Reorder nut specific codec tags and add a comment for marking them as
[ffmpeg.git] / libavcodec / mpegvideo.c
1 /*
2  * The simplest mpeg encoder (well, it was the simplest!)
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of FFmpeg.
9  *
10  * FFmpeg is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * FFmpeg is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with FFmpeg; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24
25 /**
26  * @file
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/intmath.h"
31 #include "avcodec.h"
32 #include "dsputil.h"
33 #include "mpegvideo.h"
34 #include "mpegvideo_common.h"
35 #include "mjpegenc.h"
36 #include "msmpeg4.h"
37 #include "faandct.h"
38 #include "xvmc_internal.h"
39 #include <limits.h>
40
41 //#undef NDEBUG
42 //#include <assert.h>
43
44 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
45                                    DCTELEM *block, int n, int qscale);
46 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
47                                    DCTELEM *block, int n, int qscale);
48 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
49                                    DCTELEM *block, int n, int qscale);
50 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
51                                    DCTELEM *block, int n, int qscale);
52 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
53                                    DCTELEM *block, int n, int qscale);
54 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
55                                   DCTELEM *block, int n, int qscale);
56 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
57                                   DCTELEM *block, int n, int qscale);
58
59
60 /* enable all paranoid tests for rounding, overflows, etc... */
61 //#define PARANOID
62
63 //#define DEBUG
64
65
66 static const uint8_t ff_default_chroma_qscale_table[32]={
67 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
68     0, 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
69 };
70
71 const uint8_t ff_mpeg1_dc_scale_table[128]={
72 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
73     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
74     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
75     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
76     8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
77 };
78
79 static const uint8_t mpeg2_dc_scale_table1[128]={
80 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
81     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
82     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
83     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
84     4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
85 };
86
87 static const uint8_t mpeg2_dc_scale_table2[128]={
88 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
89     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
90     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
91     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
92     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
93 };
94
95 static const uint8_t mpeg2_dc_scale_table3[128]={
96 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
97     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
98     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
99     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
100     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
101 };
102
103 const uint8_t * const ff_mpeg2_dc_scale_table[4]={
104     ff_mpeg1_dc_scale_table,
105     mpeg2_dc_scale_table1,
106     mpeg2_dc_scale_table2,
107     mpeg2_dc_scale_table3,
108 };
109
110 const enum PixelFormat ff_pixfmt_list_420[] = {
111     PIX_FMT_YUV420P,
112     PIX_FMT_NONE
113 };
114
115 const enum PixelFormat ff_hwaccel_pixfmt_list_420[] = {
116     PIX_FMT_DXVA2_VLD,
117     PIX_FMT_VAAPI_VLD,
118     PIX_FMT_YUV420P,
119     PIX_FMT_NONE
120 };
121
122 const uint8_t *ff_find_start_code(const uint8_t * restrict p, const uint8_t *end, uint32_t * restrict state){
123     int i;
124
125     assert(p<=end);
126     if(p>=end)
127         return end;
128
129     for(i=0; i<3; i++){
130         uint32_t tmp= *state << 8;
131         *state= tmp + *(p++);
132         if(tmp == 0x100 || p==end)
133             return p;
134     }
135
136     while(p<end){
137         if     (p[-1] > 1      ) p+= 3;
138         else if(p[-2]          ) p+= 2;
139         else if(p[-3]|(p[-1]-1)) p++;
140         else{
141             p++;
142             break;
143         }
144     }
145
146     p= FFMIN(p, end)-4;
147     *state= AV_RB32(p);
148
149     return p+4;
150 }
151
152 /* init common dct for both encoder and decoder */
153 av_cold int ff_dct_common_init(MpegEncContext *s)
154 {
155     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
156     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
157     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
158     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
159     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
160     if(s->flags & CODEC_FLAG_BITEXACT)
161         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
162     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
163
164 #if   HAVE_MMX
165     MPV_common_init_mmx(s);
166 #elif ARCH_ALPHA
167     MPV_common_init_axp(s);
168 #elif CONFIG_MLIB
169     MPV_common_init_mlib(s);
170 #elif HAVE_MMI
171     MPV_common_init_mmi(s);
172 #elif ARCH_ARM
173     MPV_common_init_arm(s);
174 #elif HAVE_ALTIVEC
175     MPV_common_init_altivec(s);
176 #elif ARCH_BFIN
177     MPV_common_init_bfin(s);
178 #endif
179
180     /* load & permutate scantables
181        note: only wmv uses different ones
182     */
183     if(s->alternate_scan){
184         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_alternate_vertical_scan);
185         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_alternate_vertical_scan);
186     }else{
187         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_zigzag_direct);
188         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_zigzag_direct);
189     }
190     ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
191     ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
192
193     return 0;
194 }
195
196 void ff_copy_picture(Picture *dst, Picture *src){
197     *dst = *src;
198     dst->type= FF_BUFFER_TYPE_COPY;
199 }
200
201 /**
202  * Releases a frame buffer
203  */
204 static void free_frame_buffer(MpegEncContext *s, Picture *pic)
205 {
206     s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
207     av_freep(&pic->hwaccel_picture_private);
208 }
209
210 /**
211  * Allocates a frame buffer
212  */
213 static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
214 {
215     int r;
216
217     if (s->avctx->hwaccel) {
218         assert(!pic->hwaccel_picture_private);
219         if (s->avctx->hwaccel->priv_data_size) {
220             pic->hwaccel_picture_private = av_mallocz(s->avctx->hwaccel->priv_data_size);
221             if (!pic->hwaccel_picture_private) {
222                 av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
223                 return -1;
224             }
225         }
226     }
227
228     r = s->avctx->get_buffer(s->avctx, (AVFrame*)pic);
229
230     if (r<0 || !pic->age || !pic->type || !pic->data[0]) {
231         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
232         av_freep(&pic->hwaccel_picture_private);
233         return -1;
234     }
235
236     if (s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])) {
237         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n");
238         free_frame_buffer(s, pic);
239         return -1;
240     }
241
242     if (pic->linesize[1] != pic->linesize[2]) {
243         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n");
244         free_frame_buffer(s, pic);
245         return -1;
246     }
247
248     return 0;
249 }
250
251 /**
252  * allocates a Picture
253  * The pixels are allocated/set by calling get_buffer() if shared=0
254  */
255 int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared){
256     const int big_mb_num= s->mb_stride*(s->mb_height+1) + 1; //the +1 is needed so memset(,,stride*height) does not sig11
257     const int mb_array_size= s->mb_stride*s->mb_height;
258     const int b8_array_size= s->b8_stride*s->mb_height*2;
259     const int b4_array_size= s->b4_stride*s->mb_height*4;
260     int i;
261     int r= -1;
262
263     if(shared){
264         assert(pic->data[0]);
265         assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED);
266         pic->type= FF_BUFFER_TYPE_SHARED;
267     }else{
268         assert(!pic->data[0]);
269
270         if (alloc_frame_buffer(s, pic) < 0)
271             return -1;
272
273         s->linesize  = pic->linesize[0];
274         s->uvlinesize= pic->linesize[1];
275     }
276
277     if(pic->qscale_table==NULL){
278         if (s->encoding) {
279             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_var   , mb_array_size * sizeof(int16_t)  , fail)
280             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mc_mb_var, mb_array_size * sizeof(int16_t)  , fail)
281             FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_mean  , mb_array_size * sizeof(int8_t )  , fail)
282         }
283
284         FF_ALLOCZ_OR_GOTO(s->avctx, pic->mbskip_table , mb_array_size * sizeof(uint8_t)+2, fail) //the +2 is for the slice end check
285         FF_ALLOCZ_OR_GOTO(s->avctx, pic->qscale_table , mb_array_size * sizeof(uint8_t)  , fail)
286         FF_ALLOCZ_OR_GOTO(s->avctx, pic->mb_type_base , (big_mb_num + s->mb_stride) * sizeof(uint32_t), fail)
287         pic->mb_type= pic->mb_type_base + 2*s->mb_stride+1;
288         if(s->out_format == FMT_H264){
289             for(i=0; i<2; i++){
290                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->motion_val_base[i], 2 * (b4_array_size+4)  * sizeof(int16_t), fail)
291                 pic->motion_val[i]= pic->motion_val_base[i]+4;
292                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->ref_index[i], 4*mb_array_size * sizeof(uint8_t), fail)
293             }
294             pic->motion_subsample_log2= 2;
295         }else if(s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_MV) || (s->avctx->debug_mv)){
296             for(i=0; i<2; i++){
297                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->motion_val_base[i], 2 * (b8_array_size+4) * sizeof(int16_t), fail)
298                 pic->motion_val[i]= pic->motion_val_base[i]+4;
299                 FF_ALLOCZ_OR_GOTO(s->avctx, pic->ref_index[i], 4*mb_array_size * sizeof(uint8_t), fail)
300             }
301             pic->motion_subsample_log2= 3;
302         }
303         if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
304             FF_ALLOCZ_OR_GOTO(s->avctx, pic->dct_coeff, 64 * mb_array_size * sizeof(DCTELEM)*6, fail)
305         }
306         pic->qstride= s->mb_stride;
307         FF_ALLOCZ_OR_GOTO(s->avctx, pic->pan_scan , 1 * sizeof(AVPanScan), fail)
308     }
309
310     /* It might be nicer if the application would keep track of these
311      * but it would require an API change. */
312     memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1);
313     s->prev_pict_types[0]= s->dropable ? FF_B_TYPE : s->pict_type;
314     if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == FF_B_TYPE)
315         pic->age= INT_MAX; // Skipped MBs in B-frames are quite rare in MPEG-1/2 and it is a bit tricky to skip them anyway.
316
317     return 0;
318 fail: //for the FF_ALLOCZ_OR_GOTO macro
319     if(r>=0)
320         free_frame_buffer(s, pic);
321     return -1;
322 }
323
324 /**
325  * deallocates a picture
326  */
327 static void free_picture(MpegEncContext *s, Picture *pic){
328     int i;
329
330     if(pic->data[0] && pic->type!=FF_BUFFER_TYPE_SHARED){
331         free_frame_buffer(s, pic);
332     }
333
334     av_freep(&pic->mb_var);
335     av_freep(&pic->mc_mb_var);
336     av_freep(&pic->mb_mean);
337     av_freep(&pic->mbskip_table);
338     av_freep(&pic->qscale_table);
339     av_freep(&pic->mb_type_base);
340     av_freep(&pic->dct_coeff);
341     av_freep(&pic->pan_scan);
342     pic->mb_type= NULL;
343     for(i=0; i<2; i++){
344         av_freep(&pic->motion_val_base[i]);
345         av_freep(&pic->ref_index[i]);
346     }
347
348     if(pic->type == FF_BUFFER_TYPE_SHARED){
349         for(i=0; i<4; i++){
350             pic->base[i]=
351             pic->data[i]= NULL;
352         }
353         pic->type= 0;
354     }
355 }
356
357 static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base){
358     int i;
359
360     // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264)
361     FF_ALLOCZ_OR_GOTO(s->avctx, s->allocated_edge_emu_buffer, (s->width+64)*2*21*2, fail); //(width + edge + align)*interlaced*MBsize*tolerance
362     s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;
363
364      //FIXME should be linesize instead of s->width*2 but that is not known before get_buffer()
365     FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad,  (s->width+64)*4*16*2*sizeof(uint8_t), fail)
366     s->me.temp=         s->me.scratchpad;
367     s->rd_scratchpad=   s->me.scratchpad;
368     s->b_scratchpad=    s->me.scratchpad;
369     s->obmc_scratchpad= s->me.scratchpad + 16;
370     if (s->encoding) {
371         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map      , ME_MAP_SIZE*sizeof(uint32_t), fail)
372         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map, ME_MAP_SIZE*sizeof(uint32_t), fail)
373         if(s->avctx->noise_reduction){
374             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum, 2 * 64 * sizeof(int), fail)
375         }
376     }
377     FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64*12*2 * sizeof(DCTELEM), fail)
378     s->block= s->blocks[0];
379
380     for(i=0;i<12;i++){
381         s->pblocks[i] = &s->block[i];
382     }
383     return 0;
384 fail:
385     return -1; //free() through MPV_common_end()
386 }
387
388 static void free_duplicate_context(MpegEncContext *s){
389     if(s==NULL) return;
390
391     av_freep(&s->allocated_edge_emu_buffer); s->edge_emu_buffer= NULL;
392     av_freep(&s->me.scratchpad);
393     s->me.temp=
394     s->rd_scratchpad=
395     s->b_scratchpad=
396     s->obmc_scratchpad= NULL;
397
398     av_freep(&s->dct_error_sum);
399     av_freep(&s->me.map);
400     av_freep(&s->me.score_map);
401     av_freep(&s->blocks);
402     s->block= NULL;
403 }
404
405 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src){
406 #define COPY(a) bak->a= src->a
407     COPY(allocated_edge_emu_buffer);
408     COPY(edge_emu_buffer);
409     COPY(me.scratchpad);
410     COPY(me.temp);
411     COPY(rd_scratchpad);
412     COPY(b_scratchpad);
413     COPY(obmc_scratchpad);
414     COPY(me.map);
415     COPY(me.score_map);
416     COPY(blocks);
417     COPY(block);
418     COPY(start_mb_y);
419     COPY(end_mb_y);
420     COPY(me.map_generation);
421     COPY(pb);
422     COPY(dct_error_sum);
423     COPY(dct_count[0]);
424     COPY(dct_count[1]);
425 #undef COPY
426 }
427
428 void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src){
429     MpegEncContext bak;
430     int i;
431     //FIXME copy only needed parts
432 //START_TIMER
433     backup_duplicate_context(&bak, dst);
434     memcpy(dst, src, sizeof(MpegEncContext));
435     backup_duplicate_context(dst, &bak);
436     for(i=0;i<12;i++){
437         dst->pblocks[i] = &dst->block[i];
438     }
439 //STOP_TIMER("update_duplicate_context") //about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
440 }
441
442 /**
443  * sets the given MpegEncContext to common defaults (same for encoding and decoding).
444  * the changed fields will not depend upon the prior state of the MpegEncContext.
445  */
446 void MPV_common_defaults(MpegEncContext *s){
447     s->y_dc_scale_table=
448     s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
449     s->chroma_qscale_table= ff_default_chroma_qscale_table;
450     s->progressive_frame= 1;
451     s->progressive_sequence= 1;
452     s->picture_structure= PICT_FRAME;
453
454     s->coded_picture_number = 0;
455     s->picture_number = 0;
456     s->input_picture_number = 0;
457
458     s->picture_in_gop_number = 0;
459
460     s->f_code = 1;
461     s->b_code = 1;
462 }
463
464 /**
465  * sets the given MpegEncContext to defaults for decoding.
466  * the changed fields will not depend upon the prior state of the MpegEncContext.
467  */
468 void MPV_decode_defaults(MpegEncContext *s){
469     MPV_common_defaults(s);
470 }
471
472 /**
473  * init common structure for both encoder and decoder.
474  * this assumes that some variables like width/height are already set
475  */
476 av_cold int MPV_common_init(MpegEncContext *s)
477 {
478     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y, threads;
479
480     if(s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
481         s->mb_height = (s->height + 31) / 32 * 2;
482     else
483         s->mb_height = (s->height + 15) / 16;
484
485     if(s->avctx->pix_fmt == PIX_FMT_NONE){
486         av_log(s->avctx, AV_LOG_ERROR, "decoding to PIX_FMT_NONE is not supported.\n");
487         return -1;
488     }
489
490     if(s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height)){
491         av_log(s->avctx, AV_LOG_ERROR, "too many threads\n");
492         return -1;
493     }
494
495     if((s->width || s->height) && avcodec_check_dimensions(s->avctx, s->width, s->height))
496         return -1;
497
498     dsputil_init(&s->dsp, s->avctx);
499     ff_dct_common_init(s);
500
501     s->flags= s->avctx->flags;
502     s->flags2= s->avctx->flags2;
503
504     s->mb_width  = (s->width  + 15) / 16;
505     s->mb_stride = s->mb_width + 1;
506     s->b8_stride = s->mb_width*2 + 1;
507     s->b4_stride = s->mb_width*4 + 1;
508     mb_array_size= s->mb_height * s->mb_stride;
509     mv_table_size= (s->mb_height+2) * s->mb_stride + 1;
510
511     /* set chroma shifts */
512     avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
513                                                     &(s->chroma_y_shift) );
514
515     /* set default edge pos, will be overriden in decode_header if needed */
516     s->h_edge_pos= s->mb_width*16;
517     s->v_edge_pos= s->mb_height*16;
518
519     s->mb_num = s->mb_width * s->mb_height;
520
521     s->block_wrap[0]=
522     s->block_wrap[1]=
523     s->block_wrap[2]=
524     s->block_wrap[3]= s->b8_stride;
525     s->block_wrap[4]=
526     s->block_wrap[5]= s->mb_stride;
527
528     y_size = s->b8_stride * (2 * s->mb_height + 1);
529     c_size = s->mb_stride * (s->mb_height + 1);
530     yc_size = y_size + 2 * c_size;
531
532     /* convert fourcc to upper case */
533     s->codec_tag=          toupper( s->avctx->codec_tag     &0xFF)
534                         + (toupper((s->avctx->codec_tag>>8 )&0xFF)<<8 )
535                         + (toupper((s->avctx->codec_tag>>16)&0xFF)<<16)
536                         + (toupper((s->avctx->codec_tag>>24)&0xFF)<<24);
537
538     s->stream_codec_tag=          toupper( s->avctx->stream_codec_tag     &0xFF)
539                                + (toupper((s->avctx->stream_codec_tag>>8 )&0xFF)<<8 )
540                                + (toupper((s->avctx->stream_codec_tag>>16)&0xFF)<<16)
541                                + (toupper((s->avctx->stream_codec_tag>>24)&0xFF)<<24);
542
543     s->avctx->coded_frame= (AVFrame*)&s->current_picture;
544
545     FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num+1)*sizeof(int), fail) //error ressilience code looks cleaner with this
546     for(y=0; y<s->mb_height; y++){
547         for(x=0; x<s->mb_width; x++){
548             s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride;
549         }
550     }
551     s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed?
552
553     if (s->encoding) {
554         /* Allocate MV tables */
555         FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base            , mv_table_size * 2 * sizeof(int16_t), fail)
556         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base       , mv_table_size * 2 * sizeof(int16_t), fail)
557         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base       , mv_table_size * 2 * sizeof(int16_t), fail)
558         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
559         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
560         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base     , mv_table_size * 2 * sizeof(int16_t), fail)
561         s->p_mv_table           = s->p_mv_table_base            + s->mb_stride + 1;
562         s->b_forw_mv_table      = s->b_forw_mv_table_base       + s->mb_stride + 1;
563         s->b_back_mv_table      = s->b_back_mv_table_base       + s->mb_stride + 1;
564         s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
565         s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1;
566         s->b_direct_mv_table    = s->b_direct_mv_table_base     + s->mb_stride + 1;
567
568         if(s->msmpeg4_version){
569             FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int), fail);
570         }
571         FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
572
573         /* Allocate MB type table */
574         FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type  , mb_array_size * sizeof(uint16_t), fail) //needed for encoding
575
576         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
577
578         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix  , 64*32   * sizeof(int), fail)
579         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix  , 64*32   * sizeof(int), fail)
580         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t), fail)
581         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t), fail)
582         FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
583         FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
584
585         if(s->avctx->noise_reduction){
586             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset, 2 * 64 * sizeof(uint16_t), fail)
587         }
588     }
589     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture, MAX_PICTURE_COUNT * sizeof(Picture), fail)
590     for(i = 0; i < MAX_PICTURE_COUNT; i++) {
591         avcodec_get_frame_defaults((AVFrame *)&s->picture[i]);
592     }
593
594     FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table, mb_array_size*sizeof(uint8_t), fail)
595
596     if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){
597         /* interlaced direct mode decoding tables */
598             for(i=0; i<2; i++){
599                 int j, k;
600                 for(j=0; j<2; j++){
601                     for(k=0; k<2; k++){
602                         FF_ALLOCZ_OR_GOTO(s->avctx,    s->b_field_mv_table_base[i][j][k], mv_table_size * 2 * sizeof(int16_t), fail)
603                         s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] + s->mb_stride + 1;
604                     }
605                     FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
606                     FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
607                     s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]+ s->mb_stride + 1;
608                 }
609                 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
610             }
611     }
612     if (s->out_format == FMT_H263) {
613         /* ac values */
614         FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base, yc_size * sizeof(int16_t) * 16, fail);
615         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
616         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
617         s->ac_val[2] = s->ac_val[1] + c_size;
618
619         /* cbp values */
620         FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size, fail);
621         s->coded_block= s->coded_block_base + s->b8_stride + 1;
622
623         /* cbp, ac_pred, pred_dir */
624         FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail)
625         FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail)
626     }
627
628     if (s->h263_pred || s->h263_plus || !s->encoding) {
629         /* dc values */
630         //MN: we need these for error resilience of intra-frames
631         FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
632         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
633         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
634         s->dc_val[2] = s->dc_val[1] + c_size;
635         for(i=0;i<yc_size;i++)
636             s->dc_val_base[i] = 1024;
637     }
638
639     /* which mb is a intra block */
640     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
641     memset(s->mbintra_table, 1, mb_array_size);
642
643     /* init macroblock skip table */
644     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size+2, fail);
645     //Note the +1 is for a quicker mpeg4 slice_end detection
646     FF_ALLOCZ_OR_GOTO(s->avctx, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE, fail);
647
648     s->parse_context.state= -1;
649     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
650        s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
651        s->visualization_buffer[1] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
652        s->visualization_buffer[2] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
653     }
654
655     s->context_initialized = 1;
656
657     s->thread_context[0]= s;
658     threads = s->avctx->thread_count;
659
660     for(i=1; i<threads; i++){
661         s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
662         memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
663     }
664
665     for(i=0; i<threads; i++){
666         if(init_duplicate_context(s->thread_context[i], s) < 0)
667            goto fail;
668         s->thread_context[i]->start_mb_y= (s->mb_height*(i  ) + s->avctx->thread_count/2) / s->avctx->thread_count;
669         s->thread_context[i]->end_mb_y  = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count;
670     }
671
672     return 0;
673  fail:
674     MPV_common_end(s);
675     return -1;
676 }
677
678 /* init common structure for both encoder and decoder */
679 void MPV_common_end(MpegEncContext *s)
680 {
681     int i, j, k;
682
683     for(i=0; i<s->avctx->thread_count; i++){
684         free_duplicate_context(s->thread_context[i]);
685     }
686     for(i=1; i<s->avctx->thread_count; i++){
687         av_freep(&s->thread_context[i]);
688     }
689
690     av_freep(&s->parse_context.buffer);
691     s->parse_context.buffer_size=0;
692
693     av_freep(&s->mb_type);
694     av_freep(&s->p_mv_table_base);
695     av_freep(&s->b_forw_mv_table_base);
696     av_freep(&s->b_back_mv_table_base);
697     av_freep(&s->b_bidir_forw_mv_table_base);
698     av_freep(&s->b_bidir_back_mv_table_base);
699     av_freep(&s->b_direct_mv_table_base);
700     s->p_mv_table= NULL;
701     s->b_forw_mv_table= NULL;
702     s->b_back_mv_table= NULL;
703     s->b_bidir_forw_mv_table= NULL;
704     s->b_bidir_back_mv_table= NULL;
705     s->b_direct_mv_table= NULL;
706     for(i=0; i<2; i++){
707         for(j=0; j<2; j++){
708             for(k=0; k<2; k++){
709                 av_freep(&s->b_field_mv_table_base[i][j][k]);
710                 s->b_field_mv_table[i][j][k]=NULL;
711             }
712             av_freep(&s->b_field_select_table[i][j]);
713             av_freep(&s->p_field_mv_table_base[i][j]);
714             s->p_field_mv_table[i][j]=NULL;
715         }
716         av_freep(&s->p_field_select_table[i]);
717     }
718
719     av_freep(&s->dc_val_base);
720     av_freep(&s->ac_val_base);
721     av_freep(&s->coded_block_base);
722     av_freep(&s->mbintra_table);
723     av_freep(&s->cbp_table);
724     av_freep(&s->pred_dir_table);
725
726     av_freep(&s->mbskip_table);
727     av_freep(&s->prev_pict_types);
728     av_freep(&s->bitstream_buffer);
729     s->allocated_bitstream_buffer_size=0;
730
731     av_freep(&s->avctx->stats_out);
732     av_freep(&s->ac_stats);
733     av_freep(&s->error_status_table);
734     av_freep(&s->mb_index2xy);
735     av_freep(&s->lambda_table);
736     av_freep(&s->q_intra_matrix);
737     av_freep(&s->q_inter_matrix);
738     av_freep(&s->q_intra_matrix16);
739     av_freep(&s->q_inter_matrix16);
740     av_freep(&s->input_picture);
741     av_freep(&s->reordered_input_picture);
742     av_freep(&s->dct_offset);
743
744     if(s->picture){
745         for(i=0; i<MAX_PICTURE_COUNT; i++){
746             free_picture(s, &s->picture[i]);
747         }
748     }
749     av_freep(&s->picture);
750     s->context_initialized = 0;
751     s->last_picture_ptr=
752     s->next_picture_ptr=
753     s->current_picture_ptr= NULL;
754     s->linesize= s->uvlinesize= 0;
755
756     for(i=0; i<3; i++)
757         av_freep(&s->visualization_buffer[i]);
758
759     avcodec_default_free_buffers(s->avctx);
760 }
761
762 void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
763 {
764     int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1];
765     uint8_t index_run[MAX_RUN+1];
766     int last, run, level, start, end, i;
767
768     /* If table is static, we can quit if rl->max_level[0] is not NULL */
769     if(static_store && rl->max_level[0])
770         return;
771
772     /* compute max_level[], max_run[] and index_run[] */
773     for(last=0;last<2;last++) {
774         if (last == 0) {
775             start = 0;
776             end = rl->last;
777         } else {
778             start = rl->last;
779             end = rl->n;
780         }
781
782         memset(max_level, 0, MAX_RUN + 1);
783         memset(max_run, 0, MAX_LEVEL + 1);
784         memset(index_run, rl->n, MAX_RUN + 1);
785         for(i=start;i<end;i++) {
786             run = rl->table_run[i];
787             level = rl->table_level[i];
788             if (index_run[run] == rl->n)
789                 index_run[run] = i;
790             if (level > max_level[run])
791                 max_level[run] = level;
792             if (run > max_run[level])
793                 max_run[level] = run;
794         }
795         if(static_store)
796             rl->max_level[last] = static_store[last];
797         else
798             rl->max_level[last] = av_malloc(MAX_RUN + 1);
799         memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
800         if(static_store)
801             rl->max_run[last] = static_store[last] + MAX_RUN + 1;
802         else
803             rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
804         memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
805         if(static_store)
806             rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
807         else
808             rl->index_run[last] = av_malloc(MAX_RUN + 1);
809         memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
810     }
811 }
812
813 void init_vlc_rl(RLTable *rl)
814 {
815     int i, q;
816
817     for(q=0; q<32; q++){
818         int qmul= q*2;
819         int qadd= (q-1)|1;
820
821         if(q==0){
822             qmul=1;
823             qadd=0;
824         }
825         for(i=0; i<rl->vlc.table_size; i++){
826             int code= rl->vlc.table[i][0];
827             int len = rl->vlc.table[i][1];
828             int level, run;
829
830             if(len==0){ // illegal code
831                 run= 66;
832                 level= MAX_LEVEL;
833             }else if(len<0){ //more bits needed
834                 run= 0;
835                 level= code;
836             }else{
837                 if(code==rl->n){ //esc
838                     run= 66;
839                     level= 0;
840                 }else{
841                     run=   rl->table_run  [code] + 1;
842                     level= rl->table_level[code] * qmul + qadd;
843                     if(code >= rl->last) run+=192;
844                 }
845             }
846             rl->rl_vlc[q][i].len= len;
847             rl->rl_vlc[q][i].level= level;
848             rl->rl_vlc[q][i].run= run;
849         }
850     }
851 }
852
853 int ff_find_unused_picture(MpegEncContext *s, int shared){
854     int i;
855
856     if(shared){
857         for(i=0; i<MAX_PICTURE_COUNT; i++){
858             if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i;
859         }
860     }else{
861         for(i=0; i<MAX_PICTURE_COUNT; i++){
862             if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME
863         }
864         for(i=0; i<MAX_PICTURE_COUNT; i++){
865             if(s->picture[i].data[0]==NULL) return i;
866         }
867     }
868
869     av_log(s->avctx, AV_LOG_FATAL, "Internal error, picture buffer overflow\n");
870     /* We could return -1, but the codec would crash trying to draw into a
871      * non-existing frame anyway. This is safer than waiting for a random crash.
872      * Also the return of this is never useful, an encoder must only allocate
873      * as much as allowed in the specification. This has no relationship to how
874      * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
875      * enough for such valid streams).
876      * Plus, a decoder has to check stream validity and remove frames if too
877      * many reference frames are around. Waiting for "OOM" is not correct at
878      * all. Similarly, missing reference frames have to be replaced by
879      * interpolated/MC frames, anything else is a bug in the codec ...
880      */
881     abort();
882     return -1;
883 }
884
885 static void update_noise_reduction(MpegEncContext *s){
886     int intra, i;
887
888     for(intra=0; intra<2; intra++){
889         if(s->dct_count[intra] > (1<<16)){
890             for(i=0; i<64; i++){
891                 s->dct_error_sum[intra][i] >>=1;
892             }
893             s->dct_count[intra] >>= 1;
894         }
895
896         for(i=0; i<64; i++){
897             s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1);
898         }
899     }
900 }
901
902 /**
903  * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
904  */
905 int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
906 {
907     int i;
908     Picture *pic;
909     s->mb_skipped = 0;
910
911     assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);
912
913     /* mark&release old frames */
914     if (s->pict_type != FF_B_TYPE && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->data[0]) {
915       if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
916           free_frame_buffer(s, s->last_picture_ptr);
917
918         /* release forgotten pictures */
919         /* if(mpeg124/h263) */
920         if(!s->encoding){
921             for(i=0; i<MAX_PICTURE_COUNT; i++){
922                 if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){
923                     av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
924                     free_frame_buffer(s, &s->picture[i]);
925                 }
926             }
927         }
928       }
929     }
930
931     if(!s->encoding){
932         /* release non reference frames */
933         for(i=0; i<MAX_PICTURE_COUNT; i++){
934             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
935                 free_frame_buffer(s, &s->picture[i]);
936             }
937         }
938
939         if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL)
940             pic= s->current_picture_ptr; //we already have a unused image (maybe it was set before reading the header)
941         else{
942             i= ff_find_unused_picture(s, 0);
943             pic= &s->picture[i];
944         }
945
946         pic->reference= 0;
947         if (!s->dropable){
948             if (s->codec_id == CODEC_ID_H264)
949                 pic->reference = s->picture_structure;
950             else if (s->pict_type != FF_B_TYPE)
951                 pic->reference = 3;
952         }
953
954         pic->coded_picture_number= s->coded_picture_number++;
955
956         if(ff_alloc_picture(s, pic, 0) < 0)
957             return -1;
958
959         s->current_picture_ptr= pic;
960         s->current_picture_ptr->top_field_first= s->top_field_first; //FIXME use only the vars from current_pic
961         s->current_picture_ptr->interlaced_frame= !s->progressive_frame && !s->progressive_sequence;
962     }
963
964     s->current_picture_ptr->pict_type= s->pict_type;
965 //    if(s->flags && CODEC_FLAG_QSCALE)
966   //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
967     s->current_picture_ptr->key_frame= s->pict_type == FF_I_TYPE;
968
969     ff_copy_picture(&s->current_picture, s->current_picture_ptr);
970
971     if (s->pict_type != FF_B_TYPE) {
972         s->last_picture_ptr= s->next_picture_ptr;
973         if(!s->dropable)
974             s->next_picture_ptr= s->current_picture_ptr;
975     }
976 /*    av_log(s->avctx, AV_LOG_DEBUG, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
977         s->last_picture_ptr    ? s->last_picture_ptr->data[0] : NULL,
978         s->next_picture_ptr    ? s->next_picture_ptr->data[0] : NULL,
979         s->current_picture_ptr ? s->current_picture_ptr->data[0] : NULL,
980         s->pict_type, s->dropable);*/
981
982     if(s->codec_id != CODEC_ID_H264){
983         if((s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL) && s->pict_type!=FF_I_TYPE){
984             av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
985             /* Allocate a dummy frame */
986             i= ff_find_unused_picture(s, 0);
987             s->last_picture_ptr= &s->picture[i];
988             if(ff_alloc_picture(s, s->last_picture_ptr, 0) < 0)
989                 return -1;
990         }
991         if((s->next_picture_ptr==NULL || s->next_picture_ptr->data[0]==NULL) && s->pict_type==FF_B_TYPE){
992             /* Allocate a dummy frame */
993             i= ff_find_unused_picture(s, 0);
994             s->next_picture_ptr= &s->picture[i];
995             if(ff_alloc_picture(s, s->next_picture_ptr, 0) < 0)
996                 return -1;
997         }
998     }
999
1000     if(s->last_picture_ptr) ff_copy_picture(&s->last_picture, s->last_picture_ptr);
1001     if(s->next_picture_ptr) ff_copy_picture(&s->next_picture, s->next_picture_ptr);
1002
1003     assert(s->pict_type == FF_I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0]));
1004
1005     if(s->picture_structure!=PICT_FRAME && s->out_format != FMT_H264){
1006         int i;
1007         for(i=0; i<4; i++){
1008             if(s->picture_structure == PICT_BOTTOM_FIELD){
1009                  s->current_picture.data[i] += s->current_picture.linesize[i];
1010             }
1011             s->current_picture.linesize[i] *= 2;
1012             s->last_picture.linesize[i] *=2;
1013             s->next_picture.linesize[i] *=2;
1014         }
1015     }
1016
1017     s->hurry_up= s->avctx->hurry_up;
1018     s->error_recognition= avctx->error_recognition;
1019
1020     /* set dequantizer, we can't do it during init as it might change for mpeg4
1021        and we can't do it in the header decode as init is not called for mpeg4 there yet */
1022     if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){
1023         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1024         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1025     }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
1026         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1027         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1028     }else{
1029         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1030         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1031     }
1032
1033     if(s->dct_error_sum){
1034         assert(s->avctx->noise_reduction && s->encoding);
1035
1036         update_noise_reduction(s);
1037     }
1038
1039     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration)
1040         return ff_xvmc_field_start(s, avctx);
1041
1042     return 0;
1043 }
1044
1045 /* generic function for encode/decode called after a frame has been coded/decoded */
1046 void MPV_frame_end(MpegEncContext *s)
1047 {
1048     int i;
1049     /* draw edge for correct motion prediction if outside */
1050     //just to make sure that all data is rendered.
1051     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
1052         ff_xvmc_field_end(s);
1053     }else if(!s->avctx->hwaccel
1054        && !(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
1055        && s->unrestricted_mv
1056        && s->current_picture.reference
1057        && !s->intra_only
1058        && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
1059             s->dsp.draw_edges(s->current_picture.data[0], s->linesize  , s->h_edge_pos   , s->v_edge_pos   , EDGE_WIDTH  );
1060             s->dsp.draw_edges(s->current_picture.data[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
1061             s->dsp.draw_edges(s->current_picture.data[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
1062     }
1063     emms_c();
1064
1065     s->last_pict_type    = s->pict_type;
1066     s->last_lambda_for[s->pict_type]= s->current_picture_ptr->quality;
1067     if(s->pict_type!=FF_B_TYPE){
1068         s->last_non_b_pict_type= s->pict_type;
1069     }
1070 #if 0
1071         /* copy back current_picture variables */
1072     for(i=0; i<MAX_PICTURE_COUNT; i++){
1073         if(s->picture[i].data[0] == s->current_picture.data[0]){
1074             s->picture[i]= s->current_picture;
1075             break;
1076         }
1077     }
1078     assert(i<MAX_PICTURE_COUNT);
1079 #endif
1080
1081     if(s->encoding){
1082         /* release non-reference frames */
1083         for(i=0; i<MAX_PICTURE_COUNT; i++){
1084             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
1085                 free_frame_buffer(s, &s->picture[i]);
1086             }
1087         }
1088     }
1089     // clear copies, to avoid confusion
1090 #if 0
1091     memset(&s->last_picture, 0, sizeof(Picture));
1092     memset(&s->next_picture, 0, sizeof(Picture));
1093     memset(&s->current_picture, 0, sizeof(Picture));
1094 #endif
1095     s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr;
1096 }
1097
1098 /**
1099  * draws an line from (ex, ey) -> (sx, sy).
1100  * @param w width of the image
1101  * @param h height of the image
1102  * @param stride stride/linesize of the image
1103  * @param color color of the arrow
1104  */
1105 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1106     int x, y, fr, f;
1107
1108     sx= av_clip(sx, 0, w-1);
1109     sy= av_clip(sy, 0, h-1);
1110     ex= av_clip(ex, 0, w-1);
1111     ey= av_clip(ey, 0, h-1);
1112
1113     buf[sy*stride + sx]+= color;
1114
1115     if(FFABS(ex - sx) > FFABS(ey - sy)){
1116         if(sx > ex){
1117             FFSWAP(int, sx, ex);
1118             FFSWAP(int, sy, ey);
1119         }
1120         buf+= sx + sy*stride;
1121         ex-= sx;
1122         f= ((ey-sy)<<16)/ex;
1123         for(x= 0; x <= ex; x++){
1124             y = (x*f)>>16;
1125             fr= (x*f)&0xFFFF;
1126             buf[ y   *stride + x]+= (color*(0x10000-fr))>>16;
1127             buf[(y+1)*stride + x]+= (color*         fr )>>16;
1128         }
1129     }else{
1130         if(sy > ey){
1131             FFSWAP(int, sx, ex);
1132             FFSWAP(int, sy, ey);
1133         }
1134         buf+= sx + sy*stride;
1135         ey-= sy;
1136         if(ey) f= ((ex-sx)<<16)/ey;
1137         else   f= 0;
1138         for(y= 0; y <= ey; y++){
1139             x = (y*f)>>16;
1140             fr= (y*f)&0xFFFF;
1141             buf[y*stride + x  ]+= (color*(0x10000-fr))>>16;
1142             buf[y*stride + x+1]+= (color*         fr )>>16;
1143         }
1144     }
1145 }
1146
1147 /**
1148  * draws an arrow from (ex, ey) -> (sx, sy).
1149  * @param w width of the image
1150  * @param h height of the image
1151  * @param stride stride/linesize of the image
1152  * @param color color of the arrow
1153  */
1154 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1155     int dx,dy;
1156
1157     sx= av_clip(sx, -100, w+100);
1158     sy= av_clip(sy, -100, h+100);
1159     ex= av_clip(ex, -100, w+100);
1160     ey= av_clip(ey, -100, h+100);
1161
1162     dx= ex - sx;
1163     dy= ey - sy;
1164
1165     if(dx*dx + dy*dy > 3*3){
1166         int rx=  dx + dy;
1167         int ry= -dx + dy;
1168         int length= ff_sqrt((rx*rx + ry*ry)<<8);
1169
1170         //FIXME subpixel accuracy
1171         rx= ROUNDED_DIV(rx*3<<4, length);
1172         ry= ROUNDED_DIV(ry*3<<4, length);
1173
1174         draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1175         draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1176     }
1177     draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1178 }
1179
1180 /**
1181  * prints debuging info for the given picture.
1182  */
1183 void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){
1184
1185     if(s->avctx->hwaccel || !pict || !pict->mb_type) return;
1186
1187     if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){
1188         int x,y;
1189
1190         av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
1191         switch (pict->pict_type) {
1192             case FF_I_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break;
1193             case FF_P_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break;
1194             case FF_B_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break;
1195             case FF_S_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break;
1196             case FF_SI_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break;
1197             case FF_SP_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break;
1198         }
1199         for(y=0; y<s->mb_height; y++){
1200             for(x=0; x<s->mb_width; x++){
1201                 if(s->avctx->debug&FF_DEBUG_SKIP){
1202                     int count= s->mbskip_table[x + y*s->mb_stride];
1203                     if(count>9) count=9;
1204                     av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
1205                 }
1206                 if(s->avctx->debug&FF_DEBUG_QP){
1207                     av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
1208                 }
1209                 if(s->avctx->debug&FF_DEBUG_MB_TYPE){
1210                     int mb_type= pict->mb_type[x + y*s->mb_stride];
1211                     //Type & MV direction
1212                     if(IS_PCM(mb_type))
1213                         av_log(s->avctx, AV_LOG_DEBUG, "P");
1214                     else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
1215                         av_log(s->avctx, AV_LOG_DEBUG, "A");
1216                     else if(IS_INTRA4x4(mb_type))
1217                         av_log(s->avctx, AV_LOG_DEBUG, "i");
1218                     else if(IS_INTRA16x16(mb_type))
1219                         av_log(s->avctx, AV_LOG_DEBUG, "I");
1220                     else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
1221                         av_log(s->avctx, AV_LOG_DEBUG, "d");
1222                     else if(IS_DIRECT(mb_type))
1223                         av_log(s->avctx, AV_LOG_DEBUG, "D");
1224                     else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
1225                         av_log(s->avctx, AV_LOG_DEBUG, "g");
1226                     else if(IS_GMC(mb_type))
1227                         av_log(s->avctx, AV_LOG_DEBUG, "G");
1228                     else if(IS_SKIP(mb_type))
1229                         av_log(s->avctx, AV_LOG_DEBUG, "S");
1230                     else if(!USES_LIST(mb_type, 1))
1231                         av_log(s->avctx, AV_LOG_DEBUG, ">");
1232                     else if(!USES_LIST(mb_type, 0))
1233                         av_log(s->avctx, AV_LOG_DEBUG, "<");
1234                     else{
1235                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1236                         av_log(s->avctx, AV_LOG_DEBUG, "X");
1237                     }
1238
1239                     //segmentation
1240                     if(IS_8X8(mb_type))
1241                         av_log(s->avctx, AV_LOG_DEBUG, "+");
1242                     else if(IS_16X8(mb_type))
1243                         av_log(s->avctx, AV_LOG_DEBUG, "-");
1244                     else if(IS_8X16(mb_type))
1245                         av_log(s->avctx, AV_LOG_DEBUG, "|");
1246                     else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
1247                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1248                     else
1249                         av_log(s->avctx, AV_LOG_DEBUG, "?");
1250
1251
1252                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264)
1253                         av_log(s->avctx, AV_LOG_DEBUG, "=");
1254                     else
1255                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1256                 }
1257 //                av_log(s->avctx, AV_LOG_DEBUG, " ");
1258             }
1259             av_log(s->avctx, AV_LOG_DEBUG, "\n");
1260         }
1261     }
1262
1263     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
1264         const int shift= 1 + s->quarter_sample;
1265         int mb_y;
1266         uint8_t *ptr;
1267         int i;
1268         int h_chroma_shift, v_chroma_shift, block_height;
1269         const int width = s->avctx->width;
1270         const int height= s->avctx->height;
1271         const int mv_sample_log2= 4 - pict->motion_subsample_log2;
1272         const int mv_stride= (s->mb_width << mv_sample_log2) + (s->codec_id == CODEC_ID_H264 ? 0 : 1);
1273         s->low_delay=0; //needed to see the vectors without trashing the buffers
1274
1275         avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
1276         for(i=0; i<3; i++){
1277             memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift);
1278             pict->data[i]= s->visualization_buffer[i];
1279         }
1280         pict->type= FF_BUFFER_TYPE_COPY;
1281         ptr= pict->data[0];
1282         block_height = 16>>v_chroma_shift;
1283
1284         for(mb_y=0; mb_y<s->mb_height; mb_y++){
1285             int mb_x;
1286             for(mb_x=0; mb_x<s->mb_width; mb_x++){
1287                 const int mb_index= mb_x + mb_y*s->mb_stride;
1288                 if((s->avctx->debug_mv) && pict->motion_val){
1289                   int type;
1290                   for(type=0; type<3; type++){
1291                     int direction = 0;
1292                     switch (type) {
1293                       case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=FF_P_TYPE))
1294                                 continue;
1295                               direction = 0;
1296                               break;
1297                       case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=FF_B_TYPE))
1298                                 continue;
1299                               direction = 0;
1300                               break;
1301                       case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=FF_B_TYPE))
1302                                 continue;
1303                               direction = 1;
1304                               break;
1305                     }
1306                     if(!USES_LIST(pict->mb_type[mb_index], direction))
1307                         continue;
1308
1309                     if(IS_8X8(pict->mb_type[mb_index])){
1310                       int i;
1311                       for(i=0; i<4; i++){
1312                         int sx= mb_x*16 + 4 + 8*(i&1);
1313                         int sy= mb_y*16 + 4 + 8*(i>>1);
1314                         int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1315                         int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1316                         int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1317                         draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1318                       }
1319                     }else if(IS_16X8(pict->mb_type[mb_index])){
1320                       int i;
1321                       for(i=0; i<2; i++){
1322                         int sx=mb_x*16 + 8;
1323                         int sy=mb_y*16 + 4 + 8*i;
1324                         int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1);
1325                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1326                         int my=(pict->motion_val[direction][xy][1]>>shift);
1327
1328                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1329                             my*=2;
1330
1331                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1332                       }
1333                     }else if(IS_8X16(pict->mb_type[mb_index])){
1334                       int i;
1335                       for(i=0; i<2; i++){
1336                         int sx=mb_x*16 + 4 + 8*i;
1337                         int sy=mb_y*16 + 8;
1338                         int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1);
1339                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1340                         int my=(pict->motion_val[direction][xy][1]>>shift);
1341
1342                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1343                             my*=2;
1344
1345                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1346                       }
1347                     }else{
1348                       int sx= mb_x*16 + 8;
1349                       int sy= mb_y*16 + 8;
1350                       int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2;
1351                       int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1352                       int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1353                       draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1354                     }
1355                   }
1356                 }
1357                 if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){
1358                     uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL;
1359                     int y;
1360                     for(y=0; y<block_height; y++){
1361                         *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= c;
1362                         *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= c;
1363                     }
1364                 }
1365                 if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){
1366                     int mb_type= pict->mb_type[mb_index];
1367                     uint64_t u,v;
1368                     int y;
1369 #define COLOR(theta, r)\
1370 u= (int)(128 + r*cos(theta*3.141592/180));\
1371 v= (int)(128 + r*sin(theta*3.141592/180));
1372
1373
1374                     u=v=128;
1375                     if(IS_PCM(mb_type)){
1376                         COLOR(120,48)
1377                     }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){
1378                         COLOR(30,48)
1379                     }else if(IS_INTRA4x4(mb_type)){
1380                         COLOR(90,48)
1381                     }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){
1382 //                        COLOR(120,48)
1383                     }else if(IS_DIRECT(mb_type)){
1384                         COLOR(150,48)
1385                     }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){
1386                         COLOR(170,48)
1387                     }else if(IS_GMC(mb_type)){
1388                         COLOR(190,48)
1389                     }else if(IS_SKIP(mb_type)){
1390 //                        COLOR(180,48)
1391                     }else if(!USES_LIST(mb_type, 1)){
1392                         COLOR(240,48)
1393                     }else if(!USES_LIST(mb_type, 0)){
1394                         COLOR(0,48)
1395                     }else{
1396                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1397                         COLOR(300,48)
1398                     }
1399
1400                     u*= 0x0101010101010101ULL;
1401                     v*= 0x0101010101010101ULL;
1402                     for(y=0; y<block_height; y++){
1403                         *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= u;
1404                         *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= v;
1405                     }
1406
1407                     //segmentation
1408                     if(IS_8X8(mb_type) || IS_16X8(mb_type)){
1409                         *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
1410                         *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
1411                     }
1412                     if(IS_8X8(mb_type) || IS_8X16(mb_type)){
1413                         for(y=0; y<16; y++)
1414                             pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80;
1415                     }
1416                     if(IS_8X8(mb_type) && mv_sample_log2 >= 2){
1417                         int dm= 1 << (mv_sample_log2-2);
1418                         for(i=0; i<4; i++){
1419                             int sx= mb_x*16 + 8*(i&1);
1420                             int sy= mb_y*16 + 8*(i>>1);
1421                             int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1422                             //FIXME bidir
1423                             int32_t *mv = (int32_t*)&pict->motion_val[0][xy];
1424                             if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)])
1425                                 for(y=0; y<8; y++)
1426                                     pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80;
1427                             if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)])
1428                                 *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL;
1429                         }
1430                     }
1431
1432                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){
1433                         // hmm
1434                     }
1435                 }
1436                 s->mbskip_table[mb_index]=0;
1437             }
1438         }
1439     }
1440 }
1441
1442 static inline int hpel_motion_lowres(MpegEncContext *s,
1443                                   uint8_t *dest, uint8_t *src,
1444                                   int field_based, int field_select,
1445                                   int src_x, int src_y,
1446                                   int width, int height, int stride,
1447                                   int h_edge_pos, int v_edge_pos,
1448                                   int w, int h, h264_chroma_mc_func *pix_op,
1449                                   int motion_x, int motion_y)
1450 {
1451     const int lowres= s->avctx->lowres;
1452     const int op_index= FFMIN(lowres, 2);
1453     const int s_mask= (2<<lowres)-1;
1454     int emu=0;
1455     int sx, sy;
1456
1457     if(s->quarter_sample){
1458         motion_x/=2;
1459         motion_y/=2;
1460     }
1461
1462     sx= motion_x & s_mask;
1463     sy= motion_y & s_mask;
1464     src_x += motion_x >> (lowres+1);
1465     src_y += motion_y >> (lowres+1);
1466
1467     src += src_y * stride + src_x;
1468
1469     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - w
1470        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
1471         ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
1472                             src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
1473         src= s->edge_emu_buffer;
1474         emu=1;
1475     }
1476
1477     sx= (sx << 2) >> lowres;
1478     sy= (sy << 2) >> lowres;
1479     if(field_select)
1480         src += s->linesize;
1481     pix_op[op_index](dest, src, stride, h, sx, sy);
1482     return emu;
1483 }
1484
1485 /* apply one mpeg motion vector to the three components */
1486 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1487                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1488                                int field_based, int bottom_field, int field_select,
1489                                uint8_t **ref_picture, h264_chroma_mc_func *pix_op,
1490                                int motion_x, int motion_y, int h, int mb_y)
1491 {
1492     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1493     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy;
1494     const int lowres= s->avctx->lowres;
1495     const int op_index= FFMIN(lowres, 2);
1496     const int block_s= 8>>lowres;
1497     const int s_mask= (2<<lowres)-1;
1498     const int h_edge_pos = s->h_edge_pos >> lowres;
1499     const int v_edge_pos = s->v_edge_pos >> lowres;
1500     linesize   = s->current_picture.linesize[0] << field_based;
1501     uvlinesize = s->current_picture.linesize[1] << field_based;
1502
1503     if(s->quarter_sample){ //FIXME obviously not perfect but qpel will not work in lowres anyway
1504         motion_x/=2;
1505         motion_y/=2;
1506     }
1507
1508     if(field_based){
1509         motion_y += (bottom_field - field_select)*((1<<lowres)-1);
1510     }
1511
1512     sx= motion_x & s_mask;
1513     sy= motion_y & s_mask;
1514     src_x = s->mb_x*2*block_s               + (motion_x >> (lowres+1));
1515     src_y =(   mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1));
1516
1517     if (s->out_format == FMT_H263) {
1518         uvsx = ((motion_x>>1) & s_mask) | (sx&1);
1519         uvsy = ((motion_y>>1) & s_mask) | (sy&1);
1520         uvsrc_x = src_x>>1;
1521         uvsrc_y = src_y>>1;
1522     }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
1523         mx = motion_x / 4;
1524         my = motion_y / 4;
1525         uvsx = (2*mx) & s_mask;
1526         uvsy = (2*my) & s_mask;
1527         uvsrc_x = s->mb_x*block_s               + (mx >> lowres);
1528         uvsrc_y =    mb_y*block_s               + (my >> lowres);
1529     } else {
1530         mx = motion_x / 2;
1531         my = motion_y / 2;
1532         uvsx = mx & s_mask;
1533         uvsy = my & s_mask;
1534         uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1535         uvsrc_y =(   mb_y*block_s>>field_based) + (my >> (lowres+1));
1536     }
1537
1538     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
1539     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1540     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1541
1542     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - 2*block_s
1543        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
1544             ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
1545                              src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
1546             ptr_y = s->edge_emu_buffer;
1547             if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1548                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
1549                 ff_emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based,
1550                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
1551                 ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based,
1552                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
1553                 ptr_cb= uvbuf;
1554                 ptr_cr= uvbuf+16;
1555             }
1556     }
1557
1558     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
1559         dest_y += s->linesize;
1560         dest_cb+= s->uvlinesize;
1561         dest_cr+= s->uvlinesize;
1562     }
1563
1564     if(field_select){
1565         ptr_y += s->linesize;
1566         ptr_cb+= s->uvlinesize;
1567         ptr_cr+= s->uvlinesize;
1568     }
1569
1570     sx= (sx << 2) >> lowres;
1571     sy= (sy << 2) >> lowres;
1572     pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy);
1573
1574     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1575         uvsx= (uvsx << 2) >> lowres;
1576         uvsy= (uvsy << 2) >> lowres;
1577         pix_op[op_index](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
1578         pix_op[op_index](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
1579     }
1580     //FIXME h261 lowres loop filter
1581 }
1582
1583 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1584                                      uint8_t *dest_cb, uint8_t *dest_cr,
1585                                      uint8_t **ref_picture,
1586                                      h264_chroma_mc_func *pix_op,
1587                                      int mx, int my){
1588     const int lowres= s->avctx->lowres;
1589     const int op_index= FFMIN(lowres, 2);
1590     const int block_s= 8>>lowres;
1591     const int s_mask= (2<<lowres)-1;
1592     const int h_edge_pos = s->h_edge_pos >> (lowres+1);
1593     const int v_edge_pos = s->v_edge_pos >> (lowres+1);
1594     int emu=0, src_x, src_y, offset, sx, sy;
1595     uint8_t *ptr;
1596
1597     if(s->quarter_sample){
1598         mx/=2;
1599         my/=2;
1600     }
1601
1602     /* In case of 8X8, we construct a single chroma motion vector
1603        with a special rounding */
1604     mx= ff_h263_round_chroma(mx);
1605     my= ff_h263_round_chroma(my);
1606
1607     sx= mx & s_mask;
1608     sy= my & s_mask;
1609     src_x = s->mb_x*block_s + (mx >> (lowres+1));
1610     src_y = s->mb_y*block_s + (my >> (lowres+1));
1611
1612     offset = src_y * s->uvlinesize + src_x;
1613     ptr = ref_picture[1] + offset;
1614     if(s->flags&CODEC_FLAG_EMU_EDGE){
1615         if(   (unsigned)src_x > h_edge_pos - (!!sx) - block_s
1616            || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){
1617             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
1618             ptr= s->edge_emu_buffer;
1619             emu=1;
1620         }
1621     }
1622     sx= (sx << 2) >> lowres;
1623     sy= (sy << 2) >> lowres;
1624     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1625
1626     ptr = ref_picture[2] + offset;
1627     if(emu){
1628         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
1629         ptr= s->edge_emu_buffer;
1630     }
1631     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1632 }
1633
1634 /**
1635  * motion compensation of a single macroblock
1636  * @param s context
1637  * @param dest_y luma destination pointer
1638  * @param dest_cb chroma cb/u destination pointer
1639  * @param dest_cr chroma cr/v destination pointer
1640  * @param dir direction (0->forward, 1->backward)
1641  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1642  * @param pic_op halfpel motion compensation function (average or put normally)
1643  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1644  */
1645 static inline void MPV_motion_lowres(MpegEncContext *s,
1646                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1647                               int dir, uint8_t **ref_picture,
1648                               h264_chroma_mc_func *pix_op)
1649 {
1650     int mx, my;
1651     int mb_x, mb_y, i;
1652     const int lowres= s->avctx->lowres;
1653     const int block_s= 8>>lowres;
1654
1655     mb_x = s->mb_x;
1656     mb_y = s->mb_y;
1657
1658     switch(s->mv_type) {
1659     case MV_TYPE_16X16:
1660         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1661                     0, 0, 0,
1662                     ref_picture, pix_op,
1663                     s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y);
1664         break;
1665     case MV_TYPE_8X8:
1666         mx = 0;
1667         my = 0;
1668             for(i=0;i<4;i++) {
1669                 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * s->linesize)*block_s,
1670                             ref_picture[0], 0, 0,
1671                             (2*mb_x + (i & 1))*block_s, (2*mb_y + (i >>1))*block_s,
1672                             s->width, s->height, s->linesize,
1673                             s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1674                             block_s, block_s, pix_op,
1675                             s->mv[dir][i][0], s->mv[dir][i][1]);
1676
1677                 mx += s->mv[dir][i][0];
1678                 my += s->mv[dir][i][1];
1679             }
1680
1681         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
1682             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, pix_op, mx, my);
1683         break;
1684     case MV_TYPE_FIELD:
1685         if (s->picture_structure == PICT_FRAME) {
1686             /* top field */
1687             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1688                         1, 0, s->field_select[dir][0],
1689                         ref_picture, pix_op,
1690                         s->mv[dir][0][0], s->mv[dir][0][1], block_s, mb_y);
1691             /* bottom field */
1692             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1693                         1, 1, s->field_select[dir][1],
1694                         ref_picture, pix_op,
1695                         s->mv[dir][1][0], s->mv[dir][1][1], block_s, mb_y);
1696         } else {
1697             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
1698                 ref_picture= s->current_picture_ptr->data;
1699             }
1700
1701             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1702                         0, 0, s->field_select[dir][0],
1703                         ref_picture, pix_op,
1704                         s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y>>1);
1705         }
1706         break;
1707     case MV_TYPE_16X8:
1708         for(i=0; i<2; i++){
1709             uint8_t ** ref2picture;
1710
1711             if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == FF_B_TYPE || s->first_field){
1712                 ref2picture= ref_picture;
1713             }else{
1714                 ref2picture= s->current_picture_ptr->data;
1715             }
1716
1717             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1718                         0, 0, s->field_select[dir][i],
1719                         ref2picture, pix_op,
1720                         s->mv[dir][i][0], s->mv[dir][i][1] + 2*block_s*i, block_s, mb_y>>1);
1721
1722             dest_y += 2*block_s*s->linesize;
1723             dest_cb+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
1724             dest_cr+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
1725         }
1726         break;
1727     case MV_TYPE_DMV:
1728         if(s->picture_structure == PICT_FRAME){
1729             for(i=0; i<2; i++){
1730                 int j;
1731                 for(j=0; j<2; j++){
1732                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1733                                 1, j, j^i,
1734                                 ref_picture, pix_op,
1735                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], block_s, mb_y);
1736                 }
1737                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
1738             }
1739         }else{
1740             for(i=0; i<2; i++){
1741                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1742                             0, 0, s->picture_structure != i+1,
1743                             ref_picture, pix_op,
1744                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],2*block_s, mb_y>>1);
1745
1746                 // after put we make avg of the same block
1747                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
1748
1749                 //opposite parity is always in the same frame if this is second field
1750                 if(!s->first_field){
1751                     ref_picture = s->current_picture_ptr->data;
1752                 }
1753             }
1754         }
1755     break;
1756     default: assert(0);
1757     }
1758 }
1759
1760 /* put block[] to dest[] */
1761 static inline void put_dct(MpegEncContext *s,
1762                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
1763 {
1764     s->dct_unquantize_intra(s, block, i, qscale);
1765     s->dsp.idct_put (dest, line_size, block);
1766 }
1767
1768 /* add block[] to dest[] */
1769 static inline void add_dct(MpegEncContext *s,
1770                            DCTELEM *block, int i, uint8_t *dest, int line_size)
1771 {
1772     if (s->block_last_index[i] >= 0) {
1773         s->dsp.idct_add (dest, line_size, block);
1774     }
1775 }
1776
1777 static inline void add_dequant_dct(MpegEncContext *s,
1778                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
1779 {
1780     if (s->block_last_index[i] >= 0) {
1781         s->dct_unquantize_inter(s, block, i, qscale);
1782
1783         s->dsp.idct_add (dest, line_size, block);
1784     }
1785 }
1786
1787 /**
1788  * cleans dc, ac, coded_block for the current non intra MB
1789  */
1790 void ff_clean_intra_table_entries(MpegEncContext *s)
1791 {
1792     int wrap = s->b8_stride;
1793     int xy = s->block_index[0];
1794
1795     s->dc_val[0][xy           ] =
1796     s->dc_val[0][xy + 1       ] =
1797     s->dc_val[0][xy     + wrap] =
1798     s->dc_val[0][xy + 1 + wrap] = 1024;
1799     /* ac pred */
1800     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1801     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1802     if (s->msmpeg4_version>=3) {
1803         s->coded_block[xy           ] =
1804         s->coded_block[xy + 1       ] =
1805         s->coded_block[xy     + wrap] =
1806         s->coded_block[xy + 1 + wrap] = 0;
1807     }
1808     /* chroma */
1809     wrap = s->mb_stride;
1810     xy = s->mb_x + s->mb_y * wrap;
1811     s->dc_val[1][xy] =
1812     s->dc_val[2][xy] = 1024;
1813     /* ac pred */
1814     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1815     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1816
1817     s->mbintra_table[xy]= 0;
1818 }
1819
1820 /* generic function called after a macroblock has been parsed by the
1821    decoder or after it has been encoded by the encoder.
1822
1823    Important variables used:
1824    s->mb_intra : true if intra macroblock
1825    s->mv_dir   : motion vector direction
1826    s->mv_type  : motion vector type
1827    s->mv       : motion vector
1828    s->interlaced_dct : true if interlaced dct used (mpeg2)
1829  */
1830 static av_always_inline
1831 void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
1832                             int lowres_flag, int is_mpeg12)
1833 {
1834     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1835     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
1836         ff_xvmc_decode_mb(s);//xvmc uses pblocks
1837         return;
1838     }
1839
1840     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1841        /* save DCT coefficients */
1842        int i,j;
1843        DCTELEM *dct = &s->current_picture.dct_coeff[mb_xy*64*6];
1844        for(i=0; i<6; i++)
1845            for(j=0; j<64; j++)
1846                *dct++ = block[i][s->dsp.idct_permutation[j]];
1847     }
1848
1849     s->current_picture.qscale_table[mb_xy]= s->qscale;
1850
1851     /* update DC predictors for P macroblocks */
1852     if (!s->mb_intra) {
1853         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1854             if(s->mbintra_table[mb_xy])
1855                 ff_clean_intra_table_entries(s);
1856         } else {
1857             s->last_dc[0] =
1858             s->last_dc[1] =
1859             s->last_dc[2] = 128 << s->intra_dc_precision;
1860         }
1861     }
1862     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1863         s->mbintra_table[mb_xy]=1;
1864
1865     if ((s->flags&CODEC_FLAG_PSNR) || !(s->encoding && (s->intra_only || s->pict_type==FF_B_TYPE) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc
1866         uint8_t *dest_y, *dest_cb, *dest_cr;
1867         int dct_linesize, dct_offset;
1868         op_pixels_func (*op_pix)[4];
1869         qpel_mc_func (*op_qpix)[16];
1870         const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
1871         const int uvlinesize= s->current_picture.linesize[1];
1872         const int readable= s->pict_type != FF_B_TYPE || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1873         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
1874
1875         /* avoid copy if macroblock skipped in last frame too */
1876         /* skip only during decoding as we might trash the buffers during encoding a bit */
1877         if(!s->encoding){
1878             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
1879             const int age= s->current_picture.age;
1880
1881             assert(age);
1882
1883             if (s->mb_skipped) {
1884                 s->mb_skipped= 0;
1885                 assert(s->pict_type!=FF_I_TYPE);
1886
1887                 (*mbskip_ptr) ++; /* indicate that this time we skipped it */
1888                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1889
1890                 /* if previous was skipped too, then nothing to do !  */
1891                 if (*mbskip_ptr >= age && s->current_picture.reference){
1892                     return;
1893                 }
1894             } else if(!s->current_picture.reference){
1895                 (*mbskip_ptr) ++; /* increase counter so the age can be compared cleanly */
1896                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1897             } else{
1898                 *mbskip_ptr = 0; /* not skipped */
1899             }
1900         }
1901
1902         dct_linesize = linesize << s->interlaced_dct;
1903         dct_offset =(s->interlaced_dct)? linesize : linesize*block_size;
1904
1905         if(readable){
1906             dest_y=  s->dest[0];
1907             dest_cb= s->dest[1];
1908             dest_cr= s->dest[2];
1909         }else{
1910             dest_y = s->b_scratchpad;
1911             dest_cb= s->b_scratchpad+16*linesize;
1912             dest_cr= s->b_scratchpad+32*linesize;
1913         }
1914
1915         if (!s->mb_intra) {
1916             /* motion handling */
1917             /* decoding or more than one mb_type (MC was already done otherwise) */
1918             if(!s->encoding){
1919                 if(lowres_flag){
1920                     h264_chroma_mc_func *op_pix = s->dsp.put_h264_chroma_pixels_tab;
1921
1922                     if (s->mv_dir & MV_DIR_FORWARD) {
1923                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix);
1924                         op_pix = s->dsp.avg_h264_chroma_pixels_tab;
1925                     }
1926                     if (s->mv_dir & MV_DIR_BACKWARD) {
1927                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix);
1928                     }
1929                 }else{
1930                     op_qpix= s->me.qpel_put;
1931                     if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1932                         op_pix = s->dsp.put_pixels_tab;
1933                     }else{
1934                         op_pix = s->dsp.put_no_rnd_pixels_tab;
1935                     }
1936                     if (s->mv_dir & MV_DIR_FORWARD) {
1937                         MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1938                         op_pix = s->dsp.avg_pixels_tab;
1939                         op_qpix= s->me.qpel_avg;
1940                     }
1941                     if (s->mv_dir & MV_DIR_BACKWARD) {
1942                         MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1943                     }
1944                 }
1945             }
1946
1947             /* skip dequant / idct if we are really late ;) */
1948             if(s->hurry_up>1) goto skip_idct;
1949             if(s->avctx->skip_idct){
1950                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == FF_B_TYPE)
1951                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != FF_I_TYPE)
1952                    || s->avctx->skip_idct >= AVDISCARD_ALL)
1953                     goto skip_idct;
1954             }
1955
1956             /* add dct residue */
1957             if(s->encoding || !(   s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO
1958                                 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){
1959                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
1960                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
1961                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
1962                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
1963
1964                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1965                     if (s->chroma_y_shift){
1966                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
1967                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
1968                     }else{
1969                         dct_linesize >>= 1;
1970                         dct_offset >>=1;
1971                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
1972                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
1973                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
1974                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
1975                     }
1976                 }
1977             } else if(is_mpeg12 || (s->codec_id != CODEC_ID_WMV2)){
1978                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
1979                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
1980                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
1981                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
1982
1983                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1984                     if(s->chroma_y_shift){//Chroma420
1985                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
1986                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
1987                     }else{
1988                         //chroma422
1989                         dct_linesize = uvlinesize << s->interlaced_dct;
1990                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
1991
1992                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
1993                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
1994                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
1995                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
1996                         if(!s->chroma_x_shift){//Chroma444
1997                             add_dct(s, block[8], 8, dest_cb+8, dct_linesize);
1998                             add_dct(s, block[9], 9, dest_cr+8, dct_linesize);
1999                             add_dct(s, block[10], 10, dest_cb+8+dct_offset, dct_linesize);
2000                             add_dct(s, block[11], 11, dest_cr+8+dct_offset, dct_linesize);
2001                         }
2002                     }
2003                 }//fi gray
2004             }
2005             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2006                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2007             }
2008         } else {
2009             /* dct only in intra block */
2010             if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){
2011                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2012                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2013                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2014                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2015
2016                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2017                     if(s->chroma_y_shift){
2018                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2019                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2020                     }else{
2021                         dct_offset >>=1;
2022                         dct_linesize >>=1;
2023                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2024                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2025                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2026                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2027                     }
2028                 }
2029             }else{
2030                 s->dsp.idct_put(dest_y                          , dct_linesize, block[0]);
2031                 s->dsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2032                 s->dsp.idct_put(dest_y + dct_offset             , dct_linesize, block[2]);
2033                 s->dsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2034
2035                 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2036                     if(s->chroma_y_shift){
2037                         s->dsp.idct_put(dest_cb, uvlinesize, block[4]);
2038                         s->dsp.idct_put(dest_cr, uvlinesize, block[5]);
2039                     }else{
2040
2041                         dct_linesize = uvlinesize << s->interlaced_dct;
2042                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
2043
2044                         s->dsp.idct_put(dest_cb,              dct_linesize, block[4]);
2045                         s->dsp.idct_put(dest_cr,              dct_linesize, block[5]);
2046                         s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2047                         s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2048                         if(!s->chroma_x_shift){//Chroma444
2049                             s->dsp.idct_put(dest_cb + 8,              dct_linesize, block[8]);
2050                             s->dsp.idct_put(dest_cr + 8,              dct_linesize, block[9]);
2051                             s->dsp.idct_put(dest_cb + 8 + dct_offset, dct_linesize, block[10]);
2052                             s->dsp.idct_put(dest_cr + 8 + dct_offset, dct_linesize, block[11]);
2053                         }
2054                     }
2055                 }//gray
2056             }
2057         }
2058 skip_idct:
2059         if(!readable){
2060             s->dsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2061             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2062             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2063         }
2064     }
2065 }
2066
2067 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){
2068 #if !CONFIG_SMALL
2069     if(s->out_format == FMT_MPEG1) {
2070         if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 1);
2071         else                 MPV_decode_mb_internal(s, block, 0, 1);
2072     } else
2073 #endif
2074     if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 0);
2075     else                  MPV_decode_mb_internal(s, block, 0, 0);
2076 }
2077
2078 /**
2079  *
2080  * @param h is the normal height, this will be reduced automatically if needed for the last row
2081  */
2082 void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
2083     if (s->avctx->draw_horiz_band) {
2084         AVFrame *src;
2085         const int field_pic= s->picture_structure != PICT_FRAME;
2086         int offset[4];
2087
2088         h= FFMIN(h, (s->avctx->height>>field_pic) - y);
2089
2090         if(field_pic && !(s->avctx->slice_flags&SLICE_FLAG_ALLOW_FIELD)){
2091             h <<= 1;
2092             y <<= 1;
2093             if(s->first_field) return;
2094         }
2095
2096         if(s->pict_type==FF_B_TYPE || s->low_delay || (s->avctx->slice_flags&SLICE_FLAG_CODED_ORDER))
2097             src= (AVFrame*)s->current_picture_ptr;
2098         else if(s->last_picture_ptr)
2099             src= (AVFrame*)s->last_picture_ptr;
2100         else
2101             return;
2102
2103         if(s->pict_type==FF_B_TYPE && s->picture_structure == PICT_FRAME && s->out_format != FMT_H264){
2104             offset[0]=
2105             offset[1]=
2106             offset[2]=
2107             offset[3]= 0;
2108         }else{
2109             offset[0]= y * s->linesize;
2110             offset[1]=
2111             offset[2]= (y >> s->chroma_y_shift) * s->uvlinesize;
2112             offset[3]= 0;
2113         }
2114
2115         emms_c();
2116
2117         s->avctx->draw_horiz_band(s->avctx, src, offset,
2118                                   y, s->picture_structure, h);
2119     }
2120 }
2121
2122 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2123     const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
2124     const int uvlinesize= s->current_picture.linesize[1];
2125     const int mb_size= 4 - s->avctx->lowres;
2126
2127     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2128     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2129     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2130     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2131     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2132     s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2133     //block_index is not used by mpeg2, so it is not affected by chroma_format
2134
2135     s->dest[0] = s->current_picture.data[0] + ((s->mb_x - 1) << mb_size);
2136     s->dest[1] = s->current_picture.data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
2137     s->dest[2] = s->current_picture.data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
2138
2139     if(!(s->pict_type==FF_B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2140     {
2141         if(s->picture_structure==PICT_FRAME){
2142         s->dest[0] += s->mb_y *   linesize << mb_size;
2143         s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
2144         s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
2145         }else{
2146             s->dest[0] += (s->mb_y>>1) *   linesize << mb_size;
2147             s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
2148             s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
2149             assert((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2150         }
2151     }
2152 }
2153
2154 void ff_mpeg_flush(AVCodecContext *avctx){
2155     int i;
2156     MpegEncContext *s = avctx->priv_data;
2157
2158     if(s==NULL || s->picture==NULL)
2159         return;
2160
2161     for(i=0; i<MAX_PICTURE_COUNT; i++){
2162        if(s->picture[i].data[0] && (   s->picture[i].type == FF_BUFFER_TYPE_INTERNAL
2163                                     || s->picture[i].type == FF_BUFFER_TYPE_USER))
2164         free_frame_buffer(s, &s->picture[i]);
2165     }
2166     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2167
2168     s->mb_x= s->mb_y= 0;
2169     s->closed_gop= 0;
2170
2171     s->parse_context.state= -1;
2172     s->parse_context.frame_start_found= 0;
2173     s->parse_context.overread= 0;
2174     s->parse_context.overread_index= 0;
2175     s->parse_context.index= 0;
2176     s->parse_context.last_index= 0;
2177     s->bitstream_buffer_size=0;
2178     s->pp_time=0;
2179 }
2180
2181 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
2182                                    DCTELEM *block, int n, int qscale)
2183 {
2184     int i, level, nCoeffs;
2185     const uint16_t *quant_matrix;
2186
2187     nCoeffs= s->block_last_index[n];
2188
2189     if (n < 4)
2190         block[0] = block[0] * s->y_dc_scale;
2191     else
2192         block[0] = block[0] * s->c_dc_scale;
2193     /* XXX: only mpeg1 */
2194     quant_matrix = s->intra_matrix;
2195     for(i=1;i<=nCoeffs;i++) {
2196         int j= s->intra_scantable.permutated[i];
2197         level = block[j];
2198         if (level) {
2199             if (level < 0) {
2200                 level = -level;
2201                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2202                 level = (level - 1) | 1;
2203                 level = -level;
2204             } else {
2205                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2206                 level = (level - 1) | 1;
2207             }
2208             block[j] = level;
2209         }
2210     }
2211 }
2212
2213 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
2214                                    DCTELEM *block, int n, int qscale)
2215 {
2216     int i, level, nCoeffs;
2217     const uint16_t *quant_matrix;
2218
2219     nCoeffs= s->block_last_index[n];
2220
2221     quant_matrix = s->inter_matrix;
2222     for(i=0; i<=nCoeffs; i++) {
2223         int j= s->intra_scantable.permutated[i];
2224         level = block[j];
2225         if (level) {
2226             if (level < 0) {
2227                 level = -level;
2228                 level = (((level << 1) + 1) * qscale *
2229                          ((int) (quant_matrix[j]))) >> 4;
2230                 level = (level - 1) | 1;
2231                 level = -level;
2232             } else {
2233                 level = (((level << 1) + 1) * qscale *
2234                          ((int) (quant_matrix[j]))) >> 4;
2235                 level = (level - 1) | 1;
2236             }
2237             block[j] = level;
2238         }
2239     }
2240 }
2241
2242 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
2243                                    DCTELEM *block, int n, int qscale)
2244 {
2245     int i, level, nCoeffs;
2246     const uint16_t *quant_matrix;
2247
2248     if(s->alternate_scan) nCoeffs= 63;
2249     else nCoeffs= s->block_last_index[n];
2250
2251     if (n < 4)
2252         block[0] = block[0] * s->y_dc_scale;
2253     else
2254         block[0] = block[0] * s->c_dc_scale;
2255     quant_matrix = s->intra_matrix;
2256     for(i=1;i<=nCoeffs;i++) {
2257         int j= s->intra_scantable.permutated[i];
2258         level = block[j];
2259         if (level) {
2260             if (level < 0) {
2261                 level = -level;
2262                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2263                 level = -level;
2264             } else {
2265                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2266             }
2267             block[j] = level;
2268         }
2269     }
2270 }
2271
2272 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
2273                                    DCTELEM *block, int n, int qscale)
2274 {
2275     int i, level, nCoeffs;
2276     const uint16_t *quant_matrix;
2277     int sum=-1;
2278
2279     if(s->alternate_scan) nCoeffs= 63;
2280     else nCoeffs= s->block_last_index[n];
2281
2282     if (n < 4)
2283         block[0] = block[0] * s->y_dc_scale;
2284     else
2285         block[0] = block[0] * s->c_dc_scale;
2286     quant_matrix = s->intra_matrix;
2287     for(i=1;i<=nCoeffs;i++) {
2288         int j= s->intra_scantable.permutated[i];
2289         level = block[j];
2290         if (level) {
2291             if (level < 0) {
2292                 level = -level;
2293                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2294                 level = -level;
2295             } else {
2296                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2297             }
2298             block[j] = level;
2299             sum+=level;
2300         }
2301     }
2302     block[63]^=sum&1;
2303 }
2304
2305 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
2306                                    DCTELEM *block, int n, int qscale)
2307 {
2308     int i, level, nCoeffs;
2309     const uint16_t *quant_matrix;
2310     int sum=-1;
2311
2312     if(s->alternate_scan) nCoeffs= 63;
2313     else nCoeffs= s->block_last_index[n];
2314
2315     quant_matrix = s->inter_matrix;
2316     for(i=0; i<=nCoeffs; i++) {
2317         int j= s->intra_scantable.permutated[i];
2318         level = block[j];
2319         if (level) {
2320             if (level < 0) {
2321                 level = -level;
2322                 level = (((level << 1) + 1) * qscale *
2323                          ((int) (quant_matrix[j]))) >> 4;
2324                 level = -level;
2325             } else {
2326                 level = (((level << 1) + 1) * qscale *
2327                          ((int) (quant_matrix[j]))) >> 4;
2328             }
2329             block[j] = level;
2330             sum+=level;
2331         }
2332     }
2333     block[63]^=sum&1;
2334 }
2335
2336 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
2337                                   DCTELEM *block, int n, int qscale)
2338 {
2339     int i, level, qmul, qadd;
2340     int nCoeffs;
2341
2342     assert(s->block_last_index[n]>=0);
2343
2344     qmul = qscale << 1;
2345
2346     if (!s->h263_aic) {
2347         if (n < 4)
2348             block[0] = block[0] * s->y_dc_scale;
2349         else
2350             block[0] = block[0] * s->c_dc_scale;
2351         qadd = (qscale - 1) | 1;
2352     }else{
2353         qadd = 0;
2354     }
2355     if(s->ac_pred)
2356         nCoeffs=63;
2357     else
2358         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
2359
2360     for(i=1; i<=nCoeffs; i++) {
2361         level = block[i];
2362         if (level) {
2363             if (level < 0) {
2364                 level = level * qmul - qadd;
2365             } else {
2366                 level = level * qmul + qadd;
2367             }
2368             block[i] = level;
2369         }
2370     }
2371 }
2372
2373 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
2374                                   DCTELEM *block, int n, int qscale)
2375 {
2376     int i, level, qmul, qadd;
2377     int nCoeffs;
2378
2379     assert(s->block_last_index[n]>=0);
2380
2381     qadd = (qscale - 1) | 1;
2382     qmul = qscale << 1;
2383
2384     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
2385
2386     for(i=0; i<=nCoeffs; i++) {
2387         level = block[i];
2388         if (level) {
2389             if (level < 0) {
2390                 level = level * qmul - qadd;
2391             } else {
2392                 level = level * qmul + qadd;
2393             }
2394             block[i] = level;
2395         }
2396     }
2397 }
2398
2399 /**
2400  * set qscale and update qscale dependent variables.
2401  */
2402 void ff_set_qscale(MpegEncContext * s, int qscale)
2403 {
2404     if (qscale < 1)
2405         qscale = 1;
2406     else if (qscale > 31)
2407         qscale = 31;
2408
2409     s->qscale = qscale;
2410     s->chroma_qscale= s->chroma_qscale_table[qscale];
2411
2412     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2413     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2414 }