analyse.c.svn-base

H.264 source codes

/*****************************************************************************
 * analyse.c: h264 encoder library
 *****************************************************************************
 * Copyright (C) 2003 Laurent Aimar
 * $Id: analyse.c,v 1.1 2004/06/03 19:27:08 fenrir Exp $
 *
 * Authors: Laurent Aimar <[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 <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <math.h>
#include <limits.h>

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

typedef struct
{
    /* 16x16 */
    int i_ref;
    x264_me_t me16x16;

    /* 8x8 */
    int       i_cost8x8;
    x264_me_t me8x8[4];

    /* Sub 4x4 */
    int       i_cost4x4[4]; /* cost per 8x8 partition */
    x264_me_t me4x4[4][4];

    /* Sub 8x4 */
    int       i_cost8x4[4]; /* cost per 8x8 partition */
    x264_me_t me8x4[4][2];

    /* Sub 4x8 */
    int       i_cost4x8[4]; /* cost per 8x8 partition */
    x264_me_t me4x8[4][4];

    /* 16x8 */
    int       i_cost16x8;
    x264_me_t me16x8[2];

    /* 8x16 */
    int       i_cost8x16;
    x264_me_t me8x16[2];

} x264_mb_analysis_list_t;

typedef struct
{
    /* conduct the analysis using this lamda and QP */
    int i_lambda;
    int i_qp;
    int16_t *p_cost_mv;


    /* I: Intra part */
    /* Take some shortcuts in intra search if intra is deemed unlikely */
    int b_fast_intra;

    /* Luma part 16x16 and 4x4 modes stats */
    int i_sad_i16x16;
    int i_predict16x16;

    int i_sad_i4x4;
    int i_predict4x4[4][4];

    /* Chroma part */
    int i_sad_i8x8;
    int i_predict8x8;

    /* II: Inter part P/B frame */
    x264_mb_analysis_list_t l0;
    x264_mb_analysis_list_t l1;

    int i_cost16x16bi; /* used the same ref and mv as l0 and l1 (at least for now) */
    int i_cost16x16direct;
    int i_cost8x8bi;
    int i_cost8x8direct[4];
    int i_cost16x8bi;
    int i_cost8x16bi;

    int i_mb_partition16x8[2]; /* mb_partition_e */
    int i_mb_partition8x16[2];
    int i_mb_type16x8; /* mb_class_e */
    int i_mb_type8x16;

    int b_direct_available;

} x264_mb_analysis_t;

static const int i_qp0_cost_table[52] = {
   1, 1, 1, 1, 1, 1, 1, 1,  /*  0-7 */
   1, 1, 1, 1,              /*  8-11 */
   1, 1, 1, 1, 2, 2, 2, 2,  /* 12-19 */
   3, 3, 3, 4, 4, 4, 5, 6,  /* 20-27 */
   6, 7, 8, 9,10,11,13,14,  /* 28-35 */
  16,18,20,23,25,29,32,36,  /* 36-43 */
  40,45,51,57,64,72,81,91   /* 44-51 */
};

static const uint8_t block_idx_x[16] = {
    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
};
static const uint8_t block_idx_y[16] = {
    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
};

/* TODO: calculate CABAC costs */
static const int i_mb_b_cost_table[18] = {
    9, 9, 9, 0, 0, 0, 1, 3, 7, 7, 7, 3, 7, 7, 7, 5, 9, 0
};
static const int i_mb_b16x8_cost_table[16] = {
    0, 0, 0, 0, 0, 0, 0, 5, 7, 7, 7, 5, 7, 9, 9, 9
};
static const int i_sub_mb_b_cost_table[13] = {
    7, 5, 5, 3, 7, 5, 7, 3, 7, 7, 7, 5, 1
};
static const int i_sub_mb_p_cost_table[4] = {
    5, 3, 3, 1
};

/* 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*h->param.analyse.i_mv_range + 1) * sizeof(int16_t) );
        p_cost_mv[a->i_qp] += 2*4*h->param.analyse.i_mv_range;
        for( i = 0; i <= 2*4*h->param.analyse.i_mv_range; 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 = i_qp;
    a->i_lambda = i_qp0_cost_table[i_qp];

    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;
    a->b_fast_intra = 0;

    /* I: Intra part */
    a->i_sad_i16x16 =
    a->i_sad_i4x4   =
    a->i_sad_i8x8   = COST_MAX;

    /* 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_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 ) - 8 );
        h->mb.mv_min[0] = 4*( h->mb.mv_min_fpel[0] - 16 );
        h->mb.mv_max[0] = 4*( h->mb.mv_max_fpel[0] + 16 );
        if( h->mb.i_mb_x == 0)
        {
            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 ) - 8 );
            h->mb.mv_min[1] = 4*( h->mb.mv_min_fpel[1] - 16 );
            h->mb.mv_max[1] = 4*( h->mb.mv_max_fpel[1] + 16 );
        }
#undef CLIP_FMV

        a->l0.me16x16.cost =
        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_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_cost16x16bi   =
            a->i_cost16x16direct =
            a->i_cost8x8bi     =
            a->i_cost16x8bi    =
            a->i_cost8x16bi    = COST_MAX;
        }

        /* Fast intra decision */
        if( h->mb.i_mb_xy > 4 )
        {
            const unsigned int i_neighbour = h->mb.i_neighbour;
            if(   ((i_neighbour&MB_LEFT) && IS_INTRA( h->mb.type[h->mb.i_mb_xy - 1] ))
               || ((i_neighbour&MB_TOP) && IS_INTRA( h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride] ))
               || (((i_neighbour&(MB_TOP|MB_LEFT)) == (MB_TOP|MB_LEFT)) && IS_INTRA( h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride-1 ] ))
               || ((i_neighbour&MB_TOPRIGHT) && IS_INTRA( h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride+1 ] ))
               || (h->sh.i_type == SLICE_TYPE_P && IS_INTRA( h->fref0[0]->mb_type[h->mb.i_mb_xy] ))
               || (h->mb.i_mb_xy < 3*(h->stat.frame.i_mb_count[I_4x4] + 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_LEFT|MB_TOP) ) == (MB_LEFT|MB_TOP) )
    {
        /* 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_8x8_mode_available( unsigned int i_neighbour, int *mode, int *pi_count )
{
    if( ( i_neighbour & (MB_LEFT|MB_TOP) ) == (MB_LEFT|MB_TOP) )
    {
        /* 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 = 8 */
static void predict_4x4_mode_available( unsigned int i_neighbour, int idx, int *mode, int *pi_count )
{
    int b_a, b_b, b_c;
    static const unsigned int needmb[16] =
    {
        MB_LEFT|MB_TOP, MB_TOP,
        MB_LEFT,        MB_PRIVATE,
        MB_TOP,         MB_TOP|MB_TOPRIGHT,
        0,              MB_PRIVATE,
        MB_LEFT,        0,
        MB_LEFT,        MB_PRIVATE,
        0,              MB_PRIVATE,
        0,              MB_PRIVATE
    };

    /* FIXME even when b_c == 0 there is some case where missing pixels
     * are emulated and thus more mode are available TODO
     * analysis and encode should be fixed too */
    b_a = (needmb[idx]&i_neighbour&MB_LEFT) == (needmb[idx]&MB_LEFT);
    b_b = (needmb[idx]&i_neighbour&MB_TOP) == (needmb[idx]&MB_TOP);
    b_c = (needmb[idx]&i_neighbour&(MB_TOPRIGHT|MB_PRIVATE)) == (needmb[idx]&(MB_TOPRIGHT|MB_PRIVATE));

    if( b_a && b_b )
    {
        *mode++ = I_PRED_4x4_DC;
        *mode++ = I_PRED_4x4_H;
        *mode++ = I_PRED_4x4_V;
        *mode++ = I_PRED_4x4_DDR;
        *mode++ = I_PRED_4x4_VR;
        *mode++ = I_PRED_4x4_HD;
        *mode++ = I_PRED_4x4_HU;

        *pi_count = 7;

        if( b_c )
        {
            *mode++ = I_PRED_4x4_DDL;
            *mode++ = I_PRED_4x4_VL;
            (*pi_count) += 2;
        }