generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Complete high bitdepth VAR_TX implementation.

Browse Source 
VAR_TX now works in the high bitdepth configuration.

Change-Id: I4114d7d9ed59c598f1e4d35b8e75876c07074ba7
This commit is contained in:
Geza Lore
2016-01-22 13:57:28 +00:00
parent 00380700fb
commit 432e875dce
1 changed files with 70 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

126
vp10/encoder/rdopt.c
View File

@@ -652,13 +652,8 @@ static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
return; return;
if (!is_inter_block(mbmi)) { if (!is_inter_block(mbmi)) {
#if CONFIG_VAR_TX
struct encode_b_args arg = {x, NULL, &mbmi->skip}; struct encode_b_args arg = {x, NULL, &mbmi->skip};
#if CONFIG_VP9_HIGHBITDEPTH #if CONFIG_VAR_TX
vp10_encode_block_intra(plane, block, blk_row, blk_col,
plane_bsize, tx_size, &arg);
dist_block(x, plane, block, tx_size, &dist, &sse);
#else
uint8_t *dst, *src; uint8_t *dst, *src;
int src_stride = x->plane[plane].src.stride; int src_stride = x->plane[plane].src.stride;
int dst_stride = xd->plane[plane].dst.stride; int dst_stride = xd->plane[plane].dst.stride;
@@ -680,9 +675,7 @@ static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
args->cpi->fn_ptr[txsize_to_bsize[tx_size]].vf(src, src_stride, args->cpi->fn_ptr[txsize_to_bsize[tx_size]].vf(src, src_stride,
dst, dst_stride, &tmp_sse); dst, dst_stride, &tmp_sse);
dist = (int64_t)tmp_sse * 16; dist = (int64_t)tmp_sse * 16;
#endif // CONFIG_VP9_HIGHBITDEPTH
#else #else
struct encode_b_args arg = {x, NULL, &mbmi->skip};
vp10_encode_block_intra(plane, block, blk_row, blk_col, vp10_encode_block_intra(plane, block, blk_row, blk_col,
plane_bsize, tx_size, &arg); plane_bsize, tx_size, &arg);
dist_block(x, plane, block, tx_size, &dist, &sse); dist_block(x, plane, block, tx_size, &dist, &sse);
@@ -1487,18 +1480,20 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
#if CONFIG_VAR_TX #if CONFIG_VAR_TX
const int coeff_ctx = combine_entropy_contexts(*(tempa + idx), const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy)); *(templ + idy));
#endif #endif // CONFIG_VAR_TX
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1); vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan); vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block,
#if CONFIG_VAR_TX #if CONFIG_VAR_TX
coeff_ctx, ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
so->neighbors, cpi->sf.use_fast_coef_costing);
*(tempa + idx) = !(p->eobs[block] == 0);
*(templ + idy) = !(p->eobs[block] == 0);
#else #else
tempa + idx, templ + idy, ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy,
#endif
TX_4X4, TX_4X4,
so->scan, so->neighbors, so->scan, so->neighbors,
cpi->sf.use_fast_coef_costing); cpi->sf.use_fast_coef_costing);
#endif // CONFIG_VAR_TX
if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd) if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
goto next_highbd; goto next_highbd;
vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block), vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
@@ -1511,18 +1506,19 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
#if CONFIG_VAR_TX #if CONFIG_VAR_TX
const int coeff_ctx = combine_entropy_contexts(*(tempa + idx), const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy)); *(templ + idy));
#endif #endif // CONFIG_VAR_TX
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0); vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan); vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block,
#if CONFIG_VAR_TX #if CONFIG_VAR_TX
coeff_ctx, ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
so->neighbors, cpi->sf.use_fast_coef_costing);
*(tempa + idx) = !(p->eobs[block] == 0);
*(templ + idy) = !(p->eobs[block] == 0);
#else #else
tempa + idx, templ + idy, ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy,
#endif TX_4X4, so->scan, so->neighbors,
TX_4X4,
so->scan, so->neighbors,
cpi->sf.use_fast_coef_costing); cpi->sf.use_fast_coef_costing);
#endif // CONFIG_VAR_TX
distortion += vp10_highbd_block_error( distortion += vp10_highbd_block_error(
coeff, BLOCK_OFFSET(pd->dqcoeff, block), coeff, BLOCK_OFFSET(pd->dqcoeff, block),
16, &unused, xd->bd) >> 2; 16, &unused, xd->bd) >> 2;
@@ -1555,6 +1551,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
next_highbd: next_highbd:
{} {}
} }
if (best_rd >= rd_thresh) if (best_rd >= rd_thresh)
return best_rd; return best_rd;
@@ -1604,8 +1601,8 @@ next_highbd:
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4); TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type, 0); const scan_order *so = get_scan(TX_4X4, tx_type, 0);
#if CONFIG_VAR_TX #if CONFIG_VAR_TX
int coeff_ctx = combine_entropy_contexts(*(tempa + idx), const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy)); *(templ + idy));
#endif #endif
vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1); vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan); vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
@@ -1629,8 +1626,8 @@ next_highbd:
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4); TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type, 0); const scan_order *so = get_scan(TX_4X4, tx_type, 0);
#if CONFIG_VAR_TX #if CONFIG_VAR_TX
int coeff_ctx = combine_entropy_contexts(*(tempa + idx), const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy)); *(templ + idy));
#endif #endif
vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0); vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan); vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
@@ -2321,12 +2318,6 @@ void vp10_tx_block_rd_b(const VP10_COMP *cpi, MACROBLOCK *x, TX_SIZE tx_size,
MACROBLOCKD *xd = &x->e_mbd; MACROBLOCKD *xd = &x->e_mbd;
const struct macroblock_plane *const p = &x->plane[plane]; const struct macroblock_plane *const p = &x->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane]; struct macroblockd_plane *const pd = &xd->plane[plane];
#if CONFIG_VP9_HIGHBITDEPTH
const int ss_txfrm_size = tx_size << 1;
int64_t this_sse;
int shift = tx_size == TX_32X32 ? 0 : 2;
tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
#endif
unsigned int tmp_sse = 0; unsigned int tmp_sse = 0;
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block); tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV; PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
@@ -2391,35 +2382,59 @@ void vp10_tx_block_rd_b(const VP10_COMP *cpi, MACROBLOCK *x, TX_SIZE tx_size,
cpi->fn_ptr[txm_bsize].vf(src, src_stride, rec_buffer, 32, &tmp_sse); cpi->fn_ptr[txm_bsize].vf(src, src_stride, rec_buffer, 32, &tmp_sse);
} }
#if CONFIG_VP9_HIGHBITDEPTH
*dist += vp10_highbd_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
&this_sse, xd->bd) >> shift;
*bsse += this_sse >> shift;
#else
*bsse += (int64_t)tmp_sse * 16; *bsse += (int64_t)tmp_sse * 16;
if (p->eobs[block] > 0) { if (p->eobs[block] > 0) {
switch (tx_size) { const int lossless = xd->lossless[xd->mi[0]->mbmi.segment_id];
case TX_32X32: #if CONFIG_VP9_HIGHBITDEPTH
vp10_inv_txfm_add_32x32(dqcoeff, rec_buffer, 32, p->eobs[block], if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
const int bd = xd->bd;
switch (tx_size) {
case TX_32X32:
vp10_highbd_inv_txfm_add_32x32(dqcoeff, rec_buffer, 32,
p->eobs[block], bd, tx_type);
break;
case TX_16X16:
vp10_highbd_inv_txfm_add_16x16(dqcoeff, rec_buffer, 32,
p->eobs[block], bd, tx_type);
break;
case TX_8X8:
vp10_highbd_inv_txfm_add_8x8(dqcoeff, rec_buffer, 32,
p->eobs[block], bd, tx_type);
break;
case TX_4X4:
vp10_highbd_inv_txfm_add_4x4(dqcoeff, rec_buffer, 32,
p->eobs[block], bd, tx_type, lossless);
break;
default:
assert(0 && "Invalid transform size");
break;
}
} else {
#else
{
#endif // CONFIG_VP9_HIGHBITDEPTH
switch (tx_size) {
case TX_32X32:
vp10_inv_txfm_add_32x32(dqcoeff, rec_buffer, 32, p->eobs[block],
tx_type);
break;
case TX_16X16:
vp10_inv_txfm_add_16x16(dqcoeff, rec_buffer, 32, p->eobs[block],
tx_type);
break;
case TX_8X8:
vp10_inv_txfm_add_8x8(dqcoeff, rec_buffer, 32, p->eobs[block],
tx_type); tx_type);
break; break;
case TX_16X16: case TX_4X4:
vp10_inv_txfm_add_16x16(dqcoeff, rec_buffer, 32, p->eobs[block], vp10_inv_txfm_add_4x4(dqcoeff, rec_buffer, 32, p->eobs[block],
tx_type); tx_type, lossless);
break; break;
case TX_8X8: default:
vp10_inv_txfm_add_8x8(dqcoeff, rec_buffer, 32, p->eobs[block], assert(0 && "Invalid transform size");
tx_type); break;
break; }
case TX_4X4:
vp10_inv_txfm_add_4x4(dqcoeff, rec_buffer, 32, p->eobs[block],
tx_type,
xd->lossless[xd->mi[0]->mbmi.segment_id]);
break;
default:
assert(0 && "Invalid transform size");
break;
} }
if ((bh >> 2) + blk_col > max_blocks_wide || if ((bh >> 2) + blk_col > max_blocks_wide ||
@@ -2444,7 +2459,6 @@ void vp10_tx_block_rd_b(const VP10_COMP *cpi, MACROBLOCK *x, TX_SIZE tx_size,
} }
} }
*dist += (int64_t)tmp_sse * 16; *dist += (int64_t)tmp_sse * 16;
#endif // CONFIG_VP9_HIGHBITDEPTH
*rate += cost_coeffs(x, plane, block, coeff_ctx, tx_size, *rate += cost_coeffs(x, plane, block, coeff_ctx, tx_size,
scan_order->scan, scan_order->neighbors, 0); scan_order->scan, scan_order->neighbors, 0);