vpx/vp9/encoder/vp9_bitstream.c

1525 lines
49 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include <assert.h>
#include <stdio.h>
#include <limits.h>
2010-05-18 17:58:33 +02:00
#include "vpx/vpx_encoder.h"
#include "vpx_mem/vpx_mem.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_findnearmv.h"
#include "vp9/common/vp9_tile_common.h"
#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_mvref_common.h"
#include "vp9/common/vp9_treecoder.h"
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/common/vp9_pragmas.h"
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/encoder/vp9_encodemv.h"
#include "vp9/encoder/vp9_bitstream.h"
#include "vp9/encoder/vp9_segmentation.h"
#include "vp9/encoder/vp9_subexp.h"
#include "vp9/encoder/vp9_write_bit_buffer.h"
2010-05-18 17:58:33 +02:00
#if defined(SECTIONBITS_OUTPUT)
unsigned __int64 Sectionbits[500];
#endif
#ifdef ENTROPY_STATS
int intra_mode_stats[VP9_INTRA_MODES]
[VP9_INTRA_MODES]
[VP9_INTRA_MODES];
vp9_coeff_stats tree_update_hist[TX_SIZES][BLOCK_TYPES];
2010-05-18 17:58:33 +02:00
extern unsigned int active_section;
#endif
#ifdef MODE_STATS
int64_t tx_count_32x32p_stats[TX_SIZE_CONTEXTS][TX_SIZES];
int64_t tx_count_16x16p_stats[TX_SIZE_CONTEXTS][TX_SIZES - 1];
int64_t tx_count_8x8p_stats[TX_SIZE_CONTEXTS][TX_SIZES - 2];
int64_t switchable_interp_stats[VP9_SWITCHABLE_FILTERS+1]
[VP9_SWITCHABLE_FILTERS];
void init_tx_count_stats() {
vp9_zero(tx_count_32x32p_stats);
vp9_zero(tx_count_16x16p_stats);
vp9_zero(tx_count_8x8p_stats);
}
void init_switchable_interp_stats() {
vp9_zero(switchable_interp_stats);
}
static void update_tx_count_stats(VP9_COMMON *cm) {
int i, j;
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (j = 0; j < TX_SIZES; j++) {
tx_count_32x32p_stats[i][j] += cm->fc.tx_count_32x32p[i][j];
}
}
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (j = 0; j < TX_SIZES - 1; j++) {
tx_count_16x16p_stats[i][j] += cm->fc.tx_count_16x16p[i][j];
}
}
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (j = 0; j < TX_SIZES - 2; j++) {
tx_count_8x8p_stats[i][j] += cm->fc.tx_count_8x8p[i][j];
}
}
}
static void update_switchable_interp_stats(VP9_COMMON *cm) {
int i, j;
for (i = 0; i < VP9_SWITCHABLE_FILTERS+1; ++i)
for (j = 0; j < VP9_SWITCHABLE_FILTERS; ++j) {
switchable_interp_stats[i][j] += cm->fc.switchable_interp_count[i][j];
}
}
void write_tx_count_stats() {
int i, j;
FILE *fp = fopen("tx_count.bin", "wb");
fwrite(tx_count_32x32p_stats, sizeof(tx_count_32x32p_stats), 1, fp);
fwrite(tx_count_16x16p_stats, sizeof(tx_count_16x16p_stats), 1, fp);
fwrite(tx_count_8x8p_stats, sizeof(tx_count_8x8p_stats), 1, fp);
fclose(fp);
printf(
"vp9_default_tx_count_32x32p[TX_SIZE_CONTEXTS][TX_SIZES] = {\n");
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
printf(" { ");
for (j = 0; j < TX_SIZES; j++) {
printf("%"PRId64", ", tx_count_32x32p_stats[i][j]);
}
printf("},\n");
}
printf("};\n");
printf(
"vp9_default_tx_count_16x16p[TX_SIZE_CONTEXTS][TX_SIZES-1] = {\n");
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
printf(" { ");
for (j = 0; j < TX_SIZES - 1; j++) {
printf("%"PRId64", ", tx_count_16x16p_stats[i][j]);
}
printf("},\n");
}
printf("};\n");
printf(
"vp9_default_tx_count_8x8p[TX_SIZE_CONTEXTS][TX_SIZES-2] = {\n");
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
printf(" { ");
for (j = 0; j < TX_SIZES - 2; j++) {
printf("%"PRId64", ", tx_count_8x8p_stats[i][j]);
}
printf("},\n");
}
printf("};\n");
}
void write_switchable_interp_stats() {
int i, j;
FILE *fp = fopen("switchable_interp.bin", "wb");
fwrite(switchable_interp_stats, sizeof(switchable_interp_stats), 1, fp);
fclose(fp);
printf(
"vp9_default_switchable_filter_count[VP9_SWITCHABLE_FILTERS+1]"
"[VP9_SWITCHABLE_FILTERS] = {\n");
for (i = 0; i < VP9_SWITCHABLE_FILTERS+1; i++) {
printf(" { ");
for (j = 0; j < VP9_SWITCHABLE_FILTERS; j++) {
printf("%"PRId64", ", switchable_interp_stats[i][j]);
}
printf("},\n");
}
printf("};\n");
}
#endif
static INLINE void write_be32(uint8_t *p, int value) {
p[0] = value >> 24;
p[1] = value >> 16;
p[2] = value >> 8;
p[3] = value;
}
void vp9_encode_unsigned_max(struct vp9_write_bit_buffer *wb,
int data, int max) {
vp9_wb_write_literal(wb, data, get_unsigned_bits(max));
}
static void update_mode(
vp9_writer *w,
int n,
vp9_tree tree,
vp9_prob Pnew[/* n-1 */],
vp9_prob Pcur[/* n-1 */],
unsigned int bct[/* n-1 */] [2],
const unsigned int num_events[/* n */]
) {
int i = 0;
vp9_tree_probs_from_distribution(tree, Pnew, bct, num_events, 0);
n--;
for (i = 0; i < n; ++i) {
vp9_cond_prob_diff_update(w, &Pcur[i], VP9_MODE_UPDATE_PROB, bct[i]);
}
}
static void update_mbintra_mode_probs(VP9_COMP* const cpi,
vp9_writer* const bc) {
VP9_COMMON *const cm = &cpi->common;
int j;
vp9_prob pnew[VP9_INTRA_MODES - 1];
unsigned int bct[VP9_INTRA_MODES - 1][2];
for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
update_mode(bc, VP9_INTRA_MODES, vp9_intra_mode_tree, pnew,
cm->fc.y_mode_prob[j], bct,
(unsigned int *)cpi->y_mode_count[j]);
}
static void write_selected_tx_size(const VP9_COMP *cpi, TX_SIZE tx_size,
BLOCK_SIZE_TYPE bsize, vp9_writer *w) {
const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
const vp9_prob *tx_probs = get_tx_probs2(xd, &cpi->common.fc.tx_probs);
vp9_write(w, tx_size != TX_4X4, tx_probs[0]);
if (bsize >= BLOCK_16X16 && tx_size != TX_4X4) {
vp9_write(w, tx_size != TX_8X8, tx_probs[1]);
if (bsize >= BLOCK_32X32 && tx_size != TX_8X8)
vp9_write(w, tx_size != TX_16X16, tx_probs[2]);
}
}
static int write_skip_coeff(const VP9_COMP *cpi, int segment_id, MODE_INFO *m,
vp9_writer *w) {
const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
if (vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_SKIP)) {
return 1;
} else {
const int skip_coeff = m->mbmi.mb_skip_coeff;
vp9_write(w, skip_coeff, vp9_get_pred_prob_mbskip(&cpi->common, xd));
return skip_coeff;
}
}
void vp9_update_skip_probs(VP9_COMP *cpi, vp9_writer *w) {
VP9_COMMON *cm = &cpi->common;
int k;
for (k = 0; k < MBSKIP_CONTEXTS; ++k)
vp9_cond_prob_diff_update(w, &cm->fc.mbskip_probs[k],
VP9_MODE_UPDATE_PROB, cm->counts.mbskip[k]);
}
static void write_intra_mode(vp9_writer *bc, int m, const vp9_prob *p) {
write_token(bc, vp9_intra_mode_tree, p, vp9_intra_mode_encodings + m);
}
static void update_switchable_interp_probs(VP9_COMP *const cpi,
vp9_writer* const bc) {
VP9_COMMON *const pc = &cpi->common;
unsigned int branch_ct[VP9_SWITCHABLE_FILTERS + 1]
[VP9_SWITCHABLE_FILTERS - 1][2];
vp9_prob new_prob[VP9_SWITCHABLE_FILTERS + 1][VP9_SWITCHABLE_FILTERS - 1];
int i, j;
for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j) {
vp9_tree_probs_from_distribution(
vp9_switchable_interp_tree,
new_prob[j], branch_ct[j],
pc->counts.switchable_interp[j], 0);
}
for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j) {
for (i = 0; i < VP9_SWITCHABLE_FILTERS - 1; ++i) {
vp9_cond_prob_diff_update(bc, &pc->fc.switchable_interp_prob[j][i],
VP9_MODE_UPDATE_PROB, branch_ct[j][i]);
}
}
#ifdef MODE_STATS
if (!cpi->dummy_packing)
update_switchable_interp_stats(pc);
#endif
}
static void update_inter_mode_probs(VP9_COMMON *pc, vp9_writer* const bc) {
int i, j;
for (i = 0; i < INTER_MODE_CONTEXTS; ++i) {
unsigned int branch_ct[VP9_INTER_MODES - 1][2];
vp9_prob new_prob[VP9_INTER_MODES - 1];
vp9_tree_probs_from_distribution(vp9_inter_mode_tree,
new_prob, branch_ct,
pc->counts.inter_mode[i], NEARESTMV);
for (j = 0; j < VP9_INTER_MODES - 1; ++j)
vp9_cond_prob_diff_update(bc, &pc->fc.inter_mode_probs[i][j],
VP9_MODE_UPDATE_PROB, branch_ct[j]);
}
}
static void pack_mb_tokens(vp9_writer* const bc,
TOKENEXTRA **tp,
const TOKENEXTRA *const stop) {
TOKENEXTRA *p = *tp;
2010-05-18 17:58:33 +02:00
while (p < stop) {
const int t = p->token;
const struct vp9_token *const a = vp9_coef_encodings + t;
const vp9_extra_bit *const b = vp9_extra_bits + t;
int i = 0;
const vp9_prob *pp;
int v = a->value;
int n = a->len;
vp9_prob probs[ENTROPY_NODES];
if (t == EOSB_TOKEN) {
++p;
break;
}
if (t >= TWO_TOKEN) {
vp9_model_to_full_probs(p->context_tree, probs);
pp = probs;
} else {
pp = p->context_tree;
}
assert(pp != 0);
/* skip one or two nodes */
if (p->skip_eob_node) {
n -= p->skip_eob_node;
i = 2 * p->skip_eob_node;
}
2010-05-18 17:58:33 +02:00
do {
const int bb = (v >> --n) & 1;
vp9_write(bc, bb, pp[i >> 1]);
i = vp9_coef_tree[i + bb];
} while (n);
2010-05-18 17:58:33 +02:00
if (b->base_val) {
const int e = p->extra, l = b->len;
2010-05-18 17:58:33 +02:00
if (l) {
const unsigned char *pb = b->prob;
int v = e >> 1;
int n = l; /* number of bits in v, assumed nonzero */
int i = 0;
2010-05-18 17:58:33 +02:00
do {
const int bb = (v >> --n) & 1;
vp9_write(bc, bb, pb[i >> 1]);
i = b->tree[i + bb];
} while (n);
}
2010-05-18 17:58:33 +02:00
vp9_write_bit(bc, e & 1);
2010-05-18 17:58:33 +02:00
}
++p;
}
*tp = p;
2010-05-18 17:58:33 +02:00
}
static void write_sb_mv_ref(vp9_writer *w, MB_PREDICTION_MODE mode,
const vp9_prob *p) {
assert(is_inter_mode(mode));
write_token(w, vp9_inter_mode_tree, p,
&vp9_inter_mode_encodings[mode - NEARESTMV]);
}
static void write_segment_id(vp9_writer *w, const struct segmentation *seg,
int segment_id) {
if (seg->enabled && seg->update_map)
treed_write(w, vp9_segment_tree, seg->tree_probs, segment_id, 3);
2010-05-18 17:58:33 +02:00
}
// This function encodes the reference frame
static void encode_ref_frame(VP9_COMP *cpi, vp9_writer *bc) {
VP9_COMMON *const pc = &cpi->common;
MACROBLOCK *const x = &cpi->mb;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *mi = &xd->mode_info_context->mbmi;
const int segment_id = mi->segment_id;
int seg_ref_active = vp9_segfeature_active(&xd->seg, segment_id,
SEG_LVL_REF_FRAME);
// If segment level coding of this signal is disabled...
// or the segment allows multiple reference frame options
if (!seg_ref_active) {
// does the feature use compound prediction or not
// (if not specified at the frame/segment level)
if (pc->comp_pred_mode == HYBRID_PREDICTION) {
vp9_write(bc, mi->ref_frame[1] > INTRA_FRAME,
vp9_get_pred_prob_comp_inter_inter(pc, xd));
} else {
assert((mi->ref_frame[1] <= INTRA_FRAME) ==
(pc->comp_pred_mode == SINGLE_PREDICTION_ONLY));
}
if (mi->ref_frame[1] > INTRA_FRAME) {
vp9_write(bc, mi->ref_frame[0] == GOLDEN_FRAME,
vp9_get_pred_prob_comp_ref_p(pc, xd));
} else {
vp9_write(bc, mi->ref_frame[0] != LAST_FRAME,
vp9_get_pred_prob_single_ref_p1(pc, xd));
if (mi->ref_frame[0] != LAST_FRAME)
vp9_write(bc, mi->ref_frame[0] != GOLDEN_FRAME,
vp9_get_pred_prob_single_ref_p2(pc, xd));
}
} else {
assert(mi->ref_frame[1] <= INTRA_FRAME);
assert(vp9_get_segdata(&xd->seg, segment_id, SEG_LVL_REF_FRAME) ==
mi->ref_frame[0]);
}
// if using the prediction mdoel we have nothing further to do because
// the reference frame is fully coded by the segment
}
2010-05-18 17:58:33 +02:00
static void pack_inter_mode_mvs(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc) {
VP9_COMMON *const pc = &cpi->common;
const nmv_context *nmvc = &pc->fc.nmvc;
MACROBLOCK *const x = &cpi->mb;
MACROBLOCKD *const xd = &x->e_mbd;
struct segmentation *seg = &xd->seg;
MB_MODE_INFO *const mi = &m->mbmi;
const MV_REFERENCE_FRAME rf = mi->ref_frame[0];
const MB_PREDICTION_MODE mode = mi->mode;
const int segment_id = mi->segment_id;
int skip_coeff;
const BLOCK_SIZE_TYPE bsize = mi->sb_type;
const int allow_hp = xd->allow_high_precision_mv;
x->partition_info = x->pi + (m - pc->mi);
#ifdef ENTROPY_STATS
active_section = 9;
#endif
if (seg->update_map) {
if (seg->temporal_update) {
const int pred_flag = mi->seg_id_predicted;
vp9_prob pred_prob = vp9_get_pred_prob_seg_id(xd);
vp9_write(bc, pred_flag, pred_prob);
if (!pred_flag)
write_segment_id(bc, seg, segment_id);
} else {
write_segment_id(bc, seg, segment_id);
}
}
skip_coeff = write_skip_coeff(cpi, segment_id, m, bc);
2010-05-18 17:58:33 +02:00
if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
vp9_write(bc, rf != INTRA_FRAME,
vp9_get_pred_prob_intra_inter(pc, xd));
if (bsize >= BLOCK_8X8 && pc->tx_mode == TX_MODE_SELECT &&
!(rf != INTRA_FRAME &&
(skip_coeff || vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)))) {
write_selected_tx_size(cpi, mi->txfm_size, bsize, bc);
}
if (rf == INTRA_FRAME) {
#ifdef ENTROPY_STATS
active_section = 6;
#endif
if (bsize >= BLOCK_8X8) {
write_intra_mode(bc, mode, pc->fc.y_mode_prob[size_group_lookup[bsize]]);
} else {
int idx, idy;
const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
const MB_PREDICTION_MODE bm = m->bmi[idy * 2 + idx].as_mode;
write_intra_mode(bc, bm, pc->fc.y_mode_prob[0]);
}
}
}
write_intra_mode(bc, mi->uv_mode, pc->fc.uv_mode_prob[mode]);
} else {
vp9_prob *mv_ref_p;
encode_ref_frame(cpi, bc);
mv_ref_p = cpi->common.fc.inter_mode_probs[mi->mb_mode_context[rf]];
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
active_section = 3;
2010-05-18 17:58:33 +02:00
#endif
// If segment skip is not enabled code the mode.
if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)) {
if (bsize >= BLOCK_8X8) {
write_sb_mv_ref(bc, mode, mv_ref_p);
++pc->counts.inter_mode[mi->mb_mode_context[rf]]
[inter_mode_offset(mode)];
}
}
if (pc->mcomp_filter_type == SWITCHABLE) {
const int ctx = vp9_get_pred_context_switchable_interp(xd);
write_token(bc, vp9_switchable_interp_tree,
pc->fc.switchable_interp_prob[ctx],
&vp9_switchable_interp_encodings[mi->interp_filter]);
} else {
assert(mi->interp_filter == pc->mcomp_filter_type);
}
if (bsize < BLOCK_8X8) {
int j;
MB_PREDICTION_MODE blockmode;
int_mv blockmv;
const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
int idx, idy;
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
j = idy * 2 + idx;
blockmode = x->partition_info->bmi[j].mode;
blockmv = m->bmi[j].as_mv[0];
write_sb_mv_ref(bc, blockmode, mv_ref_p);
++pc->counts.inter_mode[mi->mb_mode_context[rf]]
[inter_mode_offset(blockmode)];
if (blockmode == NEWMV) {
#ifdef ENTROPY_STATS
active_section = 11;
#endif
vp9_encode_mv(cpi, bc, &blockmv.as_mv, &mi->best_mv.as_mv,
nmvc, allow_hp);
if (mi->ref_frame[1] > INTRA_FRAME)
vp9_encode_mv(cpi, bc,
&m->bmi[j].as_mv[1].as_mv,
&mi->best_second_mv.as_mv,
nmvc, allow_hp);
}
}
}
} else if (mode == NEWMV) {
#ifdef ENTROPY_STATS
active_section = 5;
#endif
vp9_encode_mv(cpi, bc, &mi->mv[0].as_mv, &mi->best_mv.as_mv,
nmvc, allow_hp);
if (mi->ref_frame[1] > INTRA_FRAME)
vp9_encode_mv(cpi, bc, &mi->mv[1].as_mv, &mi->best_second_mv.as_mv,
nmvc, allow_hp);
}
}
2010-05-18 17:58:33 +02:00
}
static void write_mb_modes_kf(const VP9_COMP *cpi, MODE_INFO *m,
vp9_writer *bc) {
const VP9_COMMON *const c = &cpi->common;
const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
const int ym = m->mbmi.mode;
const int mis = c->mode_info_stride;
const int segment_id = m->mbmi.segment_id;
if (xd->seg.update_map)
write_segment_id(bc, &xd->seg, m->mbmi.segment_id);
write_skip_coeff(cpi, segment_id, m, bc);
if (m->mbmi.sb_type >= BLOCK_8X8 && c->tx_mode == TX_MODE_SELECT)
write_selected_tx_size(cpi, m->mbmi.txfm_size, m->mbmi.sb_type, bc);
if (m->mbmi.sb_type >= BLOCK_8X8) {
const MB_PREDICTION_MODE A = above_block_mode(m, 0, mis);
const MB_PREDICTION_MODE L = xd->left_available ?
left_block_mode(m, 0) : DC_PRED;
write_intra_mode(bc, ym, vp9_kf_y_mode_prob[A][L]);
} else {
int idx, idy;
const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[m->mbmi.sb_type];
const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[m->mbmi.sb_type];
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
const int i = idy * 2 + idx;
const MB_PREDICTION_MODE A = above_block_mode(m, i, mis);
const MB_PREDICTION_MODE L = (xd->left_available || idx) ?
left_block_mode(m, i) : DC_PRED;
const int bm = m->bmi[i].as_mode;
#ifdef ENTROPY_STATS
++intra_mode_stats[A][L][bm];
#endif
write_intra_mode(bc, bm, vp9_kf_y_mode_prob[A][L]);
}
}
}
write_intra_mode(bc, m->mbmi.uv_mode, vp9_kf_uv_mode_prob[ym]);
}
static void write_modes_b(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
TOKENEXTRA **tok, TOKENEXTRA *tok_end,
int mi_row, int mi_col) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->mb.e_mbd;
if (m->mbmi.sb_type < BLOCK_8X8)
if (xd->ab_index > 0)
return;
xd->mode_info_context = m;
set_mi_row_col(&cpi->common, xd, mi_row,
1 << mi_height_log2(m->mbmi.sb_type),
mi_col, 1 << mi_width_log2(m->mbmi.sb_type));
if ((cm->frame_type == KEY_FRAME) || cm->intra_only) {
write_mb_modes_kf(cpi, m, bc);
#ifdef ENTROPY_STATS
active_section = 8;
#endif
} else {
pack_inter_mode_mvs(cpi, m, bc);
#ifdef ENTROPY_STATS
active_section = 1;
#endif
}
assert(*tok < tok_end);
pack_mb_tokens(bc, tok, tok_end);
}
static void write_modes_sb(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
TOKENEXTRA **tok, TOKENEXTRA *tok_end,
int mi_row, int mi_col,
BLOCK_SIZE_TYPE bsize) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
const int mis = cm->mode_info_stride;
int bsl = b_width_log2(bsize);
int bs = (1 << bsl) / 4; // mode_info step for subsize
int n;
PARTITION_TYPE partition = PARTITION_NONE;
BLOCK_SIZE_TYPE subsize;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
return;
partition = partition_lookup[bsl][m->mbmi.sb_type];
if (bsize < BLOCK_8X8)
if (xd->ab_index > 0)
return;
if (bsize >= BLOCK_8X8) {
int pl;
const int idx = check_bsize_coverage(cm, mi_row, mi_col, bsize);
set_partition_seg_context(cm, xd, mi_row, mi_col);
pl = partition_plane_context(xd, bsize);
// encode the partition information
if (idx == 0)
write_token(bc, vp9_partition_tree,
cm->fc.partition_prob[cm->frame_type][pl],
vp9_partition_encodings + partition);
else if (idx > 0)
vp9_write(bc, partition == PARTITION_SPLIT,
cm->fc.partition_prob[cm->frame_type][pl][idx]);
}
subsize = get_subsize(bsize, partition);
*(get_sb_index(xd, subsize)) = 0;
switch (partition) {
case PARTITION_NONE:
write_modes_b(cpi, m, bc, tok, tok_end, mi_row, mi_col);
break;
case PARTITION_HORZ:
write_modes_b(cpi, m, bc, tok, tok_end, mi_row, mi_col);
*(get_sb_index(xd, subsize)) = 1;
if ((mi_row + bs) < cm->mi_rows)
write_modes_b(cpi, m + bs * mis, bc, tok, tok_end, mi_row + bs, mi_col);
break;
case PARTITION_VERT:
write_modes_b(cpi, m, bc, tok, tok_end, mi_row, mi_col);
*(get_sb_index(xd, subsize)) = 1;
if ((mi_col + bs) < cm->mi_cols)
write_modes_b(cpi, m + bs, bc, tok, tok_end, mi_row, mi_col + bs);
break;
case PARTITION_SPLIT:
for (n = 0; n < 4; n++) {
int j = n >> 1, i = n & 0x01;
*(get_sb_index(xd, subsize)) = n;
write_modes_sb(cpi, m + j * bs * mis + i * bs, bc, tok, tok_end,
mi_row + j * bs, mi_col + i * bs, subsize);
}
break;
default:
assert(0);
}
// update partition context
if (bsize >= BLOCK_8X8 &&
(bsize == BLOCK_8X8 || partition != PARTITION_SPLIT)) {
set_partition_seg_context(cm, xd, mi_row, mi_col);
update_partition_context(xd, subsize, bsize);
}
}
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
static void write_modes(VP9_COMP *cpi, vp9_writer* const bc,
TOKENEXTRA **tok, TOKENEXTRA *tok_end) {
VP9_COMMON *const c = &cpi->common;
const int mis = c->mode_info_stride;
MODE_INFO *m, *m_ptr = c->mi;
int mi_row, mi_col;
2010-05-18 17:58:33 +02:00
m_ptr += c->cur_tile_mi_col_start + c->cur_tile_mi_row_start * mis;
for (mi_row = c->cur_tile_mi_row_start; mi_row < c->cur_tile_mi_row_end;
mi_row += 8, m_ptr += 8 * mis) {
m = m_ptr;
vp9_zero(c->left_seg_context);
for (mi_col = c->cur_tile_mi_col_start; mi_col < c->cur_tile_mi_col_end;
mi_col += MI_BLOCK_SIZE, m += MI_BLOCK_SIZE)
write_modes_sb(cpi, m, bc, tok, tok_end, mi_row, mi_col, BLOCK_64X64);
}
2010-05-18 17:58:33 +02:00
}
/* This function is used for debugging probability trees. */
static void print_prob_tree(vp9_coeff_probs *coef_probs, int block_types) {
/* print coef probability tree */
int i, j, k, l, m;
FILE *f = fopen("enc_tree_probs.txt", "a");
fprintf(f, "{\n");
for (i = 0; i < block_types; i++) {
fprintf(f, " {\n");
for (j = 0; j < REF_TYPES; ++j) {
fprintf(f, " {\n");
for (k = 0; k < COEF_BANDS; k++) {
fprintf(f, " {\n");
for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
fprintf(f, " {");
for (m = 0; m < ENTROPY_NODES; m++) {
fprintf(f, "%3u, ",
(unsigned int)(coef_probs[i][j][k][l][m]));
}
}
fprintf(f, " }\n");
}
fprintf(f, " }\n");
}
fprintf(f, " }\n");
}
fprintf(f, "}\n");
fclose(f);
}
static void build_tree_distribution(VP9_COMP *cpi, TX_SIZE tx_size) {
vp9_coeff_probs_model *coef_probs = cpi->frame_coef_probs[tx_size];
vp9_coeff_count *coef_counts = cpi->coef_counts[tx_size];
unsigned int (*eob_branch_ct)[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS] =
cpi->common.counts.eob_branch[tx_size];
vp9_coeff_stats *coef_branch_ct = cpi->frame_branch_ct[tx_size];
vp9_prob full_probs[ENTROPY_NODES];
int i, j, k, l;
for (i = 0; i < BLOCK_TYPES; ++i) {
for (j = 0; j < REF_TYPES; ++j) {
for (k = 0; k < COEF_BANDS; ++k) {
for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
if (l >= 3 && k == 0)
continue;
vp9_tree_probs_from_distribution(vp9_coef_tree,
full_probs,
coef_branch_ct[i][j][k][l],
coef_counts[i][j][k][l], 0);
vpx_memcpy(coef_probs[i][j][k][l], full_probs,
sizeof(vp9_prob) * UNCONSTRAINED_NODES);
coef_branch_ct[i][j][k][l][0][1] = eob_branch_ct[i][j][k][l] -
coef_branch_ct[i][j][k][l][0][0];
coef_probs[i][j][k][l][0] =
get_binary_prob(coef_branch_ct[i][j][k][l][0][0],
coef_branch_ct[i][j][k][l][0][1]);
#ifdef ENTROPY_STATS
if (!cpi->dummy_packing) {
int t;
for (t = 0; t < MAX_ENTROPY_TOKENS; ++t)
context_counters[tx_size][i][j][k][l][t] +=
coef_counts[i][j][k][l][t];
context_counters[tx_size][i][j][k][l][MAX_ENTROPY_TOKENS] +=
eob_branch_ct[i][j][k][l];
}
#endif
}
}
}
}
}
static void build_coeff_contexts(VP9_COMP *cpi) {
TX_SIZE t;
for (t = TX_4X4; t <= TX_32X32; t++)
build_tree_distribution(cpi, t);
2010-05-18 17:58:33 +02:00
}
static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
TX_SIZE tx_size) {
vp9_coeff_probs_model *new_frame_coef_probs = cpi->frame_coef_probs[tx_size];
vp9_coeff_probs_model *old_frame_coef_probs =
cpi->common.fc.coef_probs[tx_size];
vp9_coeff_stats *frame_branch_ct = cpi->frame_branch_ct[tx_size];
int i, j, k, l, t;
int update[2] = {0, 0};
int savings;
const int entropy_nodes_update = UNCONSTRAINED_NODES;
const int tstart = 0;
/* dry run to see if there is any udpate at all needed */
savings = 0;
for (i = 0; i < BLOCK_TYPES; ++i) {
for (j = 0; j < REF_TYPES; ++j) {
for (k = 0; k < COEF_BANDS; ++k) {
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
// int prev_coef_savings[ENTROPY_NODES] = {0};
for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
for (t = tstart; t < entropy_nodes_update; ++t) {
vp9_prob newp = new_frame_coef_probs[i][j][k][l][t];
const vp9_prob oldp = old_frame_coef_probs[i][j][k][l][t];
const vp9_prob upd = VP9_COEF_UPDATE_PROB;
int s;
int u = 0;
if (l >= 3 && k == 0)
continue;
if (t == PIVOT_NODE)
s = vp9_prob_diff_update_savings_search_model(
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
frame_branch_ct[i][j][k][l][0],
old_frame_coef_probs[i][j][k][l], &newp, upd, i, j);
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
else
s = vp9_prob_diff_update_savings_search(
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
frame_branch_ct[i][j][k][l][t], oldp, &newp, upd);
if (s > 0 && newp != oldp)
u = 1;
if (u)
savings += s - (int)(vp9_cost_zero(upd));
else
savings -= (int)(vp9_cost_zero(upd));
update[u]++;
}
}
}
}
}
// printf("Update %d %d, savings %d\n", update[0], update[1], savings);
/* Is coef updated at all */
if (update[1] == 0 || savings < 0) {
vp9_write_bit(bc, 0);
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
return;
}
vp9_write_bit(bc, 1);
for (i = 0; i < BLOCK_TYPES; ++i) {
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
for (j = 0; j < REF_TYPES; ++j) {
for (k = 0; k < COEF_BANDS; ++k) {
// int prev_coef_savings[ENTROPY_NODES] = {0};
for (l = 0; l < PREV_COEF_CONTEXTS; ++l) {
// calc probs and branch cts for this frame only
for (t = tstart; t < entropy_nodes_update; ++t) {
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
vp9_prob newp = new_frame_coef_probs[i][j][k][l][t];
vp9_prob *oldp = old_frame_coef_probs[i][j][k][l] + t;
const vp9_prob upd = VP9_COEF_UPDATE_PROB;
int s;
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
int u = 0;
if (l >= 3 && k == 0)
continue;
if (t == PIVOT_NODE)
s = vp9_prob_diff_update_savings_search_model(
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
frame_branch_ct[i][j][k][l][0],
old_frame_coef_probs[i][j][k][l], &newp, upd, i, j);
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
else
s = vp9_prob_diff_update_savings_search(
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
frame_branch_ct[i][j][k][l][t],
*oldp, &newp, upd);
if (s > 0 && newp != *oldp)
u = 1;
vp9_write(bc, u, upd);
#ifdef ENTROPY_STATS
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
if (!cpi->dummy_packing)
++tree_update_hist[tx_size][i][j][k][l][t][u];
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
#endif
if (u) {
/* send/use new probability */
vp9_write_prob_diff_update(bc, newp, *oldp);
Modeling default coef probs with distribution Replaces the default tables for single coefficient magnitudes with those obtained from an appropriate distribution. The EOB node is left unchanged. The model is represeted as a 256-size codebook where the index corresponds to the probability of the Zero or the One node. Two variations are implemented corresponding to whether the Zero node or the One-node is used as the peg. The main advantage is that the default prob tables will become considerably smaller and manageable. Besides there is substantially less risk of over-fitting for a training set. Various distributions are tried and the one that gives the best results is the family of Generalized Gaussian distributions with shape parameter 0.75. The results are within about 0.2% of fully trained tables for the Zero peg variant, and within 0.1% of the One peg variant. The forward updates are optionally (controlled by a macro) model-based, i.e. restricted to only convey probabilities from the codebook. Backward updates can also be optionally (controlled by another macro) model-based, but is turned off by default. Currently model-based forward updates work about the same as unconstrained updates, but there is a drop in performance with backward-updates being model based. The model based approach also allows the probabilities for the key frames to be adjusted from the defaults based on the base_qindex of the frame. Currently the adjustment function is a placeholder that adjusts the prob of EOB and Zero node from the nominal one at higher quality (lower qindex) or lower quality (higher qindex) ends of the range. The rest of the probabilities are then derived based on the model from the adjusted prob of zero. Change-Id: Iae050f3cbcc6d8b3f204e8dc395ae47b3b2192c9
2013-03-13 19:03:17 +01:00
*oldp = newp;
2010-05-18 17:58:33 +02:00
}
}
}
}
}
}
}
2010-05-18 17:58:33 +02:00
static void update_coef_probs(VP9_COMP* const cpi, vp9_writer* const bc) {
const TX_MODE tx_mode = cpi->common.tx_mode;
vp9_clear_system_state();
// Build the cofficient contexts based on counts collected in encode loop
build_coeff_contexts(cpi);
update_coef_probs_common(bc, cpi, TX_4X4);
// do not do this if not even allowed
if (tx_mode > ONLY_4X4)
update_coef_probs_common(bc, cpi, TX_8X8);
if (tx_mode > ALLOW_8X8)
update_coef_probs_common(bc, cpi, TX_16X16);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
if (tx_mode > ALLOW_16X16)
update_coef_probs_common(bc, cpi, TX_32X32);
2010-05-18 17:58:33 +02:00
}
static void encode_loopfilter(struct loopfilter *lf,
struct vp9_write_bit_buffer *wb) {
int i;
// Encode the loop filter level and type
vp9_wb_write_literal(wb, lf->filter_level, 6);
vp9_wb_write_literal(wb, lf->sharpness_level, 3);
// Write out loop filter deltas applied at the MB level based on mode or
// ref frame (if they are enabled).
vp9_wb_write_bit(wb, lf->mode_ref_delta_enabled);
if (lf->mode_ref_delta_enabled) {
// Do the deltas need to be updated
vp9_wb_write_bit(wb, lf->mode_ref_delta_update);
if (lf->mode_ref_delta_update) {
// Send update
for (i = 0; i < MAX_REF_LF_DELTAS; i++) {
const int delta = lf->ref_deltas[i];
// Frame level data
if (delta != lf->last_ref_deltas[i]) {
lf->last_ref_deltas[i] = delta;
vp9_wb_write_bit(wb, 1);
assert(delta != 0);
vp9_wb_write_literal(wb, abs(delta) & 0x3F, 6);
vp9_wb_write_bit(wb, delta < 0);
} else {
vp9_wb_write_bit(wb, 0);
}
}
// Send update
for (i = 0; i < MAX_MODE_LF_DELTAS; i++) {
const int delta = lf->mode_deltas[i];
if (delta != lf->last_mode_deltas[i]) {
lf->last_mode_deltas[i] = delta;
vp9_wb_write_bit(wb, 1);
assert(delta != 0);
vp9_wb_write_literal(wb, abs(delta) & 0x3F, 6);
vp9_wb_write_bit(wb, delta < 0);
} else {
vp9_wb_write_bit(wb, 0);
}
}
}
}
}
static void write_delta_q(struct vp9_write_bit_buffer *wb, int delta_q) {
if (delta_q != 0) {
vp9_wb_write_bit(wb, 1);
vp9_wb_write_literal(wb, abs(delta_q), 4);
vp9_wb_write_bit(wb, delta_q < 0);
} else {
vp9_wb_write_bit(wb, 0);
}
}
static void encode_quantization(VP9_COMMON *cm,
struct vp9_write_bit_buffer *wb) {
vp9_wb_write_literal(wb, cm->base_qindex, QINDEX_BITS);
write_delta_q(wb, cm->y_dc_delta_q);
write_delta_q(wb, cm->uv_dc_delta_q);
write_delta_q(wb, cm->uv_ac_delta_q);
}
static void encode_segmentation(VP9_COMP *cpi,
struct vp9_write_bit_buffer *wb) {
int i, j;
struct segmentation *seg = &cpi->mb.e_mbd.seg;
vp9_wb_write_bit(wb, seg->enabled);
if (!seg->enabled)
return;
// Segmentation map
vp9_wb_write_bit(wb, seg->update_map);
if (seg->update_map) {
// Select the coding strategy (temporal or spatial)
vp9_choose_segmap_coding_method(cpi);
// Write out probabilities used to decode unpredicted macro-block segments
for (i = 0; i < SEG_TREE_PROBS; i++) {
const int prob = seg->tree_probs[i];
const int update = prob != MAX_PROB;
vp9_wb_write_bit(wb, update);
if (update)
vp9_wb_write_literal(wb, prob, 8);
}
// Write out the chosen coding method.
vp9_wb_write_bit(wb, seg->temporal_update);
if (seg->temporal_update) {
for (i = 0; i < PREDICTION_PROBS; i++) {
const int prob = seg->pred_probs[i];
const int update = prob != MAX_PROB;
vp9_wb_write_bit(wb, update);
if (update)
vp9_wb_write_literal(wb, prob, 8);
}
}
}
// Segmentation data
vp9_wb_write_bit(wb, seg->update_data);
if (seg->update_data) {
vp9_wb_write_bit(wb, seg->abs_delta);
for (i = 0; i < MAX_SEGMENTS; i++) {
for (j = 0; j < SEG_LVL_MAX; j++) {
const int active = vp9_segfeature_active(seg, i, j);
vp9_wb_write_bit(wb, active);
if (active) {
const int data = vp9_get_segdata(seg, i, j);
const int data_max = vp9_seg_feature_data_max(j);
if (vp9_is_segfeature_signed(j)) {
vp9_encode_unsigned_max(wb, abs(data), data_max);
vp9_wb_write_bit(wb, data < 0);
} else {
vp9_encode_unsigned_max(wb, data, data_max);
}
}
}
}
}
}
static void encode_txfm_probs(VP9_COMP *cpi, vp9_writer *w) {
VP9_COMMON *const cm = &cpi->common;
// Mode
vp9_write_literal(w, MIN(cm->tx_mode, ALLOW_32X32), 2);
if (cm->tx_mode >= ALLOW_32X32)
vp9_write_bit(w, cm->tx_mode == TX_MODE_SELECT);
// Probabilities
if (cm->tx_mode == TX_MODE_SELECT) {
int i, j;
unsigned int ct_8x8p[TX_SIZES - 3][2];
unsigned int ct_16x16p[TX_SIZES - 2][2];
unsigned int ct_32x32p[TX_SIZES - 1][2];
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_8x8(cm->counts.tx.p8x8[i],
ct_8x8p);
for (j = 0; j < TX_SIZES - 3; j++)
vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p8x8[i][j],
VP9_MODE_UPDATE_PROB, ct_8x8p[j]);
}
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_16x16(cm->counts.tx.p16x16[i],
ct_16x16p);
for (j = 0; j < TX_SIZES - 2; j++)
vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p16x16[i][j],
VP9_MODE_UPDATE_PROB, ct_16x16p[j]);
}
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_32x32(cm->counts.tx.p32x32[i], ct_32x32p);
for (j = 0; j < TX_SIZES - 1; j++)
vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p32x32[i][j],
VP9_MODE_UPDATE_PROB, ct_32x32p[j]);
}
#ifdef MODE_STATS
if (!cpi->dummy_packing)
update_tx_count_stats(cm);
#endif
}
}
static void write_interp_filter_type(INTERPOLATIONFILTERTYPE type,
struct vp9_write_bit_buffer *wb) {
const int type_to_literal[] = { 1, 0, 2 };
vp9_wb_write_bit(wb, type == SWITCHABLE);
if (type != SWITCHABLE)
vp9_wb_write_literal(wb, type_to_literal[type], 2);
}
static void fix_mcomp_filter_type(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
if (cm->mcomp_filter_type == SWITCHABLE) {
// Check to see if only one of the filters is actually used
int count[VP9_SWITCHABLE_FILTERS];
int i, j, c = 0;
for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
count[i] = 0;
for (j = 0; j <= VP9_SWITCHABLE_FILTERS; ++j)
count[i] += cm->counts.switchable_interp[j][i];
c += (count[i] > 0);
}
if (c == 1) {
// Only one filter is used. So set the filter at frame level
for (i = 0; i < VP9_SWITCHABLE_FILTERS; ++i) {
if (count[i]) {
cm->mcomp_filter_type = i;
break;
}
}
}
}
}
static void write_tile_info(VP9_COMMON *cm, struct vp9_write_bit_buffer *wb) {
int min_log2_tile_cols, max_log2_tile_cols, ones;
vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
// columns
ones = cm->log2_tile_cols - min_log2_tile_cols;
while (ones--)
vp9_wb_write_bit(wb, 1);
if (cm->log2_tile_cols < max_log2_tile_cols)
vp9_wb_write_bit(wb, 0);
// rows
vp9_wb_write_bit(wb, cm->log2_tile_rows != 0);
if (cm->log2_tile_rows != 0)
vp9_wb_write_bit(wb, cm->log2_tile_rows != 1);
}
static int get_refresh_mask(VP9_COMP *cpi) {
// Should the GF or ARF be updated using the transmitted frame or buffer
#if CONFIG_MULTIPLE_ARF
if (!cpi->multi_arf_enabled && cpi->refresh_golden_frame &&
!cpi->refresh_alt_ref_frame) {
#else
if (cpi->refresh_golden_frame && !cpi->refresh_alt_ref_frame) {
#endif
// Preserve the previously existing golden frame and update the frame in
// the alt ref slot instead. This is highly specific to the use of
// alt-ref as a forward reference, and this needs to be generalized as
// other uses are implemented (like RTC/temporal scaling)
//
// gld_fb_idx and alt_fb_idx need to be swapped for future frames, but
// that happens in vp9_onyx_if.c:update_reference_frames() so that it can
// be done outside of the recode loop.
return (cpi->refresh_last_frame << cpi->lst_fb_idx) |
(cpi->refresh_golden_frame << cpi->alt_fb_idx);
} else {
int arf_idx = cpi->alt_fb_idx;
#if CONFIG_MULTIPLE_ARF
// Determine which ARF buffer to use to encode this ARF frame.
if (cpi->multi_arf_enabled) {
int sn = cpi->sequence_number;
arf_idx = (cpi->frame_coding_order[sn] < 0) ?
cpi->arf_buffer_idx[sn + 1] :
cpi->arf_buffer_idx[sn];
}
#endif
return (cpi->refresh_last_frame << cpi->lst_fb_idx) |
(cpi->refresh_golden_frame << cpi->gld_fb_idx) |
(cpi->refresh_alt_ref_frame << arf_idx);
}
}
static size_t encode_tiles(VP9_COMP *cpi, uint8_t *data_ptr) {
VP9_COMMON *const cm = &cpi->common;
vp9_writer residual_bc;
int tile_row, tile_col;
TOKENEXTRA *tok[4][1 << 6], *tok_end;
size_t total_size = 0;
const int tile_cols = 1 << cm->log2_tile_cols;
const int tile_rows = 1 << cm->log2_tile_rows;
vpx_memset(cm->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
mi_cols_aligned_to_sb(cm->mi_cols));
tok[0][0] = cpi->tok;
for (tile_row = 0; tile_row < tile_rows; tile_row++) {
if (tile_row)
tok[tile_row][0] = tok[tile_row - 1][tile_cols - 1] +
cpi->tok_count[tile_row - 1][tile_cols - 1];
for (tile_col = 1; tile_col < tile_cols; tile_col++)
tok[tile_row][tile_col] = tok[tile_row][tile_col - 1] +
cpi->tok_count[tile_row][tile_col - 1];
}
for (tile_row = 0; tile_row < tile_rows; tile_row++) {
vp9_get_tile_row_offsets(cm, tile_row);
for (tile_col = 0; tile_col < tile_cols; tile_col++) {
vp9_get_tile_col_offsets(cm, tile_col);
tok_end = tok[tile_row][tile_col] + cpi->tok_count[tile_row][tile_col];
if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1)
vp9_start_encode(&residual_bc, data_ptr + total_size + 4);
else
vp9_start_encode(&residual_bc, data_ptr + total_size);
write_modes(cpi, &residual_bc, &tok[tile_row][tile_col], tok_end);
assert(tok[tile_row][tile_col] == tok_end);
vp9_stop_encode(&residual_bc);
if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1) {
// size of this tile
write_be32(data_ptr + total_size, residual_bc.pos);
total_size += 4;
}
total_size += residual_bc.pos;
}
}
return total_size;
}
static void write_display_size(VP9_COMP *cpi, struct vp9_write_bit_buffer *wb) {
VP9_COMMON *const cm = &cpi->common;
const int scaling_active = cm->width != cm->display_width ||
cm->height != cm->display_height;
vp9_wb_write_bit(wb, scaling_active);
if (scaling_active) {
vp9_wb_write_literal(wb, cm->display_width - 1, 16);
vp9_wb_write_literal(wb, cm->display_height - 1, 16);
}
}
static void write_frame_size(VP9_COMP *cpi,
struct vp9_write_bit_buffer *wb) {
VP9_COMMON *const cm = &cpi->common;
vp9_wb_write_literal(wb, cm->width - 1, 16);
vp9_wb_write_literal(wb, cm->height - 1, 16);
write_display_size(cpi, wb);
}
static void write_frame_size_with_refs(VP9_COMP *cpi,
struct vp9_write_bit_buffer *wb) {
VP9_COMMON *const cm = &cpi->common;
int refs[ALLOWED_REFS_PER_FRAME] = {cpi->lst_fb_idx, cpi->gld_fb_idx,
cpi->alt_fb_idx};
int i, found = 0;
for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[cm->ref_frame_map[refs[i]]];
found = cm->width == cfg->y_crop_width &&
cm->height == cfg->y_crop_height;
vp9_wb_write_bit(wb, found);
if (found)
break;
}
if (!found) {
vp9_wb_write_literal(wb, cm->width - 1, 16);
vp9_wb_write_literal(wb, cm->height - 1, 16);
}
write_display_size(cpi, wb);
}
static void write_sync_code(struct vp9_write_bit_buffer *wb) {
vp9_wb_write_literal(wb, SYNC_CODE_0, 8);
vp9_wb_write_literal(wb, SYNC_CODE_1, 8);
vp9_wb_write_literal(wb, SYNC_CODE_2, 8);
}
static void write_uncompressed_header(VP9_COMP *cpi,
struct vp9_write_bit_buffer *wb) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->mb.e_mbd;
2010-05-18 17:58:33 +02:00
// frame marker bits
vp9_wb_write_literal(wb, 0x2, 2);
// bitstream version.
// 00 - profile 0. 4:2:0 only
// 10 - profile 1. adds 4:4:4, 4:2:2, alpha
vp9_wb_write_bit(wb, cm->version);
vp9_wb_write_bit(wb, 0);
vp9_wb_write_bit(wb, 0);
vp9_wb_write_bit(wb, cm->frame_type);
vp9_wb_write_bit(wb, cm->show_frame);
vp9_wb_write_bit(wb, cm->error_resilient_mode);
if (cm->frame_type == KEY_FRAME) {
write_sync_code(wb);
// colorspaces
// 000 - Unknown
// 001 - BT.601
// 010 - BT.709
// 011 - SMPTE-170
// 100 - SMPTE-240
// 101 - Reserved
// 110 - Reserved
// 111 - sRGB (RGB)
vp9_wb_write_literal(wb, 0, 3);
if (1 /* colorspace != sRGB */) {
vp9_wb_write_bit(wb, 0); // 0: [16, 235] (i.e. xvYCC), 1: [0, 255]
if (cm->version == 1) {
vp9_wb_write_bit(wb, cm->subsampling_x);
vp9_wb_write_bit(wb, cm->subsampling_y);
vp9_wb_write_bit(wb, 0); // has extra plane
}
} else {
assert(cm->version == 1);
vp9_wb_write_bit(wb, 0); // has extra plane
}
2010-05-18 17:58:33 +02:00
write_frame_size(cpi, wb);
} else {
const int refs[ALLOWED_REFS_PER_FRAME] = {cpi->lst_fb_idx, cpi->gld_fb_idx,
cpi->alt_fb_idx};
if (!cm->show_frame)
vp9_wb_write_bit(wb, cm->intra_only);
if (!cm->error_resilient_mode)
vp9_wb_write_literal(wb, cm->reset_frame_context, 2);
if (cm->intra_only) {
write_sync_code(wb);
vp9_wb_write_literal(wb, get_refresh_mask(cpi), NUM_REF_FRAMES);
write_frame_size(cpi, wb);
} else {
int i;
vp9_wb_write_literal(wb, get_refresh_mask(cpi), NUM_REF_FRAMES);
for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) {
vp9_wb_write_literal(wb, refs[i], NUM_REF_FRAMES_LOG2);
vp9_wb_write_bit(wb, cm->ref_frame_sign_bias[LAST_FRAME + i]);
}
write_frame_size_with_refs(cpi, wb);
vp9_wb_write_bit(wb, xd->allow_high_precision_mv);
fix_mcomp_filter_type(cpi);
write_interp_filter_type(cm->mcomp_filter_type, wb);
}
}
2010-05-18 17:58:33 +02:00
if (!cm->error_resilient_mode) {
vp9_wb_write_bit(wb, cm->refresh_frame_context);
vp9_wb_write_bit(wb, cm->frame_parallel_decoding_mode);
}
vp9_wb_write_literal(wb, cm->frame_context_idx, NUM_FRAME_CONTEXTS_LOG2);
encode_loopfilter(&xd->lf, wb);
encode_quantization(cm, wb);
encode_segmentation(cpi, wb);
write_tile_info(cm, wb);
}
static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->mb.e_mbd;
FRAME_CONTEXT *const fc = &cm->fc;
vp9_writer header_bc;
vp9_start_encode(&header_bc, data);
if (xd->lossless)
cm->tx_mode = ONLY_4X4;
else
encode_txfm_probs(cpi, &header_bc);
update_coef_probs(cpi, &header_bc);
2010-05-18 17:58:33 +02:00
#ifdef ENTROPY_STATS
active_section = 2;
2010-05-18 17:58:33 +02:00
#endif
vp9_update_skip_probs(cpi, &header_bc);
if (cm->frame_type != KEY_FRAME) {
int i;
#ifdef ENTROPY_STATS
active_section = 1;
#endif
update_inter_mode_probs(cm, &header_bc);
vp9_zero(cm->counts.inter_mode);
if (cm->mcomp_filter_type == SWITCHABLE)
update_switchable_interp_probs(cpi, &header_bc);
for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
vp9_cond_prob_diff_update(&header_bc, &fc->intra_inter_prob[i],
VP9_MODE_UPDATE_PROB,
cpi->intra_inter_count[i]);
if (cm->allow_comp_inter_inter) {
const int comp_pred_mode = cpi->common.comp_pred_mode;
const int use_compound_pred = comp_pred_mode != SINGLE_PREDICTION_ONLY;
const int use_hybrid_pred = comp_pred_mode == HYBRID_PREDICTION;
vp9_write_bit(&header_bc, use_compound_pred);
if (use_compound_pred) {
vp9_write_bit(&header_bc, use_hybrid_pred);
if (use_hybrid_pred)
for (i = 0; i < COMP_INTER_CONTEXTS; i++)
vp9_cond_prob_diff_update(&header_bc, &fc->comp_inter_prob[i],
VP9_MODE_UPDATE_PROB,
cpi->comp_inter_count[i]);
}
}
if (cm->comp_pred_mode != COMP_PREDICTION_ONLY) {
for (i = 0; i < REF_CONTEXTS; i++) {
vp9_cond_prob_diff_update(&header_bc, &fc->single_ref_prob[i][0],
VP9_MODE_UPDATE_PROB,
cpi->single_ref_count[i][0]);
vp9_cond_prob_diff_update(&header_bc, &fc->single_ref_prob[i][1],
VP9_MODE_UPDATE_PROB,
cpi->single_ref_count[i][1]);
}
}
if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY)
for (i = 0; i < REF_CONTEXTS; i++)
vp9_cond_prob_diff_update(&header_bc, &fc->comp_ref_prob[i],
VP9_MODE_UPDATE_PROB,
cpi->comp_ref_count[i]);
update_mbintra_mode_probs(cpi, &header_bc);
for (i = 0; i < NUM_PARTITION_CONTEXTS; ++i) {
vp9_prob pnew[PARTITION_TYPES - 1];
unsigned int bct[PARTITION_TYPES - 1][2];
update_mode(&header_bc, PARTITION_TYPES,
vp9_partition_tree, pnew,
fc->partition_prob[cm->frame_type][i], bct,
(unsigned int *)cpi->partition_count[i]);
}
vp9_write_nmv_probs(cpi, xd->allow_high_precision_mv, &header_bc);
}
2010-05-18 17:58:33 +02:00
vp9_stop_encode(&header_bc);
assert(header_bc.pos <= 0xffff);
2010-05-18 17:58:33 +02:00
return header_bc.pos;
}
void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, unsigned long *size) {
uint8_t *data = dest;
size_t first_part_size;
struct vp9_write_bit_buffer wb = {data, 0};
struct vp9_write_bit_buffer saved_wb;
write_uncompressed_header(cpi, &wb);
saved_wb = wb;
vp9_wb_write_literal(&wb, 0, 16); // don't know in advance first part. size
data += vp9_rb_bytes_written(&wb);
vp9_compute_update_table();
#ifdef ENTROPY_STATS
if (pc->frame_type == INTER_FRAME)
active_section = 0;
else
active_section = 7;
#endif
vp9_clear_system_state(); // __asm emms;
first_part_size = write_compressed_header(cpi, data);
data += first_part_size;
vp9_wb_write_literal(&saved_wb, first_part_size, 16);
data += encode_tiles(cpi, data);
*size = data - dest;
2010-05-18 17:58:33 +02:00
}
#ifdef ENTROPY_STATS
static void print_tree_update_for_type(FILE *f,
vp9_coeff_stats *tree_update_hist,
int block_types, const char *header) {
int i, j, k, l, m;
fprintf(f, "const vp9_coeff_prob %s = {\n", header);
for (i = 0; i < block_types; i++) {
fprintf(f, " { \n");
for (j = 0; j < REF_TYPES; j++) {
fprintf(f, " { \n");
for (k = 0; k < COEF_BANDS; k++) {
fprintf(f, " {\n");
for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
fprintf(f, " {");
for (m = 0; m < ENTROPY_NODES; m++) {
fprintf(f, "%3d, ",
get_binary_prob(tree_update_hist[i][j][k][l][m][0],
tree_update_hist[i][j][k][l][m][1]));
}
fprintf(f, "},\n");
2010-05-18 17:58:33 +02:00
}
fprintf(f, "},\n");
}
fprintf(f, " },\n");
2010-05-18 17:58:33 +02:00
}
fprintf(f, " },\n");
}
fprintf(f, "};\n");
}
void print_tree_update_probs() {
FILE *f = fopen("coefupdprob.h", "w");
fprintf(f, "\n/* Update probabilities for token entropy tree. */\n\n");
print_tree_update_for_type(f, tree_update_hist[TX_4X4], BLOCK_TYPES,
"vp9_coef_update_probs_4x4[BLOCK_TYPES]");
print_tree_update_for_type(f, tree_update_hist[TX_8X8], BLOCK_TYPES,
"vp9_coef_update_probs_8x8[BLOCK_TYPES]");
print_tree_update_for_type(f, tree_update_hist[TX_16X16], BLOCK_TYPES,
"vp9_coef_update_probs_16x16[BLOCK_TYPES]");
print_tree_update_for_type(f, tree_update_hist[TX_32X32], BLOCK_TYPES,
"vp9_coef_update_probs_32x32[BLOCK_TYPES]");
fclose(f);
f = fopen("treeupdate.bin", "wb");
fwrite(tree_update_hist, sizeof(tree_update_hist), 1, f);
fclose(f);
2010-05-18 17:58:33 +02:00
}
#endif