Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ ca32f7f2

History | View | Annotate | Download (126 KB)

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

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

    
28
#include "libavcore/imgutils.h"
29
#include "internal.h"
30
#include "dsputil.h"
31
#include "avcodec.h"
32
#include "mpegvideo.h"
33
#include "h264.h"
34
#include "h264data.h"
35
#include "h264_mvpred.h"
36
#include "h264_parser.h"
37
#include "golomb.h"
38
#include "mathops.h"
39
#include "rectangle.h"
40
#include "vdpau_internal.h"
41
#include "libavutil/avassert.h"
42

    
43
#include "cabac.h"
44

    
45
//#undef NDEBUG
46
#include <assert.h>
47

    
48
static const uint8_t rem6[52]={
49
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,
50
};
51

    
52
static const uint8_t div6[52]={
53
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,
54
};
55

    
56
static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
57
    PIX_FMT_DXVA2_VLD,
58
    PIX_FMT_VAAPI_VLD,
59
    PIX_FMT_YUVJ420P,
60
    PIX_FMT_NONE
61
};
62

    
63
void ff_h264_write_back_intra_pred_mode(H264Context *h){
64
    int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
65

    
66
    AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
67
    mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
68
    mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
69
    mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
70
}
71

    
72
/**
73
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
74
 */
75
int ff_h264_check_intra4x4_pred_mode(H264Context *h){
76
    MpegEncContext * const s = &h->s;
77
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
78
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
79
    int i;
80

    
81
    if(!(h->top_samples_available&0x8000)){
82
        for(i=0; i<4; i++){
83
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
84
            if(status<0){
85
                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);
86
                return -1;
87
            } else if(status){
88
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
89
            }
90
        }
91
    }
92

    
93
    if((h->left_samples_available&0x8888)!=0x8888){
94
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
95
        for(i=0; i<4; i++){
96
            if(!(h->left_samples_available&mask[i])){
97
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
98
                if(status<0){
99
                    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);
100
                    return -1;
101
                } else if(status){
102
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
103
                }
104
            }
105
        }
106
    }
107

    
108
    return 0;
109
} //FIXME cleanup like ff_h264_check_intra_pred_mode
110

    
111
/**
112
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
113
 */
114
int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
115
    MpegEncContext * const s = &h->s;
116
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
117
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
118

    
119
    if(mode > 6U) {
120
        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);
121
        return -1;
122
    }
123

    
124
    if(!(h->top_samples_available&0x8000)){
125
        mode= top[ mode ];
126
        if(mode<0){
127
            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);
128
            return -1;
129
        }
130
    }
131

    
132
    if((h->left_samples_available&0x8080) != 0x8080){
133
        mode= left[ mode ];
134
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
135
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
136
        }
137
        if(mode<0){
138
            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);
139
            return -1;
140
        }
141
    }
142

    
143
    return mode;
144
}
145

    
146
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
147
    int i, si, di;
148
    uint8_t *dst;
149
    int bufidx;
150

    
151
//    src[0]&0x80;                //forbidden bit
152
    h->nal_ref_idc= src[0]>>5;
153
    h->nal_unit_type= src[0]&0x1F;
154

    
155
    src++; length--;
156
#if 0
157
    for(i=0; i<length; i++)
158
        printf("%2X ", src[i]);
159
#endif
160

    
161
#if HAVE_FAST_UNALIGNED
162
# if HAVE_FAST_64BIT
163
#   define RS 7
164
    for(i=0; i+1<length; i+=9){
165
        if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
166
# else
167
#   define RS 3
168
    for(i=0; i+1<length; i+=5){
169
        if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
170
# endif
171
            continue;
172
        if(i>0 && !src[i]) i--;
173
        while(src[i]) i++;
174
#else
175
#   define RS 0
176
    for(i=0; i+1<length; i+=2){
177
        if(src[i]) continue;
178
        if(i>0 && src[i-1]==0) i--;
179
#endif
180
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
181
            if(src[i+2]!=3){
182
                /* startcode, so we must be past the end */
183
                length=i;
184
            }
185
            break;
186
        }
187
        i-= RS;
188
    }
189

    
190
    if(i>=length-1){ //no escaped 0
191
        *dst_length= length;
192
        *consumed= length+1; //+1 for the header
193
        return src;
194
    }
195

    
196
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
197
    av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
198
    dst= h->rbsp_buffer[bufidx];
199

    
200
    if (dst == NULL){
201
        return NULL;
202
    }
203

    
204
//printf("decoding esc\n");
205
    memcpy(dst, src, i);
206
    si=di=i;
207
    while(si+2<length){
208
        //remove escapes (very rare 1:2^22)
209
        if(src[si+2]>3){
210
            dst[di++]= src[si++];
211
            dst[di++]= src[si++];
212
        }else if(src[si]==0 && src[si+1]==0){
213
            if(src[si+2]==3){ //escape
214
                dst[di++]= 0;
215
                dst[di++]= 0;
216
                si+=3;
217
                continue;
218
            }else //next start code
219
                goto nsc;
220
        }
221

    
222
        dst[di++]= src[si++];
223
    }
224
    while(si<length)
225
        dst[di++]= src[si++];
226
nsc:
227

    
228
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
229

    
230
    *dst_length= di;
231
    *consumed= si + 1;//+1 for the header
232
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
233
    return dst;
234
}
235

    
236
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
237
    int v= *src;
238
    int r;
239

    
240
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
241

    
242
    for(r=1; r<9; r++){
243
        if(v&1) return r;
244
        v>>=1;
245
    }
246
    return 0;
247
}
248

    
249
/**
250
 * IDCT transforms the 16 dc values and dequantizes them.
251
 * @param qp quantization parameter
252
 */
253
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
254
#define stride 16
255
    int i;
256
    int temp[16]; //FIXME check if this is a good idea
257
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
258
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
259

    
260
//memset(block, 64, 2*256);
261
//return;
262
    for(i=0; i<4; i++){
263
        const int offset= y_offset[i];
264
        const int z0= block[offset+stride*0] + block[offset+stride*4];
265
        const int z1= block[offset+stride*0] - block[offset+stride*4];
266
        const int z2= block[offset+stride*1] - block[offset+stride*5];
267
        const int z3= block[offset+stride*1] + block[offset+stride*5];
268

    
269
        temp[4*i+0]= z0+z3;
270
        temp[4*i+1]= z1+z2;
271
        temp[4*i+2]= z1-z2;
272
        temp[4*i+3]= z0-z3;
273
    }
274

    
275
    for(i=0; i<4; i++){
276
        const int offset= x_offset[i];
277
        const int z0= temp[4*0+i] + temp[4*2+i];
278
        const int z1= temp[4*0+i] - temp[4*2+i];
279
        const int z2= temp[4*1+i] - temp[4*3+i];
280
        const int z3= temp[4*1+i] + temp[4*3+i];
281

    
282
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
283
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
284
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
285
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
286
    }
287
}
288

    
289
#if 0
290
/**
291
 * DCT transforms the 16 dc values.
292
 * @param qp quantization parameter ??? FIXME
293
 */
294
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
295
//    const int qmul= dequant_coeff[qp][0];
296
    int i;
297
    int temp[16]; //FIXME check if this is a good idea
298
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
299
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
300

301
    for(i=0; i<4; i++){
302
        const int offset= y_offset[i];
303
        const int z0= block[offset+stride*0] + block[offset+stride*4];
304
        const int z1= block[offset+stride*0] - block[offset+stride*4];
305
        const int z2= block[offset+stride*1] - block[offset+stride*5];
306
        const int z3= block[offset+stride*1] + block[offset+stride*5];
307

308
        temp[4*i+0]= z0+z3;
309
        temp[4*i+1]= z1+z2;
310
        temp[4*i+2]= z1-z2;
311
        temp[4*i+3]= z0-z3;
312
    }
313

314
    for(i=0; i<4; i++){
315
        const int offset= x_offset[i];
316
        const int z0= temp[4*0+i] + temp[4*2+i];
317
        const int z1= temp[4*0+i] - temp[4*2+i];
318
        const int z2= temp[4*1+i] - temp[4*3+i];
319
        const int z3= temp[4*1+i] + temp[4*3+i];
320

321
        block[stride*0 +offset]= (z0 + z3)>>1;
322
        block[stride*2 +offset]= (z1 + z2)>>1;
323
        block[stride*8 +offset]= (z1 - z2)>>1;
324
        block[stride*10+offset]= (z0 - z3)>>1;
325
    }
326
}
327
#endif
328

    
329
#undef xStride
330
#undef stride
331

    
332
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
333
    const int stride= 16*2;
334
    const int xStride= 16;
335
    int a,b,c,d,e;
336

    
337
    a= block[stride*0 + xStride*0];
338
    b= block[stride*0 + xStride*1];
339
    c= block[stride*1 + xStride*0];
340
    d= block[stride*1 + xStride*1];
341

    
342
    e= a-b;
343
    a= a+b;
344
    b= c-d;
345
    c= c+d;
346

    
347
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
348
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
349
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
350
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
351
}
352

    
353
#if 0
354
static void chroma_dc_dct_c(DCTELEM *block){
355
    const int stride= 16*2;
356
    const int xStride= 16;
357
    int a,b,c,d,e;
358

359
    a= block[stride*0 + xStride*0];
360
    b= block[stride*0 + xStride*1];
361
    c= block[stride*1 + xStride*0];
362
    d= block[stride*1 + xStride*1];
363

364
    e= a-b;
365
    a= a+b;
366
    b= c-d;
367
    c= c+d;
368

369
    block[stride*0 + xStride*0]= (a+c);
370
    block[stride*0 + xStride*1]= (e+b);
371
    block[stride*1 + xStride*0]= (a-c);
372
    block[stride*1 + xStride*1]= (e-b);
373
}
374
#endif
375

    
376
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
377
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
378
                           int src_x_offset, int src_y_offset,
379
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
380
    MpegEncContext * const s = &h->s;
381
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
382
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
383
    const int luma_xy= (mx&3) + ((my&3)<<2);
384
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
385
    uint8_t * src_cb, * src_cr;
386
    int extra_width= h->emu_edge_width;
387
    int extra_height= h->emu_edge_height;
388
    int emu=0;
389
    const int full_mx= mx>>2;
390
    const int full_my= my>>2;
391
    const int pic_width  = 16*s->mb_width;
392
    const int pic_height = 16*s->mb_height >> MB_FIELD;
393

    
394
    if(mx&7) extra_width -= 3;
395
    if(my&7) extra_height -= 3;
396

    
397
    if(   full_mx < 0-extra_width
398
       || full_my < 0-extra_height
399
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
400
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
401
        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);
402
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
403
        emu=1;
404
    }
405

    
406
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
407
    if(!square){
408
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
409
    }
410

    
411
    if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
412

    
413
    if(MB_FIELD){
414
        // chroma offset when predicting from a field of opposite parity
415
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
416
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
417
    }
418
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
419
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
420

    
421
    if(emu){
422
        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);
423
            src_cb= s->edge_emu_buffer;
424
    }
425
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
426

    
427
    if(emu){
428
        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);
429
            src_cr= s->edge_emu_buffer;
430
    }
431
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
432
}
433

    
434
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
435
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
436
                           int x_offset, int y_offset,
437
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
438
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
439
                           int list0, int list1){
440
    MpegEncContext * const s = &h->s;
441
    qpel_mc_func *qpix_op=  qpix_put;
442
    h264_chroma_mc_func chroma_op= chroma_put;
443

    
444
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
445
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
446
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
447
    x_offset += 8*s->mb_x;
448
    y_offset += 8*(s->mb_y >> MB_FIELD);
449

    
450
    if(list0){
451
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
452
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
453
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
454
                           qpix_op, chroma_op);
455

    
456
        qpix_op=  qpix_avg;
457
        chroma_op= chroma_avg;
458
    }
459

    
460
    if(list1){
461
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
462
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
463
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
464
                           qpix_op, chroma_op);
465
    }
466
}
467

    
468
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
469
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
470
                           int x_offset, int y_offset,
471
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
472
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
473
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
474
                           int list0, int list1){
475
    MpegEncContext * const s = &h->s;
476

    
477
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
478
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
479
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
480
    x_offset += 8*s->mb_x;
481
    y_offset += 8*(s->mb_y >> MB_FIELD);
482

    
483
    if(list0 && list1){
484
        /* don't optimize for luma-only case, since B-frames usually
485
         * use implicit weights => chroma too. */
486
        uint8_t *tmp_cb = s->obmc_scratchpad;
487
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
488
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
489
        int refn0 = h->ref_cache[0][ scan8[n] ];
490
        int refn1 = h->ref_cache[1][ scan8[n] ];
491

    
492
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
493
                    dest_y, dest_cb, dest_cr,
494
                    x_offset, y_offset, qpix_put, chroma_put);
495
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
496
                    tmp_y, tmp_cb, tmp_cr,
497
                    x_offset, y_offset, qpix_put, chroma_put);
498

    
499
        if(h->use_weight == 2){
500
            int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
501
            int weight1 = 64 - weight0;
502
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
503
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
504
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
505
        }else{
506
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
507
                            h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
508
                            h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
509
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
510
                            h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
511
                            h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
512
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
513
                            h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
514
                            h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
515
        }
516
    }else{
517
        int list = list1 ? 1 : 0;
518
        int refn = h->ref_cache[list][ scan8[n] ];
519
        Picture *ref= &h->ref_list[list][refn];
520
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
521
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
522
                    qpix_put, chroma_put);
523

    
524
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
525
                       h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
526
        if(h->use_weight_chroma){
527
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
528
                             h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
529
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
530
                             h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
531
        }
532
    }
533
}
534

    
535
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
536
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
537
                           int x_offset, int y_offset,
538
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
539
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
540
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
541
                           int list0, int list1){
542
    if((h->use_weight==2 && list0 && list1
543
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
544
       || h->use_weight==1)
545
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
546
                         x_offset, y_offset, qpix_put, chroma_put,
547
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
548
    else
549
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
550
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
551
}
552

    
553
static inline void prefetch_motion(H264Context *h, int list){
554
    /* fetch pixels for estimated mv 4 macroblocks ahead
555
     * optimized for 64byte cache lines */
556
    MpegEncContext * const s = &h->s;
557
    const int refn = h->ref_cache[list][scan8[0]];
558
    if(refn >= 0){
559
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
560
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
561
        uint8_t **src= h->ref_list[list][refn].data;
562
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
563
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
564
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
565
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
566
    }
567
}
568

    
569
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
570
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
571
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
572
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
573
    MpegEncContext * const s = &h->s;
574
    const int mb_xy= h->mb_xy;
575
    const int mb_type= s->current_picture.mb_type[mb_xy];
576

    
577
    assert(IS_INTER(mb_type));
578

    
579
    prefetch_motion(h, 0);
580

    
581
    if(IS_16X16(mb_type)){
582
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
583
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
584
                weight_op, weight_avg,
585
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
586
    }else if(IS_16X8(mb_type)){
587
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
588
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
589
                &weight_op[1], &weight_avg[1],
590
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
591
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
592
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
593
                &weight_op[1], &weight_avg[1],
594
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
595
    }else if(IS_8X16(mb_type)){
596
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
597
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
598
                &weight_op[2], &weight_avg[2],
599
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
600
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
601
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
602
                &weight_op[2], &weight_avg[2],
603
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
604
    }else{
605
        int i;
606

    
607
        assert(IS_8X8(mb_type));
608

    
609
        for(i=0; i<4; i++){
610
            const int sub_mb_type= h->sub_mb_type[i];
611
            const int n= 4*i;
612
            int x_offset= (i&1)<<2;
613
            int y_offset= (i&2)<<1;
614

    
615
            if(IS_SUB_8X8(sub_mb_type)){
616
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
617
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
618
                    &weight_op[3], &weight_avg[3],
619
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
620
            }else if(IS_SUB_8X4(sub_mb_type)){
621
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
622
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
623
                    &weight_op[4], &weight_avg[4],
624
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
625
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
626
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
627
                    &weight_op[4], &weight_avg[4],
628
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
629
            }else if(IS_SUB_4X8(sub_mb_type)){
630
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
631
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
632
                    &weight_op[5], &weight_avg[5],
633
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
634
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
635
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
636
                    &weight_op[5], &weight_avg[5],
637
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
638
            }else{
639
                int j;
640
                assert(IS_SUB_4X4(sub_mb_type));
641
                for(j=0; j<4; j++){
642
                    int sub_x_offset= x_offset + 2*(j&1);
643
                    int sub_y_offset= y_offset +   (j&2);
644
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
645
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
646
                        &weight_op[6], &weight_avg[6],
647
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
648
                }
649
            }
650
        }
651
    }
652

    
653
    prefetch_motion(h, 1);
654
}
655

    
656

    
657
static void free_tables(H264Context *h){
658
    int i;
659
    H264Context *hx;
660
    av_freep(&h->intra4x4_pred_mode);
661
    av_freep(&h->chroma_pred_mode_table);
662
    av_freep(&h->cbp_table);
663
    av_freep(&h->mvd_table[0]);
664
    av_freep(&h->mvd_table[1]);
665
    av_freep(&h->direct_table);
666
    av_freep(&h->non_zero_count);
667
    av_freep(&h->slice_table_base);
668
    h->slice_table= NULL;
669
    av_freep(&h->list_counts);
670

    
671
    av_freep(&h->mb2b_xy);
672
    av_freep(&h->mb2br_xy);
673

    
674
    for(i = 0; i < MAX_THREADS; i++) {
675
        hx = h->thread_context[i];
676
        if(!hx) continue;
677
        av_freep(&hx->top_borders[1]);
678
        av_freep(&hx->top_borders[0]);
679
        av_freep(&hx->s.obmc_scratchpad);
680
        av_freep(&hx->rbsp_buffer[1]);
681
        av_freep(&hx->rbsp_buffer[0]);
682
        hx->rbsp_buffer_size[0] = 0;
683
        hx->rbsp_buffer_size[1] = 0;
684
        if (i) av_freep(&h->thread_context[i]);
685
    }
686
}
687

    
688
static void init_dequant8_coeff_table(H264Context *h){
689
    int i,q,x;
690
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
691
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
692

    
693
    for(i=0; i<2; i++ ){
694
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
695
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
696
            break;
697
        }
698

    
699
        for(q=0; q<52; q++){
700
            int shift = div6[q];
701
            int idx = rem6[q];
702
            for(x=0; x<64; x++)
703
                h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
704
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
705
                    h->pps.scaling_matrix8[i][x]) << shift;
706
        }
707
    }
708
}
709

    
710
static void init_dequant4_coeff_table(H264Context *h){
711
    int i,j,q,x;
712
    for(i=0; i<6; i++ ){
713
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
714
        for(j=0; j<i; j++){
715
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
716
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
717
                break;
718
            }
719
        }
720
        if(j<i)
721
            continue;
722

    
723
        for(q=0; q<52; q++){
724
            int shift = div6[q] + 2;
725
            int idx = rem6[q];
726
            for(x=0; x<16; x++)
727
                h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
728
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
729
                    h->pps.scaling_matrix4[i][x]) << shift;
730
        }
731
    }
732
}
733

    
734
static void init_dequant_tables(H264Context *h){
735
    int i,x;
736
    init_dequant4_coeff_table(h);
737
    if(h->pps.transform_8x8_mode)
738
        init_dequant8_coeff_table(h);
739
    if(h->sps.transform_bypass){
740
        for(i=0; i<6; i++)
741
            for(x=0; x<16; x++)
742
                h->dequant4_coeff[i][0][x] = 1<<6;
743
        if(h->pps.transform_8x8_mode)
744
            for(i=0; i<2; i++)
745
                for(x=0; x<64; x++)
746
                    h->dequant8_coeff[i][0][x] = 1<<6;
747
    }
748
}
749

    
750

    
751
int ff_h264_alloc_tables(H264Context *h){
752
    MpegEncContext * const s = &h->s;
753
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
754
    const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
755
    int x,y;
756

    
757
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
758

    
759
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
760
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
761
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
762

    
763
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
764
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
765
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
766
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
767
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
768

    
769
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
770
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
771

    
772
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
773
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
774
    for(y=0; y<s->mb_height; y++){
775
        for(x=0; x<s->mb_width; x++){
776
            const int mb_xy= x + y*s->mb_stride;
777
            const int b_xy = 4*x + 4*y*h->b_stride;
778

    
779
            h->mb2b_xy [mb_xy]= b_xy;
780
            h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
781
        }
782
    }
783

    
784
    s->obmc_scratchpad = NULL;
785

    
786
    if(!h->dequant4_coeff[0])
787
        init_dequant_tables(h);
788

    
789
    return 0;
790
fail:
791
    free_tables(h);
792
    return -1;
793
}
794

    
795
/**
796
 * Mimic alloc_tables(), but for every context thread.
797
 */
798
static void clone_tables(H264Context *dst, H264Context *src, int i){
799
    MpegEncContext * const s = &src->s;
800
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
801
    dst->non_zero_count           = src->non_zero_count;
802
    dst->slice_table              = src->slice_table;
803
    dst->cbp_table                = src->cbp_table;
804
    dst->mb2b_xy                  = src->mb2b_xy;
805
    dst->mb2br_xy                 = src->mb2br_xy;
806
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
807
    dst->mvd_table[0]             = src->mvd_table[0] + i*8*2*s->mb_stride;
808
    dst->mvd_table[1]             = src->mvd_table[1] + i*8*2*s->mb_stride;
809
    dst->direct_table             = src->direct_table;
810
    dst->list_counts              = src->list_counts;
811

    
812
    dst->s.obmc_scratchpad = NULL;
813
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
814
}
815

    
816
/**
817
 * Init context
818
 * Allocate buffers which are not shared amongst multiple threads.
819
 */
820
static int context_init(H264Context *h){
821
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
822
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
823

    
824
    h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
825
    h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
826

    
827
    return 0;
828
fail:
829
    return -1; // free_tables will clean up for us
830
}
831

    
832
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
833

    
834
static av_cold void common_init(H264Context *h){
835
    MpegEncContext * const s = &h->s;
836

    
837
    s->width = s->avctx->width;
838
    s->height = s->avctx->height;
839
    s->codec_id= s->avctx->codec->id;
840

    
841
    ff_h264dsp_init(&h->h264dsp);
842
    ff_h264_pred_init(&h->hpc, s->codec_id);
843

    
844
    h->dequant_coeff_pps= -1;
845
    s->unrestricted_mv=1;
846
    s->decode=1; //FIXME
847

    
848
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
849

    
850
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
851
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
852
}
853

    
854
int ff_h264_decode_extradata(H264Context *h)
855
{
856
    AVCodecContext *avctx = h->s.avctx;
857

    
858
    if(*(char *)avctx->extradata == 1){
859
        int i, cnt, nalsize;
860
        unsigned char *p = avctx->extradata;
861

    
862
        h->is_avc = 1;
863

    
864
        if(avctx->extradata_size < 7) {
865
            av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
866
            return -1;
867
        }
868
        /* sps and pps in the avcC always have length coded with 2 bytes,
869
           so put a fake nal_length_size = 2 while parsing them */
870
        h->nal_length_size = 2;
871
        // Decode sps from avcC
872
        cnt = *(p+5) & 0x1f; // Number of sps
873
        p += 6;
874
        for (i = 0; i < cnt; i++) {
875
            nalsize = AV_RB16(p) + 2;
876
            if(decode_nal_units(h, p, nalsize) < 0) {
877
                av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
878
                return -1;
879
            }
880
            p += nalsize;
881
        }
882
        // Decode pps from avcC
883
        cnt = *(p++); // Number of pps
884
        for (i = 0; i < cnt; i++) {
885
            nalsize = AV_RB16(p) + 2;
886
            if(decode_nal_units(h, p, nalsize)  != nalsize) {
887
                av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
888
                return -1;
889
            }
890
            p += nalsize;
891
        }
892
        // Now store right nal length size, that will be use to parse all other nals
893
        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
894
    } else {
895
        h->is_avc = 0;
896
        if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
897
            return -1;
898
    }
899
    return 0;
900
}
901

    
902
av_cold int ff_h264_decode_init(AVCodecContext *avctx){
903
    H264Context *h= avctx->priv_data;
904
    MpegEncContext * const s = &h->s;
905

    
906
    MPV_decode_defaults(s);
907

    
908
    s->avctx = avctx;
909
    common_init(h);
910

    
911
    s->out_format = FMT_H264;
912
    s->workaround_bugs= avctx->workaround_bugs;
913

    
914
    // set defaults
915
//    s->decode_mb= ff_h263_decode_mb;
916
    s->quarter_sample = 1;
917
    if(!avctx->has_b_frames)
918
    s->low_delay= 1;
919

    
920
    avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
921

    
922
    ff_h264_decode_init_vlc();
923

    
924
    h->thread_context[0] = h;
925
    h->outputed_poc = INT_MIN;
926
    h->prev_poc_msb= 1<<16;
927
    h->x264_build = -1;
928
    ff_h264_reset_sei(h);
929
    if(avctx->codec_id == CODEC_ID_H264){
930
        if(avctx->ticks_per_frame == 1){
931
            s->avctx->time_base.den *=2;
932
        }
933
        avctx->ticks_per_frame = 2;
934
    }
935

    
936
    if(avctx->extradata_size > 0 && avctx->extradata &&
937
        ff_h264_decode_extradata(h))
938
        return -1;
939

    
940
    if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
941
        s->avctx->has_b_frames = h->sps.num_reorder_frames;
942
        s->low_delay = 0;
943
    }
944

    
945
    return 0;
946
}
947

    
948
int ff_h264_frame_start(H264Context *h){
949
    MpegEncContext * const s = &h->s;
950
    int i;
951

    
952
    if(MPV_frame_start(s, s->avctx) < 0)
953
        return -1;
954
    ff_er_frame_start(s);
955
    /*
956
     * MPV_frame_start uses pict_type to derive key_frame.
957
     * This is incorrect for H.264; IDR markings must be used.
958
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
959
     * See decode_nal_units().
960
     */
961
    s->current_picture_ptr->key_frame= 0;
962
    s->current_picture_ptr->mmco_reset= 0;
963

    
964
    assert(s->linesize && s->uvlinesize);
965

    
966
    for(i=0; i<16; i++){
967
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
968
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
969
    }
970
    for(i=0; i<4; i++){
971
        h->block_offset[16+i]=
972
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
973
        h->block_offset[24+16+i]=
974
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
975
    }
976

    
977
    /* can't be in alloc_tables because linesize isn't known there.
978
     * FIXME: redo bipred weight to not require extra buffer? */
979
    for(i = 0; i < s->avctx->thread_count; i++)
980
        if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
981
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
982

    
983
    /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
984
    memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
985

    
986
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
987

    
988
    // We mark the current picture as non-reference after allocating it, so
989
    // that if we break out due to an error it can be released automatically
990
    // in the next MPV_frame_start().
991
    // SVQ3 as well as most other codecs have only last/next/current and thus
992
    // get released even with set reference, besides SVQ3 and others do not
993
    // mark frames as reference later "naturally".
994
    if(s->codec_id != CODEC_ID_SVQ3)
995
        s->current_picture_ptr->reference= 0;
996

    
997
    s->current_picture_ptr->field_poc[0]=
998
    s->current_picture_ptr->field_poc[1]= INT_MAX;
999
    assert(s->current_picture_ptr->long_ref==0);
1000

    
1001
    return 0;
1002
}
1003

    
1004
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
1005
    MpegEncContext * const s = &h->s;
1006
    uint8_t *top_border;
1007
    int top_idx = 1;
1008

    
1009
    src_y  -=   linesize;
1010
    src_cb -= uvlinesize;
1011
    src_cr -= uvlinesize;
1012

    
1013
    if(!simple && FRAME_MBAFF){
1014
        if(s->mb_y&1){
1015
            if(!MB_MBAFF){
1016
                top_border = h->top_borders[0][s->mb_x];
1017
                AV_COPY128(top_border, src_y + 15*linesize);
1018
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1019
                    AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1020
                    AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1021
                }
1022
            }
1023
        }else if(MB_MBAFF){
1024
            top_idx = 0;
1025
        }else
1026
            return;
1027
    }
1028

    
1029
    top_border = h->top_borders[top_idx][s->mb_x];
1030
    // There are two lines saved, the line above the the top macroblock of a pair,
1031
    // and the line above the bottom macroblock
1032
    AV_COPY128(top_border, src_y + 16*linesize);
1033

    
1034
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1035
        AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1036
        AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1037
    }
1038
}
1039

    
1040
static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
1041
    MpegEncContext * const s = &h->s;
1042
    int deblock_left;
1043
    int deblock_top;
1044
    int top_idx = 1;
1045
    uint8_t *top_border_m1;
1046
    uint8_t *top_border;
1047

    
1048
    if(!simple && FRAME_MBAFF){
1049
        if(s->mb_y&1){
1050
            if(!MB_MBAFF)
1051
                return;
1052
        }else{
1053
            top_idx = MB_MBAFF ? 0 : 1;
1054
        }
1055
    }
1056

    
1057
    if(h->deblocking_filter == 2) {
1058
        deblock_left = h->left_type[0];
1059
        deblock_top  = h->top_type;
1060
    } else {
1061
        deblock_left = (s->mb_x > 0);
1062
        deblock_top =  (s->mb_y > !!MB_FIELD);
1063
    }
1064

    
1065
    src_y  -=   linesize + 1;
1066
    src_cb -= uvlinesize + 1;
1067
    src_cr -= uvlinesize + 1;
1068

    
1069
    top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1070
    top_border    = h->top_borders[top_idx][s->mb_x];
1071

    
1072
#define XCHG(a,b,xchg)\
1073
if (xchg) AV_SWAP64(b,a);\
1074
else      AV_COPY64(b,a);
1075

    
1076
    if(deblock_top){
1077
        if(deblock_left){
1078
            XCHG(top_border_m1+8, src_y -7, 1);
1079
        }
1080
        XCHG(top_border+0, src_y +1, xchg);
1081
        XCHG(top_border+8, src_y +9, 1);
1082
        if(s->mb_x+1 < s->mb_width){
1083
            XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1084
        }
1085
    }
1086

    
1087
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1088
        if(deblock_top){
1089
            if(deblock_left){
1090
                XCHG(top_border_m1+16, src_cb -7, 1);
1091
                XCHG(top_border_m1+24, src_cr -7, 1);
1092
            }
1093
            XCHG(top_border+16, src_cb+1, 1);
1094
            XCHG(top_border+24, src_cr+1, 1);
1095
        }
1096
    }
1097
}
1098

    
1099
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1100
    MpegEncContext * const s = &h->s;
1101
    const int mb_x= s->mb_x;
1102
    const int mb_y= s->mb_y;
1103
    const int mb_xy= h->mb_xy;
1104
    const int mb_type= s->current_picture.mb_type[mb_xy];
1105
    uint8_t  *dest_y, *dest_cb, *dest_cr;
1106
    int linesize, uvlinesize /*dct_offset*/;
1107
    int i;
1108
    int *block_offset = &h->block_offset[0];
1109
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1110
    /* is_h264 should always be true if SVQ3 is disabled. */
1111
    const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1112
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1113
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1114

    
1115
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1116
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1117
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1118

    
1119
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1120
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1121

    
1122
    h->list_counts[mb_xy]= h->list_count;
1123

    
1124
    if (!simple && MB_FIELD) {
1125
        linesize   = h->mb_linesize   = s->linesize * 2;
1126
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1127
        block_offset = &h->block_offset[24];
1128
        if(mb_y&1){ //FIXME move out of this function?
1129
            dest_y -= s->linesize*15;
1130
            dest_cb-= s->uvlinesize*7;
1131
            dest_cr-= s->uvlinesize*7;
1132
        }
1133
        if(FRAME_MBAFF) {
1134
            int list;
1135
            for(list=0; list<h->list_count; list++){
1136
                if(!USES_LIST(mb_type, list))
1137
                    continue;
1138
                if(IS_16X16(mb_type)){
1139
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
1140
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1141
                }else{
1142
                    for(i=0; i<16; i+=4){
1143
                        int ref = h->ref_cache[list][scan8[i]];
1144
                        if(ref >= 0)
1145
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1146
                    }
1147
                }
1148
            }
1149
        }
1150
    } else {
1151
        linesize   = h->mb_linesize   = s->linesize;
1152
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1153
//        dct_offset = s->linesize * 16;
1154
    }
1155

    
1156
    if (!simple && IS_INTRA_PCM(mb_type)) {
1157
        for (i=0; i<16; i++) {
1158
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1159
        }
1160
        for (i=0; i<8; i++) {
1161
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1162
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1163
        }
1164
    } else {
1165
        if(IS_INTRA(mb_type)){
1166
            if(h->deblocking_filter)
1167
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1168

    
1169
            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1170
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1171
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1172
            }
1173

    
1174
            if(IS_INTRA4x4(mb_type)){
1175
                if(simple || !s->encoding){
1176
                    if(IS_8x8DCT(mb_type)){
1177
                        if(transform_bypass){
1178
                            idct_dc_add =
1179
                            idct_add    = s->dsp.add_pixels8;
1180
                        }else{
1181
                            idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1182
                            idct_add    = h->h264dsp.h264_idct8_add;
1183
                        }
1184
                        for(i=0; i<16; i+=4){
1185
                            uint8_t * const ptr= dest_y + block_offset[i];
1186
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1187
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1188
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1189
                            }else{
1190
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
1191
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1192
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
1193
                                if(nnz){
1194
                                    if(nnz == 1 && h->mb[i*16])
1195
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
1196
                                    else
1197
                                        idct_add   (ptr, h->mb + i*16, linesize);
1198
                                }
1199
                            }
1200
                        }
1201
                    }else{
1202
                        if(transform_bypass){
1203
                            idct_dc_add =
1204
                            idct_add    = s->dsp.add_pixels4;
1205
                        }else{
1206
                            idct_dc_add = h->h264dsp.h264_idct_dc_add;
1207
                            idct_add    = h->h264dsp.h264_idct_add;
1208
                        }
1209
                        for(i=0; i<16; i++){
1210
                            uint8_t * const ptr= dest_y + block_offset[i];
1211
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1212

    
1213
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1214
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1215
                            }else{
1216
                                uint8_t *topright;
1217
                                int nnz, tr;
1218
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1219
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1220
                                    assert(mb_y || linesize <= block_offset[i]);
1221
                                    if(!topright_avail){
1222
                                        tr= ptr[3 - linesize]*0x01010101;
1223
                                        topright= (uint8_t*) &tr;
1224
                                    }else
1225
                                        topright= ptr + 4 - linesize;
1226
                                }else
1227
                                    topright= NULL;
1228

    
1229
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1230
                                nnz = h->non_zero_count_cache[ scan8[i] ];
1231
                                if(nnz){
1232
                                    if(is_h264){
1233
                                        if(nnz == 1 && h->mb[i*16])
1234
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
1235
                                        else
1236
                                            idct_add   (ptr, h->mb + i*16, linesize);
1237
                                    }else
1238
                                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1239
                                }
1240
                            }
1241
                        }
1242
                    }
1243
                }
1244
            }else{
1245
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1246
                if(is_h264){
1247
                    if(!transform_bypass)
1248
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1249
                }else
1250
                    ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1251
            }
1252
            if(h->deblocking_filter)
1253
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1254
        }else if(is_h264){
1255
            hl_motion(h, dest_y, dest_cb, dest_cr,
1256
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1257
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1258
                      h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
1259
        }
1260

    
1261

    
1262
        if(!IS_INTRA4x4(mb_type)){
1263
            if(is_h264){
1264
                if(IS_INTRA16x16(mb_type)){
1265
                    if(transform_bypass){
1266
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1267
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1268
                        }else{
1269
                            for(i=0; i<16; i++){
1270
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1271
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1272
                            }
1273
                        }
1274
                    }else{
1275
                         h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1276
                    }
1277
                }else if(h->cbp&15){
1278
                    if(transform_bypass){
1279
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1280
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1281
                        for(i=0; i<16; i+=di){
1282
                            if(h->non_zero_count_cache[ scan8[i] ]){
1283
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1284
                            }
1285
                        }
1286
                    }else{
1287
                        if(IS_8x8DCT(mb_type)){
1288
                            h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1289
                        }else{
1290
                            h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1291
                        }
1292
                    }
1293
                }
1294
            }else{
1295
                for(i=0; i<16; i++){
1296
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1297
                        uint8_t * const ptr= dest_y + block_offset[i];
1298
                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1299
                    }
1300
                }
1301
            }
1302
        }
1303

    
1304
        if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1305
            uint8_t *dest[2] = {dest_cb, dest_cr};
1306
            if(transform_bypass){
1307
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1308
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1309
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1310
                }else{
1311
                    idct_add = s->dsp.add_pixels4;
1312
                    for(i=16; i<16+8; i++){
1313
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1314
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1315
                    }
1316
                }
1317
            }else{
1318
                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
1319
                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
1320
                if(is_h264){
1321
                    h->h264dsp.h264_idct_add8(dest, block_offset,
1322
                                              h->mb, uvlinesize,
1323
                                              h->non_zero_count_cache);
1324
                }else{
1325
                    for(i=16; i<16+8; i++){
1326
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1327
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1328
                            ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1329
                        }
1330
                    }
1331
                }
1332
            }
1333
        }
1334
    }
1335
    if(h->cbp || IS_INTRA(mb_type))
1336
        s->dsp.clear_blocks(h->mb);
1337
}
1338

    
1339
/**
1340
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1341
 */
1342
static void hl_decode_mb_simple(H264Context *h){
1343
    hl_decode_mb_internal(h, 1);
1344
}
1345

    
1346
/**
1347
 * Process a macroblock; this handles edge cases, such as interlacing.
1348
 */
1349
static void av_noinline hl_decode_mb_complex(H264Context *h){
1350
    hl_decode_mb_internal(h, 0);
1351
}
1352

    
1353
void ff_h264_hl_decode_mb(H264Context *h){
1354
    MpegEncContext * const s = &h->s;
1355
    const int mb_xy= h->mb_xy;
1356
    const int mb_type= s->current_picture.mb_type[mb_xy];
1357
    int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1358

    
1359
    if (is_complex)
1360
        hl_decode_mb_complex(h);
1361
    else hl_decode_mb_simple(h);
1362
}
1363

    
1364
static int pred_weight_table(H264Context *h){
1365
    MpegEncContext * const s = &h->s;
1366
    int list, i;
1367
    int luma_def, chroma_def;
1368

    
1369
    h->use_weight= 0;
1370
    h->use_weight_chroma= 0;
1371
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1372
    if(CHROMA)
1373
        h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1374
    luma_def = 1<<h->luma_log2_weight_denom;
1375
    chroma_def = 1<<h->chroma_log2_weight_denom;
1376

    
1377
    for(list=0; list<2; list++){
1378
        h->luma_weight_flag[list]   = 0;
1379
        h->chroma_weight_flag[list] = 0;
1380
        for(i=0; i<h->ref_count[list]; i++){
1381
            int luma_weight_flag, chroma_weight_flag;
1382

    
1383
            luma_weight_flag= get_bits1(&s->gb);
1384
            if(luma_weight_flag){
1385
                h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
1386
                h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
1387
                if(   h->luma_weight[i][list][0] != luma_def
1388
                   || h->luma_weight[i][list][1] != 0) {
1389
                    h->use_weight= 1;
1390
                    h->luma_weight_flag[list]= 1;
1391
                }
1392
            }else{
1393
                h->luma_weight[i][list][0]= luma_def;
1394
                h->luma_weight[i][list][1]= 0;
1395
            }
1396

    
1397
            if(CHROMA){
1398
                chroma_weight_flag= get_bits1(&s->gb);
1399
                if(chroma_weight_flag){
1400
                    int j;
1401
                    for(j=0; j<2; j++){
1402
                        h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
1403
                        h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
1404
                        if(   h->chroma_weight[i][list][j][0] != chroma_def
1405
                           || h->chroma_weight[i][list][j][1] != 0) {
1406
                            h->use_weight_chroma= 1;
1407
                            h->chroma_weight_flag[list]= 1;
1408
                        }
1409
                    }
1410
                }else{
1411
                    int j;
1412
                    for(j=0; j<2; j++){
1413
                        h->chroma_weight[i][list][j][0]= chroma_def;
1414
                        h->chroma_weight[i][list][j][1]= 0;
1415
                    }
1416
                }
1417
            }
1418
        }
1419
        if(h->slice_type_nos != FF_B_TYPE) break;
1420
    }
1421
    h->use_weight= h->use_weight || h->use_weight_chroma;
1422
    return 0;
1423
}
1424

    
1425
/**
1426
 * Initialize implicit_weight table.
1427
 * @param field  0/1 initialize the weight for interlaced MBAFF
1428
 *                -1 initializes the rest
1429
 */
1430
static void implicit_weight_table(H264Context *h, int field){
1431
    MpegEncContext * const s = &h->s;
1432
    int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
1433

    
1434
    for (i = 0; i < 2; i++) {
1435
        h->luma_weight_flag[i]   = 0;
1436
        h->chroma_weight_flag[i] = 0;
1437
    }
1438

    
1439
    if(field < 0){
1440
        cur_poc = s->current_picture_ptr->poc;
1441
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
1442
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1443
        h->use_weight= 0;
1444
        h->use_weight_chroma= 0;
1445
        return;
1446
    }
1447
        ref_start= 0;
1448
        ref_count0= h->ref_count[0];
1449
        ref_count1= h->ref_count[1];
1450
    }else{
1451
        cur_poc = s->current_picture_ptr->field_poc[field];
1452
        ref_start= 16;
1453
        ref_count0= 16+2*h->ref_count[0];
1454
        ref_count1= 16+2*h->ref_count[1];
1455
    }
1456

    
1457
    h->use_weight= 2;
1458
    h->use_weight_chroma= 2;
1459
    h->luma_log2_weight_denom= 5;
1460
    h->chroma_log2_weight_denom= 5;
1461

    
1462
    for(ref0=ref_start; ref0 < ref_count0; ref0++){
1463
        int poc0 = h->ref_list[0][ref0].poc;
1464
        for(ref1=ref_start; ref1 < ref_count1; ref1++){
1465
            int poc1 = h->ref_list[1][ref1].poc;
1466
            int td = av_clip(poc1 - poc0, -128, 127);
1467
            int w= 32;
1468
            if(td){
1469
                int tb = av_clip(cur_poc - poc0, -128, 127);
1470
                int tx = (16384 + (FFABS(td) >> 1)) / td;
1471
                int dist_scale_factor = (tb*tx + 32) >> 8;
1472
                if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
1473
                    w = 64 - dist_scale_factor;
1474
            }
1475
            if(field<0){
1476
                h->implicit_weight[ref0][ref1][0]=
1477
                h->implicit_weight[ref0][ref1][1]= w;
1478
            }else{
1479
                h->implicit_weight[ref0][ref1][field]=w;
1480
            }
1481
        }
1482
    }
1483
}
1484

    
1485
/**
1486
 * instantaneous decoder refresh.
1487
 */
1488
static void idr(H264Context *h){
1489
    ff_h264_remove_all_refs(h);
1490
    h->prev_frame_num= 0;
1491
    h->prev_frame_num_offset= 0;
1492
    h->prev_poc_msb=
1493
    h->prev_poc_lsb= 0;
1494
}
1495

    
1496
/* forget old pics after a seek */
1497
static void flush_dpb(AVCodecContext *avctx){
1498
    H264Context *h= avctx->priv_data;
1499
    int i;
1500
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1501
        if(h->delayed_pic[i])
1502
            h->delayed_pic[i]->reference= 0;
1503
        h->delayed_pic[i]= NULL;
1504
    }
1505
    h->outputed_poc= INT_MIN;
1506
    h->prev_interlaced_frame = 1;
1507
    idr(h);
1508
    if(h->s.current_picture_ptr)
1509
        h->s.current_picture_ptr->reference= 0;
1510
    h->s.first_field= 0;
1511
    ff_h264_reset_sei(h);
1512
    ff_mpeg_flush(avctx);
1513
}
1514

    
1515
static int init_poc(H264Context *h){
1516
    MpegEncContext * const s = &h->s;
1517
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1518
    int field_poc[2];
1519
    Picture *cur = s->current_picture_ptr;
1520

    
1521
    h->frame_num_offset= h->prev_frame_num_offset;
1522
    if(h->frame_num < h->prev_frame_num)
1523
        h->frame_num_offset += max_frame_num;
1524

    
1525
    if(h->sps.poc_type==0){
1526
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1527

    
1528
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1529
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1530
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1531
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1532
        else
1533
            h->poc_msb = h->prev_poc_msb;
1534
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1535
        field_poc[0] =
1536
        field_poc[1] = h->poc_msb + h->poc_lsb;
1537
        if(s->picture_structure == PICT_FRAME)
1538
            field_poc[1] += h->delta_poc_bottom;
1539
    }else if(h->sps.poc_type==1){
1540
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1541
        int i;
1542

    
1543
        if(h->sps.poc_cycle_length != 0)
1544
            abs_frame_num = h->frame_num_offset + h->frame_num;
1545
        else
1546
            abs_frame_num = 0;
1547

    
1548
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
1549
            abs_frame_num--;
1550

    
1551
        expected_delta_per_poc_cycle = 0;
1552
        for(i=0; i < h->sps.poc_cycle_length; i++)
1553
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1554

    
1555
        if(abs_frame_num > 0){
1556
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1557
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1558

    
1559
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1560
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
1561
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1562
        } else
1563
            expectedpoc = 0;
1564

    
1565
        if(h->nal_ref_idc == 0)
1566
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1567

    
1568
        field_poc[0] = expectedpoc + h->delta_poc[0];
1569
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1570

    
1571
        if(s->picture_structure == PICT_FRAME)
1572
            field_poc[1] += h->delta_poc[1];
1573
    }else{
1574
        int poc= 2*(h->frame_num_offset + h->frame_num);
1575

    
1576
        if(!h->nal_ref_idc)
1577
            poc--;
1578

    
1579
        field_poc[0]= poc;
1580
        field_poc[1]= poc;
1581
    }
1582

    
1583
    if(s->picture_structure != PICT_BOTTOM_FIELD)
1584
        s->current_picture_ptr->field_poc[0]= field_poc[0];
1585
    if(s->picture_structure != PICT_TOP_FIELD)
1586
        s->current_picture_ptr->field_poc[1]= field_poc[1];
1587
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1588

    
1589
    return 0;
1590
}
1591

    
1592

    
1593
/**
1594
 * initialize scan tables
1595
 */
1596
static void init_scan_tables(H264Context *h){
1597
    int i;
1598
    for(i=0; i<16; i++){
1599
#define T(x) (x>>2) | ((x<<2) & 0xF)
1600
        h->zigzag_scan[i] = T(zigzag_scan[i]);
1601
        h-> field_scan[i] = T( field_scan[i]);
1602
#undef T
1603
    }
1604
    for(i=0; i<64; i++){
1605
#define T(x) (x>>3) | ((x&7)<<3)
1606
        h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1607
        h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1608
        h->field_scan8x8[i]        = T(field_scan8x8[i]);
1609
        h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1610
#undef T
1611
    }
1612
    if(h->sps.transform_bypass){ //FIXME same ugly
1613
        h->zigzag_scan_q0          = zigzag_scan;
1614
        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1615
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1616
        h->field_scan_q0           = field_scan;
1617
        h->field_scan8x8_q0        = field_scan8x8;
1618
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1619
    }else{
1620
        h->zigzag_scan_q0          = h->zigzag_scan;
1621
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1622
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1623
        h->field_scan_q0           = h->field_scan;
1624
        h->field_scan8x8_q0        = h->field_scan8x8;
1625
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1626
    }
1627
}
1628

    
1629
static void field_end(H264Context *h){
1630
    MpegEncContext * const s = &h->s;
1631
    AVCodecContext * const avctx= s->avctx;
1632
    s->mb_y= 0;
1633

    
1634
    s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1635
    s->current_picture_ptr->pict_type= s->pict_type;
1636

    
1637
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1638
        ff_vdpau_h264_set_reference_frames(s);
1639

    
1640
    if(!s->dropable) {
1641
        ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1642
        h->prev_poc_msb= h->poc_msb;
1643
        h->prev_poc_lsb= h->poc_lsb;
1644
    }
1645
    h->prev_frame_num_offset= h->frame_num_offset;
1646
    h->prev_frame_num= h->frame_num;
1647

    
1648
    if (avctx->hwaccel) {
1649
        if (avctx->hwaccel->end_frame(avctx) < 0)
1650
            av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1651
    }
1652

    
1653
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1654
        ff_vdpau_h264_picture_complete(s);
1655

    
1656
    /*
1657
     * FIXME: Error handling code does not seem to support interlaced
1658
     * when slices span multiple rows
1659
     * The ff_er_add_slice calls don't work right for bottom
1660
     * fields; they cause massive erroneous error concealing
1661
     * Error marking covers both fields (top and bottom).
1662
     * This causes a mismatched s->error_count
1663
     * and a bad error table. Further, the error count goes to
1664
     * INT_MAX when called for bottom field, because mb_y is
1665
     * past end by one (callers fault) and resync_mb_y != 0
1666
     * causes problems for the first MB line, too.
1667
     */
1668
    if (!FIELD_PICTURE)
1669
        ff_er_frame_end(s);
1670

    
1671
    MPV_frame_end(s);
1672

    
1673
    h->current_slice=0;
1674
}
1675

    
1676
/**
1677
 * Replicate H264 "master" context to thread contexts.
1678
 */
1679
static void clone_slice(H264Context *dst, H264Context *src)
1680
{
1681
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1682
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1683
    dst->s.current_picture      = src->s.current_picture;
1684
    dst->s.linesize             = src->s.linesize;
1685
    dst->s.uvlinesize           = src->s.uvlinesize;
1686
    dst->s.first_field          = src->s.first_field;
1687

    
1688
    dst->prev_poc_msb           = src->prev_poc_msb;
1689
    dst->prev_poc_lsb           = src->prev_poc_lsb;
1690
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1691
    dst->prev_frame_num         = src->prev_frame_num;
1692
    dst->short_ref_count        = src->short_ref_count;
1693

    
1694
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1695
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1696
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1697
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1698

    
1699
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1700
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1701
}
1702

    
1703
/**
1704
 * decodes a slice header.
1705
 * This will also call MPV_common_init() and frame_start() as needed.
1706
 *
1707
 * @param h h264context
1708
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1709
 *
1710
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1711
 */
1712
static int decode_slice_header(H264Context *h, H264Context *h0){
1713
    MpegEncContext * const s = &h->s;
1714
    MpegEncContext * const s0 = &h0->s;
1715
    unsigned int first_mb_in_slice;
1716
    unsigned int pps_id;
1717
    int num_ref_idx_active_override_flag;
1718
    unsigned int slice_type, tmp, i, j;
1719
    int default_ref_list_done = 0;
1720
    int last_pic_structure;
1721

    
1722
    s->dropable= h->nal_ref_idc == 0;
1723

    
1724
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1725
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1726
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1727
    }else{
1728
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1729
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1730
    }
1731

    
1732
    first_mb_in_slice= get_ue_golomb(&s->gb);
1733

    
1734
    if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1735
        if(h0->current_slice && FIELD_PICTURE){
1736
            field_end(h);
1737
        }
1738

    
1739
        h0->current_slice = 0;
1740
        if (!s0->first_field)
1741
            s->current_picture_ptr= NULL;
1742
    }
1743

    
1744
    slice_type= get_ue_golomb_31(&s->gb);
1745
    if(slice_type > 9){
1746
        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);
1747
        return -1;
1748
    }
1749
    if(slice_type > 4){
1750
        slice_type -= 5;
1751
        h->slice_type_fixed=1;
1752
    }else
1753
        h->slice_type_fixed=0;
1754

    
1755
    slice_type= golomb_to_pict_type[ slice_type ];
1756
    if (slice_type == FF_I_TYPE
1757
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1758
        default_ref_list_done = 1;
1759
    }
1760
    h->slice_type= slice_type;
1761
    h->slice_type_nos= slice_type & 3;
1762

    
1763
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1764

    
1765
    pps_id= get_ue_golomb(&s->gb);
1766
    if(pps_id>=MAX_PPS_COUNT){
1767
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1768
        return -1;
1769
    }
1770
    if(!h0->pps_buffers[pps_id]) {
1771
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1772
        return -1;
1773
    }
1774
    h->pps= *h0->pps_buffers[pps_id];
1775

    
1776
    if(!h0->sps_buffers[h->pps.sps_id]) {
1777
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1778
        return -1;
1779
    }
1780
    h->sps = *h0->sps_buffers[h->pps.sps_id];
1781

    
1782
    s->avctx->profile = h->sps.profile_idc;
1783
    s->avctx->level   = h->sps.level_idc;
1784
    s->avctx->refs    = h->sps.ref_frame_count;
1785

    
1786
    if(h == h0 && h->dequant_coeff_pps != pps_id){
1787
        h->dequant_coeff_pps = pps_id;
1788
        init_dequant_tables(h);
1789
    }
1790

    
1791
    s->mb_width= h->sps.mb_width;
1792
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1793

    
1794
    h->b_stride=  s->mb_width*4;
1795

    
1796
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1797
    if(h->sps.frame_mbs_only_flag)
1798
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1799
    else
1800
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
1801

    
1802
    if (s->context_initialized
1803
        && (   s->width != s->avctx->width || s->height != s->avctx->height
1804
            || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
1805
        if(h != h0)
1806
            return -1;   // width / height changed during parallelized decoding
1807
        free_tables(h);
1808
        flush_dpb(s->avctx);
1809
        MPV_common_end(s);
1810
    }
1811
    if (!s->context_initialized) {
1812
        if(h != h0)
1813
            return -1;  // we cant (re-)initialize context during parallel decoding
1814

    
1815
        avcodec_set_dimensions(s->avctx, s->width, s->height);
1816
        s->avctx->sample_aspect_ratio= h->sps.sar;
1817
        av_assert0(s->avctx->sample_aspect_ratio.den);
1818

    
1819
        if(h->sps.video_signal_type_present_flag){
1820
            s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1821
            if(h->sps.colour_description_present_flag){
1822
                s->avctx->color_primaries = h->sps.color_primaries;
1823
                s->avctx->color_trc       = h->sps.color_trc;
1824
                s->avctx->colorspace      = h->sps.colorspace;
1825
            }
1826
        }
1827

    
1828
        if(h->sps.timing_info_present_flag){
1829
            int64_t den= h->sps.time_scale;
1830
            if(h->x264_build < 44U)
1831
                den *= 2;
1832
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1833
                      h->sps.num_units_in_tick, den, 1<<30);
1834
        }
1835
        s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
1836
                                                 s->avctx->codec->pix_fmts ?
1837
                                                 s->avctx->codec->pix_fmts :
1838
                                                 s->avctx->color_range == AVCOL_RANGE_JPEG ?
1839
                                                 hwaccel_pixfmt_list_h264_jpeg_420 :
1840
                                                 ff_hwaccel_pixfmt_list_420);
1841
        s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1842

    
1843
        if (MPV_common_init(s) < 0)
1844
            return -1;
1845
        s->first_field = 0;
1846
        h->prev_interlaced_frame = 1;
1847

    
1848
        init_scan_tables(h);
1849
        ff_h264_alloc_tables(h);
1850

    
1851
        for(i = 1; i < s->avctx->thread_count; i++) {
1852
            H264Context *c;
1853
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1854
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1855
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1856
            c->h264dsp = h->h264dsp;
1857
            c->sps = h->sps;
1858
            c->pps = h->pps;
1859
            init_scan_tables(c);
1860
            clone_tables(c, h, i);
1861
        }
1862

    
1863
        for(i = 0; i < s->avctx->thread_count; i++)
1864
            if(context_init(h->thread_context[i]) < 0)
1865
                return -1;
1866
    }
1867

    
1868
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1869

    
1870
    h->mb_mbaff = 0;
1871
    h->mb_aff_frame = 0;
1872
    last_pic_structure = s0->picture_structure;
1873
    if(h->sps.frame_mbs_only_flag){
1874
        s->picture_structure= PICT_FRAME;
1875
    }else{
1876
        if(get_bits1(&s->gb)) { //field_pic_flag
1877
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1878
        } else {
1879
            s->picture_structure= PICT_FRAME;
1880
            h->mb_aff_frame = h->sps.mb_aff;
1881
        }
1882
    }
1883
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1884

    
1885
    if(h0->current_slice == 0){
1886
        while(h->frame_num !=  h->prev_frame_num &&
1887
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1888
            Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1889
            av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1890
            if (ff_h264_frame_start(h) < 0)
1891
                return -1;
1892
            h->prev_frame_num++;
1893
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1894
            s->current_picture_ptr->frame_num= h->prev_frame_num;
1895
            ff_generate_sliding_window_mmcos(h);
1896
            ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1897
            /* Error concealment: if a ref is missing, copy the previous ref in its place.
1898
             * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
1899
             * about there being no actual duplicates.
1900
             * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
1901
             * concealing a lost frame, this probably isn't noticable by comparison, but it should
1902
             * be fixed. */
1903
            if (h->short_ref_count) {
1904
                if (prev) {
1905
                    av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
1906
                                  (const uint8_t**)prev->data, prev->linesize,
1907
                                  s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
1908
                    h->short_ref[0]->poc = prev->poc+2;
1909
                }
1910
                h->short_ref[0]->frame_num = h->prev_frame_num;
1911
            }
1912
        }
1913

    
1914
        /* See if we have a decoded first field looking for a pair... */
1915
        if (s0->first_field) {
1916
            assert(s0->current_picture_ptr);
1917
            assert(s0->current_picture_ptr->data[0]);
1918
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1919

    
1920
            /* figure out if we have a complementary field pair */
1921
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1922
                /*
1923
                 * Previous field is unmatched. Don't display it, but let it
1924
                 * remain for reference if marked as such.
1925
                 */
1926
                s0->current_picture_ptr = NULL;
1927
                s0->first_field = FIELD_PICTURE;
1928

    
1929
            } else {
1930
                if (h->nal_ref_idc &&
1931
                        s0->current_picture_ptr->reference &&
1932
                        s0->current_picture_ptr->frame_num != h->frame_num) {
1933
                    /*
1934
                     * This and previous field were reference, but had
1935
                     * different frame_nums. Consider this field first in
1936
                     * pair. Throw away previous field except for reference
1937
                     * purposes.
1938
                     */
1939
                    s0->first_field = 1;
1940
                    s0->current_picture_ptr = NULL;
1941

    
1942
                } else {
1943
                    /* Second field in complementary pair */
1944
                    s0->first_field = 0;
1945
                }
1946
            }
1947

    
1948
        } else {
1949
            /* Frame or first field in a potentially complementary pair */
1950
            assert(!s0->current_picture_ptr);
1951
            s0->first_field = FIELD_PICTURE;
1952
        }
1953

    
1954
        if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1955
            s0->first_field = 0;
1956
            return -1;
1957
        }
1958
    }
1959
    if(h != h0)
1960
        clone_slice(h, h0);
1961

    
1962
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1963

    
1964
    assert(s->mb_num == s->mb_width * s->mb_height);
1965
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1966
       first_mb_in_slice                    >= s->mb_num){
1967
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1968
        return -1;
1969
    }
1970
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1971
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1972
    if (s->picture_structure == PICT_BOTTOM_FIELD)
1973
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
1974
    assert(s->mb_y < s->mb_height);
1975

    
1976
    if(s->picture_structure==PICT_FRAME){
1977
        h->curr_pic_num=   h->frame_num;
1978
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1979
    }else{
1980
        h->curr_pic_num= 2*h->frame_num + 1;
1981
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1982
    }
1983

    
1984
    if(h->nal_unit_type == NAL_IDR_SLICE){
1985
        get_ue_golomb(&s->gb); /* idr_pic_id */
1986
    }
1987

    
1988
    if(h->sps.poc_type==0){
1989
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1990

    
1991
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1992
            h->delta_poc_bottom= get_se_golomb(&s->gb);
1993
        }
1994
    }
1995

    
1996
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1997
        h->delta_poc[0]= get_se_golomb(&s->gb);
1998

    
1999
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2000
            h->delta_poc[1]= get_se_golomb(&s->gb);
2001
    }
2002

    
2003
    init_poc(h);
2004

    
2005
    if(h->pps.redundant_pic_cnt_present){
2006
        h->redundant_pic_count= get_ue_golomb(&s->gb);
2007
    }
2008

    
2009
    //set defaults, might be overridden a few lines later
2010
    h->ref_count[0]= h->pps.ref_count[0];
2011
    h->ref_count[1]= h->pps.ref_count[1];
2012

    
2013
    if(h->slice_type_nos != FF_I_TYPE){
2014
        if(h->slice_type_nos == FF_B_TYPE){
2015
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
2016
        }
2017
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
2018

    
2019
        if(num_ref_idx_active_override_flag){
2020
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2021
            if(h->slice_type_nos==FF_B_TYPE)
2022
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2023

    
2024
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
2025
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2026
                h->ref_count[0]= h->ref_count[1]= 1;
2027
                return -1;
2028
            }
2029
        }
2030
        if(h->slice_type_nos == FF_B_TYPE)
2031
            h->list_count= 2;
2032
        else
2033
            h->list_count= 1;
2034
    }else
2035
        h->list_count= 0;
2036

    
2037
    if(!default_ref_list_done){
2038
        ff_h264_fill_default_ref_list(h);
2039
    }
2040

    
2041
    if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
2042
        return -1;
2043

    
2044
    if(h->slice_type_nos!=FF_I_TYPE){
2045
        s->last_picture_ptr= &h->ref_list[0][0];
2046
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2047
    }
2048
    if(h->slice_type_nos==FF_B_TYPE){
2049
        s->next_picture_ptr= &h->ref_list[1][0];
2050
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2051
    }
2052

    
2053
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2054
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2055
        pred_weight_table(h);
2056
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2057
        implicit_weight_table(h, -1);
2058
    }else {
2059
        h->use_weight = 0;
2060
        for (i = 0; i < 2; i++) {
2061
            h->luma_weight_flag[i]   = 0;
2062
            h->chroma_weight_flag[i] = 0;
2063
        }
2064
    }
2065

    
2066
    if(h->nal_ref_idc)
2067
        ff_h264_decode_ref_pic_marking(h0, &s->gb);
2068

    
2069
    if(FRAME_MBAFF){
2070
        ff_h264_fill_mbaff_ref_list(h);
2071

    
2072
        if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
2073
            implicit_weight_table(h, 0);
2074
            implicit_weight_table(h, 1);
2075
        }
2076
    }
2077

    
2078
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2079
        ff_h264_direct_dist_scale_factor(h);
2080
    ff_h264_direct_ref_list_init(h);
2081

    
2082
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2083
        tmp = get_ue_golomb_31(&s->gb);
2084
        if(tmp > 2){
2085
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2086
            return -1;
2087
        }
2088
        h->cabac_init_idc= tmp;
2089
    }
2090

    
2091
    h->last_qscale_diff = 0;
2092
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2093
    if(tmp>51){
2094
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2095
        return -1;
2096
    }
2097
    s->qscale= tmp;
2098
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2099
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2100
    //FIXME qscale / qp ... stuff
2101
    if(h->slice_type == FF_SP_TYPE){
2102
        get_bits1(&s->gb); /* sp_for_switch_flag */
2103
    }
2104
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2105
        get_se_golomb(&s->gb); /* slice_qs_delta */
2106
    }
2107

    
2108
    h->deblocking_filter = 1;
2109
    h->slice_alpha_c0_offset = 52;
2110
    h->slice_beta_offset = 52;
2111
    if( h->pps.deblocking_filter_parameters_present ) {
2112
        tmp= get_ue_golomb_31(&s->gb);
2113
        if(tmp > 2){
2114
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2115
            return -1;
2116
        }
2117
        h->deblocking_filter= tmp;
2118
        if(h->deblocking_filter < 2)
2119
            h->deblocking_filter^= 1; // 1<->0
2120

    
2121
        if( h->deblocking_filter ) {
2122
            h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2123
            h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2124
            if(   h->slice_alpha_c0_offset > 104U
2125
               || h->slice_beta_offset     > 104U){
2126
                av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
2127
                return -1;
2128
            }
2129
        }
2130
    }
2131

    
2132
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2133
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2134
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2135
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2136
        h->deblocking_filter= 0;
2137

    
2138
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2139
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2140
            /* Cheat slightly for speed:
2141
               Do not bother to deblock across slices. */
2142
            h->deblocking_filter = 2;
2143
        } else {
2144
            h0->max_contexts = 1;
2145
            if(!h0->single_decode_warning) {
2146
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2147
                h0->single_decode_warning = 1;
2148
            }
2149
            if(h != h0)
2150
                return 1; // deblocking switched inside frame
2151
        }
2152
    }
2153
    h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
2154

    
2155
#if 0 //FMO
2156
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2157
        slice_group_change_cycle= get_bits(&s->gb, ?);
2158
#endif
2159

    
2160
    h0->last_slice_type = slice_type;
2161
    h->slice_num = ++h0->current_slice;
2162
    if(h->slice_num >= MAX_SLICES){
2163
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2164
    }
2165

    
2166
    for(j=0; j<2; j++){
2167
        int id_list[16];
2168
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2169
        for(i=0; i<16; i++){
2170
            id_list[i]= 60;
2171
            if(h->ref_list[j][i].data[0]){
2172
                int k;
2173
                uint8_t *base= h->ref_list[j][i].base[0];
2174
                for(k=0; k<h->short_ref_count; k++)
2175
                    if(h->short_ref[k]->base[0] == base){
2176
                        id_list[i]= k;
2177
                        break;
2178
                    }
2179
                for(k=0; k<h->long_ref_count; k++)
2180
                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2181
                        id_list[i]= h->short_ref_count + k;
2182
                        break;
2183
                    }
2184
            }
2185
        }
2186

    
2187
        ref2frm[0]=
2188
        ref2frm[1]= -1;
2189
        for(i=0; i<16; i++)
2190
            ref2frm[i+2]= 4*id_list[i]
2191
                          +(h->ref_list[j][i].reference&3);
2192
        ref2frm[18+0]=
2193
        ref2frm[18+1]= -1;
2194
        for(i=16; i<48; i++)
2195
            ref2frm[i+4]= 4*id_list[(i-16)>>1]
2196
                          +(h->ref_list[j][i].reference&3);
2197
    }
2198

    
2199
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2200
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2201

    
2202
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2203
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
2204
               h->slice_num,
2205
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2206
               first_mb_in_slice,
2207
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2208
               pps_id, h->frame_num,
2209
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2210
               h->ref_count[0], h->ref_count[1],
2211
               s->qscale,
2212
               h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2213
               h->use_weight,
2214
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2215
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2216
               );
2217
    }
2218

    
2219
    return 0;
2220
}
2221

    
2222
int ff_h264_get_slice_type(const H264Context *h)
2223
{
2224
    switch (h->slice_type) {
2225
    case FF_P_TYPE:  return 0;
2226
    case FF_B_TYPE:  return 1;
2227
    case FF_I_TYPE:  return 2;
2228
    case FF_SP_TYPE: return 3;
2229
    case FF_SI_TYPE: return 4;
2230
    default:         return -1;
2231
    }
2232
}
2233

    
2234
/**
2235
 *
2236
 * @return non zero if the loop filter can be skiped
2237
 */
2238
static int fill_filter_caches(H264Context *h, int mb_type){
2239
    MpegEncContext * const s = &h->s;
2240
    const int mb_xy= h->mb_xy;
2241
    int top_xy, left_xy[2];
2242
    int top_type, left_type[2];
2243

    
2244
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
2245

    
2246
    //FIXME deblocking could skip the intra and nnz parts.
2247

    
2248
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
2249
     * stuff, I can't imagine that these complex rules are worth it. */
2250

    
2251
    left_xy[1] = left_xy[0] = mb_xy-1;
2252
    if(FRAME_MBAFF){
2253
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
2254
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
2255
        if(s->mb_y&1){
2256
            if (left_mb_field_flag != curr_mb_field_flag) {
2257
                left_xy[0] -= s->mb_stride;
2258
            }
2259
        }else{
2260
            if(curr_mb_field_flag){
2261
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
2262
            }
2263
            if (left_mb_field_flag != curr_mb_field_flag) {
2264
                left_xy[1] += s->mb_stride;
2265
            }
2266
        }
2267
    }
2268

    
2269
    h->top_mb_xy = top_xy;
2270
    h->left_mb_xy[0] = left_xy[0];
2271
    h->left_mb_xy[1] = left_xy[1];
2272
    {
2273
        //for sufficiently low qp, filtering wouldn't do anything
2274
        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
2275
        int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
2276
        int qp = s->current_picture.qscale_table[mb_xy];
2277
        if(qp <= qp_thresh
2278
           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
2279
           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
2280
            if(!FRAME_MBAFF)
2281
                return 1;
2282
            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
2283
               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
2284
                return 1;
2285
        }
2286
    }
2287

    
2288
    top_type     = s->current_picture.mb_type[top_xy]    ;
2289
    left_type[0] = s->current_picture.mb_type[left_xy[0]];
2290
    left_type[1] = s->current_picture.mb_type[left_xy[1]];
2291
    if(h->deblocking_filter == 2){
2292
        if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
2293
        if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
2294
    }else{
2295
        if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
2296
        if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
2297
    }
2298
    h->top_type    = top_type    ;
2299
    h->left_type[0]= left_type[0];
2300
    h->left_type[1]= left_type[1];
2301

    
2302
    if(IS_INTRA(mb_type))
2303
        return 0;
2304

    
2305
    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
2306
    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
2307
    AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
2308
    AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
2309
    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
2310

    
2311
    h->cbp= h->cbp_table[mb_xy];
2312

    
2313
    {
2314
        int list;
2315
        for(list=0; list<h->list_count; list++){
2316
            int8_t *ref;
2317
            int y, b_stride;
2318
            int16_t (*mv_dst)[2];
2319
            int16_t (*mv_src)[2];
2320

    
2321
            if(!USES_LIST(mb_type, list)){
2322
                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
2323
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2324
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2325
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2326
                AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2327
                continue;
2328
            }
2329

    
2330
            ref = &s->current_picture.ref_index[list][4*mb_xy];
2331
            {
2332
                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2333
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2334
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2335
                ref += 2;
2336
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2337
                AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
2338
            }
2339

    
2340
            b_stride = h->b_stride;
2341
            mv_dst   = &h->mv_cache[list][scan8[0]];
2342
            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
2343
            for(y=0; y<4; y++){
2344
                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
2345
            }
2346

    
2347
        }
2348
    }
2349

    
2350

    
2351
/*
2352
0 . T T. T T T T
2353
1 L . .L . . . .
2354
2 L . .L . . . .
2355
3 . T TL . . . .
2356
4 L . .L . . . .
2357
5 L . .. . . . .
2358
*/
2359
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
2360
    if(top_type){
2361
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
2362
    }
2363

    
2364
    if(left_type[0]){
2365
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
2366
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
2367
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
2368
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
2369
    }
2370

    
2371
    // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
2372
    if(!CABAC && h->pps.transform_8x8_mode){
2373
        if(IS_8x8DCT(top_type)){
2374
            h->non_zero_count_cache[4+8*0]=
2375
            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
2376
            h->non_zero_count_cache[6+8*0]=
2377
            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
2378
        }
2379
        if(IS_8x8DCT(left_type[0])){
2380
            h->non_zero_count_cache[3+8*1]=
2381
            h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
2382
        }
2383
        if(IS_8x8DCT(left_type[1])){
2384
            h->non_zero_count_cache[3+8*3]=
2385
            h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
2386
        }
2387

    
2388
        if(IS_8x8DCT(mb_type)){
2389
            h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
2390
            h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
2391

    
2392
            h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
2393
            h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
2394

    
2395
            h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
2396
            h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
2397

    
2398
            h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
2399
            h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
2400
        }
2401
    }
2402

    
2403
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
2404
        int list;
2405
        for(list=0; list<h->list_count; list++){
2406
            if(USES_LIST(top_type, list)){
2407
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
2408
                const int b8_xy= 4*top_xy + 2;
2409
                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2410
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
2411
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
2412
                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
2413
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
2414
                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
2415
            }else{
2416
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
2417
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
2418
            }
2419

    
2420
            if(!IS_INTERLACED(mb_type^left_type[0])){
2421
                if(USES_LIST(left_type[0], list)){
2422
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
2423
                    const int b8_xy= 4*left_xy[0] + 1;
2424
                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
2425
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
2426
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
2427
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
2428
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
2429
                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
2430
                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
2431
                    h->ref_cache[list][scan8[0] - 1 +16 ]=
2432
                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
2433
                }else{
2434
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
2435
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
2436
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
2437
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
2438
                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
2439
                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
2440
                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
2441
                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
2442
                }
2443
            }
2444
        }
2445
    }
2446

    
2447
    return 0;
2448
}
2449

    
2450
static void loop_filter(H264Context *h){
2451
    MpegEncContext * const s = &h->s;
2452
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2453
    int linesize, uvlinesize, mb_x, mb_y;
2454
    const int end_mb_y= s->mb_y + FRAME_MBAFF;
2455
    const int old_slice_type= h->slice_type;
2456

    
2457
    if(h->deblocking_filter) {
2458
        for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2459
            for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2460
                int mb_xy, mb_type;
2461
                mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2462
                h->slice_num= h->slice_table[mb_xy];
2463
                mb_type= s->current_picture.mb_type[mb_xy];
2464
                h->list_count= h->list_counts[mb_xy];
2465

    
2466
                if(FRAME_MBAFF)
2467
                    h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2468

    
2469
                s->mb_x= mb_x;
2470
                s->mb_y= mb_y;
2471
                dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2472
                dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2473
                dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2474
                    //FIXME simplify above
2475

    
2476
                if (MB_FIELD) {
2477
                    linesize   = h->mb_linesize   = s->linesize * 2;
2478
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2479
                    if(mb_y&1){ //FIXME move out of this function?
2480
                        dest_y -= s->linesize*15;
2481
                        dest_cb-= s->uvlinesize*7;
2482
                        dest_cr-= s->uvlinesize*7;
2483
                    }
2484
                } else {
2485
                    linesize   = h->mb_linesize   = s->linesize;
2486
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2487
                }
2488
                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2489
                if(fill_filter_caches(h, mb_type))
2490
                    continue;
2491
                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2492
                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2493

    
2494
                if (FRAME_MBAFF) {
2495
                    ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2496
                } else {
2497
                    ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2498
                }
2499
            }
2500
        }
2501
    }
2502
    h->slice_type= old_slice_type;
2503
    s->mb_x= 0;
2504
    s->mb_y= end_mb_y - FRAME_MBAFF;
2505
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2506
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2507
}
2508

    
2509
static void predict_field_decoding_flag(H264Context *h){
2510
    MpegEncContext * const s = &h->s;
2511
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2512
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2513
                ? s->current_picture.mb_type[mb_xy-1]
2514
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2515
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2516
                : 0;
2517
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2518
}
2519

    
2520
static int decode_slice(struct AVCodecContext *avctx, void *arg){
2521
    H264Context *h = *(void**)arg;
2522
    MpegEncContext * const s = &h->s;
2523
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2524

    
2525
    s->mb_skip_run= -1;
2526

    
2527
    h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2528
                    (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2529

    
2530
    if( h->pps.cabac ) {
2531
        /* realign */
2532
        align_get_bits( &s->gb );
2533

    
2534
        /* init cabac */
2535
        ff_init_cabac_states( &h->cabac);
2536
        ff_init_cabac_decoder( &h->cabac,
2537
                               s->gb.buffer + get_bits_count(&s->gb)/8,
2538
                               (get_bits_left(&s->gb) + 7)/8);
2539

    
2540
        ff_h264_init_cabac_states(h);
2541

    
2542
        for(;;){
2543
//START_TIMER
2544
            int ret = ff_h264_decode_mb_cabac(h);
2545
            int eos;
2546
//STOP_TIMER("decode_mb_cabac")
2547

    
2548
            if(ret>=0) ff_h264_hl_decode_mb(h);
2549

    
2550
            if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2551
                s->mb_y++;
2552

    
2553
                ret = ff_h264_decode_mb_cabac(h);
2554

    
2555
                if(ret>=0) ff_h264_hl_decode_mb(h);
2556
                s->mb_y--;
2557
            }
2558
            eos = get_cabac_terminate( &h->cabac );
2559

    
2560
            if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2561
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2562
                return 0;
2563
            }
2564
            if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2565
                av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
2566
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2567
                return -1;
2568
            }
2569

    
2570
            if( ++s->mb_x >= s->mb_width ) {
2571
                s->mb_x = 0;
2572
                loop_filter(h);
2573
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2574
                ++s->mb_y;
2575
                if(FIELD_OR_MBAFF_PICTURE) {
2576
                    ++s->mb_y;
2577
                    if(FRAME_MBAFF && s->mb_y < s->mb_height)
2578
                        predict_field_decoding_flag(h);
2579
                }
2580
            }
2581

    
2582
            if( eos || s->mb_y >= s->mb_height ) {
2583
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2584
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2585
                return 0;
2586
            }
2587
        }
2588

    
2589
    } else {
2590
        for(;;){
2591
            int ret = ff_h264_decode_mb_cavlc(h);
2592

    
2593
            if(ret>=0) ff_h264_hl_decode_mb(h);
2594

    
2595
            if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2596
                s->mb_y++;
2597
                ret = ff_h264_decode_mb_cavlc(h);
2598

    
2599
                if(ret>=0) ff_h264_hl_decode_mb(h);
2600
                s->mb_y--;
2601
            }
2602

    
2603
            if(ret<0){
2604
                av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2605
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2606

    
2607
                return -1;
2608
            }
2609

    
2610
            if(++s->mb_x >= s->mb_width){
2611
                s->mb_x=0;
2612
                loop_filter(h);
2613
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2614
                ++s->mb_y;
2615
                if(FIELD_OR_MBAFF_PICTURE) {
2616
                    ++s->mb_y;
2617
                    if(FRAME_MBAFF && s->mb_y < s->mb_height)
2618
                        predict_field_decoding_flag(h);
2619
                }
2620
                if(s->mb_y >= s->mb_height){
2621
                    tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2622

    
2623
                    if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2624
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2625

    
2626
                        return 0;
2627
                    }else{
2628
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2629

    
2630
                        return -1;
2631
                    }
2632
                }
2633
            }
2634

    
2635
            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2636
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2637
                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2638
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2639

    
2640
                    return 0;
2641
                }else{
2642
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2643

    
2644
                    return -1;
2645
                }
2646
            }
2647
        }
2648
    }
2649

    
2650
#if 0
2651
    for(;s->mb_y < s->mb_height; s->mb_y++){
2652
        for(;s->mb_x < s->mb_width; s->mb_x++){
2653
            int ret= decode_mb(h);
2654

2655
            ff_h264_hl_decode_mb(h);
2656

2657
            if(ret<0){
2658
                av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2659
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2660

2661
                return -1;
2662
            }
2663

2664
            if(++s->mb_x >= s->mb_width){
2665
                s->mb_x=0;
2666
                if(++s->mb_y >= s->mb_height){
2667
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
2668
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2669

2670
                        return 0;
2671
                    }else{
2672
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2673

2674
                        return -1;
2675
                    }
2676
                }
2677
            }
2678

2679
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2680
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
2681
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2682

2683
                    return 0;
2684
                }else{
2685
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
2686

2687
                    return -1;
2688
                }
2689
            }
2690
        }
2691
        s->mb_x=0;
2692
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
2693
    }
2694
#endif
2695
    return -1; //not reached
2696
}
2697

    
2698
/**
2699
 * Call decode_slice() for each context.
2700
 *
2701
 * @param h h264 master context
2702
 * @param context_count number of contexts to execute
2703
 */
2704
static void execute_decode_slices(H264Context *h, int context_count){
2705
    MpegEncContext * const s = &h->s;
2706
    AVCodecContext * const avctx= s->avctx;
2707
    H264Context *hx;
2708
    int i;
2709

    
2710
    if (s->avctx->hwaccel)
2711
        return;
2712
    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2713
        return;
2714
    if(context_count == 1) {
2715
        decode_slice(avctx, &h);
2716
    } else {
2717
        for(i = 1; i < context_count; i++) {
2718
            hx = h->thread_context[i];
2719
            hx->s.error_recognition = avctx->error_recognition;
2720
            hx->s.error_count = 0;
2721
        }
2722

    
2723
        avctx->execute(avctx, (void *)decode_slice,
2724
                       h->thread_context, NULL, context_count, sizeof(void*));
2725

    
2726
        /* pull back stuff from slices to master context */
2727
        hx = h->thread_context[context_count - 1];
2728
        s->mb_x = hx->s.mb_x;
2729
        s->mb_y = hx->s.mb_y;
2730
        s->dropable = hx->s.dropable;
2731
        s->picture_structure = hx->s.picture_structure;
2732
        for(i = 1; i < context_count; i++)
2733
            h->s.error_count += h->thread_context[i]->s.error_count;
2734
    }
2735
}
2736

    
2737

    
2738
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2739
    MpegEncContext * const s = &h->s;
2740
    AVCodecContext * const avctx= s->avctx;
2741
    int buf_index=0;
2742
    H264Context *hx; ///< thread context
2743
    int context_count = 0;
2744
    int next_avc= h->is_avc ? 0 : buf_size;
2745

    
2746
    h->max_contexts = avctx->thread_count;
2747
#if 0
2748
    int i;
2749
    for(i=0; i<50; i++){
2750
        av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2751
    }
2752
#endif
2753
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2754
        h->current_slice = 0;
2755
        if (!s->first_field)
2756
            s->current_picture_ptr= NULL;
2757
        ff_h264_reset_sei(h);
2758
    }
2759

    
2760
    for(;;){
2761
        int consumed;
2762
        int dst_length;
2763
        int bit_length;
2764
        const uint8_t *ptr;
2765
        int i, nalsize = 0;
2766
        int err;
2767

    
2768
        if(buf_index >= next_avc) {
2769
            if(buf_index >= buf_size) break;
2770
            nalsize = 0;
2771
            for(i = 0; i < h->nal_length_size; i++)
2772
                nalsize = (nalsize << 8) | buf[buf_index++];
2773
            if(nalsize <= 0 || nalsize > buf_size - buf_index){
2774
                av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2775
                break;
2776
            }
2777
            next_avc= buf_index + nalsize;
2778
        } else {
2779
            // start code prefix search
2780
            for(; buf_index + 3 < next_avc; buf_index++){
2781
                // This should always succeed in the first iteration.
2782
                if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2783
                    break;
2784
            }
2785

    
2786
            if(buf_index+3 >= buf_size) break;
2787

    
2788
            buf_index+=3;
2789
            if(buf_index >= next_avc) continue;
2790
        }
2791

    
2792
        hx = h->thread_context[context_count];
2793

    
2794
        ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2795
        if (ptr==NULL || dst_length < 0){
2796
            return -1;
2797
        }
2798
        i= buf_index + consumed;
2799
        if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2800
           buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2801
            s->workaround_bugs |= FF_BUG_TRUNCATED;
2802

    
2803
        if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2804
        while(ptr[dst_length - 1] == 0 && dst_length > 0)
2805
            dst_length--;
2806
        }
2807
        bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2808

    
2809
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
2810
            av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
2811
        }
2812

    
2813
        if (h->is_avc && (nalsize != consumed) && nalsize){
2814
            av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2815
        }
2816

    
2817
        buf_index += consumed;
2818

    
2819
        if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2820
           ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2821
            continue;
2822

    
2823
      again:
2824
        err = 0;
2825
        switch(hx->nal_unit_type){
2826
        case NAL_IDR_SLICE:
2827
            if (h->nal_unit_type != NAL_IDR_SLICE) {
2828
                av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2829
                return -1;
2830
            }
2831
            idr(h); //FIXME ensure we don't loose some frames if there is reordering
2832
        case NAL_SLICE:
2833
            init_get_bits(&hx->s.gb, ptr, bit_length);
2834
            hx->intra_gb_ptr=
2835
            hx->inter_gb_ptr= &hx->s.gb;
2836
            hx->s.data_partitioning = 0;
2837

    
2838
            if((err = decode_slice_header(hx, h)))
2839
               break;
2840

    
2841
            if (h->current_slice == 1) {
2842
                if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2843
                    return -1;
2844
                if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2845
                    ff_vdpau_h264_picture_start(s);
2846
            }
2847

    
2848
            s->current_picture_ptr->key_frame |=
2849
                    (hx->nal_unit_type == NAL_IDR_SLICE) ||
2850
                    (h->sei_recovery_frame_cnt >= 0);
2851
            if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2852
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2853
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2854
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2855
               && avctx->skip_frame < AVDISCARD_ALL){
2856
                if(avctx->hwaccel) {
2857
                    if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2858
                        return -1;
2859
                }else
2860
                if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2861
                    static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2862
                    ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2863
                    ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2864
                }else
2865
                    context_count++;
2866
            }
2867
            break;
2868
        case NAL_DPA:
2869
            init_get_bits(&hx->s.gb, ptr, bit_length);
2870
            hx->intra_gb_ptr=
2871
            hx->inter_gb_ptr= NULL;
2872

    
2873
            if ((err = decode_slice_header(hx, h)) < 0)
2874
                break;
2875

    
2876
            hx->s.data_partitioning = 1;
2877

    
2878
            break;
2879
        case NAL_DPB:
2880
            init_get_bits(&hx->intra_gb, ptr, bit_length);
2881
            hx->intra_gb_ptr= &hx->intra_gb;
2882
            break;
2883
        case NAL_DPC:
2884
            init_get_bits(&hx->inter_gb, ptr, bit_length);
2885
            hx->inter_gb_ptr= &hx->inter_gb;
2886

    
2887
            if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2888
               && s->context_initialized
2889
               && s->hurry_up < 5
2890
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2891
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2892
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2893
               && avctx->skip_frame < AVDISCARD_ALL)
2894
                context_count++;
2895
            break;
2896
        case NAL_SEI:
2897
            init_get_bits(&s->gb, ptr, bit_length);
2898
            ff_h264_decode_sei(h);
2899
            break;
2900
        case NAL_SPS:
2901
            init_get_bits(&s->gb, ptr, bit_length);
2902
            ff_h264_decode_seq_parameter_set(h);
2903

    
2904
            if(s->flags& CODEC_FLAG_LOW_DELAY)
2905
                s->low_delay=1;
2906

    
2907
            if(avctx->has_b_frames < 2)
2908
                avctx->has_b_frames= !s->low_delay;
2909
            break;
2910
        case NAL_PPS:
2911
            init_get_bits(&s->gb, ptr, bit_length);
2912

    
2913
            ff_h264_decode_picture_parameter_set(h, bit_length);
2914

    
2915
            break;
2916
        case NAL_AUD:
2917
        case NAL_END_SEQUENCE:
2918
        case NAL_END_STREAM:
2919
        case NAL_FILLER_DATA:
2920
        case NAL_SPS_EXT:
2921
        case NAL_AUXILIARY_SLICE:
2922
            break;
2923
        default:
2924
            av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2925
        }
2926

    
2927
        if(context_count == h->max_contexts) {
2928
            execute_decode_slices(h, context_count);
2929
            context_count = 0;
2930
        }
2931

    
2932
        if (err < 0)
2933
            av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2934
        else if(err == 1) {
2935
            /* Slice could not be decoded in parallel mode, copy down
2936
             * NAL unit stuff to context 0 and restart. Note that
2937
             * rbsp_buffer is not transferred, but since we no longer
2938
             * run in parallel mode this should not be an issue. */
2939
            h->nal_unit_type = hx->nal_unit_type;
2940
            h->nal_ref_idc   = hx->nal_ref_idc;
2941
            hx = h;
2942
            goto again;
2943
        }
2944
    }
2945
    if(context_count)
2946
        execute_decode_slices(h, context_count);
2947
    return buf_index;
2948
}
2949

    
2950
/**
2951
 * returns the number of bytes consumed for building the current frame
2952
 */
2953
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2954
        if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2955
        if(pos+10>buf_size) pos=buf_size; // oops ;)
2956

    
2957
        return pos;
2958
}
2959

    
2960
static int decode_frame(AVCodecContext *avctx,
2961
                             void *data, int *data_size,
2962
                             AVPacket *avpkt)
2963
{
2964
    const uint8_t *buf = avpkt->data;
2965
    int buf_size = avpkt->size;
2966
    H264Context *h = avctx->priv_data;
2967
    MpegEncContext *s = &h->s;
2968
    AVFrame *pict = data;
2969
    int buf_index;
2970

    
2971
    s->flags= avctx->flags;
2972
    s->flags2= avctx->flags2;
2973

    
2974
   /* end of stream, output what is still in the buffers */
2975
 out:
2976
    if (buf_size == 0) {
2977
        Picture *out;
2978
        int i, out_idx;
2979

    
2980
//FIXME factorize this with the output code below
2981
        out = h->delayed_pic[0];
2982
        out_idx = 0;
2983
        for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2984
            if(h->delayed_pic[i]->poc < out->poc){
2985
                out = h->delayed_pic[i];
2986
                out_idx = i;
2987
            }
2988

    
2989
        for(i=out_idx; h->delayed_pic[i]; i++)
2990
            h->delayed_pic[i] = h->delayed_pic[i+1];
2991

    
2992
        if(out){
2993
            *data_size = sizeof(AVFrame);
2994
            *pict= *(AVFrame*)out;
2995
        }
2996

    
2997
        return 0;
2998
    }
2999

    
3000
    buf_index=decode_nal_units(h, buf, buf_size);
3001
    if(buf_index < 0)
3002
        return -1;
3003

    
3004
    if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
3005
        buf_size = 0;
3006
        goto out;
3007
    }
3008

    
3009
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
3010
        if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
3011
        av_log(avctx, AV_LOG_ERROR, "no frame!\n");
3012
        return -1;
3013
    }
3014

    
3015
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
3016
        Picture *out = s->current_picture_ptr;
3017
        Picture *cur = s->current_picture_ptr;
3018
        int i, pics, out_of_order, out_idx;
3019

    
3020
        field_end(h);
3021

    
3022
        if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
3023
            /* Wait for second field. */
3024
            *data_size = 0;
3025

    
3026
        } else {
3027
            cur->interlaced_frame = 0;
3028
            cur->repeat_pict = 0;
3029

    
3030
            /* Signal interlacing information externally. */
3031
            /* Prioritize picture timing SEI information over used decoding process if it exists. */
3032

    
3033
            if(h->sps.pic_struct_present_flag){
3034
                switch (h->sei_pic_struct)
3035
                {
3036
                case SEI_PIC_STRUCT_FRAME:
3037
                    break;
3038
                case SEI_PIC_STRUCT_TOP_FIELD:
3039
                case SEI_PIC_STRUCT_BOTTOM_FIELD:
3040
                    cur->interlaced_frame = 1;
3041
                    break;
3042
                case SEI_PIC_STRUCT_TOP_BOTTOM:
3043
                case SEI_PIC_STRUCT_BOTTOM_TOP:
3044
                    if (FIELD_OR_MBAFF_PICTURE)
3045
                        cur->interlaced_frame = 1;
3046
                    else
3047
                        // try to flag soft telecine progressive
3048
                        cur->interlaced_frame = h->prev_interlaced_frame;
3049
                    break;
3050
                case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
3051
                case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
3052
                    // Signal the possibility of telecined film externally (pic_struct 5,6)
3053
                    // From these hints, let the applications decide if they apply deinterlacing.
3054
                    cur->repeat_pict = 1;
3055
                    break;
3056
                case SEI_PIC_STRUCT_FRAME_DOUBLING:
3057
                    // Force progressive here, as doubling interlaced frame is a bad idea.
3058
                    cur->repeat_pict = 2;
3059
                    break;
3060
                case SEI_PIC_STRUCT_FRAME_TRIPLING:
3061
                    cur->repeat_pict = 4;
3062
                    break;
3063
                }
3064

    
3065
                if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
3066
                    cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
3067
            }else{
3068
                /* Derive interlacing flag from used decoding process. */
3069
                cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
3070
            }
3071
            h->prev_interlaced_frame = cur->interlaced_frame;
3072

    
3073
            if (cur->field_poc[0] != cur->field_poc[1]){
3074
                /* Derive top_field_first from field pocs. */
3075
                cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
3076
            }else{
3077
                if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
3078
                    /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
3079
                    if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
3080
                      || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
3081
                        cur->top_field_first = 1;
3082
                    else
3083
                        cur->top_field_first = 0;
3084
                }else{
3085
                    /* Most likely progressive */
3086
                    cur->top_field_first = 0;
3087
                }
3088
            }
3089

    
3090
        //FIXME do something with unavailable reference frames
3091

    
3092
            /* Sort B-frames into display order */
3093

    
3094
            if(h->sps.bitstream_restriction_flag
3095
               && s->avctx->has_b_frames < h->sps.num_reorder_frames){
3096
                s->avctx->has_b_frames = h->sps.num_reorder_frames;
3097
                s->low_delay = 0;
3098
            }
3099

    
3100
            if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
3101
               && !h->sps.bitstream_restriction_flag){
3102
                s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
3103
                s->low_delay= 0;
3104
            }
3105

    
3106
            pics = 0;
3107
            while(h->delayed_pic[pics]) pics++;
3108

    
3109
            assert(pics <= MAX_DELAYED_PIC_COUNT);
3110

    
3111
            h->delayed_pic[pics++] = cur;
3112
            if(cur->reference == 0)
3113
                cur->reference = DELAYED_PIC_REF;
3114

    
3115
            out = h->delayed_pic[0];
3116
            out_idx = 0;
3117
            for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
3118
                if(h->delayed_pic[i]->poc < out->poc){
3119
                    out = h->delayed_pic[i];
3120
                    out_idx = i;
3121
                }
3122
            if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
3123
                h->outputed_poc= INT_MIN;
3124
            out_of_order = out->poc < h->outputed_poc;
3125

    
3126
            if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
3127
                { }
3128
            else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
3129
               || (s->low_delay &&
3130
                ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
3131
                 || cur->pict_type == FF_B_TYPE)))
3132
            {
3133
                s->low_delay = 0;
3134
                s->avctx->has_b_frames++;
3135
            }
3136

    
3137
            if(out_of_order || pics > s->avctx->has_b_frames){
3138
                out->reference &= ~DELAYED_PIC_REF;
3139
                for(i=out_idx; h->delayed_pic[i]; i++)
3140
                    h->delayed_pic[i] = h->delayed_pic[i+1];
3141
            }
3142
            if(!out_of_order && pics > s->avctx->has_b_frames){
3143
                *data_size = sizeof(AVFrame);
3144

    
3145
                if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
3146
                    h->outputed_poc = INT_MIN;
3147
                } else
3148
                    h->outputed_poc = out->poc;
3149
                *pict= *(AVFrame*)out;
3150
            }else{
3151
                av_log(avctx, AV_LOG_DEBUG, "no picture\n");
3152
            }
3153
        }
3154
    }
3155

    
3156
    assert(pict->data[0] || !*data_size);
3157
    ff_print_debug_info(s, pict);
3158
//printf("out %d\n", (int)pict->data[0]);
3159

    
3160
    return get_consumed_bytes(s, buf_index, buf_size);
3161
}
3162
#if 0
3163
static inline void fill_mb_avail(H264Context *h){
3164
    MpegEncContext * const s = &h->s;
3165
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3166

3167
    if(s->mb_y){
3168
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
3169
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
3170
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
3171
    }else{
3172
        h->mb_avail[0]=
3173
        h->mb_avail[1]=
3174
        h->mb_avail[2]= 0;
3175
    }
3176
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
3177
    h->mb_avail[4]= 1; //FIXME move out
3178
    h->mb_avail[5]= 0; //FIXME move out
3179
}
3180
#endif
3181

    
3182
#ifdef TEST
3183
#undef printf
3184
#undef random
3185
#define COUNT 8000
3186
#define SIZE (COUNT*40)
3187
int main(void){
3188
    int i;
3189
    uint8_t temp[SIZE];
3190
    PutBitContext pb;
3191
    GetBitContext gb;
3192
//    int int_temp[10000];
3193
    DSPContext dsp;
3194
    AVCodecContext avctx;
3195

    
3196
    dsputil_init(&dsp, &avctx);
3197

    
3198
    init_put_bits(&pb, temp, SIZE);
3199
    printf("testing unsigned exp golomb\n");
3200
    for(i=0; i<COUNT; i++){
3201
        START_TIMER
3202
        set_ue_golomb(&pb, i);
3203
        STOP_TIMER("set_ue_golomb");
3204
    }
3205
    flush_put_bits(&pb);
3206

    
3207
    init_get_bits(&gb, temp, 8*SIZE);
3208
    for(i=0; i<COUNT; i++){
3209
        int j, s;
3210

    
3211
        s= show_bits(&gb, 24);
3212

    
3213
        START_TIMER
3214
        j= get_ue_golomb(&gb);
3215
        if(j != i){
3216
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3217
//            return -1;
3218
        }
3219
        STOP_TIMER("get_ue_golomb");
3220
    }
3221

    
3222

    
3223
    init_put_bits(&pb, temp, SIZE);
3224
    printf("testing signed exp golomb\n");
3225
    for(i=0; i<COUNT; i++){
3226
        START_TIMER
3227
        set_se_golomb(&pb, i - COUNT/2);
3228
        STOP_TIMER("set_se_golomb");
3229
    }
3230
    flush_put_bits(&pb);
3231

    
3232
    init_get_bits(&gb, temp, 8*SIZE);
3233
    for(i=0; i<COUNT; i++){
3234
        int j, s;
3235

    
3236
        s= show_bits(&gb, 24);
3237

    
3238
        START_TIMER
3239
        j= get_se_golomb(&gb);
3240
        if(j != i - COUNT/2){
3241
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
3242
//            return -1;
3243
        }
3244
        STOP_TIMER("get_se_golomb");
3245
    }
3246

    
3247
#if 0
3248
    printf("testing 4x4 (I)DCT\n");
3249

3250
    DCTELEM block[16];
3251
    uint8_t src[16], ref[16];
3252
    uint64_t error= 0, max_error=0;
3253

3254
    for(i=0; i<COUNT; i++){
3255
        int j;
3256
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
3257
        for(j=0; j<16; j++){
3258
            ref[j]= random()%255;
3259
            src[j]= random()%255;
3260
        }
3261

3262
        h264_diff_dct_c(block, src, ref, 4);
3263

3264
        //normalize
3265
        for(j=0; j<16; j++){
3266
//            printf("%d ", block[j]);
3267
            block[j]= block[j]*4;
3268
            if(j&1) block[j]= (block[j]*4 + 2)/5;
3269
            if(j&4) block[j]= (block[j]*4 + 2)/5;
3270
        }
3271
//        printf("\n");
3272

3273
        h->h264dsp.h264_idct_add(ref, block, 4);
3274
/*        for(j=0; j<16; j++){
3275
            printf("%d ", ref[j]);
3276
        }
3277
        printf("\n");*/
3278

3279
        for(j=0; j<16; j++){
3280
            int diff= FFABS(src[j] - ref[j]);
3281

3282
            error+= diff*diff;
3283
            max_error= FFMAX(max_error, diff);
3284
        }
3285
    }
3286
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3287
    printf("testing quantizer\n");
3288
    for(qp=0; qp<52; qp++){
3289
        for(i=0; i<16; i++)
3290
            src1_block[i]= src2_block[i]= random()%255;
3291

3292
    }
3293
    printf("Testing NAL layer\n");
3294

3295
    uint8_t bitstream[COUNT];
3296
    uint8_t nal[COUNT*2];
3297
    H264Context h;
3298
    memset(&h, 0, sizeof(H264Context));
3299

3300
    for(i=0; i<COUNT; i++){
3301
        int zeros= i;
3302
        int nal_length;
3303
        int consumed;
3304
        int out_length;
3305
        uint8_t *out;
3306
        int j;
3307

3308
        for(j=0; j<COUNT; j++){
3309
            bitstream[j]= (random() % 255) + 1;
3310
        }
3311

3312
        for(j=0; j<zeros; j++){
3313
            int pos= random() % COUNT;
3314
            while(bitstream[pos] == 0){
3315
                pos++;
3316
                pos %= COUNT;
3317
            }
3318
            bitstream[pos]=0;
3319
        }
3320

3321
        START_TIMER
3322

3323
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3324
        if(nal_length<0){
3325
            printf("encoding failed\n");
3326
            return -1;
3327
        }
3328

3329
        out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3330

3331
        STOP_TIMER("NAL")
3332

3333
        if(out_length != COUNT){
3334
            printf("incorrect length %d %d\n", out_length, COUNT);
3335
            return -1;
3336
        }
3337

3338
        if(consumed != nal_length){
3339
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
3340
            return -1;
3341
        }
3342

3343
        if(memcmp(bitstream, out, COUNT)){
3344
            printf("mismatch\n");
3345
            return -1;
3346
        }
3347
    }
3348
#endif
3349

    
3350
    printf("Testing RBSP\n");
3351

    
3352

    
3353
    return 0;
3354
}
3355
#endif /* TEST */
3356

    
3357

    
3358
av_cold void ff_h264_free_context(H264Context *h)
3359
{
3360
    int i;
3361

    
3362
    free_tables(h); //FIXME cleanup init stuff perhaps
3363

    
3364
    for(i = 0; i < MAX_SPS_COUNT; i++)
3365
        av_freep(h->sps_buffers + i);
3366

    
3367
    for(i = 0; i < MAX_PPS_COUNT; i++)
3368
        av_freep(h->pps_buffers + i);
3369
}
3370

    
3371
av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3372
{
3373
    H264Context *h = avctx->priv_data;
3374
    MpegEncContext *s = &h->s;
3375

    
3376
    ff_h264_free_context(h);
3377

    
3378
    MPV_common_end(s);
3379

    
3380
//    memset(h, 0, sizeof(H264Context));
3381

    
3382
    return 0;
3383
}
3384

    
3385

    
3386
AVCodec h264_decoder = {
3387
    "h264",
3388
    AVMEDIA_TYPE_VIDEO,
3389
    CODEC_ID_H264,
3390
    sizeof(H264Context),
3391
    ff_h264_decode_init,
3392
    NULL,
3393
    ff_h264_decode_end,
3394
    decode_frame,
3395
    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3396
    .flush= flush_dpb,
3397
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3398
};
3399

    
3400
#if CONFIG_H264_VDPAU_DECODER
3401
AVCodec h264_vdpau_decoder = {
3402
    "h264_vdpau",
3403
    AVMEDIA_TYPE_VIDEO,
3404
    CODEC_ID_H264,
3405
    sizeof(H264Context),
3406
    ff_h264_decode_init,
3407
    NULL,
3408
    ff_h264_decode_end,
3409
    decode_frame,
3410
    CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3411
    .flush= flush_dpb,
3412
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3413
    .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3414
};
3415
#endif