void av1_foreach_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
foreach_transformed_block_visitor visit, void *arg) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
// transform size varies per plane, look it up in a common way.
const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi, pd) : mbmi->tx_size;
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
const uint8_t txw_unit = tx_size_wide_unit[tx_size];
const uint8_t txh_unit = tx_size_high_unit[tx_size];
const int step = txw_unit * txh_unit;
int i = 0, r, c;
// If mb_to_right_edge is < 0 we are in a situation in which
// the current block size extends into the UMV and we won't
// visit the sub blocks that are wholly within the UMV.
const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
// Keep track of the row and column of the blocks we use so that we know
// if we are in the unrestricted motion border.
for (r = 0; r < max_blocks_high; r += txh_unit) {
// Skip visiting the sub blocks that are wholly within the UMV.
for (c = 0; c < max_blocks_wide; c += txw_unit) {
visit(plane, i, r, c, plane_bsize, tx_size, arg);
i += step;
}
}
}
void vp9_foreach_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
foreach_transformed_block_visitor visit, void *arg) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const MB_MODE_INFO* mbmi = &xd->mi[0].src_mi->mbmi;
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
// transform size varies per plane, look it up in a common way.
const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi, pd)
: mbmi->tx_size;
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
const int step = 1 << (tx_size << 1);
int i;
// If mb_to_right_edge is < 0 we are in a situation in which
// the current block size extends into the UMV and we won't
// visit the sub blocks that are wholly within the UMV.
if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) {
int r, c;
int max_blocks_wide = num_4x4_w;
int max_blocks_high = num_4x4_h;
// xd->mb_to_right_edge is in units of pixels * 8. This converts
// it to 4x4 block sizes.
if (xd->mb_to_right_edge < 0)
max_blocks_wide += (xd->mb_to_right_edge >> (5 + pd->subsampling_x));
if (xd->mb_to_bottom_edge < 0)
max_blocks_high += (xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
i = 0;
// Unlike the normal case - in here we have to keep track of the
// row and column of the blocks we use so that we know if we are in
// the unrestricted motion border.
for (r = 0; r < num_4x4_h; r += (1 << tx_size)) {
for (c = 0; c < num_4x4_w; c += (1 << tx_size)) {
if (r < max_blocks_high && c < max_blocks_wide)
visit(plane, i, plane_bsize, tx_size, arg);
i += step;
}
}
} else {
void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
MACROBLOCKD *const xd = &x->e_mbd;
struct optimize_ctx ctx;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
struct encode_b_args arg = {x, &ctx, &mbmi->skip};
int plane;
for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
if (!x->skip_recode)
vp9_subtract_plane(x, bsize, plane);
if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
const struct macroblockd_plane* const pd = &xd->plane[plane];
const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi) : mbmi->tx_size;
vp9_get_entropy_contexts(bsize, tx_size, pd,
ctx.ta[plane], ctx.tl[plane]);
}
vp9_foreach_transformed_block_in_plane(xd, bsize, plane, encode_block,
&arg);
}
}
void av1_foreach_transformed_block_interleave(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit, void *arg) {
const struct macroblockd_plane *const pd_y = &xd->plane[0];
const struct macroblockd_plane *const pd_c = &xd->plane[1];
const MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
const TX_SIZE tx_log2_y = mbmi->tx_size;
const TX_SIZE tx_log2_c = get_uv_tx_size(mbmi, pd_c);
const int tx_sz_y = (1 << tx_log2_y);
const int tx_sz_c = (1 << tx_log2_c);
const BLOCK_SIZE plane_bsize_y = get_plane_block_size(bsize, pd_y);
const BLOCK_SIZE plane_bsize_c = get_plane_block_size(bsize, pd_c);
const int num_4x4_w_y = num_4x4_blocks_wide_lookup[plane_bsize_y];
const int num_4x4_w_c = num_4x4_blocks_wide_lookup[plane_bsize_c];
const int num_4x4_h_y = num_4x4_blocks_high_lookup[plane_bsize_y];
const int num_4x4_h_c = num_4x4_blocks_high_lookup[plane_bsize_c];
const int step_y = 1 << (tx_log2_y << 1);
const int step_c = 1 << (tx_log2_c << 1);
const int max_4x4_w_y =
get_max_4x4_size(num_4x4_w_y, xd->mb_to_right_edge, pd_y->subsampling_x);
const int max_4x4_h_y =
get_max_4x4_size(num_4x4_h_y, xd->mb_to_bottom_edge, pd_y->subsampling_y);
const int extra_step_y = ((num_4x4_w_y - max_4x4_w_y) >> tx_log2_y) * step_y;
const int max_4x4_w_c =
get_max_4x4_size(num_4x4_w_c, xd->mb_to_right_edge, pd_c->subsampling_x);
const int max_4x4_h_c =
get_max_4x4_size(num_4x4_h_c, xd->mb_to_bottom_edge, pd_c->subsampling_y);
const int extra_step_c = ((num_4x4_w_c - max_4x4_w_c) >> tx_log2_c) * step_c;
// The max_4x4_w/h may be smaller than tx_sz under some corner cases,
// i.e. when the SB is splitted by tile boundaries.
const int tu_num_w_y = (max_4x4_w_y + tx_sz_y - 1) / tx_sz_y;
const int tu_num_h_y = (max_4x4_h_y + tx_sz_y - 1) / tx_sz_y;
const int tu_num_w_c = (max_4x4_w_c + tx_sz_c - 1) / tx_sz_c;
const int tu_num_h_c = (max_4x4_h_c + tx_sz_c - 1) / tx_sz_c;
const int tu_num_y = tu_num_w_y * tu_num_h_y;
const int tu_num_c = tu_num_w_c * tu_num_h_c;
int tu_idx_c = 0;
int offset_y, row_y, col_y;
int offset_c, row_c, col_c;
for (row_y = 0; row_y < tu_num_h_y; row_y++) {
for (col_y = 0; col_y < tu_num_w_y; col_y++) {
// luma
offset_y = (row_y * tu_num_w_y + col_y) * step_y + row_y * extra_step_y;
visit(0, offset_y, row_y * tx_sz_y, col_y * tx_sz_y, plane_bsize_y,
tx_log2_y, arg);
// chroma
if (tu_idx_c < tu_num_c) {
row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
offset_c = tu_idx_c * step_c + (tu_idx_c / tu_num_w_c) * extra_step_c;
visit(1, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
visit(2, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
tu_idx_c++;
}
}
}
// In 422 case, it's possible that Chroma has more TUs than Luma
while (tu_idx_c < tu_num_c) {
row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
offset_c = tu_idx_c * step_c + row_c * extra_step_c;
visit(1, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
visit(2, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
tu_idx_c++;
}
}
