slicetype.c

linux下编译已经通过

/*****************************************************************************
 * slicetype.c: h264 encoder library
 *****************************************************************************
 * Copyright (C) 2005 Loren Merritt
 *
 * Authors: 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 <math.h>
#include <limits.h>

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


static void x264_lowres_context_init( x264_t *h, x264_mb_analysis_t *a )
{
    a->i_qp = 12; // arbitrary, but low because SATD scores are 1/4 normal
    a->i_lambda = i_qp0_cost_table[ a->i_qp ];
    x264_mb_analyse_load_costs( h, a );
    h->mb.i_me_method = X264_MIN( X264_ME_HEX, h->param.analyse.i_me_method ); // maybe dia?
    h->mb.i_subpel_refine = 4; // 3 should be enough, but not tweaking for speed now
    h->mb.b_chroma_me = 0;
}

int x264_slicetype_mb_cost( x264_t *h, x264_mb_analysis_t *a,
                            x264_frame_t **frames, int p0, int p1, int b,
                            int dist_scale_factor )
{
    x264_frame_t *fref0 = frames[p0];
    x264_frame_t *fref1 = frames[p1];
    x264_frame_t *fenc  = frames[b];
    const int b_bidir = (b < p1);
    const int i_mb_x = h->mb.i_mb_x;
    const int i_mb_y = h->mb.i_mb_y;
    const int i_mb_stride = h->sps->i_mb_width;
    const int i_mb_xy = i_mb_x + i_mb_y * i_mb_stride;
    const int i_stride = fenc->i_stride_lowres;
    const int i_pel_offset = 8 * ( i_mb_x + i_mb_y * i_stride );

    DECLARE_ALIGNED( uint8_t, pix1[9*FDEC_STRIDE], 8 );
    uint8_t *pix2 = pix1+8;
    x264_me_t m[2];
    int i_bcost = COST_MAX;
    int i_cost_bak;
    int l, i;

    h->mb.pic.p_fenc[0] = h->mb.pic.fenc_buf;
    h->mc.copy[PIXEL_8x8]( h->mb.pic.p_fenc[0], FENC_STRIDE, &fenc->lowres[0][i_pel_offset], i_stride, 8 );

    if( !p0 && !p1 && !b )
        goto lowres_intra_mb;

    // no need for h->mb.mv_min[]
    h->mb.mv_min_fpel[0] = -8*h->mb.i_mb_x - 4;
    h->mb.mv_max_fpel[0] = 8*( h->sps->i_mb_width - h->mb.i_mb_x - 1 ) + 4;
    h->mb.mv_min_spel[0] = 4*( h->mb.mv_min_fpel[0] - 8 );
    h->mb.mv_max_spel[0] = 4*( h->mb.mv_max_fpel[0] + 8 );
    if( h->mb.i_mb_x <= 1 )
    {
        h->mb.mv_min_fpel[1] = -8*h->mb.i_mb_y - 4;
        h->mb.mv_max_fpel[1] = 8*( h->sps->i_mb_height - h->mb.i_mb_y - 1 ) + 4;
        h->mb.mv_min_spel[1] = 4*( h->mb.mv_min_fpel[1] - 8 );
        h->mb.mv_max_spel[1] = 4*( h->mb.mv_max_fpel[1] + 8 );
    }

#define LOAD_HPELS_LUMA(dst, src) \
    { \
        (dst)[0] = &(src)[0][i_pel_offset]; \
        (dst)[1] = &(src)[1][i_pel_offset]; \
        (dst)[2] = &(src)[2][i_pel_offset]; \
        (dst)[3] = &(src)[3][i_pel_offset]; \
    }
#define SAVE_MVS( mv0, mv1 ) \
    { \
        fenc->mv[0][i_mb_xy][0] = mv0[0]; \
        fenc->mv[0][i_mb_xy][1] = mv0[1]; \
        if( b_bidir ) \
        { \
            fenc->mv[1][i_mb_xy][0] = mv1[0]; \
            fenc->mv[1][i_mb_xy][1] = mv1[1]; \
        } \
    }
#define CLIP_MV( mv ) \
    { \
        mv[0] = x264_clip3( mv[0], h->mb.mv_min_spel[0], h->mb.mv_max_spel[0] ); \
        mv[1] = x264_clip3( mv[1], h->mb.mv_min_spel[1], h->mb.mv_max_spel[1] ); \
    }
#define TRY_BIDIR( mv0, mv1, penalty ) \
    { \
        int stride2 = 16; \
        uint8_t *src2; \
        int i_cost; \
        h->mc.mc_luma( m[0].p_fref, m[0].i_stride[0], pix1, 16, \
                       (mv0)[0], (mv0)[1], 8, 8 ); \
        src2 = h->mc.get_ref( m[1].p_fref, m[1].i_stride[0], pix2, &stride2, \
                       (mv1)[0], (mv1)[1], 8, 8 ); \
        h->mc.avg[PIXEL_8x8]( pix1, 16, src2, stride2 ); \
        i_cost = penalty + h->pixf.mbcmp[PIXEL_8x8]( \
                           m[0].p_fenc[0], FENC_STRIDE, pix1, 16 ); \
        if( i_bcost > i_cost ) \
        { \
            i_bcost = i_cost; \
            SAVE_MVS( mv0, mv1 ); \
        } \
    }

    m[0].i_pixel = PIXEL_8x8;
    m[0].p_cost_mv = a->p_cost_mv;
    m[0].i_stride[0] = i_stride;
    m[0].p_fenc[0] = h->mb.pic.p_fenc[0];
    LOAD_HPELS_LUMA( m[0].p_fref, fref0->lowres );

    if( b_bidir )
    {
        int16_t *mvr = fref1->mv[0][i_mb_xy];
        int dmv[2][2];
        int mv0[2] = {0,0};

        m[1] = m[0];
        LOAD_HPELS_LUMA( m[1].p_fref, fref1->lowres );

        dmv[0][0] = ( mvr[0] * dist_scale_factor + 128 ) >> 8;
        dmv[0][1] = ( mvr[1] * dist_scale_factor + 128 ) >> 8;
        dmv[1][0] = dmv[0][0] - mvr[0];
        dmv[1][1] = dmv[0][1] - mvr[1];
        CLIP_MV( dmv[0] );
        CLIP_MV( dmv[1] );

        TRY_BIDIR( dmv[0], dmv[1], 0 );
        if( dmv[0][0] || dmv[0][1] || dmv[1][0] || dmv[1][1] )
           TRY_BIDIR( mv0, mv0, 0 );
//      if( i_bcost < 60 ) // arbitrary threshold
//          return i_bcost;
    }

    i_cost_bak = i_bcost;
    for( l = 0; l < 1 + b_bidir; l++ )
    {
        int mvc[4][2] = {{0}}, i_mvc;
        int16_t (*fenc_mv)[2] = &fenc->mv[l][i_mb_xy];
        i_mvc = 0;
#define MVC(mv) { mvc[i_mvc][0] = mv[0]; mvc[i_mvc][1] = mv[1]; i_mvc++; }
        if( i_mb_x > 0 )
            MVC(fenc_mv[-1]);
        if( i_mb_y > 0 )
        {
            MVC(fenc_mv[-i_mb_stride]);
            if( i_mb_x < h->sps->i_mb_width - 1 )
                MVC(fenc_mv[-i_mb_stride+1]);
            if( i_mb_x > 0 )
                MVC(fenc_mv[-i_mb_stride-1]);
        }
#undef MVC
        m[l].mvp[0] = x264_median( mvc[0][0], mvc[1][0], mvc[2][0] );
        m[l].mvp[1] = x264_median( mvc[0][1], mvc[1][1], mvc[2][1] );

        x264_me_search( h, &m[l], mvc, i_mvc );

        m[l].cost -= 2; // remove mvcost from skip mbs
        if( m[l].mv[0] || m[l].mv[1] )
            m[l].cost += 5;
        i_bcost = X264_MIN( i_bcost, m[l].cost );
    }

    if( b_bidir && (m[0].mv[0] || m[0].mv[1] || m[1].mv[0] || m[1].mv[1]) )
        TRY_BIDIR( m[0].mv, m[1].mv, 5 );

    if( i_bcost < i_cost_bak )
        SAVE_MVS( m[0].mv, m[1].mv );

    //FIXME intra part could be shared across multiple encodings of the frame
lowres_intra_mb:
    if( !b_bidir ) // forbid intra-mbs in B-frames, because it's rare and not worth checking
    {
        uint8_t *pix = &pix1[8+FDEC_STRIDE - 1];
        uint8_t *src = &fenc->lowres[0][i_pel_offset - 1];
        const int intra_penalty = 5;
        int satds[4], i_icost, b_intra;

        memcpy( pix-FDEC_STRIDE, src-i_stride, 17 );
        for( i=0; i<8; i++ )
            pix[i*FDEC_STRIDE] = src[i*i_stride];
        pix++;

        if( h->pixf.intra_satd_x3_8x8c && h->pixf.mbcmp[0] == h->pixf.satd[0] )
        {
            h->pixf.intra_satd_x3_8x8c( h->mb.pic.p_fenc[0], pix, satds );
            h->predict_8x8c[I_PRED_CHROMA_P]( pix );
            satds[I_PRED_CHROMA_P] =
                h->pixf.satd[PIXEL_8x8]( pix, FDEC_STRIDE, h->mb.pic.p_fenc[0], FENC_STRIDE );
        }
        else
        {
            for( i=0; i<4; i++ )
            {
                h->predict_8x8c[i]( pix );
                satds[i] = h->pixf.mbcmp[PIXEL_8x8]( pix, FDEC_STRIDE, h->mb.pic.p_fenc[0], FENC_STRIDE );
            }
        }
        i_icost = X264_MIN4( satds[0], satds[1], satds[2], satds[3] );

        if( i_icost < i_bcost * 2 )
        {
            DECLARE_ALIGNED( uint8_t, edge[33], 8 );
            x264_predict_8x8_filter( pix, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );
            for( i=3; i<9; i++ )
            {
                int satd;
                h->predict_8x8[i]( pix, edge );
                satd = h->pixf.mbcmp[PIXEL_8x8]( pix, FDEC_STRIDE, h->mb.pic.p_fenc[0], FENC_STRIDE );
                i_icost = X264_MIN( i_icost, satd );
            }
        }

        i_icost += intra_penalty;
        b_intra = i_icost < i_bcost;
        if( b_intra )
            i_bcost = i_icost;
        if(    i_mb_x > 0 && i_mb_x < h->sps->i_mb_width - 1
            && i_mb_y > 0 && i_mb_y < h->sps->i_mb_height - 1 )
        {
            fenc->i_intra_mbs[b-p0] += b_intra;
            fenc->i_cost_est[0][0] += i_icost;
        }
    }

    return i_bcost;
}
#undef TRY_BIDIR
#undef SAVE_MVS

int x264_slicetype_frame_cost( x264_t *h, x264_mb_analysis_t *a,
                               x264_frame_t **frames, int p0, int p1, int b,
                               int b_intra_penalty )
{
    int i_score = 0;

    /* Check whether we already evaluated this frame
     * If we have tried this frame as P, then we have also tried
     * the preceding frames as B. (is this still true?) */
    if( frames[b]->i_cost_est[b-p0][p1-b] >= 0 )
    {
        i_score = frames[b]->i_cost_est[b-p0][p1-b];
    }
    else
    {
        int dist_scale_factor = 128;
        int *row_satd = frames[b]->i_row_satds[b-p0][p1-b];

        /* Init MVs so that we don't have to check edge conditions when loading predictors. */
        /* FIXME: not needed every time */
        memset( frames[b]->mv[0], 0, h->sps->i_mb_height * h->sps->i_mb_width * 2*sizeof(int16_t) );