mcomp.c

/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */


#include "onyx_int.h"
#include "mcomp.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_config.h"
#include <stdio.h>
#include <limits.h>
#include <math.h>
#include "vp8/common/findnearmv.h"
#include "vp8/common/common.h"

#ifdef VP8_ENTROPY_STATS
static int mv_ref_ct [31] [4] [2];
static int mv_mode_cts [4] [2];
#endif

int vp8_mv_bit_cost(int_mv *mv, int_mv *ref, int *mvcost[2], int Weight)
{
    /* MV costing is based on the distribution of vectors in the previous
     * frame and as such will tend to over state the cost of vectors. In
     * addition coding a new vector can have a knock on effect on the cost
     * of subsequent vectors and the quality of prediction from NEAR and
     * NEAREST for subsequent blocks. The "Weight" parameter allows, to a
     * limited extent, for some account to be taken of these factors.
     */
    return ((mvcost[0][(mv->as_mv.row - ref->as_mv.row) >> 1] + mvcost[1][(mv->as_mv.col - ref->as_mv.col) >> 1]) * Weight) >> 7;
}

static int mv_err_cost(int_mv *mv, int_mv *ref, int *mvcost[2], int error_per_bit)
{
    /* Ignore mv costing if mvcost is NULL */
    if (mvcost)
        return ((mvcost[0][(mv->as_mv.row - ref->as_mv.row) >> 1] +
                 mvcost[1][(mv->as_mv.col - ref->as_mv.col) >> 1])
                 * error_per_bit + 128) >> 8;
    return 0;
}

static int mvsad_err_cost(int_mv *mv, int_mv *ref, int *mvsadcost[2], int error_per_bit)
{
    /* Calculate sad error cost on full pixel basis. */
    /* Ignore mv costing if mvsadcost is NULL */
    if (mvsadcost)
        return ((mvsadcost[0][(mv->as_mv.row - ref->as_mv.row)] +
                 mvsadcost[1][(mv->as_mv.col - ref->as_mv.col)])
                * error_per_bit + 128) >> 8;
    return 0;
}

void vp8_init_dsmotion_compensation(MACROBLOCK *x, int stride)
{
    int Len;
    int search_site_count = 0;


    /* Generate offsets for 4 search sites per step. */
    Len = MAX_FIRST_STEP;
    x->ss[search_site_count].mv.col = 0;
    x->ss[search_site_count].mv.row = 0;
    x->ss[search_site_count].offset = 0;
    search_site_count++;

    while (Len > 0)
    {

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = 0;
        x->ss[search_site_count].mv.row = -Len;
        x->ss[search_site_count].offset = -Len * stride;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = 0;
        x->ss[search_site_count].mv.row = Len;
        x->ss[search_site_count].offset = Len * stride;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = -Len;
        x->ss[search_site_count].mv.row = 0;
        x->ss[search_site_count].offset = -Len;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = Len;
        x->ss[search_site_count].mv.row = 0;
        x->ss[search_site_count].offset = Len;
        search_site_count++;

        /* Contract. */
        Len /= 2;
    }

    x->ss_count = search_site_count;
    x->searches_per_step = 4;
}

void vp8_init3smotion_compensation(MACROBLOCK *x, int stride)
{
    int Len;
    int search_site_count = 0;

    /* Generate offsets for 8 search sites per step. */
    Len = MAX_FIRST_STEP;
    x->ss[search_site_count].mv.col = 0;
    x->ss[search_site_count].mv.row = 0;
    x->ss[search_site_count].offset = 0;
    search_site_count++;

    while (Len > 0)
    {

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = 0;
        x->ss[search_site_count].mv.row = -Len;
        x->ss[search_site_count].offset = -Len * stride;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = 0;
        x->ss[search_site_count].mv.row = Len;
        x->ss[search_site_count].offset = Len * stride;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = -Len;
        x->ss[search_site_count].mv.row = 0;
        x->ss[search_site_count].offset = -Len;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = Len;
        x->ss[search_site_count].mv.row = 0;
        x->ss[search_site_count].offset = Len;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = -Len;
        x->ss[search_site_count].mv.row = -Len;
        x->ss[search_site_count].offset = -Len * stride - Len;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = Len;
        x->ss[search_site_count].mv.row = -Len;
        x->ss[search_site_count].offset = -Len * stride + Len;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = -Len;
        x->ss[search_site_count].mv.row = Len;
        x->ss[search_site_count].offset = Len * stride - Len;
        search_site_count++;

        /* Compute offsets for search sites. */
        x->ss[search_site_count].mv.col = Len;
        x->ss[search_site_count].mv.row = Len;
        x->ss[search_site_count].offset = Len * stride + Len;
        search_site_count++;


        /* Contract. */
        Len /= 2;
    }

    x->ss_count = search_site_count;
    x->searches_per_step = 8;
}

/*
 * To avoid the penalty for crossing cache-line read, preload the reference
 * area in a small buffer, which is aligned to make sure there won't be crossing
 * cache-line read while reading from this buffer. This reduced the cpu
 * cycles spent on reading ref data in sub-pixel filter functions.
 * TODO: Currently, since sub-pixel search range here is -3 ~ 3, copy 22 rows x
 * 32 cols area that is enough for 16x16 macroblock. Later, for SPLITMV, we
 * could reduce the area.
 */

/* estimated cost of a motion vector (r,c) */
#define MVC(r,c) (mvcost ? ((mvcost[0][(r)-rr] + mvcost[1][(c) - rc]) * error_per_bit + 128 )>>8 : 0)
/* pointer to predictor base of a motionvector */
#define PRE(r,c) (y + (((r)>>2) * y_stride + ((c)>>2) -(offset)))
/* convert motion vector component to offset for svf calc */
#define SP(x) (((x)&3)<<1)
/* returns subpixel variance error function. */
#define DIST(r,c) vfp->svf( PRE(r,c), y_stride, SP(c),SP(r), z,b->src_stride,&sse)
#define IFMVCV(r,c,s,e) if ( c >= minc && c <= maxc && r >= minr && r <= maxr) s else e;
/* returns distortion + motion vector cost */
#define ERR(r,c) (MVC(r,c)+DIST(r,c))
/* checks if (r,c) has better score than previous best */
#define CHECK_BETTER(v,r,c) IFMVCV(r,c,{thismse = DIST(r,c); if((v = (MVC(r,c)+thismse)) < besterr) { besterr = v; br=r; bc=c; *distortion = thismse; *sse1 = sse; }}, v=UINT_MAX;)

int vp8_find_best_sub_pixel_step_iteratively(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
                                             int_mv *bestmv, int_mv *ref_mv,
                                             int error_per_bit,
                                             const vp8_variance_fn_ptr_t *vfp,
                                             int *mvcost[2], int *distortion,
                                             unsigned int *sse1)
{
    unsigned char *z = (*(b->base_src) + b->src);

    int rr = ref_mv->as_mv.row >> 1, rc = ref_mv->as_mv.col >> 1;
    int br = bestmv->as_mv.row * 4, bc = bestmv->as_mv.col * 4;
    int tr = br, tc = bc;
    unsigned int besterr;
    unsigned int left, right, up, down, diag;
    unsigned int sse;
    unsigned int whichdir;
    unsigned int halfiters = 4;
    unsigned int quarteriters = 4;
    int thismse;

    int minc = MAX(x->mv_col_min * 4,
                   (ref_mv->as_mv.col >> 1) - ((1 << mvlong_width) - 1));
    int maxc = MIN(x->mv_col_max * 4,
                   (ref_mv->as_mv.col >> 1) + ((1 << mvlong_width) - 1));
    int minr = MAX(x->mv_row_min * 4,
                   (ref_mv->as_mv.row >> 1) - ((1 << mvlong_width) - 1));
    int maxr = MIN(x->mv_row_max * 4,
                   (ref_mv->as_mv.row >> 1) + ((1 << mvlong_width) - 1));

    int y_stride;
    int offset;
    int pre_stride = x->e_mbd.pre.y_stride;
    unsigned char *base_pre = x->e_mbd.pre.y_buffer;


#if ARCH_X86 || ARCH_X86_64
    MACROBLOCKD *xd = &x->e_mbd;
    unsigned char *y_0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
    unsigned char *y;
    int buf_r1, buf_r2, buf_c1;

    /* Clamping to avoid out-of-range data access */
    buf_r1 = ((bestmv->as_mv.row - 3) < x->mv_row_min)?(bestmv->as_mv.row - x->mv_row_min):3;
    buf_r2 = ((bestmv->as_mv.row + 3) > x->mv_row_max)?(x->mv_row_max - bestmv->as_mv.row):3;
    buf_c1 = ((bestmv->as_mv.col - 3) < x->mv_col_min)?(bestmv->as_mv.col - x->mv_col_min):3;
    y_stride = 32;

    /* Copy to intermediate buffer before searching. */
    vfp->copymem(y_0 - buf_c1 - pre_stride*buf_r1, pre_stride, xd->y_buf, y_stride, 16+buf_r1+buf_r2);
    y = xd->y_buf + y_stride*buf_r1 +buf_c1;
#else
    unsigned char *y = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
    y_stride = pre_stride;
#endif

    offset = (bestmv->as_mv.row) * y_stride + bestmv->as_mv.col;

    /* central mv */
    bestmv->as_mv.row *= 8;
    bestmv->as_mv.col *= 8;

    /* calculate central point error */
    besterr = vfp->vf(y, y_stride, z, b->src_stride, sse1);
    *distortion = besterr;
    besterr += mv_err_cost(bestmv, ref_mv, mvcost, error_per_bit);

    /* TODO: Each subsequent iteration checks at least one point in common
     * with the last iteration could be 2 ( if diag selected)
     */
    while (--halfiters)
    {
        /* 1/2 pel */
        CHECK_BETTER(left, tr, tc - 2);
        CHECK_BETTER(right, tr, tc + 2);
        CHECK_BETTER(up, tr - 2, tc);
        CHECK_BETTER(down, tr + 2, tc);

        whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);

        switch (whichdir)
        {
        case 0:
            CHECK_BETTER(diag, tr - 2, tc - 2);
            break;
        case 1:
            CHECK_BETTER(diag, tr - 2, tc + 2);
            break;
        case 2:
            CHECK_BETTER(diag, tr + 2, tc - 2);
            break;
        case 3:
            CHECK_BETTER(diag, tr + 2, tc + 2);
            break;
        }

        /* no reason to check the same one again. */
        if (tr == br && tc == bc)
            break;

        tr = br;
        tc = bc;
    }

    /* TODO: Each subsequent iteration checks at least one point in common
     * with the last iteration could be 2 ( if diag selected)
     */

    /* 1/4 pel */
    while (--quarteriters)
    {
        CHECK_BETTER(left, tr, tc - 1);
        CHECK_BETTER(right, tr, tc + 1);
        CHECK_BETTER(up, tr - 1, tc);
        CHECK_BETTER(down, tr + 1, tc);

        whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);

        switch (whichdir)
        {
        case 0:
            CHECK_BETTER(diag, tr - 1, tc - 1);
            break;
        case 1:
            CHECK_BETTER(diag, tr - 1, tc + 1);
            break;
        case 2:
            CHECK_BETTER(diag, tr + 1, tc - 1);
            break;
        case 3:
            CHECK_BETTER(diag, tr + 1, tc + 1);
            break;
        }

        /* no reason to check the same one again. */
        if (tr == br && tc == bc)
            break;

        tr = br;
        tc = bc;
    }

    bestmv->as_mv.row = br * 2;
    bestmv->as_mv.col = bc * 2;

    if ((abs(bestmv->as_mv.col - ref_mv->as_mv.col) > (MAX_FULL_PEL_VAL<<3)) ||
        (abs(bestmv->as_mv.row - ref_mv->as_mv.row) > (MAX_FULL_PEL_VAL<<3)))
        return INT_MAX;

    return besterr;
}
#undef MVC
#undef PRE
#undef SP
#undef DIST
#undef IFMVCV
#undef ERR
#undef CHECK_BETTER

int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
                                 int_mv *bestmv, int_mv *ref_mv,
                                 int error_per_bit,
                                 const vp8_variance_fn_ptr_t *vfp,
                                 int *mvcost[2], int *distortion,
                                 unsigned int *sse1)
{
    int bestmse = INT_MAX;
    int_mv startmv;
    int_mv this_mv;
    unsigned char *z = (*(b->base_src) + b->src);
    int left, right, up, down, diag;
    unsigned int sse;
    int whichdir ;
    int thismse;
    int y_stride;
    int pre_stride = x->e_mbd.pre.y_stride;
    unsigned char *base_pre = x->e_mbd.pre.y_buffer;

#if ARCH_X86 || ARCH_X86_64
    MACROBLOCKD *xd = &x->e_mbd;
    unsigned char *y_0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
    unsigned char *y;

    y_stride = 32;
    /* Copy 18 rows x 32 cols area to intermediate buffer before searching. */
     vfp->copymem(y_0 - 1 - pre_stride, pre_stride, xd->y_buf, y_stride, 18);
     y = xd->y_buf + y_stride + 1;
#else
     unsigned char *y = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
     y_stride = pre_stride;
#endif

    /* central mv */
    bestmv->as_mv.row *= 8;
    bestmv->as_mv.col *= 8;
    startmv = *bestmv;

    /* calculate central point error */
    bestmse = vfp->vf(y, y_stride, z, b->src_stride, sse1);
    *distortion = bestmse;
    bestmse += mv_err_cost(bestmv, ref_mv, mvcost, error_per_bit);

    /* go left then right and check error */
    this_mv.as_mv.row = startmv.as_mv.row;
    this_mv.as_mv.col = ((startmv.as_mv.col - 8) | 4);
    thismse = vfp->svf_halfpix_h(y - 1, y_stride, z, b->src_stride, &sse);
    left = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (left < bestmse)
    {
        *bestmv = this_mv;
        bestmse = left;
        *distortion = thismse;
        *sse1 = sse;
    }

    this_mv.as_mv.col += 8;
    thismse = vfp->svf_halfpix_h(y, y_stride, z, b->src_stride, &sse);
    right = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (right < bestmse)
    {
        *bestmv = this_mv;
        bestmse = right;
        *distortion = thismse;
        *sse1 = sse;
    }

    /* go up then down and check error */
    this_mv.as_mv.col = startmv.as_mv.col;
    this_mv.as_mv.row = ((startmv.as_mv.row - 8) | 4);
    thismse =  vfp->svf_halfpix_v(y - y_stride, y_stride, z, b->src_stride, &sse);
    up = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (up < bestmse)
    {
        *bestmv = this_mv;
        bestmse = up;
        *distortion = thismse;
        *sse1 = sse;
    }

    this_mv.as_mv.row += 8;
    thismse = vfp->svf_halfpix_v(y, y_stride, z, b->src_stride, &sse);
    down = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (down < bestmse)
    {
        *bestmv = this_mv;
        bestmse = down;
        *distortion = thismse;
        *sse1 = sse;
    }


    /* now check 1 more diagonal */
    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
    this_mv = startmv;

    switch (whichdir)
    {
    case 0:
        this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
        this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
        thismse = vfp->svf_halfpix_hv(y - 1 - y_stride, y_stride, z, b->src_stride, &sse);
        break;
    case 1:
        this_mv.as_mv.col += 4;
        this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
        thismse = vfp->svf_halfpix_hv(y - y_stride, y_stride, z, b->src_stride, &sse);
        break;
    case 2:
        this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
        this_mv.as_mv.row += 4;
        thismse = vfp->svf_halfpix_hv(y - 1, y_stride, z, b->src_stride, &sse);
        break;
    case 3:
    default:
        this_mv.as_mv.col += 4;
        this_mv.as_mv.row += 4;
        thismse = vfp->svf_halfpix_hv(y, y_stride, z, b->src_stride, &sse);
        break;
    }

    diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (diag < bestmse)
    {
        *bestmv = this_mv;
        bestmse = diag;
        *distortion = thismse;
        *sse1 = sse;
    }


    /* time to check quarter pels. */
    if (bestmv->as_mv.row < startmv.as_mv.row)
        y -= y_stride;

    if (bestmv->as_mv.col < startmv.as_mv.col)
        y--;

    startmv = *bestmv;



    /* go left then right and check error */
    this_mv.as_mv.row = startmv.as_mv.row;

    if (startmv.as_mv.col & 7)
    {
        this_mv.as_mv.col = startmv.as_mv.col - 2;
        thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
    }
    else
    {
        this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
        thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
    }

    left = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (left < bestmse)
    {
        *bestmv = this_mv;
        bestmse = left;
        *distortion = thismse;
        *sse1 = sse;
    }

    this_mv.as_mv.col += 4;
    thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
    right = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (right < bestmse)
    {
        *bestmv = this_mv;
        bestmse = right;
        *distortion = thismse;
        *sse1 = sse;
    }

    /* go up then down and check error */
    this_mv.as_mv.col = startmv.as_mv.col;

    if (startmv.as_mv.row & 7)
    {
        this_mv.as_mv.row = startmv.as_mv.row - 2;
        thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
    }
    else
    {
        this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
        thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
    }

    up = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (up < bestmse)
    {
        *bestmv = this_mv;
        bestmse = up;
        *distortion = thismse;
        *sse1 = sse;
    }

    this_mv.as_mv.row += 4;
    thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
    down = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (down < bestmse)
    {
        *bestmv = this_mv;
        bestmse = down;
        *distortion = thismse;
        *sse1 = sse;
    }


    /* now check 1 more diagonal */
    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);

    this_mv = startmv;

    switch (whichdir)
    {
    case 0:

        if (startmv.as_mv.row & 7)
        {
            this_mv.as_mv.row -= 2;

            if (startmv.as_mv.col & 7)
            {
                this_mv.as_mv.col -= 2;
                thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
            }
            else
            {
                this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
                thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);;
            }
        }
        else
        {
            this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;

            if (startmv.as_mv.col & 7)
            {
                this_mv.as_mv.col -= 2;
                thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
            }
            else
            {
                this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
                thismse = vfp->svf(y - y_stride - 1, y_stride, 6, 6, z, b->src_stride, &sse);
            }
        }

        break;
    case 1:
        this_mv.as_mv.col += 2;

        if (startmv.as_mv.row & 7)
        {
            this_mv.as_mv.row -= 2;
            thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
        }
        else
        {
            this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
            thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
        }

        break;
    case 2:
        this_mv.as_mv.row += 2;

        if (startmv.as_mv.col & 7)
        {
            this_mv.as_mv.col -= 2;
            thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
        }
        else
        {
            this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
            thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
        }

        break;
    case 3:
        this_mv.as_mv.col += 2;
        this_mv.as_mv.row += 2;
        thismse = vfp->svf(y, y_stride,  this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
        break;
    }

    diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (diag < bestmse)
    {
        *bestmv = this_mv;
        bestmse = diag;
        *distortion = thismse;
        *sse1 = sse;
    }

    return bestmse;
}

int vp8_find_best_half_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
                                  int_mv *bestmv, int_mv *ref_mv,
                                  int error_per_bit,
                                  const vp8_variance_fn_ptr_t *vfp,
                                  int *mvcost[2], int *distortion,
                                  unsigned int *sse1)
{
    int bestmse = INT_MAX;
    int_mv startmv;
    int_mv this_mv;
    unsigned char *z = (*(b->base_src) + b->src);
    int left, right, up, down, diag;
    unsigned int sse;
    int whichdir ;
    int thismse;
    int y_stride;
    int pre_stride = x->e_mbd.pre.y_stride;
    unsigned char *base_pre = x->e_mbd.pre.y_buffer;

#if ARCH_X86 || ARCH_X86_64
    MACROBLOCKD *xd = &x->e_mbd;
    unsigned char *y_0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
    unsigned char *y;

    y_stride = 32;
    /* Copy 18 rows x 32 cols area to intermediate buffer before searching. */
    vfp->copymem(y_0 - 1 - pre_stride, pre_stride, xd->y_buf, y_stride, 18);
    y = xd->y_buf + y_stride + 1;
#else
    unsigned char *y = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
    y_stride = pre_stride;
#endif

    /* central mv */
    bestmv->as_mv.row *= 8;
    bestmv->as_mv.col *= 8;
    startmv = *bestmv;

    /* calculate central point error */
    bestmse = vfp->vf(y, y_stride, z, b->src_stride, sse1);
    *distortion = bestmse;
    bestmse += mv_err_cost(bestmv, ref_mv, mvcost, error_per_bit);

    /* go left then right and check error */
    this_mv.as_mv.row = startmv.as_mv.row;
    this_mv.as_mv.col = ((startmv.as_mv.col - 8) | 4);
    thismse = vfp->svf_halfpix_h(y - 1, y_stride, z, b->src_stride, &sse);
    left = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (left < bestmse)
    {
        *bestmv = this_mv;
        bestmse = left;
        *distortion = thismse;
        *sse1 = sse;
    }

    this_mv.as_mv.col += 8;
    thismse = vfp->svf_halfpix_h(y, y_stride, z, b->src_stride, &sse);
    right = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (right < bestmse)
    {
        *bestmv = this_mv;
        bestmse = right;
        *distortion = thismse;
        *sse1 = sse;
    }

    /* go up then down and check error */
    this_mv.as_mv.col = startmv.as_mv.col;
    this_mv.as_mv.row = ((startmv.as_mv.row - 8) | 4);
    thismse = vfp->svf_halfpix_v(y - y_stride, y_stride, z, b->src_stride, &sse);
    up = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (up < bestmse)
    {
        *bestmv = this_mv;
        bestmse = up;
        *distortion = thismse;
        *sse1 = sse;
    }

    this_mv.as_mv.row += 8;
    thismse = vfp->svf_halfpix_v(y, y_stride, z, b->src_stride, &sse);
    down = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (down < bestmse)
    {
        *bestmv = this_mv;
        bestmse = down;
        *distortion = thismse;
        *sse1 = sse;
    }

    /* now check 1 more diagonal - */
    whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
    this_mv = startmv;

    switch (whichdir)
    {
    case 0:
        this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
        this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
        thismse = vfp->svf_halfpix_hv(y - 1 - y_stride, y_stride, z, b->src_stride, &sse);
        break;
    case 1:
        this_mv.as_mv.col += 4;
        this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
        thismse = vfp->svf_halfpix_hv(y - y_stride, y_stride, z, b->src_stride, &sse);
        break;
    case 2:
        this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
        this_mv.as_mv.row += 4;
        thismse = vfp->svf_halfpix_hv(y - 1, y_stride, z, b->src_stride, &sse);
        break;
    case 3:
    default:
        this_mv.as_mv.col += 4;
        this_mv.as_mv.row += 4;
        thismse = vfp->svf_halfpix_hv(y, y_stride, z, b->src_stride, &sse);
        break;
    }

    diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);

    if (diag < bestmse)
    {
        *bestmv = this_mv;
        bestmse = diag;
        *distortion = thismse;
        *sse1 = sse;
    }

    return bestmse;
}

#define CHECK_BOUNDS(range) \
{\
    all_in = 1;\
    all_in &= ((br-range) >= x->mv_row_min);\
    all_in &= ((br+range) <= x->mv_row_max);\
    all_in &= ((bc-range) >= x->mv_col_min);\
    all_in &= ((bc+range) <= x->mv_col_max);\
}

#define CHECK_POINT \
{\
    if (this_mv.as_mv.col < x->mv_col_min) continue;\
    if (this_mv.as_mv.col > x->mv_col_max) continue;\
    if (this_mv.as_mv.row < x->mv_row_min) continue;\
    if (this_mv.as_mv.row > x->mv_row_max) continue;\
}

#define CHECK_BETTER \
{\
    if (thissad < bestsad)\
    {\
        thissad += mvsad_err_cost(&this_mv, &fcenter_mv, mvsadcost, sad_per_bit);\
        if (thissad < bestsad)\
        {\
            bestsad = thissad;\
            best_site = i;\
        }\
    }\
}

static const MV next_chkpts[6][3] =
{
    {{ -2, 0}, { -1, -2}, {1, -2}},
    {{ -1, -2}, {1, -2}, {2, 0}},
    {{1, -2}, {2, 0}, {1, 2}},
    {{2, 0}, {1, 2}, { -1, 2}},
    {{1, 2}, { -1, 2}, { -2, 0}},
    {{ -1, 2}, { -2, 0}, { -1, -2}}
};

int vp8_hex_search
(
    MACROBLOCK *x,
    BLOCK *b,
    BLOCKD *d,
    int_mv *ref_mv,
    int_mv *best_mv,
    int search_param,
    int sad_per_bit,
    const vp8_variance_fn_ptr_t *vfp,
    int *mvsadcost[2],
    int *mvcost[2],
    int_mv *center_mv
)
{
    MV hex[6] = { { -1, -2}, {1, -2}, {2, 0}, {1, 2}, { -1, 2}, { -2, 0} } ;
    MV neighbors[4] = {{0, -1}, { -1, 0}, {1, 0}, {0, 1}} ;
    int i, j;

    unsigned char *what = (*(b->base_src) + b->src);
    int what_stride = b->src_stride;
    int pre_stride = x->e_mbd.pre.y_stride;
    unsigned char *base_pre = x->e_mbd.pre.y_buffer;

    int in_what_stride = pre_stride;
    int br, bc;
    int_mv this_mv;
    unsigned int bestsad;
    unsigned int thissad;
    unsigned char *base_offset;
    unsigned char *this_offset;
    int k = -1;
    int all_in;
    int best_site = -1;
    int hex_range = 127;
    int dia_range = 8;

    int_mv fcenter_mv;
    fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
    fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;

    /* adjust ref_mv to make sure it is within MV range */
    vp8_clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
    br = ref_mv->as_mv.row;
    bc = ref_mv->as_mv.col;

    /* Work out the start point for the search */
    base_offset = (unsigned char *)(base_pre + d->offset);
    this_offset = base_offset + (br * (pre_stride)) + bc;
    this_mv.as_mv.row = br;
    this_mv.as_mv.col = bc;
    bestsad = vfp->sdf(what, what_stride, this_offset, in_what_stride, UINT_MAX)
            + mvsad_err_cost(&this_mv, &fcenter_mv, mvsadcost, sad_per_bit);

#if CONFIG_MULTI_RES_ENCODING
    /* Lower search range based on prediction info */
    if (search_param >= 6) goto cal_neighbors;
    else if (search_param >= 5) hex_range = 4;
    else if (search_param >= 4) hex_range = 6;
    else if (search_param >= 3) hex_range = 15;
    else if (search_param >= 2) hex_range = 31;
    else if (search_param >= 1) hex_range = 63;

    dia_range = 8;
#endif

    /* hex search */
    CHECK_BOUNDS(2)

    if(all_in)
    {
        for (i = 0; i < 6; i++)
        {
            this_mv.as_mv.row = br + hex[i].row;
            this_mv.as_mv.col = bc + hex[i].col;
            this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) + this_mv.as_mv.col;
            thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
            CHECK_BETTER
        }
    }else
    {
        for (i = 0; i < 6; i++)
        {
            this_mv.as_mv.row = br + hex[i].row;
            this_mv.as_mv.col = bc + hex[i].col;
            CHECK_POINT
            this_offset = base_offset + (this_mv.as_mv.row * in_what_stride) + this_mv.as_mv.col;
            thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
            CHECK_BETTER
        }
    }

    if (best_site == -1)
        goto cal_neighbors;
    else
    {
        br += hex[best_site].row;
        bc += hex[best_site].col;
        k = best_site;
    }

    for (j = 1; j < hex_range; j++)
    {
        best_site = -1;
        CHECK_BOUNDS(2)

        if(all_in)
        {
            for (i = 0; i < 3; i++)
            {
                this_mv.as_mv.row = br + next_chkpts[k][i].row;
                this_mv.as_mv.col = bc + next_chkpts[k][i].col;
                this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) + this_mv.as_mv.col;
                thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
                CHECK_BETTER
            }
        }else
        {
            for (i = 0; i < 3; i++)
            {
                this_mv.as_mv.row = br + next_chkpts[k][i].row;
                this_mv.as_mv.col = bc + next_chkpts[k][i].col;
                CHECK_POINT
                this_offset = base_offset + (this_mv.as_mv.row * (in_what_stride)) + this_mv.as_mv.col;
                thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride, bestsad);
                CHECK_BETTER
            }
        }

        if (best_site == -1)
            break;
        else
        {
            br += next_chkpts[k][best_site].row;
            bc += next_chkpts[k][best_site].col;
            k += 5 + best_site;
            if (k >= 12