Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 6d7e6b26

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
#if ARCH_X86
43
#include "x86/h264_i386.h"
44
#endif
45

    
46
//#undef NDEBUG
47
#include <assert.h>
48

    
49
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
50
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
51

    
52
static const uint8_t rem6[52]={
53
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,
54
};
55

    
56
static const uint8_t div6[52]={
57
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,
58
};
59

    
60
void ff_h264_write_back_intra_pred_mode(H264Context *h){
61
    const int mb_xy= h->mb_xy;
62

    
63
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
64
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
65
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
66
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
67
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
68
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
69
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
70
}
71

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

    
81
    if(!(h->top_samples_available&0x8000)){
82
        for(i=0; i<4; i++){
83
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
84
            if(status<0){
85
                av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
86
                return -1;
87
            } else if(status){
88
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
89
            }
90
        }
91
    }
92

    
93
    if((h->left_samples_available&0x8888)!=0x8888){
94
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
95
        for(i=0; i<4; i++){
96
            if(!(h->left_samples_available&mask[i])){
97
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
98
                if(status<0){
99
                    av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
100
                    return -1;
101
                } else if(status){
102
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
103
                }
104
            }
105
        }
106
    }
107

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

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

    
119
    if(mode > 6U) {
120
        av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
121
        return -1;
122
    }
123

    
124
    if(!(h->top_samples_available&0x8000)){
125
        mode= top[ mode ];
126
        if(mode<0){
127
            av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
128
            return -1;
129
        }
130
    }
131

    
132
    if((h->left_samples_available&0x8080) != 0x8080){
133
        mode= left[ mode ];
134
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
135
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
136
        }
137
        if(mode<0){
138
            av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
139
            return -1;
140
        }
141
    }
142

    
143
    return mode;
144
}
145

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

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

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

    
161
#if HAVE_FAST_UNALIGNED
162
# if HAVE_FAST_64BIT
163
#   define RS 7
164
    for(i=0; i+1<length; i+=9){
165
        if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
166
# else
167
#   define RS 3
168
    for(i=0; i+1<length; i+=5){
169
        if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
170
# endif
171
            continue;
172
        if(i>0 && !src[i]) i--;
173
        while(src[i]) i++;
174
#else
175
#   define RS 0
176
    for(i=0; i+1<length; i+=2){
177
        if(src[i]) continue;
178
        if(i>0 && src[i-1]==0) i--;
179
#endif
180
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
181
            if(src[i+2]!=3){
182
                /* startcode, so we must be past the end */
183
                length=i;
184
            }
185
            break;
186
        }
187
        i-= RS;
188
    }
189

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
329
#undef xStride
330
#undef stride
331

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

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

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

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

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

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

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

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

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

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

    
397
    if(   full_mx < 0-extra_width
398
       || full_my < 0-extra_height
399
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
400
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
401
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
402
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
403
        emu=1;
404
    }
405

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

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

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

    
421
    if(emu){
422
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
423
            src_cb= s->edge_emu_buffer;
424
    }
425
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
426

    
427
    if(emu){
428
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
429
            src_cr= s->edge_emu_buffer;
430
    }
431
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
432
}
433

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
577
    assert(IS_INTER(mb_type));
578

    
579
    prefetch_motion(h, 0);
580

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

    
607
        assert(IS_8X8(mb_type));
608

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

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

    
653
    prefetch_motion(h, 1);
654
}
655

    
656

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

    
671
    av_freep(&h->mb2b_xy);
672
    av_freep(&h->mb2b8_xy);
673

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

    
688
static void init_dequant8_coeff_table(H264Context *h){
689
    int i,q,x;
690
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
691
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
692
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
693

    
694
    for(i=0; i<2; i++ ){
695
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
696
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
697
            break;
698
        }
699

    
700
        for(q=0; q<52; q++){
701
            int shift = div6[q];
702
            int idx = rem6[q];
703
            for(x=0; x<64; x++)
704
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
705
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
706
                    h->pps.scaling_matrix8[i][x]) << shift;
707
        }
708
    }
709
}
710

    
711
static void init_dequant4_coeff_table(H264Context *h){
712
    int i,j,q,x;
713
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
714
    for(i=0; i<6; i++ ){
715
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
716
        for(j=0; j<i; j++){
717
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
718
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
719
                break;
720
            }
721
        }
722
        if(j<i)
723
            continue;
724

    
725
        for(q=0; q<52; q++){
726
            int shift = div6[q] + 2;
727
            int idx = rem6[q];
728
            for(x=0; x<16; x++)
729
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
730
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
731
                    h->pps.scaling_matrix4[i][x]) << shift;
732
        }
733
    }
734
}
735

    
736
static void init_dequant_tables(H264Context *h){
737
    int i,x;
738
    init_dequant4_coeff_table(h);
739
    if(h->pps.transform_8x8_mode)
740
        init_dequant8_coeff_table(h);
741
    if(h->sps.transform_bypass){
742
        for(i=0; i<6; i++)
743
            for(x=0; x<16; x++)
744
                h->dequant4_coeff[i][0][x] = 1<<6;
745
        if(h->pps.transform_8x8_mode)
746
            for(i=0; i<2; i++)
747
                for(x=0; x<64; x++)
748
                    h->dequant8_coeff[i][0][x] = 1<<6;
749
    }
750
}
751

    
752

    
753
int ff_h264_alloc_tables(H264Context *h){
754
    MpegEncContext * const s = &h->s;
755
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
756
    int x,y;
757

    
758
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t), fail)
759

    
760
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
761
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
762
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
763

    
764
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
765
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t), fail);
766
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t), fail);
767
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 32*big_mb_num * sizeof(uint8_t) , fail);
768
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
769

    
770
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
771
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
772

    
773
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy  , big_mb_num * sizeof(uint32_t), fail);
774
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b8_xy , big_mb_num * sizeof(uint32_t), fail);
775
    for(y=0; y<s->mb_height; y++){
776
        for(x=0; x<s->mb_width; x++){
777
            const int mb_xy= x + y*s->mb_stride;
778
            const int b_xy = 4*x + 4*y*h->b_stride;
779
            const int b8_xy= 2*x + 2*y*h->b8_stride;
780

    
781
            h->mb2b_xy [mb_xy]= b_xy;
782
            h->mb2b8_xy[mb_xy]= b8_xy;
783
        }
784
    }
785

    
786
    s->obmc_scratchpad = NULL;
787

    
788
    if(!h->dequant4_coeff[0])
789
        init_dequant_tables(h);
790

    
791
    return 0;
792
fail:
793
    free_tables(h);
794
    return -1;
795
}
796

    
797
/**
798
 * Mimic alloc_tables(), but for every context thread.
799
 */
800
static void clone_tables(H264Context *dst, H264Context *src){
801
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
802
    dst->non_zero_count           = src->non_zero_count;
803
    dst->slice_table              = src->slice_table;
804
    dst->cbp_table                = src->cbp_table;
805
    dst->mb2b_xy                  = src->mb2b_xy;
806
    dst->mb2b8_xy                 = src->mb2b8_xy;
807
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
808
    dst->mvd_table[0]             = src->mvd_table[0];
809
    dst->mvd_table[1]             = src->mvd_table[1];
810
    dst->direct_table             = src->direct_table;
811

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

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

    
824
    return 0;
825
fail:
826
    return -1; // free_tables will clean up for us
827
}
828

    
829
static av_cold void common_init(H264Context *h){
830
    MpegEncContext * const s = &h->s;
831

    
832
    s->width = s->avctx->width;
833
    s->height = s->avctx->height;
834
    s->codec_id= s->avctx->codec->id;
835

    
836
    ff_h264_pred_init(&h->hpc, s->codec_id);
837

    
838
    h->dequant_coeff_pps= -1;
839
    s->unrestricted_mv=1;
840
    s->decode=1; //FIXME
841

    
842
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
843

    
844
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
845
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
846
}
847

    
848
av_cold int ff_h264_decode_init(AVCodecContext *avctx){
849
    H264Context *h= avctx->priv_data;
850
    MpegEncContext * const s = &h->s;
851

    
852
    MPV_decode_defaults(s);
853

    
854
    s->avctx = avctx;
855
    common_init(h);
856

    
857
    s->out_format = FMT_H264;
858
    s->workaround_bugs= avctx->workaround_bugs;
859

    
860
    // set defaults
861
//    s->decode_mb= ff_h263_decode_mb;
862
    s->quarter_sample = 1;
863
    if(!avctx->has_b_frames)
864
    s->low_delay= 1;
865

    
866
    avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
867

    
868
    ff_h264_decode_init_vlc();
869

    
870
    if(avctx->extradata_size > 0 && avctx->extradata &&
871
       *(char *)avctx->extradata == 1){
872
        h->is_avc = 1;
873
        h->got_avcC = 0;
874
    } else {
875
        h->is_avc = 0;
876
    }
877

    
878
    h->thread_context[0] = h;
879
    h->outputed_poc = INT_MIN;
880
    h->prev_poc_msb= 1<<16;
881
    ff_h264_reset_sei(h);
882
    if(avctx->codec_id == CODEC_ID_H264){
883
        if(avctx->ticks_per_frame == 1){
884
            s->avctx->time_base.den *=2;
885
        }
886
        avctx->ticks_per_frame = 2;
887
    }
888
    return 0;
889
}
890

    
891
int ff_h264_frame_start(H264Context *h){
892
    MpegEncContext * const s = &h->s;
893
    int i;
894

    
895
    if(MPV_frame_start(s, s->avctx) < 0)
896
        return -1;
897
    ff_er_frame_start(s);
898
    /*
899
     * MPV_frame_start uses pict_type to derive key_frame.
900
     * This is incorrect for H.264; IDR markings must be used.
901
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
902
     * See decode_nal_units().
903
     */
904
    s->current_picture_ptr->key_frame= 0;
905
    s->current_picture_ptr->mmco_reset= 0;
906

    
907
    assert(s->linesize && s->uvlinesize);
908

    
909
    for(i=0; i<16; i++){
910
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
911
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
912
    }
913
    for(i=0; i<4; i++){
914
        h->block_offset[16+i]=
915
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
916
        h->block_offset[24+16+i]=
917
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
918
    }
919

    
920
    /* can't be in alloc_tables because linesize isn't known there.
921
     * FIXME: redo bipred weight to not require extra buffer? */
922
    for(i = 0; i < s->avctx->thread_count; i++)
923
        if(!h->thread_context[i]->s.obmc_scratchpad)
924
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
925

    
926
    /* some macroblocks will be accessed before they're available */
927
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
928
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
929

    
930
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
931

    
932
    // We mark the current picture as non-reference after allocating it, so
933
    // that if we break out due to an error it can be released automatically
934
    // in the next MPV_frame_start().
935
    // SVQ3 as well as most other codecs have only last/next/current and thus
936
    // get released even with set reference, besides SVQ3 and others do not
937
    // mark frames as reference later "naturally".
938
    if(s->codec_id != CODEC_ID_SVQ3)
939
        s->current_picture_ptr->reference= 0;
940

    
941
    s->current_picture_ptr->field_poc[0]=
942
    s->current_picture_ptr->field_poc[1]= INT_MAX;
943
    assert(s->current_picture_ptr->long_ref==0);
944

    
945
    return 0;
946
}
947

    
948
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){
949
    MpegEncContext * const s = &h->s;
950
    int top_idx = 1;
951

    
952
    src_y  -=   linesize;
953
    src_cb -= uvlinesize;
954
    src_cr -= uvlinesize;
955

    
956
    if(!simple && FRAME_MBAFF){
957
        if(s->mb_y&1){
958
            if(!MB_MBAFF){
959
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
960
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
961
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
962
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
963
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
964
                }
965
            }
966
        }else if(MB_MBAFF){
967
            top_idx = 0;
968
        }else
969
            return;
970
    }
971

    
972
    // There are two lines saved, the line above the the top macroblock of a pair,
973
    // and the line above the bottom macroblock
974

    
975
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
976
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
977

    
978
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
979
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
980
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
981
    }
982
}
983

    
984
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){
985
    MpegEncContext * const s = &h->s;
986
    int temp8, i;
987
    uint64_t temp64;
988
    int deblock_left;
989
    int deblock_top;
990
    int mb_xy;
991
    int top_idx = 1;
992

    
993
    if(!simple && FRAME_MBAFF){
994
        if(s->mb_y&1){
995
            if(!MB_MBAFF)
996
                return;
997
        }else{
998
            top_idx = MB_MBAFF ? 0 : 1;
999
        }
1000
    }
1001

    
1002
    if(h->deblocking_filter == 2) {
1003
        mb_xy = h->mb_xy;
1004
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1005
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1006
    } else {
1007
        deblock_left = (s->mb_x > 0);
1008
        deblock_top =  (s->mb_y > !!MB_FIELD);
1009
    }
1010

    
1011
    src_y  -=   linesize + 1;
1012
    src_cb -= uvlinesize + 1;
1013
    src_cr -= uvlinesize + 1;
1014

    
1015
#define XCHG(a,b,t,xchg)\
1016
t= a;\
1017
if(xchg)\
1018
    a= b;\
1019
b= t;
1020

    
1021
    if(deblock_top){
1022
        if(deblock_left){
1023
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x-1]+8), *(uint64_t*)(src_y -7), temp64, 1);
1024
        }
1025
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
1026
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
1027
        if(s->mb_x+1 < s->mb_width){
1028
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
1029
        }
1030
    }
1031

    
1032
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1033
        if(deblock_top){
1034
            if(deblock_left){
1035
                XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x-1]+16), *(uint64_t*)(src_cb -7), temp64, 1);
1036
                XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x-1]+24), *(uint64_t*)(src_cr -7), temp64, 1);
1037
            }
1038
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
1039
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
1040
        }
1041
    }
1042
}
1043

    
1044
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1045
    MpegEncContext * const s = &h->s;
1046
    const int mb_x= s->mb_x;
1047
    const int mb_y= s->mb_y;
1048
    const int mb_xy= h->mb_xy;
1049
    const int mb_type= s->current_picture.mb_type[mb_xy];
1050
    uint8_t  *dest_y, *dest_cb, *dest_cr;
1051
    int linesize, uvlinesize /*dct_offset*/;
1052
    int i;
1053
    int *block_offset = &h->block_offset[0];
1054
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1055
    /* is_h264 should always be true if SVQ3 is disabled. */
1056
    const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1057
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1058
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1059

    
1060
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1061
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1062
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1063

    
1064
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1065
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1066

    
1067
    h->list_counts[mb_xy]= h->list_count;
1068

    
1069
    if (!simple && MB_FIELD) {
1070
        linesize   = h->mb_linesize   = s->linesize * 2;
1071
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1072
        block_offset = &h->block_offset[24];
1073
        if(mb_y&1){ //FIXME move out of this function?
1074
            dest_y -= s->linesize*15;
1075
            dest_cb-= s->uvlinesize*7;
1076
            dest_cr-= s->uvlinesize*7;
1077
        }
1078
        if(FRAME_MBAFF) {
1079
            int list;
1080
            for(list=0; list<h->list_count; list++){
1081
                if(!USES_LIST(mb_type, list))
1082
                    continue;
1083
                if(IS_16X16(mb_type)){
1084
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
1085
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1086
                }else{
1087
                    for(i=0; i<16; i+=4){
1088
                        int ref = h->ref_cache[list][scan8[i]];
1089
                        if(ref >= 0)
1090
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1091
                    }
1092
                }
1093
            }
1094
        }
1095
    } else {
1096
        linesize   = h->mb_linesize   = s->linesize;
1097
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1098
//        dct_offset = s->linesize * 16;
1099
    }
1100

    
1101
    if (!simple && IS_INTRA_PCM(mb_type)) {
1102
        for (i=0; i<16; i++) {
1103
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1104
        }
1105
        for (i=0; i<8; i++) {
1106
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1107
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1108
        }
1109
    } else {
1110
        if(IS_INTRA(mb_type)){
1111
            if(h->deblocking_filter)
1112
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1113

    
1114
            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1115
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1116
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1117
            }
1118

    
1119
            if(IS_INTRA4x4(mb_type)){
1120
                if(simple || !s->encoding){
1121
                    if(IS_8x8DCT(mb_type)){
1122
                        if(transform_bypass){
1123
                            idct_dc_add =
1124
                            idct_add    = s->dsp.add_pixels8;
1125
                        }else{
1126
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
1127
                            idct_add    = s->dsp.h264_idct8_add;
1128
                        }
1129
                        for(i=0; i<16; i+=4){
1130
                            uint8_t * const ptr= dest_y + block_offset[i];
1131
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1132
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1133
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1134
                            }else{
1135
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
1136
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1137
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
1138
                                if(nnz){
1139
                                    if(nnz == 1 && h->mb[i*16])
1140
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
1141
                                    else
1142
                                        idct_add   (ptr, h->mb + i*16, linesize);
1143
                                }
1144
                            }
1145
                        }
1146
                    }else{
1147
                        if(transform_bypass){
1148
                            idct_dc_add =
1149
                            idct_add    = s->dsp.add_pixels4;
1150
                        }else{
1151
                            idct_dc_add = s->dsp.h264_idct_dc_add;
1152
                            idct_add    = s->dsp.h264_idct_add;
1153
                        }
1154
                        for(i=0; i<16; i++){
1155
                            uint8_t * const ptr= dest_y + block_offset[i];
1156
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1157

    
1158
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1159
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1160
                            }else{
1161
                                uint8_t *topright;
1162
                                int nnz, tr;
1163
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1164
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1165
                                    assert(mb_y || linesize <= block_offset[i]);
1166
                                    if(!topright_avail){
1167
                                        tr= ptr[3 - linesize]*0x01010101;
1168
                                        topright= (uint8_t*) &tr;
1169
                                    }else
1170
                                        topright= ptr + 4 - linesize;
1171
                                }else
1172
                                    topright= NULL;
1173

    
1174
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1175
                                nnz = h->non_zero_count_cache[ scan8[i] ];
1176
                                if(nnz){
1177
                                    if(is_h264){
1178
                                        if(nnz == 1 && h->mb[i*16])
1179
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
1180
                                        else
1181
                                            idct_add   (ptr, h->mb + i*16, linesize);
1182
                                    }else
1183
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1184
                                }
1185
                            }
1186
                        }
1187
                    }
1188
                }
1189
            }else{
1190
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1191
                if(is_h264){
1192
                    if(!transform_bypass)
1193
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1194
                }else
1195
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1196
            }
1197
            if(h->deblocking_filter)
1198
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1199
        }else if(is_h264){
1200
            hl_motion(h, dest_y, dest_cb, dest_cr,
1201
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1202
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1203
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1204
        }
1205

    
1206

    
1207
        if(!IS_INTRA4x4(mb_type)){
1208
            if(is_h264){
1209
                if(IS_INTRA16x16(mb_type)){
1210
                    if(transform_bypass){
1211
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1212
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1213
                        }else{
1214
                            for(i=0; i<16; i++){
1215
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1216
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1217
                            }
1218
                        }
1219
                    }else{
1220
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1221
                    }
1222
                }else if(h->cbp&15){
1223
                    if(transform_bypass){
1224
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1225
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1226
                        for(i=0; i<16; i+=di){
1227
                            if(h->non_zero_count_cache[ scan8[i] ]){
1228
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1229
                            }
1230
                        }
1231
                    }else{
1232
                        if(IS_8x8DCT(mb_type)){
1233
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1234
                        }else{
1235
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1236
                        }
1237
                    }
1238
                }
1239
            }else{
1240
                for(i=0; i<16; i++){
1241
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1242
                        uint8_t * const ptr= dest_y + block_offset[i];
1243
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1244
                    }
1245
                }
1246
            }
1247
        }
1248

    
1249
        if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1250
            uint8_t *dest[2] = {dest_cb, dest_cr};
1251
            if(transform_bypass){
1252
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1253
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1254
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1255
                }else{
1256
                    idct_add = s->dsp.add_pixels4;
1257
                    for(i=16; i<16+8; i++){
1258
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1259
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1260
                    }
1261
                }
1262
            }else{
1263
                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]);
1264
                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]);
1265
                if(is_h264){
1266
                    idct_add = s->dsp.h264_idct_add;
1267
                    idct_dc_add = s->dsp.h264_idct_dc_add;
1268
                    for(i=16; i<16+8; i++){
1269
                        if(h->non_zero_count_cache[ scan8[i] ])
1270
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1271
                        else if(h->mb[i*16])
1272
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1273
                    }
1274
                }else{
1275
                    for(i=16; i<16+8; i++){
1276
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1277
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1278
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1279
                        }
1280
                    }
1281
                }
1282
            }
1283
        }
1284
    }
1285
    if(h->cbp || IS_INTRA(mb_type))
1286
        s->dsp.clear_blocks(h->mb);
1287

    
1288
    if(h->deblocking_filter && 0) {
1289
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1290
        fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
1291
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1292
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1293
        if (!simple && FRAME_MBAFF) {
1294
            ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1295
        } else {
1296
            ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1297
        }
1298
    }
1299
}
1300

    
1301
/**
1302
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1303
 */
1304
static void hl_decode_mb_simple(H264Context *h){
1305
    hl_decode_mb_internal(h, 1);
1306
}
1307

    
1308
/**
1309
 * Process a macroblock; this handles edge cases, such as interlacing.
1310
 */
1311
static void av_noinline hl_decode_mb_complex(H264Context *h){
1312
    hl_decode_mb_internal(h, 0);
1313
}
1314

    
1315
void ff_h264_hl_decode_mb(H264Context *h){
1316
    MpegEncContext * const s = &h->s;
1317
    const int mb_xy= h->mb_xy;
1318
    const int mb_type= s->current_picture.mb_type[mb_xy];
1319
    int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1320

    
1321
    if (is_complex)
1322
        hl_decode_mb_complex(h);
1323
    else hl_decode_mb_simple(h);
1324
}
1325

    
1326
static int pred_weight_table(H264Context *h){
1327
    MpegEncContext * const s = &h->s;
1328
    int list, i;
1329
    int luma_def, chroma_def;
1330

    
1331
    h->use_weight= 0;
1332
    h->use_weight_chroma= 0;
1333
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1334
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1335
    luma_def = 1<<h->luma_log2_weight_denom;
1336
    chroma_def = 1<<h->chroma_log2_weight_denom;
1337

    
1338
    for(list=0; list<2; list++){
1339
        h->luma_weight_flag[list]   = 0;
1340
        h->chroma_weight_flag[list] = 0;
1341
        for(i=0; i<h->ref_count[list]; i++){
1342
            int luma_weight_flag, chroma_weight_flag;
1343

    
1344
            luma_weight_flag= get_bits1(&s->gb);
1345
            if(luma_weight_flag){
1346
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
1347
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
1348
                if(   h->luma_weight[list][i] != luma_def
1349
                   || h->luma_offset[list][i] != 0) {
1350
                    h->use_weight= 1;
1351
                    h->luma_weight_flag[list]= 1;
1352
                }
1353
            }else{
1354
                h->luma_weight[list][i]= luma_def;
1355
                h->luma_offset[list][i]= 0;
1356
            }
1357

    
1358
            if(CHROMA){
1359
                chroma_weight_flag= get_bits1(&s->gb);
1360
                if(chroma_weight_flag){
1361
                    int j;
1362
                    for(j=0; j<2; j++){
1363
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1364
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1365
                        if(   h->chroma_weight[list][i][j] != chroma_def
1366
                           || h->chroma_offset[list][i][j] != 0) {
1367
                            h->use_weight_chroma= 1;
1368
                            h->chroma_weight_flag[list]= 1;
1369
                        }
1370
                    }
1371
                }else{
1372
                    int j;
1373
                    for(j=0; j<2; j++){
1374
                        h->chroma_weight[list][i][j]= chroma_def;
1375
                        h->chroma_offset[list][i][j]= 0;
1376
                    }
1377
                }
1378
            }
1379
        }
1380
        if(h->slice_type_nos != FF_B_TYPE) break;
1381
    }
1382
    h->use_weight= h->use_weight || h->use_weight_chroma;
1383
    return 0;
1384
}
1385

    
1386
static void implicit_weight_table(H264Context *h){
1387
    MpegEncContext * const s = &h->s;
1388
    int ref0, ref1, i;
1389
    int cur_poc = s->current_picture_ptr->poc;
1390

    
1391
    for (i = 0; i < 2; i++) {
1392
        h->luma_weight_flag[i]   = 0;
1393
        h->chroma_weight_flag[i] = 0;
1394
    }
1395

    
1396
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1397
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1398
        h->use_weight= 0;
1399
        h->use_weight_chroma= 0;
1400
        return;
1401
    }
1402

    
1403
    h->use_weight= 2;
1404
    h->use_weight_chroma= 2;
1405
    h->luma_log2_weight_denom= 5;
1406
    h->chroma_log2_weight_denom= 5;
1407

    
1408
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1409
        int poc0 = h->ref_list[0][ref0].poc;
1410
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1411
            int poc1 = h->ref_list[1][ref1].poc;
1412
            int td = av_clip(poc1 - poc0, -128, 127);
1413
            if(td){
1414
                int tb = av_clip(cur_poc - poc0, -128, 127);
1415
                int tx = (16384 + (FFABS(td) >> 1)) / td;
1416
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1417
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
1418
                    h->implicit_weight[ref0][ref1] = 32;
1419
                else
1420
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1421
            }else
1422
                h->implicit_weight[ref0][ref1] = 32;
1423
        }
1424
    }
1425
}
1426

    
1427
/**
1428
 * instantaneous decoder refresh.
1429
 */
1430
static void idr(H264Context *h){
1431
    ff_h264_remove_all_refs(h);
1432
    h->prev_frame_num= 0;
1433
    h->prev_frame_num_offset= 0;
1434
    h->prev_poc_msb=
1435
    h->prev_poc_lsb= 0;
1436
}
1437

    
1438
/* forget old pics after a seek */
1439
static void flush_dpb(AVCodecContext *avctx){
1440
    H264Context *h= avctx->priv_data;
1441
    int i;
1442
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1443
        if(h->delayed_pic[i])
1444
            h->delayed_pic[i]->reference= 0;
1445
        h->delayed_pic[i]= NULL;
1446
    }
1447
    h->outputed_poc= INT_MIN;
1448
    h->prev_interlaced_frame = 1;
1449
    idr(h);
1450
    if(h->s.current_picture_ptr)
1451
        h->s.current_picture_ptr->reference= 0;
1452
    h->s.first_field= 0;
1453
    ff_h264_reset_sei(h);
1454
    ff_mpeg_flush(avctx);
1455
}
1456

    
1457
static int init_poc(H264Context *h){
1458
    MpegEncContext * const s = &h->s;
1459
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1460
    int field_poc[2];
1461
    Picture *cur = s->current_picture_ptr;
1462

    
1463
    h->frame_num_offset= h->prev_frame_num_offset;
1464
    if(h->frame_num < h->prev_frame_num)
1465
        h->frame_num_offset += max_frame_num;
1466

    
1467
    if(h->sps.poc_type==0){
1468
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1469

    
1470
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1471
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1472
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1473
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1474
        else
1475
            h->poc_msb = h->prev_poc_msb;
1476
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1477
        field_poc[0] =
1478
        field_poc[1] = h->poc_msb + h->poc_lsb;
1479
        if(s->picture_structure == PICT_FRAME)
1480
            field_poc[1] += h->delta_poc_bottom;
1481
    }else if(h->sps.poc_type==1){
1482
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1483
        int i;
1484

    
1485
        if(h->sps.poc_cycle_length != 0)
1486
            abs_frame_num = h->frame_num_offset + h->frame_num;
1487
        else
1488
            abs_frame_num = 0;
1489

    
1490
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
1491
            abs_frame_num--;
1492

    
1493
        expected_delta_per_poc_cycle = 0;
1494
        for(i=0; i < h->sps.poc_cycle_length; i++)
1495
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1496

    
1497
        if(abs_frame_num > 0){
1498
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1499
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1500

    
1501
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1502
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
1503
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1504
        } else
1505
            expectedpoc = 0;
1506

    
1507
        if(h->nal_ref_idc == 0)
1508
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1509

    
1510
        field_poc[0] = expectedpoc + h->delta_poc[0];
1511
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1512

    
1513
        if(s->picture_structure == PICT_FRAME)
1514
            field_poc[1] += h->delta_poc[1];
1515
    }else{
1516
        int poc= 2*(h->frame_num_offset + h->frame_num);
1517

    
1518
        if(!h->nal_ref_idc)
1519
            poc--;
1520

    
1521
        field_poc[0]= poc;
1522
        field_poc[1]= poc;
1523
    }
1524

    
1525
    if(s->picture_structure != PICT_BOTTOM_FIELD)
1526
        s->current_picture_ptr->field_poc[0]= field_poc[0];
1527
    if(s->picture_structure != PICT_TOP_FIELD)
1528
        s->current_picture_ptr->field_poc[1]= field_poc[1];
1529
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1530

    
1531
    return 0;
1532
}
1533

    
1534

    
1535
/**
1536
 * initialize scan tables
1537
 */
1538
static void init_scan_tables(H264Context *h){
1539
    MpegEncContext * const s = &h->s;
1540
    int i;
1541
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1542
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1543
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1544
    }else{
1545
        for(i=0; i<16; i++){
1546
#define T(x) (x>>2) | ((x<<2) & 0xF)
1547
            h->zigzag_scan[i] = T(zigzag_scan[i]);
1548
            h-> field_scan[i] = T( field_scan[i]);
1549
#undef T
1550
        }
1551
    }
1552
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1553
        memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1554
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1555
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1556
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1557
    }else{
1558
        for(i=0; i<64; i++){
1559
#define T(x) (x>>3) | ((x&7)<<3)
1560
            h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1561
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1562
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
1563
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1564
#undef T
1565
        }
1566
    }
1567
    if(h->sps.transform_bypass){ //FIXME same ugly
1568
        h->zigzag_scan_q0          = zigzag_scan;
1569
        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1570
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1571
        h->field_scan_q0           = field_scan;
1572
        h->field_scan8x8_q0        = field_scan8x8;
1573
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1574
    }else{
1575
        h->zigzag_scan_q0          = h->zigzag_scan;
1576
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1577
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1578
        h->field_scan_q0           = h->field_scan;
1579
        h->field_scan8x8_q0        = h->field_scan8x8;
1580
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1581
    }
1582
}
1583

    
1584
static void field_end(H264Context *h){
1585
    MpegEncContext * const s = &h->s;
1586
    AVCodecContext * const avctx= s->avctx;
1587
    s->mb_y= 0;
1588

    
1589
    s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1590
    s->current_picture_ptr->pict_type= s->pict_type;
1591

    
1592
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1593
        ff_vdpau_h264_set_reference_frames(s);
1594

    
1595
    if(!s->dropable) {
1596
        ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1597
        h->prev_poc_msb= h->poc_msb;
1598
        h->prev_poc_lsb= h->poc_lsb;
1599
    }
1600
    h->prev_frame_num_offset= h->frame_num_offset;
1601
    h->prev_frame_num= h->frame_num;
1602

    
1603
    if (avctx->hwaccel) {
1604
        if (avctx->hwaccel->end_frame(avctx) < 0)
1605
            av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1606
    }
1607

    
1608
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1609
        ff_vdpau_h264_picture_complete(s);
1610

    
1611
    /*
1612
     * FIXME: Error handling code does not seem to support interlaced
1613
     * when slices span multiple rows
1614
     * The ff_er_add_slice calls don't work right for bottom
1615
     * fields; they cause massive erroneous error concealing
1616
     * Error marking covers both fields (top and bottom).
1617
     * This causes a mismatched s->error_count
1618
     * and a bad error table. Further, the error count goes to
1619
     * INT_MAX when called for bottom field, because mb_y is
1620
     * past end by one (callers fault) and resync_mb_y != 0
1621
     * causes problems for the first MB line, too.
1622
     */
1623
    if (!FIELD_PICTURE)
1624
        ff_er_frame_end(s);
1625

    
1626
    MPV_frame_end(s);
1627

    
1628
    h->current_slice=0;
1629
}
1630

    
1631
/**
1632
 * Replicates H264 "master" context to thread contexts.
1633
 */
1634
static void clone_slice(H264Context *dst, H264Context *src)
1635
{
1636
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1637
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1638
    dst->s.current_picture      = src->s.current_picture;
1639
    dst->s.linesize             = src->s.linesize;
1640
    dst->s.uvlinesize           = src->s.uvlinesize;
1641
    dst->s.first_field          = src->s.first_field;
1642

    
1643
    dst->prev_poc_msb           = src->prev_poc_msb;
1644
    dst->prev_poc_lsb           = src->prev_poc_lsb;
1645
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1646
    dst->prev_frame_num         = src->prev_frame_num;
1647
    dst->short_ref_count        = src->short_ref_count;
1648

    
1649
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1650
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1651
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1652
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1653

    
1654
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1655
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1656
}
1657

    
1658
/**
1659
 * decodes a slice header.
1660
 * This will also call MPV_common_init() and frame_start() as needed.
1661
 *
1662
 * @param h h264context
1663
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1664
 *
1665
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1666
 */
1667
static int decode_slice_header(H264Context *h, H264Context *h0){
1668
    MpegEncContext * const s = &h->s;
1669
    MpegEncContext * const s0 = &h0->s;
1670
    unsigned int first_mb_in_slice;
1671
    unsigned int pps_id;
1672
    int num_ref_idx_active_override_flag;
1673
    unsigned int slice_type, tmp, i, j;
1674
    int default_ref_list_done = 0;
1675
    int last_pic_structure;
1676

    
1677
    s->dropable= h->nal_ref_idc == 0;
1678

    
1679
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1680
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1681
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1682
    }else{
1683
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1684
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1685
    }
1686

    
1687
    first_mb_in_slice= get_ue_golomb(&s->gb);
1688

    
1689
    if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1690
        if(h0->current_slice && FIELD_PICTURE){
1691
            field_end(h);
1692
        }
1693

    
1694
        h0->current_slice = 0;
1695
        if (!s0->first_field)
1696
            s->current_picture_ptr= NULL;
1697
    }
1698

    
1699
    slice_type= get_ue_golomb_31(&s->gb);
1700
    if(slice_type > 9){
1701
        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);
1702
        return -1;
1703
    }
1704
    if(slice_type > 4){
1705
        slice_type -= 5;
1706
        h->slice_type_fixed=1;
1707
    }else
1708
        h->slice_type_fixed=0;
1709

    
1710
    slice_type= golomb_to_pict_type[ slice_type ];
1711
    if (slice_type == FF_I_TYPE
1712
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1713
        default_ref_list_done = 1;
1714
    }
1715
    h->slice_type= slice_type;
1716
    h->slice_type_nos= slice_type & 3;
1717

    
1718
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1719
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1720
        av_log(h->s.avctx, AV_LOG_ERROR,
1721
               "B picture before any references, skipping\n");
1722
        return -1;
1723
    }
1724

    
1725
    pps_id= get_ue_golomb(&s->gb);
1726
    if(pps_id>=MAX_PPS_COUNT){
1727
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1728
        return -1;
1729
    }
1730
    if(!h0->pps_buffers[pps_id]) {
1731
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1732
        return -1;
1733
    }
1734
    h->pps= *h0->pps_buffers[pps_id];
1735

    
1736
    if(!h0->sps_buffers[h->pps.sps_id]) {
1737
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1738
        return -1;
1739
    }
1740
    h->sps = *h0->sps_buffers[h->pps.sps_id];
1741

    
1742
    if(h == h0 && h->dequant_coeff_pps != pps_id){
1743
        h->dequant_coeff_pps = pps_id;
1744
        init_dequant_tables(h);
1745
    }
1746

    
1747
    s->mb_width= h->sps.mb_width;
1748
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1749

    
1750
    h->b_stride=  s->mb_width*4;
1751
    h->b8_stride= s->mb_width*2;
1752

    
1753
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1754
    if(h->sps.frame_mbs_only_flag)
1755
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1756
    else
1757
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1758

    
1759
    if (s->context_initialized
1760
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1761
        if(h != h0)
1762
            return -1;   // width / height changed during parallelized decoding
1763
        free_tables(h);
1764
        flush_dpb(s->avctx);
1765
        MPV_common_end(s);
1766
    }
1767
    if (!s->context_initialized) {
1768
        if(h != h0)
1769
            return -1;  // we cant (re-)initialize context during parallel decoding
1770

    
1771
        avcodec_set_dimensions(s->avctx, s->width, s->height);
1772
        s->avctx->sample_aspect_ratio= h->sps.sar;
1773
        if(!s->avctx->sample_aspect_ratio.den)
1774
            s->avctx->sample_aspect_ratio.den = 1;
1775

    
1776
        if(h->sps.video_signal_type_present_flag){
1777
            s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1778
            if(h->sps.colour_description_present_flag){
1779
                s->avctx->color_primaries = h->sps.color_primaries;
1780
                s->avctx->color_trc       = h->sps.color_trc;
1781
                s->avctx->colorspace      = h->sps.colorspace;
1782
            }
1783
        }
1784

    
1785
        if(h->sps.timing_info_present_flag){
1786
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
1787
            if(h->x264_build > 0 && h->x264_build < 44)
1788
                s->avctx->time_base.den *= 2;
1789
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1790
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
1791
        }
1792
        s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1793
        s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1794

    
1795
        if (MPV_common_init(s) < 0)
1796
            return -1;
1797
        s->first_field = 0;
1798
        h->prev_interlaced_frame = 1;
1799

    
1800
        init_scan_tables(h);
1801
        ff_h264_alloc_tables(h);
1802

    
1803
        for(i = 1; i < s->avctx->thread_count; i++) {
1804
            H264Context *c;
1805
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1806
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1807
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1808
            c->sps = h->sps;
1809
            c->pps = h->pps;
1810
            init_scan_tables(c);
1811
            clone_tables(c, h);
1812
        }
1813

    
1814
        for(i = 0; i < s->avctx->thread_count; i++)
1815
            if(context_init(h->thread_context[i]) < 0)
1816
                return -1;
1817
    }
1818

    
1819
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1820

    
1821
    h->mb_mbaff = 0;
1822
    h->mb_aff_frame = 0;
1823
    last_pic_structure = s0->picture_structure;
1824
    if(h->sps.frame_mbs_only_flag){
1825
        s->picture_structure= PICT_FRAME;
1826
    }else{
1827
        if(get_bits1(&s->gb)) { //field_pic_flag
1828
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1829
        } else {
1830
            s->picture_structure= PICT_FRAME;
1831
            h->mb_aff_frame = h->sps.mb_aff;
1832
        }
1833
    }
1834
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1835

    
1836
    if(h0->current_slice == 0){
1837
        while(h->frame_num !=  h->prev_frame_num &&
1838
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1839
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1840
            if (ff_h264_frame_start(h) < 0)
1841
                return -1;
1842
            h->prev_frame_num++;
1843
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1844
            s->current_picture_ptr->frame_num= h->prev_frame_num;
1845
            ff_h264_execute_ref_pic_marking(h, NULL, 0);
1846
        }
1847

    
1848
        /* See if we have a decoded first field looking for a pair... */
1849
        if (s0->first_field) {
1850
            assert(s0->current_picture_ptr);
1851
            assert(s0->current_picture_ptr->data[0]);
1852
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1853

    
1854
            /* figure out if we have a complementary field pair */
1855
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1856
                /*
1857
                 * Previous field is unmatched. Don't display it, but let it
1858
                 * remain for reference if marked as such.
1859
                 */
1860
                s0->current_picture_ptr = NULL;
1861
                s0->first_field = FIELD_PICTURE;
1862

    
1863
            } else {
1864
                if (h->nal_ref_idc &&
1865
                        s0->current_picture_ptr->reference &&
1866
                        s0->current_picture_ptr->frame_num != h->frame_num) {
1867
                    /*
1868
                     * This and previous field were reference, but had
1869
                     * different frame_nums. Consider this field first in
1870
                     * pair. Throw away previous field except for reference
1871
                     * purposes.
1872
                     */
1873
                    s0->first_field = 1;
1874
                    s0->current_picture_ptr = NULL;
1875

    
1876
                } else {
1877
                    /* Second field in complementary pair */
1878
                    s0->first_field = 0;
1879
                }
1880
            }
1881

    
1882
        } else {
1883
            /* Frame or first field in a potentially complementary pair */
1884
            assert(!s0->current_picture_ptr);
1885
            s0->first_field = FIELD_PICTURE;
1886
        }
1887

    
1888
        if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1889
            s0->first_field = 0;
1890
            return -1;
1891
        }
1892
    }
1893
    if(h != h0)
1894
        clone_slice(h, h0);
1895

    
1896
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1897

    
1898
    assert(s->mb_num == s->mb_width * s->mb_height);
1899
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1900
       first_mb_in_slice                    >= s->mb_num){
1901
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1902
        return -1;
1903
    }
1904
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1905
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1906
    if (s->picture_structure == PICT_BOTTOM_FIELD)
1907
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
1908
    assert(s->mb_y < s->mb_height);
1909

    
1910
    if(s->picture_structure==PICT_FRAME){
1911
        h->curr_pic_num=   h->frame_num;
1912
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1913
    }else{
1914
        h->curr_pic_num= 2*h->frame_num + 1;
1915
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1916
    }
1917

    
1918
    if(h->nal_unit_type == NAL_IDR_SLICE){
1919
        get_ue_golomb(&s->gb); /* idr_pic_id */
1920
    }
1921

    
1922
    if(h->sps.poc_type==0){
1923
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1924

    
1925
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1926
            h->delta_poc_bottom= get_se_golomb(&s->gb);
1927
        }
1928
    }
1929

    
1930
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1931
        h->delta_poc[0]= get_se_golomb(&s->gb);
1932

    
1933
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1934
            h->delta_poc[1]= get_se_golomb(&s->gb);
1935
    }
1936

    
1937
    init_poc(h);
1938

    
1939
    if(h->pps.redundant_pic_cnt_present){
1940
        h->redundant_pic_count= get_ue_golomb(&s->gb);
1941
    }
1942

    
1943
    //set defaults, might be overridden a few lines later
1944
    h->ref_count[0]= h->pps.ref_count[0];
1945
    h->ref_count[1]= h->pps.ref_count[1];
1946

    
1947
    if(h->slice_type_nos != FF_I_TYPE){
1948
        if(h->slice_type_nos == FF_B_TYPE){
1949
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
1950
        }
1951
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
1952

    
1953
        if(num_ref_idx_active_override_flag){
1954
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1955
            if(h->slice_type_nos==FF_B_TYPE)
1956
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1957

    
1958
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1959
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1960
                h->ref_count[0]= h->ref_count[1]= 1;
1961
                return -1;
1962
            }
1963
        }
1964
        if(h->slice_type_nos == FF_B_TYPE)
1965
            h->list_count= 2;
1966
        else
1967
            h->list_count= 1;
1968
    }else
1969
        h->list_count= 0;
1970

    
1971
    if(!default_ref_list_done){
1972
        ff_h264_fill_default_ref_list(h);
1973
    }
1974

    
1975
    if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1976
        return -1;
1977

    
1978
    if(h->slice_type_nos!=FF_I_TYPE){
1979
        s->last_picture_ptr= &h->ref_list[0][0];
1980
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1981
    }
1982
    if(h->slice_type_nos==FF_B_TYPE){
1983
        s->next_picture_ptr= &h->ref_list[1][0];
1984
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1985
    }
1986

    
1987
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
1988
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1989
        pred_weight_table(h);
1990
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1991
        implicit_weight_table(h);
1992
    else {
1993
        h->use_weight = 0;
1994
        for (i = 0; i < 2; i++) {
1995
            h->luma_weight_flag[i]   = 0;
1996
            h->chroma_weight_flag[i] = 0;
1997
        }
1998
    }
1999

    
2000
    if(h->nal_ref_idc)
2001
        ff_h264_decode_ref_pic_marking(h0, &s->gb);
2002

    
2003
    if(FRAME_MBAFF)
2004
        ff_h264_fill_mbaff_ref_list(h);
2005

    
2006
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2007
        ff_h264_direct_dist_scale_factor(h);
2008
    ff_h264_direct_ref_list_init(h);
2009

    
2010
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2011
        tmp = get_ue_golomb_31(&s->gb);
2012
        if(tmp > 2){
2013
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2014
            return -1;
2015
        }
2016
        h->cabac_init_idc= tmp;
2017
    }
2018

    
2019
    h->last_qscale_diff = 0;
2020
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2021
    if(tmp>51){
2022
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2023
        return -1;
2024
    }
2025
    s->qscale= tmp;
2026
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2027
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2028
    //FIXME qscale / qp ... stuff
2029
    if(h->slice_type == FF_SP_TYPE){
2030
        get_bits1(&s->gb); /* sp_for_switch_flag */
2031
    }
2032
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2033
        get_se_golomb(&s->gb); /* slice_qs_delta */
2034
    }
2035

    
2036
    h->deblocking_filter = 1;
2037
    h->slice_alpha_c0_offset = 0;
2038
    h->slice_beta_offset = 0;
2039
    if( h->pps.deblocking_filter_parameters_present ) {
2040
        tmp= get_ue_golomb_31(&s->gb);
2041
        if(tmp > 2){
2042
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2043
            return -1;
2044
        }
2045
        h->deblocking_filter= tmp;
2046
        if(h->deblocking_filter < 2)
2047
            h->deblocking_filter^= 1; // 1<->0
2048

    
2049
        if( h->deblocking_filter ) {
2050
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
2051
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
2052
        }
2053
    }
2054

    
2055
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2056
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2057
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2058
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2059
        h->deblocking_filter= 0;
2060

    
2061
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2062
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2063
            /* Cheat slightly for speed:
2064
               Do not bother to deblock across slices. */
2065
            h->deblocking_filter = 2;
2066
        } else {
2067
            h0->max_contexts = 1;
2068
            if(!h0->single_decode_warning) {
2069
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2070
                h0->single_decode_warning = 1;
2071
            }
2072
            if(h != h0)
2073
                return 1; // deblocking switched inside frame
2074
        }
2075
    }
2076
    h->qp_thresh= 15 - 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]);
2077

    
2078
#if 0 //FMO
2079
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2080
        slice_group_change_cycle= get_bits(&s->gb, ?);
2081
#endif
2082

    
2083
    h0->last_slice_type = slice_type;
2084
    h->slice_num = ++h0->current_slice;
2085
    if(h->slice_num >= MAX_SLICES){
2086
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2087
    }
2088

    
2089
    for(j=0; j<2; j++){
2090
        int id_list[16];
2091
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2092
        for(i=0; i<16; i++){
2093
            id_list[i]= 60;
2094
            if(h->ref_list[j][i].data[0]){
2095
                int k;
2096
                uint8_t *base= h->ref_list[j][i].base[0];
2097
                for(k=0; k<h->short_ref_count; k++)
2098
                    if(h->short_ref[k]->base[0] == base){
2099
                        id_list[i]= k;
2100
                        break;
2101
                    }
2102
                for(k=0; k<h->long_ref_count; k++)
2103
                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2104
                        id_list[i]= h->short_ref_count + k;
2105
                        break;
2106
                    }
2107
            }
2108
        }
2109

    
2110
        ref2frm[0]=
2111
        ref2frm[1]= -1;
2112
        for(i=0; i<16; i++)
2113
            ref2frm[i+2]= 4*id_list[i]
2114
                          +(h->ref_list[j][i].reference&3);
2115
        ref2frm[18+0]=
2116
        ref2frm[18+1]= -1;
2117
        for(i=16; i<48; i++)
2118
            ref2frm[i+4]= 4*id_list[(i-16)>>1]
2119
                          +(h->ref_list[j][i].reference&3);
2120
    }
2121

    
2122
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2123
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2124

    
2125
    s->avctx->refs= h->sps.ref_frame_count;
2126

    
2127
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2128
        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",
2129
               h->slice_num,
2130
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2131
               first_mb_in_slice,
2132
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2133
               pps_id, h->frame_num,
2134
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2135
               h->ref_count[0], h->ref_count[1],
2136
               s->qscale,
2137
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
2138
               h->use_weight,
2139
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2140
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2141
               );
2142
    }
2143

    
2144
    return 0;
2145
}
2146

    
2147
int ff_h264_get_slice_type(H264Context *h)
2148
{
2149
    switch (h->slice_type) {
2150
    case FF_P_TYPE:  return 0;
2151
    case FF_B_TYPE:  return 1;
2152
    case FF_I_TYPE:  return 2;
2153
    case FF_SP_TYPE: return 3;
2154
    case FF_SI_TYPE: return 4;
2155
    default:         return -1;
2156
    }
2157
}
2158

    
2159
static void loop_filter(H264Context *h){
2160
    MpegEncContext * const s = &h->s;
2161
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2162
    int linesize, uvlinesize, mb_x, mb_y;
2163
    const int end_mb_y= s->mb_y + FRAME_MBAFF;
2164
    const int old_slice_type= h->slice_type;
2165

    
2166
    if(h->deblocking_filter) {
2167
        for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2168
            for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2169
                int list, mb_xy, mb_type, is_complex;
2170
                mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2171
                h->slice_num= h->slice_table[mb_xy];
2172
                mb_type= s->current_picture.mb_type[mb_xy];
2173
                h->list_count= h->list_counts[mb_xy];
2174
                if(h->list_count==2){
2175
                    h->slice_type= h->slice_type_nos= FF_B_TYPE;
2176
                }else if(h->list_count==1){
2177
                    h->slice_type= h->slice_type_nos= FF_P_TYPE;
2178
                }else
2179
                    h->slice_type= h->slice_type_nos= FF_I_TYPE;
2180

    
2181
                if(FRAME_MBAFF)
2182
                    h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2183

    
2184
                is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0; //FIXME qscale might be wrong
2185

    
2186
                s->mb_x= mb_x;
2187
                s->mb_y= mb_y;
2188
                dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2189
                dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2190
                dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2191
                    //FIXME simplify above
2192

    
2193
                if (MB_FIELD) {
2194
                    linesize   = h->mb_linesize   = s->linesize * 2;
2195
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2196
                    if(mb_y&1){ //FIXME move out of this function?
2197
                        dest_y -= s->linesize*15;
2198
                        dest_cb-= s->uvlinesize*7;
2199
                        dest_cr-= s->uvlinesize*7;
2200
                    }
2201
                } else {
2202
                    linesize   = h->mb_linesize   = s->linesize;
2203
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2204
                }
2205
                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, !is_complex);
2206
                if(fill_filter_caches(h, mb_type) < 0)
2207
                    continue;
2208
                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2209
                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2210

    
2211
                if (is_complex && FRAME_MBAFF) {
2212
                    ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2213
                } else {
2214
                    ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2215
                }
2216
            }
2217
        }
2218
    }
2219
    h->slice_type= old_slice_type;
2220
    s->mb_x= 0;
2221
    s->mb_y= end_mb_y - FRAME_MBAFF;
2222
}
2223

    
2224
static int decode_slice(struct AVCodecContext *avctx, void *arg){
2225
    H264Context *h = *(void**)arg;
2226
    MpegEncContext * const s = &h->s;
2227
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2228

    
2229
    s->mb_skip_run= -1;
2230

    
2231
    h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2232
                    (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2233

    
2234
    if( h->pps.cabac ) {
2235
        /* realign */
2236
        align_get_bits( &s->gb );
2237

    
2238
        /* init cabac */
2239
        ff_init_cabac_states( &h->cabac);
2240
        ff_init_cabac_decoder( &h->cabac,
2241
                               s->gb.buffer + get_bits_count(&s->gb)/8,
2242
                               (get_bits_left(&s->gb) + 7)/8);
2243

    
2244
        ff_h264_init_cabac_states(h);
2245

    
2246
        for(;;){
2247
//START_TIMER
2248
            int ret = ff_h264_decode_mb_cabac(h);
2249
            int eos;
2250
//STOP_TIMER("decode_mb_cabac")
2251

    
2252
            if(ret>=0) ff_h264_hl_decode_mb(h);
2253

    
2254
            if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2255
                s->mb_y++;
2256

    
2257
                ret = ff_h264_decode_mb_cabac(h);
2258

    
2259
                if(ret>=0) ff_h264_hl_decode_mb(h);
2260
                s->mb_y--;
2261
            }
2262
            eos = get_cabac_terminate( &h->cabac );
2263

    
2264
            if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2265
                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);
2266
                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);
2267
                return -1;
2268
            }
2269

    
2270
            if( ++s->mb_x >= s->mb_width ) {
2271
                s->mb_x = 0;
2272
                loop_filter(h);
2273
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2274
                ++s->mb_y;
2275
                if(FIELD_OR_MBAFF_PICTURE) {
2276
                    ++s->mb_y;
2277
                }
2278
            }
2279

    
2280
            if( eos || s->mb_y >= s->mb_height ) {
2281
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2282
                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);
2283
                return 0;
2284
            }
2285
        }
2286

    
2287
    } else {
2288
        for(;;){
2289
            int ret = ff_h264_decode_mb_cavlc(h);
2290

    
2291
            if(ret>=0) ff_h264_hl_decode_mb(h);
2292

    
2293
            if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2294
                s->mb_y++;
2295
                ret = ff_h264_decode_mb_cavlc(h);
2296

    
2297
                if(ret>=0) ff_h264_hl_decode_mb(h);
2298
                s->mb_y--;
2299
            }
2300

    
2301
            if(ret<0){
2302
                av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
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

    
2305
                return -1;
2306
            }
2307

    
2308
            if(++s->mb_x >= s->mb_width){
2309
                s->mb_x=0;
2310
                loop_filter(h);
2311
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2312
                ++s->mb_y;
2313
                if(FIELD_OR_MBAFF_PICTURE) {
2314
                    ++s->mb_y;
2315
                }
2316
                if(s->mb_y >= s->mb_height){
2317
                    tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2318

    
2319
                    if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2320
                        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);
2321

    
2322
                        return 0;
2323
                    }else{
2324
                        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);
2325

    
2326
                        return -1;
2327
                    }
2328
                }
2329
            }
2330

    
2331
            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2332
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2333
                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2334
                    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);
2335

    
2336
                    return 0;
2337
                }else{
2338
                    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);
2339

    
2340
                    return -1;
2341
                }
2342
            }
2343
        }
2344
    }
2345

    
2346
#if 0
2347
    for(;s->mb_y < s->mb_height; s->mb_y++){
2348
        for(;s->mb_x < s->mb_width; s->mb_x++){
2349
            int ret= decode_mb(h);
2350

2351
            ff_h264_hl_decode_mb(h);
2352

2353
            if(ret<0){
2354
                av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2355
                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);
2356

2357
                return -1;
2358
            }
2359

2360
            if(++s->mb_x >= s->mb_width){
2361
                s->mb_x=0;
2362
                if(++s->mb_y >= s->mb_height){
2363
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
2364
                        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);
2365

2366
                        return 0;
2367
                    }else{
2368
                        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);
2369

2370
                        return -1;
2371
                    }
2372
                }
2373
            }
2374

2375
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2376
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
2377
                    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);
2378

2379
                    return 0;
2380
                }else{
2381
                    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);
2382

2383
                    return -1;
2384
                }
2385
            }
2386
        }
2387
        s->mb_x=0;
2388
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
2389
    }
2390
#endif
2391
    return -1; //not reached
2392
}
2393

    
2394
/**
2395
 * Call decode_slice() for each context.
2396
 *
2397
 * @param h h264 master context
2398
 * @param context_count number of contexts to execute
2399
 */
2400
static void execute_decode_slices(H264Context *h, int context_count){
2401
    MpegEncContext * const s = &h->s;
2402
    AVCodecContext * const avctx= s->avctx;
2403
    H264Context *hx;
2404
    int i;
2405

    
2406
    if (s->avctx->hwaccel)
2407
        return;
2408
    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2409
        return;
2410
    if(context_count == 1) {
2411
        decode_slice(avctx, &h);
2412
    } else {
2413
        for(i = 1; i < context_count; i++) {
2414
            hx = h->thread_context[i];
2415
            hx->s.error_recognition = avctx->error_recognition;
2416
            hx->s.error_count = 0;
2417
        }
2418

    
2419
        avctx->execute(avctx, (void *)decode_slice,
2420
                       h->thread_context, NULL, context_count, sizeof(void*));
2421

    
2422
        /* pull back stuff from slices to master context */
2423
        hx = h->thread_context[context_count - 1];
2424
        s->mb_x = hx->s.mb_x;
2425
        s->mb_y = hx->s.mb_y;
2426
        s->dropable = hx->s.dropable;
2427
        s->picture_structure = hx->s.picture_structure;
2428
        for(i = 1; i < context_count; i++)
2429
            h->s.error_count += h->thread_context[i]->s.error_count;
2430
    }
2431
}
2432

    
2433

    
2434
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2435
    MpegEncContext * const s = &h->s;
2436
    AVCodecContext * const avctx= s->avctx;
2437
    int buf_index=0;
2438
    H264Context *hx; ///< thread context
2439
    int context_count = 0;
2440
    int next_avc= h->is_avc ? 0 : buf_size;
2441

    
2442
    h->max_contexts = avctx->thread_count;
2443
#if 0
2444
    int i;
2445
    for(i=0; i<50; i++){
2446
        av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2447
    }
2448
#endif
2449
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2450
        h->current_slice = 0;
2451
        if (!s->first_field)
2452
            s->current_picture_ptr= NULL;
2453
        ff_h264_reset_sei(h);
2454
    }
2455

    
2456
    for(;;){
2457
        int consumed;
2458
        int dst_length;
2459
        int bit_length;
2460
        const uint8_t *ptr;
2461
        int i, nalsize = 0;
2462
        int err;
2463

    
2464
        if(buf_index >= next_avc) {
2465
            if(buf_index >= buf_size) break;
2466
            nalsize = 0;
2467
            for(i = 0; i < h->nal_length_size; i++)
2468
                nalsize = (nalsize << 8) | buf[buf_index++];
2469
            if(nalsize <= 1 || nalsize > buf_size - buf_index){
2470
                if(nalsize == 1){
2471
                    buf_index++;
2472
                    continue;
2473
                }else{
2474
                    av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2475
                    break;
2476
                }
2477
            }
2478
            next_avc= buf_index + nalsize;
2479
        } else {
2480
            // start code prefix search
2481
            for(; buf_index + 3 < next_avc; buf_index++){
2482
                // This should always succeed in the first iteration.
2483
                if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2484
                    break;
2485
            }
2486

    
2487
            if(buf_index+3 >= buf_size) break;
2488

    
2489
            buf_index+=3;
2490
            if(buf_index >= next_avc) continue;
2491
        }
2492

    
2493
        hx = h->thread_context[context_count];
2494

    
2495
        ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2496
        if (ptr==NULL || dst_length < 0){
2497
            return -1;
2498
        }
2499
        while(ptr[dst_length - 1] == 0 && dst_length > 0)
2500
            dst_length--;
2501
        bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2502

    
2503
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
2504
            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);
2505
        }
2506

    
2507
        if (h->is_avc && (nalsize != consumed) && nalsize){
2508
            av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2509
        }
2510

    
2511
        buf_index += consumed;
2512

    
2513
        if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2514
           ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2515
            continue;
2516

    
2517
      again:
2518
        err = 0;
2519
        switch(hx->nal_unit_type){
2520
        case NAL_IDR_SLICE:
2521
            if (h->nal_unit_type != NAL_IDR_SLICE) {
2522
                av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2523
                return -1;
2524
            }
2525
            idr(h); //FIXME ensure we don't loose some frames if there is reordering
2526
        case NAL_SLICE:
2527
            init_get_bits(&hx->s.gb, ptr, bit_length);
2528
            hx->intra_gb_ptr=
2529
            hx->inter_gb_ptr= &hx->s.gb;
2530
            hx->s.data_partitioning = 0;
2531

    
2532
            if((err = decode_slice_header(hx, h)))
2533
               break;
2534

    
2535
            if (s->avctx->hwaccel && h->current_slice == 1) {
2536
                if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2537
                    return -1;
2538
            }
2539

    
2540
            s->current_picture_ptr->key_frame |=
2541
                    (hx->nal_unit_type == NAL_IDR_SLICE) ||
2542
                    (h->sei_recovery_frame_cnt >= 0);
2543
            if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2544
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2545
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2546
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2547
               && avctx->skip_frame < AVDISCARD_ALL){
2548
                if(avctx->hwaccel) {
2549
                    if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2550
                        return -1;
2551
                }else
2552
                if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2553
                    static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2554
                    ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2555
                    ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2556
                }else
2557
                    context_count++;
2558
            }
2559
            break;
2560
        case NAL_DPA:
2561
            init_get_bits(&hx->s.gb, ptr, bit_length);
2562
            hx->intra_gb_ptr=
2563
            hx->inter_gb_ptr= NULL;
2564

    
2565
            if ((err = decode_slice_header(hx, h)) < 0)
2566
                break;
2567

    
2568
            hx->s.data_partitioning = 1;
2569

    
2570
            break;
2571
        case NAL_DPB:
2572
            init_get_bits(&hx->intra_gb, ptr, bit_length);
2573
            hx->intra_gb_ptr= &hx->intra_gb;
2574
            break;
2575
        case NAL_DPC:
2576
            init_get_bits(&hx->inter_gb, ptr, bit_length);
2577
            hx->inter_gb_ptr= &hx->inter_gb;
2578

    
2579
            if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2580
               && s->context_initialized
2581
               && s->hurry_up < 5
2582
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2583
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2584
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2585
               && avctx->skip_frame < AVDISCARD_ALL)
2586
                context_count++;
2587
            break;
2588
        case NAL_SEI:
2589
            init_get_bits(&s->gb, ptr, bit_length);
2590
            ff_h264_decode_sei(h);
2591
            break;
2592
        case NAL_SPS:
2593
            init_get_bits(&s->gb, ptr, bit_length);
2594
            ff_h264_decode_seq_parameter_set(h);
2595

    
2596
            if(s->flags& CODEC_FLAG_LOW_DELAY)
2597
                s->low_delay=1;
2598

    
2599
            if(avctx->has_b_frames < 2)
2600
                avctx->has_b_frames= !s->low_delay;
2601
            break;
2602
        case NAL_PPS:
2603
            init_get_bits(&s->gb, ptr, bit_length);
2604

    
2605
            ff_h264_decode_picture_parameter_set(h, bit_length);
2606

    
2607
            break;
2608
        case NAL_AUD:
2609
        case NAL_END_SEQUENCE:
2610
        case NAL_END_STREAM:
2611
        case NAL_FILLER_DATA:
2612
        case NAL_SPS_EXT:
2613
        case NAL_AUXILIARY_SLICE:
2614
            break;
2615
        default:
2616
            av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2617
        }
2618

    
2619
        if(context_count == h->max_contexts) {
2620
            execute_decode_slices(h, context_count);
2621
            context_count = 0;
2622
        }
2623

    
2624
        if (err < 0)
2625
            av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2626
        else if(err == 1) {
2627
            /* Slice could not be decoded in parallel mode, copy down
2628
             * NAL unit stuff to context 0 and restart. Note that
2629
             * rbsp_buffer is not transferred, but since we no longer
2630
             * run in parallel mode this should not be an issue. */
2631
            h->nal_unit_type = hx->nal_unit_type;
2632
            h->nal_ref_idc   = hx->nal_ref_idc;
2633
            hx = h;
2634
            goto again;
2635
        }
2636
    }
2637
    if(context_count)
2638
        execute_decode_slices(h, context_count);
2639
    return buf_index;
2640
}
2641

    
2642
/**
2643
 * returns the number of bytes consumed for building the current frame
2644
 */
2645
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2646
        if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2647
        if(pos+10>buf_size) pos=buf_size; // oops ;)
2648

    
2649
        return pos;
2650
}
2651

    
2652
static int decode_frame(AVCodecContext *avctx,
2653
                             void *data, int *data_size,
2654
                             AVPacket *avpkt)
2655
{
2656
    const uint8_t *buf = avpkt->data;
2657
    int buf_size = avpkt->size;
2658
    H264Context *h = avctx->priv_data;
2659
    MpegEncContext *s = &h->s;
2660
    AVFrame *pict = data;
2661
    int buf_index;
2662

    
2663
    s->flags= avctx->flags;
2664
    s->flags2= avctx->flags2;
2665

    
2666
   /* end of stream, output what is still in the buffers */
2667
    if (buf_size == 0) {
2668
        Picture *out;
2669
        int i, out_idx;
2670

    
2671
//FIXME factorize this with the output code below
2672
        out = h->delayed_pic[0];
2673
        out_idx = 0;
2674
        for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2675
            if(h->delayed_pic[i]->poc < out->poc){
2676
                out = h->delayed_pic[i];
2677
                out_idx = i;
2678
            }
2679

    
2680
        for(i=out_idx; h->delayed_pic[i]; i++)
2681
            h->delayed_pic[i] = h->delayed_pic[i+1];
2682

    
2683
        if(out){
2684
            *data_size = sizeof(AVFrame);
2685
            *pict= *(AVFrame*)out;
2686
        }
2687

    
2688
        return 0;
2689
    }
2690

    
2691
    if(h->is_avc && !h->got_avcC) {
2692
        int i, cnt, nalsize;
2693
        unsigned char *p = avctx->extradata;
2694
        if(avctx->extradata_size < 7) {
2695
            av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2696
            return -1;
2697
        }
2698
        if(*p != 1) {
2699
            av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2700
            return -1;
2701
        }
2702
        /* sps and pps in the avcC always have length coded with 2 bytes,
2703
           so put a fake nal_length_size = 2 while parsing them */
2704
        h->nal_length_size = 2;
2705
        // Decode sps from avcC
2706
        cnt = *(p+5) & 0x1f; // Number of sps
2707
        p += 6;
2708
        for (i = 0; i < cnt; i++) {
2709
            nalsize = AV_RB16(p) + 2;
2710
            if(decode_nal_units(h, p, nalsize) < 0) {
2711
                av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2712
                return -1;
2713
            }
2714
            p += nalsize;
2715
        }
2716
        // Decode pps from avcC
2717
        cnt = *(p++); // Number of pps
2718
        for (i = 0; i < cnt; i++) {
2719
            nalsize = AV_RB16(p) + 2;
2720
            if(decode_nal_units(h, p, nalsize)  != nalsize) {
2721
                av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2722
                return -1;
2723
            }
2724
            p += nalsize;
2725
        }
2726
        // Now store right nal length size, that will be use to parse all other nals
2727
        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2728
        // Do not reparse avcC
2729
        h->got_avcC = 1;
2730
    }
2731

    
2732
    if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2733
        if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2734
            return -1;
2735
        h->got_avcC = 1;
2736
    }
2737

    
2738
    buf_index=decode_nal_units(h, buf, buf_size);
2739
    if(buf_index < 0)
2740
        return -1;
2741

    
2742
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2743
        if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2744
        av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2745
        return -1;
2746
    }
2747

    
2748
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2749
        Picture *out = s->current_picture_ptr;
2750
        Picture *cur = s->current_picture_ptr;
2751
        int i, pics, out_of_order, out_idx;
2752

    
2753
        field_end(h);
2754

    
2755
        if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2756
            /* Wait for second field. */
2757
            *data_size = 0;
2758

    
2759
        } else {
2760
            cur->interlaced_frame = 0;
2761
            cur->repeat_pict = 0;
2762

    
2763
            /* Signal interlacing information externally. */
2764
            /* Prioritize picture timing SEI information over used decoding process if it exists. */
2765

    
2766
            if(h->sps.pic_struct_present_flag){
2767
                switch (h->sei_pic_struct)
2768
                {
2769
                case SEI_PIC_STRUCT_FRAME:
2770
                    break;
2771
                case SEI_PIC_STRUCT_TOP_FIELD:
2772
                case SEI_PIC_STRUCT_BOTTOM_FIELD:
2773
                    cur->interlaced_frame = 1;
2774
                    break;
2775
                case SEI_PIC_STRUCT_TOP_BOTTOM:
2776
                case SEI_PIC_STRUCT_BOTTOM_TOP:
2777
                    if (FIELD_OR_MBAFF_PICTURE)
2778
                        cur->interlaced_frame = 1;
2779
                    else
2780
                        // try to flag soft telecine progressive
2781
                        cur->interlaced_frame = h->prev_interlaced_frame;
2782
                    break;
2783
                case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2784
                case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2785
                    // Signal the possibility of telecined film externally (pic_struct 5,6)
2786
                    // From these hints, let the applications decide if they apply deinterlacing.
2787
                    cur->repeat_pict = 1;
2788
                    break;
2789
                case SEI_PIC_STRUCT_FRAME_DOUBLING:
2790
                    // Force progressive here, as doubling interlaced frame is a bad idea.
2791
                    cur->repeat_pict = 2;
2792
                    break;
2793
                case SEI_PIC_STRUCT_FRAME_TRIPLING:
2794
                    cur->repeat_pict = 4;
2795
                    break;
2796
                }
2797

    
2798
                if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2799
                    cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2800
            }else{
2801
                /* Derive interlacing flag from used decoding process. */
2802
                cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2803
            }
2804
            h->prev_interlaced_frame = cur->interlaced_frame;
2805

    
2806
            if (cur->field_poc[0] != cur->field_poc[1]){
2807
                /* Derive top_field_first from field pocs. */
2808
                cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2809
            }else{
2810
                if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2811
                    /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2812
                    if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2813
                      || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2814
                        cur->top_field_first = 1;
2815
                    else
2816
                        cur->top_field_first = 0;
2817
                }else{
2818
                    /* Most likely progressive */
2819
                    cur->top_field_first = 0;
2820
                }
2821
            }
2822

    
2823
        //FIXME do something with unavailable reference frames
2824

    
2825
            /* Sort B-frames into display order */
2826

    
2827
            if(h->sps.bitstream_restriction_flag
2828
               && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2829
                s->avctx->has_b_frames = h->sps.num_reorder_frames;
2830
                s->low_delay = 0;
2831
            }
2832

    
2833
            if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
2834
               && !h->sps.bitstream_restriction_flag){
2835
                s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
2836
                s->low_delay= 0;
2837
            }
2838

    
2839
            pics = 0;
2840
            while(h->delayed_pic[pics]) pics++;
2841

    
2842
            assert(pics <= MAX_DELAYED_PIC_COUNT);
2843

    
2844
            h->delayed_pic[pics++] = cur;
2845
            if(cur->reference == 0)
2846
                cur->reference = DELAYED_PIC_REF;
2847

    
2848
            out = h->delayed_pic[0];
2849
            out_idx = 0;
2850
            for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2851
                if(h->delayed_pic[i]->poc < out->poc){
2852
                    out = h->delayed_pic[i];
2853
                    out_idx = i;
2854
                }
2855
            if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2856
                h->outputed_poc= INT_MIN;
2857
            out_of_order = out->poc < h->outputed_poc;
2858

    
2859
            if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2860
                { }
2861
            else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2862
               || (s->low_delay &&
2863
                ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2864
                 || cur->pict_type == FF_B_TYPE)))
2865
            {
2866
                s->low_delay = 0;
2867
                s->avctx->has_b_frames++;
2868
            }
2869

    
2870
            if(out_of_order || pics > s->avctx->has_b_frames){
2871
                out->reference &= ~DELAYED_PIC_REF;
2872
                for(i=out_idx; h->delayed_pic[i]; i++)
2873
                    h->delayed_pic[i] = h->delayed_pic[i+1];
2874
            }
2875
            if(!out_of_order && pics > s->avctx->has_b_frames){
2876
                *data_size = sizeof(AVFrame);
2877

    
2878
                if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2879
                    h->outputed_poc = INT_MIN;
2880
                } else
2881
                    h->outputed_poc = out->poc;
2882
                *pict= *(AVFrame*)out;
2883
            }else{
2884
                av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2885
            }
2886
        }
2887
    }
2888

    
2889
    assert(pict->data[0] || !*data_size);
2890
    ff_print_debug_info(s, pict);
2891
//printf("out %d\n", (int)pict->data[0]);
2892

    
2893
    return get_consumed_bytes(s, buf_index, buf_size);
2894
}
2895
#if 0
2896
static inline void fill_mb_avail(H264Context *h){
2897
    MpegEncContext * const s = &h->s;
2898
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2899

2900
    if(s->mb_y){
2901
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2902
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2903
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2904
    }else{
2905
        h->mb_avail[0]=
2906
        h->mb_avail[1]=
2907
        h->mb_avail[2]= 0;
2908
    }
2909
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2910
    h->mb_avail[4]= 1; //FIXME move out
2911
    h->mb_avail[5]= 0; //FIXME move out
2912
}
2913
#endif
2914

    
2915
#ifdef TEST
2916
#undef printf
2917
#undef random
2918
#define COUNT 8000
2919
#define SIZE (COUNT*40)
2920
int main(void){
2921
    int i;
2922
    uint8_t temp[SIZE];
2923
    PutBitContext pb;
2924
    GetBitContext gb;
2925
//    int int_temp[10000];
2926
    DSPContext dsp;
2927
    AVCodecContext avctx;
2928

    
2929
    dsputil_init(&dsp, &avctx);
2930

    
2931
    init_put_bits(&pb, temp, SIZE);
2932
    printf("testing unsigned exp golomb\n");
2933
    for(i=0; i<COUNT; i++){
2934
        START_TIMER
2935
        set_ue_golomb(&pb, i);
2936
        STOP_TIMER("set_ue_golomb");
2937
    }
2938
    flush_put_bits(&pb);
2939

    
2940
    init_get_bits(&gb, temp, 8*SIZE);
2941
    for(i=0; i<COUNT; i++){
2942
        int j, s;
2943

    
2944
        s= show_bits(&gb, 24);
2945

    
2946
        START_TIMER
2947
        j= get_ue_golomb(&gb);
2948
        if(j != i){
2949
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2950
//            return -1;
2951
        }
2952
        STOP_TIMER("get_ue_golomb");
2953
    }
2954

    
2955

    
2956
    init_put_bits(&pb, temp, SIZE);
2957
    printf("testing signed exp golomb\n");
2958
    for(i=0; i<COUNT; i++){
2959
        START_TIMER
2960
        set_se_golomb(&pb, i - COUNT/2);
2961
        STOP_TIMER("set_se_golomb");
2962
    }
2963
    flush_put_bits(&pb);
2964

    
2965
    init_get_bits(&gb, temp, 8*SIZE);
2966
    for(i=0; i<COUNT; i++){
2967
        int j, s;
2968

    
2969
        s= show_bits(&gb, 24);
2970

    
2971
        START_TIMER
2972
        j= get_se_golomb(&gb);
2973
        if(j != i - COUNT/2){
2974
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2975
//            return -1;
2976
        }
2977
        STOP_TIMER("get_se_golomb");
2978
    }
2979

    
2980
#if 0
2981
    printf("testing 4x4 (I)DCT\n");
2982

2983
    DCTELEM block[16];
2984
    uint8_t src[16], ref[16];
2985
    uint64_t error= 0, max_error=0;
2986

2987
    for(i=0; i<COUNT; i++){
2988
        int j;
2989
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2990
        for(j=0; j<16; j++){
2991
            ref[j]= random()%255;
2992
            src[j]= random()%255;
2993
        }
2994

2995
        h264_diff_dct_c(block, src, ref, 4);
2996

2997
        //normalize
2998
        for(j=0; j<16; j++){
2999
//            printf("%d ", block[j]);
3000
            block[j]= block[j]*4;
3001
            if(j&1) block[j]= (block[j]*4 + 2)/5;
3002
            if(j&4) block[j]= (block[j]*4 + 2)/5;
3003
        }
3004
//        printf("\n");
3005

3006
        s->dsp.h264_idct_add(ref, block, 4);
3007
/*        for(j=0; j<16; j++){
3008
            printf("%d ", ref[j]);
3009
        }
3010
        printf("\n");*/
3011

3012
        for(j=0; j<16; j++){
3013
            int diff= FFABS(src[j] - ref[j]);
3014

3015
            error+= diff*diff;
3016
            max_error= FFMAX(max_error, diff);
3017
        }
3018
    }
3019
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3020
    printf("testing quantizer\n");
3021
    for(qp=0; qp<52; qp++){
3022
        for(i=0; i<16; i++)
3023
            src1_block[i]= src2_block[i]= random()%255;
3024

3025
    }
3026
    printf("Testing NAL layer\n");
3027

3028
    uint8_t bitstream[COUNT];
3029
    uint8_t nal[COUNT*2];
3030
    H264Context h;
3031
    memset(&h, 0, sizeof(H264Context));
3032

3033
    for(i=0; i<COUNT; i++){
3034
        int zeros= i;
3035
        int nal_length;
3036
        int consumed;
3037
        int out_length;
3038
        uint8_t *out;
3039
        int j;
3040

3041
        for(j=0; j<COUNT; j++){
3042
            bitstream[j]= (random() % 255) + 1;
3043
        }
3044

3045
        for(j=0; j<zeros; j++){
3046
            int pos= random() % COUNT;
3047
            while(bitstream[pos] == 0){
3048
                pos++;
3049
                pos %= COUNT;
3050
            }
3051
            bitstream[pos]=0;
3052
        }
3053

3054
        START_TIMER
3055

3056
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3057
        if(nal_length<0){
3058
            printf("encoding failed\n");
3059
            return -1;
3060
        }
3061

3062
        out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3063

3064
        STOP_TIMER("NAL")
3065

3066
        if(out_length != COUNT){
3067
            printf("incorrect length %d %d\n", out_length, COUNT);
3068
            return -1;
3069
        }
3070

3071
        if(consumed != nal_length){
3072
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
3073
            return -1;
3074
        }
3075

3076
        if(memcmp(bitstream, out, COUNT)){
3077
            printf("mismatch\n");
3078
            return -1;
3079
        }
3080
    }
3081
#endif
3082

    
3083
    printf("Testing RBSP\n");
3084

    
3085

    
3086
    return 0;
3087
}
3088
#endif /* TEST */
3089

    
3090

    
3091
av_cold void ff_h264_free_context(H264Context *h)
3092
{
3093
    int i;
3094

    
3095
    free_tables(h); //FIXME cleanup init stuff perhaps
3096

    
3097
    for(i = 0; i < MAX_SPS_COUNT; i++)
3098
        av_freep(h->sps_buffers + i);
3099

    
3100
    for(i = 0; i < MAX_PPS_COUNT; i++)
3101
        av_freep(h->pps_buffers + i);
3102
}
3103

    
3104
av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3105
{
3106
    H264Context *h = avctx->priv_data;
3107
    MpegEncContext *s = &h->s;
3108

    
3109
    ff_h264_free_context(h);
3110

    
3111
    MPV_common_end(s);
3112

    
3113
//    memset(h, 0, sizeof(H264Context));
3114

    
3115
    return 0;
3116
}
3117

    
3118

    
3119
AVCodec h264_decoder = {
3120
    "h264",
3121
    CODEC_TYPE_VIDEO,
3122
    CODEC_ID_H264,
3123
    sizeof(H264Context),
3124
    ff_h264_decode_init,
3125
    NULL,
3126
    ff_h264_decode_end,
3127
    decode_frame,
3128
    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3129
    .flush= flush_dpb,
3130
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3131
    .pix_fmts= ff_hwaccel_pixfmt_list_420,
3132
};
3133

    
3134
#if CONFIG_H264_VDPAU_DECODER
3135
AVCodec h264_vdpau_decoder = {
3136
    "h264_vdpau",
3137
    CODEC_TYPE_VIDEO,
3138
    CODEC_ID_H264,
3139
    sizeof(H264Context),
3140
    ff_h264_decode_init,
3141
    NULL,
3142
    ff_h264_decode_end,
3143
    decode_frame,
3144
    CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3145
    .flush= flush_dpb,
3146
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3147
    .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3148
};
3149
#endif