FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
motion_est.c
Go to the documentation of this file.
1 /*
2  * Motion estimation
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer
5  *
6  * new motion estimation (X1/EPZS) 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  * Motion estimation.
28  */
29 
30 #include <stdlib.h>
31 #include <stdio.h>
32 #include <limits.h>
33 
34 #include "avcodec.h"
35 #include "mathops.h"
36 #include "mpegutils.h"
37 #include "mpegvideo.h"
38 
39 #undef NDEBUG
40 #include <assert.h>
41 
42 #define P_LEFT P[1]
43 #define P_TOP P[2]
44 #define P_TOPRIGHT P[3]
45 #define P_MEDIAN P[4]
46 #define P_MV1 P[9]
47 
48 #define ME_MAP_SHIFT 3
49 #define ME_MAP_MV_BITS 11
50 
51 static int sad_hpel_motion_search(MpegEncContext * s,
52  int *mx_ptr, int *my_ptr, int dmin,
53  int src_index, int ref_index,
54  int size, int h);
55 
56 static inline unsigned update_map_generation(MotionEstContext *c)
57 {
58  c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
59  if(c->map_generation==0){
60  c->map_generation= 1<<(ME_MAP_MV_BITS*2);
61  memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
62  }
63  return c->map_generation;
64 }
65 
66 /* shape adaptive search stuff */
67 typedef struct Minima{
68  int height;
69  int x, y;
70  int checked;
71 }Minima;
72 
73 static int minima_cmp(const void *a, const void *b){
74  const Minima *da = (const Minima *) a;
75  const Minima *db = (const Minima *) b;
76 
77  return da->height - db->height;
78 }
79 
80 #define FLAG_QPEL 1 //must be 1
81 #define FLAG_CHROMA 2
82 #define FLAG_DIRECT 4
83 
84 static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
85  const int offset[3]= {
86  y*c-> stride + x,
87  ((y*c->uvstride + x)>>1),
88  ((y*c->uvstride + x)>>1),
89  };
90  int i;
91  for(i=0; i<3; i++){
92  c->src[0][i]= src [i] + offset[i];
93  c->ref[0][i]= ref [i] + offset[i];
94  }
95  if(ref_index){
96  for(i=0; i<3; i++){
97  c->ref[ref_index][i]= ref2[i] + offset[i];
98  }
99  }
100 }
101 
102 static int get_flags(MotionEstContext *c, int direct, int chroma){
103  return ((c->avctx->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
104  + (direct ? FLAG_DIRECT : 0)
105  + (chroma ? FLAG_CHROMA : 0);
106 }
107 
108 static av_always_inline int cmp_direct_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
109  const int size, const int h, int ref_index, int src_index,
110  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel){
111  MotionEstContext * const c= &s->me;
112  const int stride= c->stride;
113  const int hx= subx + (x<<(1+qpel));
114  const int hy= suby + (y<<(1+qpel));
115  uint8_t * const * const ref= c->ref[ref_index];
116  uint8_t * const * const src= c->src[src_index];
117  int d;
118  //FIXME check chroma 4mv, (no crashes ...)
119  av_assert2(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));
120  if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
121  const int time_pp= s->pp_time;
122  const int time_pb= s->pb_time;
123  const int mask= 2*qpel+1;
124  if(s->mv_type==MV_TYPE_8X8){
125  int i;
126  for(i=0; i<4; i++){
127  int fx = c->direct_basis_mv[i][0] + hx;
128  int fy = c->direct_basis_mv[i][1] + hy;
129  int bx = hx ? fx - c->co_located_mv[i][0] : c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(qpel+4));
130  int by = hy ? fy - c->co_located_mv[i][1] : c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(qpel+4));
131  int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
132  int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
133 
134  uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);
135  if(qpel){
136  c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);
137  c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);
138  }else{
139  c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);
140  c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);
141  }
142  }
143  }else{
144  int fx = c->direct_basis_mv[0][0] + hx;
145  int fy = c->direct_basis_mv[0][1] + hy;
146  int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
147  int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
148  int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
149  int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
150 
151  if(qpel){
152  c->qpel_put[1][fxy](c->temp , ref[0] + (fx>>2) + (fy>>2)*stride , stride);
153  c->qpel_put[1][fxy](c->temp + 8 , ref[0] + (fx>>2) + (fy>>2)*stride + 8 , stride);
154  c->qpel_put[1][fxy](c->temp + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8*stride, stride);
155  c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);
156  c->qpel_avg[1][bxy](c->temp , ref[8] + (bx>>2) + (by>>2)*stride , stride);
157  c->qpel_avg[1][bxy](c->temp + 8 , ref[8] + (bx>>2) + (by>>2)*stride + 8 , stride);
158  c->qpel_avg[1][bxy](c->temp + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8*stride, stride);
159  c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);
160  }else{
161  av_assert2((fx>>1) + 16*s->mb_x >= -16);
162  av_assert2((fy>>1) + 16*s->mb_y >= -16);
163  av_assert2((fx>>1) + 16*s->mb_x <= s->width);
164  av_assert2((fy>>1) + 16*s->mb_y <= s->height);
165  av_assert2((bx>>1) + 16*s->mb_x >= -16);
166  av_assert2((by>>1) + 16*s->mb_y >= -16);
167  av_assert2((bx>>1) + 16*s->mb_x <= s->width);
168  av_assert2((by>>1) + 16*s->mb_y <= s->height);
169 
170  c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);
171  c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);
172  }
173  }
174  d = cmp_func(s, c->temp, src[0], stride, 16);
175  }else
176  d= 256*256*256*32;
177  return d;
178 }
179 
180 static av_always_inline int cmp_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
181  const int size, const int h, int ref_index, int src_index,
182  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel, int chroma){
183  MotionEstContext * const c= &s->me;
184  const int stride= c->stride;
185  const int uvstride= c->uvstride;
186  const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
187  const int hx= subx + (x<<(1+qpel));
188  const int hy= suby + (y<<(1+qpel));
189  uint8_t * const * const ref= c->ref[ref_index];
190  uint8_t * const * const src= c->src[src_index];
191  int d;
192  //FIXME check chroma 4mv, (no crashes ...)
193  int uvdxy; /* no, it might not be used uninitialized */
194  if(dxy){
195  if(qpel){
196  c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
197  if(chroma){
198  int cx= hx/2;
199  int cy= hy/2;
200  cx= (cx>>1)|(cx&1);
201  cy= (cy>>1)|(cy&1);
202  uvdxy= (cx&1) + 2*(cy&1);
203  //FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264
204  }
205  }else{
206  c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
207  if(chroma)
208  uvdxy= dxy | (x&1) | (2*(y&1));
209  }
210  d = cmp_func(s, c->temp, src[0], stride, h);
211  }else{
212  d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
213  if(chroma)
214  uvdxy= (x&1) + 2*(y&1);
215  }
216  if(chroma){
217  uint8_t * const uvtemp= c->temp + 16*stride;
218  c->hpel_put[size+1][uvdxy](uvtemp , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
219  c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
220  d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1);
221  d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
222  }
223  return d;
224 }
225 
226 static int cmp_simple(MpegEncContext *s, const int x, const int y,
227  int ref_index, int src_index,
228  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){
229  return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0);
230 }
231 
232 static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y,
233  const int size, const int h, int ref_index, int src_index,
234  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
235  if(flags&FLAG_DIRECT){
236  return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
237  }else{
238  return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
239  }
240 }
241 
242 static int cmp_internal(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
243  const int size, const int h, int ref_index, int src_index,
244  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
245  if(flags&FLAG_DIRECT){
246  return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
247  }else{
248  return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL, flags&FLAG_CHROMA);
249  }
250 }
251 
252 /** @brief compares a block (either a full macroblock or a partition thereof)
253  against a proposed motion-compensated prediction of that block
254  */
255 static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
256  const int size, const int h, int ref_index, int src_index,
257  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
258  if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size)
259  && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
260  && flags==0 && h==16 && size==0 && subx==0 && suby==0){
261  return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func);
262  }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
263  && subx==0 && suby==0){
264  return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags);
265  }else{
266  return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags);
267  }
268 }
269 
270 static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
271  const int size, const int h, int ref_index, int src_index,
272  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
273  if(flags&FLAG_DIRECT){
274  return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0);
275  }else{
276  return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
277  }
278 }
279 
280 static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
281  const int size, const int h, int ref_index, int src_index,
282  me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
283  if(flags&FLAG_DIRECT){
284  return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1);
285  }else{
286  return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA);
287  }
288 }
289 
290 #include "motion_est_template.c"
291 
292 static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
293  int stride, int h)
294 {
295  return 0;
296 }
297 
298 static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){
299 }
300 
301 int ff_init_me(MpegEncContext *s){
302  MotionEstContext * const c= &s->me;
303  int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
304  int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
305 
306  if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -FFMIN(ME_MAP_SIZE, MAX_SAB_SIZE)){
307  av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n");
308  return -1;
309  }
310  //special case of snow is needed because snow uses its own iterative ME code
311  if(s->me_method!=ME_ZERO && s->me_method!=ME_EPZS && s->me_method!=ME_X1 && s->avctx->codec_id != AV_CODEC_ID_SNOW){
312  av_log(s->avctx, AV_LOG_ERROR, "me_method is only allowed to be set to zero and epzs; for hex,umh,full and others see dia_size\n");
313  return -1;
314  }
315 
316  c->avctx= s->avctx;
317 
318  if(cache_size < 2*dia_size && !c->stride){
319  av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
320  }
321 
322  ff_set_cmp(&s->mecc, s->mecc.me_pre_cmp, c->avctx->me_pre_cmp);
323  ff_set_cmp(&s->mecc, s->mecc.me_cmp, c->avctx->me_cmp);
324  ff_set_cmp(&s->mecc, s->mecc.me_sub_cmp, c->avctx->me_sub_cmp);
325  ff_set_cmp(&s->mecc, s->mecc.mb_cmp, c->avctx->mb_cmp);
326 
327  c->flags = get_flags(c, 0, c->avctx->me_cmp &FF_CMP_CHROMA);
328  c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
329  c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp &FF_CMP_CHROMA);
330 
331 /*FIXME s->no_rounding b_type*/
332  if(s->flags&CODEC_FLAG_QPEL){
333  c->sub_motion_search= qpel_motion_search;
334  c->qpel_avg = s->qdsp.avg_qpel_pixels_tab;
335  if (s->no_rounding)
336  c->qpel_put = s->qdsp.put_no_rnd_qpel_pixels_tab;
337  else
338  c->qpel_put = s->qdsp.put_qpel_pixels_tab;
339  }else{
340  if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
341  c->sub_motion_search= hpel_motion_search;
342  else if( c->avctx->me_sub_cmp == FF_CMP_SAD
343  && c->avctx-> me_cmp == FF_CMP_SAD
344  && c->avctx-> mb_cmp == FF_CMP_SAD)
345  c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
346  else
347  c->sub_motion_search= hpel_motion_search;
348  }
349  c->hpel_avg = s->hdsp.avg_pixels_tab;
350  if (s->no_rounding)
351  c->hpel_put = s->hdsp.put_no_rnd_pixels_tab;
352  else
353  c->hpel_put = s->hdsp.put_pixels_tab;
354 
355  if(s->linesize){
356  c->stride = s->linesize;
357  c->uvstride= s->uvlinesize;
358  }else{
359  c->stride = 16*s->mb_width + 32;
360  c->uvstride= 8*s->mb_width + 16;
361  }
362 
363  /* 8x8 fullpel search would need a 4x4 chroma compare, which we do
364  * not have yet, and even if we had, the motion estimation code
365  * does not expect it. */
366  if (s->codec_id != AV_CODEC_ID_SNOW) {
367  if ((c->avctx->me_cmp & FF_CMP_CHROMA) /* && !s->mecc.me_cmp[2] */)
368  s->mecc.me_cmp[2] = zero_cmp;
369  if ((c->avctx->me_sub_cmp & FF_CMP_CHROMA) && !s->mecc.me_sub_cmp[2])
370  s->mecc.me_sub_cmp[2] = zero_cmp;
371  c->hpel_put[2][0]= c->hpel_put[2][1]=
372  c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
373  }
374 
375  if(s->codec_id == AV_CODEC_ID_H261){
376  c->sub_motion_search= no_sub_motion_search;
377  }
378 
379  return 0;
380 }
381 
382 #define CHECK_SAD_HALF_MV(suffix, x, y) \
383 {\
384  d = s->mecc.pix_abs[size][(x ? 1 : 0) + (y ? 2 : 0)](NULL, pix, ptr + ((x) >> 1), stride, h); \
385  d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
386  COPY3_IF_LT(dminh, d, dx, x, dy, y)\
387 }
388 
389 static int sad_hpel_motion_search(MpegEncContext * s,
390  int *mx_ptr, int *my_ptr, int dmin,
391  int src_index, int ref_index,
392  int size, int h)
393 {
394  MotionEstContext * const c= &s->me;
395  const int penalty_factor= c->sub_penalty_factor;
396  int mx, my, dminh;
397  uint8_t *pix, *ptr;
398  int stride= c->stride;
399  LOAD_COMMON
400 
401  av_assert2(c->sub_flags == 0);
402 
403  if(c->skip){
404  *mx_ptr = 0;
405  *my_ptr = 0;
406  return dmin;
407  }
408 
409  pix = c->src[src_index][0];
410 
411  mx = *mx_ptr;
412  my = *my_ptr;
413  ptr = c->ref[ref_index][0] + (my * stride) + mx;
414 
415  dminh = dmin;
416 
417  if (mx > xmin && mx < xmax &&
418  my > ymin && my < ymax) {
419  int dx=0, dy=0;
420  int d, pen_x, pen_y;
421  const int index= (my<<ME_MAP_SHIFT) + mx;
422  const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
423  const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
424  const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)];
425  const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
426  mx<<=1;
427  my<<=1;
428 
429 
430  pen_x= pred_x + mx;
431  pen_y= pred_y + my;
432 
433  ptr-= stride;
434  if(t<=b){
435  CHECK_SAD_HALF_MV(y2 , 0, -1)
436  if(l<=r){
437  CHECK_SAD_HALF_MV(xy2, -1, -1)
438  if(t+r<=b+l){
439  CHECK_SAD_HALF_MV(xy2, +1, -1)
440  ptr+= stride;
441  }else{
442  ptr+= stride;
443  CHECK_SAD_HALF_MV(xy2, -1, +1)
444  }
445  CHECK_SAD_HALF_MV(x2 , -1, 0)
446  }else{
447  CHECK_SAD_HALF_MV(xy2, +1, -1)
448  if(t+l<=b+r){
449  CHECK_SAD_HALF_MV(xy2, -1, -1)
450  ptr+= stride;
451  }else{
452  ptr+= stride;
453  CHECK_SAD_HALF_MV(xy2, +1, +1)
454  }
455  CHECK_SAD_HALF_MV(x2 , +1, 0)
456  }
457  }else{
458  if(l<=r){
459  if(t+l<=b+r){
460  CHECK_SAD_HALF_MV(xy2, -1, -1)
461  ptr+= stride;
462  }else{
463  ptr+= stride;
464  CHECK_SAD_HALF_MV(xy2, +1, +1)
465  }
466  CHECK_SAD_HALF_MV(x2 , -1, 0)
467  CHECK_SAD_HALF_MV(xy2, -1, +1)
468  }else{
469  if(t+r<=b+l){
470  CHECK_SAD_HALF_MV(xy2, +1, -1)
471  ptr+= stride;
472  }else{
473  ptr+= stride;
474  CHECK_SAD_HALF_MV(xy2, -1, +1)
475  }
476  CHECK_SAD_HALF_MV(x2 , +1, 0)
477  CHECK_SAD_HALF_MV(xy2, +1, +1)
478  }
479  CHECK_SAD_HALF_MV(y2 , 0, +1)
480  }
481  mx+=dx;
482  my+=dy;
483 
484  }else{
485  mx<<=1;
486  my<<=1;
487  }
488 
489  *mx_ptr = mx;
490  *my_ptr = my;
491  return dminh;
492 }
493 
494 static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
495 {
496  const int xy= s->mb_x + s->mb_y*s->mb_stride;
497 
498  s->p_mv_table[xy][0] = mx;
499  s->p_mv_table[xy][1] = my;
500 
501  /* has already been set to the 4 MV if 4MV is done */
502  if(mv4){
503  int mot_xy= s->block_index[0];
504 
505  s->current_picture.motion_val[0][mot_xy ][0] = mx;
506  s->current_picture.motion_val[0][mot_xy ][1] = my;
507  s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
508  s->current_picture.motion_val[0][mot_xy + 1][1] = my;
509 
510  mot_xy += s->b8_stride;
511  s->current_picture.motion_val[0][mot_xy ][0] = mx;
512  s->current_picture.motion_val[0][mot_xy ][1] = my;
513  s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
514  s->current_picture.motion_val[0][mot_xy + 1][1] = my;
515  }
516 }
517 
518 /**
519  * get fullpel ME search limits.
520  */
521 static inline void get_limits(MpegEncContext *s, int x, int y)
522 {
523  MotionEstContext * const c= &s->me;
524  int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
525  int max_range = MAX_MV >> (1 + !!(c->flags&FLAG_QPEL));
526 /*
527  if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
528  else c->range= 16;
529 */
530  if (s->unrestricted_mv) {
531  c->xmin = - x - 16;
532  c->ymin = - y - 16;
533  c->xmax = - x + s->width;
534  c->ymax = - y + s->height;
535  } else if (s->out_format == FMT_H261){
536  // Search range of H261 is different from other codec standards
537  c->xmin = (x > 15) ? - 15 : 0;
538  c->ymin = (y > 15) ? - 15 : 0;
539  c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
540  c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
541  } else {
542  c->xmin = - x;
543  c->ymin = - y;
544  c->xmax = - x + s->mb_width *16 - 16;
545  c->ymax = - y + s->mb_height*16 - 16;
546  }
547  if(!range || range > max_range)
548  range = max_range;
549  if(range){
550  c->xmin = FFMAX(c->xmin,-range);
551  c->xmax = FFMIN(c->xmax, range);
552  c->ymin = FFMAX(c->ymin,-range);
553  c->ymax = FFMIN(c->ymax, range);
554  }
555 }
556 
557 static inline void init_mv4_ref(MotionEstContext *c){
558  const int stride= c->stride;
559 
560  c->ref[1][0] = c->ref[0][0] + 8;
561  c->ref[2][0] = c->ref[0][0] + 8*stride;
562  c->ref[3][0] = c->ref[2][0] + 8;
563  c->src[1][0] = c->src[0][0] + 8;
564  c->src[2][0] = c->src[0][0] + 8*stride;
565  c->src[3][0] = c->src[2][0] + 8;
566 }
567 
568 static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
569 {
570  MotionEstContext * const c= &s->me;
571  const int size= 1;
572  const int h=8;
573  int block;
574  int P[10][2];
575  int dmin_sum=0, mx4_sum=0, my4_sum=0, i;
576  int same=1;
577  const int stride= c->stride;
578  uint8_t *mv_penalty= c->current_mv_penalty;
579  int saftey_cliping= s->unrestricted_mv && (s->width&15) && (s->height&15);
580 
581  init_mv4_ref(c);
582 
583  for(block=0; block<4; block++){
584  int mx4, my4;
585  int pred_x4, pred_y4;
586  int dmin4;
587  static const int off[4]= {2, 1, 1, -1};
588  const int mot_stride = s->b8_stride;
589  const int mot_xy = s->block_index[block];
590 
591  if(saftey_cliping){
592  c->xmax = - 16*s->mb_x + s->width - 8*(block &1);
593  c->ymax = - 16*s->mb_y + s->height - 8*(block>>1);
594  }
595 
596  P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
597  P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
598 
599  if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
600 
601  /* special case for first line */
602  if (s->first_slice_line && block<2) {
603  c->pred_x= pred_x4= P_LEFT[0];
604  c->pred_y= pred_y4= P_LEFT[1];
605  } else {
606  P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
607  P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
608  P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
609  P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];
610  if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
611  if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
612  if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
613  if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
614 
615  P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
616  P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
617 
618  c->pred_x= pred_x4 = P_MEDIAN[0];
619  c->pred_y= pred_y4 = P_MEDIAN[1];
620  }
621  P_MV1[0]= mx;
622  P_MV1[1]= my;
623  if(saftey_cliping)
624  for(i=1; i<10; i++){
625  if (s->first_slice_line && block<2 && i>1 && i<9)
626  continue;
627  if (i>4 && i<9)
628  continue;
629  if(P[i][0] > (c->xmax<<shift)) P[i][0]= (c->xmax<<shift);
630  if(P[i][1] > (c->ymax<<shift)) P[i][1]= (c->ymax<<shift);
631  }
632 
633  dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
634 
635  dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
636 
637  if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
638  int dxy;
639  const int offset= ((block&1) + (block>>1)*stride)*8;
640  uint8_t *dest_y = c->scratchpad + offset;
641  if(s->quarter_sample){
642  uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
643  dxy = ((my4 & 3) << 2) | (mx4 & 3);
644 
645  if(s->no_rounding)
646  s->qdsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y, ref, stride);
647  else
648  s->qdsp.put_qpel_pixels_tab[1][dxy](dest_y, ref, stride);
649  }else{
650  uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
651  dxy = ((my4 & 1) << 1) | (mx4 & 1);
652 
653  if(s->no_rounding)
654  s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , stride, h);
655  else
656  s->hdsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
657  }
658  dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
659  }else
660  dmin_sum+= dmin4;
661 
662  if(s->quarter_sample){
663  mx4_sum+= mx4/2;
664  my4_sum+= my4/2;
665  }else{
666  mx4_sum+= mx4;
667  my4_sum+= my4;
668  }
669 
670  s->current_picture.motion_val[0][s->block_index[block]][0] = mx4;
671  s->current_picture.motion_val[0][s->block_index[block]][1] = my4;
672 
673  if(mx4 != mx || my4 != my) same=0;
674  }
675 
676  if(same)
677  return INT_MAX;
678 
679  if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
680  dmin_sum += s->mecc.mb_cmp[0](s,
681  s->new_picture.f->data[0] +
682  s->mb_x * 16 + s->mb_y * 16 * stride,
683  c->scratchpad, stride, 16);
684  }
685 
686  if(c->avctx->mb_cmp&FF_CMP_CHROMA){
687  int dxy;
688  int mx, my;
689  int offset;
690 
691  mx= ff_h263_round_chroma(mx4_sum);
692  my= ff_h263_round_chroma(my4_sum);
693  dxy = ((my & 1) << 1) | (mx & 1);
694 
695  offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
696 
697  if(s->no_rounding){
698  s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.f->data[1] + offset, s->uvlinesize, 8);
699  s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f->data[2] + offset, s->uvlinesize, 8);
700  }else{
701  s->hdsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.f->data[1] + offset, s->uvlinesize, 8);
702  s->hdsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_picture.f->data[2] + offset, s->uvlinesize, 8);
703  }
704 
705  dmin_sum += s->mecc.mb_cmp[1](s, s->new_picture.f->data[1] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad, s->uvlinesize, 8);
706  dmin_sum += s->mecc.mb_cmp[1](s, s->new_picture.f->data[2] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad + 8, s->uvlinesize, 8);
707  }
708 
709  c->pred_x= mx;
710  c->pred_y= my;
711 
712  switch(c->avctx->mb_cmp&0xFF){
713  /*case FF_CMP_SSE:
714  return dmin_sum+ 32*s->qscale*s->qscale;*/
715  case FF_CMP_RD:
716  return dmin_sum;
717  default:
718  return dmin_sum+ 11*c->mb_penalty_factor;
719  }
720 }
721 
722 static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
723  MotionEstContext * const c= &s->me;
724 
725  c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
726  c->src[1][0] = c->src[0][0] + s->linesize;
727  if(c->flags & FLAG_CHROMA){
728  c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
729  c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
730  c->src[1][1] = c->src[0][1] + s->uvlinesize;
731  c->src[1][2] = c->src[0][2] + s->uvlinesize;
732  }
733 }
734 
735 static int interlaced_search(MpegEncContext *s, int ref_index,
736  int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
737 {
738  MotionEstContext * const c= &s->me;
739  const int size=0;
740  const int h=8;
741  int block;
742  int P[10][2];
743  uint8_t * const mv_penalty= c->current_mv_penalty;
744  int same=1;
745  const int stride= 2*s->linesize;
746  int dmin_sum= 0;
747  const int mot_stride= s->mb_stride;
748  const int xy= s->mb_x + s->mb_y*mot_stride;
749 
750  c->ymin>>=1;
751  c->ymax>>=1;
752  c->stride<<=1;
753  c->uvstride<<=1;
754  init_interlaced_ref(s, ref_index);
755 
756  for(block=0; block<2; block++){
757  int field_select;
758  int best_dmin= INT_MAX;
759  int best_field= -1;
760 
761  for(field_select=0; field_select<2; field_select++){
762  int dmin, mx_i, my_i;
763  int16_t (*mv_table)[2]= mv_tables[block][field_select];
764 
765  if(user_field_select){
766  av_assert1(field_select==0 || field_select==1);
767  av_assert1(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);
768  if(field_select_tables[block][xy] != field_select)
769  continue;
770  }
771 
772  P_LEFT[0] = mv_table[xy - 1][0];
773  P_LEFT[1] = mv_table[xy - 1][1];
774  if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
775 
776  c->pred_x= P_LEFT[0];
777  c->pred_y= P_LEFT[1];
778 
779  if(!s->first_slice_line){
780  P_TOP[0] = mv_table[xy - mot_stride][0];
781  P_TOP[1] = mv_table[xy - mot_stride][1];
782  P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
783  P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
784  if(P_TOP[1] > (c->ymax<<1)) P_TOP[1] = (c->ymax<<1);
785  if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
786  if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
787  if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
788 
789  P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
790  P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
791  }
792  P_MV1[0]= mx; //FIXME not correct if block != field_select
793  P_MV1[1]= my / 2;
794 
795  dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);
796 
797  dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
798 
799  mv_table[xy][0]= mx_i;
800  mv_table[xy][1]= my_i;
801 
802  if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
803  int dxy;
804 
805  //FIXME chroma ME
806  uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
807  dxy = ((my_i & 1) << 1) | (mx_i & 1);
808 
809  if(s->no_rounding){
810  s->hdsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref , stride, h);
811  }else{
812  s->hdsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
813  }
814  dmin = s->mecc.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
815  dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
816  }else
817  dmin+= c->mb_penalty_factor; //field_select bits
818 
819  dmin += field_select != block; //slightly prefer same field
820 
821  if(dmin < best_dmin){
822  best_dmin= dmin;
823  best_field= field_select;
824  }
825  }
826  {
827  int16_t (*mv_table)[2]= mv_tables[block][best_field];
828 
829  if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all
830  if(mv_table[xy][1]&1) same=0;
831  if(mv_table[xy][1]*2 != my) same=0;
832  if(best_field != block) same=0;
833  }
834 
835  field_select_tables[block][xy]= best_field;
836  dmin_sum += best_dmin;
837  }
838 
839  c->ymin<<=1;
840  c->ymax<<=1;
841  c->stride>>=1;
842  c->uvstride>>=1;
843 
844  if(same)
845  return INT_MAX;
846 
847  switch(c->avctx->mb_cmp&0xFF){
848  /*case FF_CMP_SSE:
849  return dmin_sum+ 32*s->qscale*s->qscale;*/
850  case FF_CMP_RD:
851  return dmin_sum;
852  default:
853  return dmin_sum+ 11*c->mb_penalty_factor;
854  }
855 }
856 
857 static inline int get_penalty_factor(int lambda, int lambda2, int type){
858  switch(type&0xFF){
859  default:
860  case FF_CMP_SAD:
861  return lambda>>FF_LAMBDA_SHIFT;
862  case FF_CMP_DCT:
863  return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
864  case FF_CMP_W53:
865  return (4*lambda)>>(FF_LAMBDA_SHIFT);
866  case FF_CMP_W97:
867  return (2*lambda)>>(FF_LAMBDA_SHIFT);
868  case FF_CMP_SATD:
869  case FF_CMP_DCT264:
870  return (2*lambda)>>FF_LAMBDA_SHIFT;
871  case FF_CMP_RD:
872  case FF_CMP_PSNR:
873  case FF_CMP_SSE:
874  case FF_CMP_NSSE:
875  return lambda2>>FF_LAMBDA_SHIFT;
876  case FF_CMP_BIT:
877  return 1;
878  }
879 }
880 
881 void ff_estimate_p_frame_motion(MpegEncContext * s,
882  int mb_x, int mb_y)
883 {
884  MotionEstContext * const c= &s->me;
885  uint8_t *pix, *ppix;
886  int sum, mx, my, dmin;
887  int varc; ///< the variance of the block (sum of squared (p[y][x]-average))
888  int vard; ///< sum of squared differences with the estimated motion vector
889  int P[10][2];
890  const int shift= 1+s->quarter_sample;
891  int mb_type=0;
892  Picture * const pic= &s->current_picture;
893 
894  init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
895 
896  av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
897  av_assert0(s->linesize == c->stride);
898  av_assert0(s->uvlinesize == c->uvstride);
899 
900  c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
901  c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
902  c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
903  c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
904 
905  get_limits(s, 16*mb_x, 16*mb_y);
906  c->skip=0;
907 
908  /* intra / predictive decision */
909  pix = c->src[0][0];
910  sum = s->mpvencdsp.pix_sum(pix, s->linesize);
911  varc = s->mpvencdsp.pix_norm1(pix, s->linesize) -
912  (((unsigned) sum * sum) >> 8) + 500;
913 
914  pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
915  pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
916  c->mb_var_sum_temp += (varc+128)>>8;
917 
918  switch(s->me_method) {
919  case ME_ZERO:
920  default:
921  mx = 0;
922  my = 0;
923  dmin = 0;
924  break;
925  case ME_X1:
926  case ME_EPZS:
927  {
928  const int mot_stride = s->b8_stride;
929  const int mot_xy = s->block_index[0];
930 
931  P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
932  P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
933 
934  if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
935 
936  if(!s->first_slice_line) {
937  P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
938  P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
939  P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
940  P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
941  if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
942  if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
943  if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
944 
945  P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
946  P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
947 
948  if(s->out_format == FMT_H263){
949  c->pred_x = P_MEDIAN[0];
950  c->pred_y = P_MEDIAN[1];
951  }else { /* mpeg1 at least */
952  c->pred_x= P_LEFT[0];
953  c->pred_y= P_LEFT[1];
954  }
955  }else{
956  c->pred_x= P_LEFT[0];
957  c->pred_y= P_LEFT[1];
958  }
959 
960  }
961  dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
962 
963  break;
964  }
965 
966  /* At this point (mx,my) are full-pell and the relative displacement */
967  ppix = c->ref[0][0] + (my * s->linesize) + mx;
968 
969  vard = s->mecc.sse[0](NULL, pix, ppix, s->linesize, 16);
970 
971  pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
972  c->mc_mb_var_sum_temp += (vard+128)>>8;
973 
974  if(mb_type){
975  int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
976  int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
977  c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
978 
979  if(mb_type == CANDIDATE_MB_TYPE_INTER){
980  c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
981  set_p_mv_tables(s, mx, my, 1);
982  }else{
983  mx <<=shift;
984  my <<=shift;
985  }
986  if(mb_type == CANDIDATE_MB_TYPE_INTER4V){
987  h263_mv4_search(s, mx, my, shift);
988 
989  set_p_mv_tables(s, mx, my, 0);
990  }
991  if(mb_type == CANDIDATE_MB_TYPE_INTER_I){
992  interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);
993  }
994  }else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
995  int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
996  int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
997  c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
998 
999  if (vard*2 + 200*256 > varc)
1000  mb_type|= CANDIDATE_MB_TYPE_INTRA;
1001  if (varc*2 + 200*256 > vard || s->qscale > 24){
1002 // if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){
1003  mb_type|= CANDIDATE_MB_TYPE_INTER;
1004  c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1005  if (s->mpv_flags & FF_MPV_FLAG_MV0)
1006  if(mx || my)
1007  mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference
1008  }else{
1009  mx <<=shift;
1010  my <<=shift;
1011  }
1012  if((s->flags&CODEC_FLAG_4MV)
1013  && !c->skip && varc>50<<8 && vard>10<<8){
1014  if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
1015  mb_type|=CANDIDATE_MB_TYPE_INTER4V;
1016 
1017  set_p_mv_tables(s, mx, my, 0);
1018  }else
1019  set_p_mv_tables(s, mx, my, 1);
1020  if((s->flags&CODEC_FLAG_INTERLACED_ME)
1021  && !c->skip){ //FIXME varc/d checks
1022  if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
1023  mb_type |= CANDIDATE_MB_TYPE_INTER_I;
1024  }
1025  }else{
1026  int intra_score, i;
1027  mb_type= CANDIDATE_MB_TYPE_INTER;
1028 
1029  dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1030  if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1031  dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1032 
1033  if((s->flags&CODEC_FLAG_4MV)
1034  && !c->skip && varc>50<<8 && vard>10<<8){
1035  int dmin4= h263_mv4_search(s, mx, my, shift);
1036  if(dmin4 < dmin){
1037  mb_type= CANDIDATE_MB_TYPE_INTER4V;
1038  dmin=dmin4;
1039  }
1040  }
1041  if((s->flags&CODEC_FLAG_INTERLACED_ME)
1042  && !c->skip){ //FIXME varc/d checks
1043  int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
1044  if(dmin_i < dmin){
1045  mb_type = CANDIDATE_MB_TYPE_INTER_I;
1046  dmin= dmin_i;
1047  }
1048  }
1049 
1050  set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
1051 
1052  /* get intra luma score */
1053  if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
1054  intra_score= varc - 500;
1055  }else{
1056  unsigned mean = (sum+128)>>8;
1057  mean*= 0x01010101;
1058 
1059  for(i=0; i<16; i++){
1060  *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
1061  *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
1062  *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
1063  *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
1064  }
1065 
1066  intra_score= s->mecc.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
1067  }
1068  intra_score += c->mb_penalty_factor*16;
1069 
1070  if(intra_score < dmin){
1071  mb_type= CANDIDATE_MB_TYPE_INTRA;
1072  s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
1073  }else
1074  s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;
1075 
1076  {
1077  int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1078  int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1079  c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1080  }
1081  }
1082 
1083  s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;
1084 }
1085 
1086 int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1087  int mb_x, int mb_y)
1088 {
1089  MotionEstContext * const c= &s->me;
1090  int mx, my, dmin;
1091  int P[10][2];
1092  const int shift= 1+s->quarter_sample;
1093  const int xy= mb_x + mb_y*s->mb_stride;
1094  init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
1095 
1096  av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
1097 
1098  c->pre_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
1099  c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1100 
1101  get_limits(s, 16*mb_x, 16*mb_y);
1102  c->skip=0;
1103 
1104  P_LEFT[0] = s->p_mv_table[xy + 1][0];
1105  P_LEFT[1] = s->p_mv_table[xy + 1][1];
1106 
1107  if(P_LEFT[0] < (c->xmin<<shift)) P_LEFT[0] = (c->xmin<<shift);
1108 
1109  /* special case for first line */
1110  if (s->first_slice_line) {
1111  c->pred_x= P_LEFT[0];
1112  c->pred_y= P_LEFT[1];
1113  P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
1114  P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
1115  } else {
1116  P_TOP[0] = s->p_mv_table[xy + s->mb_stride ][0];
1117  P_TOP[1] = s->p_mv_table[xy + s->mb_stride ][1];
1118  P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
1119  P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
1120  if(P_TOP[1] < (c->ymin<<shift)) P_TOP[1] = (c->ymin<<shift);
1121  if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
1122  if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);
1123 
1124  P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1125  P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1126 
1127  c->pred_x = P_MEDIAN[0];
1128  c->pred_y = P_MEDIAN[1];
1129  }
1130 
1131  dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1132 
1133  s->p_mv_table[xy][0] = mx<<shift;
1134  s->p_mv_table[xy][1] = my<<shift;
1135 
1136  return dmin;
1137 }
1138 
1139 static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,
1140  int16_t (*mv_table)[2], int ref_index, int f_code)
1141 {
1142  MotionEstContext * const c= &s->me;
1143  int mx, my, dmin;
1144  int P[10][2];
1145  const int shift= 1+s->quarter_sample;
1146  const int mot_stride = s->mb_stride;
1147  const int mot_xy = mb_y*mot_stride + mb_x;
1148  uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;
1149  int mv_scale;
1150 
1151  c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1152  c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1153  c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1154  c->current_mv_penalty= mv_penalty;
1155 
1156  get_limits(s, 16*mb_x, 16*mb_y);
1157 
1158  switch(s->me_method) {
1159  case ME_ZERO:
1160  default:
1161  mx = 0;
1162  my = 0;
1163  dmin = 0;
1164  break;
1165  case ME_X1:
1166  case ME_EPZS:
1167  P_LEFT[0] = mv_table[mot_xy - 1][0];
1168  P_LEFT[1] = mv_table[mot_xy - 1][1];
1169 
1170  if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);
1171 
1172  /* special case for first line */
1173  if (!s->first_slice_line) {
1174  P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
1175  P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
1176  P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
1177  P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
1178  if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);
1179  if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);
1180  if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);
1181 
1182  P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1183  P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1184  }
1185  c->pred_x = P_LEFT[0];
1186  c->pred_y = P_LEFT[1];
1187 
1188  if(mv_table == s->b_forw_mv_table){
1189  mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);
1190  }else{
1191  mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
1192  }
1193 
1194  dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
1195 
1196  break;
1197  }
1198 
1199  dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
1200 
1201  if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1202  dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
1203 
1204 // s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
1205  mv_table[mot_xy][0]= mx;
1206  mv_table[mot_xy][1]= my;
1207 
1208  return dmin;
1209 }
1210 
1211 static inline int check_bidir_mv(MpegEncContext * s,
1212  int motion_fx, int motion_fy,
1213  int motion_bx, int motion_by,
1214  int pred_fx, int pred_fy,
1215  int pred_bx, int pred_by,
1216  int size, int h)
1217 {
1218  //FIXME optimize?
1219  //FIXME better f_code prediction (max mv & distance)
1220  //FIXME pointers
1221  MotionEstContext * const c= &s->me;
1222  uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
1223  uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame
1224  int stride= c->stride;
1225  uint8_t *dest_y = c->scratchpad;
1226  uint8_t *ptr;
1227  int dxy;
1228  int src_x, src_y;
1229  int fbmin;
1230  uint8_t **src_data= c->src[0];
1231  uint8_t **ref_data= c->ref[0];
1232  uint8_t **ref2_data= c->ref[2];
1233 
1234  if(s->quarter_sample){
1235  dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1236  src_x = motion_fx >> 2;
1237  src_y = motion_fy >> 2;
1238 
1239  ptr = ref_data[0] + (src_y * stride) + src_x;
1240  s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr, stride);
1241 
1242  dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1243  src_x = motion_bx >> 2;
1244  src_y = motion_by >> 2;
1245 
1246  ptr = ref2_data[0] + (src_y * stride) + src_x;
1247  s->qdsp.avg_qpel_pixels_tab[size][dxy](dest_y, ptr, stride);
1248  }else{
1249  dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1250  src_x = motion_fx >> 1;
1251  src_y = motion_fy >> 1;
1252 
1253  ptr = ref_data[0] + (src_y * stride) + src_x;
1254  s->hdsp.put_pixels_tab[size][dxy](dest_y , ptr , stride, h);
1255 
1256  dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1257  src_x = motion_bx >> 1;
1258  src_y = motion_by >> 1;
1259 
1260  ptr = ref2_data[0] + (src_y * stride) + src_x;
1261  s->hdsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
1262  }
1263 
1264  fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
1265  +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
1266  + s->mecc.mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic
1267 
1268  if(c->avctx->mb_cmp&FF_CMP_CHROMA){
1269  }
1270  //FIXME CHROMA !!!
1271 
1272  return fbmin;
1273 }
1274 
1275 /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
1276 static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
1277 {
1278  MotionEstContext * const c= &s->me;
1279  const int mot_stride = s->mb_stride;
1280  const int xy = mb_y *mot_stride + mb_x;
1281  int fbmin;
1282  int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];
1283  int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];
1284  int pred_bx= s->b_bidir_back_mv_table[xy-1][0];
1285  int pred_by= s->b_bidir_back_mv_table[xy-1][1];
1286  int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];
1287  int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];
1288  int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];
1289  int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];
1290  const int flags= c->sub_flags;
1291  const int qpel= flags&FLAG_QPEL;
1292  const int shift= 1+qpel;
1293  const int xmin= c->xmin<<shift;
1294  const int ymin= c->ymin<<shift;
1295  const int xmax= c->xmax<<shift;
1296  const int ymax= c->ymax<<shift;
1297 #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
1298 #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
1299  int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);
1300  uint8_t map[256] = { 0 };
1301 
1302  map[hashidx&255] = 1;
1303 
1304  fbmin= check_bidir_mv(s, motion_fx, motion_fy,
1305  motion_bx, motion_by,
1306  pred_fx, pred_fy,
1307  pred_bx, pred_by,
1308  0, 16);
1309 
1310  if(s->avctx->bidir_refine){
1311  int end;
1312  static const uint8_t limittab[5]={0,8,32,64,80};
1313  const int limit= limittab[s->avctx->bidir_refine];
1314  static const int8_t vect[][4]={
1315 { 0, 0, 0, 1}, { 0, 0, 0,-1}, { 0, 0, 1, 0}, { 0, 0,-1, 0}, { 0, 1, 0, 0}, { 0,-1, 0, 0}, { 1, 0, 0, 0}, {-1, 0, 0, 0},
1316 
1317 { 0, 0, 1, 1}, { 0, 0,-1,-1}, { 0, 1, 1, 0}, { 0,-1,-1, 0}, { 1, 1, 0, 0}, {-1,-1, 0, 0}, { 1, 0, 0, 1}, {-1, 0, 0,-1},
1318 { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
1319 { 0, 0,-1, 1}, { 0, 0, 1,-1}, { 0,-1, 1, 0}, { 0, 1,-1, 0}, {-1, 1, 0, 0}, { 1,-1, 0, 0}, { 1, 0, 0,-1}, {-1, 0, 0, 1},
1320 { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
1321 
1322 { 0, 1, 1, 1}, { 0,-1,-1,-1}, { 1, 1, 1, 0}, {-1,-1,-1, 0}, { 1, 1, 0, 1}, {-1,-1, 0,-1}, { 1, 0, 1, 1}, {-1, 0,-1,-1},
1323 { 0,-1, 1, 1}, { 0, 1,-1,-1}, {-1, 1, 1, 0}, { 1,-1,-1, 0}, { 1, 1, 0,-1}, {-1,-1, 0, 1}, { 1, 0,-1, 1}, {-1, 0, 1,-1},
1324 { 0, 1,-1, 1}, { 0,-1, 1,-1}, { 1,-1, 1, 0}, {-1, 1,-1, 0}, {-1, 1, 0, 1}, { 1,-1, 0,-1}, { 1, 0, 1,-1}, {-1, 0,-1, 1},
1325 { 0, 1, 1,-1}, { 0,-1,-1, 1}, { 1, 1,-1, 0}, {-1,-1, 1, 0}, { 1,-1, 0, 1}, {-1, 1, 0,-1}, {-1, 0, 1, 1}, { 1, 0,-1,-1},
1326 
1327 { 1, 1, 1, 1}, {-1,-1,-1,-1},
1328 { 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},
1329 { 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},
1330  };
1331  static const uint8_t hash[]={
1332 HASH8( 0, 0, 0, 1), HASH8( 0, 0, 0,-1), HASH8( 0, 0, 1, 0), HASH8( 0, 0,-1, 0), HASH8( 0, 1, 0, 0), HASH8( 0,-1, 0, 0), HASH8( 1, 0, 0, 0), HASH8(-1, 0, 0, 0),
1333 
1334 HASH8( 0, 0, 1, 1), HASH8( 0, 0,-1,-1), HASH8( 0, 1, 1, 0), HASH8( 0,-1,-1, 0), HASH8( 1, 1, 0, 0), HASH8(-1,-1, 0, 0), HASH8( 1, 0, 0, 1), HASH8(-1, 0, 0,-1),
1335 HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),
1336 HASH8( 0, 0,-1, 1), HASH8( 0, 0, 1,-1), HASH8( 0,-1, 1, 0), HASH8( 0, 1,-1, 0), HASH8(-1, 1, 0, 0), HASH8( 1,-1, 0, 0), HASH8( 1, 0, 0,-1), HASH8(-1, 0, 0, 1),
1337 HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),
1338 
1339 HASH8( 0, 1, 1, 1), HASH8( 0,-1,-1,-1), HASH8( 1, 1, 1, 0), HASH8(-1,-1,-1, 0), HASH8( 1, 1, 0, 1), HASH8(-1,-1, 0,-1), HASH8( 1, 0, 1, 1), HASH8(-1, 0,-1,-1),
1340 HASH8( 0,-1, 1, 1), HASH8( 0, 1,-1,-1), HASH8(-1, 1, 1, 0), HASH8( 1,-1,-1, 0), HASH8( 1, 1, 0,-1), HASH8(-1,-1, 0, 1), HASH8( 1, 0,-1, 1), HASH8(-1, 0, 1,-1),
1341 HASH8( 0, 1,-1, 1), HASH8( 0,-1, 1,-1), HASH8( 1,-1, 1, 0), HASH8(-1, 1,-1, 0), HASH8(-1, 1, 0, 1), HASH8( 1,-1, 0,-1), HASH8( 1, 0, 1,-1), HASH8(-1, 0,-1, 1),
1342 HASH8( 0, 1, 1,-1), HASH8( 0,-1,-1, 1), HASH8( 1, 1,-1, 0), HASH8(-1,-1, 1, 0), HASH8( 1,-1, 0, 1), HASH8(-1, 1, 0,-1), HASH8(-1, 0, 1, 1), HASH8( 1, 0,-1,-1),
1343 
1344 HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),
1345 HASH8( 1, 1, 1,-1), HASH8(-1,-1,-1, 1), HASH8( 1, 1,-1, 1), HASH8(-1,-1, 1,-1), HASH8( 1,-1, 1, 1), HASH8(-1, 1,-1,-1), HASH8(-1, 1, 1, 1), HASH8( 1,-1,-1,-1),
1346 HASH8( 1, 1,-1,-1), HASH8(-1,-1, 1, 1), HASH8( 1,-1,-1, 1), HASH8(-1, 1, 1,-1), HASH8( 1,-1, 1,-1), HASH8(-1, 1,-1, 1),
1347 };
1348 
1349 #define CHECK_BIDIR(fx,fy,bx,by)\
1350  if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
1351  &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1352  &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1353  int score;\
1354  map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
1355  score= check_bidir_mv(s, motion_fx+fx, motion_fy+fy, motion_bx+bx, motion_by+by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);\
1356  if(score < fbmin){\
1357  hashidx += HASH(fx,fy,bx,by);\
1358  fbmin= score;\
1359  motion_fx+=fx;\
1360  motion_fy+=fy;\
1361  motion_bx+=bx;\
1362  motion_by+=by;\
1363  end=0;\
1364  }\
1365  }
1366 #define CHECK_BIDIR2(a,b,c,d)\
1367 CHECK_BIDIR(a,b,c,d)\
1368 CHECK_BIDIR(-(a),-(b),-(c),-(d))
1369 
1370  do{
1371  int i;
1372  int borderdist=0;
1373  end=1;
1374 
1375  CHECK_BIDIR2(0,0,0,1)
1376  CHECK_BIDIR2(0,0,1,0)
1377  CHECK_BIDIR2(0,1,0,0)
1378  CHECK_BIDIR2(1,0,0,0)
1379 
1380  for(i=8; i<limit; i++){
1381  int fx= motion_fx+vect[i][0];
1382  int fy= motion_fy+vect[i][1];
1383  int bx= motion_bx+vect[i][2];
1384  int by= motion_by+vect[i][3];
1385  if(borderdist<=0){
1386  int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);
1387  int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);
1388  if((a|b) < 0)
1389  map[(hashidx+hash[i])&255] = 1;
1390  }
1391  if(!map[(hashidx+hash[i])&255]){
1392  int score;
1393  map[(hashidx+hash[i])&255] = 1;
1394  score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
1395  if(score < fbmin){
1396  hashidx += hash[i];
1397  fbmin= score;
1398  motion_fx=fx;
1399  motion_fy=fy;
1400  motion_bx=bx;
1401  motion_by=by;
1402  end=0;
1403  borderdist--;
1404  if(borderdist<=0){
1405  int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);
1406  int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);
1407  borderdist= FFMIN(a,b);
1408  }
1409  }
1410  }
1411  }
1412  }while(!end);
1413  }
1414 
1415  s->b_bidir_forw_mv_table[xy][0]= motion_fx;
1416  s->b_bidir_forw_mv_table[xy][1]= motion_fy;
1417  s->b_bidir_back_mv_table[xy][0]= motion_bx;
1418  s->b_bidir_back_mv_table[xy][1]= motion_by;
1419 
1420  return fbmin;
1421 }
1422 
1423 static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
1424 {
1425  MotionEstContext * const c= &s->me;
1426  int P[10][2];
1427  const int mot_stride = s->mb_stride;
1428  const int mot_xy = mb_y*mot_stride + mb_x;
1429  const int shift= 1+s->quarter_sample;
1430  int dmin, i;
1431  const int time_pp= s->pp_time;
1432  const int time_pb= s->pb_time;
1433  int mx, my, xmin, xmax, ymin, ymax;
1434  int16_t (*mv_table)[2]= s->b_direct_mv_table;
1435 
1436  c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;
1437  ymin= xmin=(-32)>>shift;
1438  ymax= xmax= 31>>shift;
1439 
1440  if (IS_8X8(s->next_picture.mb_type[mot_xy])) {
1441  s->mv_type= MV_TYPE_8X8;
1442  }else{
1443  s->mv_type= MV_TYPE_16X16;
1444  }
1445 
1446  for(i=0; i<4; i++){
1447  int index= s->block_index[i];
1448  int min, max;
1449 
1450  c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];
1451  c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];
1452  c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
1453  c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
1454 // c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
1455 // c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
1456 
1457  max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1458  min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1459  max+= 16*mb_x + 1; // +-1 is for the simpler rounding
1460  min+= 16*mb_x - 1;
1461  xmax= FFMIN(xmax, s->width - max);
1462  xmin= FFMAX(xmin, - 16 - min);
1463 
1464  max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1465  min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1466  max+= 16*mb_y + 1; // +-1 is for the simpler rounding
1467  min+= 16*mb_y - 1;
1468  ymax= FFMIN(ymax, s->height - max);
1469  ymin= FFMAX(ymin, - 16 - min);
1470 
1471  if(s->mv_type == MV_TYPE_16X16) break;
1472  }
1473 
1474  av_assert2(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1475 
1476  if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1477  s->b_direct_mv_table[mot_xy][0]= 0;
1478  s->b_direct_mv_table[mot_xy][1]= 0;
1479 
1480  return 256*256*256*64;
1481  }
1482 
1483  c->xmin= xmin;
1484  c->ymin= ymin;
1485  c->xmax= xmax;
1486  c->ymax= ymax;
1487  c->flags |= FLAG_DIRECT;
1488  c->sub_flags |= FLAG_DIRECT;
1489  c->pred_x=0;
1490  c->pred_y=0;
1491 
1492  P_LEFT[0] = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
1493  P_LEFT[1] = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
1494 
1495  /* special case for first line */
1496  if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped
1497  P_TOP[0] = av_clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
1498  P_TOP[1] = av_clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
1499  P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
1500  P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, ymax<<shift);
1501 
1502  P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1503  P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1504  }
1505 
1506  dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
1507  if(c->sub_flags&FLAG_QPEL)
1508  dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1509  else
1510  dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1511 
1512  if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1513  dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1514 
1515  get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
1516 
1517  mv_table[mot_xy][0]= mx;
1518  mv_table[mot_xy][1]= my;
1519  c->flags &= ~FLAG_DIRECT;
1520  c->sub_flags &= ~FLAG_DIRECT;
1521 
1522  return dmin;
1523 }
1524 
1525 void ff_estimate_b_frame_motion(MpegEncContext * s,
1526  int mb_x, int mb_y)
1527 {
1528  MotionEstContext * const c= &s->me;
1529  const int penalty_factor= c->mb_penalty_factor;
1530  int fmin, bmin, dmin, fbmin, bimin, fimin;
1531  int type=0;
1532  const int xy = mb_y*s->mb_stride + mb_x;
1533  init_ref(c, s->new_picture.f->data, s->last_picture.f->data,
1534  s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);
1535 
1536  get_limits(s, 16*mb_x, 16*mb_y);
1537 
1538  c->skip=0;
1539 
1540  if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {
1541  int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
1542 
1543  score= ((unsigned)(score*score + 128*256))>>16;
1544  c->mc_mb_var_sum_temp += score;
1545  s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1546  s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
1547 
1548  return;
1549  }
1550 
1551  if (s->codec_id == AV_CODEC_ID_MPEG4)
1552  dmin= direct_search(s, mb_x, mb_y);
1553  else
1554  dmin= INT_MAX;
1555 //FIXME penalty stuff for non mpeg4
1556  c->skip=0;
1557  fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +
1558  3 * penalty_factor;
1559 
1560  c->skip=0;
1561  bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +
1562  2 * penalty_factor;
1563  av_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
1564 
1565  c->skip=0;
1566  fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
1567  av_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);
1568 
1569  if(s->flags & CODEC_FLAG_INTERLACED_ME){
1570 //FIXME mb type penalty
1571  c->skip=0;
1572  c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1573  fimin= interlaced_search(s, 0,
1574  s->b_field_mv_table[0], s->b_field_select_table[0],
1575  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
1576  c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
1577  bimin= interlaced_search(s, 2,
1578  s->b_field_mv_table[1], s->b_field_select_table[1],
1579  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);
1580  }else
1581  fimin= bimin= INT_MAX;
1582 
1583  {
1584  int score= fmin;
1585  type = CANDIDATE_MB_TYPE_FORWARD;
1586 
1587  if (dmin <= score){
1588  score = dmin;
1589  type = CANDIDATE_MB_TYPE_DIRECT;
1590  }
1591  if(bmin<score){
1592  score=bmin;
1593  type= CANDIDATE_MB_TYPE_BACKWARD;
1594  }
1595  if(fbmin<score){
1596  score=fbmin;
1597  type= CANDIDATE_MB_TYPE_BIDIR;
1598  }
1599  if(fimin<score){
1600  score=fimin;
1601  type= CANDIDATE_MB_TYPE_FORWARD_I;
1602  }
1603  if(bimin<score){
1604  score=bimin;
1605  type= CANDIDATE_MB_TYPE_BACKWARD_I;
1606  }
1607 
1608  score= ((unsigned)(score*score + 128*256))>>16;
1609  c->mc_mb_var_sum_temp += score;
1610  s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1611  }
1612 
1613  if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1614  type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
1615  if(fimin < INT_MAX)
1616  type |= CANDIDATE_MB_TYPE_FORWARD_I;
1617  if(bimin < INT_MAX)
1618  type |= CANDIDATE_MB_TYPE_BACKWARD_I;
1619  if(fimin < INT_MAX && bimin < INT_MAX){
1620  type |= CANDIDATE_MB_TYPE_BIDIR_I;
1621  }
1622  //FIXME something smarter
1623  if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB
1624  if (s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT &&
1625  s->mpv_flags & FF_MPV_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
1626  type |= CANDIDATE_MB_TYPE_DIRECT0;
1627  }
1628 
1629  s->mb_type[mb_y*s->mb_stride + mb_x]= type;
1630 }
1631 
1632 /* find best f_code for ME which do unlimited searches */
1633 int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)
1634 {
1635  if(s->me_method>=ME_EPZS){
1636  int score[8];
1637  int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);
1638  uint8_t * fcode_tab= s->fcode_tab;
1639  int best_fcode=-1;
1640  int best_score=-10000000;
1641 
1642  if(s->msmpeg4_version)
1643  range= FFMIN(range, 16);
1644  else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
1645  range= FFMIN(range, 256);
1646 
1647  for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);
1648 
1649  for(y=0; y<s->mb_height; y++){
1650  int x;
1651  int xy= y*s->mb_stride;
1652  for(x=0; x<s->mb_width; x++){
1653  if(s->mb_type[xy] & type){
1654  int mx= mv_table[xy][0];
1655  int my= mv_table[xy][1];
1656  int fcode= FFMAX(fcode_tab[mx + MAX_MV],
1657  fcode_tab[my + MAX_MV]);
1658  int j;
1659 
1660  if(mx >= range || mx < -range ||
1661  my >= range || my < -range)
1662  continue;
1663 
1664  for(j=0; j<fcode && j<8; j++){
1665  if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
1666  score[j]-= 170;
1667  }
1668  }
1669  xy++;
1670  }
1671  }
1672 
1673  for(i=1; i<8; i++){
1674  if(score[i] > best_score){
1675  best_score= score[i];
1676  best_fcode= i;
1677  }
1678  }
1679 
1680  return best_fcode;
1681  }else{
1682  return 1;
1683  }
1684 }
1685 
1686 void ff_fix_long_p_mvs(MpegEncContext * s)
1687 {
1688  MotionEstContext * const c= &s->me;
1689  const int f_code= s->f_code;
1690  int y, range;
1691  av_assert0(s->pict_type==AV_PICTURE_TYPE_P);
1692 
1693  range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1694 
1695  av_assert0(range <= 16 || !s->msmpeg4_version);
1696  av_assert0(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
1697 
1698  if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1699 
1700  if(s->flags&CODEC_FLAG_4MV){
1701  const int wrap= s->b8_stride;
1702 
1703  /* clip / convert to intra 8x8 type MVs */
1704  for(y=0; y<s->mb_height; y++){
1705  int xy= y*2*wrap;
1706  int i= y*s->mb_stride;
1707  int x;
1708 
1709  for(x=0; x<s->mb_width; x++){
1710  if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){
1711  int block;
1712  for(block=0; block<4; block++){
1713  int off= (block& 1) + (block>>1)*wrap;
1714  int mx = s->current_picture.motion_val[0][ xy + off ][0];
1715  int my = s->current_picture.motion_val[0][ xy + off ][1];
1716 
1717  if( mx >=range || mx <-range
1718  || my >=range || my <-range){
1719  s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;
1720  s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;
1721  s->current_picture.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;
1722  }
1723  }
1724  }
1725  xy+=2;
1726  i++;
1727  }
1728  }
1729  }
1730 }
1731 
1732 /**
1733  *
1734  * @param truncate 1 for truncation, 0 for using intra
1735  */
1736 void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
1737  int16_t (*mv_table)[2], int f_code, int type, int truncate)
1738 {
1739  MotionEstContext * const c= &s->me;
1740  int y, h_range, v_range;
1741 
1742  // RAL: 8 in MPEG-1, 16 in MPEG-4
1743  int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1744 
1745  if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1746 
1747  h_range= range;
1748  v_range= field_select_table ? range>>1 : range;
1749 
1750  /* clip / convert to intra 16x16 type MVs */
1751  for(y=0; y<s->mb_height; y++){
1752  int x;
1753  int xy= y*s->mb_stride;
1754  for(x=0; x<s->mb_width; x++){
1755  if (s->mb_type[xy] & type){ // RAL: "type" test added...
1756  if (!field_select_table || field_select_table[xy] == field_select) {
1757  if( mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
1758  || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
1759 
1760  if(truncate){
1761  if (mv_table[xy][0] > h_range-1) mv_table[xy][0]= h_range-1;
1762  else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
1763  if (mv_table[xy][1] > v_range-1) mv_table[xy][1]= v_range-1;
1764  else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
1765  }else{
1766  s->mb_type[xy] &= ~type;
1767  s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;
1768  mv_table[xy][0]=
1769  mv_table[xy][1]= 0;
1770  }
1771  }
1772  }
1773  }
1774  xy++;
1775  }
1776  }
1777 }
Generated on Sun Sep 14 2014 18:55:59 for FFmpeg by   doxygen 1.8.2