Commit 9fdd90c9 authored by John Koleszar's avatar John Koleszar Committed by Code Review

Merge "arm: remove duplicate functions"

parents 91d927f9 d330a587
......@@ -295,22 +295,8 @@ void vp8_predict_intra4x4(BLOCKD *x,
}
}
// copy 4 bytes from the above right down so that the 4x4 prediction modes using pixels above and
// to the right prediction have filled in pixels to use.
void vp8_intra_prediction_down_copy(MACROBLOCKD *x)
{
unsigned char *above_right = *(x->block[0].base_dst) + x->block[0].dst - x->block[0].dst_stride + 16;
unsigned int *src_ptr = (unsigned int *)above_right;
unsigned int *dst_ptr0 = (unsigned int *)(above_right + 4 * x->block[0].dst_stride);
unsigned int *dst_ptr1 = (unsigned int *)(above_right + 8 * x->block[0].dst_stride);
unsigned int *dst_ptr2 = (unsigned int *)(above_right + 12 * x->block[0].dst_stride);
*dst_ptr0 = *src_ptr;
*dst_ptr1 = *src_ptr;
*dst_ptr2 = *src_ptr;
}
extern void vp8_intra_prediction_down_copy(MACROBLOCKD *x);
/*
......
......@@ -21,8 +21,6 @@ static int mv_ref_ct [31] [4] [2];
static int mv_mode_cts [4] [2];
#endif
static int mv_bits_sadcost[256];
extern unsigned int vp8_sub_pixel_variance16x16s_neon
(
unsigned char *src_ptr,
......@@ -58,292 +56,7 @@ extern unsigned int vp8_sub_pixel_variance16x16s_4_4_neon
unsigned int *sse
);
void vp8cx_init_mv_bits_sadcost()
{
int i;
for (i = 0; i < 256; i++)
{
mv_bits_sadcost[i] = (int)sqrt(i * 16);
}
}
int vp8_mv_bit_cost(MV *mv, 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->row - ref->row) >> 1] + mvcost[1][(mv->col - ref->col) >> 1]) * Weight) >> 7;
}
int vp8_mv_err_cost(MV *mv, MV *ref, int *mvcost[2], int error_per_bit)
{
//int i;
//return ((mvcost[0][(mv->row - ref->row)>>1] + mvcost[1][(mv->col - ref->col)>>1] + 128) * error_per_bit) >> 8;
//return ( (vp8_mv_bit_cost(mv, ref, mvcost, 100) + 128) * error_per_bit) >> 8;
//i = (vp8_mv_bit_cost(mv, ref, mvcost, 100) * error_per_bit + 128) >> 8;
return ((mvcost[0][(mv->row - ref->row) >> 1] + mvcost[1][(mv->col - ref->col) >> 1]) * error_per_bit + 128) >> 8;
//return (vp8_mv_bit_cost(mv, ref, mvcost, 128) * error_per_bit + 128) >> 8;
}
static int mv_bits(MV *mv, MV *ref, int *mvcost[2])
{
// get the estimated number of bits for a motion vector, to be used for costing in SAD based
// motion estimation
return ((mvcost[0][(mv->row - ref->row) >> 1] + mvcost[1][(mv->col - ref->col)>> 1]) + 128) >> 8;
}
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;
}
#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
#define SP(x) (((x)&3)<<1) // convert motion vector component to offset for svf calc
#define DIST(r,c) svf( PRE(r,c), d->pre_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,{if((v = ERR(r,c)) < besterr) { besterr = v; br=r; bc=c; }}, 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, MV *bestmv, MV *ref_mv, int error_per_bit, vp8_subpixvariance_fn_t svf, vp8_variance_fn_t vf, int *mvcost[2])
{
unsigned char *y = *(d->base_pre) + d->pre + (bestmv->row) * d->pre_stride + bestmv->col;
unsigned char *z = (*(b->base_src) + b->src);
int rr = ref_mv->row >> 1, rc = ref_mv->col >> 1;
int br = bestmv->row << 2, bc = bestmv->col << 2;
int tr = br, tc = bc;
unsigned int besterr = INT_MAX;
unsigned int left, right, up, down, diag;
unsigned int sse;
unsigned int whichdir;
unsigned int halfiters = 4;
unsigned int quarteriters = 4;
int minc = MAX(x->mv_col_min << 2, (ref_mv->col >> 1) - ((1 << mvlong_width) - 1));
int maxc = MIN(x->mv_col_max << 2, (ref_mv->col >> 1) + ((1 << mvlong_width) - 1));
int minr = MAX(x->mv_row_min << 2, (ref_mv->row >> 1) - ((1 << mvlong_width) - 1));
int maxr = MIN(x->mv_row_max << 2, (ref_mv->row >> 1) + ((1 << mvlong_width) - 1));
// central mv
bestmv->row <<= 3;
bestmv->col <<= 3;
// calculate central point error
besterr = vf(y, d->pre_stride, z, b->src_stride, &sse);
besterr += vp8_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->row = br << 1;
bestmv->col = bc << 1;
if ((abs(bestmv->col - ref_mv->col) > MAX_FULL_PEL_VAL) || (abs(bestmv->row - ref_mv->row) > MAX_FULL_PEL_VAL))
return INT_MAX;
return besterr;
}
#undef MVC
#undef PRE
#undef SP
#undef DIST
#undef ERR
#undef CHECK_BETTER
#undef MIN
#undef MAX
int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d, MV *bestmv, MV *ref_mv, int error_per_bit, vp8_subpixvariance_fn_t svf, vp8_variance_fn_t vf, int *mvcost[2])
{
int bestmse = INT_MAX;
......@@ -935,335 +648,6 @@ cal_neighbors:
#undef ERR
#undef CHECK_BETTER
int vp8_diamond_search_sad
(
MACROBLOCK *x,
BLOCK *b,
BLOCKD *d,
MV *ref_mv,
MV *best_mv,
int search_param,
int error_per_bit,
int *num00,
vp8_variance_fn_ptr_t *fn_ptr,
int *mvsadcost[2],
int *mvcost[2]
)
{
int i, j, step;
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;
unsigned char *best_address;
int tot_steps;
MV this_mv;
int bestsad = INT_MAX;
int best_site = 0;
int last_site = 0;
int ref_row = ref_mv->row >> 3;
int ref_col = ref_mv->col >> 3;
int this_row_offset;
int this_col_offset;
search_site *ss;
unsigned char *check_here;
int thissad;
// Work out the start point for the search
in_what = (unsigned char *)(*(d->base_pre) + d->pre + (ref_row * (d->pre_stride)) + ref_col);
best_address = in_what;
// We need to check that the starting point for the search (as indicated by ref_mv) is within the buffer limits
if ((ref_col > x->mv_col_min) && (ref_col < x->mv_col_max) &&
(ref_row > x->mv_row_min) && (ref_row < x->mv_row_max))
{
// Check the starting position
bestsad = fn_ptr->sdf(what, what_stride, in_what, in_what_stride, 0x7fffffff) + vp8_mv_err_cost(ref_mv, ref_mv, mvsadcost, error_per_bit);
}
// search_param determines the length of the initial step and hence the number of iterations
// 0 = initial step (MAX_FIRST_STEP) pel : 1 = (MAX_FIRST_STEP/2) pel, 2 = (MAX_FIRST_STEP/4) pel... etc.
ss = &x->ss[search_param * x->searches_per_step];
tot_steps = (x->ss_count / x->searches_per_step) - search_param;
i = 1;
best_mv->row = ref_row;
best_mv->col = ref_col;
*num00 = 0;
for (step = 0; step < tot_steps ; step++)
{
for (j = 0 ; j < x->searches_per_step ; j++)
{
// Trap illegal vectors
this_row_offset = best_mv->row + ss[i].mv.row;
this_col_offset = best_mv->col + ss[i].mv.col;
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))
{
check_here = ss[i].offset + best_address;
thissad = fn_ptr->sdf(what, what_stride, check_here , in_what_stride, bestsad);
if (thissad < 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)
{
bestsad = thissad;
best_site = i;
}
}
}
i++;
}
if (best_site != last_site)
{
best_mv->row += ss[best_site].mv.row;
best_mv->col += ss[best_site].mv.col;
best_address += ss[best_site].offset;
last_site = best_site;
}
else if (best_address == in_what)
(*num00)++;
}
this_mv.row = best_mv->row << 3;
this_mv.col = best_mv->col << 3;
if (bestsad == INT_MAX)
return INT_MAX;
return fn_ptr->vf(what, what_stride, best_address, in_what_stride, (unsigned int *)(&thissad))
+ vp8_mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
}
#if !(CONFIG_REALTIME_ONLY)
int vp8_full_search_sad(MACROBLOCK *x, BLOCK *b, BLOCKD *d, MV *ref_mv, int error_per_bit, int distance, vp8_variance_fn_ptr_t *fn_ptr, int *mvcost[2], int *mvsadcost[2])
{
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;
unsigned char *bestaddress;
MV *best_mv = &d->bmi.mv.as_mv;
MV this_mv;
int bestsad = INT_MAX;
int r, c;
unsigned char *check_here;
int thissad;
int ref_row = ref_mv->row >> 3;
int ref_col = ref_mv->col >> 3;
int row_min = ref_row - distance;
int row_max = ref_row + distance;
int col_min = ref_col - distance;
int col_max = ref_col + distance;
// 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;
best_mv->row = ref_row;
best_mv->col = ref_col;
// We need to check that the starting point for the search (as indicated by ref_mv) is within the buffer limits
if ((ref_col > x->mv_col_min) && (ref_col < x->mv_col_max) &&
(ref_row > x->mv_row_min) && (ref_row < x->mv_row_max))
{
// Baseline value at the centre
//bestsad = fn_ptr->sf( what,what_stride,bestaddress,in_what_stride) + (int)sqrt(vp8_mv_err_cost(ref_mv,ref_mv, mvcost,error_per_bit*14));
bestsad = fn_ptr->sdf(what, what_stride, bestaddress, in_what_stride, 0x7fffffff) + vp8_mv_err_cost(ref_mv, ref_mv, mvsadcost, error_per_bit);
}
// Apply further limits to prevent us looking using vectors that stretch beyiond the UMV border
if (col_min < x->mv_col_min)
col_min = x->mv_col_min;
if (col_max > x->mv_col_max)
col_max = x->mv_col_max;
if (row_min < x->mv_row_min)
row_min = x->mv_row_min;
if (row_max > x->mv_row_max)
row_max = x->mv_row_max;
for (r = row_min; r < row_max ; r++)
{
this_mv.row = r << 3;
check_here = r * mv_stride + in_what + col_min;
for (c = col_min; c < col_max; c++)
{
thissad = fn_ptr->sdf(what, what_stride, check_here , in_what_stride, bestsad);
this_mv.col = c << 3;
//thissad += (int)sqrt(vp8_mv_err_cost(&this_mv,ref_mv, mvcost,error_per_bit*14));
//thissad += error_per_bit * mv_bits_sadcost[mv_bits(&this_mv, ref_mv, mvcost)];
thissad += vp8_mv_err_cost(&this_mv, ref_mv, mvsadcost, error_per_bit); //mv_bits(error_per_bit, &this_mv, ref_mv, mvsadcost);
if (thissad < bestsad)
{
bestsad = thissad;
best_mv->row = r;
best_mv->col = c;
bestaddress = check_here;
}
check_here++;
}
}
this_mv.row = best_mv->row << 3;
this_mv.col = best_mv->col << 3;
if (bestsad < INT_MAX)
return fn_ptr->vf(what, what_stride, bestaddress, in_what_stride, (unsigned int *)(&thissad))
+ vp8_mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
else
return INT_MAX;
}
int vp8_full_search_sadx3(MACROBLOCK *x, BLOCK *b, BLOCKD *d, MV *ref_mv, int error_per_bit, int distance, vp8_variance_fn_ptr_t *fn_ptr, int *mvcost[2], int *mvsadcost[2])
{
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;
unsigned char *bestaddress;
MV *best_mv = &d->bmi.mv.as_mv;
MV this_mv;
int bestsad = INT_MAX;
int r, c;
unsigned char *check_here;
unsigned int thissad;
int ref_row = ref_mv->row >> 3;
int ref_col = ref_mv->col >> 3;
int row_min = ref_row - distance;
int row_max = ref_row + distance;
int col_min = ref_col - distance;
int col_max = ref_col + distance;
unsigned int sad_array[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;
best_mv->row = ref_row;
best_mv->col = ref_col;
// We need to check that the starting point for the search (as indicated by ref_mv) is within the buffer limits
if ((ref_col > x->mv_col_min) && (ref_col < x->mv_col_max) &&
(ref_row > x->mv_row_min) && (ref_row < x->mv_row_max))
{
// Baseline value at the centre
bestsad = fn_ptr->sdf(what, what_stride, bestaddress, in_what_stride, 0x7fffffff) + vp8_mv_err_cost(ref_mv, ref_mv, mvsadcost, error_per_bit);
}
// Apply further limits to prevent us looking using vectors that stretch beyiond the UMV border
if (col_min < x->mv_col_min)
col_min = x->mv_col_min;
if (col_max > x->mv_col_max)
col_max = x->mv_col_max;
if (row_min < x->mv_row_min)
row_min = x->mv_row_min;
if (row_max > x->mv_row_max)
row_max = x->mv_row_max;
for (r = row_min; r < row_max ; r++)
{
this_mv.row = r << 3;
check_here = r * mv_stride + in_what + col_min;
c = col_min;
while ((c + 3) < col_max)
{
int i;
fn_ptr->sdx3f(what, what_stride, check_here , in_what_stride, sad_array);
for (i = 0; i < 3; i++)
{
thissad = sad_array[i];
if (thissad < bestsad)
{
this_mv.col = c << 3;
thissad += vp8_mv_err_cost(&this_mv, ref_mv, mvsadcost, error_per_bit);
if (thissad < bestsad)
{
bestsad = thissad;
best_mv->row = r;
best_mv->col = c;
bestaddress = check_here;
}
}
check_here++;
c++;
}
}