hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-27 19:24:06 +00:00
Code Issues Actions Projects Releases Wiki Activity

Merge branch 'intra-fix-up' into 'master'

Browse source 
Intra fix up

See merge request cs/ultravideo/vvc/uvg266!12
This commit is contained in:
Joose Sainio 2022-07-08 15:17:51 +03:00
parent 456f69f785 ea32ef33ac
commit 98322c943e
8 changed files with 126 additions and 78 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
src/encode_coding_tree.c
View file

@ -166,7 +166,8 @@ static bool can_use_lfnst_with_isp(const int width, const int height, const int
const int x, const int x,
const int y, const int y,
enum uvg_tree_type tree_type, enum uvg_tree_type tree_type,
const color_t color) const color_t color,
const lcu_t* lcu)
{ {
if (state->encoder_control->cfg.lfnst && pred_cu->type == CU_INTRA) { if (state->encoder_control->cfg.lfnst && pred_cu->type == CU_INTRA) {
const int isp_mode = 0; // ISP_TODO: assign proper ISP mode when ISP is implemented const int isp_mode = 0; // ISP_TODO: assign proper ISP mode when ISP is implemented
@ -201,12 +202,12 @@ static bool can_use_lfnst_with_isp(const int width, const int height, const int
const int tu_height = tu_width; // TODO: height for non-square blocks const int tu_height = tu_width; // TODO: height for non-square blocks
// TODO: chroma transform skip // TODO: chroma transform skip
if (tree_type != UVG_BOTH_T) { if (color == COLOR_Y) {
for (int i = 0; i < num_transform_units; i++) { for (int i = 0; i < num_transform_units; i++) {
// TODO: this works only for square blocks // TODO: this works only for square blocks
const int pu_x = x + ((i % tu_row_length) * tu_width); const int tu_x = x + ((i % tu_row_length) * tu_width);
const int pu_y = y + ((i / tu_row_length) * tu_height); const int tu_y = y + ((i / tu_row_length) * tu_height);
const cu_info_t* cur_tu = uvg_cu_array_at_const(frame->cu_array, pu_x, pu_y); const cu_info_t* cur_tu = lcu ? LCU_GET_CU_AT_PX(lcu, tu_x, tu_y) : uvg_cu_array_at_const(frame->cu_array, tu_x, tu_y);
assert(cur_tu != NULL && "NULL transform unit."); assert(cur_tu != NULL && "NULL transform unit.");
bool cbf_set = cbf_is_set(cur_tu->cbf, tr_depth, COLOR_Y); bool cbf_set = cbf_is_set(cur_tu->cbf, tr_depth, COLOR_Y);
@ -239,7 +240,7 @@ static bool encode_lfnst_idx(
const color_t color) const color_t color)
{ {
if (uvg_is_lfnst_allowed(state, pred_cu, width, height, x, y, tree_type, color)) { if (uvg_is_lfnst_allowed(state, pred_cu, width, height, x, y, tree_type, color, NULL)) {
// Getting separate tree bool from block size is a temporary fix until a proper dual tree check is possible (there is no dual tree structure at time of writing this). // Getting separate tree bool from block size is a temporary fix until a proper dual tree check is possible (there is no dual tree structure at time of writing this).
// VTM seems to force explicit dual tree structure for small 4x4 blocks // VTM seems to force explicit dual tree structure for small 4x4 blocks
bool is_separate_tree = depth == 4 || tree_type != UVG_BOTH_T; bool is_separate_tree = depth == 4 || tree_type != UVG_BOTH_T;
@ -399,7 +400,7 @@ void uvg_encode_ts_residual(encoder_state_t* const state,
belowPixel = pos_y > 0 ? coeff[pos_x + (pos_y - 1) * width] : 0; belowPixel = pos_y > 0 ? coeff[pos_x + (pos_y - 1) * width] : 0;
absLevel = uvg_derive_mod_coeff(rightPixel, belowPixel, abs(coeff[blk_pos]), 0); absLevel = uvg_derive_mod_coeff(rightPixel, belowPixel, abs(coeff[blk_pos]), 0);
cutoffVal = 2; cutoffVal = 2;
for (i = 0; i < numGtBins; i++) for (int j = 0; j < numGtBins; j++)
{ {
if (absLevel >= cutoffVal) if (absLevel >= cutoffVal)
{ {
@ -1738,7 +1739,6 @@ void uvg_encode_coding_tree(
encode_transform_coeff(state, x, y, depth, 0, 0, 0, 1, coeff, tree_type); encode_transform_coeff(state, x, y, depth, 0, 0, 0, 1, coeff, tree_type);
// Write LFNST only once for single tree structure // Write LFNST only once for single tree structure
encode_lfnst_idx(state, cabac, tmp, x, y, depth, cu_width, cu_height, tree_type, COLOR_UV); encode_lfnst_idx(state, cabac, tmp, x, y, depth, cu_width, cu_height, tree_type, COLOR_UV);
} }
} }

3
src/encode_coding_tree.h
View file

@ -49,7 +49,8 @@ bool uvg_is_lfnst_allowed(
const int x, const int x,
const int y, const int y,
enum uvg_tree_type tree_type, enum uvg_tree_type tree_type,
const color_t color); const color_t color,
const lcu_t* lcu);
void uvg_encode_coding_tree( void uvg_encode_coding_tree(
encoder_state_t * const state, encoder_state_t * const state,

43
src/search.c
View file

@ -393,6 +393,8 @@ double uvg_cu_rd_cost_chroma(const encoder_state_t *const state,
// block and return 0 cost for all others. // block and return 0 cost for all others.
return 0; return 0;
} }
int u_is_set = pred_cu->joint_cb_cr ? (pred_cu->joint_cb_cr & 2) >> 1 : cbf_is_set(pred_cu->cbf, depth, COLOR_U);
int v_is_set = pred_cu->joint_cb_cr ? (pred_cu->joint_cb_cr & 1) : cbf_is_set(pred_cu->cbf, depth, COLOR_V);
// See luma for why the second condition // See luma for why the second condition
if (!skip_residual_coding) { if (!skip_residual_coding) {
@ -401,13 +403,10 @@ double uvg_cu_rd_cost_chroma(const encoder_state_t *const state,
cabac_ctx_t *ctx = &(cabac->ctx.qt_cbf_model_cb[0]); cabac_ctx_t *ctx = &(cabac->ctx.qt_cbf_model_cb[0]);
cabac->cur_ctx = ctx; cabac->cur_ctx = ctx;
if (tr_depth == 0 || cbf_is_set(pred_cu->cbf, depth - 1, COLOR_U)) { if (tr_depth == 0 || cbf_is_set(pred_cu->cbf, depth - 1, COLOR_U)) {
int u_is_set = cbf_is_set(pred_cu->cbf, depth, COLOR_U);
CABAC_FBITS_UPDATE(cabac, ctx, u_is_set, tr_tree_bits, "cbf_cb_search"); CABAC_FBITS_UPDATE(cabac, ctx, u_is_set, tr_tree_bits, "cbf_cb_search");
} }
int is_set = cbf_is_set(pred_cu->cbf, depth, COLOR_U); ctx = &(cabac->ctx.qt_cbf_model_cr[u_is_set]);
ctx = &(cabac->ctx.qt_cbf_model_cr[is_set]);
if (tr_depth == 0 || cbf_is_set(pred_cu->cbf, depth - 1, COLOR_V)) { if (tr_depth == 0 || cbf_is_set(pred_cu->cbf, depth - 1, COLOR_V)) {
int v_is_set = cbf_is_set(pred_cu->cbf, depth, COLOR_V);
CABAC_FBITS_UPDATE(cabac, ctx, v_is_set, tr_tree_bits, "cbf_cb_search"); CABAC_FBITS_UPDATE(cabac, ctx, v_is_set, tr_tree_bits, "cbf_cb_search");
} }
} }
@ -426,7 +425,7 @@ double uvg_cu_rd_cost_chroma(const encoder_state_t *const state,
} }
if (state->encoder_control->cfg.jccr) { if (state->encoder_control->cfg.jccr) {
int cbf_mask = cbf_is_set(pred_cu->cbf, depth, COLOR_U) * 2 + cbf_is_set(pred_cu->cbf, depth, COLOR_V) - 1; int cbf_mask = u_is_set * 2 + v_is_set - 1;
cabac_ctx_t* ctx = NULL; cabac_ctx_t* ctx = NULL;
if (cbf_mask != -1) { if (cbf_mask != -1) {
cabac_data_t* cabac = (cabac_data_t*)&state->search_cabac; cabac_data_t* cabac = (cabac_data_t*)&state->search_cabac;
@ -448,14 +447,19 @@ double uvg_cu_rd_cost_chroma(const encoder_state_t *const state,
ssd = ssd_u + ssd_v; ssd = ssd_u + ssd_v;
} }
if (!skip_residual_coding) if (!skip_residual_coding) {
{
int8_t scan_order = uvg_get_scan_order(pred_cu->type, pred_cu->intra.mode_chroma, depth); int8_t scan_order = uvg_get_scan_order(pred_cu->type, pred_cu->intra.mode_chroma, depth);
const int index = xy_to_zorder(LCU_WIDTH_C, lcu_px.x, lcu_px.y); const int index = xy_to_zorder(LCU_WIDTH_C, lcu_px.x, lcu_px.y);
if((pred_cu->joint_cb_cr & 3) == 0){
coeff_bits += uvg_get_coeff_cost(state, &lcu->coeff.u[index], NULL, width, 2, scan_order, 0); coeff_bits += uvg_get_coeff_cost(state, &lcu->coeff.u[index], NULL, width, 2, scan_order, 0);
coeff_bits += uvg_get_coeff_cost(state, &lcu->coeff.v[index], NULL, width, 2, scan_order, 0); coeff_bits += uvg_get_coeff_cost(state, &lcu->coeff.v[index], NULL, width, 2, scan_order, 0);
} }
else {
coeff_bits += uvg_get_coeff_cost(state, &lcu->coeff.joint_uv[index], NULL, width, 2, scan_order, 0);
}
}
double bits = tr_tree_bits + coeff_bits; double bits = tr_tree_bits + coeff_bits;
@ -568,7 +572,7 @@ static double cu_rd_cost_tr_split_accurate(
} }
if(depth == 4 || tree_type == UVG_LUMA_T) { if(depth == 4 || tree_type == UVG_LUMA_T) {
if (uvg_is_lfnst_allowed(state, tr_cu, width, width, x_px, y_px, tree_type, COLOR_Y)) { if (uvg_is_lfnst_allowed(state, tr_cu, width, width, x_px, y_px, tree_type, COLOR_Y, lcu)) {
const int lfnst_idx = tr_cu->lfnst_idx; const int lfnst_idx = tr_cu->lfnst_idx;
CABAC_FBITS_UPDATE( CABAC_FBITS_UPDATE(
cabac, cabac,
@ -620,10 +624,10 @@ static double cu_rd_cost_tr_split_accurate(
else { else {
{ {
int index = lcu_px.y * LCU_WIDTH_C + lcu_px.x; int index = lcu_px.y * LCU_WIDTH_C + lcu_px.x;
int ssd_u_joint = uvg_pixels_calc_ssd(&lcu->ref.u[index], &lcu->rec.joint_u[index], int ssd_u_joint = uvg_pixels_calc_ssd(&lcu->ref.u[index], &lcu->rec.u[index],
LCU_WIDTH_C, LCU_WIDTH_C, LCU_WIDTH_C, LCU_WIDTH_C,
chroma_width); chroma_width);
int ssd_v_joint = uvg_pixels_calc_ssd(&lcu->ref.v[index], &lcu->rec.joint_v[index], int ssd_v_joint = uvg_pixels_calc_ssd(&lcu->ref.v[index], &lcu->rec.v[index],
LCU_WIDTH_C, LCU_WIDTH_C, LCU_WIDTH_C, LCU_WIDTH_C,
chroma_width); chroma_width);
chroma_ssd = ssd_u_joint + ssd_v_joint; chroma_ssd = ssd_u_joint + ssd_v_joint;
@ -635,7 +639,7 @@ static double cu_rd_cost_tr_split_accurate(
} }
} }
if (uvg_is_lfnst_allowed(state, tr_cu, width, width, x_px, y_px, tree_type, depth == 4 || tree_type == UVG_CHROMA_T ? COLOR_UV : COLOR_Y)) { if (uvg_is_lfnst_allowed(state, tr_cu, width, width, x_px, y_px, tree_type, depth == 4 || tree_type == UVG_CHROMA_T ? COLOR_UV : COLOR_Y, lcu)) {
const int lfnst_idx = (depth != 4 && tree_type != UVG_CHROMA_T) ? tr_cu->lfnst_idx : tr_cu->cr_lfnst_idx; const int lfnst_idx = (depth != 4 && tree_type != UVG_CHROMA_T) ? tr_cu->lfnst_idx : tr_cu->cr_lfnst_idx;
CABAC_FBITS_UPDATE( CABAC_FBITS_UPDATE(
cabac, cabac,
@ -956,7 +960,7 @@ static double search_cu(
intra_search.pred_cu.type = CU_INTRA; intra_search.pred_cu.type = CU_INTRA;
} }
intra_search.pred_cu.intra.mode_chroma = intra_search.pred_cu.intra.mode; intra_search.pred_cu.intra.mode_chroma = intra_search.pred_cu.intra.mode;
if (ctrl->cfg.rdo >= 3 || ctrl->cfg.jccr || ctrl->cfg.lfnst) { if (ctrl->cfg.rdo >= 2 || ctrl->cfg.jccr || ctrl->cfg.lfnst) {
uvg_search_cu_intra_chroma(state, x, y, depth, lcu, &intra_search, tree_type); uvg_search_cu_intra_chroma(state, x, y, depth, lcu, &intra_search, tree_type);
if (intra_search.pred_cu.joint_cb_cr == 0) { if (intra_search.pred_cu.joint_cb_cr == 0) {
@ -1150,11 +1154,11 @@ static double search_cu(
memcpy(&state->search_cabac, &pre_search_cabac, sizeof(post_seach_cabac)); memcpy(&state->search_cabac, &pre_search_cabac, sizeof(post_seach_cabac));
state->search_cabac.update = 1;
double split_bits = 0; double split_bits = 0;
if (depth < MAX_DEPTH) { if (depth < MAX_DEPTH) {
state->search_cabac.update = 1;
// Add cost of cu_split_flag. // Add cost of cu_split_flag.
const cu_info_t* left_cu = NULL, * above_cu = NULL; const cu_info_t* left_cu = NULL, * above_cu = NULL;
if (x) { if (x) {
@ -1197,14 +1201,13 @@ static double search_cu(
// the split costs at least as much as not splitting. // the split costs at least as much as not splitting.
if (cur_cu->type == CU_NOTSET || cbf || state->encoder_control->cfg.cu_split_termination == UVG_CU_SPLIT_TERMINATION_OFF) { if (cur_cu->type == CU_NOTSET || cbf || state->encoder_control->cfg.cu_split_termination == UVG_CU_SPLIT_TERMINATION_OFF) {
if (split_cost < cost) split_cost += search_cu(state, x, y, depth + 1, work_tree, tree_type); if (split_cost < cost) split_cost += search_cu(state, x, y, depth + 1, work_tree, tree_type);
if (split_cost < cost) split_cost += search_cu(state, x + half_cu, y, depth + 1, work_tree, tree_type); if (split_cost < cost || 1) split_cost += search_cu(state, x + half_cu, y, depth + 1, work_tree, tree_type);
if (split_cost < cost) split_cost += search_cu(state, x, y + half_cu, depth + 1, work_tree, tree_type); if (split_cost < cost || 1) split_cost += search_cu(state, x, y + half_cu, depth + 1, work_tree, tree_type);
if (split_cost < cost) split_cost += search_cu(state, x + half_cu, y + half_cu, depth + 1, work_tree, tree_type); if (split_cost < cost || 1) split_cost += search_cu(state, x + half_cu, y + half_cu, depth + 1, work_tree, tree_type);
} else { } else {
split_cost = INT_MAX; split_cost = INT_MAX;
} }
// If no search is not performed for this depth, try just the best mode // If no search is not performed for this depth, try just the best mode
// of the top left CU from the next depth. This should ensure that 64x64 // of the top left CU from the next depth. This should ensure that 64x64
// gets used, at least in the most obvious cases, while avoiding any // gets used, at least in the most obvious cases, while avoiding any
@ -1230,7 +1233,7 @@ static double search_cu(
x > 0 ? LCU_GET_CU_AT_PX(lcu, SUB_SCU(x) - 1, SUB_SCU(y)) : NULL, x > 0 ? LCU_GET_CU_AT_PX(lcu, SUB_SCU(x) - 1, SUB_SCU(y)) : NULL,
y > 0 ? LCU_GET_CU_AT_PX(lcu, SUB_SCU(x), SUB_SCU(y) - 1) : NULL, y > 0 ? LCU_GET_CU_AT_PX(lcu, SUB_SCU(x), SUB_SCU(y) - 1) : NULL,
0, depth, cu_width, x, y, tree_type, 0, depth, cu_width, x, y, tree_type,
& split_bits); &bits);
cur_cu->intra = cu_d1->intra; cur_cu->intra = cu_d1->intra;
cur_cu->type = CU_INTRA; cur_cu->type = CU_INTRA;
@ -1238,6 +1241,8 @@ static double search_cu(
// Disable MRL in this case // Disable MRL in this case
cur_cu->intra.multi_ref_idx = 0; cur_cu->intra.multi_ref_idx = 0;
cur_cu->lfnst_idx = 0;
cur_cu->cr_lfnst_idx = 0;
uvg_lcu_fill_trdepth(lcu, x, y, depth, cur_cu->tr_depth, tree_type); uvg_lcu_fill_trdepth(lcu, x, y, depth, cur_cu->tr_depth, tree_type);
lcu_fill_cu_info(lcu, x_local, y_local, cu_width, cu_width, cur_cu); lcu_fill_cu_info(lcu, x_local, y_local, cu_width, cu_width, cur_cu);

80
src/search_intra.c
View file

@ -314,7 +314,7 @@ static double search_intra_trdepth(
int best_tr_idx = 0; int best_tr_idx = 0;
int best_lfnst_idx = 0; int best_lfnst_idx = 0;
int trafo; uint8_t trafo;
int num_transforms = 1; int num_transforms = 1;
if (pred_cu->tr_idx != MTS_TR_NUM) if (pred_cu->tr_idx != MTS_TR_NUM)
{ {
@ -362,16 +362,14 @@ static double search_intra_trdepth(
for (trafo = mts_start; trafo < num_transforms; trafo++) { for (trafo = mts_start; trafo < num_transforms; trafo++) {
pred_cu->tr_idx = trafo; pred_cu->tr_idx = trafo;
bool constraints[2] = { pred_cu->violates_lfnst_constrained_luma, pred_cu->tr_skip = trafo == MTS_SKIP;
pred_cu->lfnst_last_scan_pos }; bool constraints[2] = { false, false};
if (mts_enabled) { if (mts_enabled) {
pred_cu->mts_last_scan_pos = 0; pred_cu->mts_last_scan_pos = 0;
pred_cu->violates_mts_coeff_constraint = 0; pred_cu->violates_mts_coeff_constraint = 0;
if (trafo == MTS_SKIP && width > ( if ((trafo == MTS_SKIP && width > (1 << state->encoder_control->cfg.trskip_max_size))
1 << state->encoder_control->cfg.trskip_max_size)) { || !state->encoder_control->cfg.trskip_enable) {
//TODO: parametrize that this is not hardcoded
// TODO: this probably should currently trip for chroma?
continue; continue;
} }
} }
@ -380,16 +378,20 @@ static double search_intra_trdepth(
continue; continue;
} }
uvg_intra_recon_cu(
uvg_intra_recon_cu(state, state,
x_px, y_px, x_px,
depth, search_data, y_px,
depth,
search_data,
pred_cu, pred_cu,
lcu, lcu,
UVG_LUMA_T,true,false); UVG_LUMA_T,
true,
false);
if (trafo != 0 && !cbf_is_set(pred_cu->cbf, depth, COLOR_Y)) continue; if (trafo != 0 && !cbf_is_set(pred_cu->cbf, depth, COLOR_Y)) continue;
// TODO: Not sure if this should be 0 or 1 but at least seems to work with 1
derive_mts_constraints(pred_cu, lcu, depth, lcu_px); derive_mts_constraints(pred_cu, lcu, depth, lcu_px);
if (pred_cu->tr_idx > 1) { if (pred_cu->tr_idx > 1) {
if (pred_cu->violates_mts_coeff_constraint || !pred_cu-> if (pred_cu->violates_mts_coeff_constraint || !pred_cu->
@ -398,8 +400,12 @@ static double search_intra_trdepth(
} }
} }
const unsigned scan_offset = xy_to_zorder(LCU_WIDTH, lcu_px.x, lcu_px.y); const unsigned scan_offset = xy_to_zorder(
LCU_WIDTH,
lcu_px.x,
lcu_px.y);
if (trafo != MTS_SKIP && end_idx != 0) {
uvg_derive_lfnst_constraints( uvg_derive_lfnst_constraints(
pred_cu, pred_cu,
depth, depth,
@ -408,7 +414,7 @@ static double search_intra_trdepth(
width, width,
height height
); );
// Temp constraints. Updating the actual pred_cu constraints here will break things later }
if (!constraints[1] && cbf_is_set(pred_cu->cbf, depth, COLOR_Y)) { if (!constraints[1] && cbf_is_set(pred_cu->cbf, depth, COLOR_Y)) {
//end_idx = 0; //end_idx = 0;
@ -424,15 +430,21 @@ static double search_intra_trdepth(
pred_cu, pred_cu,
lcu); lcu);
double transform_bits = 0; double transform_bits = 0;
if(state->encoder_control->cfg.lfnst && depth == pred_cu->tr_depth) { if (state->encoder_control->cfg.lfnst && depth == pred_cu->tr_depth &&
if(!constraints[0] && constraints[1]) { trafo != MTS_SKIP) {
transform_bits += CTX_ENTROPY_FBITS(&state->search_cabac.ctx.lfnst_idx_model[tr_cu->depth == 4 || tree_type == UVG_LUMA_T], lfnst_idx != 0); if (!constraints[0] && constraints[1]) {
if(lfnst_idx > 0) { transform_bits += CTX_ENTROPY_FBITS(
transform_bits += CTX_ENTROPY_FBITS(&state->search_cabac.ctx.lfnst_idx_model[2], lfnst_idx == 2); &state->search_cabac.ctx.lfnst_idx_model[tr_cu->depth == 4 ||
tree_type == UVG_LUMA_T],
lfnst_idx != 0);
if (lfnst_idx > 0) {
transform_bits += CTX_ENTROPY_FBITS(
&state->search_cabac.ctx.lfnst_idx_model[2],
lfnst_idx == 2);
} }
} }
} }
if (num_transforms > 1 && trafo != MTS_SKIP && width <= 32 if (num_transforms > 2 && trafo != MTS_SKIP && width <= 32
/*&& height <= 32*/ /*&& height <= 32*/
&& !pred_cu->violates_mts_coeff_constraint && pred_cu-> && !pred_cu->violates_mts_coeff_constraint && pred_cu->
mts_last_scan_pos && lfnst_idx == 0) { mts_last_scan_pos && lfnst_idx == 0) {
@ -454,7 +466,9 @@ static double search_intra_trdepth(
} }
rd_cost += transform_bits * state->frame->lambda; rd_cost += transform_bits * state->frame->lambda;
search_data->lfnst_costs[lfnst_idx] = MIN(search_data->lfnst_costs[lfnst_idx], rd_cost); search_data->lfnst_costs[lfnst_idx] = MIN(
search_data->lfnst_costs[lfnst_idx],
rd_cost);
if (rd_cost < best_rd_cost) { if (rd_cost < best_rd_cost) {
best_rd_cost = rd_cost; best_rd_cost = rd_cost;
best_lfnst_idx = pred_cu->lfnst_idx; best_lfnst_idx = pred_cu->lfnst_idx;
@ -468,7 +482,10 @@ static double search_intra_trdepth(
pred_cu->intra.mode_chroma = chroma_mode; pred_cu->intra.mode_chroma = chroma_mode;
pred_cu->joint_cb_cr = 4; pred_cu->joint_cb_cr = 4;
// TODO: Maybe check the jccr mode here also but holy shit is the interface of search_intra_rdo bad currently // TODO: Maybe check the jccr mode here also but holy shit is the interface of search_intra_rdo bad currently
const unsigned scan_offset = xy_to_zorder(LCU_WIDTH_C, lcu_px.x, lcu_px.y); const unsigned scan_offset = xy_to_zorder(
LCU_WIDTH_C,
lcu_px.x,
lcu_px.y);
uvg_intra_recon_cu( uvg_intra_recon_cu(
state, state,
x_px, x_px,
@ -477,7 +494,9 @@ static double search_intra_trdepth(
search_data, search_data,
pred_cu, pred_cu,
lcu, lcu,
UVG_BOTH_T,false,true); UVG_BOTH_T,
false,
true);
best_rd_cost += uvg_cu_rd_cost_chroma( best_rd_cost += uvg_cu_rd_cost_chroma(
state, state,
lcu_px.x, lcu_px.x,
@ -497,8 +516,8 @@ static double search_intra_trdepth(
depth, depth,
constraints, constraints,
&lcu->coeff.u[scan_offset], &lcu->coeff.u[scan_offset],
width_c,
width_c width_c
, width_c
); );
if (constraints[0] || !constraints[1]) { if (constraints[0] || !constraints[1]) {
best_lfnst_idx = 0; best_lfnst_idx = 0;
@ -509,8 +528,8 @@ static double search_intra_trdepth(
depth, depth,
constraints, constraints,
&lcu->coeff.u[scan_offset], &lcu->coeff.u[scan_offset],
width_c,
width_c width_c
, width_c
); );
if (constraints[0] || !constraints[1]) { if (constraints[0] || !constraints[1]) {
best_lfnst_idx = 0; best_lfnst_idx = 0;
@ -1621,7 +1640,7 @@ int8_t uvg_search_cu_intra_chroma(
// const int8_t modes_in_depth[5] = { 1, 1, 1, 1, 2 }; // const int8_t modes_in_depth[5] = { 1, 1, 1, 1, 2 };
int num_modes = 1; int num_modes = 1;
if (state->encoder_control->cfg.rdo >= 3) { if (state->encoder_control->cfg.rdo >= 2 || tree_type == UVG_CHROMA_T) {
num_modes = total_modes; num_modes = total_modes;
} }
@ -1666,9 +1685,12 @@ int8_t uvg_search_cu_intra_chroma(
tree_type); tree_type);
} }
if (num_modes > 1) { if (num_modes > 1 || state->encoder_control->cfg.jccr) {
uvg_search_intra_chroma_rdo(state, x_px, y_px, depth, num_modes, lcu, chroma_data, intra_mode, tree_type); uvg_search_intra_chroma_rdo(state, x_px, y_px, depth, num_modes, lcu, chroma_data, intra_mode, tree_type);
} }
else if(cur_pu->lfnst_idx) {
chroma_data[0].pred_cu.cr_lfnst_idx = cur_pu->lfnst_idx;
}
*search_data = chroma_data[0]; *search_data = chroma_data[0];
return chroma_data[0].pred_cu.intra.mode_chroma; return chroma_data[0].pred_cu.intra.mode_chroma;
} }
@ -1844,6 +1866,7 @@ void uvg_search_cu_intra(
search_data[i].cost = MAX_INT; search_data[i].cost = MAX_INT;
} }
number_of_modes = UVG_NUM_INTRA_MODES; number_of_modes = UVG_NUM_INTRA_MODES;
num_regular_modes = UVG_NUM_INTRA_MODES;
} }
uint8_t num_mrl_modes = 0; uint8_t num_mrl_modes = 0;
@ -1969,7 +1992,6 @@ void uvg_search_cu_intra(
} }
} }
// TODO: if rough search is implemented for MIP, sort mip_modes here.
search_intra_rdo( search_intra_rdo(
state, state,
x_px, x_px,

20
src/strategies/avx2/picture-avx2.c
View file

@ -233,7 +233,10 @@ static unsigned satd_4x4_8bit_avx2(const uint8_t *org, const uint8_t *cur)
row3 = _mm_add_epi16(row3, _mm_shuffle_epi32(row3, _MM_SHUFFLE(0, 1, 0, 1) )); row3 = _mm_add_epi16(row3, _mm_shuffle_epi32(row3, _MM_SHUFFLE(0, 1, 0, 1) ));
row3 = _mm_add_epi16(row3, _mm_shufflelo_epi16(row3, _MM_SHUFFLE(0, 1, 0, 1) )); row3 = _mm_add_epi16(row3, _mm_shufflelo_epi16(row3, _MM_SHUFFLE(0, 1, 0, 1) ));
const int16_t temp1 = _mm_extract_epi16(row2, 0);
const int dc1 = abs(temp1);
unsigned sum = _mm_extract_epi16(row3, 0); unsigned sum = _mm_extract_epi16(row3, 0);
sum -= dc1 - (dc1 >> 2);
unsigned satd = (sum + 1) >> 1; unsigned satd = (sum + 1) >> 1;
return satd; return satd;
@ -280,10 +283,16 @@ static void satd_8bit_4x4_dual_avx2(
row3 = _mm256_add_epi16(row3, _mm256_shuffle_epi32(row3, _MM_SHUFFLE(0, 1, 0, 1) )); row3 = _mm256_add_epi16(row3, _mm256_shuffle_epi32(row3, _MM_SHUFFLE(0, 1, 0, 1) ));
row3 = _mm256_add_epi16(row3, _mm256_shufflelo_epi16(row3, _MM_SHUFFLE(0, 1, 0, 1) )); row3 = _mm256_add_epi16(row3, _mm256_shufflelo_epi16(row3, _MM_SHUFFLE(0, 1, 0, 1) ));
const int16_t temp2 = _mm256_extract_epi16(row2, 0);
const int dc1 = abs(temp2);
unsigned sum1 = _mm_extract_epi16(_mm256_castsi256_si128(row3), 0); unsigned sum1 = _mm_extract_epi16(_mm256_castsi256_si128(row3), 0);
sum1 -= dc1 - (dc1 >> 2);
sum1 = (sum1 + 1) >> 1; sum1 = (sum1 + 1) >> 1;
const int16_t temp3 = _mm256_extract_epi16(row2, 8);
const int dc2 = abs(temp3);
unsigned sum2 = _mm_extract_epi16(_mm256_extracti128_si256(row3, 1), 0); unsigned sum2 = _mm_extract_epi16(_mm256_extracti128_si256(row3, 1), 0);
sum2 -= dc2 - (dc2 >> 2);
sum2 = (sum2 + 1) >> 1; sum2 = (sum2 + 1) >> 1;
satds_out[0] = sum1; satds_out[0] = sum1;
@ -522,6 +531,13 @@ static void uvg_satd_8bit_8x8_general_dual_avx2(const uint8_t * buf1, unsigned s
sum_block_dual_avx2(temp, sum0, sum1); sum_block_dual_avx2(temp, sum0, sum1);
const int16_t temp2 = _mm256_extract_epi16(temp[0], 0);
const int dc1 = abs(temp2);
const int16_t temp3 = _mm256_extract_epi16(temp[0], 8);
const int dc2 = abs(temp3);
*sum0 -= dc1 - (dc1 >> 2);
*sum1 -= dc2 - (dc2 >> 2);
*sum0 = (*sum0 + 2) >> 2; *sum0 = (*sum0 + 2) >> 2;
*sum1 = (*sum1 + 2) >> 2; *sum1 = (*sum1 + 2) >> 2;
} }
@ -558,6 +574,10 @@ static unsigned satd_8x8_subblock_8bit_avx2(const uint8_t * buf1, unsigned strid
unsigned sad = sum_block_avx2(temp); unsigned sad = sum_block_avx2(temp);
const int16_t temp1 = _mm_extract_epi16(temp[0], 0);
const int dc1 = abs(temp1);
sad -= dc1 - (dc1 >> 2);
unsigned result = (sad + 2) >> 2; unsigned result = (sad + 2) >> 2;
return result; return result;
} }

2
src/transform.c
View file

@ -703,7 +703,7 @@ void uvg_chroma_transform_search(
transforms[i] == CHROMA_TS); transforms[i] == CHROMA_TS);
} }
if((depth == 4 || tree_type == UVG_CHROMA_T) && state->encoder_control->cfg.lfnst && 0) { if((depth == 4 || tree_type == UVG_CHROMA_T) && state->encoder_control->cfg.lfnst && 0) {
if(uvg_is_lfnst_allowed(state, pred_cu, width, height, 0, 0 , UVG_CHROMA_T, COLOR_UV)) { if(uvg_is_lfnst_allowed(state, pred_cu, width, height, 0, 0 , UVG_CHROMA_T, COLOR_UV, lcu)) {
const int lfnst_idx = pred_cu->cr_lfnst_idx; const int lfnst_idx = pred_cu->cr_lfnst_idx;
CABAC_FBITS_UPDATE( CABAC_FBITS_UPDATE(
&state->search_cabac, &state->search_cabac,

6
tests/satd_tests.c
View file

@ -119,7 +119,7 @@ static void satd_tear_down_tests()
TEST satd_test_black_and_white(void) TEST satd_test_black_and_white(void)
{ {
const int satd_results[5] = {2040, 4080, 16320, 65280, 261120}; const int satd_results[5] = {510, 1020, 4080, 16320, 65280};
const int test = 0; const int test = 0;
@ -137,7 +137,7 @@ TEST satd_test_black_and_white(void)
TEST satd_test_checkers(void) TEST satd_test_checkers(void)
{ {
const int satd_checkers_results[5] = { 2040, 4080, 16320, 65280, 261120 }; const int satd_checkers_results[5] = { 1278, 2556, 10224, 40896, 163584 };
const int test = 1; const int test = 1;
@ -156,7 +156,7 @@ TEST satd_test_checkers(void)
TEST satd_test_gradient(void) TEST satd_test_gradient(void)
{ {
const int satd_gradient_results[5] = {3140,9004,20481,67262,258672}; const int satd_gradient_results[5] = {2728,7158,10775,23399,72780};
const int test = 2; const int test = 2;

4
tests/tests_main.c
View file

@ -38,7 +38,7 @@ GREATEST_MAIN_DEFS();
#if UVG_BIT_DEPTH == 8 #if UVG_BIT_DEPTH == 8
extern SUITE(sad_tests); extern SUITE(sad_tests);
extern SUITE(intra_sad_tests); extern SUITE(intra_sad_tests);
// extern SUITE(satd_tests); extern SUITE(satd_tests);
extern SUITE(speed_tests); extern SUITE(speed_tests);
extern SUITE(dct_tests); extern SUITE(dct_tests);
extern SUITE(mts_tests); extern SUITE(mts_tests);
@ -56,7 +56,7 @@ int main(int argc, char **argv)
#if UVG_BIT_DEPTH == 8 #if UVG_BIT_DEPTH == 8
RUN_SUITE(sad_tests); RUN_SUITE(sad_tests);
RUN_SUITE(intra_sad_tests); RUN_SUITE(intra_sad_tests);
// RUN_SUITE(satd_tests); RUN_SUITE(satd_tests);
RUN_SUITE(dct_tests); RUN_SUITE(dct_tests);
RUN_SUITE(mts_tests); RUN_SUITE(mts_tests);