Commit d6da7b8e authored by Yunqing Wang's avatar Yunqing Wang
Browse files

Improve bounds checking in vp8_diamond_search_sadx4()

In order to know if all 4/8 neighbor points are within the bounds,
4 bounds checking are enough instead of checking 4 bounds for
each points (16/32 checkings). This improvement reduces cost of
vp8_diamond_search_sadx4() by 30%, and gives encoder a 1.5%
performance gain (test options: 1 pass, good, speed=4).

Change-Id: Ie8da29d18a6ecfc9829e74ac02f6fa70e042331a
parent e50f5d40
......@@ -1035,84 +1035,73 @@ int vp8_diamond_search_sadx4
for (step = 0; step < tot_steps ; step++)
{
int check_row_min, check_col_min, check_row_max, check_col_max;
int all_in = 1, t;
check_row_min = x->mv_row_min - best_mv->row;
check_row_max = x->mv_row_max - best_mv->row;
check_col_min = x->mv_col_min - best_mv->col;
check_col_max = x->mv_col_max - best_mv->col;
// To know if all neighbor points are within the bounds, 4 bounds checking are enough instead of
// checking 4 bounds for each points.
all_in &= ((best_mv->row + ss[i].mv.row)> x->mv_row_min);
all_in &= ((best_mv->row + ss[i+1].mv.row) < x->mv_row_max);
all_in &= ((best_mv->col + ss[i+2].mv.col) > x->mv_col_min);
all_in &= ((best_mv->col + ss[i+3].mv.col) < x->mv_col_max);
for (j = 0 ; j < x->searches_per_step ; j += 4)
if (all_in)
{
unsigned char *block_offset[4];
unsigned int valid_block[4];
int all_in = 1, t;
unsigned int sad_array[4];
for (t = 0; t < 4; t++)
for (j = 0 ; j < x->searches_per_step ; j += 4)
{
valid_block [t] = (ss[t+i].mv.col > check_col_min);
valid_block [t] &= (ss[t+i].mv.col < check_col_max);
valid_block [t] &= (ss[t+i].mv.row > check_row_min);
valid_block [t] &= (ss[t+i].mv.row < check_row_max);
unsigned char *block_offset[4];
all_in &= valid_block[t];
block_offset[t] = ss[i+t].offset + best_address;
}
if (all_in)
{
unsigned int sad_array[4];
for (t = 0; t < 4; t++)
block_offset[t] = ss[i+t].offset + best_address;
fn_ptr->sdx4df(what, what_stride, block_offset, in_what_stride, sad_array);
for (t = 0; t < 4; t++, i++)
{
thissad = sad_array[t];
if (thissad < bestsad)
if (sad_array[t] < bestsad)
{
this_mv.row = (best_mv->row + ss[i].mv.row) << 3;
this_mv.col = (best_mv->col + ss[i].mv.col) << 3;
thissad += vp8_mv_err_cost(&this_mv, ref_mv, mvsadcost, error_per_bit);
sad_array[t] += vp8_mv_err_cost(&this_mv, ref_mv, mvsadcost, error_per_bit);
if (thissad < bestsad)
if (sad_array[t] < bestsad)
{
bestsad = thissad;
bestsad = sad_array[t];
best_site = i;
}
}
}
}
else
}
else
{
for (j = 0 ; j < x->searches_per_step ; j++)
{
int t;
// Trap illegal vectors
this_row_offset = best_mv->row + ss[i].mv.row;
this_col_offset = best_mv->col + ss[i].mv.col;
for (t = 0; t < 4; i++, t++)
if ((this_col_offset > x->mv_col_min) && (this_col_offset < x->mv_col_max) &&
(this_row_offset > x->mv_row_min) && (this_row_offset < x->mv_row_max))
{
// Trap illegal vectors
if (valid_block[t])
check_here = ss[i].offset + best_address;
thissad = fn_ptr->sdf(what, what_stride, check_here , in_what_stride, bestsad);
if (thissad < bestsad)
{
check_here = block_offset[t];
thissad = fn_ptr->sdf(what, what_stride, check_here , in_what_stride, bestsad);
this_mv.row = this_row_offset << 3;
this_mv.col = this_col_offset << 3;
thissad += vp8_mv_err_cost(&this_mv, ref_mv, mvsadcost, error_per_bit);
if (thissad < bestsad)
{
this_row_offset = best_mv->row + ss[i].mv.row;
this_col_offset = best_mv->col + ss[i].mv.col;
this_mv.row = this_row_offset << 3;
this_mv.col = this_col_offset << 3;
thissad += vp8_mv_err_cost(&this_mv, ref_mv, mvsadcost, error_per_bit);
if (thissad < bestsad)
{
bestsad = thissad;
best_site = i;
}
bestsad = thissad;
best_site = i;
}
}
}
i++;
}
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment