Commit b3cc3a07 authored by hui su's avatar hui su

Enable ADST for UV channel

derflr +0.202%
hevclf +0.207%
hevcmr +0.095%
hevchr +0.077%

Tested locally on several derf sequences, speed (encoder + decoder)
is slower by less than 1%.

It is part of the EXT_TX experiment, which is to be continued to
explore different transform variants.

Change-Id: I05d44994a62106538a9a241ed8d89bd7c5d14761
parent 26addefc
......@@ -264,6 +264,7 @@ EXPERIMENT_LIST="
spatial_svc
fp_mb_stats
emulate_hardware
ext_tx
"
CONFIG_LIST="
dependency_tracking
......
......@@ -219,15 +219,21 @@ static const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = {
};
static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type, const MACROBLOCKD *xd,
int block_idx) {
int block_idx, TX_SIZE tx_size) {
const MODE_INFO *const mi = xd->mi[0];
const MB_MODE_INFO *const mbmi = &mi->mbmi;
#if CONFIG_EXT_TX
if (xd->lossless || is_inter_block(mbmi) || tx_size >= TX_32X32)
return DCT_DCT;
return intra_mode_to_tx_type_lookup[plane_type == PLANE_TYPE_Y ?
get_y_mode(mi, block_idx) : mbmi->uv_mode];
#else
if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(mbmi) ||
mbmi->tx_size >= TX_32X32)
tx_size >= TX_32X32)
return DCT_DCT;
return intra_mode_to_tx_type_lookup[get_y_mode(mi, block_idx)];
#endif // CONFIG_EXT_TX
}
void vp10_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y);
......
......@@ -188,7 +188,7 @@ static void inverse_transform_block_inter(MACROBLOCKD* xd, int plane,
uint8_t *dst, int stride,
int eob, int block) {
struct macroblockd_plane *const pd = &xd->plane[plane];
TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block);
TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block, tx_size);
if (eob > 0) {
tran_low_t *const dqcoeff = pd->dqcoeff;
#if CONFIG_VP9_HIGHBITDEPTH
......@@ -347,7 +347,7 @@ static void predict_and_reconstruct_intra_block(MACROBLOCKD *const xd,
col, row, plane);
if (!mbmi->skip) {
TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx);
TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
const scan_order *sc = get_scan(tx_size, tx_type);
const int eob = vp10_decode_block_tokens(xd, plane, sc, col, row, tx_size,
r, mbmi->segment_id);
......@@ -362,7 +362,7 @@ static int reconstruct_inter_block(MACROBLOCKD *const xd, vpx_reader *r,
struct macroblockd_plane *const pd = &xd->plane[plane];
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
int block_idx = (row << 1) + col;
TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx);
TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size);
const scan_order *sc = get_scan(tx_size, tx_type);
const int eob = vp10_decode_block_tokens(xd, plane, sc, col, row, tx_size, r,
mbmi->segment_id);
......
......@@ -104,7 +104,7 @@ static int optimize_b(MACROBLOCK *mb, int plane, int block,
const int mul = 1 + (tx_size == TX_32X32);
const int16_t *dequant_ptr = pd->dequant;
const uint8_t *const band_translate = get_band_translate(tx_size);
TX_TYPE tx_type = get_tx_type(type, xd, block);
TX_TYPE tx_type = get_tx_type(type, xd, block, tx_size);
const scan_order *const so = get_scan(tx_size, tx_type);
const int16_t *const scan = so->scan;
const int16_t *const nb = so->neighbors;
......@@ -329,7 +329,7 @@ void vp10_xform_quant_fp(MACROBLOCK *x, int plane, int block,
const struct macroblock_plane *const p = &x->plane[plane];
const struct macroblockd_plane *const pd = &xd->plane[plane];
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
const scan_order *const scan_order = get_scan(tx_size, tx_type);
tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
......@@ -642,7 +642,7 @@ void vp10_xform_quant(MACROBLOCK *x, int plane, int block,
const struct macroblock_plane *const p = &x->plane[plane];
const struct macroblockd_plane *const pd = &xd->plane[plane];
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
const scan_order *const scan_order = get_scan(tx_size, tx_type);
tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
......@@ -741,7 +741,7 @@ static void encode_block(int plane, int block, BLOCK_SIZE plane_bsize,
int i, j;
uint8_t *dst;
ENTROPY_CONTEXT *a, *l;
TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block);
TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block, tx_size);
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
a = &ctx->ta[plane][i];
......@@ -928,7 +928,7 @@ void vp10_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
tran_low_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
const scan_order *const scan_order = get_scan(tx_size, tx_type);
PREDICTION_MODE mode;
const int bwl = b_width_log2_lookup[plane_bsize];
......
......@@ -601,7 +601,7 @@ static void block_yrd(VP10_COMP *cpi, MACROBLOCK *x, int *rate, int64_t *dist,
for (c = 0; c < num_4x4_w; c += block_step) {
if (c < max_blocks_wide) {
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
const scan_order *const scan_order = get_scan(tx_size, tx_type);
tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
......
......@@ -570,7 +570,7 @@ static void txfm_rd_in_plane(MACROBLOCK *x,
vp10_get_entropy_contexts(bsize, tx_size, pd, args.t_above, args.t_left);
tx_type = get_tx_type(pd->plane_type, xd, 0);
tx_type = get_tx_type(pd->plane_type, xd, 0, tx_size);
args.so = get_scan(tx_size, tx_type);
vp10_foreach_transformed_block_in_plane(xd, bsize, plane,
......@@ -804,7 +804,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
vpx_highbd_subtract_block(4, 4, src_diff, 8, src, src_stride,
dst, dst_stride, xd->bd);
if (xd->lossless) {
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type);
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT,
vp10_highbd_fwht4x4);
......@@ -820,7 +820,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
vp10_highbd_iwht4x4_add);
} else {
int64_t unused;
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type);
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type,
vpx_highbd_fdct4x4);
......@@ -909,7 +909,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
vpx_subtract_block(4, 4, src_diff, 8, src, src_stride, dst, dst_stride);
if (xd->lossless) {
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type);
vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, vp10_fwht4x4);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
......@@ -923,7 +923,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
vp10_iwht4x4_add);
} else {
int64_t unused;
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type);
vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, vpx_fdct4x4);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
......@@ -1316,7 +1316,7 @@ static int64_t encode_inter_mb_segment(VP10_COMP *cpi,
pd->dst.stride)];
int64_t thisdistortion = 0, thissse = 0;
int thisrate = 0, ref;
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, i);
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, i, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type);
const int is_compound = has_second_ref(&mi->mbmi);
const InterpKernel *kernel = vp10_filter_kernels[mi->mbmi.interp_filter];
......
......@@ -507,7 +507,7 @@ static void tokenize_b(int plane, int block, BLOCK_SIZE plane_bsize,
const tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
const int segment_id = mbmi->segment_id;
const int16_t *scan, *nb;
const TX_TYPE tx_type = get_tx_type(type, xd, block);
const TX_TYPE tx_type = get_tx_type(type, xd, block, tx_size);
const scan_order *const so = get_scan(tx_size, tx_type);
const int ref = is_inter_block(mbmi);
unsigned int (*const counts)[COEFF_CONTEXTS][ENTROPY_TOKENS] =
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment