42 #define P_TOPRIGHT P[3]
46 #define ME_MAP_SHIFT 3
47 #define ME_MAP_MV_BITS 11
50 int *mx_ptr,
int *my_ptr,
int dmin,
51 int src_index,
int ref_index,
57 if(
c->map_generation==0){
61 return c->map_generation;
78 #define FLAG_QPEL 1 //must be 1
83 uint8_t *
const ref[3], uint8_t *
const ref2[3],
84 int x,
int y,
int ref_index)
88 ((y*
c->uvstride + x)>>1),
89 ((y*
c->uvstride + x)>>1),
110 const int size,
const int h,
int ref_index,
int src_index,
114 const int hx = subx + x * (1 << (1 + qpel));
115 const int hy = suby + y * (1 << (1 + qpel));
116 const uint8_t *
const *
const ref =
c->ref[ref_index];
117 const uint8_t *
const *
const src =
c->src[src_index];
120 av_assert2(x >=
c->xmin && hx <= c->xmax<<(qpel+1) && y >=
c->ymin && hy <= c->ymax<<(qpel+1));
121 if(x >=
c->xmin && hx <= c->xmax<<(qpel+1) && y >=
c->ymin && hy <= c->ymax<<(qpel+1)){
122 const int time_pp=
s->pp_time;
123 const int time_pb=
s->pb_time;
124 const int mask= 2*qpel+1;
128 int fx =
c->direct_basis_mv[
i][0] + hx;
129 int fy =
c->direct_basis_mv[
i][1] + hy;
130 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));
131 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));
132 int fxy= (fx&
mask) + ((fy&
mask)<<(qpel+1));
133 int bxy= (bx&
mask) + ((by&
mask)<<(qpel+1));
135 uint8_t *dst=
c->temp + 8*(
i&1) + 8*
stride*(
i>>1);
145 int fx =
c->direct_basis_mv[0][0] + hx;
146 int fy =
c->direct_basis_mv[0][1] + hy;
147 int bx = hx ? fx -
c->co_located_mv[0][0] : (
c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
148 int by = hy ? fy -
c->co_located_mv[0][1] : (
c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
149 int fxy= (fx&
mask) + ((fy&
mask)<<(qpel+1));
150 int bxy= (bx&
mask) + ((by&
mask)<<(qpel+1));
154 c->qpel_put[1][fxy](
c->temp + 8 ,
ref[0] + (fx>>2) + (fy>>2)*
stride + 8 ,
stride);
158 c->qpel_avg[1][bxy](
c->temp + 8 ,
ref[8] + (bx>>2) + (by>>2)*
stride + 8 ,
stride);
182 const int size,
const int h,
int ref_index,
int src_index,
186 const int uvstride=
c->uvstride;
187 const int dxy= subx + (suby<<(1+qpel));
188 const int hx= subx + x*(1<<(1+qpel));
189 const int hy= suby + y*(1<<(1+qpel));
190 const uint8_t *
const *
const ref =
c->ref[ref_index];
191 const uint8_t *
const *
const src =
c->src[src_index];
197 if (
h <<
size == 16) {
199 }
else if (
size == 0 &&
h == 8) {
209 uvdxy= (cx&1) + 2*(cy&1);
215 uvdxy= dxy | (x&1) | (2*(y&1));
221 uvdxy= (x&1) + 2*(y&1);
224 uint8_t *
const uvtemp=
c->temp + 16*
stride;
225 c->hpel_put[
size+1][uvdxy](uvtemp ,
ref[1] + (x>>1) + (y>>1)*uvstride, uvstride,
h>>1);
226 c->hpel_put[
size+1][uvdxy](uvtemp+8,
ref[2] + (x>>1) + (y>>1)*uvstride, uvstride,
h>>1);
227 d += chroma_cmp_func(
s, uvtemp ,
src[1], uvstride,
h>>1);
228 d += chroma_cmp_func(
s, uvtemp+8,
src[2], uvstride,
h>>1);
234 int ref_index,
int src_index,
236 return cmp_inline(
s,x,y,0,0,0,16,ref_index,src_index,
cmp_func, chroma_cmp_func, 0, 0);
240 const int size,
const int h,
int ref_index,
int src_index,
243 return cmp_direct_inline(
s,x,y,0,0,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func,
flags&
FLAG_QPEL);
245 return cmp_inline(
s,x,y,0,0,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 0,
flags&
FLAG_CHROMA);
250 const int size,
const int h,
int ref_index,
int src_index,
253 return cmp_direct_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func,
flags&
FLAG_QPEL);
255 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);
263 const int size,
const int h,
int ref_index,
int src_index,
267 &&
flags==0 &&
h==16 &&
size==0 && subx==0 && suby==0){
270 && subx==0 && suby==0){
273 return cmp_internal(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func,
flags);
278 const int size,
const int h,
int ref_index,
int src_index,
281 return cmp_direct_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 0);
283 return cmp_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 0,
flags&
FLAG_CHROMA);
288 const int size,
const int h,
int ref_index,
int src_index,
291 return cmp_direct_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 1);
293 return cmp_inline(
s,x,y,subx,suby,
size,
h,ref_index,src_index,
cmp_func, chroma_cmp_func, 1,
flags&
FLAG_CHROMA);
311 int dia_size=
FFMAX(
FFABS(
s->avctx->dia_size)&255,
FFABS(
s->avctx->pre_dia_size)&255);
321 c->avctx->me_sub_cmp =
c->avctx->me_cmp;
323 if(cache_size < 2*dia_size && !c->
stride){
324 av_log(
s->avctx,
AV_LOG_INFO,
"ME_MAP size may be a little small for the selected diamond size\n");
327 ff_set_cmp(&
s->mecc,
s->mecc.me_pre_cmp,
c->avctx->me_pre_cmp);
329 ff_set_cmp(&
s->mecc,
s->mecc.me_sub_cmp,
c->avctx->me_sub_cmp);
339 c->qpel_avg =
s->qdsp.avg_qpel_pixels_tab;
341 c->qpel_put =
s->qdsp.put_no_rnd_qpel_pixels_tab;
343 c->qpel_put =
s->qdsp.put_qpel_pixels_tab;
354 c->hpel_avg =
s->hdsp.avg_pixels_tab;
356 c->hpel_put =
s->hdsp.put_no_rnd_pixels_tab;
358 c->hpel_put =
s->hdsp.put_pixels_tab;
361 c->stride =
s->linesize;
362 c->uvstride=
s->uvlinesize;
364 c->stride = 16*
s->mb_width + 32;
365 c->uvstride= 8*
s->mb_width + 16;
376 c->hpel_put[2][0]=
c->hpel_put[2][1]=
387 #define CHECK_SAD_HALF_MV(suffix, x, y) \
389 d = s->mecc.pix_abs[size][(x ? 1 : 0) + (y ? 2 : 0)](NULL, pix, ptr + ((x) >> 1), stride, h); \
390 d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
391 COPY3_IF_LT(dminh, d, dx, x, dy, y)\
395 int *mx_ptr,
int *my_ptr,
int dmin,
396 int src_index,
int ref_index,
400 const int penalty_factor=
c->sub_penalty_factor;
402 const uint8_t *pix, *ptr;
414 pix =
c->src[src_index][0];
418 ptr =
c->ref[ref_index][0] + (my *
stride) + mx;
422 if (mx > xmin && mx < xmax &&
423 my > ymin && my < ymax) {
501 const int xy=
s->mb_x +
s->mb_y*
s->mb_stride;
503 s->p_mv_table[xy][0] = mx;
504 s->p_mv_table[xy][1] = my;
508 int mot_xy=
s->block_index[0];
510 s->current_picture.motion_val[0][mot_xy ][0] = mx;
511 s->current_picture.motion_val[0][mot_xy ][1] = my;
512 s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
513 s->current_picture.motion_val[0][mot_xy + 1][1] = my;
515 mot_xy +=
s->b8_stride;
516 s->current_picture.motion_val[0][mot_xy ][0] = mx;
517 s->current_picture.motion_val[0][mot_xy ][1] = my;
518 s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
519 s->current_picture.motion_val[0][mot_xy + 1][1] = my;
529 int range=
c->avctx->me_range >> (1 + !!(
c->flags&
FLAG_QPEL));
535 if (
s->unrestricted_mv) {
538 c->xmax = - x +
s->width;
539 c->ymax = - y +
s->height;
542 c->xmin = (x > 15) ? - 15 : 0;
543 c->ymin = (y > 15) ? - 15 : 0;
544 c->xmax = (x <
s->mb_width * 16 - 16) ? 15 : 0;
545 c->ymax = (y <
s->mb_height * 16 - 16) ? 15 : 0;
549 c->xmax = - x +
s->mb_width *16 - 16;
550 c->ymax = - y +
s->mb_height*16 - 16;
552 if(!range || range > max_range)
555 c->xmin =
FFMAX(
c->xmin,-range);
556 c->xmax =
FFMIN(
c->xmax, range);
557 c->ymin =
FFMAX(
c->ymin,-range);
558 c->ymax =
FFMIN(
c->ymax, range);
565 c->ref[1][0] =
c->ref[0][0] + 8;
566 c->ref[2][0] =
c->ref[0][0] + 8*
stride;
567 c->ref[3][0] =
c->ref[2][0] + 8;
568 c->src[1][0] =
c->src[0][0] + 8;
569 c->src[2][0] =
c->src[0][0] + 8*
stride;
570 c->src[3][0] =
c->src[2][0] + 8;
580 int dmin_sum=0, mx4_sum=0, my4_sum=0,
i;
584 int safety_clipping=
s->unrestricted_mv && (
s->width&15) && (
s->height&15);
590 int pred_x4, pred_y4;
592 static const int off[4]= {2, 1, 1, -1};
593 const int mot_stride =
s->b8_stride;
594 const int mot_xy =
s->block_index[
block];
597 c->xmax = - 16*
s->mb_x +
s->width - 8*(
block &1);
598 c->ymax = - 16*
s->mb_y +
s->height - 8*(
block>>1);
601 P_LEFT[0] =
s->current_picture.motion_val[0][mot_xy - 1][0];
602 P_LEFT[1] =
s->current_picture.motion_val[0][mot_xy - 1][1];
607 if (
s->first_slice_line &&
block<2) {
611 P_TOP[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
612 P_TOP[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
613 P_TOPRIGHT[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride + off[
block]][0];
614 P_TOPRIGHT[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride + off[
block]][1];
630 if (
s->first_slice_line && block<2 && i>1 &&
i<9)
642 if (
s->mecc.me_sub_cmp[0] !=
s->mecc.mb_cmp[0]) {
645 uint8_t *dest_y =
c->scratchpad +
offset;
646 if(
s->quarter_sample){
648 dxy = ((my4 & 3) << 2) | (mx4 & 3);
651 s->qdsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y,
ref,
stride);
653 s->qdsp.put_qpel_pixels_tab[1][dxy](dest_y,
ref,
stride);
656 dxy = ((my4 & 1) << 1) | (mx4 & 1);
659 s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y ,
ref ,
stride,
h);
661 s->hdsp.put_pixels_tab [1][dxy](dest_y ,
ref ,
stride,
h);
667 if(
s->quarter_sample){
675 s->current_picture.motion_val[0][
s->block_index[
block]][0] = mx4;
676 s->current_picture.motion_val[0][
s->block_index[
block]][1] = my4;
678 if(mx4 != mx || my4 != my) same=0;
684 if (
s->mecc.me_sub_cmp[0] !=
s->mecc.mb_cmp[0]) {
685 dmin_sum +=
s->mecc.mb_cmp[0](
s,
686 s->new_picture->data[0] +
687 s->mb_x * 16 +
s->mb_y * 16 *
stride,
698 dxy = ((my & 1) << 1) | (mx & 1);
700 offset= (
s->mb_x*8 + (mx>>1)) + (
s->mb_y*8 + (my>>1))*
s->uvlinesize;
703 s->hdsp.put_no_rnd_pixels_tab[1][dxy](
c->scratchpad ,
s->last_picture.f->data[1] +
offset,
s->uvlinesize, 8);
704 s->hdsp.put_no_rnd_pixels_tab[1][dxy](
c->scratchpad + 8,
s->last_picture.f->data[2] +
offset,
s->uvlinesize, 8);
706 s->hdsp.put_pixels_tab [1][dxy](
c->scratchpad ,
s->last_picture.f->data[1] +
offset,
s->uvlinesize, 8);
707 s->hdsp.put_pixels_tab [1][dxy](
c->scratchpad + 8,
s->last_picture.f->data[2] +
offset,
s->uvlinesize, 8);
710 dmin_sum +=
s->mecc.mb_cmp[1](
s,
s->new_picture->data[1] +
s->mb_x * 8 +
s->mb_y * 8 *
s->uvlinesize,
c->scratchpad,
s->uvlinesize, 8);
711 dmin_sum +=
s->mecc.mb_cmp[1](
s,
s->new_picture->data[2] +
s->mb_x * 8 +
s->mb_y * 8 *
s->uvlinesize,
c->scratchpad + 8,
s->uvlinesize, 8);
717 switch(
c->avctx->mb_cmp&0xFF){
723 return dmin_sum+ 11*
c->mb_penalty_factor;
730 c->ref[1+ref_index][0] =
c->ref[0+ref_index][0] +
s->linesize;
731 c->src[1][0] =
c->src[0][0] +
s->linesize;
733 c->ref[1+ref_index][1] =
c->ref[0+ref_index][1] +
s->uvlinesize;
734 c->ref[1+ref_index][2] =
c->ref[0+ref_index][2] +
s->uvlinesize;
735 c->src[1][1] =
c->src[0][1] +
s->uvlinesize;
736 c->src[1][2] =
c->src[0][2] +
s->uvlinesize;
741 int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2],
int mx,
int my,
int user_field_select)
748 const uint8_t *
const mv_penalty =
c->current_mv_penalty;
750 const int stride= 2*
s->linesize;
752 const int mot_stride=
s->mb_stride;
753 const int xy=
s->mb_x +
s->mb_y*mot_stride;
763 int best_dmin= INT_MAX;
766 for(field_select=0; field_select<2; field_select++){
767 int dmin, mx_i, my_i;
768 int16_t (*mv_table)[2]= mv_tables[
block][field_select];
770 if(user_field_select){
771 av_assert1(field_select==0 || field_select==1);
773 if(field_select_tables[
block][xy] != field_select)
777 P_LEFT[0] = mv_table[xy - 1][0];
778 P_LEFT[1] = mv_table[xy - 1][1];
784 if(!
s->first_slice_line){
785 P_TOP[0] = mv_table[xy - mot_stride][0];
786 P_TOP[1] = mv_table[xy - mot_stride][1];
787 P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
788 P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
802 dmin=
c->sub_motion_search(
s, &mx_i, &my_i, dmin,
block, field_select+ref_index,
size,
h);
804 mv_table[xy][0]= mx_i;
805 mv_table[xy][1]= my_i;
807 if (
s->mecc.me_sub_cmp[0] !=
s->mecc.mb_cmp[0]) {
811 const uint8_t *
ref =
c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*
stride;
812 dxy = ((my_i & 1) << 1) | (mx_i & 1);
822 dmin+=
c->mb_penalty_factor;
824 dmin += field_select !=
block;
826 if(dmin < best_dmin){
828 best_field= field_select;
832 int16_t (*mv_table)[2]= mv_tables[
block][best_field];
834 if(mv_table[xy][0] != mx) same=0;
835 if(mv_table[xy][1]&1) same=0;
836 if(mv_table[xy][1]*2 != my) same=0;
837 if(best_field !=
block) same=0;
840 field_select_tables[
block][xy]= best_field;
841 dmin_sum += best_dmin;
852 switch(
c->avctx->mb_cmp&0xFF){
858 return dmin_sum+ 11*
c->mb_penalty_factor;
891 const uint8_t *pix, *ppix;
892 int sum, mx = 0, my = 0, dmin = 0;
896 const int shift= 1+
s->quarter_sample;
899 init_ref(
c,
s->new_picture->data,
s->last_picture.f->data,
NULL, 16*mb_x, 16*mb_y, 0);
901 av_assert0(
s->quarter_sample==0 ||
s->quarter_sample==1);
908 c->current_mv_penalty=
c->mv_penalty[
s->f_code] +
MAX_DMV;
915 sum =
s->mpvencdsp.pix_sum(pix,
s->linesize);
916 varc =
s->mpvencdsp.pix_norm1(pix,
s->linesize) -
917 (((unsigned) sum * sum) >> 8) + 500;
919 s->mb_mean[
s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
920 s->mb_var [
s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
921 c->mb_var_sum_temp += (varc+128)>>8;
924 const int mot_stride =
s->b8_stride;
925 const int mot_xy =
s->block_index[0];
927 P_LEFT[0] =
s->current_picture.motion_val[0][mot_xy - 1][0];
928 P_LEFT[1] =
s->current_picture.motion_val[0][mot_xy - 1][1];
933 if (!
s->first_slice_line) {
934 P_TOP[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][0];
935 P_TOP[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride ][1];
936 P_TOPRIGHT[0] =
s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
937 P_TOPRIGHT[1] =
s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
963 ppix =
c->ref[0][0] + (my *
s->linesize) + mx;
965 vard =
s->mecc.sse[0](
NULL, pix, ppix,
s->linesize, 16);
967 s->mc_mb_var[
s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
968 c->mc_mb_var_sum_temp += (vard+128)>>8;
975 if (vard*2 + 200*256 > varc && !
s->intra_penalty)
977 if (varc*2 + 200*256 > vard ||
s->qscale > 24){
980 c->sub_motion_search(
s, &mx, &my, dmin, 0, 0, 0, 16);
989 && !
c->skip && varc>50<<8 && vard>10<<8){
1005 dmin=
c->sub_motion_search(
s, &mx, &my, dmin, 0, 0, 0, 16);
1006 if(
c->avctx->me_sub_cmp !=
c->avctx->mb_cmp && !
c->skip)
1010 && !
c->skip && varc>50<<8 && vard>10<<8){
1030 intra_score= varc - 500;
1032 unsigned mean = (sum+128)>>8;
1035 for(
i=0;
i<16;
i++){
1036 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+ 0]) =
mean;
1037 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+ 4]) =
mean;
1038 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+ 8]) =
mean;
1039 *(uint32_t*)(&
c->scratchpad[
i*
s->linesize+12]) =
mean;
1042 intra_score=
s->mecc.mb_cmp[0](
s,
c->scratchpad, pix,
s->linesize, 16);
1044 intra_score +=
c->mb_penalty_factor*16 +
s->intra_penalty;
1046 if(intra_score < dmin){
1050 s->current_picture.mb_type[mb_y*
s->mb_stride + mb_x] = 0;
1059 s->mb_type[mb_y*
s->mb_stride + mb_x]= mb_type;
1068 const int shift= 1+
s->quarter_sample;
1069 const int xy= mb_x + mb_y*
s->mb_stride;
1070 init_ref(
c,
s->new_picture->data,
s->last_picture.f->data,
NULL, 16*mb_x, 16*mb_y, 0);
1072 av_assert0(
s->quarter_sample==0 ||
s->quarter_sample==1);
1075 c->current_mv_penalty=
c->mv_penalty[
s->f_code] +
MAX_DMV;
1080 P_LEFT[0] =
s->p_mv_table[xy + 1][0];
1081 P_LEFT[1] =
s->p_mv_table[xy + 1][1];
1086 if (
s->first_slice_line) {
1092 P_TOP[0] =
s->p_mv_table[xy +
s->mb_stride ][0];
1093 P_TOP[1] =
s->p_mv_table[xy +
s->mb_stride ][1];
1094 P_TOPRIGHT[0] =
s->p_mv_table[xy +
s->mb_stride - 1][0];
1095 P_TOPRIGHT[1] =
s->p_mv_table[xy +
s->mb_stride - 1][1];
1109 s->p_mv_table[xy][0] = mx<<
shift;
1110 s->p_mv_table[xy][1] = my<<
shift;
1116 int16_t (*mv_table)[2],
int ref_index,
int f_code)
1119 int mx = 0, my = 0, dmin = 0;
1121 const int shift= 1+
s->quarter_sample;
1122 const int mot_stride =
s->mb_stride;
1123 const int mot_xy = mb_y*mot_stride + mb_x;
1135 P_LEFT[0] = mv_table[mot_xy - 1][0];
1136 P_LEFT[1] = mv_table[mot_xy - 1][1];
1141 if (!
s->first_slice_line) {
1142 P_TOP[0] = mv_table[mot_xy - mot_stride ][0];
1143 P_TOP[1] = mv_table[mot_xy - mot_stride ][1];
1144 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
1145 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
1156 if(mv_table ==
s->b_forw_mv_table){
1165 dmin=
c->sub_motion_search(
s, &mx, &my, dmin, 0, ref_index, 0, 16);
1167 if(
c->avctx->me_sub_cmp !=
c->avctx->mb_cmp && !
c->skip)
1171 mv_table[mot_xy][0]= mx;
1172 mv_table[mot_xy][1]= my;
1178 int motion_fx,
int motion_fy,
1179 int motion_bx,
int motion_by,
1180 int pred_fx,
int pred_fy,
1181 int pred_bx,
int pred_by,
1188 const uint8_t *
const mv_penalty_f =
c->mv_penalty[
s->f_code] +
MAX_DMV;
1189 const uint8_t *
const mv_penalty_b =
c->mv_penalty[
s->b_code] +
MAX_DMV;
1191 uint8_t *dest_y =
c->scratchpad;
1196 const uint8_t *
const *src_data =
c->src[0];
1197 const uint8_t *
const *ref_data =
c->ref[0];
1198 const uint8_t *
const *ref2_data =
c->ref[2];
1200 if(
s->quarter_sample){
1201 dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1202 src_x = motion_fx >> 2;
1203 src_y = motion_fy >> 2;
1205 ptr = ref_data[0] + (src_y *
stride) + src_x;
1206 s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr,
stride);
1208 dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1209 src_x = motion_bx >> 2;
1210 src_y = motion_by >> 2;
1212 ptr = ref2_data[0] + (src_y *
stride) + src_x;
1213 s->qdsp.avg_qpel_pixels_tab[
size][dxy](dest_y, ptr,
stride);
1215 dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1216 src_x = motion_fx >> 1;
1217 src_y = motion_fy >> 1;
1219 ptr = ref_data[0] + (src_y *
stride) + src_x;
1220 s->hdsp.put_pixels_tab[
size][dxy](dest_y , ptr ,
stride,
h);
1222 dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1223 src_x = motion_bx >> 1;
1224 src_y = motion_by >> 1;
1226 ptr = ref2_data[0] + (src_y *
stride) + src_x;
1227 s->hdsp.avg_pixels_tab[
size][dxy](dest_y , ptr ,
stride,
h);
1230 fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*
c->mb_penalty_factor
1231 +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*
c->mb_penalty_factor
1245 const int mot_stride =
s->mb_stride;
1246 const int xy = mb_y *mot_stride + mb_x;
1248 int pred_fx=
s->b_bidir_forw_mv_table[xy-1][0];
1249 int pred_fy=
s->b_bidir_forw_mv_table[xy-1][1];
1250 int pred_bx=
s->b_bidir_back_mv_table[xy-1][0];
1251 int pred_by=
s->b_bidir_back_mv_table[xy-1][1];
1252 int motion_fx=
s->b_bidir_forw_mv_table[xy][0]=
s->b_forw_mv_table[xy][0];
1253 int motion_fy=
s->b_bidir_forw_mv_table[xy][1]=
s->b_forw_mv_table[xy][1];
1254 int motion_bx=
s->b_bidir_back_mv_table[xy][0]=
s->b_back_mv_table[xy][0];
1255 int motion_by=
s->b_bidir_back_mv_table[xy][1]=
s->b_back_mv_table[xy][1];
1256 const int flags=
c->sub_flags;
1258 const int shift= 1+qpel;
1259 const int xmin=
c->xmin * (1 <<
shift);
1260 const int ymin=
c->ymin * (1 <<
shift);
1261 const int xmax=
c->xmax<<
shift;
1262 const int ymax=
c->ymax<<
shift;
1263 #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
1264 #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
1265 int hashidx=
HASH(motion_fx,motion_fy, motion_bx, motion_by);
1266 uint8_t
map[256] = { 0 };
1268 map[hashidx&255] = 1;
1271 motion_bx, motion_by,
1276 if(
s->avctx->bidir_refine){
1278 static const uint8_t limittab[5]={0,8,32,64,80};
1279 const int limit= limittab[
s->avctx->bidir_refine];
1280 static const int8_t vect[][4]={
1281 { 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},
1283 { 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},
1284 { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
1285 { 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},
1286 { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
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},
1291 { 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},
1293 { 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}, {-1, 1, 1, 1}, { 1,-1,-1,-1},
1295 { 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 static const uint8_t
hash[]={
1298 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),
1300 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),
1301 HASH8( 0, 1, 0, 1),
HASH8( 0,-1, 0,-1),
HASH8( 1, 0, 1, 0),
HASH8(-1, 0,-1, 0),
1302 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),
1303 HASH8( 0,-1, 0, 1),
HASH8( 0, 1, 0,-1),
HASH8(-1, 0, 1, 0),
HASH8( 1, 0,-1, 0),
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),
1308 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),
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),
HASH8(-1, 1, 1, 1),
HASH8( 1,-1,-1,-1),
1312 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),
1315 #define CHECK_BIDIR(fx,fy,bx,by)\
1316 if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
1317 &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1318 &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1320 map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
1321 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);\
1323 hashidx += HASH(fx,fy,bx,by);\
1332 #define CHECK_BIDIR2(a,b,c,d)\
1333 CHECK_BIDIR(a,b,c,d)\
1334 CHECK_BIDIR(-(a),-(b),-(c),-(d))
1347 int fx= motion_fx+vect[
i][0];
1348 int fy= motion_fy+vect[
i][1];
1349 int bx= motion_bx+vect[
i][2];
1350 int by= motion_by+vect[
i][3];
1352 int a= (xmax -
FFMAX(fx,bx))|(
FFMIN(fx,bx) - xmin);
1353 int b= (ymax -
FFMAX(fy,by))|(
FFMIN(fy,by) - ymin);
1360 score=
check_bidir_mv(
s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
1381 s->b_bidir_forw_mv_table[xy][0]= motion_fx;
1382 s->b_bidir_forw_mv_table[xy][1]= motion_fy;
1383 s->b_bidir_back_mv_table[xy][0]= motion_bx;
1384 s->b_bidir_back_mv_table[xy][1]= motion_by;
1393 const int mot_stride =
s->mb_stride;
1394 const int mot_xy = mb_y*mot_stride + mb_x;
1395 const int shift= 1+
s->quarter_sample;
1397 const int time_pp=
s->pp_time;
1398 const int time_pb=
s->pb_time;
1399 int mx, my, xmin, xmax, ymin, ymax;
1400 int16_t (*mv_table)[2]=
s->b_direct_mv_table;
1402 c->current_mv_penalty=
c->mv_penalty[1] +
MAX_DMV;
1403 ymin= xmin=(-32)>>
shift;
1404 ymax= xmax= 31>>
shift;
1406 if (
IS_8X8(
s->next_picture.mb_type[mot_xy])) {
1416 c->co_located_mv[
i][0] =
s->next_picture.motion_val[0][
index][0];
1417 c->co_located_mv[
i][1] =
s->next_picture.motion_val[0][
index][1];
1418 c->direct_basis_mv[
i][0]=
c->co_located_mv[
i][0]*time_pb/time_pp + ((
i& 1)<<(
shift+3));
1419 c->direct_basis_mv[
i][1]=
c->co_located_mv[
i][1]*time_pb/time_pp + ((
i>>1)<<(
shift+3));
1423 max=
FFMAX(
c->direct_basis_mv[
i][0],
c->direct_basis_mv[
i][0] -
c->co_located_mv[
i][0])>>
shift;
1424 min=
FFMIN(
c->direct_basis_mv[
i][0],
c->direct_basis_mv[
i][0] -
c->co_located_mv[
i][0])>>
shift;
1430 max=
FFMAX(
c->direct_basis_mv[
i][1],
c->direct_basis_mv[
i][1] -
c->co_located_mv[
i][1])>>
shift;
1431 min=
FFMIN(
c->direct_basis_mv[
i][1],
c->direct_basis_mv[
i][1] -
c->co_located_mv[
i][1])>>
shift;
1440 av_assert2(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1442 if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1443 s->b_direct_mv_table[mot_xy][0]= 0;
1444 s->b_direct_mv_table[mot_xy][1]= 0;
1446 return 256*256*256*64;
1462 if (!
s->first_slice_line) {
1478 if(
c->avctx->me_sub_cmp !=
c->avctx->mb_cmp && !
c->skip)
1483 mv_table[mot_xy][0]= mx;
1484 mv_table[mot_xy][1]= my;
1495 const int penalty_factor=
c->mb_penalty_factor;
1496 int fmin, bmin, dmin, fbmin, bimin, fimin;
1498 const int xy = mb_y*
s->mb_stride + mb_x;
1499 init_ref(
c,
s->new_picture->data,
s->last_picture.f->data,
1500 s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);
1509 score= ((unsigned)(score*score + 128*256))>>16;
1510 c->mc_mb_var_sum_temp += score;
1511 s->mc_mb_var[mb_y*
s->mb_stride + mb_x] = score;
1529 ff_dlog(
s,
" %d %d ",
s->b_forw_mv_table[xy][0],
s->b_forw_mv_table[xy][1]);
1538 c->current_mv_penalty=
c->mv_penalty[
s->f_code] +
MAX_DMV;
1540 s->b_field_mv_table[0],
s->b_field_select_table[0],
1541 s->b_forw_mv_table[xy][0],
s->b_forw_mv_table[xy][1], 0);
1542 c->current_mv_penalty=
c->mv_penalty[
s->b_code] +
MAX_DMV;
1544 s->b_field_mv_table[1],
s->b_field_select_table[1],
1545 s->b_back_mv_table[xy][0],
s->b_back_mv_table[xy][1], 0);
1547 fimin= bimin= INT_MAX;
1574 score= ((unsigned)(score*score + 128*256))>>16;
1575 c->mc_mb_var_sum_temp += score;
1576 s->mc_mb_var[mb_y*
s->mb_stride + mb_x] = score;
1585 if(fimin < INT_MAX && bimin < INT_MAX){
1595 s->mb_type[mb_y*
s->mb_stride + mb_x]=
type;
1603 int i, y, range=
s->avctx->me_range ?
s->avctx->me_range : (INT_MAX/2);
1606 int best_score=-10000000;
1608 if(
s->msmpeg4_version)
1609 range=
FFMIN(range, 16);
1611 range=
FFMIN(range, 256);
1613 for(
i=0;
i<8;
i++) score[
i]=
s->mb_num*(8-
i);
1615 for(y=0; y<
s->mb_height; y++){
1617 int xy= y*
s->mb_stride;
1618 for(x=0; x<
s->mb_width; x++, xy++){
1619 if(
s->mb_type[xy] &
type){
1620 int mx= mv_table[xy][0];
1621 int my= mv_table[xy][1];
1626 if (mx >= range || mx < -range ||
1627 my >= range || my < -range)
1630 for(j=0; j<fcode && j<8; j++){
1632 s->mc_mb_var[xy] <
s->mb_var[xy])
1640 if(score[
i] > best_score){
1641 best_score= score[
i];
1655 const int f_code=
s->f_code;
1659 range = (((
s->out_format ==
FMT_MPEG1 ||
s->msmpeg4_version) ? 8 : 16) << f_code);
1661 av_assert0(range <= 16 || !s->msmpeg4_version);
1664 if(
c->avctx->me_range && range >
c->avctx->me_range) range=
c->avctx->me_range;
1667 const int wrap=
s->b8_stride;
1670 for(y=0; y<
s->mb_height; y++){
1672 int i= y*
s->mb_stride;
1675 for(x=0; x<
s->mb_width; x++){
1680 int mx =
s->current_picture.motion_val[0][ xy + off ][0];
1681 int my =
s->current_picture.motion_val[0][ xy + off ][1];
1683 if( mx >=range || mx <-range
1684 || my >=range || my <-range){
1687 s->current_picture.mb_type[
i] =
type;
1702 int16_t (*mv_table)[2],
int f_code,
int type,
int truncate)
1705 int y, h_range, v_range;
1708 int range = (((
s->out_format ==
FMT_MPEG1 ||
s->msmpeg4_version) ? 8 : 16) << f_code);
1710 if(
c->avctx->me_range && range >
c->avctx->me_range) range=
c->avctx->me_range;
1713 v_range= field_select_table ? range>>1 : range;
1716 for(y=0; y<
s->mb_height; y++){
1718 int xy= y*
s->mb_stride;
1719 for(x=0; x<
s->mb_width; x++){
1720 if (
s->mb_type[xy] &
type){
1721 if (!field_select_table || field_select_table[xy] == field_select) {
1722 if( mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
1723 || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
1726 if (mv_table[xy][0] > h_range-1) mv_table[xy][0]= h_range-1;
1727 else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
1728 if (mv_table[xy][1] > v_range-1) mv_table[xy][1]= v_range-1;
1729 else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
1731 s->mb_type[xy] &= ~
type;