Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ fb823b77

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
    dst->list_counts              = src->list_counts;
812

    
813
    dst->s.obmc_scratchpad = NULL;
814
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
815
}
816

    
817
/**
818
 * Init context
819
 * Allocate buffers which are not shared amongst multiple threads.
820
 */
821
static int context_init(H264Context *h){
822
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
823
    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
824

    
825
    return 0;
826
fail:
827
    return -1; // free_tables will clean up for us
828
}
829

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

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

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

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

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

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

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

    
853
    MPV_decode_defaults(s);
854

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

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

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

    
867
    avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
868

    
869
    ff_h264_decode_init_vlc();
870

    
871
    if(avctx->extradata_size > 0 && avctx->extradata &&
872
       *(char *)avctx->extradata == 1){
873
        h->is_avc = 1;
874
        h->got_avcC = 0;
875
    } else {
876
        h->is_avc = 0;
877
    }
878

    
879
    h->thread_context[0] = h;
880
    h->outputed_poc = INT_MIN;
881
    h->prev_poc_msb= 1<<16;
882
    ff_h264_reset_sei(h);
883
    if(avctx->codec_id == CODEC_ID_H264){
884
        if(avctx->ticks_per_frame == 1){
885
            s->avctx->time_base.den *=2;
886
        }
887
        avctx->ticks_per_frame = 2;
888
    }
889
    return 0;
890
}
891

    
892
int ff_h264_frame_start(H264Context *h){
893
    MpegEncContext * const s = &h->s;
894
    int i;
895

    
896
    if(MPV_frame_start(s, s->avctx) < 0)
897
        return -1;
898
    ff_er_frame_start(s);
899
    /*
900
     * MPV_frame_start uses pict_type to derive key_frame.
901
     * This is incorrect for H.264; IDR markings must be used.
902
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
903
     * See decode_nal_units().
904
     */
905
    s->current_picture_ptr->key_frame= 0;
906
    s->current_picture_ptr->mmco_reset= 0;
907

    
908
    assert(s->linesize && s->uvlinesize);
909

    
910
    for(i=0; i<16; i++){
911
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
912
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
913
    }
914
    for(i=0; i<4; i++){
915
        h->block_offset[16+i]=
916
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
917
        h->block_offset[24+16+i]=
918
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
919
    }
920

    
921
    /* can't be in alloc_tables because linesize isn't known there.
922
     * FIXME: redo bipred weight to not require extra buffer? */
923
    for(i = 0; i < s->avctx->thread_count; i++)
924
        if(!h->thread_context[i]->s.obmc_scratchpad)
925
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
926

    
927
    /* some macroblocks will be accessed before they're available */
928
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
929
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
930

    
931
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
932

    
933
    // We mark the current picture as non-reference after allocating it, so
934
    // that if we break out due to an error it can be released automatically
935
    // in the next MPV_frame_start().
936
    // SVQ3 as well as most other codecs have only last/next/current and thus
937
    // get released even with set reference, besides SVQ3 and others do not
938
    // mark frames as reference later "naturally".
939
    if(s->codec_id != CODEC_ID_SVQ3)
940
        s->current_picture_ptr->reference= 0;
941

    
942
    s->current_picture_ptr->field_poc[0]=
943
    s->current_picture_ptr->field_poc[1]= INT_MAX;
944
    assert(s->current_picture_ptr->long_ref==0);
945

    
946
    return 0;
947
}
948

    
949
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){
950
    MpegEncContext * const s = &h->s;
951
    int top_idx = 1;
952

    
953
    src_y  -=   linesize;
954
    src_cb -= uvlinesize;
955
    src_cr -= uvlinesize;
956

    
957
    if(!simple && FRAME_MBAFF){
958
        if(s->mb_y&1){
959
            if(!MB_MBAFF){
960
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
961
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
962
                if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
963
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
964
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
965
                }
966
            }
967
        }else if(MB_MBAFF){
968
            top_idx = 0;
969
        }else
970
            return;
971
    }
972

    
973
    // There are two lines saved, the line above the the top macroblock of a pair,
974
    // and the line above the bottom macroblock
975

    
976
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
977
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
978

    
979
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
980
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
981
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
982
    }
983
}
984

    
985
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){
986
    MpegEncContext * const s = &h->s;
987
    int temp8, i;
988
    uint64_t temp64;
989
    int deblock_left;
990
    int deblock_top;
991
    int mb_xy;
992
    int top_idx = 1;
993

    
994
    if(!simple && FRAME_MBAFF){
995
        if(s->mb_y&1){
996
            if(!MB_MBAFF)
997
                return;
998
        }else{
999
            top_idx = MB_MBAFF ? 0 : 1;
1000
        }
1001
    }
1002

    
1003
    if(h->deblocking_filter == 2) {
1004
        mb_xy = h->mb_xy;
1005
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
1006
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
1007
    } else {
1008
        deblock_left = (s->mb_x > 0);
1009
        deblock_top =  (s->mb_y > !!MB_FIELD);
1010
    }
1011

    
1012
    src_y  -=   linesize + 1;
1013
    src_cb -= uvlinesize + 1;
1014
    src_cr -= uvlinesize + 1;
1015

    
1016
#define XCHG(a,b,t,xchg)\
1017
t= a;\
1018
if(xchg)\
1019
    a= b;\
1020
b= t;
1021

    
1022
    if(deblock_top){
1023
        if(deblock_left){
1024
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x-1]+8), *(uint64_t*)(src_y -7), temp64, 1);
1025
        }
1026
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
1027
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
1028
        if(s->mb_x+1 < s->mb_width){
1029
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
1030
        }
1031
    }
1032

    
1033
    if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1034
        if(deblock_top){
1035
            if(deblock_left){
1036
                XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x-1]+16), *(uint64_t*)(src_cb -7), temp64, 1);
1037
                XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x-1]+24), *(uint64_t*)(src_cr -7), temp64, 1);
1038
            }
1039
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
1040
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
1041
        }
1042
    }
1043
}
1044

    
1045
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
1046
    MpegEncContext * const s = &h->s;
1047
    const int mb_x= s->mb_x;
1048
    const int mb_y= s->mb_y;
1049
    const int mb_xy= h->mb_xy;
1050
    const int mb_type= s->current_picture.mb_type[mb_xy];
1051
    uint8_t  *dest_y, *dest_cb, *dest_cr;
1052
    int linesize, uvlinesize /*dct_offset*/;
1053
    int i;
1054
    int *block_offset = &h->block_offset[0];
1055
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1056
    /* is_h264 should always be true if SVQ3 is disabled. */
1057
    const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1058
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1059
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1060

    
1061
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
1062
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
1063
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
1064

    
1065
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
1066
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
1067

    
1068
    h->list_counts[mb_xy]= h->list_count;
1069

    
1070
    if (!simple && MB_FIELD) {
1071
        linesize   = h->mb_linesize   = s->linesize * 2;
1072
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1073
        block_offset = &h->block_offset[24];
1074
        if(mb_y&1){ //FIXME move out of this function?
1075
            dest_y -= s->linesize*15;
1076
            dest_cb-= s->uvlinesize*7;
1077
            dest_cr-= s->uvlinesize*7;
1078
        }
1079
        if(FRAME_MBAFF) {
1080
            int list;
1081
            for(list=0; list<h->list_count; list++){
1082
                if(!USES_LIST(mb_type, list))
1083
                    continue;
1084
                if(IS_16X16(mb_type)){
1085
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
1086
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1087
                }else{
1088
                    for(i=0; i<16; i+=4){
1089
                        int ref = h->ref_cache[list][scan8[i]];
1090
                        if(ref >= 0)
1091
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1092
                    }
1093
                }
1094
            }
1095
        }
1096
    } else {
1097
        linesize   = h->mb_linesize   = s->linesize;
1098
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1099
//        dct_offset = s->linesize * 16;
1100
    }
1101

    
1102
    if (!simple && IS_INTRA_PCM(mb_type)) {
1103
        for (i=0; i<16; i++) {
1104
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
1105
        }
1106
        for (i=0; i<8; i++) {
1107
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
1108
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
1109
        }
1110
    } else {
1111
        if(IS_INTRA(mb_type)){
1112
            if(h->deblocking_filter)
1113
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
1114

    
1115
            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1116
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1117
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
1118
            }
1119

    
1120
            if(IS_INTRA4x4(mb_type)){
1121
                if(simple || !s->encoding){
1122
                    if(IS_8x8DCT(mb_type)){
1123
                        if(transform_bypass){
1124
                            idct_dc_add =
1125
                            idct_add    = s->dsp.add_pixels8;
1126
                        }else{
1127
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
1128
                            idct_add    = s->dsp.h264_idct8_add;
1129
                        }
1130
                        for(i=0; i<16; i+=4){
1131
                            uint8_t * const ptr= dest_y + block_offset[i];
1132
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1133
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1134
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
1135
                            }else{
1136
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
1137
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1138
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
1139
                                if(nnz){
1140
                                    if(nnz == 1 && h->mb[i*16])
1141
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
1142
                                    else
1143
                                        idct_add   (ptr, h->mb + i*16, linesize);
1144
                                }
1145
                            }
1146
                        }
1147
                    }else{
1148
                        if(transform_bypass){
1149
                            idct_dc_add =
1150
                            idct_add    = s->dsp.add_pixels4;
1151
                        }else{
1152
                            idct_dc_add = s->dsp.h264_idct_dc_add;
1153
                            idct_add    = s->dsp.h264_idct_add;
1154
                        }
1155
                        for(i=0; i<16; i++){
1156
                            uint8_t * const ptr= dest_y + block_offset[i];
1157
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1158

    
1159
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1160
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
1161
                            }else{
1162
                                uint8_t *topright;
1163
                                int nnz, tr;
1164
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1165
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1166
                                    assert(mb_y || linesize <= block_offset[i]);
1167
                                    if(!topright_avail){
1168
                                        tr= ptr[3 - linesize]*0x01010101;
1169
                                        topright= (uint8_t*) &tr;
1170
                                    }else
1171
                                        topright= ptr + 4 - linesize;
1172
                                }else
1173
                                    topright= NULL;
1174

    
1175
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1176
                                nnz = h->non_zero_count_cache[ scan8[i] ];
1177
                                if(nnz){
1178
                                    if(is_h264){
1179
                                        if(nnz == 1 && h->mb[i*16])
1180
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
1181
                                        else
1182
                                            idct_add   (ptr, h->mb + i*16, linesize);
1183
                                    }else
1184
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
1185
                                }
1186
                            }
1187
                        }
1188
                    }
1189
                }
1190
            }else{
1191
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1192
                if(is_h264){
1193
                    if(!transform_bypass)
1194
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
1195
                }else
1196
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
1197
            }
1198
            if(h->deblocking_filter)
1199
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
1200
        }else if(is_h264){
1201
            hl_motion(h, dest_y, dest_cb, dest_cr,
1202
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1203
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1204
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
1205
        }
1206

    
1207

    
1208
        if(!IS_INTRA4x4(mb_type)){
1209
            if(is_h264){
1210
                if(IS_INTRA16x16(mb_type)){
1211
                    if(transform_bypass){
1212
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1213
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
1214
                        }else{
1215
                            for(i=0; i<16; i++){
1216
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1217
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
1218
                            }
1219
                        }
1220
                    }else{
1221
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1222
                    }
1223
                }else if(h->cbp&15){
1224
                    if(transform_bypass){
1225
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1226
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1227
                        for(i=0; i<16; i+=di){
1228
                            if(h->non_zero_count_cache[ scan8[i] ]){
1229
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
1230
                            }
1231
                        }
1232
                    }else{
1233
                        if(IS_8x8DCT(mb_type)){
1234
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1235
                        }else{
1236
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
1237
                        }
1238
                    }
1239
                }
1240
            }else{
1241
                for(i=0; i<16; i++){
1242
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
1243
                        uint8_t * const ptr= dest_y + block_offset[i];
1244
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1245
                    }
1246
                }
1247
            }
1248
        }
1249

    
1250
        if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
1251
            uint8_t *dest[2] = {dest_cb, dest_cr};
1252
            if(transform_bypass){
1253
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
1254
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
1255
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
1256
                }else{
1257
                    idct_add = s->dsp.add_pixels4;
1258
                    for(i=16; i<16+8; i++){
1259
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
1260
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1261
                    }
1262
                }
1263
            }else{
1264
                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]);
1265
                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]);
1266
                if(is_h264){
1267
                    idct_add = s->dsp.h264_idct_add;
1268
                    idct_dc_add = s->dsp.h264_idct_dc_add;
1269
                    for(i=16; i<16+8; i++){
1270
                        if(h->non_zero_count_cache[ scan8[i] ])
1271
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1272
                        else if(h->mb[i*16])
1273
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
1274
                    }
1275
                }else{
1276
                    for(i=16; i<16+8; i++){
1277
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
1278
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
1279
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
1280
                        }
1281
                    }
1282
                }
1283
            }
1284
        }
1285
    }
1286
    if(h->cbp || IS_INTRA(mb_type))
1287
        s->dsp.clear_blocks(h->mb);
1288

    
1289
    if(h->deblocking_filter && 0) {
1290
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
1291
        fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
1292
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
1293
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
1294
        if (!simple && FRAME_MBAFF) {
1295
            ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1296
        } else {
1297
            ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
1298
        }
1299
    }
1300
}
1301

    
1302
/**
1303
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
1304
 */
1305
static void hl_decode_mb_simple(H264Context *h){
1306
    hl_decode_mb_internal(h, 1);
1307
}
1308

    
1309
/**
1310
 * Process a macroblock; this handles edge cases, such as interlacing.
1311
 */
1312
static void av_noinline hl_decode_mb_complex(H264Context *h){
1313
    hl_decode_mb_internal(h, 0);
1314
}
1315

    
1316
void ff_h264_hl_decode_mb(H264Context *h){
1317
    MpegEncContext * const s = &h->s;
1318
    const int mb_xy= h->mb_xy;
1319
    const int mb_type= s->current_picture.mb_type[mb_xy];
1320
    int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1321

    
1322
    if (is_complex)
1323
        hl_decode_mb_complex(h);
1324
    else hl_decode_mb_simple(h);
1325
}
1326

    
1327
static int pred_weight_table(H264Context *h){
1328
    MpegEncContext * const s = &h->s;
1329
    int list, i;
1330
    int luma_def, chroma_def;
1331

    
1332
    h->use_weight= 0;
1333
    h->use_weight_chroma= 0;
1334
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
1335
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
1336
    luma_def = 1<<h->luma_log2_weight_denom;
1337
    chroma_def = 1<<h->chroma_log2_weight_denom;
1338

    
1339
    for(list=0; list<2; list++){
1340
        h->luma_weight_flag[list]   = 0;
1341
        h->chroma_weight_flag[list] = 0;
1342
        for(i=0; i<h->ref_count[list]; i++){
1343
            int luma_weight_flag, chroma_weight_flag;
1344

    
1345
            luma_weight_flag= get_bits1(&s->gb);
1346
            if(luma_weight_flag){
1347
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
1348
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
1349
                if(   h->luma_weight[list][i] != luma_def
1350
                   || h->luma_offset[list][i] != 0) {
1351
                    h->use_weight= 1;
1352
                    h->luma_weight_flag[list]= 1;
1353
                }
1354
            }else{
1355
                h->luma_weight[list][i]= luma_def;
1356
                h->luma_offset[list][i]= 0;
1357
            }
1358

    
1359
            if(CHROMA){
1360
                chroma_weight_flag= get_bits1(&s->gb);
1361
                if(chroma_weight_flag){
1362
                    int j;
1363
                    for(j=0; j<2; j++){
1364
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
1365
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
1366
                        if(   h->chroma_weight[list][i][j] != chroma_def
1367
                           || h->chroma_offset[list][i][j] != 0) {
1368
                            h->use_weight_chroma= 1;
1369
                            h->chroma_weight_flag[list]= 1;
1370
                        }
1371
                    }
1372
                }else{
1373
                    int j;
1374
                    for(j=0; j<2; j++){
1375
                        h->chroma_weight[list][i][j]= chroma_def;
1376
                        h->chroma_offset[list][i][j]= 0;
1377
                    }
1378
                }
1379
            }
1380
        }
1381
        if(h->slice_type_nos != FF_B_TYPE) break;
1382
    }
1383
    h->use_weight= h->use_weight || h->use_weight_chroma;
1384
    return 0;
1385
}
1386

    
1387
static void implicit_weight_table(H264Context *h){
1388
    MpegEncContext * const s = &h->s;
1389
    int ref0, ref1, i;
1390
    int cur_poc = s->current_picture_ptr->poc;
1391

    
1392
    for (i = 0; i < 2; i++) {
1393
        h->luma_weight_flag[i]   = 0;
1394
        h->chroma_weight_flag[i] = 0;
1395
    }
1396

    
1397
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
1398
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
1399
        h->use_weight= 0;
1400
        h->use_weight_chroma= 0;
1401
        return;
1402
    }
1403

    
1404
    h->use_weight= 2;
1405
    h->use_weight_chroma= 2;
1406
    h->luma_log2_weight_denom= 5;
1407
    h->chroma_log2_weight_denom= 5;
1408

    
1409
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
1410
        int poc0 = h->ref_list[0][ref0].poc;
1411
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
1412
            int poc1 = h->ref_list[1][ref1].poc;
1413
            int td = av_clip(poc1 - poc0, -128, 127);
1414
            if(td){
1415
                int tb = av_clip(cur_poc - poc0, -128, 127);
1416
                int tx = (16384 + (FFABS(td) >> 1)) / td;
1417
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
1418
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
1419
                    h->implicit_weight[ref0][ref1] = 32;
1420
                else
1421
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
1422
            }else
1423
                h->implicit_weight[ref0][ref1] = 32;
1424
        }
1425
    }
1426
}
1427

    
1428
/**
1429
 * instantaneous decoder refresh.
1430
 */
1431
static void idr(H264Context *h){
1432
    ff_h264_remove_all_refs(h);
1433
    h->prev_frame_num= 0;
1434
    h->prev_frame_num_offset= 0;
1435
    h->prev_poc_msb=
1436
    h->prev_poc_lsb= 0;
1437
}
1438

    
1439
/* forget old pics after a seek */
1440
static void flush_dpb(AVCodecContext *avctx){
1441
    H264Context *h= avctx->priv_data;
1442
    int i;
1443
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
1444
        if(h->delayed_pic[i])
1445
            h->delayed_pic[i]->reference= 0;
1446
        h->delayed_pic[i]= NULL;
1447
    }
1448
    h->outputed_poc= INT_MIN;
1449
    h->prev_interlaced_frame = 1;
1450
    idr(h);
1451
    if(h->s.current_picture_ptr)
1452
        h->s.current_picture_ptr->reference= 0;
1453
    h->s.first_field= 0;
1454
    ff_h264_reset_sei(h);
1455
    ff_mpeg_flush(avctx);
1456
}
1457

    
1458
static int init_poc(H264Context *h){
1459
    MpegEncContext * const s = &h->s;
1460
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
1461
    int field_poc[2];
1462
    Picture *cur = s->current_picture_ptr;
1463

    
1464
    h->frame_num_offset= h->prev_frame_num_offset;
1465
    if(h->frame_num < h->prev_frame_num)
1466
        h->frame_num_offset += max_frame_num;
1467

    
1468
    if(h->sps.poc_type==0){
1469
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
1470

    
1471
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
1472
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1473
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
1474
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1475
        else
1476
            h->poc_msb = h->prev_poc_msb;
1477
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
1478
        field_poc[0] =
1479
        field_poc[1] = h->poc_msb + h->poc_lsb;
1480
        if(s->picture_structure == PICT_FRAME)
1481
            field_poc[1] += h->delta_poc_bottom;
1482
    }else if(h->sps.poc_type==1){
1483
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1484
        int i;
1485

    
1486
        if(h->sps.poc_cycle_length != 0)
1487
            abs_frame_num = h->frame_num_offset + h->frame_num;
1488
        else
1489
            abs_frame_num = 0;
1490

    
1491
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
1492
            abs_frame_num--;
1493

    
1494
        expected_delta_per_poc_cycle = 0;
1495
        for(i=0; i < h->sps.poc_cycle_length; i++)
1496
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
1497

    
1498
        if(abs_frame_num > 0){
1499
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1500
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1501

    
1502
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1503
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
1504
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
1505
        } else
1506
            expectedpoc = 0;
1507

    
1508
        if(h->nal_ref_idc == 0)
1509
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1510

    
1511
        field_poc[0] = expectedpoc + h->delta_poc[0];
1512
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1513

    
1514
        if(s->picture_structure == PICT_FRAME)
1515
            field_poc[1] += h->delta_poc[1];
1516
    }else{
1517
        int poc= 2*(h->frame_num_offset + h->frame_num);
1518

    
1519
        if(!h->nal_ref_idc)
1520
            poc--;
1521

    
1522
        field_poc[0]= poc;
1523
        field_poc[1]= poc;
1524
    }
1525

    
1526
    if(s->picture_structure != PICT_BOTTOM_FIELD)
1527
        s->current_picture_ptr->field_poc[0]= field_poc[0];
1528
    if(s->picture_structure != PICT_TOP_FIELD)
1529
        s->current_picture_ptr->field_poc[1]= field_poc[1];
1530
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
1531

    
1532
    return 0;
1533
}
1534

    
1535

    
1536
/**
1537
 * initialize scan tables
1538
 */
1539
static void init_scan_tables(H264Context *h){
1540
    MpegEncContext * const s = &h->s;
1541
    int i;
1542
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
1543
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
1544
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
1545
    }else{
1546
        for(i=0; i<16; i++){
1547
#define T(x) (x>>2) | ((x<<2) & 0xF)
1548
            h->zigzag_scan[i] = T(zigzag_scan[i]);
1549
            h-> field_scan[i] = T( field_scan[i]);
1550
#undef T
1551
        }
1552
    }
1553
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
1554
        memcpy(h->zigzag_scan8x8,       ff_zigzag_direct,     64*sizeof(uint8_t));
1555
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
1556
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
1557
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
1558
    }else{
1559
        for(i=0; i<64; i++){
1560
#define T(x) (x>>3) | ((x&7)<<3)
1561
            h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
1562
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
1563
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
1564
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
1565
#undef T
1566
        }
1567
    }
1568
    if(h->sps.transform_bypass){ //FIXME same ugly
1569
        h->zigzag_scan_q0          = zigzag_scan;
1570
        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
1571
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
1572
        h->field_scan_q0           = field_scan;
1573
        h->field_scan8x8_q0        = field_scan8x8;
1574
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
1575
    }else{
1576
        h->zigzag_scan_q0          = h->zigzag_scan;
1577
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
1578
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
1579
        h->field_scan_q0           = h->field_scan;
1580
        h->field_scan8x8_q0        = h->field_scan8x8;
1581
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
1582
    }
1583
}
1584

    
1585
static void field_end(H264Context *h){
1586
    MpegEncContext * const s = &h->s;
1587
    AVCodecContext * const avctx= s->avctx;
1588
    s->mb_y= 0;
1589

    
1590
    s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
1591
    s->current_picture_ptr->pict_type= s->pict_type;
1592

    
1593
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1594
        ff_vdpau_h264_set_reference_frames(s);
1595

    
1596
    if(!s->dropable) {
1597
        ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1598
        h->prev_poc_msb= h->poc_msb;
1599
        h->prev_poc_lsb= h->poc_lsb;
1600
    }
1601
    h->prev_frame_num_offset= h->frame_num_offset;
1602
    h->prev_frame_num= h->frame_num;
1603

    
1604
    if (avctx->hwaccel) {
1605
        if (avctx->hwaccel->end_frame(avctx) < 0)
1606
            av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
1607
    }
1608

    
1609
    if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1610
        ff_vdpau_h264_picture_complete(s);
1611

    
1612
    /*
1613
     * FIXME: Error handling code does not seem to support interlaced
1614
     * when slices span multiple rows
1615
     * The ff_er_add_slice calls don't work right for bottom
1616
     * fields; they cause massive erroneous error concealing
1617
     * Error marking covers both fields (top and bottom).
1618
     * This causes a mismatched s->error_count
1619
     * and a bad error table. Further, the error count goes to
1620
     * INT_MAX when called for bottom field, because mb_y is
1621
     * past end by one (callers fault) and resync_mb_y != 0
1622
     * causes problems for the first MB line, too.
1623
     */
1624
    if (!FIELD_PICTURE)
1625
        ff_er_frame_end(s);
1626

    
1627
    MPV_frame_end(s);
1628

    
1629
    h->current_slice=0;
1630
}
1631

    
1632
/**
1633
 * Replicates H264 "master" context to thread contexts.
1634
 */
1635
static void clone_slice(H264Context *dst, H264Context *src)
1636
{
1637
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
1638
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
1639
    dst->s.current_picture      = src->s.current_picture;
1640
    dst->s.linesize             = src->s.linesize;
1641
    dst->s.uvlinesize           = src->s.uvlinesize;
1642
    dst->s.first_field          = src->s.first_field;
1643

    
1644
    dst->prev_poc_msb           = src->prev_poc_msb;
1645
    dst->prev_poc_lsb           = src->prev_poc_lsb;
1646
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
1647
    dst->prev_frame_num         = src->prev_frame_num;
1648
    dst->short_ref_count        = src->short_ref_count;
1649

    
1650
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
1651
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
1652
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1653
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
1654

    
1655
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
1656
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
1657
}
1658

    
1659
/**
1660
 * decodes a slice header.
1661
 * This will also call MPV_common_init() and frame_start() as needed.
1662
 *
1663
 * @param h h264context
1664
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
1665
 *
1666
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1667
 */
1668
static int decode_slice_header(H264Context *h, H264Context *h0){
1669
    MpegEncContext * const s = &h->s;
1670
    MpegEncContext * const s0 = &h0->s;
1671
    unsigned int first_mb_in_slice;
1672
    unsigned int pps_id;
1673
    int num_ref_idx_active_override_flag;
1674
    unsigned int slice_type, tmp, i, j;
1675
    int default_ref_list_done = 0;
1676
    int last_pic_structure;
1677

    
1678
    s->dropable= h->nal_ref_idc == 0;
1679

    
1680
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
1681
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
1682
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
1683
    }else{
1684
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
1685
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
1686
    }
1687

    
1688
    first_mb_in_slice= get_ue_golomb(&s->gb);
1689

    
1690
    if(first_mb_in_slice == 0){ //FIXME better field boundary detection
1691
        if(h0->current_slice && FIELD_PICTURE){
1692
            field_end(h);
1693
        }
1694

    
1695
        h0->current_slice = 0;
1696
        if (!s0->first_field)
1697
            s->current_picture_ptr= NULL;
1698
    }
1699

    
1700
    slice_type= get_ue_golomb_31(&s->gb);
1701
    if(slice_type > 9){
1702
        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);
1703
        return -1;
1704
    }
1705
    if(slice_type > 4){
1706
        slice_type -= 5;
1707
        h->slice_type_fixed=1;
1708
    }else
1709
        h->slice_type_fixed=0;
1710

    
1711
    slice_type= golomb_to_pict_type[ slice_type ];
1712
    if (slice_type == FF_I_TYPE
1713
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
1714
        default_ref_list_done = 1;
1715
    }
1716
    h->slice_type= slice_type;
1717
    h->slice_type_nos= slice_type & 3;
1718

    
1719
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
1720
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
1721
        av_log(h->s.avctx, AV_LOG_ERROR,
1722
               "B picture before any references, skipping\n");
1723
        return -1;
1724
    }
1725

    
1726
    pps_id= get_ue_golomb(&s->gb);
1727
    if(pps_id>=MAX_PPS_COUNT){
1728
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
1729
        return -1;
1730
    }
1731
    if(!h0->pps_buffers[pps_id]) {
1732
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
1733
        return -1;
1734
    }
1735
    h->pps= *h0->pps_buffers[pps_id];
1736

    
1737
    if(!h0->sps_buffers[h->pps.sps_id]) {
1738
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
1739
        return -1;
1740
    }
1741
    h->sps = *h0->sps_buffers[h->pps.sps_id];
1742

    
1743
    if(h == h0 && h->dequant_coeff_pps != pps_id){
1744
        h->dequant_coeff_pps = pps_id;
1745
        init_dequant_tables(h);
1746
    }
1747

    
1748
    s->mb_width= h->sps.mb_width;
1749
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1750

    
1751
    h->b_stride=  s->mb_width*4;
1752
    h->b8_stride= s->mb_width*2;
1753

    
1754
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
1755
    if(h->sps.frame_mbs_only_flag)
1756
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
1757
    else
1758
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
1759

    
1760
    if (s->context_initialized
1761
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
1762
        if(h != h0)
1763
            return -1;   // width / height changed during parallelized decoding
1764
        free_tables(h);
1765
        flush_dpb(s->avctx);
1766
        MPV_common_end(s);
1767
    }
1768
    if (!s->context_initialized) {
1769
        if(h != h0)
1770
            return -1;  // we cant (re-)initialize context during parallel decoding
1771

    
1772
        avcodec_set_dimensions(s->avctx, s->width, s->height);
1773
        s->avctx->sample_aspect_ratio= h->sps.sar;
1774
        if(!s->avctx->sample_aspect_ratio.den)
1775
            s->avctx->sample_aspect_ratio.den = 1;
1776

    
1777
        if(h->sps.video_signal_type_present_flag){
1778
            s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
1779
            if(h->sps.colour_description_present_flag){
1780
                s->avctx->color_primaries = h->sps.color_primaries;
1781
                s->avctx->color_trc       = h->sps.color_trc;
1782
                s->avctx->colorspace      = h->sps.colorspace;
1783
            }
1784
        }
1785

    
1786
        if(h->sps.timing_info_present_flag){
1787
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
1788
            if(h->x264_build > 0 && h->x264_build < 44)
1789
                s->avctx->time_base.den *= 2;
1790
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
1791
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
1792
        }
1793
        s->avctx->pix_fmt = s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts);
1794
        s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
1795

    
1796
        if (MPV_common_init(s) < 0)
1797
            return -1;
1798
        s->first_field = 0;
1799
        h->prev_interlaced_frame = 1;
1800

    
1801
        init_scan_tables(h);
1802
        ff_h264_alloc_tables(h);
1803

    
1804
        for(i = 1; i < s->avctx->thread_count; i++) {
1805
            H264Context *c;
1806
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
1807
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
1808
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
1809
            c->sps = h->sps;
1810
            c->pps = h->pps;
1811
            init_scan_tables(c);
1812
            clone_tables(c, h);
1813
        }
1814

    
1815
        for(i = 0; i < s->avctx->thread_count; i++)
1816
            if(context_init(h->thread_context[i]) < 0)
1817
                return -1;
1818
    }
1819

    
1820
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
1821

    
1822
    h->mb_mbaff = 0;
1823
    h->mb_aff_frame = 0;
1824
    last_pic_structure = s0->picture_structure;
1825
    if(h->sps.frame_mbs_only_flag){
1826
        s->picture_structure= PICT_FRAME;
1827
    }else{
1828
        if(get_bits1(&s->gb)) { //field_pic_flag
1829
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
1830
        } else {
1831
            s->picture_structure= PICT_FRAME;
1832
            h->mb_aff_frame = h->sps.mb_aff;
1833
        }
1834
    }
1835
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
1836

    
1837
    if(h0->current_slice == 0){
1838
        while(h->frame_num !=  h->prev_frame_num &&
1839
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
1840
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
1841
            if (ff_h264_frame_start(h) < 0)
1842
                return -1;
1843
            h->prev_frame_num++;
1844
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
1845
            s->current_picture_ptr->frame_num= h->prev_frame_num;
1846
            ff_h264_execute_ref_pic_marking(h, NULL, 0);
1847
        }
1848

    
1849
        /* See if we have a decoded first field looking for a pair... */
1850
        if (s0->first_field) {
1851
            assert(s0->current_picture_ptr);
1852
            assert(s0->current_picture_ptr->data[0]);
1853
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
1854

    
1855
            /* figure out if we have a complementary field pair */
1856
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
1857
                /*
1858
                 * Previous field is unmatched. Don't display it, but let it
1859
                 * remain for reference if marked as such.
1860
                 */
1861
                s0->current_picture_ptr = NULL;
1862
                s0->first_field = FIELD_PICTURE;
1863

    
1864
            } else {
1865
                if (h->nal_ref_idc &&
1866
                        s0->current_picture_ptr->reference &&
1867
                        s0->current_picture_ptr->frame_num != h->frame_num) {
1868
                    /*
1869
                     * This and previous field were reference, but had
1870
                     * different frame_nums. Consider this field first in
1871
                     * pair. Throw away previous field except for reference
1872
                     * purposes.
1873
                     */
1874
                    s0->first_field = 1;
1875
                    s0->current_picture_ptr = NULL;
1876

    
1877
                } else {
1878
                    /* Second field in complementary pair */
1879
                    s0->first_field = 0;
1880
                }
1881
            }
1882

    
1883
        } else {
1884
            /* Frame or first field in a potentially complementary pair */
1885
            assert(!s0->current_picture_ptr);
1886
            s0->first_field = FIELD_PICTURE;
1887
        }
1888

    
1889
        if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
1890
            s0->first_field = 0;
1891
            return -1;
1892
        }
1893
    }
1894
    if(h != h0)
1895
        clone_slice(h, h0);
1896

    
1897
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
1898

    
1899
    assert(s->mb_num == s->mb_width * s->mb_height);
1900
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
1901
       first_mb_in_slice                    >= s->mb_num){
1902
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1903
        return -1;
1904
    }
1905
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
1906
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
1907
    if (s->picture_structure == PICT_BOTTOM_FIELD)
1908
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
1909
    assert(s->mb_y < s->mb_height);
1910

    
1911
    if(s->picture_structure==PICT_FRAME){
1912
        h->curr_pic_num=   h->frame_num;
1913
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
1914
    }else{
1915
        h->curr_pic_num= 2*h->frame_num + 1;
1916
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
1917
    }
1918

    
1919
    if(h->nal_unit_type == NAL_IDR_SLICE){
1920
        get_ue_golomb(&s->gb); /* idr_pic_id */
1921
    }
1922

    
1923
    if(h->sps.poc_type==0){
1924
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
1925

    
1926
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
1927
            h->delta_poc_bottom= get_se_golomb(&s->gb);
1928
        }
1929
    }
1930

    
1931
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
1932
        h->delta_poc[0]= get_se_golomb(&s->gb);
1933

    
1934
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
1935
            h->delta_poc[1]= get_se_golomb(&s->gb);
1936
    }
1937

    
1938
    init_poc(h);
1939

    
1940
    if(h->pps.redundant_pic_cnt_present){
1941
        h->redundant_pic_count= get_ue_golomb(&s->gb);
1942
    }
1943

    
1944
    //set defaults, might be overridden a few lines later
1945
    h->ref_count[0]= h->pps.ref_count[0];
1946
    h->ref_count[1]= h->pps.ref_count[1];
1947

    
1948
    if(h->slice_type_nos != FF_I_TYPE){
1949
        if(h->slice_type_nos == FF_B_TYPE){
1950
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
1951
        }
1952
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
1953

    
1954
        if(num_ref_idx_active_override_flag){
1955
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
1956
            if(h->slice_type_nos==FF_B_TYPE)
1957
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
1958

    
1959
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
1960
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
1961
                h->ref_count[0]= h->ref_count[1]= 1;
1962
                return -1;
1963
            }
1964
        }
1965
        if(h->slice_type_nos == FF_B_TYPE)
1966
            h->list_count= 2;
1967
        else
1968
            h->list_count= 1;
1969
    }else
1970
        h->list_count= 0;
1971

    
1972
    if(!default_ref_list_done){
1973
        ff_h264_fill_default_ref_list(h);
1974
    }
1975

    
1976
    if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
1977
        return -1;
1978

    
1979
    if(h->slice_type_nos!=FF_I_TYPE){
1980
        s->last_picture_ptr= &h->ref_list[0][0];
1981
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1982
    }
1983
    if(h->slice_type_nos==FF_B_TYPE){
1984
        s->next_picture_ptr= &h->ref_list[1][0];
1985
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1986
    }
1987

    
1988
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
1989
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
1990
        pred_weight_table(h);
1991
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
1992
        implicit_weight_table(h);
1993
    else {
1994
        h->use_weight = 0;
1995
        for (i = 0; i < 2; i++) {
1996
            h->luma_weight_flag[i]   = 0;
1997
            h->chroma_weight_flag[i] = 0;
1998
        }
1999
    }
2000

    
2001
    if(h->nal_ref_idc)
2002
        ff_h264_decode_ref_pic_marking(h0, &s->gb);
2003

    
2004
    if(FRAME_MBAFF)
2005
        ff_h264_fill_mbaff_ref_list(h);
2006

    
2007
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
2008
        ff_h264_direct_dist_scale_factor(h);
2009
    ff_h264_direct_ref_list_init(h);
2010

    
2011
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
2012
        tmp = get_ue_golomb_31(&s->gb);
2013
        if(tmp > 2){
2014
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2015
            return -1;
2016
        }
2017
        h->cabac_init_idc= tmp;
2018
    }
2019

    
2020
    h->last_qscale_diff = 0;
2021
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
2022
    if(tmp>51){
2023
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
2024
        return -1;
2025
    }
2026
    s->qscale= tmp;
2027
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2028
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
2029
    //FIXME qscale / qp ... stuff
2030
    if(h->slice_type == FF_SP_TYPE){
2031
        get_bits1(&s->gb); /* sp_for_switch_flag */
2032
    }
2033
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
2034
        get_se_golomb(&s->gb); /* slice_qs_delta */
2035
    }
2036

    
2037
    h->deblocking_filter = 1;
2038
    h->slice_alpha_c0_offset = 0;
2039
    h->slice_beta_offset = 0;
2040
    if( h->pps.deblocking_filter_parameters_present ) {
2041
        tmp= get_ue_golomb_31(&s->gb);
2042
        if(tmp > 2){
2043
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
2044
            return -1;
2045
        }
2046
        h->deblocking_filter= tmp;
2047
        if(h->deblocking_filter < 2)
2048
            h->deblocking_filter^= 1; // 1<->0
2049

    
2050
        if( h->deblocking_filter ) {
2051
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
2052
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
2053
        }
2054
    }
2055

    
2056
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
2057
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
2058
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
2059
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
2060
        h->deblocking_filter= 0;
2061

    
2062
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
2063
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
2064
            /* Cheat slightly for speed:
2065
               Do not bother to deblock across slices. */
2066
            h->deblocking_filter = 2;
2067
        } else {
2068
            h0->max_contexts = 1;
2069
            if(!h0->single_decode_warning) {
2070
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
2071
                h0->single_decode_warning = 1;
2072
            }
2073
            if(h != h0)
2074
                return 1; // deblocking switched inside frame
2075
        }
2076
    }
2077
    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]);
2078

    
2079
#if 0 //FMO
2080
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2081
        slice_group_change_cycle= get_bits(&s->gb, ?);
2082
#endif
2083

    
2084
    h0->last_slice_type = slice_type;
2085
    h->slice_num = ++h0->current_slice;
2086
    if(h->slice_num >= MAX_SLICES){
2087
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
2088
    }
2089

    
2090
    for(j=0; j<2; j++){
2091
        int id_list[16];
2092
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
2093
        for(i=0; i<16; i++){
2094
            id_list[i]= 60;
2095
            if(h->ref_list[j][i].data[0]){
2096
                int k;
2097
                uint8_t *base= h->ref_list[j][i].base[0];
2098
                for(k=0; k<h->short_ref_count; k++)
2099
                    if(h->short_ref[k]->base[0] == base){
2100
                        id_list[i]= k;
2101
                        break;
2102
                    }
2103
                for(k=0; k<h->long_ref_count; k++)
2104
                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
2105
                        id_list[i]= h->short_ref_count + k;
2106
                        break;
2107
                    }
2108
            }
2109
        }
2110

    
2111
        ref2frm[0]=
2112
        ref2frm[1]= -1;
2113
        for(i=0; i<16; i++)
2114
            ref2frm[i+2]= 4*id_list[i]
2115
                          +(h->ref_list[j][i].reference&3);
2116
        ref2frm[18+0]=
2117
        ref2frm[18+1]= -1;
2118
        for(i=16; i<48; i++)
2119
            ref2frm[i+4]= 4*id_list[(i-16)>>1]
2120
                          +(h->ref_list[j][i].reference&3);
2121
    }
2122

    
2123
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
2124
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
2125

    
2126
    s->avctx->refs= h->sps.ref_frame_count;
2127

    
2128
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2129
        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",
2130
               h->slice_num,
2131
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
2132
               first_mb_in_slice,
2133
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2134
               pps_id, h->frame_num,
2135
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2136
               h->ref_count[0], h->ref_count[1],
2137
               s->qscale,
2138
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
2139
               h->use_weight,
2140
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
2141
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
2142
               );
2143
    }
2144

    
2145
    return 0;
2146
}
2147

    
2148
int ff_h264_get_slice_type(H264Context *h)
2149
{
2150
    switch (h->slice_type) {
2151
    case FF_P_TYPE:  return 0;
2152
    case FF_B_TYPE:  return 1;
2153
    case FF_I_TYPE:  return 2;
2154
    case FF_SP_TYPE: return 3;
2155
    case FF_SI_TYPE: return 4;
2156
    default:         return -1;
2157
    }
2158
}
2159

    
2160
static void loop_filter(H264Context *h){
2161
    MpegEncContext * const s = &h->s;
2162
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2163
    int linesize, uvlinesize, mb_x, mb_y;
2164
    const int end_mb_y= s->mb_y + FRAME_MBAFF;
2165
    const int old_slice_type= h->slice_type;
2166

    
2167
    if(h->deblocking_filter) {
2168
        for(mb_x= 0; mb_x<s->mb_width; mb_x++){
2169
            for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
2170
                int list, mb_xy, mb_type, is_complex;
2171
                mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
2172
                h->slice_num= h->slice_table[mb_xy];
2173
                mb_type= s->current_picture.mb_type[mb_xy];
2174
                h->list_count= h->list_counts[mb_xy];
2175
                if(h->list_count==2){
2176
                    h->slice_type= h->slice_type_nos= FF_B_TYPE;
2177
                }else if(h->list_count==1){
2178
                    h->slice_type= h->slice_type_nos= FF_P_TYPE;
2179
                }else
2180
                    h->slice_type= h->slice_type_nos= FF_I_TYPE;
2181

    
2182
                if(FRAME_MBAFF)
2183
                    h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2184

    
2185
                is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0; //FIXME qscale might be wrong
2186

    
2187
                s->mb_x= mb_x;
2188
                s->mb_y= mb_y;
2189
                dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2190
                dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2191
                dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2192
                    //FIXME simplify above
2193

    
2194
                if (MB_FIELD) {
2195
                    linesize   = h->mb_linesize   = s->linesize * 2;
2196
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2197
                    if(mb_y&1){ //FIXME move out of this function?
2198
                        dest_y -= s->linesize*15;
2199
                        dest_cb-= s->uvlinesize*7;
2200
                        dest_cr-= s->uvlinesize*7;
2201
                    }
2202
                } else {
2203
                    linesize   = h->mb_linesize   = s->linesize;
2204
                    uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2205
                }
2206
                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, !is_complex);
2207
                if(fill_filter_caches(h, mb_type) < 0)
2208
                    continue;
2209
                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2210
                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2211

    
2212
                if (is_complex && FRAME_MBAFF) {
2213
                    ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2214
                } else {
2215
                    ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2216
                }
2217
            }
2218
        }
2219
    }
2220
    h->slice_type= old_slice_type;
2221
    s->mb_x= 0;
2222
    s->mb_y= end_mb_y - FRAME_MBAFF;
2223
}
2224

    
2225
static int decode_slice(struct AVCodecContext *avctx, void *arg){
2226
    H264Context *h = *(void**)arg;
2227
    MpegEncContext * const s = &h->s;
2228
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
2229

    
2230
    s->mb_skip_run= -1;
2231

    
2232
    h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME || s->codec_id != CODEC_ID_H264 ||
2233
                    (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
2234

    
2235
    if( h->pps.cabac ) {
2236
        /* realign */
2237
        align_get_bits( &s->gb );
2238

    
2239
        /* init cabac */
2240
        ff_init_cabac_states( &h->cabac);
2241
        ff_init_cabac_decoder( &h->cabac,
2242
                               s->gb.buffer + get_bits_count(&s->gb)/8,
2243
                               (get_bits_left(&s->gb) + 7)/8);
2244

    
2245
        ff_h264_init_cabac_states(h);
2246

    
2247
        for(;;){
2248
//START_TIMER
2249
            int ret = ff_h264_decode_mb_cabac(h);
2250
            int eos;
2251
//STOP_TIMER("decode_mb_cabac")
2252

    
2253
            if(ret>=0) ff_h264_hl_decode_mb(h);
2254

    
2255
            if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
2256
                s->mb_y++;
2257

    
2258
                ret = ff_h264_decode_mb_cabac(h);
2259

    
2260
                if(ret>=0) ff_h264_hl_decode_mb(h);
2261
                s->mb_y--;
2262
            }
2263
            eos = get_cabac_terminate( &h->cabac );
2264

    
2265
            if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2266
                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);
2267
                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);
2268
                return -1;
2269
            }
2270

    
2271
            if( ++s->mb_x >= s->mb_width ) {
2272
                s->mb_x = 0;
2273
                loop_filter(h);
2274
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2275
                ++s->mb_y;
2276
                if(FIELD_OR_MBAFF_PICTURE) {
2277
                    ++s->mb_y;
2278
                }
2279
            }
2280

    
2281
            if( eos || s->mb_y >= s->mb_height ) {
2282
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2283
                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);
2284
                return 0;
2285
            }
2286
        }
2287

    
2288
    } else {
2289
        for(;;){
2290
            int ret = ff_h264_decode_mb_cavlc(h);
2291

    
2292
            if(ret>=0) ff_h264_hl_decode_mb(h);
2293

    
2294
            if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
2295
                s->mb_y++;
2296
                ret = ff_h264_decode_mb_cavlc(h);
2297

    
2298
                if(ret>=0) ff_h264_hl_decode_mb(h);
2299
                s->mb_y--;
2300
            }
2301

    
2302
            if(ret<0){
2303
                av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2304
                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);
2305

    
2306
                return -1;
2307
            }
2308

    
2309
            if(++s->mb_x >= s->mb_width){
2310
                s->mb_x=0;
2311
                loop_filter(h);
2312
                ff_draw_horiz_band(s, 16*s->mb_y, 16);
2313
                ++s->mb_y;
2314
                if(FIELD_OR_MBAFF_PICTURE) {
2315
                    ++s->mb_y;
2316
                }
2317
                if(s->mb_y >= s->mb_height){
2318
                    tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2319

    
2320
                    if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
2321
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
2322

    
2323
                        return 0;
2324
                    }else{
2325
                        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);
2326

    
2327
                        return -1;
2328
                    }
2329
                }
2330
            }
2331

    
2332
            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
2333
                tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
2334
                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
2335
                    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);
2336

    
2337
                    return 0;
2338
                }else{
2339
                    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);
2340

    
2341
                    return -1;
2342
                }
2343
            }
2344
        }
2345
    }
2346

    
2347
#if 0
2348
    for(;s->mb_y < s->mb_height; s->mb_y++){
2349
        for(;s->mb_x < s->mb_width; s->mb_x++){
2350
            int ret= decode_mb(h);
2351

2352
            ff_h264_hl_decode_mb(h);
2353

2354
            if(ret<0){
2355
                av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
2356
                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);
2357

2358
                return -1;
2359
            }
2360

2361
            if(++s->mb_x >= s->mb_width){
2362
                s->mb_x=0;
2363
                if(++s->mb_y >= s->mb_height){
2364
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
2365
                        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);
2366

2367
                        return 0;
2368
                    }else{
2369
                        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);
2370

2371
                        return -1;
2372
                    }
2373
                }
2374
            }
2375

2376
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
2377
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
2378
                    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);
2379

2380
                    return 0;
2381
                }else{
2382
                    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);
2383

2384
                    return -1;
2385
                }
2386
            }
2387
        }
2388
        s->mb_x=0;
2389
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
2390
    }
2391
#endif
2392
    return -1; //not reached
2393
}
2394

    
2395
/**
2396
 * Call decode_slice() for each context.
2397
 *
2398
 * @param h h264 master context
2399
 * @param context_count number of contexts to execute
2400
 */
2401
static void execute_decode_slices(H264Context *h, int context_count){
2402
    MpegEncContext * const s = &h->s;
2403
    AVCodecContext * const avctx= s->avctx;
2404
    H264Context *hx;
2405
    int i;
2406

    
2407
    if (s->avctx->hwaccel)
2408
        return;
2409
    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2410
        return;
2411
    if(context_count == 1) {
2412
        decode_slice(avctx, &h);
2413
    } else {
2414
        for(i = 1; i < context_count; i++) {
2415
            hx = h->thread_context[i];
2416
            hx->s.error_recognition = avctx->error_recognition;
2417
            hx->s.error_count = 0;
2418
        }
2419

    
2420
        avctx->execute(avctx, (void *)decode_slice,
2421
                       h->thread_context, NULL, context_count, sizeof(void*));
2422

    
2423
        /* pull back stuff from slices to master context */
2424
        hx = h->thread_context[context_count - 1];
2425
        s->mb_x = hx->s.mb_x;
2426
        s->mb_y = hx->s.mb_y;
2427
        s->dropable = hx->s.dropable;
2428
        s->picture_structure = hx->s.picture_structure;
2429
        for(i = 1; i < context_count; i++)
2430
            h->s.error_count += h->thread_context[i]->s.error_count;
2431
    }
2432
}
2433

    
2434

    
2435
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
2436
    MpegEncContext * const s = &h->s;
2437
    AVCodecContext * const avctx= s->avctx;
2438
    int buf_index=0;
2439
    H264Context *hx; ///< thread context
2440
    int context_count = 0;
2441
    int next_avc= h->is_avc ? 0 : buf_size;
2442

    
2443
    h->max_contexts = avctx->thread_count;
2444
#if 0
2445
    int i;
2446
    for(i=0; i<50; i++){
2447
        av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
2448
    }
2449
#endif
2450
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
2451
        h->current_slice = 0;
2452
        if (!s->first_field)
2453
            s->current_picture_ptr= NULL;
2454
        ff_h264_reset_sei(h);
2455
    }
2456

    
2457
    for(;;){
2458
        int consumed;
2459
        int dst_length;
2460
        int bit_length;
2461
        const uint8_t *ptr;
2462
        int i, nalsize = 0;
2463
        int err;
2464

    
2465
        if(buf_index >= next_avc) {
2466
            if(buf_index >= buf_size) break;
2467
            nalsize = 0;
2468
            for(i = 0; i < h->nal_length_size; i++)
2469
                nalsize = (nalsize << 8) | buf[buf_index++];
2470
            if(nalsize <= 1 || nalsize > buf_size - buf_index){
2471
                if(nalsize == 1){
2472
                    buf_index++;
2473
                    continue;
2474
                }else{
2475
                    av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
2476
                    break;
2477
                }
2478
            }
2479
            next_avc= buf_index + nalsize;
2480
        } else {
2481
            // start code prefix search
2482
            for(; buf_index + 3 < next_avc; buf_index++){
2483
                // This should always succeed in the first iteration.
2484
                if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
2485
                    break;
2486
            }
2487

    
2488
            if(buf_index+3 >= buf_size) break;
2489

    
2490
            buf_index+=3;
2491
            if(buf_index >= next_avc) continue;
2492
        }
2493

    
2494
        hx = h->thread_context[context_count];
2495

    
2496
        ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
2497
        if (ptr==NULL || dst_length < 0){
2498
            return -1;
2499
        }
2500
        while(ptr[dst_length - 1] == 0 && dst_length > 0)
2501
            dst_length--;
2502
        bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
2503

    
2504
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
2505
            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);
2506
        }
2507

    
2508
        if (h->is_avc && (nalsize != consumed) && nalsize){
2509
            av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
2510
        }
2511

    
2512
        buf_index += consumed;
2513

    
2514
        if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
2515
           ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
2516
            continue;
2517

    
2518
      again:
2519
        err = 0;
2520
        switch(hx->nal_unit_type){
2521
        case NAL_IDR_SLICE:
2522
            if (h->nal_unit_type != NAL_IDR_SLICE) {
2523
                av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
2524
                return -1;
2525
            }
2526
            idr(h); //FIXME ensure we don't loose some frames if there is reordering
2527
        case NAL_SLICE:
2528
            init_get_bits(&hx->s.gb, ptr, bit_length);
2529
            hx->intra_gb_ptr=
2530
            hx->inter_gb_ptr= &hx->s.gb;
2531
            hx->s.data_partitioning = 0;
2532

    
2533
            if((err = decode_slice_header(hx, h)))
2534
               break;
2535

    
2536
            if (s->avctx->hwaccel && h->current_slice == 1) {
2537
                if (s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
2538
                    return -1;
2539
            }
2540

    
2541
            s->current_picture_ptr->key_frame |=
2542
                    (hx->nal_unit_type == NAL_IDR_SLICE) ||
2543
                    (h->sei_recovery_frame_cnt >= 0);
2544
            if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
2545
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2546
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2547
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2548
               && avctx->skip_frame < AVDISCARD_ALL){
2549
                if(avctx->hwaccel) {
2550
                    if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
2551
                        return -1;
2552
                }else
2553
                if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU){
2554
                    static const uint8_t start_code[] = {0x00, 0x00, 0x01};
2555
                    ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
2556
                    ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
2557
                }else
2558
                    context_count++;
2559
            }
2560
            break;
2561
        case NAL_DPA:
2562
            init_get_bits(&hx->s.gb, ptr, bit_length);
2563
            hx->intra_gb_ptr=
2564
            hx->inter_gb_ptr= NULL;
2565

    
2566
            if ((err = decode_slice_header(hx, h)) < 0)
2567
                break;
2568

    
2569
            hx->s.data_partitioning = 1;
2570

    
2571
            break;
2572
        case NAL_DPB:
2573
            init_get_bits(&hx->intra_gb, ptr, bit_length);
2574
            hx->intra_gb_ptr= &hx->intra_gb;
2575
            break;
2576
        case NAL_DPC:
2577
            init_get_bits(&hx->inter_gb, ptr, bit_length);
2578
            hx->inter_gb_ptr= &hx->inter_gb;
2579

    
2580
            if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
2581
               && s->context_initialized
2582
               && s->hurry_up < 5
2583
               && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
2584
               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
2585
               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
2586
               && avctx->skip_frame < AVDISCARD_ALL)
2587
                context_count++;
2588
            break;
2589
        case NAL_SEI:
2590
            init_get_bits(&s->gb, ptr, bit_length);
2591
            ff_h264_decode_sei(h);
2592
            break;
2593
        case NAL_SPS:
2594
            init_get_bits(&s->gb, ptr, bit_length);
2595
            ff_h264_decode_seq_parameter_set(h);
2596

    
2597
            if(s->flags& CODEC_FLAG_LOW_DELAY)
2598
                s->low_delay=1;
2599

    
2600
            if(avctx->has_b_frames < 2)
2601
                avctx->has_b_frames= !s->low_delay;
2602
            break;
2603
        case NAL_PPS:
2604
            init_get_bits(&s->gb, ptr, bit_length);
2605

    
2606
            ff_h264_decode_picture_parameter_set(h, bit_length);
2607

    
2608
            break;
2609
        case NAL_AUD:
2610
        case NAL_END_SEQUENCE:
2611
        case NAL_END_STREAM:
2612
        case NAL_FILLER_DATA:
2613
        case NAL_SPS_EXT:
2614
        case NAL_AUXILIARY_SLICE:
2615
            break;
2616
        default:
2617
            av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
2618
        }
2619

    
2620
        if(context_count == h->max_contexts) {
2621
            execute_decode_slices(h, context_count);
2622
            context_count = 0;
2623
        }
2624

    
2625
        if (err < 0)
2626
            av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
2627
        else if(err == 1) {
2628
            /* Slice could not be decoded in parallel mode, copy down
2629
             * NAL unit stuff to context 0 and restart. Note that
2630
             * rbsp_buffer is not transferred, but since we no longer
2631
             * run in parallel mode this should not be an issue. */
2632
            h->nal_unit_type = hx->nal_unit_type;
2633
            h->nal_ref_idc   = hx->nal_ref_idc;
2634
            hx = h;
2635
            goto again;
2636
        }
2637
    }
2638
    if(context_count)
2639
        execute_decode_slices(h, context_count);
2640
    return buf_index;
2641
}
2642

    
2643
/**
2644
 * returns the number of bytes consumed for building the current frame
2645
 */
2646
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
2647
        if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
2648
        if(pos+10>buf_size) pos=buf_size; // oops ;)
2649

    
2650
        return pos;
2651
}
2652

    
2653
static int decode_frame(AVCodecContext *avctx,
2654
                             void *data, int *data_size,
2655
                             AVPacket *avpkt)
2656
{
2657
    const uint8_t *buf = avpkt->data;
2658
    int buf_size = avpkt->size;
2659
    H264Context *h = avctx->priv_data;
2660
    MpegEncContext *s = &h->s;
2661
    AVFrame *pict = data;
2662
    int buf_index;
2663

    
2664
    s->flags= avctx->flags;
2665
    s->flags2= avctx->flags2;
2666

    
2667
   /* end of stream, output what is still in the buffers */
2668
    if (buf_size == 0) {
2669
        Picture *out;
2670
        int i, out_idx;
2671

    
2672
//FIXME factorize this with the output code below
2673
        out = h->delayed_pic[0];
2674
        out_idx = 0;
2675
        for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2676
            if(h->delayed_pic[i]->poc < out->poc){
2677
                out = h->delayed_pic[i];
2678
                out_idx = i;
2679
            }
2680

    
2681
        for(i=out_idx; h->delayed_pic[i]; i++)
2682
            h->delayed_pic[i] = h->delayed_pic[i+1];
2683

    
2684
        if(out){
2685
            *data_size = sizeof(AVFrame);
2686
            *pict= *(AVFrame*)out;
2687
        }
2688

    
2689
        return 0;
2690
    }
2691

    
2692
    if(h->is_avc && !h->got_avcC) {
2693
        int i, cnt, nalsize;
2694
        unsigned char *p = avctx->extradata;
2695
        if(avctx->extradata_size < 7) {
2696
            av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
2697
            return -1;
2698
        }
2699
        if(*p != 1) {
2700
            av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
2701
            return -1;
2702
        }
2703
        /* sps and pps in the avcC always have length coded with 2 bytes,
2704
           so put a fake nal_length_size = 2 while parsing them */
2705
        h->nal_length_size = 2;
2706
        // Decode sps from avcC
2707
        cnt = *(p+5) & 0x1f; // Number of sps
2708
        p += 6;
2709
        for (i = 0; i < cnt; i++) {
2710
            nalsize = AV_RB16(p) + 2;
2711
            if(decode_nal_units(h, p, nalsize) < 0) {
2712
                av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
2713
                return -1;
2714
            }
2715
            p += nalsize;
2716
        }
2717
        // Decode pps from avcC
2718
        cnt = *(p++); // Number of pps
2719
        for (i = 0; i < cnt; i++) {
2720
            nalsize = AV_RB16(p) + 2;
2721
            if(decode_nal_units(h, p, nalsize)  != nalsize) {
2722
                av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
2723
                return -1;
2724
            }
2725
            p += nalsize;
2726
        }
2727
        // Now store right nal length size, that will be use to parse all other nals
2728
        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
2729
        // Do not reparse avcC
2730
        h->got_avcC = 1;
2731
    }
2732

    
2733
    if(!h->got_avcC && !h->is_avc && s->avctx->extradata_size){
2734
        if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)
2735
            return -1;
2736
        h->got_avcC = 1;
2737
    }
2738

    
2739
    buf_index=decode_nal_units(h, buf, buf_size);
2740
    if(buf_index < 0)
2741
        return -1;
2742

    
2743
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
2744
        if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
2745
        av_log(avctx, AV_LOG_ERROR, "no frame!\n");
2746
        return -1;
2747
    }
2748

    
2749
    if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
2750
        Picture *out = s->current_picture_ptr;
2751
        Picture *cur = s->current_picture_ptr;
2752
        int i, pics, out_of_order, out_idx;
2753

    
2754
        field_end(h);
2755

    
2756
        if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
2757
            /* Wait for second field. */
2758
            *data_size = 0;
2759

    
2760
        } else {
2761
            cur->interlaced_frame = 0;
2762
            cur->repeat_pict = 0;
2763

    
2764
            /* Signal interlacing information externally. */
2765
            /* Prioritize picture timing SEI information over used decoding process if it exists. */
2766

    
2767
            if(h->sps.pic_struct_present_flag){
2768
                switch (h->sei_pic_struct)
2769
                {
2770
                case SEI_PIC_STRUCT_FRAME:
2771
                    break;
2772
                case SEI_PIC_STRUCT_TOP_FIELD:
2773
                case SEI_PIC_STRUCT_BOTTOM_FIELD:
2774
                    cur->interlaced_frame = 1;
2775
                    break;
2776
                case SEI_PIC_STRUCT_TOP_BOTTOM:
2777
                case SEI_PIC_STRUCT_BOTTOM_TOP:
2778
                    if (FIELD_OR_MBAFF_PICTURE)
2779
                        cur->interlaced_frame = 1;
2780
                    else
2781
                        // try to flag soft telecine progressive
2782
                        cur->interlaced_frame = h->prev_interlaced_frame;
2783
                    break;
2784
                case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2785
                case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2786
                    // Signal the possibility of telecined film externally (pic_struct 5,6)
2787
                    // From these hints, let the applications decide if they apply deinterlacing.
2788
                    cur->repeat_pict = 1;
2789
                    break;
2790
                case SEI_PIC_STRUCT_FRAME_DOUBLING:
2791
                    // Force progressive here, as doubling interlaced frame is a bad idea.
2792
                    cur->repeat_pict = 2;
2793
                    break;
2794
                case SEI_PIC_STRUCT_FRAME_TRIPLING:
2795
                    cur->repeat_pict = 4;
2796
                    break;
2797
                }
2798

    
2799
                if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2800
                    cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
2801
            }else{
2802
                /* Derive interlacing flag from used decoding process. */
2803
                cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
2804
            }
2805
            h->prev_interlaced_frame = cur->interlaced_frame;
2806

    
2807
            if (cur->field_poc[0] != cur->field_poc[1]){
2808
                /* Derive top_field_first from field pocs. */
2809
                cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
2810
            }else{
2811
                if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
2812
                    /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
2813
                    if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
2814
                      || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2815
                        cur->top_field_first = 1;
2816
                    else
2817
                        cur->top_field_first = 0;
2818
                }else{
2819
                    /* Most likely progressive */
2820
                    cur->top_field_first = 0;
2821
                }
2822
            }
2823

    
2824
        //FIXME do something with unavailable reference frames
2825

    
2826
            /* Sort B-frames into display order */
2827

    
2828
            if(h->sps.bitstream_restriction_flag
2829
               && s->avctx->has_b_frames < h->sps.num_reorder_frames){
2830
                s->avctx->has_b_frames = h->sps.num_reorder_frames;
2831
                s->low_delay = 0;
2832
            }
2833

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

    
2840
            pics = 0;
2841
            while(h->delayed_pic[pics]) pics++;
2842

    
2843
            assert(pics <= MAX_DELAYED_PIC_COUNT);
2844

    
2845
            h->delayed_pic[pics++] = cur;
2846
            if(cur->reference == 0)
2847
                cur->reference = DELAYED_PIC_REF;
2848

    
2849
            out = h->delayed_pic[0];
2850
            out_idx = 0;
2851
            for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
2852
                if(h->delayed_pic[i]->poc < out->poc){
2853
                    out = h->delayed_pic[i];
2854
                    out_idx = i;
2855
                }
2856
            if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
2857
                h->outputed_poc= INT_MIN;
2858
            out_of_order = out->poc < h->outputed_poc;
2859

    
2860
            if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
2861
                { }
2862
            else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
2863
               || (s->low_delay &&
2864
                ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
2865
                 || cur->pict_type == FF_B_TYPE)))
2866
            {
2867
                s->low_delay = 0;
2868
                s->avctx->has_b_frames++;
2869
            }
2870

    
2871
            if(out_of_order || pics > s->avctx->has_b_frames){
2872
                out->reference &= ~DELAYED_PIC_REF;
2873
                for(i=out_idx; h->delayed_pic[i]; i++)
2874
                    h->delayed_pic[i] = h->delayed_pic[i+1];
2875
            }
2876
            if(!out_of_order && pics > s->avctx->has_b_frames){
2877
                *data_size = sizeof(AVFrame);
2878

    
2879
                if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
2880
                    h->outputed_poc = INT_MIN;
2881
                } else
2882
                    h->outputed_poc = out->poc;
2883
                *pict= *(AVFrame*)out;
2884
            }else{
2885
                av_log(avctx, AV_LOG_DEBUG, "no picture\n");
2886
            }
2887
        }
2888
    }
2889

    
2890
    assert(pict->data[0] || !*data_size);
2891
    ff_print_debug_info(s, pict);
2892
//printf("out %d\n", (int)pict->data[0]);
2893

    
2894
    return get_consumed_bytes(s, buf_index, buf_size);
2895
}
2896
#if 0
2897
static inline void fill_mb_avail(H264Context *h){
2898
    MpegEncContext * const s = &h->s;
2899
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2900

2901
    if(s->mb_y){
2902
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
2903
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
2904
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
2905
    }else{
2906
        h->mb_avail[0]=
2907
        h->mb_avail[1]=
2908
        h->mb_avail[2]= 0;
2909
    }
2910
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
2911
    h->mb_avail[4]= 1; //FIXME move out
2912
    h->mb_avail[5]= 0; //FIXME move out
2913
}
2914
#endif
2915

    
2916
#ifdef TEST
2917
#undef printf
2918
#undef random
2919
#define COUNT 8000
2920
#define SIZE (COUNT*40)
2921
int main(void){
2922
    int i;
2923
    uint8_t temp[SIZE];
2924
    PutBitContext pb;
2925
    GetBitContext gb;
2926
//    int int_temp[10000];
2927
    DSPContext dsp;
2928
    AVCodecContext avctx;
2929

    
2930
    dsputil_init(&dsp, &avctx);
2931

    
2932
    init_put_bits(&pb, temp, SIZE);
2933
    printf("testing unsigned exp golomb\n");
2934
    for(i=0; i<COUNT; i++){
2935
        START_TIMER
2936
        set_ue_golomb(&pb, i);
2937
        STOP_TIMER("set_ue_golomb");
2938
    }
2939
    flush_put_bits(&pb);
2940

    
2941
    init_get_bits(&gb, temp, 8*SIZE);
2942
    for(i=0; i<COUNT; i++){
2943
        int j, s;
2944

    
2945
        s= show_bits(&gb, 24);
2946

    
2947
        START_TIMER
2948
        j= get_ue_golomb(&gb);
2949
        if(j != i){
2950
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2951
//            return -1;
2952
        }
2953
        STOP_TIMER("get_ue_golomb");
2954
    }
2955

    
2956

    
2957
    init_put_bits(&pb, temp, SIZE);
2958
    printf("testing signed exp golomb\n");
2959
    for(i=0; i<COUNT; i++){
2960
        START_TIMER
2961
        set_se_golomb(&pb, i - COUNT/2);
2962
        STOP_TIMER("set_se_golomb");
2963
    }
2964
    flush_put_bits(&pb);
2965

    
2966
    init_get_bits(&gb, temp, 8*SIZE);
2967
    for(i=0; i<COUNT; i++){
2968
        int j, s;
2969

    
2970
        s= show_bits(&gb, 24);
2971

    
2972
        START_TIMER
2973
        j= get_se_golomb(&gb);
2974
        if(j != i - COUNT/2){
2975
            printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
2976
//            return -1;
2977
        }
2978
        STOP_TIMER("get_se_golomb");
2979
    }
2980

    
2981
#if 0
2982
    printf("testing 4x4 (I)DCT\n");
2983

2984
    DCTELEM block[16];
2985
    uint8_t src[16], ref[16];
2986
    uint64_t error= 0, max_error=0;
2987

2988
    for(i=0; i<COUNT; i++){
2989
        int j;
2990
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
2991
        for(j=0; j<16; j++){
2992
            ref[j]= random()%255;
2993
            src[j]= random()%255;
2994
        }
2995

2996
        h264_diff_dct_c(block, src, ref, 4);
2997

2998
        //normalize
2999
        for(j=0; j<16; j++){
3000
//            printf("%d ", block[j]);
3001
            block[j]= block[j]*4;
3002
            if(j&1) block[j]= (block[j]*4 + 2)/5;
3003
            if(j&4) block[j]= (block[j]*4 + 2)/5;
3004
        }
3005
//        printf("\n");
3006

3007
        s->dsp.h264_idct_add(ref, block, 4);
3008
/*        for(j=0; j<16; j++){
3009
            printf("%d ", ref[j]);
3010
        }
3011
        printf("\n");*/
3012

3013
        for(j=0; j<16; j++){
3014
            int diff= FFABS(src[j] - ref[j]);
3015

3016
            error+= diff*diff;
3017
            max_error= FFMAX(max_error, diff);
3018
        }
3019
    }
3020
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
3021
    printf("testing quantizer\n");
3022
    for(qp=0; qp<52; qp++){
3023
        for(i=0; i<16; i++)
3024
            src1_block[i]= src2_block[i]= random()%255;
3025

3026
    }
3027
    printf("Testing NAL layer\n");
3028

3029
    uint8_t bitstream[COUNT];
3030
    uint8_t nal[COUNT*2];
3031
    H264Context h;
3032
    memset(&h, 0, sizeof(H264Context));
3033

3034
    for(i=0; i<COUNT; i++){
3035
        int zeros= i;
3036
        int nal_length;
3037
        int consumed;
3038
        int out_length;
3039
        uint8_t *out;
3040
        int j;
3041

3042
        for(j=0; j<COUNT; j++){
3043
            bitstream[j]= (random() % 255) + 1;
3044
        }
3045

3046
        for(j=0; j<zeros; j++){
3047
            int pos= random() % COUNT;
3048
            while(bitstream[pos] == 0){
3049
                pos++;
3050
                pos %= COUNT;
3051
            }
3052
            bitstream[pos]=0;
3053
        }
3054

3055
        START_TIMER
3056

3057
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
3058
        if(nal_length<0){
3059
            printf("encoding failed\n");
3060
            return -1;
3061
        }
3062

3063
        out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
3064

3065
        STOP_TIMER("NAL")
3066

3067
        if(out_length != COUNT){
3068
            printf("incorrect length %d %d\n", out_length, COUNT);
3069
            return -1;
3070
        }
3071

3072
        if(consumed != nal_length){
3073
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
3074
            return -1;
3075
        }
3076

3077
        if(memcmp(bitstream, out, COUNT)){
3078
            printf("mismatch\n");
3079
            return -1;
3080
        }
3081
    }
3082
#endif
3083

    
3084
    printf("Testing RBSP\n");
3085

    
3086

    
3087
    return 0;
3088
}
3089
#endif /* TEST */
3090

    
3091

    
3092
av_cold void ff_h264_free_context(H264Context *h)
3093
{
3094
    int i;
3095

    
3096
    free_tables(h); //FIXME cleanup init stuff perhaps
3097

    
3098
    for(i = 0; i < MAX_SPS_COUNT; i++)
3099
        av_freep(h->sps_buffers + i);
3100

    
3101
    for(i = 0; i < MAX_PPS_COUNT; i++)
3102
        av_freep(h->pps_buffers + i);
3103
}
3104

    
3105
av_cold int ff_h264_decode_end(AVCodecContext *avctx)
3106
{
3107
    H264Context *h = avctx->priv_data;
3108
    MpegEncContext *s = &h->s;
3109

    
3110
    ff_h264_free_context(h);
3111

    
3112
    MPV_common_end(s);
3113

    
3114
//    memset(h, 0, sizeof(H264Context));
3115

    
3116
    return 0;
3117
}
3118

    
3119

    
3120
AVCodec h264_decoder = {
3121
    "h264",
3122
    CODEC_TYPE_VIDEO,
3123
    CODEC_ID_H264,
3124
    sizeof(H264Context),
3125
    ff_h264_decode_init,
3126
    NULL,
3127
    ff_h264_decode_end,
3128
    decode_frame,
3129
    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
3130
    .flush= flush_dpb,
3131
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
3132
    .pix_fmts= ff_hwaccel_pixfmt_list_420,
3133
};
3134

    
3135
#if CONFIG_H264_VDPAU_DECODER
3136
AVCodec h264_vdpau_decoder = {
3137
    "h264_vdpau",
3138
    CODEC_TYPE_VIDEO,
3139
    CODEC_ID_H264,
3140
    sizeof(H264Context),
3141
    ff_h264_decode_init,
3142
    NULL,
3143
    ff_h264_decode_end,
3144
    decode_frame,
3145
    CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
3146
    .flush= flush_dpb,
3147
    .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
3148
    .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
3149
};
3150
#endif