// Identical to aom_clpf_detect_c() apart from "rec" and "org".
void aom_clpf_detect_hbd_c(const uint16_t *rec, const uint16_t *org,
int rstride, int ostride, int x0, int y0, int width,
int height, int *sum0, int *sum1,
unsigned int strength, int size, unsigned int bd) {
const int shift = bd - 8;
int x, y;
for (y = y0; y < y0 + size; y++) {
for (x = x0; x < x0 + size; x++) {
const int O = org[y * ostride + x] >> shift;
const int X = rec[y * rstride + x] >> shift;
const int A = rec[AOMMAX(0, y - 2) * rstride + x] >> shift;
const int B = rec[AOMMAX(0, y - 1) * rstride + x] >> shift;
const int C = rec[y * rstride + AOMMAX(0, x - 2)] >> shift;
const int D = rec[y * rstride + AOMMAX(0, x - 1)] >> shift;
const int E = rec[y * rstride + AOMMIN(width - 1, x + 1)] >> shift;
const int F = rec[y * rstride + AOMMIN(width - 1, x + 2)] >> shift;
const int G = rec[AOMMIN(height - 1, y + 1) * rstride + x] >> shift;
const int H = rec[AOMMIN(height - 1, y + 2) * rstride + x] >> shift;
const int delta = av1_clpf_sample(X, A, B, C, D, E, F, G, H,
strength >> shift, bd - shift);
const int Y = X + delta;
*sum0 += (O - X) * (O - X);
*sum1 += (O - Y) * (O - Y);
}
}
}
void aom_clpf_detect_multi_c(const uint8_t *rec, const uint8_t *org,
int rstride, int ostride, int x0, int y0,
int width, int height, int *sum, int size,
unsigned int bd) {
int x, y;
for (y = y0; y < y0 + size; y++) {
for (x = x0; x < x0 + size; x++) {
const int O = org[y * ostride + x];
const int X = rec[y * rstride + x];
const int A = rec[AOMMAX(0, y - 2) * rstride + x];
const int B = rec[AOMMAX(0, y - 1) * rstride + x];
const int C = rec[y * rstride + AOMMAX(0, x - 2)];
const int D = rec[y * rstride + AOMMAX(0, x - 1)];
const int E = rec[y * rstride + AOMMIN(width - 1, x + 1)];
const int F = rec[y * rstride + AOMMIN(width - 1, x + 2)];
const int G = rec[AOMMIN(height - 1, y + 1) * rstride + x];
const int H = rec[AOMMIN(height - 1, y + 2) * rstride + x];
const int delta1 = av1_clpf_sample(X, A, B, C, D, E, F, G, H, 1, bd);
const int delta2 = av1_clpf_sample(X, A, B, C, D, E, F, G, H, 2, bd);
const int delta3 = av1_clpf_sample(X, A, B, C, D, E, F, G, H, 4, bd);
const int F1 = X + delta1;
const int F2 = X + delta2;
const int F3 = X + delta3;
sum[0] += (O - X) * (O - X);
sum[1] += (O - F1) * (O - F1);
sum[2] += (O - F2) * (O - F2);
sum[3] += (O - F3) * (O - F3);
}
}
}
void aom_noise_tx_filter(struct aom_noise_tx_t *noise_tx, const float *psd) {
const int block_size = noise_tx->block_size;
const float kBeta = 1.1f;
const float kEps = 1e-6f;
for (int y = 0; y < block_size; ++y) {
for (int x = 0; x < block_size; ++x) {
int i = y * block_size + x;
float *c = noise_tx->tx_block + 2 * i;
const float p = c[0] * c[0] + c[1] * c[1];
if (p > kBeta * psd[i] && p > 1e-6) {
noise_tx->tx_block[2 * i + 0] *= (p - psd[i]) / AOMMAX(p, kEps);
noise_tx->tx_block[2 * i + 1] *= (p - psd[i]) / AOMMAX(p, kEps);
} else {
noise_tx->tx_block[2 * i + 0] *= (kBeta - 1.0f) / kBeta;
noise_tx->tx_block[2 * i + 1] *= (kBeta - 1.0f) / kBeta;
}
}
}
}
static int cost_and_tokenize_map(Av1ColorMapParam *param, TOKENEXTRA **t,
int plane, int calc_rate, int allow_update_cdf,
FRAME_COUNTS *counts, MapCdf map_pb_cdf) {
const uint8_t *const color_map = param->color_map;
MapCdf map_cdf = param->map_cdf;
ColorCost color_cost = param->color_cost;
const int plane_block_width = param->plane_width;
const int rows = param->rows;
const int cols = param->cols;
const int n = param->n_colors;
const int palette_size_idx = n - PALETTE_MIN_SIZE;
int this_rate = 0;
uint8_t color_order[PALETTE_MAX_SIZE];
(void)plane;
(void)counts;
for (int k = 1; k < rows + cols - 1; ++k) {
for (int j = AOMMIN(k, cols - 1); j >= AOMMAX(0, k - rows + 1); --j) {
int i = k - j;
int color_new_idx;
const int color_ctx = av1_get_palette_color_index_context(
color_map, plane_block_width, i, j, n, color_order, &color_new_idx);
assert(color_new_idx >= 0 && color_new_idx < n);
if (calc_rate) {
this_rate += (*color_cost)[palette_size_idx][color_ctx][color_new_idx];
} else {
(*t)->token = color_new_idx;
(*t)->color_map_cdf = map_pb_cdf[palette_size_idx][color_ctx];
++(*t);
if (allow_update_cdf)
update_cdf(map_cdf[palette_size_idx][color_ctx], color_new_idx, n);
#if CONFIG_ENTROPY_STATS
if (plane) {
++counts->palette_uv_color_index[palette_size_idx][color_ctx]
[color_new_idx];
} else {
++counts->palette_y_color_index[palette_size_idx][color_ctx]
[color_new_idx];
}
#endif
}
}
}
if (calc_rate) return this_rate;
return 0;
}
void av1_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
struct macroblockd_plane *planes,
int frame_filter_level,
#if CONFIG_LOOPFILTER_LEVEL
int frame_filter_level_r,
#endif
int y_only, int partial_frame, AVxWorker *workers,
int num_workers, AV1LfSync *lf_sync) {
int start_mi_row, end_mi_row, mi_rows_to_filter;
if (!frame_filter_level) return;
start_mi_row = 0;
mi_rows_to_filter = cm->mi_rows;
if (partial_frame && cm->mi_rows > 8) {
start_mi_row = cm->mi_rows >> 1;
start_mi_row &= 0xfffffff8;
mi_rows_to_filter = AOMMAX(cm->mi_rows / 8, 8);
}
void av1_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
int plane, TX_SIZE tx_size, int has_eob, int aoff,
int loff) {
ENTROPY_CONTEXT *const a = pd->above_context + aoff;
ENTROPY_CONTEXT *const l = pd->left_context + loff;
const int txs_wide = tx_size_wide_unit[tx_size];
const int txs_high = tx_size_high_unit[tx_size];
const BLOCK_SIZE bsize = AOMMAX(xd->mi[0]->mbmi.sb_type, BLOCK_8X8);
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
// above
if (has_eob && xd->mb_to_right_edge < 0) {
int i;
const int blocks_wide = max_block_wide(xd, plane_bsize, plane);
int above_contexts = txs_wide;
if (above_contexts + aoff > blocks_wide)
above_contexts = blocks_wide - aoff;
for (i = 0; i < above_contexts; ++i) a[i] = has_eob;
for (i = above_contexts; i < txs_wide; ++i) a[i] = 0;
} else {
memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * txs_wide);
}
// left
if (has_eob && xd->mb_to_bottom_edge < 0) {
int i;
const int blocks_high = max_block_high(xd, plane_bsize, plane);
int left_contexts = txs_high;
if (left_contexts + loff > blocks_high) left_contexts = blocks_high - loff;
for (i = 0; i < left_contexts; ++i) l[i] = has_eob;
for (i = left_contexts; i < txs_high; ++i) l[i] = 0;
} else {
memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * txs_high);
}
}
static void setup_ref_mv_list(const AV1_COMMON *cm, const MACROBLOCKD *xd,
MV_REFERENCE_FRAME ref_frame,
uint8_t *refmv_count,
CANDIDATE_MV *ref_mv_stack,
int_mv *mv_ref_list,
int block, int mi_row, int mi_col,
int16_t *mode_context) {
int idx, nearest_refmv_count = 0;
uint8_t newmv_count = 0;
CANDIDATE_MV tmp_mv;
int len, nr_len;
const MV_REF *const prev_frame_mvs_base = cm->use_prev_frame_mvs ?
cm->prev_frame->mvs + mi_row * cm->mi_cols + mi_col : NULL;
const int bs = AOMMAX(xd->n8_w, xd->n8_h);
const int has_tr = has_top_right(xd, mi_row, mi_col, bs);
MV_REFERENCE_FRAME rf[2];
av1_set_ref_frame(rf, ref_frame);
mode_context[ref_frame] = 0;
*refmv_count = 0;
// Scan the first above row mode info.
newmv_count += scan_row_mbmi(cm, xd, mi_row, mi_col, block, rf,
-1, ref_mv_stack, refmv_count);
// Scan the first left column mode info.
newmv_count += scan_col_mbmi(cm, xd, mi_row, mi_col, block, rf,
-1, ref_mv_stack, refmv_count);
// Check top-right boundary
if (has_tr)
newmv_count += scan_blk_mbmi(cm, xd, mi_row, mi_col, block, rf,
-1, 1, ref_mv_stack, refmv_count);
nearest_refmv_count = *refmv_count;
for (idx = 0; idx < nearest_refmv_count; ++idx)
ref_mv_stack[idx].weight += REF_CAT_LEVEL;
if (prev_frame_mvs_base && cm->show_frame && cm->last_show_frame &&
rf[1] == NONE) {
int blk_row, blk_col;
int coll_blk_count = 0;
for (blk_row = 0; blk_row < xd->n8_h; ++blk_row) {
for (blk_col = 0; blk_col < xd->n8_w; ++blk_col) {
coll_blk_count += add_col_ref_mv(cm, prev_frame_mvs_base, xd,
mi_row, mi_col, ref_frame,
blk_row, blk_col,
refmv_count, ref_mv_stack,
mode_context);
}
}
if (coll_blk_count == 0)
mode_context[ref_frame] |= (1 << ZEROMV_OFFSET);
} else {
mode_context[ref_frame] |= (1 << ZEROMV_OFFSET);
}
// Scan the second outer area.
for (idx = 2; idx <= 4; ++idx) {
scan_row_mbmi(cm, xd, mi_row, mi_col, block, rf,
-idx, ref_mv_stack, refmv_count);
scan_col_mbmi(cm, xd, mi_row, mi_col, block, rf,
-idx, ref_mv_stack, refmv_count);
}
switch (nearest_refmv_count) {
case 0:
mode_context[ref_frame] |= 0;
if (*refmv_count >= 1)
mode_context[ref_frame] |= 1;
if (*refmv_count == 1)
mode_context[ref_frame] |= (1 << REFMV_OFFSET);
else if (*refmv_count >= 2)
mode_context[ref_frame] |= (2 << REFMV_OFFSET);
break;
case 1:
mode_context[ref_frame] |= (newmv_count > 0) ? 2 : 3;
if (*refmv_count == 1)
mode_context[ref_frame] |= (3 << REFMV_OFFSET);
else if (*refmv_count >= 2)
mode_context[ref_frame] |= (4 << REFMV_OFFSET);
break;
case 2:
default:
if (newmv_count >= 2)
mode_context[ref_frame] |= 4;
else if (newmv_count == 1)
mode_context[ref_frame] |= 5;
else
mode_context[ref_frame] |= 6;
mode_context[ref_frame] |= (5 << REFMV_OFFSET);
break;
}
// Rank the likelihood and assign nearest and near mvs.
len = nearest_refmv_count;
while (len > 0) {
nr_len = 0;
for (idx = 1; idx < len; ++idx) {
if (ref_mv_stack[idx - 1].weight < ref_mv_stack[idx].weight) {
tmp_mv = ref_mv_stack[idx - 1];
ref_mv_stack[idx - 1] = ref_mv_stack[idx];
ref_mv_stack[idx] = tmp_mv;
nr_len = idx;
}
}
len = nr_len;
}
len = *refmv_count;
while (len > nearest_refmv_count) {
nr_len = nearest_refmv_count;
for (idx = nearest_refmv_count + 1; idx < len; ++idx) {
if (ref_mv_stack[idx - 1].weight < ref_mv_stack[idx].weight) {
tmp_mv = ref_mv_stack[idx - 1];
ref_mv_stack[idx - 1] = ref_mv_stack[idx];
ref_mv_stack[idx] = tmp_mv;
nr_len = idx;
}
}
len = nr_len;
}
if (xd->is_sec_rect) {
if (xd->n8_w < xd->n8_h) {
const MODE_INFO *const candidate_mi = xd->mi[-1];
const MB_MODE_INFO *const candidate_mbmi = &candidate_mi->mbmi;
handle_sec_rect_block(candidate_mbmi, nearest_refmv_count, ref_mv_stack,
ref_frame, mode_context);
}
if (xd->n8_w > xd->n8_h) {
const MODE_INFO *const candidate_mi = xd->mi[-xd->mi_stride];
const MB_MODE_INFO *const candidate_mbmi = &candidate_mi->mbmi;
handle_sec_rect_block(candidate_mbmi, nearest_refmv_count, ref_mv_stack,
ref_frame, mode_context);
}
}
if (rf[1] > NONE) {
for (idx = 0; idx < *refmv_count; ++idx) {
clamp_mv_ref(&ref_mv_stack[idx].this_mv.as_mv,
xd->n8_w << 3 , xd->n8_h << 3, xd);
clamp_mv_ref(&ref_mv_stack[idx].comp_mv.as_mv,
xd->n8_w << 3 , xd->n8_h << 3, xd);
}
} else {
for (idx = 0; idx < AOMMIN(MAX_MV_REF_CANDIDATES, *refmv_count); ++idx) {
mv_ref_list[idx].as_int = ref_mv_stack[idx].this_mv.as_int;
clamp_mv_ref(&mv_ref_list[idx].as_mv,
xd->n8_w << 3, xd->n8_h << 3, xd);
}
}
}
static int ransac(const int *matched_points, int npoints,
int *num_inliers_by_motion, double *params_by_motion,
int num_desired_motions, const int minpts,
IsDegenerateFunc is_degenerate,
FindTransformationFunc find_transformation,
ProjectPointsDoubleFunc projectpoints) {
static const double PROBABILITY_REQUIRED = 0.9;
static const double EPS = 1e-12;
int N = 10000, trial_count = 0;
int i = 0;
int ret_val = 0;
unsigned int seed = (unsigned int)npoints;
int indices[MAX_MINPTS] = { 0 };
double *points1, *points2;
double *corners1, *corners2;
double *image1_coord;
// Store information for the num_desired_motions best transformations found
// and the worst motion among them, as well as the motion currently under
// consideration.
RANSAC_MOTION *motions, *worst_kept_motion = NULL;
RANSAC_MOTION current_motion;
// Store the parameters and the indices of the inlier points for the motion
// currently under consideration.
double params_this_motion[MAX_PARAMDIM];
double *cnp1, *cnp2;
for (i = 0; i < num_desired_motions; ++i) {
num_inliers_by_motion[i] = 0;
}
if (npoints < minpts * MINPTS_MULTIPLIER || npoints == 0) {
return 1;
}
points1 = (double *)aom_malloc(sizeof(*points1) * npoints * 2);
points2 = (double *)aom_malloc(sizeof(*points2) * npoints * 2);
corners1 = (double *)aom_malloc(sizeof(*corners1) * npoints * 2);
corners2 = (double *)aom_malloc(sizeof(*corners2) * npoints * 2);
image1_coord = (double *)aom_malloc(sizeof(*image1_coord) * npoints * 2);
motions =
(RANSAC_MOTION *)aom_malloc(sizeof(RANSAC_MOTION) * num_desired_motions);
for (i = 0; i < num_desired_motions; ++i) {
motions[i].inlier_indices =
(int *)aom_malloc(sizeof(*motions->inlier_indices) * npoints);
clear_motion(motions + i, npoints);
}
current_motion.inlier_indices =
(int *)aom_malloc(sizeof(*current_motion.inlier_indices) * npoints);
clear_motion(¤t_motion, npoints);
worst_kept_motion = motions;
if (!(points1 && points2 && corners1 && corners2 && image1_coord && motions &&
current_motion.inlier_indices)) {
ret_val = 1;
goto finish_ransac;
}
cnp1 = corners1;
cnp2 = corners2;
for (i = 0; i < npoints; ++i) {
*(cnp1++) = *(matched_points++);
*(cnp1++) = *(matched_points++);
*(cnp2++) = *(matched_points++);
*(cnp2++) = *(matched_points++);
}
while (N > trial_count) {
double sum_distance = 0.0;
double sum_distance_squared = 0.0;
clear_motion(¤t_motion, npoints);
int degenerate = 1;
int num_degenerate_iter = 0;
while (degenerate) {
num_degenerate_iter++;
if (!get_rand_indices(npoints, minpts, indices, &seed)) {
ret_val = 1;
goto finish_ransac;
}
copy_points_at_indices(points1, corners1, indices, minpts);
copy_points_at_indices(points2, corners2, indices, minpts);
degenerate = is_degenerate(points1);
if (num_degenerate_iter > MAX_DEGENERATE_ITER) {
ret_val = 1;
goto finish_ransac;
}
}
if (find_transformation(minpts, points1, points2, params_this_motion)) {
trial_count++;
continue;
}
projectpoints(params_this_motion, corners1, image1_coord, npoints, 2, 2);
for (i = 0; i < npoints; ++i) {
double dx = image1_coord[i * 2] - corners2[i * 2];
double dy = image1_coord[i * 2 + 1] - corners2[i * 2 + 1];
double distance = sqrt(dx * dx + dy * dy);
if (distance < INLIER_THRESHOLD) {
current_motion.inlier_indices[current_motion.num_inliers++] = i;
sum_distance += distance;
sum_distance_squared += distance * distance;
}
}
if (current_motion.num_inliers >= worst_kept_motion->num_inliers &&
current_motion.num_inliers > 1) {
int temp;
double fracinliers, pNoOutliers, mean_distance, dtemp;
mean_distance = sum_distance / ((double)current_motion.num_inliers);
current_motion.variance =
sum_distance_squared / ((double)current_motion.num_inliers - 1.0) -
mean_distance * mean_distance * ((double)current_motion.num_inliers) /
((double)current_motion.num_inliers - 1.0);
if (is_better_motion(¤t_motion, worst_kept_motion)) {
// This motion is better than the worst currently kept motion. Remember
// the inlier points and variance. The parameters for each kept motion
// will be recomputed later using only the inliers.
worst_kept_motion->num_inliers = current_motion.num_inliers;
worst_kept_motion->variance = current_motion.variance;
memcpy(worst_kept_motion->inlier_indices, current_motion.inlier_indices,
sizeof(*current_motion.inlier_indices) * npoints);
assert(npoints > 0);
fracinliers = (double)current_motion.num_inliers / (double)npoints;
pNoOutliers = 1 - pow(fracinliers, minpts);
pNoOutliers = fmax(EPS, pNoOutliers);
pNoOutliers = fmin(1 - EPS, pNoOutliers);
dtemp = log(1.0 - PROBABILITY_REQUIRED) / log(pNoOutliers);
temp = (dtemp > (double)INT32_MAX)
? INT32_MAX
: dtemp < (double)INT32_MIN ? INT32_MIN : (int)dtemp;
if (temp > 0 && temp < N) {
N = AOMMAX(temp, MIN_TRIALS);
}
// Determine the new worst kept motion and its num_inliers and variance.
for (i = 0; i < num_desired_motions; ++i) {
if (is_better_motion(worst_kept_motion, &motions[i])) {
worst_kept_motion = &motions[i];
}
}
}
}
trial_count++;
}
// Sort the motions, best first.
qsort(motions, num_desired_motions, sizeof(RANSAC_MOTION), compare_motions);
// Recompute the motions using only the inliers.
for (i = 0; i < num_desired_motions; ++i) {
if (motions[i].num_inliers >= minpts) {
copy_points_at_indices(points1, corners1, motions[i].inlier_indices,
motions[i].num_inliers);
copy_points_at_indices(points2, corners2, motions[i].inlier_indices,
motions[i].num_inliers);
find_transformation(motions[i].num_inliers, points1, points2,
params_by_motion + (MAX_PARAMDIM - 1) * i);
}
num_inliers_by_motion[i] = motions[i].num_inliers;
}
finish_ransac:
aom_free(points1);
aom_free(points2);
aom_free(corners1);
aom_free(corners2);
aom_free(image1_coord);
aom_free(current_motion.inlier_indices);
for (i = 0; i < num_desired_motions; ++i) {
aom_free(motions[i].inlier_indices);
}
aom_free(motions);
return ret_val;
}
void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
MACROBLOCKD *xd, int global_level) {
int r, c;
int sbr, sbc;
int nhsb, nvsb;
int16_t src[OD_DERING_INBUF_SIZE];
int16_t *linebuf[3];
int16_t colbuf[3][OD_BSIZE_MAX + 2 * OD_FILT_VBORDER][OD_FILT_HBORDER];
dering_list dlist[MAX_MIB_SIZE * MAX_MIB_SIZE];
unsigned char *row_dering, *prev_row_dering, *curr_row_dering;
int dering_count;
int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = { { 0 } };
int stride;
int bsize[3];
int dec[3];
int pli;
int dering_left;
int coeff_shift = AOMMAX(cm->bit_depth - 8, 0);
int nplanes;
if (xd->plane[1].subsampling_x == xd->plane[1].subsampling_y &&
xd->plane[2].subsampling_x == xd->plane[2].subsampling_y)
nplanes = 3;
else
nplanes = 1;
nvsb = (cm->mi_rows + MAX_MIB_SIZE - 1) / MAX_MIB_SIZE;
nhsb = (cm->mi_cols + MAX_MIB_SIZE - 1) / MAX_MIB_SIZE;
av1_setup_dst_planes(xd->plane, frame, 0, 0);
row_dering = aom_malloc(sizeof(*row_dering) * nhsb * 2);
memset(row_dering, 1, sizeof(*row_dering) * (nhsb + 2) * 2);
prev_row_dering = row_dering + 1;
curr_row_dering = prev_row_dering + nhsb + 2;
for (pli = 0; pli < nplanes; pli++) {
dec[pli] = xd->plane[pli].subsampling_x;
bsize[pli] = OD_DERING_SIZE_LOG2 - dec[pli];
}
stride = (cm->mi_cols << bsize[0]) + 2 * OD_FILT_HBORDER;
for (pli = 0; pli < nplanes; pli++) {
linebuf[pli] = aom_malloc(sizeof(*linebuf) * OD_FILT_VBORDER * stride);
}
for (sbr = 0; sbr < nvsb; sbr++) {
for (pli = 0; pli < nplanes; pli++) {
for (r = 0; r < (MAX_MIB_SIZE << bsize[pli]) + 2 * OD_FILT_VBORDER; r++) {
for (c = 0; c < OD_FILT_HBORDER; c++) {
colbuf[pli][r][c] = OD_DERING_VERY_LARGE;
}
}
}
dering_left = 1;
for (sbc = 0; sbc < nhsb; sbc++) {
int level;
int nhb, nvb;
int cstart = 0;
if (!dering_left) cstart = -OD_FILT_HBORDER;
nhb = AOMMIN(MAX_MIB_SIZE, cm->mi_cols - MAX_MIB_SIZE * sbc);
nvb = AOMMIN(MAX_MIB_SIZE, cm->mi_rows - MAX_MIB_SIZE * sbr);
level = compute_level_from_index(
global_level, cm->mi_grid_visible[MAX_MIB_SIZE * sbr * cm->mi_stride +
MAX_MIB_SIZE * sbc]
->mbmi.dering_gain);
curr_row_dering[sbc] = 0;
if (level == 0 ||
(dering_count = sb_compute_dering_list(
cm, sbr * MAX_MIB_SIZE, sbc * MAX_MIB_SIZE, dlist)) == 0) {
dering_left = 0;
continue;
}
curr_row_dering[sbc] = 1;
for (pli = 0; pli < nplanes; pli++) {
int16_t dst[OD_BSIZE_MAX * OD_BSIZE_MAX];
int threshold;
int coffset;
int rend, cend;
if (sbc == nhsb - 1)
cend = (nhb << bsize[pli]);
else
cend = (nhb << bsize[pli]) + OD_FILT_HBORDER;
if (sbr == nvsb - 1)
rend = (nvb << bsize[pli]);
else
rend = (nvb << bsize[pli]) + OD_FILT_VBORDER;
coffset = sbc * MAX_MIB_SIZE << bsize[pli];
if (sbc == nhsb - 1) {
/* On the last superblock column, fill in the right border with
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = cend; c < (nhb << bsize[pli]) + OD_FILT_HBORDER; ++c) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (sbr == nvsb - 1) {
/* On the last superblock row, fill in the bottom border with
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
for (r = rend; r < rend + OD_FILT_VBORDER; r++) {
for (c = 0; c < (nhb << bsize[pli]) + 2 * OD_FILT_HBORDER; c++) {
src[(r + OD_FILT_VBORDER) * OD_FILT_BSTRIDE + c] =
OD_DERING_VERY_LARGE;
}
}
}
/* Copy in the pixels we need from the current superblock for
deringing.*/
copy_sb8_16(
cm,
&src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER + cstart],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr, coffset + cstart,
xd->plane[pli].dst.stride, rend, cend - cstart);
if (!prev_row_dering[sbc]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER], OD_FILT_BSTRIDE,
xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
nhb << bsize[pli]);
} else if (sbr > 0) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = 0; c < nhb << bsize[pli]; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
linebuf[pli][r * stride + coffset + c];
}
}
} else {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = 0; c < nhb << bsize[pli]; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (!prev_row_dering[sbc - 1]) {
copy_sb8_16(cm, src, OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset - OD_FILT_HBORDER, xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc > 0) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = -OD_FILT_HBORDER; c < 0; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
linebuf[pli][r * stride + coffset + c];
}
}
} else {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = -OD_FILT_HBORDER; c < 0; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (!prev_row_dering[sbc + 1]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER + (nhb << bsize[pli])],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset + (nhb << bsize[pli]), xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc < nhsb - 1) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = nhb << bsize[pli];
c < (nhb << bsize[pli]) + OD_FILT_HBORDER; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
linebuf[pli][r * stride + coffset + c];
}
}
} else {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = nhb << bsize[pli];
c < (nhb << bsize[pli]) + OD_FILT_HBORDER; c++) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
if (dering_left) {
/* If we deringed the superblock on the left then we need to copy in
saved pixels. */
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = 0; c < OD_FILT_HBORDER; c++) {
src[r * OD_FILT_BSTRIDE + c] = colbuf[pli][r][c];
}
}
}
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = 0; c < OD_FILT_HBORDER; c++) {
/* Saving pixels in case we need to dering the superblock on the
right. */
colbuf[pli][r][c] =
src[r * OD_FILT_BSTRIDE + c + (nhb << bsize[pli])];
}
}
copy_sb8_16(cm, &linebuf[pli][coffset], stride, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * (sbr + 1) - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
(nhb << bsize[pli]));
/* FIXME: This is a temporary hack that uses more conservative
deringing for chroma. */
if (pli)
threshold = (level * 5 + 4) >> 3 << coeff_shift;
else
threshold = level << coeff_shift;
if (threshold == 0) continue;
od_dering(
dst, &src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER],
dec[pli], dir, pli, dlist, dering_count, threshold, coeff_shift);
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
copy_dering_16bit_to_16bit(
(int16_t *)&CONVERT_TO_SHORTPTR(
xd->plane[pli]
.dst.buf)[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << bsize[pli]) +
(sbc * MAX_MIB_SIZE << bsize[pli])],
xd->plane[pli].dst.stride, dst, dlist, dering_count,
3 - dec[pli]);
} else {
#endif
copy_dering_16bit_to_8bit(
&xd->plane[pli].dst.buf[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << bsize[pli]) +
(sbc * MAX_MIB_SIZE << bsize[pli])],
xd->plane[pli].dst.stride, dst, dlist, dering_count, bsize[pli]);
#if CONFIG_AOM_HIGHBITDEPTH
}
#endif
}
dering_left = 1;
}
{
unsigned char *tmp;
tmp = prev_row_dering;
prev_row_dering = curr_row_dering;
curr_row_dering = tmp;
}
}
