Crop parameters are unsigned, having them negative could be bad and lead
[ffmpeg.git] / libavcodec / mpegvideo_common.h
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 mpegvideo_common.h
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #ifndef FFMPEG_MPEGVIDEO_COMMON_H
31 #define FFMPEG_MPEGVIDEO_COMMON_H
32
33 #include "avcodec.h"
34 #include "dsputil.h"
35 #include "mpegvideo.h"
36 #include "mjpegenc.h"
37 #include "msmpeg4.h"
38 #include "faandct.h"
39 #include <limits.h>
40
41 int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
42 int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
43 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block);
44 void copy_picture(Picture *dst, Picture *src);
45
46 /**
47  * allocates a Picture
48  * The pixels are allocated/set by calling get_buffer() if shared=0
49  */
50 int alloc_picture(MpegEncContext *s, Picture *pic, int shared);
51
52 /**
53  * sets the given MpegEncContext to common defaults (same for encoding and decoding).
54  * the changed fields will not depend upon the prior state of the MpegEncContext.
55  */
56 void MPV_common_defaults(MpegEncContext *s);
57
58 static inline void gmc1_motion(MpegEncContext *s,
59                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
60                                uint8_t **ref_picture)
61 {
62     uint8_t *ptr;
63     int offset, src_x, src_y, linesize, uvlinesize;
64     int motion_x, motion_y;
65     int emu=0;
66
67     motion_x= s->sprite_offset[0][0];
68     motion_y= s->sprite_offset[0][1];
69     src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
70     src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
71     motion_x<<=(3-s->sprite_warping_accuracy);
72     motion_y<<=(3-s->sprite_warping_accuracy);
73     src_x = av_clip(src_x, -16, s->width);
74     if (src_x == s->width)
75         motion_x =0;
76     src_y = av_clip(src_y, -16, s->height);
77     if (src_y == s->height)
78         motion_y =0;
79
80     linesize = s->linesize;
81     uvlinesize = s->uvlinesize;
82
83     ptr = ref_picture[0] + (src_y * linesize) + src_x;
84
85     if(s->flags&CODEC_FLAG_EMU_EDGE){
86         if(   (unsigned)src_x >= s->h_edge_pos - 17
87            || (unsigned)src_y >= s->v_edge_pos - 17){
88             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
89             ptr= s->edge_emu_buffer;
90         }
91     }
92
93     if((motion_x|motion_y)&7){
94         s->dsp.gmc1(dest_y  , ptr  , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
95         s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
96     }else{
97         int dxy;
98
99         dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
100         if (s->no_rounding){
101             s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
102         }else{
103             s->dsp.put_pixels_tab       [0][dxy](dest_y, ptr, linesize, 16);
104         }
105     }
106
107     if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
108
109     motion_x= s->sprite_offset[1][0];
110     motion_y= s->sprite_offset[1][1];
111     src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
112     src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
113     motion_x<<=(3-s->sprite_warping_accuracy);
114     motion_y<<=(3-s->sprite_warping_accuracy);
115     src_x = av_clip(src_x, -8, s->width>>1);
116     if (src_x == s->width>>1)
117         motion_x =0;
118     src_y = av_clip(src_y, -8, s->height>>1);
119     if (src_y == s->height>>1)
120         motion_y =0;
121
122     offset = (src_y * uvlinesize) + src_x;
123     ptr = ref_picture[1] + offset;
124     if(s->flags&CODEC_FLAG_EMU_EDGE){
125         if(   (unsigned)src_x >= (s->h_edge_pos>>1) - 9
126            || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
127             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
128             ptr= s->edge_emu_buffer;
129             emu=1;
130         }
131     }
132     s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
133
134     ptr = ref_picture[2] + offset;
135     if(emu){
136         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
137         ptr= s->edge_emu_buffer;
138     }
139     s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
140
141     return;
142 }
143
144 static inline void gmc_motion(MpegEncContext *s,
145                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
146                                uint8_t **ref_picture)
147 {
148     uint8_t *ptr;
149     int linesize, uvlinesize;
150     const int a= s->sprite_warping_accuracy;
151     int ox, oy;
152
153     linesize = s->linesize;
154     uvlinesize = s->uvlinesize;
155
156     ptr = ref_picture[0];
157
158     ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
159     oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
160
161     s->dsp.gmc(dest_y, ptr, linesize, 16,
162            ox,
163            oy,
164            s->sprite_delta[0][0], s->sprite_delta[0][1],
165            s->sprite_delta[1][0], s->sprite_delta[1][1],
166            a+1, (1<<(2*a+1)) - s->no_rounding,
167            s->h_edge_pos, s->v_edge_pos);
168     s->dsp.gmc(dest_y+8, ptr, linesize, 16,
169            ox + s->sprite_delta[0][0]*8,
170            oy + s->sprite_delta[1][0]*8,
171            s->sprite_delta[0][0], s->sprite_delta[0][1],
172            s->sprite_delta[1][0], s->sprite_delta[1][1],
173            a+1, (1<<(2*a+1)) - s->no_rounding,
174            s->h_edge_pos, s->v_edge_pos);
175
176     if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
177
178     ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
179     oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
180
181     ptr = ref_picture[1];
182     s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
183            ox,
184            oy,
185            s->sprite_delta[0][0], s->sprite_delta[0][1],
186            s->sprite_delta[1][0], s->sprite_delta[1][1],
187            a+1, (1<<(2*a+1)) - s->no_rounding,
188            s->h_edge_pos>>1, s->v_edge_pos>>1);
189
190     ptr = ref_picture[2];
191     s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
192            ox,
193            oy,
194            s->sprite_delta[0][0], s->sprite_delta[0][1],
195            s->sprite_delta[1][0], s->sprite_delta[1][1],
196            a+1, (1<<(2*a+1)) - s->no_rounding,
197            s->h_edge_pos>>1, s->v_edge_pos>>1);
198 }
199
200 static inline int hpel_motion(MpegEncContext *s,
201                                   uint8_t *dest, uint8_t *src,
202                                   int field_based, int field_select,
203                                   int src_x, int src_y,
204                                   int width, int height, int stride,
205                                   int h_edge_pos, int v_edge_pos,
206                                   int w, int h, op_pixels_func *pix_op,
207                                   int motion_x, int motion_y)
208 {
209     int dxy;
210     int emu=0;
211
212     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
213     src_x += motion_x >> 1;
214     src_y += motion_y >> 1;
215
216     /* WARNING: do no forget half pels */
217     src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
218     if (src_x == width)
219         dxy &= ~1;
220     src_y = av_clip(src_y, -16, height);
221     if (src_y == height)
222         dxy &= ~2;
223     src += src_y * stride + src_x;
224
225     if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
226         if(   (unsigned)src_x > h_edge_pos - (motion_x&1) - w
227            || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
228             ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
229                              src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
230             src= s->edge_emu_buffer;
231             emu=1;
232         }
233     }
234     if(field_select)
235         src += s->linesize;
236     pix_op[dxy](dest, src, stride, h);
237     return emu;
238 }
239
240 /* apply one mpeg motion vector to the three components */
241 static av_always_inline
242 void mpeg_motion(MpegEncContext *s,
243                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
244                  int field_based, int bottom_field, int field_select,
245                  uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
246                  int motion_x, int motion_y, int h)
247 {
248     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
249     int dxy, uvdxy, mx, my, src_x, src_y,
250         uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
251
252 #if 0
253 if(s->quarter_sample)
254 {
255     motion_x>>=1;
256     motion_y>>=1;
257 }
258 #endif
259
260     v_edge_pos = s->v_edge_pos >> field_based;
261     linesize   = s->current_picture.linesize[0] << field_based;
262     uvlinesize = s->current_picture.linesize[1] << field_based;
263
264     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
265     src_x = s->mb_x* 16               + (motion_x >> 1);
266     src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1);
267
268     if (s->out_format == FMT_H263) {
269         if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
270             mx = (motion_x>>1)|(motion_x&1);
271             my = motion_y >>1;
272             uvdxy = ((my & 1) << 1) | (mx & 1);
273             uvsrc_x = s->mb_x* 8               + (mx >> 1);
274             uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
275         }else{
276             uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
277             uvsrc_x = src_x>>1;
278             uvsrc_y = src_y>>1;
279         }
280     }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
281         mx = motion_x / 4;
282         my = motion_y / 4;
283         uvdxy = 0;
284         uvsrc_x = s->mb_x*8 + mx;
285         uvsrc_y = s->mb_y*8 + my;
286     } else {
287         if(s->chroma_y_shift){
288             mx = motion_x / 2;
289             my = motion_y / 2;
290             uvdxy = ((my & 1) << 1) | (mx & 1);
291             uvsrc_x = s->mb_x* 8               + (mx >> 1);
292             uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
293         } else {
294             if(s->chroma_x_shift){
295             //Chroma422
296                 mx = motion_x / 2;
297                 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
298                 uvsrc_x = s->mb_x* 8           + (mx >> 1);
299                 uvsrc_y = src_y;
300             } else {
301             //Chroma444
302                 uvdxy = dxy;
303                 uvsrc_x = src_x;
304                 uvsrc_y = src_y;
305             }
306         }
307     }
308
309     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
310     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
311     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
312
313     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16
314        || (unsigned)src_y >    v_edge_pos - (motion_y&1) - h){
315             if(s->codec_id == CODEC_ID_MPEG2VIDEO ||
316                s->codec_id == CODEC_ID_MPEG1VIDEO){
317                 av_log(s->avctx,AV_LOG_DEBUG,
318                         "MPEG motion vector out of boundary\n");
319                 return ;
320             }
321             ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
322                                 17, 17+field_based,
323                                 src_x, src_y<<field_based,
324                                 s->h_edge_pos, s->v_edge_pos);
325             ptr_y = s->edge_emu_buffer;
326             if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
327                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
328                 ff_emulated_edge_mc(uvbuf ,
329                                     ptr_cb, s->uvlinesize,
330                                     9, 9+field_based,
331                                     uvsrc_x, uvsrc_y<<field_based,
332                                     s->h_edge_pos>>1, s->v_edge_pos>>1);
333                 ff_emulated_edge_mc(uvbuf+16,
334                                     ptr_cr, s->uvlinesize,
335                                     9, 9+field_based,
336                                     uvsrc_x, uvsrc_y<<field_based,
337                                     s->h_edge_pos>>1, s->v_edge_pos>>1);
338                 ptr_cb= uvbuf;
339                 ptr_cr= uvbuf+16;
340             }
341     }
342
343     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
344         dest_y += s->linesize;
345         dest_cb+= s->uvlinesize;
346         dest_cr+= s->uvlinesize;
347     }
348
349     if(field_select){
350         ptr_y += s->linesize;
351         ptr_cb+= s->uvlinesize;
352         ptr_cr+= s->uvlinesize;
353     }
354
355     pix_op[0][dxy](dest_y, ptr_y, linesize, h);
356
357     if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
358         pix_op[s->chroma_x_shift][uvdxy]
359                 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
360         pix_op[s->chroma_x_shift][uvdxy]
361                 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
362     }
363     if((ENABLE_H261_ENCODER || ENABLE_H261_DECODER) &&
364          s->out_format == FMT_H261){
365         ff_h261_loop_filter(s);
366     }
367 }
368
369 //FIXME move to dsputil, avg variant, 16x16 version
370 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
371     int x;
372     uint8_t * const top   = src[1];
373     uint8_t * const left  = src[2];
374     uint8_t * const mid   = src[0];
375     uint8_t * const right = src[3];
376     uint8_t * const bottom= src[4];
377 #define OBMC_FILTER(x, t, l, m, r, b)\
378     dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
379 #define OBMC_FILTER4(x, t, l, m, r, b)\
380     OBMC_FILTER(x         , t, l, m, r, b);\
381     OBMC_FILTER(x+1       , t, l, m, r, b);\
382     OBMC_FILTER(x  +stride, t, l, m, r, b);\
383     OBMC_FILTER(x+1+stride, t, l, m, r, b);
384
385     x=0;
386     OBMC_FILTER (x  , 2, 2, 4, 0, 0);
387     OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
388     OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
389     OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
390     OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
391     OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
392     x+= stride;
393     OBMC_FILTER (x  , 1, 2, 5, 0, 0);
394     OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
395     OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
396     OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
397     x+= stride;
398     OBMC_FILTER4(x  , 1, 2, 5, 0, 0);
399     OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
400     OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
401     OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
402     x+= 2*stride;
403     OBMC_FILTER4(x  , 0, 2, 5, 0, 1);
404     OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
405     OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
406     OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
407     x+= 2*stride;
408     OBMC_FILTER (x  , 0, 2, 5, 0, 1);
409     OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
410     OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
411     OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
412     OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
413     OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
414     x+= stride;
415     OBMC_FILTER (x  , 0, 2, 4, 0, 2);
416     OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
417     OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
418     OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
419 }
420
421 /* obmc for 1 8x8 luma block */
422 static inline void obmc_motion(MpegEncContext *s,
423                                uint8_t *dest, uint8_t *src,
424                                int src_x, int src_y,
425                                op_pixels_func *pix_op,
426                                int16_t mv[5][2]/* mid top left right bottom*/)
427 #define MID    0
428 {
429     int i;
430     uint8_t *ptr[5];
431
432     assert(s->quarter_sample==0);
433
434     for(i=0; i<5; i++){
435         if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
436             ptr[i]= ptr[MID];
437         }else{
438             ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
439             hpel_motion(s, ptr[i], src, 0, 0,
440                         src_x, src_y,
441                         s->width, s->height, s->linesize,
442                         s->h_edge_pos, s->v_edge_pos,
443                         8, 8, pix_op,
444                         mv[i][0], mv[i][1]);
445         }
446     }
447
448     put_obmc(dest, ptr, s->linesize);
449 }
450
451 static inline void qpel_motion(MpegEncContext *s,
452                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
453                                int field_based, int bottom_field, int field_select,
454                                uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
455                                qpel_mc_func (*qpix_op)[16],
456                                int motion_x, int motion_y, int h)
457 {
458     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
459     int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
460
461     dxy = ((motion_y & 3) << 2) | (motion_x & 3);
462     src_x = s->mb_x *  16                 + (motion_x >> 2);
463     src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
464
465     v_edge_pos = s->v_edge_pos >> field_based;
466     linesize = s->linesize << field_based;
467     uvlinesize = s->uvlinesize << field_based;
468
469     if(field_based){
470         mx= motion_x/2;
471         my= motion_y>>1;
472     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
473         static const int rtab[8]= {0,0,1,1,0,0,0,1};
474         mx= (motion_x>>1) + rtab[motion_x&7];
475         my= (motion_y>>1) + rtab[motion_y&7];
476     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
477         mx= (motion_x>>1)|(motion_x&1);
478         my= (motion_y>>1)|(motion_y&1);
479     }else{
480         mx= motion_x/2;
481         my= motion_y/2;
482     }
483     mx= (mx>>1)|(mx&1);
484     my= (my>>1)|(my&1);
485
486     uvdxy= (mx&1) | ((my&1)<<1);
487     mx>>=1;
488     my>>=1;
489
490     uvsrc_x = s->mb_x *  8                 + mx;
491     uvsrc_y = s->mb_y * (8 >> field_based) + my;
492
493     ptr_y  = ref_picture[0] +   src_y *   linesize +   src_x;
494     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
495     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
496
497     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16
498        || (unsigned)src_y >    v_edge_pos - (motion_y&3) - h  ){
499         ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
500                             17, 17+field_based, src_x, src_y<<field_based,
501                             s->h_edge_pos, s->v_edge_pos);
502         ptr_y= s->edge_emu_buffer;
503         if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
504             uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
505             ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
506                                 9, 9 + field_based,
507                                 uvsrc_x, uvsrc_y<<field_based,
508                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
509             ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
510                                 9, 9 + field_based,
511                                 uvsrc_x, uvsrc_y<<field_based,
512                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
513             ptr_cb= uvbuf;
514             ptr_cr= uvbuf + 16;
515         }
516     }
517
518     if(!field_based)
519         qpix_op[0][dxy](dest_y, ptr_y, linesize);
520     else{
521         if(bottom_field){
522             dest_y += s->linesize;
523             dest_cb+= s->uvlinesize;
524             dest_cr+= s->uvlinesize;
525         }
526
527         if(field_select){
528             ptr_y  += s->linesize;
529             ptr_cb += s->uvlinesize;
530             ptr_cr += s->uvlinesize;
531         }
532         //damn interlaced mode
533         //FIXME boundary mirroring is not exactly correct here
534         qpix_op[1][dxy](dest_y  , ptr_y  , linesize);
535         qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
536     }
537     if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
538         pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
539         pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
540     }
541 }
542
543 /**
544  * h263 chroma 4mv motion compensation.
545  */
546 static inline void chroma_4mv_motion(MpegEncContext *s,
547                                      uint8_t *dest_cb, uint8_t *dest_cr,
548                                      uint8_t **ref_picture,
549                                      op_pixels_func *pix_op,
550                                      int mx, int my){
551     int dxy, emu=0, src_x, src_y, offset;
552     uint8_t *ptr;
553
554     /* In case of 8X8, we construct a single chroma motion vector
555        with a special rounding */
556     mx= ff_h263_round_chroma(mx);
557     my= ff_h263_round_chroma(my);
558
559     dxy = ((my & 1) << 1) | (mx & 1);
560     mx >>= 1;
561     my >>= 1;
562
563     src_x = s->mb_x * 8 + mx;
564     src_y = s->mb_y * 8 + my;
565     src_x = av_clip(src_x, -8, s->width/2);
566     if (src_x == s->width/2)
567         dxy &= ~1;
568     src_y = av_clip(src_y, -8, s->height/2);
569     if (src_y == s->height/2)
570         dxy &= ~2;
571
572     offset = (src_y * (s->uvlinesize)) + src_x;
573     ptr = ref_picture[1] + offset;
574     if(s->flags&CODEC_FLAG_EMU_EDGE){
575         if(   (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
576            || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
577             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
578                                 9, 9, src_x, src_y,
579                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
580             ptr= s->edge_emu_buffer;
581             emu=1;
582         }
583     }
584     pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
585
586     ptr = ref_picture[2] + offset;
587     if(emu){
588         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
589                             9, 9, src_x, src_y,
590                             s->h_edge_pos>>1, s->v_edge_pos>>1);
591         ptr= s->edge_emu_buffer;
592     }
593     pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
594 }
595
596 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
597     /* fetch pixels for estimated mv 4 macroblocks ahead
598      * optimized for 64byte cache lines */
599     const int shift = s->quarter_sample ? 2 : 1;
600     const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
601     const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
602     int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
603     s->dsp.prefetch(pix[0]+off, s->linesize, 4);
604     off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
605     s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
606 }
607
608 /**
609  * motion compensation of a single macroblock
610  * @param s context
611  * @param dest_y luma destination pointer
612  * @param dest_cb chroma cb/u destination pointer
613  * @param dest_cr chroma cr/v destination pointer
614  * @param dir direction (0->forward, 1->backward)
615  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
616  * @param pic_op halfpel motion compensation function (average or put normally)
617  * @param pic_op qpel motion compensation function (average or put normally)
618  * the motion vectors are taken from s->mv and the MV type from s->mv_type
619  */
620 static inline void MPV_motion(MpegEncContext *s,
621                               uint8_t *dest_y, uint8_t *dest_cb,
622                               uint8_t *dest_cr, int dir,
623                               uint8_t **ref_picture,
624                               op_pixels_func (*pix_op)[4],
625                               qpel_mc_func (*qpix_op)[16])
626 {
627     int dxy, mx, my, src_x, src_y, motion_x, motion_y;
628     int mb_x, mb_y, i;
629     uint8_t *ptr, *dest;
630
631     mb_x = s->mb_x;
632     mb_y = s->mb_y;
633
634     prefetch_motion(s, ref_picture, dir);
635
636     if(s->obmc && s->pict_type != FF_B_TYPE){
637         int16_t mv_cache[4][4][2];
638         const int xy= s->mb_x + s->mb_y*s->mb_stride;
639         const int mot_stride= s->b8_stride;
640         const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
641
642         assert(!s->mb_skipped);
643
644         memcpy(mv_cache[1][1], s->current_picture.motion_val[0][mot_xy           ], sizeof(int16_t)*4);
645         memcpy(mv_cache[2][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
646         memcpy(mv_cache[3][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
647
648         if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){
649             memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
650         }else{
651             memcpy(mv_cache[0][1], s->current_picture.motion_val[0][mot_xy-mot_stride], sizeof(int16_t)*4);
652         }
653
654         if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){
655             *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1];
656             *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1];
657         }else{
658             *(int32_t*)mv_cache[1][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1];
659             *(int32_t*)mv_cache[2][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1+mot_stride];
660         }
661
662         if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){
663             *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2];
664             *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2];
665         }else{
666             *(int32_t*)mv_cache[1][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2];
667             *(int32_t*)mv_cache[2][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2+mot_stride];
668         }
669
670         mx = 0;
671         my = 0;
672         for(i=0;i<4;i++) {
673             const int x= (i&1)+1;
674             const int y= (i>>1)+1;
675             int16_t mv[5][2]= {
676                 {mv_cache[y][x  ][0], mv_cache[y][x  ][1]},
677                 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
678                 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
679                 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
680                 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
681             //FIXME cleanup
682             obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
683                         ref_picture[0],
684                         mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
685                         pix_op[1],
686                         mv);
687
688             mx += mv[0][0];
689             my += mv[0][1];
690         }
691         if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY))
692             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
693
694         return;
695     }
696
697     switch(s->mv_type) {
698     case MV_TYPE_16X16:
699         if(s->mcsel){
700             if(s->real_sprite_warping_points==1){
701                 gmc1_motion(s, dest_y, dest_cb, dest_cr,
702                             ref_picture);
703             }else{
704                 gmc_motion(s, dest_y, dest_cb, dest_cr,
705                             ref_picture);
706             }
707         }else if(s->quarter_sample){
708             qpel_motion(s, dest_y, dest_cb, dest_cr,
709                         0, 0, 0,
710                         ref_picture, pix_op, qpix_op,
711                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
712         }else if(ENABLE_WMV2 && s->mspel){
713             ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
714                         ref_picture, pix_op,
715                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
716         }else
717         {
718             mpeg_motion(s, dest_y, dest_cb, dest_cr,
719                         0, 0, 0,
720                         ref_picture, pix_op,
721                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
722         }
723         break;
724     case MV_TYPE_8X8:
725         mx = 0;
726         my = 0;
727         if(s->quarter_sample){
728             for(i=0;i<4;i++) {
729                 motion_x = s->mv[dir][i][0];
730                 motion_y = s->mv[dir][i][1];
731
732                 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
733                 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
734                 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
735
736                 /* WARNING: do no forget half pels */
737                 src_x = av_clip(src_x, -16, s->width);
738                 if (src_x == s->width)
739                     dxy &= ~3;
740                 src_y = av_clip(src_y, -16, s->height);
741                 if (src_y == s->height)
742                     dxy &= ~12;
743
744                 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
745                 if(s->flags&CODEC_FLAG_EMU_EDGE){
746                     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8
747                        || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){
748                         ff_emulated_edge_mc(s->edge_emu_buffer, ptr,
749                                             s->linesize, 9, 9,
750                                             src_x, src_y,
751                                             s->h_edge_pos, s->v_edge_pos);
752                         ptr= s->edge_emu_buffer;
753                     }
754                 }
755                 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
756                 qpix_op[1][dxy](dest, ptr, s->linesize);
757
758                 mx += s->mv[dir][i][0]/2;
759                 my += s->mv[dir][i][1]/2;
760             }
761         }else{
762             for(i=0;i<4;i++) {
763                 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
764                             ref_picture[0], 0, 0,
765                             mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
766                             s->width, s->height, s->linesize,
767                             s->h_edge_pos, s->v_edge_pos,
768                             8, 8, pix_op[1],
769                             s->mv[dir][i][0], s->mv[dir][i][1]);
770
771                 mx += s->mv[dir][i][0];
772                 my += s->mv[dir][i][1];
773             }
774         }
775
776         if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY))
777             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
778         break;
779     case MV_TYPE_FIELD:
780         if (s->picture_structure == PICT_FRAME) {
781             if(s->quarter_sample){
782                 for(i=0; i<2; i++){
783                     qpel_motion(s, dest_y, dest_cb, dest_cr,
784                                 1, i, s->field_select[dir][i],
785                                 ref_picture, pix_op, qpix_op,
786                                 s->mv[dir][i][0], s->mv[dir][i][1], 8);
787                 }
788             }else{
789                 /* top field */
790                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
791                             1, 0, s->field_select[dir][0],
792                             ref_picture, pix_op,
793                             s->mv[dir][0][0], s->mv[dir][0][1], 8);
794                 /* bottom field */
795                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
796                             1, 1, s->field_select[dir][1],
797                             ref_picture, pix_op,
798                             s->mv[dir][1][0], s->mv[dir][1][1], 8);
799             }
800         } else {
801             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
802                 ref_picture= s->current_picture_ptr->data;
803             }
804
805             mpeg_motion(s, dest_y, dest_cb, dest_cr,
806                         0, 0, s->field_select[dir][0],
807                         ref_picture, pix_op,
808                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
809         }
810         break;
811     case MV_TYPE_16X8:
812         for(i=0; i<2; i++){
813             uint8_t ** ref2picture;
814
815             if(s->picture_structure == s->field_select[dir][i] + 1
816                || s->pict_type == FF_B_TYPE || s->first_field){
817                 ref2picture= ref_picture;
818             }else{
819                 ref2picture= s->current_picture_ptr->data;
820             }
821
822             mpeg_motion(s, dest_y, dest_cb, dest_cr,
823                         0, 0, s->field_select[dir][i],
824                         ref2picture, pix_op,
825                         s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8);
826
827             dest_y += 16*s->linesize;
828             dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
829             dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
830         }
831         break;
832     case MV_TYPE_DMV:
833         if(s->picture_structure == PICT_FRAME){
834             for(i=0; i<2; i++){
835                 int j;
836                 for(j=0; j<2; j++){
837                     mpeg_motion(s, dest_y, dest_cb, dest_cr,
838                                 1, j, j^i,
839                                 ref_picture, pix_op,
840                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8);
841                 }
842                 pix_op = s->dsp.avg_pixels_tab;
843             }
844         }else{
845             for(i=0; i<2; i++){
846                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
847                             0, 0, s->picture_structure != i+1,
848                             ref_picture, pix_op,
849                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],16);
850
851                 // after put we make avg of the same block
852                 pix_op=s->dsp.avg_pixels_tab;
853
854                 //opposite parity is always in the same frame if this is second field
855                 if(!s->first_field){
856                     ref_picture = s->current_picture_ptr->data;
857                 }
858             }
859         }
860     break;
861     default: assert(0);
862     }
863 }
864
865 #endif /* FFMPEG_MPEGVIDEO_COMMON_H */