me_fullsearch.c

This program can encode the YUV vdieo format to H.264 and decode it.

/*!
*************************************************************************************
* \file me_fullsearch.c
*
* \brief
*    Motion Estimation using Fullsearch
*
* \author
*    Main contributors (see contributors.h for copyright, address and affiliation details)
*      - Alexis Michael Tourapis <[e-mail]>
*      - Athanasios Leontaris    <[e-mail]>
*
*************************************************************************************
*/

// Includes
#include "contributors.h"
#include <stdlib.h>
#include <limits.h>
#include <string.h>

#include "global.h"
#include "image.h"
#include "memalloc.h"
#include "mb_access.h"
#include "refbuf.h"

#include "me_distortion.h"
#include "me_fullsearch.h"

// Define Global Parameters
extern int *mvbits;
extern short*  spiral_search_x;
extern short*  spiral_search_y;
extern short*  spiral_hpel_search_x;
extern short*  spiral_hpel_search_y;

// Functions
/*!
 ***********************************************************************
 * \brief
 *    Full pixel block motion search
 ***********************************************************************
 */
int                                               //  ==> minimum motion cost after search
FullPelBlockMotionSearch (imgpel*   orig_pic,     // <--  original pixel values for the AxB block
                          short     ref,          // <--  reference frame (0... or -1 (backward))
                          int       list,         // <--  current list
                          int       pic_pix_x,    // <--  absolute x-coordinate of regarded AxB block
                          int       pic_pix_y,    // <--  absolute y-coordinate of regarded AxB block
                          int       blocktype,    // <--  block type (1-16x16 ... 7-4x4)
                          short     pred_mv_x,    // <--  motion vector predictor (x) in sub-pel units
                          short     pred_mv_y,    // <--  motion vector predictor (y) in sub-pel units
                          short*    mv_x,         // <--> in: search center (x) / out: motion vector (x) - in pel units
                          short*    mv_y,         // <--> in: search center (y) / out: motion vector (y) - in pel units
                          int       search_range, // <--  1-d search range in pel units
                          int       min_mcost,    // <--  minimum motion cost (cost for center or huge value)
                          int       lambda_factor)       // <--  lagrangian parameter for determining motion cost
{
  int   pos, cand_x, cand_y, mcost;

  int   list_offset   = img->mb_data[img->current_mb_nr].list_offset;
  StorablePicture *ref_picture = listX[list+list_offset][ref];

  int   best_pos      = 0;                                        // position with minimum motion cost
  int   max_pos       = (2*search_range+1)*(2*search_range+1);    // number of search positions
  int   blocksize_y   = input->blc_size[blocktype][1];            // vertical block size
  int   blocksize_x   = input->blc_size[blocktype][0];            // horizontal block size

  int   pred_x        = (pic_pix_x << 2) + pred_mv_x;       // predicted position x (in sub-pel units)
  int   pred_y        = (pic_pix_y << 2) + pred_mv_y;       // predicted position y (in sub-pel units)
  int   center_x      = pic_pix_x + *mv_x;                        // center position x (in pel units)
  int   center_y      = pic_pix_y + *mv_y;                        // center position y (in pel units)
  int   check_for_00  = (blocktype==1 && !input->rdopt && img->type!=B_SLICE && ref==0);
  int   apply_weights = ( (active_pps->weighted_pred_flag && (img->type == P_SLICE || img->type == SP_SLICE)) ||
    (active_pps->weighted_bipred_idc && (img->type == B_SLICE))) && input->UseWeightedReferenceME;
  int   dist_method = F_PEL + 3 * apply_weights;

  ref_pic_sub.luma = ref_picture->curr_imgY_sub;

  img_width  = ref_picture->size_x;
  img_height = ref_picture->size_y;
  width_pad  = ref_picture->size_x_pad;
  height_pad = ref_picture->size_y_pad;

  if (apply_weights)
  {
    weight_luma = wp_weight[list + list_offset][ref][0];
    offset_luma = wp_offset[list + list_offset][ref][0];
  }

  if (ChromaMEEnable)
  {
    ref_pic_sub.crcb[0] = ref_picture->imgUV_sub[0];
    ref_pic_sub.crcb[1] = ref_picture->imgUV_sub[1];
    width_pad_cr  = ref_picture->size_x_cr_pad;
    height_pad_cr = ref_picture->size_y_cr_pad;

    if (apply_weights)
    {
      weight_cr[0] = wp_weight[list + list_offset][ref][1];
      weight_cr[1] = wp_weight[list + list_offset][ref][2];
      offset_cr[0] = wp_offset[list + list_offset][ref][1];
      offset_cr[1] = wp_offset[list + list_offset][ref][2];
    }
  }

  //===== set function for getting reference picture lines =====
  if ((center_x > search_range) && (center_x < img_width -1-search_range-blocksize_x) &&
    (center_y > search_range) && (center_y < img_height-1-search_range-blocksize_y)   )
  {
    ref_access_method = FAST_ACCESS;
  }
  else
  {
    ref_access_method = UMV_ACCESS;
  }

  //===== loop over all search positions =====
  for (pos=0; pos<max_pos; pos++)
  {
    //--- set candidate position (absolute position in pel units) ---
    cand_x = (center_x + spiral_search_x[pos])<<2;
    cand_y = (center_y + spiral_search_y[pos])<<2;

    //--- initialize motion cost (cost for motion vector) and check ---
    mcost = MV_COST_SMP (lambda_factor, cand_x, cand_y, pred_x, pred_y);
    if (check_for_00 && cand_x==pic_pix_x && cand_y==pic_pix_y)
    {
      mcost -= WEIGHTED_COST (lambda_factor, 16);
    }
    if (mcost >= min_mcost)   continue;

    //--- add residual cost to motion cost ---
    mcost += computeUniPred[dist_method](orig_pic, blocksize_y, blocksize_x,
      min_mcost - mcost, cand_x + IMG_PAD_SIZE_TIMES4, cand_y + IMG_PAD_SIZE_TIMES4);

    //--- check if motion cost is less than minimum cost ---
    if (mcost < min_mcost)
    {
      best_pos  = pos;
      min_mcost = mcost;
    }
  }


  //===== set best motion vector and return minimum motion cost =====
  if (best_pos)
  {
    *mv_x += spiral_search_x[best_pos];
    *mv_y += spiral_search_y[best_pos];
  }
  return min_mcost;
}

/*!
 ***********************************************************************
 * \brief
 *    Full pixel block motion search
 ***********************************************************************
 */
int                                                //  ==> minimum motion cost after search
FullPelBlockMotionBiPred (imgpel*   orig_pic,      // <--  original pixel values for the AxB block
                          short       ref,         // <--  reference frame (0... or -1 (backward))
                          int       list,
                          int       pic_pix_x,     // <--  absolute x-coordinate of regarded AxB block
                          int       pic_pix_y,     // <--  absolute y-coordinate of regarded AxB block
                          int       blocktype,     // <--  block type (1-16x16 ... 7-4x4)
                          short     pred_mv_x1,    // <--  motion vector predictor (x) in sub-pel units
                          short     pred_mv_y1,    // <--  motion vector predictor (y) in sub-pel units
                          short     pred_mv_x2,    // <--  motion vector predictor (x) in sub-pel units
                          short     pred_mv_y2,    // <--  motion vector predictor (y) in sub-pel units
                          short*    mv_x,          // <--> in: search center (x) / out: motion vector (x) - in pel units
                          short*    mv_y,          // <--> in: search center (y) / out: motion vector (y) - in pel units
                          short*    s_mv_x,        // <--> in: search center (x) / out: motion vector (x) - in pel units
                          short*    s_mv_y,        // <--> in: search center (y) / out: motion vector (y) - in pel units
                          int       search_range,  // <--  1-d search range in pel units
                          int       min_mcost,     // <--  minimum motion cost (cost for center or huge value)
                          int       lambda_factor) // <--  lagrangian parameter for determining motion cost
{
  int   pos, cand_x, cand_y, mcost;


  int   list_offset   = img->mb_data[img->current_mb_nr].list_offset;

  int   best_pos      = 0;                                        // position with minimum motion cost
  int   max_pos       = (2*search_range+1)*(2*search_range+1);    // number of search positions
  int   blocksize_y   = input->blc_size[blocktype][1];            // vertical block size
  int   blocksize_x   = input->blc_size[blocktype][0];            // horizontal block size

  int   pred_x1        = (pic_pix_x << 2) + pred_mv_x1;       // predicted position x (in sub-pel units)
  int   pred_y1        = (pic_pix_y << 2) + pred_mv_y1;       // predicted position y (in sub-pel units)
  int   pred_x2        = (pic_pix_x << 2) + pred_mv_x2;       // predicted position x (in sub-pel units)
  int   pred_y2        = (pic_pix_y << 2) + pred_mv_y2;       // predicted position y (in sub-pel units)
  short center_x      = pic_pix_x + *mv_x;                      // center position x (in pel units)
  short center_y      = pic_pix_y + *mv_y;                      // center position y (in pel units)
  short ref1_center_x = pic_pix_x + *s_mv_x;                      // mvx of second pred (in pel units)
  short ref1_center_y = pic_pix_y + *s_mv_y;                      // mvy of second pred (in pel units)


  short apply_weights = (active_pps->weighted_bipred_idc>0);

  short offset1 = (apply_weights ? (list == 0?  wp_offset[list_offset    ][ref][0]:  wp_offset[list_offset + 1][0  ][ref]) : 0);
  short offset2 = (apply_weights ? (list == 0?  wp_offset[list_offset + 1][ref][0]:  wp_offset[list_offset    ][0  ][ref]) : 0);

  StorablePicture *ref_picture1 = listX[list       + list_offset][ref];
  StorablePicture *ref_picture2 = listX[(list ^ 1) + list_offset][0];

  ref_pic1_sub.luma = ref_picture1->curr_imgY_sub;
  ref_pic2_sub.luma = ref_picture2->curr_imgY_sub;

  img_width     = ref_picture1->size_x;
  img_height    = ref_picture1->size_y;
  width_pad    = ref_picture1->size_x_pad;
  height_pad   = ref_picture1->size_y_pad;

  if (apply_weights)
  {
    weight1 = list == 0
      ? wbp_weight[list_offset         ][ref][0][0]
      : wbp_weight[list_offset + LIST_1][0  ][ref][0];
    weight2 = list == 0
      ? wbp_weight[list_offset + LIST_1][ref][0][0]
      : wbp_weight[list_offset         ][0  ][ref][0];
    offsetBi=(offset1 + offset2 + 1)>>1;
    computeBiPred = computeBiPred2[F_PEL];
  }
  else
  {
    weight1 = 1<<luma_log_weight_denom;
    weight2 = 1<<luma_log_weight_denom;
    offsetBi = 0;
    computeBiPred = computeBiPred1[F_PEL];
  }

  if (ChromaMEEnable )
  {
    ref_pic1_sub.crcb[0] = ref_picture1->imgUV_sub[0];
    ref_pic1_sub.crcb[1] = ref_picture1->imgUV_sub[1];
    ref_pic2_sub.crcb[0] = ref_picture2->imgUV_sub[0];
    ref_pic2_sub.crcb[1] = ref_picture2->imgUV_sub[1];
    width_pad_cr  = ref_picture1->size_x_cr_pad;
    height_pad_cr = ref_picture1->size_y_cr_pad;
    if (apply_weights)
    {
      weight1_cr[0] = list == 0
        ? wbp_weight[list_offset         ][ref][0][1]
        : wbp_weight[list_offset + LIST_1][0  ][ref][1];
      weight1_cr[1] = list == 0
        ? wbp_weight[list_offset         ][ref][0][2]
        : wbp_weight[list_offset + LIST_1][0  ][ref][2];