void vp10_dering_frame(YV12_BUFFER_CONFIG *frame, VP10_COMMON *cm,
MACROBLOCKD *xd, int global_level) {
int r, c;
int sbr, sbc;
int nhsb, nvsb;
od_dering_in *src[3];
unsigned char *bskip;
int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = {{0}};
int stride;
int bsize[3];
int dec[3];
int pli;
int coeff_shift = VPXMAX(cm->bit_depth - 8, 0);
nvsb = (cm->mi_rows + MI_BLOCK_SIZE - 1)/MI_BLOCK_SIZE;
nhsb = (cm->mi_cols + MI_BLOCK_SIZE - 1)/MI_BLOCK_SIZE;
bskip = vpx_malloc(sizeof(*bskip)*cm->mi_rows*cm->mi_cols);
vp10_setup_dst_planes(xd->plane, frame, 0, 0);
for (pli = 0; pli < 3; pli++) {
dec[pli] = xd->plane[pli].subsampling_x;
bsize[pli] = 8 >> dec[pli];
}
stride = bsize[0]*cm->mi_cols;
for (pli = 0; pli < 3; pli++) {
src[pli] = vpx_malloc(sizeof(*src)*cm->mi_rows*cm->mi_cols*64);
for (r = 0; r < bsize[pli]*cm->mi_rows; ++r) {
for (c = 0; c < bsize[pli]*cm->mi_cols; ++c) {
#if CONFIG_VPX_HIGHBITDEPTH
if (cm->use_highbitdepth) {
src[pli][r * stride + c] =
CONVERT_TO_SHORTPTR(xd->plane[pli].dst.buf)
[r * xd->plane[pli].dst.stride + c];
} else {
#endif
src[pli][r * stride + c] =
xd->plane[pli].dst.buf[r * xd->plane[pli].dst.stride + c];
#if CONFIG_VPX_HIGHBITDEPTH
}
#endif
}
}
}
for (r = 0; r < cm->mi_rows; ++r) {
for (c = 0; c < cm->mi_cols; ++c) {
const MB_MODE_INFO *mbmi =
&cm->mi_grid_visible[r * cm->mi_stride + c]->mbmi;
bskip[r * cm->mi_cols + c] = mbmi->skip;
}
}
for (sbr = 0; sbr < nvsb; sbr++) {
for (sbc = 0; sbc < nhsb; sbc++) {
int level;
int nhb, nvb;
nhb = VPXMIN(MI_BLOCK_SIZE, cm->mi_cols - MI_BLOCK_SIZE*sbc);
nvb = VPXMIN(MI_BLOCK_SIZE, cm->mi_rows - MI_BLOCK_SIZE*sbr);
for (pli = 0; pli < 3; pli++) {
int16_t dst[MI_BLOCK_SIZE*MI_BLOCK_SIZE*8*8];
int threshold;
#if DERING_REFINEMENT
level = compute_level_from_index(
global_level,
cm->mi_grid_visible[MI_BLOCK_SIZE*sbr*cm->mi_stride +
MI_BLOCK_SIZE*sbc]->mbmi.dering_gain);
#else
level = global_level;
#endif
/* FIXME: This is a temporary hack that uses more conservative
deringing for chroma. */
if (pli) level = (level*5 + 4) >> 3;
if (sb_all_skip(cm, sbr*MI_BLOCK_SIZE, sbc*MI_BLOCK_SIZE)) level = 0;
threshold = level << coeff_shift;
od_dering(
&OD_DERING_VTBL_C,
dst,
MI_BLOCK_SIZE*bsize[pli],
&src[pli][sbr*stride*bsize[pli]*MI_BLOCK_SIZE +
sbc*bsize[pli]*MI_BLOCK_SIZE],
stride, nhb, nvb, sbc, sbr, nhsb, nvsb, dec[pli], dir, pli,
&bskip[MI_BLOCK_SIZE*sbr*cm->mi_cols + MI_BLOCK_SIZE*sbc],
cm->mi_cols, threshold, OD_DERING_NO_CHECK_OVERLAP, coeff_shift);
for (r = 0; r < bsize[pli]*nvb; ++r) {
for (c = 0; c < bsize[pli]*nhb; ++c) {
#if CONFIG_VPX_HIGHBITDEPTH
if (cm->use_highbitdepth) {
CONVERT_TO_SHORTPTR(xd->plane[pli].dst.buf)
[xd->plane[pli].dst.stride*(bsize[pli]*MI_BLOCK_SIZE*sbr + r)
+ sbc*bsize[pli]*MI_BLOCK_SIZE + c] =
dst[r * MI_BLOCK_SIZE * bsize[pli] + c];
} else {
#endif
xd->plane[pli].dst.buf[xd->plane[pli].dst.stride*
(bsize[pli]*MI_BLOCK_SIZE*sbr + r) +
sbc*bsize[pli]*MI_BLOCK_SIZE + c] =
dst[r * MI_BLOCK_SIZE * bsize[pli] + c];
#if CONFIG_VPX_HIGHBITDEPTH
}
#endif
}
}
}
}
}
for (pli = 0; pli < 3; pli++) {
vpx_free(src[pli]);
}
vpx_free(bskip);
}
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;
}
}