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

[mtt] remove depth from cbf

Browse source
This commit is contained in:
Joose Sainio 2022-09-14 11:47:26 +03:00 committed by Marko Viitanen
parent e3dbeda7f7
commit 9a29d9ded3
10 changed files with 98 additions and 99 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

37
src/cu.h
View file

@ -132,6 +132,8 @@ typedef struct
uint16_t cbf; uint16_t cbf;
uint8_t root_cbf;
uint32_t split_tree : 3 * 9; uint32_t split_tree : 3 * 9;
/** /**
@ -536,34 +538,31 @@ static INLINE unsigned xy_to_zorder(unsigned width, unsigned x, unsigned y)
} while(0) } while(0)
#define NUM_CBF_DEPTHS 5
static const uint16_t cbf_masks[NUM_CBF_DEPTHS] = { 0x1f, 0x0f, 0x07, 0x03, 0x1 };
/** /**
* Check if CBF in a given level >= depth is true. * Check if CBF in a given level >= depth is true.
*/ */
static INLINE int cbf_is_set(uint16_t cbf, int depth, color_t plane) static INLINE int cbf_is_set(uint16_t cbf, color_t plane)
{ {
return (cbf & (cbf_masks[depth] << (NUM_CBF_DEPTHS * plane))) != 0; return (cbf & (1 << (plane))) != 0;
} }
/** /**
* Check if CBF in a given level >= depth is true. * Check if CBF in a given level >= depth is true.
*/ */
static INLINE int cbf_is_set_any(uint16_t cbf, int depth) static INLINE int cbf_is_set_any(uint16_t cbf)
{ {
return cbf_is_set(cbf, depth, COLOR_Y) || return cbf_is_set(cbf, COLOR_Y) ||
cbf_is_set(cbf, depth, COLOR_U) || cbf_is_set(cbf, COLOR_U) ||
cbf_is_set(cbf, depth, COLOR_V); cbf_is_set(cbf, COLOR_V);
} }
/** /**
* Set CBF in a level to true. * Set CBF in a level to true.
*/ */
static INLINE void cbf_set(uint16_t *cbf, int depth, color_t plane) static INLINE void cbf_set(uint16_t *cbf, color_t plane)
{ {
// Return value of the bit corresponding to the level. // Return value of the bit corresponding to the level.
*cbf |= (0x10 >> depth) << (NUM_CBF_DEPTHS * plane); *cbf |= (1) << (plane);
} }
/** /**
@ -572,20 +571,20 @@ static INLINE void cbf_set(uint16_t *cbf, int depth, color_t plane)
*/ */
static INLINE void cbf_set_conditionally(uint16_t *cbf, uint16_t child_cbfs[3], int depth, color_t plane) static INLINE void cbf_set_conditionally(uint16_t *cbf, uint16_t child_cbfs[3], int depth, color_t plane)
{ {
bool child_cbf_set = cbf_is_set(child_cbfs[0], depth + 1, plane) || bool child_cbf_set = cbf_is_set(child_cbfs[0], plane) ||
cbf_is_set(child_cbfs[1], depth + 1, plane) || cbf_is_set(child_cbfs[1], plane) ||
cbf_is_set(child_cbfs[2], depth + 1, plane); cbf_is_set(child_cbfs[2], plane);
if (child_cbf_set) { if (child_cbf_set) {
cbf_set(cbf, depth, plane); cbf_set(cbf, plane);
} }
} }
/** /**
* Set CBF in a levels <= depth to false. * Set CBF in a levels <= depth to false.
*/ */
static INLINE void cbf_clear(uint16_t *cbf, int depth, color_t plane) static INLINE void cbf_clear(uint16_t *cbf, color_t plane)
{ {
*cbf &= ~(cbf_masks[depth] << (NUM_CBF_DEPTHS * plane)); *cbf &= ~(1 << (plane));
} }
/** /**
@ -593,8 +592,8 @@ static INLINE void cbf_clear(uint16_t *cbf, int depth, color_t plane)
*/ */
static INLINE void cbf_copy(uint16_t *cbf, uint16_t src, color_t plane) static INLINE void cbf_copy(uint16_t *cbf, uint16_t src, color_t plane)
{ {
cbf_clear(cbf, 0, plane); cbf_clear(cbf, plane);
*cbf |= src & (cbf_masks[0] << (NUM_CBF_DEPTHS * plane)); *cbf |= src & (1 << plane);
} }
#define GET_SPLITDATA(CU,curDepth) (((CU)->split_tree >> (curDepth)) & 7) #define GET_SPLITDATA(CU,curDepth) (((CU)->split_tree >> (curDepth)) & 7)

22
src/encode_coding_tree.c
View file

@ -466,7 +466,7 @@ static void encode_chroma_tu(
uvg_get_sub_coeff(coeff_u, coeff->u, x_local, y_local, cu_loc->chroma_width, cu_loc->chroma_height, LCU_WIDTH_C); uvg_get_sub_coeff(coeff_u, coeff->u, x_local, y_local, cu_loc->chroma_width, cu_loc->chroma_height, LCU_WIDTH_C);
uvg_get_sub_coeff(coeff_v, coeff->v, x_local, y_local, cu_loc->chroma_width, cu_loc->chroma_height, LCU_WIDTH_C); uvg_get_sub_coeff(coeff_v, coeff->v, x_local, y_local, cu_loc->chroma_width, cu_loc->chroma_height, LCU_WIDTH_C);
if (cbf_is_set(cur_pu->cbf, depth, COLOR_U)) { if (cbf_is_set(cur_pu->cbf, COLOR_U)) {
// TODO: height for this check and the others below // TODO: height for this check and the others below
if(state->encoder_control->cfg.trskip_enable && width_c <= (1 << state->encoder_control->cfg.trskip_max_size)){ if(state->encoder_control->cfg.trskip_enable && width_c <= (1 << state->encoder_control->cfg.trskip_max_size)){
cabac->cur_ctx = &cabac->ctx.transform_skip_model_chroma; cabac->cur_ctx = &cabac->ctx.transform_skip_model_chroma;
@ -477,7 +477,7 @@ static void encode_chroma_tu(
uvg_encode_coeff_nxn(state, &state->cabac, coeff_u, cu_loc, COLOR_U, *scan_idx, cur_pu, NULL); uvg_encode_coeff_nxn(state, &state->cabac, coeff_u, cu_loc, COLOR_U, *scan_idx, cur_pu, NULL);
} }
if (cbf_is_set(cur_pu->cbf, depth, COLOR_V)) { if (cbf_is_set(cur_pu->cbf, COLOR_V)) {
if (state->encoder_control->cfg.trskip_enable && width_c <= (1 << state->encoder_control->cfg.trskip_max_size)) { if (state->encoder_control->cfg.trskip_enable && width_c <= (1 << state->encoder_control->cfg.trskip_max_size)) {
cabac->cur_ctx = &cabac->ctx.transform_skip_model_chroma; cabac->cur_ctx = &cabac->ctx.transform_skip_model_chroma;
CABAC_BIN(cabac, (cur_pu->tr_skip >> COLOR_V) & 1, "transform_skip_flag"); CABAC_BIN(cabac, (cur_pu->tr_skip >> COLOR_V) & 1, "transform_skip_flag");
@ -526,7 +526,7 @@ static void encode_transform_unit(
int8_t scan_idx = uvg_get_scan_order(cur_pu->type, cur_pu->intra.mode, depth); int8_t scan_idx = uvg_get_scan_order(cur_pu->type, cur_pu->intra.mode, depth);
int cbf_y = cbf_is_set(cur_pu->cbf, depth, COLOR_Y); int cbf_y = cbf_is_set(cur_pu->cbf, COLOR_Y);
if (cbf_y && !only_chroma) { if (cbf_y && !only_chroma) {
int x_local = x % LCU_WIDTH; int x_local = x % LCU_WIDTH;
@ -573,8 +573,8 @@ static void encode_transform_unit(
} }
} }
bool chroma_cbf_set = cbf_is_set(cur_pu->cbf, depth, COLOR_U) || bool chroma_cbf_set = cbf_is_set(cur_pu->cbf, COLOR_U) ||
cbf_is_set(cur_pu->cbf, depth, COLOR_V); cbf_is_set(cur_pu->cbf, COLOR_V);
if ((chroma_cbf_set || joint_chroma) && last_split) { if ((chroma_cbf_set || joint_chroma) && last_split) {
//Need to drop const to get lfnst constraints //Need to drop const to get lfnst constraints
// Use original dimensions instead of ISP split dimensions // Use original dimensions instead of ISP split dimensions
@ -644,9 +644,9 @@ static void encode_transform_coeff(
int8_t split = (cu_loc->width > TR_MAX_WIDTH || cu_loc->height > TR_MAX_WIDTH); int8_t split = (cu_loc->width > TR_MAX_WIDTH || cu_loc->height > TR_MAX_WIDTH);
const int cb_flag_y = tree_type != UVG_CHROMA_T ? cbf_is_set(cur_pu->cbf, depth, COLOR_Y) : 0; const int cb_flag_y = tree_type != UVG_CHROMA_T ? cbf_is_set(cur_pu->cbf, COLOR_Y) : 0;
const int cb_flag_u = tree_type != UVG_LUMA_T ?( cur_pu->joint_cb_cr ? (cur_pu->joint_cb_cr >> 1) & 1 : cbf_is_set(cur_cu->cbf, depth, COLOR_U)) : 0; const int cb_flag_u = tree_type != UVG_LUMA_T ?( cur_pu->joint_cb_cr ? (cur_pu->joint_cb_cr >> 1) & 1 : cbf_is_set(cur_cu->cbf, COLOR_U)) : 0;
const int cb_flag_v = tree_type != UVG_LUMA_T ? (cur_pu->joint_cb_cr ? cur_pu->joint_cb_cr & 1 : cbf_is_set(cur_cu->cbf, depth, COLOR_V)) : 0; const int cb_flag_v = tree_type != UVG_LUMA_T ? (cur_pu->joint_cb_cr ? cur_pu->joint_cb_cr & 1 : cbf_is_set(cur_cu->cbf, COLOR_V)) : 0;
// The split_transform_flag is not signaled when: // The split_transform_flag is not signaled when:
// - transform size is greater than 32 (depth == 0) // - transform size is greater than 32 (depth == 0)
@ -1610,15 +1610,15 @@ void uvg_encode_coding_tree(
} }
{ {
int cbf = cbf_is_set_any(cur_cu->cbf, depth);
// Only need to signal coded block flag if not skipped or merged // Only need to signal coded block flag if not skipped or merged
// skip = no coded residual, merge = coded residual // skip = no coded residual, merge = coded residual
const bool has_coeffs = cur_pu->root_cbf || cur_pu->cbf;
if (!cur_cu->merged) { if (!cur_cu->merged) {
cabac->cur_ctx = &(cabac->ctx.cu_qt_root_cbf_model); cabac->cur_ctx = &(cabac->ctx.cu_qt_root_cbf_model);
CABAC_BIN(cabac, cbf, "rqt_root_cbf"); CABAC_BIN(cabac, has_coeffs, "rqt_root_cbf");
} }
// Code (possible) coeffs to bitstream // Code (possible) coeffs to bitstream
if (cbf) { if (has_coeffs) {
int luma_cbf_ctx = 0; int luma_cbf_ctx = 0;
encode_transform_coeff(state, cu_loc, depth, 0, 0, 0, 0, coeff, tree_type, true, false, &luma_cbf_ctx, cu_loc); encode_transform_coeff(state, cu_loc, depth, 0, 0, 0, 0, coeff, tree_type, true, false, &luma_cbf_ctx, cu_loc);
} }

8
src/encoderstate.c
View file

@ -667,12 +667,12 @@ static void set_cu_qps(encoder_state_t *state, const cu_loc_t* const cu_loc, int
for (int y_scu = cu_loc->y; !cbf_found && y_scu < y_limit; y_scu += tu_width) { for (int y_scu = cu_loc->y; !cbf_found && y_scu < y_limit; y_scu += tu_width) {
for (int x_scu = cu_loc->x; !cbf_found && x_scu < x_limit; x_scu += tu_width) { for (int x_scu = cu_loc->x; !cbf_found && x_scu < x_limit; x_scu += tu_width) {
cu_info_t *tu = uvg_cu_array_at(state->tile->frame->cu_array, x_scu, y_scu); cu_info_t *tu = uvg_cu_array_at(state->tile->frame->cu_array, x_scu, y_scu);
if (cbf_is_set_any(tu->cbf, cu->depth)) { if (cbf_is_set_any(tu->cbf)) {
cbf_found = true; cbf_found = true;
} }
} }
} }
} else if (cbf_is_set_any(cu->cbf, cu->depth)) { } else if (cbf_is_set_any(cu->cbf)) {
cbf_found = true; cbf_found = true;
} }
@ -2045,9 +2045,9 @@ static void _encode_one_frame_add_bitstream_deps(const encoder_state_t * const s
void uvg_encode_one_frame(encoder_state_t * const state, uvg_picture* frame) void uvg_encode_one_frame(encoder_state_t * const state, uvg_picture* frame)
{ {
#if UVG_DEBUG_PRINT_CABAC == 1 #if UVG_DEBUG_PRINT_CABAC == 1
uvg_cabac_bins_count = 0; // uvg_cabac_bins_count = 0;
if (state->frame->num == 0) uvg_cabac_bins_verbose = true; if (state->frame->num == 0) uvg_cabac_bins_verbose = true;
else uvg_cabac_bins_verbose = false; // else uvg_cabac_bins_verbose = false;
#endif #endif

12
src/filter.c
View file

@ -757,8 +757,8 @@ static void filter_deblock_edge_luma(encoder_state_t * const state,
cu_q = uvg_cu_array_at(frame->cu_array, x_coord, y); cu_q = uvg_cu_array_at(frame->cu_array, x_coord, y);
} }
bool nonzero_coeffs = cbf_is_set(cu_q->cbf, cu_q->tr_depth, COLOR_Y) bool nonzero_coeffs = cbf_is_set(cu_q->cbf, COLOR_Y)
|| cbf_is_set(cu_p->cbf, cu_p->tr_depth, COLOR_Y); || cbf_is_set(cu_p->cbf, COLOR_Y);
// Filter strength // Filter strength
strength = 0; strength = 0;
@ -1122,10 +1122,10 @@ static void filter_deblock_edge_chroma(encoder_state_t * const state,
c_strength[1] = 2; c_strength[1] = 2;
} }
else if (tu_boundary){ //TODO: Add ciip/IBC related stuff else if (tu_boundary){ //TODO: Add ciip/IBC related stuff
bool nonzero_coeffs_U = cbf_is_set(cu_q->cbf, cu_q->tr_depth, COLOR_U) bool nonzero_coeffs_U = cbf_is_set(cu_q->cbf, COLOR_U)
|| cbf_is_set(cu_p->cbf, cu_p->tr_depth, COLOR_U); || cbf_is_set(cu_p->cbf, COLOR_U);
bool nonzero_coeffs_V = cbf_is_set(cu_q->cbf, cu_q->tr_depth, COLOR_V) bool nonzero_coeffs_V = cbf_is_set(cu_q->cbf, COLOR_V)
|| cbf_is_set(cu_p->cbf, cu_p->tr_depth, COLOR_V); || cbf_is_set(cu_p->cbf, COLOR_V);
c_strength[0] = nonzero_coeffs_U ? 1 : 0; c_strength[0] = nonzero_coeffs_U ? 1 : 0;
c_strength[1] = nonzero_coeffs_V ? 1 : 0; c_strength[1] = nonzero_coeffs_V ? 1 : 0;
} }

22
src/intra.c
View file

@ -1792,11 +1792,11 @@ void uvg_intra_recon_cu(
// Reset CBFs because CBFs might have been set // Reset CBFs because CBFs might have been set
// for depth earlier // for depth earlier
if (recon_luma) { if (recon_luma) {
cbf_clear(&cur_cu->cbf, depth, COLOR_Y); cbf_clear(&cur_cu->cbf, COLOR_Y);
} }
if (recon_chroma) { if (recon_chroma) {
cbf_clear(&cur_cu->cbf, depth, COLOR_U); cbf_clear(&cur_cu->cbf, COLOR_U);
cbf_clear(&cur_cu->cbf, depth, COLOR_V); cbf_clear(&cur_cu->cbf, COLOR_V);
} }
if (width > TR_MAX_WIDTH || height > TR_MAX_WIDTH) { if (width > TR_MAX_WIDTH || height > TR_MAX_WIDTH) {
@ -1820,13 +1820,13 @@ void uvg_intra_recon_cu(
LCU_GET_CU_AT_PX(lcu, (lcu_px.x + half_width) >> (tree_type == UVG_CHROMA_T), (lcu_px.y + half_height) >> (tree_type == UVG_CHROMA_T))->cbf, LCU_GET_CU_AT_PX(lcu, (lcu_px.x + half_width) >> (tree_type == UVG_CHROMA_T), (lcu_px.y + half_height) >> (tree_type == UVG_CHROMA_T))->cbf,
}; };
if (recon_luma && depth <= MAX_DEPTH) { //if (recon_luma && depth <= MAX_DEPTH) {
cbf_set_conditionally(&cur_cu->cbf, child_cbfs, depth, COLOR_Y); // cbf_set_conditionally(&cur_cu->cbf, child_cbfs, depth, COLOR_Y);
} //}
if (recon_chroma && depth <= MAX_DEPTH) { //if (recon_chroma && depth <= MAX_DEPTH) {
cbf_set_conditionally(&cur_cu->cbf, child_cbfs, depth, COLOR_U); // cbf_set_conditionally(&cur_cu->cbf, child_cbfs, depth, COLOR_U);
cbf_set_conditionally(&cur_cu->cbf, child_cbfs, depth, COLOR_V); // cbf_set_conditionally(&cur_cu->cbf, child_cbfs, depth, COLOR_V);
} //}
return; return;
} }
if (search_data->pred_cu.intra.isp_mode != ISP_MODE_NO_ISP && recon_luma ) { if (search_data->pred_cu.intra.isp_mode != ISP_MODE_NO_ISP && recon_luma ) {
@ -1848,7 +1848,7 @@ void uvg_intra_recon_cu(
uvg_quantize_lcu_residual(state, true, false, false, uvg_quantize_lcu_residual(state, true, false, false,
&tu_loc, depth, cur_cu, lcu, &tu_loc, depth, cur_cu, lcu,
false, tree_type); false, tree_type);
search_data->best_isp_cbfs |= cbf_is_set(cur_cu->cbf, depth, COLOR_Y) << i; search_data->best_isp_cbfs |= cbf_is_set(cur_cu->cbf, COLOR_Y) << i;
} }
} }
const bool has_luma = recon_luma && search_data->pred_cu.intra.isp_mode == ISP_MODE_NO_ISP; const bool has_luma = recon_luma && search_data->pred_cu.intra.isp_mode == ISP_MODE_NO_ISP;

2
src/rdo.c
View file

@ -1732,7 +1732,7 @@ void uvg_rdoq(
assert(0); assert(0);
} }
// This cbf should work even with non-square blocks // This cbf should work even with non-square blocks
ctx_cbf = ( color != COLOR_V ? 0 : cbf_is_set(cbf, 5 - uvg_math_floor_log2(width), COLOR_U)); ctx_cbf = ( color != COLOR_V ? 0 : cbf_is_set(cbf, COLOR_U));
best_cost = block_uncoded_cost + lambda * CTX_ENTROPY_BITS(&base_cbf_model[ctx_cbf],0); best_cost = block_uncoded_cost + lambda * CTX_ENTROPY_BITS(&base_cbf_model[ctx_cbf],0);
base_cost += lambda * CTX_ENTROPY_BITS(&base_cbf_model[ctx_cbf],1); base_cost += lambda * CTX_ENTROPY_BITS(&base_cbf_model[ctx_cbf],1);
} }

42
src/search.c
View file

@ -361,11 +361,11 @@ double uvg_cu_rd_cost_luma(
if (pred_cu->type == CU_INTER || pred_cu->intra.isp_mode == ISP_MODE_NO_ISP) { if (pred_cu->type == CU_INTER || pred_cu->intra.isp_mode == ISP_MODE_NO_ISP) {
const int depth = 6 - uvg_g_convert_to_log2[cu_loc->width]; const int depth = 6 - uvg_g_convert_to_log2[cu_loc->width];
const int is_tr_split = tr_cu->tr_depth - tr_cu->depth; const int is_tr_split = tr_cu->tr_depth - tr_cu->depth;
int is_set = cbf_is_set(pred_cu->cbf, depth, COLOR_Y); int is_set = cbf_is_set(pred_cu->cbf, COLOR_Y);
if (pred_cu->type == CU_INTRA || if (pred_cu->type == CU_INTRA ||
is_tr_split || is_tr_split ||
cbf_is_set(tr_cu->cbf, depth, COLOR_U) || cbf_is_set(tr_cu->cbf, COLOR_U) ||
cbf_is_set(tr_cu->cbf, depth, COLOR_V)) cbf_is_set(tr_cu->cbf, COLOR_V))
{ {
cabac_ctx_t* ctx = &(cabac->ctx.qt_cbf_model_luma[0]); cabac_ctx_t* ctx = &(cabac->ctx.qt_cbf_model_luma[0]);
@ -455,8 +455,8 @@ double uvg_cu_rd_cost_chroma(
} }
const int depth = 6 - uvg_g_convert_to_log2[cu_loc->width]; const int depth = 6 - uvg_g_convert_to_log2[cu_loc->width];
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 u_is_set = pred_cu->joint_cb_cr ? (pred_cu->joint_cb_cr & 2) >> 1 : cbf_is_set(pred_cu->cbf, 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); int v_is_set = pred_cu->joint_cb_cr ? (pred_cu->joint_cb_cr & 1) : cbf_is_set(pred_cu->cbf, COLOR_V);
if (cu_loc->width > TR_MAX_WIDTH || cu_loc->height > TR_MAX_WIDTH) { if (cu_loc->width > TR_MAX_WIDTH || cu_loc->height > TR_MAX_WIDTH) {
double sum = 0; double sum = 0;
@ -482,11 +482,11 @@ double uvg_cu_rd_cost_chroma(
cabac_data_t* cabac = (cabac_data_t*)&state->search_cabac; cabac_data_t* cabac = (cabac_data_t*)&state->search_cabac;
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, 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");
} }
ctx = &(cabac->ctx.qt_cbf_model_cr[u_is_set]); ctx = &(cabac->ctx.qt_cbf_model_cr[u_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, 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");
} }
} }
@ -558,13 +558,13 @@ static double cu_rd_cost_tr_split_accurate(
const int depth = 6 - uvg_g_convert_to_log2[cu_loc->width]; const int depth = 6 - uvg_g_convert_to_log2[cu_loc->width];
const int cb_flag_u = tr_cu->joint_cb_cr ? tr_cu->joint_cb_cr >> 1 : cbf_is_set(tr_cu->cbf, depth, COLOR_U); const int cb_flag_u = tr_cu->joint_cb_cr ? tr_cu->joint_cb_cr >> 1 : cbf_is_set(tr_cu->cbf, COLOR_U);
const int cb_flag_v = tr_cu->joint_cb_cr ? tr_cu->joint_cb_cr & 1 : cbf_is_set(tr_cu->cbf, depth, COLOR_V); const int cb_flag_v = tr_cu->joint_cb_cr ? tr_cu->joint_cb_cr & 1 : cbf_is_set(tr_cu->cbf, COLOR_V);
cabac_data_t* cabac = (cabac_data_t*)&state->search_cabac; cabac_data_t* cabac = (cabac_data_t*)&state->search_cabac;
{ {
int cbf = cbf_is_set_any(tr_cu->cbf, depth); int cbf = cbf_is_set_any(tr_cu->cbf);
// Only need to signal coded block flag if not skipped or merged // Only need to signal coded block flag if not skipped or merged
// skip = no coded residual, merge = coded residual // skip = no coded residual, merge = coded residual
if (pred_cu->type != CU_INTRA && (!pred_cu->merged)) { if (pred_cu->type != CU_INTRA && (!pred_cu->merged)) {
@ -596,7 +596,7 @@ static double cu_rd_cost_tr_split_accurate(
CABAC_FBITS_UPDATE(cabac, &(cabac->ctx.qt_cbf_model_cr[cb_flag_u]), cb_flag_v, tr_tree_bits, "cbf_cr"); CABAC_FBITS_UPDATE(cabac, &(cabac->ctx.qt_cbf_model_cr[cb_flag_u]), cb_flag_v, tr_tree_bits, "cbf_cr");
} }
const int cb_flag_y = cbf_is_set(tr_cu->cbf, depth, COLOR_Y) && tree_type != UVG_CHROMA_T; const int cb_flag_y = cbf_is_set(tr_cu->cbf, COLOR_Y) && tree_type != UVG_CHROMA_T;
const bool is_isp = !(pred_cu->type == CU_INTER || pred_cu->intra.isp_mode == ISP_MODE_NO_ISP); const bool is_isp = !(pred_cu->type == CU_INTER || pred_cu->intra.isp_mode == ISP_MODE_NO_ISP);
// Add transform_tree cbf_luma bit cost. // Add transform_tree cbf_luma bit cost.
@ -1182,8 +1182,8 @@ static double search_cu(
const int split_type = intra_search.pred_cu.intra.isp_mode; const int split_type = intra_search.pred_cu.intra.isp_mode;
const int split_num = split_type == ISP_MODE_NO_ISP ? 0 : uvg_get_isp_split_num(cu_width, cu_height, split_type, true); const int split_num = split_type == ISP_MODE_NO_ISP ? 0 : uvg_get_isp_split_num(cu_width, cu_height, split_type, true);
const int cbf_cb = cbf_is_set(cur_cu->cbf, split_tree.current_depth, COLOR_U); const int cbf_cb = cbf_is_set(cur_cu->cbf, COLOR_U);
const int cbf_cr = cbf_is_set(cur_cu->cbf, split_tree.current_depth, COLOR_V); const int cbf_cr = cbf_is_set(cur_cu->cbf, COLOR_V);
const int jccr = cur_cu->joint_cb_cr; const int jccr = cur_cu->joint_cb_cr;
for (int i = 0; i < split_num; ++i) { for (int i = 0; i < split_num; ++i) {
cu_loc_t isp_loc; cu_loc_t isp_loc;
@ -1195,14 +1195,14 @@ static double search_cu(
uvg_get_isp_cu_arr_coords(&tmp_x, &tmp_y); uvg_get_isp_cu_arr_coords(&tmp_x, &tmp_y);
cu_info_t* split_cu = LCU_GET_CU_AT_PX(lcu, tmp_x % LCU_WIDTH, tmp_y % LCU_WIDTH); cu_info_t* split_cu = LCU_GET_CU_AT_PX(lcu, tmp_x % LCU_WIDTH, tmp_y % LCU_WIDTH);
bool cur_cbf = (intra_search.best_isp_cbfs >> i) & 1; bool cur_cbf = (intra_search.best_isp_cbfs >> i) & 1;
cbf_clear(&split_cu->cbf, split_tree.current_depth, COLOR_Y); cbf_clear(&split_cu->cbf, COLOR_Y);
cbf_clear(&split_cu->cbf, split_tree.current_depth, COLOR_U); cbf_clear(&split_cu->cbf, COLOR_U);
cbf_clear(&split_cu->cbf, split_tree.current_depth, COLOR_V); cbf_clear(&split_cu->cbf, COLOR_V);
if (cur_cbf) { if (cur_cbf) {
cbf_set(&split_cu->cbf, split_tree.current_depth, COLOR_Y); cbf_set(&split_cu->cbf, COLOR_Y);
} }
if(cbf_cb) cbf_set(&split_cu->cbf, split_tree.current_depth, COLOR_U); if(cbf_cb) cbf_set(&split_cu->cbf, COLOR_U);
if(cbf_cr) cbf_set(&split_cu->cbf, split_tree.current_depth, COLOR_V); if(cbf_cr) cbf_set(&split_cu->cbf, COLOR_V);
split_cu->joint_cb_cr = jccr; split_cu->joint_cb_cr = jccr;
} }
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);
@ -1241,7 +1241,7 @@ static double search_cu(
false, false,
tree_type); tree_type);
int cbf = cbf_is_set_any(cur_cu->cbf, split_tree.current_depth); int cbf = cbf_is_set_any(cur_cu->cbf);
if (cur_cu->merged && !cbf) { if (cur_cu->merged && !cbf) {
cur_cu->merged = 0; cur_cu->merged = 0;
@ -1327,7 +1327,7 @@ static double search_cu(
int half_cu = cu_width >> (tree_type != UVG_CHROMA_T); int half_cu = cu_width >> (tree_type != UVG_CHROMA_T);
double split_cost = 0.0; double split_cost = 0.0;
int cbf = cbf_is_set_any(cur_cu->cbf, split_tree.current_depth); int cbf = cbf_is_set_any(cur_cu->cbf);
cabac_data_t post_seach_cabac; cabac_data_t post_seach_cabac;
memcpy(&post_seach_cabac, &state->search_cabac, sizeof(post_seach_cabac)); memcpy(&post_seach_cabac, &state->search_cabac, sizeof(post_seach_cabac));
memcpy(&state->search_cabac, &pre_search_cabac, sizeof(post_seach_cabac)); memcpy(&state->search_cabac, &pre_search_cabac, sizeof(post_seach_cabac));

18
src/search_inter.c
View file

@ -1816,7 +1816,7 @@ static void search_pu_inter(
uvg_quantize_lcu_residual(state, true, false, false, cu_loc, depth, cur_pu, lcu, true, UVG_BOTH_T); uvg_quantize_lcu_residual(state, true, false, false, cu_loc, depth, cur_pu, lcu, true, UVG_BOTH_T);
if (cbf_is_set(cur_pu->cbf, depth, COLOR_Y)) { if (cbf_is_set(cur_pu->cbf, COLOR_Y)) {
continue; continue;
} }
else if (has_chroma) { else if (has_chroma) {
@ -1827,7 +1827,7 @@ static void search_pu_inter(
cu_loc, depth, cur_pu, lcu, cu_loc, depth, cur_pu, lcu,
true, true,
UVG_BOTH_T); UVG_BOTH_T);
if (!cbf_is_set_any(cur_pu->cbf, depth)) { if (!cbf_is_set_any(cur_pu->cbf)) {
cur_pu->type = CU_INTER; cur_pu->type = CU_INTER;
cur_pu->merge_idx = merge_idx; cur_pu->merge_idx = merge_idx;
cur_pu->skipped = true; cur_pu->skipped = true;
@ -2228,20 +2228,20 @@ void uvg_cu_cost_inter_rd2(
v_resi, v_resi,
&chorma_ts_out, &chorma_ts_out,
UVG_BOTH_T); UVG_BOTH_T);
cbf_clear(&cur_cu->cbf, depth, COLOR_U); cbf_clear(&cur_cu->cbf, COLOR_U);
cbf_clear(&cur_cu->cbf, depth, COLOR_V); cbf_clear(&cur_cu->cbf, COLOR_V);
if (chorma_ts_out.best_u_cost + chorma_ts_out.best_v_cost < chorma_ts_out.best_combined_cost) { if (chorma_ts_out.best_u_cost + chorma_ts_out.best_v_cost < chorma_ts_out.best_combined_cost) {
cur_cu->joint_cb_cr = 0; cur_cu->joint_cb_cr = 0;
cur_cu->tr_skip |= (chorma_ts_out.best_u_index == CHROMA_TS) << COLOR_U; cur_cu->tr_skip |= (chorma_ts_out.best_u_index == CHROMA_TS) << COLOR_U;
cur_cu->tr_skip |= (chorma_ts_out.best_v_index == CHROMA_TS) << COLOR_V; cur_cu->tr_skip |= (chorma_ts_out.best_v_index == CHROMA_TS) << COLOR_V;
if(chorma_ts_out.best_u_index != NO_RESIDUAL) cbf_set(&cur_cu->cbf, depth, COLOR_U); if(chorma_ts_out.best_u_index != NO_RESIDUAL) cbf_set(&cur_cu->cbf, COLOR_U);
if(chorma_ts_out.best_v_index != NO_RESIDUAL) cbf_set(&cur_cu->cbf, depth, COLOR_V); if(chorma_ts_out.best_v_index != NO_RESIDUAL) cbf_set(&cur_cu->cbf, COLOR_V);
chroma_cost += chorma_ts_out.best_u_cost + chorma_ts_out.best_v_cost; chroma_cost += chorma_ts_out.best_u_cost + chorma_ts_out.best_v_cost;
} }
else { else {
cur_cu->joint_cb_cr = chorma_ts_out.best_combined_index; cur_cu->joint_cb_cr = chorma_ts_out.best_combined_index;
if (chorma_ts_out.best_combined_index & 2) cbf_set(&cur_cu->cbf, depth, COLOR_U); if (chorma_ts_out.best_combined_index & 2) cbf_set(&cur_cu->cbf, COLOR_U);
if (chorma_ts_out.best_combined_index & 1) cbf_set(&cur_cu->cbf, depth, COLOR_V); if (chorma_ts_out.best_combined_index & 1) cbf_set(&cur_cu->cbf, COLOR_V);
chroma_cost += chorma_ts_out.best_combined_cost; chroma_cost += chorma_ts_out.best_combined_cost;
} }
} }
@ -2257,7 +2257,7 @@ void uvg_cu_cost_inter_rd2(
UVG_BOTH_T); UVG_BOTH_T);
} }
int cbf = cbf_is_set_any(cur_cu->cbf, depth); int cbf = cbf_is_set_any(cur_cu->cbf);
if(cbf) { if(cbf) {
*inter_cost = uvg_cu_rd_cost_luma(state, cu_loc, cur_cu, lcu, 0); *inter_cost = uvg_cu_rd_cost_luma(state, cu_loc, cur_cu, lcu, 0);

12
src/search_intra.c
View file

@ -304,10 +304,10 @@ static double search_intra_trdepth(
nosplit_cost = 0.0; nosplit_cost = 0.0;
cbf_clear(&pred_cu->cbf, depth, COLOR_Y); cbf_clear(&pred_cu->cbf, COLOR_Y);
if (reconstruct_chroma) { if (reconstruct_chroma) {
cbf_clear(&pred_cu->cbf, depth, COLOR_U); cbf_clear(&pred_cu->cbf, COLOR_U);
cbf_clear(&pred_cu->cbf, depth, COLOR_V); cbf_clear(&pred_cu->cbf, COLOR_V);
} }
const int8_t chroma_mode = reconstruct_chroma ? (!pred_cu->intra.mip_flag ? pred_cu->intra.mode : 0) : -1; const int8_t chroma_mode = reconstruct_chroma ? (!pred_cu->intra.mip_flag ? pred_cu->intra.mode : 0) : -1;
@ -398,7 +398,7 @@ static double search_intra_trdepth(
); );
if (pred_cu->intra.isp_mode != ISP_MODE_NO_ISP && search_data->best_isp_cbfs == 0) continue; if (pred_cu->intra.isp_mode != ISP_MODE_NO_ISP && search_data->best_isp_cbfs == 0) continue;
if (trafo != 0 && !cbf_is_set(pred_cu->cbf, depth, COLOR_Y)) continue; if (trafo != 0 && !cbf_is_set(pred_cu->cbf, COLOR_Y)) continue;
derive_mts_constraints(pred_cu, lcu, width, height, lcu_px); derive_mts_constraints(pred_cu, lcu, width, height, lcu_px);
if (pred_cu->tr_idx > 1) { if (pred_cu->tr_idx > 1) {
@ -424,7 +424,7 @@ static double search_intra_trdepth(
COLOR_Y); COLOR_Y);
} }
if (!constraints[1] && cbf_is_set(pred_cu->cbf, depth, COLOR_Y)) { if (!constraints[1] && cbf_is_set(pred_cu->cbf, COLOR_Y)) {
//end_idx = 0; //end_idx = 0;
if (pred_cu->lfnst_idx > 0) { if (pred_cu->lfnst_idx > 0) {
continue; continue;
@ -1347,7 +1347,7 @@ static int8_t search_intra_rdo(
best_isp_mode = isp_mode; best_isp_mode = isp_mode;
best_bits = search_data[mode].bits; best_bits = search_data[mode].bits;
} }
if (state->encoder_control->cfg.intra_rdo_et && !cbf_is_set_any(search_data[mode].pred_cu.cbf, depth)) { if (state->encoder_control->cfg.intra_rdo_et && !cbf_is_set_any(search_data[mode].pred_cu.cbf)) {
modes_to_check = mode + 1; modes_to_check = mode + 1;
break; break;
} }

22
src/transform.c
View file

@ -446,7 +446,7 @@ static void quantize_chroma(
if (transforms[i] == DCT7_CHROMA) { if (transforms[i] == DCT7_CHROMA) {
uint16_t temp_cbf = 0; uint16_t temp_cbf = 0;
if (*u_has_coeffs)cbf_set(&temp_cbf, depth, 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[i * trans_offset], v_quant_coeff, width, height, COLOR_V,
scan_order, CU_INTRA, temp_cbf, lfnst_idx); scan_order, CU_INTRA, temp_cbf, lfnst_idx);
@ -1289,7 +1289,7 @@ static void quantize_tr_residual(
for (int j = 0; j < tr_height; ++j) { for (int j = 0; j < tr_height; ++j) {
memcpy(&dst_coeff[j * lcu_width], &coeff[j * tr_width], tr_width * sizeof(coeff_t)); memcpy(&dst_coeff[j * lcu_width], &coeff[j * tr_width], tr_width * sizeof(coeff_t));
} }
cbf_set(&cur_pu->cbf, depth, color); cbf_set(&cur_pu->cbf, color);
} }
else { else {
for (int j = 0; j < tr_height; ++j) { for (int j = 0; j < tr_height; ++j) {
@ -1318,13 +1318,12 @@ static void quantize_tr_residual(
} }
// ISP_TODO: does this cu point to correct cbf when ISP is used for small blocks? cbf_clear(&cur_pu->cbf, color);
cbf_clear(&cur_pu->cbf, depth, color);
if (has_coeffs) { if (has_coeffs) {
for (int j = 0; j < tr_height; ++j) { for (int j = 0; j < tr_height; ++j) {
memcpy(&dst_coeff[j * lcu_width], &coeff[j * tr_width], tr_width * sizeof(coeff_t)); memcpy(&dst_coeff[j * lcu_width], &coeff[j * tr_width], tr_width * sizeof(coeff_t));
} }
cbf_set(&cur_pu->cbf, depth, color); cbf_set(&cur_pu->cbf, color);
} }
else { else {
for (int j = 0; j < tr_height; ++j) { for (int j = 0; j < tr_height; ++j) {
@ -1387,11 +1386,11 @@ void uvg_quantize_lcu_residual(
// for depth earlier // for depth earlier
// ISP_TODO: does this cur_cu point to the correct place when ISP is used for small blocks? // ISP_TODO: does this cur_cu point to the correct place when ISP is used for small blocks?
if (luma) { if (luma) {
cbf_clear(&cur_pu->cbf, depth, COLOR_Y); cbf_clear(&cur_pu->cbf, COLOR_Y);
} }
if (chroma || jccr) { if (chroma || jccr) {
cbf_clear(&cur_pu->cbf, depth, COLOR_U); cbf_clear(&cur_pu->cbf, COLOR_U);
cbf_clear(&cur_pu->cbf, depth, COLOR_V); cbf_clear(&cur_pu->cbf, COLOR_V);
} }
if (depth == 0 || cur_pu->tr_depth > depth) { if (depth == 0 || cur_pu->tr_depth > depth) {
@ -1423,9 +1422,10 @@ void uvg_quantize_lcu_residual(
}; };
if (depth <= MAX_DEPTH) { if (depth <= MAX_DEPTH) {
cbf_set_conditionally(&cur_pu->cbf, child_cbfs, depth, COLOR_Y); cur_pu->root_cbf = cbf_is_set_any(cur_pu->cbf)
cbf_set_conditionally(&cur_pu->cbf, child_cbfs, depth, COLOR_U); || cbf_is_set_any(child_cbfs[0])
cbf_set_conditionally(&cur_pu->cbf, child_cbfs, depth, COLOR_V); || cbf_is_set_any(child_cbfs[1])
|| cbf_is_set_any(child_cbfs[2]);
} }
} else { } else {