mirror of
https://github.com/ultravideo/uvg266.git
synced 2024-11-27 19:24:06 +00:00
Merge branch 'smp-merge-analysis'
This commit is contained in:
Ari Lemmetti
commit
4e94d60552
101
src/inter.c
101
src/inter.c
|
|
@ -519,48 +519,75 @@ void kvz_inter_recon_cu(const encoder_state_t * const state,
|
|||
bool predict_chroma)
|
||||
{
|
||||
cu_info_t *cu = LCU_GET_CU_AT_PX(lcu, SUB_SCU(x), SUB_SCU(y));
|
||||
|
||||
const int num_pu = kvz_part_mode_num_parts[cu->part_size];
|
||||
for (int i = 0; i < num_pu; ++i) {
|
||||
const int pu_x = PU_GET_X(cu->part_size, width, x, i);
|
||||
const int pu_y = PU_GET_Y(cu->part_size, width, y, i);
|
||||
const int pu_w = PU_GET_W(cu->part_size, width, i);
|
||||
const int pu_h = PU_GET_H(cu->part_size, width, i);
|
||||
kvz_inter_pred_pu(state, lcu, x, y, width, predict_luma, predict_chroma, i);
|
||||
}
|
||||
}
|
||||
|
||||
cu_info_t *pu = LCU_GET_CU_AT_PX(lcu, SUB_SCU(pu_x), SUB_SCU(pu_y));
|
||||
/**
|
||||
* Predict a single PU.
|
||||
*
|
||||
* The PU may use either uniprediction or biprediction.
|
||||
*
|
||||
* \param state encoder state
|
||||
* \param lcu containing LCU
|
||||
* \param x x-coordinate of the CU in pixels
|
||||
* \param y y-coordinate of the CU in pixels
|
||||
* \param width CU width
|
||||
* \param predict_luma Enable or disable luma prediction for this call.
|
||||
* \param predict_chroma Enable or disable chroma prediction for this call.
|
||||
* \param i_pu Index of the PU. Always zero for 2Nx2N. Used for SMP+AMP.
|
||||
*/
|
||||
void kvz_inter_pred_pu(const encoder_state_t * const state,
|
||||
lcu_t *lcu,
|
||||
int32_t x,
|
||||
int32_t y,
|
||||
int32_t width,
|
||||
bool predict_luma,
|
||||
bool predict_chroma,
|
||||
int i_pu)
|
||||
|
||||
if (pu->inter.mv_dir == 3) {
|
||||
const kvz_picture *const refs[2] = {
|
||||
state->frame->ref->images[
|
||||
state->frame->ref_LX[0][
|
||||
pu->inter.mv_ref[0]]],
|
||||
state->frame->ref->images[
|
||||
state->frame->ref_LX[1][
|
||||
pu->inter.mv_ref[1]]],
|
||||
};
|
||||
kvz_inter_recon_bipred(state,
|
||||
refs[0], refs[1],
|
||||
pu_x, pu_y,
|
||||
pu_w, pu_h,
|
||||
pu->inter.mv,
|
||||
lcu,
|
||||
predict_luma, predict_chroma);
|
||||
} else {
|
||||
const int mv_idx = pu->inter.mv_dir - 1;
|
||||
const kvz_picture *const ref =
|
||||
state->frame->ref->images[
|
||||
state->frame->ref_LX[mv_idx][
|
||||
pu->inter.mv_ref[mv_idx]]];
|
||||
{
|
||||
cu_info_t *cu = LCU_GET_CU_AT_PX(lcu, SUB_SCU(x), SUB_SCU(y));
|
||||
const int pu_x = PU_GET_X(cu->part_size, width, x, i_pu);
|
||||
const int pu_y = PU_GET_Y(cu->part_size, width, y, i_pu);
|
||||
const int pu_w = PU_GET_W(cu->part_size, width, i_pu);
|
||||
const int pu_h = PU_GET_H(cu->part_size, width, i_pu);
|
||||
cu_info_t *pu = LCU_GET_CU_AT_PX(lcu, SUB_SCU(pu_x), SUB_SCU(pu_y));
|
||||
|
||||
inter_recon_unipred(state,
|
||||
ref,
|
||||
pu_x, pu_y,
|
||||
pu_w, pu_h,
|
||||
pu->inter.mv[mv_idx],
|
||||
lcu,
|
||||
NULL,
|
||||
predict_luma, predict_chroma);
|
||||
}
|
||||
if (pu->inter.mv_dir == 3) {
|
||||
const kvz_picture *const refs[2] = {
|
||||
state->frame->ref->images[
|
||||
state->frame->ref_LX[0][
|
||||
pu->inter.mv_ref[0]]],
|
||||
state->frame->ref->images[
|
||||
state->frame->ref_LX[1][
|
||||
pu->inter.mv_ref[1]]],
|
||||
};
|
||||
kvz_inter_recon_bipred(state,
|
||||
refs[0], refs[1],
|
||||
pu_x, pu_y,
|
||||
pu_w, pu_h,
|
||||
pu->inter.mv,
|
||||
lcu,
|
||||
predict_luma, predict_chroma);
|
||||
}
|
||||
else {
|
||||
const int mv_idx = pu->inter.mv_dir - 1;
|
||||
const kvz_picture *const ref =
|
||||
state->frame->ref->images[
|
||||
state->frame->ref_LX[mv_idx][
|
||||
pu->inter.mv_ref[mv_idx]]];
|
||||
|
||||
inter_recon_unipred(state,
|
||||
ref,
|
||||
pu_x, pu_y,
|
||||
pu_w, pu_h,
|
||||
pu->inter.mv[mv_idx],
|
||||
lcu,
|
||||
NULL,
|
||||
predict_luma, predict_chroma);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -48,6 +48,15 @@ void kvz_inter_recon_cu(const encoder_state_t * const state,
|
|||
bool predict_luma,
|
||||
bool predict_chroma);
|
||||
|
||||
void kvz_inter_pred_pu(const encoder_state_t * const state,
|
||||
lcu_t *lcu,
|
||||
int32_t x,
|
||||
int32_t y,
|
||||
int32_t width,
|
||||
bool predict_luma,
|
||||
bool predict_chroma,
|
||||
int i_pu);
|
||||
|
||||
void kvz_inter_recon_bipred(const encoder_state_t * const state,
|
||||
const kvz_picture * ref1,
|
||||
const kvz_picture * ref2,
|
||||
|
|
|
|||
|
|
@ -1551,87 +1551,88 @@ static void search_pu_inter(encoder_state_t * const state,
|
|||
mrg_costs[i] = MAX_DOUBLE;
|
||||
}
|
||||
|
||||
if (cur_cu->part_size == SIZE_2Nx2N) {
|
||||
int num_rdo_cands = 0;
|
||||
|
||||
int num_rdo_cands = 0;
|
||||
// Check motion vector constraints and perform rough search
|
||||
for (int merge_idx = 0; merge_idx < info.num_merge_cand; ++merge_idx) {
|
||||
|
||||
// Check motion vector constraints and perform rough search
|
||||
for (int merge_idx = 0; merge_idx < info.num_merge_cand; ++merge_idx) {
|
||||
inter_merge_cand_t *cur_cand = &info.merge_cand[merge_idx];
|
||||
cur_cu->inter.mv_dir = cur_cand->dir;
|
||||
cur_cu->inter.mv_ref[0] = cur_cand->ref[0];
|
||||
cur_cu->inter.mv_ref[1] = cur_cand->ref[1];
|
||||
cur_cu->inter.mv[0][0] = cur_cand->mv[0][0];
|
||||
cur_cu->inter.mv[0][1] = cur_cand->mv[0][1];
|
||||
cur_cu->inter.mv[1][0] = cur_cand->mv[1][0];
|
||||
cur_cu->inter.mv[1][1] = cur_cand->mv[1][1];
|
||||
|
||||
inter_merge_cand_t *cur_cand = &info.merge_cand[merge_idx];
|
||||
cur_cu->inter.mv_dir = cur_cand->dir;
|
||||
cur_cu->inter.mv_ref[0] = cur_cand->ref[0];
|
||||
cur_cu->inter.mv_ref[1] = cur_cand->ref[1];
|
||||
cur_cu->inter.mv[0][0] = cur_cand->mv[0][0];
|
||||
cur_cu->inter.mv[0][1] = cur_cand->mv[0][1];
|
||||
cur_cu->inter.mv[1][0] = cur_cand->mv[1][0];
|
||||
cur_cu->inter.mv[1][1] = cur_cand->mv[1][1];
|
||||
// If bipred is not enabled, do not try candidates with mv_dir == 3.
|
||||
// Bipred is also forbidden for 4x8 and 8x4 blocks by the standard.
|
||||
if (cur_cu->inter.mv_dir == 3 && !state->encoder_control->cfg.bipred) continue;
|
||||
if (cur_cu->inter.mv_dir == 3 && !(width + height > 12)) continue;
|
||||
|
||||
if (cur_cu->inter.mv_dir == 3 && !state->encoder_control->cfg.bipred) continue;
|
||||
bool is_duplicate = merge_candidate_in_list(info.merge_cand, cur_cand,
|
||||
mrg_cands,
|
||||
num_rdo_cands);
|
||||
bool is_duplicate = merge_candidate_in_list(info.merge_cand, cur_cand,
|
||||
mrg_cands,
|
||||
num_rdo_cands);
|
||||
|
||||
// Don't try merge candidates that don't satisfy mv constraints.
|
||||
// Don't add duplicates to list
|
||||
if (!fracmv_within_tile(&info, cur_cu->inter.mv[0][0], cur_cu->inter.mv[0][1]) ||
|
||||
!fracmv_within_tile(&info, cur_cu->inter.mv[1][0], cur_cu->inter.mv[1][1]) ||
|
||||
is_duplicate)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
kvz_inter_recon_cu(state, lcu, x, y, width, true, false);
|
||||
mrg_costs[num_rdo_cands] = kvz_satd_any_size(width, height,
|
||||
lcu->rec.y + y_local * LCU_WIDTH + x_local, LCU_WIDTH,
|
||||
lcu->ref.y + y_local * LCU_WIDTH + x_local, LCU_WIDTH);
|
||||
|
||||
mrg_cands[num_rdo_cands] = merge_idx;
|
||||
num_rdo_cands++;
|
||||
// Don't try merge candidates that don't satisfy mv constraints.
|
||||
// Don't add duplicates to list
|
||||
if (!fracmv_within_tile(&info, cur_cu->inter.mv[0][0], cur_cu->inter.mv[0][1]) ||
|
||||
!fracmv_within_tile(&info, cur_cu->inter.mv[1][0], cur_cu->inter.mv[1][1]) ||
|
||||
is_duplicate)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
// Sort candidates by cost
|
||||
kvz_sort_modes(mrg_cands, mrg_costs, num_rdo_cands);
|
||||
kvz_inter_pred_pu(state, lcu, x_cu, y_cu, width_cu, true, false, i_pu);
|
||||
mrg_costs[num_rdo_cands] = kvz_satd_any_size(width, height,
|
||||
lcu->rec.y + y_local * LCU_WIDTH + x_local, LCU_WIDTH,
|
||||
lcu->ref.y + y_local * LCU_WIDTH + x_local, LCU_WIDTH);
|
||||
|
||||
// Limit by availability
|
||||
// TODO: Do not limit to just 1
|
||||
num_rdo_cands = MIN(1, num_rdo_cands);
|
||||
mrg_cands[num_rdo_cands] = merge_idx;
|
||||
num_rdo_cands++;
|
||||
}
|
||||
|
||||
// Sort candidates by cost
|
||||
kvz_sort_modes(mrg_cands, mrg_costs, num_rdo_cands);
|
||||
|
||||
// Limit by availability
|
||||
// TODO: Do not limit to just 1
|
||||
num_rdo_cands = MIN(1, num_rdo_cands);
|
||||
|
||||
// Early Skip Mode Decision
|
||||
bool has_chroma = state->encoder_control->chroma_format != KVZ_CSP_400;
|
||||
if (cfg->early_skip) {
|
||||
for (int merge_rdo_idx = 0; merge_rdo_idx < num_rdo_cands; ++merge_rdo_idx) {
|
||||
// Early Skip Mode Decision
|
||||
bool has_chroma = state->encoder_control->chroma_format != KVZ_CSP_400;
|
||||
if (cfg->early_skip && cur_cu->part_size == SIZE_2Nx2N) {
|
||||
for (int merge_rdo_idx = 0; merge_rdo_idx < num_rdo_cands; ++merge_rdo_idx) {
|
||||
|
||||
// Reconstruct blocks with merge candidate.
|
||||
// Check luma CBF. Then, check chroma CBFs if luma CBF is not set
|
||||
// and chroma exists.
|
||||
// Early terminate if merge candidate with zero CBF is found.
|
||||
int merge_idx = mrg_cands[merge_rdo_idx];
|
||||
cur_cu->inter.mv_dir = info.merge_cand[merge_idx].dir;
|
||||
cur_cu->inter.mv_ref[0] = info.merge_cand[merge_idx].ref[0];
|
||||
cur_cu->inter.mv_ref[1] = info.merge_cand[merge_idx].ref[1];
|
||||
cur_cu->inter.mv[0][0] = info.merge_cand[merge_idx].mv[0][0];
|
||||
cur_cu->inter.mv[0][1] = info.merge_cand[merge_idx].mv[0][1];
|
||||
cur_cu->inter.mv[1][0] = info.merge_cand[merge_idx].mv[1][0];
|
||||
cur_cu->inter.mv[1][1] = info.merge_cand[merge_idx].mv[1][1];
|
||||
kvz_lcu_fill_trdepth(lcu, x, y, depth, MAX(1, depth));
|
||||
kvz_inter_recon_cu(state, lcu, x, y, width, true, false);
|
||||
kvz_quantize_lcu_residual(state, true, false, x, y, depth, cur_cu, lcu);
|
||||
// Reconstruct blocks with merge candidate.
|
||||
// Check luma CBF. Then, check chroma CBFs if luma CBF is not set
|
||||
// and chroma exists.
|
||||
// Early terminate if merge candidate with zero CBF is found.
|
||||
int merge_idx = mrg_cands[merge_rdo_idx];
|
||||
cur_cu->inter.mv_dir = info.merge_cand[merge_idx].dir;
|
||||
cur_cu->inter.mv_ref[0] = info.merge_cand[merge_idx].ref[0];
|
||||
cur_cu->inter.mv_ref[1] = info.merge_cand[merge_idx].ref[1];
|
||||
cur_cu->inter.mv[0][0] = info.merge_cand[merge_idx].mv[0][0];
|
||||
cur_cu->inter.mv[0][1] = info.merge_cand[merge_idx].mv[0][1];
|
||||
cur_cu->inter.mv[1][0] = info.merge_cand[merge_idx].mv[1][0];
|
||||
cur_cu->inter.mv[1][1] = info.merge_cand[merge_idx].mv[1][1];
|
||||
kvz_lcu_fill_trdepth(lcu, x, y, depth, MAX(1, depth));
|
||||
kvz_inter_recon_cu(state, lcu, x, y, width, true, false);
|
||||
kvz_quantize_lcu_residual(state, true, false, x, y, depth, cur_cu, lcu);
|
||||
|
||||
if (cbf_is_set(cur_cu->cbf, depth, COLOR_Y)) {
|
||||
continue;
|
||||
}
|
||||
else if (has_chroma) {
|
||||
kvz_inter_recon_cu(state, lcu, x, y, width, false, has_chroma);
|
||||
kvz_quantize_lcu_residual(state, false, has_chroma, x, y, depth, cur_cu, lcu);
|
||||
if (!cbf_is_set_any(cur_cu->cbf, depth)) {
|
||||
cur_cu->type = CU_INTER;
|
||||
cur_cu->merge_idx = merge_idx;
|
||||
cur_cu->skipped = true;
|
||||
*inter_cost = 0.0; // TODO: Check this
|
||||
*inter_bitcost = 0; // TODO: Check this
|
||||
return;
|
||||
}
|
||||
if (cbf_is_set(cur_cu->cbf, depth, COLOR_Y)) {
|
||||
continue;
|
||||
}
|
||||
else if (has_chroma) {
|
||||
kvz_inter_recon_cu(state, lcu, x, y, width, false, has_chroma);
|
||||
kvz_quantize_lcu_residual(state, false, has_chroma, x, y, depth, cur_cu, lcu);
|
||||
if (!cbf_is_set_any(cur_cu->cbf, depth)) {
|
||||
cur_cu->type = CU_INTER;
|
||||
cur_cu->merge_idx = merge_idx;
|
||||
cur_cu->skipped = true;
|
||||
*inter_cost = 0.0; // TODO: Check this
|
||||
*inter_bitcost = 0; // TODO: Check this
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in a new issue