analyse.c

绝对好的源码

/*****************************************************************************
 * analyse.c: h264 encoder library
 *****************************************************************************
 * Copyright (C) 2003 x264 project
 * $Id: analyse.c,v 1.1 2004/06/03 19:27:08 fenrir Exp $
 *
 * Authors: Laurent Aimar <[e-mail]>
 *          Loren Merritt <[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, USA.
 *****************************************************************************/

#include <stdio.h>
#include <string.h>
#include <math.h>
#include <limits.h>

#include "common/common.h"
#include "macroblock.h"

#include "ratecontrol.h"
#include "analyse.h"
#include "rdo.c"


static void x264_analyse_update_cache( x264_t *h, x264_mb_analysis_t *a );

/* initialize an array of lambda*nbits for all possible mvs */
static void x264_mb_analyse_load_costs( x264_t *h, x264_mb_analysis_t *a )
{
    static int16_t *p_cost_mv[52];

    if( !p_cost_mv[a->i_qp] )
    {
        /* could be faster, but isn't called many times */
        /* factor of 4 from qpel, 2 from sign, and 2 because mv can be opposite from mvp */
        int i;
        p_cost_mv[a->i_qp] = x264_malloc( (4*4*2048 + 1) * sizeof(int16_t) );
        p_cost_mv[a->i_qp] += 2*4*2048;
        for( i = 0; i <= 2*4*2048; i++ )
        {
            p_cost_mv[a->i_qp][-i] =
            p_cost_mv[a->i_qp][i]  = a->i_lambda * bs_size_se( i );
        }
    }

    a->p_cost_mv = p_cost_mv[a->i_qp];
}

static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int i_qp )
{
    memset( a, 0, sizeof( x264_mb_analysis_t ) );

    /* conduct the analysis using this lamda and QP */
    a->i_qp = h->mb.i_qp = i_qp;
    a->i_lambda = i_qp0_cost_table[i_qp];
    a->i_lambda2 = i_qp0_cost2_table[i_qp];
    a->b_mbrd = h->param.analyse.i_subpel_refine >= 6 &&
                ( h->sh.i_type != SLICE_TYPE_B || h->param.analyse.b_bframe_rdo );

    h->mb.i_me_method = h->param.analyse.i_me_method;
    h->mb.i_subpel_refine = h->param.analyse.i_subpel_refine;
    h->mb.b_chroma_me = h->param.analyse.b_chroma_me && h->sh.i_type == SLICE_TYPE_P
                        && h->mb.i_subpel_refine >= 5;
    h->mb.b_trellis = h->param.analyse.i_trellis > 1 && a->b_mbrd;
    h->mb.b_transform_8x8 = 0;
    h->mb.b_noise_reduction = 0;

    /* I: Intra part */
    a->i_satd_i16x16 =
    a->i_satd_i8x8   =
    a->i_satd_i4x4   =
    a->i_satd_i8x8chroma = COST_MAX;

    a->b_fast_intra = 0;

    /* II: Inter part P/B frame */
    if( h->sh.i_type != SLICE_TYPE_I )
    {
        int i;
        int i_fmv_range = h->param.analyse.i_mv_range - 16;

        /* Calculate max allowed MV range */
#define CLIP_FMV(mv) x264_clip3( mv, -i_fmv_range, i_fmv_range )
        h->mb.mv_min[0] = 4*( -16*h->mb.i_mb_x - 24 );
        h->mb.mv_max[0] = 4*( 16*( h->sps->i_mb_width - h->mb.i_mb_x - 1 ) + 24 );
        h->mb.mv_min_fpel[0] = CLIP_FMV( -16*h->mb.i_mb_x - 8 );
        h->mb.mv_max_fpel[0] = CLIP_FMV( 16*( h->sps->i_mb_width - h->mb.i_mb_x - 1 ) + 8 );
        h->mb.mv_min_spel[0] = 4*( h->mb.mv_min_fpel[0] - 16 );
        h->mb.mv_max_spel[0] = 4*( h->mb.mv_max_fpel[0] + 16 );
        if( h->mb.i_mb_x == 0)
        {
            h->mb.mv_min[1] = 4*( -16*h->mb.i_mb_y - 24 );
            h->mb.mv_max[1] = 4*( 16*( h->sps->i_mb_height - h->mb.i_mb_y - 1 ) + 24 );
            h->mb.mv_min_fpel[1] = CLIP_FMV( -16*h->mb.i_mb_y - 8 );
            h->mb.mv_max_fpel[1] = CLIP_FMV( 16*( h->sps->i_mb_height - h->mb.i_mb_y - 1 ) + 8 );
            h->mb.mv_min_spel[1] = 4*( h->mb.mv_min_fpel[1] - 16 );
            h->mb.mv_max_spel[1] = 4*( h->mb.mv_max_fpel[1] + 16 );
        }
#undef CLIP_FMV

        a->l0.me16x16.cost =
        a->l0.i_rd16x16    =
        a->l0.i_cost8x8    = COST_MAX;

        for( i = 0; i < 4; i++ )
        {
            a->l0.i_cost4x4[i] =
            a->l0.i_cost8x4[i] =
            a->l0.i_cost4x8[i] = COST_MAX;
        }

        a->l0.i_cost16x8   =
        a->l0.i_cost8x16   = COST_MAX;
        if( h->sh.i_type == SLICE_TYPE_B )
        {
            a->l1.me16x16.cost =
            a->l1.i_rd16x16    =
            a->l1.i_cost8x8    = COST_MAX;

            for( i = 0; i < 4; i++ )
            {
                a->l1.i_cost4x4[i] =
                a->l1.i_cost8x4[i] =
                a->l1.i_cost4x8[i] =
                a->i_cost8x8direct[i] = COST_MAX;
            }

            a->l1.i_cost16x8   =
            a->l1.i_cost8x16   =
            a->i_rd16x16bi     =
            a->i_rd16x16direct =
            a->i_rd8x8bi       =
            a->i_rd16x8bi      =
            a->i_rd8x16bi      =
            a->i_cost16x16bi   =
            a->i_cost16x16direct =
            a->i_cost8x8bi     =
            a->i_cost16x8bi    =
            a->i_cost8x16bi    = COST_MAX;
        }

        /* Fast intra decision */
        if( h->mb.i_mb_xy - h->sh.i_first_mb > 4 )
        {
            if(   IS_INTRA( h->mb.i_mb_type_left )
               || IS_INTRA( h->mb.i_mb_type_top )
               || IS_INTRA( h->mb.i_mb_type_topleft )
               || IS_INTRA( h->mb.i_mb_type_topright )
               || (h->sh.i_type == SLICE_TYPE_P && IS_INTRA( h->fref0[0]->mb_type[h->mb.i_mb_xy] ))
               || (h->mb.i_mb_xy - h->sh.i_first_mb < 3*(h->stat.frame.i_mb_count[I_4x4] + h->stat.frame.i_mb_count[I_8x8] + h->stat.frame.i_mb_count[I_16x16])) )
            { /* intra is likely */ }
            else
            {
                a->b_fast_intra = 1;
            }
        }
    }
}



/*
 * Handle intra mb
 */
/* Max = 4 */
static void predict_16x16_mode_available( unsigned int i_neighbour, int *mode, int *pi_count )
{
    if( i_neighbour & MB_TOPLEFT )
    {
        /* top and left avaible */
        *mode++ = I_PRED_16x16_V;
        *mode++ = I_PRED_16x16_H;
        *mode++ = I_PRED_16x16_DC;
        *mode++ = I_PRED_16x16_P;
        *pi_count = 4;
    }
    else if( i_neighbour & MB_LEFT )
    {
        /* left available*/
        *mode++ = I_PRED_16x16_DC_LEFT;
        *mode++ = I_PRED_16x16_H;
        *pi_count = 2;
    }
    else if( i_neighbour & MB_TOP )
    {
        /* top available*/
        *mode++ = I_PRED_16x16_DC_TOP;
        *mode++ = I_PRED_16x16_V;
        *pi_count = 2;
    }
    else
    {
        /* none avaible */
        *mode = I_PRED_16x16_DC_128;
        *pi_count = 1;
    }
}

/* Max = 4 */
static void predict_8x8chroma_mode_available( unsigned int i_neighbour, int *mode, int *pi_count )
{
    if( i_neighbour & MB_TOPLEFT )
    {
        /* top and left avaible */
        *mode++ = I_PRED_CHROMA_V;
        *mode++ = I_PRED_CHROMA_H;
        *mode++ = I_PRED_CHROMA_DC;
        *mode++ = I_PRED_CHROMA_P;
        *pi_count = 4;
    }
    else if( i_neighbour & MB_LEFT )
    {
        /* left available*/
        *mode++ = I_PRED_CHROMA_DC_LEFT;
        *mode++ = I_PRED_CHROMA_H;
        *pi_count = 2;
    }
    else if( i_neighbour & MB_TOP )
    {
        /* top available*/
        *mode++ = I_PRED_CHROMA_DC_TOP;
        *mode++ = I_PRED_CHROMA_V;
        *pi_count = 2;
    }
    else
    {
        /* none avaible */
        *mode = I_PRED_CHROMA_DC_128;
        *pi_count = 1;
    }
}

/* MAX = 9 */
static void predict_4x4_mode_available( unsigned int i_neighbour,
                                        int *mode, int *pi_count )
{
    int b_l = i_neighbour & MB_LEFT;
    int b_t = i_neighbour & MB_TOP;

    if( b_l && b_t )
    {
        *pi_count = 6;
        *mode++ = I_PRED_4x4_DC;
        *mode++ = I_PRED_4x4_H;
        *mode++ = I_PRED_4x4_V;
        *mode++ = I_PRED_4x4_DDL;
        if( i_neighbour & MB_TOPLEFT )
        {
            *mode++ = I_PRED_4x4_DDR;
            *mode++ = I_PRED_4x4_VR;
            *mode++ = I_PRED_4x4_HD;
            *pi_count += 3;
        }
        *mode++ = I_PRED_4x4_VL;
        *mode++ = I_PRED_4x4_HU;
    }
    else if( b_l )
    {
        *mode++ = I_PRED_4x4_DC_LEFT;
        *mode++ = I_PRED_4x4_H;
        *mode++ = I_PRED_4x4_HU;
        *pi_count = 3;
    }
    else if( b_t )
    {
        *mode++ = I_PRED_4x4_DC_TOP;
        *mode++ = I_PRED_4x4_V;
        *mode++ = I_PRED_4x4_DDL;
        *mode++ = I_PRED_4x4_VL;
        *pi_count = 4;
    }
    else
    {
        *mode++ = I_PRED_4x4_DC_128;
        *pi_count = 1;
    }
}

static void x264_mb_analyse_intra_chroma( x264_t *h, x264_mb_analysis_t *a )
{
    int i;

    int i_max;
    int predict_mode[9];

    uint8_t *p_dstc[2], *p_srcc[2];

    if( a->i_satd_i8x8chroma < COST_MAX )
        return;

    /* 8x8 prediction selection for chroma */
    p_dstc[0] = h->mb.pic.p_fdec[1];
    p_dstc[1] = h->mb.pic.p_fdec[2];
    p_srcc[0] = h->mb.pic.p_fenc[1];
    p_srcc[1] = h->mb.pic.p_fenc[2];

    predict_8x8chroma_mode_available( h->mb.i_neighbour, predict_mode, &i_max );
    a->i_satd_i8x8chroma = COST_MAX;
    if( i_max == 4 && h->pixf.intra_satd_x3_8x8c && h->pixf.mbcmp[0] == h->pixf.satd[0] )
    {
        int satdu[4], satdv[4];
        h->pixf.intra_satd_x3_8x8c( p_srcc[0], p_dstc[0], satdu );
        h->pixf.intra_satd_x3_8x8c( p_srcc[1], p_dstc[1], satdv );
        h->predict_8x8c[I_PRED_CHROMA_P]( p_dstc[0] );
        h->predict_8x8c[I_PRED_CHROMA_P]( p_dstc[1] );
        satdu[I_PRED_CHROMA_P] =
            h->pixf.mbcmp[PIXEL_8x8]( p_dstc[0], FDEC_STRIDE, p_srcc[0], FENC_STRIDE );
        satdv[I_PRED_CHROMA_P] =
            h->pixf.mbcmp[PIXEL_8x8]( p_dstc[1], FDEC_STRIDE, p_srcc[1], FENC_STRIDE );
        
        for( i=0; i<i_max; i++ )
        {
            int i_mode = predict_mode[i];
            int i_satd = satdu[i_mode] + satdv[i_mode]
                       + a->i_lambda * bs_size_ue(i_mode);
            COPY2_IF_LT( a->i_satd_i8x8chroma, i_satd, a->i_predict8x8chroma, i_mode );
        }
    }
    else
    {
        for( i=0; i<i_max; i++ )
        {
            int i_satd;
            int i_mode = predict_mode[i];

            /* we do the prediction */
            h->predict_8x8c[i_mode]( p_dstc[0] );
            h->predict_8x8c[i_mode]( p_dstc[1] );

            /* we calculate the cost */
            i_satd = h->pixf.mbcmp[PIXEL_8x8]( p_dstc[0], FDEC_STRIDE,
                                               p_srcc[0], FENC_STRIDE ) +
                     h->pixf.mbcmp[PIXEL_8x8]( p_dstc[1], FDEC_STRIDE,
                                               p_srcc[1], FENC_STRIDE ) +
                     a->i_lambda * bs_size_ue( x264_mb_pred_mode8x8c_fix[i_mode] );

            COPY2_IF_LT( a->i_satd_i8x8chroma, i_satd, a->i_predict8x8chroma, i_mode );
        }
    }

    h->mb.i_chroma_pred_mode = a->i_predict8x8chroma;
}

static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_inter )
{
    const unsigned int flags = h->sh.i_type == SLICE_TYPE_I ? h->param.analyse.intra : h->param.analyse.inter;
    uint8_t  *p_src = h->mb.pic.p_fenc[0];
    uint8_t  *p_dst = h->mb.pic.p_fdec[0];

    int i, idx;
    int i_max;
    int predict_mode[9];
    int b_merged_satd = h->pixf.intra_satd_x3_16x16 && h->pixf.mbcmp[0] == h->pixf.satd[0];

    /*---------------- Try all mode and calculate their score ---------------*/