Imported Debian version 2.5.3~trusty1
[deb_ffmpeg.git] / ffmpeg / libavcodec / motion_est.c
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 if (h << size == 16) {
197 c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
198 } else if (size == 0 && h == 8) {
199 c->qpel_put[1][dxy](c->temp , ref[0] + x + y*stride , stride);
200 c->qpel_put[1][dxy](c->temp + 8, ref[0] + x + y*stride + 8, stride);
201 } else
202 av_assert2(0);
203 if(chroma){
204 int cx= hx/2;
205 int cy= hy/2;
206 cx= (cx>>1)|(cx&1);
207 cy= (cy>>1)|(cy&1);
208 uvdxy= (cx&1) + 2*(cy&1);
209 //FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264
210 }
211 }else{
212 c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
213 if(chroma)
214 uvdxy= dxy | (x&1) | (2*(y&1));
215 }
216 d = cmp_func(s, c->temp, src[0], stride, h);
217 }else{
218 d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
219 if(chroma)
220 uvdxy= (x&1) + 2*(y&1);
221 }
222 if(chroma){
223 uint8_t * const uvtemp= c->temp + 16*stride;
224 c->hpel_put[size+1][uvdxy](uvtemp , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
225 c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
226 d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1);
227 d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
228 }
229 return d;
230 }
231
232 static int cmp_simple(MpegEncContext *s, const int x, const int y,
233 int ref_index, int src_index,
234 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){
235 return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0);
236 }
237
238 static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y,
239 const int size, const int h, int ref_index, int src_index,
240 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
241 if(flags&FLAG_DIRECT){
242 return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
243 }else{
244 return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
245 }
246 }
247
248 static int cmp_internal(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
249 const int size, const int h, int ref_index, int src_index,
250 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
251 if(flags&FLAG_DIRECT){
252 return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
253 }else{
254 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);
255 }
256 }
257
258 /** @brief compares a block (either a full macroblock or a partition thereof)
259 against a proposed motion-compensated prediction of that block
260 */
261 static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
262 const int size, const int h, int ref_index, int src_index,
263 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
264 if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size)
265 && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
266 && flags==0 && h==16 && size==0 && subx==0 && suby==0){
267 return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func);
268 }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
269 && subx==0 && suby==0){
270 return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags);
271 }else{
272 return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags);
273 }
274 }
275
276 static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
277 const int size, const int h, int ref_index, int src_index,
278 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
279 if(flags&FLAG_DIRECT){
280 return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0);
281 }else{
282 return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
283 }
284 }
285
286 static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
287 const int size, const int h, int ref_index, int src_index,
288 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
289 if(flags&FLAG_DIRECT){
290 return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1);
291 }else{
292 return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA);
293 }
294 }
295
296 #include "motion_est_template.c"
297
298 static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
299 ptrdiff_t stride, int h)
300 {
301 return 0;
302 }
303
304 static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){
305 }
306
307 int ff_init_me(MpegEncContext *s){
308 MotionEstContext * const c= &s->me;
309 int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
310 int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
311
312 if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -FFMIN(ME_MAP_SIZE, MAX_SAB_SIZE)){
313 av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n");
314 return -1;
315 }
316 //special case of snow is needed because snow uses its own iterative ME code
317 if(s->me_method!=ME_ZERO && s->me_method!=ME_EPZS && s->me_method!=ME_X1 && s->avctx->codec_id != AV_CODEC_ID_SNOW){
318 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");
319 return -1;
320 }
321
322 c->avctx= s->avctx;
323
324 if(cache_size < 2*dia_size && !c->stride){
325 av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
326 }
327
328 ff_set_cmp(&s->mecc, s->mecc.me_pre_cmp, c->avctx->me_pre_cmp);
329 ff_set_cmp(&s->mecc, s->mecc.me_cmp, c->avctx->me_cmp);
330 ff_set_cmp(&s->mecc, s->mecc.me_sub_cmp, c->avctx->me_sub_cmp);
331 ff_set_cmp(&s->mecc, s->mecc.mb_cmp, c->avctx->mb_cmp);
332
333 c->flags = get_flags(c, 0, c->avctx->me_cmp &FF_CMP_CHROMA);
334 c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
335 c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp &FF_CMP_CHROMA);
336
337 /*FIXME s->no_rounding b_type*/
338 if(s->flags&CODEC_FLAG_QPEL){
339 c->sub_motion_search= qpel_motion_search;
340 c->qpel_avg = s->qdsp.avg_qpel_pixels_tab;
341 if (s->no_rounding)
342 c->qpel_put = s->qdsp.put_no_rnd_qpel_pixels_tab;
343 else
344 c->qpel_put = s->qdsp.put_qpel_pixels_tab;
345 }else{
346 if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
347 c->sub_motion_search= hpel_motion_search;
348 else if( c->avctx->me_sub_cmp == FF_CMP_SAD
349 && c->avctx-> me_cmp == FF_CMP_SAD
350 && c->avctx-> mb_cmp == FF_CMP_SAD)
351 c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
352 else
353 c->sub_motion_search= hpel_motion_search;
354 }
355 c->hpel_avg = s->hdsp.avg_pixels_tab;
356 if (s->no_rounding)
357 c->hpel_put = s->hdsp.put_no_rnd_pixels_tab;
358 else
359 c->hpel_put = s->hdsp.put_pixels_tab;
360
361 if(s->linesize){
362 c->stride = s->linesize;
363 c->uvstride= s->uvlinesize;
364 }else{
365 c->stride = 16*s->mb_width + 32;
366 c->uvstride= 8*s->mb_width + 16;
367 }
368
369 /* 8x8 fullpel search would need a 4x4 chroma compare, which we do
370 * not have yet, and even if we had, the motion estimation code
371 * does not expect it. */
372 if (s->codec_id != AV_CODEC_ID_SNOW) {
373 if ((c->avctx->me_cmp & FF_CMP_CHROMA) /* && !s->mecc.me_cmp[2] */)
374 s->mecc.me_cmp[2] = zero_cmp;
375 if ((c->avctx->me_sub_cmp & FF_CMP_CHROMA) && !s->mecc.me_sub_cmp[2])
376 s->mecc.me_sub_cmp[2] = zero_cmp;
377 c->hpel_put[2][0]= c->hpel_put[2][1]=
378 c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
379 }
380
381 if(s->codec_id == AV_CODEC_ID_H261){
382 c->sub_motion_search= no_sub_motion_search;
383 }
384
385 return 0;
386 }
387
388 #define CHECK_SAD_HALF_MV(suffix, x, y) \
389 {\
390 d = s->mecc.pix_abs[size][(x ? 1 : 0) + (y ? 2 : 0)](NULL, pix, ptr + ((x) >> 1), stride, h); \
391 d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
392 COPY3_IF_LT(dminh, d, dx, x, dy, y)\
393 }
394
395 static int sad_hpel_motion_search(MpegEncContext * s,
396 int *mx_ptr, int *my_ptr, int dmin,
397 int src_index, int ref_index,
398 int size, int h)
399 {
400 MotionEstContext * const c= &s->me;
401 const int penalty_factor= c->sub_penalty_factor;
402 int mx, my, dminh;
403 uint8_t *pix, *ptr;
404 int stride= c->stride;
405 LOAD_COMMON
406
407 av_assert2(c->sub_flags == 0);
408
409 if(c->skip){
410 *mx_ptr = 0;
411 *my_ptr = 0;
412 return dmin;
413 }
414
415 pix = c->src[src_index][0];
416
417 mx = *mx_ptr;
418 my = *my_ptr;
419 ptr = c->ref[ref_index][0] + (my * stride) + mx;
420
421 dminh = dmin;
422
423 if (mx > xmin && mx < xmax &&
424 my > ymin && my < ymax) {
425 int dx=0, dy=0;
426 int d, pen_x, pen_y;
427 const int index= (my<<ME_MAP_SHIFT) + mx;
428 const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
429 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
430 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)];
431 const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
432 mx<<=1;
433 my<<=1;
434
435
436 pen_x= pred_x + mx;
437 pen_y= pred_y + my;
438
439 ptr-= stride;
440 if(t<=b){
441 CHECK_SAD_HALF_MV(y2 , 0, -1)
442 if(l<=r){
443 CHECK_SAD_HALF_MV(xy2, -1, -1)
444 if(t+r<=b+l){
445 CHECK_SAD_HALF_MV(xy2, +1, -1)
446 ptr+= stride;
447 }else{
448 ptr+= stride;
449 CHECK_SAD_HALF_MV(xy2, -1, +1)
450 }
451 CHECK_SAD_HALF_MV(x2 , -1, 0)
452 }else{
453 CHECK_SAD_HALF_MV(xy2, +1, -1)
454 if(t+l<=b+r){
455 CHECK_SAD_HALF_MV(xy2, -1, -1)
456 ptr+= stride;
457 }else{
458 ptr+= stride;
459 CHECK_SAD_HALF_MV(xy2, +1, +1)
460 }
461 CHECK_SAD_HALF_MV(x2 , +1, 0)
462 }
463 }else{
464 if(l<=r){
465 if(t+l<=b+r){
466 CHECK_SAD_HALF_MV(xy2, -1, -1)
467 ptr+= stride;
468 }else{
469 ptr+= stride;
470 CHECK_SAD_HALF_MV(xy2, +1, +1)
471 }
472 CHECK_SAD_HALF_MV(x2 , -1, 0)
473 CHECK_SAD_HALF_MV(xy2, -1, +1)
474 }else{
475 if(t+r<=b+l){
476 CHECK_SAD_HALF_MV(xy2, +1, -1)
477 ptr+= stride;
478 }else{
479 ptr+= stride;
480 CHECK_SAD_HALF_MV(xy2, -1, +1)
481 }
482 CHECK_SAD_HALF_MV(x2 , +1, 0)
483 CHECK_SAD_HALF_MV(xy2, +1, +1)
484 }
485 CHECK_SAD_HALF_MV(y2 , 0, +1)
486 }
487 mx+=dx;
488 my+=dy;
489
490 }else{
491 mx<<=1;
492 my<<=1;
493 }
494
495 *mx_ptr = mx;
496 *my_ptr = my;
497 return dminh;
498 }
499
500 static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
501 {
502 const int xy= s->mb_x + s->mb_y*s->mb_stride;
503
504 s->p_mv_table[xy][0] = mx;
505 s->p_mv_table[xy][1] = my;
506
507 /* has already been set to the 4 MV if 4MV is done */
508 if(mv4){
509 int mot_xy= s->block_index[0];
510
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 mot_xy += s->b8_stride;
517 s->current_picture.motion_val[0][mot_xy ][0] = mx;
518 s->current_picture.motion_val[0][mot_xy ][1] = my;
519 s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
520 s->current_picture.motion_val[0][mot_xy + 1][1] = my;
521 }
522 }
523
524 /**
525 * get fullpel ME search limits.
526 */
527 static inline void get_limits(MpegEncContext *s, int x, int y)
528 {
529 MotionEstContext * const c= &s->me;
530 int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
531 int max_range = MAX_MV >> (1 + !!(c->flags&FLAG_QPEL));
532 /*
533 if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
534 else c->range= 16;
535 */
536 if (s->unrestricted_mv) {
537 c->xmin = - x - 16;
538 c->ymin = - y - 16;
539 c->xmax = - x + s->width;
540 c->ymax = - y + s->height;
541 } else if (s->out_format == FMT_H261){
542 // Search range of H261 is different from other codec standards
543 c->xmin = (x > 15) ? - 15 : 0;
544 c->ymin = (y > 15) ? - 15 : 0;
545 c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
546 c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
547 } else {
548 c->xmin = - x;
549 c->ymin = - y;
550 c->xmax = - x + s->mb_width *16 - 16;
551 c->ymax = - y + s->mb_height*16 - 16;
552 }
553 if(!range || range > max_range)
554 range = max_range;
555 if(range){
556 c->xmin = FFMAX(c->xmin,-range);
557 c->xmax = FFMIN(c->xmax, range);
558 c->ymin = FFMAX(c->ymin,-range);
559 c->ymax = FFMIN(c->ymax, range);
560 }
561 }
562
563 static inline void init_mv4_ref(MotionEstContext *c){
564 const int stride= c->stride;
565
566 c->ref[1][0] = c->ref[0][0] + 8;
567 c->ref[2][0] = c->ref[0][0] + 8*stride;
568 c->ref[3][0] = c->ref[2][0] + 8;
569 c->src[1][0] = c->src[0][0] + 8;
570 c->src[2][0] = c->src[0][0] + 8*stride;
571 c->src[3][0] = c->src[2][0] + 8;
572 }
573
574 static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
575 {
576 MotionEstContext * const c= &s->me;
577 const int size= 1;
578 const int h=8;
579 int block;
580 int P[10][2];
581 int dmin_sum=0, mx4_sum=0, my4_sum=0, i;
582 int same=1;
583 const int stride= c->stride;
584 uint8_t *mv_penalty= c->current_mv_penalty;
585 int saftey_cliping= s->unrestricted_mv && (s->width&15) && (s->height&15);
586
587 init_mv4_ref(c);
588
589 for(block=0; block<4; block++){
590 int mx4, my4;
591 int pred_x4, pred_y4;
592 int dmin4;
593 static const int off[4]= {2, 1, 1, -1};
594 const int mot_stride = s->b8_stride;
595 const int mot_xy = s->block_index[block];
596
597 if(saftey_cliping){
598 c->xmax = - 16*s->mb_x + s->width - 8*(block &1);
599 c->ymax = - 16*s->mb_y + s->height - 8*(block>>1);
600 }
601
602 P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
603 P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
604
605 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
606
607 /* special case for first line */
608 if (s->first_slice_line && block<2) {
609 c->pred_x= pred_x4= P_LEFT[0];
610 c->pred_y= pred_y4= P_LEFT[1];
611 } else {
612 P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
613 P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
614 P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
615 P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];
616 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
617 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
618 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
619 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
620
621 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
622 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
623
624 c->pred_x= pred_x4 = P_MEDIAN[0];
625 c->pred_y= pred_y4 = P_MEDIAN[1];
626 }
627 P_MV1[0]= mx;
628 P_MV1[1]= my;
629 if(saftey_cliping)
630 for(i=1; i<10; i++){
631 if (s->first_slice_line && block<2 && i>1 && i<9)
632 continue;
633 if (i>4 && i<9)
634 continue;
635 if(P[i][0] > (c->xmax<<shift)) P[i][0]= (c->xmax<<shift);
636 if(P[i][1] > (c->ymax<<shift)) P[i][1]= (c->ymax<<shift);
637 }
638
639 dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
640
641 dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
642
643 if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
644 int dxy;
645 const int offset= ((block&1) + (block>>1)*stride)*8;
646 uint8_t *dest_y = c->scratchpad + offset;
647 if(s->quarter_sample){
648 uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
649 dxy = ((my4 & 3) << 2) | (mx4 & 3);
650
651 if(s->no_rounding)
652 s->qdsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y, ref, stride);
653 else
654 s->qdsp.put_qpel_pixels_tab[1][dxy](dest_y, ref, stride);
655 }else{
656 uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
657 dxy = ((my4 & 1) << 1) | (mx4 & 1);
658
659 if(s->no_rounding)
660 s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , stride, h);
661 else
662 s->hdsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
663 }
664 dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
665 }else
666 dmin_sum+= dmin4;
667
668 if(s->quarter_sample){
669 mx4_sum+= mx4/2;
670 my4_sum+= my4/2;
671 }else{
672 mx4_sum+= mx4;
673 my4_sum+= my4;
674 }
675
676 s->current_picture.motion_val[0][s->block_index[block]][0] = mx4;
677 s->current_picture.motion_val[0][s->block_index[block]][1] = my4;
678
679 if(mx4 != mx || my4 != my) same=0;
680 }
681
682 if(same)
683 return INT_MAX;
684
685 if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
686 dmin_sum += s->mecc.mb_cmp[0](s,
687 s->new_picture.f->data[0] +
688 s->mb_x * 16 + s->mb_y * 16 * stride,
689 c->scratchpad, stride, 16);
690 }
691
692 if(c->avctx->mb_cmp&FF_CMP_CHROMA){
693 int dxy;
694 int mx, my;
695 int offset;
696
697 mx= ff_h263_round_chroma(mx4_sum);
698 my= ff_h263_round_chroma(my4_sum);
699 dxy = ((my & 1) << 1) | (mx & 1);
700
701 offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
702
703 if(s->no_rounding){
704 s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.f->data[1] + offset, s->uvlinesize, 8);
705 s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f->data[2] + offset, s->uvlinesize, 8);
706 }else{
707 s->hdsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.f->data[1] + offset, s->uvlinesize, 8);
708 s->hdsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_picture.f->data[2] + offset, s->uvlinesize, 8);
709 }
710
711 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);
712 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);
713 }
714
715 c->pred_x= mx;
716 c->pred_y= my;
717
718 switch(c->avctx->mb_cmp&0xFF){
719 /*case FF_CMP_SSE:
720 return dmin_sum+ 32*s->qscale*s->qscale;*/
721 case FF_CMP_RD:
722 return dmin_sum;
723 default:
724 return dmin_sum+ 11*c->mb_penalty_factor;
725 }
726 }
727
728 static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
729 MotionEstContext * const c= &s->me;
730
731 c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
732 c->src[1][0] = c->src[0][0] + s->linesize;
733 if(c->flags & FLAG_CHROMA){
734 c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
735 c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
736 c->src[1][1] = c->src[0][1] + s->uvlinesize;
737 c->src[1][2] = c->src[0][2] + s->uvlinesize;
738 }
739 }
740
741 static int interlaced_search(MpegEncContext *s, int ref_index,
742 int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
743 {
744 MotionEstContext * const c= &s->me;
745 const int size=0;
746 const int h=8;
747 int block;
748 int P[10][2];
749 uint8_t * const mv_penalty= c->current_mv_penalty;
750 int same=1;
751 const int stride= 2*s->linesize;
752 int dmin_sum= 0;
753 const int mot_stride= s->mb_stride;
754 const int xy= s->mb_x + s->mb_y*mot_stride;
755
756 c->ymin>>=1;
757 c->ymax>>=1;
758 c->stride<<=1;
759 c->uvstride<<=1;
760 init_interlaced_ref(s, ref_index);
761
762 for(block=0; block<2; block++){
763 int field_select;
764 int best_dmin= INT_MAX;
765 int best_field= -1;
766
767 for(field_select=0; field_select<2; field_select++){
768 int dmin, mx_i, my_i;
769 int16_t (*mv_table)[2]= mv_tables[block][field_select];
770
771 if(user_field_select){
772 av_assert1(field_select==0 || field_select==1);
773 av_assert1(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);
774 if(field_select_tables[block][xy] != field_select)
775 continue;
776 }
777
778 P_LEFT[0] = mv_table[xy - 1][0];
779 P_LEFT[1] = mv_table[xy - 1][1];
780 if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
781
782 c->pred_x= P_LEFT[0];
783 c->pred_y= P_LEFT[1];
784
785 if(!s->first_slice_line){
786 P_TOP[0] = mv_table[xy - mot_stride][0];
787 P_TOP[1] = mv_table[xy - mot_stride][1];
788 P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
789 P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
790 if(P_TOP[1] > (c->ymax<<1)) P_TOP[1] = (c->ymax<<1);
791 if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
792 if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
793 if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
794
795 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
796 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
797 }
798 P_MV1[0]= mx; //FIXME not correct if block != field_select
799 P_MV1[1]= my / 2;
800
801 dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);
802
803 dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
804
805 mv_table[xy][0]= mx_i;
806 mv_table[xy][1]= my_i;
807
808 if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
809 int dxy;
810
811 //FIXME chroma ME
812 uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
813 dxy = ((my_i & 1) << 1) | (mx_i & 1);
814
815 if(s->no_rounding){
816 s->hdsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref , stride, h);
817 }else{
818 s->hdsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
819 }
820 dmin = s->mecc.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
821 dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
822 }else
823 dmin+= c->mb_penalty_factor; //field_select bits
824
825 dmin += field_select != block; //slightly prefer same field
826
827 if(dmin < best_dmin){
828 best_dmin= dmin;
829 best_field= field_select;
830 }
831 }
832 {
833 int16_t (*mv_table)[2]= mv_tables[block][best_field];
834
835 if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all
836 if(mv_table[xy][1]&1) same=0;
837 if(mv_table[xy][1]*2 != my) same=0;
838 if(best_field != block) same=0;
839 }
840
841 field_select_tables[block][xy]= best_field;
842 dmin_sum += best_dmin;
843 }
844
845 c->ymin<<=1;
846 c->ymax<<=1;
847 c->stride>>=1;
848 c->uvstride>>=1;
849
850 if(same)
851 return INT_MAX;
852
853 switch(c->avctx->mb_cmp&0xFF){
854 /*case FF_CMP_SSE:
855 return dmin_sum+ 32*s->qscale*s->qscale;*/
856 case FF_CMP_RD:
857 return dmin_sum;
858 default:
859 return dmin_sum+ 11*c->mb_penalty_factor;
860 }
861 }
862
863 static inline int get_penalty_factor(int lambda, int lambda2, int type){
864 switch(type&0xFF){
865 default:
866 case FF_CMP_SAD:
867 return lambda>>FF_LAMBDA_SHIFT;
868 case FF_CMP_DCT:
869 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
870 case FF_CMP_W53:
871 return (4*lambda)>>(FF_LAMBDA_SHIFT);
872 case FF_CMP_W97:
873 return (2*lambda)>>(FF_LAMBDA_SHIFT);
874 case FF_CMP_SATD:
875 case FF_CMP_DCT264:
876 return (2*lambda)>>FF_LAMBDA_SHIFT;
877 case FF_CMP_RD:
878 case FF_CMP_PSNR:
879 case FF_CMP_SSE:
880 case FF_CMP_NSSE:
881 return lambda2>>FF_LAMBDA_SHIFT;
882 case FF_CMP_BIT:
883 return 1;
884 }
885 }
886
887 void ff_estimate_p_frame_motion(MpegEncContext * s,
888 int mb_x, int mb_y)
889 {
890 MotionEstContext * const c= &s->me;
891 uint8_t *pix, *ppix;
892 int sum, mx, my, dmin;
893 int varc; ///< the variance of the block (sum of squared (p[y][x]-average))
894 int vard; ///< sum of squared differences with the estimated motion vector
895 int P[10][2];
896 const int shift= 1+s->quarter_sample;
897 int mb_type=0;
898 Picture * const pic= &s->current_picture;
899
900 init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
901
902 av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
903 av_assert0(s->linesize == c->stride);
904 av_assert0(s->uvlinesize == c->uvstride);
905
906 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
907 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
908 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
909 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
910
911 get_limits(s, 16*mb_x, 16*mb_y);
912 c->skip=0;
913
914 /* intra / predictive decision */
915 pix = c->src[0][0];
916 sum = s->mpvencdsp.pix_sum(pix, s->linesize);
917 varc = s->mpvencdsp.pix_norm1(pix, s->linesize) -
918 (((unsigned) sum * sum) >> 8) + 500;
919
920 pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
921 pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
922 c->mb_var_sum_temp += (varc+128)>>8;
923
924 switch(s->me_method) {
925 case ME_ZERO:
926 default:
927 mx = 0;
928 my = 0;
929 dmin = 0;
930 break;
931 case ME_X1:
932 case ME_EPZS:
933 {
934 const int mot_stride = s->b8_stride;
935 const int mot_xy = s->block_index[0];
936
937 P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
938 P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
939
940 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
941
942 if(!s->first_slice_line) {
943 P_TOP[0] = s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
944 P_TOP[1] = s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
945 P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
946 P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
947 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
948 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
949 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
950
951 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
952 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
953
954 if(s->out_format == FMT_H263){
955 c->pred_x = P_MEDIAN[0];
956 c->pred_y = P_MEDIAN[1];
957 }else { /* mpeg1 at least */
958 c->pred_x= P_LEFT[0];
959 c->pred_y= P_LEFT[1];
960 }
961 }else{
962 c->pred_x= P_LEFT[0];
963 c->pred_y= P_LEFT[1];
964 }
965
966 }
967 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
968
969 break;
970 }
971
972 /* At this point (mx,my) are full-pell and the relative displacement */
973 ppix = c->ref[0][0] + (my * s->linesize) + mx;
974
975 vard = s->mecc.sse[0](NULL, pix, ppix, s->linesize, 16);
976
977 pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
978 c->mc_mb_var_sum_temp += (vard+128)>>8;
979
980 if (c->avctx->mb_decision > FF_MB_DECISION_SIMPLE) {
981 int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
982 int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
983 c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
984
985 if (vard*2 + 200*256 > varc)
986 mb_type|= CANDIDATE_MB_TYPE_INTRA;
987 if (varc*2 + 200*256 > vard || s->qscale > 24){
988 // if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){
989 mb_type|= CANDIDATE_MB_TYPE_INTER;
990 c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
991 if (s->mpv_flags & FF_MPV_FLAG_MV0)
992 if(mx || my)
993 mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference
994 }else{
995 mx <<=shift;
996 my <<=shift;
997 }
998 if((s->flags&CODEC_FLAG_4MV)
999 && !c->skip && varc>50<<8 && vard>10<<8){
1000 if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
1001 mb_type|=CANDIDATE_MB_TYPE_INTER4V;
1002
1003 set_p_mv_tables(s, mx, my, 0);
1004 }else
1005 set_p_mv_tables(s, mx, my, 1);
1006 if((s->flags&CODEC_FLAG_INTERLACED_ME)
1007 && !c->skip){ //FIXME varc/d checks
1008 if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
1009 mb_type |= CANDIDATE_MB_TYPE_INTER_I;
1010 }
1011 }else{
1012 int intra_score, i;
1013 mb_type= CANDIDATE_MB_TYPE_INTER;
1014
1015 dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1016 if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1017 dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1018
1019 if((s->flags&CODEC_FLAG_4MV)
1020 && !c->skip && varc>50<<8 && vard>10<<8){
1021 int dmin4= h263_mv4_search(s, mx, my, shift);
1022 if(dmin4 < dmin){
1023 mb_type= CANDIDATE_MB_TYPE_INTER4V;
1024 dmin=dmin4;
1025 }
1026 }
1027 if((s->flags&CODEC_FLAG_INTERLACED_ME)
1028 && !c->skip){ //FIXME varc/d checks
1029 int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
1030 if(dmin_i < dmin){
1031 mb_type = CANDIDATE_MB_TYPE_INTER_I;
1032 dmin= dmin_i;
1033 }
1034 }
1035
1036 set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
1037
1038 /* get intra luma score */
1039 if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
1040 intra_score= varc - 500;
1041 }else{
1042 unsigned mean = (sum+128)>>8;
1043 mean*= 0x01010101;
1044
1045 for(i=0; i<16; i++){
1046 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
1047 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
1048 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
1049 *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
1050 }
1051
1052 intra_score= s->mecc.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
1053 }
1054 intra_score += c->mb_penalty_factor*16;
1055
1056 if(intra_score < dmin){
1057 mb_type= CANDIDATE_MB_TYPE_INTRA;
1058 s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
1059 }else
1060 s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;
1061
1062 {
1063 int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1064 int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1065 c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1066 }
1067 }
1068
1069 s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;
1070 }
1071
1072 int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1073 int mb_x, int mb_y)
1074 {
1075 MotionEstContext * const c= &s->me;
1076 int mx, my, dmin;
1077 int P[10][2];
1078 const int shift= 1+s->quarter_sample;
1079 const int xy= mb_x + mb_y*s->mb_stride;
1080 init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
1081
1082 av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
1083
1084 c->pre_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
1085 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1086
1087 get_limits(s, 16*mb_x, 16*mb_y);
1088 c->skip=0;
1089
1090 P_LEFT[0] = s->p_mv_table[xy + 1][0];
1091 P_LEFT[1] = s->p_mv_table[xy + 1][1];
1092
1093 if(P_LEFT[0] < (c->xmin<<shift)) P_LEFT[0] = (c->xmin<<shift);
1094
1095 /* special case for first line */
1096 if (s->first_slice_line) {
1097 c->pred_x= P_LEFT[0];
1098 c->pred_y= P_LEFT[1];
1099 P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
1100 P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
1101 } else {
1102 P_TOP[0] = s->p_mv_table[xy + s->mb_stride ][0];
1103 P_TOP[1] = s->p_mv_table[xy + s->mb_stride ][1];
1104 P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
1105 P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
1106 if(P_TOP[1] < (c->ymin<<shift)) P_TOP[1] = (c->ymin<<shift);
1107 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
1108 if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);
1109
1110 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1111 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1112
1113 c->pred_x = P_MEDIAN[0];
1114 c->pred_y = P_MEDIAN[1];
1115 }
1116
1117 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1118
1119 s->p_mv_table[xy][0] = mx<<shift;
1120 s->p_mv_table[xy][1] = my<<shift;
1121
1122 return dmin;
1123 }
1124
1125 static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,
1126 int16_t (*mv_table)[2], int ref_index, int f_code)
1127 {
1128 MotionEstContext * const c= &s->me;
1129 int mx, my, dmin;
1130 int P[10][2];
1131 const int shift= 1+s->quarter_sample;
1132 const int mot_stride = s->mb_stride;
1133 const int mot_xy = mb_y*mot_stride + mb_x;
1134 uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;
1135 int mv_scale;
1136
1137 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1138 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1139 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1140 c->current_mv_penalty= mv_penalty;
1141
1142 get_limits(s, 16*mb_x, 16*mb_y);
1143
1144 switch(s->me_method) {
1145 case ME_ZERO:
1146 default:
1147 mx = 0;
1148 my = 0;
1149 dmin = 0;
1150 break;
1151 case ME_X1:
1152 case ME_EPZS:
1153 P_LEFT[0] = mv_table[mot_xy - 1][0];
1154 P_LEFT[1] = mv_table[mot_xy - 1][1];
1155
1156 if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);
1157
1158 /* special case for first line */
1159 if (!s->first_slice_line) {
1160 P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
1161 P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
1162 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
1163 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
1164 if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);
1165 if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);
1166 if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);
1167
1168 P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1169 P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1170 }
1171 c->pred_x = P_LEFT[0];
1172 c->pred_y = P_LEFT[1];
1173
1174 if(mv_table == s->b_forw_mv_table){
1175 mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);
1176 }else{
1177 mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
1178 }
1179
1180 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
1181
1182 break;
1183 }
1184
1185 dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
1186
1187 if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1188 dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
1189
1190 // s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
1191 mv_table[mot_xy][0]= mx;
1192 mv_table[mot_xy][1]= my;
1193
1194 return dmin;
1195 }
1196
1197 static inline int check_bidir_mv(MpegEncContext * s,
1198 int motion_fx, int motion_fy,
1199 int motion_bx, int motion_by,
1200 int pred_fx, int pred_fy,
1201 int pred_bx, int pred_by,
1202 int size, int h)
1203 {
1204 //FIXME optimize?
1205 //FIXME better f_code prediction (max mv & distance)
1206 //FIXME pointers
1207 MotionEstContext * const c= &s->me;
1208 uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
1209 uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame
1210 int stride= c->stride;
1211 uint8_t *dest_y = c->scratchpad;
1212 uint8_t *ptr;
1213 int dxy;
1214 int src_x, src_y;
1215 int fbmin;
1216 uint8_t **src_data= c->src[0];
1217 uint8_t **ref_data= c->ref[0];
1218 uint8_t **ref2_data= c->ref[2];
1219
1220 if(s->quarter_sample){
1221 dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1222 src_x = motion_fx >> 2;
1223 src_y = motion_fy >> 2;
1224
1225 ptr = ref_data[0] + (src_y * stride) + src_x;
1226 s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr, stride);
1227
1228 dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1229 src_x = motion_bx >> 2;
1230 src_y = motion_by >> 2;
1231
1232 ptr = ref2_data[0] + (src_y * stride) + src_x;
1233 s->qdsp.avg_qpel_pixels_tab[size][dxy](dest_y, ptr, stride);
1234 }else{
1235 dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1236 src_x = motion_fx >> 1;
1237 src_y = motion_fy >> 1;
1238
1239 ptr = ref_data[0] + (src_y * stride) + src_x;
1240 s->hdsp.put_pixels_tab[size][dxy](dest_y , ptr , stride, h);
1241
1242 dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1243 src_x = motion_bx >> 1;
1244 src_y = motion_by >> 1;
1245
1246 ptr = ref2_data[0] + (src_y * stride) + src_x;
1247 s->hdsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
1248 }
1249
1250 fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
1251 +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
1252 + s->mecc.mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic
1253
1254 if(c->avctx->mb_cmp&FF_CMP_CHROMA){
1255 }
1256 //FIXME CHROMA !!!
1257
1258 return fbmin;
1259 }
1260
1261 /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
1262 static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
1263 {
1264 MotionEstContext * const c= &s->me;
1265 const int mot_stride = s->mb_stride;
1266 const int xy = mb_y *mot_stride + mb_x;
1267 int fbmin;
1268 int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];
1269 int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];
1270 int pred_bx= s->b_bidir_back_mv_table[xy-1][0];
1271 int pred_by= s->b_bidir_back_mv_table[xy-1][1];
1272 int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];
1273 int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];
1274 int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];
1275 int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];
1276 const int flags= c->sub_flags;
1277 const int qpel= flags&FLAG_QPEL;
1278 const int shift= 1+qpel;
1279 const int xmin= c->xmin<<shift;
1280 const int ymin= c->ymin<<shift;
1281 const int xmax= c->xmax<<shift;
1282 const int ymax= c->ymax<<shift;
1283 #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
1284 #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
1285 int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);
1286 uint8_t map[256] = { 0 };
1287
1288 map[hashidx&255] = 1;
1289
1290 fbmin= check_bidir_mv(s, motion_fx, motion_fy,
1291 motion_bx, motion_by,
1292 pred_fx, pred_fy,
1293 pred_bx, pred_by,
1294 0, 16);
1295
1296 if(s->avctx->bidir_refine){
1297 int end;
1298 static const uint8_t limittab[5]={0,8,32,64,80};
1299 const int limit= limittab[s->avctx->bidir_refine];
1300 static const int8_t vect[][4]={
1301 { 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},
1302
1303 { 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},
1304 { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
1305 { 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},
1306 { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
1307
1308 { 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},
1309 { 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},
1310 { 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},
1311 { 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},
1312
1313 { 1, 1, 1, 1}, {-1,-1,-1,-1},
1314 { 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},
1315 { 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},
1316 };
1317 static const uint8_t hash[]={
1318 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),
1319
1320 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),
1321 HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),
1322 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),
1323 HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),
1324
1325 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),
1326 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),
1327 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),
1328 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),
1329
1330 HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),
1331 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),
1332 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),
1333 };
1334
1335 #define CHECK_BIDIR(fx,fy,bx,by)\
1336 if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
1337 &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1338 &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1339 int score;\
1340 map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
1341 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);\
1342 if(score < fbmin){\
1343 hashidx += HASH(fx,fy,bx,by);\
1344 fbmin= score;\
1345 motion_fx+=fx;\
1346 motion_fy+=fy;\
1347 motion_bx+=bx;\
1348 motion_by+=by;\
1349 end=0;\
1350 }\
1351 }
1352 #define CHECK_BIDIR2(a,b,c,d)\
1353 CHECK_BIDIR(a,b,c,d)\
1354 CHECK_BIDIR(-(a),-(b),-(c),-(d))
1355
1356 do{
1357 int i;
1358 int borderdist=0;
1359 end=1;
1360
1361 CHECK_BIDIR2(0,0,0,1)
1362 CHECK_BIDIR2(0,0,1,0)
1363 CHECK_BIDIR2(0,1,0,0)
1364 CHECK_BIDIR2(1,0,0,0)
1365
1366 for(i=8; i<limit; i++){
1367 int fx= motion_fx+vect[i][0];
1368 int fy= motion_fy+vect[i][1];
1369 int bx= motion_bx+vect[i][2];
1370 int by= motion_by+vect[i][3];
1371 if(borderdist<=0){
1372 int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);
1373 int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);
1374 if((a|b) < 0)
1375 map[(hashidx+hash[i])&255] = 1;
1376 }
1377 if(!map[(hashidx+hash[i])&255]){
1378 int score;
1379 map[(hashidx+hash[i])&255] = 1;
1380 score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
1381 if(score < fbmin){
1382 hashidx += hash[i];
1383 fbmin= score;
1384 motion_fx=fx;
1385 motion_fy=fy;
1386 motion_bx=bx;
1387 motion_by=by;
1388 end=0;
1389 borderdist--;
1390 if(borderdist<=0){
1391 int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);
1392 int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);
1393 borderdist= FFMIN(a,b);
1394 }
1395 }
1396 }
1397 }
1398 }while(!end);
1399 }
1400
1401 s->b_bidir_forw_mv_table[xy][0]= motion_fx;
1402 s->b_bidir_forw_mv_table[xy][1]= motion_fy;
1403 s->b_bidir_back_mv_table[xy][0]= motion_bx;
1404 s->b_bidir_back_mv_table[xy][1]= motion_by;
1405
1406 return fbmin;
1407 }
1408
1409 static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
1410 {
1411 MotionEstContext * const c= &s->me;
1412 int P[10][2];
1413 const int mot_stride = s->mb_stride;
1414 const int mot_xy = mb_y*mot_stride + mb_x;
1415 const int shift= 1+s->quarter_sample;
1416 int dmin, i;
1417 const int time_pp= s->pp_time;
1418 const int time_pb= s->pb_time;
1419 int mx, my, xmin, xmax, ymin, ymax;
1420 int16_t (*mv_table)[2]= s->b_direct_mv_table;
1421
1422 c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;
1423 ymin= xmin=(-32)>>shift;
1424 ymax= xmax= 31>>shift;
1425
1426 if (IS_8X8(s->next_picture.mb_type[mot_xy])) {
1427 s->mv_type= MV_TYPE_8X8;
1428 }else{
1429 s->mv_type= MV_TYPE_16X16;
1430 }
1431
1432 for(i=0; i<4; i++){
1433 int index= s->block_index[i];
1434 int min, max;
1435
1436 c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];
1437 c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];
1438 c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
1439 c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
1440 // c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
1441 // c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
1442
1443 max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1444 min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1445 max+= 16*mb_x + 1; // +-1 is for the simpler rounding
1446 min+= 16*mb_x - 1;
1447 xmax= FFMIN(xmax, s->width - max);
1448 xmin= FFMAX(xmin, - 16 - min);
1449
1450 max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1451 min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1452 max+= 16*mb_y + 1; // +-1 is for the simpler rounding
1453 min+= 16*mb_y - 1;
1454 ymax= FFMIN(ymax, s->height - max);
1455 ymin= FFMAX(ymin, - 16 - min);
1456
1457 if(s->mv_type == MV_TYPE_16X16) break;
1458 }
1459
1460 av_assert2(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1461
1462 if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1463 s->b_direct_mv_table[mot_xy][0]= 0;
1464 s->b_direct_mv_table[mot_xy][1]= 0;
1465
1466 return 256*256*256*64;
1467 }
1468
1469 c->xmin= xmin;
1470 c->ymin= ymin;
1471 c->xmax= xmax;
1472 c->ymax= ymax;
1473 c->flags |= FLAG_DIRECT;
1474 c->sub_flags |= FLAG_DIRECT;
1475 c->pred_x=0;
1476 c->pred_y=0;
1477
1478 P_LEFT[0] = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
1479 P_LEFT[1] = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
1480
1481 /* special case for first line */
1482 if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped
1483 P_TOP[0] = av_clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
1484 P_TOP[1] = av_clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
1485 P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
1486 P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, ymax<<shift);
1487
1488 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1489 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1490 }
1491
1492 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
1493 if(c->sub_flags&FLAG_QPEL)
1494 dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1495 else
1496 dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1497
1498 if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1499 dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1500
1501 get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
1502
1503 mv_table[mot_xy][0]= mx;
1504 mv_table[mot_xy][1]= my;
1505 c->flags &= ~FLAG_DIRECT;
1506 c->sub_flags &= ~FLAG_DIRECT;
1507
1508 return dmin;
1509 }
1510
1511 void ff_estimate_b_frame_motion(MpegEncContext * s,
1512 int mb_x, int mb_y)
1513 {
1514 MotionEstContext * const c= &s->me;
1515 const int penalty_factor= c->mb_penalty_factor;
1516 int fmin, bmin, dmin, fbmin, bimin, fimin;
1517 int type=0;
1518 const int xy = mb_y*s->mb_stride + mb_x;
1519 init_ref(c, s->new_picture.f->data, s->last_picture.f->data,
1520 s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);
1521
1522 get_limits(s, 16*mb_x, 16*mb_y);
1523
1524 c->skip=0;
1525
1526 if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {
1527 int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
1528
1529 score= ((unsigned)(score*score + 128*256))>>16;
1530 c->mc_mb_var_sum_temp += score;
1531 s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1532 s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
1533
1534 return;
1535 }
1536
1537 if (s->codec_id == AV_CODEC_ID_MPEG4)
1538 dmin= direct_search(s, mb_x, mb_y);
1539 else
1540 dmin= INT_MAX;
1541 //FIXME penalty stuff for non mpeg4
1542 c->skip=0;
1543 fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +
1544 3 * penalty_factor;
1545
1546 c->skip=0;
1547 bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +
1548 2 * penalty_factor;
1549 av_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
1550
1551 c->skip=0;
1552 fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
1553 av_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);
1554
1555 if(s->flags & CODEC_FLAG_INTERLACED_ME){
1556 //FIXME mb type penalty
1557 c->skip=0;
1558 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1559 fimin= interlaced_search(s, 0,
1560 s->b_field_mv_table[0], s->b_field_select_table[0],
1561 s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
1562 c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
1563 bimin= interlaced_search(s, 2,
1564 s->b_field_mv_table[1], s->b_field_select_table[1],
1565 s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);
1566 }else
1567 fimin= bimin= INT_MAX;
1568
1569 {
1570 int score= fmin;
1571 type = CANDIDATE_MB_TYPE_FORWARD;
1572
1573 if (dmin <= score){
1574 score = dmin;
1575 type = CANDIDATE_MB_TYPE_DIRECT;
1576 }
1577 if(bmin<score){
1578 score=bmin;
1579 type= CANDIDATE_MB_TYPE_BACKWARD;
1580 }
1581 if(fbmin<score){
1582 score=fbmin;
1583 type= CANDIDATE_MB_TYPE_BIDIR;
1584 }
1585 if(fimin<score){
1586 score=fimin;
1587 type= CANDIDATE_MB_TYPE_FORWARD_I;
1588 }
1589 if(bimin<score){
1590 score=bimin;
1591 type= CANDIDATE_MB_TYPE_BACKWARD_I;
1592 }
1593
1594 score= ((unsigned)(score*score + 128*256))>>16;
1595 c->mc_mb_var_sum_temp += score;
1596 s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1597 }
1598
1599 if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1600 type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
1601 if(fimin < INT_MAX)
1602 type |= CANDIDATE_MB_TYPE_FORWARD_I;
1603 if(bimin < INT_MAX)
1604 type |= CANDIDATE_MB_TYPE_BACKWARD_I;
1605 if(fimin < INT_MAX && bimin < INT_MAX){
1606 type |= CANDIDATE_MB_TYPE_BIDIR_I;
1607 }
1608 //FIXME something smarter
1609 if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB
1610 if (s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT &&
1611 s->mpv_flags & FF_MPV_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
1612 type |= CANDIDATE_MB_TYPE_DIRECT0;
1613 }
1614
1615 s->mb_type[mb_y*s->mb_stride + mb_x]= type;
1616 }
1617
1618 /* find best f_code for ME which do unlimited searches */
1619 int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)
1620 {
1621 if(s->me_method>=ME_EPZS){
1622 int score[8];
1623 int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);
1624 uint8_t * fcode_tab= s->fcode_tab;
1625 int best_fcode=-1;
1626 int best_score=-10000000;
1627
1628 if(s->msmpeg4_version)
1629 range= FFMIN(range, 16);
1630 else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
1631 range= FFMIN(range, 256);
1632
1633 for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);
1634
1635 for(y=0; y<s->mb_height; y++){
1636 int x;
1637 int xy= y*s->mb_stride;
1638 for(x=0; x<s->mb_width; x++){
1639 if(s->mb_type[xy] & type){
1640 int mx= mv_table[xy][0];
1641 int my= mv_table[xy][1];
1642 int fcode= FFMAX(fcode_tab[mx + MAX_MV],
1643 fcode_tab[my + MAX_MV]);
1644 int j;
1645
1646 if(mx >= range || mx < -range ||
1647 my >= range || my < -range)
1648 continue;
1649
1650 for(j=0; j<fcode && j<8; j++){
1651 if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
1652 score[j]-= 170;
1653 }
1654 }
1655 xy++;
1656 }
1657 }
1658
1659 for(i=1; i<8; i++){
1660 if(score[i] > best_score){
1661 best_score= score[i];
1662 best_fcode= i;
1663 }
1664 }
1665
1666 return best_fcode;
1667 }else{
1668 return 1;
1669 }
1670 }
1671
1672 void ff_fix_long_p_mvs(MpegEncContext * s)
1673 {
1674 MotionEstContext * const c= &s->me;
1675 const int f_code= s->f_code;
1676 int y, range;
1677 av_assert0(s->pict_type==AV_PICTURE_TYPE_P);
1678
1679 range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1680
1681 av_assert0(range <= 16 || !s->msmpeg4_version);
1682 av_assert0(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
1683
1684 if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1685
1686 if(s->flags&CODEC_FLAG_4MV){
1687 const int wrap= s->b8_stride;
1688
1689 /* clip / convert to intra 8x8 type MVs */
1690 for(y=0; y<s->mb_height; y++){
1691 int xy= y*2*wrap;
1692 int i= y*s->mb_stride;
1693 int x;
1694
1695 for(x=0; x<s->mb_width; x++){
1696 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){
1697 int block;
1698 for(block=0; block<4; block++){
1699 int off= (block& 1) + (block>>1)*wrap;
1700 int mx = s->current_picture.motion_val[0][ xy + off ][0];
1701 int my = s->current_picture.motion_val[0][ xy + off ][1];
1702
1703 if( mx >=range || mx <-range
1704 || my >=range || my <-range){
1705 s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;
1706 s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;
1707 s->current_picture.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;
1708 }
1709 }
1710 }
1711 xy+=2;
1712 i++;
1713 }
1714 }
1715 }
1716 }
1717
1718 /**
1719 *
1720 * @param truncate 1 for truncation, 0 for using intra
1721 */
1722 void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
1723 int16_t (*mv_table)[2], int f_code, int type, int truncate)
1724 {
1725 MotionEstContext * const c= &s->me;
1726 int y, h_range, v_range;
1727
1728 // RAL: 8 in MPEG-1, 16 in MPEG-4
1729 int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1730
1731 if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1732
1733 h_range= range;
1734 v_range= field_select_table ? range>>1 : range;
1735
1736 /* clip / convert to intra 16x16 type MVs */
1737 for(y=0; y<s->mb_height; y++){
1738 int x;
1739 int xy= y*s->mb_stride;
1740 for(x=0; x<s->mb_width; x++){
1741 if (s->mb_type[xy] & type){ // RAL: "type" test added...
1742 if (!field_select_table || field_select_table[xy] == field_select) {
1743 if( mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
1744 || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
1745
1746 if(truncate){
1747 if (mv_table[xy][0] > h_range-1) mv_table[xy][0]= h_range-1;
1748 else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
1749 if (mv_table[xy][1] > v_range-1) mv_table[xy][1]= v_range-1;
1750 else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
1751 }else{
1752 s->mb_type[xy] &= ~type;
1753 s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;
1754 mv_table[xy][0]=
1755 mv_table[xy][1]= 0;
1756 }
1757 }
1758 }
1759 }
1760 xy++;
1761 }
1762 }
1763 }