*/
+#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"
#ifdef ENTROPY_STATS
static int mv_ref_ct [31] [4] [2];
static int mv_err_cost(int_mv *mv, int_mv *ref, int *mvcost[2], int error_per_bit)
{
- 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;
+ // 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. */
- 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;
+ // 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)
x->searches_per_step = 8;
}
-
-#define MVC(r,c) (((mvcost[0][(r)-rr] + mvcost[1][(c) - rc]) * error_per_bit + 128 )>>8 ) // estimated cost of a motion vector (r,c)
-#define PRE(r,c) (*(d->base_pre) + d->pre + ((r)>>2) * d->pre_stride + ((c)>>2)) // pointer to predictor base of a motionvector
+/*
+ * 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.
+ */
+#define MVC(r,c) (mvcost ? ((mvcost[0][(r)-rr] + mvcost[1][(c) - rc]) * error_per_bit + 128 )>>8 : 0) // estimated cost of a motion vector (r,c)
+#define PRE(r,c) (y + (((r)>>2) * y_stride + ((c)>>2) -(offset))) // pointer to predictor base of a motionvector
#define SP(x) (((x)&3)<<1) // convert motion vector component to offset for svf calc
-#define DIST(r,c) vfp->svf( PRE(r,c), d->pre_stride, SP(c),SP(r), z,b->src_stride,&sse) // 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) // returns subpixel variance error function.
#define IFMVCV(r,c,s,e) if ( c >= minc && c <= maxc && r >= minr && r <= maxr) s else e;
#define ERR(r,c) (MVC(r,c)+DIST(r,c)) // returns distortion + motion vector cost
#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=INT_MAX;)// checks if (r,c) has better score than previous best
-#define MIN(x,y) (((x)<(y))?(x):(y))
-#define MAX(x,y) (((x)>(y))?(x):(y))
-
-//#define CHECK_BETTER(v,r,c) if((v = ERR(r,c)) < besterr) { besterr = v; br=r; bc=c; }
int vp8_find_best_sub_pixel_step_iteratively(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
int_mv *bestmv, int_mv *ref_mv,
int *mvcost[2], int *distortion,
unsigned int *sse1)
{
- unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
unsigned char *z = (*(b->base_src) + b->src);
int rr = ref_mv->as_mv.row >> 1, rc = ref_mv->as_mv.col >> 1;
int minr = MAX(x->mv_row_min << 2, (ref_mv->as_mv.row >> 1) - ((1 << mvlong_width) - 1));
int maxr = MIN(x->mv_row_max << 2, (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 *y0 = base_pre + d->offset + (bestmv->as_mv.row) * pre_stride + bestmv->as_mv.col;
+ unsigned char *y;
+ int buf_r1, buf_r2, buf_c1, buf_c2;
+
+ // 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;
+ buf_c2 = ((bestmv->as_mv.col + 3) > x->mv_col_max)?(x->mv_col_max - bestmv->as_mv.col):3;
+ y_stride = 32;
+
+ /* Copy to intermediate buffer before searching. */
+ vfp->copymem(y0 - 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 <<= 3;
bestmv->as_mv.col <<= 3;
// calculate central point error
- besterr = vfp->vf(y, d->pre_stride, z, b->src_stride, sse1);
+ besterr = vfp->vf(y, y_stride, z, b->src_stride, sse1);
*distortion = besterr;
besterr += mv_err_cost(bestmv, ref_mv, mvcost, error_per_bit);
#undef PRE
#undef SP
#undef DIST
+#undef IFMVCV
#undef ERR
#undef CHECK_BETTER
-#undef MIN
-#undef MAX
+
int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
int_mv *bestmv, int_mv *ref_mv,
int error_per_bit,
int bestmse = INT_MAX;
int_mv startmv;
int_mv this_mv;
- unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
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 *y0 = 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(y0 - 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 <<= 3;
startmv = *bestmv;
// calculate central point error
- bestmse = vfp->vf(y, d->pre_stride, z, b->src_stride, sse1);
+ 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, d->pre_stride, z, b->src_stride, &sse);
+ 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)
}
this_mv.as_mv.col += 8;
- thismse = vfp->svf_halfpix_h(y, d->pre_stride, z, b->src_stride, &sse);
+ 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)
// 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 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+ 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)
}
this_mv.as_mv.row += 8;
- thismse = vfp->svf_halfpix_v(y, d->pre_stride, z, b->src_stride, &sse);
+ 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)
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 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+ 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 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+ 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, d->pre_stride, z, b->src_stride, &sse);
+ 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, d->pre_stride, z, b->src_stride, &sse);
+ thismse = vfp->svf_halfpix_hv(y, y_stride, z, b->src_stride, &sse);
break;
}
// time to check quarter pels.
if (bestmv->as_mv.row < startmv.as_mv.row)
- y -= d->pre_stride;
+ y -= y_stride;
if (bestmv->as_mv.col < startmv.as_mv.col)
y--;
if (startmv.as_mv.col & 7)
{
this_mv.as_mv.col = startmv.as_mv.col - 2;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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, d->pre_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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);
}
this_mv.as_mv.col += 4;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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)
if (startmv.as_mv.row & 7)
{
this_mv.as_mv.row = startmv.as_mv.row - 2;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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 - d->pre_stride, d->pre_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
+ 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);
}
this_mv.as_mv.row += 4;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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)
if (startmv.as_mv.col & 7)
{
this_mv.as_mv.col -= 2;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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, d->pre_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);;
+ thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);;
}
}
else
if (startmv.as_mv.col & 7)
{
this_mv.as_mv.col -= 2;
- thismse = vfp->svf(y - d->pre_stride, d->pre_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
+ 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 - d->pre_stride - 1, d->pre_stride, 6, 6, z, b->src_stride, &sse);
+ thismse = vfp->svf(y - y_stride - 1, y_stride, 6, 6, z, b->src_stride, &sse);
}
}
if (startmv.as_mv.row & 7)
{
this_mv.as_mv.row -= 2;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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 - d->pre_stride, d->pre_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
+ thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
}
break;
if (startmv.as_mv.col & 7)
{
this_mv.as_mv.col -= 2;
- thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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, d->pre_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ 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, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+ thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
break;
}
return bestmse;
}
-int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
+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 bestmse = INT_MAX;
int_mv startmv;
int_mv this_mv;
- unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
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 *y0 = 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(y0 - 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 <<= 3;
startmv = *bestmv;
// calculate central point error
- bestmse = vfp->vf(y, d->pre_stride, z, b->src_stride, sse1);
+ 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, d->pre_stride, z, b->src_stride, &sse);
+ 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)
}
this_mv.as_mv.col += 8;
- thismse = vfp->svf_halfpix_h(y, d->pre_stride, z, b->src_stride, &sse);
+ 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)
// 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 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+ 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)
}
this_mv.as_mv.row += 8;
- thismse = vfp->svf_halfpix_v(y, d->pre_stride, z, b->src_stride, &sse);
+ 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)
*sse1 = sse;
}
- // somewhat strangely not doing all the diagonals for half pel is slower than doing them.
-#if 0
// now check 1 more diagonal -
whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
this_mv = startmv;
switch (whichdir)
{
case 0:
- this_mv.col = (this_mv.col - 8) | 4;
- this_mv.row = (this_mv.row - 8) | 4;
- diag = vfp->svf(y - 1 - d->pre_stride, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+ 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.col += 4;
- this_mv.row = (this_mv.row - 8) | 4;
- diag = vfp->svf(y - d->pre_stride, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+ 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.col = (this_mv.col - 8) | 4;
- this_mv.row += 4;
- diag = vfp->svf(y - 1, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+ 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:
- this_mv.col += 4;
- this_mv.row += 4;
- diag = vfp->svf(y, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+ 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 += mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
-
- if (diag < bestmse)
- {
- *bestmv = this_mv;
- bestmse = diag;
- }
-
-#else
- 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 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
- 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;
- }
-
- this_mv.as_mv.col += 8;
- thismse = vfp->svf_halfpix_hv(y - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
if (diag < bestmse)
*sse1 = sse;
}
- this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
- this_mv.as_mv.row = startmv.as_mv.row + 4;
- thismse = vfp->svf_halfpix_hv(y - 1, d->pre_stride, z, b->src_stride, &sse);
- 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;
- }
-
- this_mv.as_mv.col += 8;
- thismse = vfp->svf_halfpix_hv(y, d->pre_stride, z, b->src_stride, &sse);
- 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;
- }
-
-#endif
return bestmse;
}
unsigned char *what = (*(b->base_src) + b->src);
int what_stride = b->src_stride;
- int in_what_stride = d->pre_stride;
- int br = ref_mv->as_mv.row, bc = ref_mv->as_mv.col;
+ 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 = 0x7fffffff;
unsigned int thissad;
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 *)(*(d->base_pre) + d->pre);
- this_offset = base_offset + (br * (d->pre_stride)) + bc;
+ 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, 0x7fffffff)
+ 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
//j=0
CHECK_BOUNDS(2)
k = best_site;
}
- for (j = 1; j < 127; j++)
+ for (j = 1; j < hex_range; j++)
{
best_site = -1;
CHECK_BOUNDS(2)
// check 4 1-away neighbors
cal_neighbors:
- for (j = 0; j < 32; j++)
+ for (j = 0; j < dia_range; j++)
{
best_site = -1;
CHECK_BOUNDS(1)
#undef CHECK_POINT
#undef CHECK_BETTER
-int vp8_diamond_search_sad
+int vp8_diamond_search_sad_c
(
MACROBLOCK *x,
BLOCK *b,
unsigned char *what = (*(b->base_src) + b->src);
int what_stride = b->src_stride;
unsigned char *in_what;
- int in_what_stride = d->pre_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;
unsigned char *best_address;
int tot_steps;
int best_site = 0;
int last_site = 0;
- int ref_row = ref_mv->as_mv.row;
- int ref_col = ref_mv->as_mv.col;
+ int ref_row;
+ int ref_col;
int this_row_offset;
int this_col_offset;
search_site *ss;
fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+ vp8_clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+ ref_row = ref_mv->as_mv.row;
+ ref_col = ref_mv->as_mv.col;
*num00 = 0;
-
best_mv->as_mv.row = ref_row;
best_mv->as_mv.col = ref_col;
// Work out the start point for the search
- in_what = (unsigned char *)(*(d->base_pre) + d->pre + (ref_row * (d->pre_stride)) + ref_col);
+ in_what = (unsigned char *)(base_pre + d->offset + (ref_row * pre_stride) + ref_col);
best_address = in_what;
// Check the starting position
unsigned char *what = (*(b->base_src) + b->src);
int what_stride = b->src_stride;
unsigned char *in_what;
- int in_what_stride = d->pre_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;
unsigned char *best_address;
int tot_steps;
int_mv this_mv;
- int bestsad = INT_MAX;
+ unsigned int bestsad = UINT_MAX;
int best_site = 0;
int last_site = 0;
- int ref_row = ref_mv->as_mv.row;
- int ref_col = ref_mv->as_mv.col;
+ int ref_row;
+ int ref_col;
int this_row_offset;
int this_col_offset;
search_site *ss;
fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
+ vp8_clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+ ref_row = ref_mv->as_mv.row;
+ ref_col = ref_mv->as_mv.col;
*num00 = 0;
best_mv->as_mv.row = ref_row;
best_mv->as_mv.col = ref_col;
// Work out the start point for the search
- in_what = (unsigned char *)(*(d->base_pre) + d->pre + (ref_row * (d->pre_stride)) + ref_col);
+ in_what = (unsigned char *)(base_pre + d->offset + (ref_row * pre_stride) + ref_col);
best_address = in_what;
// Check the starting position
+ mv_err_cost(&this_mv, center_mv, mvcost, x->errorperbit);
}
-int vp8_full_search_sad(MACROBLOCK *x, BLOCK *b, BLOCKD *d, int_mv *ref_mv,
+int vp8_full_search_sad_c(MACROBLOCK *x, BLOCK *b, BLOCKD *d, int_mv *ref_mv,
int sad_per_bit, int distance,
vp8_variance_fn_ptr_t *fn_ptr, int *mvcost[2],
int_mv *center_mv)
unsigned char *what = (*(b->base_src) + b->src);
int what_stride = b->src_stride;
unsigned char *in_what;
- int in_what_stride = d->pre_stride;
- int mv_stride = d->pre_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 mv_stride = pre_stride;
unsigned char *bestaddress;
int_mv *best_mv = &d->bmi.mv;
int_mv this_mv;
fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
// Work out the mid point for the search
- in_what = *(d->base_pre) + d->pre;
- bestaddress = in_what + (ref_row * d->pre_stride) + ref_col;
+ in_what = base_pre + d->offset;
+ bestaddress = in_what + (ref_row * pre_stride) + ref_col;
best_mv->as_mv.row = ref_row;
best_mv->as_mv.col = ref_col;
unsigned char *what = (*(b->base_src) + b->src);
int what_stride = b->src_stride;
unsigned char *in_what;
- int in_what_stride = d->pre_stride;
- int mv_stride = d->pre_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 mv_stride = pre_stride;
unsigned char *bestaddress;
int_mv *best_mv = &d->bmi.mv;
int_mv this_mv;
- int bestsad = INT_MAX;
+ unsigned int bestsad = UINT_MAX;
int r, c;
unsigned char *check_here;
fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
// Work out the mid point for the search
- in_what = *(d->base_pre) + d->pre;
- bestaddress = in_what + (ref_row * d->pre_stride) + ref_col;
+ in_what = base_pre + d->offset;
+ bestaddress = in_what + (ref_row * pre_stride) + ref_col;
best_mv->as_mv.row = ref_row;
best_mv->as_mv.col = ref_col;
{
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;
unsigned char *in_what;
- int in_what_stride = d->pre_stride;
- int mv_stride = d->pre_stride;
+ int in_what_stride = pre_stride;
+ int mv_stride = pre_stride;
unsigned char *bestaddress;
int_mv *best_mv = &d->bmi.mv;
int_mv this_mv;
- int bestsad = INT_MAX;
+ unsigned int bestsad = UINT_MAX;
int r, c;
unsigned char *check_here;
fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
// Work out the mid point for the search
- in_what = *(d->base_pre) + d->pre;
- bestaddress = in_what + (ref_row * d->pre_stride) + ref_col;
+ in_what = base_pre + d->offset;
+ bestaddress = in_what + (ref_row * pre_stride) + ref_col;
best_mv->as_mv.row = ref_row;
best_mv->as_mv.col = ref_col;
return INT_MAX;
}
-int vp8_refining_search_sad(MACROBLOCK *x, BLOCK *b, BLOCKD *d, int_mv *ref_mv,
+int vp8_refining_search_sad_c(MACROBLOCK *x, BLOCK *b, BLOCKD *d, int_mv *ref_mv,
int error_per_bit, int search_range,
vp8_variance_fn_ptr_t *fn_ptr, int *mvcost[2],
int_mv *center_mv)
short this_row_offset, this_col_offset;
int what_stride = b->src_stride;
- int in_what_stride = d->pre_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;
unsigned char *what = (*(b->base_src) + b->src);
- unsigned char *best_address = (unsigned char *)(*(d->base_pre) + d->pre +
- (ref_mv->as_mv.row * (d->pre_stride)) + ref_mv->as_mv.col);
+ unsigned char *best_address = (unsigned char *)(base_pre + d->offset +
+ (ref_mv->as_mv.row * pre_stride) + ref_mv->as_mv.col);
unsigned char *check_here;
unsigned int thissad;
int_mv this_mv;
short this_row_offset, this_col_offset;
int what_stride = b->src_stride;
- int in_what_stride = d->pre_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;
unsigned char *what = (*(b->base_src) + b->src);
- unsigned char *best_address = (unsigned char *)(*(d->base_pre) + d->pre +
- (ref_mv->as_mv.row * (d->pre_stride)) + ref_mv->as_mv.col);
+ unsigned char *best_address = (unsigned char *)(base_pre + d->offset +
+ (ref_mv->as_mv.row * pre_stride) + ref_mv->as_mv.col);
unsigned char *check_here;
unsigned int thissad;
int_mv this_mv;
projects / profile / ivi / libvpx.git / blobdiff