hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-27 11:24:05 +00:00
Code Issues Actions Projects Releases Wiki Activity

[transform] Simplify chroma transform search a bit

Browse source
This commit is contained in:
Joose Sainio 2022-11-08 13:01:27 +02:00 committed by Marko Viitanen
parent 58c6af8c87
commit 274e71dff6
1 changed files with 26 additions and 31 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

57
src/transform.c
View file

@ -415,14 +415,11 @@ static void generate_jccr_transforms(
static void quantize_chroma( static void quantize_chroma(
encoder_state_t* const state, encoder_state_t* const state,
int depth,
int8_t width, int8_t width,
int8_t height, int8_t height,
coeff_t u_coeff[5120], coeff_t u_coeff[5120],
coeff_t v_coeff[2048], coeff_t v_coeff[2048],
enum uvg_chroma_transforms transforms[5], enum uvg_chroma_transforms transform,
const int trans_offset,
int i,
coeff_t u_quant_coeff[1024], coeff_t u_quant_coeff[1024],
coeff_t v_quant_coeff[1024], coeff_t v_quant_coeff[1024],
const coeff_scan_order_t scan_order, const coeff_scan_order_t scan_order,
@ -431,9 +428,9 @@ static void quantize_chroma(
uint8_t lfnst_idx) uint8_t lfnst_idx)
{ {
if (state->encoder_control->cfg.rdoq_enable && if (state->encoder_control->cfg.rdoq_enable &&
(transforms[i] != CHROMA_TS || !state->encoder_control->cfg.rdoq_skip)) (transform != CHROMA_TS || !state->encoder_control->cfg.rdoq_skip))
{ {
uvg_rdoq(state, &u_coeff[i * trans_offset], u_quant_coeff, width, height, transforms[i] != JCCR_1 ? COLOR_U : COLOR_V, uvg_rdoq(state, u_coeff, u_quant_coeff, width, height, transform != JCCR_1 ? COLOR_U : COLOR_V,
scan_order, CU_INTRA, 0, lfnst_idx); scan_order, CU_INTRA, 0, lfnst_idx);
int j; int j;
@ -444,25 +441,25 @@ static void quantize_chroma(
} }
} }
if (transforms[i] == DCT7_CHROMA) { if (transform == DCT7_CHROMA) {
uint16_t temp_cbf = 0; uint16_t temp_cbf = 0;
if (*u_has_coeffs)cbf_set(&temp_cbf, COLOR_U); if (*u_has_coeffs)cbf_set(&temp_cbf, COLOR_U);
uvg_rdoq(state, &v_coeff[i * trans_offset], v_quant_coeff, width, height, COLOR_V, uvg_rdoq(state, v_coeff, v_quant_coeff, width, height, COLOR_V,
scan_order, CU_INTRA, temp_cbf, lfnst_idx); scan_order, CU_INTRA, temp_cbf, lfnst_idx);
} }
} }
else if (state->encoder_control->cfg.rdoq_enable && transforms[i] == CHROMA_TS) { else if (state->encoder_control->cfg.rdoq_enable && transform == CHROMA_TS) {
uvg_ts_rdoq(state, &u_coeff[i * trans_offset], u_quant_coeff, width, height, COLOR_U, scan_order); uvg_ts_rdoq(state, u_coeff, u_quant_coeff, width, height, COLOR_U, scan_order);
uvg_ts_rdoq(state, &v_coeff[i * trans_offset], v_quant_coeff, width, height, COLOR_V, scan_order); uvg_ts_rdoq(state, v_coeff, v_quant_coeff, width, height, COLOR_V, scan_order);
} }
else { else {
uvg_quant(state, &u_coeff[i * trans_offset], u_quant_coeff, width, height, transforms[i] != JCCR_1 ? COLOR_U : COLOR_V, uvg_quant(state, u_coeff, u_quant_coeff, width, height, transform != JCCR_1 ? COLOR_U : COLOR_V,
scan_order, CU_INTRA, transforms[i] == CHROMA_TS, lfnst_idx); scan_order, CU_INTRA, transform == CHROMA_TS, lfnst_idx);
if (!IS_JCCR_MODE(transforms[i])) { if (!IS_JCCR_MODE(transform)) {
uvg_quant(state, &v_coeff[i * trans_offset], v_quant_coeff, width, height, COLOR_V, uvg_quant(state, v_coeff, v_quant_coeff, width, height, COLOR_V,
scan_order, CU_INTRA, transforms[i] == CHROMA_TS, lfnst_idx); scan_order, CU_INTRA, transform == CHROMA_TS, lfnst_idx);
} }
} }
@ -472,7 +469,7 @@ static void quantize_chroma(
break; break;
} }
} }
if (!IS_JCCR_MODE(transforms[i])) { if (!IS_JCCR_MODE(transform)) {
for (int j = 0; j < width * height; ++j) { for (int j = 0; j < width * height; ++j) {
if (v_quant_coeff[j]) { if (v_quant_coeff[j]) {
*v_has_coeffs = 1; *v_has_coeffs = 1;
@ -498,7 +495,7 @@ void uvg_chroma_transform_search(
{ {
ALIGNED(64) coeff_t u_coeff[LCU_WIDTH_C * LCU_WIDTH_C * 5]; ALIGNED(64) coeff_t u_coeff[LCU_WIDTH_C * LCU_WIDTH_C * 5];
ALIGNED(64) uint8_t u_recon[LCU_WIDTH_C * LCU_WIDTH_C * 5]; ALIGNED(64) uint8_t u_recon[LCU_WIDTH_C * LCU_WIDTH_C * 5];
ALIGNED(64) coeff_t v_coeff[LCU_WIDTH_C * LCU_WIDTH_C * 2]; ALIGNED(64) coeff_t v_coeff[LCU_WIDTH_C * LCU_WIDTH_C * 2]; // In case of JCCR the v channel does not have coefficients
ALIGNED(64) uint8_t v_recon[LCU_WIDTH_C * LCU_WIDTH_C * 5]; ALIGNED(64) uint8_t v_recon[LCU_WIDTH_C * LCU_WIDTH_C * 5];
const int width = cu_loc->chroma_width; const int width = cu_loc->chroma_width;
const int height = cu_loc->chroma_height; const int height = cu_loc->chroma_height;
@ -553,22 +550,20 @@ void uvg_chroma_transform_search(
int16_t v_recon_resi[LCU_WIDTH_C * LCU_WIDTH_C]; int16_t v_recon_resi[LCU_WIDTH_C * LCU_WIDTH_C];
bool u_has_coeffs = false; bool u_has_coeffs = false;
bool v_has_coeffs = false; bool v_has_coeffs = false;
bool is_jccr = IS_JCCR_MODE(transforms[i]);
if(pred_cu->cr_lfnst_idx) { if(pred_cu->cr_lfnst_idx) {
uvg_fwd_lfnst(pred_cu, width, height, COLOR_U, pred_cu->cr_lfnst_idx, &u_coeff[i * trans_offset], tree_type); uvg_fwd_lfnst(pred_cu, width, height, COLOR_U, pred_cu->cr_lfnst_idx, &u_coeff[i * trans_offset], tree_type);
if (!IS_JCCR_MODE(transforms[i])) { if (!is_jccr) {
uvg_fwd_lfnst(pred_cu, width, height, COLOR_V, pred_cu->cr_lfnst_idx, &v_coeff[i * trans_offset], tree_type); uvg_fwd_lfnst(pred_cu, width, height, COLOR_V, pred_cu->cr_lfnst_idx, &v_coeff[i * trans_offset], tree_type);
} }
} }
quantize_chroma( quantize_chroma(
state, state,
depth,
width, width,
height, height,
u_coeff, &u_coeff[i * trans_offset],
v_coeff, &v_coeff[i * trans_offset],
transforms, transforms[i],
trans_offset,
i,
u_quant_coeff, u_quant_coeff,
v_quant_coeff, v_quant_coeff,
SCAN_DIAG, SCAN_DIAG,
@ -580,13 +575,13 @@ void uvg_chroma_transform_search(
if(pred_cu->type == CU_INTRA && transforms[i] != CHROMA_TS && (cu_loc->width == 4 || tree_type == UVG_CHROMA_T)) { if(pred_cu->type == CU_INTRA && transforms[i] != CHROMA_TS && (cu_loc->width == 4 || tree_type == UVG_CHROMA_T)) {
bool constraints[2] = { false, false }; bool constraints[2] = { false, false };
uvg_derive_lfnst_constraints(pred_cu, constraints, u_quant_coeff, width, height, NULL, COLOR_U); uvg_derive_lfnst_constraints(pred_cu, constraints, u_quant_coeff, width, height, NULL, COLOR_U);
if(!IS_JCCR_MODE(transforms[i])) { if(!is_jccr) {
uvg_derive_lfnst_constraints(pred_cu, constraints, v_quant_coeff, width, height, NULL, COLOR_V); uvg_derive_lfnst_constraints(pred_cu, constraints, v_quant_coeff, width, height, NULL, COLOR_V);
} }
if (!constraints[1] && (u_has_coeffs || v_has_coeffs) && pred_cu->cr_lfnst_idx != 0) continue; if (!constraints[1] && (u_has_coeffs || v_has_coeffs) && pred_cu->cr_lfnst_idx != 0) continue;
} }
if (IS_JCCR_MODE(transforms[i]) && !u_has_coeffs) continue; if (is_jccr && !u_has_coeffs) continue;
if (u_has_coeffs) { if (u_has_coeffs) {
uvg_dequant(state, u_quant_coeff, &u_coeff[i * trans_offset], width, width, transforms[i] != JCCR_1 ? COLOR_U : COLOR_V, uvg_dequant(state, u_quant_coeff, &u_coeff[i * trans_offset], width, width, transforms[i] != JCCR_1 ? COLOR_U : COLOR_V,
@ -619,7 +614,7 @@ void uvg_chroma_transform_search(
} }
if (v_has_coeffs && !(IS_JCCR_MODE(transforms[i]))) { if (v_has_coeffs && !is_jccr) {
uvg_dequant(state, v_quant_coeff, &v_coeff[i * trans_offset], width, width, COLOR_V, uvg_dequant(state, v_quant_coeff, &v_coeff[i * trans_offset], width, width, COLOR_V,
pred_cu->type, transforms[i] == CHROMA_TS); pred_cu->type, transforms[i] == CHROMA_TS);
@ -638,7 +633,7 @@ void uvg_chroma_transform_search(
v_recon[trans_offset * i + j] = CLIP_TO_PIXEL(v_pred[j] + v_recon_resi[j]); v_recon[trans_offset * i + j] = CLIP_TO_PIXEL(v_pred[j] + v_recon_resi[j]);
} }
} }
else if (u_has_coeffs && IS_JCCR_MODE(transforms[i])) { else if (u_has_coeffs && is_jccr) {
if (transforms[i] == JCCR_1) { if (transforms[i] == JCCR_1) {
for (int j = 0; j < width * height; j++) { for (int j = 0; j < width * height; j++) {
v_recon[trans_offset * i + j] = CLIP_TO_PIXEL(v_pred[j] + u_recon_resi[j]); v_recon[trans_offset * i + j] = CLIP_TO_PIXEL(v_pred[j] + u_recon_resi[j]);
@ -706,7 +701,7 @@ void uvg_chroma_transform_search(
COEFF_ORDER_LINEAR); COEFF_ORDER_LINEAR);
u_bits += coeff_cost; u_bits += coeff_cost;
} }
if (cbf_v && !IS_JCCR_MODE(transforms[i])) { if (cbf_v && !is_jccr) {
if (can_use_tr_skip) { if (can_use_tr_skip) {
CABAC_FBITS_UPDATE(&state->search_cabac, &state->search_cabac.ctx.transform_skip_model_chroma, CABAC_FBITS_UPDATE(&state->search_cabac, &state->search_cabac.ctx.transform_skip_model_chroma,
transforms[i] == CHROMA_TS, v_bits, "tr_skip_v" transforms[i] == CHROMA_TS, v_bits, "tr_skip_v"
@ -743,7 +738,7 @@ void uvg_chroma_transform_search(
pred_cu->lfnst_last_scan_pos = false; pred_cu->lfnst_last_scan_pos = false;
pred_cu->violates_lfnst_constrained_chroma = false; pred_cu->violates_lfnst_constrained_chroma = false;
} }
if (!IS_JCCR_MODE(transforms[i])) { if (!is_jccr) {
double u_cost = UVG_CHROMA_MULT * ssd_u + u_bits * state->frame->lambda; double u_cost = UVG_CHROMA_MULT * ssd_u + u_bits * state->frame->lambda;
double v_cost = UVG_CHROMA_MULT * ssd_v + v_bits * state->frame->lambda; double v_cost = UVG_CHROMA_MULT * ssd_v + v_bits * state->frame->lambda;
if (u_cost < chorma_ts_out->best_u_cost) { if (u_cost < chorma_ts_out->best_u_cost) {