Refactor recursive transform block scheme
This commit re-designs the recursive transform block partition
rate-distortion optimization framework. It allows the encoder to
improve speed by 10%.
Change-Id: I6dd3a7dd428a530d8012e5c6ddc40e650c8b392b
diff --git a/vp10/common/pred_common.h b/vp10/common/pred_common.h
index 571783e..314bf1e 100644
--- a/vp10/common/pred_common.h
+++ b/vp10/common/pred_common.h
@@ -189,6 +189,7 @@
if (tx_size == plane_tx_size) {
++get_tx_counts(max_tx_size, ctx, tx_counts)[tx_size];
+ mbmi->tx_size = tx_size;
} else {
int bsl = b_width_log2_lookup[bsize];
int i;
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index bc142e9..ffd84e8 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -1286,7 +1286,6 @@
static void select_tx_block(const VP10_COMP *cpi, MACROBLOCK *x,
int blk_row, int blk_col, int plane, int block,
TX_SIZE tx_size, BLOCK_SIZE plane_bsize,
- BLOCK_SIZE txb_bsize,
ENTROPY_CONTEXT *ta, ENTROPY_CONTEXT *tl,
int *rate, int64_t *dist,
int64_t *bsse, int *skip) {
@@ -1299,34 +1298,38 @@
int max_blocks_high = num_4x4_blocks_high_lookup[plane_bsize];
int max_blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize];
int64_t this_rd = INT64_MAX;
- ENTROPY_CONTEXT ctxa[16], ctxl[16];
ENTROPY_CONTEXT *pta = ta + blk_col;
ENTROPY_CONTEXT *ptl = tl + blk_row;
+ ENTROPY_CONTEXT stxa = 0, stxl = 0;
int coeff_ctx, i;
-
- memcpy(ctxa, ta, sizeof(ENTROPY_CONTEXT) * max_blocks_wide);
- memcpy(ctxl, tl, sizeof(ENTROPY_CONTEXT) * max_blocks_high);
+ int64_t sum_dist = 0, sum_bsse = 0;
+ int64_t sum_rd = INT64_MAX;
+ int sum_rate = vp10_cost_bit(128, 1);
+ int all_skip = 1;
+ TX_SIZE swap_tx_size = TX_SIZES;
switch (tx_size) {
case TX_4X4:
+ stxa = pta[0];
+ stxl = ptl[0];
break;
case TX_8X8:
- pta[0] = !!*(const uint16_t *)&pta[0];
- ptl[0] = !!*(const uint16_t *)&ptl[0];
+ stxa = !!*(const uint16_t *)&pta[0];
+ stxl = !!*(const uint16_t *)&ptl[0];
break;
case TX_16X16:
- pta[0] = !!*(const uint32_t *)&pta[0];
- ptl[0] = !!*(const uint32_t *)&ptl[0];
+ stxa = !!*(const uint32_t *)&pta[0];
+ stxl = !!*(const uint32_t *)&ptl[0];
break;
case TX_32X32:
- pta[0] = !!*(const uint64_t *)&pta[0];
- ptl[0] = !!*(const uint64_t *)&ptl[0];
+ stxa = !!*(const uint64_t *)&pta[0];
+ stxl = !!*(const uint64_t *)&ptl[0];
break;
default:
assert(0 && "Invalid transform size.");
break;
}
- coeff_ctx = combine_entropy_contexts(pta[0], ptl[0]);
+ coeff_ctx = combine_entropy_contexts(stxa, stxl);
if (xd->mb_to_bottom_edge < 0)
max_blocks_high += xd->mb_to_bottom_edge >> (5 + pd->subsampling_y);
@@ -1341,38 +1344,23 @@
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide)
return;
- mbmi->inter_tx_size[tx_idx] = tx_size;
- mbmi->tx_size = tx_size;
-
- if (cpi->common.tx_mode == TX_MODE_SELECT || tx_size == TX_4X4) {
- tx_block_rd_b(x, tx_size, blk_row, blk_col, plane, block,
- plane_bsize, coeff_ctx, rate, dist, bsse, skip);
- if (tx_size > TX_4X4)
- *rate += vp10_cost_bit(128, 0);
- this_rd = RDCOST(x->rdmult, x->rddiv, *rate, *dist);
- for (i = 0; i < (1 << tx_size); ++i) {
- pta[i] = !(p->eobs[block] == 0);
- ptl[i] = !(p->eobs[block] == 0);
- }
- }
-
if (tx_size > TX_4X4) {
BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
- int bh = num_4x4_blocks_high_lookup[bsize];
+ int bsl = b_height_log2_lookup[bsize];
int sub_step = 1 << (2 * (tx_size - 1));
int i;
- int this_rate, sum_rate = vp10_cost_bit(128, 1);
- int64_t this_dist, sum_dist = 0;
- int64_t this_bsse, sum_bsse = 0;
- int this_skip, all_skip = 1;
- int64_t sum_rd;
+ int this_rate;
+ int64_t this_dist;
+ int64_t this_bsse;
+ int this_skip;
+
+ --bsl;
for (i = 0; i < 4; ++i) {
- int offsetr = (i >> 1) * bh / 2;
- int offsetc = (i & 0x01) * bh / 2;
+ int offsetr = (i >> 1) << bsl;
+ int offsetc = (i & 0x01) << bsl;
select_tx_block(cpi, x, blk_row + offsetr, blk_col + offsetc,
plane, block + i * sub_step, tx_size - 1,
- plane_bsize, txsize_to_bsize[tx_size - 1],
- ctxa, ctxl, &this_rate, &this_dist,
+ plane_bsize, ta, tl, &this_rate, &this_dist,
&this_bsse, &this_skip);
sum_rate += this_rate;
sum_dist += this_dist;
@@ -1380,24 +1368,29 @@
all_skip &= this_skip;
}
sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
+ }
- if (this_rd < sum_rd) {
- int idx, idy;
- for (idy = blk_row; idy < blk_row + bh; idy += 2)
- for (idx = blk_col; idx < blk_col + bh; idx += 2)
- mbmi->inter_tx_size[(idy / 2) * 8 + (idx / 2)] = tx_size;
- mbmi->tx_size = tx_size;
- } else {
- *rate = sum_rate;
- *dist = sum_dist;
- *bsse = sum_bsse;
- *skip = all_skip;
+ if (cpi->common.tx_mode == TX_MODE_SELECT || tx_size == TX_4X4) {
+ swap_tx_size = mbmi->inter_tx_size[tx_idx];
+ mbmi->inter_tx_size[tx_idx] = tx_size;
- memcpy(pta, ctxa + (blk_col >> pd->subsampling_x),
- sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide_lookup[txb_bsize]);
- memcpy(ptl, ctxl + (blk_row >> pd->subsampling_y),
- sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_high_lookup[txb_bsize]);
- }
+ tx_block_rd_b(x, tx_size, blk_row, blk_col, plane, block,
+ plane_bsize, coeff_ctx, rate, dist, bsse, skip);
+ if (tx_size > TX_4X4)
+ *rate += vp10_cost_bit(128, 0);
+ this_rd = RDCOST(x->rdmult, x->rddiv, *rate, *dist);
+ }
+
+ if (this_rd < sum_rd) {
+ for (i = 0; i < (1 << tx_size); ++i)
+ pta[i] = ptl[i] = !(p->eobs[block] == 0);
+ mbmi->tx_size = tx_size;
+ } else {
+ *rate = sum_rate;
+ *dist = sum_dist;
+ *bsse = sum_bsse;
+ *skip = all_skip;
+ mbmi->inter_tx_size[tx_idx] = swap_tx_size;
}
}
@@ -1437,7 +1430,7 @@
for (idy = 0; idy < mi_height; idy += bh) {
for (idx = 0; idx < mi_width; idx += bh) {
select_tx_block(cpi, x, idy, idx, 0, block,
- max_txsize_lookup[plane_bsize], plane_bsize, txb_size,
+ max_txsize_lookup[plane_bsize], plane_bsize,
ctxa, ctxl, &pnrate, &pndist, &pnsse, &pnskip);
*rate += pnrate;
*distortion += pndist;
