43 #define P_TOPRIGHT P[3]
47 #define ME_MAP_SHIFT 3
48 #define ME_MAP_MV_BITS 11
51 int *mx_ptr,
int *my_ptr,
int dmin,
52 int src_index,
int ref_index,
58 if(
c->map_generation==0){
62 return c->map_generation;
79 #define FLAG_QPEL 1 //must be 1
86 ((y*
c->uvstride + x)>>1),
87 ((y*
c->uvstride + x)>>1),
108 const int size,
const int h,
int ref_index,
int src_index,
112 const int hx= subx + (x<<(1+qpel));
113 const int hy= suby + (y<<(1+qpel));
118 av_assert2(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;
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));
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));
152 c->qpel_put[1][fxy](
c->temp + 8 ,
ref[0] + (fx>>2) + (fy>>2)*
stride + 8 ,
stride);
156 c->qpel_avg[1][bxy](
c->temp + 8 ,
ref[8] + (bx>>2) + (by>>2)*
stride + 8 ,
stride);
180 const int size,
const int h,
int ref_index,
int src_index,
184 const int uvstride=
c->uvstride;
185 const int dxy= subx + (suby<<(1+qpel));
186 const int hx= subx + x*(1<<(1+qpel));
187 const int hy= suby + y*(1<<(1+qpel));
195 if (
h <<
size == 16) {
197 }
else if (
size == 0 &&
h == 8) {
207 uvdxy= (cx&1) + 2*(cy&1);
213 uvdxy= dxy | (x&1) | (2*(y&1));
219 uvdxy= (x&1) + 2*(y&1);
223 c->hpel_put[
size+1][uvdxy](uvtemp ,
ref[1] + (x>>1) + (y>>1)*uvstride, uvstride,
h>>1);
224 c->hpel_put[
size+1][uvdxy](uvtemp+8,
ref[2] + (x>>1) + (y>>1)*uvstride, uvstride,
h>>1);
225 d += chroma_cmp_func(
s, uvtemp ,
src[1], uvstride,
h>>1);
226 d += chroma_cmp_func(
s, uvtemp+8,
src[2], uvstride,
h>>1);
232 int ref_index,
int src_index,
234 return cmp_inline(
s,x,y,0,0,0,16,ref_index,src_index,
cmp_func, chroma_cmp_func, 0, 0);
238 const int size,
const int h,
int ref_index,
int src_index,
241 return cmp_direct_inline(
s,x,y,0,0,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func,
flags&
FLAG_QPEL);
243 return cmp_inline(
s,x,y,0,0,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 0,
flags&
FLAG_CHROMA);
248 const int size,
const int h,
int ref_index,
int src_index,
251 return cmp_direct_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func,
flags&
FLAG_QPEL);
253 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);
261 const int size,
const int h,
int ref_index,
int src_index,
265 &&
flags==0 &&
h==16 &&
size==0 && subx==0 && suby==0){
268 && subx==0 && suby==0){
271 return cmp_internal(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func,
flags);
276 const int size,
const int h,
int ref_index,
int src_index,
279 return cmp_direct_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 0);
281 return cmp_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 0,
flags&
FLAG_CHROMA);
286 const int size,
const int h,
int ref_index,
int src_index,
289 return cmp_direct_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 1);
291 return cmp_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 1,
flags&
FLAG_CHROMA);
309 int dia_size=
FFMAX(
FFABS(
s->avctx->dia_size)&255,
FFABS(
s->avctx->pre_dia_size)&255);
319 c->avctx->me_sub_cmp =
c->avctx->me_cmp;
321 if(cache_size < 2*dia_size && !c->
stride){
322 av_log(
s->avctx,
AV_LOG_INFO,
"ME_MAP size may be a little small for the selected diamond size\n");
325 ff_set_cmp(&
s->mecc,
s->mecc.me_pre_cmp,
c->avctx->me_pre_cmp);
327 ff_set_cmp(&
s->mecc,
s->mecc.me_sub_cmp,
c->avctx->me_sub_cmp);
337 c->qpel_avg =
s->qdsp.avg_qpel_pixels_tab;
339 c->qpel_put =
s->qdsp.put_no_rnd_qpel_pixels_tab;
341 c->qpel_put =
s->qdsp.put_qpel_pixels_tab;
352 c->hpel_avg =
s->hdsp.avg_pixels_tab;
354 c->hpel_put =
s->hdsp.put_no_rnd_pixels_tab;
356 c->hpel_put =
s->hdsp.put_pixels_tab;
359 c->stride =
s->linesize;
360 c->uvstride=
s->uvlinesize;
362 c->stride = 16*
s->mb_width + 32;
363 c->uvstride= 8*
s->mb_width + 16;
374 c->hpel_put[2][0]=
c->hpel_put[2][1]=
385 #define CHECK_SAD_HALF_MV(suffix, x, y) \
387 d = s->mecc.pix_abs[size][(x ? 1 : 0) + (y ? 2 : 0)](NULL, pix, ptr + ((x) >> 1), stride, h); \
388 d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
389 COPY3_IF_LT(dminh, d, dx, x, dy, y)\
393 int *mx_ptr,
int *my_ptr,
int dmin,
394 int src_index,
int ref_index,
398 const int penalty_factor=
c->sub_penalty_factor;
412 pix =
c->src[src_index][0];
416 ptr =
c->ref[ref_index][0] + (my *
stride) + mx;
420 if (mx > xmin && mx < xmax &&
421 my > ymin && my < ymax) {
499 const int xy=
s->mb_x +
s->mb_y*
s->mb_stride;
501 s->p_mv_table[xy][0] = mx;
502 s->p_mv_table[xy][1] = my;
506 int mot_xy=
s->block_index[0];
508 s->current_picture.motion_val[0][mot_xy ][0] = mx;
509 s->current_picture.motion_val[0][mot_xy ][1] = my;
510 s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
511 s->current_picture.motion_val[0][mot_xy + 1][1] = my;
513 mot_xy +=
s->b8_stride;
514 s->current_picture.motion_val[0][mot_xy ][0] = mx;
515 s->current_picture.motion_val[0][mot_xy ][1] = my;
516 s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
517 s->current_picture.motion_val[0][mot_xy + 1][1] = my;
527 int range=
c->avctx->me_range >> (1 + !!(
c->flags&
FLAG_QPEL));
533 if (
s->unrestricted_mv) {
536 c->xmax = - x +
s->width;
537 c->ymax = - y +
s->height;
540 c->xmin = (x > 15) ? - 15 : 0;
541 c->ymin = (y > 15) ? - 15 : 0;
542 c->xmax = (x <
s->mb_width * 16 - 16) ? 15 : 0;
543 c->ymax = (y <
s->mb_height * 16 - 16) ? 15 : 0;
547 c->xmax = - x +
s->mb_width *16 - 16;
548 c->ymax = - y +
s->mb_height*16 - 16;
550 if(!range || range > max_range)
553 c->xmin =
FFMAX(
c->xmin,-range);
554 c->xmax =
FFMIN(
c->xmax, range);
555 c->ymin =
FFMAX(
c->ymin,-range);
556 c->ymax =
FFMIN(
c->ymax, range);
563 c->ref[1][0] =
c->ref[0][0] + 8;
564 c->ref[2][0] =
c->ref[0][0] + 8*
stride;
565 c->ref[3][0] =
c->ref[2][0] + 8;
566 c->src[1][0] =
c->src[0][0] + 8;
567 c->src[2][0] =
c->src[0][0] + 8*
stride;
568 c->src[3][0] =
c->src[2][0] + 8;
578 int dmin_sum=0, mx4_sum=0, my4_sum=0,
i;
582 int safety_clipping=
s->unrestricted_mv && (
s->width&15) && (
s->height&15);
588 int pred_x4, pred_y4;
590 static const int off[4]= {2, 1, 1, -1};
591 const int mot_stride =
s->b8_stride;
592 const int mot_xy =
s->block_index[
block];
595 c->xmax = - 16*
s->mb_x +
s->width - 8*(
block &1);
596 c->ymax = - 16*
s->mb_y +
s->height - 8*(
block>>1);
599 P_LEFT[0] =
s->current_picture.motion_val[0][mot_xy - 1][0];
600 P_LEFT[1] =
s->current_picture.motion_val[0][mot_xy - 1][1];
605 if (
s->first_slice_line &&
block<2) {
609 P_TOP[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
610 P_TOP[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
611 P_TOPRIGHT[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride + off[
block]][0];
612 P_TOPRIGHT[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride + off[
block]][1];
628 if (
s->first_slice_line && block<2 && i>1 &&
i<9)
640 if (
s->mecc.me_sub_cmp[0] !=
s->mecc.mb_cmp[0]) {
644 if(
s->quarter_sample){
646 dxy = ((my4 & 3) << 2) | (mx4 & 3);
649 s->qdsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y,
ref,
stride);
651 s->qdsp.put_qpel_pixels_tab[1][dxy](dest_y,
ref,
stride);
654 dxy = ((my4 & 1) << 1) | (mx4 & 1);
657 s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y ,
ref ,
stride,
h);
659 s->hdsp.put_pixels_tab [1][dxy](dest_y ,
ref ,
stride,
h);
665 if(
s->quarter_sample){
673 s->current_picture.motion_val[0][
s->block_index[
block]][0] = mx4;
674 s->current_picture.motion_val[0][
s->block_index[
block]][1] = my4;
676 if(mx4 != mx || my4 != my) same=0;
682 if (
s->mecc.me_sub_cmp[0] !=
s->mecc.mb_cmp[0]) {
683 dmin_sum +=
s->mecc.mb_cmp[0](
s,
684 s->new_picture.f->data[0] +
685 s->mb_x * 16 +
s->mb_y * 16 *
stride,
696 dxy = ((my & 1) << 1) | (mx & 1);
698 offset= (
s->mb_x*8 + (mx>>1)) + (
s->mb_y*8 + (my>>1))*
s->uvlinesize;
701 s->hdsp.put_no_rnd_pixels_tab[1][dxy](
c->scratchpad ,
s->last_picture.f->data[1] +
offset,
s->uvlinesize, 8);
702 s->hdsp.put_no_rnd_pixels_tab[1][dxy](
c->scratchpad + 8,
s->last_picture.f->data[2] +
offset,
s->uvlinesize, 8);
704 s->hdsp.put_pixels_tab [1][dxy](
c->scratchpad ,
s->last_picture.f->data[1] +
offset,
s->uvlinesize, 8);
705 s->hdsp.put_pixels_tab [1][dxy](
c->scratchpad + 8,
s->last_picture.f->data[2] +
offset,
s->uvlinesize, 8);
708 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);
709 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);
715 switch(
c->avctx->mb_cmp&0xFF){
721 return dmin_sum+ 11*
c->mb_penalty_factor;
728 c->ref[1+ref_index][0] =
c->ref[0+ref_index][0] +
s->linesize;
729 c->src[1][0] =
c->src[0][0] +
s->linesize;
731 c->ref[1+ref_index][1] =
c->ref[0+ref_index][1] +
s->uvlinesize;
732 c->ref[1+ref_index][2] =
c->ref[0+ref_index][2] +
s->uvlinesize;
733 c->src[1][1] =
c->src[0][1] +
s->uvlinesize;
734 c->src[1][2] =
c->src[0][2] +
s->uvlinesize;
739 int16_t (*mv_tables[2][2])[2],
uint8_t *field_select_tables[2],
int mx,
int my,
int user_field_select)
748 const int stride= 2*
s->linesize;
750 const int mot_stride=
s->mb_stride;
751 const int xy=
s->mb_x +
s->mb_y*mot_stride;
761 int best_dmin= INT_MAX;
764 for(field_select=0; field_select<2; field_select++){
765 int dmin, mx_i, my_i;
766 int16_t (*mv_table)[2]= mv_tables[
block][field_select];
768 if(user_field_select){
769 av_assert1(field_select==0 || field_select==1);
771 if(field_select_tables[
block][xy] != field_select)
775 P_LEFT[0] = mv_table[xy - 1][0];
776 P_LEFT[1] = mv_table[xy - 1][1];
782 if(!
s->first_slice_line){
783 P_TOP[0] = mv_table[xy - mot_stride][0];
784 P_TOP[1] = mv_table[xy - mot_stride][1];
785 P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
786 P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
800 dmin=
c->sub_motion_search(
s, &mx_i, &my_i, dmin,
block, field_select+ref_index,
size,
h);
802 mv_table[xy][0]= mx_i;
803 mv_table[xy][1]= my_i;
805 if (
s->mecc.me_sub_cmp[0] !=
s->mecc.mb_cmp[0]) {
810 dxy = ((my_i & 1) << 1) | (mx_i & 1);
820 dmin+=
c->mb_penalty_factor;
822 dmin += field_select !=
block;
824 if(dmin < best_dmin){
826 best_field= field_select;
830 int16_t (*mv_table)[2]= mv_tables[
block][best_field];
832 if(mv_table[xy][0] != mx) same=0;
833 if(mv_table[xy][1]&1) same=0;
834 if(mv_table[xy][1]*2 != my) same=0;
835 if(best_field !=
block) same=0;
838 field_select_tables[
block][xy]= best_field;
839 dmin_sum += best_dmin;
850 switch(
c->avctx->mb_cmp&0xFF){
856 return dmin_sum+ 11*
c->mb_penalty_factor;
890 int sum, mx = 0, my = 0, dmin = 0;
894 const int shift= 1+
s->quarter_sample;
896 Picture *
const pic= &
s->current_picture;
898 init_ref(
c,
s->new_picture.f->data,
s->last_picture.f->data,
NULL, 16*mb_x, 16*mb_y, 0);
900 av_assert0(
s->quarter_sample==0 ||
s->quarter_sample==1);
907 c->current_mv_penalty=
c->mv_penalty[
s->f_code] +
MAX_DMV;
914 sum =
s->mpvencdsp.pix_sum(pix,
s->linesize);
915 varc =
s->mpvencdsp.pix_norm1(pix,
s->linesize) -
916 (((unsigned) sum * sum) >> 8) + 500;
918 pic->
mb_mean[
s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
919 pic->
mb_var [
s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
920 c->mb_var_sum_temp += (varc+128)>>8;
923 const int mot_stride =
s->b8_stride;
924 const int mot_xy =
s->block_index[0];
926 P_LEFT[0] =
s->current_picture.motion_val[0][mot_xy - 1][0];
927 P_LEFT[1] =
s->current_picture.motion_val[0][mot_xy - 1][1];
932 if (!
s->first_slice_line) {
933 P_TOP[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
934 P_TOP[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
935 P_TOPRIGHT[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
936 P_TOPRIGHT[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
962 ppix =
c->ref[0][0] + (my *
s->linesize) + mx;
964 vard =
s->mecc.sse[0](
NULL, pix, ppix,
s->linesize, 16);
966 pic->
mc_mb_var[
s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
967 c->mc_mb_var_sum_temp += (vard+128)>>8;
974 if (vard*2 + 200*256 > varc && !
s->intra_penalty)
976 if (varc*2 + 200*256 > vard ||
s->qscale > 24){
979 c->sub_motion_search(
s, &mx, &my, dmin, 0, 0, 0, 16);
988 && !
c->skip && varc>50<<8 && vard>10<<8){
1004 dmin=
c->sub_motion_search(
s, &mx, &my, dmin, 0, 0, 0, 16);
1005 if(
c->avctx->me_sub_cmp !=
c->avctx->mb_cmp && !
c->skip)
1009 && !
c->skip && varc>50<<8 && vard>10<<8){
1029 intra_score= varc - 500;
1031 unsigned mean = (sum+128)>>8;
1034 for(
i=0;
i<16;
i++){
1035 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+ 0]) = mean;
1036 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+ 4]) = mean;
1037 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+ 8]) = mean;
1038 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+12]) = mean;
1041 intra_score=
s->mecc.mb_cmp[0](
s,
c->scratchpad, pix,
s->linesize, 16);
1043 intra_score +=
c->mb_penalty_factor*16 +
s->intra_penalty;
1045 if(intra_score < dmin){
1049 s->current_picture.mb_type[mb_y*
s->mb_stride + mb_x] = 0;
1058 s->mb_type[mb_y*
s->mb_stride + mb_x]= mb_type;
1067 const int shift= 1+
s->quarter_sample;
1068 const int xy= mb_x + mb_y*
s->mb_stride;
1069 init_ref(
c,
s->new_picture.f->data,
s->last_picture.f->data,
NULL, 16*mb_x, 16*mb_y, 0);
1071 av_assert0(
s->quarter_sample==0 ||
s->quarter_sample==1);
1074 c->current_mv_penalty=
c->mv_penalty[
s->f_code] +
MAX_DMV;
1079 P_LEFT[0] =
s->p_mv_table[xy + 1][0];
1080 P_LEFT[1] =
s->p_mv_table[xy + 1][1];
1085 if (
s->first_slice_line) {
1091 P_TOP[0] =
s->p_mv_table[xy +
s->mb_stride ][0];
1092 P_TOP[1] =
s->p_mv_table[xy +
s->mb_stride ][1];
1093 P_TOPRIGHT[0] =
s->p_mv_table[xy +
s->mb_stride - 1][0];
1094 P_TOPRIGHT[1] =
s->p_mv_table[xy +
s->mb_stride - 1][1];
1108 s->p_mv_table[xy][0] = mx<<
shift;
1109 s->p_mv_table[xy][1] = my<<
shift;
1115 int16_t (*mv_table)[2],
int ref_index,
int f_code)
1118 int mx = 0, my = 0, dmin = 0;
1120 const int shift= 1+
s->quarter_sample;
1121 const int mot_stride =
s->mb_stride;
1122 const int mot_xy = mb_y*mot_stride + mb_x;
1134 P_LEFT[0] = mv_table[mot_xy - 1][0];
1135 P_LEFT[1] = mv_table[mot_xy - 1][1];
1140 if (!
s->first_slice_line) {
1141 P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
1142 P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
1143 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
1144 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
1155 if(mv_table ==
s->b_forw_mv_table){
1164 dmin=
c->sub_motion_search(
s, &mx, &my, dmin, 0, ref_index, 0, 16);
1166 if(
c->avctx->me_sub_cmp !=
c->avctx->mb_cmp && !
c->skip)
1170 mv_table[mot_xy][0]= mx;
1171 mv_table[mot_xy][1]= my;
1177 int motion_fx,
int motion_fy,
1178 int motion_bx,
int motion_by,
1179 int pred_fx,
int pred_fy,
1180 int pred_bx,
int pred_by,
1199 if(
s->quarter_sample){
1200 dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1201 src_x = motion_fx >> 2;
1202 src_y = motion_fy >> 2;
1204 ptr = ref_data[0] + (src_y *
stride) + src_x;
1205 s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr,
stride);
1207 dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1208 src_x = motion_bx >> 2;
1209 src_y = motion_by >> 2;
1211 ptr = ref2_data[0] + (src_y *
stride) + src_x;
1212 s->qdsp.avg_qpel_pixels_tab[
size][dxy](dest_y, ptr,
stride);
1214 dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1215 src_x = motion_fx >> 1;
1216 src_y = motion_fy >> 1;
1218 ptr = ref_data[0] + (src_y *
stride) + src_x;
1219 s->hdsp.put_pixels_tab[
size][dxy](dest_y , ptr ,
stride,
h);
1221 dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1222 src_x = motion_bx >> 1;
1223 src_y = motion_by >> 1;
1225 ptr = ref2_data[0] + (src_y *
stride) + src_x;
1226 s->hdsp.avg_pixels_tab[
size][dxy](dest_y , ptr ,
stride,
h);
1229 fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*
c->mb_penalty_factor
1230 +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*
c->mb_penalty_factor
1244 const int mot_stride =
s->mb_stride;
1245 const int xy = mb_y *mot_stride + mb_x;
1247 int pred_fx=
s->b_bidir_forw_mv_table[xy-1][0];
1248 int pred_fy=
s->b_bidir_forw_mv_table[xy-1][1];
1249 int pred_bx=
s->b_bidir_back_mv_table[xy-1][0];
1250 int pred_by=
s->b_bidir_back_mv_table[xy-1][1];
1251 int motion_fx=
s->b_bidir_forw_mv_table[xy][0]=
s->b_forw_mv_table[xy][0];
1252 int motion_fy=
s->b_bidir_forw_mv_table[xy][1]=
s->b_forw_mv_table[xy][1];
1253 int motion_bx=
s->b_bidir_back_mv_table[xy][0]=
s->b_back_mv_table[xy][0];
1254 int motion_by=
s->b_bidir_back_mv_table[xy][1]=
s->b_back_mv_table[xy][1];
1255 const int flags=
c->sub_flags;
1257 const int shift= 1+qpel;
1258 const int xmin=
c->xmin<<
shift;
1259 const int ymin=
c->ymin<<
shift;
1260 const int xmax=
c->xmax<<
shift;
1261 const int ymax=
c->ymax<<
shift;
1262 #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
1263 #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
1264 int hashidx=
HASH(motion_fx,motion_fy, motion_bx, motion_by);
1267 map[hashidx&255] = 1;
1270 motion_bx, motion_by,
1275 if(
s->avctx->bidir_refine){
1277 static const uint8_t limittab[5]={0,8,32,64,80};
1278 const int limit= limittab[
s->avctx->bidir_refine];
1279 static const int8_t vect[][4]={
1280 { 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},
1282 { 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},
1283 { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
1284 { 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},
1285 { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
1287 { 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},
1288 { 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},
1289 { 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},
1290 { 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},
1292 { 1, 1, 1, 1}, {-1,-1,-1,-1},
1293 { 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},
1294 { 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},
1297 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),
1299 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),
1300 HASH8( 0, 1, 0, 1),
HASH8( 0,-1, 0,-1),
HASH8( 1, 0, 1, 0),
HASH8(-1, 0,-1, 0),
1301 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),
1302 HASH8( 0,-1, 0, 1),
HASH8( 0, 1, 0,-1),
HASH8(-1, 0, 1, 0),
HASH8( 1, 0,-1, 0),
1304 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),
1305 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),
1306 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),
1307 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),
1310 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),
1311 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),
1314 #define CHECK_BIDIR(fx,fy,bx,by)\
1315 if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
1316 &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1317 &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1319 map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
1320 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);\
1322 hashidx += HASH(fx,fy,bx,by);\
1331 #define CHECK_BIDIR2(a,b,c,d)\
1332 CHECK_BIDIR(a,b,c,d)\
1333 CHECK_BIDIR(-(a),-(b),-(c),-(d))
1345 for(
i=8;
i<limit;
i++){
1346 int fx= motion_fx+vect[
i][0];
1347 int fy= motion_fy+vect[
i][1];
1348 int bx= motion_bx+vect[
i][2];
1349 int by= motion_by+vect[
i][3];
1351 int a= (xmax -
FFMAX(fx,bx))|(
FFMIN(fx,bx) - xmin);
1352 int b= (ymax -
FFMAX(fy,by))|(
FFMIN(fy,by) - ymin);
1359 score=
check_bidir_mv(
s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
1380 s->b_bidir_forw_mv_table[xy][0]= motion_fx;
1381 s->b_bidir_forw_mv_table[xy][1]= motion_fy;
1382 s->b_bidir_back_mv_table[xy][0]= motion_bx;
1383 s->b_bidir_back_mv_table[xy][1]= motion_by;
1392 const int mot_stride =
s->mb_stride;
1393 const int mot_xy = mb_y*mot_stride + mb_x;
1394 const int shift= 1+
s->quarter_sample;
1396 const int time_pp=
s->pp_time;
1397 const int time_pb=
s->pb_time;
1398 int mx, my, xmin, xmax, ymin, ymax;
1399 int16_t (*mv_table)[2]=
s->b_direct_mv_table;
1401 c->current_mv_penalty=
c->mv_penalty[1] +
MAX_DMV;
1402 ymin= xmin=(-32)>>
shift;
1403 ymax= xmax= 31>>
shift;
1405 if (
IS_8X8(
s->next_picture.mb_type[mot_xy])) {
1415 c->co_located_mv[
i][0] =
s->next_picture.motion_val[0][
index][0];
1416 c->co_located_mv[
i][1] =
s->next_picture.motion_val[0][
index][1];
1417 c->direct_basis_mv[
i][0]=
c->co_located_mv[
i][0]*time_pb/time_pp + ((
i& 1)<<(
shift+3));
1418 c->direct_basis_mv[
i][1]=
c->co_located_mv[
i][1]*time_pb/time_pp + ((
i>>1)<<(
shift+3));
1422 max=
FFMAX(
c->direct_basis_mv[
i][0],
c->direct_basis_mv[
i][0] -
c->co_located_mv[
i][0])>>
shift;
1423 min=
FFMIN(
c->direct_basis_mv[
i][0],
c->direct_basis_mv[
i][0] -
c->co_located_mv[
i][0])>>
shift;
1429 max=
FFMAX(
c->direct_basis_mv[
i][1],
c->direct_basis_mv[
i][1] -
c->co_located_mv[
i][1])>>
shift;
1430 min=
FFMIN(
c->direct_basis_mv[
i][1],
c->direct_basis_mv[
i][1] -
c->co_located_mv[
i][1])>>
shift;
1439 av_assert2(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1441 if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1442 s->b_direct_mv_table[mot_xy][0]= 0;
1443 s->b_direct_mv_table[mot_xy][1]= 0;
1445 return 256*256*256*64;
1461 if (!
s->first_slice_line) {
1462 P_TOP[0] = av_clip(mv_table[mot_xy - mot_stride ][0], xmin<<
shift, xmax<<
shift);
1463 P_TOP[1] = av_clip(mv_table[mot_xy - mot_stride ][1], ymin<<
shift, ymax<<
shift);
1477 if(
c->avctx->me_sub_cmp !=
c->avctx->mb_cmp && !
c->skip)
1482 mv_table[mot_xy][0]= mx;
1483 mv_table[mot_xy][1]= my;
1494 const int penalty_factor=
c->mb_penalty_factor;
1495 int fmin, bmin, dmin, fbmin, bimin, fimin;
1497 const int xy = mb_y*
s->mb_stride + mb_x;
1498 init_ref(
c,
s->new_picture.f->data,
s->last_picture.f->data,
1499 s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);
1508 score= ((unsigned)(score*score + 128*256))>>16;
1509 c->mc_mb_var_sum_temp += score;
1510 s->current_picture.mc_mb_var[mb_y*
s->mb_stride + mb_x] = score;
1528 ff_dlog(
s,
" %d %d ",
s->b_forw_mv_table[xy][0],
s->b_forw_mv_table[xy][1]);
1537 c->current_mv_penalty=
c->mv_penalty[
s->f_code] +
MAX_DMV;
1539 s->b_field_mv_table[0],
s->b_field_select_table[0],
1540 s->b_forw_mv_table[xy][0],
s->b_forw_mv_table[xy][1], 0);
1541 c->current_mv_penalty=
c->mv_penalty[
s->b_code] +
MAX_DMV;
1543 s->b_field_mv_table[1],
s->b_field_select_table[1],
1544 s->b_back_mv_table[xy][0],
s->b_back_mv_table[xy][1], 0);
1546 fimin= bimin= INT_MAX;
1573 score= ((unsigned)(score*score + 128*256))>>16;
1574 c->mc_mb_var_sum_temp += score;
1575 s->current_picture.mc_mb_var[mb_y*
s->mb_stride + mb_x] = score;
1584 if(fimin < INT_MAX && bimin < INT_MAX){
1594 s->mb_type[mb_y*
s->mb_stride + mb_x]=
type;
1602 int i, y, range=
s->avctx->me_range ?
s->avctx->me_range : (INT_MAX/2);
1605 int best_score=-10000000;
1607 if(
s->msmpeg4_version)
1608 range=
FFMIN(range, 16);
1610 range=
FFMIN(range, 256);
1612 for(
i=0;
i<8;
i++) score[
i]=
s->mb_num*(8-
i);
1614 for(y=0; y<
s->mb_height; y++){
1616 int xy= y*
s->mb_stride;
1617 for(x=0; x<
s->mb_width; x++, xy++){
1618 if(
s->mb_type[xy] &
type){
1619 int mx= mv_table[xy][0];
1620 int my= mv_table[xy][1];
1625 if (mx >= range || mx < -range ||
1626 my >= range || my < -range)
1629 for(j=0; j<fcode && j<8; j++){
1630 if(
s->pict_type==
AV_PICTURE_TYPE_B ||
s->current_picture.mc_mb_var[xy] <
s->current_picture.mb_var[xy])
1638 if(score[
i] > best_score){
1639 best_score= score[
i];
1653 const int f_code=
s->f_code;
1657 range = (((
s->out_format ==
FMT_MPEG1 ||
s->msmpeg4_version) ? 8 : 16) << f_code);
1659 av_assert0(range <= 16 || !s->msmpeg4_version);
1662 if(
c->avctx->me_range && range >
c->avctx->me_range) range=
c->avctx->me_range;
1665 const int wrap=
s->b8_stride;
1668 for(y=0; y<
s->mb_height; y++){
1670 int i= y*
s->mb_stride;
1673 for(x=0; x<
s->mb_width; x++){
1678 int mx =
s->current_picture.motion_val[0][ xy + off ][0];
1679 int my =
s->current_picture.motion_val[0][ xy + off ][1];
1681 if( mx >=range || mx <-range
1682 || my >=range || my <-range){
1685 s->current_picture.mb_type[
i] =
type;
1700 int16_t (*mv_table)[2],
int f_code,
int type,
int truncate)
1703 int y, h_range, v_range;
1706 int range = (((
s->out_format ==
FMT_MPEG1 ||
s->msmpeg4_version) ? 8 : 16) << f_code);
1708 if(
c->avctx->me_range && range >
c->avctx->me_range) range=
c->avctx->me_range;
1711 v_range= field_select_table ? range>>1 : range;
1714 for(y=0; y<
s->mb_height; y++){
1716 int xy= y*
s->mb_stride;
1717 for(x=0; x<
s->mb_width; x++){
1718 if (
s->mb_type[xy] &
type){
1719 if (!field_select_table || field_select_table[xy] == field_select) {
1720 if( mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
1721 || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
1724 if (mv_table[xy][0] > h_range-1) mv_table[xy][0]= h_range-1;
1725 else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
1726 if (mv_table[xy][1] > v_range-1) mv_table[xy][1]= v_range-1;
1727 else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
1729 s->mb_type[xy] &= ~
type;