t264enc.c

T264是中国的视频编码自由组织合力开发的264编解码程序

/*****************************************************************************
 *
 *  T264 AVC CODEC
 *
 *  Copyright(C) 2004-2005 llcc <[e-mail]>
 *               2004-2005 visionany <[e-mail]>
 *
 *  This program is free software ; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation ; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY ; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program ; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 *
 ****************************************************************************/
#include "config.h"
#include "stdio.h"
#ifndef CHIP_DM642
#include "memory.h"
#endif

#include "T264.h"
#include "utility.h"
#include "intra.h"
#include "cavlc.h"
#include "inter.h"
#include "inter_b.h"
#include "interpolate.h"
#include "estimation.h"
#include "deblock.h"
#include "ratecontrol.h"
#include "stat.h"
#ifndef CHIP_DM642
#include "sse2.h"
#endif
#include "math.h"
#include "dct.h"
#include "predict.h"
#include "bitstream.h"
#include "rbsp.h"
#include "assert.h"
#include "typedecision.h"
//for CABAC
#include "cabac_engine.h"
#include "cabac.h"

const int32_t chroma_qp[] =
{
    0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 
    11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 
    21, 22, 23, 24, 25, 26, 27, 28, 29,
    29, 30, 31, 32, 32, 33, 34, 34, 35, 35,
    36, 36, 37, 37, 37, 38, 38, 38, 39, 39, 39, 39
};

//! convert from H.263 QP to H.26L quant given by: quant=pow(2,QP/6)
static const int32_t qp_cost[52]=
{
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 2, 2, 2, 2,
    3, 3, 3, 4, 4, 4, 5, 6,
    6, 7, 8, 9,10,11,13,14,
    16,18,20,23,25,29,32,36,
    40,45,51,57,64,72,81,91
};

void
T264_mb_load_context(T264_t* t, int32_t mb_y, int32_t mb_x)
{
    int32_t qpc;
    int32_t i, j;

    t->mb.mb_x = mb_x;
    t->mb.mb_y = mb_y;
    t->mb.mb_xy = t->mb.mb_y * t->mb_stride + t->mb.mb_x;
    t->mb.mb_neighbour = 0;
    if (mb_x != 0)
        t->mb.mb_neighbour |= MB_LEFT;
    if (mb_y != 0)
    {
        t->mb.mb_neighbour |= MB_TOP;
        //if (mb_x != t->mb_stride - 1) //just tell MB is not the top-most
            t->mb.mb_neighbour |= MB_TOPRIGHT;
    }
    t->mb.src_y = t->cur.Y[0]  + (mb_y << 4) * t->stride    + (mb_x << 4);
    t->mb.dst_y  = t->rec->Y[0] + (mb_y << 4) * t->edged_stride    + (mb_x << 4);
    t->mb.src_u = t->cur.U     + (mb_y << 3) * t->stride_uv + (mb_x << 3);
    t->mb.dst_u = t->rec->U + (mb_y << 3) * t->edged_stride_uv + (mb_x << 3);
    t->mb.src_v = t->cur.V     + (mb_y << 3) * t->stride_uv + (mb_x << 3);
    t->mb.dst_v = t->rec->V + (mb_y << 3) * t->edged_stride_uv + (mb_x << 3);

    t->mb.mb_qp_delta = 0;
    /* t->ps.chroma_qp_index_offset maybe modify in ratecontrol */
    qpc = clip3(t->ps.chroma_qp_index_offset + t->qp_y, 0, 51);
    t->qp_uv = chroma_qp[qpc];
    t->mb.lambda = qp_cost[t->qp_y];

    t->mb.context = &t->rec->mb[t->mb.mb_xy];

	/*for CABAC, init the mvd_ref */
	memset(&(t->mb.mvd_ref[0][0][0]), 0, sizeof(t->mb.mvd_ref));
#define INITINVALIDVEC(vec) vec.refno = -2; vec.x = vec.y = 0;
    INITINVALIDVEC(t->mb.vec_ref[0].vec[0]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4].vec[0]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4 + 8].vec[0]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4 + 16].vec[0]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4 + 24].vec[0]);
    INITINVALIDVEC(t->mb.vec_ref[0].vec[1]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4].vec[1]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4 + 8].vec[1]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4 + 16].vec[1]);
    INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 4 + 24].vec[1]);
    
    t->mb.vec_ref[0].part = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4].part      = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4 + 8].part  = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4 + 16].part = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4 + 24].part = -1;
    
    t->mb.vec_ref[0].subpart = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4].subpart      = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4 + 8].subpart  = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4 + 16].subpart = -1;
    t->mb.vec_ref[IPM_LUMA - 8 + 4 + 24].subpart = -1;
	
    memset(t->mb.submb_part, -1, sizeof(uint8_t) * 16);//t->mb.submb_part));
    t->mb.mb_part = -1;
    for(i = 0 ; i < 2 ; i ++)
    {
        for(j = 0 ; j < 16 ; j ++)
        {
            INITINVALIDVEC(t->mb.vec[i][j]);
        }
    }
    t->mb.sad_ref[0] = t->mb.sad_ref[1] = t->mb.sad_ref[2] = 0;

    //intra_4x4 prediction modes and non-zero counts
	if( mb_y > 0 )
    {
        int16_t top_xy  = t->mb.mb_xy - t->mb_stride;
        /* intra 4x4 pred mode layout
         	? x x x x
            x
            x
            x
            x
         */        
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 0] = t->rec->mb[top_xy].mode_i4x4[10];
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 1] = t->rec->mb[top_xy].mode_i4x4[11];
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 2] = t->rec->mb[top_xy].mode_i4x4[14];
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 3] = t->rec->mb[top_xy].mode_i4x4[15];

        t->mb.vec_ref[IPM_LUMA - 8 + 0].vec[0] = t->rec->mb[top_xy].vec[0][12];
        t->mb.vec_ref[IPM_LUMA - 8 + 1].vec[0] = t->rec->mb[top_xy].vec[0][13];
        t->mb.vec_ref[IPM_LUMA - 8 + 2].vec[0] = t->rec->mb[top_xy].vec[0][14];
        t->mb.vec_ref[IPM_LUMA - 8 + 3].vec[0] = t->rec->mb[top_xy].vec[0][15];
        t->mb.vec_ref[IPM_LUMA - 8 + 0].vec[1] = t->rec->mb[top_xy].vec[1][12];
        t->mb.vec_ref[IPM_LUMA - 8 + 1].vec[1] = t->rec->mb[top_xy].vec[1][13];
        t->mb.vec_ref[IPM_LUMA - 8 + 2].vec[1] = t->rec->mb[top_xy].vec[1][14];
        t->mb.vec_ref[IPM_LUMA - 8 + 3].vec[1] = t->rec->mb[top_xy].vec[1][15];

        t->mb.vec_ref[IPM_LUMA - 8 + 0].part = 
        t->mb.vec_ref[IPM_LUMA - 8 + 1].part = 
        t->mb.vec_ref[IPM_LUMA - 8 + 2].part = 
        t->mb.vec_ref[IPM_LUMA - 8 + 3].part = t->rec->mb[top_xy].mb_part;

        t->mb.vec_ref[IPM_LUMA - 8 + 0].subpart = t->rec->mb[top_xy].submb_part[12];
        t->mb.vec_ref[IPM_LUMA - 8 + 1].subpart = t->rec->mb[top_xy].submb_part[13];
        t->mb.vec_ref[IPM_LUMA - 8 + 2].subpart = t->rec->mb[top_xy].submb_part[14];
        t->mb.vec_ref[IPM_LUMA - 8 + 3].subpart = t->rec->mb[top_xy].submb_part[15];

        t->mb.sad_ref[1] = t->rec->mb[top_xy].sad;

		//for CABAC, load mvd
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 0][0] = t->rec->mb[top_xy].mvd[0][12][0];
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 0][1] = t->rec->mb[top_xy].mvd[0][12][1];

		t->mb.mvd_ref[0][IPM_LUMA - 8 + 1][0] = t->rec->mb[top_xy].mvd[0][13][0];
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 1][1] = t->rec->mb[top_xy].mvd[0][13][1];

		t->mb.mvd_ref[0][IPM_LUMA - 8 + 2][0] = t->rec->mb[top_xy].mvd[0][14][0];
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 2][1] = t->rec->mb[top_xy].mvd[0][14][1];

		t->mb.mvd_ref[0][IPM_LUMA - 8 + 3][0] = t->rec->mb[top_xy].mvd[0][15][0];
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 3][1] = t->rec->mb[top_xy].mvd[0][15][1];

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 0][0] = t->rec->mb[top_xy].mvd[1][12][0];
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 0][1] = t->rec->mb[top_xy].mvd[1][12][1];

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 1][0] = t->rec->mb[top_xy].mvd[1][13][0];
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 1][1] = t->rec->mb[top_xy].mvd[1][13][1];

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 2][0] = t->rec->mb[top_xy].mvd[1][14][0];
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 2][1] = t->rec->mb[top_xy].mvd[1][14][1];

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 3][0] = t->rec->mb[top_xy].mvd[1][15][0];
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 3][1] = t->rec->mb[top_xy].mvd[1][15][1];

        if (mb_x != t->mb_stride - 1)
        {
            int32_t righttop_xy = top_xy + 1;
            t->mb.vec_ref[IPM_LUMA - 8 + 4].vec[0]     = t->rec->mb[righttop_xy].vec[0][12];
            t->mb.vec_ref[IPM_LUMA - 8 + 4].vec[1]     = t->rec->mb[righttop_xy].vec[1][12];
            t->mb.vec_ref[IPM_LUMA - 8 + 4].part    = t->rec->mb[righttop_xy].mb_part;
            t->mb.vec_ref[IPM_LUMA - 8 + 4].subpart = t->rec->mb[righttop_xy].submb_part[12];
            t->mb.sad_ref[2] = t->rec->mb[righttop_xy].sad;
        }
        /* nnz layout:
          ? x x x x ? x x
          x         x
          x         x
          x         ? x x
          x         x
                    x
         */
        t->mb.nnz_ref[NNZ_LUMA - 8 + 0] = t->rec->mb[top_xy].nnz[12];
        t->mb.nnz_ref[NNZ_LUMA - 8 + 1] = t->rec->mb[top_xy].nnz[13];
        t->mb.nnz_ref[NNZ_LUMA - 8 + 2] = t->rec->mb[top_xy].nnz[14];
        t->mb.nnz_ref[NNZ_LUMA - 8 + 3] = t->rec->mb[top_xy].nnz[15];

        t->mb.nnz_ref[NNZ_CHROMA0 - 8 + 0] = t->rec->mb[top_xy].nnz[18];
        t->mb.nnz_ref[NNZ_CHROMA0 - 8 + 1] = t->rec->mb[top_xy].nnz[19];
        t->mb.nnz_ref[NNZ_CHROMA1 - 8 + 0] = t->rec->mb[top_xy].nnz[22];
        t->mb.nnz_ref[NNZ_CHROMA1 - 8 + 1] = t->rec->mb[top_xy].nnz[23];
    }
    else
    {
        /* load intra4x4 */
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 0] = 
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 1] = 
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 2] = 
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 8 + 3] = -1;

        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 0].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 1].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 2].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 3].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 0].vec[1]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 1].vec[1]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 2].vec[1]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 8 + 3].vec[1]);

        t->mb.vec_ref[IPM_LUMA - 8 + 0].part = 
        t->mb.vec_ref[IPM_LUMA - 8 + 1].part = 
        t->mb.vec_ref[IPM_LUMA - 8 + 2].part = 
        t->mb.vec_ref[IPM_LUMA - 8 + 3].part = -1;

        t->mb.vec_ref[IPM_LUMA - 8 + 0].subpart = 
        t->mb.vec_ref[IPM_LUMA - 8 + 1].subpart = 
        t->mb.vec_ref[IPM_LUMA - 8 + 2].subpart = 
        t->mb.vec_ref[IPM_LUMA - 8 + 3].subpart = -1;

		//for CABAC, load mvd
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 0][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 0][1] = 0;

		t->mb.mvd_ref[0][IPM_LUMA - 8 + 1][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 1][1] = 0;

		t->mb.mvd_ref[0][IPM_LUMA - 8 + 2][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 2][1] = 0;

		t->mb.mvd_ref[0][IPM_LUMA - 8 + 3][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 8 + 3][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 0][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 0][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 1][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 1][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 2][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 2][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 8 + 3][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 8 + 3][1] = 0;

        t->mb.nnz_ref[NNZ_LUMA - 8 + 0] =
        t->mb.nnz_ref[NNZ_LUMA - 8 + 1] =
        t->mb.nnz_ref[NNZ_LUMA - 8 + 2] =
        t->mb.nnz_ref[NNZ_LUMA - 8 + 3] = 0x80;

        t->mb.nnz_ref[NNZ_CHROMA0 - 8 + 0] =
        t->mb.nnz_ref[NNZ_CHROMA0 - 8 + 1] =
        t->mb.nnz_ref[NNZ_CHROMA1 - 8 + 0] =
        t->mb.nnz_ref[NNZ_CHROMA1 - 8 + 1] = 0x80;
    }

    if( mb_x > 0 )
    {
        int16_t left_xy  = t->mb.mb_xy - 1;

        /* load intra4x4 */
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 0] = t->rec->mb[left_xy].mode_i4x4[5];
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 8] = t->rec->mb[left_xy].mode_i4x4[7];
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 16] = t->rec->mb[left_xy].mode_i4x4[13];
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 24] = t->rec->mb[left_xy].mode_i4x4[15];

        t->mb.vec_ref[IPM_LUMA - 1 + 0].vec[0] = t->rec->mb[left_xy].vec[0][3];
        t->mb.vec_ref[IPM_LUMA - 1 + 8].vec[0] = t->rec->mb[left_xy].vec[0][7];
        t->mb.vec_ref[IPM_LUMA - 1 + 16].vec[0] = t->rec->mb[left_xy].vec[0][11];
        t->mb.vec_ref[IPM_LUMA - 1 + 24].vec[0] = t->rec->mb[left_xy].vec[0][15];
        t->mb.vec_ref[IPM_LUMA - 1 + 0].vec[1] = t->rec->mb[left_xy].vec[1][3];
        t->mb.vec_ref[IPM_LUMA - 1 + 8].vec[1] = t->rec->mb[left_xy].vec[1][7];
        t->mb.vec_ref[IPM_LUMA - 1 + 16].vec[1] = t->rec->mb[left_xy].vec[1][11];
        t->mb.vec_ref[IPM_LUMA - 1 + 24].vec[1] = t->rec->mb[left_xy].vec[1][15];

        t->mb.vec_ref[IPM_LUMA - 1 + 0].part = 
        t->mb.vec_ref[IPM_LUMA - 1 + 8].part = 
        t->mb.vec_ref[IPM_LUMA - 1 + 16].part =
        t->mb.vec_ref[IPM_LUMA - 1 + 24].part = t->rec->mb[left_xy].mb_part;

        t->mb.vec_ref[IPM_LUMA - 8 + 0].subpart = t->rec->mb[left_xy].submb_part[3];
        t->mb.vec_ref[IPM_LUMA - 8 + 8].subpart = t->rec->mb[left_xy].submb_part[7];
        t->mb.vec_ref[IPM_LUMA - 8 + 16].subpart = t->rec->mb[left_xy].submb_part[11];
        t->mb.vec_ref[IPM_LUMA - 8 + 24].subpart = t->rec->mb[left_xy].submb_part[15];

        t->mb.sad_ref[0] = t->rec->mb[left_xy].sad;

		//for CABAC, load mvd
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 0][0] = t->rec->mb[left_xy].mvd[0][3][0];
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 0][1] = t->rec->mb[left_xy].mvd[0][3][1];

		t->mb.mvd_ref[0][IPM_LUMA - 1 + 8][0] = t->rec->mb[left_xy].mvd[0][7][0];
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 8][1] = t->rec->mb[left_xy].mvd[0][7][1];
		
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 16][0] = t->rec->mb[left_xy].mvd[0][11][0];
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 16][1] = t->rec->mb[left_xy].mvd[0][11][1];

		t->mb.mvd_ref[0][IPM_LUMA - 1 + 24][0] = t->rec->mb[left_xy].mvd[0][15][0];
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 24][1] = t->rec->mb[left_xy].mvd[0][15][1];

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 0][0] = t->rec->mb[left_xy].mvd[1][3][0];
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 0][1] = t->rec->mb[left_xy].mvd[1][3][1];

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 8][0] = t->rec->mb[left_xy].mvd[1][7][0];
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 8][1] = t->rec->mb[left_xy].mvd[1][7][1];

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 16][0] = t->rec->mb[left_xy].mvd[1][11][0];
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 16][1] = t->rec->mb[left_xy].mvd[1][11][1];

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 24][0] = t->rec->mb[left_xy].mvd[1][15][0];
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 24][1] = t->rec->mb[left_xy].mvd[1][15][1];


        /* load non_zero_count */
        t->mb.nnz_ref[NNZ_LUMA - 1 + 0] = t->rec->mb[left_xy].nnz[3];
        t->mb.nnz_ref[NNZ_LUMA - 1 + 8] = t->rec->mb[left_xy].nnz[7];
        t->mb.nnz_ref[NNZ_LUMA - 1 + 16] = t->rec->mb[left_xy].nnz[11];
        t->mb.nnz_ref[NNZ_LUMA - 1 + 24] = t->rec->mb[left_xy].nnz[15];

        t->mb.nnz_ref[NNZ_CHROMA0 - 1 + 0] = t->rec->mb[left_xy].nnz[17];
        t->mb.nnz_ref[NNZ_CHROMA0 - 1 + 8] = t->rec->mb[left_xy].nnz[19];
        t->mb.nnz_ref[NNZ_CHROMA1 - 1 + 0] = t->rec->mb[left_xy].nnz[21];
        t->mb.nnz_ref[NNZ_CHROMA1 - 1 + 8] = t->rec->mb[left_xy].nnz[23];
    }
    else
    {
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 0]  = 
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 8]  = 
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 16] =
        t->mb.i4x4_pred_mode_ref[IPM_LUMA - 1 + 24] = -1;

        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 0].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 8].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 16].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 24].vec[0]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 0].vec[1]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 8].vec[1]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 16].vec[1]);
        INITINVALIDVEC(t->mb.vec_ref[IPM_LUMA - 1 + 24].vec[1]);

        t->mb.vec_ref[IPM_LUMA - 1 + 0].part  = 
        t->mb.vec_ref[IPM_LUMA - 1 + 8].part  = 
        t->mb.vec_ref[IPM_LUMA - 1 + 16].part =
        t->mb.vec_ref[IPM_LUMA - 1 + 24].part = -1;

        t->mb.vec_ref[IPM_LUMA - 1 + 0].subpart  = 
        t->mb.vec_ref[IPM_LUMA - 1 + 8].subpart  = 
        t->mb.vec_ref[IPM_LUMA - 1 + 16].subpart =
        t->mb.vec_ref[IPM_LUMA - 1 + 24].subpart = -1;

		//for CABAC, load mvd
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 0][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 0][1] = 0;

		t->mb.mvd_ref[0][IPM_LUMA - 1 + 8][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 8][1] = 0;

		t->mb.mvd_ref[0][IPM_LUMA - 1 + 16][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 16][1] = 0;

		t->mb.mvd_ref[0][IPM_LUMA - 1 + 24][0] = 0;
		t->mb.mvd_ref[0][IPM_LUMA - 1 + 24][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 0][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 0][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 8][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 8][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 16][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 16][1] = 0;

		t->mb.mvd_ref[1][IPM_LUMA - 1 + 24][0] = 0;
		t->mb.mvd_ref[1][IPM_LUMA - 1 + 24][1] = 0;

        t->mb.nnz_ref[NNZ_LUMA - 1 + 0]  =