libavcodec.hg: motion_est.c comparison

comparison motion_est.c @ 1962:1d5abf80fa41 libavcodec

minor motion estimation cleanup 4mv motion vector passthrough

author	michael
date	Sat, 24 Apr 2004 03:36:37 +0000
parents	5dafb10e0252
children	2fd0c3f311bf

comparison

equal deleted inserted replaced

-:603d4a0e974c
+:1d5abf80fa41
 static int get_flags(MpegEncContext *s, int direct, int chroma){
 return   ((s->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
 + (direct ? FLAG_DIRECT : 0)
 + (chroma ? FLAG_CHROMA : 0);
-}
-static inline void init_mc(MpegEncContext *s, int size, int flags){
-MotionEstContext * const c= &s->me;
-/*FIXME s->no_rounding b_type*/
-if(flags & FLAG_CHROMA){
-if(s->no_rounding) c->chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];
-else               c->chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
-}
-if(flags & FLAG_QPEL){
-c->qpel_avg= &s->dsp.avg_qpel_pixels_tab[size];
-if(s->no_rounding) c->qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];
-else               c->qpel_put= &s->dsp.put_qpel_pixels_tab[size];
-}else{
-c->hpel_avg= &s->dsp.avg_pixels_tab[size];
-if(s->no_rounding) c->hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];
-else               c->hpel_put= &s->dsp.put_pixels_tab[size];
-}
-c->temp= c->scratchpad;
 }
 static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
 const int size, const int h, int ref_index, int src_index,
 me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
 return (s->qscale*s->qscale*185 + 64)>>7;
 }
 }
 void ff_init_me(MpegEncContext *s){
+MotionEstContext * const c= &s->me;
 ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, s->avctx->me_pre_cmp);
 ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
 ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
 ff_set_cmp(&s->dsp, s->dsp.mb_cmp, s->avctx->mb_cmp);
 s->me.flags    = get_flags(s, 0, s->avctx->me_cmp    &FF_CMP_CHROMA);
 s->me.sub_flags= get_flags(s, 0, s->avctx->me_sub_cmp&FF_CMP_CHROMA);
 s->me.mb_flags = get_flags(s, 0, s->avctx->mb_cmp    &FF_CMP_CHROMA);
+/*FIXME s->no_rounding b_type*/
 if(s->flags&CODEC_FLAG_QPEL){
 s->me.sub_motion_search= qpel_motion_search;
+c->qpel_avg= s->dsp.avg_qpel_pixels_tab;
+if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab;
+else               c->qpel_put= s->dsp.put_qpel_pixels_tab;
 }else{
 if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)
 s->me.sub_motion_search= hpel_motion_search;
 else if(   s->avctx->me_sub_cmp == FF_CMP_SAD
 && s->avctx->    me_cmp == FF_CMP_SAD
 && s->avctx->    mb_cmp == FF_CMP_SAD)
 s->me.sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
 else
 s->me.sub_motion_search= hpel_motion_search;
+c->hpel_avg= s->dsp.avg_pixels_tab;
+if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab;
+else               c->hpel_put= s->dsp.put_pixels_tab;
 }
 if(s->linesize){
 s->me.stride  = s->linesize;
 s->me.uvstride= s->uvlinesize;
 }else{
 s->me.stride  = 16*s->mb_width + 32;
 s->me.uvstride=  8*s->mb_width + 16;
 }
+c->temp= c->scratchpad;
 }
 #if 0
 static int pix_dev(uint8_t * pix, int line_size, int mean)
 {
 s->me.xmax = - x + s->mb_width *16 - 16;
 s->me.ymax = - y + s->mb_height*16 - 16;
 }
 }
+static inline void init_mv4_ref(MpegEncContext *s){
+MotionEstContext * const c= &s->me;
+const int stride= s->linesize;
+c->ref[1][0] = c->ref[0][0] + 8;
+c->ref[2][0] = c->ref[0][0] + 8*stride;
+c->ref[3][0] = c->ref[2][0] + 8;
+c->src[1][0] = c->src[0][0] + 8;
+c->src[2][0] = c->src[0][0] + 8*stride;
+c->src[3][0] = c->src[2][0] + 8;
+}
 static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
 {
 MotionEstContext * const c= &s->me;
 const int size= 1;
 const int h=8;
 int same=1;
 const int stride= s->linesize;
 const int uvstride= s->uvlinesize;
 uint8_t *mv_penalty= s->me.current_mv_penalty;
-c->ref[1][0] = c->ref[0][0] + 8;
+init_mv4_ref(s);
-c->ref[2][0] = c->ref[0][0] + 8*stride;
-c->ref[3][0] = c->ref[2][0] + 8;
-c->src[1][0] = c->src[0][0] + 8;
-c->src[2][0] = c->src[0][0] + 8*stride;
-c->src[3][0] = c->src[2][0] + 8;
-init_mc(s, size, c->flags);
 for(block=0; block<4; block++){
 int mx4, my4;
 int pred_x4, pred_y4;
 int dmin4;
 default:
 return dmin_sum+ 11*s->me.mb_penalty_factor;
 }
 }
+static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
+MotionEstContext * const c= &s->me;
+c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
+c->src[1][0] = c->src[0][0] + s->linesize;
+if(c->flags & FLAG_CHROMA){
+c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
+c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
+c->src[1][1] = c->src[0][1] + s->uvlinesize;
+c->src[1][2] = c->src[0][2] + s->uvlinesize;
+}
+}
 static int interlaced_search(MpegEncContext *s, int ref_index,
 int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my)
 {
 MotionEstContext * const c= &s->me;
 const int size=0;
 c->ymin>>=1;
 c->ymax>>=1;
 c->stride<<=1;
 c->uvstride<<=1;
-c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
+init_interlaced_ref(s, ref_index);
-c->src[1][0] = c->src[0][0] + s->linesize;
-if(c->flags & FLAG_CHROMA){
-c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
-c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
-c->src[1][1] = c->src[0][1] + s->uvlinesize;
-c->src[1][2] = c->src[0][2] + s->uvlinesize;
-}
-init_mc(s, size, c->flags);
 for(block=0; block<2; block++){
 int field_select;
 int best_dmin= INT_MAX;
 int best_field= -1;
 int xy= 2*mb_x + 2*mb_y*s->b8_stride;
 int mb_type= s->current_picture.mb_type[mb_xy];
 int flags= c->flags;
 int shift= (flags&FLAG_QPEL) + 1;
 int mask= (1<<shift)-1;
-int x, y;
+int x, y, i;
 int d=0;
 me_cmp_func cmpf= s->dsp.sse[0];
 me_cmp_func chroma_cmpf= s->dsp.sse[1];
 assert(p_type==0 || !USES_LIST(mb_type, 1));
 if(IS_INTERLACED(mb_type)){
 int xy2= xy  + s->b8_stride;
 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
 c->stride<<=1;
 c->uvstride<<=1;
-c->ref[1][0] = c->ref[0][0] + s->linesize;
+init_interlaced_ref(s, 2);
-c->ref[3][0] = c->ref[2][0] + s->linesize;
-c->src[1][0] = c->src[0][0] + s->linesize;
-if(c->flags & FLAG_CHROMA){
-c->ref[1][1] = c->ref[0][1] + s->uvlinesize;
-c->ref[1][2] = c->ref[0][2] + s->uvlinesize;
-c->ref[3][1] = c->ref[2][1] + s->uvlinesize;
-c->ref[3][2] = c->ref[2][2] + s->uvlinesize;
-c->src[1][1] = c->src[0][1] + s->uvlinesize;
-c->src[1][2] = c->src[0][2] + s->uvlinesize;
-}
 if(USES_LIST(mb_type, 0)){
 int field_select0= p->ref_index[0][xy ];
 int field_select1= p->ref_index[0][xy2];
 assert(field_select0==0 ||field_select0==1);
 assert(field_select1==0 ||field_select1==1);
 d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1+2, 1, cmpf, chroma_cmpf, flags);
 //FIXME bidir scores
 }
 c->stride>>=1;
 c->uvstride>>=1;
+}else if(IS_8X8(mb_type)){
+cmpf= s->dsp.sse[1];
+chroma_cmpf= s->dsp.sse[1];
+init_mv4_ref(s);
+for(i=0; i<4; i++){
+xy= s->block_index[i];
+x= p->motion_val[0][xy][0];
+y= p->motion_val[0][xy][1];
+d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 1, 8, i, i, cmpf, chroma_cmpf, flags);
+}
+s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER4V;
 }else{
 if(USES_LIST(mb_type, 0)){
 if(p_type){
 *(uint32_t*)s->p_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER;
 const int shift= 1+s->quarter_sample;
 int mb_type=0;
 Picture * const pic= &s->current_picture;
 init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
-init_mc(s, 0, c->flags);
 assert(s->quarter_sample==0 || s->quarter_sample==1);
 assert(s->linesize == s->me.stride);
 assert(s->uvlinesize == s->me.uvstride);
 int mx, my, dmin;
 int P[10][2];
 const int shift= 1+s->quarter_sample;
 const int xy= mb_x + mb_y*s->mb_stride;
 init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
-init_mc(s, 0, c->flags);
 assert(s->quarter_sample==0 || s->quarter_sample==1);
 s->me.pre_penalty_factor    = get_penalty_factor(s, s->avctx->me_pre_cmp);
 s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;
 {
 const int penalty_factor= s->me.mb_penalty_factor;
 int fmin, bmin, dmin, fbmin, bimin, fimin;
 int type=0;
 init_ref(s, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
-init_mc(s, 0, s->me.flags);
 s->me.skip=0;
 if(s->avctx->me_threshold){
 int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;

Mercurial > libavcodec.hg

comparison motion_est.c @ 1962:1d5abf80fa41 libavcodec