md_low.c

H.264编码实现

/*!
 ***************************************************************************
 * \file md_low.c
 *
 * \brief
 *    Main macroblock mode decision functions and helpers
 *
 **************************************************************************
 */

#include <math.h>
#include <limits.h>
#include <float.h>

#include "global.h"
#include "rdopt_coding_state.h"
#include "mb_access.h"
#include "intrarefresh.h"
#include "image.h"
#include "transform8x8.h"
#include "ratectl.h"
#include "mode_decision.h"
#include "fmo.h"
#include "me_umhex.h"
#include "me_umhexsmp.h"
#include "macroblock.h"
#include "q_around.h"


//==== MODULE PARAMETERS ====
static imgpel temp_imgY[16][16]; // to temp store the Y data for 8x8 transform
static imgpel temp_imgU[16][16];
static imgpel temp_imgV[16][16];


/*!
*************************************************************************************
* \brief
*    Mode Decision for a macroblock
*************************************************************************************
*/
void encode_one_macroblock_low (Macroblock *currMB)
{

  int         block, mode, i=0, j, k, dummy;
  char        best_pdir;
  RD_PARAMS   enc_mb;
  char        best_ref[2] = {0, -1};
  int         bmcost[5] = {INT_MAX};
  double      rd_cost = 0, min_rd_cost = 1e30;
  int         cost = 0;
  int         min_cost = INT_MAX, cost_direct=0, have_direct=0, i16mode=0;
  int         intra1 = 0;
  int         temp_cpb = 0;
  int         best_transform_flag = 0;
  int         cost8x8_direct = 0;
  short       islice      = (short) (img->type==I_SLICE);
  short       bslice      = (short) (img->type==B_SLICE);
  short       pslice      = (short) ((img->type==P_SLICE) || (img->type==SP_SLICE));
  short       intra       = (short) (islice || (pslice && img->mb_y==img->mb_y_upd && img->mb_y_upd!=img->mb_y_intra));
  int         lambda_mf[3];
  int         prev_mb_nr  = FmoGetPreviousMBNr(img->current_mb_nr);
  Macroblock* prevMB      = (prev_mb_nr >= 0) ? &img->mb_data[prev_mb_nr]:NULL ;
  Block8x8Info *b8x8info   = img->b8x8info;

  char   **ipredmodes = img->ipredmode;
  short   *allmvs = params->IntraProfile ? NULL: img->all_mv[0][0][0][0][0];
  int     ****i4p;  //for non-RD-opt. mode
  imgpel  (*mb_pred)[16] = img->mb_pred[0];
  int    is_cavlc = (img->currentSlice->symbol_mode == CAVLC);

  int tmp_8x8_flag, tmp_no_mbpart;
  // Fast Mode Decision
  short bipred_me, inter_skip = 0;


  if(params->SearchMode == UM_HEX)
  {
    UMHEX_decide_intrabk_SAD();
  }
  else if (params->SearchMode == UM_HEX_SIMPLE)
  {
    smpUMHEX_decide_intrabk_SAD();
  }

  intra |= RandomIntra (img->current_mb_nr);    // Forced Pseudo-Random Intra

  //===== Setup Macroblock encoding parameters =====
  init_enc_mb_params(currMB, &enc_mb, intra, bslice);

  // reset chroma intra predictor to default
  currMB->c_ipred_mode = DC_PRED_8;

  //=====   S T O R E   C O D I N G   S T A T E   =====
  //---------------------------------------------------
  store_coding_state (currMB, cs_cm);

  if (!intra)
  {
    //===== set direct motion vectors =====
    best_mode = 1;
    if (bslice)
    {
      Get_Direct_Motion_Vectors (currMB);
    }

    if (params->CtxAdptLagrangeMult == 1)
    {
      get_initial_mb16x16_cost(currMB);
    }

    //===== MOTION ESTIMATION FOR 16x16, 16x8, 8x16 BLOCKS =====
    for (min_cost=INT_MAX, mode=1; mode<4; mode++)
    {
      bipred_me = 0;
      b8x8info->bipred8x8me[mode][0] = 0;
      if (enc_mb.valid[mode] && !inter_skip)
      {
        for (cost=0, block=0; block<(mode==1?1:2); block++)
        {
          update_lambda_costs(&enc_mb, lambda_mf);
          PartitionMotionSearch (currMB, mode, block, lambda_mf);

          //--- set 4x4 block indizes (for getting MV) ---
          j = (block==1 && mode==2 ? 2 : 0);
          i = (block==1 && mode==3 ? 2 : 0);

          //--- get cost and reference frame for List 0 prediction ---
          bmcost[LIST_0] = INT_MAX;
          list_prediction_cost(currMB, LIST_0, block, mode, &enc_mb, bmcost, best_ref);

          if (bslice)
          {
            //--- get cost and reference frame for List 1 prediction ---
            bmcost[LIST_1] = INT_MAX;
            list_prediction_cost(currMB, LIST_1, block, mode, &enc_mb, bmcost, best_ref);

            // Compute bipredictive cost between best list 0 and best list 1 references
            list_prediction_cost(currMB, BI_PRED, block, mode, &enc_mb, bmcost, best_ref);

            // currently Bi prediction ME is only supported for modes 1, 2, 3
            if (is_bipred_enabled(mode))
            {
              list_prediction_cost(currMB, BI_PRED_L0, block, mode, &enc_mb, bmcost, 0);
              list_prediction_cost(currMB, BI_PRED_L1, block, mode, &enc_mb, bmcost, 0);
            }
            else
            {
              bmcost[BI_PRED_L0] = INT_MAX;
              bmcost[BI_PRED_L1] = INT_MAX;
            }

            // Determine prediction list based on mode cost
            determine_prediction_list(mode, bmcost, best_ref, &best_pdir, &cost, &bipred_me);
          }
          else // if (bslice)
          {
            best_pdir  = 0;
            cost      += bmcost[LIST_0];
          }

          assign_enc_picture_params(mode, best_pdir, block, enc_mb.list_offset[LIST_0], best_ref[LIST_0], best_ref[LIST_1], bslice, bipred_me);          
          //----- set reference frame and direction parameters -----
          set_block8x8_info(b8x8info, mode, block, best_ref, best_pdir, bipred_me);

          //--- set reference frames and motion vectors ---
          if (mode>1 && block==0)
            SetRefAndMotionVectors (currMB, block, mode, best_pdir, best_ref[LIST_0], best_ref[LIST_1], bipred_me);
        } // for (block=0; block<(mode==1?1:2); block++)

        currMB->luma_transform_size_8x8_flag = 0;
        if (params->Transform8x8Mode) //for inter rd-off, set 8x8 to do 8x8 transform
        {
          SetModesAndRefframeForBlocks(currMB, mode);
          currMB->luma_transform_size_8x8_flag = TransformDecision(currMB, -1, &cost);
        }

        if ((!inter_skip) && (cost < min_cost))
        {
          best_mode = (short) mode;
          min_cost  = cost;
          best_transform_flag = currMB->luma_transform_size_8x8_flag;

          if (params->CtxAdptLagrangeMult == 1)
          {
            adjust_mb16x16_cost(cost);
          }
        }
      } // if (enc_mb.valid[mode])
    } // for (mode=1; mode<4; mode++)

    if ((!inter_skip) && enc_mb.valid[P8x8])
    {
      giRDOpt_B8OnlyFlag = 1;

      tr8x8.mb_p8x8_cost = INT_MAX;
      tr4x4.mb_p8x8_cost = INT_MAX;
      //===== store coding state of macroblock =====
      store_coding_state (currMB, cs_mb);

      currMB->all_blk_8x8 = -1;

      if (params->Transform8x8Mode)
      {
        tr8x8.mb_p8x8_cost = 0;
        //===========================================================
        // Check 8x8 partition with transform size 8x8
        //===========================================================
        //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
        for (cost_direct=cbp8x8=cbp_blk8x8=cnt_nonz_8x8=0, block = 0; block < 4; block++)
        {
          submacroblock_mode_decision(&enc_mb, &tr8x8, currMB, cofAC8x8ts[0][block], cofAC8x8ts[1][block], cofAC8x8ts[2][block],
            &have_direct, bslice, block, &cost_direct, &cost, &cost8x8_direct, 1, is_cavlc);
          set_subblock8x8_info(b8x8info, P8x8, block, &tr8x8);
        }

        // following params could be added in RD_8x8DATA structure
        cbp8_8x8ts      = cbp8x8;
        cbp_blk8_8x8ts  = cbp_blk8x8;
        cnt_nonz8_8x8ts = cnt_nonz_8x8;
        currMB->luma_transform_size_8x8_flag = 0; //switch to 4x4 transform size

        //--- re-set coding state (as it was before 8x8 block coding) ---
        //reset_coding_state (currMB, cs_mb);
      }// if (params->Transform8x8Mode)


      if (params->Transform8x8Mode != 2)
      {
        tr4x4.mb_p8x8_cost = 0;
        //=================================================================
        // Check 8x8, 8x4, 4x8 and 4x4 partitions with transform size 4x4
        //=================================================================
        //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
        for (cost_direct=cbp8x8=cbp_blk8x8=cnt_nonz_8x8=0, block=0; block<4; block++)
        {
          submacroblock_mode_decision(&enc_mb, &tr4x4, currMB, cofAC8x8[block], cofAC8x8CbCr[0][block], cofAC8x8CbCr[1][block],
            &have_direct, bslice, block, &cost_direct, &cost, &cost8x8_direct, 0, is_cavlc);
          set_subblock8x8_info(b8x8info, P8x8, block, &tr4x4);
        }
        //--- re-set coding state (as it was before 8x8 block coding) ---
        // reset_coding_state (currMB, cs_mb);
      }// if (params->Transform8x8Mode != 2)

      //--- re-set coding state (as it was before 8x8 block coding) ---
      reset_coding_state (currMB, cs_mb);


      // This is not enabled yet since mpr has reverse order.
      if (params->RCEnable)
        rc_store_diff(img->opix_x, img->opix_y, mb_pred);

      //check cost for P8x8 for non-rdopt mode
      if (tr4x4.mb_p8x8_cost < min_cost || tr8x8.mb_p8x8_cost < min_cost)
      {
        best_mode = P8x8;

        if (img->AdaptiveRounding)
          store_adaptive_rounding_parameters_luma (currMB, best_mode);		

        if (params->Transform8x8Mode == 2)
        {
          min_cost = tr8x8.mb_p8x8_cost;
          currMB->luma_transform_size_8x8_flag=1;
        }
        else if (params->Transform8x8Mode)
        {
          if (tr8x8.mb_p8x8_cost < tr4x4.mb_p8x8_cost)
          {
            min_cost = tr8x8.mb_p8x8_cost;
            currMB->luma_transform_size_8x8_flag=1;
          }
          else if(tr4x4.mb_p8x8_cost < tr8x8.mb_p8x8_cost)
          {
            min_cost = tr4x4.mb_p8x8_cost;
            currMB->luma_transform_size_8x8_flag=0;
          }
          else
          {
            if (GetBestTransformP8x8() == 0)
            {
              min_cost = tr4x4.mb_p8x8_cost;
              currMB->luma_transform_size_8x8_flag=0;
            }
            else
            {
              min_cost = tr8x8.mb_p8x8_cost;
              currMB->luma_transform_size_8x8_flag=1;
            }
          }
        }
        else
        {
          min_cost = tr4x4.mb_p8x8_cost;
          currMB->luma_transform_size_8x8_flag=0;
        }
      }// if ((tr4x4.mb_p8x8_cost < min_cost || tr8x8.mb_p8x8_cost < min_cost))
      giRDOpt_B8OnlyFlag = 0;
    }
    else // if (enc_mb.valid[P8x8])
    {
      tr4x4.mb_p8x8_cost = INT_MAX;
    }

    // Find a motion vector for the Skip mode
    if(pslice)
      FindSkipModeMotionVector (currMB);
  }
  else // if (!intra)
  {
    min_cost = INT_MAX;
  }

  //========= C H O O S E   B E S T   M A C R O B L O C K   M O D E =========
  //-------------------------------------------------------------------------
  tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;  //save 8x8_flag
  tmp_no_mbpart = currMB->NoMbPartLessThan8x8Flag;      //save no-part-less
  if ((img->yuv_format != YUV400) && (img->yuv_format != YUV444))
    // precompute all chroma intra prediction modes
    IntraChromaPrediction(currMB, NULL, NULL, NULL);

  if (enc_mb.valid[0] && bslice) // check DIRECT MODE
  {
    if(have_direct)
    {
      switch(params->Transform8x8Mode)
      {
      case 1: // Mixture of 8x8 & 4x4 transform
        cost = ((cost8x8_direct < cost_direct) || !(enc_mb.valid[5] && enc_mb.valid[6] && enc_mb.valid[7]))
          ? cost8x8_direct : cost_direct;
        break;
      case 2: // 8x8 Transform only
        cost = cost8x8_direct;
        break;
      default: // 4x4 Transform only
        cost = cost_direct;
        break;
      }
    }
    else
    { //!have_direct
      cost = GetDirectCostMB (currMB, bslice);
    }
    if (cost!=INT_MAX)
    {
      cost -= (int)floor(16*enc_mb.lambda_md+0.4999);
    }

    if (cost <= min_cost)
    {
      if(active_sps->direct_8x8_inference_flag && params->Transform8x8Mode)
      {