Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 26b4fe82

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 "common.h"
30
#include "dsputil.h"
31
#include "avcodec.h"
32
#include "mpegvideo.h"
33
#include "h264.h"
34
#include "h264data.h"
35
#include "h264_parser.h"
36
#include "golomb.h"
37

    
38
#include "cabac.h"
39

    
40
//#undef NDEBUG
41
#include <assert.h>
42

    
43
static VLC coeff_token_vlc[4];
44
static VLC chroma_dc_coeff_token_vlc;
45

    
46
static VLC total_zeros_vlc[15];
47
static VLC chroma_dc_total_zeros_vlc[3];
48

    
49
static VLC run_vlc[6];
50
static VLC run7_vlc;
51

    
52
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
53
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
54
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);
55
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);
56

    
57
static av_always_inline uint32_t pack16to32(int a, int b){
58
#ifdef WORDS_BIGENDIAN
59
   return (b&0xFFFF) + (a<<16);
60
#else
61
   return (a&0xFFFF) + (b<<16);
62
#endif
63
}
64

    
65
const uint8_t ff_rem6[52]={
66
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,
67
};
68

    
69
const uint8_t ff_div6[52]={
70
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,
71
};
72

    
73

    
74
/**
75
 * fill a rectangle.
76
 * @param h height of the rectangle, should be a constant
77
 * @param w width of the rectangle, should be a constant
78
 * @param size the size of val (1 or 4), should be a constant
79
 */
80
static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
81
    uint8_t *p= (uint8_t*)vp;
82
    assert(size==1 || size==4);
83
    assert(w<=4);
84

    
85
    w      *= size;
86
    stride *= size;
87

    
88
    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
89
    assert((stride&(w-1))==0);
90
    if(w==2){
91
        const uint16_t v= size==4 ? val : val*0x0101;
92
        *(uint16_t*)(p + 0*stride)= v;
93
        if(h==1) return;
94
        *(uint16_t*)(p + 1*stride)= v;
95
        if(h==2) return;
96
        *(uint16_t*)(p + 2*stride)=
97
        *(uint16_t*)(p + 3*stride)= v;
98
    }else if(w==4){
99
        const uint32_t v= size==4 ? val : val*0x01010101;
100
        *(uint32_t*)(p + 0*stride)= v;
101
        if(h==1) return;
102
        *(uint32_t*)(p + 1*stride)= v;
103
        if(h==2) return;
104
        *(uint32_t*)(p + 2*stride)=
105
        *(uint32_t*)(p + 3*stride)= v;
106
    }else if(w==8){
107
    //gcc can't optimize 64bit math on x86_32
108
#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
109
        const uint64_t v= val*0x0100000001ULL;
110
        *(uint64_t*)(p + 0*stride)= v;
111
        if(h==1) return;
112
        *(uint64_t*)(p + 1*stride)= v;
113
        if(h==2) return;
114
        *(uint64_t*)(p + 2*stride)=
115
        *(uint64_t*)(p + 3*stride)= v;
116
    }else if(w==16){
117
        const uint64_t v= val*0x0100000001ULL;
118
        *(uint64_t*)(p + 0+0*stride)=
119
        *(uint64_t*)(p + 8+0*stride)=
120
        *(uint64_t*)(p + 0+1*stride)=
121
        *(uint64_t*)(p + 8+1*stride)= v;
122
        if(h==2) return;
123
        *(uint64_t*)(p + 0+2*stride)=
124
        *(uint64_t*)(p + 8+2*stride)=
125
        *(uint64_t*)(p + 0+3*stride)=
126
        *(uint64_t*)(p + 8+3*stride)= v;
127
#else
128
        *(uint32_t*)(p + 0+0*stride)=
129
        *(uint32_t*)(p + 4+0*stride)= val;
130
        if(h==1) return;
131
        *(uint32_t*)(p + 0+1*stride)=
132
        *(uint32_t*)(p + 4+1*stride)= val;
133
        if(h==2) return;
134
        *(uint32_t*)(p + 0+2*stride)=
135
        *(uint32_t*)(p + 4+2*stride)=
136
        *(uint32_t*)(p + 0+3*stride)=
137
        *(uint32_t*)(p + 4+3*stride)= val;
138
    }else if(w==16){
139
        *(uint32_t*)(p + 0+0*stride)=
140
        *(uint32_t*)(p + 4+0*stride)=
141
        *(uint32_t*)(p + 8+0*stride)=
142
        *(uint32_t*)(p +12+0*stride)=
143
        *(uint32_t*)(p + 0+1*stride)=
144
        *(uint32_t*)(p + 4+1*stride)=
145
        *(uint32_t*)(p + 8+1*stride)=
146
        *(uint32_t*)(p +12+1*stride)= val;
147
        if(h==2) return;
148
        *(uint32_t*)(p + 0+2*stride)=
149
        *(uint32_t*)(p + 4+2*stride)=
150
        *(uint32_t*)(p + 8+2*stride)=
151
        *(uint32_t*)(p +12+2*stride)=
152
        *(uint32_t*)(p + 0+3*stride)=
153
        *(uint32_t*)(p + 4+3*stride)=
154
        *(uint32_t*)(p + 8+3*stride)=
155
        *(uint32_t*)(p +12+3*stride)= val;
156
#endif
157
    }else
158
        assert(0);
159
    assert(h==4);
160
}
161

    
162
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
163
    MpegEncContext * const s = &h->s;
164
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
165
    int topleft_xy, top_xy, topright_xy, left_xy[2];
166
    int topleft_type, top_type, topright_type, left_type[2];
167
    int left_block[8];
168
    int i;
169

    
170
    //FIXME deblocking could skip the intra and nnz parts.
171
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
172
        return;
173

    
174
    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
175

    
176
    top_xy     = mb_xy  - s->mb_stride;
177
    topleft_xy = top_xy - 1;
178
    topright_xy= top_xy + 1;
179
    left_xy[1] = left_xy[0] = mb_xy-1;
180
    left_block[0]= 0;
181
    left_block[1]= 1;
182
    left_block[2]= 2;
183
    left_block[3]= 3;
184
    left_block[4]= 7;
185
    left_block[5]= 10;
186
    left_block[6]= 8;
187
    left_block[7]= 11;
188
    if(FRAME_MBAFF){
189
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
190
        const int top_pair_xy      = pair_xy     - s->mb_stride;
191
        const int topleft_pair_xy  = top_pair_xy - 1;
192
        const int topright_pair_xy = top_pair_xy + 1;
193
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
194
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
195
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
196
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
197
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
198
        const int bottom = (s->mb_y & 1);
199
        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);
200
        if (bottom
201
                ? !curr_mb_frame_flag // bottom macroblock
202
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
203
                ) {
204
            top_xy -= s->mb_stride;
205
        }
206
        if (bottom
207
                ? !curr_mb_frame_flag // bottom macroblock
208
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
209
                ) {
210
            topleft_xy -= s->mb_stride;
211
        }
212
        if (bottom
213
                ? !curr_mb_frame_flag // bottom macroblock
214
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
215
                ) {
216
            topright_xy -= s->mb_stride;
217
        }
218
        if (left_mb_frame_flag != curr_mb_frame_flag) {
219
            left_xy[1] = left_xy[0] = pair_xy - 1;
220
            if (curr_mb_frame_flag) {
221
                if (bottom) {
222
                    left_block[0]= 2;
223
                    left_block[1]= 2;
224
                    left_block[2]= 3;
225
                    left_block[3]= 3;
226
                    left_block[4]= 8;
227
                    left_block[5]= 11;
228
                    left_block[6]= 8;
229
                    left_block[7]= 11;
230
                } else {
231
                    left_block[0]= 0;
232
                    left_block[1]= 0;
233
                    left_block[2]= 1;
234
                    left_block[3]= 1;
235
                    left_block[4]= 7;
236
                    left_block[5]= 10;
237
                    left_block[6]= 7;
238
                    left_block[7]= 10;
239
                }
240
            } else {
241
                left_xy[1] += s->mb_stride;
242
                //left_block[0]= 0;
243
                left_block[1]= 2;
244
                left_block[2]= 0;
245
                left_block[3]= 2;
246
                //left_block[4]= 7;
247
                left_block[5]= 10;
248
                left_block[6]= 7;
249
                left_block[7]= 10;
250
            }
251
        }
252
    }
253

    
254
    h->top_mb_xy = top_xy;
255
    h->left_mb_xy[0] = left_xy[0];
256
    h->left_mb_xy[1] = left_xy[1];
257
    if(for_deblock){
258
        topleft_type = 0;
259
        topright_type = 0;
260
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
261
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
262
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
263

    
264
        if(FRAME_MBAFF && !IS_INTRA(mb_type)){
265
            int list;
266
            int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
267
            for(i=0; i<16; i++)
268
                h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
269
            for(list=0; list<h->list_count; list++){
270
                if(USES_LIST(mb_type,list)){
271
                    uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
272
                    uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
273
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
274
                    for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
275
                        dst[0] = src[0];
276
                        dst[1] = src[1];
277
                        dst[2] = src[2];
278
                        dst[3] = src[3];
279
                    }
280
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
281
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
282
                    ref += h->b8_stride;
283
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
284
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
285
                }else{
286
                    fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
287
                    fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
288
                }
289
            }
290
        }
291
    }else{
292
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
293
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
294
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
295
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
296
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
297
    }
298

    
299
    if(IS_INTRA(mb_type)){
300
        h->topleft_samples_available=
301
        h->top_samples_available=
302
        h->left_samples_available= 0xFFFF;
303
        h->topright_samples_available= 0xEEEA;
304

    
305
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
306
            h->topleft_samples_available= 0xB3FF;
307
            h->top_samples_available= 0x33FF;
308
            h->topright_samples_available= 0x26EA;
309
        }
310
        for(i=0; i<2; i++){
311
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
312
                h->topleft_samples_available&= 0xDF5F;
313
                h->left_samples_available&= 0x5F5F;
314
            }
315
        }
316

    
317
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
318
            h->topleft_samples_available&= 0x7FFF;
319

    
320
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
321
            h->topright_samples_available&= 0xFBFF;
322

    
323
        if(IS_INTRA4x4(mb_type)){
324
            if(IS_INTRA4x4(top_type)){
325
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
326
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
327
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
328
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
329
            }else{
330
                int pred;
331
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
332
                    pred= -1;
333
                else{
334
                    pred= 2;
335
                }
336
                h->intra4x4_pred_mode_cache[4+8*0]=
337
                h->intra4x4_pred_mode_cache[5+8*0]=
338
                h->intra4x4_pred_mode_cache[6+8*0]=
339
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
340
            }
341
            for(i=0; i<2; i++){
342
                if(IS_INTRA4x4(left_type[i])){
343
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
344
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
345
                }else{
346
                    int pred;
347
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
348
                        pred= -1;
349
                    else{
350
                        pred= 2;
351
                    }
352
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
353
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
354
                }
355
            }
356
        }
357
    }
358

    
359

    
360
/*
361
0 . T T. T T T T
362
1 L . .L . . . .
363
2 L . .L . . . .
364
3 . T TL . . . .
365
4 L . .L . . . .
366
5 L . .. . . . .
367
*/
368
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
369
    if(top_type){
370
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
371
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
372
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
373
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
374

    
375
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
376
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
377

    
378
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
379
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
380

    
381
    }else{
382
        h->non_zero_count_cache[4+8*0]=
383
        h->non_zero_count_cache[5+8*0]=
384
        h->non_zero_count_cache[6+8*0]=
385
        h->non_zero_count_cache[7+8*0]=
386

    
387
        h->non_zero_count_cache[1+8*0]=
388
        h->non_zero_count_cache[2+8*0]=
389

    
390
        h->non_zero_count_cache[1+8*3]=
391
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
392

    
393
    }
394

    
395
    for (i=0; i<2; i++) {
396
        if(left_type[i]){
397
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
398
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
399
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
400
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
401
        }else{
402
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
403
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
404
            h->non_zero_count_cache[0+8*1 +   8*i]=
405
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
406
        }
407
    }
408

    
409
    if( h->pps.cabac ) {
410
        // top_cbp
411
        if(top_type) {
412
            h->top_cbp = h->cbp_table[top_xy];
413
        } else if(IS_INTRA(mb_type)) {
414
            h->top_cbp = 0x1C0;
415
        } else {
416
            h->top_cbp = 0;
417
        }
418
        // left_cbp
419
        if (left_type[0]) {
420
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
421
        } else if(IS_INTRA(mb_type)) {
422
            h->left_cbp = 0x1C0;
423
        } else {
424
            h->left_cbp = 0;
425
        }
426
        if (left_type[0]) {
427
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
428
        }
429
        if (left_type[1]) {
430
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
431
        }
432
    }
433

    
434
#if 1
435
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
436
        int list;
437
        for(list=0; list<h->list_count; list++){
438
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
439
                /*if(!h->mv_cache_clean[list]){
440
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
441
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
442
                    h->mv_cache_clean[list]= 1;
443
                }*/
444
                continue;
445
            }
446
            h->mv_cache_clean[list]= 0;
447

    
448
            if(USES_LIST(top_type, list)){
449
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
450
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
451
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
452
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
453
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
454
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
455
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
456
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
457
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
458
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
459
            }else{
460
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
461
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
462
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
463
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
464
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
465
            }
466

    
467
            for(i=0; i<2; i++){
468
                int cache_idx = scan8[0] - 1 + i*2*8;
469
                if(USES_LIST(left_type[i], list)){
470
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
471
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
472
                    *(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]];
473
                    *(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]];
474
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
475
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
476
                }else{
477
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
478
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
479
                    h->ref_cache[list][cache_idx  ]=
480
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
481
                }
482
            }
483

    
484
            if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
485
                continue;
486

    
487
            if(USES_LIST(topleft_type, list)){
488
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
489
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
490
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
491
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
492
            }else{
493
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
494
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
495
            }
496

    
497
            if(USES_LIST(topright_type, list)){
498
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
499
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
500
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
501
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
502
            }else{
503
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
504
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
505
            }
506

    
507
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
508
                continue;
509

    
510
            h->ref_cache[list][scan8[5 ]+1] =
511
            h->ref_cache[list][scan8[7 ]+1] =
512
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
513
            h->ref_cache[list][scan8[4 ]] =
514
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
515
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
516
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
517
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
518
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
519
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
520

    
521
            if( h->pps.cabac ) {
522
                /* XXX beurk, Load mvd */
523
                if(USES_LIST(top_type, list)){
524
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
525
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
526
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
527
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
528
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
529
                }else{
530
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
531
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
532
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
533
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
534
                }
535
                if(USES_LIST(left_type[0], list)){
536
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
537
                    *(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]];
538
                    *(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]];
539
                }else{
540
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
541
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
542
                }
543
                if(USES_LIST(left_type[1], list)){
544
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
545
                    *(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]];
546
                    *(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]];
547
                }else{
548
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
549
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
550
                }
551
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
552
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
553
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
554
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
555
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
556

    
557
                if(h->slice_type == B_TYPE){
558
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
559

    
560
                    if(IS_DIRECT(top_type)){
561
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
562
                    }else if(IS_8X8(top_type)){
563
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
564
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
565
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
566
                    }else{
567
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
568
                    }
569

    
570
                    if(IS_DIRECT(left_type[0]))
571
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
572
                    else if(IS_8X8(left_type[0]))
573
                        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)];
574
                    else
575
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
576

    
577
                    if(IS_DIRECT(left_type[1]))
578
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
579
                    else if(IS_8X8(left_type[1]))
580
                        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)];
581
                    else
582
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
583
                }
584
            }
585

    
586
            if(FRAME_MBAFF){
587
#define MAP_MVS\
588
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
589
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
590
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
591
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
592
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
593
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
594
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
595
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
596
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
597
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
598
                if(MB_FIELD){
599
#define MAP_F2F(idx, mb_type)\
600
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
601
                        h->ref_cache[list][idx] <<= 1;\
602
                        h->mv_cache[list][idx][1] /= 2;\
603
                        h->mvd_cache[list][idx][1] /= 2;\
604
                    }
605
                    MAP_MVS
606
#undef MAP_F2F
607
                }else{
608
#define MAP_F2F(idx, mb_type)\
609
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
610
                        h->ref_cache[list][idx] >>= 1;\
611
                        h->mv_cache[list][idx][1] <<= 1;\
612
                        h->mvd_cache[list][idx][1] <<= 1;\
613
                    }
614
                    MAP_MVS
615
#undef MAP_F2F
616
                }
617
            }
618
        }
619
    }
620
#endif
621

    
622
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
623
}
624

    
625
static inline void write_back_intra_pred_mode(H264Context *h){
626
    MpegEncContext * const s = &h->s;
627
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
628

    
629
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
630
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
631
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
632
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
633
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
634
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
635
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
636
}
637

    
638
/**
639
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
640
 */
641
static inline int check_intra4x4_pred_mode(H264Context *h){
642
    MpegEncContext * const s = &h->s;
643
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
644
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
645
    int i;
646

    
647
    if(!(h->top_samples_available&0x8000)){
648
        for(i=0; i<4; i++){
649
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
650
            if(status<0){
651
                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);
652
                return -1;
653
            } else if(status){
654
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
655
            }
656
        }
657
    }
658

    
659
    if(!(h->left_samples_available&0x8000)){
660
        for(i=0; i<4; i++){
661
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
662
            if(status<0){
663
                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);
664
                return -1;
665
            } else if(status){
666
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
667
            }
668
        }
669
    }
670

    
671
    return 0;
672
} //FIXME cleanup like next
673

    
674
/**
675
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
676
 */
677
static inline int check_intra_pred_mode(H264Context *h, int mode){
678
    MpegEncContext * const s = &h->s;
679
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
680
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
681

    
682
    if(mode > 6U) {
683
        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);
684
        return -1;
685
    }
686

    
687
    if(!(h->top_samples_available&0x8000)){
688
        mode= top[ mode ];
689
        if(mode<0){
690
            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);
691
            return -1;
692
        }
693
    }
694

    
695
    if(!(h->left_samples_available&0x8000)){
696
        mode= left[ mode ];
697
        if(mode<0){
698
            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);
699
            return -1;
700
        }
701
    }
702

    
703
    return mode;
704
}
705

    
706
/**
707
 * gets the predicted intra4x4 prediction mode.
708
 */
709
static inline int pred_intra_mode(H264Context *h, int n){
710
    const int index8= scan8[n];
711
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
712
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
713
    const int min= FFMIN(left, top);
714

    
715
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
716

    
717
    if(min<0) return DC_PRED;
718
    else      return min;
719
}
720

    
721
static inline void write_back_non_zero_count(H264Context *h){
722
    MpegEncContext * const s = &h->s;
723
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
724

    
725
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
726
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
727
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
728
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
729
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
730
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
731
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
732

    
733
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
734
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
735
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
736

    
737
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
738
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
739
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
740

    
741
    if(FRAME_MBAFF){
742
        // store all luma nnzs, for deblocking
743
        int v = 0, i;
744
        for(i=0; i<16; i++)
745
            v += (!!h->non_zero_count_cache[scan8[i]]) << i;
746
        *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
747
    }
748
}
749

    
750
/**
751
 * gets the predicted number of non zero coefficients.
752
 * @param n block index
753
 */
754
static inline int pred_non_zero_count(H264Context *h, int n){
755
    const int index8= scan8[n];
756
    const int left= h->non_zero_count_cache[index8 - 1];
757
    const int top = h->non_zero_count_cache[index8 - 8];
758
    int i= left + top;
759

    
760
    if(i<64) i= (i+1)>>1;
761

    
762
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
763

    
764
    return i&31;
765
}
766

    
767
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
768
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
769
    MpegEncContext *s = &h->s;
770

    
771
    /* there is no consistent mapping of mvs to neighboring locations that will
772
     * make mbaff happy, so we can't move all this logic to fill_caches */
773
    if(FRAME_MBAFF){
774
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
775
        const int16_t *mv;
776
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
777
        *C = h->mv_cache[list][scan8[0]-2];
778

    
779
        if(!MB_FIELD
780
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
781
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
782
            if(IS_INTERLACED(mb_types[topright_xy])){
783
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
784
                const int x4 = X4, y4 = Y4;\
785
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
786
                if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
787
                    return LIST_NOT_USED;\
788
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
789
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
790
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
791
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
792

    
793
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
794
            }
795
        }
796
        if(topright_ref == PART_NOT_AVAILABLE
797
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
798
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
799
            if(!MB_FIELD
800
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
801
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
802
            }
803
            if(MB_FIELD
804
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
805
               && i >= scan8[0]+8){
806
                // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
807
                SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
808
            }
809
        }
810
#undef SET_DIAG_MV
811
    }
812

    
813
    if(topright_ref != PART_NOT_AVAILABLE){
814
        *C= h->mv_cache[list][ i - 8 + part_width ];
815
        return topright_ref;
816
    }else{
817
        tprintf(s->avctx, "topright MV not available\n");
818

    
819
        *C= h->mv_cache[list][ i - 8 - 1 ];
820
        return h->ref_cache[list][ i - 8 - 1 ];
821
    }
822
}
823

    
824
/**
825
 * gets the predicted MV.
826
 * @param n the block index
827
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
828
 * @param mx the x component of the predicted motion vector
829
 * @param my the y component of the predicted motion vector
830
 */
831
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
832
    const int index8= scan8[n];
833
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
834
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
835
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
836
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
837
    const int16_t * C;
838
    int diagonal_ref, match_count;
839

    
840
    assert(part_width==1 || part_width==2 || part_width==4);
841

    
842
/* mv_cache
843
  B . . A T T T T
844
  U . . L . . , .
845
  U . . L . . . .
846
  U . . L . . , .
847
  . . . L . . . .
848
*/
849

    
850
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
851
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
852
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
853
    if(match_count > 1){ //most common
854
        *mx= mid_pred(A[0], B[0], C[0]);
855
        *my= mid_pred(A[1], B[1], C[1]);
856
    }else if(match_count==1){
857
        if(left_ref==ref){
858
            *mx= A[0];
859
            *my= A[1];
860
        }else if(top_ref==ref){
861
            *mx= B[0];
862
            *my= B[1];
863
        }else{
864
            *mx= C[0];
865
            *my= C[1];
866
        }
867
    }else{
868
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
869
            *mx= A[0];
870
            *my= A[1];
871
        }else{
872
            *mx= mid_pred(A[0], B[0], C[0]);
873
            *my= mid_pred(A[1], B[1], C[1]);
874
        }
875
    }
876

    
877
    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);
878
}
879

    
880
/**
881
 * gets the directionally predicted 16x8 MV.
882
 * @param n the block index
883
 * @param mx the x component of the predicted motion vector
884
 * @param my the y component of the predicted motion vector
885
 */
886
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
887
    if(n==0){
888
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
889
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
890

    
891
        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);
892

    
893
        if(top_ref == ref){
894
            *mx= B[0];
895
            *my= B[1];
896
            return;
897
        }
898
    }else{
899
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
900
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
901

    
902
        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);
903

    
904
        if(left_ref == ref){
905
            *mx= A[0];
906
            *my= A[1];
907
            return;
908
        }
909
    }
910

    
911
    //RARE
912
    pred_motion(h, n, 4, list, ref, mx, my);
913
}
914

    
915
/**
916
 * gets the directionally predicted 8x16 MV.
917
 * @param n the block index
918
 * @param mx the x component of the predicted motion vector
919
 * @param my the y component of the predicted motion vector
920
 */
921
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
922
    if(n==0){
923
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
924
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
925

    
926
        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);
927

    
928
        if(left_ref == ref){
929
            *mx= A[0];
930
            *my= A[1];
931
            return;
932
        }
933
    }else{
934
        const int16_t * C;
935
        int diagonal_ref;
936

    
937
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
938

    
939
        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);
940

    
941
        if(diagonal_ref == ref){
942
            *mx= C[0];
943
            *my= C[1];
944
            return;
945
        }
946
    }
947

    
948
    //RARE
949
    pred_motion(h, n, 2, list, ref, mx, my);
950
}
951

    
952
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
953
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
954
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
955

    
956
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
957

    
958
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
959
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
960
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
961

    
962
        *mx = *my = 0;
963
        return;
964
    }
965

    
966
    pred_motion(h, 0, 4, 0, 0, mx, my);
967

    
968
    return;
969
}
970

    
971
static inline void direct_dist_scale_factor(H264Context * const h){
972
    const int poc = h->s.current_picture_ptr->poc;
973
    const int poc1 = h->ref_list[1][0].poc;
974
    int i;
975
    for(i=0; i<h->ref_count[0]; i++){
976
        int poc0 = h->ref_list[0][i].poc;
977
        int td = av_clip(poc1 - poc0, -128, 127);
978
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
979
            h->dist_scale_factor[i] = 256;
980
        }else{
981
            int tb = av_clip(poc - poc0, -128, 127);
982
            int tx = (16384 + (FFABS(td) >> 1)) / td;
983
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
984
        }
985
    }
986
    if(FRAME_MBAFF){
987
        for(i=0; i<h->ref_count[0]; i++){
988
            h->dist_scale_factor_field[2*i] =
989
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
990
        }
991
    }
992
}
993
static inline void direct_ref_list_init(H264Context * const h){
994
    MpegEncContext * const s = &h->s;
995
    Picture * const ref1 = &h->ref_list[1][0];
996
    Picture * const cur = s->current_picture_ptr;
997
    int list, i, j;
998
    if(cur->pict_type == I_TYPE)
999
        cur->ref_count[0] = 0;
1000
    if(cur->pict_type != B_TYPE)
1001
        cur->ref_count[1] = 0;
1002
    for(list=0; list<2; list++){
1003
        cur->ref_count[list] = h->ref_count[list];
1004
        for(j=0; j<h->ref_count[list]; j++)
1005
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1006
    }
1007
    if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1008
        return;
1009
    for(list=0; list<2; list++){
1010
        for(i=0; i<ref1->ref_count[list]; i++){
1011
            const int poc = ref1->ref_poc[list][i];
1012
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
1013
            for(j=0; j<h->ref_count[list]; j++)
1014
                if(h->ref_list[list][j].poc == poc){
1015
                    h->map_col_to_list0[list][i] = j;
1016
                    break;
1017
                }
1018
        }
1019
    }
1020
    if(FRAME_MBAFF){
1021
        for(list=0; list<2; list++){
1022
            for(i=0; i<ref1->ref_count[list]; i++){
1023
                j = h->map_col_to_list0[list][i];
1024
                h->map_col_to_list0_field[list][2*i] = 2*j;
1025
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1026
            }
1027
        }
1028
    }
1029
}
1030

    
1031
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1032
    MpegEncContext * const s = &h->s;
1033
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1034
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1035
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1036
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1037
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1038
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1039
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1040
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1041
    const int is_b8x8 = IS_8X8(*mb_type);
1042
    unsigned int sub_mb_type;
1043
    int i8, i4;
1044

    
1045
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
1046
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1047
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1048
         * so we know exactly what block size to use */
1049
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1050
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1051
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
1052
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1053
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1054
    }else{
1055
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1056
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
1057
    }
1058
    if(!is_b8x8)
1059
        *mb_type |= MB_TYPE_DIRECT2;
1060
    if(MB_FIELD)
1061
        *mb_type |= MB_TYPE_INTERLACED;
1062

    
1063
    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);
1064

    
1065
    if(h->direct_spatial_mv_pred){
1066
        int ref[2];
1067
        int mv[2][2];
1068
        int list;
1069

    
1070
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1071

    
1072
        /* ref = min(neighbors) */
1073
        for(list=0; list<2; list++){
1074
            int refa = h->ref_cache[list][scan8[0] - 1];
1075
            int refb = h->ref_cache[list][scan8[0] - 8];
1076
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1077
            if(refc == -2)
1078
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1079
            ref[list] = refa;
1080
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1081
                ref[list] = refb;
1082
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1083
                ref[list] = refc;
1084
            if(ref[list] < 0)
1085
                ref[list] = -1;
1086
        }
1087

    
1088
        if(ref[0] < 0 && ref[1] < 0){
1089
            ref[0] = ref[1] = 0;
1090
            mv[0][0] = mv[0][1] =
1091
            mv[1][0] = mv[1][1] = 0;
1092
        }else{
1093
            for(list=0; list<2; list++){
1094
                if(ref[list] >= 0)
1095
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1096
                else
1097
                    mv[list][0] = mv[list][1] = 0;
1098
            }
1099
        }
1100

    
1101
        if(ref[1] < 0){
1102
            *mb_type &= ~MB_TYPE_P0L1;
1103
            sub_mb_type &= ~MB_TYPE_P0L1;
1104
        }else if(ref[0] < 0){
1105
            *mb_type &= ~MB_TYPE_P0L0;
1106
            sub_mb_type &= ~MB_TYPE_P0L0;
1107
        }
1108

    
1109
        if(IS_16X16(*mb_type)){
1110
            int a=0, b=0;
1111

    
1112
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1113
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1114
            if(!IS_INTRA(mb_type_col)
1115
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1116
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1117
                       && (h->x264_build>33 || !h->x264_build)))){
1118
                if(ref[0] > 0)
1119
                    a= pack16to32(mv[0][0],mv[0][1]);
1120
                if(ref[1] > 0)
1121
                    b= pack16to32(mv[1][0],mv[1][1]);
1122
            }else{
1123
                a= pack16to32(mv[0][0],mv[0][1]);
1124
                b= pack16to32(mv[1][0],mv[1][1]);
1125
            }
1126
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1127
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1128
        }else{
1129
            for(i8=0; i8<4; i8++){
1130
                const int x8 = i8&1;
1131
                const int y8 = i8>>1;
1132

    
1133
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1134
                    continue;
1135
                h->sub_mb_type[i8] = sub_mb_type;
1136

    
1137
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1138
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1139
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1140
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1141

    
1142
                /* col_zero_flag */
1143
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
1144
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1145
                                                  && (h->x264_build>33 || !h->x264_build)))){
1146
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1147
                    if(IS_SUB_8X8(sub_mb_type)){
1148
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1149
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1150
                            if(ref[0] == 0)
1151
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1152
                            if(ref[1] == 0)
1153
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1154
                        }
1155
                    }else
1156
                    for(i4=0; i4<4; i4++){
1157
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1158
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1159
                            if(ref[0] == 0)
1160
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1161
                            if(ref[1] == 0)
1162
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1163
                        }
1164
                    }
1165
                }
1166
            }
1167
        }
1168
    }else{ /* direct temporal mv pred */
1169
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1170
        const int *dist_scale_factor = h->dist_scale_factor;
1171

    
1172
        if(FRAME_MBAFF){
1173
            if(IS_INTERLACED(*mb_type)){
1174
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
1175
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
1176
                dist_scale_factor = h->dist_scale_factor_field;
1177
            }
1178
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1179
                /* FIXME assumes direct_8x8_inference == 1 */
1180
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1181
                int mb_types_col[2];
1182
                int y_shift;
1183

    
1184
                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1185
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1186
                         | (*mb_type & MB_TYPE_INTERLACED);
1187
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1188

    
1189
                if(IS_INTERLACED(*mb_type)){
1190
                    /* frame to field scaling */
1191
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1192
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1193
                    if(s->mb_y&1){
1194
                        l1ref0 -= 2*h->b8_stride;
1195
                        l1ref1 -= 2*h->b8_stride;
1196
                        l1mv0 -= 4*h->b_stride;
1197
                        l1mv1 -= 4*h->b_stride;
1198
                    }
1199
                    y_shift = 0;
1200

    
1201
                    if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1202
                       && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1203
                       && !is_b8x8)
1204
                        *mb_type |= MB_TYPE_16x8;
1205
                    else
1206
                        *mb_type |= MB_TYPE_8x8;
1207
                }else{
1208
                    /* field to frame scaling */
1209
                    /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1210
                     * but in MBAFF, top and bottom POC are equal */
1211
                    int dy = (s->mb_y&1) ? 1 : 2;
1212
                    mb_types_col[0] =
1213
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1214
                    l1ref0 += dy*h->b8_stride;
1215
                    l1ref1 += dy*h->b8_stride;
1216
                    l1mv0 += 2*dy*h->b_stride;
1217
                    l1mv1 += 2*dy*h->b_stride;
1218
                    y_shift = 2;
1219

    
1220
                    if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1221
                       && !is_b8x8)
1222
                        *mb_type |= MB_TYPE_16x16;
1223
                    else
1224
                        *mb_type |= MB_TYPE_8x8;
1225
                }
1226

    
1227
                for(i8=0; i8<4; i8++){
1228
                    const int x8 = i8&1;
1229
                    const int y8 = i8>>1;
1230
                    int ref0, scale;
1231
                    const int16_t (*l1mv)[2]= l1mv0;
1232

    
1233
                    if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1234
                        continue;
1235
                    h->sub_mb_type[i8] = sub_mb_type;
1236

    
1237
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1238
                    if(IS_INTRA(mb_types_col[y8])){
1239
                        fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1240
                        fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1241
                        fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1242
                        continue;
1243
                    }
1244

    
1245
                    ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1246
                    if(ref0 >= 0)
1247
                        ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1248
                    else{
1249
                        ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1250
                        l1mv= l1mv1;
1251
                    }
1252
                    scale = dist_scale_factor[ref0];
1253
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1254

    
1255
                    {
1256
                        const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1257
                        int my_col = (mv_col[1]<<y_shift)/2;
1258
                        int mx = (scale * mv_col[0] + 128) >> 8;
1259
                        int my = (scale * my_col + 128) >> 8;
1260
                        fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1261
                        fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1262
                    }
1263
                }
1264
                return;
1265
            }
1266
        }
1267

    
1268
        /* one-to-one mv scaling */
1269

    
1270
        if(IS_16X16(*mb_type)){
1271
            int ref, mv0, mv1;
1272

    
1273
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1274
            if(IS_INTRA(mb_type_col)){
1275
                ref=mv0=mv1=0;
1276
            }else{
1277
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1278
                                                : map_col_to_list0[1][l1ref1[0]];
1279
                const int scale = dist_scale_factor[ref0];
1280
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1281
                int mv_l0[2];
1282
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1283
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1284
                ref= ref0;
1285
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1286
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1287
            }
1288
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1289
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1290
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1291
        }else{
1292
            for(i8=0; i8<4; i8++){
1293
                const int x8 = i8&1;
1294
                const int y8 = i8>>1;
1295
                int ref0, scale;
1296
                const int16_t (*l1mv)[2]= l1mv0;
1297

    
1298
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1299
                    continue;
1300
                h->sub_mb_type[i8] = sub_mb_type;
1301
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1302
                if(IS_INTRA(mb_type_col)){
1303
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1304
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1305
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1306
                    continue;
1307
                }
1308

    
1309
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1310
                if(ref0 >= 0)
1311
                    ref0 = map_col_to_list0[0][ref0];
1312
                else{
1313
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1314
                    l1mv= l1mv1;
1315
                }
1316
                scale = dist_scale_factor[ref0];
1317

    
1318
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1319
                if(IS_SUB_8X8(sub_mb_type)){
1320
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1321
                    int mx = (scale * mv_col[0] + 128) >> 8;
1322
                    int my = (scale * mv_col[1] + 128) >> 8;
1323
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1324
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1325
                }else
1326
                for(i4=0; i4<4; i4++){
1327
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1328
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1329
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1330
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1331
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1332
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1333
                }
1334
            }
1335
        }
1336
    }
1337
}
1338

    
1339
static inline void write_back_motion(H264Context *h, int mb_type){
1340
    MpegEncContext * const s = &h->s;
1341
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1342
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1343
    int list;
1344

    
1345
    if(!USES_LIST(mb_type, 0))
1346
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1347

    
1348
    for(list=0; list<h->list_count; list++){
1349
        int y;
1350
        if(!USES_LIST(mb_type, list))
1351
            continue;
1352

    
1353
        for(y=0; y<4; y++){
1354
            *(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];
1355
            *(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];
1356
        }
1357
        if( h->pps.cabac ) {
1358
            if(IS_SKIP(mb_type))
1359
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1360
            else
1361
            for(y=0; y<4; y++){
1362
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1363
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1364
            }
1365
        }
1366

    
1367
        {
1368
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1369
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1370
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1371
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1372
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1373
        }
1374
    }
1375

    
1376
    if(h->slice_type == B_TYPE && h->pps.cabac){
1377
        if(IS_8X8(mb_type)){
1378
            uint8_t *direct_table = &h->direct_table[b8_xy];
1379
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1380
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1381
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1382
        }
1383
    }
1384
}
1385

    
1386
/**
1387
 * Decodes a network abstraction layer unit.
1388
 * @param consumed is the number of bytes used as input
1389
 * @param length is the length of the array
1390
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1391
 * @returns decoded bytes, might be src+1 if no escapes
1392
 */
1393
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1394
    int i, si, di;
1395
    uint8_t *dst;
1396

    
1397
//    src[0]&0x80;                //forbidden bit
1398
    h->nal_ref_idc= src[0]>>5;
1399
    h->nal_unit_type= src[0]&0x1F;
1400

    
1401
    src++; length--;
1402
#if 0
1403
    for(i=0; i<length; i++)
1404
        printf("%2X ", src[i]);
1405
#endif
1406
    for(i=0; i+1<length; i+=2){
1407
        if(src[i]) continue;
1408
        if(i>0 && src[i-1]==0) i--;
1409
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1410
            if(src[i+2]!=3){
1411
                /* startcode, so we must be past the end */
1412
                length=i;
1413
            }
1414
            break;
1415
        }
1416
    }
1417

    
1418
    if(i>=length-1){ //no escaped 0
1419
        *dst_length= length;
1420
        *consumed= length+1; //+1 for the header
1421
        return src;
1422
    }
1423

    
1424
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1425
    dst= h->rbsp_buffer;
1426

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
1551
#undef xStride
1552
#undef stride
1553

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

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

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

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

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

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

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

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

    
1598
/**
1599
 * gets the chroma qp.
1600
 */
1601
static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1602

    
1603
    return chroma_qp[av_clip(qscale + chroma_qp_index_offset, 0, 51)];
1604
}
1605

    
1606
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1607
//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1608
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
1609
    int i;
1610
    const int * const quant_table= quant_coeff[qscale];
1611
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1612
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1613
    const unsigned int threshold2= (threshold1<<1);
1614
    int last_non_zero;
1615

    
1616
    if(separate_dc){
1617
        if(qscale<=18){
1618
            //avoid overflows
1619
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1620
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1621
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1622

    
1623
            int level= block[0]*quant_coeff[qscale+18][0];
1624
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1625
                if(level>0){
1626
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1627
                    block[0]= level;
1628
                }else{
1629
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1630
                    block[0]= -level;
1631
                }
1632
//                last_non_zero = i;
1633
            }else{
1634
                block[0]=0;
1635
            }
1636
        }else{
1637
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1638
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1639
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1640

    
1641
            int level= block[0]*quant_table[0];
1642
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1643
                if(level>0){
1644
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1645
                    block[0]= level;
1646
                }else{
1647
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1648
                    block[0]= -level;
1649
                }
1650
//                last_non_zero = i;
1651
            }else{
1652
                block[0]=0;
1653
            }
1654
        }
1655
        last_non_zero= 0;
1656
        i=1;
1657
    }else{
1658
        last_non_zero= -1;
1659
        i=0;
1660
    }
1661

    
1662
    for(; i<16; i++){
1663
        const int j= scantable[i];
1664
        int level= block[j]*quant_table[j];
1665

    
1666
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1667
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1668
        if(((unsigned)(level+threshold1))>threshold2){
1669
            if(level>0){
1670
                level= (bias + level)>>QUANT_SHIFT;
1671
                block[j]= level;
1672
            }else{
1673
                level= (bias - level)>>QUANT_SHIFT;
1674
                block[j]= -level;
1675
            }
1676
            last_non_zero = i;
1677
        }else{
1678
            block[j]=0;
1679
        }
1680
    }
1681

    
1682
    return last_non_zero;
1683
}
1684

    
1685
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1686
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1687
    ((uint32_t*)(src+0*stride))[0]= a;
1688
    ((uint32_t*)(src+1*stride))[0]= a;
1689
    ((uint32_t*)(src+2*stride))[0]= a;
1690
    ((uint32_t*)(src+3*stride))[0]= a;
1691
}
1692

    
1693
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1694
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1695
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1696
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1697
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1698
}
1699

    
1700
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1701
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1702
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1703

    
1704
    ((uint32_t*)(src+0*stride))[0]=
1705
    ((uint32_t*)(src+1*stride))[0]=
1706
    ((uint32_t*)(src+2*stride))[0]=
1707
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1708
}
1709

    
1710
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1711
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1712

    
1713
    ((uint32_t*)(src+0*stride))[0]=
1714
    ((uint32_t*)(src+1*stride))[0]=
1715
    ((uint32_t*)(src+2*stride))[0]=
1716
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1717
}
1718

    
1719
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1720
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1721

    
1722
    ((uint32_t*)(src+0*stride))[0]=
1723
    ((uint32_t*)(src+1*stride))[0]=
1724
    ((uint32_t*)(src+2*stride))[0]=
1725
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1726
}
1727

    
1728
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1729
    ((uint32_t*)(src+0*stride))[0]=
1730
    ((uint32_t*)(src+1*stride))[0]=
1731
    ((uint32_t*)(src+2*stride))[0]=
1732
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1733
}
1734

    
1735

    
1736
#define LOAD_TOP_RIGHT_EDGE\
1737
    const int t4= topright[0];\
1738
    const int t5= topright[1];\
1739
    const int t6= topright[2];\
1740
    const int t7= topright[3];\
1741

    
1742
#define LOAD_LEFT_EDGE\
1743
    const int l0= src[-1+0*stride];\
1744
    const int l1= src[-1+1*stride];\
1745
    const int l2= src[-1+2*stride];\
1746
    const int l3= src[-1+3*stride];\
1747

    
1748
#define LOAD_TOP_EDGE\
1749
    const int t0= src[ 0-1*stride];\
1750
    const int t1= src[ 1-1*stride];\
1751
    const int t2= src[ 2-1*stride];\
1752
    const int t3= src[ 3-1*stride];\
1753

    
1754
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1755
    const int lt= src[-1-1*stride];
1756
    LOAD_TOP_EDGE
1757
    LOAD_LEFT_EDGE
1758

    
1759
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1760
    src[0+2*stride]=
1761
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1762
    src[0+1*stride]=
1763
    src[1+2*stride]=
1764
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1765
    src[0+0*stride]=
1766
    src[1+1*stride]=
1767
    src[2+2*stride]=
1768
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1769
    src[1+0*stride]=
1770
    src[2+1*stride]=
1771
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1772
    src[2+0*stride]=
1773
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1774
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1775
}
1776

    
1777
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1778
    LOAD_TOP_EDGE
1779
    LOAD_TOP_RIGHT_EDGE
1780
//    LOAD_LEFT_EDGE
1781

    
1782
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1783
    src[1+0*stride]=
1784
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1785
    src[2+0*stride]=
1786
    src[1+1*stride]=
1787
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1788
    src[3+0*stride]=
1789
    src[2+1*stride]=
1790
    src[1+2*stride]=
1791
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1792
    src[3+1*stride]=
1793
    src[2+2*stride]=
1794
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1795
    src[3+2*stride]=
1796
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1797
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1798
}
1799

    
1800
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1801
    const int lt= src[-1-1*stride];
1802
    LOAD_TOP_EDGE
1803
    LOAD_LEFT_EDGE
1804
    const __attribute__((unused)) int unu= l3;
1805

    
1806
    src[0+0*stride]=
1807
    src[1+2*stride]=(lt + t0 + 1)>>1;
1808
    src[1+0*stride]=
1809
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1810
    src[2+0*stride]=
1811
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1812
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1813
    src[0+1*stride]=
1814
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1815
    src[1+1*stride]=
1816
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1817
    src[2+1*stride]=
1818
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1819
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1820
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1821
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1822
}
1823

    
1824
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1825
    LOAD_TOP_EDGE
1826
    LOAD_TOP_RIGHT_EDGE
1827
    const __attribute__((unused)) int unu= t7;
1828

    
1829
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1830
    src[1+0*stride]=
1831
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1832
    src[2+0*stride]=
1833
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1834
    src[3+0*stride]=
1835
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1836
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1837
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1838
    src[1+1*stride]=
1839
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1840
    src[2+1*stride]=
1841
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1842
    src[3+1*stride]=
1843
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1844
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1845
}
1846

    
1847
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1848
    LOAD_LEFT_EDGE
1849

    
1850
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1851
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1852
    src[2+0*stride]=
1853
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1854
    src[3+0*stride]=
1855
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1856
    src[2+1*stride]=
1857
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1858
    src[3+1*stride]=
1859
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1860
    src[3+2*stride]=
1861
    src[1+3*stride]=
1862
    src[0+3*stride]=
1863
    src[2+2*stride]=
1864
    src[2+3*stride]=
1865
    src[3+3*stride]=l3;
1866
}
1867

    
1868
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1869
    const int lt= src[-1-1*stride];
1870
    LOAD_TOP_EDGE
1871
    LOAD_LEFT_EDGE
1872
    const __attribute__((unused)) int unu= t3;
1873

    
1874
    src[0+0*stride]=
1875
    src[2+1*stride]=(lt + l0 + 1)>>1;
1876
    src[1+0*stride]=
1877
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1878
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1879
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1880
    src[0+1*stride]=
1881
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1882
    src[1+1*stride]=
1883
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1884
    src[0+2*stride]=
1885
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1886
    src[1+2*stride]=
1887
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1888
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1889
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1890
}
1891

    
1892
void ff_pred16x16_vertical_c(uint8_t *src, int stride){
1893
    int i;
1894
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1895
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1896
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1897
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1898

    
1899
    for(i=0; i<16; i++){
1900
        ((uint32_t*)(src+i*stride))[0]= a;
1901
        ((uint32_t*)(src+i*stride))[1]= b;
1902
        ((uint32_t*)(src+i*stride))[2]= c;
1903
        ((uint32_t*)(src+i*stride))[3]= d;
1904
    }
1905
}
1906

    
1907
void ff_pred16x16_horizontal_c(uint8_t *src, int stride){
1908
    int i;
1909

    
1910
    for(i=0; i<16; i++){
1911
        ((uint32_t*)(src+i*stride))[0]=
1912
        ((uint32_t*)(src+i*stride))[1]=
1913
        ((uint32_t*)(src+i*stride))[2]=
1914
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1915
    }
1916
}
1917

    
1918
void ff_pred16x16_dc_c(uint8_t *src, int stride){
1919
    int i, dc=0;
1920

    
1921
    for(i=0;i<16; i++){
1922
        dc+= src[-1+i*stride];
1923
    }
1924

    
1925
    for(i=0;i<16; i++){
1926
        dc+= src[i-stride];
1927
    }
1928

    
1929
    dc= 0x01010101*((dc + 16)>>5);
1930

    
1931
    for(i=0; i<16; i++){
1932
        ((uint32_t*)(src+i*stride))[0]=
1933
        ((uint32_t*)(src+i*stride))[1]=
1934
        ((uint32_t*)(src+i*stride))[2]=
1935
        ((uint32_t*)(src+i*stride))[3]= dc;
1936
    }
1937
}
1938

    
1939
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1940
    int i, dc=0;
1941

    
1942
    for(i=0;i<16; i++){
1943
        dc+= src[-1+i*stride];
1944
    }
1945

    
1946
    dc= 0x01010101*((dc + 8)>>4);
1947

    
1948
    for(i=0; i<16; i++){
1949
        ((uint32_t*)(src+i*stride))[0]=
1950
        ((uint32_t*)(src+i*stride))[1]=
1951
        ((uint32_t*)(src+i*stride))[2]=
1952
        ((uint32_t*)(src+i*stride))[3]= dc;
1953
    }
1954
}
1955

    
1956
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1957
    int i, dc=0;
1958

    
1959
    for(i=0;i<16; i++){
1960
        dc+= src[i-stride];
1961
    }
1962
    dc= 0x01010101*((dc + 8)>>4);
1963

    
1964
    for(i=0; i<16; i++){
1965
        ((uint32_t*)(src+i*stride))[0]=
1966
        ((uint32_t*)(src+i*stride))[1]=
1967
        ((uint32_t*)(src+i*stride))[2]=
1968
        ((uint32_t*)(src+i*stride))[3]= dc;
1969
    }
1970
}
1971

    
1972
void ff_pred16x16_128_dc_c(uint8_t *src, int stride){
1973
    int i;
1974

    
1975
    for(i=0; i<16; i++){
1976
        ((uint32_t*)(src+i*stride))[0]=
1977
        ((uint32_t*)(src+i*stride))[1]=
1978
        ((uint32_t*)(src+i*stride))[2]=
1979
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1980
    }
1981
}
1982

    
1983
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1984
  int i, j, k;
1985
  int a;
1986
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
1987
  const uint8_t * const src0 = src+7-stride;
1988
  const uint8_t *src1 = src+8*stride-1;
1989
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1990
  int H = src0[1] - src0[-1];
1991
  int V = src1[0] - src2[ 0];
1992
  for(k=2; k<=8; ++k) {
1993
    src1 += stride; src2 -= stride;
1994
    H += k*(src0[k] - src0[-k]);
1995
    V += k*(src1[0] - src2[ 0]);
1996
  }
1997
  if(svq3){
1998
    H = ( 5*(H/4) ) / 16;
1999
    V = ( 5*(V/4) ) / 16;
2000

    
2001
    /* required for 100% accuracy */
2002
    i = H; H = V; V = i;
2003
  }else{
2004
    H = ( 5*H+32 ) >> 6;
2005
    V = ( 5*V+32 ) >> 6;
2006
  }
2007

    
2008
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2009
  for(j=16; j>0; --j) {
2010
    int b = a;
2011
    a += V;
2012
    for(i=-16; i<0; i+=4) {
2013
      src[16+i] = cm[ (b    ) >> 5 ];
2014
      src[17+i] = cm[ (b+  H) >> 5 ];
2015
      src[18+i] = cm[ (b+2*H) >> 5 ];
2016
      src[19+i] = cm[ (b+3*H) >> 5 ];
2017
      b += 4*H;
2018
    }
2019
    src += stride;
2020
  }
2021
}
2022

    
2023
void ff_pred16x16_plane_c(uint8_t *src, int stride){
2024
    pred16x16_plane_compat_c(src, stride, 0);
2025
}
2026

    
2027
void ff_pred8x8_vertical_c(uint8_t *src, int stride){
2028
    int i;
2029
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2030
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2031

    
2032
    for(i=0; i<8; i++){
2033
        ((uint32_t*)(src+i*stride))[0]= a;
2034
        ((uint32_t*)(src+i*stride))[1]= b;
2035
    }
2036
}
2037

    
2038
void ff_pred8x8_horizontal_c(uint8_t *src, int stride){
2039
    int i;
2040

    
2041
    for(i=0; i<8; i++){
2042
        ((uint32_t*)(src+i*stride))[0]=
2043
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2044
    }
2045
}
2046

    
2047
void ff_pred8x8_128_dc_c(uint8_t *src, int stride){
2048
    int i;
2049

    
2050
    for(i=0; i<8; i++){
2051
        ((uint32_t*)(src+i*stride))[0]=
2052
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2053
    }
2054
}
2055

    
2056
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2057
    int i;
2058
    int dc0, dc2;
2059

    
2060
    dc0=dc2=0;
2061
    for(i=0;i<4; i++){
2062
        dc0+= src[-1+i*stride];
2063
        dc2+= src[-1+(i+4)*stride];
2064
    }
2065
    dc0= 0x01010101*((dc0 + 2)>>2);
2066
    dc2= 0x01010101*((dc2 + 2)>>2);
2067

    
2068
    for(i=0; i<4; i++){
2069
        ((uint32_t*)(src+i*stride))[0]=
2070
        ((uint32_t*)(src+i*stride))[1]= dc0;
2071
    }
2072
    for(i=4; i<8; i++){
2073
        ((uint32_t*)(src+i*stride))[0]=
2074
        ((uint32_t*)(src+i*stride))[1]= dc2;
2075
    }
2076
}
2077

    
2078
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2079
    int i;
2080
    int dc0, dc1;
2081

    
2082
    dc0=dc1=0;
2083
    for(i=0;i<4; i++){
2084
        dc0+= src[i-stride];
2085
        dc1+= src[4+i-stride];
2086
    }
2087
    dc0= 0x01010101*((dc0 + 2)>>2);
2088
    dc1= 0x01010101*((dc1 + 2)>>2);
2089

    
2090
    for(i=0; i<4; i++){
2091
        ((uint32_t*)(src+i*stride))[0]= dc0;
2092
        ((uint32_t*)(src+i*stride))[1]= dc1;
2093
    }
2094
    for(i=4; i<8; i++){
2095
        ((uint32_t*)(src+i*stride))[0]= dc0;
2096
        ((uint32_t*)(src+i*stride))[1]= dc1;
2097
    }
2098
}
2099

    
2100

    
2101
void ff_pred8x8_dc_c(uint8_t *src, int stride){
2102
    int i;
2103
    int dc0, dc1, dc2, dc3;
2104

    
2105
    dc0=dc1=dc2=0;
2106
    for(i=0;i<4; i++){
2107
        dc0+= src[-1+i*stride] + src[i-stride];
2108
        dc1+= src[4+i-stride];
2109
        dc2+= src[-1+(i+4)*stride];
2110
    }
2111
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2112
    dc0= 0x01010101*((dc0 + 4)>>3);
2113
    dc1= 0x01010101*((dc1 + 2)>>2);
2114
    dc2= 0x01010101*((dc2 + 2)>>2);
2115

    
2116
    for(i=0; i<4; i++){
2117
        ((uint32_t*)(src+i*stride))[0]= dc0;
2118
        ((uint32_t*)(src+i*stride))[1]= dc1;
2119
    }
2120
    for(i=4; i<8; i++){
2121
        ((uint32_t*)(src+i*stride))[0]= dc2;
2122
        ((uint32_t*)(src+i*stride))[1]= dc3;
2123
    }
2124
}
2125

    
2126
void ff_pred8x8_plane_c(uint8_t *src, int stride){
2127
  int j, k;
2128
  int a;
2129
  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
2130
  const uint8_t * const src0 = src+3-stride;
2131
  const uint8_t *src1 = src+4*stride-1;
2132
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2133
  int H = src0[1] - src0[-1];
2134
  int V = src1[0] - src2[ 0];
2135
  for(k=2; k<=4; ++k) {
2136
    src1 += stride; src2 -= stride;
2137
    H += k*(src0[k] - src0[-k]);
2138
    V += k*(src1[0] - src2[ 0]);
2139
  }
2140
  H = ( 17*H+16 ) >> 5;
2141
  V = ( 17*V+16 ) >> 5;
2142

    
2143
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2144
  for(j=8; j>0; --j) {
2145
    int b = a;
2146
    a += V;
2147
    src[0] = cm[ (b    ) >> 5 ];
2148
    src[1] = cm[ (b+  H) >> 5 ];
2149
    src[2] = cm[ (b+2*H) >> 5 ];
2150
    src[3] = cm[ (b+3*H) >> 5 ];
2151
    src[4] = cm[ (b+4*H) >> 5 ];
2152
    src[5] = cm[ (b+5*H) >> 5 ];
2153
    src[6] = cm[ (b+6*H) >> 5 ];
2154
    src[7] = cm[ (b+7*H) >> 5 ];
2155
    src += stride;
2156
  }
2157
}
2158

    
2159
#define SRC(x,y) src[(x)+(y)*stride]
2160
#define PL(y) \
2161
    const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2162
#define PREDICT_8x8_LOAD_LEFT \
2163
    const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2164
                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2165
    PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2166
    const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2167

    
2168
#define PT(x) \
2169
    const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2170
#define PREDICT_8x8_LOAD_TOP \
2171
    const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2172
                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2173
    PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2174
    const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2175
                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2176

    
2177
#define PTR(x) \
2178
    t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2179
#define PREDICT_8x8_LOAD_TOPRIGHT \
2180
    int t8, t9, t10, t11, t12, t13, t14, t15; \
2181
    if(has_topright) { \
2182
        PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2183
        t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2184
    } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2185

    
2186
#define PREDICT_8x8_LOAD_TOPLEFT \
2187
    const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2188

    
2189
#define PREDICT_8x8_DC(v) \
2190
    int y; \
2191
    for( y = 0; y < 8; y++ ) { \
2192
        ((uint32_t*)src)[0] = \
2193
        ((uint32_t*)src)[1] = v; \
2194
        src += stride; \
2195
    }
2196

    
2197
static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2198
{
2199
    PREDICT_8x8_DC(0x80808080);
2200
}
2201
static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2202
{
2203
    PREDICT_8x8_LOAD_LEFT;
2204
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2205
    PREDICT_8x8_DC(dc);
2206
}
2207
static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2208
{
2209
    PREDICT_8x8_LOAD_TOP;
2210
    const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2211
    PREDICT_8x8_DC(dc);
2212
}
2213
static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2214
{
2215
    PREDICT_8x8_LOAD_LEFT;
2216
    PREDICT_8x8_LOAD_TOP;
2217
    const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2218
                         +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2219
    PREDICT_8x8_DC(dc);
2220
}
2221
static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2222
{
2223
    PREDICT_8x8_LOAD_LEFT;
2224
#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2225
               ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2226
    ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2227
#undef ROW
2228
}
2229
static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2230
{
2231
    int y;
2232
    PREDICT_8x8_LOAD_TOP;
2233
    src[0] = t0;
2234
    src[1] = t1;
2235
    src[2] = t2;
2236
    src[3] = t3;
2237
    src[4] = t4;
2238
    src[5] = t5;
2239
    src[6] = t6;
2240
    src[7] = t7;
2241
    for( y = 1; y < 8; y++ )
2242
        *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2243
}
2244
static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2245
{
2246
    PREDICT_8x8_LOAD_TOP;
2247
    PREDICT_8x8_LOAD_TOPRIGHT;
2248
    SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2249
    SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2250
    SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2251
    SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2252
    SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2253
    SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2254
    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;
2255
    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;
2256
    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;
2257
    SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2258
    SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2259
    SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2260
    SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2261
    SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2262
    SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2263
}
2264
static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2265
{
2266
    PREDICT_8x8_LOAD_TOP;
2267
    PREDICT_8x8_LOAD_LEFT;
2268
    PREDICT_8x8_LOAD_TOPLEFT;
2269
    SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2270
    SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2271
    SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2272
    SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2273
    SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2274
    SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2275
    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;
2276
    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;
2277
    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;
2278
    SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2279
    SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2280
    SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2281
    SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2282
    SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2283
    SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2284

    
2285
}
2286
static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2287
{
2288
    PREDICT_8x8_LOAD_TOP;
2289
    PREDICT_8x8_LOAD_LEFT;
2290
    PREDICT_8x8_LOAD_TOPLEFT;
2291
    SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2292
    SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2293
    SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2294
    SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2295
    SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2296
    SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2297
    SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2298
    SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2299
    SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2300
    SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2301
    SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2302
    SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2303
    SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2304
    SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2305
    SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2306
    SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2307
    SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2308
    SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2309
    SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2310
    SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2311
    SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2312
    SRC(7,0)= (t6 + t7 + 1) >> 1;
2313
}
2314
static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2315
{
2316
    PREDICT_8x8_LOAD_TOP;
2317
    PREDICT_8x8_LOAD_LEFT;
2318
    PREDICT_8x8_LOAD_TOPLEFT;
2319
    SRC(0,7)= (l6 + l7 + 1) >> 1;
2320
    SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2321
    SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2322
    SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2323
    SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2324
    SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2325
    SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2326
    SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2327
    SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2328
    SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2329
    SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2330
    SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2331
    SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2332
    SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2333
    SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2334
    SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2335
    SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2336
    SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2337
    SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2338
    SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2339
    SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2340
    SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2341
}
2342
static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2343
{
2344
    PREDICT_8x8_LOAD_TOP;
2345
    PREDICT_8x8_LOAD_TOPRIGHT;
2346
    SRC(0,0)= (t0 + t1 + 1) >> 1;
2347
    SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2348
    SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2349
    SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2350
    SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2351
    SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2352
    SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2353
    SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2354
    SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2355
    SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2356
    SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2357
    SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2358
    SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2359
    SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2360
    SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2361
    SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2362
    SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2363
    SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2364
    SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2365
    SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2366
    SRC(7,6)= (t10 + t11 + 1) >> 1;
2367
    SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2368
}
2369
static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2370
{
2371
    PREDICT_8x8_LOAD_LEFT;
2372
    SRC(0,0)= (l0 + l1 + 1) >> 1;
2373
    SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2374
    SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2375
    SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2376
    SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2377
    SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2378
    SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2379
    SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2380
    SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2381
    SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2382
    SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2383
    SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2384
    SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2385
    SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2386
    SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2387
    SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2388
    SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2389
    SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2390
}
2391
#undef PREDICT_8x8_LOAD_LEFT
2392
#undef PREDICT_8x8_LOAD_TOP
2393
#undef PREDICT_8x8_LOAD_TOPLEFT
2394
#undef PREDICT_8x8_LOAD_TOPRIGHT
2395
#undef PREDICT_8x8_DC
2396
#undef PTR
2397
#undef PT
2398
#undef PL
2399
#undef SRC
2400

    
2401
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2402
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2403
                           int src_x_offset, int src_y_offset,
2404
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2405
    MpegEncContext * const s = &h->s;
2406
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2407
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2408
    const int luma_xy= (mx&3) + ((my&3)<<2);
2409
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2410
    uint8_t * src_cb, * src_cr;
2411
    int extra_width= h->emu_edge_width;
2412
    int extra_height= h->emu_edge_height;
2413
    int emu=0;
2414
    const int full_mx= mx>>2;
2415
    const int full_my= my>>2;
2416
    const int pic_width  = 16*s->mb_width;
2417
    const int pic_height = 16*s->mb_height >> MB_MBAFF;
2418

    
2419
    if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
2420
        return;
2421

    
2422
    if(mx&7) extra_width -= 3;
2423
    if(my&7) extra_height -= 3;
2424

    
2425
    if(   full_mx < 0-extra_width
2426
       || full_my < 0-extra_height
2427
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
2428
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
2429
        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);
2430
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
2431
        emu=1;
2432
    }
2433

    
2434
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
2435
    if(!square){
2436
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
2437
    }
2438

    
2439
    if(s->flags&CODEC_FLAG_GRAY) return;
2440

    
2441
    if(MB_MBAFF){
2442
        // chroma offset when predicting from a field of opposite parity
2443
        my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2444
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2445
    }
2446
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2447
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2448

    
2449
    if(emu){
2450
        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);
2451
            src_cb= s->edge_emu_buffer;
2452
    }
2453
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2454

    
2455
    if(emu){
2456
        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);
2457
            src_cr= s->edge_emu_buffer;
2458
    }
2459
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
2460
}
2461

    
2462
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2463
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2464
                           int x_offset, int y_offset,
2465
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2466
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2467
                           int list0, int list1){
2468
    MpegEncContext * const s = &h->s;
2469
    qpel_mc_func *qpix_op=  qpix_put;
2470
    h264_chroma_mc_func chroma_op= chroma_put;
2471

    
2472
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2473
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2474
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2475
    x_offset += 8*s->mb_x;
2476
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2477

    
2478
    if(list0){
2479
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2480
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2481
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2482
                           qpix_op, chroma_op);
2483

    
2484
        qpix_op=  qpix_avg;
2485
        chroma_op= chroma_avg;
2486
    }
2487

    
2488
    if(list1){
2489
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2490
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2491
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2492
                           qpix_op, chroma_op);
2493
    }
2494
}
2495

    
2496
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2497
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2498
                           int x_offset, int y_offset,
2499
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2500
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2501
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2502
                           int list0, int list1){
2503
    MpegEncContext * const s = &h->s;
2504

    
2505
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
2506
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
2507
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
2508
    x_offset += 8*s->mb_x;
2509
    y_offset += 8*(s->mb_y >> MB_MBAFF);
2510

    
2511
    if(list0 && list1){
2512
        /* don't optimize for luma-only case, since B-frames usually
2513
         * use implicit weights => chroma too. */
2514
        uint8_t *tmp_cb = s->obmc_scratchpad;
2515
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2516
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
2517
        int refn0 = h->ref_cache[0][ scan8[n] ];
2518
        int refn1 = h->ref_cache[1][ scan8[n] ];
2519

    
2520
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2521
                    dest_y, dest_cb, dest_cr,
2522
                    x_offset, y_offset, qpix_put, chroma_put);
2523
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2524
                    tmp_y, tmp_cb, tmp_cr,
2525
                    x_offset, y_offset, qpix_put, chroma_put);
2526

    
2527
        if(h->use_weight == 2){
2528
            int weight0 = h->implicit_weight[refn0][refn1];
2529
            int weight1 = 64 - weight0;
2530
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
2531
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2532
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
2533
        }else{
2534
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
2535
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2536
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2537
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2538
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2539
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2540
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2541
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2542
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2543
        }
2544
    }else{
2545
        int list = list1 ? 1 : 0;
2546
        int refn = h->ref_cache[list][ scan8[n] ];
2547
        Picture *ref= &h->ref_list[list][refn];
2548
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2549
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2550
                    qpix_put, chroma_put);
2551

    
2552
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
2553
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2554
        if(h->use_weight_chroma){
2555
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2556
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2557
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
2558
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2559
        }
2560
    }
2561
}
2562

    
2563
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2564
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2565
                           int x_offset, int y_offset,
2566
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2567
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2568
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2569
                           int list0, int list1){
2570
    if((h->use_weight==2 && list0 && list1
2571
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2572
       || h->use_weight==1)
2573
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2574
                         x_offset, y_offset, qpix_put, chroma_put,
2575
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2576
    else
2577
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2578
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2579
}
2580

    
2581
static inline void prefetch_motion(H264Context *h, int list){
2582
    /* fetch pixels for estimated mv 4 macroblocks ahead
2583
     * optimized for 64byte cache lines */
2584
    MpegEncContext * const s = &h->s;
2585
    const int refn = h->ref_cache[list][scan8[0]];
2586
    if(refn >= 0){
2587
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2588
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2589
        uint8_t **src= h->ref_list[list][refn].data;
2590
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
2591
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
2592
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2593
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2594
    }
2595
}
2596

    
2597
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2598
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2599
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2600
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2601
    MpegEncContext * const s = &h->s;
2602
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2603
    const int mb_type= s->current_picture.mb_type[mb_xy];
2604

    
2605
    assert(IS_INTER(mb_type));
2606

    
2607
    prefetch_motion(h, 0);
2608

    
2609
    if(IS_16X16(mb_type)){
2610
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2611
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2612
                &weight_op[0], &weight_avg[0],
2613
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2614
    }else if(IS_16X8(mb_type)){
2615
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2616
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2617
                &weight_op[1], &weight_avg[1],
2618
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2619
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2620
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2621
                &weight_op[1], &weight_avg[1],
2622
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2623
    }else if(IS_8X16(mb_type)){
2624
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
2625
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2626
                &weight_op[2], &weight_avg[2],
2627
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2628
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
2629
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2630
                &weight_op[2], &weight_avg[2],
2631
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2632
    }else{
2633
        int i;
2634

    
2635
        assert(IS_8X8(mb_type));
2636

    
2637
        for(i=0; i<4; i++){
2638
            const int sub_mb_type= h->sub_mb_type[i];
2639
            const int n= 4*i;
2640
            int x_offset= (i&1)<<2;
2641
            int y_offset= (i&2)<<1;
2642

    
2643
            if(IS_SUB_8X8(sub_mb_type)){
2644
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2645
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2646
                    &weight_op[3], &weight_avg[3],
2647
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2648
            }else if(IS_SUB_8X4(sub_mb_type)){
2649
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2650
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2651
                    &weight_op[4], &weight_avg[4],
2652
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2653
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2654
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2655
                    &weight_op[4], &weight_avg[4],
2656
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2657
            }else if(IS_SUB_4X8(sub_mb_type)){
2658
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2659
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2660
                    &weight_op[5], &weight_avg[5],
2661
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2662
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2663
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2664
                    &weight_op[5], &weight_avg[5],
2665
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2666
            }else{
2667
                int j;
2668
                assert(IS_SUB_4X4(sub_mb_type));
2669
                for(j=0; j<4; j++){
2670
                    int sub_x_offset= x_offset + 2*(j&1);
2671
                    int sub_y_offset= y_offset +   (j&2);
2672
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2673
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2674
                        &weight_op[6], &weight_avg[6],
2675
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2676
                }
2677
            }
2678
        }
2679
    }
2680

    
2681
    prefetch_motion(h, 1);
2682
}
2683

    
2684
static void decode_init_vlc(void){
2685
    static int done = 0;
2686

    
2687
    if (!done) {
2688
        int i;
2689
        done = 1;
2690

    
2691
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2692
                 &chroma_dc_coeff_token_len [0], 1, 1,
2693
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2694

    
2695
        for(i=0; i<4; i++){
2696
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2697
                     &coeff_token_len [i][0], 1, 1,
2698
                     &coeff_token_bits[i][0], 1, 1, 1);
2699
        }
2700

    
2701
        for(i=0; i<3; i++){
2702
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2703
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2704
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2705
        }
2706
        for(i=0; i<15; i++){
2707
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2708
                     &total_zeros_len [i][0], 1, 1,
2709
                     &total_zeros_bits[i][0], 1, 1, 1);
2710
        }
2711

    
2712
        for(i=0; i<6; i++){
2713
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2714
                     &run_len [i][0], 1, 1,
2715
                     &run_bits[i][0], 1, 1, 1);
2716
        }
2717
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2718
                 &run_len [6][0], 1, 1,
2719
                 &run_bits[6][0], 1, 1, 1);
2720
    }
2721
}
2722

    
2723
/**
2724
 * Sets the intra prediction function pointers.
2725
 */
2726
static void init_pred_ptrs(H264Context *h){
2727
//    MpegEncContext * const s = &h->s;
2728

    
2729
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2730
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2731
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2732
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2733
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2734
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2735
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2736
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2737
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2738
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2739
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2740
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2741

    
2742
    h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;
2743
    h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;
2744
    h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;
2745
    h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2746
    h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2747
    h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;
2748
    h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;
2749
    h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;
2750
    h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;
2751
    h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;
2752
    h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;
2753
    h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;
2754

    
2755
    h->pred8x8[DC_PRED8x8     ]= ff_pred8x8_dc_c;
2756
    h->pred8x8[VERT_PRED8x8   ]= ff_pred8x8_vertical_c;
2757
    h->pred8x8[HOR_PRED8x8    ]= ff_pred8x8_horizontal_c;
2758
    h->pred8x8[PLANE_PRED8x8  ]= ff_pred8x8_plane_c;
2759
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2760
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2761
    h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c;
2762

    
2763
    h->pred16x16[DC_PRED8x8     ]= ff_pred16x16_dc_c;
2764
    h->pred16x16[VERT_PRED8x8   ]= ff_pred16x16_vertical_c;
2765
    h->pred16x16[HOR_PRED8x8    ]= ff_pred16x16_horizontal_c;
2766
    h->pred16x16[PLANE_PRED8x8  ]= ff_pred16x16_plane_c;
2767
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2768
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2769
    h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c;
2770
}
2771

    
2772
static void free_tables(H264Context *h){
2773
    av_freep(&h->intra4x4_pred_mode);
2774
    av_freep(&h->chroma_pred_mode_table);
2775
    av_freep(&h->cbp_table);
2776
    av_freep(&h->mvd_table[0]);
2777
    av_freep(&h->mvd_table[1]);
2778
    av_freep(&h->direct_table);
2779
    av_freep(&h->non_zero_count);
2780
    av_freep(&h->slice_table_base);
2781
    av_freep(&h->top_borders[1]);
2782
    av_freep(&h->top_borders[0]);
2783
    h->slice_table= NULL;
2784

    
2785
    av_freep(&h->mb2b_xy);
2786
    av_freep(&h->mb2b8_xy);
2787

    
2788
    av_freep(&h->s.obmc_scratchpad);
2789
}
2790

    
2791
static void init_dequant8_coeff_table(H264Context *h){
2792
    int i,q,x;
2793
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2794
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2795
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2796

    
2797
    for(i=0; i<2; i++ ){
2798
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2799
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2800
            break;
2801
        }
2802

    
2803
        for(q=0; q<52; q++){
2804
            int shift = ff_div6[q];
2805
            int idx = ff_rem6[q];
2806
            for(x=0; x<64; x++)
2807
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2808
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2809
                    h->pps.scaling_matrix8[i][x]) << shift;
2810
        }
2811
    }
2812
}
2813

    
2814
static void init_dequant4_coeff_table(H264Context *h){
2815
    int i,j,q,x;
2816
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2817
    for(i=0; i<6; i++ ){
2818
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2819
        for(j=0; j<i; j++){
2820
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2821
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2822
                break;
2823
            }
2824
        }
2825
        if(j<i)
2826
            continue;
2827

    
2828
        for(q=0; q<52; q++){
2829
            int shift = ff_div6[q] + 2;
2830
            int idx = ff_rem6[q];
2831
            for(x=0; x<16; x++)
2832
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2833
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2834
                    h->pps.scaling_matrix4[i][x]) << shift;
2835
        }
2836
    }
2837
}
2838

    
2839
static void init_dequant_tables(H264Context *h){
2840
    int i,x;
2841
    init_dequant4_coeff_table(h);
2842
    if(h->pps.transform_8x8_mode)
2843
        init_dequant8_coeff_table(h);
2844
    if(h->sps.transform_bypass){
2845
        for(i=0; i<6; i++)
2846
            for(x=0; x<16; x++)
2847
                h->dequant4_coeff[i][0][x] = 1<<6;
2848
        if(h->pps.transform_8x8_mode)
2849
            for(i=0; i<2; i++)
2850
                for(x=0; x<64; x++)
2851
                    h->dequant8_coeff[i][0][x] = 1<<6;
2852
    }
2853
}
2854

    
2855

    
2856
/**
2857
 * allocates tables.
2858
 * needs width/height
2859
 */
2860
static int alloc_tables(H264Context *h){
2861
    MpegEncContext * const s = &h->s;
2862
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2863
    int x,y;
2864

    
2865
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2866

    
2867
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2868
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
2869
    CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2870
    CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))
2871
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2872

    
2873
    if( h->pps.cabac ) {
2874
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2875
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2876
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2877
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2878
    }
2879

    
2880
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));
2881
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2882

    
2883
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2884
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2885
    for(y=0; y<s->mb_height; y++){
2886
        for(x=0; x<s->mb_width; x++){
2887
            const int mb_xy= x + y*s->mb_stride;
2888
            const int b_xy = 4*x + 4*y*h->b_stride;
2889
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2890

    
2891
            h->mb2b_xy [mb_xy]= b_xy;
2892
            h->mb2b8_xy[mb_xy]= b8_xy;
2893
        }
2894
    }
2895

    
2896
    s->obmc_scratchpad = NULL;
2897

    
2898
    if(!h->dequant4_coeff[0])
2899
        init_dequant_tables(h);
2900

    
2901
    return 0;
2902
fail:
2903
    free_tables(h);
2904
    return -1;
2905
}
2906

    
2907
static void common_init(H264Context *h){
2908
    MpegEncContext * const s = &h->s;
2909

    
2910
    s->width = s->avctx->width;
2911
    s->height = s->avctx->height;
2912
    s->codec_id= s->avctx->codec->id;
2913

    
2914
    init_pred_ptrs(h);
2915

    
2916
    h->dequant_coeff_pps= -1;
2917
    s->unrestricted_mv=1;
2918
    s->decode=1; //FIXME
2919

    
2920
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2921
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2922
}
2923

    
2924
static int decode_init(AVCodecContext *avctx){
2925
    H264Context *h= avctx->priv_data;
2926
    MpegEncContext * const s = &h->s;
2927

    
2928
    MPV_decode_defaults(s);
2929

    
2930
    s->avctx = avctx;
2931
    common_init(h);
2932

    
2933
    s->out_format = FMT_H264;
2934
    s->workaround_bugs= avctx->workaround_bugs;
2935

    
2936
    // set defaults
2937
//    s->decode_mb= ff_h263_decode_mb;
2938
    s->low_delay= 1;
2939
    avctx->pix_fmt= PIX_FMT_YUV420P;
2940

    
2941
    decode_init_vlc();
2942

    
2943
    if(avctx->extradata_size > 0 && avctx->extradata &&
2944
       *(char *)avctx->extradata == 1){
2945
        h->is_avc = 1;
2946
        h->got_avcC = 0;
2947
    } else {
2948
        h->is_avc = 0;
2949
    }
2950

    
2951
    return 0;
2952
}
2953

    
2954
static int frame_start(H264Context *h){
2955
    MpegEncContext * const s = &h->s;
2956
    int i;
2957

    
2958
    if(MPV_frame_start(s, s->avctx) < 0)
2959
        return -1;
2960
    ff_er_frame_start(s);
2961

    
2962
    assert(s->linesize && s->uvlinesize);
2963

    
2964
    for(i=0; i<16; i++){
2965
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2966
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2967
    }
2968
    for(i=0; i<4; i++){
2969
        h->block_offset[16+i]=
2970
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2971
        h->block_offset[24+16+i]=
2972
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2973
    }
2974

    
2975
    /* can't be in alloc_tables because linesize isn't known there.
2976
     * FIXME: redo bipred weight to not require extra buffer? */
2977
    if(!s->obmc_scratchpad)
2978
        s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2979

    
2980
    /* some macroblocks will be accessed before they're available */
2981
    if(FRAME_MBAFF)
2982
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
2983

    
2984
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2985
    return 0;
2986
}
2987

    
2988
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2989
    MpegEncContext * const s = &h->s;
2990
    int i;
2991

    
2992
    src_y  -=   linesize;
2993
    src_cb -= uvlinesize;
2994
    src_cr -= uvlinesize;
2995

    
2996
    // There are two lines saved, the line above the the top macroblock of a pair,
2997
    // and the line above the bottom macroblock
2998
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
2999
    for(i=1; i<17; i++){
3000
        h->left_border[i]= src_y[15+i*  linesize];
3001
    }
3002

    
3003
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
3004
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3005

    
3006
    if(!(s->flags&CODEC_FLAG_GRAY)){
3007
        h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];
3008
        h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3009
        for(i=1; i<9; i++){
3010
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
3011
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3012
        }
3013
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3014
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3015
    }
3016
}
3017

    
3018
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){
3019
    MpegEncContext * const s = &h->s;
3020
    int temp8, i;
3021
    uint64_t temp64;
3022
    int deblock_left = (s->mb_x > 0);
3023
    int deblock_top  = (s->mb_y > 0);
3024

    
3025
    src_y  -=   linesize + 1;
3026
    src_cb -= uvlinesize + 1;
3027
    src_cr -= uvlinesize + 1;
3028

    
3029
#define XCHG(a,b,t,xchg)\
3030
t= a;\
3031
if(xchg)\
3032
    a= b;\
3033
b= t;
3034

    
3035
    if(deblock_left){
3036
        for(i = !deblock_top; i<17; i++){
3037
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3038
        }
3039
    }
3040

    
3041
    if(deblock_top){
3042
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3043
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3044
        if(s->mb_x+1 < s->mb_width){
3045
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3046
        }
3047
    }
3048

    
3049
    if(!(s->flags&CODEC_FLAG_GRAY)){
3050
        if(deblock_left){
3051
            for(i = !deblock_top; i<9; i++){
3052
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
3053
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3054
            }
3055
        }
3056
        if(deblock_top){
3057
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3058
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3059
        }
3060
    }
3061
}
3062

    
3063
static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3064
    MpegEncContext * const s = &h->s;
3065
    int i;
3066

    
3067
    src_y  -= 2 *   linesize;
3068
    src_cb -= 2 * uvlinesize;
3069
    src_cr -= 2 * uvlinesize;
3070

    
3071
    // There are two lines saved, the line above the the top macroblock of a pair,
3072
    // and the line above the bottom macroblock
3073
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
3074
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
3075
    for(i=2; i<34; i++){
3076
        h->left_border[i]= src_y[15+i*  linesize];
3077
    }
3078

    
3079
    *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);
3080
    *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3081
    *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);
3082
    *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3083

    
3084
    if(!(s->flags&CODEC_FLAG_GRAY)){
3085
        h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];
3086
        h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];
3087
        h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];
3088
        h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3089
        for(i=2; i<18; i++){
3090
            h->left_border[i+34   ]= src_cb[7+i*uvlinesize];
3091
            h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3092
        }
3093
        *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3094
        *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3095
        *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3096
        *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3097
    }
3098
}
3099

    
3100
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){
3101
    MpegEncContext * const s = &h->s;
3102
    int temp8, i;
3103
    uint64_t temp64;
3104
    int deblock_left = (s->mb_x > 0);
3105
    int deblock_top  = (s->mb_y > 1);
3106

    
3107
    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);
3108

    
3109
    src_y  -= 2 *   linesize + 1;
3110
    src_cb -= 2 * uvlinesize + 1;
3111
    src_cr -= 2 * uvlinesize + 1;
3112

    
3113
#define XCHG(a,b,t,xchg)\
3114
t= a;\
3115
if(xchg)\
3116
    a= b;\
3117
b= t;
3118

    
3119
    if(deblock_left){
3120
        for(i = (!deblock_top)<<1; i<34; i++){
3121
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
3122
        }
3123
    }
3124

    
3125
    if(deblock_top){
3126
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3127
        XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3128
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3129
        XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3130
        if(s->mb_x+1 < s->mb_width){
3131
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3132
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
3133
        }
3134
    }
3135

    
3136
    if(!(s->flags&CODEC_FLAG_GRAY)){
3137
        if(deblock_left){
3138
            for(i = (!deblock_top) << 1; i<18; i++){
3139
                XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);
3140
                XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3141
            }
3142
        }
3143
        if(deblock_top){
3144
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3145
            XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3146
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3147
            XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3148
        }
3149
    }
3150
}
3151

    
3152
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
3153
    MpegEncContext * const s = &h->s;
3154
    const int mb_x= s->mb_x;
3155
    const int mb_y= s->mb_y;
3156
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3157
    const int mb_type= s->current_picture.mb_type[mb_xy];
3158
    uint8_t  *dest_y, *dest_cb, *dest_cr;
3159
    int linesize, uvlinesize /*dct_offset*/;
3160
    int i;
3161
    int *block_offset = &h->block_offset[0];
3162
    const unsigned int bottom = mb_y & 1;
3163
    const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
3164
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3165
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
3166

    
3167
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3168
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3169
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3170

    
3171
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
3172
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
3173

    
3174
    if (!simple && MB_FIELD) {
3175
        linesize   = h->mb_linesize   = s->linesize * 2;
3176
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3177
        block_offset = &h->block_offset[24];
3178
        if(mb_y&1){ //FIXME move out of this func?
3179
            dest_y -= s->linesize*15;
3180
            dest_cb-= s->uvlinesize*7;
3181
            dest_cr-= s->uvlinesize*7;
3182
        }
3183
        if(FRAME_MBAFF) {
3184
            int list;
3185
            for(list=0; list<h->list_count; list++){
3186
                if(!USES_LIST(mb_type, list))
3187
                    continue;
3188
                if(IS_16X16(mb_type)){
3189
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
3190
                    fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);
3191
                }else{
3192
                    for(i=0; i<16; i+=4){
3193
                        //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
3194
                        int ref = h->ref_cache[list][scan8[i]];
3195
                        if(ref >= 0)
3196
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);
3197
                    }
3198
                }
3199
            }
3200
        }
3201
    } else {
3202
        linesize   = h->mb_linesize   = s->linesize;
3203
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3204
//        dct_offset = s->linesize * 16;
3205
    }
3206

    
3207
    if(transform_bypass){
3208
        idct_dc_add =
3209
        idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3210
    }else if(IS_8x8DCT(mb_type)){
3211
        idct_dc_add = s->dsp.h264_idct8_dc_add;
3212
        idct_add = s->dsp.h264_idct8_add;
3213
    }else{
3214
        idct_dc_add = s->dsp.h264_idct_dc_add;
3215
        idct_add = s->dsp.h264_idct_add;
3216
    }
3217

    
3218
    if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
3219
       && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
3220
        int mbt_y = mb_y&~1;
3221
        uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;
3222
        uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3223
        uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
3224
        xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
3225
    }
3226

    
3227
    if (!simple && IS_INTRA_PCM(mb_type)) {
3228
        unsigned int x, y;
3229

    
3230
        // The pixels are stored in h->mb array in the same order as levels,
3231
        // copy them in output in the correct order.
3232
        for(i=0; i<16; i++) {
3233
            for (y=0; y<4; y++) {
3234
                for (x=0; x<4; x++) {
3235
                    *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3236
                }
3237
            }
3238
        }
3239
        for(i=16; i<16+4; i++) {
3240
            for (y=0; y<4; y++) {
3241
                for (x=0; x<4; x++) {
3242
                    *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3243
                }
3244
            }
3245
        }
3246
        for(i=20; i<20+4; i++) {
3247
            for (y=0; y<4; y++) {
3248
                for (x=0; x<4; x++) {
3249
                    *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3250
                }
3251
            }
3252
        }
3253
    } else {
3254
        if(IS_INTRA(mb_type)){
3255
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3256
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3257

    
3258
            if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3259
                h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3260
                h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3261
            }
3262

    
3263
            if(IS_INTRA4x4(mb_type)){
3264
                if(simple || !s->encoding){
3265
                    if(IS_8x8DCT(mb_type)){
3266
                        for(i=0; i<16; i+=4){
3267
                            uint8_t * const ptr= dest_y + block_offset[i];
3268
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3269
                            const int nnz = h->non_zero_count_cache[ scan8[i] ];
3270
                            h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3271
                                                   (h->topright_samples_available<<i)&0x4000, linesize);
3272
                            if(nnz){
3273
                                if(nnz == 1 && h->mb[i*16])
3274
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3275
                                else
3276
                                    idct_add(ptr, h->mb + i*16, linesize);
3277
                            }
3278
                        }
3279
                    }else
3280
                    for(i=0; i<16; i++){
3281
                        uint8_t * const ptr= dest_y + block_offset[i];
3282
                        uint8_t *topright;
3283
                        const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3284
                        int nnz, tr;
3285

    
3286
                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3287
                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3288
                            assert(mb_y || linesize <= block_offset[i]);
3289
                            if(!topright_avail){
3290
                                tr= ptr[3 - linesize]*0x01010101;
3291
                                topright= (uint8_t*) &tr;
3292
                            }else
3293
                                topright= ptr + 4 - linesize;
3294
                        }else
3295
                            topright= NULL;
3296

    
3297
                        h->pred4x4[ dir ](ptr, topright, linesize);
3298
                        nnz = h->non_zero_count_cache[ scan8[i] ];
3299
                        if(nnz){
3300
                            if(is_h264){
3301
                                if(nnz == 1 && h->mb[i*16])
3302
                                    idct_dc_add(ptr, h->mb + i*16, linesize);
3303
                                else
3304
                                    idct_add(ptr, h->mb + i*16, linesize);
3305
                            }else
3306
                                svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3307
                        }
3308
                    }
3309
                }
3310
            }else{
3311
                h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3312
                if(is_h264){
3313
                    if(!transform_bypass)
3314
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
3315
                }else
3316
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3317
            }
3318
            if(h->deblocking_filter && (simple || !FRAME_MBAFF))
3319
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3320
        }else if(is_h264){
3321
            hl_motion(h, dest_y, dest_cb, dest_cr,
3322
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
3323
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
3324
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3325
        }
3326

    
3327

    
3328
        if(!IS_INTRA4x4(mb_type)){
3329
            if(is_h264){
3330
                if(IS_INTRA16x16(mb_type)){
3331
                    for(i=0; i<16; i++){
3332
                        if(h->non_zero_count_cache[ scan8[i] ])
3333
                            idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3334
                        else if(h->mb[i*16])
3335
                            idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3336
                    }
3337
                }else{
3338
                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3339
                    for(i=0; i<16; i+=di){
3340
                        int nnz = h->non_zero_count_cache[ scan8[i] ];
3341
                        if(nnz){
3342
                            if(nnz==1 && h->mb[i*16])
3343
                                idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3344
                            else
3345
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3346
                        }
3347
                    }
3348
                }
3349
            }else{
3350
                for(i=0; i<16; i++){
3351
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3352
                        uint8_t * const ptr= dest_y + block_offset[i];
3353
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3354
                    }
3355
                }
3356
            }
3357
        }
3358

    
3359
        if(simple || !(s->flags&CODEC_FLAG_GRAY)){
3360
            uint8_t *dest[2] = {dest_cb, dest_cr};
3361
            if(transform_bypass){
3362
                idct_add = idct_dc_add = s->dsp.add_pixels4;
3363
            }else{
3364
                idct_add = s->dsp.h264_idct_add;
3365
                idct_dc_add = s->dsp.h264_idct_dc_add;
3366
                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]);
3367
                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]);
3368
            }
3369
            if(is_h264){
3370
                for(i=16; i<16+8; i++){
3371
                    if(h->non_zero_count_cache[ scan8[i] ])
3372
                        idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3373
                    else if(h->mb[i*16])
3374
                        idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
3375
                }
3376
            }else{
3377
                for(i=16; i<16+8; i++){
3378
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3379
                        uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
3380
                        svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3381
                    }
3382
                }
3383
            }
3384
        }
3385
    }
3386
    if(h->deblocking_filter) {
3387
        if (!simple && FRAME_MBAFF) {
3388
            //FIXME try deblocking one mb at a time?
3389
            // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
3390
            const int mb_y = s->mb_y - 1;
3391
            uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3392
            const int mb_xy= mb_x + mb_y*s->mb_stride;
3393
            const int mb_type_top   = s->current_picture.mb_type[mb_xy];
3394
            const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3395
            if (!bottom) return;
3396
            pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
3397
            pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3398
            pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3399

    
3400
            if(IS_INTRA(mb_type_top | mb_type_bottom))
3401
                xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3402

    
3403
            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3404
            // deblock a pair
3405
            // top
3406
            s->mb_y--;
3407
            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);
3408
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3409
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);
3410
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3411
            // bottom
3412
            s->mb_y++;
3413
            tprintf(h->s.avctx, "call mbaff filter_mb\n");
3414
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3415
            h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
3416
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3417
        } else {
3418
            tprintf(h->s.avctx, "call filter_mb\n");
3419
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3420
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3421
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3422
        }
3423
    }
3424
}
3425

    
3426
/**
3427
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
3428
 */
3429
static void hl_decode_mb_simple(H264Context *h){
3430
    hl_decode_mb_internal(h, 1);
3431
}
3432

    
3433
/**
3434
 * Process a macroblock; this handles edge cases, such as interlacing.
3435
 */
3436
static void av_noinline hl_decode_mb_complex(H264Context *h){
3437
    hl_decode_mb_internal(h, 0);
3438
}
3439

    
3440
static void hl_decode_mb(H264Context *h){
3441
    MpegEncContext * const s = &h->s;
3442
    const int mb_x= s->mb_x;
3443
    const int mb_y= s->mb_y;
3444
    const int mb_xy= mb_x + mb_y*s->mb_stride;
3445
    const int mb_type= s->current_picture.mb_type[mb_xy];
3446
    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;
3447

    
3448
    if(!s->decode)
3449
        return;
3450

    
3451
    if (is_complex)
3452
        hl_decode_mb_complex(h);
3453
    else hl_decode_mb_simple(h);
3454
}
3455

    
3456
/**
3457
 * fills the default_ref_list.
3458
 */
3459
static int fill_default_ref_list(H264Context *h){
3460
    MpegEncContext * const s = &h->s;
3461
    int i;
3462
    int smallest_poc_greater_than_current = -1;
3463
    Picture sorted_short_ref[32];
3464

    
3465
    if(h->slice_type==B_TYPE){
3466
        int out_i;
3467
        int limit= INT_MIN;
3468

    
3469
        /* sort frame according to poc in B slice */
3470
        for(out_i=0; out_i<h->short_ref_count; out_i++){
3471
            int best_i=INT_MIN;
3472
            int best_poc=INT_MAX;
3473

    
3474
            for(i=0; i<h->short_ref_count; i++){
3475
                const int poc= h->short_ref[i]->poc;
3476
                if(poc > limit && poc < best_poc){
3477
                    best_poc= poc;
3478
                    best_i= i;
3479
                }
3480
            }
3481

    
3482
            assert(best_i != INT_MIN);
3483

    
3484
            limit= best_poc;
3485
            sorted_short_ref[out_i]= *h->short_ref[best_i];
3486
            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);
3487
            if (-1 == smallest_poc_greater_than_current) {
3488
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3489
                    smallest_poc_greater_than_current = out_i;
3490
                }
3491
            }
3492
        }
3493
    }
3494

    
3495
    if(s->picture_structure == PICT_FRAME){
3496
        if(h->slice_type==B_TYPE){
3497
            int list;
3498
            tprintf(h->s.avctx, "current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3499

    
3500
            // find the largest poc
3501
            for(list=0; list<2; list++){
3502
                int index = 0;
3503
                int j= -99;
3504
                int step= list ? -1 : 1;
3505

    
3506
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3507
                    while(j<0 || j>= h->short_ref_count){
3508
                        if(j != -99 && step == (list ? -1 : 1))
3509
                            return -1;
3510
                        step = -step;
3511
                        j= smallest_poc_greater_than_current + (step>>1);
3512
                    }
3513
                    if(sorted_short_ref[j].reference != 3) continue;
3514
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
3515
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3516
                }
3517

    
3518
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3519
                    if(h->long_ref[i] == NULL) continue;
3520
                    if(h->long_ref[i]->reference != 3) continue;
3521

    
3522
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
3523
                    h->default_ref_list[ list ][index++].pic_id= i;;
3524
                }
3525

    
3526
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3527
                    // swap the two first elements of L1 when
3528
                    // L0 and L1 are identical
3529
                    Picture temp= h->default_ref_list[1][0];
3530
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
3531
                    h->default_ref_list[1][1] = temp;
3532
                }
3533

    
3534
                if(index < h->ref_count[ list ])
3535
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3536
            }
3537
        }else{
3538
            int index=0;
3539
            for(i=0; i<h->short_ref_count; i++){
3540
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3541
                h->default_ref_list[0][index  ]= *h->short_ref[i];
3542
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3543
            }
3544
            for(i = 0; i < 16; i++){
3545
                if(h->long_ref[i] == NULL) continue;
3546
                if(h->long_ref[i]->reference != 3) continue;
3547
                h->default_ref_list[0][index  ]= *h->long_ref[i];
3548
                h->default_ref_list[0][index++].pic_id= i;;
3549
            }
3550
            if(index < h->ref_count[0])
3551
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3552
        }
3553
    }else{ //FIELD
3554
        if(h->slice_type==B_TYPE){
3555
        }else{
3556
            //FIXME second field balh
3557
        }
3558
    }
3559
#ifdef TRACE
3560
    for (i=0; i<h->ref_count[0]; i++) {
3561
        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]);
3562
    }
3563
    if(h->slice_type==B_TYPE){
3564
        for (i=0; i<h->ref_count[1]; i++) {
3565
            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]);
3566
        }
3567
    }
3568
#endif
3569
    return 0;
3570
}
3571

    
3572
static void print_short_term(H264Context *h);
3573
static void print_long_term(H264Context *h);
3574

    
3575
static int decode_ref_pic_list_reordering(H264Context *h){
3576
    MpegEncContext * const s = &h->s;
3577
    int list, index;
3578

    
3579
    print_short_term(h);
3580
    print_long_term(h);
3581
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3582

    
3583
    for(list=0; list<h->list_count; list++){
3584
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3585

    
3586
        if(get_bits1(&s->gb)){
3587
            int pred= h->curr_pic_num;
3588

    
3589
            for(index=0; ; index++){
3590
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3591
                unsigned int pic_id;
3592
                int i;
3593
                Picture *ref = NULL;
3594

    
3595
                if(reordering_of_pic_nums_idc==3)
3596
                    break;
3597

    
3598
                if(index >= h->ref_count[list]){
3599
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3600
                    return -1;
3601
                }
3602

    
3603
                if(reordering_of_pic_nums_idc<3){
3604
                    if(reordering_of_pic_nums_idc<2){
3605
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3606

    
3607
                        if(abs_diff_pic_num >= h->max_pic_num){
3608
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3609
                            return -1;
3610
                        }
3611

    
3612
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3613
                        else                                pred+= abs_diff_pic_num;
3614
                        pred &= h->max_pic_num - 1;
3615

    
3616
                        for(i= h->short_ref_count-1; i>=0; i--){
3617
                            ref = h->short_ref[i];
3618
                            assert(ref->reference == 3);
3619
                            assert(!ref->long_ref);
3620
                            if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3621
                                break;
3622
                        }
3623
                        if(i>=0)
3624
                            ref->pic_id= ref->frame_num;
3625
                    }else{
3626
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3627
                        if(pic_id>31){
3628
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
3629
                            return -1;
3630
                        }
3631
                        ref = h->long_ref[pic_id];
3632
                        if(ref){
3633
                            ref->pic_id= pic_id;
3634
                            assert(ref->reference == 3);
3635
                            assert(ref->long_ref);
3636
                            i=0;
3637
                        }else{
3638
                            i=-1;
3639
                        }
3640
                    }
3641

    
3642
                    if (i < 0) {
3643
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3644
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3645
                    } else {
3646
                        for(i=index; i+1<h->ref_count[list]; i++){
3647
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3648
                                break;
3649
                        }
3650
                        for(; i > index; i--){
3651
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3652
                        }
3653
                        h->ref_list[list][index]= *ref;
3654
                    }
3655
                }else{
3656
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3657
                    return -1;
3658
                }
3659
            }
3660
        }
3661
    }
3662
    for(list=0; list<h->list_count; list++){
3663
        for(index= 0; index < h->ref_count[list]; index++){
3664
            if(!h->ref_list[list][index].data[0])
3665
                h->ref_list[list][index]= s->current_picture;
3666
        }
3667
    }
3668

    
3669
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3670
        direct_dist_scale_factor(h);
3671
    direct_ref_list_init(h);
3672
    return 0;
3673
}
3674

    
3675
static void fill_mbaff_ref_list(H264Context *h){
3676
    int list, i, j;
3677
    for(list=0; list<2; list++){ //FIXME try list_count
3678
        for(i=0; i<h->ref_count[list]; i++){
3679
            Picture *frame = &h->ref_list[list][i];
3680
            Picture *field = &h->ref_list[list][16+2*i];
3681
            field[0] = *frame;
3682
            for(j=0; j<3; j++)
3683
                field[0].linesize[j] <<= 1;
3684
            field[1] = field[0];
3685
            for(j=0; j<3; j++)
3686
                field[1].data[j] += frame->linesize[j];
3687

    
3688
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
3689
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
3690
            for(j=0; j<2; j++){
3691
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
3692
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
3693
            }
3694
        }
3695
    }
3696
    for(j=0; j<h->ref_count[1]; j++){
3697
        for(i=0; i<h->ref_count[0]; i++)
3698
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
3699
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
3700
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
3701
    }
3702
}
3703

    
3704
static int pred_weight_table(H264Context *h){
3705
    MpegEncContext * const s = &h->s;
3706
    int list, i;
3707
    int luma_def, chroma_def;
3708

    
3709
    h->use_weight= 0;
3710
    h->use_weight_chroma= 0;
3711
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3712
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3713
    luma_def = 1<<h->luma_log2_weight_denom;
3714
    chroma_def = 1<<h->chroma_log2_weight_denom;
3715

    
3716
    for(list=0; list<2; list++){
3717
        for(i=0; i<h->ref_count[list]; i++){
3718
            int luma_weight_flag, chroma_weight_flag;
3719

    
3720
            luma_weight_flag= get_bits1(&s->gb);
3721
            if(luma_weight_flag){
3722
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3723
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3724
                if(   h->luma_weight[list][i] != luma_def
3725
                   || h->luma_offset[list][i] != 0)
3726
                    h->use_weight= 1;
3727
            }else{
3728
                h->luma_weight[list][i]= luma_def;
3729
                h->luma_offset[list][i]= 0;
3730
            }
3731

    
3732
            chroma_weight_flag= get_bits1(&s->gb);
3733
            if(chroma_weight_flag){
3734
                int j;
3735
                for(j=0; j<2; j++){
3736
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3737
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3738
                    if(   h->chroma_weight[list][i][j] != chroma_def
3739
                       || h->chroma_offset[list][i][j] != 0)
3740
                        h->use_weight_chroma= 1;
3741
                }
3742
            }else{
3743
                int j;
3744
                for(j=0; j<2; j++){
3745
                    h->chroma_weight[list][i][j]= chroma_def;
3746
                    h->chroma_offset[list][i][j]= 0;
3747
                }
3748
            }
3749
        }
3750
        if(h->slice_type != B_TYPE) break;
3751
    }
3752
    h->use_weight= h->use_weight || h->use_weight_chroma;
3753
    return 0;
3754
}
3755

    
3756
static void implicit_weight_table(H264Context *h){
3757
    MpegEncContext * const s = &h->s;
3758
    int ref0, ref1;
3759
    int cur_poc = s->current_picture_ptr->poc;
3760

    
3761
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3762
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3763
        h->use_weight= 0;
3764
        h->use_weight_chroma= 0;
3765
        return;
3766
    }
3767

    
3768
    h->use_weight= 2;
3769
    h->use_weight_chroma= 2;
3770
    h->luma_log2_weight_denom= 5;
3771
    h->chroma_log2_weight_denom= 5;
3772

    
3773
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3774
        int poc0 = h->ref_list[0][ref0].poc;
3775
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3776
            int poc1 = h->ref_list[1][ref1].poc;
3777
            int td = av_clip(poc1 - poc0, -128, 127);
3778
            if(td){
3779
                int tb = av_clip(cur_poc - poc0, -128, 127);
3780
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3781
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3782
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3783
                    h->implicit_weight[ref0][ref1] = 32;
3784
                else
3785
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3786
            }else
3787
                h->implicit_weight[ref0][ref1] = 32;
3788
        }
3789
    }
3790
}
3791

    
3792
static inline void unreference_pic(H264Context *h, Picture *pic){
3793
    int i;
3794
    pic->reference=0;
3795
    if(pic == h->delayed_output_pic)
3796
        pic->reference=1;
3797
    else{
3798
        for(i = 0; h->delayed_pic[i]; i++)
3799
            if(pic == h->delayed_pic[i]){
3800
                pic->reference=1;
3801
                break;
3802
            }
3803
    }
3804
}
3805

    
3806
/**
3807
 * instantaneous decoder refresh.
3808
 */
3809
static void idr(H264Context *h){
3810
    int i;
3811

    
3812
    for(i=0; i<16; i++){
3813
        if (h->long_ref[i] != NULL) {
3814
            unreference_pic(h, h->long_ref[i]);
3815
            h->long_ref[i]= NULL;
3816
        }
3817
    }
3818
    h->long_ref_count=0;
3819

    
3820
    for(i=0; i<h->short_ref_count; i++){
3821
        unreference_pic(h, h->short_ref[i]);
3822
        h->short_ref[i]= NULL;
3823
    }
3824
    h->short_ref_count=0;
3825
}
3826

    
3827
/* forget old pics after a seek */
3828
static void flush_dpb(AVCodecContext *avctx){
3829
    H264Context *h= avctx->priv_data;
3830
    int i;
3831
    for(i=0; i<16; i++) {
3832
        if(h->delayed_pic[i])
3833
            h->delayed_pic[i]->reference= 0;
3834
        h->delayed_pic[i]= NULL;
3835
    }
3836
    if(h->delayed_output_pic)
3837
        h->delayed_output_pic->reference= 0;
3838
    h->delayed_output_pic= NULL;
3839
    idr(h);
3840
    if(h->s.current_picture_ptr)
3841
        h->s.current_picture_ptr->reference= 0;
3842
}
3843

    
3844
/**
3845
 *
3846
 * @return the removed picture or NULL if an error occurs
3847
 */
3848
static Picture * remove_short(H264Context *h, int frame_num){
3849
    MpegEncContext * const s = &h->s;
3850
    int i;
3851

    
3852
    if(s->avctx->debug&FF_DEBUG_MMCO)
3853
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3854

    
3855
    for(i=0; i<h->short_ref_count; i++){
3856
        Picture *pic= h->short_ref[i];
3857
        if(s->avctx->debug&FF_DEBUG_MMCO)
3858
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3859
        if(pic->frame_num == frame_num){
3860
            h->short_ref[i]= NULL;
3861
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3862
            h->short_ref_count--;
3863
            return pic;
3864
        }
3865
    }
3866
    return NULL;
3867
}
3868

    
3869
/**
3870
 *
3871
 * @return the removed picture or NULL if an error occurs
3872
 */
3873
static Picture * remove_long(H264Context *h, int i){
3874
    Picture *pic;
3875

    
3876
    pic= h->long_ref[i];
3877
    h->long_ref[i]= NULL;
3878
    if(pic) h->long_ref_count--;
3879

    
3880
    return pic;
3881
}
3882

    
3883
/**
3884
 * print short term list
3885
 */
3886
static void print_short_term(H264Context *h) {
3887
    uint32_t i;
3888
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3889
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3890
        for(i=0; i<h->short_ref_count; i++){
3891
            Picture *pic= h->short_ref[i];
3892
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3893
        }
3894
    }
3895
}
3896

    
3897
/**
3898
 * print long term list
3899
 */
3900
static void print_long_term(H264Context *h) {
3901
    uint32_t i;
3902
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3903
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3904
        for(i = 0; i < 16; i++){
3905
            Picture *pic= h->long_ref[i];
3906
            if (pic) {
3907
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3908
            }
3909
        }
3910
    }
3911
}
3912

    
3913
/**
3914
 * Executes the reference picture marking (memory management control operations).
3915
 */
3916
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3917
    MpegEncContext * const s = &h->s;
3918
    int i, j;
3919
    int current_is_long=0;
3920
    Picture *pic;
3921

    
3922
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3923
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3924

    
3925
    for(i=0; i<mmco_count; i++){
3926
        if(s->avctx->debug&FF_DEBUG_MMCO)
3927
            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);
3928

    
3929
        switch(mmco[i].opcode){
3930
        case MMCO_SHORT2UNUSED:
3931
            pic= remove_short(h, mmco[i].short_frame_num);
3932
            if(pic)
3933
                unreference_pic(h, pic);
3934
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3935
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");
3936
            break;
3937
        case MMCO_SHORT2LONG:
3938
            pic= remove_long(h, mmco[i].long_index);
3939
            if(pic) unreference_pic(h, pic);
3940

    
3941
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3942
            if (h->long_ref[ mmco[i].long_index ]){
3943
                h->long_ref[ mmco[i].long_index ]->long_ref=1;
3944
                h->long_ref_count++;
3945
            }
3946
            break;
3947
        case MMCO_LONG2UNUSED:
3948
            pic= remove_long(h, mmco[i].long_index);
3949
            if(pic)
3950
                unreference_pic(h, pic);
3951
            else if(s->avctx->debug&FF_DEBUG_MMCO)
3952
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");
3953
            break;
3954
        case MMCO_LONG:
3955
            pic= remove_long(h, mmco[i].long_index);
3956
            if(pic) unreference_pic(h, pic);
3957

    
3958
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3959
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3960
            h->long_ref_count++;
3961

    
3962
            current_is_long=1;
3963
            break;
3964
        case MMCO_SET_MAX_LONG:
3965
            assert(mmco[i].long_index <= 16);
3966
            // just remove the long term which index is greater than new max
3967
            for(j = mmco[i].long_index; j<16; j++){
3968
                pic = remove_long(h, j);
3969
                if (pic) unreference_pic(h, pic);
3970
            }
3971
            break;
3972
        case MMCO_RESET:
3973
            while(h->short_ref_count){
3974
                pic= remove_short(h, h->short_ref[0]->frame_num);
3975
                if(pic) unreference_pic(h, pic);
3976
            }
3977
            for(j = 0; j < 16; j++) {
3978
                pic= remove_long(h, j);
3979
                if(pic) unreference_pic(h, pic);
3980
            }
3981
            break;
3982
        default: assert(0);
3983
        }
3984
    }
3985

    
3986
    if(!current_is_long){
3987
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3988
        if(pic){
3989
            unreference_pic(h, pic);
3990
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3991
        }
3992

    
3993
        if(h->short_ref_count)
3994
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3995

    
3996
        h->short_ref[0]= s->current_picture_ptr;
3997
        h->short_ref[0]->long_ref=0;
3998
        h->short_ref_count++;
3999
    }
4000

    
4001
    print_short_term(h);
4002
    print_long_term(h);
4003
    return 0;
4004
}
4005

    
4006
static int decode_ref_pic_marking(H264Context *h){
4007
    MpegEncContext * const s = &h->s;
4008
    int i;
4009

    
4010
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
4011
        s->broken_link= get_bits1(&s->gb) -1;
4012
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
4013
        if(h->mmco[0].long_index == -1)
4014
            h->mmco_index= 0;
4015
        else{
4016
            h->mmco[0].opcode= MMCO_LONG;
4017
            h->mmco_index= 1;
4018
        }
4019
    }else{
4020
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
4021
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
4022
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
4023

    
4024
                h->mmco[i].opcode= opcode;
4025
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
4026
                    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
4027
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
4028
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
4029
                        return -1;
4030
                    }*/
4031
                }
4032
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
4033
                    unsigned int long_index= get_ue_golomb(&s->gb);
4034
                    if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ long_index >= 16){
4035
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
4036
                        return -1;
4037
                    }
4038
                    h->mmco[i].long_index= long_index;
4039
                }
4040

    
4041
                if(opcode > (unsigned)MMCO_LONG){
4042
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
4043
                    return -1;
4044
                }
4045
                if(opcode == MMCO_END)
4046
                    break;
4047
            }
4048
            h->mmco_index= i;
4049
        }else{
4050
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
4051

    
4052
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
4053
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
4054
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
4055
                h->mmco_index= 1;
4056
            }else
4057
                h->mmco_index= 0;
4058
        }
4059
    }
4060

    
4061
    return 0;
4062
}
4063

    
4064
static int init_poc(H264Context *h){
4065
    MpegEncContext * const s = &h->s;
4066
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
4067
    int field_poc[2];
4068

    
4069
    if(h->nal_unit_type == NAL_IDR_SLICE){
4070
        h->frame_num_offset= 0;
4071
    }else{
4072
        if(h->frame_num < h->prev_frame_num)
4073
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
4074
        else
4075
            h->frame_num_offset= h->prev_frame_num_offset;
4076
    }
4077

    
4078
    if(h->sps.poc_type==0){
4079
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
4080

    
4081
        if(h->nal_unit_type == NAL_IDR_SLICE){
4082
             h->prev_poc_msb=
4083
             h->prev_poc_lsb= 0;
4084
        }
4085

    
4086
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
4087
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
4088
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
4089
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
4090
        else
4091
            h->poc_msb = h->prev_poc_msb;
4092
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
4093
        field_poc[0] =
4094
        field_poc[1] = h->poc_msb + h->poc_lsb;
4095
        if(s->picture_structure == PICT_FRAME)
4096
            field_poc[1] += h->delta_poc_bottom;
4097
    }else if(h->sps.poc_type==1){
4098
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
4099
        int i;
4100

    
4101
        if(h->sps.poc_cycle_length != 0)
4102
            abs_frame_num = h->frame_num_offset + h->frame_num;
4103
        else
4104
            abs_frame_num = 0;
4105

    
4106
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
4107
            abs_frame_num--;
4108

    
4109
        expected_delta_per_poc_cycle = 0;
4110
        for(i=0; i < h->sps.poc_cycle_length; i++)
4111
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
4112

    
4113
        if(abs_frame_num > 0){
4114
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
4115
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
4116

    
4117
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
4118
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
4119
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
4120
        } else
4121
            expectedpoc = 0;
4122

    
4123
        if(h->nal_ref_idc == 0)
4124
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
4125

    
4126
        field_poc[0] = expectedpoc + h->delta_poc[0];
4127
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
4128

    
4129
        if(s->picture_structure == PICT_FRAME)
4130
            field_poc[1] += h->delta_poc[1];
4131
    }else{
4132
        int poc;
4133
        if(h->nal_unit_type == NAL_IDR_SLICE){
4134
            poc= 0;
4135
        }else{
4136
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
4137
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
4138
        }
4139
        field_poc[0]= poc;
4140
        field_poc[1]= poc;
4141
    }
4142

    
4143
    if(s->picture_structure != PICT_BOTTOM_FIELD)
4144
        s->current_picture_ptr->field_poc[0]= field_poc[0];
4145
    if(s->picture_structure != PICT_TOP_FIELD)
4146
        s->current_picture_ptr->field_poc[1]= field_poc[1];
4147
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
4148
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
4149

    
4150
    return 0;
4151
}
4152

    
4153
/**
4154
 * decodes a slice header.
4155
 * this will allso call MPV_common_init() and frame_start() as needed
4156
 */
4157
static int decode_slice_header(H264Context *h){
4158
    MpegEncContext * const s = &h->s;
4159
    unsigned int first_mb_in_slice;
4160
    unsigned int pps_id;
4161
    int num_ref_idx_active_override_flag;
4162
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
4163
    unsigned int slice_type, tmp;
4164
    int default_ref_list_done = 0;
4165

    
4166
    s->current_picture.reference= h->nal_ref_idc != 0;
4167
    s->dropable= h->nal_ref_idc == 0;
4168

    
4169
    first_mb_in_slice= get_ue_golomb(&s->gb);
4170

    
4171
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
4172
        h->slice_num = 0;
4173
        s->current_picture_ptr= NULL;
4174
    }
4175

    
4176
    slice_type= get_ue_golomb(&s->gb);
4177
    if(slice_type > 9){
4178
        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);
4179
        return -1;
4180
    }
4181
    if(slice_type > 4){
4182
        slice_type -= 5;
4183
        h->slice_type_fixed=1;
4184
    }else
4185
        h->slice_type_fixed=0;
4186

    
4187
    slice_type= slice_type_map[ slice_type ];
4188
    if (slice_type == I_TYPE
4189
        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
4190
        default_ref_list_done = 1;
4191
    }
4192
    h->slice_type= slice_type;
4193

    
4194
    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
4195

    
4196
    pps_id= get_ue_golomb(&s->gb);
4197
    if(pps_id>=MAX_PPS_COUNT){
4198
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
4199
        return -1;
4200
    }
4201
    h->pps= h->pps_buffer[pps_id];
4202
    if(h->pps.slice_group_count == 0){
4203
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
4204
        return -1;
4205
    }
4206

    
4207
    h->sps= h->sps_buffer[ h->pps.sps_id ];
4208
    if(h->sps.log2_max_frame_num == 0){
4209
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
4210
        return -1;
4211
    }
4212

    
4213
    if(h->dequant_coeff_pps != pps_id){
4214
        h->dequant_coeff_pps = pps_id;
4215
        init_dequant_tables(h);
4216
    }
4217

    
4218
    s->mb_width= h->sps.mb_width;
4219
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
4220

    
4221
    h->b_stride=  s->mb_width*4;
4222
    h->b8_stride= s->mb_width*2;
4223

    
4224
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
4225
    if(h->sps.frame_mbs_only_flag)
4226
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
4227
    else
4228
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
4229

    
4230
    if (s->context_initialized
4231
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
4232
        free_tables(h);
4233
        MPV_common_end(s);
4234
    }
4235
    if (!s->context_initialized) {
4236
        if (MPV_common_init(s) < 0)
4237
            return -1;
4238

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