Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ e1c88a21

History | View | Annotate | Download (115 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 libavcodec/h264.c
24
 * H.264 / AVC / MPEG4 part10 codec.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

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

    
41
#include "cabac.h"
42

    
43
//#undef NDEBUG
44
#include <assert.h>
45

    
46
static const uint8_t rem6[52]={
47
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,
48
};
49

    
50
static const uint8_t div6[52]={
51
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,
52
};
53

    
54
void ff_h264_write_back_intra_pred_mode(H264Context *h){
55
    const int mb_xy= h->mb_xy;
56

    
57
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
58
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
59
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
60
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
61
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
62
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
63
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
64
}
65

    
66
/**
67
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
68
 */
69
int ff_h264_check_intra4x4_pred_mode(H264Context *h){
70
    MpegEncContext * const s = &h->s;
71
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
72
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
73
    int i;
74

    
75
    if(!(h->top_samples_available&0x8000)){
76
        for(i=0; i<4; i++){
77
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
78
            if(status<0){
79
                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);
80
                return -1;
81
            } else if(status){
82
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
83
            }
84
        }
85
    }
86

    
87
    if((h->left_samples_available&0x8888)!=0x8888){
88
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
89
        for(i=0; i<4; i++){
90
            if(!(h->left_samples_available&mask[i])){
91
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
92
                if(status<0){
93
                    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);
94
                    return -1;
95
                } else if(status){
96
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
97
                }
98
            }
99
        }
100
    }
101

    
102
    return 0;
103
} //FIXME cleanup like ff_h264_check_intra_pred_mode
104

    
105
/**
106
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
107
 */
108
int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
109
    MpegEncContext * const s = &h->s;
110
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
111
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
112

    
113
    if(mode > 6U) {
114
        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);
115
        return -1;
116
    }
117

    
118
    if(!(h->top_samples_available&0x8000)){
119
        mode= top[ mode ];
120
        if(mode<0){
121
            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);
122
            return -1;
123
        }
124
    }
125

    
126
    if((h->left_samples_available&0x8080) != 0x8080){
127
        mode= left[ mode ];
128
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
129
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
130
        }
131
        if(mode<0){
132
            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);
133
            return -1;
134
        }
135
    }
136

    
137
    return mode;
138
}
139

    
140
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
141
    int i, si, di;
142
    uint8_t *dst;
143
    int bufidx;
144

    
145
//    src[0]&0x80;                //forbidden bit
146
    h->nal_ref_idc= src[0]>>5;
147
    h->nal_unit_type= src[0]&0x1F;
148

    
149
    src++; length--;
150
#if 0
151
    for(i=0; i<length; i++)
152
        printf("%2X ", src[i]);
153
#endif
154

    
155
#if HAVE_FAST_UNALIGNED
156
# if HAVE_FAST_64BIT
157
#   define RS 7
158
    for(i=0; i+1<length; i+=9){
159
        if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
160
# else
161
#   define RS 3
162
    for(i=0; i+1<length; i+=5){
163
        if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
164
# endif
165
            continue;
166
        if(i>0 && !src[i]) i--;
167
        while(src[i]) i++;
168
#else
169
#   define RS 0
170
    for(i=0; i+1<length; i+=2){
171
        if(src[i]) continue;
172
        if(i>0 && src[i-1]==0) i--;
173
#endif
174
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
175
            if(src[i+2]!=3){
176
                /* startcode, so we must be past the end */
177
                length=i;
178
            }
179
            break;
180
        }
181
        i-= RS;
182
    }
183

    
184
    if(i>=length-1){ //no escaped 0
185
        *dst_length= length;
186
        *consumed= length+1; //+1 for the header
187
        return src;
188
    }
189

    
190
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
191
    av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
192
    dst= h->rbsp_buffer[bufidx];
193

    
194
    if (dst == NULL){
195
        return NULL;
196
    }
197

    
198
//printf("decoding esc\n");
199
    memcpy(dst, src, i);
200
    si=di=i;
201
    while(si+2<length){
202
        //remove escapes (very rare 1:2^22)
203
        if(src[si+2]>3){
204
            dst[di++]= src[si++];
205
            dst[di++]= src[si++];
206
        }else if(src[si]==0 && src[si+1]==0){
207
            if(src[si+2]==3){ //escape
208
                dst[di++]= 0;
209
                dst[di++]= 0;
210
                si+=3;
211
                continue;
212
            }else //next start code
213
                goto nsc;
214
        }
215

    
216
        dst[di++]= src[si++];
217
    }
218
    while(si<length)
219
        dst[di++]= src[si++];
220
nsc:
221

    
222
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
223

    
224
    *dst_length= di;
225
    *consumed= si + 1;//+1 for the header
226
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
227
    return dst;
228
}
229

    
230
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
231
    int v= *src;
232
    int r;
233

    
234
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
235

    
236
    for(r=1; r<9; r++){
237
        if(v&1) return r;
238
        v>>=1;
239
    }
240
    return 0;
241
}
242

    
243
/**
244
 * IDCT transforms the 16 dc values and dequantizes them.
245
 * @param qp quantization parameter
246
 */
247
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
248
#define stride 16
249
    int i;
250
    int temp[16]; //FIXME check if this is a good idea
251
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
252
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
253

    
254
//memset(block, 64, 2*256);
255
//return;
256
    for(i=0; i<4; i++){
257
        const int offset= y_offset[i];
258
        const int z0= block[offset+stride*0] + block[offset+stride*4];
259
        const int z1= block[offset+stride*0] - block[offset+stride*4];
260
        const int z2= block[offset+stride*1] - block[offset+stride*5];
261
        const int z3= block[offset+stride*1] + block[offset+stride*5];
262

    
263
        temp[4*i+0]= z0+z3;
264
        temp[4*i+1]= z1+z2;
265
        temp[4*i+2]= z1-z2;
266
        temp[4*i+3]= z0-z3;
267
    }
268

    
269
    for(i=0; i<4; i++){
270
        const int offset= x_offset[i];
271
        const int z0= temp[4*0+i] + temp[4*2+i];
272
        const int z1= temp[4*0+i] - temp[4*2+i];
273
        const int z2= temp[4*1+i] - temp[4*3+i];
274
        const int z3= temp[4*1+i] + temp[4*3+i];
275

    
276
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
277
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
278
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
279
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
280
    }
281
}
282

    
283
#if 0
284
/**
285
 * DCT transforms the 16 dc values.
286
 * @param qp quantization parameter ??? FIXME
287
 */
288
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
289
//    const int qmul= dequant_coeff[qp][0];
290
    int i;
291
    int temp[16]; //FIXME check if this is a good idea
292
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
293
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
294

295
    for(i=0; i<4; i++){
296
        const int offset= y_offset[i];
297
        const int z0= block[offset+stride*0] + block[offset+stride*4];
298
        const int z1= block[offset+stride*0] - block[offset+stride*4];
299
        const int z2= block[offset+stride*1] - block[offset+stride*5];
300
        const int z3= block[offset+stride*1] + block[offset+stride*5];
301

302
        temp[4*i+0]= z0+z3;
303
        temp[4*i+1]= z1+z2;
304
        temp[4*i+2]= z1-z2;
305
        temp[4*i+3]= z0-z3;
306
    }
307

308
    for(i=0; i<4; i++){
309
        const int offset= x_offset[i];
310
        const int z0= temp[4*0+i] + temp[4*2+i];
311
        const int z1= temp[4*0+i] - temp[4*2+i];
312
        const int z2= temp[4*1+i] - temp[4*3+i];
313
        const int z3= temp[4*1+i] + temp[4*3+i];
314

315
        block[stride*0 +offset]= (z0 + z3)>>1;
316
        block[stride*2 +offset]= (z1 + z2)>>1;
317
        block[stride*8 +offset]= (z1 - z2)>>1;
318
        block[stride*10+offset]= (z0 - z3)>>1;
319
    }
320
}
321
#endif
322

    
323
#undef xStride
324
#undef stride
325

    
326
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
327
    const int stride= 16*2;
328
    const int xStride= 16;
329
    int a,b,c,d,e;
330

    
331
    a= block[stride*0 + xStride*0];
332
    b= block[stride*0 + xStride*1];
333
    c= block[stride*1 + xStride*0];
334
    d= block[stride*1 + xStride*1];
335

    
336
    e= a-b;
337
    a= a+b;
338
    b= c-d;
339
    c= c+d;
340

    
341
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
342
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
343
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
344
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
345
}
346

    
347
#if 0
348
static void chroma_dc_dct_c(DCTELEM *block){
349
    const int stride= 16*2;
350
    const int xStride= 16;
351
    int a,b,c,d,e;
352

353
    a= block[stride*0 + xStride*0];
354
    b= block[stride*0 + xStride*1];
355
    c= block[stride*1 + xStride*0];
356
    d= block[stride*1 + xStride*1];
357

358
    e= a-b;
359
    a= a+b;
360
    b= c-d;
361
    c= c+d;
362

363
    block[stride*0 + xStride*0]= (a+c);
364
    block[stride*0 + xStride*1]= (e+b);
365
    block[stride*1 + xStride*0]= (a-c);
366
    block[stride*1 + xStride*1]= (e-b);
367
}
368
#endif
369

    
370
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
371
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
372
                           int src_x_offset, int src_y_offset,
373
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
374
    MpegEncContext * const s = &h->s;
375
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
376
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
377
    const int luma_xy= (mx&3) + ((my&3)<<2);
378
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
379
    uint8_t * src_cb, * src_cr;
380
    int extra_width= h->emu_edge_width;
381
    int extra_height= h->emu_edge_height;
382
    int emu=0;
383
    const int full_mx= mx>>2;
384
    const int full_my= my>>2;
385
    const int pic_width  = 16*s->mb_width;
386
    const int pic_height = 16*s->mb_height >> MB_FIELD;
387

    
388
    if(mx&7) extra_width -= 3;
389
    if(my&7) extra_height -= 3;
390

    
391
    if(   full_mx < 0-extra_width
392
       || full_my < 0-extra_height
393
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
394
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
395
        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);
396
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
397
        emu=1;
398
    }
399

    
400
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
401
    if(!square){
402
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
403
    }
404

    
405
    if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
406

    
407
    if(MB_FIELD){
408
        // chroma offset when predicting from a field of opposite parity
409
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
410
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
411
    }
412
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
413
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
414

    
415
    if(emu){
416
        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);
417
            src_cb= s->edge_emu_buffer;
418
    }
419
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
420

    
421
    if(emu){
422
        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);
423
            src_cr= s->edge_emu_buffer;
424
    }
425
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
426
}
427

    
428
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
429
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
430
                           int x_offset, int y_offset,
431
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
432
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
433
                           int list0, int list1){
434
    MpegEncContext * const s = &h->s;
435
    qpel_mc_func *qpix_op=  qpix_put;
436
    h264_chroma_mc_func chroma_op= chroma_put;
437

    
438
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
439
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
440
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
441
    x_offset += 8*s->mb_x;
442
    y_offset += 8*(s->mb_y >> MB_FIELD);
443

    
444
    if(list0){
445
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
446
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
447
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
448
                           qpix_op, chroma_op);
449

    
450
        qpix_op=  qpix_avg;
451
        chroma_op= chroma_avg;
452
    }
453

    
454
    if(list1){
455
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
456
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
457
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
458
                           qpix_op, chroma_op);
459
    }
460
}
461

    
462
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
463
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
464
                           int x_offset, int y_offset,
465
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
466
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
467
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
468
                           int list0, int list1){
469
    MpegEncContext * const s = &h->s;
470

    
471
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
472
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
473
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
474
    x_offset += 8*s->mb_x;
475
    y_offset += 8*(s->mb_y >> MB_FIELD);
476

    
477
    if(list0 && list1){
478
        /* don't optimize for luma-only case, since B-frames usually
479
         * use implicit weights => chroma too. */
480
        uint8_t *tmp_cb = s->obmc_scratchpad;
481
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
482
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
483
        int refn0 = h->ref_cache[0][ scan8[n] ];
484
        int refn1 = h->ref_cache[1][ scan8[n] ];
485

    
486
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
487
                    dest_y, dest_cb, dest_cr,
488
                    x_offset, y_offset, qpix_put, chroma_put);
489
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
490
                    tmp_y, tmp_cb, tmp_cr,
491
                    x_offset, y_offset, qpix_put, chroma_put);
492

    
493
        if(h->use_weight == 2){
494
            int weight0 = h->implicit_weight[refn0][refn1];
495
            int weight1 = 64 - weight0;
496
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
497
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
498
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
499
        }else{
500
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
501
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
502
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
503
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
504
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
505
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
506
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
507
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
508
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
509
        }
510
    }else{
511
        int list = list1 ? 1 : 0;
512
        int refn = h->ref_cache[list][ scan8[n] ];
513
        Picture *ref= &h->ref_list[list][refn];
514
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
515
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
516
                    qpix_put, chroma_put);
517

    
518
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
519
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
520
        if(h->use_weight_chroma){
521
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
522
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
523
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
524
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
525
        }
526
    }
527
}
528

    
529
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
530
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
531
                           int x_offset, int y_offset,
532
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
533
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
534
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
535
                           int list0, int list1){
536
    if((h->use_weight==2 && list0 && list1
537
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
538
       || h->use_weight==1)
539
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
540
                         x_offset, y_offset, qpix_put, chroma_put,
541
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
542
    else
543
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
544
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
545
}
546

    
547
static inline void prefetch_motion(H264Context *h, int list){
548
    /* fetch pixels for estimated mv 4 macroblocks ahead
549
     * optimized for 64byte cache lines */
550
    MpegEncContext * const s = &h->s;
551
    const int refn = h->ref_cache[list][scan8[0]];
552
    if(refn >= 0){
553
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
554
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
555
        uint8_t **src= h->ref_list[list][refn].data;
556
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
557
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
558
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
559
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
560
    }
561
}
562

    
563
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
564
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
565
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
566
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
567
    MpegEncContext * const s = &h->s;
568
    const int mb_xy= h->mb_xy;
569
    const int mb_type= s->current_picture.mb_type[mb_xy];
570

    
571
    assert(IS_INTER(mb_type));
572

    
573
    prefetch_motion(h, 0);
574

    
575
    if(IS_16X16(mb_type)){
576
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
577
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
578
                weight_op, weight_avg,
579
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
580
    }else if(IS_16X8(mb_type)){
581
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
582
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
583
                &weight_op[1], &weight_avg[1],
584
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
585
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
586
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
587
                &weight_op[1], &weight_avg[1],
588
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
589
    }else if(IS_8X16(mb_type)){
590
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
591
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
592
                &weight_op[2], &weight_avg[2],
593
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
594
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
595
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
596
                &weight_op[2], &weight_avg[2],
597
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
598
    }else{
599
        int i;
600

    
601
        assert(IS_8X8(mb_type));
602

    
603
        for(i=0; i<4; i++){
604
            const int sub_mb_type= h->sub_mb_type[i];
605
            const int n= 4*i;
606
            int x_offset= (i&1)<<2;
607
            int y_offset= (i&2)<<1;
608

    
609
            if(IS_SUB_8X8(sub_mb_type)){
610
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
611
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
612
                    &weight_op[3], &weight_avg[3],
613
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
614
            }else if(IS_SUB_8X4(sub_mb_type)){
615
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
616
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
617
                    &weight_op[4], &weight_avg[4],
618
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
619
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
620
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
621
                    &weight_op[4], &weight_avg[4],
622
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
623
            }else if(IS_SUB_4X8(sub_mb_type)){
624
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
625
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
626
                    &weight_op[5], &weight_avg[5],
627
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
628
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
629
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
630
                    &weight_op[5], &weight_avg[5],
631
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
632
            }else{
633
                int j;
634
                assert(IS_SUB_4X4(sub_mb_type));
635
                for(j=0; j<4; j++){
636
                    int sub_x_offset= x_offset + 2*(j&1);
637
                    int sub_y_offset= y_offset +   (j&2);
638
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
639
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
640
                        &weight_op[6], &weight_avg[6],
641
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
642
                }
643
            }
644
        }
645
    }
646

    
647
    prefetch_motion(h, 1);
648
}
649

    
650

    
651
static void free_tables(H264Context *h){
652
    int i;
653
    H264Context *hx;
654
    av_freep(&h->intra4x4_pred_mode);
655
    av_freep(&h->chroma_pred_mode_table);
656
    av_freep(&h->cbp_table);
657
    av_freep(&h->mvd_table[0]);
658
    av_freep(&h->mvd_table[1]);
659
    av_freep(&h->direct_table);
660
    av_freep(&h->non_zero_count);
661
    av_freep(&h->slice_table_base);
662
    h->slice_table= NULL;
663
    av_freep(&h->list_counts);
664

    
665
    av_freep(&h->mb2b_xy);
666
    av_freep(&h->mb2br_xy);
667
    av_freep(&h->mb2b8_xy);
668

    
669
    for(i = 0; i < MAX_THREADS; i++) {
670
        hx = h->thread_context[i];
671
        if(!hx) continue;
672
        av_freep(&hx->top_borders[1]);
673
        av_freep(&hx->top_borders[0]);
674
        av_freep(&hx->s.obmc_scratchpad);
675
        av_freep(&hx->rbsp_buffer[1]);
676
        av_freep(&hx->rbsp_buffer[0]);
677
        hx->rbsp_buffer_size[0] = 0;
678
        hx->rbsp_buffer_size[1] = 0;
679
        if (i) av_freep(&h->thread_context[i]);
680
    }
681
}
682

    
683
static void init_dequant8_coeff_table(H264Context *h){
684
    int i,q,x;
685
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
686
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
687
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
688

    
689
    for(i=0; i<2; i++ ){
690
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
691
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
692
            break;
693
        }
694

    
695
        for(q=0; q<52; q++){
696
            int shift = div6[q];
697
            int idx = rem6[q];
698
            for(x=0; x<64; x++)
699
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
700
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
701
                    h->pps.scaling_matrix8[i][x]) << shift;
702
        }
703
    }
704
}
705

    
706
static void init_dequant4_coeff_table(H264Context *h){
707
    int i,j,q,x;
708
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
709
    for(i=0; i<6; i++ ){
710
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
711
        for(j=0; j<i; j++){
712
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
713
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
714
                break;
715
            }
716
        }
717
        if(j<i)
718
            continue;
719

    
720
        for(q=0; q<52; q++){
721
            int shift = div6[q] + 2;
722
            int idx = rem6[q];
723
            for(x=0; x<16; x++)
724
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
725
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
726
                    h->pps.scaling_matrix4[i][x]) << shift;
727
        }
728
    }
729
}
730

    
731
static void init_dequant_tables(H264Context *h){
732
    int i,x;
733
    init_dequant4_coeff_table(h);
734
    if(h->pps.transform_8x8_mode)
735
        init_dequant8_coeff_table(h);
736
    if(h->sps.transform_bypass){
737
        for(i=0; i<6; i++)
738
            for(x=0; x<16; x++)
739
                h->dequant4_coeff[i][0][x] = 1<<6;
740
        if(h->pps.transform_8x8_mode)
741
            for(i=0; i<2; i++)
742
                for(x=0; x<64; x++)
743
                    h->dequant8_coeff[i][0][x] = 1<<6;
744
    }
745
}
746

    
747

    
748
int ff_h264_alloc_tables(H264Context *h){
749
    MpegEncContext * const s = &h->s;
750
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
751
    int x,y;
752

    
753
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
754

    
755
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
756
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
757
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
758

    
759
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
760
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*big_mb_num * sizeof(uint8_t), fail);
761
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*big_mb_num * sizeof(uint8_t), fail);
762
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
763
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
764

    
765
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
766
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
767

    
768
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
769
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
770
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
771
    for(y=0; y<s->mb_height; y++){
772
        for(x=0; x<s->mb_width; x++){
773
            const int mb_xy= x + y*s->mb_stride;
774
            const int b_xy = 4*x + 4*y*h->b_stride;
775
            const int b8_xy= 2*x + 2*y*h->b8_stride;
776

    
777
            h->mb2b_xy [mb_xy]= b_xy;
778
            h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
779
            h->mb2b8_xy[mb_xy]= b8_xy;
780
        }
781
    }
782

    
783
    s->obmc_scratchpad = NULL;
784

    
785
    if(!h->dequant4_coeff[0])
786
        init_dequant_tables(h);
787

    
788
    return 0;
789
fail:
790
    free_tables(h);
791
    return -1;
792
}
793

    
794
/**
795
 * Mimic alloc_tables(), but for every context thread.
796
 */
797
static void clone_tables(H264Context *dst, H264Context *src){
798
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
799
    dst->non_zero_count           = src->non_zero_count;
800
    dst->slice_table              = src->slice_table;
801
    dst->cbp_table                = src->cbp_table;
802
    dst->mb2b_xy                  = src->mb2b_xy;
803
    dst->mb2br_xy                 = src->mb2br_xy;
804
    dst->mb2b8_xy                 = src->mb2b8_xy;
805
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
806
    dst->mvd_table[0]             = src->mvd_table[0];
807
    dst->mvd_table[1]             = src->mvd_table[1];
808
    dst->direct_table             = src->direct_table;
809
    dst->list_counts              = src->list_counts;
810

    
811
    dst->s.obmc_scratchpad = NULL;
812
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
813
}
814

    
815
/**
816
 * Init context
817
 * Allocate buffers which are not shared amongst multiple threads.
818
 */
819
static int context_init(H264Context *h){
820
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
821
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
822

    
823
    return 0;
824
fail:
825
    return -1; // free_tables will clean up for us
826
}
827

    
828
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
829

    
830
static av_cold void common_init(H264Context *h){
831
    MpegEncContext * const s = &h->s;
832

    
833
    s->width = s->avctx->width;
834
    s->height = s->avctx->height;
835
    s->codec_id= s->avctx->codec->id;
836

    
837
    ff_h264_pred_init(&h->hpc, s->codec_id);
838

    
839
    h->dequant_coeff_pps= -1;
840
    s->unrestricted_mv=1;
841
    s->decode=1; //FIXME
842

    
843
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
844

    
845
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
846
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
847
}
848

    
849
av_cold int ff_h264_decode_init(AVCodecContext *avctx){
850
    H264Context *h= avctx->priv_data;
851
    MpegEncContext * const s = &h->s;
852

    
853
    MPV_decode_defaults(s);
854

    
855
    s->avctx = avctx;
856
    common_init(h);
857

    
858
    s->out_format = FMT_H264;
859
    s->workaround_bugs= avctx->workaround_bugs;
860

    
861
    // set defaults
862
//    s->decode_mb= ff_h263_decode_mb;
863
    s->quarter_sample = 1;
864
    if(!avctx->has_b_frames)
865
    s->low_delay= 1;
866

    
867
    avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
868

    
869
    ff_h264_decode_init_vlc();
870

    
871
    h->thread_context[0] = h;
872
    h->outputed_poc = INT_MIN;
873
    h->prev_poc_msb= 1<<16;
874
    h->x264_build = -1;
875
    ff_h264_reset_sei(h);
876
    if(avctx->codec_id == CODEC_ID_H264){
877
        if(avctx->ticks_per_frame == 1){
878
            s->avctx->time_base.den *=2;
879
        }
880
        avctx->ticks_per_frame = 2;
881
    }
882

    
883
    if(avctx->extradata_size > 0 && avctx->extradata && *(char *)avctx->extradata == 1){
884
        int i, cnt, nalsize;
885
        unsigned char *p = avctx->extradata;
886

    
887
        h->is_avc = 1;
888

    
889
        if(avctx->extradata_size < 7) {
890
            av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
891
            return -1;
892
        }
893
        /* sps and pps in the avcC always have length coded with 2 bytes,
894
           so put a fake nal_length_size = 2 while parsing them */
895
        h->nal_length_size = 2;
896
        // Decode sps from avcC
897
        cnt = *(p+5) & 0x1f; // Number of sps
898
        p += 6;
899
        for (i = 0; i < cnt; i++) {
900
            nalsize = AV_RB16(p) + 2;
901
            if(decode_nal_units(h, p, nalsize) < 0) {
902
                av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
903
                return -1;
904
            }
905
            p += nalsize;
906
        }
907
        // Decode pps from avcC
908
        cnt = *(p++); // Number of pps
909
        for (i = 0; i < cnt; i++) {
910
            nalsize = AV_RB16(p) + 2;
911
            if(decode_nal_units(h, p, nalsize)  != nalsize) {
912
                av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
913
                return -1;
914
            }
915
            p += nalsize;
916
        }
917
        // Now store right nal length size, that will be use to parse all other nals
918
        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
919
    } else {
920
        h->is_avc = 0;
921
        if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
922
            return -1;
923
    }
924
    if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
925
        s->avctx->has_b_frames = h->sps.num_reorder_frames;
926
        s->low_delay = 0;
927
    }
928

    
929
    return 0;
930
}
931

    
932
int ff_h264_frame_start(H264Context *h){
933
    MpegEncContext * const s = &h->s;
934
    int i;
935

    
936
    if(MPV_frame_start(s, s->avctx) < 0)
937
        return -1;
938
    ff_er_frame_start(s);
939
    /*
940
     * MPV_frame_start uses pict_type to derive key_frame.
941
     * This is incorrect for H.264; IDR markings must be used.
942
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
943
     * See decode_nal_units().
944
     */
945
    s->current_picture_ptr->key_frame= 0;
946
    s->current_picture_ptr->mmco_reset= 0;
947

    
948
    assert(s->linesize && s->uvlinesize);
949

    
950
    for(i=0; i<16; i++){
951
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
952
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
953
    }
954
    for(i=0; i<4; i++){
955
        h->block_offset[16+i]=
956
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
957
        h->block_offset[24+16+i]=
958
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
959
    }
960

    
961
    /* can't be in alloc_tables because linesize isn't known there.
962
     * FIXME: redo bipred weight to not require extra buffer? */
963
    for(i = 0; i < s->avctx->thread_count; i++)
964
        if(!h->thread_context[i]->s.obmc_scratchpad)
965
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
966

    
967
    /* some macroblocks will be accessed before they're available */
968
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
969
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
970

    
971
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
972

    
973
    // We mark the current picture as non-reference after allocating it, so
974
    // that if we break out due to an error it can be released automatically
975
    // in the next MPV_frame_start().
976
    // SVQ3 as well as most other codecs have only last/next/current and thus
977
    // get released even with set reference, besides SVQ3 and others do not
978
    // mark frames as reference later "naturally".
979
    if(s->codec_id != CODEC_ID_SVQ3)
980
        s->current_picture_ptr->reference= 0;
981

    
982
    s->current_picture_ptr->field_poc[0]=
983
    s->current_picture_ptr->field_poc[1]= INT_MAX;
984
    assert(s->current_picture_ptr->long_ref==0);
985

    
986
    return 0;
987
}
988

    
989
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){
990
    MpegEncContext * const s = &h->s;
991
    uint8_t *top_border;
992
    int top_idx = 1;
993

    
994
    src_y  -=   linesize;
995
    src_cb -= uvlinesize;
996
    src_cr -= uvlinesize;
997

    
998
    if(!simple && FRAME_MBAFF){
999
        if(s->mb_y&1){
1000
            if(!MB_MBAFF){
1001
                top_border = h->top_borders[0][s->mb_x];
1002
                AV_COPY128(top_border, src_y + 15*linesize);
1003
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1004
                    AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1005
                    AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1006
                }
1007
            }
1008
        }else if(MB_MBAFF){
1009
            top_idx = 0;
1010
        }else
1011
            return;
1012
    }
1013

    
1014
    top_border = h->top_borders[top_idx][s->mb_x];
1015
    // There are two lines saved, the line above the the top macroblock of a pair,
1016
    // and the line above the bottom macroblock
1017
    AV_COPY128(top_border, src_y + 16*linesize);
1018

    
1019
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1020
        AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1021
        AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1022
    }
1023
}
1024

    
1025
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){
1026
    MpegEncContext * const s = &h->s;
1027
    int deblock_left;
1028
    int deblock_top;
1029
    int top_idx = 1;
1030
    uint8_t *top_border_m1;
1031
    uint8_t *top_border;
1032

    
1033
    if(!simple && FRAME_MBAFF){
1034
        if(s->mb_y&1){
1035
            if(!MB_MBAFF)
1036
                return;
1037
        }else{
1038
            top_idx = MB_MBAFF ? 0 : 1;
1039
        }
1040
    }
1041

    
1042
    if(h->deblocking_filter == 2) {
1043
        deblock_left = h->left_type[0];
1044
        deblock_top  = h->top_type;
1045
    } else {
1046
        deblock_left = (s->mb_x > 0);
1047
        deblock_top =  (s->mb_y > !!MB_FIELD);
1048
    }
1049

    
1050
    src_y  -=   linesize + 1;
1051
    src_cb -= uvlinesize + 1;
1052
    src_cr -= uvlinesize + 1;
1053

    
1054
    top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1055
    top_border    = h->top_borders[top_idx][s->mb_x];
1056

    
1057
#define XCHG(a,b,xchg)\
1058
if (xchg) AV_SWAP64(b,a);\
1059
else      AV_COPY64(b,a);
1060

    
1061
    if(deblock_top){
1062
        if(deblock_left){
1063
            XCHG(top_border_m1+8, src_y -7, 1);
1064
        }
1065
        XCHG(top_border+0, src_y +1, xchg);
1066
        XCHG(top_border+8, src_y +9, 1);
1067
        if(s->mb_x+1 < s->mb_width){
1068
            XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
1069
        }
1070
    }
1071

    
1072
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1073
        if(deblock_top){
1074
            if(deblock_left){
1075
                XCHG(top_border_m1+16, src_cb -7, 1);
1076
                XCHG(top_border_m1+24, src_cr -7, 1);
1077
            }
1078
            XCHG(top_border+16, src_cb+1, 1);
1079
            XCHG(top_border+24, src_cr+1, 1);
1080
        }
1081
    }
1082
}
1083

    
1084
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1085
    MpegEncContext * const s = &h->s;
1086
    const int mb_x= s->mb_x;
1087
    const int mb_y= s->mb_y;
1088
    const int mb_xy= h->mb_xy;
1089
    const int mb_type= s->current_picture.mb_type[mb_xy];
1090
    uint8_t  *dest_y, *dest_cb, *dest_cr;
1091
    int linesize, uvlinesize /*dct_offset*/;
1092
    int i;
1093
    int *block_offset = &h->block_offset[0];
1094
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1095
    /* is_h264 should always be true if SVQ3 is disabled. */
1096
    const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1097
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1098
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1099

    
1100
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1101
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1102
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1103

    
1104
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1105
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1106

    
1107
    h->list_counts[mb_xy]= h->list_count;
1108

    
1109
    if (!simple && MB_FIELD) {
1110
        linesize   = h->mb_linesize   = s->linesize * 2;
1111
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1112
        block_offset = &h->block_offset[24];
1113
        if(mb_y&1){ //FIXME move out of this function?
1114
            dest_y -= s->linesize*15;
1115
            dest_cb-= s->uvlinesize*7;
1116
            dest_cr-= s->uvlinesize*7;
1117
        }
1118
        if(FRAME_MBAFF) {
1119
            int list;
1120
            for(list=0; list<h->list_count; list++){
1121
                if(!USES_LIST(mb_type, list))
1122
                    continue;
1123
                if(IS_16X16(mb_type)){
1124
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
1125
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1126
                }else{
1127
                    for(i=0; i<16; i+=4){
1128
                        int ref = h->ref_cache[list][scan8[i]];
1129
                        if(ref >= 0)
1130
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1131
                    }
1132
                }
1133
            }
1134
        }
1135
    } else {
1136
        linesize   = h->mb_linesize   = s->linesize;
1137
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1138
//        dct_offset = s->linesize * 16;
1139
    }
1140

    
1141
    if (!simple && IS_INTRA_PCM(mb_type)) {
1142
        for (i=0; i<16; i++) {
1143
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1144
        }
1145
        for (i=0; i<8; i++) {
1146
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1147
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1148
        }
1149
    } else {
1150
        if(IS_INTRA(mb_type)){
1151
            if(h->deblocking_filter)
1152
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1153

    
1154
            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1155
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1156
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1157
            }
1158

    
1159
            if(IS_INTRA4x4(mb_type)){
1160
                if(simple || !s->encoding){
1161
                    if(IS_8x8DCT(mb_type)){
1162
                        if(transform_bypass){
1163
                            idct_dc_add =
1164
                            idct_add    = s->dsp.add_pixels8;
1165
                        }else{
1166
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
1167
                            idct_add    = s->dsp.h264_idct8_add;
1168
                        }
1169
                        for(i=0; i<16; i+=4){
1170
                            uint8_t * const ptr= dest_y + block_offset[i];
1171
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1172
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1173
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1174
                            }else{
1175
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
1176
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1177
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
1178
                                if(nnz){
1179
                                    if(nnz == 1 && h->mb[i*16])
1180
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
1181
                                    else
1182
                                        idct_add   (ptr, h->mb + i*16, linesize);
1183
                                }
1184
                            }
1185
                        }
1186
                    }else{
1187
                        if(transform_bypass){
1188
                            idct_dc_add =
1189
                            idct_add    = s->dsp.add_pixels4;
1190
                        }else{
1191
                            idct_dc_add = s->dsp.h264_idct_dc_add;
1192
                            idct_add    = s->dsp.h264_idct_add;
1193
                        }
1194
                        for(i=0; i<16; i++){
1195
                            uint8_t * const ptr= dest_y + block_offset[i];
1196
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1197

    
1198
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1199
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1200
                            }else{
1201
                                uint8_t *topright;
1202
                                int nnz, tr;
1203
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1204
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1205
                                    assert(mb_y || linesize <= block_offset[i]);
1206
                                    if(!topright_avail){
1207
                                        tr= ptr[3 - linesize]*0x01010101;
1208
                                        topright= (uint8_t*) &tr;
1209
                                    }else
1210
                                        topright= ptr + 4 - linesize;
1211
                                }else
1212
                                    topright= NULL;
1213

    
1214
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1215
                                nnz = h->non_zero_count_cache[ scan8[i] ];
1216
                                if(nnz){
1217
                                    if(is_h264){
1218
                                        if(nnz == 1 && h->mb[i*16])
1219
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
1220
                                        else
1221
                                            idct_add   (ptr, h->mb + i*16, linesize);
1222
                                    }else
1223
                                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1224
                                }
1225
                            }
1226
                        }
1227
                    }
1228
                }
1229
            }else{
1230
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1231
                if(is_h264){
1232
                    if(!transform_bypass)
1233
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1234
                }else
1235
                    ff_svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1236
            }
1237
            if(h->deblocking_filter)
1238
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1239
        }else if(is_h264){
1240
            hl_motion(h, dest_y, dest_cb, dest_cr,
1241
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1242
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1243
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1244
        }
1245

    
1246

    
1247
        if(!IS_INTRA4x4(mb_type)){
1248
            if(is_h264){
1249
                if(IS_INTRA16x16(mb_type)){
1250
                    if(transform_bypass){
1251
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1252
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1253
                        }else{
1254
                            for(i=0; i<16; i++){
1255
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1256
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1257
                            }
1258
                        }
1259
                    }else{
1260
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1261
                    }
1262
                }else if(h->cbp&15){
1263
                    if(transform_bypass){
1264
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1265
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1266
                        for(i=0; i<16; i+=di){
1267
                            if(h->non_zero_count_cache[ scan8[i] ]){
1268
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1269
                            }
1270
                        }
1271
                    }else{
1272
                        if(IS_8x8DCT(mb_type)){
1273
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1274
                        }else{
1275
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1276
                        }
1277
                    }
1278
                }
1279
            }else{
1280
                for(i=0; i<16; i++){
1281
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1282
                        uint8_t * const ptr= dest_y + block_offset[i];
1283
                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1284
                    }
1285
                }
1286
            }
1287
        }
1288

    
1289
        if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1290
            uint8_t *dest[2] = {dest_cb, dest_cr};
1291
            if(transform_bypass){
1292
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1293
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1294
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1295
                }else{
1296
                    idct_add = s->dsp.add_pixels4;
1297
                    for(i=16; i<16+8; i++){
1298
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1299
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1300
                    }
1301
                }
1302
            }else{
1303
                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]);
1304
                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]);
1305
                if(is_h264){
1306
                    idct_add = s->dsp.h264_idct_add;
1307
                    idct_dc_add = s->dsp.h264_idct_dc_add;
1308
                    for(i=16; i<16+8; i++){
1309
                        if(h->non_zero_count_cache[ scan8[i] ])
1310
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1311
                        else if(h->mb[i*16])
1312
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1313
                    }
1314
                }else{
1315
                    for(i=16; i<16+8; i++){
1316
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1317
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1318
                            ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1319
                        }
1320
                    }
1321
                }
1322
            }
1323
        }
1324
    }
1325
    if(h->cbp || IS_INTRA(mb_type))
1326
        s->dsp.clear_blocks(h->mb);
1327
}
1328

    
1329
/**
1330
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1331
 */
1332
static void hl_decode_mb_simple(H264Context *h){
1333
    hl_decode_mb_internal(h, 1);
1334
}
1335

    
1336
/**
1337
 * Process a macroblock; this handles edge cases, such as interlacing.
1338
 */
1339
static void av_noinline hl_decode_mb_complex(H264Context *h){
1340
    hl_decode_mb_internal(h, 0);
1341
}
1342

    
1343
void ff_h264_hl_decode_mb(H264Context *h){
1344
    MpegEncContext * const s = &h->s;
1345
    const int mb_xy= h->mb_xy;
1346
    const int mb_type= s->current_picture.mb_type[mb_xy];
1347
    int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1348

    
1349
    if (is_complex)
1350
        hl_decode_mb_complex(h);
1351
    else hl_decode_mb_simple(h);
1352
}
1353

    
1354
static int pred_weight_table(H264Context *h){
1355
    MpegEncContext * const s = &h->s;
1356
    int list, i;
1357
    int luma_def, chroma_def;
1358

    
1359
    h->use_weight= 0;
1360
    h->use_weight_chroma= 0;
1361
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1362
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1363
    luma_def = 1<<h->luma_log2_weight_denom;
1364
    chroma_def = 1<<h->chroma_log2_weight_denom;
1365

    
1366
    for(list=0; list<2; list++){
1367
        h->luma_weight_flag[list]   = 0;
1368
        h->chroma_weight_flag[list] = 0;
1369
        for(i=0; i<h->ref_count[list]; i++){
1370
            int luma_weight_flag, chroma_weight_flag;
1371

    
1372
            luma_weight_flag= get_bits1(&s->gb);
1373
            if(luma_weight_flag){
1374
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
1375
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
1376
                if(   h->luma_weight[list][i] != luma_def
1377
                   || h->luma_offset[list][i] != 0) {
1378
                    h->use_weight= 1;
1379
                    h->luma_weight_flag[list]= 1;
1380
                }
1381
            }else{
1382
                h->luma_weight[list][i]= luma_def;
1383
                h->luma_offset[list][i]= 0;
1384
            }
1385

    
1386
            if(CHROMA){
1387
                chroma_weight_flag= get_bits1(&s->gb);
1388
                if(chroma_weight_flag){
1389
                    int j;
1390
                    for(j=0; j<2; j++){
1391
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1392
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1393
                        if(   h->chroma_weight[list][i][j] != chroma_def
1394
                           || h->chroma_offset[list][i][j] != 0) {
1395
                            h->use_weight_chroma= 1;
1396
                            h->chroma_weight_flag[list]= 1;
1397
                        }
1398
                    }
1399
                }else{
1400
                    int j;
1401
                    for(j=0; j<2; j++){
1402
                        h->chroma_weight[list][i][j]= chroma_def;
1403
                        h->chroma_offset[list][i][j]= 0;
1404
                    }
1405
                }
1406
            }
1407
        }
1408
        if(h->slice_type_nos != FF_B_TYPE) break;
1409
    }
1410
    h->use_weight= h->use_weight || h->use_weight_chroma;
1411
    return 0;
1412
}
1413

    
1414
static void implicit_weight_table(H264Context *h){
1415
    MpegEncContext * const s = &h->s;
1416
    int ref0, ref1, i;
1417
    int cur_poc = s->current_picture_ptr->poc;
1418

    
1419
    for (i = 0; i < 2; i++) {
1420
        h->luma_weight_flag[i]   = 0;
1421
        h->chroma_weight_flag[i] = 0;
1422
    }
1423

    
1424
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1425
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1426
        h->use_weight= 0;
1427
        h->use_weight_chroma= 0;
1428
        return;
1429
    }
1430

    
1431
    h->use_weight= 2;
1432
    h->use_weight_chroma= 2;
1433
    h->luma_log2_weight_denom= 5;
1434
    h->chroma_log2_weight_denom= 5;
1435

    
1436
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1437
        int poc0 = h->ref_list[0][ref0].poc;
1438
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1439
            int poc1 = h->ref_list[1][ref1].poc;
1440
            int td = av_clip(poc1 - poc0, -128, 127);
1441
            if(td){
1442
                int tb = av_clip(cur_poc - poc0, -128, 127);
1443
                int tx = (16384 + (FFABS(td) >> 1)) / td;
1444
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1445
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
1446
                    h->implicit_weight[ref0][ref1] = 32;
1447
                else
1448
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1449
            }else
1450
                h->implicit_weight[ref0][ref1] = 32;
1451
        }
1452
    }
1453
}
1454

    
1455
/**
1456
 * instantaneous decoder refresh.
1457
 */
1458
static void idr(H264Context *h){
1459
    ff_h264_remove_all_refs(h);
1460
    h->prev_frame_num= 0;
1461
    h->prev_frame_num_offset= 0;
1462
    h->prev_poc_msb=
1463
    h->prev_poc_lsb= 0;
1464
}
1465

    
1466
/* forget old pics after a seek */
1467
static void flush_dpb(AVCodecContext *avctx){
1468
    H264Context *h= avctx->priv_data;
1469
    int i;
1470
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1471
        if(h->delayed_pic[i])
1472
            h->delayed_pic[i]->reference= 0;
1473
        h->delayed_pic[i]= NULL;
1474
    }
1475
    h->outputed_poc= INT_MIN;
1476
    h->prev_interlaced_frame = 1;
1477
    idr(h);
1478
    if(h->s.current_picture_ptr)
1479
        h->s.current_picture_ptr->reference= 0;
1480
    h->s.first_field= 0;
1481
    ff_h264_reset_sei(h);
1482
    ff_mpeg_flush(avctx);
1483
}
1484

    
1485
static int init_poc(H264Context *h){
1486
    MpegEncContext * const s = &h->s;
1487
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1488
    int field_poc[2];
1489
    Picture *cur = s->current_picture_ptr;
1490

    
1491
    h->frame_num_offset= h->prev_frame_num_offset;
1492
    if(h->frame_num < h->prev_frame_num)
1493
        h->frame_num_offset += max_frame_num;
1494

    
1495
    if(h->sps.poc_type==0){
1496
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1497

    
1498
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1499
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1500
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1501
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1502
        else
1503
            h->poc_msb = h->prev_poc_msb;
1504
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1505
        field_poc[0] =
1506
        field_poc[1] = h->poc_msb + h->poc_lsb;
1507
        if(s->picture_structure == PICT_FRAME)
1508
            field_poc[1] += h->delta_poc_bottom;
1509
    }else if(h->sps.poc_type==1){
1510
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1511
        int i;
1512

    
1513
        if(h->sps.poc_cycle_length != 0)
1514
            abs_frame_num = h->frame_num_offset + h->frame_num;
1515
        else
1516
            abs_frame_num = 0;
1517

    
1518
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
1519
            abs_frame_num--;
1520

    
1521
        expected_delta_per_poc_cycle = 0;
1522
        for(i=0; i < h->sps.poc_cycle_length; i++)
1523
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1524

    
1525
        if(abs_frame_num > 0){
1526
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1527
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1528

    
1529
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1530
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
1531
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1532
        } else
1533
            expectedpoc = 0;
1534

    
1535
        if(h->nal_ref_idc == 0)
1536
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1537

    
1538
        field_poc[0] = expectedpoc + h->delta_poc[0];
1539
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1540

    
1541
        if(s->picture_structure == PICT_FRAME)
1542
            field_poc[1] += h->delta_poc[1];
1543
    }else{
1544
        int poc= 2*(h->frame_num_offset + h->frame_num);
1545

    
1546
        if(!h->nal_ref_idc)
1547
            poc--;
1548

    
1549
        field_poc[0]= poc;
1550
        field_poc[1]= poc;
1551
    }
1552

    
1553
    if(s->picture_structure != PICT_BOTTOM_FIELD)
1554
        s->current_picture_ptr->field_poc[0]= field_poc[0];
1555
    if(s->picture_structure != PICT_TOP_FIELD)
1556
        s->current_picture_ptr->field_poc[1]= field_poc[1];
1557
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1558

    
1559
    return 0;
1560
}
1561

    
1562

    
1563
/**
1564
 * initialize scan tables
1565
 */
1566
static void init_scan_tables(H264Context *h){
1567
    MpegEncContext * const s = &h->s;
1568
    int i;
1569
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1570
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1571
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1572
    }else{
1573
        for(i=0; i<16; i++){
1574
#define T(x) (x>>2) | ((x<<2) & 0xF)
1575
            h->zigzag_scan[i] = T(zigzag_scan[i]);
1576
            h-> field_scan[i] = T( field_scan[i]);
1577
#undef T
1578
        }
1579
    }
1580
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1581
        memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1582
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1583
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1584
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1585
    }else{
1586
        for(i=0; i<64; i++){
1587
#define T(x) (x>>3) | ((x&7)<<3)
1588
            h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1589
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1590
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
1591
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1592
#undef T
1593
        }
1594
    }
1595
    if(h->sps.transform_bypass){ //FIXME same ugly
1596
        h->zigzag_scan_q0          = zigzag_scan;
1597
        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1598
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1599
        h->field_scan_q0           = field_scan;
1600
        h->field_scan8x8_q0        = field_scan8x8;
1601
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1602
    }else{
1603
        h->zigzag_scan_q0          = h->zigzag_scan;
1604
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1605
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1606
        h->field_scan_q0           = h->field_scan;
1607
        h->field_scan8x8_q0        = h->field_scan8x8;
1608
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1609
    }
1610
}
1611

    
1612
static void field_end(H264Context *h){
1613
    MpegEncContext * const s = &h->s;
1614
    AVCodecContext * const avctx= s->avctx;
1615
    s->mb_y= 0;
1616

    
1617
    s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1618
    s->current_picture_ptr->pict_type= s->pict_type;
1619

    
1620
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1621
        ff_vdpau_h264_set_reference_frames(s);
1622

    
1623
    if(!s->dropable) {
1624
        ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1625
        h->prev_poc_msb= h->poc_msb;
1626
        h->prev_poc_lsb= h->poc_lsb;
1627
    }
1628
    h->prev_frame_num_offset= h->frame_num_offset;
1629
    h->prev_frame_num= h->frame_num;
1630

    
1631
    if (avctx->hwaccel) {
1632
        if (avctx->hwaccel->end_frame(avctx) < 0)
1633
            av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1634
    }
1635

    
1636
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1637
        ff_vdpau_h264_picture_complete(s);
1638

    
1639
    /*
1640
     * FIXME: Error handling code does not seem to support interlaced
1641
     * when slices span multiple rows
1642
     * The ff_er_add_slice calls don't work right for bottom
1643
     * fields; they cause massive erroneous error concealing
1644
     * Error marking covers both fields (top and bottom).
1645
     * This causes a mismatched s->error_count
1646
     * and a bad error table. Further, the error count goes to
1647
     * INT_MAX when called for bottom field, because mb_y is
1648
     * past end by one (callers fault) and resync_mb_y != 0
1649
     * causes problems for the first MB line, too.
1650
     */
1651
    if (!FIELD_PICTURE)
1652
        ff_er_frame_end(s);
1653

    
1654
    MPV_frame_end(s);
1655

    
1656
    h->current_slice=0;
1657
}
1658

    
1659
/**
1660
 * Replicates H264 "master" context to thread contexts.
1661
 */
1662
static void clone_slice(H264Context *dst, H264Context *src)
1663
{
1664
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1665
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1666
    dst->s.current_picture      = src->s.current_picture;
1667
    dst->s.linesize             = src->s.linesize;
1668
    dst->s.uvlinesize           = src->s.uvlinesize;
1669
    dst->s.first_field          = src->s.first_field;
1670

    
1671
    dst->prev_poc_msb           = src->prev_poc_msb;
1672
    dst->prev_poc_lsb           = src->prev_poc_lsb;
1673
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1674
    dst->prev_frame_num         = src->prev_frame_num;
1675
    dst->short_ref_count        = src->short_ref_count;
1676

    
1677
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1678
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1679
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1680
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1681

    
1682
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1683
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1684
}
1685

    
1686
/**
1687
 * decodes a slice header.
1688
 * This will also call MPV_common_init() and frame_start() as needed.
1689
 *
1690
 * @param h h264context
1691
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1692
 *
1693
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1694
 */
1695
static int decode_slice_header(H264Context *h, H264Context *h0){
1696
    MpegEncContext * const s = &h->s;
1697
    MpegEncContext * const s0 = &h0->s;
1698
    unsigned int first_mb_in_slice;
1699
    unsigned int pps_id;
1700
    int num_ref_idx_active_override_flag;
1701
    unsigned int slice_type, tmp, i, j;
1702
    int default_ref_list_done = 0;
1703
    int last_pic_structure;
1704

    
1705
    s->dropable= h->nal_ref_idc == 0;
1706

    
1707
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1708
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1709
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1710
    }else{
1711
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1712
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1713
    }
1714

    
1715
    first_mb_in_slice= get_ue_golomb(&s->gb);
1716

    
1717
    if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1718
        if(h0->current_slice && FIELD_PICTURE){
1719
            field_end(h);
1720
        }
1721

    
1722
        h0->current_slice = 0;
1723
        if (!s0->first_field)
1724
            s->current_picture_ptr= NULL;
1725
    }
1726

    
1727
    slice_type= get_ue_golomb_31(&s->gb);
1728
    if(slice_type > 9){
1729
        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);
1730
        return -1;
1731
    }
1732
    if(slice_type > 4){
1733
        slice_type -= 5;
1734
        h->slice_type_fixed=1;
1735
    }else
1736
        h->slice_type_fixed=0;
1737

    
1738
    slice_type= golomb_to_pict_type[ slice_type ];
1739
    if (slice_type == FF_I_TYPE
1740
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1741
        default_ref_list_done = 1;
1742
    }
1743
    h->slice_type= slice_type;
1744
    h->slice_type_nos= slice_type & 3;
1745

    
1746
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1747

    
1748
    pps_id= get_ue_golomb(&s->gb);
1749
    if(pps_id>=MAX_PPS_COUNT){
1750
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1751
        return -1;
1752
    }
1753
    if(!h0->pps_buffers[pps_id]) {
1754
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1755
        return -1;
1756
    }
1757
    h->pps= *h0->pps_buffers[pps_id];
1758

    
1759
    if(!h0->sps_buffers[h->pps.sps_id]) {
1760
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1761
        return -1;
1762
    }
1763
    h->sps = *h0->sps_buffers[h->pps.sps_id];
1764

    
1765
    if(h == h0 && h->dequant_coeff_pps != pps_id){
1766
        h->dequant_coeff_pps = pps_id;
1767
        init_dequant_tables(h);
1768
    }
1769

    
1770
    s->mb_width= h->sps.mb_width;
1771
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1772

    
1773
    h->b_stride=  s->mb_width*4;
1774
    h->b8_stride= s->mb_width*2;
1775

    
1776
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1777
    if(h->sps.frame_mbs_only_flag)
1778
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1779
    else
1780
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1781

    
1782
    if (s->context_initialized
1783
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1784
        if(h != h0)
1785
            return -1;   // width / height changed during parallelized decoding
1786
        free_tables(h);
1787
        flush_dpb(s->avctx);
1788
        MPV_common_end(s);
1789
    }
1790
    if (!s->context_initialized) {
1791
        if(h != h0)
1792
            return -1;  // we cant (re-)initialize context during parallel decoding
1793

    
1794
        avcodec_set_dimensions(s->avctx, s->width, s->height);
1795
        s->avctx->sample_aspect_ratio= h->sps.sar;
1796
        if(!s->avctx->sample_aspect_ratio.den)
1797
            s->avctx->sample_aspect_ratio.den = 1;
1798

    
1799
        if(h->sps.video_signal_type_present_flag){
1800
            s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1801
            if(h->sps.colour_description_present_flag){
1802
                s->avctx->color_primaries = h->sps.color_primaries;
1803
                s->avctx->color_trc       = h->sps.color_trc;
1804
                s->avctx->colorspace      = h->sps.colorspace;
1805
            }
1806
        }
1807

    
1808
        if(h->sps.timing_info_present_flag){
1809
            int64_t den= h->sps.time_scale;
1810
            if(h->x264_build < 44U)
1811
                den *= 2;
1812
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1813
                      h->sps.num_units_in_tick, den, 1<<30);
1814
        }
1815
        s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1816
        s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1817

    
1818
        if (MPV_common_init(s) < 0)
1819
            return -1;
1820
        s->first_field = 0;
1821
        h->prev_interlaced_frame = 1;
1822

    
1823
        init_scan_tables(h);
1824
        ff_h264_alloc_tables(h);
1825

    
1826
        for(i = 1; i < s->avctx->thread_count; i++) {
1827
            H264Context *c;
1828
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1829
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1830
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1831
            c->sps = h->sps;
1832
            c->pps = h->pps;
1833
            init_scan_tables(c);
1834
            clone_tables(c, h);
1835
        }
1836

    
1837
        for(i = 0; i < s->avctx->thread_count; i++)
1838
            if(context_init(h->thread_context[i]) < 0)
1839
                return -1;
1840
    }
1841

    
1842
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1843

    
1844
    h->mb_mbaff = 0;
1845
    h->mb_aff_frame = 0;
1846
    last_pic_structure = s0->picture_structure;
1847
    if(h->sps.frame_mbs_only_flag){
1848
        s->picture_structure= PICT_FRAME;
1849
    }else{
1850
        if(get_bits1(&s->gb)) { //field_pic_flag
1851
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1852
        } else {
1853
            s->picture_structure= PICT_FRAME;
1854
            h->mb_aff_frame = h->sps.mb_aff;
1855
        }
1856
    }
1857
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1858

    
1859
    if(h0->current_slice == 0){
1860
        while(h->frame_num !=  h->prev_frame_num &&
1861
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1862
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1863
            if (ff_h264_frame_start(h) < 0)
1864
                return -1;
1865
            h->prev_frame_num++;
1866
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1867
            s->current_picture_ptr->frame_num= h->prev_frame_num;
1868
            ff_h264_execute_ref_pic_marking(h, NULL, 0);
1869
        }
1870

    
1871
        /* See if we have a decoded first field looking for a pair... */
1872
        if (s0->first_field) {
1873
            assert(s0->current_picture_ptr);
1874
            assert(s0->current_picture_ptr->data[0]);
1875
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1876

    
1877
            /* figure out if we have a complementary field pair */
1878
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1879
                /*
1880
                 * Previous field is unmatched. Don't display it, but let it
1881
                 * remain for reference if marked as such.
1882
                 */
1883
                s0->current_picture_ptr = NULL;
1884
                s0->first_field = FIELD_PICTURE;
1885

    
1886
            } else {
1887
                if (h->nal_ref_idc &&
1888
                        s0->current_picture_ptr->reference &&
1889
                        s0->current_picture_ptr->frame_num != h->frame_num) {
1890
                    /*
1891
                     * This and previous field were reference, but had
1892
                     * different frame_nums. Consider this field first in
1893
                     * pair. Throw away previous field except for reference
1894
                     * purposes.
1895
                     */
1896
                    s0->first_field = 1;
1897
                    s0->current_picture_ptr = NULL;
1898

    
1899
                } else {
1900
                    /* Second field in complementary pair */
1901
                    s0->first_field = 0;
1902
                }
1903
            }
1904

    
1905
        } else {
1906
            /* Frame or first field in a potentially complementary pair */
1907
            assert(!s0->current_picture_ptr);
1908
            s0->first_field = FIELD_PICTURE;
1909
        }
1910

    
1911
        if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1912
            s0->first_field = 0;
1913
            return -1;
1914
        }
1915
    }
1916
    if(h != h0)
1917
        clone_slice(h, h0);
1918

    
1919
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1920

    
1921
    assert(s->mb_num == s->mb_width * s->mb_height);
1922
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1923
       first_mb_in_slice                    >= s->mb_num){
1924
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1925
        return -1;
1926
    }
1927
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1928
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1929
    if (s->picture_structure == PICT_BOTTOM_FIELD)
1930
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
1931
    assert(s->mb_y < s->mb_height);
1932

    
1933
    if(s->picture_structure==PICT_FRAME){
1934
        h->curr_pic_num=   h->frame_num;
1935
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1936
    }else{
1937
        h->curr_pic_num= 2*h->frame_num + 1;
1938
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1939
    }
1940

    
1941
    if(h->nal_unit_type == NAL_IDR_SLICE){
1942
        get_ue_golomb(&s->gb); /* idr_pic_id */
1943
    }
1944

    
1945
    if(h->sps.poc_type==0){
1946
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1947

    
1948
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1949
            h->delta_poc_bottom= get_se_golomb(&s->gb);
1950
        }
1951
    }
1952

    
1953
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1954
        h->delta_poc[0]= get_se_golomb(&s->gb);
1955

    
1956
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1957
            h->delta_poc[1]= get_se_golomb(&s->gb);
1958
    }
1959

    
1960
    init_poc(h);
1961

    
1962
    if(h->pps.redundant_pic_cnt_present){
1963
        h->redundant_pic_count= get_ue_golomb(&s->gb);
1964
    }
1965

    
1966
    //set defaults, might be overridden a few lines later
1967
    h->ref_count[0]= h->pps.ref_count[0];
1968
    h->ref_count[1]= h->pps.ref_count[1];
1969

    
1970
    if(h->slice_type_nos != FF_I_TYPE){
1971
        if(h->slice_type_nos == FF_B_TYPE){
1972
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
1973
        }
1974
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
1975

    
1976
        if(num_ref_idx_active_override_flag){
1977
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1978
            if(h->slice_type_nos==FF_B_TYPE)
1979
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1980

    
1981
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1982
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1983
                h->ref_count[0]= h->ref_count[1]= 1;
1984
                return -1;
1985
            }
1986
        }
1987
        if(h->slice_type_nos == FF_B_TYPE)
1988
            h->list_count= 2;
1989
        else
1990
            h->list_count= 1;
1991
    }else
1992
        h->list_count= 0;
1993

    
1994
    if(!default_ref_list_done){
1995
        ff_h264_fill_default_ref_list(h);
1996
    }
1997

    
1998
    if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1999
        return -1;
2000

    
2001
    if(h->slice_type_nos!=FF_I_TYPE){
2002
        s->last_picture_ptr= &h->ref_list[0][0];
2003
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
2004
    }
2005
    if(h->slice_type_nos==FF_B_TYPE){
2006
        s->next_picture_ptr= &h->ref_list[1][0];
2007
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
2008
    }
2009

    
2010
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
2011
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
2012
        pred_weight_table(h);
2013
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
2014
        implicit_weight_table(h);
2015
    else {
2016
        h->use_weight = 0;
2017
        for (i = 0; i < 2; i++) {
2018
            h->luma_weight_flag[i]   = 0;
2019
            h->chroma_weight_flag[i] = 0;
2020
        }
2021
    }
2022

    
2023
    if(h->nal_ref_idc)
2024
        ff_h264_decode_ref_pic_marking(h0, &s->gb);
2025

    
2026
    if(FRAME_MBAFF)
2027
        ff_h264_fill_mbaff_ref_list(h);
2028

    
2029
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2030
        ff_h264_direct_dist_scale_factor(h);
2031
    ff_h264_direct_ref_list_init(h);
2032

    
2033
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2034
        tmp = get_ue_golomb_31(&s->gb);
2035
        if(tmp > 2){
2036
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2037
            return -1;
2038
        }
2039
        h->cabac_init_idc= tmp;
2040
    }
2041

    
2042
    h->last_qscale_diff = 0;
2043
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2044
    if(tmp>51){
2045
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2046
        return -1;
2047
    }
2048
    s->qscale= tmp;
2049
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2050
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2051
    //FIXME qscale / qp ... stuff
2052
    if(h->slice_type == FF_SP_TYPE){
2053
        get_bits1(&s->gb); /* sp_for_switch_flag */
2054
    }
2055
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2056
        get_se_golomb(&s->gb); /* slice_qs_delta */
2057
    }
2058

    
2059
    h->deblocking_filter = 1;
2060
    h->slice_alpha_c0_offset = 52;
2061
    h->slice_beta_offset = 52;
2062
    if( h->pps.deblocking_filter_parameters_present ) {
2063
        tmp= get_ue_golomb_31(&s->gb);
2064
        if(tmp > 2){
2065
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2066
            return -1;
2067
        }
2068
        h->deblocking_filter= tmp;
2069
        if(h->deblocking_filter < 2)
2070
            h->deblocking_filter^= 1; // 1<->0
2071

    
2072
        if( h->deblocking_filter ) {
2073
            h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2074
            h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
2075
            if(   h->slice_alpha_c0_offset > 104U
2076
               || h->slice_beta_offset     > 104U){
2077
                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);
2078
                return -1;
2079
            }
2080
        }
2081
    }
2082

    
2083
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2084
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2085
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2086
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2087
        h->deblocking_filter= 0;
2088

    
2089
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2090
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2091
            /* Cheat slightly for speed:
2092
               Do not bother to deblock across slices. */
2093
            h->deblocking_filter = 2;
2094
        } else {
2095
            h0->max_contexts = 1;
2096
            if(!h0->single_decode_warning) {
2097
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2098
                h0->single_decode_warning = 1;
2099
            }
2100
            if(h != h0)
2101
                return 1; // deblocking switched inside frame
2102
        }
2103
    }
2104
    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]);
2105

    
2106
#if 0 //FMO
2107
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2108
        slice_group_change_cycle= get_bits(&s->gb, ?);
2109
#endif
2110

    
2111
    h0->last_slice_type = slice_type;
2112
    h->slice_num = ++h0->current_slice;
2113
    if(h->slice_num >= MAX_SLICES){
2114
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2115
    }
2116

    
2117
    for(j=0; j<2; j++){
2118
        int id_list[16];
2119
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2120
        for(i=0; i<16; i++){
2121
            id_list[i]= 60;
2122
            if(h->ref_list[j][i].data[0]){
2123
                int k;
2124
                uint8_t *base= h->ref_list[j][i].base[0];
2125
                for(k=0; k<h->short_ref_count; k++)
2126
                    if(h->short_ref[k]->base[0] == base){
2127
                        id_list[i]= k;
2128
                        break;
2129
                    }
2130
                for(k=0; k<h->long_ref_count; k++)
2131
                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2132
                        id_list[i]= h->short_ref_count + k;
2133
                        break;
2134
                    }
2135
            }
2136
        }
2137

    
2138
        ref2frm[0]=
2139
        ref2frm[1]= -1;
2140
        for(i=0; i<16; i++)
2141
            ref2frm[i+2]= 4*id_list[i]
2142
                          +(h->ref_list[j][i].reference&3);
2143
        ref2frm[18+0]=
2144
        ref2frm[18+1]= -1;
2145
        for(i=16; i<48; i++)
2146
            ref2frm[i+4]= 4*id_list[(i-16)>>1]
2147
                          +(h->ref_list[j][i].reference&3);
2148
    }
2149

    
2150
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2151
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2152

    
2153
    s->avctx->refs= h->sps.ref_frame_count;
2154

    
2155
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2156
        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",
2157
               h->slice_num,
2158
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2159
               first_mb_in_slice,
2160
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2161
               pps_id, h->frame_num,
2162
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2163
               h->ref_count[0], h->ref_count[1],
2164
               s->qscale,
2165
               h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
2166
               h->use_weight,
2167
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2168
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2169
               );
2170
    }
2171

    
2172
    return 0;
2173
}
2174

    
2175
int ff_h264_get_slice_type(const H264Context *h)
2176
{
2177
    switch (h->slice_type) {
2178
    case FF_P_TYPE:  return 0;
2179
    case FF_B_TYPE:  return 1;
2180
    case FF_I_TYPE:  return 2;
2181
    case FF_SP_TYPE: return 3;
2182
    case FF_SI_TYPE: return 4;
2183
    default:         return -1;
2184
    }
2185
}
2186

    
2187
static void loop_filter(H264Context *h){
2188
    MpegEncContext * const s = &h->s;
2189
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2190
    int linesize, uvlinesize, mb_x, mb_y;
2191
    const int end_mb_y= s->mb_y + FRAME_MBAFF;
2192
    const int old_slice_type= h->slice_type;
2193

    
2194
    if(h->deblocking_filter) {
2195
        for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2196
            for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2197
                int mb_xy, mb_type;
2198
                mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2199
                h->slice_num= h->slice_table[mb_xy];
2200
                mb_type= s->current_picture.mb_type[mb_xy];
2201
                h->list_count= h->list_counts[mb_xy];
2202

    
2203
                if(FRAME_MBAFF)
2204
                    h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2205

    
2206
                s->mb_x= mb_x;
2207
                s->mb_y= mb_y;
2208
                dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2209
                dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2210
                dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2211
                    //FIXME simplify above
2212

    
2213
                if (MB_FIELD) {
2214
                    linesize   = h->mb_linesize   = s->linesize * 2;
2215
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2216
                    if(mb_y&1){ //FIXME move out of this function?
2217
                        dest_y -= s->linesize*15;
2218
                        dest_cb-= s->uvlinesize*7;
2219
                        dest_cr-= s->uvlinesize*7;
2220
                    }
2221
                } else {
2222
                    linesize   = h->mb_linesize   = s->linesize;
2223
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2224
                }
2225
                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2226
                if(fill_filter_caches(h, mb_type))
2227
                    continue;
2228
                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2229
                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2230

    
2231
                if (FRAME_MBAFF) {
2232
                    ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2233
                } else {
2234
                    ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2235
                }
2236
            }
2237
        }
2238
    }
2239
    h->slice_type= old_slice_type;
2240
    s->mb_x= 0;
2241
    s->mb_y= end_mb_y - FRAME_MBAFF;
2242
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2243
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2244
}
2245

    
2246
static void predict_field_decoding_flag(H264Context *h){
2247
    MpegEncContext * const s = &h->s;
2248
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2249
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
2250
                ? s->current_picture.mb_type[mb_xy-1]
2251
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
2252
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
2253
                : 0;
2254
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2255
}
2256

    
2257
static int decode_slice(struct AVCodecContext *avctx, void *arg){
2258
    H264Context *h = *(void**)arg;
2259
    MpegEncContext * const s = &h->s;
2260
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2261

    
2262
    s->mb_skip_run= -1;
2263

    
2264
    h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2265
                    (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2266

    
2267
    if( h->pps.cabac ) {
2268
        /* realign */
2269
        align_get_bits( &s->gb );
2270

    
2271
        /* init cabac */
2272
        ff_init_cabac_states( &h->cabac);
2273
        ff_init_cabac_decoder( &h->cabac,
2274
                               s->gb.buffer + get_bits_count(&s->gb)/8,
2275
                               (get_bits_left(&s->gb) + 7)/8);
2276

    
2277
        ff_h264_init_cabac_states(h);
2278

    
2279
        for(;;){
2280
//START_TIMER
2281
            int ret = ff_h264_decode_mb_cabac(h);
2282
            int eos;
2283
//STOP_TIMER("decode_mb_cabac")
2284

    
2285
            if(ret>=0) ff_h264_hl_decode_mb(h);
2286

    
2287
            if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2288
                s->mb_y++;
2289

    
2290
                ret = ff_h264_decode_mb_cabac(h);
2291

    
2292
                if(ret>=0) ff_h264_hl_decode_mb(h);
2293
                s->mb_y--;
2294
            }
2295
            eos = get_cabac_terminate( &h->cabac );
2296

    
2297
            if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
2298
                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);
2299
                return 0;
2300
            }
2301
            if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2302
                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);
2303
                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);
2304
                return -1;
2305
            }
2306

    
2307
            if( ++s->mb_x >= s->mb_width ) {
2308
                s->mb_x = 0;
2309
                loop_filter(h);
2310
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2311
                ++s->mb_y;
2312
                if(FIELD_OR_MBAFF_PICTURE) {
2313
                    ++s->mb_y;
2314
                    if(FRAME_MBAFF && s->mb_y < s->mb_height)
2315
                        predict_field_decoding_flag(h);
2316
                }
2317
            }
2318

    
2319
            if( eos || s->mb_y >= s->mb_height ) {
2320
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2321
                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);
2322
                return 0;
2323
            }
2324
        }
2325

    
2326
    } else {
2327
        for(;;){
2328
            int ret = ff_h264_decode_mb_cavlc(h);
2329

    
2330
            if(ret>=0) ff_h264_hl_decode_mb(h);
2331

    
2332
            if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2333
                s->mb_y++;
2334
                ret = ff_h264_decode_mb_cavlc(h);
2335

    
2336
                if(ret>=0) ff_h264_hl_decode_mb(h);
2337
                s->mb_y--;
2338
            }
2339

    
2340
            if(ret<0){
2341
                av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2342
                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);
2343

    
2344
                return -1;
2345
            }
2346

    
2347
            if(++s->mb_x >= s->mb_width){
2348
                s->mb_x=0;
2349
                loop_filter(h);
2350
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2351
                ++s->mb_y;
2352
                if(FIELD_OR_MBAFF_PICTURE) {
2353
                    ++s->mb_y;
2354
                    if(FRAME_MBAFF && s->mb_y < s->mb_height)
2355
                        predict_field_decoding_flag(h);
2356
                }
2357
                if(s->mb_y >= s->mb_height){
2358
                    tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2359

    
2360
                    if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2361
                        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);
2362

    
2363
                        return 0;
2364
                    }else{
2365
                        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);
2366

    
2367
                        return -1;
2368
                    }
2369
                }
2370
            }
2371

    
2372
            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2373
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2374
                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2375
                    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);
2376

    
2377
                    return 0;
2378
                }else{
2379
                    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);
2380

    
2381
                    return -1;
2382
                }
2383
            }
2384
        }
2385
    }
2386

    
2387
#if 0
2388
    for(;s->mb_y < s->mb_height; s->mb_y++){
2389
        for(;s->mb_x < s->mb_width; s->mb_x++){
2390
            int ret= decode_mb(h);
2391

2392
            ff_h264_hl_decode_mb(h);
2393

2394
            if(ret<0){
2395
                av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2396
                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);
2397

2398
                return -1;
2399
            }
2400

2401
            if(++s->mb_x >= s->mb_width){
2402
                s->mb_x=0;
2403
                if(++s->mb_y >= s->mb_height){
2404
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
2405
                        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);
2406

2407
                        return 0;
2408
                    }else{
2409
                        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);
2410

2411
                        return -1;
2412
                    }
2413
                }
2414
            }
2415

2416
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2417
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
2418
                    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);
2419

2420
                    return 0;
2421
                }else{
2422
                    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);
2423

2424
                    return -1;
2425
                }
2426
            }
2427
        }
2428
        s->mb_x=0;
2429
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
2430
    }
2431
#endif
2432
    return -1; //not reached
2433
}
2434

    
2435
/**
2436
 * Call decode_slice() for each context.
2437
 *
2438
 * @param h h264 master context
2439
 * @param context_count number of contexts to execute
2440
 */
2441
static void execute_decode_slices(H264Context *h, int context_count){
2442
    MpegEncContext * const s = &h->s;
2443
    AVCodecContext * const avctx= s->avctx;
2444
    H264Context *hx;
2445
    int i;
2446

    
2447
    if (s->avctx->hwaccel)
2448
        return;
2449
    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2450
        return;
2451
    if(context_count == 1) {
2452
        decode_slice(avctx, &h);
2453
    } else {
2454
        for(i = 1; i < context_count; i++) {
2455
            hx = h->thread_context[i];
2456
            hx->s.error_recognition = avctx->error_recognition;
2457
            hx->s.error_count = 0;
2458
        }
2459

    
2460
        avctx->execute(avctx, (void *)decode_slice,
2461
                       h->thread_context, NULL, context_count, sizeof(void*));
2462

    
2463
        /* pull back stuff from slices to master context */
2464
        hx = h->thread_context[context_count - 1];
2465
        s->mb_x = hx->s.mb_x;
2466
        s->mb_y = hx->s.mb_y;
2467
        s->dropable = hx->s.dropable;
2468
        s->picture_structure = hx->s.picture_structure;
2469
        for(i = 1; i < context_count; i++)
2470
            h->s.error_count += h->thread_context[i]->s.error_count;
2471
    }
2472
}
2473

    
2474

    
2475
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2476
    MpegEncContext * const s = &h->s;
2477
    AVCodecContext * const avctx= s->avctx;
2478
    int buf_index=0;
2479
    H264Context *hx; ///< thread context
2480
    int context_count = 0;
2481
    int next_avc= h->is_avc ? 0 : buf_size;
2482

    
2483
    h->max_contexts = avctx->thread_count;
2484
#if 0
2485
    int i;
2486
    for(i=0; i<50; i++){
2487
        av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2488
    }
2489
#endif
2490
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2491
        h->current_slice = 0;
2492
        if (!s->first_field)
2493
            s->current_picture_ptr= NULL;
2494
        ff_h264_reset_sei(h);
2495
    }
2496

    
2497
    for(;;){
2498
        int consumed;
2499
        int dst_length;
2500
        int bit_length;
2501
        const uint8_t *ptr;
2502
        int i, nalsize = 0;
2503
        int err;
2504

    
2505
        if(buf_index >= next_avc) {
2506
            if(buf_index >= buf_size) break;
2507
            nalsize = 0;
2508
            for(i = 0; i < h->nal_length_size; i++)
2509
                nalsize = (nalsize << 8) | buf[buf_index++];
2510
            if(nalsize <= 1 || nalsize > buf_size - buf_index){
2511
                if(nalsize == 1){
2512
                    buf_index++;
2513
                    continue;
2514
                }else{
2515
                    av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2516
                    break;
2517
                }
2518
            }
2519
            next_avc= buf_index + nalsize;
2520
        } else {
2521
            // start code prefix search
2522
            for(; buf_index + 3 < next_avc; buf_index++){
2523
                // This should always succeed in the first iteration.
2524
                if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2525
                    break;
2526
            }
2527

    
2528
            if(buf_index+3 >= buf_size) break;
2529

    
2530
            buf_index+=3;
2531
            if(buf_index >= next_avc) continue;
2532
        }
2533

    
2534
        hx = h->thread_context[context_count];
2535

    
2536
        ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2537
        if (ptr==NULL || dst_length < 0){
2538
            return -1;
2539
        }
2540
        i= buf_index + consumed;
2541
        if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
2542
           buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
2543
            s->workaround_bugs |= FF_BUG_TRUNCATED;
2544

    
2545
        if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
2546
        while(ptr[dst_length - 1] == 0 && dst_length > 0)
2547
            dst_length--;
2548
        }
2549
        bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2550

    
2551
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
2552
            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);
2553
        }
2554

    
2555
        if (h->is_avc && (nalsize != consumed) && nalsize){
2556
            av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2557
        }
2558

    
2559
        buf_index += consumed;
2560

    
2561
        if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2562
           ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2563
            continue;
2564

    
2565
      again:
2566
        err = 0;
2567
        switch(hx->nal_unit_type){
2568
        case NAL_IDR_SLICE:
2569
            if (h->nal_unit_type != NAL_IDR_SLICE) {
2570
                av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2571
                return -1;
2572
            }
2573
            idr(h); //FIXME ensure we don't loose some frames if there is reordering
2574
        case NAL_SLICE:
2575
            init_get_bits(&hx->s.gb, ptr, bit_length);
2576
            hx->intra_gb_ptr=
2577
            hx->inter_gb_ptr= &hx->s.gb;
2578
            hx->s.data_partitioning = 0;
2579

    
2580
            if((err = decode_slice_header(hx, h)))
2581
               break;
2582

    
2583
            avctx->profile = hx->sps.profile_idc;
2584
            avctx->level   = hx->sps.level_idc;
2585

    
2586
            if (s->avctx->hwaccel && h->current_slice == 1) {
2587
                if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2588
                    return -1;
2589
            }
2590

    
2591
            s->current_picture_ptr->key_frame |=
2592
                    (hx->nal_unit_type == NAL_IDR_SLICE) ||
2593
                    (h->sei_recovery_frame_cnt >= 0);
2594
            if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2595
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2596
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2597
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2598
               && avctx->skip_frame < AVDISCARD_ALL){
2599
                if(avctx->hwaccel) {
2600
                    if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2601
                        return -1;
2602
                }else
2603
                if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2604
                    static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2605
                    ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2606
                    ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2607
                }else
2608
                    context_count++;
2609
            }
2610
            break;
2611
        case NAL_DPA:
2612
            init_get_bits(&hx->s.gb, ptr, bit_length);
2613
            hx->intra_gb_ptr=
2614
            hx->inter_gb_ptr= NULL;
2615

    
2616
            if ((err = decode_slice_header(hx, h)) < 0)
2617
                break;
2618

    
2619
            avctx->profile = hx->sps.profile_idc;
2620
            avctx->level   = hx->sps.level_idc;
2621

    
2622
            hx->s.data_partitioning = 1;
2623

    
2624
            break;
2625
        case NAL_DPB:
2626
            init_get_bits(&hx->intra_gb, ptr, bit_length);
2627
            hx->intra_gb_ptr= &hx->intra_gb;
2628
            break;
2629
        case NAL_DPC:
2630
            init_get_bits(&hx->inter_gb, ptr, bit_length);
2631
            hx->inter_gb_ptr= &hx->inter_gb;
2632

    
2633
            if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2634
               && s->context_initialized
2635
               && s->hurry_up < 5
2636
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2637
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2638
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2639
               && avctx->skip_frame < AVDISCARD_ALL)
2640
                context_count++;
2641
            break;
2642
        case NAL_SEI:
2643
            init_get_bits(&s->gb, ptr, bit_length);
2644
            ff_h264_decode_sei(h);
2645
            break;
2646
        case NAL_SPS:
2647
            init_get_bits(&s->gb, ptr, bit_length);
2648
            ff_h264_decode_seq_parameter_set(h);
2649

    
2650
            if(s->flags& CODEC_FLAG_LOW_DELAY)
2651
                s->low_delay=1;
2652

    
2653
            if(avctx->has_b_frames < 2)
2654
                avctx->has_b_frames= !s->low_delay;
2655
            break;
2656
        case NAL_PPS:
2657
            init_get_bits(&s->gb, ptr, bit_length);
2658

    
2659
            ff_h264_decode_picture_parameter_set(h, bit_length);
2660

    
2661
            break;
2662
        case NAL_AUD:
2663
        case NAL_END_SEQUENCE:
2664
        case NAL_END_STREAM:
2665
        case NAL_FILLER_DATA:
2666
        case NAL_SPS_EXT:
2667
        case NAL_AUXILIARY_SLICE:
2668
            break;
2669
        default:
2670
            av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2671
        }
2672

    
2673
        if(context_count == h->max_contexts) {
2674
            execute_decode_slices(h, context_count);
2675
            context_count = 0;
2676
        }
2677

    
2678
        if (err < 0)
2679
            av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2680
        else if(err == 1) {
2681
            /* Slice could not be decoded in parallel mode, copy down
2682
             * NAL unit stuff to context 0 and restart. Note that
2683
             * rbsp_buffer is not transferred, but since we no longer
2684
             * run in parallel mode this should not be an issue. */
2685
            h->nal_unit_type = hx->nal_unit_type;
2686
            h->nal_ref_idc   = hx->nal_ref_idc;
2687
            hx = h;
2688
            goto again;
2689
        }
2690
    }
2691
    if(context_count)
2692
        execute_decode_slices(h, context_count);
2693
    return buf_index;
2694
}
2695

    
2696
/**
2697
 * returns the number of bytes consumed for building the current frame
2698
 */
2699
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2700
        if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2701
        if(pos+10>buf_size) pos=buf_size; // oops ;)
2702

    
2703
        return pos;
2704
}
2705

    
2706
static int decode_frame(AVCodecContext *avctx,
2707
                             void *data, int *data_size,
2708
                             AVPacket *avpkt)
2709
{
2710
    const uint8_t *buf = avpkt->data;
2711
    int buf_size = avpkt->size;
2712
    H264Context *h = avctx->priv_data;
2713
    MpegEncContext *s = &h->s;
2714
    AVFrame *pict = data;
2715
    int buf_index;
2716

    
2717
    s->flags= avctx->flags;
2718
    s->flags2= avctx->flags2;
2719

    
2720
   /* end of stream, output what is still in the buffers */
2721
    if (buf_size == 0) {
2722
        Picture *out;
2723
        int i, out_idx;
2724

    
2725
//FIXME factorize this with the output code below
2726
        out = h->delayed_pic[0];
2727
        out_idx = 0;
2728
        for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2729
            if(h->delayed_pic[i]->poc < out->poc){
2730
                out = h->delayed_pic[i];
2731
                out_idx = i;
2732
            }
2733

    
2734
        for(i=out_idx; h->delayed_pic[i]; i++)
2735
            h->delayed_pic[i] = h->delayed_pic[i+1];
2736

    
2737
        if(out){
2738
            *data_size = sizeof(AVFrame);
2739
            *pict= *(AVFrame*)out;
2740
        }
2741

    
2742
        return 0;
2743
    }
2744

    
2745
    buf_index=decode_nal_units(h, buf, buf_size);
2746
    if(buf_index < 0)
2747
        return -1;
2748

    
2749
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2750
        if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2751
        av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2752
        return -1;
2753
    }
2754

    
2755
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2756
        Picture *out = s->current_picture_ptr;
2757
        Picture *cur = s->current_picture_ptr;
2758
        int i, pics, out_of_order, out_idx;
2759

    
2760
        field_end(h);
2761

    
2762
        if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2763
            /* Wait for second field. */
2764
            *data_size = 0;
2765

    
2766
        } else {
2767
            cur->interlaced_frame = 0;
2768
            cur->repeat_pict = 0;
2769

    
2770
            /* Signal interlacing information externally. */
2771
            /* Prioritize picture timing SEI information over used decoding process if it exists. */
2772

    
2773
            if(h->sps.pic_struct_present_flag){
2774
                switch (h->sei_pic_struct)
2775
                {
2776
                case SEI_PIC_STRUCT_FRAME:
2777
                    break;
2778
                case SEI_PIC_STRUCT_TOP_FIELD:
2779
                case SEI_PIC_STRUCT_BOTTOM_FIELD:
2780
                    cur->interlaced_frame = 1;
2781
                    break;
2782
                case SEI_PIC_STRUCT_TOP_BOTTOM:
2783
                case SEI_PIC_STRUCT_BOTTOM_TOP:
2784
                    if (FIELD_OR_MBAFF_PICTURE)
2785
                        cur->interlaced_frame = 1;
2786
                    else
2787
                        // try to flag soft telecine progressive
2788
                        cur->interlaced_frame = h->prev_interlaced_frame;
2789
                    break;
2790
                case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2791
                case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2792
                    // Signal the possibility of telecined film externally (pic_struct 5,6)
2793
                    // From these hints, let the applications decide if they apply deinterlacing.
2794
                    cur->repeat_pict = 1;
2795
                    break;
2796
                case SEI_PIC_STRUCT_FRAME_DOUBLING:
2797
                    // Force progressive here, as doubling interlaced frame is a bad idea.
2798
                    cur->repeat_pict = 2;
2799
                    break;
2800
                case SEI_PIC_STRUCT_FRAME_TRIPLING:
2801
                    cur->repeat_pict = 4;
2802
                    break;
2803
                }
2804

    
2805
                if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2806
                    cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2807
            }else{
2808
                /* Derive interlacing flag from used decoding process. */
2809
                cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2810
            }
2811
            h->prev_interlaced_frame = cur->interlaced_frame;
2812

    
2813
            if (cur->field_poc[0] != cur->field_poc[1]){
2814
                /* Derive top_field_first from field pocs. */
2815
                cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2816
            }else{
2817
                if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2818
                    /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2819
                    if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2820
                      || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2821
                        cur->top_field_first = 1;
2822
                    else
2823
                        cur->top_field_first = 0;
2824
                }else{
2825
                    /* Most likely progressive */
2826
                    cur->top_field_first = 0;
2827
                }
2828
            }
2829

    
2830
        //FIXME do something with unavailable reference frames
2831

    
2832
            /* Sort B-frames into display order */
2833

    
2834
            if(h->sps.bitstream_restriction_flag
2835
               && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2836
                s->avctx->has_b_frames = h->sps.num_reorder_frames;
2837
                s->low_delay = 0;
2838
            }
2839

    
2840
            if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2841
               && !h->sps.bitstream_restriction_flag){
2842
                s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2843
                s->low_delay= 0;
2844
            }
2845

    
2846
            pics = 0;
2847
            while(h->delayed_pic[pics]) pics++;
2848

    
2849
            assert(pics <= MAX_DELAYED_PIC_COUNT);
2850

    
2851
            h->delayed_pic[pics++] = cur;
2852
            if(cur->reference == 0)
2853
                cur->reference = DELAYED_PIC_REF;
2854

    
2855
            out = h->delayed_pic[0];
2856
            out_idx = 0;
2857
            for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2858
                if(h->delayed_pic[i]->poc < out->poc){
2859
                    out = h->delayed_pic[i];
2860
                    out_idx = i;
2861
                }
2862
            if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2863
                h->outputed_poc= INT_MIN;
2864
            out_of_order = out->poc < h->outputed_poc;
2865

    
2866
            if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2867
                { }
2868
            else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2869
               || (s->low_delay &&
2870
                ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2871
                 || cur->pict_type == FF_B_TYPE)))
2872
            {
2873
                s->low_delay = 0;
2874
                s->avctx->has_b_frames++;
2875
            }
2876

    
2877
            if(out_of_order || pics > s->avctx->has_b_frames){
2878
                out->reference &= ~DELAYED_PIC_REF;
2879
                for(i=out_idx; h->delayed_pic[i]; i++)
2880
                    h->delayed_pic[i] = h->delayed_pic[i+1];
2881
            }
2882
            if(!out_of_order && pics > s->avctx->has_b_frames){
2883
                *data_size = sizeof(AVFrame);
2884

    
2885
                if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2886
                    h->outputed_poc = INT_MIN;
2887
                } else
2888
                    h->outputed_poc = out->poc;
2889
                *pict= *(AVFrame*)out;
2890
            }else{
2891
                av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2892
            }
2893
        }
2894
    }
2895

    
2896
    assert(pict->data[0] || !*data_size);
2897
    ff_print_debug_info(s, pict);
2898
//printf("out %d\n", (int)pict->data[0]);
2899

    
2900
    return get_consumed_bytes(s, buf_index, buf_size);
2901
}
2902
#if 0
2903
static inline void fill_mb_avail(H264Context *h){
2904
    MpegEncContext * const s = &h->s;
2905
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2906

2907
    if(s->mb_y){
2908
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2909
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2910
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2911
    }else{
2912
        h->mb_avail[0]=
2913
        h->mb_avail[1]=
2914
        h->mb_avail[2]= 0;
2915
    }
2916
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2917
    h->mb_avail[4]= 1; //FIXME move out
2918
    h->mb_avail[5]= 0; //FIXME move out
2919
}
2920
#endif
2921

    
2922
#ifdef TEST
2923
#undef printf
2924
#undef random
2925
#define COUNT 8000
2926
#define SIZE (COUNT*40)
2927
int main(void){
2928
    int i;
2929
    uint8_t temp[SIZE];
2930
    PutBitContext pb;
2931
    GetBitContext gb;
2932
//    int int_temp[10000];
2933
    DSPContext dsp;
2934
    AVCodecContext avctx;
2935

    
2936
    dsputil_init(&dsp, &avctx);
2937

    
2938
    init_put_bits(&pb, temp, SIZE);
2939
    printf("testing unsigned exp golomb\n");
2940
    for(i=0; i<COUNT; i++){
2941
        START_TIMER
2942
        set_ue_golomb(&pb, i);
2943
        STOP_TIMER("set_ue_golomb");
2944
    }
2945
    flush_put_bits(&pb);
2946

    
2947
    init_get_bits(&gb, temp, 8*SIZE);
2948
    for(i=0; i<COUNT; i++){
2949
        int j, s;
2950

    
2951
        s= show_bits(&gb, 24);
2952

    
2953
        START_TIMER
2954
        j= get_ue_golomb(&gb);
2955
        if(j != i){
2956
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2957
//            return -1;
2958
        }
2959
        STOP_TIMER("get_ue_golomb");
2960
    }
2961

    
2962

    
2963
    init_put_bits(&pb, temp, SIZE);
2964
    printf("testing signed exp golomb\n");
2965
    for(i=0; i<COUNT; i++){
2966
        START_TIMER
2967
        set_se_golomb(&pb, i - COUNT/2);
2968
        STOP_TIMER("set_se_golomb");
2969
    }
2970
    flush_put_bits(&pb);
2971

    
2972
    init_get_bits(&gb, temp, 8*SIZE);
2973
    for(i=0; i<COUNT; i++){
2974
        int j, s;
2975

    
2976
        s= show_bits(&gb, 24);
2977

    
2978
        START_TIMER
2979
        j= get_se_golomb(&gb);
2980
        if(j != i - COUNT/2){
2981
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2982
//            return -1;
2983
        }
2984
        STOP_TIMER("get_se_golomb");
2985
    }
2986

    
2987
#if 0
2988
    printf("testing 4x4 (I)DCT\n");
2989

2990
    DCTELEM block[16];
2991
    uint8_t src[16], ref[16];
2992
    uint64_t error= 0, max_error=0;
2993

2994
    for(i=0; i<COUNT; i++){
2995
        int j;
2996
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2997
        for(j=0; j<16; j++){
2998
            ref[j]= random()%255;
2999
            src[j]= random()%255;
3000
        }
3001

3002
        h264_diff_dct_c(block, src, ref, 4);
3003

3004
        //normalize
3005
        for(j=0; j<16; j++){
3006
//            printf("%d ", block[j]);
3007
            block[j]= block[j]*4;
3008
            if(j&1) block[j]= (block[j]*4 + 2)/5;
3009
            if(j&4) block[j]= (block[j]*4 + 2)/5;
3010
        }
3011
//        printf("\n");
3012

3013
        s->dsp.h264_idct_add(ref, block, 4);
3014
/*        for(j=0; j<16; j++){
3015
            printf("%d ", ref[j]);
3016
        }
3017
        printf("\n");*/
3018

3019
        for(j=0; j<16; j++){
3020
            int diff= FFABS(src[j] - ref[j]);
3021

3022
            error+= diff*diff;
3023
            max_error= FFMAX(max_error, diff);
3024
        }
3025
    }
3026
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3027
    printf("testing quantizer\n");
3028
    for(qp=0; qp<52; qp++){
3029
        for(i=0; i<16; i++)
3030
            src1_block[i]= src2_block[i]= random()%255;
3031

3032
    }
3033
    printf("Testing NAL layer\n");
3034

3035
    uint8_t bitstream[COUNT];
3036
    uint8_t nal[COUNT*2];
3037
    H264Context h;
3038
    memset(&h, 0, sizeof(H264Context));
3039

3040
    for(i=0; i<COUNT; i++){
3041
        int zeros= i;
3042
        int nal_length;
3043
        int consumed;
3044
        int out_length;
3045
        uint8_t *out;
3046
        int j;
3047

3048
        for(j=0; j<COUNT; j++){
3049
            bitstream[j]= (random() % 255) + 1;
3050
        }
3051

3052
        for(j=0; j<zeros; j++){
3053
            int pos= random() % COUNT;
3054
            while(bitstream[pos] == 0){
3055
                pos++;
3056
                pos %= COUNT;
3057
            }
3058
            bitstream[pos]=0;
3059
        }
3060

3061
        START_TIMER
3062

3063
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3064
        if(nal_length<0){
3065
            printf("encoding failed\n");
3066
            return -1;
3067
        }
3068

3069
        out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3070

3071
        STOP_TIMER("NAL")
3072

3073
        if(out_length != COUNT){
3074
            printf("incorrect length %d %d\n", out_length, COUNT);
3075
            return -1;
3076
        }
3077

3078
        if(consumed != nal_length){
3079
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
3080
            return -1;
3081
        }
3082

3083
        if(memcmp(bitstream, out, COUNT)){
3084
            printf("mismatch\n");
3085
            return -1;
3086
        }
3087
    }
3088
#endif
3089

    
3090
    printf("Testing RBSP\n");
3091

    
3092

    
3093
    return 0;
3094
}
3095
#endif /* TEST */
3096

    
3097

    
3098
av_cold void ff_h264_free_context(H264Context *h)
3099
{
3100
    int i;
3101

    
3102
    free_tables(h); //FIXME cleanup init stuff perhaps
3103

    
3104
    for(i = 0; i < MAX_SPS_COUNT; i++)
3105
        av_freep(h->sps_buffers + i);
3106

    
3107
    for(i = 0; i < MAX_PPS_COUNT; i++)
3108
        av_freep(h->pps_buffers + i);
3109
}
3110

    
3111
av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3112
{
3113
    H264Context *h = avctx->priv_data;
3114
    MpegEncContext *s = &h->s;
3115

    
3116
    ff_h264_free_context(h);
3117

    
3118
    MPV_common_end(s);
3119

    
3120
//    memset(h, 0, sizeof(H264Context));
3121

    
3122
    return 0;
3123
}
3124

    
3125

    
3126
AVCodec h264_decoder = {
3127
    "h264",
3128
    CODEC_TYPE_VIDEO,
3129
    CODEC_ID_H264,
3130
    sizeof(H264Context),
3131
    ff_h264_decode_init,
3132
    NULL,
3133
    ff_h264_decode_end,
3134
    decode_frame,
3135
    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3136
    .flush= flush_dpb,
3137
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3138
    .pix_fmts= ff_hwaccel_pixfmt_list_420,
3139
};
3140

    
3141
#if CONFIG_H264_VDPAU_DECODER
3142
AVCodec h264_vdpau_decoder = {
3143
    "h264_vdpau",
3144
    CODEC_TYPE_VIDEO,
3145
    CODEC_ID_H264,
3146
    sizeof(H264Context),
3147
    ff_h264_decode_init,
3148
    NULL,
3149
    ff_h264_decode_end,
3150
    decode_frame,
3151
    CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3152
    .flush= flush_dpb,
3153
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3154
    .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3155
};
3156
#endif