me.c

H.264编码器

/* me.c: h264 encoder library (Motion Estimation) */

#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include "stdint.h"

#include "common.h"

static/* inline */ int me_mv_test( x264_t *h,
                              int *pi_sad, int *pi_mvx, int *pi_mvy,
                              int i_sad_mode, uint8_t *img, int i_img, uint8_t *test, int i_test_stride,
                              int mvx, int mvy,
                              int i_lambda_motion, int mvxp, int mvyp )
{
    int i_sad;

    i_sad = h->pixf.sad[i_sad_mode]( img, i_img, &test[mvy*i_test_stride+mvx], i_test_stride );
    if( i_sad + i_lambda_motion * ( bs_size_se((mvx<<2) - mvxp)+bs_size_se((mvy<<2) - mvyp) ) <
       *pi_sad + i_lambda_motion * ( bs_size_se(((*pi_mvx)<<2) - mvxp)+bs_size_se(((*pi_mvy)<<2) - mvyp) ) )
    {
        *pi_sad = i_sad;
        *pi_mvx = mvx;
        *pi_mvy = mvy;
        return 1;
    }

    return 0;
}

static/* inline */ int me_mv_test_buffer( x264_t *h,
                                     int *pi_sad, int *pi_mvx, int *pi_mvy,
                                     int i_sad_mode, uint8_t *img, int i_img, uint8_t *test, int i_test_stride,
                                     int mvx, int mvy,
                                      int i_lambda_motion, int mvxp, int mvyp )
{
    int i_sad;

    i_sad = h->pixf.satd[i_sad_mode]( img, i_img, test, i_test_stride );
    if( i_sad  + i_lambda_motion * ( bs_size_se(mvx - mvxp)     + bs_size_se(mvy - mvyp) ) <
       *pi_sad + i_lambda_motion * ( bs_size_se((*pi_mvx)-mvxp) + bs_size_se((*pi_mvy) - mvyp) ) )
    {
        *pi_sad = i_sad;
        *pi_mvx = mvx;
        *pi_mvy = mvy;
        return 1;
    }

    return 0;
}

/* Range : -16:16 */
static int x264_me_umhexagons( x264_t *h, uint8_t *p_ref, int i_ref,
                               uint8_t *p_img, int i_img,
                               int i_sad_mode,
                               int i_lambda_motion,
                               int *pi_mvx, int *pi_mvy )
{
    int i, n, dx, dy;

    int i_bsatd;
    int i_bmx,i_bmy;
    int mvx, mvy;
    int bw, bh;

    int ok;

    static const int hgrid_x[16]= { -4,-4,-4,-2, 0, 2, 4, 4, 4, 4, 4, 2, 0,-2,-4,-4 };
    static const int hgrid_y[16]= {  0, 1, 2, 3, 4, 3, 2, 1, 0,-1,-2,-3,-4,-3, 2, 1 };

    static const int hexa_x[6] = { -2,-1, 1, 2, 1,-1 };
    static const int hexa_y[6] = { 0,  2, 2, 0, 2, 2 };

    static const int star_x[4] = { -1, 0, 1,  0 };
    static const int star_y[4] = {  0, 1, 0, -1 };

    /* block size */
    switch( i_sad_mode )
    {
        case PIXEL_16x16:
            bw = bh = 16;
            break;
        case PIXEL_16x8:
            bw = 16; bh = 8;
            break;
        case PIXEL_8x16:
            bw = 8; bh = 16;
            break;
        case PIXEL_8x8:
            bw = 8; bh = 8;
            break;
        case PIXEL_8x4:
            bw = 8; bh = 4;
            break;
        case PIXEL_4x8:
            bw = 4; bh = 8;
            break;
        case PIXEL_4x4:
        default:
            bw = 4; bh = 4;
            break;
    }

    /* mv start (clipped) */
    mvx = x264_clip3( (*pi_mvx)>>2, -16, 16 );
    mvy = x264_clip3( (*pi_mvy)>>2, -16, 16 );

    /* init with center */
    i_bmx = mvx;
    i_bmy = mvy;
    i_bsatd = h->pixf.sad[i_sad_mode]( p_img, i_img, &p_ref[mvy*i_ref+mvx], i_ref );

    /* First step: Unsymetrical cross edge */
    for( dx = mvx - 16; dx <= mvx + 16; dx += 2 )
    {
        if( dx >= -16 && dx <= 16 )
        {
            me_mv_test( h, &i_bsatd, &i_bmx, &i_bmy,
                        i_sad_mode, p_img, i_img, p_ref, i_ref, dx, mvy,
                        i_lambda_motion, *pi_mvx, *pi_mvy );
        }
    }
    for( dy = mvy - 8; dy <= mvy + 8; dy += 2 )
    {
        if( dy >= - 16 && dy <= 16 )
        {
            me_mv_test( h, &i_bsatd, &i_bmx, &i_bmy,
                        i_sad_mode, p_img, i_img, p_ref, i_ref, mvx, dy,
                        i_lambda_motion, *pi_mvx, *pi_mvy );
        }
    }

    /* Second step-part 1: full search -2/2 */
    mvx = i_bmx; mvy = i_bmy;
    for( dy = -2; dy <= 2; dy++ )
    {
        for( dx = -2; dx <= 2; dx++ )
        {
            int mx, my;
            mx = mvx+dx;
            my = mvy+dy;

            if( mx >= -16 && mx <= 16 && my >= -16 && my < 16 )
            {
                me_mv_test( h, &i_bsatd, &i_bmx, &i_bmy,
                            i_sad_mode, p_img, i_img, p_ref, i_ref, mx, my,
                            i_lambda_motion, *pi_mvx, *pi_mvy );
            }
        }
    }
    /* Second step-part 2: multi hexa research */
    ok = 1;
    mvx = i_bmx; mvy = i_bmy;
    for( n = 1; n <= 4; n++ )
    {
        for( i = 0; i < 16; i++ )
        {
            dx = mvx + hgrid_x[i] * n;
            dy = mvy + hgrid_y[i] * n;
            if( dx >= -16 && dx <= 16 && dy >= -16 && dy <= 16 )
            {
                if( me_mv_test( h, &i_bsatd, &i_bmx, &i_bmy,
                                i_sad_mode, p_img, i_img, p_ref, i_ref, dx, dy,
                                i_lambda_motion, *pi_mvx, *pi_mvy ) )
                {
                    ok = 0; /* we will have to do EHS step */
                }
            }
        }
    }

    /* Third step: EHS */
    if( !ok )
    {
        do
        {
            ok = 1;
            mvx = i_bmx; mvy = i_bmy;

            for( i = 0; i < 6; i++ )
            {
                dx = mvx + hexa_x[i];
                dy = mvy + hexa_y[i];
                if( dx >= -16 && dx <= 16 && dy >= -16 && dy <= 16 )
                {
                    if( me_mv_test( h, &i_bsatd, &i_bmx, &i_bmy,
                                    i_sad_mode, p_img, i_img, p_ref, i_ref, dx, dy,
                                    i_lambda_motion, *pi_mvx, *pi_mvy ) )
                    {
                        ok = 0;
                    }
                }
            }
        } while( !ok );

        do
        {
            ok = 1;
            mvx = i_bmx; mvy = i_bmy;

            for( i = 0; i < 4; i++ )
            {
                dx = mvx + star_x[i];
                dy = mvy + star_y[i];
                if( dx >= -16 && dx <= 16 && dy >= -16 && dy <= 16 )
                {
                    if( me_mv_test( h, &i_bsatd, &i_bmx, &i_bmy,
                                    i_sad_mode, p_img, i_img, p_ref, i_ref, dx, dy,
                                    i_lambda_motion, *pi_mvx, *pi_mvy ) )
                    {
                        ok = 0;
                    }
                }
            }
        } while( !ok );
    }

    /* 1/4 pixel rafinement using diamond research */
    i_bsatd = h->pixf.satd[i_sad_mode]( p_img, i_img, &p_ref[i_bmx+i_bmy*i_ref], i_ref ) +
                i_lambda_motion * ( bs_size_se((i_bmx<<2) - (*pi_mvx))+bs_size_se((i_bmy<<2) - (*pi_mvy)) );

    i_bmx <<= 2;
    i_bmy <<= 2;
    {
        uint8_t pix[16*16];
        int satd[4];
        int i, i_b;

#define TEST_MV( s, mx, my ) \
        h->mc[MC_LUMA]( p_ref, i_ref, pix, bw, (mx), (my), bw, bh ); \
        (s) = h->pixf.satd[i_sad_mode]( p_img, i_img, pix, bw ) + \
            i_lambda_motion * ( bs_size_se((mx) - (*pi_mvx))+bs_size_se((my) - (*pi_mvy)) )

        TEST_MV( satd[0], i_bmx,   i_bmy-1 );
        TEST_MV( satd[1], i_bmx,   i_bmy+1 );
        TEST_MV( satd[2], i_bmx-1, i_bmy );
        TEST_MV( satd[3], i_bmx+1, i_bmy );
        for( i = 1, i_b = 0; i < 4; i++ )
        {
            if( satd[i] < satd[i_b] )
            {
                i_b = i;
            }
        }
        if( satd[i_b] < i_bsatd )
        {
            if( i_b == 0 )      i_bmy--;
            else if( i_b == 1 ) i_bmy++;
            else if( i_b == 2 ) i_bmx--;
            else if( i_b == 3 ) i_bmx++;
            i_bsatd = satd[i_b];

            if( i_b != 1 ) { TEST_MV( satd[0], i_bmx,   i_bmy-1 ); } else { satd[0] = i_bsatd; }
            if( i_b != 0 ) { TEST_MV( satd[1], i_bmx,   i_bmy+1 ); } else { satd[1] = i_bsatd; }
            if( i_b != 3 ) { TEST_MV( satd[2], i_bmx-1, i_bmy   ); } else { satd[2] = i_bsatd; }
            if( i_b != 2 ) { TEST_MV( satd[3], i_bmx+1, i_bmy   ); } else { satd[3] = i_bsatd; }

            for( i = 1, i_b = 0; i < 4; i++ )
            {
                if( satd[i] < satd[i_b] )
                {
                    i_b = i;
                }
            }
            if( satd[i_b] < i_bsatd )
            {
                if( i_b == 0 )      i_bmy--;
                else if( i_b == 1 ) i_bmy++;
                else if( i_b == 2 ) i_bmx--;
                else if( i_b == 3 ) i_bmx++;
                i_bsatd = satd[i_b];
            }
        }
#undef TEST_MV
    }
    *pi_mvx = i_bmx;
    *pi_mvy = i_bmy;

    return i_bsatd;
}

static int x264_me_diamond( x264_t *h,
                            uint8_t *p_ref, int i_ref,
                            uint8_t *p_img, int i_img,
                            int i_sad_mode,
                            int i_lambda_motion,
                            int *pi_mvx, int *pi_mvy )
{
    int bw, bh;
    int i_bsatd;
    int i_bmx, i_bmy;

    /* block size */
    switch( i_sad_mode )
    {
        case PIXEL_16x16:
            bw = bh = 16;
            break;
        case PIXEL_16x8:
            bw = 16; bh = 8;
            break;
        case PIXEL_8x16:
            bw = 8; bh = 16;
            break;
        case PIXEL_8x8:
            bw = 8; bh = 8;
            break;
        case PIXEL_8x4:
            bw = 8; bh = 4;
            break;
        case PIXEL_4x8:
            bw = 4; bh = 8;
            break;
        case PIXEL_4x4:
        default:
            bw = 4; bh = 4;
            break;
    }

    /* init with center */
    i_bmx = x264_clip3( (*pi_mvx)>>2, -16, 16 );
    i_bmy = x264_clip3( (*pi_mvy)>>2, -16, 16 );
    i_bsatd = h->pixf.sad[i_sad_mode]( p_img, i_img, &p_ref[i_bmy*i_ref+i_bmx], i_ref );

    for( ;; )
    {
        int satd[4];
        int i, i_b;
        if( i_bmx < -16 || i_bmx > 16 || i_bmy < -16 || i_bmy > 16 )
        {
            break;
        }
#define TEST_MV( mx, my ) \
        h->pixf.sad[i_sad_mode]( p_img, i_img, &p_ref[(my)*i_ref+(mx)], i_ref ) + \
            i_lambda_motion * ( bs_size_se(((mx)<<2) - (*pi_mvx))+bs_size_se(((my)<<2) - (*pi_mvy)) )
        satd[0] = TEST_MV( i_bmx,   i_bmy-1 );
        satd[1] = TEST_MV( i_bmx,   i_bmy+1 );
        satd[2] = TEST_MV( i_bmx-1, i_bmy );
        satd[3] = TEST_MV( i_bmx+1, i_bmy );
#undef TEST_MV

        for( i = 1, i_b = 0; i < 4; i++ )
        {
            if( satd[i] < satd[i_b] )
            {
                i_b = i;
            }
        }
        if( satd[i_b] >= i_bsatd )
        {
            break;
        }

        if( i_b == 0 )      i_bmy--;
        else if( i_b == 1 ) i_bmy++;
        else if( i_b == 2 ) i_bmx--;
        else if( i_b == 3 ) i_bmx++;
        i_bsatd = satd[i_b];
    }

    /* 1/4 pixel rafinement using diamond research */
    i_bsatd = h->pixf.satd[i_sad_mode]( p_img, i_img, &p_ref[i_bmx+i_bmy*i_ref], i_ref ) +
                i_lambda_motion * ( bs_size_se((i_bmx<<2) - (*pi_mvx))+bs_size_se((i_bmy<<2) - (*pi_mvy)) );
    i_bmx <<= 2;
    i_bmy <<= 2;
    {
        uint8_t pix[16*16];
        int satd[4];
        int i, i_b;

#define TEST_MV( s, mx, my ) \
        h->mc[MC_LUMA]( p_ref, i_ref, pix, bw, (mx), (my), bw, bh ); \
        (s) = h->pixf.satd[i_sad_mode]( p_img, i_img, pix, bw ) + \
            i_lambda_motion * ( bs_size_se((mx) - (*pi_mvx))+bs_size_se((my) - (*pi_mvy)) )

        TEST_MV( satd[0], i_bmx,   i_bmy-1 );
        TEST_MV( satd[1], i_bmx,   i_bmy+1 );
        TEST_MV( satd[2], i_bmx-1, i_bmy );
        TEST_MV( satd[3], i_bmx+1, i_bmy );
        for( i = 1, i_b = 0; i < 4; i++ )
        {
            if( satd[i] < satd[i_b] )
            {
                i_b = i;
            }
        }
        if( satd[i_b] < i_bsatd )
        {
            if( i_b == 0 )      i_bmy--;
            else if( i_b == 1 ) i_bmy++;
            else if( i_b == 2 ) i_bmx--;
            else if( i_b == 3 ) i_bmx++;
            i_bsatd = satd[i_b];

            if( i_b != 1 ) { TEST_MV( satd[0], i_bmx,   i_bmy-1 ); } else { satd[0] = i_bsatd; }
            if( i_b != 0 ) { TEST_MV( satd[1], i_bmx,   i_bmy+1 ); } else { satd[1] = i_bsatd; }
            if( i_b != 3 ) { TEST_MV( satd[2], i_bmx-1, i_bmy   ); } else { satd[2] = i_bsatd; }
            if( i_b != 2 ) { TEST_MV( satd[3], i_bmx+1, i_bmy   ); } else { satd[3] = i_bsatd; }

            for( i = 1, i_b = 0; i < 4; i++ )
            {
                if( satd[i] < satd[i_b] )
                {
                    i_b = i;
                }
            }
            if( satd[i_b] < i_bsatd )
            {
                if( i_b == 0 )      i_bmy--;
                else if( i_b == 1 ) i_bmy++;
                else if( i_b == 2 ) i_bmx--;
                else if( i_b == 3 ) i_bmx++;
                i_bsatd = satd[i_b];
            }
        }
#undef TEST_MV
    }


    *pi_mvx = i_bmx;
    *pi_mvy = i_bmy;

    return i_bsatd;
}

void x264_me_init(  x264_me_t pf[2] )
{
    pf[X264_ME_UMHEXAGONS] = x264_me_umhexagons;
    pf[X264_ME_DIAMOND]    = x264_me_diamond;
}