blob: dc2353e8371e7bff661f4e10a98979209f333d86 [file] [log] [blame]
/*
* Copyright (c) 2021, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 3-Clause Clear License
* and the Alliance for Open Media Patent License 1.0. If the BSD 3-Clause Clear
* License was not distributed with this source code in the LICENSE file, you
* can obtain it at aomedia.org/license/software-license/bsd-3-c-c/. If the
* Alliance for Open Media Patent License 1.0 was not distributed with this
* source code in the PATENTS file, you can obtain it at
* aomedia.org/license/patent-license/.
*/
#include "av1/encoder/encodetxb.h"
#include "aom_ports/mem.h"
#include "av1/common/blockd.h"
#include "av1/common/idct.h"
#include "av1/common/pred_common.h"
#include "av1/common/scan.h"
#include "av1/common/reconintra.h"
#include "av1/encoder/bitstream.h"
#include "av1/encoder/cost.h"
#include "av1/encoder/encodeframe.h"
#include "av1/encoder/hash.h"
#include "av1/encoder/rdopt.h"
#include "av1/encoder/tokenize.h"
#if CONFIG_PAR_HIDING
typedef struct {
tran_low_t qc;
tran_low_t dqc; // dequantized qc
int64_t delta_cost; // cost change between between coding up and low level
int delta_rate; // rate change between coding up and low level
bool upround; // is quantized into up level
bool tunable; // tunable mark
} coeff_info;
// set rd related information for the coefficient at current position.
void set_coeff_info(tran_low_t qc_low, tran_low_t dqc_low, tran_low_t qc_up,
tran_low_t dqc_up, int64_t cost_low, int64_t cost_up,
int rate_low, int rate_up, bool upround,
coeff_info *coef_info, const int scan_idx) {
if (!scan_idx) {
return;
}
coef_info[scan_idx].upround = upround;
if (upround) {
coef_info[scan_idx].qc = qc_low;
coef_info[scan_idx].dqc = dqc_low;
} else {
coef_info[scan_idx].qc = qc_up;
coef_info[scan_idx].dqc = dqc_up;
}
coef_info[scan_idx].tunable =
(abs(coef_info[scan_idx].qc) < MAX_BASE_BR_RANGE) ||
((abs(qc_up) == MAX_BASE_BR_RANGE) && upround);
if (coef_info[scan_idx].tunable) {
if (upround) {
coef_info[scan_idx].delta_cost = (cost_low - cost_up);
coef_info[scan_idx].delta_rate = (rate_low - rate_up);
} else {
coef_info[scan_idx].delta_cost = (cost_up - cost_low);
coef_info[scan_idx].delta_rate = (rate_up - rate_low);
}
}
}
#endif // CONFIG_PAR_HIDING
typedef struct LevelDownStats {
int update;
tran_low_t low_qc;
tran_low_t low_dqc;
int64_t dist0;
int rate;
int rate_low;
int64_t dist;
int64_t dist_low;
int64_t rd;
int64_t rd_low;
int64_t nz_rd;
int64_t rd_diff;
int cost_diff;
int64_t dist_diff;
int new_eob;
} LevelDownStats;
static INLINE int get_dqv(const int32_t *dequant, int coeff_idx,
const qm_val_t *iqmatrix) {
int dqv = dequant[!!coeff_idx];
if (iqmatrix != NULL)
dqv =
((iqmatrix[coeff_idx] * dqv) + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
return dqv;
}
void av1_alloc_txb_buf(AV1_COMP *cpi) {
AV1_COMMON *cm = &cpi->common;
int size = ((cm->mi_params.mi_rows >> cm->seq_params.mib_size_log2) + 1) *
((cm->mi_params.mi_cols >> cm->seq_params.mib_size_log2) + 1);
av1_free_txb_buf(cpi);
// TODO(jingning): This should be further reduced.
CHECK_MEM_ERROR(cm, cpi->coeff_buffer_base,
aom_memalign(32, sizeof(*cpi->coeff_buffer_base) * size));
}
void av1_free_txb_buf(AV1_COMP *cpi) { aom_free(cpi->coeff_buffer_base); }
static void write_golomb(aom_writer *w, int level) {
int x = level + 1;
int length = 0;
#if CONFIG_BYPASS_IMPROVEMENT
length = get_msb(x) + 1;
assert(length > 0);
aom_write_literal(w, 0, length - 1);
aom_write_literal(w, x, length);
#else
int i = x;
while (i) {
i >>= 1;
++length;
}
assert(length > 0);
for (i = 0; i < length - 1; ++i) aom_write_bit(w, 0);
for (i = length - 1; i >= 0; --i) aom_write_bit(w, (x >> i) & 0x01);
#endif
}
static INLINE int64_t get_coeff_dist(tran_low_t tcoeff, tran_low_t dqcoeff,
int shift) {
const int64_t diff = (tcoeff - dqcoeff) * (1 << shift);
const int64_t error = diff * diff;
return error;
}
static const int8_t eob_to_pos_small[33] = {
0, 1, 2, // 0-2
3, 3, // 3-4
4, 4, 4, 4, // 5-8
5, 5, 5, 5, 5, 5, 5, 5, // 9-16
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6 // 17-32
};
static const int8_t eob_to_pos_large[17] = {
6, // place holder
7, // 33-64
8, 8, // 65-128
9, 9, 9, 9, // 129-256
10, 10, 10, 10, 10, 10, 10, 10, // 257-512
11 // 513-
};
static INLINE int get_eob_pos_token(const int eob, int *const extra) {
int t;
if (eob < 33) {
t = eob_to_pos_small[eob];
} else {
const int e = AOMMIN((eob - 1) >> 5, 16);
t = eob_to_pos_large[e];
}
*extra = eob - av1_eob_group_start[t];
return t;
}
#if CONFIG_ENTROPY_STATS
void av1_update_eob_context(int cdf_idx, int eob, TX_SIZE tx_size,
TX_CLASS tx_class, PLANE_TYPE plane,
FRAME_CONTEXT *ec_ctx, FRAME_COUNTS *counts,
uint8_t allow_update_cdf) {
#else
void av1_update_eob_context(int eob, TX_SIZE tx_size, TX_CLASS tx_class,
PLANE_TYPE plane, FRAME_CONTEXT *ec_ctx,
uint8_t allow_update_cdf) {
#endif
int eob_extra;
const int eob_pt = get_eob_pos_token(eob, &eob_extra);
TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
const int eob_multi_size = txsize_log2_minus4[tx_size];
const int eob_multi_ctx = (tx_class == TX_CLASS_2D) ? 0 : 1;
switch (eob_multi_size) {
case 0:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi16[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf)
update_cdf(ec_ctx->eob_flag_cdf16[plane][eob_multi_ctx], eob_pt - 1, 5);
break;
case 1:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi32[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf)
update_cdf(ec_ctx->eob_flag_cdf32[plane][eob_multi_ctx], eob_pt - 1, 6);
break;
case 2:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi64[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf)
update_cdf(ec_ctx->eob_flag_cdf64[plane][eob_multi_ctx], eob_pt - 1, 7);
break;
case 3:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi128[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf) {
update_cdf(ec_ctx->eob_flag_cdf128[plane][eob_multi_ctx], eob_pt - 1,
8);
}
break;
case 4:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi256[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf) {
update_cdf(ec_ctx->eob_flag_cdf256[plane][eob_multi_ctx], eob_pt - 1,
9);
}
break;
case 5:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi512[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf) {
update_cdf(ec_ctx->eob_flag_cdf512[plane][eob_multi_ctx], eob_pt - 1,
10);
}
break;
case 6:
default:
#if CONFIG_ENTROPY_STATS
++counts->eob_multi1024[cdf_idx][plane][eob_multi_ctx][eob_pt - 1];
#endif
if (allow_update_cdf) {
update_cdf(ec_ctx->eob_flag_cdf1024[plane][eob_multi_ctx], eob_pt - 1,
11);
}
break;
}
if (av1_eob_offset_bits[eob_pt] > 0) {
int eob_ctx = eob_pt - 3;
int eob_shift = av1_eob_offset_bits[eob_pt] - 1;
int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
#if CONFIG_ENTROPY_STATS
counts->eob_extra[cdf_idx][txs_ctx][plane][eob_pt][bit]++;
#endif // CONFIG_ENTROPY_STATS
if (allow_update_cdf)
update_cdf(ec_ctx->eob_extra_cdf[txs_ctx][plane][eob_ctx], bit, 2);
}
}
static int get_eob_cost(int eob, const LV_MAP_EOB_COST *txb_eob_costs,
const LV_MAP_COEFF_COST *txb_costs, TX_CLASS tx_class) {
int eob_extra;
const int eob_pt = get_eob_pos_token(eob, &eob_extra);
int eob_cost = 0;
const int eob_multi_ctx = (tx_class == TX_CLASS_2D) ? 0 : 1;
eob_cost = txb_eob_costs->eob_cost[eob_multi_ctx][eob_pt - 1];
if (av1_eob_offset_bits[eob_pt] > 0) {
const int eob_ctx = eob_pt - 3;
const int eob_shift = av1_eob_offset_bits[eob_pt] - 1;
const int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
eob_cost += txb_costs->eob_extra_cost[eob_ctx][bit];
const int offset_bits = av1_eob_offset_bits[eob_pt];
if (offset_bits > 1) eob_cost += av1_cost_literal(offset_bits - 1);
}
return eob_cost;
}
static const int golomb_bits_cost[32] = {
0, 512, 512 * 3, 512 * 3, 512 * 5, 512 * 5, 512 * 5, 512 * 5,
512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7,
512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9,
512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9
};
static const int golomb_cost_diff[32] = {
0, 512, 512 * 2, 0, 512 * 2, 0, 0, 0, 512 * 2, 0, 0, 0, 0, 0, 0, 0,
512 * 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
};
static INLINE int get_golomb_cost(int abs_qc) {
if (abs_qc >= 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
const int r = abs_qc - COEFF_BASE_RANGE - NUM_BASE_LEVELS;
const int length = get_msb(r) + 1;
return av1_cost_literal(2 * length - 1);
}
return 0;
}
#if CONFIG_ATC_COEFCODING
// Golomb cost of coding bypass coded level values in the
// low-frequency region.
static INLINE int get_golomb_cost_lf(int abs_qc) {
if (abs_qc >= 1 + LF_NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
const int r = abs_qc - COEFF_BASE_RANGE - LF_NUM_BASE_LEVELS;
const int length = get_msb(r) + 1;
return av1_cost_literal(2 * length - 1);
}
return 0;
}
// Base range cost of coding level values in the
// low-frequency region, includes the bypass cost.
static INLINE int get_br_lf_cost(tran_low_t level, const int *coeff_lps) {
const int base_range =
AOMMIN(level - 1 - LF_NUM_BASE_LEVELS, COEFF_BASE_RANGE);
return coeff_lps[base_range] + get_golomb_cost_lf(level);
}
// Calculates differential cost for base range coding in the low-frequency
// region for encoder coefficient level optimization.
static INLINE int get_br_lf_cost_with_diff(tran_low_t level,
const int *coeff_lps, int *diff) {
const int base_range =
AOMMIN(level - 1 - LF_NUM_BASE_LEVELS, COEFF_BASE_RANGE);
int golomb_bits = 0;
if (level <= COEFF_BASE_RANGE + 1 + LF_NUM_BASE_LEVELS)
*diff += coeff_lps[base_range + COEFF_BASE_RANGE + 1];
if (level >= COEFF_BASE_RANGE + 1 + LF_NUM_BASE_LEVELS) {
int r = level - COEFF_BASE_RANGE - LF_NUM_BASE_LEVELS;
if (r < 32) {
golomb_bits = golomb_bits_cost[r];
*diff += golomb_cost_diff[r];
} else {
golomb_bits = get_golomb_cost_lf(level);
*diff += (r & (r - 1)) == 0 ? 1024 : 0;
}
}
return coeff_lps[base_range] + golomb_bits;
}
#endif // CONFIG_ATC_COEFCODING
static INLINE int get_br_cost_with_diff(tran_low_t level, const int *coeff_lps,
int *diff) {
const int base_range = AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
int golomb_bits = 0;
if (level <= COEFF_BASE_RANGE + 1 + NUM_BASE_LEVELS)
*diff += coeff_lps[base_range + COEFF_BASE_RANGE + 1];
if (level >= COEFF_BASE_RANGE + 1 + NUM_BASE_LEVELS) {
int r = level - COEFF_BASE_RANGE - NUM_BASE_LEVELS;
if (r < 32) {
golomb_bits = golomb_bits_cost[r];
*diff += golomb_cost_diff[r];
} else {
golomb_bits = get_golomb_cost(level);
*diff += (r & (r - 1)) == 0 ? 1024 : 0;
}
}
return coeff_lps[base_range] + golomb_bits;
}
static INLINE int get_br_cost(tran_low_t level, const int *coeff_lps) {
const int base_range = AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
return coeff_lps[base_range] + get_golomb_cost(level);
}
static INLINE int get_nz_map_ctx(const uint8_t *const levels,
const int coeff_idx, const int bwl,
const int height, const int scan_idx,
const int is_eob,
#if !CONFIG_ATC_COEFCODING
const TX_SIZE tx_size,
#endif // !CONFIG_ATC_COEFCODING
const TX_CLASS tx_class
#if CONFIG_ATC_COEFCODING
,
const int plane) {
#else
) {
#endif // CONFIG_ATC_COEFCODING
if (is_eob) {
if (scan_idx == 0) return 0;
if (scan_idx <= (height << bwl) / 8) return 1;
if (scan_idx <= (height << bwl) / 4) return 2;
return 3;
}
#if CONFIG_ATC_COEFCODING
int stats = 0;
const int row = coeff_idx >> bwl;
const int col = coeff_idx - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
stats =
get_nz_mag_lf(levels + get_padded_idx(coeff_idx, bwl), bwl, tx_class);
return get_nz_map_ctx_from_stats_lf(stats, coeff_idx, bwl, tx_class);
} else {
stats = get_nz_mag(levels + get_padded_idx(coeff_idx, bwl), bwl, tx_class);
return get_nz_map_ctx_from_stats(stats, coeff_idx, bwl, tx_class);
}
#else
const int stats =
get_nz_mag(levels + get_padded_idx(coeff_idx, bwl), bwl, tx_class);
return get_nz_map_ctx_from_stats(stats, coeff_idx, bwl, tx_size, tx_class);
#endif // CONFIG_ATC_COEFCODING
}
static INLINE int get_nz_map_ctx_skip(const uint8_t *const levels,
const int coeff_idx, const int bwl) {
const int stats =
get_nz_mag_skip(levels + get_padded_idx_left(coeff_idx, bwl), bwl);
return get_nz_map_ctx_from_stats_skip(stats, coeff_idx, bwl);
}
void av1_txb_init_levels_signs_c(const tran_low_t *const coeff, const int width,
const int height, uint8_t *const levels,
int8_t *const signs) {
const int stride = width + TX_PAD_LEFT;
int8_t *si = signs;
uint8_t *ls = levels;
// bottom 4 pad
memset(levels + stride * (height + TX_PAD_TOP), 0,
sizeof(*levels) * (TX_PAD_BOTTOM * stride + TX_PAD_END));
memset(signs + stride * (height + TX_PAD_TOP), 0,
sizeof(*signs) * (TX_PAD_BOTTOM * stride + TX_PAD_END));
// top 4 pad
memset(levels, 0, sizeof(*levels) * (TX_PAD_TOP * stride));
ls += TX_PAD_TOP * stride;
memset(signs, 0, sizeof(*signs) * (TX_PAD_TOP * stride));
si += TX_PAD_TOP * stride;
for (int i = 0; i < height; i++) {
// left 4 pad
for (int j = 0; j < TX_PAD_LEFT; j++) {
*ls++ = 0;
}
for (int j = 0; j < width; j++) {
*si++ = (int8_t)(coeff[i * width + j] > 0) ? 1 : -1;
*ls++ = (uint8_t)clamp(abs(coeff[i * width + j]), 0, INT8_MAX);
}
// right 4 pad
for (int j = 0; j < TX_PAD_RIGHT; j++) {
*si++ = 0;
}
}
}
void av1_txb_init_levels_skip_c(const tran_low_t *const coeff, const int width,
const int height, uint8_t *const levels) {
const int stride = width + TX_PAD_LEFT;
uint8_t *ls = levels;
// bottom 4 padded region
memset(levels + stride * (height + TX_PAD_TOP), 0,
sizeof(*levels) * (TX_PAD_BOTTOM * stride + TX_PAD_END));
// top 4 padded region
memset(levels, 0, sizeof(*levels) * (TX_PAD_TOP * stride));
ls += TX_PAD_TOP * stride;
for (int i = 0; i < height; i++) {
// left 4 padded region for each row
for (int j = 0; j < TX_PAD_LEFT; j++) {
*ls++ = 0;
}
for (int j = 0; j < width; j++) {
*ls++ = (uint8_t)clamp(abs(coeff[i * width + j]), 0, INT8_MAX);
}
}
}
void av1_txb_init_levels_c(const tran_low_t *const coeff, const int width,
const int height, uint8_t *const levels) {
const int stride = width + TX_PAD_HOR;
uint8_t *ls = levels;
memset(levels + stride * height, 0,
sizeof(*levels) * (TX_PAD_BOTTOM * stride + TX_PAD_END));
for (int i = 0; i < height; i++) {
for (int j = 0; j < width; j++) {
*ls++ = (uint8_t)clamp(abs(coeff[i * width + j]), 0, INT8_MAX);
}
for (int j = 0; j < TX_PAD_HOR; j++) {
*ls++ = 0;
}
}
}
void av1_get_nz_map_contexts_c(const uint8_t *const levels,
const int16_t *const scan, const uint16_t eob,
const TX_SIZE tx_size, const TX_CLASS tx_class,
int8_t *const coeff_contexts
#if CONFIG_ATC_COEFCODING
,
const int plane) {
#else
) {
#endif // CONFIG_ATC_COEFCODING
const int bwl = get_txb_bwl(tx_size);
const int height = get_txb_high(tx_size);
for (int i = 0; i < eob; ++i) {
const int pos = scan[i];
coeff_contexts[pos] =
get_nz_map_ctx(levels, pos, bwl, height, i, i == eob - 1,
#if !CONFIG_ATC_COEFCODING
tx_size,
#endif // !CONFIG_ATC_COEFCODING
tx_class
#if CONFIG_ATC_COEFCODING
,
plane);
#else
);
#endif // CONFIG_ATC_COEFCODING
}
}
void av1_get_nz_map_contexts_skip_c(const uint8_t *const levels,
const int16_t *const scan,
const uint16_t eob, const TX_SIZE tx_size,
int8_t *const coeff_contexts) {
const int bwl = get_txb_bwl(tx_size);
for (int i = 0; i < eob; ++i) {
const int pos = scan[i];
coeff_contexts[pos] = get_nz_map_ctx_skip(levels, pos, bwl);
}
}
int av1_write_sig_txtype(const AV1_COMMON *const cm, MACROBLOCK *const x,
aom_writer *w, int blk_row, int blk_col, int plane,
int block, TX_SIZE tx_size) {
MACROBLOCKD *xd = &x->e_mbd;
const CB_COEFF_BUFFER *cb_coef_buff = x->cb_coef_buff;
const int txb_offset =
x->mbmi_ext_frame->cb_offset[plane] / (TX_SIZE_W_MIN * TX_SIZE_H_MIN);
#if CONFIG_CONTEXT_DERIVATION
const int width = get_txb_wide(tx_size);
const int height = get_txb_high(tx_size);
if (plane == AOM_PLANE_U)
memset(xd->tmp_sign, 0, width * height * sizeof(int32_t));
#endif // CONFIG_CONTEXT_DERIVATION
const uint16_t *eob_txb = cb_coef_buff->eobs[plane] + txb_offset;
const uint16_t eob = eob_txb[block];
const uint8_t *entropy_ctx = cb_coef_buff->entropy_ctx[plane] + txb_offset;
#if CONFIG_CONTEXT_DERIVATION
int txb_skip_ctx = (entropy_ctx[block] & TXB_SKIP_CTX_MASK);
if (plane == AOM_PLANE_V) {
txb_skip_ctx += (xd->eob_u_flag ? V_TXB_SKIP_CONTEXT_OFFSET : 0);
}
#else
const int txb_skip_ctx = entropy_ctx[block] & TXB_SKIP_CTX_MASK;
#endif // CONFIG_CONTEXT_DERIVATION
const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
const PLANE_TYPE plane_type = get_plane_type(plane);
const TX_TYPE tx_type =
av1_get_tx_type(xd, plane_type, blk_row, blk_col, tx_size,
cm->features.reduced_tx_set_used);
#if CONFIG_CROSS_CHROMA_TX && CCTX_C2_DROPPED
if (plane == AOM_PLANE_V && is_cctx_allowed(cm, xd)) {
CctxType cctx_type = av1_get_cctx_type(xd, blk_row, blk_col);
if (!keep_chroma_c2(cctx_type)) return 0;
}
#endif // CONFIG_CROSS_CHROMA_TX && CCTX_C2_DROPPED
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_U) {
xd->eob_u_flag = eob ? 1 : 0;
}
if (plane == AOM_PLANE_Y || plane == AOM_PLANE_U) {
aom_write_symbol(w, eob == 0, ec_ctx->txb_skip_cdf[txs_ctx][txb_skip_ctx],
2);
} else {
aom_write_symbol(w, eob == 0, ec_ctx->v_txb_skip_cdf[txb_skip_ctx], 2);
}
#else
aom_write_symbol(w, eob == 0, ec_ctx->txb_skip_cdf[txs_ctx][txb_skip_ctx], 2);
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CROSS_CHROMA_TX
if (plane == AOM_PLANE_U && is_cctx_allowed(cm, xd)) {
CctxType cctx_type = av1_get_cctx_type(xd, blk_row, blk_col);
if (eob > 0) av1_write_cctx_type(cm, xd, cctx_type, tx_size, w);
}
#endif // CONFIG_CROSS_CHROMA_TX
if (eob == 0) return 0;
if (plane == 0) { // Only y plane's tx_type is transmitted
av1_write_tx_type(cm, xd, tx_type, tx_size, w);
}
return 1;
}
void av1_write_coeffs_txb_skip(const AV1_COMMON *const cm, MACROBLOCK *const x,
aom_writer *w, int blk_row, int blk_col,
int plane, int block, TX_SIZE tx_size) {
MACROBLOCKD *xd = &x->e_mbd;
const CB_COEFF_BUFFER *cb_coef_buff = x->cb_coef_buff;
const uint16_t eob = av1_get_max_eob(tx_size);
FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
const PLANE_TYPE plane_type = get_plane_type(plane);
const TX_TYPE tx_type =
av1_get_tx_type(xd, plane_type, blk_row, blk_col, tx_size,
cm->features.reduced_tx_set_used);
const int width = get_txb_wide(tx_size);
const int height = get_txb_high(tx_size);
uint8_t levels_buf[TX_PAD_2D];
int8_t signs_buf[TX_PAD_2D];
const tran_low_t *tcoeff_txb =
cb_coef_buff->tcoeff[plane] + x->mbmi_ext_frame->cb_offset[plane];
const tran_low_t *tcoeff = tcoeff_txb + BLOCK_OFFSET(block);
av1_txb_init_levels_signs(tcoeff, width, height, levels_buf, signs_buf);
uint8_t *const levels = set_levels(levels_buf, width);
int8_t *const signs = set_signs(signs_buf, width);
const SCAN_ORDER *const scan_order = get_scan(tx_size, tx_type);
const int16_t *const scan = scan_order->scan;
const int bwl = get_txb_bwl(tx_size);
DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);
av1_get_nz_map_contexts_skip(levels, scan, eob, tx_size, coeff_contexts);
for (int c = 0; c < eob; ++c) {
const int pos = scan[c];
const int coeff_ctx = coeff_contexts[pos];
const tran_low_t v = tcoeff[pos];
const tran_low_t level = abs(v);
aom_write_symbol(w, AOMMIN(level, 3),
ec_ctx->coeff_base_cdf_idtx[coeff_ctx], 4);
if (level > NUM_BASE_LEVELS) {
// level is above 1.
const int base_range = level - 1 - NUM_BASE_LEVELS;
const int br_ctx = get_br_ctx_skip(levels, pos, bwl);
aom_cdf_prob *cdf = ec_ctx->coeff_br_cdf_idtx[br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
}
// Loop to code all signs, bypass levels in the transform block
for (int c = eob - 1; c >= 0; --c) {
const int pos = scan[c];
const tran_low_t v = tcoeff[pos];
const tran_low_t level = abs(v);
const int sign = (v < 0) ? 1 : 0;
if (level) {
int idtx_sign_ctx = get_sign_ctx_skip(signs, levels, pos, bwl);
aom_write_symbol(w, sign, ec_ctx->idtx_sign_cdf[idtx_sign_ctx], 2);
if (level > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
}
}
}
#if CONFIG_PAR_HIDING
static INLINE void write_coeff_hidden(aom_writer *w, TX_CLASS tx_class,
const int16_t *scan, int bwl,
uint8_t *levels, const int level,
base_cdf_arr base_cdf_ph,
br_cdf_arr br_cdf_ph) {
const int q_index = (level >> 1);
const int pos = scan[0];
int ctx_id = get_base_ctx_ph(levels, pos, bwl, tx_class);
aom_write_symbol(w, AOMMIN(q_index, 3), base_cdf_ph[ctx_id], 4);
if (q_index > NUM_BASE_LEVELS) {
ctx_id = get_par_br_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf_br = br_cdf_ph[ctx_id];
const int base_range = q_index - 1 - NUM_BASE_LEVELS;
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf_br, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
}
#endif // CONFIG_PAR_HIDING
void av1_write_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCK *const x,
aom_writer *w, int blk_row, int blk_col, int plane,
int block, TX_SIZE tx_size) {
MACROBLOCKD *xd = &x->e_mbd;
const CB_COEFF_BUFFER *cb_coef_buff = x->cb_coef_buff;
#if CONFIG_CONTEXT_DERIVATION
const int width = get_txb_wide(tx_size);
const int height = get_txb_high(tx_size);
if (plane == AOM_PLANE_U)
memset(xd->tmp_sign, 0, width * height * sizeof(int32_t));
#endif // CONFIG_CONTEXT_DERIVATION
const int txb_offset =
x->mbmi_ext_frame->cb_offset[plane] / (TX_SIZE_W_MIN * TX_SIZE_H_MIN);
const uint16_t *eob_txb = cb_coef_buff->eobs[plane] + txb_offset;
const uint16_t eob = eob_txb[block];
const uint8_t *entropy_ctx = cb_coef_buff->entropy_ctx[plane] + txb_offset;
const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
#if CONFIG_PC_WIENER
if (!is_global_intrabc_allowed(cm) && !cm->features.coded_lossless) {
// Assert only when LR is enabled.
assert((eob == 0) == av1_get_txk_skip(cm, xd->mi_row, xd->mi_col, plane,
blk_row, blk_col));
}
#endif // CONFIG_PC_WIENER
if (eob == 0) return;
const PLANE_TYPE plane_type = get_plane_type(plane);
const TX_TYPE tx_type =
av1_get_tx_type(xd, plane_type, blk_row, blk_col, tx_size,
cm->features.reduced_tx_set_used);
#if DEBUG_EXTQUANT
fprintf(cm->fEncCoeffLog, "\nblk_row=%d,blk_col=%d,plane=%d,tx_size=%d",
blk_row, blk_col, plane, tx_size);
#endif
int eob_extra;
const int eob_pt = get_eob_pos_token(eob, &eob_extra);
const int eob_multi_size = txsize_log2_minus4[tx_size];
const TX_CLASS tx_class = tx_type_to_class[get_primary_tx_type(tx_type)];
const int eob_multi_ctx = (tx_class == TX_CLASS_2D) ? 0 : 1;
switch (eob_multi_size) {
case 0:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf16[plane_type][eob_multi_ctx], 5);
break;
case 1:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf32[plane_type][eob_multi_ctx], 6);
break;
case 2:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf64[plane_type][eob_multi_ctx], 7);
break;
case 3:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf128[plane_type][eob_multi_ctx], 8);
break;
case 4:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf256[plane_type][eob_multi_ctx], 9);
break;
case 5:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf512[plane_type][eob_multi_ctx], 10);
break;
default:
aom_write_symbol(w, eob_pt - 1,
ec_ctx->eob_flag_cdf1024[plane_type][eob_multi_ctx], 11);
break;
}
const int eob_offset_bits = av1_eob_offset_bits[eob_pt];
if (eob_offset_bits > 0) {
const int eob_ctx = eob_pt - 3;
int eob_shift = eob_offset_bits - 1;
int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
aom_write_symbol(w, bit,
ec_ctx->eob_extra_cdf[txs_ctx][plane_type][eob_ctx], 2);
#if CONFIG_BYPASS_IMPROVEMENT
// Zero out top bit; write (eob_offset_bits - 1) lsb bits.
eob_extra &= (1 << (eob_offset_bits - 1)) - 1;
aom_write_literal(w, eob_extra, eob_offset_bits - 1);
#else
for (int i = 1; i < eob_offset_bits; i++) {
eob_shift = eob_offset_bits - 1 - i;
bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
aom_write_bit(w, bit);
}
#endif
}
// write sec_tx_type here
// Only y plane's sec_tx_type is transmitted
if ((plane == AOM_PLANE_Y) && (cm->seq_params.enable_ist)) {
av1_write_sec_tx_type(cm, xd, tx_type, tx_size, eob, w);
}
#if DEBUG_EXTQUANT
fprintf(cm->fEncCoeffLog, "tx_type=%d, eob=%d\n", tx_type, eob);
#endif
#if !CONFIG_CONTEXT_DERIVATION
const int width = get_txb_wide(tx_size);
const int height = get_txb_high(tx_size);
#endif // CONFIG_CONTEXT_DERIVATION
uint8_t levels_buf[TX_PAD_2D];
uint8_t *const levels = set_levels(levels_buf, width);
const tran_low_t *tcoeff_txb =
cb_coef_buff->tcoeff[plane] + x->mbmi_ext_frame->cb_offset[plane];
const tran_low_t *tcoeff = tcoeff_txb + BLOCK_OFFSET(block);
av1_txb_init_levels(tcoeff, width, height, levels);
const SCAN_ORDER *const scan_order = get_scan(tx_size, tx_type);
const int16_t *const scan = scan_order->scan;
DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);
av1_get_nz_map_contexts(levels, scan, eob, tx_size, tx_class, coeff_contexts
#if CONFIG_ATC_COEFCODING
,
plane
#endif // CONFIG_ATC_COEFCODING
);
const int bwl = get_txb_bwl(tx_size);
#if CONFIG_PAR_HIDING
bool enable_parity_hiding = cm->features.allow_parity_hiding &&
!xd->lossless[xd->mi[0]->segment_id] &&
plane == PLANE_TYPE_Y &&
get_primary_tx_type(tx_type) < IDTX;
for (int c = eob - 1; c > 0; --c) {
#else
for (int c = eob - 1; c >= 0; --c) {
#endif // CONFIG_PAR_HIDING
const int pos = scan[c];
const int coeff_ctx = coeff_contexts[pos];
const tran_low_t v = tcoeff[pos];
const tran_low_t level = abs(v);
#if CONFIG_ATC_COEFCODING
if (c == eob - 1) {
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
aom_write_symbol(
w, AOMMIN(level, LF_BASE_SYMBOLS - 1) - 1,
ec_ctx->coeff_base_lf_eob_cdf[txs_ctx][plane_type][coeff_ctx],
LF_BASE_SYMBOLS - 1);
} else {
aom_write_symbol(
w, AOMMIN(level, 3) - 1,
ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx], 3);
}
} else {
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
aom_write_symbol(
w, AOMMIN(level, LF_BASE_SYMBOLS - 1),
ec_ctx->coeff_base_lf_cdf[txs_ctx][plane_type][coeff_ctx],
LF_BASE_SYMBOLS);
} else {
aom_write_symbol(w, AOMMIN(level, 3),
ec_ctx->coeff_base_cdf[txs_ctx][plane_type][coeff_ctx],
4);
}
#else
if (c == eob - 1) {
aom_write_symbol(
w, AOMMIN(level, 3) - 1,
ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx], 3);
} else {
aom_write_symbol(w, AOMMIN(level, 3),
ec_ctx->coeff_base_cdf[txs_ctx][plane_type][coeff_ctx],
4);
#endif // CONFIG_ATC_COEFCODING
}
#if CONFIG_ATC_COEFCODING
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
if (level > LF_NUM_BASE_LEVELS) {
const int base_range =
level - 1 - LF_NUM_BASE_LEVELS; // level is above 1.
const int br_ctx = get_br_lf_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf = ec_ctx->coeff_br_lf_cdf[plane_type][br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
} else {
if (level > NUM_BASE_LEVELS) {
const int base_range = level - 1 - NUM_BASE_LEVELS;
const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf = ec_ctx->coeff_br_cdf[plane_type][br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
}
#else
if (level > NUM_BASE_LEVELS) {
// level is above 1.
const int base_range = level - 1 - NUM_BASE_LEVELS;
const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf =
ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
#endif // CONFIG_ATC_COEFCODING
}
#if CONFIG_PAR_HIDING
int num_nz = 0;
bool is_hidden = false;
if (enable_parity_hiding) {
for (int c = eob - 1; c > 0; --c) {
const int pos = scan[c];
num_nz += !!tcoeff[pos];
}
is_hidden = num_nz >= PHTHRESH;
}
if (is_hidden) {
const int pos = scan[0];
const tran_low_t v = tcoeff[pos];
const tran_low_t level = abs(v);
write_coeff_hidden(w, tx_class, scan, bwl, levels, level,
ec_ctx->coeff_base_ph_cdf, ec_ctx->coeff_br_ph_cdf);
} else {
const int c = 0;
const int pos = scan[c];
const int coeff_ctx = coeff_contexts[pos];
const tran_low_t v = tcoeff[pos];
const tran_low_t level = abs(v);
#if CONFIG_ATC_COEFCODING
if (c == eob - 1) {
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
aom_write_symbol(
w, AOMMIN(level, LF_BASE_SYMBOLS - 1) - 1,
ec_ctx->coeff_base_lf_eob_cdf[txs_ctx][plane_type][coeff_ctx],
LF_BASE_SYMBOLS - 1);
} else {
aom_write_symbol(
w, AOMMIN(level, 3) - 1,
ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx], 3);
}
} else {
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
aom_write_symbol(
w, AOMMIN(level, LF_BASE_SYMBOLS - 1),
ec_ctx->coeff_base_lf_cdf[txs_ctx][plane_type][coeff_ctx],
LF_BASE_SYMBOLS);
} else {
aom_write_symbol(w, AOMMIN(level, 3),
ec_ctx->coeff_base_cdf[txs_ctx][plane_type][coeff_ctx],
4);
}
#else
if (c == eob - 1) {
aom_write_symbol(
w, AOMMIN(level, 3) - 1,
ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx], 3);
} else {
aom_write_symbol(w, AOMMIN(level, 3),
ec_ctx->coeff_base_cdf[txs_ctx][plane_type][coeff_ctx],
4);
#endif // CONFIG_ATC_COEFCODING
}
#if CONFIG_ATC_COEFCODING
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
if (level > LF_NUM_BASE_LEVELS) {
const int base_range =
level - 1 - LF_NUM_BASE_LEVELS; // level is above 1.
const int br_ctx = get_br_lf_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf = ec_ctx->coeff_br_lf_cdf[plane_type][br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
} else {
if (level > NUM_BASE_LEVELS) {
const int base_range = level - 1 - NUM_BASE_LEVELS;
const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf = ec_ctx->coeff_br_cdf[plane_type][br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
}
#else
if (level > NUM_BASE_LEVELS) {
// level is above 1.
const int base_range = level - 1 - NUM_BASE_LEVELS;
const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
aom_cdf_prob *cdf =
ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx];
for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
if (k < BR_CDF_SIZE - 1) break;
}
}
#endif // CONFIG_ATC_COEFCODING
}
#endif // CONFIG_PAR_HIDING
#if DEBUG_EXTQUANT
for (int c = 0; c < eob; ++c) {
const tran_low_t v = tcoeff[scan[c]];
const tran_low_t level = abs(v);
fprintf(cm->fEncCoeffLog, "c=%d,pos=%d,level=%d,dq_coeff=%d\n", c, scan[c],
level, v);
}
#endif
// Loop to code all signs in the transform block,
// starting with the sign of DC (if applicable)
for (int c = 0; c < eob; ++c) {
const tran_low_t v = tcoeff[scan[c]];
const tran_low_t level = abs(v);
const int sign = (v < 0) ? 1 : 0;
if (level) {
if (c == 0) {
const int dc_sign_ctx =
(entropy_ctx[block] >> DC_SIGN_CTX_SHIFT) & DC_SIGN_CTX_MASK;
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_U) xd->tmp_sign[0] = (sign ? 2 : 1);
if (plane == AOM_PLANE_V) {
aom_write_symbol(
w, sign, ec_ctx->v_dc_sign_cdf[xd->tmp_sign[0]][dc_sign_ctx], 2);
} else {
aom_write_symbol(w, sign,
ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx], 2);
}
#else
aom_write_symbol(w, sign, ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx],
2);
#endif // CONFIG_CONTEXT_DERIVATION
} else {
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_U) xd->tmp_sign[scan[c]] = (sign ? 2 : 1);
if (plane == AOM_PLANE_Y || plane == AOM_PLANE_U)
aom_write_bit(w, sign);
else
aom_write_symbol(w, sign,
ec_ctx->v_ac_sign_cdf[xd->tmp_sign[scan[c]]], 2);
#else
aom_write_bit(w, sign);
#endif // CONFIG_CONTEXT_DERIVATION
}
#if CONFIG_PAR_HIDING
#if CONFIG_ATC_COEFCODING
if (is_hidden && c == 0) {
int q_index = level >> 1;
if (q_index > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, q_index - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
} else {
const int pos = scan[c];
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
if (level > COEFF_BASE_RANGE + LF_NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - LF_NUM_BASE_LEVELS);
} else {
if (level > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
}
}
#else
if (is_hidden && c == 0) {
int q_index = level >> 1;
if (q_index > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, q_index - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
} else {
if (level > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
}
#endif // CONFIG_ATC_COEFCODING
#else
#if CONFIG_ATC_COEFCODING
const int pos = scan[c];
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
if (level > COEFF_BASE_RANGE + LF_NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - LF_NUM_BASE_LEVELS);
} else {
if (level > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
}
#else
if (level > COEFF_BASE_RANGE + NUM_BASE_LEVELS)
write_golomb(w, level - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
#endif // CONFIG_ATC_COEFCODING
#endif // CONFIG_PAR_HIDING
}
}
}
typedef struct encode_txb_args {
const AV1_COMMON *cm;
MACROBLOCK *x;
aom_writer *w;
} ENCODE_TXB_ARGS;
void av1_write_intra_coeffs_mb(const AV1_COMMON *const cm, MACROBLOCK *x,
aom_writer *w, BLOCK_SIZE bsize) {
MACROBLOCKD *xd = &x->e_mbd;
const MB_MODE_INFO *const mbmi = xd->mi[0];
const int is_inter = is_inter_block(mbmi, xd->tree_type);
int block[MAX_MB_PLANE] = { 0 };
int row, col;
assert(bsize == get_plane_block_size(bsize, xd->plane[0].subsampling_x,
xd->plane[0].subsampling_y));
const int max_blocks_wide = max_block_wide(xd, bsize, 0);
const int max_blocks_high = max_block_high(xd, bsize, 0);
const BLOCK_SIZE max_unit_bsize = BLOCK_64X64;
int mu_blocks_wide = mi_size_wide[max_unit_bsize];
int mu_blocks_high = mi_size_high[max_unit_bsize];
mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);
for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
const int plane_start = get_partition_plane_start(xd->tree_type);
const int plane_end =
get_partition_plane_end(xd->tree_type, av1_num_planes(cm));
for (int plane = plane_start; plane < plane_end; ++plane) {
if (plane && !xd->is_chroma_ref) break;
#if CONFIG_CROSS_CHROMA_TX
if (plane == AOM_PLANE_U && is_cctx_allowed(cm, xd)) continue;
#endif // CONFIG_CROSS_CHROMA_TX
const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
const int stepr = tx_size_high_unit[tx_size];
const int stepc = tx_size_wide_unit[tx_size];
const int step = stepr * stepc;
const struct macroblockd_plane *const pd = &xd->plane[plane];
const int ss_x = pd->subsampling_x;
const int ss_y = pd->subsampling_y;
const BLOCK_SIZE plane_bsize =
get_mb_plane_block_size(xd, mbmi, plane, ss_x, ss_y);
const int plane_unit_height =
get_plane_tx_unit_height(xd, plane_bsize, plane, row, ss_y);
const int plane_unit_width =
get_plane_tx_unit_width(xd, plane_bsize, plane, col, ss_x);
for (int blk_row = row >> ss_y; blk_row < plane_unit_height;
blk_row += stepr) {
for (int blk_col = col >> ss_x; blk_col < plane_unit_width;
blk_col += stepc) {
#if CONFIG_CROSS_CHROMA_TX
// Loop order for the two chroma planes is changed for CCTX
// because the transform information for both planes are needed at
// once at the decoder side.
if (plane == AOM_PLANE_V && is_cctx_allowed(cm, xd)) {
const int code_rest =
av1_write_sig_txtype(cm, x, w, blk_row, blk_col, AOM_PLANE_U,
block[AOM_PLANE_U], tx_size);
if (code_rest)
av1_write_coeffs_txb(cm, x, w, blk_row, blk_col, AOM_PLANE_U,
block[AOM_PLANE_U], tx_size);
block[AOM_PLANE_U] += step;
}
#endif // CONFIG_CROSS_CHROMA_TX
const int code_rest = av1_write_sig_txtype(
cm, x, w, blk_row, blk_col, plane, block[plane], tx_size);
const TX_TYPE tx_type =
av1_get_tx_type(xd, get_plane_type(plane), blk_row, blk_col,
tx_size, cm->features.reduced_tx_set_used);
if (code_rest) {
if ((mbmi->fsc_mode[xd->tree_type == CHROMA_PART] &&
get_primary_tx_type(tx_type) == IDTX &&
plane == PLANE_TYPE_Y) ||
use_inter_fsc(cm, plane, tx_type, is_inter)) {
av1_write_coeffs_txb_skip(cm, x, w, blk_row, blk_col, plane,
block[plane], tx_size);
} else {
av1_write_coeffs_txb(cm, x, w, blk_row, blk_col, plane,
block[plane], tx_size);
}
}
block[plane] += step;
}
}
}
}
}
}
#if CONFIG_CROSS_CHROMA_TX
int get_cctx_type_cost(const AV1_COMMON *cm, const MACROBLOCK *x,
const MACROBLOCKD *xd, int plane, TX_SIZE tx_size,
int block, CctxType cctx_type) {
if (plane == AOM_PLANE_U && x->plane[plane].eobs[block] &&
is_cctx_allowed(cm, xd)) {
const TX_SIZE square_tx_size = txsize_sqr_map[tx_size];
int above_cctx, left_cctx;
#if CONFIG_EXT_RECUR_PARTITIONS
get_above_and_left_cctx_type(cm, xd, &above_cctx, &left_cctx);
#else
get_above_and_left_cctx_type(cm, xd, tx_size, &above_cctx, &left_cctx);
#endif // CONFIG_EXT_RECUR_PARTITIONS
const int cctx_ctx = get_cctx_context(xd, &above_cctx, &left_cctx);
return x->mode_costs.cctx_type_cost[square_tx_size][cctx_ctx][cctx_type];
} else {
return 0;
}
}
#endif // CONFIG_CROSS_CHROMA_TX
// TODO(angiebird): use this function whenever it's possible
static int get_tx_type_cost(const MACROBLOCK *x, const MACROBLOCKD *xd,
int plane, TX_SIZE tx_size, TX_TYPE tx_type,
int reduced_tx_set_used, int eob) {
if (plane > 0) return 0;
const TX_SIZE square_tx_size = txsize_sqr_map[tx_size];
const MB_MODE_INFO *mbmi = xd->mi[0];
if (mbmi->fsc_mode[xd->tree_type == CHROMA_PART] &&
!is_inter_block(mbmi, xd->tree_type) && plane == PLANE_TYPE_Y) {
return 0;
}
const int is_inter = is_inter_block(mbmi, xd->tree_type);
if (get_ext_tx_types(tx_size, is_inter, reduced_tx_set_used) > 1 &&
!xd->lossless[xd->mi[0]->segment_id]) {
const int ext_tx_set =
get_ext_tx_set(tx_size, is_inter, reduced_tx_set_used);
if (is_inter) {
if (ext_tx_set > 0)
return x->mode_costs
.inter_tx_type_costs[ext_tx_set][square_tx_size][tx_type];
} else {
if (ext_tx_set > 0) {
PREDICTION_MODE intra_dir;
if (mbmi->filter_intra_mode_info.use_filter_intra)
intra_dir = fimode_to_intradir[mbmi->filter_intra_mode_info
.filter_intra_mode];
else
intra_dir = mbmi->mode;
TX_TYPE primary_tx_type = get_primary_tx_type(tx_type);
int tx_type_cost =
x->mode_costs.intra_tx_type_costs[ext_tx_set][square_tx_size]
[intra_dir][primary_tx_type];
if (block_signals_sec_tx_type(xd, tx_size, tx_type, eob) &&
xd->enable_ist) {
tx_type_cost +=
x->mode_costs.stx_flag_cost[square_tx_size]
[get_secondary_tx_type(tx_type)];
}
return tx_type_cost;
}
}
} else if (!is_inter && !xd->lossless[xd->mi[0]->segment_id]) {
if (block_signals_sec_tx_type(xd, tx_size, tx_type, eob) &&
xd->enable_ist) {
return x->mode_costs
.stx_flag_cost[square_tx_size][get_secondary_tx_type(tx_type)];
}
}
return 0;
}
static INLINE void update_coeff_eob_fast(int *eob, int shift,
const int32_t *dequant_ptr,
const int16_t *scan,
const tran_low_t *coeff_ptr,
tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr) {
// TODO(sarahparker) make this work for aomqm
int eob_out = *eob;
int zbin[2] = { dequant_ptr[0] + ROUND_POWER_OF_TWO(dequant_ptr[0] * 70,
7 + QUANT_TABLE_BITS),
dequant_ptr[1] + ROUND_POWER_OF_TWO(dequant_ptr[1] * 70,
7 + QUANT_TABLE_BITS) };
for (int i = *eob - 1; i >= 0; i--) {
const int rc = scan[i];
const int qcoeff = qcoeff_ptr[rc];
const int coeff = coeff_ptr[rc];
const int coeff_sign = AOMSIGN(coeff);
int64_t abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
if (((abs_coeff << (1 + shift)) < zbin[rc != 0]) || (qcoeff == 0)) {
eob_out--;
qcoeff_ptr[rc] = 0;
dqcoeff_ptr[rc] = 0;
} else {
break;
}
}
*eob = eob_out;
}
static AOM_FORCE_INLINE int warehouse_efficients_txb_skip(
#if CONFIG_CROSS_CHROMA_TX
const AV1_COMMON *cm,
#endif // CONFIG_CROSS_CHROMA_TX
const MACROBLOCK *x, const int plane, const int block,
const TX_SIZE tx_size, const TXB_CTX *const txb_ctx,
const struct macroblock_plane *p, const int eob,
const LV_MAP_COEFF_COST *const coeff_costs, const MACROBLOCKD *const xd,
const TX_TYPE tx_type,
#if CONFIG_CROSS_CHROMA_TX
const CctxType cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
int reduced_tx_set_used) {
const tran_low_t *const qcoeff = p->qcoeff + BLOCK_OFFSET(block);
const int txb_skip_ctx = txb_ctx->txb_skip_ctx;
const int bwl = get_txb_bwl(tx_size);
const int width = get_txb_wide(tx_size);
const int height = get_txb_high(tx_size);
const SCAN_ORDER *const scan_order = get_scan(tx_size, tx_type);
const int16_t *const scan = scan_order->scan;
uint8_t levels_buf[TX_PAD_2D];
uint8_t *const levels = set_levels(levels_buf, width);
int cost = coeff_costs->txb_skip_cost[txb_skip_ctx][0];
int8_t signs_buf[TX_PAD_2D];
int8_t *const signs = set_signs(signs_buf, width);
av1_txb_init_levels_signs(qcoeff, width, height, levels_buf, signs_buf);
cost += get_tx_type_cost(x, xd, plane, tx_size, tx_type, reduced_tx_set_used,
eob);
#if CONFIG_CROSS_CHROMA_TX
cost += get_cctx_type_cost(cm, x, xd, plane, tx_size, block, cctx_type);
#endif // CONFIG_CROSS_CHROMA_TX
DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);
av1_get_nz_map_contexts_skip(levels, scan, eob, tx_size, coeff_contexts);
const int(*lps_cost)[COEFF_BASE_RANGE + 1 + COEFF_BASE_RANGE + 1] =
coeff_costs->lps_cost_skip;
const int(*base_cost)[8] = coeff_costs->idtx_base_cost;
for (int c = 0; c < eob; c++) {
const int pos = scan[c];
const int coeff_ctx = coeff_contexts[pos];
const tran_low_t v = qcoeff[pos];
const int level = abs(v);
cost += base_cost[coeff_ctx][AOMMIN(level, 3)];
if (v) {
if (level > NUM_BASE_LEVELS) {
const int ctx = get_br_ctx_skip(levels, pos, bwl);
cost += get_br_cost(level, lps_cost[ctx]);
}
}
}
for (int c = eob - 1; c >= 0; --c) {
const int pos = scan[c];
const tran_low_t v = qcoeff[pos];
const tran_low_t level = abs(v);
const int sign = (v < 0) ? 1 : 0;
if (level) {
int idtx_sign_ctx = get_sign_ctx_skip(signs, levels, pos, bwl);
cost += coeff_costs->idtx_sign_cost[idtx_sign_ctx][sign];
}
}
return cost;
}
static AOM_FORCE_INLINE int warehouse_efficients_txb(
#if CONFIG_CROSS_CHROMA_TX
const AV1_COMMON *cm,
#endif // CONFIG_CROSS_CHROMA_TX
const MACROBLOCK *x, const int plane, const int block,
const TX_SIZE tx_size, const TXB_CTX *const txb_ctx,
const struct macroblock_plane *p, const int eob,
const PLANE_TYPE plane_type, const LV_MAP_COEFF_COST *const coeff_costs,
const MACROBLOCKD *const xd, const TX_TYPE tx_type,
#if CONFIG_CROSS_CHROMA_TX
const CctxType cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
const TX_CLASS tx_class, int reduced_tx_set_used
#if CONFIG_PAR_HIDING
,
bool enable_parity_hiding, const LV_MAP_COEFF_COST *const coeff_costs_ph
#endif
) {
const tran_low_t *const qcoeff = p->qcoeff + BLOCK_OFFSET(block);
#if CONFIG_CONTEXT_DERIVATION
const struct macroblock_plane *pu = &x->plane[AOM_PLANE_U];
int txb_skip_ctx = txb_ctx->txb_skip_ctx;
if (plane == AOM_PLANE_V) {
txb_skip_ctx += (pu->eobs[block] ? V_TXB_SKIP_CONTEXT_OFFSET : 0);
}
#else
const int txb_skip_ctx = txb_ctx->txb_skip_ctx;
#endif // CONFIG_CONTEXT_DERIVATION
const int bwl = get_txb_bwl(tx_size);
const int width = get_txb_wide(tx_size);
const int height = get_txb_high(tx_size);
const SCAN_ORDER *const scan_order = get_scan(tx_size, tx_type);
const int16_t *const scan = scan_order->scan;
uint8_t levels_buf[TX_PAD_2D];
uint8_t *const levels = set_levels(levels_buf, width);
DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);
const int eob_multi_size = txsize_log2_minus4[tx_size];
const LV_MAP_EOB_COST *const eob_costs =
&x->coeff_costs.eob_costs[eob_multi_size][plane_type];
#if CONFIG_CONTEXT_DERIVATION
int cost;
if (plane == AOM_PLANE_V) {
cost = coeff_costs->v_txb_skip_cost[txb_skip_ctx][0];
} else {
cost = coeff_costs->txb_skip_cost[txb_skip_ctx][0];
}
#else
int cost = coeff_costs->txb_skip_cost[txb_skip_ctx][0];
#endif // CONFIG_CONTEXT_DERIVATION
av1_txb_init_levels(qcoeff, width, height, levels);
cost += get_tx_type_cost(x, xd, plane, tx_size, tx_type, reduced_tx_set_used,
eob);
#if CONFIG_CROSS_CHROMA_TX
cost += get_cctx_type_cost(cm, x, xd, plane, tx_size, block, cctx_type);
#endif // CONFIG_CROSS_CHROMA_TX
cost += get_eob_cost(eob, eob_costs, coeff_costs, tx_class);
av1_get_nz_map_contexts(levels, scan, eob, tx_size, tx_class, coeff_contexts
#if CONFIG_ATC_COEFCODING
,
plane
#endif // CONFIG_ATC_COEFCODING
);
const int(*lps_cost)[COEFF_BASE_RANGE + 1 + COEFF_BASE_RANGE + 1] =
coeff_costs->lps_cost;
#if CONFIG_ATC_COEFCODING
const int(*lps_lf_cost)[COEFF_BASE_RANGE + 1 + COEFF_BASE_RANGE + 1] =
coeff_costs->lps_lf_cost;
#endif // CONFIG_ATC_COEFCODING
int c = eob - 1;
{
const int pos = scan[c];
const tran_low_t v = qcoeff[pos];
const int sign = AOMSIGN(v);
const int level = (v ^ sign) - sign;
const int coeff_ctx = coeff_contexts[pos];
#if CONFIG_ATC_COEFCODING
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost +=
coeff_costs->base_lf_eob_cost[coeff_ctx]
[AOMMIN(level, LF_BASE_SYMBOLS - 1) - 1];
} else {
cost += coeff_costs->base_eob_cost[coeff_ctx][AOMMIN(level, 3) - 1];
}
#else
cost += coeff_costs->base_eob_cost[coeff_ctx][AOMMIN(level, 3) - 1];
#endif // CONFIG_ATC_COEFCODING
if (v) {
// sign bit cost
#if CONFIG_ATC_COEFCODING
if (limits) {
if (level > LF_NUM_BASE_LEVELS) {
const int ctx = get_br_ctx_lf_eob(pos, tx_class);
cost += get_br_lf_cost(level, lps_lf_cost[ctx]);
}
} else {
if (level > NUM_BASE_LEVELS) {
const int ctx = 7; /* get_lf_ctx_eob */
cost += get_br_cost(level, lps_cost[ctx]);
}
}
#else
if (level > NUM_BASE_LEVELS) {
const int ctx = get_br_ctx_eob(pos, bwl, tx_class);
cost += get_br_cost(level, lps_cost[ctx]);
}
#endif // CONFIG_ATC_COEFCODING
if (c) {
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V) {
const int sign01 = (sign ^ sign) - sign;
cost += coeff_costs->v_ac_sign_cost[xd->tmp_sign[pos]][sign01];
} else {
cost += av1_cost_literal(1);
}
#else
cost += av1_cost_literal(1);
#endif // CONFIG_CONTEXT_DERIVATION
} else {
const int sign01 = (sign ^ sign) - sign;
const int dc_sign_ctx = txb_ctx->dc_sign_ctx;
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V) {
cost +=
coeff_costs->v_dc_sign_cost[xd->tmp_sign[0]][dc_sign_ctx][sign01];
} else {
cost += coeff_costs->dc_sign_cost[dc_sign_ctx][sign01];
}
#else
cost += coeff_costs->dc_sign_cost[dc_sign_ctx][sign01];
#endif // CONFIG_CONTEXT_DERIVATION
return cost;
}
}
}
#if CONFIG_ATC_COEFCODING
const int(*base_lf_cost)[LF_BASE_SYMBOLS * 2] = coeff_costs->base_lf_cost;
#endif // CONFIG_ATC_COEFCODING
const int(*base_cost)[8] = coeff_costs->base_cost;
for (c = eob - 2; c >= 1; --c) {
const int pos = scan[c];
const int coeff_ctx = coeff_contexts[pos];
const tran_low_t v = qcoeff[pos];
const int level = abs(v);
#if CONFIG_ATC_COEFCODING
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost += base_lf_cost[coeff_ctx][AOMMIN(level, LF_BASE_SYMBOLS - 1)];
} else {
cost += base_cost[coeff_ctx][AOMMIN(level, 3)];
}
#else
cost += base_cost[coeff_ctx][AOMMIN(level, 3)];
#endif // CONFIG_ATC_COEFCODING
if (v) {
// sign bit cost
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V) {
const int sign = AOMSIGN(v);
const int sign01 = (sign ^ sign) - sign;
cost += coeff_costs->v_ac_sign_cost[xd->tmp_sign[pos]][sign01];
} else {
cost += av1_cost_literal(1);
}
#else
cost += av1_cost_literal(1);
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_ATC_COEFCODING
if (limits) {
if (level > LF_NUM_BASE_LEVELS) {
const int ctx = get_br_lf_ctx(levels, pos, bwl, tx_class);
cost += get_br_lf_cost(level, lps_lf_cost[ctx]);
}
} else {
if (level > NUM_BASE_LEVELS) {
const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
cost += get_br_cost(level, lps_cost[ctx]);
}
}
#else
if (level > NUM_BASE_LEVELS) {
const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
cost += get_br_cost(level, lps_cost[ctx]);
}
#endif // CONFIG_ATC_COEFCODING
}
}
// c == 0 after previous loop
#if CONFIG_PAR_HIDING
int num_nz = 0;
for (c = eob - 1; c > 0; --c) {
const int pos = scan[c];
num_nz += !!qcoeff[pos];
}
c = 0;
if (num_nz >= PHTHRESH && enable_parity_hiding) {
const int pos = scan[c];
const tran_low_t v = qcoeff[pos];
const int level = abs(v);
const int q_index = level >> 1;
cost += coeff_costs_ph->base_ph_cost[get_base_ctx_ph(
levels, pos, bwl, tx_class)][AOMMIN(q_index, 3)];
if (v) {
const int dc_sign_ctx = txb_ctx->dc_sign_ctx;
cost += coeff_costs->dc_sign_cost[dc_sign_ctx][v < 0];
if (q_index > NUM_BASE_LEVELS) {
const int ctx = get_par_br_ctx(levels, pos, bwl, tx_class);
cost += get_br_cost(q_index, coeff_costs_ph->lps_ph_cost[ctx]);
}
}
return cost;
}
#endif
{
const int pos = scan[c];
const tran_low_t v = qcoeff[pos];
const int coeff_ctx = coeff_contexts[pos];
const int sign = AOMSIGN(v);
const int level = (v ^ sign) - sign;
#if CONFIG_ATC_COEFCODING
const int row = pos >> bwl;
const int col = pos - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost += base_lf_cost[coeff_ctx][AOMMIN(level, LF_BASE_SYMBOLS - 1)];
} else {
cost += base_cost[coeff_ctx][AOMMIN(level, 3)];
}
#else
cost += base_cost[coeff_ctx][AOMMIN(level, 3)];
#endif // CONFIG_ATC_COEFCODING
if (v) {
// sign bit cost
const int sign01 = (sign ^ sign) - sign;
const int dc_sign_ctx = txb_ctx->dc_sign_ctx;
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V) {
cost +=
coeff_costs->v_dc_sign_cost[xd->tmp_sign[0]][dc_sign_ctx][sign01];
} else {
cost += coeff_costs->dc_sign_cost[dc_sign_ctx][sign01];
}
#else
cost += coeff_costs->dc_sign_cost[dc_sign_ctx][sign01];
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_ATC_COEFCODING
if (limits) {
if (level > LF_NUM_BASE_LEVELS) {
const int ctx = get_br_lf_ctx(levels, pos, bwl, tx_class);
cost += get_br_lf_cost(level, lps_lf_cost[ctx]);
}
} else {
if (level > NUM_BASE_LEVELS) {
const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
cost += get_br_cost(level, lps_cost[ctx]);
}
}
#else
if (level > NUM_BASE_LEVELS) {
const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
cost += get_br_cost(level, lps_cost[ctx]);
}
#endif // CONFIG_ATC_COEFCODING
}
}
return cost;
}
static AOM_FORCE_INLINE int warehouse_efficients_txb_laplacian(
const AV1_COMMON *cm, const MACROBLOCK *x, const int plane, const int block,
const TX_SIZE tx_size, const TXB_CTX *const txb_ctx, const int eob,
const PLANE_TYPE plane_type, const LV_MAP_COEFF_COST *const coeff_costs,
const MACROBLOCKD *const xd, const TX_TYPE tx_type,
#if CONFIG_CROSS_CHROMA_TX
const CctxType cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
const TX_CLASS tx_class, int reduced_tx_set_used) {
#if CONFIG_CONTEXT_DERIVATION
int txb_skip_ctx = txb_ctx->txb_skip_ctx;
if (plane == AOM_PLANE_V) {
txb_skip_ctx +=
(x->plane[AOM_PLANE_U].eobs[block] ? V_TXB_SKIP_CONTEXT_OFFSET : 0);
}
#else
const int txb_skip_ctx = txb_ctx->txb_skip_ctx;
#endif // CONFIG_CONTEXT_DERIVATION
const int eob_multi_size = txsize_log2_minus4[tx_size];
const LV_MAP_EOB_COST *const eob_costs =
&x->coeff_costs.eob_costs[eob_multi_size][plane_type];
#if CONFIG_CONTEXT_DERIVATION
int cost;
if (plane == AOM_PLANE_V) {
cost = coeff_costs->v_txb_skip_cost[txb_skip_ctx][0];
} else {
cost = coeff_costs->txb_skip_cost[txb_skip_ctx][0];
}
#else
int cost = coeff_costs->txb_skip_cost[txb_skip_ctx][0];
#endif // CONFIG_CONTEXT_DERIVATION
cost += get_tx_type_cost(x, xd, plane, tx_size, tx_type, reduced_tx_set_used,
eob);
#if CONFIG_CROSS_CHROMA_TX
cost += get_cctx_type_cost(cm, x, xd, plane, tx_size, block, cctx_type);
#endif // CONFIG_CROSS_CHROMA_TX
const MB_MODE_INFO *mbmi = xd->mi[0];
if ((mbmi->fsc_mode[xd->tree_type == CHROMA_PART] &&
get_primary_tx_type(tx_type) == IDTX && plane == PLANE_TYPE_Y) ||
use_inter_fsc(cm, plane, tx_type, is_inter_block(mbmi, xd->tree_type))) {
cost +=
av1_cost_coeffs_txb_skip_estimate(x, plane, block, tx_size, tx_type);
} else {
cost += get_eob_cost(eob, eob_costs, coeff_costs, tx_class);
cost += av1_cost_coeffs_txb_estimate(x, plane, block, tx_size, tx_type);
}
return cost;
}
// Look up table of individual cost of coefficient by its quantization level.
// determined based on Laplacian distribution conditioned on estimated context
static const int costLUT[15] = { -1143, 53, 545, 825, 1031,
1209, 1393, 1577, 1763, 1947,
2132, 2317, 2501, 2686, 2871 };
static const int const_term = (1 << AV1_PROB_COST_SHIFT);
static const int loge_par = ((14427 << AV1_PROB_COST_SHIFT) + 5000) / 10000;
int av1_cost_coeffs_txb_estimate(const MACROBLOCK *x, const int plane,
const int block, const TX_SIZE tx_size,
const TX_TYPE tx_type) {
assert(plane == 0);
int cost = 0;
const struct macroblock_plane *p = &x->plane[plane];
const SCAN_ORDER *scan_order = get_scan(tx_size, tx_type);
const int16_t *scan = scan_order->scan;
tran_low_t *qcoeff = p->qcoeff + BLOCK_OFFSET(block);
int eob = p->eobs[block];
// coeffs
int c = eob - 1;
// eob
{
const int pos = scan[c];
const tran_low_t v = abs(qcoeff[pos]) - 1;
cost += (v << (AV1_PROB_COST_SHIFT + 2));
}
// other coeffs
for (c = eob - 2; c >= 0; c--) {
const int pos = scan[c];
const tran_low_t v = abs(qcoeff[pos]);
const int idx = AOMMIN(v, 14);
cost += costLUT[idx];
}
// const_term does not contain DC, and log(e) does not contain eob, so both
// (eob-1)
cost += (const_term + loge_par) * (eob - 1);
return cost;
}
int av1_cost_coeffs_txb_skip_estimate(const MACROBLOCK *x, const int plane,
const int block, const TX_SIZE tx_size,
const TX_TYPE tx_type) {
assert(plane == PLANE_TYPE_Y);
int cost = 0;
const struct macroblock_plane *p = &x->plane[plane];
const SCAN_ORDER *scan_order = get_scan(tx_size, tx_type);
const int16_t *scan = scan_order->scan;
tran_low_t *qcoeff = p->qcoeff + BLOCK_OFFSET(block);
int eob = p->eobs[block];
assert(eob == av1_get_max_eob(tx_size));
// coeffs
for (int c = 0; c < eob; c++) {
const int pos = scan[c];
const tran_low_t v = abs(qcoeff[pos]);
const int idx = AOMMIN(v, 14);
cost += costLUT[idx];
}
cost += (const_term + loge_par) * (eob - 1);
return cost;
}
int av1_cost_coeffs_txb(const AV1_COMMON *cm, const MACROBLOCK *x,
const int plane, const int block, const TX_SIZE tx_size,
const TX_TYPE tx_type,
#if CONFIG_CROSS_CHROMA_TX
const CctxType cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
const TXB_CTX *const txb_ctx, int reduced_tx_set_used) {
const struct macroblock_plane *p = &x->plane[plane];
const int eob = p->eobs[block];
const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
const PLANE_TYPE plane_type = get_plane_type(plane);
const LV_MAP_COEFF_COST *const coeff_costs =
&x->coeff_costs.coeff_costs[txs_ctx][plane_type];
const MACROBLOCKD *const xd = &x->e_mbd;
if (eob == 0) {
int skip_cost = 0;
#if CONFIG_CROSS_CHROMA_TX && CCTX_C2_DROPPED
if (plane == AOM_PLANE_V && !keep_chroma_c2(cctx_type) &&
is_cctx_allowed(cm, xd))
return 0;
#endif // CONFIG_CROSS_CHROMA_TX && CCTX_C2_DROPPED
#if CONFIG_CONTEXT_DERIVATION
int txb_skip_ctx = txb_ctx->txb_skip_ctx;
if (plane == AOM_PLANE_Y || plane == AOM_PLANE_U) {
skip_cost += coeff_costs->txb_skip_cost[txb_skip_ctx][1];
} else {
txb_skip_ctx +=
(x->plane[AOM_PLANE_U].eobs[block] ? V_TXB_SKIP_CONTEXT_OFFSET : 0);
skip_cost += coeff_costs->v_txb_skip_cost[txb_skip_ctx][1];
}
#else
skip_cost += coeff_costs->txb_skip_cost[txb_ctx->txb_skip_ctx][1];
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CROSS_CHROMA_TX
skip_cost +=
get_cctx_type_cost(cm, x, xd, plane, tx_size, block, cctx_type);
#endif // CONFIG_CROSS_CHROMA_TX
return skip_cost;
}
const TX_CLASS tx_class = tx_type_to_class[get_primary_tx_type(tx_type)];
#if CONFIG_PAR_HIDING
bool enable_parity_hiding = cm->features.allow_parity_hiding &&
!xd->lossless[xd->mi[0]->segment_id] &&
plane == PLANE_TYPE_Y &&
get_primary_tx_type(tx_type) < IDTX;
#endif // CONFIG_PAR_HIDING
const MB_MODE_INFO *mbmi = xd->mi[0];
if ((mbmi->fsc_mode[xd->tree_type == CHROMA_PART] &&
get_primary_tx_type(tx_type) == IDTX && plane == PLANE_TYPE_Y) ||
use_inter_fsc(cm, plane, tx_type, is_inter_block(mbmi, xd->tree_type))) {
return warehouse_efficients_txb_skip(
#if CONFIG_CROSS_CHROMA_TX
cm,
#endif // CONFIG_CROSS_CHROMA_TX
x, plane, block, tx_size, txb_ctx, p, eob, coeff_costs, xd, tx_type,
#if CONFIG_CROSS_CHROMA_TX
cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
reduced_tx_set_used);
} else {
return warehouse_efficients_txb(
#if CONFIG_CROSS_CHROMA_TX
cm,
#endif // CONFIG_CROSS_CHROMA_TX
x, plane, block, tx_size, txb_ctx, p, eob, plane_type, coeff_costs, xd,
tx_type,
#if CONFIG_CROSS_CHROMA_TX
cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
tx_class, reduced_tx_set_used
#if CONFIG_PAR_HIDING
,
enable_parity_hiding, &x->coeff_costs.coeff_costs[0][0]
#endif // CONFIG_PAR_HIDING
);
}
}
int av1_cost_coeffs_txb_laplacian(const AV1_COMMON *cm, const MACROBLOCK *x,
const int plane, const int block,
const TX_SIZE tx_size, const TX_TYPE tx_type,
#if CONFIG_CROSS_CHROMA_TX
const CctxType cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
const TXB_CTX *const txb_ctx,
const int reduced_tx_set_used,
const int adjust_eob) {
const struct macroblock_plane *p = &x->plane[plane];
int eob = p->eobs[block];
if (adjust_eob) {
const SCAN_ORDER *scan_order = get_scan(tx_size, tx_type);
const int16_t *scan = scan_order->scan;
tran_low_t *tcoeff = p->coeff + BLOCK_OFFSET(block);
tran_low_t *qcoeff = p->qcoeff + BLOCK_OFFSET(block);
tran_low_t *dqcoeff = p->dqcoeff + BLOCK_OFFSET(block);
update_coeff_eob_fast(&eob, av1_get_tx_scale(tx_size), p->dequant_QTX, scan,
tcoeff, qcoeff, dqcoeff);
p->eobs[block] = eob;
}
const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
const PLANE_TYPE plane_type = get_plane_type(plane);
const LV_MAP_COEFF_COST *const coeff_costs =
&x->coeff_costs.coeff_costs[txs_ctx][plane_type];
const MACROBLOCKD *const xd = &x->e_mbd;
if (eob == 0) {
int skip_cost = 0;
#if CONFIG_CONTEXT_DERIVATION
int txb_skip_ctx = txb_ctx->txb_skip_ctx;
if (plane == AOM_PLANE_Y || plane == AOM_PLANE_U) {
skip_cost += coeff_costs->txb_skip_cost[txb_skip_ctx][1];
} else {
txb_skip_ctx +=
(x->plane[AOM_PLANE_U].eobs[block] ? V_TXB_SKIP_CONTEXT_OFFSET : 0);
skip_cost += coeff_costs->v_txb_skip_cost[txb_skip_ctx][1];
}
#else
skip_cost += coeff_costs->txb_skip_cost[txb_ctx->txb_skip_ctx][1];
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CROSS_CHROMA_TX
skip_cost +=
get_cctx_type_cost(cm, x, xd, plane, tx_size, block, cctx_type);
#endif // CONFIG_CROSS_CHROMA_TX
return skip_cost;
}
const TX_CLASS tx_class = tx_type_to_class[get_primary_tx_type(tx_type)];
return warehouse_efficients_txb_laplacian(cm, x, plane, block, tx_size,
txb_ctx, eob, plane_type,
coeff_costs, xd, tx_type,
#if CONFIG_CROSS_CHROMA_TX
cctx_type,
#endif // CONFIG_CROSS_CHROMA_TX
tx_class, reduced_tx_set_used);
}
static AOM_FORCE_INLINE int get_two_coeff_cost_simple(
#if CONFIG_ATC_COEFCODING
int plane,
#endif // CONFIG_ATC_COEFCODING
int ci, tran_low_t abs_qc, int coeff_ctx,
const LV_MAP_COEFF_COST *txb_costs, int bwl, TX_CLASS tx_class,
const uint8_t *levels, int *cost_low) {
// this simple version assumes the coeff's scan_idx is not DC (scan_idx != 0)
// and not the last (scan_idx != eob - 1)
assert(ci > 0);
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int cost = 0;
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost +=
txb_costs->base_lf_cost[coeff_ctx][AOMMIN(abs_qc, LF_BASE_SYMBOLS - 1)];
} else {
cost += txb_costs->base_cost[coeff_ctx][AOMMIN(abs_qc, 3)];
}
#else
int cost = txb_costs->base_cost[coeff_ctx][AOMMIN(abs_qc, 3)];
#endif // CONFIG_ATC_COEFCODING
int diff = 0;
#if CONFIG_ATC_COEFCODING
if (limits) {
if (abs_qc <= (LF_BASE_SYMBOLS - 1)) {
if (abs_qc == 0) {
diff = 0;
} else if (abs_qc == 1) {
diff = txb_costs->base_lf_cost[coeff_ctx][1] + av1_cost_literal(1) -
txb_costs->base_lf_cost[coeff_ctx][0];
} else if (abs_qc == 2) {
diff = txb_costs->base_lf_cost[coeff_ctx][2] -
txb_costs->base_lf_cost[coeff_ctx][1];
} else if (abs_qc == 3) {
diff = txb_costs->base_lf_cost[coeff_ctx][3] -
txb_costs->base_lf_cost[coeff_ctx][2];
} else if (abs_qc == 4) {
diff = txb_costs->base_lf_cost[coeff_ctx][4] -
txb_costs->base_lf_cost[coeff_ctx][3];
} else {
diff = txb_costs->base_lf_cost[coeff_ctx][5] -
txb_costs->base_lf_cost[coeff_ctx][4];
}
}
} else {
if (abs_qc <= 3) {
if (abs_qc == 0) {
diff = 0;
} else if (abs_qc == 1) {
diff = txb_costs->base_cost[coeff_ctx][1] + av1_cost_literal(1) -
txb_costs->base_cost[coeff_ctx][0];
} else if (abs_qc == 2) {
diff = txb_costs->base_cost[coeff_ctx][2] -
txb_costs->base_cost[coeff_ctx][1];
} else {
diff = txb_costs->base_cost[coeff_ctx][3] -
txb_costs->base_cost[coeff_ctx][2];
}
}
}
#else
if (abs_qc <= 3) diff = txb_costs->base_cost[coeff_ctx][abs_qc + 4];
#endif // CONFIG_ATC_COEFCODING
if (abs_qc) {
cost += av1_cost_literal(1);
#if CONFIG_ATC_COEFCODING
if (limits) {
if (abs_qc > LF_NUM_BASE_LEVELS) {
const int br_ctx = get_br_lf_ctx(levels, ci, bwl, tx_class);
int brcost_diff = 0;
cost += get_br_lf_cost_with_diff(abs_qc, txb_costs->lps_lf_cost[br_ctx],
&brcost_diff);
diff += brcost_diff;
}
} else {
if (abs_qc > NUM_BASE_LEVELS) {
const int br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
int brcost_diff = 0;
cost += get_br_cost_with_diff(abs_qc, txb_costs->lps_cost[br_ctx],
&brcost_diff);
diff += brcost_diff;
}
}
#else
if (abs_qc > NUM_BASE_LEVELS) {
const int br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
int brcost_diff = 0;
cost += get_br_cost_with_diff(abs_qc, txb_costs->lps_cost[br_ctx],
&brcost_diff);
diff += brcost_diff;
}
#endif // CONFIG_ATC_COEFCODING
}
*cost_low = cost - diff;
return cost;
}
static INLINE int get_coeff_cost_eob(int ci, tran_low_t abs_qc, int sign,
int coeff_ctx, int dc_sign_ctx,
const LV_MAP_COEFF_COST *txb_costs,
int bwl, TX_CLASS tx_class
#if CONFIG_CONTEXT_DERIVATION
,
int32_t *tmp_sign
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
,
int plane
#endif // CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
) {
int cost = 0;
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost +=
txb_costs->base_lf_eob_cost[coeff_ctx]
[AOMMIN(abs_qc, LF_BASE_SYMBOLS - 1) - 1];
} else {
cost += txb_costs->base_eob_cost[coeff_ctx][AOMMIN(abs_qc, 3) - 1];
}
#else
cost += txb_costs->base_eob_cost[coeff_ctx][AOMMIN(abs_qc, 3) - 1];
#endif // CONFIG_ATC_COEFCODING
if (abs_qc != 0) {
if (ci == 0) {
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V)
cost += txb_costs->v_dc_sign_cost[tmp_sign[0]][dc_sign_ctx][sign];
else
cost += txb_costs->dc_sign_cost[dc_sign_ctx][sign];
#else
cost += txb_costs->dc_sign_cost[dc_sign_ctx][sign];
#endif // CONFIG_CONTEXT_DERIVATION
} else {
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V)
cost += txb_costs->v_ac_sign_cost[tmp_sign[ci]][sign];
else
cost += av1_cost_literal(1);
#else
cost += av1_cost_literal(1);
#endif // CONFIG_CONTEXT_DERIVATION
}
#if CONFIG_ATC_COEFCODING
if (limits) {
if (abs_qc > LF_NUM_BASE_LEVELS) {
int br_ctx;
br_ctx = get_br_ctx_lf_eob(ci, tx_class);
cost += get_br_lf_cost(abs_qc, txb_costs->lps_lf_cost[br_ctx]);
}
} else {
if (abs_qc > NUM_BASE_LEVELS) {
int br_ctx = 7; /* get_br_ctx_eob */
cost += get_br_cost(abs_qc, txb_costs->lps_cost[br_ctx]);
}
}
#else
if (abs_qc > NUM_BASE_LEVELS) {
int br_ctx;
br_ctx = get_br_ctx_eob(ci, bwl, tx_class);
cost += get_br_cost(abs_qc, txb_costs->lps_cost[br_ctx]);
}
#endif // CONFIG_ATC_COEFCODING
}
return cost;
}
static INLINE int get_coeff_cost_general(int is_last, int ci, tran_low_t abs_qc,
int sign, int coeff_ctx,
int dc_sign_ctx,
const LV_MAP_COEFF_COST *txb_costs,
int bwl, TX_CLASS tx_class,
const uint8_t *levels
#if CONFIG_CONTEXT_DERIVATION
,
int32_t *tmp_sign
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
,
int plane
#endif // CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
) {
int cost = 0;
if (is_last) {
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost +=
txb_costs->base_lf_eob_cost[coeff_ctx]
[AOMMIN(abs_qc, LF_BASE_SYMBOLS - 1) - 1];
} else {
cost += txb_costs->base_eob_cost[coeff_ctx][AOMMIN(abs_qc, 3) - 1];
}
#else
cost += txb_costs->base_eob_cost[coeff_ctx][AOMMIN(abs_qc, 3) - 1];
#endif // CONFIG_ATC_COEFCODING
} else {
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
cost +=
txb_costs
->base_lf_cost[coeff_ctx][AOMMIN(abs_qc, LF_BASE_SYMBOLS - 1)];
} else {
cost += txb_costs->base_cost[coeff_ctx][AOMMIN(abs_qc, 3)];
}
#else
cost += txb_costs->base_cost[coeff_ctx][AOMMIN(abs_qc, 3)];
#endif // CONFIG_ATC_COEFCODING
}
if (abs_qc != 0) {
if (ci == 0) {
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V)
cost += txb_costs->v_dc_sign_cost[tmp_sign[0]][dc_sign_ctx][sign];
else
cost += txb_costs->dc_sign_cost[dc_sign_ctx][sign];
#else
cost += txb_costs->dc_sign_cost[dc_sign_ctx][sign];
#endif // CONFIG_CONTEXT_DERIVATION
} else {
#if CONFIG_CONTEXT_DERIVATION
if (plane == AOM_PLANE_V)
cost += txb_costs->v_ac_sign_cost[tmp_sign[ci]][sign];
else
cost += av1_cost_literal(1);
#else
cost += av1_cost_literal(1);
#endif // CONFIG_CONTEXT_DERIVATION
}
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
if (abs_qc > LF_NUM_BASE_LEVELS) {
int br_ctx;
if (is_last)
br_ctx = get_br_ctx_lf_eob(ci, tx_class);
else
br_ctx = get_br_lf_ctx(levels, ci, bwl, tx_class);
cost += get_br_lf_cost(abs_qc, txb_costs->lps_lf_cost[br_ctx]);
}
} else {
if (abs_qc > NUM_BASE_LEVELS) {
int br_ctx;
if (is_last)
br_ctx = 0; /*get_br_ctx_eob*/
else
br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
cost += get_br_cost(abs_qc, txb_costs->lps_cost[br_ctx]);
}
}
#else
if (abs_qc > NUM_BASE_LEVELS) {
int br_ctx;
if (is_last)
br_ctx = get_br_ctx_eob(ci, bwl, tx_class);
else
br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
cost += get_br_cost(abs_qc, txb_costs->lps_cost[br_ctx]);
}
#endif // CONFIG_ATC_COEFCODING
}
return cost;
}
static INLINE void get_qc_dqc_low(tran_low_t abs_qc, int sign, int dqv,
int shift, tran_low_t *qc_low,
tran_low_t *dqc_low) {
tran_low_t abs_qc_low = abs_qc - 1;
*qc_low = (-sign ^ abs_qc_low) + sign;
assert((sign ? -abs_qc_low : abs_qc_low) == *qc_low);
tran_low_t abs_dqc_low =
(tran_low_t)(ROUND_POWER_OF_TWO_64((tran_high_t)abs_qc_low * dqv,
QUANT_TABLE_BITS) >>
shift);
*dqc_low = (-sign ^ abs_dqc_low) + sign;
assert((sign ? -abs_dqc_low : abs_dqc_low) == *dqc_low);
}
static INLINE void update_coeff_general(
int *accu_rate, int64_t *accu_dist, int si, int eob,
#if !CONFIG_ATC_COEFCODING
TX_SIZE tx_size,
#endif // !CONFIG_ATC_COEFCODING
TX_CLASS tx_class, int bwl, int height, int64_t rdmult, int shift,
int dc_sign_ctx, const int32_t *dequant, const int16_t *scan,
const LV_MAP_COEFF_COST *txb_costs, const tran_low_t *tcoeff,
tran_low_t *qcoeff, tran_low_t *dqcoeff, uint8_t *levels,
const qm_val_t *iqmatrix
#if CONFIG_CONTEXT_DERIVATION
,
int32_t *tmp_sign
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
,
int plane
#endif // CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
#if CONFIG_PAR_HIDING
,
coeff_info *coef_info, bool enable_parity_hiding
#endif // CONFIG_PAR_HIDING
) {
const int dqv = get_dqv(dequant, scan[si], iqmatrix);
const int ci = scan[si];
const tran_low_t qc = qcoeff[ci];
const int is_last = si == (eob - 1);
const int coeff_ctx =
get_lower_levels_ctx_general(is_last, si, bwl, height, levels, ci,
#if !CONFIG_ATC_COEFCODING
tx_size,
#endif // !CONFIG_ATC_COEFCODING
tx_class
#if CONFIG_ATC_COEFCODING
,
plane
#endif // CONFIG_ATC_COEFCODING
);
if (qc == 0) {
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
*accu_rate += txb_costs->base_lf_cost[coeff_ctx][0];
} else {
*accu_rate += txb_costs->base_cost[coeff_ctx][0];
}
#else
*accu_rate += txb_costs->base_cost[coeff_ctx][0];
#endif // CONFIG_ATC_COEFCODING
} else {
const int sign = (qc < 0) ? 1 : 0;
const tran_low_t abs_qc = abs(qc);
const tran_low_t tqc = tcoeff[ci];
const tran_low_t dqc = dqcoeff[ci];
const int64_t dist = get_coeff_dist(tqc, dqc, shift);
const int64_t dist0 = get_coeff_dist(tqc, 0, shift);
const int rate =
get_coeff_cost_general(is_last, ci, abs_qc, sign, coeff_ctx,
dc_sign_ctx, txb_costs, bwl, tx_class, levels
#if CONFIG_CONTEXT_DERIVATION
,
tmp_sign
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
,
plane
#endif // CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
);
const int64_t rd = RDCOST(rdmult, rate, dist);
tran_low_t qc_low, dqc_low;
tran_low_t abs_qc_low;
int64_t dist_low, rd_low;
int rate_low;
if (abs_qc == 1) {
abs_qc_low = qc_low = dqc_low = 0;
dist_low = dist0;
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
if (limits) {
rate_low = txb_costs->base_lf_cost[coeff_ctx][0];
} else {
rate_low = txb_costs->base_cost[coeff_ctx][0];
}
#else
rate_low = txb_costs->base_cost[coeff_ctx][0];
#endif // CONFIG_ATC_COEFCODING
} else {
get_qc_dqc_low(abs_qc, sign, dqv, shift, &qc_low, &dqc_low);
abs_qc_low = abs_qc - 1;
dist_low = get_coeff_dist(tqc, dqc_low, shift);
rate_low =
get_coeff_cost_general(is_last, ci, abs_qc_low, sign, coeff_ctx,
dc_sign_ctx, txb_costs, bwl, tx_class, levels
#if CONFIG_CONTEXT_DERIVATION
,
tmp_sign
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
,
plane
#endif // CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
);
}
rd_low = RDCOST(rdmult, rate_low, dist_low);
if (rd_low < rd) {
qcoeff[ci] = qc_low;
dqcoeff[ci] = dqc_low;
levels[get_padded_idx(ci, bwl)] = AOMMIN(abs_qc_low, INT8_MAX);
*accu_rate += rate_low;
*accu_dist += dist_low - dist0;
#if CONFIG_PAR_HIDING
if (enable_parity_hiding)
set_coeff_info(qc_low, dqc_low, qc, dqc, rd_low, rd, rate_low, rate,
false, coef_info, si);
#endif // CONFIG_PAR_HIDING
} else {
*accu_rate += rate;
*accu_dist += dist - dist0;
#if CONFIG_PAR_HIDING
if (enable_parity_hiding)
set_coeff_info(qc_low, dqc_low, qc, dqc, rd_low, rd, rate_low, rate,
true, coef_info, si);
#endif // CONFIG_PAR_HIDING
}
}
}
static AOM_FORCE_INLINE void update_coeff_simple(
int *accu_rate, int si, int eob,
#if !CONFIG_ATC_COEFCODING
TX_SIZE tx_size,
#endif // !CONFIG_ATC_COEFCODING
TX_CLASS tx_class, int bwl, int64_t rdmult, int shift,
const int32_t *dequant, const int16_t *scan,
const LV_MAP_COEFF_COST *txb_costs, const tran_low_t *tcoeff,
tran_low_t *qcoeff, tran_low_t *dqcoeff, uint8_t *levels,
const qm_val_t *iqmatrix
#if CONFIG_PAR_HIDING
,
coeff_info *coef_info, bool enable_parity_hiding
#endif // CONFIG_PAR_HIDING
#if CONFIG_ATC_COEFCODING
,
int plane) {
#else
) {
#endif // CONFIG_ATC_COEFCODING
const int dqv = get_dqv(dequant, scan[si], iqmatrix);
(void)eob;
// this simple version assumes the coeff's scan_idx is not DC (scan_idx != 0)
// and not the last (scan_idx != eob - 1)
assert(si != eob - 1);
assert(si > 0);
const int ci = scan[si];
const tran_low_t qc = qcoeff[ci];
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
int coeff_ctx = 0;
if (limits) {
coeff_ctx = get_lower_levels_lf_ctx(levels, ci, bwl, tx_class);
} else {
coeff_ctx = get_lower_levels_ctx(levels, ci, bwl, tx_class);
}
#else
const int coeff_ctx =
get_lower_levels_ctx(levels, ci, bwl, tx_size, tx_class);
#endif // CONFIG_ATC_COEFCODING
if (qc == 0) {
#if CONFIG_ATC_COEFCODING
if (limits) {
*accu_rate += txb_costs->base_lf_cost[coeff_ctx][0];
} else {
*accu_rate += txb_costs->base_cost[coeff_ctx][0];
}
#else
*accu_rate += txb_costs->base_cost[coeff_ctx][0];
#endif // CONFIG_ATC_COEFCODING
} else {
const tran_low_t abs_qc = abs(qc);
const tran_low_t abs_tqc = abs(tcoeff[ci]);
const tran_low_t abs_dqc = abs(dqcoeff[ci]);
int rate_low = 0;
const int rate = get_two_coeff_cost_simple(
#if CONFIG_ATC_COEFCODING
plane,
#endif // CONFIG_ATC_COEFCODING
ci, abs_qc, coeff_ctx, txb_costs, bwl, tx_class, levels, &rate_low);
if (abs_dqc < abs_tqc) {
*accu_rate += rate;
return;
}
const int64_t dist = get_coeff_dist(abs_tqc, abs_dqc, shift);
const int64_t rd = RDCOST(rdmult, rate, dist);
const tran_low_t abs_qc_low = abs_qc - 1;
const tran_low_t abs_dqc_low =
(tran_low_t)ROUND_POWER_OF_TWO_64((tran_high_t)abs_qc_low * dqv,
QUANT_TABLE_BITS) >>
shift;
const int64_t dist_low = get_coeff_dist(abs_tqc, abs_dqc_low, shift);
const int64_t rd_low = RDCOST(rdmult, rate_low, dist_low);
if (rd_low < rd) {
#if CONFIG_PAR_HIDING
tran_low_t qc_low = qc < 0 ? -abs_qc_low : abs_qc_low;
tran_low_t dqc_low = qc < 0 ? -abs_dqc_low : abs_dqc_low;
if (enable_parity_hiding)
set_coeff_info(qc_low, dqc_low, qc, dqcoeff[ci], rd_low, rd, rate_low,
rate, false, coef_info, si);
qcoeff[ci] = qc_low;
dqcoeff[ci] = dqc_low;
#else
const int sign = (qc < 0) ? 1 : 0;
qcoeff[ci] = (-sign ^ abs_qc_low) + sign;
dqcoeff[ci] = (-sign ^ abs_dqc_low) + sign;
#endif // CONFIG_PAR_HIDING
levels[get_padded_idx(ci, bwl)] = AOMMIN(abs_qc_low, INT8_MAX);
*accu_rate += rate_low;
} else {
*accu_rate += rate;
#if CONFIG_PAR_HIDING
tran_low_t qc_low = qc < 0 ? -abs_qc_low : abs_qc_low;
tran_low_t dqc_low = qc < 0 ? -abs_dqc_low : abs_dqc_low;
if (enable_parity_hiding)
set_coeff_info(qc_low, dqc_low, qc, dqcoeff[ci], rd_low, rd, rate_low,
rate, true, coef_info, si);
#endif // CONFIG_PAR_HIDING
}
}
}
static AOM_FORCE_INLINE void update_coeff_eob(
int *accu_rate, int64_t *accu_dist, int *eob, int *nz_num, int *nz_ci,
int si,
#if !CONFIG_ATC_COEFCODING
TX_SIZE tx_size,
#endif // !CONFIG_ATC_COEFCODING
TX_CLASS tx_class, int bwl, int height, int dc_sign_ctx, int64_t rdmult,
int shift, const int32_t *dequant, const int16_t *scan,
const LV_MAP_EOB_COST *txb_eob_costs, const LV_MAP_COEFF_COST *txb_costs,
const tran_low_t *tcoeff, tran_low_t *qcoeff, tran_low_t *dqcoeff,
uint8_t *levels, int sharpness, const qm_val_t *iqmatrix
#if CONFIG_CONTEXT_DERIVATION
,
int32_t *tmp_sign
#endif // CONFIG_CONTEXT_DERIVATION
#if CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
,
int plane
#endif // CONFIG_CONTEXT_DERIVATION || CONFIG_ATC_COEFCODING
#if CONFIG_PAR_HIDING
,
coeff_info *coef_info, bool enable_parity_hiding
#endif // CONFIG_PAR_HIDING
) {
const int dqv = get_dqv(dequant, scan[si], iqmatrix);
assert(si != *eob - 1);
const int ci = scan[si];
const tran_low_t qc = qcoeff[ci];
#if CONFIG_ATC_COEFCODING
const int row = ci >> bwl;
const int col = ci - (row << bwl);
int limits = get_lf_limits(row, col, tx_class, plane);
int coeff_ctx = 0;
if (limits) {
coeff_ctx = get_lower_levels_lf_ctx(levels, ci, bwl, tx_class);
} else {
coeff_ctx = get_lower_levels_ctx(levels, ci, bwl, tx_class);
}
#else
const int coeff_ctx =
get_lower_levels_ctx(levels, ci, bwl, tx_size, tx_class);
#endif // CONFIG_ATC_COEFCODING
if (qc == 0) {
#if CONFIG_ATC_COEFCODING
if (limits) {
*accu_rate += txb_costs->base_lf_cost[coeff_ctx][0];
} else {
*accu_rate += txb_costs->base_cost[coeff_ctx][0];
}
#else
*accu_rate += txb_costs->base_cost[coeff_ctx][0];
#endif // CONFIG_ATC_COEFCODING
} else {
#if CONFIG_PAR_HIDING
int64_t rd_eob_low = INT64_MAX >> 1;
int rate_eob_low = INT32_MAX >> 1;
#endif // CONFIG_PAR_HIDING
int lower_level = 0;
const tran_low_t abs_qc = abs(qc);
const tran_low_t tqc = tcoeff[ci];
const tran_low_t dqc = dqcoeff[ci];
const int sign = (qc < 0) ? 1 : 0;
const int64_t dist0 = get_coeff_dist(tqc, 0, shift);
int64_t dist = get_coeff_dist(tqc, dqc, shift) - dist0;