3 * Copyright (c) 2000,2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer
6 * new motion estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
8 * This file is part of FFmpeg.
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.
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.
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
33 #include "libavutil/intmath.h"
37 #include "mpegvideo.h"
42 #define SQ(a) ((a)*(a))
46 #define P_TOPRIGHT P[3]
50 static inline int sad_hpel_motion_search(MpegEncContext * s,
51 int *mx_ptr, int *my_ptr, int dmin,
52 int src_index, int ref_index,
55 static inline unsigned update_map_generation(MotionEstContext *c)
57 c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
58 if(c->map_generation==0){
59 c->map_generation= 1<<(ME_MAP_MV_BITS*2);
60 memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
62 return c->map_generation;
65 /* shape adaptive search stuff */
66 typedef struct Minima{
72 static int minima_cmp(const void *a, const void *b){
73 const Minima *da = (const Minima *) a;
74 const Minima *db = (const Minima *) b;
76 return da->height - db->height;
79 #define FLAG_QPEL 1 //must be 1
83 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){
84 const int offset[3]= {
86 ((y*c->uvstride + x)>>1),
87 ((y*c->uvstride + x)>>1),
91 c->src[0][i]= src [i] + offset[i];
92 c->ref[0][i]= ref [i] + offset[i];
96 c->ref[ref_index][i]= ref2[i] + offset[i];
101 static int get_flags(MotionEstContext *c, int direct, int chroma){
102 return ((c->avctx->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
103 + (direct ? FLAG_DIRECT : 0)
104 + (chroma ? FLAG_CHROMA : 0);
107 static av_always_inline int cmp_direct_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
108 const int size, const int h, int ref_index, int src_index,
109 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel){
110 MotionEstContext * const c= &s->me;
111 const int stride= c->stride;
112 const int hx= subx + (x<<(1+qpel));
113 const int hy= suby + (y<<(1+qpel));
114 uint8_t * const * const ref= c->ref[ref_index];
115 uint8_t * const * const src= c->src[src_index];
117 //FIXME check chroma 4mv, (no crashes ...)
118 assert(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));
119 if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
120 const int time_pp= s->pp_time;
121 const int time_pb= s->pb_time;
122 const int mask= 2*qpel+1;
123 if(s->mv_type==MV_TYPE_8X8){
126 int fx = c->direct_basis_mv[i][0] + hx;
127 int fy = c->direct_basis_mv[i][1] + hy;
128 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));
129 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));
130 int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
131 int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
133 uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);
135 c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);
136 c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);
138 c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);
139 c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);
143 int fx = c->direct_basis_mv[0][0] + hx;
144 int fy = c->direct_basis_mv[0][1] + hy;
145 int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
146 int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
147 int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
148 int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
151 c->qpel_put[1][fxy](c->temp , ref[0] + (fx>>2) + (fy>>2)*stride , stride);
152 c->qpel_put[1][fxy](c->temp + 8 , ref[0] + (fx>>2) + (fy>>2)*stride + 8 , stride);
153 c->qpel_put[1][fxy](c->temp + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8*stride, stride);
154 c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);
155 c->qpel_avg[1][bxy](c->temp , ref[8] + (bx>>2) + (by>>2)*stride , stride);
156 c->qpel_avg[1][bxy](c->temp + 8 , ref[8] + (bx>>2) + (by>>2)*stride + 8 , stride);
157 c->qpel_avg[1][bxy](c->temp + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8*stride, stride);
158 c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);
160 assert((fx>>1) + 16*s->mb_x >= -16);
161 assert((fy>>1) + 16*s->mb_y >= -16);
162 assert((fx>>1) + 16*s->mb_x <= s->width);
163 assert((fy>>1) + 16*s->mb_y <= s->height);
164 assert((bx>>1) + 16*s->mb_x >= -16);
165 assert((by>>1) + 16*s->mb_y >= -16);
166 assert((bx>>1) + 16*s->mb_x <= s->width);
167 assert((by>>1) + 16*s->mb_y <= s->height);
169 c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);
170 c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);
173 d = cmp_func(s, c->temp, src[0], stride, 16);
179 static av_always_inline int cmp_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
180 const int size, const int h, int ref_index, int src_index,
181 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel, int chroma){
182 MotionEstContext * const c= &s->me;
183 const int stride= c->stride;
184 const int uvstride= c->uvstride;
185 const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
186 const int hx= subx + (x<<(1+qpel));
187 const int hy= suby + (y<<(1+qpel));
188 uint8_t * const * const ref= c->ref[ref_index];
189 uint8_t * const * const src= c->src[src_index];
191 //FIXME check chroma 4mv, (no crashes ...)
192 int uvdxy; /* no, it might not be used uninitialized */
195 c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
201 uvdxy= (cx&1) + 2*(cy&1);
202 //FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264
205 c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
207 uvdxy= dxy | (x&1) | (2*(y&1));
209 d = cmp_func(s, c->temp, src[0], stride, h);
211 d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
213 uvdxy= (x&1) + 2*(y&1);
216 uint8_t * const uvtemp= c->temp + 16*stride;
217 c->hpel_put[size+1][uvdxy](uvtemp , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
218 c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
219 d += chroma_cmp_func(s, uvtemp , src[1], uvstride, h>>1);
220 d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
225 static int cmp_simple(MpegEncContext *s, const int x, const int y,
226 int ref_index, int src_index,
227 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){
228 return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0);
231 static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y,
232 const int size, const int h, int ref_index, int src_index,
233 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
234 if(flags&FLAG_DIRECT){
235 return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
237 return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
241 static int cmp_internal(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
242 const int size, const int h, int ref_index, int src_index,
243 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
244 if(flags&FLAG_DIRECT){
245 return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
247 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);
251 /** @brief compares a block (either a full macroblock or a partition thereof)
252 against a proposed motion-compensated prediction of that block
254 static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
255 const int size, const int h, int ref_index, int src_index,
256 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
257 if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size)
258 && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
259 && flags==0 && h==16 && size==0 && subx==0 && suby==0){
260 return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func);
261 }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
262 && subx==0 && suby==0){
263 return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags);
265 return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags);
269 static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
270 const int size, const int h, int ref_index, int src_index,
271 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
272 if(flags&FLAG_DIRECT){
273 return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0);
275 return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
279 static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
280 const int size, const int h, int ref_index, int src_index,
281 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
282 if(flags&FLAG_DIRECT){
283 return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1);
285 return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA);
289 #include "motion_est_template.c"
291 static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){
295 static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){
298 int ff_init_me(MpegEncContext *s){
299 MotionEstContext * const c= &s->me;
300 int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
301 int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
303 if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -ME_MAP_SIZE){
304 av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n");
307 //special case of snow is needed because snow uses its own iterative ME code
308 if(s->me_method!=ME_ZERO && s->me_method!=ME_EPZS && s->me_method!=ME_X1 && s->avctx->codec_id != CODEC_ID_SNOW){
309 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");
315 if(cache_size < 2*dia_size && !c->stride){
316 av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
319 ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp);
320 ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp);
321 ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp);
322 ff_set_cmp(&s->dsp, s->dsp.mb_cmp, c->avctx->mb_cmp);
324 c->flags = get_flags(c, 0, c->avctx->me_cmp &FF_CMP_CHROMA);
325 c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
326 c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp &FF_CMP_CHROMA);
328 /*FIXME s->no_rounding b_type*/
329 if(s->flags&CODEC_FLAG_QPEL){
330 c->sub_motion_search= qpel_motion_search;
331 c->qpel_avg= s->dsp.avg_qpel_pixels_tab;
332 if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab;
333 else c->qpel_put= s->dsp.put_qpel_pixels_tab;
335 if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
336 c->sub_motion_search= hpel_motion_search;
337 else if( c->avctx->me_sub_cmp == FF_CMP_SAD
338 && c->avctx-> me_cmp == FF_CMP_SAD
339 && c->avctx-> mb_cmp == FF_CMP_SAD)
340 c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
342 c->sub_motion_search= hpel_motion_search;
344 c->hpel_avg= s->dsp.avg_pixels_tab;
345 if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab;
346 else c->hpel_put= s->dsp.put_pixels_tab;
349 c->stride = s->linesize;
350 c->uvstride= s->uvlinesize;
352 c->stride = 16*s->mb_width + 32;
353 c->uvstride= 8*s->mb_width + 16;
356 /* 8x8 fullpel search would need a 4x4 chroma compare, which we do
357 * not have yet, and even if we had, the motion estimation code
358 * does not expect it. */
359 if(s->codec_id != CODEC_ID_SNOW){
360 if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){
361 s->dsp.me_cmp[2]= zero_cmp;
363 if((c->avctx->me_sub_cmp&FF_CMP_CHROMA) && !s->dsp.me_sub_cmp[2]){
364 s->dsp.me_sub_cmp[2]= zero_cmp;
366 c->hpel_put[2][0]= c->hpel_put[2][1]=
367 c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
370 if(s->codec_id == CODEC_ID_H261){
371 c->sub_motion_search= no_sub_motion_search;
377 static inline void no_motion_search(MpegEncContext * s,
378 int *mx_ptr, int *my_ptr)
380 *mx_ptr = 16 * s->mb_x;
381 *my_ptr = 16 * s->mb_y;
384 #define Z_THRESHOLD 256
386 #define CHECK_SAD_HALF_MV(suffix, x, y) \
388 d= s->dsp.pix_abs[size][(x?1:0)+(y?2:0)](NULL, pix, ptr+((x)>>1), stride, h);\
389 d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
390 COPY3_IF_LT(dminh, d, dx, x, dy, y)\
393 static inline int sad_hpel_motion_search(MpegEncContext * s,
394 int *mx_ptr, int *my_ptr, int dmin,
395 int src_index, int ref_index,
398 MotionEstContext * const c= &s->me;
399 const int penalty_factor= c->sub_penalty_factor;
402 int stride= c->stride;
403 const int flags= c->sub_flags;
416 pix = c->src[src_index][0];
420 ptr = c->ref[ref_index][0] + (my * stride) + mx;
424 if (mx > xmin && mx < xmax &&
425 my > ymin && my < ymax) {
428 const int index= (my<<ME_MAP_SHIFT) + mx;
429 const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
430 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
431 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)];
432 const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
442 CHECK_SAD_HALF_MV(y2 , 0, -1)
444 CHECK_SAD_HALF_MV(xy2, -1, -1)
446 CHECK_SAD_HALF_MV(xy2, +1, -1)
450 CHECK_SAD_HALF_MV(xy2, -1, +1)
452 CHECK_SAD_HALF_MV(x2 , -1, 0)
454 CHECK_SAD_HALF_MV(xy2, +1, -1)
456 CHECK_SAD_HALF_MV(xy2, -1, -1)
460 CHECK_SAD_HALF_MV(xy2, +1, +1)
462 CHECK_SAD_HALF_MV(x2 , +1, 0)
467 CHECK_SAD_HALF_MV(xy2, -1, -1)
471 CHECK_SAD_HALF_MV(xy2, +1, +1)
473 CHECK_SAD_HALF_MV(x2 , -1, 0)
474 CHECK_SAD_HALF_MV(xy2, -1, +1)
477 CHECK_SAD_HALF_MV(xy2, +1, -1)
481 CHECK_SAD_HALF_MV(xy2, -1, +1)
483 CHECK_SAD_HALF_MV(x2 , +1, 0)
484 CHECK_SAD_HALF_MV(xy2, +1, +1)
486 CHECK_SAD_HALF_MV(y2 , 0, +1)
501 static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
503 const int xy= s->mb_x + s->mb_y*s->mb_stride;
505 s->p_mv_table[xy][0] = mx;
506 s->p_mv_table[xy][1] = my;
508 /* has already been set to the 4 MV if 4MV is done */
510 int mot_xy= s->block_index[0];
512 s->current_picture.f.motion_val[0][mot_xy ][0] = mx;
513 s->current_picture.f.motion_val[0][mot_xy ][1] = my;
514 s->current_picture.f.motion_val[0][mot_xy + 1][0] = mx;
515 s->current_picture.f.motion_val[0][mot_xy + 1][1] = my;
517 mot_xy += s->b8_stride;
518 s->current_picture.f.motion_val[0][mot_xy ][0] = mx;
519 s->current_picture.f.motion_val[0][mot_xy ][1] = my;
520 s->current_picture.f.motion_val[0][mot_xy + 1][0] = mx;
521 s->current_picture.f.motion_val[0][mot_xy + 1][1] = my;
526 * get fullpel ME search limits.
528 static inline void get_limits(MpegEncContext *s, int x, int y)
530 MotionEstContext * const c= &s->me;
531 int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
533 if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
536 if (s->unrestricted_mv) {
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;
550 c->xmax = - x + s->mb_width *16 - 16;
551 c->ymax = - y + s->mb_height*16 - 16;
554 c->xmin = FFMAX(c->xmin,-range);
555 c->xmax = FFMIN(c->xmax, range);
556 c->ymin = FFMAX(c->ymin,-range);
557 c->ymax = FFMIN(c->ymax, range);
561 static inline void init_mv4_ref(MotionEstContext *c){
562 const int stride= c->stride;
564 c->ref[1][0] = c->ref[0][0] + 8;
565 c->ref[2][0] = c->ref[0][0] + 8*stride;
566 c->ref[3][0] = c->ref[2][0] + 8;
567 c->src[1][0] = c->src[0][0] + 8;
568 c->src[2][0] = c->src[0][0] + 8*stride;
569 c->src[3][0] = c->src[2][0] + 8;
572 static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
574 MotionEstContext * const c= &s->me;
579 int dmin_sum=0, mx4_sum=0, my4_sum=0, i;
581 const int stride= c->stride;
582 uint8_t *mv_penalty= c->current_mv_penalty;
583 int saftey_cliping= s->unrestricted_mv && (s->width&15) && (s->height&15);
587 for(block=0; block<4; block++){
589 int pred_x4, pred_y4;
591 static const int off[4]= {2, 1, 1, -1};
592 const int mot_stride = s->b8_stride;
593 const int mot_xy = s->block_index[block];
596 c->xmax = - 16*s->mb_x + s->width - 8*(block &1);
597 c->ymax = - 16*s->mb_y + s->height - 8*(block>>1);
600 P_LEFT[0] = s->current_picture.f.motion_val[0][mot_xy - 1][0];
601 P_LEFT[1] = s->current_picture.f.motion_val[0][mot_xy - 1][1];
603 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
605 /* special case for first line */
606 if (s->first_slice_line && block<2) {
607 c->pred_x= pred_x4= P_LEFT[0];
608 c->pred_y= pred_y4= P_LEFT[1];
610 P_TOP[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][0];
611 P_TOP[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][1];
612 P_TOPRIGHT[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + off[block]][0];
613 P_TOPRIGHT[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + off[block]][1];
614 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
615 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
616 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
617 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
619 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
620 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
622 c->pred_x= pred_x4 = P_MEDIAN[0];
623 c->pred_y= pred_y4 = P_MEDIAN[1];
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);
633 dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
635 dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
637 if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
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);
646 s->dsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y , ref , stride);
648 s->dsp.put_qpel_pixels_tab [1][dxy](dest_y , ref , stride);
650 uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
651 dxy = ((my4 & 1) << 1) | (mx4 & 1);
654 s->dsp.put_no_rnd_pixels_tab[1][dxy](dest_y , ref , stride, h);
656 s->dsp.put_pixels_tab [1][dxy](dest_y , ref , stride, h);
658 dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
662 if(s->quarter_sample){
670 s->current_picture.f.motion_val[0][s->block_index[block]][0] = mx4;
671 s->current_picture.f.motion_val[0][s->block_index[block]][1] = my4;
673 if(mx4 != mx || my4 != my) same=0;
679 if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
680 dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);
683 if(c->avctx->mb_cmp&FF_CMP_CHROMA){
688 mx= ff_h263_round_chroma(mx4_sum);
689 my= ff_h263_round_chroma(my4_sum);
690 dxy = ((my & 1) << 1) | (mx & 1);
692 offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
695 s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad , s->last_picture.f.data[1] + offset, s->uvlinesize, 8);
696 s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8);
698 s->dsp.put_pixels_tab [1][dxy](c->scratchpad , s->last_picture.f.data[1] + offset, s->uvlinesize, 8);
699 s->dsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8);
702 dmin_sum += s->dsp.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);
703 dmin_sum += s->dsp.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);
709 switch(c->avctx->mb_cmp&0xFF){
711 return dmin_sum+ 32*s->qscale*s->qscale;*/
715 return dmin_sum+ 11*c->mb_penalty_factor;
719 static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
720 MotionEstContext * const c= &s->me;
722 c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
723 c->src[1][0] = c->src[0][0] + s->linesize;
724 if(c->flags & FLAG_CHROMA){
725 c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
726 c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
727 c->src[1][1] = c->src[0][1] + s->uvlinesize;
728 c->src[1][2] = c->src[0][2] + s->uvlinesize;
732 static int interlaced_search(MpegEncContext *s, int ref_index,
733 int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
735 MotionEstContext * const c= &s->me;
740 uint8_t * const mv_penalty= c->current_mv_penalty;
742 const int stride= 2*s->linesize;
744 const int mot_stride= s->mb_stride;
745 const int xy= s->mb_x + s->mb_y*mot_stride;
751 init_interlaced_ref(s, ref_index);
753 for(block=0; block<2; block++){
755 int best_dmin= INT_MAX;
758 for(field_select=0; field_select<2; field_select++){
759 int dmin, mx_i, my_i;
760 int16_t (*mv_table)[2]= mv_tables[block][field_select];
762 if(user_field_select){
763 assert(field_select==0 || field_select==1);
764 assert(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);
765 if(field_select_tables[block][xy] != field_select)
769 P_LEFT[0] = mv_table[xy - 1][0];
770 P_LEFT[1] = mv_table[xy - 1][1];
771 if(P_LEFT[0] > (c->xmax<<1)) P_LEFT[0] = (c->xmax<<1);
773 c->pred_x= P_LEFT[0];
774 c->pred_y= P_LEFT[1];
776 if(!s->first_slice_line){
777 P_TOP[0] = mv_table[xy - mot_stride][0];
778 P_TOP[1] = mv_table[xy - mot_stride][1];
779 P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
780 P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
781 if(P_TOP[1] > (c->ymax<<1)) P_TOP[1] = (c->ymax<<1);
782 if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
783 if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
784 if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
786 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
787 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
789 P_MV1[0]= mx; //FIXME not correct if block != field_select
792 dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);
794 dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
796 mv_table[xy][0]= mx_i;
797 mv_table[xy][1]= my_i;
799 if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
803 uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
804 dxy = ((my_i & 1) << 1) | (mx_i & 1);
807 s->dsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref , stride, h);
809 s->dsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
811 dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
812 dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
814 dmin+= c->mb_penalty_factor; //field_select bits
816 dmin += field_select != block; //slightly prefer same field
818 if(dmin < best_dmin){
820 best_field= field_select;
824 int16_t (*mv_table)[2]= mv_tables[block][best_field];
826 if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all
827 if(mv_table[xy][1]&1) same=0;
828 if(mv_table[xy][1]*2 != my) same=0;
829 if(best_field != block) same=0;
832 field_select_tables[block][xy]= best_field;
833 dmin_sum += best_dmin;
844 switch(c->avctx->mb_cmp&0xFF){
846 return dmin_sum+ 32*s->qscale*s->qscale;*/
850 return dmin_sum+ 11*c->mb_penalty_factor;
854 static void clip_input_mv(MpegEncContext * s, int16_t *mv, int interlaced){
855 int ymax= s->me.ymax>>interlaced;
856 int ymin= s->me.ymin>>interlaced;
858 if(mv[0] < s->me.xmin) mv[0] = s->me.xmin;
859 if(mv[0] > s->me.xmax) mv[0] = s->me.xmax;
860 if(mv[1] < ymin) mv[1] = ymin;
861 if(mv[1] > ymax) mv[1] = ymax;
864 static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int p_type){
865 MotionEstContext * const c= &s->me;
866 Picture *p= s->current_picture_ptr;
867 int mb_xy= mb_x + mb_y*s->mb_stride;
868 int xy= 2*mb_x + 2*mb_y*s->b8_stride;
869 int mb_type= s->current_picture.f.mb_type[mb_xy];
871 int shift= (flags&FLAG_QPEL) + 1;
872 int mask= (1<<shift)-1;
875 me_cmp_func cmpf= s->dsp.sse[0];
876 me_cmp_func chroma_cmpf= s->dsp.sse[1];
878 if(p_type && USES_LIST(mb_type, 1)){
879 av_log(c->avctx, AV_LOG_ERROR, "backward motion vector in P frame\n");
882 assert(IS_INTRA(mb_type) || USES_LIST(mb_type,0) || USES_LIST(mb_type,1));
885 int xy= s->block_index[i];
886 clip_input_mv(s, p->f.motion_val[0][xy], !!IS_INTERLACED(mb_type));
887 clip_input_mv(s, p->f.motion_val[1][xy], !!IS_INTERLACED(mb_type));
890 if(IS_INTERLACED(mb_type)){
891 int xy2= xy + s->b8_stride;
892 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
896 if(!(s->flags & CODEC_FLAG_INTERLACED_ME)){
897 av_log(c->avctx, AV_LOG_ERROR, "Interlaced macroblock selected but interlaced motion estimation disabled\n");
901 if(USES_LIST(mb_type, 0)){
902 int field_select0= p->f.ref_index[0][4*mb_xy ];
903 int field_select1= p->f.ref_index[0][4*mb_xy+2];
904 assert(field_select0==0 ||field_select0==1);
905 assert(field_select1==0 ||field_select1==1);
906 init_interlaced_ref(s, 0);
909 s->p_field_select_table[0][mb_xy]= field_select0;
910 s->p_field_select_table[1][mb_xy]= field_select1;
911 *(uint32_t*)s->p_field_mv_table[0][field_select0][mb_xy] = *(uint32_t*)p->f.motion_val[0][xy ];
912 *(uint32_t*)s->p_field_mv_table[1][field_select1][mb_xy] = *(uint32_t*)p->f.motion_val[0][xy2];
913 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER_I;
915 s->b_field_select_table[0][0][mb_xy]= field_select0;
916 s->b_field_select_table[0][1][mb_xy]= field_select1;
917 *(uint32_t*)s->b_field_mv_table[0][0][field_select0][mb_xy] = *(uint32_t*)p->f.motion_val[0][xy ];
918 *(uint32_t*)s->b_field_mv_table[0][1][field_select1][mb_xy] = *(uint32_t*)p->f.motion_val[0][xy2];
919 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_FORWARD_I;
922 x = p->f.motion_val[0][xy ][0];
923 y = p->f.motion_val[0][xy ][1];
924 d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0, 0, cmpf, chroma_cmpf, flags);
925 x = p->f.motion_val[0][xy2][0];
926 y = p->f.motion_val[0][xy2][1];
927 d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1, 1, cmpf, chroma_cmpf, flags);
929 if(USES_LIST(mb_type, 1)){
930 int field_select0 = p->f.ref_index[1][4 * mb_xy ];
931 int field_select1 = p->f.ref_index[1][4 * mb_xy + 2];
932 assert(field_select0==0 ||field_select0==1);
933 assert(field_select1==0 ||field_select1==1);
934 init_interlaced_ref(s, 2);
936 s->b_field_select_table[1][0][mb_xy]= field_select0;
937 s->b_field_select_table[1][1][mb_xy]= field_select1;
938 *(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy] = *(uint32_t*)p->f.motion_val[1][xy ];
939 *(uint32_t*)s->b_field_mv_table[1][1][field_select1][mb_xy] = *(uint32_t*)p->f.motion_val[1][xy2];
940 if(USES_LIST(mb_type, 0)){
941 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BIDIR_I;
943 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BACKWARD_I;
946 x = p->f.motion_val[1][xy ][0];
947 y = p->f.motion_val[1][xy ][1];
948 d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0+2, 0, cmpf, chroma_cmpf, flags);
949 x = p->f.motion_val[1][xy2][0];
950 y = p->f.motion_val[1][xy2][1];
951 d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1+2, 1, cmpf, chroma_cmpf, flags);
956 }else if(IS_8X8(mb_type)){
957 if(!(s->flags & CODEC_FLAG_4MV)){
958 av_log(c->avctx, AV_LOG_ERROR, "4MV macroblock selected but 4MV encoding disabled\n");
962 chroma_cmpf= s->dsp.sse[1];
965 xy= s->block_index[i];
966 x= p->f.motion_val[0][xy][0];
967 y= p->f.motion_val[0][xy][1];
968 d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 1, 8, i, i, cmpf, chroma_cmpf, flags);
970 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER4V;
972 if(USES_LIST(mb_type, 0)){
974 *(uint32_t*)s->p_mv_table[mb_xy] = *(uint32_t*)p->f.motion_val[0][xy];
975 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER;
976 }else if(USES_LIST(mb_type, 1)){
977 *(uint32_t*)s->b_bidir_forw_mv_table[mb_xy] = *(uint32_t*)p->f.motion_val[0][xy];
978 *(uint32_t*)s->b_bidir_back_mv_table[mb_xy] = *(uint32_t*)p->f.motion_val[1][xy];
979 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BIDIR;
981 *(uint32_t*)s->b_forw_mv_table[mb_xy] = *(uint32_t*)p->f.motion_val[0][xy];
982 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_FORWARD;
984 x = p->f.motion_val[0][xy][0];
985 y = p->f.motion_val[0][xy][1];
986 d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 0, 0, cmpf, chroma_cmpf, flags);
987 }else if(USES_LIST(mb_type, 1)){
988 *(uint32_t*)s->b_back_mv_table[mb_xy] = *(uint32_t*)p->f.motion_val[1][xy];
989 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BACKWARD;
991 x = p->f.motion_val[1][xy][0];
992 y = p->f.motion_val[1][xy][1];
993 d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 2, 0, cmpf, chroma_cmpf, flags);
995 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
1000 void ff_estimate_p_frame_motion(MpegEncContext * s,
1003 MotionEstContext * const c= &s->me;
1004 uint8_t *pix, *ppix;
1005 int sum, mx, my, dmin;
1006 int varc; ///< the variance of the block (sum of squared (p[y][x]-average))
1007 int vard; ///< sum of squared differences with the estimated motion vector
1009 const int shift= 1+s->quarter_sample;
1011 Picture * const pic= &s->current_picture;
1013 init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);
1015 assert(s->quarter_sample==0 || s->quarter_sample==1);
1016 assert(s->linesize == c->stride);
1017 assert(s->uvlinesize == c->uvstride);
1019 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1020 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1021 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1022 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1024 get_limits(s, 16*mb_x, 16*mb_y);
1027 /* intra / predictive decision */
1029 sum = s->dsp.pix_sum(pix, s->linesize);
1030 varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500;
1032 pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1033 pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
1034 c->mb_var_sum_temp += (varc+128)>>8;
1036 if(c->avctx->me_threshold){
1037 vard= check_input_motion(s, mb_x, mb_y, 1);
1039 if((vard+128)>>8 < c->avctx->me_threshold){
1040 int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1041 int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1042 pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
1043 c->mc_mb_var_sum_temp += (vard+128)>>8;
1044 c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1047 if((vard+128)>>8 < c->avctx->mb_threshold)
1048 mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];
1051 switch(s->me_method) {
1054 no_motion_search(s, &mx, &my);
1062 const int mot_stride = s->b8_stride;
1063 const int mot_xy = s->block_index[0];
1065 P_LEFT[0] = s->current_picture.f.motion_val[0][mot_xy - 1][0];
1066 P_LEFT[1] = s->current_picture.f.motion_val[0][mot_xy - 1][1];
1068 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
1070 if(!s->first_slice_line) {
1071 P_TOP[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][0];
1072 P_TOP[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride ][1];
1073 P_TOPRIGHT[0] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + 2][0];
1074 P_TOPRIGHT[1] = s->current_picture.f.motion_val[0][mot_xy - mot_stride + 2][1];
1075 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
1076 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
1077 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
1079 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1080 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1082 if(s->out_format == FMT_H263){
1083 c->pred_x = P_MEDIAN[0];
1084 c->pred_y = P_MEDIAN[1];
1085 }else { /* mpeg1 at least */
1086 c->pred_x= P_LEFT[0];
1087 c->pred_y= P_LEFT[1];
1090 c->pred_x= P_LEFT[0];
1091 c->pred_y= P_LEFT[1];
1095 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1100 /* At this point (mx,my) are full-pell and the relative displacement */
1101 ppix = c->ref[0][0] + (my * s->linesize) + mx;
1103 vard = s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16);
1105 pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
1106 // pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
1107 c->mc_mb_var_sum_temp += (vard+128)>>8;
1110 int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1111 int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1112 c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1114 if(mb_type == CANDIDATE_MB_TYPE_INTER){
1115 c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1116 set_p_mv_tables(s, mx, my, 1);
1121 if(mb_type == CANDIDATE_MB_TYPE_INTER4V){
1122 h263_mv4_search(s, mx, my, shift);
1124 set_p_mv_tables(s, mx, my, 0);
1126 if(mb_type == CANDIDATE_MB_TYPE_INTER_I){
1127 interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);
1129 }else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1130 int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1131 int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1132 c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1134 if (vard*2 + 200*256 > varc)
1135 mb_type|= CANDIDATE_MB_TYPE_INTRA;
1136 if (varc*2 + 200*256 > vard || s->qscale > 24){
1137 // if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){
1138 mb_type|= CANDIDATE_MB_TYPE_INTER;
1139 c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1140 if(s->flags&CODEC_FLAG_MV0)
1142 mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference
1147 if((s->flags&CODEC_FLAG_4MV)
1148 && !c->skip && varc>50<<8 && vard>10<<8){
1149 if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
1150 mb_type|=CANDIDATE_MB_TYPE_INTER4V;
1152 set_p_mv_tables(s, mx, my, 0);
1154 set_p_mv_tables(s, mx, my, 1);
1155 if((s->flags&CODEC_FLAG_INTERLACED_ME)
1156 && !c->skip){ //FIXME varc/d checks
1157 if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
1158 mb_type |= CANDIDATE_MB_TYPE_INTER_I;
1162 mb_type= CANDIDATE_MB_TYPE_INTER;
1164 dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1165 if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1166 dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1168 if((s->flags&CODEC_FLAG_4MV)
1169 && !c->skip && varc>50<<8 && vard>10<<8){
1170 int dmin4= h263_mv4_search(s, mx, my, shift);
1172 mb_type= CANDIDATE_MB_TYPE_INTER4V;
1176 if((s->flags&CODEC_FLAG_INTERLACED_ME)
1177 && !c->skip){ //FIXME varc/d checks
1178 int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
1180 mb_type = CANDIDATE_MB_TYPE_INTER_I;
1185 // pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
1186 set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
1188 /* get intra luma score */
1189 if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
1190 intra_score= varc - 500;
1192 unsigned mean = (sum+128)>>8;
1195 for(i=0; i<16; i++){
1196 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
1197 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
1198 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
1199 *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
1202 intra_score= s->dsp.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
1204 intra_score += c->mb_penalty_factor*16;
1206 if(intra_score < dmin){
1207 mb_type= CANDIDATE_MB_TYPE_INTRA;
1208 s->current_picture.f.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
1210 s->current_picture.f.mb_type[mb_y*s->mb_stride + mb_x] = 0;
1213 int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1214 int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1215 c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1219 s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;
1222 int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1225 MotionEstContext * const c= &s->me;
1228 const int shift= 1+s->quarter_sample;
1229 const int xy= mb_x + mb_y*s->mb_stride;
1230 init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);
1232 assert(s->quarter_sample==0 || s->quarter_sample==1);
1234 c->pre_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
1235 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1237 get_limits(s, 16*mb_x, 16*mb_y);
1240 P_LEFT[0] = s->p_mv_table[xy + 1][0];
1241 P_LEFT[1] = s->p_mv_table[xy + 1][1];
1243 if(P_LEFT[0] < (c->xmin<<shift)) P_LEFT[0] = (c->xmin<<shift);
1245 /* special case for first line */
1246 if (s->first_slice_line) {
1247 c->pred_x= P_LEFT[0];
1248 c->pred_y= P_LEFT[1];
1249 P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
1250 P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
1252 P_TOP[0] = s->p_mv_table[xy + s->mb_stride ][0];
1253 P_TOP[1] = s->p_mv_table[xy + s->mb_stride ][1];
1254 P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
1255 P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
1256 if(P_TOP[1] < (c->ymin<<shift)) P_TOP[1] = (c->ymin<<shift);
1257 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
1258 if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);
1260 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1261 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1263 c->pred_x = P_MEDIAN[0];
1264 c->pred_y = P_MEDIAN[1];
1267 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1269 s->p_mv_table[xy][0] = mx<<shift;
1270 s->p_mv_table[xy][1] = my<<shift;
1275 static int ff_estimate_motion_b(MpegEncContext * s,
1276 int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)
1278 MotionEstContext * const c= &s->me;
1281 const int shift= 1+s->quarter_sample;
1282 const int mot_stride = s->mb_stride;
1283 const int mot_xy = mb_y*mot_stride + mb_x;
1284 uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;
1287 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1288 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1289 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1290 c->current_mv_penalty= mv_penalty;
1292 get_limits(s, 16*mb_x, 16*mb_y);
1294 switch(s->me_method) {
1297 no_motion_search(s, &mx, &my);
1304 P_LEFT[0] = mv_table[mot_xy - 1][0];
1305 P_LEFT[1] = mv_table[mot_xy - 1][1];
1307 if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);
1309 /* special case for first line */
1310 if (!s->first_slice_line) {
1311 P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
1312 P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
1313 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
1314 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
1315 if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);
1316 if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);
1317 if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);
1319 P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1320 P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1322 c->pred_x = P_LEFT[0];
1323 c->pred_y = P_LEFT[1];
1325 if(mv_table == s->b_forw_mv_table){
1326 mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);
1328 mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
1331 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
1336 dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
1338 if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1339 dmin= ff_get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
1341 //printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);
1342 // s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
1343 mv_table[mot_xy][0]= mx;
1344 mv_table[mot_xy][1]= my;
1349 static inline int check_bidir_mv(MpegEncContext * s,
1350 int motion_fx, int motion_fy,
1351 int motion_bx, int motion_by,
1352 int pred_fx, int pred_fy,
1353 int pred_bx, int pred_by,
1357 //FIXME better f_code prediction (max mv & distance)
1359 MotionEstContext * const c= &s->me;
1360 uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
1361 uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame
1362 int stride= c->stride;
1363 uint8_t *dest_y = c->scratchpad;
1368 uint8_t **src_data= c->src[0];
1369 uint8_t **ref_data= c->ref[0];
1370 uint8_t **ref2_data= c->ref[2];
1372 if(s->quarter_sample){
1373 dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1374 src_x = motion_fx >> 2;
1375 src_y = motion_fy >> 2;
1377 ptr = ref_data[0] + (src_y * stride) + src_x;
1378 s->dsp.put_qpel_pixels_tab[0][dxy](dest_y , ptr , stride);
1380 dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1381 src_x = motion_bx >> 2;
1382 src_y = motion_by >> 2;
1384 ptr = ref2_data[0] + (src_y * stride) + src_x;
1385 s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y , ptr , stride);
1387 dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1388 src_x = motion_fx >> 1;
1389 src_y = motion_fy >> 1;
1391 ptr = ref_data[0] + (src_y * stride) + src_x;
1392 s->dsp.put_pixels_tab[size][dxy](dest_y , ptr , stride, h);
1394 dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1395 src_x = motion_bx >> 1;
1396 src_y = motion_by >> 1;
1398 ptr = ref2_data[0] + (src_y * stride) + src_x;
1399 s->dsp.avg_pixels_tab[size][dxy](dest_y , ptr , stride, h);
1402 fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
1403 +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
1404 + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic
1406 if(c->avctx->mb_cmp&FF_CMP_CHROMA){
1413 /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
1414 static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
1416 MotionEstContext * const c= &s->me;
1417 const int mot_stride = s->mb_stride;
1418 const int xy = mb_y *mot_stride + mb_x;
1420 int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];
1421 int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];
1422 int pred_bx= s->b_bidir_back_mv_table[xy-1][0];
1423 int pred_by= s->b_bidir_back_mv_table[xy-1][1];
1424 int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];
1425 int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];
1426 int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];
1427 int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];
1428 const int flags= c->sub_flags;
1429 const int qpel= flags&FLAG_QPEL;
1430 const int shift= 1+qpel;
1431 const int xmin= c->xmin<<shift;
1432 const int ymin= c->ymin<<shift;
1433 const int xmax= c->xmax<<shift;
1434 const int ymax= c->ymax<<shift;
1435 #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
1436 #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
1437 int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);
1438 uint8_t map[256] = { 0 };
1440 map[hashidx&255] = 1;
1442 fbmin= check_bidir_mv(s, motion_fx, motion_fy,
1443 motion_bx, motion_by,
1448 if(s->avctx->bidir_refine){
1450 static const uint8_t limittab[5]={0,8,32,64,80};
1451 const int limit= limittab[s->avctx->bidir_refine];
1452 static const int8_t vect[][4]={
1453 { 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},
1455 { 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},
1456 { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
1457 { 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},
1458 { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
1460 { 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},
1461 { 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},
1462 { 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},
1463 { 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},
1465 { 1, 1, 1, 1}, {-1,-1,-1,-1},
1466 { 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},
1467 { 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},
1469 static const uint8_t hash[]={
1470 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),
1472 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),
1473 HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),
1474 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),
1475 HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),
1477 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),
1478 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),
1479 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),
1480 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),
1482 HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),
1483 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),
1484 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),
1487 #define CHECK_BIDIR(fx,fy,bx,by)\
1488 if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
1489 &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1490 &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1492 map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
1493 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);\
1495 hashidx += HASH(fx,fy,bx,by);\
1504 #define CHECK_BIDIR2(a,b,c,d)\
1505 CHECK_BIDIR(a,b,c,d)\
1506 CHECK_BIDIR(-(a),-(b),-(c),-(d))
1513 CHECK_BIDIR2(0,0,0,1)
1514 CHECK_BIDIR2(0,0,1,0)
1515 CHECK_BIDIR2(0,1,0,0)
1516 CHECK_BIDIR2(1,0,0,0)
1518 for(i=8; i<limit; i++){
1519 int fx= motion_fx+vect[i][0];
1520 int fy= motion_fy+vect[i][1];
1521 int bx= motion_bx+vect[i][2];
1522 int by= motion_by+vect[i][3];
1524 int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);
1525 int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);
1527 map[(hashidx+hash[i])&255] = 1;
1529 if(!map[(hashidx+hash[i])&255]){
1531 map[(hashidx+hash[i])&255] = 1;
1532 score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
1543 int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);
1544 int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);
1545 borderdist= FFMIN(a,b);
1553 s->b_bidir_forw_mv_table[xy][0]= motion_fx;
1554 s->b_bidir_forw_mv_table[xy][1]= motion_fy;
1555 s->b_bidir_back_mv_table[xy][0]= motion_bx;
1556 s->b_bidir_back_mv_table[xy][1]= motion_by;
1561 static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
1563 MotionEstContext * const c= &s->me;
1565 const int mot_stride = s->mb_stride;
1566 const int mot_xy = mb_y*mot_stride + mb_x;
1567 const int shift= 1+s->quarter_sample;
1569 const int time_pp= s->pp_time;
1570 const int time_pb= s->pb_time;
1571 int mx, my, xmin, xmax, ymin, ymax;
1572 int16_t (*mv_table)[2]= s->b_direct_mv_table;
1574 c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;
1575 ymin= xmin=(-32)>>shift;
1576 ymax= xmax= 31>>shift;
1578 if (IS_8X8(s->next_picture.f.mb_type[mot_xy])) {
1579 s->mv_type= MV_TYPE_8X8;
1581 s->mv_type= MV_TYPE_16X16;
1585 int index= s->block_index[i];
1588 c->co_located_mv[i][0] = s->next_picture.f.motion_val[0][index][0];
1589 c->co_located_mv[i][1] = s->next_picture.f.motion_val[0][index][1];
1590 c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
1591 c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
1592 // c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
1593 // c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
1595 max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1596 min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1597 max+= 16*mb_x + 1; // +-1 is for the simpler rounding
1599 xmax= FFMIN(xmax, s->width - max);
1600 xmin= FFMAX(xmin, - 16 - min);
1602 max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1603 min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1604 max+= 16*mb_y + 1; // +-1 is for the simpler rounding
1606 ymax= FFMIN(ymax, s->height - max);
1607 ymin= FFMAX(ymin, - 16 - min);
1609 if(s->mv_type == MV_TYPE_16X16) break;
1612 assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1614 if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1615 s->b_direct_mv_table[mot_xy][0]= 0;
1616 s->b_direct_mv_table[mot_xy][1]= 0;
1618 return 256*256*256*64;
1625 c->flags |= FLAG_DIRECT;
1626 c->sub_flags |= FLAG_DIRECT;
1630 P_LEFT[0] = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
1631 P_LEFT[1] = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
1633 /* special case for first line */
1634 if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped
1635 P_TOP[0] = av_clip(mv_table[mot_xy - mot_stride ][0], xmin<<shift, xmax<<shift);
1636 P_TOP[1] = av_clip(mv_table[mot_xy - mot_stride ][1], ymin<<shift, ymax<<shift);
1637 P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1 ][0], xmin<<shift, xmax<<shift);
1638 P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1 ][1], ymin<<shift, ymax<<shift);
1640 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1641 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1644 dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
1645 if(c->sub_flags&FLAG_QPEL)
1646 dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1648 dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1650 if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1651 dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1653 get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
1655 mv_table[mot_xy][0]= mx;
1656 mv_table[mot_xy][1]= my;
1657 c->flags &= ~FLAG_DIRECT;
1658 c->sub_flags &= ~FLAG_DIRECT;
1663 void ff_estimate_b_frame_motion(MpegEncContext * s,
1666 MotionEstContext * const c= &s->me;
1667 const int penalty_factor= c->mb_penalty_factor;
1668 int fmin, bmin, dmin, fbmin, bimin, fimin;
1670 const int xy = mb_y*s->mb_stride + mb_x;
1671 init_ref(c, s->new_picture.f.data, s->last_picture.f.data,
1672 s->next_picture.f.data, 16 * mb_x, 16 * mb_y, 2);
1674 get_limits(s, 16*mb_x, 16*mb_y);
1678 if (s->codec_id == CODEC_ID_MPEG4 && s->next_picture.f.mbskip_table[xy]) {
1679 int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
1681 score= ((unsigned)(score*score + 128*256))>>16;
1682 c->mc_mb_var_sum_temp += score;
1683 s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1684 s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
1689 if(c->avctx->me_threshold){
1690 int vard= check_input_motion(s, mb_x, mb_y, 0);
1692 if((vard+128)>>8 < c->avctx->me_threshold){
1693 // pix = c->src[0][0];
1694 // sum = s->dsp.pix_sum(pix, s->linesize);
1695 // varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500;
1697 // pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
1698 s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
1699 /* pic->mb_mean [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1700 c->mb_var_sum_temp += (varc+128)>>8;*/
1701 c->mc_mb_var_sum_temp += (vard+128)>>8;
1702 /* if (vard <= 64<<8 || vard < varc) {
1703 c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
1705 c->scene_change_score+= s->qscale * s->avctx->scenechange_factor;
1709 if((vard+128)>>8 < c->avctx->mb_threshold){
1710 type= s->mb_type[mb_y*s->mb_stride + mb_x];
1711 if(type == CANDIDATE_MB_TYPE_DIRECT){
1712 direct_search(s, mb_x, mb_y);
1714 if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){
1716 ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code);
1718 if(type == CANDIDATE_MB_TYPE_BACKWARD || type == CANDIDATE_MB_TYPE_BIDIR){
1720 ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code);
1722 if(type == CANDIDATE_MB_TYPE_FORWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
1724 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1725 interlaced_search(s, 0,
1726 s->b_field_mv_table[0], s->b_field_select_table[0],
1727 s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 1);
1729 if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
1731 c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
1732 interlaced_search(s, 2,
1733 s->b_field_mv_table[1], s->b_field_select_table[1],
1734 s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 1);
1740 if (s->codec_id == CODEC_ID_MPEG4)
1741 dmin= direct_search(s, mb_x, mb_y);
1744 //FIXME penalty stuff for non mpeg4
1746 fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;
1749 bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor;
1750 //printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
1753 fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
1754 //printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);
1756 if(s->flags & CODEC_FLAG_INTERLACED_ME){
1757 //FIXME mb type penalty
1759 c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1760 fimin= interlaced_search(s, 0,
1761 s->b_field_mv_table[0], s->b_field_select_table[0],
1762 s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
1763 c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
1764 bimin= interlaced_search(s, 2,
1765 s->b_field_mv_table[1], s->b_field_select_table[1],
1766 s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);
1768 fimin= bimin= INT_MAX;
1772 type = CANDIDATE_MB_TYPE_FORWARD;
1776 type = CANDIDATE_MB_TYPE_DIRECT;
1780 type= CANDIDATE_MB_TYPE_BACKWARD;
1784 type= CANDIDATE_MB_TYPE_BIDIR;
1788 type= CANDIDATE_MB_TYPE_FORWARD_I;
1792 type= CANDIDATE_MB_TYPE_BACKWARD_I;
1795 score= ((unsigned)(score*score + 128*256))>>16;
1796 c->mc_mb_var_sum_temp += score;
1797 s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1800 if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1801 type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
1803 type |= CANDIDATE_MB_TYPE_FORWARD_I;
1805 type |= CANDIDATE_MB_TYPE_BACKWARD_I;
1806 if(fimin < INT_MAX && bimin < INT_MAX){
1807 type |= CANDIDATE_MB_TYPE_BIDIR_I;
1809 //FIXME something smarter
1810 if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB
1811 if(s->codec_id == CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT && s->flags&CODEC_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
1812 type |= CANDIDATE_MB_TYPE_DIRECT0;
1815 s->mb_type[mb_y*s->mb_stride + mb_x]= type;
1818 /* find best f_code for ME which do unlimited searches */
1819 int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)
1821 if(s->me_method>=ME_EPZS){
1823 int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);
1824 uint8_t * fcode_tab= s->fcode_tab;
1826 int best_score=-10000000;
1828 if(s->msmpeg4_version)
1829 range= FFMIN(range, 16);
1830 else if(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
1831 range= FFMIN(range, 256);
1833 for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);
1835 for(y=0; y<s->mb_height; y++){
1837 int xy= y*s->mb_stride;
1838 for(x=0; x<s->mb_width; x++){
1839 if(s->mb_type[xy] & type){
1840 int mx= mv_table[xy][0];
1841 int my= mv_table[xy][1];
1842 int fcode= FFMAX(fcode_tab[mx + MAX_MV],
1843 fcode_tab[my + MAX_MV]);
1846 if(mx >= range || mx < -range ||
1847 my >= range || my < -range)
1850 for(j=0; j<fcode && j<8; j++){
1851 if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
1860 if(score[i] > best_score){
1861 best_score= score[i];
1864 // printf("%d %d\n", i, score[i]);
1867 // printf("fcode: %d type: %d\n", i, s->pict_type);
1869 /* for(i=0; i<=MAX_FCODE; i++){
1870 printf("%d ", mv_num[i]);
1878 void ff_fix_long_p_mvs(MpegEncContext * s)
1880 MotionEstContext * const c= &s->me;
1881 const int f_code= s->f_code;
1883 assert(s->pict_type==AV_PICTURE_TYPE_P);
1885 range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1887 assert(range <= 16 || !s->msmpeg4_version);
1888 assert(range <=256 || !(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
1890 if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1892 //printf("%d no:%d %d//\n", clip, noclip, f_code);
1893 if(s->flags&CODEC_FLAG_4MV){
1894 const int wrap= s->b8_stride;
1896 /* clip / convert to intra 8x8 type MVs */
1897 for(y=0; y<s->mb_height; y++){
1899 int i= y*s->mb_stride;
1902 for(x=0; x<s->mb_width; x++){
1903 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){
1905 for(block=0; block<4; block++){
1906 int off= (block& 1) + (block>>1)*wrap;
1907 int mx = s->current_picture.f.motion_val[0][ xy + off ][0];
1908 int my = s->current_picture.f.motion_val[0][ xy + off ][1];
1910 if( mx >=range || mx <-range
1911 || my >=range || my <-range){
1912 s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;
1913 s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;
1914 s->current_picture.f.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;
1927 * @param truncate 1 for truncation, 0 for using intra
1929 void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
1930 int16_t (*mv_table)[2], int f_code, int type, int truncate)
1932 MotionEstContext * const c= &s->me;
1933 int y, h_range, v_range;
1935 // RAL: 8 in MPEG-1, 16 in MPEG-4
1936 int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1938 if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1941 v_range= field_select_table ? range>>1 : range;
1943 /* clip / convert to intra 16x16 type MVs */
1944 for(y=0; y<s->mb_height; y++){
1946 int xy= y*s->mb_stride;
1947 for(x=0; x<s->mb_width; x++){
1948 if (s->mb_type[xy] & type){ // RAL: "type" test added...
1949 if(field_select_table==NULL || field_select_table[xy] == field_select){
1950 if( mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
1951 || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
1954 if (mv_table[xy][0] > h_range-1) mv_table[xy][0]= h_range-1;
1955 else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
1956 if (mv_table[xy][1] > v_range-1) mv_table[xy][1]= v_range-1;
1957 else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
1959 s->mb_type[xy] &= ~type;
1960 s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;