// Iterate over blocks within a superblock
static void vp10_clpf_sb(const YV12_BUFFER_CONFIG *frame_buffer,
const VP10_COMMON *cm, MACROBLOCKD *xd,
MODE_INFO *const *mi_8x8, int xpos, int ypos) {
// Temporary buffer (to allow SIMD parallelism)
uint8_t buf_unaligned[BS * BS + 15];
uint8_t *buf = (uint8_t *)(((intptr_t)buf_unaligned + 15) & ~15);
int x, y, p;
for (p = 0; p < (CLPF_FILTER_ALL_PLANES ? MAX_MB_PLANE : 1); p++) {
for (y = 0; y < MI_BLOCK_SIZE && ypos + y < cm->mi_rows; y++) {
for (x = 0; x < MI_BLOCK_SIZE && xpos + x < cm->mi_cols; x++) {
const MB_MODE_INFO *mbmi =
&mi_8x8[(ypos + y) * cm->mi_stride + xpos + x]->mbmi;
// Do not filter if there is no residual
if (!mbmi->skip) {
// Do not filter frame edges
int has_top = ypos + y > 0;
int has_left = xpos + x > 0;
int has_bottom = ypos + y < cm->mi_rows - 1;
int has_right = xpos + x < cm->mi_cols - 1;
#if CLPF_ALLOW_BLOCK_PARALLELISM
// Do not filter superblock edges
has_top &= !!y;
has_left &= !!x;
has_bottom &= y != MI_BLOCK_SIZE - 1;
has_right &= x != MI_BLOCK_SIZE - 1;
#endif
vp10_setup_dst_planes(xd->plane, frame_buffer, ypos + y, xpos + x);
clpf_block(
xd->plane[p].dst.buf, CLPF_ALLOW_PIXEL_PARALLELISM
? buf + y * MI_SIZE * BS + x * MI_SIZE
: xd->plane[p].dst.buf,
xd->plane[p].dst.stride,
CLPF_ALLOW_PIXEL_PARALLELISM ? BS : xd->plane[p].dst.stride,
has_top, has_left, has_bottom, has_right,
MI_SIZE >> xd->plane[p].subsampling_x,
MI_SIZE >> xd->plane[p].subsampling_y);
}
}
}
#if CLPF_ALLOW_PIXEL_PARALLELISM
for (y = 0; y < MI_BLOCK_SIZE && ypos + y < cm->mi_rows; y++) {
for (x = 0; x < MI_BLOCK_SIZE && xpos + x < cm->mi_cols; x++) {
const MB_MODE_INFO *mbmi =
&mi_8x8[(ypos + y) * cm->mi_stride + xpos + x]->mbmi;
vp10_setup_dst_planes(xd->plane, frame_buffer, ypos + y, xpos + x);
if (!mbmi->skip) {
int i = 0;
for (i = 0; i<MI_SIZE>> xd->plane[p].subsampling_y; i++)
memcpy(xd->plane[p].dst.buf + i * xd->plane[p].dst.stride,
buf + (y * MI_SIZE + i) * BS + x * MI_SIZE,
MI_SIZE >> xd->plane[p].subsampling_x);
}
}
}
// Implement row loopfiltering for each thread.
static INLINE
void thread_loop_filter_rows(const YV12_BUFFER_CONFIG *const frame_buffer,
VP10_COMMON *const cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only,
VP9LfSync *const lf_sync) {
const int num_planes = y_only ? 1 : MAX_MB_PLANE;
const int sb_cols = mi_cols_aligned_to_sb(cm->mi_cols) >> MI_BLOCK_SIZE_LOG2;
int mi_row, mi_col;
enum lf_path path;
if (y_only)
path = LF_PATH_444;
else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1)
path = LF_PATH_420;
else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0)
path = LF_PATH_444;
else
path = LF_PATH_SLOW;
for (mi_row = start; mi_row < stop;
mi_row += lf_sync->num_workers * MI_BLOCK_SIZE) {
MODE_INFO **const mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) {
const int r = mi_row >> MI_BLOCK_SIZE_LOG2;
const int c = mi_col >> MI_BLOCK_SIZE_LOG2;
LOOP_FILTER_MASK lfm;
int plane;
sync_read(lf_sync, r, c);
vp10_setup_dst_planes(planes, frame_buffer, mi_row, mi_col);
// TODO(JBB): Make setup_mask work for non 420.
vp10_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride,
&lfm);
vp10_filter_block_plane_ss00(cm, &planes[0], mi_row, &lfm);
for (plane = 1; plane < num_planes; ++plane) {
switch (path) {
case LF_PATH_420:
vp10_filter_block_plane_ss11(cm, &planes[plane], mi_row, &lfm);
break;
case LF_PATH_444:
vp10_filter_block_plane_ss00(cm, &planes[plane], mi_row, &lfm);
break;
case LF_PATH_SLOW:
vp10_filter_block_plane_non420(cm, &planes[plane], mi + mi_col,
mi_row, mi_col);
break;
}
}
sync_write(lf_sync, r, c, sb_cols);
}
}
}
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);
}
