📄 mv-search.c
字号:
if(input->AllowTransform8x8)
mcost += find_SATD (curr_diff, input->hadamard, blocktype);
if (mcost < min_mcost)
{
min_mcost = mcost;
best_pos = pos;
}
}
if (best_pos)
{
*mv_x += (spiral_search_x [best_pos] << 1);
*mv_y += (spiral_search_y [best_pos] << 1);
}
/************************************
***** *****
***** QUARTER-PEL REFINEMENT *****
***** *****
************************************/
//===== set function for getting pixel values =====
if ((pic4_pix_x + *mv_x > 1) && (pic4_pix_x + *mv_x < max_pos_x4 - 1) &&
(pic4_pix_y + *mv_y > 1) && (pic4_pix_y + *mv_y < max_pos_y4 - 1) )
{
PelY_14 = FastPelY_14;
}
else
{
PelY_14 = UMVPelY_14;
}
//===== loop over search positions =====
for (best_pos = 0, pos = 1; pos < search_pos4; pos++)
{
cand_mv_x = *mv_x + spiral_search_x[pos]; // quarter-pel units
cand_mv_y = *mv_y + spiral_search_y[pos]; // quarter-pel units
//----- set motion vector cost -----
mcost = MV_COST (lambda_factor, mv_shift, cand_mv_x, cand_mv_y, pred_mv_x, pred_mv_y);
if (mcost >= min_mcost) continue;
//----- add up SATD -----
for (y0=0, abort_search=0; y0<blocksize_y && !abort_search; y0+=4)
{
ry0 = ((pic_pix_y+y0)<<2) + cand_mv_y;
for (x0=0; x0<blocksize_x; x0+=4)
{
rx0 = ((pic_pix_x+x0)<<2) + cand_mv_x;
d = diff;
orig_line = orig_pic [y0 ]; ry=ry0;
*d++ = orig_line[x0 ] - PelY_14 (ref_pic, ry, rx0 , img_height, img_width);
*d++ = orig_line[x0+1] - PelY_14 (ref_pic, ry, rx0+ 4, img_height, img_width);
*d++ = orig_line[x0+2] - PelY_14 (ref_pic, ry, rx0+ 8, img_height, img_width);
*d++ = orig_line[x0+3] - PelY_14 (ref_pic, ry, rx0+12, img_height, img_width);
orig_line = orig_pic [y0+1]; ry=ry0+4;
*d++ = orig_line[x0 ] - PelY_14 (ref_pic, ry, rx0 , img_height, img_width);
*d++ = orig_line[x0+1] - PelY_14 (ref_pic, ry, rx0+ 4, img_height, img_width);
*d++ = orig_line[x0+2] - PelY_14 (ref_pic, ry, rx0+ 8, img_height, img_width);
*d++ = orig_line[x0+3] - PelY_14 (ref_pic, ry, rx0+12, img_height, img_width);
orig_line = orig_pic [y0+2]; ry=ry0+8;
*d++ = orig_line[x0 ] - PelY_14 (ref_pic, ry, rx0 , img_height, img_width);
*d++ = orig_line[x0+1] - PelY_14 (ref_pic, ry, rx0+ 4, img_height, img_width);
*d++ = orig_line[x0+2] - PelY_14 (ref_pic, ry, rx0+ 8, img_height, img_width);
*d++ = orig_line[x0+3] - PelY_14 (ref_pic, ry, rx0+12, img_height, img_width);
orig_line = orig_pic [y0+3]; ry=ry0+12;
*d++ = orig_line[x0 ] - PelY_14 (ref_pic, ry, rx0 , img_height, img_width);
*d++ = orig_line[x0+1] - PelY_14 (ref_pic, ry, rx0+ 4, img_height, img_width);
*d++ = orig_line[x0+2] - PelY_14 (ref_pic, ry, rx0+ 8, img_height, img_width);
*d = orig_line[x0+3] - PelY_14 (ref_pic, ry, rx0+12, img_height, img_width);
if (!input->AllowTransform8x8)
{
if ((mcost += SATD (diff, input->hadamard)) > min_mcost)
{
abort_search = 1;
break;
}
}
else
{
for(j=0, k=0; j<4; j++)
for(i=0; i<4; i++, k++)
curr_diff[y0+j][x0+i] = diff[k];
}
}
}
if(input->AllowTransform8x8)
mcost += find_SATD (curr_diff, input->hadamard, blocktype);
if (mcost < min_mcost)
{
min_mcost = mcost;
best_pos = pos;
}
}
if (best_pos)
{
*mv_x += spiral_search_x [best_pos];
*mv_y += spiral_search_y [best_pos];
}
//===== return minimum motion cost =====
return min_mcost;
}
/*!
***********************************************************************
* \brief
* Block motion search
***********************************************************************
*/
int //!< minimum motion cost after search
BlockMotionSearch (int ref, //!< reference idx
int list, //!< reference pciture list
int mb_x, //!< x-coordinate inside macroblock
int mb_y, //!< y-coordinate inside macroblock
int blocktype, //!< block type (1-16x16 ... 7-4x4)
int search_range, //!< 1-d search range for integer-position search
double lambda //!< lagrangian parameter for determining motion cost
)
{
static pel_t orig_val [256];
static pel_t *orig_pic [16] = {orig_val, orig_val+ 16, orig_val+ 32, orig_val+ 48,
orig_val+ 64, orig_val+ 80, orig_val+ 96, orig_val+112,
orig_val+128, orig_val+144, orig_val+160, orig_val+176,
orig_val+192, orig_val+208, orig_val+224, orig_val+240};
int pred_mv_x, pred_mv_y, mv_x, mv_y, i, j;
int max_value = INT_MAX;
int min_mcost = max_value;
int block_x = (mb_x>>2);
int block_y = (mb_y>>2);
int bsx = input->blc_size[blocktype][0];
int bsy = input->blc_size[blocktype][1];
int pic_pix_x = img->opix_x + mb_x;
int pic_pix_y = img->opix_y + mb_y;
int* pred_mv;
int*** mv_array = enc_picture->mv[list];
int****** all_mv = img->all_mv;
#ifdef WIN32
struct _timeb tstruct1;
struct _timeb tstruct2;
#else
struct timeb tstruct1;
struct timeb tstruct2;
#endif
int me_tmp_time;
int N_Bframe=0, n_Bframe=0;
if(input->FMEnable)
{
N_Bframe = input->successive_Bframe;
n_Bframe =(N_Bframe) ? ((Bframe_ctr%N_Bframe)+1) : 0 ;
}
pred_mv = img->pred_mv[block_x][block_y][list][ref][blocktype];
//==================================
//===== GET ORIGINAL BLOCK =====
//==================================
for (j = 0; j < bsy; j++)
{
for (i = 0; i < bsx; i++)
{
orig_pic[j][i] = imgY_org[pic_pix_y+j][pic_pix_x+i];
}
}
if(input->FMEnable)
{
if(blocktype>6)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][5][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][5][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][5][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][5][0]);
pred_SAD_uplayer /= 2;
}
else if(blocktype>4)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][4][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][4][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][4][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][4][0]);
pred_SAD_uplayer /= 2;
}
else if(blocktype == 4)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][2][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][2][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][2][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][2][0]);
pred_SAD_uplayer /= 2;
}
else if(blocktype > 1)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][1][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][1][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][1][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][1][0]);
pred_SAD_uplayer /= 2;
}
if ((img->type==B_SLICE)&& (img->nal_reference_idc>0))
{
if(blocktype>6)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][5][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][5][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][5][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][5][0]);
pred_SAD_uplayer /= 2;
}
else if(blocktype>4)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][4][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][4][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][4][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][4][0]);
pred_SAD_uplayer /= 2;
}
else if(blocktype == 4)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][2][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][2][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][2][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][2][0]);
pred_SAD_uplayer /= 2;
}
else if(blocktype > 1)
{
pred_MV_uplayer[0] = all_mv[block_x][block_y][list][ref][1][0];
pred_MV_uplayer[1] = all_mv[block_x][block_y][list][ref][1][1];
pred_SAD_uplayer = (list==1) ? (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][1][0]) : (all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][1][0]);
pred_SAD_uplayer /= 2;
}
}
pred_SAD_uplayer = flag_intra_SAD ? 0 : pred_SAD_uplayer;// for irregular motion
//Coordinate prediction
if (img->number > ref+1)
{
pred_SAD_time = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][0];
pred_MV_time[0] = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][1];
pred_MV_time[1] = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][2];
}
if(list==1 && (Bframe_ctr%N_Bframe) > 1)
{
pred_SAD_time = all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][0];
pred_MV_time[0] = (int)(all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][blocktype][1] * ((n_Bframe==1) ? (N_Bframe) : (N_Bframe-n_Bframe+1.0)/(N_Bframe-n_Bframe+2.0)) );//should add a factor
pred_MV_time[1] = (int)(all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][blocktype][2] *((n_Bframe==1) ? (N_Bframe) : (N_Bframe-n_Bframe+1.0)/(N_Bframe-n_Bframe+2.0)) );//should add a factor
}
if (input->PicInterlace == FIELD_CODING)
{
if (img->type == P_SLICE && ref > 1)
{
pred_SAD_ref = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][(ref-2)][blocktype][0];
pred_SAD_ref = flag_intra_SAD ? 0 : pred_SAD_ref;//add this for irregular motion
pred_MV_ref[0] = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][(ref-2)][blocktype][1];
pred_MV_ref[0] = (int)(pred_MV_ref[0]*((ref>>1)+1)/(float)((ref>>1)));
pred_MV_ref[1] = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][(ref-2)][blocktype][2];
pred_MV_ref[1] = (int)(pred_MV_ref[1]*((ref>>1)+1)/(float)((ref>>1)));
}
if (img->type == B_SLICE && list==0 && (ref==0 || ref==1) )
{
pred_SAD_ref = all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][blocktype][0];
pred_SAD_ref = flag_intra_SAD ? 0 : pred_SAD_ref;//add this for irregular motion
pred_MV_ref[0] =(int) (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][1]*(-n_Bframe)/(N_Bframe-n_Bframe+1.0f)); //should add a factor
pred_MV_ref[1] =(int) ( all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][2]*(-n_Bframe)/(N_Bframe-n_Bframe+1.0f));
}
}
else //frame case
{
if (ref > 0)
{//field_mode top_field
pred_SAD_ref = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][(ref-1)][blocktype][0];
pred_SAD_ref = flag_intra_SAD ? 0 : pred_SAD_ref;//add this for irregular motion
pred_MV_ref[0] = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][(ref-1)][blocktype][1];
pred_MV_ref[0] = (int)(pred_MV_ref[0]*(ref+1)/(float)(ref));
pred_MV_ref[1] = all_mincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][(ref-1)][blocktype][2];
pred_MV_ref[1] = (int)(pred_MV_ref[1]*(ref+1)/(float)(ref));
}
if (img->type == B_SLICE && (list==0 && ref==0)) //B frame forward prediction, first ref
{
pred_SAD_ref = all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][0][blocktype][0];
pred_SAD_ref = flag_intra_SAD ? 0 : pred_SAD_ref;//add this for irregular motion
pred_MV_ref[0] =(int) (all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][1]*(-n_Bframe)/(N_Bframe-n_Bframe+1.0f)); //should add a factor
pred_MV_ref[1] =(int) ( all_bwmincost[(img->pix_x>>2)+block_x][(img->pix_y>>2)+block_y][ref][blocktype][2]*(-n_Bframe)/(N_Bframe-n_Bframe+1.0f));
}
}
}
//===========================================
//===== GET MOTION VECTOR PREDICTOR =====
//===========================================
if(input->FMEnable)
FME_blocktype=blocktype;
SetMotionVectorPredictor (pred_mv, enc_picture->ref_idx, enc_picture->mv, ref, list, block_x, block_y, bsx, bsy);
pred_mv_x = pred_mv[0];
pred_mv_y = pred_mv[1];
#ifdef WIN32
_ftime( &tstruct1 ); // start time ms
#else
ftime(&tstruct1);
#endif
//==================================
//===== INTEGER-PEL SEARCH =====
//==================================
if(input->FMEnable)
{
mv_x = pred_mv_x / 4;
mv_y = pred_mv_y / 4;
if (!input->rdopt)
{
//--- adjust search center so that the (0,0)-vector is inside ---
mv_x = max (-search_range, min (search_range, mv_x));
mv_y = max (-search_range, min (search_range, mv_y));
}
min_mcost = FastIntegerPelBlockMotionSearch(orig_pic, ref, list, pic_pix_x, pic_pix_y, blocktype,
pred_mv_x, pred_mv_y, &mv_x, &mv_y, search_range,
min_mcost, lambda);
//FAST MOTION ESTIMATION. ZHIBO CHEN 2003.3
for (i=0; i < (bsx>>2); i++)
{
for ⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -