hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-30 12:44:07 +00:00
Code Issues Actions Projects Releases Wiki Activity

Moving ALF globals to alf_info struct inserted in videoframe_t.

Browse source 
g_alf_covariance and g_alf_covariance_frame moved.
This commit is contained in:
Arttu Makinen 2020-12-02 18:19:49 +02:00
parent 76cf8a16d9
commit 335ce2bdda
6 changed files with 292 additions and 198 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

324
src/alf.c
View file

@ -858,7 +858,7 @@ double get_unfiltered_distortion_cov_classes(alf_covariance* cov, const int num_
return dist; return dist;
} }
void get_frame_stats(channel_type channel, const int32_t num_ctus) void get_frame_stats(alf_info_t *alf_info, channel_type channel, const int32_t num_ctus)
{ {
bool is_luma = channel == CHANNEL_TYPE_LUMA ? true : false; bool is_luma = channel == CHANNEL_TYPE_LUMA ? true : false;
int num_classes = is_luma ? MAX_NUM_ALF_CLASSES : 1; int num_classes = is_luma ? MAX_NUM_ALF_CLASSES : 1;
@ -870,16 +870,17 @@ void get_frame_stats(channel_type channel, const int32_t num_ctus)
{ {
for (int i = 0; i < num_classes; i++) for (int i = 0; i < num_classes; i++)
{ {
reset_alf_covariance(&g_alf_covariance_frame[channel][is_luma ? i : alt_idx], MAX_ALF_NUM_CLIPPING_VALUES); is_luma ? reset_alf_covariance(&alf_info->alf_covariance_frame_luma[i], MAX_ALF_NUM_CLIPPING_VALUES) :
reset_alf_covariance(&alf_info->alf_covariance_frame_chroma[alt_idx], MAX_ALF_NUM_CLIPPING_VALUES);
} }
if (is_luma) if (is_luma)
{ {
get_frame_stat(g_alf_covariance_frame[CHANNEL_TYPE_LUMA], g_alf_covariance[COMPONENT_Y], g_ctu_enable_flag[COMPONENT_Y], NULL, num_classes, alt_idx, num_ctus); get_frame_stat(alf_info->alf_covariance_frame_luma, alf_info->alf_covariance_y, g_ctu_enable_flag[COMPONENT_Y], NULL, num_classes, alt_idx, num_ctus);
} }
else else
{ {
get_frame_stat(g_alf_covariance_frame[CHANNEL_TYPE_CHROMA], g_alf_covariance[COMPONENT_Cb], g_ctu_enable_flag[COMPONENT_Cb], g_ctu_alternative[COMPONENT_Cb], num_classes, alt_idx, num_ctus); get_frame_stat(alf_info->alf_covariance_frame_chroma, alf_info->alf_covariance_u, g_ctu_enable_flag[COMPONENT_Cb], g_ctu_alternative[COMPONENT_Cb], num_classes, alt_idx, num_ctus);
get_frame_stat(g_alf_covariance_frame[CHANNEL_TYPE_CHROMA], g_alf_covariance[COMPONENT_Cr], g_ctu_enable_flag[COMPONENT_Cr], g_ctu_alternative[COMPONENT_Cr], num_classes, alt_idx, num_ctus); get_frame_stat(alf_info->alf_covariance_frame_chroma, alf_info->alf_covariance_v, g_ctu_enable_flag[COMPONENT_Cr], g_ctu_alternative[COMPONENT_Cr], num_classes, alt_idx, num_ctus);
} }
} }
/*#else /*#else
@ -899,11 +900,7 @@ void get_frame_stats(channel_type channel, const int32_t num_ctus)
#endif*/ #endif*/
} }
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB void get_frame_stat(alf_covariance* frame_cov, alf_covariance* ctb_cov, uint8_t* ctb_enable_flags, uint8_t* ctb_alt_idx, const int num_classes, int alt_idx, const int32_t num_ctus)
void get_frame_stat(alf_covariance* frame_cov, alf_covariance** ctb_cov, uint8_t* ctb_enable_flags, uint8_t* ctb_alt_idx, const int num_classes, int alt_idx, int ctu_idx)
/*#else
void get_frame_stat(alf_covariance* frame_cov, alf_covariance** ctb_cov, uint8_t* ctb_enable_flags, const int num_classes)
#endif*/
{ {
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB
const channel_type channel = (!ctb_alt_idx ? CHANNEL_TYPE_LUMA : CHANNEL_TYPE_CHROMA); const channel_type channel = (!ctb_alt_idx ? CHANNEL_TYPE_LUMA : CHANNEL_TYPE_CHROMA);
@ -916,7 +913,7 @@ void get_frame_stat(alf_covariance* frame_cov, alf_covariance** ctb_cov, uint8_t
{ {
if (is_luma || alt_idx == ctb_alt_idx[ctu_idx]) if (is_luma || alt_idx == ctb_alt_idx[ctu_idx])
{ {
add_alf_cov(&frame_cov[is_luma ? class_idx : alt_idx], &ctb_cov[ctu_idx][class_idx]); add_alf_cov(&frame_cov[is_luma ? class_idx : alt_idx], &ctb_cov[(ctu_idx * num_classes) + class_idx]);
} }
} }
} }
@ -1112,6 +1109,14 @@ void reset_alf_covariance(alf_covariance *alf, int num_bins) {
memset(alf->ee, 0, sizeof(alf->ee)); memset(alf->ee, 0, sizeof(alf->ee));
} }
void init_alf_covariance(alf_covariance *alf, int num_coeffs) {
alf->num_coeff = num_coeffs;
alf->num_bins = MAX_ALF_NUM_CLIPPING_VALUES;
alf->pix_acc = 0;
memset(alf->y, 0, sizeof(alf->y));
memset(alf->ee, 0, sizeof(alf->ee));
}
void reset_cc_alf_aps_param(cc_alf_filter_param *cc_alf) { void reset_cc_alf_aps_param(cc_alf_filter_param *cc_alf) {
memset(cc_alf->cc_alf_filter_enabled, false, sizeof(cc_alf->cc_alf_filter_enabled)); memset(cc_alf->cc_alf_filter_enabled, false, sizeof(cc_alf->cc_alf_filter_enabled));
memset(cc_alf->cc_alf_filter_idx_enabled, false, sizeof(cc_alf->cc_alf_filter_idx_enabled)); memset(cc_alf->cc_alf_filter_idx_enabled, false, sizeof(cc_alf->cc_alf_filter_idx_enabled));
@ -1849,19 +1854,16 @@ double kvz_alf_derive_ctb_alf_enable_flags(encoder_state_t * const state,
} }
#endif*/ #endif*/
alf_covariance* alf_cov;
for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++) for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++)
{ {
for (int comp_id = comp_id_first; comp_id <= comp_id_last; comp_id++) for (int comp_id = comp_id_first; comp_id <= comp_id_last; comp_id++)
{ {
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB alf_cov = (is_luma ? state->tile->frame->alf_info->alf_covariance_y :
//#if ENABLE_QPA (comp_id == COMPONENT_Cb ? state->tile->frame->alf_info->alf_covariance_u : state->tile->frame->alf_info->alf_covariance_v));
const double ctu_lambda = chroma_weight > 0.0 ? (is_luma ? 0/*cs.picture->m_uEnerHpCtu[ctuIdx]*/ : 0/*cs.picture->m_uEnerHpCtu[ctuIdx]*/ / chroma_weight) : g_lambda[comp_id];
/*#else
const double ctu_lambda = m_lambda[comp_id];
#endif
#endif*/
double dist_unfilter_ctu = get_unfiltered_distortion_cov_classes(g_alf_covariance[comp_id][ctu_idx], num_classes); const double ctu_lambda = chroma_weight > 0.0 ? (is_luma ? 0/*cs.picture->m_uEnerHpCtu[ctuIdx]*/ : 0/*cs.picture->m_uEnerHpCtu[ctuIdx]*/ / chroma_weight) : lambda;
double dist_unfilter_ctu = get_unfiltered_distortion_cov_classes(&alf_cov[ctu_idx * num_classes], num_classes);
//ctxTempStart = AlfCtx(m_CABACEstimator->getCtx()); //ctxTempStart = AlfCtx(m_CABACEstimator->getCtx());
memcpy(&ctx_temp_start, &cabac_estimator, sizeof(ctx_temp_start)); memcpy(&ctx_temp_start, &cabac_estimator, sizeof(ctx_temp_start));
@ -1896,7 +1898,7 @@ double kvz_alf_derive_ctb_alf_enable_flags(encoder_state_t * const state,
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB
if (is_luma) if (is_luma)
{ {
cost_on += get_filtered_distortion(g_alf_covariance[comp_id][ctu_idx], num_classes, g_alf_aps_temp.num_luma_filters - 1, num_coeff, bit_depth); cost_on += get_filtered_distortion(&alf_cov[ctu_idx * num_classes], num_classes, g_alf_aps_temp.num_luma_filters - 1, num_coeff, bit_depth);
} }
else else
{ {
@ -1919,7 +1921,7 @@ double kvz_alf_derive_ctb_alf_enable_flags(encoder_state_t * const state,
double r_altCost = ctu_lambda * (23 - cabac_estimator.bits_left) + (cabac_estimator.num_buffered_bytes << 3); //frac_bits_scale * 0/*m_CABACEstimator->getEstFracBits()*/; double r_altCost = ctu_lambda * (23 - cabac_estimator.bits_left) + (cabac_estimator.num_buffered_bytes << 3); //frac_bits_scale * 0/*m_CABACEstimator->getEstFracBits()*/;
double alt_dist = 0.; double alt_dist = 0.;
alt_dist += calc_error_for_coeffs(&g_alf_covariance[comp_id][ctu_idx][0], g_filter_clipp_set[alt_idx], g_filter_coeff_set[alt_idx], num_coeff, bit_depth); alt_dist += calc_error_for_coeffs(&alf_cov[ctu_idx * num_classes], g_filter_clipp_set[alt_idx], g_filter_coeff_set[alt_idx], num_coeff, bit_depth);
double alt_cost = alt_dist + r_altCost; double alt_cost = alt_dist + r_altCost;
if (alt_cost < best_alt_cost) if (alt_cost < best_alt_cost)
@ -1985,6 +1987,67 @@ double kvz_alf_derive_ctb_alf_enable_flags(encoder_state_t * const state,
} }
void kvz_alf_create(videoframe_t *frame, enum kvz_chroma_format chroma_format)
{
const int num_ctus_in_pic = frame->width_in_lcu * frame->height_in_lcu;
const int luma_coeffs = 13;
const int chroma_coeffs = 7;
int num_classes = 0;
if (chroma_format != KVZ_CSP_400) {
for (int comp_idx = 0; comp_idx < MAX_NUM_COMPONENT; comp_idx++)
{
num_classes += comp_idx ? 1 : MAX_NUM_ALF_CLASSES;
}
}
else
{
num_classes = MAX_NUM_ALF_CLASSES;
}
const int num_covs = num_ctus_in_pic * num_classes;
const int num_luma_covs = num_ctus_in_pic * MAX_NUM_ALF_CLASSES;
frame->alf_info->alf_covariance = malloc(num_covs * sizeof(alf_covariance));
frame->alf_info->alf_covariance_y = &frame->alf_info->alf_covariance[0];
for (int indx = 0; indx < num_luma_covs; indx++)
{
init_alf_covariance(&frame->alf_info->alf_covariance_y[indx], luma_coeffs);
}
for (int k = 0; k < MAX_NUM_ALF_CLASSES; k++)
{
init_alf_covariance(&frame->alf_info->alf_covariance_frame_luma[k], luma_coeffs);
}
if (chroma_format != KVZ_CSP_400) {
const int num_chroma_covs = num_ctus_in_pic;
frame->alf_info->alf_covariance_u = &frame->alf_info->alf_covariance[num_luma_covs];
frame->alf_info->alf_covariance_v = &frame->alf_info->alf_covariance[num_luma_covs + num_chroma_covs];
for (int k = 0; k < num_chroma_covs; k++)
{
init_alf_covariance(&frame->alf_info->alf_covariance_u[k], chroma_coeffs);
init_alf_covariance(&frame->alf_info->alf_covariance_v[k], chroma_coeffs);
}
for (int k = 0; k < MAX_NUM_ALF_ALTERNATIVES_CHROMA; k++)
{
init_alf_covariance(&frame->alf_info->alf_covariance_frame_chroma[k], chroma_coeffs);
}
}
}
void kvz_alf_enc_create(encoder_state_t * const state) void kvz_alf_enc_create(encoder_state_t * const state)
{ {
if (g_created) { if (g_created) {
@ -2062,7 +2125,7 @@ void kvz_alf_enc_create(encoder_state_t * const state)
#endif*/ #endif*/
int num_coeffs = channel_idx ? 7 : 13; int num_coeffs = channel_idx ? 7 : 13;
//m_alfCovarianceFrame[ch_type] = new AlfCovariance*[m_filterShapes[ch_type].size()]; //m_alfCovarianceFrame[ch_type] = new AlfCovariance*[m_filterShapes[ch_type].size()];
g_alf_covariance_frame[ch_type] = malloc(num_classes * sizeof(alf_covariance)); /*g_alf_covariance_frame[ch_type] = malloc(num_classes * sizeof(alf_covariance));
for (int k = 0; k < num_classes; k++) for (int k = 0; k < num_classes; k++)
{ {
g_alf_covariance_frame[ch_type][k].num_coeff = num_coeffs; g_alf_covariance_frame[ch_type][k].num_coeff = num_coeffs;
@ -2070,7 +2133,7 @@ void kvz_alf_enc_create(encoder_state_t * const state)
g_alf_covariance_frame[ch_type][k].pix_acc = 0; g_alf_covariance_frame[ch_type][k].pix_acc = 0;
memset(g_alf_covariance_frame[ch_type][k].y, 0, sizeof(g_alf_covariance_frame[ch_type][k].y)); memset(g_alf_covariance_frame[ch_type][k].y, 0, sizeof(g_alf_covariance_frame[ch_type][k].y));
memset(g_alf_covariance_frame[ch_type][k].ee, 0, sizeof(g_alf_covariance_frame[ch_type][k].ee)); memset(g_alf_covariance_frame[ch_type][k].ee, 0, sizeof(g_alf_covariance_frame[ch_type][k].ee));
} }*/
} }
for (int comp_idx = 0; comp_idx < MAX_NUM_COMPONENT; comp_idx++) for (int comp_idx = 0; comp_idx < MAX_NUM_COMPONENT; comp_idx++)
@ -2099,7 +2162,7 @@ void kvz_alf_enc_create(encoder_state_t * const state)
int num_coeffs = comp_idx ? 7 : 13; int num_coeffs = comp_idx ? 7 : 13;
int alf_num_clipping_values = MAX_ALF_NUM_CLIPPING_VALUES; int alf_num_clipping_values = MAX_ALF_NUM_CLIPPING_VALUES;
g_alf_covariance[comp_idx] = malloc(num_ctus_in_pic * sizeof(**g_alf_covariance[comp_idx])); /*g_alf_covariance[comp_idx] = malloc(num_ctus_in_pic * sizeof(**g_alf_covariance[comp_idx]));
for (int j = 0; j < num_ctus_in_pic; j++) for (int j = 0; j < num_ctus_in_pic; j++)
{ {
g_alf_covariance[comp_idx][j] = malloc(num_classes * sizeof(alf_covariance)); g_alf_covariance[comp_idx][j] = malloc(num_classes * sizeof(alf_covariance));
@ -2111,7 +2174,7 @@ void kvz_alf_enc_create(encoder_state_t * const state)
memset(g_alf_covariance[comp_idx][j][k].y, 0, sizeof(g_alf_covariance[comp_idx][j][k].y)); memset(g_alf_covariance[comp_idx][j][k].y, 0, sizeof(g_alf_covariance[comp_idx][j][k].y));
memset(g_alf_covariance[comp_idx][j][k].ee, 0, sizeof(g_alf_covariance[comp_idx][j][k].ee)); memset(g_alf_covariance[comp_idx][j][k].ee, 0, sizeof(g_alf_covariance[comp_idx][j][k].ee));
} }
} }*/
} }
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB
@ -2273,6 +2336,26 @@ void kvz_alf_reconstruct(encoder_state_t * const state)
} }
} }
void kvz_alf_destroy(videoframe_t * const frame)
{
if (frame->alf_info->alf_covariance_y)
{
frame->alf_info->alf_covariance_y = NULL;
}
if (frame->alf_info->alf_covariance_u)
{
frame->alf_info->alf_covariance_u = NULL;
}
if (frame->alf_info->alf_covariance_v)
{
frame->alf_info->alf_covariance_v = NULL;
}
if (frame->alf_info->alf_covariance)
{
FREE_POINTER(frame->alf_info->alf_covariance);
}
}
void kvz_alf_enc_destroy(videoframe_t * const frame) void kvz_alf_enc_destroy(videoframe_t * const frame)
{ {
if (!g_created) if (!g_created)
@ -2284,10 +2367,10 @@ void kvz_alf_enc_destroy(videoframe_t * const frame)
for (int channel_idx = 0; channel_idx < MAX_NUM_CHANNEL_TYPE; channel_idx++) for (int channel_idx = 0; channel_idx < MAX_NUM_CHANNEL_TYPE; channel_idx++)
{ {
if (g_alf_covariance_frame[channel_idx]) /*if (g_alf_covariance_frame[channel_idx])
{ {
FREE_POINTER(g_alf_covariance_frame[channel_idx]); FREE_POINTER(g_alf_covariance_frame[channel_idx]);
} }*/
} }
for (int comp_idx = 0; comp_idx < MAX_NUM_COMPONENT; comp_idx++) for (int comp_idx = 0; comp_idx < MAX_NUM_COMPONENT; comp_idx++)
@ -2313,14 +2396,14 @@ void kvz_alf_enc_destroy(videoframe_t * const frame)
} }
//#endif //#endif
if (g_alf_covariance[comp_idx]) /*if (g_alf_covariance[comp_idx])
{ {
for (int k = 0; k < num_ctus_in_pic; k++) for (int k = 0; k < num_ctus_in_pic; k++)
{ {
FREE_POINTER(g_alf_covariance[comp_idx][k]); FREE_POINTER(g_alf_covariance[comp_idx][k]);
} }
FREE_POINTER(g_alf_covariance[comp_idx]); FREE_POINTER(g_alf_covariance[comp_idx]);
} }*/
} }
if (g_filter_coeff_set) if (g_filter_coeff_set)
@ -2455,6 +2538,7 @@ void kvz_alf_encoder(encoder_state_t * const state,
const double lambda_chroma_weight // = 0.0 const double lambda_chroma_weight // = 0.0
) )
{ {
alf_info_t *alf_info = state->tile->frame->alf_info;
//const TempCtx ctxStart(m_CtxCache, AlfCtx(m_CABACEstimator->getCtx())); //const TempCtx ctxStart(m_CtxCache, AlfCtx(m_CABACEstimator->getCtx()));
cabac_data_t ctx_start; cabac_data_t ctx_start;
memcpy(&ctx_start, &cabac_estimator, sizeof(ctx_start)); memcpy(&ctx_start, &cabac_estimator, sizeof(ctx_start));
@ -2462,6 +2546,7 @@ void kvz_alf_encoder(encoder_state_t * const state,
cabac_data_t ctx_best; cabac_data_t ctx_best;
bool is_luma = channel == CHANNEL_TYPE_LUMA ? 1 : 0; bool is_luma = channel == CHANNEL_TYPE_LUMA ? 1 : 0;
alf_covariance *alf_cov_frame = is_luma ? alf_info->alf_covariance_frame_luma : alf_info->alf_covariance_frame_chroma;
kvz_config cfg = state->encoder_control->cfg; kvz_config cfg = state->encoder_control->cfg;
double cost_min = MAX_DOUBLE; double cost_min = MAX_DOUBLE;
@ -2479,7 +2564,7 @@ void kvz_alf_encoder(encoder_state_t * const state,
{ {
g_alf_aps_temp.num_alternatives_chroma = 1; g_alf_aps_temp.num_alternatives_chroma = 1;
} }
double cost = get_unfiltered_distortion_cov_channel(g_alf_covariance_frame[channel], channel); double cost = get_unfiltered_distortion_cov_channel(alf_cov_frame, channel);
cost /= 1.001; // slight preference for unfiltered choice cost /= 1.001; // slight preference for unfiltered choice
if (cost < cost_min) if (cost < cost_min)
@ -2736,6 +2821,7 @@ void kvz_alf_get_avai_aps_ids_luma(encoder_state_t * const state,
void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state) void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state)
{ {
alf_info_t *alf_info = state->tile->frame->alf_info;
enum kvz_chroma_format chroma_fmt = state->encoder_control->chroma_format; enum kvz_chroma_format chroma_fmt = state->encoder_control->chroma_format;
bool chroma_scale_x = (chroma_fmt == KVZ_CSP_444) ? 0 : 1; bool chroma_scale_x = (chroma_fmt == KVZ_CSP_444) ? 0 : 1;
bool chroma_scale_y = (chroma_fmt != KVZ_CSP_420) ? 0 : 1; bool chroma_scale_y = (chroma_fmt != KVZ_CSP_420) ? 0 : 1;
@ -2752,18 +2838,20 @@ void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state)
const int number_of_components = (chroma_fmt == KVZ_CSP_400) ? 1 : MAX_NUM_COMPONENT; const int number_of_components = (chroma_fmt == KVZ_CSP_400) ? 1 : MAX_NUM_COMPONENT;
// init CTU stats buffers // init CTU stats buffers
for (int comp_idx = 0; comp_idx < number_of_components; comp_idx++)
{
bool is_luma = comp_idx == 0 ? 1 : 0;
const int num_classes = is_luma ? MAX_NUM_ALF_CLASSES : 1;
for (int class_idx = 0; class_idx < num_classes; class_idx++)
{ {
for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++) for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++)
{ {
reset_alf_covariance(&g_alf_covariance[comp_idx][ctu_idx][class_idx], MAX_ALF_NUM_CLIPPING_VALUES); for (int class_idx = 0; class_idx < MAX_NUM_ALF_CLASSES; class_idx++)
{
reset_alf_covariance(&state->tile->frame->alf_info->alf_covariance_y[(ctu_idx * MAX_NUM_ALF_CLASSES) + class_idx], MAX_ALF_NUM_CLIPPING_VALUES);
} }
} }
for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++)
{
reset_alf_covariance(&state->tile->frame->alf_info->alf_covariance_u[ctu_idx], MAX_ALF_NUM_CLIPPING_VALUES);
reset_alf_covariance(&state->tile->frame->alf_info->alf_covariance_v[ctu_idx], MAX_ALF_NUM_CLIPPING_VALUES);
}
} }
//kerran jossain muualla (kai?) //kerran jossain muualla (kai?)
@ -2775,24 +2863,37 @@ void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state)
const int num_classes = channel_id == CHANNEL_TYPE_LUMA ? MAX_NUM_ALF_CLASSES : 1; const int num_classes = channel_id == CHANNEL_TYPE_LUMA ? MAX_NUM_ALF_CLASSES : 1;
for (int class_idx = 0; class_idx < num_classes; class_idx++) for (int class_idx = 0; class_idx < num_classes; class_idx++)
{ {
reset_alf_covariance(&g_alf_covariance_frame[channel_idx][class_idx], MAX_ALF_NUM_CLIPPING_VALUES); reset_alf_covariance(&alf_info->alf_covariance_frame_luma[class_idx], MAX_ALF_NUM_CLIPPING_VALUES);
}
reset_alf_covariance(&alf_info->alf_covariance_frame_chroma[0], MAX_ALF_NUM_CLIPPING_VALUES);
} }
}
int max_cu_height = LCU_WIDTH; alf_covariance* alf_cov;
alf_covariance* alf_cov_frame;
#if FULL_FRAME
for (int y_pos = 0; y_pos < pic_height; y_pos += LCU_WIDTH) for (int y_pos = 0; y_pos < pic_height; y_pos += LCU_WIDTH)
{ {
for (int x_pos = 0; x_pos < pic_width; x_pos += LCU_WIDTH) for (int x_pos = 0; x_pos < pic_width; x_pos += LCU_WIDTH)
{ {
const int width = (x_pos + LCU_WIDTH > pic_width) ? (pic_width - x_pos) : LCU_WIDTH; const int width = (x_pos + LCU_WIDTH > pic_width) ? (pic_width - x_pos) : LCU_WIDTH;
const int height = (y_pos + LCU_WIDTH > pic_height) ? (pic_height - y_pos) : LCU_WIDTH; const int height = (y_pos + LCU_WIDTH > pic_height) ? (pic_height - y_pos) : LCU_WIDTH;
{
for (int comp_idx = 0; comp_idx < number_of_components; comp_idx++) for (int comp_idx = 0; comp_idx < number_of_components; comp_idx++)
{ {
if (comp_idx == COMPONENT_Y)
{
alf_cov = state->tile->frame->alf_info->alf_covariance_y;
}
else if (comp_idx == COMPONENT_Cb || comp_idx == COMPONENT_Cr)
{
alf_cov = comp_idx == COMPONENT_Cb ? state->tile->frame->alf_info->alf_covariance_u : state->tile->frame->alf_info->alf_covariance_v;
}
else
{
assert(0);
}
const bool is_luma = comp_idx == COMPONENT_Y ? 1 : 0; const bool is_luma = comp_idx == COMPONENT_Y ? 1 : 0;
channel_type ch_type = is_luma ? CHANNEL_TYPE_LUMA : CHANNEL_TYPE_CHROMA; channel_type ch_type = is_luma ? CHANNEL_TYPE_LUMA : CHANNEL_TYPE_CHROMA;
alf_cov_frame = is_luma ? alf_info->alf_covariance_frame_luma : alf_info->alf_covariance_frame_chroma;
int blk_w = is_luma ? width : width >> chroma_scale_x; int blk_w = is_luma ? width : width >> chroma_scale_x;
int blk_h = is_luma ? height : height >> chroma_scale_y; int blk_h = is_luma ? height : height >> chroma_scale_y;
@ -2805,24 +2906,23 @@ void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state)
kvz_pixel *org = comp_idx ? (comp_idx - 1 ? &state->tile->frame->source->v[pos_x + pos_y * org_stride] : &state->tile->frame->source->u[pos_x + pos_y * org_stride]) : &state->tile->frame->source->y[pos_x + pos_y * org_stride]; kvz_pixel *org = comp_idx ? (comp_idx - 1 ? &state->tile->frame->source->v[pos_x + pos_y * org_stride] : &state->tile->frame->source->u[pos_x + pos_y * org_stride]) : &state->tile->frame->source->y[pos_x + pos_y * org_stride];
kvz_pixel *rec = comp_idx ? (comp_idx - 1 ? &state->tile->frame->rec->v[pos_x + pos_y * rec_stride] : &state->tile->frame->rec->u[pos_x + pos_y * rec_stride]) : &state->tile->frame->rec->y[pos_x + pos_y * rec_stride]; kvz_pixel *rec = comp_idx ? (comp_idx - 1 ? &state->tile->frame->rec->v[pos_x + pos_y * rec_stride] : &state->tile->frame->rec->u[pos_x + pos_y * rec_stride]) : &state->tile->frame->rec->y[pos_x + pos_y * rec_stride];
const int num_classes = is_luma ? MAX_NUM_ALF_CLASSES : 1;
const int cov_index = ctu_rs_addr * num_classes;
kvz_alf_get_blk_stats(state, ch_type, kvz_alf_get_blk_stats(state, ch_type,
&g_alf_covariance[comp_idx][ctu_rs_addr], &alf_cov[cov_index],
comp_idx ? NULL : g_classifier, comp_idx ? NULL : g_classifier,
org, org_stride, rec, rec_stride, pos_x, pos_y, pos_x, pos_y, blk_w, blk_h, org, org_stride, rec, rec_stride, pos_x, pos_y, pos_x, pos_y, blk_w, blk_h,
(is_luma ? alf_vb_luma_ctu_height : alf_vb_chma_ctu_height), (is_luma ? alf_vb_luma_ctu_height : alf_vb_chma_ctu_height),
(is_luma) ? alf_vb_luma_pos : alf_vb_chma_pos (is_luma) ? alf_vb_luma_pos : alf_vb_chma_pos
); );
const int num_classes = is_luma ? MAX_NUM_ALF_CLASSES : 1;
for (int class_idx = 0; class_idx < num_classes; class_idx++) for (int class_idx = 0; class_idx < num_classes; class_idx++)
{ {
add_alf_cov(&g_alf_covariance_frame[ch_type][is_luma ? class_idx : 0], add_alf_cov(&alf_cov_frame[is_luma ? class_idx : 0],
&g_alf_covariance[comp_idx][ctu_rs_addr][class_idx] &alf_cov[cov_index + class_idx]
); );
} }
} }
}
ctu_rs_addr++; ctu_rs_addr++;
} }
} }
@ -2830,7 +2930,7 @@ void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state)
void kvz_alf_get_blk_stats(encoder_state_t * const state, void kvz_alf_get_blk_stats(encoder_state_t * const state,
channel_type channel, channel_type channel,
alf_covariance **alf_covariance, alf_covariance *alf_covariance,
alf_classifier **g_classifier, alf_classifier **g_classifier,
kvz_pixel *org, kvz_pixel *org,
int32_t org_stride, int32_t org_stride,
@ -2887,11 +2987,11 @@ void kvz_alf_get_blk_stats(encoder_state_t * const state,
{ {
if (0/*m_alfWSSD*/) if (0/*m_alfWSSD*/)
{ {
(*alf_covariance)[class_idx].ee[b0][b1][k][l] += weight * (e_local[k][b0] * (double)e_local[l][b1]); alf_covariance[class_idx].ee[b0][b1][k][l] += weight * (e_local[k][b0] * (double)e_local[l][b1]);
} }
else else
{ {
(*alf_covariance)[class_idx].ee[b0][b1][k][l] += e_local[k][b0] * (double)e_local[l][b1]; alf_covariance[class_idx].ee[b0][b1][k][l] += e_local[k][b0] * (double)e_local[l][b1];
} }
} }
} }
@ -2900,21 +3000,21 @@ void kvz_alf_get_blk_stats(encoder_state_t * const state,
{ {
if (0/*m_alfWSSD*/) if (0/*m_alfWSSD*/)
{ {
(*alf_covariance)[class_idx].y[b][k] += weight * (e_local[k][b] * (double)y_local); alf_covariance[class_idx].y[b][k] += weight * (e_local[k][b] * (double)y_local);
} }
else else
{ {
(*alf_covariance)[class_idx].y[b][k] += e_local[k][b] * (double)y_local; alf_covariance[class_idx].y[b][k] += e_local[k][b] * (double)y_local;
} }
} }
} }
if (0/*m_alfWSSD*/) if (0/*m_alfWSSD*/)
{ {
(*alf_covariance)[class_idx].pix_acc += weight * (y_local * (double)y_local); alf_covariance[class_idx].pix_acc += weight * (y_local * (double)y_local);
} }
else else
{ {
(*alf_covariance)[class_idx].pix_acc += y_local * (double)y_local; alf_covariance[class_idx].pix_acc += y_local * (double)y_local;
} }
} }
org += org_stride; org += org_stride;
@ -2932,7 +3032,7 @@ void kvz_alf_get_blk_stats(encoder_state_t * const state,
{ {
for (int b1 = 0; b1 < num_bins; b1++) for (int b1 = 0; b1 < num_bins; b1++)
{ {
(*alf_covariance)[class_idx].ee[b0][b1][k][l] = (*alf_covariance)[class_idx].ee[b1][b0][l][k]; alf_covariance[class_idx].ee[b0][b1][k][l] = alf_covariance[class_idx].ee[b1][b0][l][k];
} }
} }
} }
@ -2947,6 +3047,24 @@ void kvz_alf_calc_covariance(int16_t e_local[MAX_NUM_ALF_LUMA_COEFF][MAX_ALF_NUM
const int transpose_idx, const int transpose_idx,
int vb_distance) int vb_distance)
{ {
static const int alf_pattern_5[13] = {
0,
1, 2, 3,
4, 5, 6, 5, 4,
3, 2, 1,
0
};
static const int alf_pattern_7[25] = {
0,
1, 2, 3,
4, 5, 6, 7, 8,
9, 10, 11, 12, 11, 10, 9,
8, 7, 6, 5, 4,
3, 2, 1,
0
};
int clip_top_row = -4; int clip_top_row = -4;
int clip_bot_row = 4; int clip_bot_row = 4;
if (vb_distance >= -3 && vb_distance < 0) if (vb_distance >= -3 && vb_distance < 0)
@ -3075,17 +3193,20 @@ double kvz_alf_get_filter_coeff_and_cost(encoder_state_t * const state,
bool b_re_collect_stat, bool b_re_collect_stat,
bool only_filter_cost) bool only_filter_cost)
{ {
alf_info_t *alf_info = state->tile->frame->alf_info;
bool is_luma = channel == CHANNEL_TYPE_LUMA ? 1 : 0; bool is_luma = channel == CHANNEL_TYPE_LUMA ? 1 : 0;
const int num_coeff = channel == CHANNEL_TYPE_LUMA ? 13 : 7; const int num_coeff = channel == CHANNEL_TYPE_LUMA ? 13 : 7;
double lambda = state->frame->lambda; double lambda = state->frame->lambda;
int clip_merged[MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_LUMA_COEFF]; int clip_merged[MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_LUMA_COEFF];
const int8_t bit_depth = state->encoder_control->bitdepth; const int8_t bit_depth = state->encoder_control->bitdepth;
const int32_t num_ctus_in_pic = state->lcu_order_count; const int32_t num_ctus_in_pic = state->lcu_order_count;
alf_covariance *alf_cov_frame = is_luma ? alf_info->alf_covariance_frame_luma : alf_info->alf_covariance_frame_chroma;
//collect stat based on CTU decision //collect stat based on CTU decision
if (b_re_collect_stat) if (b_re_collect_stat)
{ {
get_frame_stats(channel, num_ctus_in_pic); get_frame_stats(state->tile->frame->alf_info, channel, num_ctus_in_pic);
} }
double dist = dist_unfilter; double dist = dist_unfilter;
@ -3115,7 +3236,7 @@ double kvz_alf_get_filter_coeff_and_cost(encoder_state_t * const state,
reset_alf_covariance(&g_alf_covariance_merged[MAX_NUM_ALF_CLASSES], MAX_ALF_NUM_CLIPPING_VALUES); reset_alf_covariance(&g_alf_covariance_merged[MAX_NUM_ALF_CLASSES], MAX_ALF_NUM_CLIPPING_VALUES);
reset_alf_covariance(&g_alf_covariance_merged[MAX_NUM_ALF_CLASSES + 1], MAX_ALF_NUM_CLIPPING_VALUES); reset_alf_covariance(&g_alf_covariance_merged[MAX_NUM_ALF_CLASSES + 1], MAX_ALF_NUM_CLIPPING_VALUES);
//distortion //distortion
dist += kvz_alf_merge_filters_and_cost(state, &g_alf_aps_temp, channel, ui_coeff_bits, g_alf_covariance_frame[channel], g_alf_covariance_merged, clip_merged); dist += kvz_alf_merge_filters_and_cost(state, &g_alf_aps_temp, channel, ui_coeff_bits, alf_cov_frame, g_alf_covariance_merged, clip_merged);
} }
else else
{ {
@ -3137,7 +3258,7 @@ double kvz_alf_get_filter_coeff_and_cost(encoder_state_t * const state,
for (int alt_idx = 0; alt_idx < g_alf_aps_temp.num_alternatives_chroma; ++alt_idx) for (int alt_idx = 0; alt_idx < g_alf_aps_temp.num_alternatives_chroma; ++alt_idx)
{ {
assert(num_coeff == g_alf_covariance_frame[channel][alt_idx].num_coeff); assert(num_coeff == alf_cov_frame[alt_idx].num_coeff);
alf_aps best_slice_param; alf_aps best_slice_param;
double best_cost = MAX_DOUBLE; double best_cost = MAX_DOUBLE;
double best_dist = MAX_DOUBLE; double best_dist = MAX_DOUBLE;
@ -3157,7 +3278,7 @@ double kvz_alf_get_filter_coeff_and_cost(encoder_state_t * const state,
g_filter_clipp_set[alt_idx][i] = fill_val; g_filter_clipp_set[alt_idx][i] = fill_val;
} }
double dist = g_alf_covariance_frame[channel][alt_idx].pix_acc + kvz_alf_derive_coeff_quant(channel, g_filter_clipp_set[alt_idx], g_filter_coeff_set[alt_idx], &g_alf_covariance_frame[channel][alt_idx], bit_depth, non_linear_flag); double dist = alf_cov_frame[alt_idx].pix_acc + kvz_alf_derive_coeff_quant(channel, g_filter_clipp_set[alt_idx], g_filter_coeff_set[alt_idx], &alf_cov_frame[alt_idx], bit_depth, non_linear_flag);
for (int i = 0; i < MAX_NUM_ALF_CHROMA_COEFF; i++) for (int i = 0; i < MAX_NUM_ALF_CHROMA_COEFF; i++)
{ {
g_alf_aps_temp.chroma_coeff[alt_idx][i] = g_filter_coeff_set[alt_idx][i]; g_alf_aps_temp.chroma_coeff[alt_idx][i] = g_filter_coeff_set[alt_idx][i];
@ -3749,7 +3870,7 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
alf_aps *aps, alf_aps *aps,
const double lambda_chroma_weight) const double lambda_chroma_weight)
{ {
//TempCtx ctxStart(m_CtxCache, AlfCtx(m_CABACEstimator->getCtx())); alf_info_t *alf_info = state->tile->frame->alf_info;
cabac_data_t ctx_start; cabac_data_t ctx_start;
memcpy(&ctx_start, &cabac_estimator, sizeof(ctx_start)); memcpy(&ctx_start, &cabac_estimator, sizeof(ctx_start));
//TempCtx ctxBest(m_CtxCache); //TempCtx ctxBest(m_CtxCache);
@ -3773,6 +3894,8 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
int size_of_aps_ids = 0; int size_of_aps_ids = 0;
const int32_t num_ctus_in_pic = state->lcu_order_count; const int32_t num_ctus_in_pic = state->lcu_order_count;
alf_aps alf_aps_temp_nl; alf_aps alf_aps_temp_nl;
alf_covariance *alf_cov_chroma;
int cov_indx = 0;
//AlfSliceParam alfSliceParamNewFiltersBest = alfSliceParamNewFilters; //AlfSliceParam alfSliceParamNewFiltersBest = alfSliceParamNewFilters;
alf_aps alf_aps_new_filters_best; alf_aps alf_aps_new_filters_best;
@ -3782,22 +3905,21 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
bool has_new_filters[2] = { aps->enabled_flag[COMPONENT_Y] , aps->enabled_flag[COMPONENT_Cb] || aps->enabled_flag[COMPONENT_Cr] }; bool has_new_filters[2] = { aps->enabled_flag[COMPONENT_Y] , aps->enabled_flag[COMPONENT_Cb] || aps->enabled_flag[COMPONENT_Cr] };
//initDistortion(); //initDistortion();
for (int comp = 0; comp < MAX_NUM_COMPONENT; comp++)
{
for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++) for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++)
{ {
g_ctb_distortion_unfilter[comp][ctu_idx] = get_unfiltered_distortion_cov_classes(g_alf_covariance[comp][ctu_idx], comp == 0 ? MAX_NUM_ALF_CLASSES : 1); g_ctb_distortion_unfilter[COMPONENT_Y][ctu_idx] = get_unfiltered_distortion_cov_classes(&alf_info->alf_covariance_y[ctu_idx], MAX_NUM_ALF_CLASSES);
} g_ctb_distortion_unfilter[COMPONENT_Cb][ctu_idx] = get_unfiltered_distortion_cov_classes(&alf_info->alf_covariance_u[ctu_idx], 1);
g_ctb_distortion_unfilter[COMPONENT_Cr][ctu_idx] = get_unfiltered_distortion_cov_classes(&alf_info->alf_covariance_v[ctu_idx], 1);
} }
//luma //luma
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB
copy_alf_param(&g_alf_aps_temp, aps); copy_alf_param(&g_alf_aps_temp, aps);
memset(g_ctu_enable_flag[COMPONENT_Y], 1, sizeof(uint8_t) * num_ctus_in_pic); memset(g_ctu_enable_flag[COMPONENT_Y], 1, sizeof(uint8_t) * num_ctus_in_pic);
get_frame_stats(CHANNEL_TYPE_LUMA, num_ctus_in_pic); get_frame_stats(alf_info, CHANNEL_TYPE_LUMA, num_ctus_in_pic);
memset(g_ctu_enable_flag[COMPONENT_Y], 0, sizeof(uint8_t) * num_ctus_in_pic); memset(g_ctu_enable_flag[COMPONENT_Y], 0, sizeof(uint8_t) * num_ctus_in_pic);
double cost_off = get_unfiltered_distortion_cov_channel(g_alf_covariance_frame[CHANNEL_TYPE_LUMA], CHANNEL_TYPE_LUMA); double cost_off = get_unfiltered_distortion_cov_channel(alf_info->alf_covariance_frame_luma, CHANNEL_TYPE_LUMA);
int new_aps_id; int new_aps_id;
int aps_ids[ALF_CTB_MAX_NUM_APS]; int aps_ids[ALF_CTB_MAX_NUM_APS];
@ -3863,6 +3985,7 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
{ {
blocks_using_new_filter++; blocks_using_new_filter++;
d_dist_org_new_filter += g_ctb_distortion_unfilter[COMPONENT_Y][ctu_idx]; d_dist_org_new_filter += g_ctb_distortion_unfilter[COMPONENT_Y][ctu_idx];
cov_indx = ctu_idx * MAX_NUM_ALF_CLASSES;
for (int class_idx = 0; class_idx < MAX_NUM_ALF_CLASSES; class_idx++) for (int class_idx = 0; class_idx < MAX_NUM_ALF_CLASSES; class_idx++)
{ {
short* p_coeff = g_coeff_final; short* p_coeff = g_coeff_final;
@ -3872,7 +3995,7 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
g_filter_tmp[i] = p_coeff[class_idx * MAX_NUM_ALF_LUMA_COEFF + i]; g_filter_tmp[i] = p_coeff[class_idx * MAX_NUM_ALF_LUMA_COEFF + i];
g_clip_tmp[i] = p_clipp[class_idx * MAX_NUM_ALF_LUMA_COEFF + i]; g_clip_tmp[i] = p_clipp[class_idx * MAX_NUM_ALF_LUMA_COEFF + i];
} }
d_dist_org_new_filter += calc_error_for_coeffs(&g_alf_covariance[COMPONENT_Y][ctu_idx][class_idx], g_clip_tmp, g_filter_tmp, MAX_NUM_ALF_LUMA_COEFF, bit_depth); d_dist_org_new_filter += calc_error_for_coeffs(&alf_info->alf_covariance_y[cov_indx + class_idx], g_clip_tmp, g_filter_tmp, MAX_NUM_ALF_LUMA_COEFF, bit_depth);
} }
} }
} //for ctb } //for ctb
@ -3948,12 +4071,13 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
double rate_on = (23 - cabac_estimator.bits_left) + (cabac_estimator.num_buffered_bytes << 3); //frac_bits_scale * 0; /*(double)m_CABACEstimator->getEstFracBits()*/ ; double rate_on = (23 - cabac_estimator.bits_left) + (cabac_estimator.num_buffered_bytes << 3); //frac_bits_scale * 0; /*(double)m_CABACEstimator->getEstFracBits()*/ ;
//distortion //distortion
double dist = dist_unfilter_ctb; double dist = dist_unfilter_ctb;
cov_indx = ctu_idx * MAX_NUM_ALF_CLASSES;
for (int class_idx = 0; class_idx < MAX_NUM_ALF_CLASSES; class_idx++) for (int class_idx = 0; class_idx < MAX_NUM_ALF_CLASSES; class_idx++)
{ {
if (filter_set_idx < ALF_NUM_FIXED_FILTER_SETS) if (filter_set_idx < ALF_NUM_FIXED_FILTER_SETS)
{ {
int filter_idx = g_class_to_filter_mapping[filter_set_idx][class_idx]; int filter_idx = g_class_to_filter_mapping[filter_set_idx][class_idx];
dist += calc_error_for_coeffs(&g_alf_covariance[COMPONENT_Y][ctu_idx][class_idx], clip_default, g_fixed_filter_set_coeff[filter_idx], MAX_NUM_ALF_LUMA_COEFF, bit_depth); dist += calc_error_for_coeffs(&alf_info->alf_covariance_y[cov_indx + class_idx], clip_default, g_fixed_filter_set_coeff[filter_idx], MAX_NUM_ALF_LUMA_COEFF, bit_depth);
} }
else else
{ {
@ -3979,7 +4103,7 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
g_filter_tmp[i] = p_coeff[class_idx * MAX_NUM_ALF_LUMA_COEFF + i]; g_filter_tmp[i] = p_coeff[class_idx * MAX_NUM_ALF_LUMA_COEFF + i];
g_clip_tmp[i] = p_clipp[class_idx * MAX_NUM_ALF_LUMA_COEFF + i]; g_clip_tmp[i] = p_clipp[class_idx * MAX_NUM_ALF_LUMA_COEFF + i];
} }
dist += calc_error_for_coeffs(&g_alf_covariance[COMPONENT_Y][ctu_idx][class_idx], g_clip_tmp, g_filter_tmp, MAX_NUM_ALF_LUMA_COEFF, bit_depth); dist += calc_error_for_coeffs(&alf_info->alf_covariance_y[cov_indx + class_idx], g_clip_tmp, g_filter_tmp, MAX_NUM_ALF_LUMA_COEFF, bit_depth);
} }
} }
//cost //cost
@ -4170,8 +4294,8 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
g_ctu_alternative[COMPONENT_Cr][ctu_idx] = 0; g_ctu_alternative[COMPONENT_Cr][ctu_idx] = 0;
} }
set_ctu_enable_flag(g_ctu_enable_flag, CHANNEL_TYPE_CHROMA, 1, num_ctus_in_pic); set_ctu_enable_flag(g_ctu_enable_flag, CHANNEL_TYPE_CHROMA, 1, num_ctus_in_pic);
get_frame_stats(CHANNEL_TYPE_CHROMA, num_ctus_in_pic); get_frame_stats(state->tile->frame->alf_info, CHANNEL_TYPE_CHROMA, num_ctus_in_pic);
cost_off = get_unfiltered_distortion_cov_channel(g_alf_covariance_frame[CHANNEL_TYPE_CHROMA], CHANNEL_TYPE_CHROMA); cost_off = get_unfiltered_distortion_cov_channel(alf_info->alf_covariance_frame_chroma, CHANNEL_TYPE_CHROMA);
cost_min = MAX_DOUBLE; cost_min = MAX_DOUBLE;
//m_CABACEstimator->getCtx() = AlfCtx(ctxBest); //m_CABACEstimator->getCtx() = AlfCtx(ctxBest);
memcpy(&cabac_estimator, &ctx_best, sizeof(cabac_estimator)); memcpy(&cabac_estimator, &ctx_best, sizeof(cabac_estimator));
@ -4246,38 +4370,26 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
g_alf_aps_temp.enabled_flag[comp_id] = true; g_alf_aps_temp.enabled_flag[comp_id] = true;
//for (int ctb_idx = 0; ctb_idx < g_num_ctus_in_pic; ctb_idx++) //for (int ctb_idx = 0; ctb_idx < g_num_ctus_in_pic; ctb_idx++)
{ {
double dist_unfilter_ctu = g_ctb_distortion_unfilter[comp_id][ctu_idx]; copy_alf_param(&g_alf_aps_temp, aps);
//cost on cur_cost += lambda * g_bits_new_filter[CHANNEL_TYPE_CHROMA];
g_ctu_enable_flag[comp_id][ctu_idx] = 1; }
//ctxTempStart = AlfCtx(m_CABACEstimator->getCtx()); else if (cur_aps && cur_aps->temporal_id <= state->slice->id && cur_aps->new_filter_flag[CHANNEL_TYPE_CHROMA])
memcpy(&ctx_temp_start, &cabac_estimator, sizeof(ctx_temp_start)); {
ctx_temp_start.only_count = 1; //g_alf_slice_aps_temp = cur_aps;
//rate copy_alf_param(&g_alf_aps_temp, cur_aps);
//m_CABACEstimator->getCtx() = AlfCtx(ctxTempStart); }
memcpy(&cabac_estimator, &ctx_temp_start, sizeof(cabac_estimator)); else
//m_CABACEstimator->resetBits(); {
kvz_cabac_reset_bits(&cabac_estimator); continue;
//ctb flag }
code_alf_ctu_enable_flag(state, &cabac_estimator, ctu_idx, comp_id, &g_alf_aps_temp); kvz_alf_reconstruct_coeff(state, &g_alf_aps_temp, CHANNEL_TYPE_CHROMA, true, true);
double rate_on = (23 - cabac_estimator.bits_left) + (cabac_estimator.num_buffered_bytes << 3); //frac_bits_scale*(double)838/*m_CABACEstimator->getEstFracBits()*/; //m_CABACEstimator->getCtx() = AlfCtx(ctxStart);
memcpy(&cabac_estimator, &ctx_start, sizeof(cabac_estimator));
//#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB for (int comp_id = 1; comp_id < MAX_NUM_COMPONENT; comp_id++)
//#if ENABLE_QPA {
const double ctu_lambda = lambda_chroma_weight > 0.0 ? 0/*cs.picture->m_uEnerHpCtu[ctbIdx]*/ / lambda_chroma_weight : g_lambda[comp_id]; alf_cov_chroma = comp_id == COMPONENT_Cb ? alf_info->alf_covariance_u : alf_info->alf_covariance_v;
/*#else g_alf_aps_temp.enabled_flag[comp_id] = true;
const double ctu_lambda = m_lambda[compId]; for (int ctu_idx = 0; ctu_idx < num_ctus_in_pic; ctu_idx++)
#endif*/
double dist = MAX_DOUBLE;
int num_alts = g_alf_aps_temp.num_alternatives_chroma;
//ctxTempBest = AlfCtx(m_CABACEstimator->getCtx());
memcpy(&ctx_temp_best, &cabac_estimator, sizeof(ctx_temp_best));
ctx_temp_best.only_count = 1;
double best_alt_rate = 0;
double best_alt_cost = MAX_DOUBLE;
int best_alt_idx = -1;
//ctxTempAltStart = AlfCtx(ctxTempBest);
memcpy(&ctx_temp_alt_start, &ctx_temp_best, sizeof(ctx_temp_alt_start));
for (int alt_idx = 0; alt_idx < num_alts; ++alt_idx)
{ {
double dist_unfilter_ctu = g_ctb_distortion_unfilter[comp_id][ctu_idx]; double dist_unfilter_ctu = g_ctb_distortion_unfilter[comp_id][ctu_idx];
//cost on //cost on
@ -4321,7 +4433,7 @@ void kvz_alf_encoder_ctb(encoder_state_t * const state,
g_filter_tmp[i] = g_chroma_coeff_final[alt_idx][i]; g_filter_tmp[i] = g_chroma_coeff_final[alt_idx][i];
g_clip_tmp[i] = g_chroma_clipp_final[alt_idx][i]; g_clip_tmp[i] = g_chroma_clipp_final[alt_idx][i];
} }
double alt_dist = calc_error_for_coeffs(&g_alf_covariance[comp_id][ctu_idx][0], g_clip_tmp, g_filter_tmp, MAX_NUM_ALF_CHROMA_COEFF, bit_depth); double alt_dist = calc_error_for_coeffs(&alf_cov_chroma[ctu_idx], g_clip_tmp, g_filter_tmp, MAX_NUM_ALF_CHROMA_COEFF, bit_depth);
double alt_cost = alt_dist + r_alt_cost; double alt_cost = alt_dist + r_alt_cost;
if (alt_cost < best_alt_cost) if (alt_cost < best_alt_cost)
{ {

81
src/alf.h
View file

@ -140,42 +140,6 @@ static const int g_class_to_filter_mapping[ALF_NUM_FIXED_FILTER_SETS][MAX_NUM_AL
{ 16, 31, 32, 15, 60, 30, 4, 17, 19, 25, 22, 20, 4, 53, 19, 21, 22, 46, 25, 55, 26, 48, 63, 58, 55 }, { 16, 31, 32, 15, 60, 30, 4, 17, 19, 25, 22, 20, 4, 53, 19, 21, 22, 46, 25, 55, 26, 48, 63, 58, 55 },
}; };
static const int alf_pattern_5[13] = {
0,
1, 2, 3,
4, 5, 6, 5, 4,
3, 2, 1,
0
};
static const int alf_weights_5[8] = {
2,
2, 2, 2,
2, 2, 1, 1
};
/*#if !JVET_O0216_ALF_COEFF_EG3 || !JVET_O0064_SIMP_ALF_CLIP_CODING
static const int alf_golomb_idx_5[8] = {
0,
0, 1, 0,
0, 1, 2, 2
};
#endif*/
static const int alf_pattern_7[25] = {
0,
1, 2, 3,
4, 5, 6, 7, 8,
9, 10, 11, 12, 11, 10, 9,
8, 7, 6, 5, 4,
3, 2, 1,
0
};
static const int alf_weights_7[14] = {
2,
2, 2, 2,
2, 2, 2, 2, 2,
2, 2, 2, 1, 1
};
//-------------------------typedef enums---------------------------- //-------------------------typedef enums----------------------------
typedef enum { ALF_FILTER_5X5 = 0, ALF_FILTER_7X7 = 1, ALF_NUM_OF_FILTER_TYPES = 2 } alf_filter_type; typedef enum { ALF_FILTER_5X5 = 0, ALF_FILTER_7X7 = 1, ALF_NUM_OF_FILTER_TYPES = 2 } alf_filter_type;
@ -238,19 +202,22 @@ typedef struct alf_classifier {
} alf_classifier; } alf_classifier;
typedef struct alf_info_t { typedef struct alf_info_t {
alf_covariance** g_alf_covariance[MAX_NUM_COMPONENT]; //[component_id][ctu_idx][class_idx] alf_covariance* alf_covariance; //Covariances of each CTU for luma and chroma components //[ctu_idx][class_idx]
alf_covariance* g_alf_covariance_frame[MAX_NUM_CHANNEL_TYPE]; //[channel][class_idx] alf_covariance* alf_covariance_y; //Pointer to the first luma covaraince //[ctu_idx][class_idx]
alf_covariance g_alf_covariance_merged[ALF_NUM_OF_FILTER_TYPES][MAX_NUM_ALF_CLASSES + 2]; alf_covariance* alf_covariance_u; //Pointer to the first Cb covariance //[ctu_idx][class_idx]
int g_alf_clip_merged[ALF_NUM_OF_FILTER_TYPES][MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_LUMA_COEFF]; alf_covariance* alf_covariance_v; //Pointer tot he first Cr covariance //[ctu_idx][class_idx]
uint8_t* g_ctu_enable_flag[MAX_NUM_COMPONENT]; alf_covariance alf_covariance_frame_luma[MAX_NUM_ALF_CLASSES]; //[class_idx]
uint8_t* g_ctu_alternative[MAX_NUM_COMPONENT]; //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB alf_covariance alf_covariance_frame_chroma[MAX_NUM_ALF_ALTERNATIVES_CHROMA]; //[class_idx]
double *g_ctb_distortion_unfilter[MAX_NUM_COMPONENT]; alf_covariance alf_covariance_merged[ALF_NUM_OF_FILTER_TYPES][MAX_NUM_ALF_CLASSES + 2];
int g_aps_id_start; int alf_clip_merged[ALF_NUM_OF_FILTER_TYPES][MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_CLASSES][MAX_NUM_ALF_LUMA_COEFF];
int** g_diff_filter_coeff; // [lumaClassIdx][coeffIdx] uint8_t* ctu_alternative[MAX_NUM_COMPONENT];
int** g_filter_coeff_set; // [lumaClassIdx][coeffIdx] double *ctb_distortion_unfilter[MAX_NUM_COMPONENT];
int** g_filter_clipp_set; // [lumaClassIdx][coeffIdx] int aps_id_start;
short* g_alf_ctb_filter_set_index_tmp; //g_num_ctus_in_pic //voisi olla lokaali muuttuja? int** diff_filter_coeff; // [lumaClassIdx][coeffIdx]
short* g_alf_ctb_filter_index; //g_num_ctus_in_pic int** filter_coeff_set; // [lumaClassIdx][coeffIdx]
int** filter_clipp_set; // [lumaClassIdx][coeffIdx]
short* alf_ctb_filter_set_index_tmp; //g_num_ctus_in_pic //voisi olla lokaali muuttuja?
short* alf_ctb_filter_index; //g_num_ctus_in_pic
} alf_info_t; } alf_info_t;
typedef struct cc_alf_filter_param { typedef struct cc_alf_filter_param {
@ -353,8 +320,9 @@ int g_aps_id_cc_alf_start[2];
int g_reuse_aps_id[2]; int g_reuse_aps_id[2];
//once per frame //once per frame
alf_covariance** g_alf_covariance[MAX_NUM_COMPONENT]; //[component_id][ctu_idx][class_idx] //alf_covariance** g_alf_covariance[MAX_NUM_COMPONENT]; //[component_id][ctu_idx][class_idx]
alf_covariance* g_alf_covariance_frame[MAX_NUM_CHANNEL_TYPE]; //[channel][class_idx] //alf_covariance* g_alf_covariance_frame[MAX_NUM_CHANNEL_TYPE]; //[channel][class_idx]
alf_covariance g_alf_covariance_merged[MAX_NUM_ALF_CLASSES + 2]; alf_covariance g_alf_covariance_merged[MAX_NUM_ALF_CLASSES + 2];
uint8_t* g_ctu_enable_flag[MAX_NUM_COMPONENT]; uint8_t* g_ctu_enable_flag[MAX_NUM_COMPONENT];
uint8_t* g_ctu_alternative[MAX_NUM_COMPONENT]; //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB uint8_t* g_ctu_alternative[MAX_NUM_COMPONENT]; //#if JVET_O0090_ALF_CHROMA_FILTER_ALTERNATIVES_CTB
@ -426,8 +394,8 @@ int get_chroma_coeff_rate(alf_aps* aps, int alt_idx);
double get_filtered_distortion(alf_covariance* cov, const int num_classes, const int num_filters_minus1, const int num_coeff); double get_filtered_distortion(alf_covariance* cov, const int num_classes, const int num_filters_minus1, const int num_coeff);
double get_unfiltered_distortion_cov_channel(alf_covariance* cov, channel_type channel); double get_unfiltered_distortion_cov_channel(alf_covariance* cov, channel_type channel);
double get_unfiltered_distortion_cov_classes(alf_covariance* cov, const int num_classes); double get_unfiltered_distortion_cov_classes(alf_covariance* cov, const int num_classes);
void get_frame_stats(channel_type channel, const int32_t num_ctus); void get_frame_stats(alf_info_t *alf_info, channel_type channel, const int32_t num_ctus);
void get_frame_stat(alf_covariance* frame_cov, alf_covariance** ctb_cov, uint8_t* ctb_enable_flags, uint8_t* ctb_alt_idx, const int num_classes, int alt_idx, const int32_t num_ctus); void get_frame_stat(alf_covariance* frame_cov, alf_covariance* ctb_cov, uint8_t* ctb_enable_flags, uint8_t* ctb_alt_idx, const int num_classes, int alt_idx, const int32_t num_ctus);
void copy_cov(alf_covariance *dst, alf_covariance *src); void copy_cov(alf_covariance *dst, alf_covariance *src);
void copy_alf_param(alf_aps *dst, alf_aps *src); void copy_alf_param(alf_aps *dst, alf_aps *src);
@ -439,6 +407,7 @@ void reset_alf_param(alf_aps *src);
void add_alf_cov(alf_covariance *dst, alf_covariance *src); void add_alf_cov(alf_covariance *dst, alf_covariance *src);
void add_alf_cov_lhs_rhs(alf_covariance *dst, alf_covariance *lhs, alf_covariance *rhs); void add_alf_cov_lhs_rhs(alf_covariance *dst, alf_covariance *lhs, alf_covariance *rhs);
void reset_alf_covariance(alf_covariance *alf, int num_bins); void reset_alf_covariance(alf_covariance *alf, int num_bins);
void init_alf_covariance(alf_covariance *alf, int num_bins);
void reset_cc_alf_aps_param(cc_alf_filter_param *cc_alf); void reset_cc_alf_aps_param(cc_alf_filter_param *cc_alf);
void copy_pixels(kvz_pixel *src, int x_src_start, int y_src_start, int src_stride, void copy_pixels(kvz_pixel *src, int x_src_start, int y_src_start, int src_stride,
kvz_pixel *dst, int x_dst_start, int y_dst_start, int dst_stride, kvz_pixel *dst, int x_dst_start, int y_dst_start, int dst_stride,
@ -468,10 +437,14 @@ double kvz_alf_derive_ctb_alf_enable_flags(encoder_state_t * const state,
const double chroma_weight const double chroma_weight
); );
void kvz_alf_create(videoframe_t *frame, enum kvz_chroma_format chroma_format);
void kvz_alf_enc_create(encoder_state_t * const state); void kvz_alf_enc_create(encoder_state_t * const state);
void kvz_alf_reconstruct(encoder_state_t * const state); void kvz_alf_reconstruct(encoder_state_t * const state);
void kvz_alf_destroy(videoframe_t * const frame);
void kvz_alf_enc_destroy(videoframe_t * const frame); void kvz_alf_enc_destroy(videoframe_t * const frame);
void kvz_alf_encoder(encoder_state_t * const state, void kvz_alf_encoder(encoder_state_t * const state,
@ -491,7 +464,7 @@ void kvz_alf_derive_stats_for_filtering(encoder_state_t * const state);
//mikä on alf_WSSD? //mikä on alf_WSSD?
void kvz_alf_get_blk_stats(encoder_state_t * const state, void kvz_alf_get_blk_stats(encoder_state_t * const state,
channel_type channel, channel_type channel,
alf_covariance **alfCovariace, alf_covariance *alfCovariace,
alf_classifier **g_classifier, alf_classifier **g_classifier,
kvz_pixel *org, kvz_pixel *org,
int32_t org_stride, int32_t org_stride,

20
src/encoder_state-bitstream.c
View file

@ -1086,20 +1086,20 @@ static void kvz_encoder_state_write_bitstream_picture_header(
} }
else else
{ {
state->tile->frame->alf_info->g_ctu_enable_flag[COMPONENT_Y] = true; /*state->tile->frame->ctu_enable_flag[COMPONENT_Y] = true;
state->tile->frame->alf_info->g_ctu_enable_flag[COMPONENT_Cb] = true; state->tile->frame->ctu_enable_flag[COMPONENT_Cb] = true;
state->tile->frame->alf_info->g_ctu_enable_flag[COMPONENT_Cr] = true; state->tile->frame->ctu_enable_flag[COMPONENT_Cr] = true;
state->tile->frame->alf_info->g_alf_cc_enable_flag[COMPONENT_Cb] = encoder->cfg.alf_type == 2; state->tile->frame->alf_cc_enable_flag[COMPONENT_Cb] = encoder->cfg.alf_type == 2 ? 1 : 0;
state->tile->frame->alf_info->g_alf_cc_enable_flag[COMPONENT_Cr] = encoder->cfg.alf_type == 2; state->tile->frame->alf_cc_enable_flag[COMPONENT_Cr] = encoder->cfg.alf_type == 2 ? 1 : 0;*/
} }
} }
else else
{ {
state->tile->frame->alf_info->g_ctu_enable_flag[COMPONENT_Y] = false; /*state->tile->frame->ctu_enable_flag[COMPONENT_Y] = false;
state->tile->frame->alf_info->g_ctu_enable_flag[COMPONENT_Cb] = false; state->tile->frame->ctu_enable_flag[COMPONENT_Cb] = false;
state->tile->frame->alf_info->g_ctu_enable_flag[COMPONENT_Cr] = false; state->tile->frame->ctu_enable_flag[COMPONENT_Cr] = false;
state->tile->frame->alf_info->g_alf_cc_enable_flag[COMPONENT_Cb] = false; state->tile->frame->alf_cc_enable_flag[COMPONENT_Cb] = false;
state->tile->frame->alf_info->g_alf_cc_enable_flag[COMPONENT_Cr] = false; state->tile->frame->alf_cc_enable_flag[COMPONENT_Cr] = false;*/
} }
WRITE_U(stream, state->encoder_control->cfg.tmvp_enable, 1, "pic_temporal_mvp_enabled_flag"); WRITE_U(stream, state->encoder_control->cfg.tmvp_enable, 1, "pic_temporal_mvp_enabled_flag");

2
src/encoder_state-ctors_dtors.c
View file

@ -102,7 +102,7 @@ static int encoder_state_config_tile_init(encoder_state_t * const state,
const int width, const int height, const int width_in_lcu, const int height_in_lcu) { const int width, const int height, const int width_in_lcu, const int height_in_lcu) {
const encoder_control_t * const encoder = state->encoder_control; const encoder_control_t * const encoder = state->encoder_control;
state->tile->frame = kvz_videoframe_alloc(width, height, state->encoder_control->chroma_format); state->tile->frame = kvz_videoframe_alloc(width, height, state->encoder_control->chroma_format, encoder->cfg.alf_type);
state->tile->frame->rec = NULL; state->tile->frame->rec = NULL;

11
src/videoframe.c
View file

@ -34,7 +34,8 @@
*/ */
videoframe_t * kvz_videoframe_alloc(int32_t width, videoframe_t * kvz_videoframe_alloc(int32_t width,
int32_t height, int32_t height,
enum kvz_chroma_format chroma_format) enum kvz_chroma_format chroma_format,
enum kvz_alf alf_type)
{ {
videoframe_t *frame = calloc(1, sizeof(videoframe_t)); videoframe_t *frame = calloc(1, sizeof(videoframe_t));
if (!frame) return 0; if (!frame) return 0;
@ -49,7 +50,11 @@ videoframe_t * kvz_videoframe_alloc(int32_t width,
frame->sao_chroma = MALLOC(sao_info_t, frame->width_in_lcu * frame->height_in_lcu); frame->sao_chroma = MALLOC(sao_info_t, frame->width_in_lcu * frame->height_in_lcu);
} }
if (alf_type) {
frame->alf_info = MALLOC(alf_info_t, 1); frame->alf_info = MALLOC(alf_info_t, 1);
kvz_alf_create(frame, chroma_format);
}
return frame; return frame;
} }
@ -71,8 +76,12 @@ int kvz_videoframe_free(videoframe_t * const frame)
FREE_POINTER(frame->sao_luma); FREE_POINTER(frame->sao_luma);
FREE_POINTER(frame->sao_chroma); FREE_POINTER(frame->sao_chroma);
if (frame->alf_info)
{
kvz_alf_destroy(frame);
kvz_alf_enc_destroy(frame); kvz_alf_enc_destroy(frame);
FREE_POINTER(frame->alf_info); FREE_POINTER(frame->alf_info);
}
free(frame); free(frame);

2
src/videoframe.h
View file

@ -52,7 +52,7 @@ typedef struct videoframe
} videoframe_t; } videoframe_t;
videoframe_t *kvz_videoframe_alloc(int32_t width, int32_t height, enum kvz_chroma_format chroma_format); videoframe_t *kvz_videoframe_alloc(int32_t width, int32_t height, enum kvz_chroma_format chroma_format, enum kvz_alf alf_type);
int kvz_videoframe_free(videoframe_t * const frame); int kvz_videoframe_free(videoframe_t * const frame);
void kvz_videoframe_set_poc(videoframe_t * frame, int32_t poc); void kvz_videoframe_set_poc(videoframe_t * frame, int32_t poc);