
        for(j=0; j<i; j++){
2819
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2820
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2821
                break;
2822
            }
2823
        }
2824
        if(j<i)
2825
            continue;
2826

    
2827
        for(q=0; q<52; q++){
2828
            int shift = ff_div6[q] + 2;
2829
            int idx = ff_rem6[q];
2830
            for(x=0; x<16; x++)
2831
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2832
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2833
                    h->pps.scaling_matrix4[i][x]) << shift;
2834
        }
2835
    }
2836
}
2837

    
2838
static void init_dequant_tables(H264Context *h){
2839
    int i,x;
2840
    init_dequant4_coeff_table(h);
2841
    if(h->pps.transform_8x8_mode)
2842
        init_dequant8_coeff_table(h);
2843
    if(h->sps.transform_bypass){
2844
        for(i=0; i<6; i++)
2845
            for(x=0; x<16; x++)
2846
                h->dequant4_coeff[i][0][x] = 1<<6;
2847
        if(h->pps.transform_8x8_mode)
2848
            for(i=0; i<2; i++)
2849
                for(x=0; x<64; x++)
2850
                    h->dequant8_coeff[i][0][x] = 1<<6;
2851
    }
2852
}
2853

    
2854

    
2855
/**
2856
 * allocates tables.
2857
 * needs width/height
2858
 */
2859
static int alloc_tables(H264Context *h){
2860
    MpegEncContext * const s = &h->s;
2861
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2862
    int x,y;
2863

    
2864
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2865

    
2866
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2867
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2868
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2869
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2870
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2871

    
2872
    if( h->pps.cabac ) {
2873
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2874
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2875
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2876
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2877
    }
2878

    
2879
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2880
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2881

    
2882
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2883
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2884
    for(y=0; y<s->mb_height; y++){
2885
        for(x=0; x<s->mb_width; x++){
2886
            const int mb_xy= x + y*s->mb_stride;
2887
            const int b_xy = 4*x + 4*y*h->b_stride;
2888
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2889

    
2890
            h->mb2b_xy [mb_xy]= b_xy;
2891
            h->mb2b8_xy[mb_xy]= b8_xy;
2892
        }
2893
    }
2894

    
2895
    s->obmc_scratchpad = NULL;
2896

    
2897
    if(!h->dequant4_coeff[0])
2898
        init_dequant_tables(h);
2899

    
2900
    return 0;
2901
fail:
2902
    free_tables(h);
2903
    return -1;
2904
}
2905

    
2906
static void common_init(H264Context *h){
2907
    MpegEncContext * const s = &h->s;
2908

    
2909
    s->width = s->avctx->width;
2910
    s->height = s->avctx->height;
2911
    s->codec_id= s->avctx->codec->id;
2912

    
2913
    init_pred_ptrs(h);
2914

    
2915
    h->dequant_coeff_pps= -1;
2916
    s->unrestricted_mv=1;
2917
    s->decode=1; //FIXME
2918

    
2919
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2920
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2921
}
2922

    
2923
static int decode_init(AVCodecContext *avctx){
2924
    H264Context *h= avctx->priv_data;
2925
    MpegEncContext * const s = &h->s;
2926

    
2927
    MPV_decode_defaults(s);
2928

    
2929
    s->avctx = avctx;
2930
    common_init(h);
2931

    
2932
    s->out_format = FMT_H264;
2933
    s->workaround_bugs= avctx->workaround_bugs;
2934

    
2935
    // set defaults
2936
//    s->decode_mb= ff_h263_decode_mb;
2937
    s->low_delay= 1;
2938
    avctx->pix_fmt= PIX_FMT_YUV420P;
2939

    
2940
    decode_init_vlc();
2941

    
2942
    if(avctx->extradata_size > 0 && avctx->extradata &&
2943
       *(char *)avctx->extradata == 1){
2944
        h->is_avc = 1;
2945
        h->got_avcC = 0;
2946
    } else {
2947
        h->is_avc = 0;
2948
    }
2949

    
2950
    return 0;
2951
}
2952

    
2953
static int frame_start(H264Context *h){
2954
    MpegEncContext * const s = &h->s;
2955
    int i;
2956

    
2957
    if(MPV_frame_start(s, s->avctx) < 0)
2958
        return -1;
2959
    ff_er_frame_start(s);
2960

    
2961
    assert(s->linesize && s->uvlinesize);
2962

    
2963
    for(i=0; i<16; i++){
2964
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2965
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2966
    }
2967
    for(i=0; i<4; i++){
2968
        h->block_offset[16+i]=
2969
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2970
        h->block_offset[24+16+i]=
2971
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2972
    }
2973

    
2974
    /* can't be in alloc_tables because linesize isn't known there.
2975
     * FIXME: redo bipred weight to not require extra buffer? */
2976
    if(!s->obmc_scratchpad)
2977
        s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2978

    
2979
    /* some macroblocks will be accessed before they're available */
2980
    if(FRAME_MBAFF)
2981
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2982

    
2983
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2984
    return 0;
2985
}
2986

    
2987
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2988
    MpegEncContext * const s = &h->s;
2989
    int i;
2990

    
2991
    src_y  -=   linesize;
2992
    src_cb -= uvlinesize;
2993
    src_cr -= uvlinesize;
2994

    
2995
    // There are two lines saved, the line above the the top macroblock of a pair,
2996
    // and the line above the bottom macroblock
2997
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2998
    for(i=1; i<17; i++){
2999
        h->left_border[i]= src_y[15+i*  linesize];
3000
    }
3001

    
3002
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3003
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3004

    
3005
    if(!(s->flags&CODEC_FLAG_GRAY)){
3006
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3007
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3008
        for(i=1; i<9; i++){
3009
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3010
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3011
        }
3012
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3013
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3014
    }
3015
}
3016

    
3017
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){
3018
    MpegEncContext * const s = &h->s;
3019
    int temp8, i;
3020
    uint64_t temp64;
3021
    int deblock_left = (s->mb_x > 0);
3022
    int deblock_top  = (s->mb_y > 0);
3023

    
3024
    src_y  -=   linesize + 1;
3025
    src_cb -= uvlinesize + 1;
3026
    src_cr -= uvlinesize + 1;
3027

    
3028
#define XCHG(a,b,t,xchg)\
3029
t= a;\
3030
if(xchg)\
3031
    a= b;\
3032
b= t;
3033

    
3034
    if(deblock_left){
3035
        for(i = !deblock_top; i<17; i++){
3036
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3037
        }
3038
    }
3039

    
3040
    if(deblock_top){
3041
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3042
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3043
        if(s->mb_x+1 < s->mb_width){
3044
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3045
        }
3046
    }
3047

    
3048
    if(!(s->flags&CODEC_FLAG_GRAY)){
3049
        if(deblock_left){
3050
            for(i = !deblock_top; i<9; i++){
3051
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3052
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3053
            }
3054
        }
3055
        if(deblock_top){
3056
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3057
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3058
        }
3059
    }
3060
}
3061

    
3062
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3063
    MpegEncContext * const s = &h->s;
3064
    int i;
3065

    
3066
    src_y  -= 2 *   linesize;
3067
    src_cb -= 2 * uvlinesize;
3068
    src_cr -= 2 * uvlinesize;
3069

    
3070
    // There are two lines saved, the line above the the top macroblock of a pair,
3071
    // and the line above the bottom macroblock
3072
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3073
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3074
    for(i=2; i<34; i++){
3075
        h->left_border[i]= src_y[15+i*  linesize];
3076
    }
3077

    
3078
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3079
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3080
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3081
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3082

    
3083
    if(!(s->flags&CODEC_FLAG_GRAY)){
3084
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3085
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3086
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3087
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3088
        for(i=2; i<18; i++){
3089
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3090
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3091
        }
3092
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3093
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3094
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3095
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3096
    }
3097
}
3098

    
3099
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){
3100
    MpegEncContext * const s = &h->s;
3101
    int temp8, i;
3102
    uint64_t temp64;
3103
    int deblock_left = (s->mb_x > 0);
3104
    int deblock_top  = (s->mb_y > 1);
3105

    
3106
    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);
3107

    
3108
    src_y  -= 2 *   linesize + 1;
3109
    src_cb -= 2 * uvlinesize + 1;
3110
    src_cr -= 2 * uvlinesize + 1;
3111

    
3112
#define XCHG(a,b,t,xchg)\
3113
t= a;\
3114
if(xchg)\
3115
    a= b;\
3116
b= t;
3117

    
3118
    if(deblock_left){
3119
        for(i = (!deblock_top)<<1; i<34; i++){
3120
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3121
        }
3122
    }
3123

    
3124
    if(deblock_top){
3125
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3126
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3127
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3128
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3129
        if(s->mb_x+1 < s->mb_width){
3130
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3131
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3132
        }
3133
    }
3134

    
3135
    if(!(s->flags&CODEC_FLAG_GRAY)){
3136
        if(deblock_left){
3137
            for(i = (!deblock_top) << 1; i<18; i++){
3138
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3139
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3140
            }
3141
        }
3142
        if(deblock_top){
3143
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3144
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3145
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3146
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3147
        }
3148
    }
3149
}
3150

    
3151
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
3152
    MpegEncContext * const s = &h->s;
3153
    const int mb_x= s->mb_x;
3154
    const int mb_y= s->mb_y;
3155
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3156
    const int mb_type= s->current_picture.mb_type[mb_xy];
3157
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3158
    int linesize, uvlinesize /*dct_offset*/;
3159
    int i;
3160
    int *block_offset = &h->block_offset[0];
3161
    const unsigned int bottom = mb_y & 1;
3162
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
3163
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3164
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3165

    
3166
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3167
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3168
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3169

    
3170
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
3171
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
3172

    
3173
    if (!simple && MB_FIELD) {
3174
        linesize   = h->mb_linesize   = s->linesize * 2;
3175
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3176
        block_offset = &h->block_offset[24];
3177
        if(mb_y&1){ //FIXME move out of this func?
3178
            dest_y -= s->linesize*15;
3179
            dest_cb-= s->uvlinesize*7;
3180
            dest_cr-= s->uvlinesize*7;
3181
        }
3182
        if(FRAME_MBAFF) {
3183
            int list;
3184
            for(list=0; list<h->list_count; list++){
3185
                if(!USES_LIST(mb_type, list))
3186
                    continue;
3187
                if(IS_16X16(mb_type)){
3188
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
3189
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3190
                }else{
3191
                    for(i=0; i<16; i+=4){
3192
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3193
                        int ref = h->ref_cache[list][scan8[i]];
3194
                        if(ref >= 0)
3195
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3196
                    }
3197
                }
3198
            }
3199
        }
3200
    } else {
3201
        linesize   = h->mb_linesize   = s->linesize;
3202
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3203
//        dct_offset = s->linesize * 16;
3204
    }
3205

    
3206
    if(transform_bypass){
3207
        idct_dc_add =
3208
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3209
    }else if(IS_8x8DCT(mb_type)){
3210
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3211
        idct_add = s->dsp.h264_idct8_add;
3212
    }else{
3213
        idct_dc_add = s->dsp.h264_idct_dc_add;
3214
        idct_add = s->dsp.h264_idct_add;
3215
    }
3216

    
3217
    if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3218
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3219
        int mbt_y = mb_y&~1;
3220
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
3221
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3222
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3223
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3224
    }
3225

    
3226
    if (!simple && IS_INTRA_PCM(mb_type)) {
3227
        unsigned int x, y;
3228

    
3229
        // The pixels are stored in h->mb array in the same order as levels,
3230
        // copy them in output in the correct order.
3231
        for(i=0; i<16; i++) {
3232
            for (y=0; y<4; y++) {
3233
                for (x=0; x<4; x++) {
3234
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3235
                }
3236
            }
3237
        }
3238
        for(i=16; i<16+4; i++) {
3239
            for (y=0; y<4; y++) {
3240
                for (x=0; x<4; x++) {
3241
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3242
                }
3243
            }
3244
        }
3245
        for(i=20; i<20+4; i++) {
3246
            for (y=0; y<4; y++) {
3247
                for (x=0; x<4; x++) {
3248
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3249
                }
3250
            }
3251
        }
3252
    } else {
3253
        if(IS_INTRA(mb_type)){
3254
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3255
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3256

    
3257
            if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3258
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3259
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3260
            }
3261

    
3262
            if(IS_INTRA4x4(mb_type)){
3263
                if(simple || !s->encoding){
3264
                    if(IS_8x8DCT(mb_type)){
3265
                        for(i=0; i<16; i+=4){
3266
                            uint8_t * const ptr= dest_y + block_offset[i];
3267
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3268
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3269
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3270
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
3271
                            if(nnz){
3272
                                if(nnz == 1 && h->mb[i*16])
3273
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3274
                                else
3275
                                    idct_add(ptr, h->mb + i*16, linesize);
3276
                            }
3277
                        }
3278
                    }else
3279
                    for(i=0; i<16; i++){
3280
                        uint8_t * const ptr= dest_y + block_offset[i];
3281
                        uint8_t *topright;
3282
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3283
                        int nnz, tr;
3284

    
3285
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3286
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3287
                            assert(mb_y || linesize <= block_offset[i]);
3288
                            if(!topright_avail){
3289
                                tr= ptr[3 - linesize]*0x01010101;
3290
                                topright= (uint8_t*) &tr;
3291
                            }else
3292
                                topright= ptr + 4 - linesize;
3293
                        }else
3294
                            topright= NULL;
3295

    
3296
                        h->pred4x4[ dir ](ptr, topright, linesize);
3297
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3298
                        if(nnz){
3299
                            if(is_h264){
3300
                                if(nnz == 1 && h->mb[i*16])
3301
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3302
                                else
3303
                                    idct_add(ptr, h->mb + i*16, linesize);
3304
                            }else
3305
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3306
                        }
3307
                    }
3308
                }
3309
            }else{
3310
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3311
                if(is_h264){
3312
                    if(!transform_bypass)
3313
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3314
                }else
3315
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3316
            }
3317
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3318
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3319
        }else if(is_h264){
3320
            hl_motion(h, dest_y, dest_cb, dest_cr,
3321
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
3322
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
3323
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3324
        }
3325

    
3326

    
3327
        if(!IS_INTRA4x4(mb_type)){
3328
            if(is_h264){
3329
                if(IS_INTRA16x16(mb_type)){
3330
                    for(i=0; i<16; i++){
3331
                        if(h->non_zero_count_cache[ scan8[i] ])
3332
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3333
                        else if(h->mb[i*16])
3334
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3335
                    }
3336
                }else{
3337
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3338
                    for(i=0; i<16; i+=di){
3339
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3340
                        if(nnz){
3341
                            if(nnz==1 && h->mb[i*16])
3342
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3343
                            else
3344
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3345
                        }
3346
                    }
3347
                }
3348
            }else{
3349
                for(i=0; i<16; i++){
3350
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3351
                        uint8_t * const ptr= dest_y + block_offset[i];
3352
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3353
                    }
3354
                }
3355
            }
3356
        }
3357

    
3358
        if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3359
            uint8_t *dest[2] = {dest_cb, dest_cr};
3360
            if(transform_bypass){
3361
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3362
            }else{
3363
                idct_add = s->dsp.h264_idct_add;
3364
                idct_dc_add = s->dsp.h264_idct_dc_add;
3365
                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]);
3366
                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]);
3367
            }
3368
            if(is_h264){
3369
                for(i=16; i<16+8; i++){
3370
                    if(h->non_zero_count_cache[ scan8[i] ])
3371
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3372
                    else if(h->mb[i*16])
3373
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3374
                }
3375
            }else{
3376
                for(i=16; i<16+8; i++){
3377
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3378
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3379
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3380
                    }
3381
                }
3382
            }
3383
        }
3384
    }
3385
    if(h->deblocking_filter) {
3386
        if (!simple && FRAME_MBAFF) {
3387
            //FIXME try deblocking one mb at a time?
3388
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3389
            const int mb_y = s->mb_y - 1;
3390
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3391
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3392
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3393
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3394
            if (!bottom) return;
3395
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3396
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3397
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3398

    
3399
            if(IS_INTRA(mb_type_top | mb_type_bottom))
3400
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3401

    
3402
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3403
            // deblock a pair
3404
            // top
3405
            s->mb_y--;
3406
            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);
3407
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3408
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
3409
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3410
            // bottom
3411
            s->mb_y++;
3412
            tprintf(h->s.avctx, "call mbaff filter_mb\n");
3413
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3414
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3415
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3416
        } else {
3417
            tprintf(h->s.avctx, "call filter_mb\n");
3418
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3419
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3420
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3421
        }
3422
    }
3423
}
3424

    
3425
/**
3426
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
3427
 */
3428
static void hl_decode_mb_simple(H264Context *h){
3429
    hl_decode_mb_internal(h, 1);
3430
}
3431

    
3432
/**
3433
 * Process a macroblock; this handles edge cases, such as interlacing.
3434
 */
3435
static void av_noinline hl_decode_mb_complex(H264Context *h){
3436
    hl_decode_mb_internal(h, 0);
3437
}
3438

    
3439
static void hl_decode_mb(H264Context *h){
3440
    MpegEncContext * const s = &h->s;
3441
    const int mb_x= s->mb_x;
3442
    const int mb_y= s->mb_y;
3443
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3444
    const int mb_type= s->current_picture.mb_type[mb_xy];
3445
    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;
3446

    
3447
    if(!s->decode)
3448
        return;
3449

    
3450
    if (is_complex)
3451
        hl_decode_mb_complex(h);
3452
    else hl_decode_mb_simple(h);
3453
}
3454

    
3455
/**
3456
 * fills the default_ref_list.
3457
 */
3458
static int fill_default_ref_list(H264Context *h){
3459
    MpegEncContext * const s = &h->s;
3460
    int i;
3461
    int smallest_poc_greater_than_current = -1;
3462
    Picture sorted_short_ref[32];
3463

    
3464
    if(h->slice_type==B_TYPE){
3465
        int out_i;
3466
        int limit= INT_MIN;
3467

    
3468
        /* sort frame according to poc in B slice */
3469
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3470
            int best_i=INT_MIN;
3471
            int best_poc=INT_MAX;
3472

    
3473
            for(i=0; i<h->short_ref_count; i++){
3474
                const int poc= h->short_ref[i]->poc;
3475
                if(poc > limit && poc < best_poc){
3476
                    best_poc= poc;
3477
                    best_i= i;
3478
                }
3479
            }
3480

    
3481
            assert(best_i != INT_MIN);
3482

    
3483
            limit= best_poc;
3484
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3485
            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);
3486
            if (-1 == smallest_poc_greater_than_current) {
3487
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3488
                    smallest_poc_greater_than_current = out_i;
3489
                }
3490
            }
3491
        }
3492
    }
3493

    
3494
    if(s->picture_structure == PICT_FRAME){
3495
        if(h->slice_type==B_TYPE){
3496
            int list;
3497
            tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3498

    
3499
            // find the largest poc
3500
            for(list=0; list<2; list++){
3501
                int index = 0;
3502
                int j= -99;
3503
                int step= list ? -1 : 1;
3504

    
3505
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3506
                    while(j<0 || j>= h->short_ref_count){
3507
                        if(j != -99 && step == (list ? -1 : 1))
3508
                            return -1;
3509
                        step = -step;
3510
                        j= smallest_poc_greater_than_current + (step>>1);
3511
                    }
3512
                    if(sorted_short_ref[j].reference != 3) continue;
3513
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3514
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3515
                }
3516

    
3517
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3518
                    if(h->long_ref[i] == NULL) continue;
3519
                    if(h->long_ref[i]->reference != 3) continue;
3520

    
3521
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3522
                    h->default_ref_list[ list ][index++].pic_id= i;;
3523
                }
3524

    
3525
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3526
                    // swap the two first elements of L1 when
3527
                    // L0 and L1 are identical
3528
                    Picture temp= h->default_ref_list[1][0];
3529
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3530
                    h->default_ref_list[1][1] = temp;
3531
                }
3532

    
3533
                if(index < h->ref_count[ list ])
3534
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3535
            }
3536
        }else{
3537
            int index=0;
3538
            for(i=0; i<h->short_ref_count; i++){
3539
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3540
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3541
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3542
            }
3543
            for(i = 0; i < 16; i++){
3544
                if(h->long_ref[i] == NULL) continue;
3545
                if(h->long_ref[i]->reference != 3) continue;
3546
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3547
                h->default_ref_list[0][index++].pic_id= i;;
3548
            }
3549
            if(index < h->ref_count[0])
3550
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3551
        }
3552
    }else{ //FIELD
3553
        if(h->slice_type==B_TYPE){
3554
        }else{
3555
            //FIXME second field balh
3556
        }
3557
    }
3558
#ifdef TRACE
3559
    for (i=0; i<h->ref_count[0]; i++) {
3560
        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]);
3561
    }
3562
    if(h->slice_type==B_TYPE){
3563
        for (i=0; i<h->ref_count[1]; i++) {
3564
            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]);
3565
        }
3566
    }
3567
#endif
3568
    return 0;
3569
}
3570

    
3571
static void print_short_term(H264Context *h);
3572
static void print_long_term(H264Context *h);
3573

    
3574
static int decode_ref_pic_list_reordering(H264Context *h){
3575
    MpegEncContext * const s = &h->s;
3576
    int list, index;
3577

    
3578
    print_short_term(h);
3579
    print_long_term(h);
3580
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3581

    
3582
    for(list=0; list<h->list_count; list++){
3583
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3584

    
3585
        if(get_bits1(&s->gb)){
3586
            int pred= h->curr_pic_num;
3587

    
3588
            for(index=0; ; index++){
3589
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3590
                unsigned int pic_id;
3591
                int i;
3592
                Picture *ref = NULL;
3593

    
3594
                if(reordering_of_pic_nums_idc==3)
3595
                    break;
3596

    
3597
                if(index >= h->ref_count[list]){
3598
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3599
                    return -1;
3600
                }
3601

    
3602
                if(reordering_of_pic_nums_idc<3){
3603
                    if(reordering_of_pic_nums_idc<2){
3604
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3605

    
3606
                        if(abs_diff_pic_num >= h->max_pic_num){
3607
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3608
                            return -1;
3609
                        }
3610

    
3611
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3612
                        else                                pred+= abs_diff_pic_num;
3613
                        pred &= h->max_pic_num - 1;
3614

    
3615
                        for(i= h->short_ref_count-1; i>=0; i--){
3616
                            ref = h->short_ref[i];
3617
                            assert(ref->reference == 3);
3618
                            assert(!ref->long_ref);
3619
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3620
                                break;
3621
                        }
3622
                        if(i>=0)
3623
                            ref->pic_id= ref->frame_num;
3624
                    }else{
3625
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3626
                        if(pic_id>31){
3627
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
3628
                            return -1;
3629
                        }
3630
                        ref = h->long_ref[pic_id];
3631
                        if(ref){
3632
                            ref->pic_id= pic_id;
3633
                            assert(ref->reference == 3);
3634
                            assert(ref->long_ref);
3635
                            i=0;
3636
                        }else{
3637
                            i=-1;
3638
                        }
3639
                    }
3640

    
3641
                    if (i < 0) {
3642
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3643
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3644
                    } else {
3645
                        for(i=index; i+1<h->ref_count[list]; i++){
3646
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3647
                                break;
3648
                        }
3649
                        for(; i > index; i--){
3650
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3651
                        }
3652
                        h->ref_list[list][index]= *ref;
3653
                    }
3654
                }else{
3655
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3656
                    return -1;
3657
                }
3658
            }
3659
        }
3660
    }
3661
    for(list=0; list<h->list_count; list++){
3662
        for(index= 0; index < h->ref_count[list]; index++){
3663
            if(!h->ref_list[list][index].data[0])
3664
                h->ref_list[list][index]= s->current_picture;
3665
        }
3666
    }
3667

    
3668
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3669
        direct_dist_scale_factor(h);
3670
    direct_ref_list_init(h);
3671
    return 0;
3672
}
3673

    
3674
static void fill_mbaff_ref_list(H264Context *h){
3675
    int list, i, j;
3676
    for(list=0; list<2; list++){ //FIXME try list_count
3677
        for(i=0; i<h->ref_count[list]; i++){
3678
            Picture *frame = &h->ref_list[list][i];
3679
            Picture *field = &h->ref_list[list][16+2*i];
3680
            field[0] = *frame;
3681
            for(j=0; j<3; j++)
3682
                field[0].linesize[j] <<= 1;
3683
            field[1] = field[0];
3684
            for(j=0; j<3; j++)
3685
                field[1].data[j] += frame->linesize[j];
3686

    
3687
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
3688
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
3689
            for(j=0; j<2; j++){
3690
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
3691
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
3692
            }
3693
        }
3694
    }
3695
    for(j=0; j<h->ref_count[1]; j++){
3696
        for(i=0; i<h->ref_count[0]; i++)
3697
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
3698
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
3699
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
3700
    }
3701
}
3702

    
3703
static int pred_weight_table(H264Context *h){
3704
    MpegEncContext * const s = &h->s;
3705
    int list, i;
3706
    int luma_def, chroma_def;
3707

    
3708
    h->use_weight= 0;
3709
    h->use_weight_chroma= 0;
3710
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3711
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3712
    luma_def = 1<<h->luma_log2_weight_denom;
3713
    chroma_def = 1<<h->chroma_log2_weight_denom;
3714

    
3715
    for(list=0; list<2; list++){
3716
        for(i=0; i<h->ref_count[list]; i++){
3717
            int luma_weight_flag, chroma_weight_flag;
3718

    
3719
            luma_weight_flag= get_bits1(&s->gb);
3720
            if(luma_weight_flag){
3721
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3722
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3723
                if(   h->luma_weight[list][i] != luma_def
3724
                   || h->luma_offset[list][i] != 0)
3725
                    h->use_weight= 1;
3726
            }else{
3727
                h->luma_weight[list][i]= luma_def;
3728
                h->luma_offset[list][i]= 0;
3729
            }
3730

    
3731
            chroma_weight_flag= get_bits1(&s->gb);
3732
            if(chroma_weight_flag){
3733
                int j;
3734
                for(j=0; j<2; j++){
3735
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3736
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3737
                    if(   h->chroma_weight[list][i][j] != chroma_def
3738
                       || h->chroma_offset[list][i][j] != 0)
3739
                        h->use_weight_chroma= 1;
3740
                }
3741
            }else{
3742
                int j;
3743
                for(j=0; j<2; j++){
3744
                    h->chroma_weight[list][i][j]= chroma_def;
3745
                    h->chroma_offset[list][i][j]= 0;
3746
                }
3747
            }
3748
        }
3749
        if(h->slice_type != B_TYPE) break;
3750
    }
3751
    h->use_weight= h->use_weight || h->use_weight_chroma;
3752
    return 0;
3753
}
3754

    
3755
static void implicit_weight_table(H264Context *h){
3756
    MpegEncContext * const s = &h->s;
3757
    int ref0, ref1;
3758
    int cur_poc = s->current_picture_ptr->poc;
3759

    
3760
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3761
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3762
        h->use_weight= 0;
3763
        h->use_weight_chroma= 0;
3764
        return;
3765
    }
3766

    
3767
    h->use_weight= 2;
3768
    h->use_weight_chroma= 2;
3769
    h->luma_log2_weight_denom= 5;
3770
    h->chroma_log2_weight_denom= 5;
3771

    
3772
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3773
        int poc0 = h->ref_list[0][ref0].poc;
3774
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3775
            int poc1 = h->ref_list[1][ref1].poc;
3776
            int td = av_clip(poc1 - poc0, -128, 127);
3777
            if(td){
3778
                int tb = av_clip(cur_poc - poc0, -128, 127);
3779
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3780
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3781
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3782
                    h->implicit_weight[ref0][ref1] = 32;
3783
                else
3784
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3785
            }else
3786
                h->implicit_weight[ref0][ref1] = 32;
3787
        }
3788
    }
3789
}
3790

    
3791
static inline void unreference_pic(H264Context *h, Picture *pic){
3792
    int i;
3793
    pic->reference=0;
3794
    if(pic == h->delayed_output_pic)
3795
        pic->reference=1;
3796
    else{
3797
        for(i = 0; h->delayed_pic[i]; i++)
3798
            if(pic == h->delayed_pic[i]){
3799
                pic->reference=1;
3800
                break;
3801
            }
3802
    }
3803
}
3804

    
3805
/**
3806
 * instantaneous decoder refresh.
3807
 */
3808
static void idr(H264Context *h){
3809
    int i;
3810

    
3811
    for(i=0; i<16; i++){
3812
        if (h->long_ref[i] != NULL) {
3813
            unreference_pic(h, h->long_ref[i]);
3814
            h->long_ref[i]= NULL;
3815
        }
3816
    }
3817
    h->long_ref_count=0;
3818

    
3819
    for(i=0; i<h->short_ref_count; i++){
3820
        unreference_pic(h, h->short_ref[i]);
3821
        h->short_ref[i]= NULL;
3822
    }
3823
    h->short_ref_count=0;
3824
}
3825

    
3826
/* forget old pics after a seek */
3827
static void flush_dpb(AVCodecContext *avctx){
3828
    H264Context *h= avctx->priv_data;
3829
    int i;
3830
    for(i=0; i<16; i++) {
3831
        if(h->delayed_pic[i])
3832
            h->delayed_pic[i]->reference= 0;
3833
        h->delayed_pic[i]= NULL;
3834
    }
3835
    if(h->delayed_output_pic)
3836
        h->delayed_output_pic->reference= 0;
3837
    h->delayed_output_pic= NULL;
3838
    idr(h);
3839
    if(h->s.current_picture_ptr)
3840
        h->s.current_picture_ptr->reference= 0;
3841
}
3842

    
3843
/**
3844
 *
3845
 * @return the removed picture or NULL if an error occurs
3846
 */
3847
static Picture * remove_short(H264Context *h, int frame_num){
3848
    MpegEncContext * const s = &h->s;
3849
    int i;
3850

    
3851
    if(s->avctx->debug&FF_DEBUG_MMCO)
3852
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3853

    
3854
    for(i=0; i<h->short_ref_count; i++){
3855
        Picture *pic= h->short_ref[i];
3856
        if(s->avctx->debug&FF_DEBUG_MMCO)
3857
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3858
        if(pic->frame_num == frame_num){
3859
            h->short_ref[i]= NULL;
3860
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3861
            h->short_ref_count--;
3862
            return pic;
3863
        }
3864
    }
3865
    return NULL;
3866
}
3867

    
3868
/**
3869
 *
3870
 * @return the removed picture or NULL if an error occurs
3871
 */
3872
static Picture * remove_long(H264Context *h, int i){
3873
    Picture *pic;
3874

    
3875
    pic= h->long_ref[i];
3876
    h->long_ref[i]= NULL;
3877
    if(pic) h->long_ref_count--;
3878

    
3879
    return pic;
3880
}
3881

    
3882
/**
3883
 * print short term list
3884
 */
3885
static void print_short_term(H264Context *h) {
3886
    uint32_t i;
3887
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3888
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3889
        for(i=0; i<h->short_ref_count; i++){
3890
            Picture *pic= h->short_ref[i];
3891
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3892
        }
3893
    }
3894
}
3895

    
3896
/**
3897
 * print long term list
3898
 */
3899
static void print_long_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, "long term list:\n");
3903
        for(i = 0; i < 16; i++){
3904
            Picture *pic= h->long_ref[i];
3905
            if (pic) {
3906
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3907
            }
3908
        }
3909
    }
3910
}
3911

    
3912
/**
3913
 * Executes the reference picture marking (memory management control operations).
3914
 */
3915
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3916
    MpegEncContext * const s = &h->s;
3917
    int i, j;
3918
    int current_is_long=0;
3919
    Picture *pic;
3920

    
3921
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3922
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3923

    
3924
    for(i=0; i<mmco_count; i++){
3925
        if(s->avctx->debug&FF_DEBUG_MMCO)
3926
            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);
3927

    
3928
        switch(mmco[i].opcode){
3929
        case MMCO_SHORT2UNUSED:
3930
            pic= remove_short(h, mmco[i].short_frame_num);
3931
            if(pic)
3932
                unreference_pic(h, pic);
3933
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3934
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
3935
            break;
3936
        case MMCO_SHORT2LONG:
3937
            pic= remove_long(h, mmco[i].long_index);
3938
            if(pic) unreference_pic(h, pic);
3939

    
3940
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3941
            if (h->long_ref[ mmco[i].long_index ]){
3942
                h->long_ref[ mmco[i].long_index ]->long_ref=1;
3943
                h->long_ref_count++;
3944
            }
3945
            break;
3946
        case MMCO_LONG2UNUSED:
3947
            pic= remove_long(h, mmco[i].long_index);
3948
            if(pic)
3949
                unreference_pic(h, pic);
3950
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3951
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
3952
            break;
3953
        case MMCO_LONG:
3954
            pic= remove_long(h, mmco[i].long_index);
3955
            if(pic) unreference_pic(h, pic);
3956

    
3957
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3958
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3959
            h->long_ref_count++;
3960

    
3961
            current_is_long=1;
3962
            break;
3963
        case MMCO_SET_MAX_LONG:
3964
            assert(mmco[i].long_index <= 16);
3965
            // just remove the long term which index is greater than new max
3966
            for(j = mmco[i].long_index; j<16; j++){
3967
                pic = remove_long(h, j);
3968
                if (pic) unreference_pic(h, pic);
3969
            }
3970
            break;
3971
        case MMCO_RESET:
3972
            while(h->short_ref_count){
3973
                pic= remove_short(h, h->short_ref[0]->frame_num);
3974
                if(pic) unreference_pic(h, pic);
3975
            }
3976
            for(j = 0; j < 16; j++) {
3977
                pic= remove_long(h, j);
3978
                if(pic) unreference_pic(h, pic);
3979
            }
3980
            break;
3981
        default: assert(0);
3982
        }
3983
    }
3984

    
3985
    if(!current_is_long){
3986
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3987
        if(pic){
3988
            unreference_pic(h, pic);
3989
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3990
        }
3991

    
3992
        if(h->short_ref_count)
3993
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3994

    
3995
        h->short_ref[0]= s->current_picture_ptr;
3996
        h->short_ref[0]->long_ref=0;
3997
        h->short_ref_count++;
3998
    }
3999

    
4000
    print_short_term(h);
4001
    print_long_term(h);
4002
    return 0;
4003
}
4004

    
4005
static int decode_ref_pic_marking(H264Context *h){
4006
    MpegEncContext * const s = &h->s;
4007
    int i;
4008

    
4009
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
4010
        s->broken_link= get_bits1(&s->gb) -1;
4011
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
4012
        if(h->mmco[0].long_index == -1)
4013
            h->mmco_index= 0;
4014
        else{
4015
            h->mmco[0].opcode= MMCO_LONG;
4016
            h->mmco_index= 1;
4017
        }
4018
    }else{
4019
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
4020
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
4021
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
4022

    
4023
                h->mmco[i].opcode= opcode;
4024
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
4025
                    h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
4026
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
4027
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
4028
                        return -1;
4029
                    }*/
4030
                }
4031
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
4032
                    unsigned int long_index= get_ue_golomb(&s->gb);
4033
                    if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ long_index >= 16){
4034
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
4035
                        return -1;
4036
                    }
4037
                    h->mmco[i].long_index= long_index;
4038
                }
4039

    
4040
                if(opcode > (unsigned)MMCO_LONG){
4041
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4042
                    return -1;
4043
                }
4044
                if(opcode == MMCO_END)
4045
                    break;
4046
            }
4047
            h->mmco_index= i;
4048
        }else{
4049
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4050

    
4051
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4052
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4053
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4054
                h->mmco_index= 1;
4055
            }else
4056
                h->mmco_index= 0;
4057
        }
4058
    }
4059

    
4060
    return 0;
4061
}
4062

    
4063
static int init_poc(H264Context *h){
4064
    MpegEncContext * const s = &h->s;
4065
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4066
    int field_poc[2];
4067

    
4068
    if(h->nal_unit_type == NAL_IDR_SLICE){
4069
        h->frame_num_offset= 0;
4070
    }else{
4071
        if(h->frame_num < h->prev_frame_num)
4072
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4073
        else
4074
            h->frame_num_offset= h->prev_frame_num_offset;
4075
    }
4076

    
4077
    if(h->sps.poc_type==0){
4078
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4079

    
4080
        if(h->nal_unit_type == NAL_IDR_SLICE){
4081
             h->prev_poc_msb=
4082
             h->prev_poc_lsb= 0;
4083
        }
4084

    
4085
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4086
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4087
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4088
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4089
        else
4090
            h->poc_msb = h->prev_poc_msb;
4091
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4092
        field_poc[0] =
4093
        field_poc[1] = h->poc_msb + h->poc_lsb;
4094
        if(s->picture_structure == PICT_FRAME)
4095
            field_poc[1] += h->delta_poc_bottom;
4096
    }else if(h->sps.poc_type==1){
4097
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4098
        int i;
4099

    
4100
        if(h->sps.poc_cycle_length != 0)
4101
            abs_frame_num = h->frame_num_offset + h->frame_num;
4102
        else
4103
            abs_frame_num = 0;
4104

    
4105
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4106
            abs_frame_num--;
4107

    
4108
        expected_delta_per_poc_cycle = 0;
4109
        for(i=0; i < h->sps.poc_cycle_length; i++)
4110
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4111

    
4112
        if(abs_frame_num > 0){
4113
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4114
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4115

    
4116
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4117
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4118
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4119
        } else
4120
            expectedpoc = 0;
4121

    
4122
        if(h->nal_ref_idc == 0)
4123
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4124

    
4125
        field_poc[0] = expectedpoc + h->delta_poc[0];
4126
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4127

    
4128
        if(s->picture_structure == PICT_FRAME)
4129
            field_poc[1] += h->delta_poc[1];
4130
    }else{
4131
        int poc;
4132
        if(h->nal_unit_type == NAL_IDR_SLICE){
4133
            poc= 0;
4134
        }else{
4135
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4136
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4137
        }
4138
        field_poc[0]= poc;
4139
        field_poc[1]= poc;
4140
    }
4141

    
4142
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4143
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4144
    if(s->picture_structure != PICT_TOP_FIELD)
4145
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4146
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4147
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4148

    
4149
    return 0;
4150
}
4151

    
4152
/**
4153
 * decodes a slice header.
4154
 * this will allso call MPV_common_init() and frame_start() as needed
4155
 */
4156
static int decode_slice_header(H264Context *h){
4157
    MpegEncContext * const s = &h->s;
4158
    unsigned int first_mb_in_slice;
4159
    unsigned int pps_id;
4160
    int num_ref_idx_active_override_flag;
4161
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4162
    unsigned int slice_type, tmp;
4163
    int default_ref_list_done = 0;
4164

    
4165
    s->current_picture.reference= h->nal_ref_idc != 0;
4166
    s->dropable= h->nal_ref_idc == 0;
4167

    
4168
    first_mb_in_slice= get_ue_golomb(&s->gb);
4169

    
4170
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
4171
        h->slice_num = 0;
4172
        s->current_picture_ptr= NULL;
4173
    }
4174

    
4175
    slice_type= get_ue_golomb(&s->gb);
4176
    if(slice_type > 9){
4177
        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);
4178
        return -1;
4179
    }
4180
    if(slice_type > 4){
4181
        slice_type -= 5;
4182
        h->slice_type_fixed=1;
4183
    }else
4184
        h->slice_type_fixed=0;
4185

    
4186
    slice_type= slice_type_map[ slice_type ];
4187
    if (slice_type == I_TYPE
4188
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4189
        default_ref_list_done = 1;
4190
    }
4191
    h->slice_type= slice_type;
4192

    
4193
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4194

    
4195
    pps_id= get_ue_golomb(&s->gb);
4196
    if(pps_id>=MAX_PPS_COUNT){
4197
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4198
        return -1;
4199
    }
4200
    h->pps= h->pps_buffer[pps_id];
4201
    if(h->pps.slice_group_count == 0){
4202
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4203
        return -1;
4204
    }
4205

    
4206
    h->sps= h->sps_buffer[ h->pps.sps_id ];
4207
    if(h->sps.log2_max_frame_num == 0){
4208
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4209
        return -1;
4210
    }
4211

    
4212
    if(h->dequant_coeff_pps != pps_id){
4213
        h->dequant_coeff_pps = pps_id;
4214
        init_dequant_tables(h);
4215
    }
4216

    
4217
    s->mb_width= h->sps.mb_width;
4218
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
4219

    
4220
    h->b_stride=  s->mb_width*4;
4221
    h->b8_stride= s->mb_width*2;
4222

    
4223
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
4224
    if(h->sps.frame_mbs_only_flag)
4225
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
4226
    else
4227
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
4228

    
4229
    if (s->context_initialized
4230
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
4231
        free_tables(h);
4232
        MPV_common_end(s);
4233
    }
4234
    if (!s->context_initialized) {
4235
        if (MPV_common_init(s) < 0)
4236
            return -1;
4237

    
4238
        if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
4239
            memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
4240
            memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
4241
        }else{
4242
            int i;
4243
            for(i=0; i<16; i++){
4244
#define T(x) (x>>2) | ((x<<2) & 0xF)
4245
                h->zigzag_scan[i] = T(zigzag_scan[i]);
4246
                h-> field_scan[i] = T( field_scan[i]);
4247
#undef T
4248
            }
4249
        }
4250
        if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
4251
            memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
4252
            memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
4253
            memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
4254
            memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
4255
        }else{
4256
            int i;
4257
            for(i=0; i<64; i++){
4258
#define T(x) (x>>3) | ((x&7)<<3)
4259
                h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
4260
                h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
4261
                h->field_scan8x8[i]        = T(field_scan8x8[i]);
4262
                h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
4263
#undef T
4264
            }
4265
        }
4266
        if(h->sps.transform_bypass){ //FIXME same ugly
4267
            h->zigzag_scan_q0          = zigzag_scan;
4268
            h->zigzag_scan8x8_q0       = zigzag_scan8x8;
4269
            h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
4270
            h->field_scan_q0           = field_scan;
4271
            h->field_scan8x8_q0        = field_scan8x8;
4272
            h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
4273
        }else{
4274
            h->zigzag_scan_q0          = h->zigzag_scan;
4275
            h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
4276
            h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
4277
            h->field_scan_q0           = h->field_scan;