aom/vp9/encoder/vp9_onyx_if.c

3872 строки
122 KiB
C
Исходник Обычный вид История

2010-05-18 19:58:33 +04:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 19:58:33 +04:00
*
* 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 19:58:33 +04:00
*/
#include <math.h>
#include <stdio.h>
#include <limits.h>
#include "./vpx_config.h"
#include "./vpx_scale_rtcd.h"
#include "vpx/internal/vpx_psnr.h"
#include "vpx_ports/vpx_timer.h"
2010-05-18 19:58:33 +04:00
#include "vp9/common/vp9_alloccommon.h"
#include "vp9/common/vp9_filter.h"
#include "vp9/common/vp9_idct.h"
#if CONFIG_VP9_POSTPROC
#include "vp9/common/vp9_postproc.h"
#endif
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/common/vp9_tile_common.h"
#include "vp9/encoder/vp9_aq_cyclicrefresh.h"
#include "vp9/encoder/vp9_aq_variance.h"
#include "vp9/encoder/vp9_bitstream.h"
#include "vp9/encoder/vp9_encodeframe.h"
#include "vp9/encoder/vp9_encodemv.h"
#include "vp9/encoder/vp9_firstpass.h"
#include "vp9/encoder/vp9_mbgraph.h"
#include "vp9/encoder/vp9_onyx_int.h"
#include "vp9/encoder/vp9_picklpf.h"
#include "vp9/encoder/vp9_ratectrl.h"
#include "vp9/encoder/vp9_rdopt.h"
#include "vp9/encoder/vp9_segmentation.h"
#include "vp9/encoder/vp9_temporal_filter.h"
#include "vp9/encoder/vp9_resize.h"
#include "vp9/encoder/vp9_svc_layercontext.h"
#define ALL_INTRA_MODES 0x3FF
#define INTRA_DC_ONLY 0x01
#define INTRA_DC_TM ((1 << TM_PRED) | (1 << DC_PRED))
#define INTRA_DC_H_V ((1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED))
#define INTRA_DC_TM_H_V (INTRA_DC_TM | (1 << V_PRED) | (1 << H_PRED))
void vp9_coef_tree_initialize();
#define DEFAULT_INTERP_FILTER SWITCHABLE
#define SHARP_FILTER_QTHRESH 0 /* Q threshold for 8-tap sharp filter */
#define ALTREF_HIGH_PRECISION_MV 1 // Whether to use high precision mv
// for altref computation.
#define HIGH_PRECISION_MV_QTHRESH 200 // Q threshold for high precision
// mv. Choose a very high value for
// now so that HIGH_PRECISION is always
// chosen.
// Masks for partially or completely disabling split mode
#define DISABLE_ALL_SPLIT 0x3F
#define DISABLE_ALL_INTER_SPLIT 0x1F
#define DISABLE_COMPOUND_SPLIT 0x18
#define LAST_AND_INTRA_SPLIT_ONLY 0x1E
// Max rate target for 1080P and below encodes under normal circumstances
// (1920 * 1080 / (16 * 16)) * MAX_MB_RATE bits per MB
#define MAX_MB_RATE 250
#define MAXRATE_1080P 2025000
#if CONFIG_INTERNAL_STATS
extern double vp9_calc_ssim(YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *dest, int lumamask,
double *weight);
2010-05-18 19:58:33 +04:00
extern double vp9_calc_ssimg(YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *dest, double *ssim_y,
double *ssim_u, double *ssim_v);
2010-05-18 19:58:33 +04:00
#endif
// #define OUTPUT_YUV_REC
2010-05-18 19:58:33 +04:00
#ifdef OUTPUT_YUV_SRC
FILE *yuv_file;
#endif
#ifdef OUTPUT_YUV_REC
FILE *yuv_rec_file;
#endif
2010-05-18 19:58:33 +04:00
#if 0
FILE *framepsnr;
FILE *kf_list;
2010-05-18 19:58:33 +04:00
FILE *keyfile;
#endif
void vp9_init_quantizer(VP9_COMP *cpi);
static const double in_frame_q_adj_ratio[MAX_SEGMENTS] =
{1.0, 2.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0};
static INLINE void Scale2Ratio(VPX_SCALING mode, int *hr, int *hs) {
switch (mode) {
case NORMAL:
*hr = 1;
*hs = 1;
break;
case FOURFIVE:
*hr = 4;
*hs = 5;
break;
case THREEFIVE:
*hr = 3;
*hs = 5;
break;
case ONETWO:
*hr = 1;
*hs = 2;
break;
default:
*hr = 1;
*hs = 1;
assert(0);
break;
}
}
static void set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv) {
MACROBLOCK *const mb = &cpi->mb;
cpi->common.allow_high_precision_mv = allow_high_precision_mv;
if (cpi->common.allow_high_precision_mv) {
mb->mvcost = mb->nmvcost_hp;
mb->mvsadcost = mb->nmvsadcost_hp;
} else {
mb->mvcost = mb->nmvcost;
mb->mvsadcost = mb->nmvsadcost;
}
}
void vp9_initialize_enc() {
static int init_done = 0;
if (!init_done) {
vp9_initialize_common();
vp9_coef_tree_initialize();
vp9_tokenize_initialize();
vp9_init_quant_tables();
vp9_init_me_luts();
vp9_rc_init_minq_luts();
// init_base_skip_probs();
vp9_entropy_mv_init();
vp9_entropy_mode_init();
init_done = 1;
}
2010-05-18 19:58:33 +04:00
}
static void dealloc_compressor_data(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
// Delete sementation map
vpx_free(cpi->segmentation_map);
cpi->segmentation_map = NULL;
vpx_free(cm->last_frame_seg_map);
cm->last_frame_seg_map = NULL;
vpx_free(cpi->coding_context.last_frame_seg_map_copy);
cpi->coding_context.last_frame_seg_map_copy = NULL;
2010-05-18 19:58:33 +04:00
vpx_free(cpi->complexity_map);
cpi->complexity_map = 0;
vpx_free(cpi->cyclic_refresh.map);
cpi->cyclic_refresh.map = 0;
vpx_free(cpi->active_map);
cpi->active_map = 0;
2010-05-18 19:58:33 +04:00
vp9_free_frame_buffers(cm);
2010-05-18 19:58:33 +04:00
vp9_free_frame_buffer(&cpi->last_frame_uf);
vp9_free_frame_buffer(&cpi->scaled_source);
vp9_free_frame_buffer(&cpi->alt_ref_buffer);
vp9_lookahead_destroy(cpi->lookahead);
2010-05-18 19:58:33 +04:00
vpx_free(cpi->tok);
cpi->tok = 0;
2010-05-18 19:58:33 +04:00
// Activity mask based per mb zbin adjustments
vpx_free(cpi->mb_activity_map);
cpi->mb_activity_map = 0;
vpx_free(cpi->mb_norm_activity_map);
cpi->mb_norm_activity_map = 0;
vpx_free(cpi->above_context[0]);
cpi->above_context[0] = NULL;
2010-05-18 19:58:33 +04:00
}
// Computes a q delta (in "q index" terms) to get from a starting q value
// to a target q value
int vp9_compute_qdelta(const VP9_COMP *cpi, double qstart, double qtarget) {
const RATE_CONTROL *const rc = &cpi->rc;
int start_index = rc->worst_quality;
int target_index = rc->worst_quality;
int i;
// Convert the average q value to an index.
for (i = rc->best_quality; i < rc->worst_quality; ++i) {
start_index = i;
if (vp9_convert_qindex_to_q(i) >= qstart)
break;
}
// Convert the q target to an index
for (i = rc->best_quality; i < rc->worst_quality; ++i) {
target_index = i;
if (vp9_convert_qindex_to_q(i) >= qtarget)
break;
}
return target_index - start_index;
}
// Computes a q delta (in "q index" terms) to get from a starting q value
// to a value that should equate to the given rate ratio.
int vp9_compute_qdelta_by_rate(VP9_COMP *cpi, int base_q_index,
double rate_target_ratio) {
int i;
int target_index = cpi->rc.worst_quality;
// Look up the current projected bits per block for the base index
const int base_bits_per_mb = vp9_rc_bits_per_mb(cpi->common.frame_type,
base_q_index, 1.0);
// Find the target bits per mb based on the base value and given ratio.
const int target_bits_per_mb = (int)(rate_target_ratio * base_bits_per_mb);
// Convert the q target to an index
for (i = cpi->rc.best_quality; i < cpi->rc.worst_quality; ++i) {
target_index = i;
if (vp9_rc_bits_per_mb(cpi->common.frame_type, i, 1.0) <=
target_bits_per_mb )
break;
}
return target_index - base_q_index;
}
// This function sets up a set of segments with delta Q values around
// the baseline frame quantizer.
static void setup_in_frame_q_adj(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
struct segmentation *const seg = &cm->seg;
// Make SURE use of floating point in this function is safe.
vp9_clear_system_state();
if (cm->frame_type == KEY_FRAME ||
cpi->refresh_alt_ref_frame ||
(cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref)) {
int segment;
// Clear down the segment map
vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
// Clear down the complexity map used for rd
vpx_memset(cpi->complexity_map, 0, cm->mi_rows * cm->mi_cols);
vp9_enable_segmentation(seg);
vp9_clearall_segfeatures(seg);
// Select delta coding method
seg->abs_delta = SEGMENT_DELTADATA;
// Segment 0 "Q" feature is disabled so it defaults to the baseline Q
vp9_disable_segfeature(seg, 0, SEG_LVL_ALT_Q);
// Use some of the segments for in frame Q adjustment
for (segment = 1; segment < 2; segment++) {
const int qindex_delta =
vp9_compute_qdelta_by_rate(cpi,
cm->base_qindex,
in_frame_q_adj_ratio[segment]);
vp9_enable_segfeature(seg, segment, SEG_LVL_ALT_Q);
vp9_set_segdata(seg, segment, SEG_LVL_ALT_Q, qindex_delta);
}
}
}
static void configure_static_seg_features(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
const RATE_CONTROL *const rc = &cpi->rc;
struct segmentation *const seg = &cm->seg;
int high_q = (int)(rc->avg_q > 48.0);
int qi_delta;
// Disable and clear down for KF
if (cm->frame_type == KEY_FRAME) {
// Clear down the global segmentation map
vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
seg->update_map = 0;
seg->update_data = 0;
cpi->static_mb_pct = 0;
// Disable segmentation
vp9_disable_segmentation(seg);
// Clear down the segment features.
vp9_clearall_segfeatures(seg);
} else if (cpi->refresh_alt_ref_frame) {
// If this is an alt ref frame
// Clear down the global segmentation map
vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
seg->update_map = 0;
seg->update_data = 0;
cpi->static_mb_pct = 0;
// Disable segmentation and individual segment features by default
vp9_disable_segmentation(seg);
vp9_clearall_segfeatures(seg);
// Scan frames from current to arf frame.
// This function re-enables segmentation if appropriate.
vp9_update_mbgraph_stats(cpi);
// If segmentation was enabled set those features needed for the
// arf itself.
if (seg->enabled) {
seg->update_map = 1;
seg->update_data = 1;
qi_delta = vp9_compute_qdelta(cpi, rc->avg_q, rc->avg_q * 0.875);
vp9_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2);
vp9_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
vp9_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
vp9_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
2011-10-05 14:26:00 +04:00
// Where relevant assume segment data is delta data
seg->abs_delta = SEGMENT_DELTADATA;
}
} else if (seg->enabled) {
// All other frames if segmentation has been enabled
// First normal frame in a valid gf or alt ref group
if (rc->frames_since_golden == 0) {
// Set up segment features for normal frames in an arf group
if (rc->source_alt_ref_active) {
seg->update_map = 0;
seg->update_data = 1;
seg->abs_delta = SEGMENT_DELTADATA;
qi_delta = vp9_compute_qdelta(cpi, rc->avg_q, rc->avg_q * 1.125);
vp9_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2);
vp9_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
vp9_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
vp9_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
// Segment coding disabled for compred testing
if (high_q || (cpi->static_mb_pct == 100)) {
vp9_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
vp9_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
vp9_enable_segfeature(seg, 1, SEG_LVL_SKIP);
}
} else {
// Disable segmentation and clear down features if alt ref
// is not active for this group
vp9_disable_segmentation(seg);
vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
seg->update_map = 0;
seg->update_data = 0;
vp9_clearall_segfeatures(seg);
}
} else if (rc->is_src_frame_alt_ref) {
// Special case where we are coding over the top of a previous
// alt ref frame.
// Segment coding disabled for compred testing
// Enable ref frame features for segment 0 as well
vp9_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME);
vp9_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
// All mbs should use ALTREF_FRAME
vp9_clear_segdata(seg, 0, SEG_LVL_REF_FRAME);
vp9_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME);
vp9_clear_segdata(seg, 1, SEG_LVL_REF_FRAME);
vp9_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
// Skip all MBs if high Q (0,0 mv and skip coeffs)
if (high_q) {
vp9_enable_segfeature(seg, 0, SEG_LVL_SKIP);
vp9_enable_segfeature(seg, 1, SEG_LVL_SKIP);
}
// Enable data update
seg->update_data = 1;
} else {
// All other frames.
// No updates.. leave things as they are.
seg->update_map = 0;
seg->update_data = 0;
}
}
}
// DEBUG: Print out the segment id of each MB in the current frame.
static void print_seg_map(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
int row, col;
int map_index = 0;
FILE *statsfile = fopen("segmap.stt", "a");
fprintf(statsfile, "%10d\n", cm->current_video_frame);
for (row = 0; row < cpi->common.mi_rows; row++) {
for (col = 0; col < cpi->common.mi_cols; col++) {
fprintf(statsfile, "%10d", cpi->segmentation_map[map_index]);
map_index++;
}
fprintf(statsfile, "\n");
}
fprintf(statsfile, "\n");
fclose(statsfile);
}
static void update_reference_segmentation_map(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
MODE_INFO **mi_8x8_ptr = cm->mi_grid_visible;
uint8_t *cache_ptr = cm->last_frame_seg_map;
int row, col;
for (row = 0; row < cm->mi_rows; row++) {
MODE_INFO **mi_8x8 = mi_8x8_ptr;
uint8_t *cache = cache_ptr;
for (col = 0; col < cm->mi_cols; col++, mi_8x8++, cache++)
cache[0] = mi_8x8[0]->mbmi.segment_id;
mi_8x8_ptr += cm->mode_info_stride;
cache_ptr += cm->mi_cols;
}
}
static int is_slowest_mode(int mode) {
return (mode == MODE_SECONDPASS_BEST || mode == MODE_BESTQUALITY);
}
static void set_rd_speed_thresholds(VP9_COMP *cpi) {
SPEED_FEATURES *sf = &cpi->sf;
int i;
// Set baseline threshold values
for (i = 0; i < MAX_MODES; ++i)
sf->thresh_mult[i] = is_slowest_mode(cpi->oxcf.mode) ? -500 : 0;
sf->thresh_mult[THR_NEARESTMV] = 0;
sf->thresh_mult[THR_NEARESTG] = 0;
sf->thresh_mult[THR_NEARESTA] = 0;
sf->thresh_mult[THR_DC] += 1000;
sf->thresh_mult[THR_NEWMV] += 1000;
sf->thresh_mult[THR_NEWA] += 1000;
sf->thresh_mult[THR_NEWG] += 1000;
sf->thresh_mult[THR_NEARMV] += 1000;
sf->thresh_mult[THR_NEARA] += 1000;
sf->thresh_mult[THR_COMP_NEARESTLA] += 1000;
sf->thresh_mult[THR_COMP_NEARESTGA] += 1000;
sf->thresh_mult[THR_TM] += 1000;
sf->thresh_mult[THR_COMP_NEARLA] += 1500;
sf->thresh_mult[THR_COMP_NEWLA] += 2000;
sf->thresh_mult[THR_NEARG] += 1000;
sf->thresh_mult[THR_COMP_NEARGA] += 1500;
sf->thresh_mult[THR_COMP_NEWGA] += 2000;
sf->thresh_mult[THR_ZEROMV] += 2000;
sf->thresh_mult[THR_ZEROG] += 2000;
sf->thresh_mult[THR_ZEROA] += 2000;
sf->thresh_mult[THR_COMP_ZEROLA] += 2500;
sf->thresh_mult[THR_COMP_ZEROGA] += 2500;
sf->thresh_mult[THR_H_PRED] += 2000;
sf->thresh_mult[THR_V_PRED] += 2000;
sf->thresh_mult[THR_D45_PRED ] += 2500;
sf->thresh_mult[THR_D135_PRED] += 2500;
sf->thresh_mult[THR_D117_PRED] += 2500;
sf->thresh_mult[THR_D153_PRED] += 2500;
sf->thresh_mult[THR_D207_PRED] += 2500;
sf->thresh_mult[THR_D63_PRED] += 2500;
/* disable frame modes if flags not set */
if (!(cpi->ref_frame_flags & VP9_LAST_FLAG)) {
sf->thresh_mult[THR_NEWMV ] = INT_MAX;
sf->thresh_mult[THR_NEARESTMV] = INT_MAX;
sf->thresh_mult[THR_ZEROMV ] = INT_MAX;
sf->thresh_mult[THR_NEARMV ] = INT_MAX;
}
if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
sf->thresh_mult[THR_NEARESTG ] = INT_MAX;
sf->thresh_mult[THR_ZEROG ] = INT_MAX;
sf->thresh_mult[THR_NEARG ] = INT_MAX;
sf->thresh_mult[THR_NEWG ] = INT_MAX;
}
if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
sf->thresh_mult[THR_NEARESTA ] = INT_MAX;
sf->thresh_mult[THR_ZEROA ] = INT_MAX;
sf->thresh_mult[THR_NEARA ] = INT_MAX;
sf->thresh_mult[THR_NEWA ] = INT_MAX;
}
if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
(VP9_LAST_FLAG | VP9_ALT_FLAG)) {
sf->thresh_mult[THR_COMP_ZEROLA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARLA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWLA ] = INT_MAX;
}
if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
(VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
sf->thresh_mult[THR_COMP_ZEROGA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
sf->thresh_mult[THR_COMP_NEARGA ] = INT_MAX;
sf->thresh_mult[THR_COMP_NEWGA ] = INT_MAX;
}
}
static void set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
SPEED_FEATURES *sf = &cpi->sf;
int i;
for (i = 0; i < MAX_REFS; ++i)
sf->thresh_mult_sub8x8[i] = is_slowest_mode(cpi->oxcf.mode) ? -500 : 0;
sf->thresh_mult_sub8x8[THR_LAST] += 2500;
sf->thresh_mult_sub8x8[THR_GOLD] += 2500;
sf->thresh_mult_sub8x8[THR_ALTR] += 2500;
sf->thresh_mult_sub8x8[THR_INTRA] += 2500;
sf->thresh_mult_sub8x8[THR_COMP_LA] += 4500;
sf->thresh_mult_sub8x8[THR_COMP_GA] += 4500;
// Check for masked out split cases.
for (i = 0; i < MAX_REFS; i++) {
if (sf->disable_split_mask & (1 << i))
sf->thresh_mult_sub8x8[i] = INT_MAX;
}
// disable mode test if frame flag is not set
if (!(cpi->ref_frame_flags & VP9_LAST_FLAG))
sf->thresh_mult_sub8x8[THR_LAST] = INT_MAX;
if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG))
sf->thresh_mult_sub8x8[THR_GOLD] = INT_MAX;
if (!(cpi->ref_frame_flags & VP9_ALT_FLAG))
sf->thresh_mult_sub8x8[THR_ALTR] = INT_MAX;
if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
(VP9_LAST_FLAG | VP9_ALT_FLAG))
sf->thresh_mult_sub8x8[THR_COMP_LA] = INT_MAX;
if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
(VP9_GOLD_FLAG | VP9_ALT_FLAG))
sf->thresh_mult_sub8x8[THR_COMP_GA] = INT_MAX;
}
static void set_good_speed_feature(VP9_COMMON *cm,
SPEED_FEATURES *sf,
int speed) {
int i;
sf->adaptive_rd_thresh = 1;
sf->recode_loop = ((speed < 1) ? ALLOW_RECODE : ALLOW_RECODE_KFMAXBW);
if (speed == 1) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method = frame_is_intra_only(cm)
? USE_FULL_RD : USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ?
DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 1;
sf->auto_mv_step_size = 1;
sf->adaptive_rd_thresh = 2;
sf->recode_loop = ALLOW_RECODE_KFARFGF;
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
}
if (speed == 2) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method = frame_is_intra_only(cm)
? USE_FULL_RD : USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ?
DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 2;
sf->reference_masking = 1;
sf->auto_mv_step_size = 1;
sf->disable_filter_search_var_thresh = 50;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
sf->adjust_partitioning_from_last_frame = 1;
sf->last_partitioning_redo_frequency = 3;
sf->adaptive_rd_thresh = 2;
sf->recode_loop = ALLOW_RECODE_KFARFGF;
sf->use_lp32x32fdct = 1;
sf->mode_skip_start = 11;
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
}
if (speed == 3) {
sf->use_square_partition_only = 1;
sf->tx_size_search_method = USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = DISABLE_ALL_SPLIT;
else
sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 2;
sf->reference_masking = 1;
sf->auto_mv_step_size = 1;
sf->disable_split_var_thresh = 32;
sf->disable_filter_search_var_thresh = 100;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->adjust_partitioning_from_last_frame = 1;
sf->last_partitioning_redo_frequency = 3;
sf->use_uv_intra_rd_estimate = 1;
sf->skip_encode_sb = 1;
sf->use_lp32x32fdct = 1;
sf->subpel_iters_per_step = 1;
sf->use_fast_coef_updates = 2;
sf->use_fast_coef_costing = 1;
sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
if (speed == 4) {
sf->use_square_partition_only = 1;
sf->tx_size_search_method = USE_LARGESTALL;
sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_COMP_REFMISMATCH |
FLAG_SKIP_INTRA_LOWVAR |
FLAG_EARLY_TERMINATE;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 2;
sf->reference_masking = 1;
sf->auto_mv_step_size = 1;
sf->disable_split_var_thresh = 64;
sf->disable_filter_search_var_thresh = 200;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->adjust_partitioning_from_last_frame = 1;
sf->last_partitioning_redo_frequency = 3;
sf->use_uv_intra_rd_estimate = 1;
sf->skip_encode_sb = 1;
sf->use_lp32x32fdct = 1;
sf->subpel_iters_per_step = 1;
sf->use_fast_coef_updates = 2;
sf->use_fast_coef_costing = 1;
sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
if (speed >= 5) {
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->partition_search_type = FIXED_PARTITION;
sf->tx_size_search_method = frame_is_intra_only(cm) ?
USE_FULL_RD : USE_LARGESTALL;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_COMP_REFMISMATCH |
FLAG_SKIP_INTRA_LOWVAR |
FLAG_EARLY_TERMINATE;
sf->use_rd_breakout = 1;
sf->use_lp32x32fdct = 1;
sf->optimize_coefficients = 0;
sf->auto_mv_step_size = 1;
sf->reference_masking = 1;
sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->search_method = HEX;
sf->subpel_iters_per_step = 1;
sf->disable_split_var_thresh = 64;
sf->disable_filter_search_var_thresh = 500;
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = INTRA_DC_ONLY;
sf->intra_uv_mode_mask[i] = INTRA_DC_ONLY;
}
sf->use_fast_coef_updates = 2;
sf->use_fast_coef_costing = 1;
sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
}
static void set_rt_speed_feature(VP9_COMMON *cm,
SPEED_FEATURES *sf,
int speed) {
sf->static_segmentation = 0;
sf->adaptive_rd_thresh = 1;
sf->recode_loop = ((speed < 1) ? ALLOW_RECODE : ALLOW_RECODE_KFMAXBW);
sf->encode_breakout_thresh = 1;
sf->use_fast_coef_costing = 1;
if (speed == 1) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method =
frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ?
DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 1;
sf->auto_mv_step_size = 1;
sf->adaptive_rd_thresh = 2;
sf->recode_loop = ALLOW_RECODE_KFARFGF;
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
sf->encode_breakout_thresh = 8;
}
if (speed >= 2) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method =
frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ?
DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH
| FLAG_SKIP_INTRA_BESTINTER | FLAG_SKIP_COMP_BESTINTRA
| FLAG_SKIP_INTRA_LOWVAR;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 2;
sf->auto_mv_step_size = 1;
sf->reference_masking = 1;
sf->disable_filter_search_var_thresh = 50;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
sf->adjust_partitioning_from_last_frame = 1;
sf->last_partitioning_redo_frequency = 3;
sf->adaptive_rd_thresh = 2;
sf->recode_loop = ALLOW_RECODE_KFARFGF;
sf->use_lp32x32fdct = 1;
sf->mode_skip_start = 11;
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
sf->encode_breakout_thresh = 200;
}
if (speed >= 3) {
sf->use_square_partition_only = 1;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = DISABLE_ALL_SPLIT;
else
sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH
| FLAG_SKIP_INTRA_BESTINTER | FLAG_SKIP_COMP_BESTINTRA
| FLAG_SKIP_INTRA_LOWVAR;
sf->disable_filter_search_var_thresh = 100;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->use_uv_intra_rd_estimate = 1;
sf->skip_encode_sb = 1;
sf->subpel_iters_per_step = 1;
sf->use_fast_coef_updates = 2;
sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
sf->encode_breakout_thresh = 400;
}
if (speed >= 4) {
sf->optimize_coefficients = 0;
sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->lpf_pick = LPF_PICK_FROM_Q;
sf->encode_breakout_thresh = 700;
}
if (speed >= 5) {
int i;
sf->last_partitioning_redo_frequency = 4;
sf->adaptive_rd_thresh = 5;
sf->use_fast_coef_costing = 0;
sf->auto_min_max_partition_size = STRICT_NEIGHBORING_MIN_MAX;
sf->adjust_partitioning_from_last_frame =
cm->last_frame_type != cm->frame_type || (0 ==
(cm->current_video_frame + 1) % sf->last_partitioning_redo_frequency);
sf->subpel_force_stop = 1;
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[i] = INTRA_DC_ONLY;
}
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_ONLY;
sf->frame_parameter_update = 0;
sf->encode_breakout_thresh = 1000;
sf->search_method = FAST_HEX;
sf->disable_inter_mode_mask[BLOCK_32X32] = 1 << INTER_OFFSET(ZEROMV);
sf->disable_inter_mode_mask[BLOCK_32X64] = ~(1 << INTER_OFFSET(NEARESTMV));
sf->disable_inter_mode_mask[BLOCK_64X32] = ~(1 << INTER_OFFSET(NEARESTMV));
sf->disable_inter_mode_mask[BLOCK_64X64] = ~(1 << INTER_OFFSET(NEARESTMV));
sf->max_intra_bsize = BLOCK_32X32;
}
if (speed >= 6) {
sf->partition_check =
(cm->current_video_frame % sf->last_partitioning_redo_frequency == 1);
sf->partition_search_type = REFERENCE_PARTITION;
sf->use_nonrd_pick_mode = 1;
sf->search_method = FAST_DIAMOND;
}
if (speed >= 7) {
sf->partition_search_type = VAR_BASED_FIXED_PARTITION;
sf->use_nonrd_pick_mode = 1;
sf->search_method = FAST_DIAMOND;
}
if (speed >= 8) {
int i;
for (i = 0; i < BLOCK_SIZES; ++i)
sf->disable_inter_mode_mask[i] = 14; // only search NEARESTMV (0)
}
}
void vp9_set_speed_features(VP9_COMP *cpi) {
SPEED_FEATURES *sf = &cpi->sf;
VP9_COMMON *cm = &cpi->common;
int speed = cpi->speed;
int i;
// Convert negative speed to positive
if (speed < 0)
speed = -speed;
#if CONFIG_INTERNAL_STATS
for (i = 0; i < MAX_MODES; ++i)
cpi->mode_chosen_counts[i] = 0;
#endif
// best quality defaults
sf->frame_parameter_update = 1;
sf->search_method = NSTEP;
sf->recode_loop = ALLOW_RECODE;
sf->subpel_search_method = SUBPEL_TREE;
sf->subpel_iters_per_step = 2;
sf->subpel_force_stop = 0;
sf->optimize_coefficients = !cpi->oxcf.lossless;
sf->reduce_first_step_size = 0;
sf->auto_mv_step_size = 0;
sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
sf->comp_inter_joint_search_thresh = BLOCK_4X4;
sf->adaptive_rd_thresh = 0;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_OFF;
Tx size selection enhancements (1) Refines the modeling function and uses that to add some speed features. Specifically, intead of using a flag use_largest_txfm as a speed feature, an enum tx_size_search_method is used, of which two of the types are USE_FULL_RD and USE_LARGESTALL. Two other new types are added: USE_LARGESTINTRA (use largest only for intra) USE_LARGESTINTRA_MODELINTER (use largest for intra, and model for inter) (2) Another change is that the framework for deciding transform type is simplified to use a heuristic count based method rather than an rd based method using txfm_cache. In practice the new method is found to work just as well - with derf only -0.01 down. The new method is more compatible with the new framework where certain rd costs are based on full rd and certain others are based on modeled rd or are not computed. In this patch the existing rd based method is still kept for use in the USE_FULL_RD mode. In the other modes, the count based method is used. However the recommendation is to remove it eventually since the benefit is limited, and will remove a lot of complications in the code (3) Finally a bug is fixed with the existing use_largest_txfm speed feature that causes mismatches when the lossless mode and 4x4 WH transform is forced. Results on derf: USE_FULL_RD: +0.03% (due to change in the tables), 0% encode time reduction USE_LARGESTINTRA: -0.21%, 15% encode time reduction (this one is a pretty good compromise) USE_LARGESTINTRA_MODELINTER: -0.98%, 22% encode time reduction (currently the benefit of modeling is limited for txfm size selection, but keeping this enum as a placeholder) . USE_LARGESTALL: -1.05%, 27% encode-time reduction (same as existing use_largest_txfm speed feature). Change-Id: I4d60a5f9ce78fbc90cddf2f97ed91d8bc0d4f936
2013-06-22 03:31:12 +04:00
sf->tx_size_search_method = USE_FULL_RD;
sf->use_lp32x32fdct = 0;
sf->adaptive_motion_search = 0;
sf->adaptive_pred_interp_filter = 0;
sf->reference_masking = 0;
sf->partition_search_type = SEARCH_PARTITION;
sf->less_rectangular_check = 0;
sf->use_square_partition_only = 0;
sf->auto_min_max_partition_size = NOT_IN_USE;
sf->max_partition_size = BLOCK_64X64;
sf->min_partition_size = BLOCK_4X4;
sf->adjust_partitioning_from_last_frame = 0;
sf->last_partitioning_redo_frequency = 4;
sf->disable_split_mask = 0;
sf->mode_search_skip_flags = 0;
sf->disable_split_var_thresh = 0;
sf->disable_filter_search_var_thresh = 0;
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = ALL_INTRA_MODES;
sf->intra_uv_mode_mask[i] = ALL_INTRA_MODES;
}
sf->use_rd_breakout = 0;
sf->skip_encode_sb = 0;
sf->use_uv_intra_rd_estimate = 0;
sf->lpf_pick = LPF_PICK_FROM_FULL_IMAGE;
sf->use_fast_coef_updates = 0;
sf->use_fast_coef_costing = 0;
sf->mode_skip_start = MAX_MODES; // Mode index at which mode skip mask set
sf->use_nonrd_pick_mode = 0;
sf->encode_breakout_thresh = 0;
for (i = 0; i < BLOCK_SIZES; ++i)
sf->disable_inter_mode_mask[i] = 0;
sf->max_intra_bsize = BLOCK_64X64;
// This setting only takes effect when partition_search_type is set
// to FIXED_PARTITION.
sf->always_this_block_size = BLOCK_16X16;
switch (cpi->oxcf.mode) {
case MODE_BESTQUALITY:
case MODE_SECONDPASS_BEST: // This is the best quality mode.
cpi->diamond_search_sad = vp9_full_range_search;
break;
case MODE_FIRSTPASS:
case MODE_GOODQUALITY:
case MODE_SECONDPASS:
set_good_speed_feature(cm, sf, speed);
Tx size selection enhancements (1) Refines the modeling function and uses that to add some speed features. Specifically, intead of using a flag use_largest_txfm as a speed feature, an enum tx_size_search_method is used, of which two of the types are USE_FULL_RD and USE_LARGESTALL. Two other new types are added: USE_LARGESTINTRA (use largest only for intra) USE_LARGESTINTRA_MODELINTER (use largest for intra, and model for inter) (2) Another change is that the framework for deciding transform type is simplified to use a heuristic count based method rather than an rd based method using txfm_cache. In practice the new method is found to work just as well - with derf only -0.01 down. The new method is more compatible with the new framework where certain rd costs are based on full rd and certain others are based on modeled rd or are not computed. In this patch the existing rd based method is still kept for use in the USE_FULL_RD mode. In the other modes, the count based method is used. However the recommendation is to remove it eventually since the benefit is limited, and will remove a lot of complications in the code (3) Finally a bug is fixed with the existing use_largest_txfm speed feature that causes mismatches when the lossless mode and 4x4 WH transform is forced. Results on derf: USE_FULL_RD: +0.03% (due to change in the tables), 0% encode time reduction USE_LARGESTINTRA: -0.21%, 15% encode time reduction (this one is a pretty good compromise) USE_LARGESTINTRA_MODELINTER: -0.98%, 22% encode time reduction (currently the benefit of modeling is limited for txfm size selection, but keeping this enum as a placeholder) . USE_LARGESTALL: -1.05%, 27% encode-time reduction (same as existing use_largest_txfm speed feature). Change-Id: I4d60a5f9ce78fbc90cddf2f97ed91d8bc0d4f936
2013-06-22 03:31:12 +04:00
break;
case MODE_REALTIME:
set_rt_speed_feature(cm, sf, speed);
break;
}; /* switch */
// Set rd thresholds based on mode and speed setting
set_rd_speed_thresholds(cpi);
set_rd_speed_thresholds_sub8x8(cpi);
// Slow quant, dct and trellis not worthwhile for first pass
// so make sure they are always turned off.
if (cpi->pass == 1) {
sf->optimize_coefficients = 0;
}
// No recode for 1 pass.
if (cpi->pass == 0) {
sf->recode_loop = DISALLOW_RECODE;
sf->optimize_coefficients = 0;
}
cpi->mb.fwd_txm4x4 = vp9_fdct4x4;
if (cpi->oxcf.lossless || cpi->mb.e_mbd.lossless) {
cpi->mb.fwd_txm4x4 = vp9_fwht4x4;
}
2010-05-18 19:58:33 +04:00
if (cpi->sf.subpel_search_method == SUBPEL_TREE) {
cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree;
cpi->find_fractional_mv_step_comp = vp9_find_best_sub_pixel_comp_tree;
}
2010-05-18 19:58:33 +04:00
cpi->mb.optimize = cpi->sf.optimize_coefficients == 1 && cpi->pass != 1;
if (cpi->encode_breakout && cpi->oxcf.mode == MODE_REALTIME &&
sf->encode_breakout_thresh > cpi->encode_breakout)
cpi->encode_breakout = sf->encode_breakout_thresh;
if (sf->disable_split_mask == DISABLE_ALL_SPLIT)
sf->adaptive_pred_interp_filter = 0;
2010-05-18 19:58:33 +04:00
}
static void alloc_raw_frame_buffers(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
const VP9_CONFIG *oxcf = &cpi->oxcf;
cpi->lookahead = vp9_lookahead_init(oxcf->width, oxcf->height,
cm->subsampling_x, cm->subsampling_y,
oxcf->lag_in_frames);
if (!cpi->lookahead)
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate lag buffers");
2010-05-18 19:58:33 +04:00
if (vp9_realloc_frame_buffer(&cpi->alt_ref_buffer,
oxcf->width, oxcf->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate altref buffer");
2010-05-18 19:58:33 +04:00
}
void vp9_alloc_compressor_data(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
2010-05-18 19:58:33 +04:00
if (vp9_alloc_frame_buffers(cm, cm->width, cm->height))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffers");
2010-05-18 19:58:33 +04:00
if (vp9_alloc_frame_buffer(&cpi->last_frame_uf,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate last frame buffer");
2010-05-18 19:58:33 +04:00
if (vp9_alloc_frame_buffer(&cpi->scaled_source,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate scaled source buffer");
2010-05-18 19:58:33 +04:00
vpx_free(cpi->tok);
2010-05-18 19:58:33 +04:00
{
unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols);
2010-05-18 19:58:33 +04:00
CHECK_MEM_ERROR(cm, cpi->tok, vpx_calloc(tokens, sizeof(*cpi->tok)));
}
2010-05-18 19:58:33 +04:00
vpx_free(cpi->mb_activity_map);
CHECK_MEM_ERROR(cm, cpi->mb_activity_map,
vpx_calloc(sizeof(unsigned int),
cm->mb_rows * cm->mb_cols));
vpx_free(cpi->mb_norm_activity_map);
CHECK_MEM_ERROR(cm, cpi->mb_norm_activity_map,
vpx_calloc(sizeof(unsigned int),
cm->mb_rows * cm->mb_cols));
// 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
// block where mi unit size is 8x8.
vpx_free(cpi->above_context[0]);
CHECK_MEM_ERROR(cm, cpi->above_context[0],
vpx_calloc(2 * mi_cols_aligned_to_sb(cm->mi_cols) *
MAX_MB_PLANE,
sizeof(*cpi->above_context[0])));
2010-05-18 19:58:33 +04:00
}
static void update_frame_size(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
vp9_update_frame_size(cm);
// Update size of buffers local to this frame
if (vp9_realloc_frame_buffer(&cpi->last_frame_uf,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to reallocate last frame buffer");
if (vp9_realloc_frame_buffer(&cpi->scaled_source,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to reallocate scaled source buffer");
{
int y_stride = cpi->scaled_source.y_stride;
if (cpi->sf.search_method == NSTEP) {
vp9_init3smotion_compensation(&cpi->mb, y_stride);
} else if (cpi->sf.search_method == DIAMOND) {
vp9_init_dsmotion_compensation(&cpi->mb, y_stride);
}
}
{
int i;
for (i = 1; i < MAX_MB_PLANE; ++i) {
cpi->above_context[i] = cpi->above_context[0] +
i * sizeof(*cpi->above_context[0]) * 2 *
mi_cols_aligned_to_sb(cm->mi_cols);
cpi->mb.e_mbd.above_seg_context = cpi->common.above_seg_context;
}
}
}
// Table that converts 0-63 Q range values passed in outside to the Qindex
// range used internally.
const int q_trans[] = {
0, 4, 8, 12, 16, 20, 24, 28,
32, 36, 40, 44, 48, 52, 56, 60,
64, 68, 72, 76, 80, 84, 88, 92,
96, 100, 104, 108, 112, 116, 120, 124,
128, 132, 136, 140, 144, 148, 152, 156,
160, 164, 168, 172, 176, 180, 184, 188,
192, 196, 200, 204, 208, 212, 216, 220,
224, 228, 232, 236, 240, 244, 249, 255,
};
2010-05-18 19:58:33 +04:00
int vp9_reverse_trans(int x) {
int i;
2010-05-18 19:58:33 +04:00
for (i = 0; i < 64; i++)
if (q_trans[i] >= x)
return i;
2010-05-18 19:58:33 +04:00
return 63;
2010-05-18 19:58:33 +04:00
};
void vp9_new_framerate(VP9_COMP *cpi, double framerate) {
VP9_COMMON *const cm = &cpi->common;
RATE_CONTROL *const rc = &cpi->rc;
VP9_CONFIG *const oxcf = &cpi->oxcf;
int vbr_max_bits;
oxcf->framerate = framerate < 0.1 ? 30 : framerate;
cpi->output_framerate = cpi->oxcf.framerate;
rc->av_per_frame_bandwidth = (int)(oxcf->target_bandwidth /
cpi->output_framerate);
rc->min_frame_bandwidth = (int)(rc->av_per_frame_bandwidth *
oxcf->two_pass_vbrmin_section / 100);
2010-05-18 19:58:33 +04:00
rc->min_frame_bandwidth = MAX(rc->min_frame_bandwidth, FRAME_OVERHEAD_BITS);
// A maximum bitrate for a frame is defined.
// The baseline for this aligns with HW implementations that
// can support decode of 1080P content up to a bitrate of MAX_MB_RATE bits
// per 16x16 MB (averaged over a frame). However this limit is extended if
// a very high rate is given on the command line or the the rate cannnot
// be acheived because of a user specificed max q (e.g. when the user
// specifies lossless encode.
//
vbr_max_bits = (int)(((int64_t)rc->av_per_frame_bandwidth *
oxcf->two_pass_vbrmax_section) / 100);
rc->max_frame_bandwidth = MAX(MAX((cm->MBs * MAX_MB_RATE), MAXRATE_1080P),
vbr_max_bits);
// Set Maximum gf/arf interval
rc->max_gf_interval = 16;
2010-05-18 19:58:33 +04:00
// Extended interval for genuinely static scenes
rc->static_scene_max_gf_interval = cpi->key_frame_frequency >> 1;
2010-05-18 19:58:33 +04:00
// Special conditions when alt ref frame enabled in lagged compress mode
if (oxcf->play_alternate && oxcf->lag_in_frames) {
if (rc->max_gf_interval > oxcf->lag_in_frames - 1)
rc->max_gf_interval = oxcf->lag_in_frames - 1;
if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
rc->static_scene_max_gf_interval = oxcf->lag_in_frames - 1;
}
if (rc->max_gf_interval > rc->static_scene_max_gf_interval)
rc->max_gf_interval = rc->static_scene_max_gf_interval;
2010-05-18 19:58:33 +04:00
}
int64_t vp9_rescale(int64_t val, int64_t num, int denom) {
int64_t llnum = num;
int64_t llden = denom;
int64_t llval = val;
return (llval * llnum / llden);
}
static void set_tile_limits(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
int min_log2_tile_cols, max_log2_tile_cols;
vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
cm->log2_tile_cols = clamp(cpi->oxcf.tile_columns,
min_log2_tile_cols, max_log2_tile_cols);
cm->log2_tile_rows = cpi->oxcf.tile_rows;
}
static void init_config(struct VP9_COMP *cpi, VP9_CONFIG *oxcf) {
[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 21:35:28 +04:00
VP9_COMMON *const cm = &cpi->common;
RATE_CONTROL *const rc = &cpi->rc;
int i;
2010-05-18 19:58:33 +04:00
cpi->oxcf = *oxcf;
2010-05-18 19:58:33 +04:00
cm->version = oxcf->version;
2010-05-18 19:58:33 +04:00
cm->width = oxcf->width;
cm->height = oxcf->height;
cm->subsampling_x = 0;
cm->subsampling_y = 0;
vp9_alloc_compressor_data(cpi);
// Spatial scalability.
cpi->svc.number_spatial_layers = oxcf->ss_number_layers;
// Temporal scalability.
cpi->svc.number_temporal_layers = oxcf->ts_number_layers;
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
vp9_init_layer_context(cpi);
}
// change includes all joint functionality
vp9_change_config(cpi, oxcf);
2010-05-18 19:58:33 +04:00
// Initialize active best and worst q and average q values.
if (cpi->pass == 0 && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
rc->avg_frame_qindex[0] = cpi->oxcf.worst_allowed_q;
rc->avg_frame_qindex[1] = cpi->oxcf.worst_allowed_q;
rc->avg_frame_qindex[2] = cpi->oxcf.worst_allowed_q;
} else {
rc->avg_frame_qindex[0] = (cpi->oxcf.worst_allowed_q +
cpi->oxcf.best_allowed_q) / 2;
rc->avg_frame_qindex[1] = (cpi->oxcf.worst_allowed_q +
cpi->oxcf.best_allowed_q) / 2;
rc->avg_frame_qindex[2] = (cpi->oxcf.worst_allowed_q +
cpi->oxcf.best_allowed_q) / 2;
}
rc->last_q[0] = cpi->oxcf.best_allowed_q;
rc->last_q[1] = cpi->oxcf.best_allowed_q;
rc->last_q[2] = cpi->oxcf.best_allowed_q;
2010-05-18 19:58:33 +04:00
// Initialise the starting buffer levels
rc->buffer_level = cpi->oxcf.starting_buffer_level;
rc->bits_off_target = cpi->oxcf.starting_buffer_level;
2010-05-18 19:58:33 +04:00
rc->rolling_target_bits = rc->av_per_frame_bandwidth;
rc->rolling_actual_bits = rc->av_per_frame_bandwidth;
rc->long_rolling_target_bits = rc->av_per_frame_bandwidth;
rc->long_rolling_actual_bits = rc->av_per_frame_bandwidth;
2010-05-18 19:58:33 +04:00
rc->total_actual_bits = 0;
rc->total_target_vs_actual = 0;
2010-05-18 19:58:33 +04:00
cpi->static_mb_pct = 0;
cpi->lst_fb_idx = 0;
cpi->gld_fb_idx = 1;
cpi->alt_fb_idx = 2;
set_tile_limits(cpi);
[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 21:35:28 +04:00
cpi->fixed_divide[0] = 0;
for (i = 1; i < 512; i++)
cpi->fixed_divide[i] = 0x80000 / i;
2010-05-18 19:58:33 +04:00
}
void vp9_change_config(struct VP9_COMP *cpi, VP9_CONFIG *oxcf) {
[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 21:35:28 +04:00
VP9_COMMON *const cm = &cpi->common;
2010-05-18 19:58:33 +04:00
if (!cpi || !oxcf)
return;
2010-05-18 19:58:33 +04:00
if (cm->version != oxcf->version) {
cm->version = oxcf->version;
}
2010-05-18 19:58:33 +04:00
cpi->oxcf = *oxcf;
2010-05-18 19:58:33 +04:00
if (cpi->oxcf.cpu_used == -6)
cpi->oxcf.play_alternate = 0;
switch (cpi->oxcf.mode) {
// Real time and one pass deprecated in test code base
case MODE_GOODQUALITY:
cpi->pass = 0;
cpi->oxcf.cpu_used = clamp(cpi->oxcf.cpu_used, -5, 5);
break;
case MODE_BESTQUALITY:
cpi->pass = 0;
break;
2010-05-18 19:58:33 +04:00
case MODE_FIRSTPASS:
cpi->pass = 1;
break;
2010-05-18 19:58:33 +04:00
case MODE_SECONDPASS:
cpi->pass = 2;
cpi->oxcf.cpu_used = clamp(cpi->oxcf.cpu_used, -5, 5);
break;
2010-05-18 19:58:33 +04:00
case MODE_SECONDPASS_BEST:
cpi->pass = 2;
break;
case MODE_REALTIME:
cpi->pass = 0;
break;
}
2010-05-18 19:58:33 +04:00
cpi->oxcf.worst_allowed_q = q_trans[oxcf->worst_allowed_q];
cpi->oxcf.best_allowed_q = q_trans[oxcf->best_allowed_q];
cpi->oxcf.cq_level = q_trans[cpi->oxcf.cq_level];
2010-05-18 19:58:33 +04:00
cpi->oxcf.lossless = oxcf->lossless;
if (cpi->oxcf.lossless) {
// In lossless mode, make sure right quantizer range and correct transform
// is set.
cpi->oxcf.worst_allowed_q = 0;
cpi->oxcf.best_allowed_q = 0;
cpi->mb.e_mbd.itxm_add = vp9_iwht4x4_add;
} else {
cpi->mb.e_mbd.itxm_add = vp9_idct4x4_add;
}
cpi->rc.baseline_gf_interval = DEFAULT_GF_INTERVAL;
cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
2010-05-18 19:58:33 +04:00
cpi->refresh_golden_frame = 0;
cpi->refresh_last_frame = 1;
cm->refresh_frame_context = 1;
cm->reset_frame_context = 0;
2010-05-18 19:58:33 +04:00
vp9_reset_segment_features(&cm->seg);
set_high_precision_mv(cpi, 0);
2010-05-18 19:58:33 +04:00
{
int i;
2010-05-18 19:58:33 +04:00
for (i = 0; i < MAX_SEGMENTS; i++)
cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
}
cpi->encode_breakout = cpi->oxcf.encode_breakout;
// local file playback mode == really big buffer
if (cpi->oxcf.end_usage == USAGE_LOCAL_FILE_PLAYBACK) {
cpi->oxcf.starting_buffer_level = 60000;
cpi->oxcf.optimal_buffer_level = 60000;
cpi->oxcf.maximum_buffer_size = 240000;
}
// Convert target bandwidth from Kbit/s to Bit/s
cpi->oxcf.target_bandwidth *= 1000;
cpi->oxcf.starting_buffer_level =
vp9_rescale(cpi->oxcf.starting_buffer_level,
cpi->oxcf.target_bandwidth, 1000);
// Set or reset optimal and maximum buffer levels.
if (cpi->oxcf.optimal_buffer_level == 0)
cpi->oxcf.optimal_buffer_level = cpi->oxcf.target_bandwidth / 8;
else
cpi->oxcf.optimal_buffer_level =
vp9_rescale(cpi->oxcf.optimal_buffer_level,
cpi->oxcf.target_bandwidth, 1000);
if (cpi->oxcf.maximum_buffer_size == 0)
cpi->oxcf.maximum_buffer_size = cpi->oxcf.target_bandwidth / 8;
else
cpi->oxcf.maximum_buffer_size =
vp9_rescale(cpi->oxcf.maximum_buffer_size,
cpi->oxcf.target_bandwidth, 1000);
// Under a configuration change, where maximum_buffer_size may change,
// keep buffer level clipped to the maximum allowed buffer size.
cpi->rc.bits_off_target = MIN(cpi->rc.bits_off_target,
cpi->oxcf.maximum_buffer_size);
cpi->rc.buffer_level = MIN(cpi->rc.buffer_level,
cpi->oxcf.maximum_buffer_size);
// Set up frame rate and related parameters rate control values.
vp9_new_framerate(cpi, cpi->oxcf.framerate);
// Set absolute upper and lower quality limits
cpi->rc.worst_quality = cpi->oxcf.worst_allowed_q;
cpi->rc.best_quality = cpi->oxcf.best_allowed_q;
// active values should only be modified if out of new range
cpi->cq_target_quality = cpi->oxcf.cq_level;
cm->interp_filter = DEFAULT_INTERP_FILTER;
2010-05-18 19:58:33 +04:00
cm->display_width = cpi->oxcf.width;
cm->display_height = cpi->oxcf.height;
2010-05-18 19:58:33 +04:00
// VP8 sharpness level mapping 0-7 (vs 0-10 in general VPx dialogs)
cpi->oxcf.sharpness = MIN(7, cpi->oxcf.sharpness);
2010-05-18 19:58:33 +04:00
cpi->common.lf.sharpness_level = cpi->oxcf.sharpness;
2010-05-18 19:58:33 +04:00
if (cpi->initial_width) {
// Increasing the size of the frame beyond the first seen frame, or some
// otherwise signaled maximum size, is not supported.
// TODO(jkoleszar): exit gracefully.
assert(cm->width <= cpi->initial_width);
assert(cm->height <= cpi->initial_height);
}
update_frame_size(cpi);
2010-05-18 19:58:33 +04:00
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
vp9_update_layer_context_change_config(cpi,
(int)cpi->oxcf.target_bandwidth);
}
cpi->speed = abs(cpi->oxcf.cpu_used);
2010-05-18 19:58:33 +04:00
// Limit on lag buffers as these are not currently dynamically allocated.
if (cpi->oxcf.lag_in_frames > MAX_LAG_BUFFERS)
cpi->oxcf.lag_in_frames = MAX_LAG_BUFFERS;
2010-05-18 19:58:33 +04:00
#if CONFIG_MULTIPLE_ARF
vp9_zero(cpi->alt_ref_source);
#else
cpi->alt_ref_source = NULL;
#endif
cpi->rc.is_src_frame_alt_ref = 0;
2010-05-18 19:58:33 +04:00
#if 0
// Experimental RD Code
cpi->frame_distortion = 0;
cpi->last_frame_distortion = 0;
2010-05-18 19:58:33 +04:00
#endif
set_tile_limits(cpi);
cpi->ext_refresh_frame_flags_pending = 0;
cpi->ext_refresh_frame_context_pending = 0;
2010-05-18 19:58:33 +04:00
}
#define M_LOG2_E 0.693147180559945309417
#define log2f(x) (log (x) / (float) M_LOG2_E)
static void cal_nmvjointsadcost(int *mvjointsadcost) {
mvjointsadcost[0] = 600;
mvjointsadcost[1] = 300;
mvjointsadcost[2] = 300;
mvjointsadcost[0] = 300;
}
static void cal_nmvsadcosts(int *mvsadcost[2]) {
int i = 1;
mvsadcost[0][0] = 0;
mvsadcost[1][0] = 0;
do {
double z = 256 * (2 * (log2f(8 * i) + .6));
mvsadcost[0][i] = (int)z;
mvsadcost[1][i] = (int)z;
mvsadcost[0][-i] = (int)z;
mvsadcost[1][-i] = (int)z;
} while (++i <= MV_MAX);
}
static void cal_nmvsadcosts_hp(int *mvsadcost[2]) {
int i = 1;
mvsadcost[0][0] = 0;
mvsadcost[1][0] = 0;
do {
double z = 256 * (2 * (log2f(8 * i) + .6));
mvsadcost[0][i] = (int)z;
mvsadcost[1][i] = (int)z;
mvsadcost[0][-i] = (int)z;
mvsadcost[1][-i] = (int)z;
} while (++i <= MV_MAX);
}
static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk,
PICK_MODE_CONTEXT *ctx) {
int num_pix = num_4x4_blk << 4;
int i, k;
ctx->num_4x4_blk = num_4x4_blk;
CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
for (i = 0; i < MAX_MB_PLANE; ++i) {
for (k = 0; k < 3; ++k) {
CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
vpx_memalign(16, num_pix * sizeof(int16_t)));
CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k],
vpx_memalign(16, num_pix * sizeof(int16_t)));
CHECK_MEM_ERROR(cm, ctx->dqcoeff[i][k],
vpx_memalign(16, num_pix * sizeof(int16_t)));
CHECK_MEM_ERROR(cm, ctx->eobs[i][k],
vpx_memalign(16, num_pix * sizeof(uint16_t)));
ctx->coeff_pbuf[i][k] = ctx->coeff[i][k];
ctx->qcoeff_pbuf[i][k] = ctx->qcoeff[i][k];
ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k];
ctx->eobs_pbuf[i][k] = ctx->eobs[i][k];
}
}
}
static void free_mode_context(PICK_MODE_CONTEXT *ctx) {
int i, k;
vpx_free(ctx->zcoeff_blk);
ctx->zcoeff_blk = 0;
for (i = 0; i < MAX_MB_PLANE; ++i) {
for (k = 0; k < 3; ++k) {
vpx_free(ctx->coeff[i][k]);
ctx->coeff[i][k] = 0;
vpx_free(ctx->qcoeff[i][k]);
ctx->qcoeff[i][k] = 0;
vpx_free(ctx->dqcoeff[i][k]);
ctx->dqcoeff[i][k] = 0;
vpx_free(ctx->eobs[i][k]);
ctx->eobs[i][k] = 0;
}
}
}
static void init_pick_mode_context(VP9_COMP *cpi) {
int i;
VP9_COMMON *const cm = &cpi->common;
MACROBLOCK *const x = &cpi->mb;
for (i = 0; i < BLOCK_SIZES; ++i) {
const int num_4x4_w = num_4x4_blocks_wide_lookup[i];
const int num_4x4_h = num_4x4_blocks_high_lookup[i];
const int num_4x4_blk = MAX(4, num_4x4_w * num_4x4_h);
if (i < BLOCK_16X16) {
for (x->sb_index = 0; x->sb_index < 4; ++x->sb_index) {
for (x->mb_index = 0; x->mb_index < 4; ++x->mb_index) {
for (x->b_index = 0; x->b_index < 16 / num_4x4_blk; ++x->b_index) {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
alloc_mode_context(cm, num_4x4_blk, ctx);
}
}
}
} else if (i < BLOCK_32X32) {
for (x->sb_index = 0; x->sb_index < 4; ++x->sb_index) {
for (x->mb_index = 0; x->mb_index < 64 / num_4x4_blk; ++x->mb_index) {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
ctx->num_4x4_blk = num_4x4_blk;
alloc_mode_context(cm, num_4x4_blk, ctx);
}
}
} else if (i < BLOCK_64X64) {
for (x->sb_index = 0; x->sb_index < 256 / num_4x4_blk; ++x->sb_index) {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
ctx->num_4x4_blk = num_4x4_blk;
alloc_mode_context(cm, num_4x4_blk, ctx);
}
} else {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
ctx->num_4x4_blk = num_4x4_blk;
alloc_mode_context(cm, num_4x4_blk, ctx);
}
}
}
static void free_pick_mode_context(MACROBLOCK *x) {
int i;
for (i = 0; i < BLOCK_SIZES; ++i) {
const int num_4x4_w = num_4x4_blocks_wide_lookup[i];
const int num_4x4_h = num_4x4_blocks_high_lookup[i];
const int num_4x4_blk = MAX(4, num_4x4_w * num_4x4_h);
if (i < BLOCK_16X16) {
for (x->sb_index = 0; x->sb_index < 4; ++x->sb_index) {
for (x->mb_index = 0; x->mb_index < 4; ++x->mb_index) {
for (x->b_index = 0; x->b_index < 16 / num_4x4_blk; ++x->b_index) {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
free_mode_context(ctx);
}
}
}
} else if (i < BLOCK_32X32) {
for (x->sb_index = 0; x->sb_index < 4; ++x->sb_index) {
for (x->mb_index = 0; x->mb_index < 64 / num_4x4_blk; ++x->mb_index) {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
free_mode_context(ctx);
}
}
} else if (i < BLOCK_64X64) {
for (x->sb_index = 0; x->sb_index < 256 / num_4x4_blk; ++x->sb_index) {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
free_mode_context(ctx);
}
} else {
PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
free_mode_context(ctx);
}
}
}
VP9_COMP *vp9_create_compressor(VP9_CONFIG *oxcf) {
int i, j;
VP9_COMP *const cpi = vpx_memalign(32, sizeof(VP9_COMP));
VP9_COMMON *const cm = cpi != NULL ? &cpi->common : NULL;
RATE_CONTROL *const rc = cpi != NULL ? &cpi->rc : NULL;
2010-05-18 19:58:33 +04:00
if (!cm)
return NULL;
2010-05-18 19:58:33 +04:00
vp9_zero(*cpi);
2010-05-18 19:58:33 +04:00
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp9_remove_compressor(cpi);
return 0;
}
cm->error.setjmp = 1;
CHECK_MEM_ERROR(cm, cpi->mb.ss, vpx_calloc(sizeof(search_site),
(MAX_MVSEARCH_STEPS * 8) + 1));
vp9_rtcd();
cpi->use_svc = 0;
init_config(cpi, oxcf);
init_pick_mode_context(cpi);
cm->current_video_frame = 0;
// Set reference frame sign bias for ALTREF frame to 1 (for now)
cm->ref_frame_sign_bias[ALTREF_FRAME] = 1;
rc->baseline_gf_interval = DEFAULT_GF_INTERVAL;
cpi->gold_is_last = 0;
cpi->alt_is_last = 0;
cpi->gold_is_alt = 0;
// Create the encoder segmentation map and set all entries to 0
CHECK_MEM_ERROR(cm, cpi->segmentation_map,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
// Create a complexity map used for rd adjustment
CHECK_MEM_ERROR(cm, cpi->complexity_map,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
// Create a map used for cyclic background refresh.
CHECK_MEM_ERROR(cm, cpi->cyclic_refresh.map,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
// And a place holder structure is the coding context
// for use if we want to save and restore it
CHECK_MEM_ERROR(cm, cpi->coding_context.last_frame_seg_map_copy,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
CHECK_MEM_ERROR(cm, cpi->active_map, vpx_calloc(cm->MBs, 1));
vpx_memset(cpi->active_map, 1, cm->MBs);
cpi->active_map_enabled = 0;
for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
sizeof(cpi->mbgraph_stats[0])); i++) {
CHECK_MEM_ERROR(cm, cpi->mbgraph_stats[i].mb_stats,
vpx_calloc(cm->MBs *
sizeof(*cpi->mbgraph_stats[i].mb_stats), 1));
}
2011-10-05 14:26:00 +04:00
/*Initialize the feed-forward activity masking.*/
cpi->activity_avg = 90 << 12;
cpi->key_frame_frequency = cpi->oxcf.key_freq;
2010-05-18 19:58:33 +04:00
rc->frames_since_key = 8; // Sensible default for first frame.
rc->this_key_frame_forced = 0;
rc->next_key_frame_forced = 0;
rc->source_alt_ref_pending = 0;
rc->source_alt_ref_active = 0;
cpi->refresh_alt_ref_frame = 0;
2010-05-18 19:58:33 +04:00
#if CONFIG_MULTIPLE_ARF
// Turn multiple ARF usage on/off. This is a quick hack for the initial test
// version. It should eventually be set via the codec API.
cpi->multi_arf_enabled = 1;
if (cpi->multi_arf_enabled) {
cpi->sequence_number = 0;
cpi->frame_coding_order_period = 0;
vp9_zero(cpi->frame_coding_order);
vp9_zero(cpi->arf_buffer_idx);
}
#endif
cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
#if CONFIG_INTERNAL_STATS
cpi->b_calculate_ssimg = 0;
cpi->count = 0;
cpi->bytes = 0;
if (cpi->b_calculate_psnr) {
cpi->total_y = 0.0;
cpi->total_u = 0.0;
cpi->total_v = 0.0;
cpi->total = 0.0;
cpi->total_sq_error = 0;
cpi->total_samples = 0;
cpi->totalp_y = 0.0;
cpi->totalp_u = 0.0;
cpi->totalp_v = 0.0;
cpi->totalp = 0.0;
cpi->totalp_sq_error = 0;
cpi->totalp_samples = 0;
cpi->tot_recode_hits = 0;
cpi->summed_quality = 0;
cpi->summed_weights = 0;
cpi->summedp_quality = 0;
cpi->summedp_weights = 0;
}
if (cpi->b_calculate_ssimg) {
cpi->total_ssimg_y = 0;
cpi->total_ssimg_u = 0;
cpi->total_ssimg_v = 0;
cpi->total_ssimg_all = 0;
}
2010-05-18 19:58:33 +04:00
#endif
cpi->first_time_stamp_ever = INT64_MAX;
2010-05-18 19:58:33 +04:00
rc->frames_till_gf_update_due = 0;
2010-05-18 19:58:33 +04:00
rc->ni_av_qi = cpi->oxcf.worst_allowed_q;
rc->ni_tot_qi = 0;
rc->ni_frames = 0;
rc->tot_q = 0.0;
rc->avg_q = vp9_convert_qindex_to_q(cpi->oxcf.worst_allowed_q);
2010-05-18 19:58:33 +04:00
rc->rate_correction_factor = 1.0;
rc->key_frame_rate_correction_factor = 1.0;
rc->gf_rate_correction_factor = 1.0;
2010-05-18 19:58:33 +04:00
cal_nmvjointsadcost(cpi->mb.nmvjointsadcost);
cpi->mb.nmvcost[0] = &cpi->mb.nmvcosts[0][MV_MAX];
cpi->mb.nmvcost[1] = &cpi->mb.nmvcosts[1][MV_MAX];
cpi->mb.nmvsadcost[0] = &cpi->mb.nmvsadcosts[0][MV_MAX];
cpi->mb.nmvsadcost[1] = &cpi->mb.nmvsadcosts[1][MV_MAX];
cal_nmvsadcosts(cpi->mb.nmvsadcost);
cpi->mb.nmvcost_hp[0] = &cpi->mb.nmvcosts_hp[0][MV_MAX];
cpi->mb.nmvcost_hp[1] = &cpi->mb.nmvcosts_hp[1][MV_MAX];
cpi->mb.nmvsadcost_hp[0] = &cpi->mb.nmvsadcosts_hp[0][MV_MAX];
cpi->mb.nmvsadcost_hp[1] = &cpi->mb.nmvsadcosts_hp[1][MV_MAX];
cal_nmvsadcosts_hp(cpi->mb.nmvsadcost_hp);
2010-05-18 19:58:33 +04:00
#ifdef OUTPUT_YUV_SRC
yuv_file = fopen("bd.yuv", "ab");
2010-05-18 19:58:33 +04:00
#endif
#ifdef OUTPUT_YUV_REC
yuv_rec_file = fopen("rec.yuv", "wb");
#endif
2010-05-18 19:58:33 +04:00
#if 0
framepsnr = fopen("framepsnr.stt", "a");
kf_list = fopen("kf_list.stt", "w");
2010-05-18 19:58:33 +04:00
#endif
cpi->output_pkt_list = oxcf->output_pkt_list;
2010-05-18 19:58:33 +04:00
cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
if (cpi->pass == 1) {
vp9_init_first_pass(cpi);
} else if (cpi->pass == 2) {
const size_t packet_sz = sizeof(FIRSTPASS_STATS);
const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
2010-05-18 19:58:33 +04:00
cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
cpi->twopass.stats_in = cpi->twopass.stats_in_start;
cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1];
vp9_init_second_pass(cpi);
}
2010-05-18 19:58:33 +04:00
vp9_set_speed_features(cpi);
// Default rd threshold factors for mode selection
for (i = 0; i < BLOCK_SIZES; ++i) {
for (j = 0; j < MAX_MODES; ++j)
cpi->rd_thresh_freq_fact[i][j] = 32;
for (j = 0; j < MAX_REFS; ++j)
cpi->rd_thresh_freq_sub8x8[i][j] = 32;
}
2010-05-18 19:58:33 +04:00
#define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SVFHH, SVFHV, SVFHHV, \
SDX3F, SDX8F, SDX4DF)\
cpi->fn_ptr[BT].sdf = SDF; \
cpi->fn_ptr[BT].sdaf = SDAF; \
cpi->fn_ptr[BT].vf = VF; \
cpi->fn_ptr[BT].svf = SVF; \
cpi->fn_ptr[BT].svaf = SVAF; \
cpi->fn_ptr[BT].svf_halfpix_h = SVFHH; \
cpi->fn_ptr[BT].svf_halfpix_v = SVFHV; \
cpi->fn_ptr[BT].svf_halfpix_hv = SVFHHV; \
cpi->fn_ptr[BT].sdx3f = SDX3F; \
cpi->fn_ptr[BT].sdx8f = SDX8F; \
cpi->fn_ptr[BT].sdx4df = SDX4DF;
BFP(BLOCK_32X16, vp9_sad32x16, vp9_sad32x16_avg,
vp9_variance32x16, vp9_sub_pixel_variance32x16,
vp9_sub_pixel_avg_variance32x16, NULL, NULL,
NULL, NULL, NULL,
vp9_sad32x16x4d)
BFP(BLOCK_16X32, vp9_sad16x32, vp9_sad16x32_avg,
vp9_variance16x32, vp9_sub_pixel_variance16x32,
vp9_sub_pixel_avg_variance16x32, NULL, NULL,
NULL, NULL, NULL,
vp9_sad16x32x4d)
BFP(BLOCK_64X32, vp9_sad64x32, vp9_sad64x32_avg,
vp9_variance64x32, vp9_sub_pixel_variance64x32,
vp9_sub_pixel_avg_variance64x32, NULL, NULL,
NULL, NULL, NULL,
vp9_sad64x32x4d)
BFP(BLOCK_32X64, vp9_sad32x64, vp9_sad32x64_avg,
vp9_variance32x64, vp9_sub_pixel_variance32x64,
vp9_sub_pixel_avg_variance32x64, NULL, NULL,
NULL, NULL, NULL,
vp9_sad32x64x4d)
BFP(BLOCK_32X32, vp9_sad32x32, vp9_sad32x32_avg,
vp9_variance32x32, vp9_sub_pixel_variance32x32,
vp9_sub_pixel_avg_variance32x32, vp9_variance_halfpixvar32x32_h,
vp9_variance_halfpixvar32x32_v,
vp9_variance_halfpixvar32x32_hv, vp9_sad32x32x3, vp9_sad32x32x8,
vp9_sad32x32x4d)
BFP(BLOCK_64X64, vp9_sad64x64, vp9_sad64x64_avg,
vp9_variance64x64, vp9_sub_pixel_variance64x64,
vp9_sub_pixel_avg_variance64x64, vp9_variance_halfpixvar64x64_h,
vp9_variance_halfpixvar64x64_v,
vp9_variance_halfpixvar64x64_hv, vp9_sad64x64x3, vp9_sad64x64x8,
vp9_sad64x64x4d)
BFP(BLOCK_16X16, vp9_sad16x16, vp9_sad16x16_avg,
vp9_variance16x16, vp9_sub_pixel_variance16x16,
vp9_sub_pixel_avg_variance16x16, vp9_variance_halfpixvar16x16_h,
vp9_variance_halfpixvar16x16_v,
vp9_variance_halfpixvar16x16_hv, vp9_sad16x16x3, vp9_sad16x16x8,
vp9_sad16x16x4d)
BFP(BLOCK_16X8, vp9_sad16x8, vp9_sad16x8_avg,
vp9_variance16x8, vp9_sub_pixel_variance16x8,
vp9_sub_pixel_avg_variance16x8, NULL, NULL, NULL,
vp9_sad16x8x3, vp9_sad16x8x8, vp9_sad16x8x4d)
BFP(BLOCK_8X16, vp9_sad8x16, vp9_sad8x16_avg,
vp9_variance8x16, vp9_sub_pixel_variance8x16,
vp9_sub_pixel_avg_variance8x16, NULL, NULL, NULL,
vp9_sad8x16x3, vp9_sad8x16x8, vp9_sad8x16x4d)
BFP(BLOCK_8X8, vp9_sad8x8, vp9_sad8x8_avg,
vp9_variance8x8, vp9_sub_pixel_variance8x8,
vp9_sub_pixel_avg_variance8x8, NULL, NULL, NULL,
vp9_sad8x8x3, vp9_sad8x8x8, vp9_sad8x8x4d)
BFP(BLOCK_8X4, vp9_sad8x4, vp9_sad8x4_avg,
vp9_variance8x4, vp9_sub_pixel_variance8x4,
vp9_sub_pixel_avg_variance8x4, NULL, NULL,
NULL, NULL, vp9_sad8x4x8,
vp9_sad8x4x4d)
BFP(BLOCK_4X8, vp9_sad4x8, vp9_sad4x8_avg,
vp9_variance4x8, vp9_sub_pixel_variance4x8,
vp9_sub_pixel_avg_variance4x8, NULL, NULL,
NULL, NULL, vp9_sad4x8x8,
vp9_sad4x8x4d)
BFP(BLOCK_4X4, vp9_sad4x4, vp9_sad4x4_avg,
vp9_variance4x4, vp9_sub_pixel_variance4x4,
vp9_sub_pixel_avg_variance4x4, NULL, NULL, NULL,
vp9_sad4x4x3, vp9_sad4x4x8, vp9_sad4x4x4d)
2010-05-18 19:58:33 +04:00
cpi->full_search_sad = vp9_full_search_sad;
cpi->diamond_search_sad = vp9_diamond_search_sad;
cpi->refining_search_sad = vp9_refining_search_sad;
2010-05-18 19:58:33 +04:00
/* vp9_init_quantizer() is first called here. Add check in
* vp9_frame_init_quantizer() so that vp9_init_quantizer is only
* called later when needed. This will avoid unnecessary calls of
* vp9_init_quantizer() for every frame.
*/
vp9_init_quantizer(cpi);
vp9_loop_filter_init(cm);
cm->error.setjmp = 0;
vp9_zero(cpi->common.counts.uv_mode);
#ifdef MODE_TEST_HIT_STATS
vp9_zero(cpi->mode_test_hits);
#endif
return cpi;
2010-05-18 19:58:33 +04:00
}
void vp9_remove_compressor(VP9_COMP *cpi) {
int i;
2010-05-18 19:58:33 +04:00
if (!cpi)
return;
2010-05-18 19:58:33 +04:00
if (cpi && (cpi->common.current_video_frame > 0)) {
#if CONFIG_INTERNAL_STATS
2010-05-18 19:58:33 +04:00
vp9_clear_system_state();
// printf("\n8x8-4x4:%d-%d\n", cpi->t8x8_count, cpi->t4x4_count);
if (cpi->pass != 1) {
FILE *f = fopen("opsnr.stt", "a");
double time_encoded = (cpi->last_end_time_stamp_seen
- cpi->first_time_stamp_ever) / 10000000.000;
double total_encode_time = (cpi->time_receive_data +
cpi->time_compress_data) / 1000.000;
double dr = (double)cpi->bytes * (double) 8 / (double)1000
/ time_encoded;
if (cpi->b_calculate_psnr) {
const double total_psnr =
vpx_sse_to_psnr((double)cpi->total_samples, 255.0,
(double)cpi->total_sq_error);
const double totalp_psnr =
vpx_sse_to_psnr((double)cpi->totalp_samples, 255.0,
(double)cpi->totalp_sq_error);
const double total_ssim = 100 * pow(cpi->summed_quality /
cpi->summed_weights, 8.0);
const double totalp_ssim = 100 * pow(cpi->summedp_quality /
cpi->summedp_weights, 8.0);
fprintf(f, "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
"VPXSSIM\tVPSSIMP\t Time(ms)\n");
fprintf(f, "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t%8.0f\n",
dr, cpi->total / cpi->count, total_psnr,
cpi->totalp / cpi->count, totalp_psnr, total_ssim, totalp_ssim,
total_encode_time);
}
2010-05-18 19:58:33 +04:00
if (cpi->b_calculate_ssimg) {
fprintf(f, "BitRate\tSSIM_Y\tSSIM_U\tSSIM_V\tSSIM_A\t Time(ms)\n");
fprintf(f, "%7.2f\t%6.4f\t%6.4f\t%6.4f\t%6.4f\t%8.0f\n", dr,
cpi->total_ssimg_y / cpi->count,
cpi->total_ssimg_u / cpi->count,
cpi->total_ssimg_v / cpi->count,
cpi->total_ssimg_all / cpi->count, total_encode_time);
}
2010-05-18 19:58:33 +04:00
fclose(f);
}
2010-05-18 19:58:33 +04:00
#endif
#ifdef MODE_TEST_HIT_STATS
if (cpi->pass != 1) {
double norm_per_pixel_mode_tests = 0;
double norm_counts[BLOCK_SIZES];
int i;
int sb64_per_frame;
int norm_factors[BLOCK_SIZES] =
{256, 128, 128, 64, 32, 32, 16, 8, 8, 4, 2, 2, 1};
FILE *f = fopen("mode_hit_stats.stt", "a");
// On average, how many mode tests do we do
for (i = 0; i < BLOCK_SIZES; ++i) {
norm_counts[i] = (double)cpi->mode_test_hits[i] /
(double)norm_factors[i];
norm_per_pixel_mode_tests += norm_counts[i];
}
// Convert to a number per 64x64 and per frame
sb64_per_frame = ((cpi->common.height + 63) / 64) *
((cpi->common.width + 63) / 64);
norm_per_pixel_mode_tests =
norm_per_pixel_mode_tests /
(double)(cpi->common.current_video_frame * sb64_per_frame);
fprintf(f, "%6.4f\n", norm_per_pixel_mode_tests);
fclose(f);
}
#endif
2010-05-18 19:58:33 +04:00
#if 0
{
printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
printf("\n_frames recive_data encod_mb_row compress_frame Total\n");
printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame,
cpi->time_receive_data / 1000, cpi->time_encode_sb_row / 1000,
cpi->time_compress_data / 1000,
(cpi->time_receive_data + cpi->time_compress_data) / 1000);
}
2010-05-18 19:58:33 +04:00
#endif
}
2010-05-18 19:58:33 +04:00
free_pick_mode_context(&cpi->mb);
dealloc_compressor_data(cpi);
vpx_free(cpi->mb.ss);
vpx_free(cpi->tok);
2010-05-18 19:58:33 +04:00
for (i = 0; i < sizeof(cpi->mbgraph_stats) /
sizeof(cpi->mbgraph_stats[0]); ++i) {
vpx_free(cpi->mbgraph_stats[i].mb_stats);
}
2011-10-05 14:26:00 +04:00
vp9_remove_common(&cpi->common);
vpx_free(cpi);
2010-05-18 19:58:33 +04:00
#ifdef OUTPUT_YUV_SRC
fclose(yuv_file);
2010-05-18 19:58:33 +04:00
#endif
#ifdef OUTPUT_YUV_REC
fclose(yuv_rec_file);
#endif
2010-05-18 19:58:33 +04:00
#if 0
if (keyfile)
fclose(keyfile);
2010-05-18 19:58:33 +04:00
if (framepsnr)
fclose(framepsnr);
2010-05-18 19:58:33 +04:00
if (kf_list)
fclose(kf_list);
2010-05-18 19:58:33 +04:00
#endif
}
static uint64_t calc_plane_error(const uint8_t *orig, int orig_stride,
const uint8_t *recon, int recon_stride,
unsigned int cols, unsigned int rows) {
unsigned int row, col;
uint64_t total_sse = 0;
int diff;
for (row = 0; row + 16 <= rows; row += 16) {
for (col = 0; col + 16 <= cols; col += 16) {
unsigned int sse;
vp9_mse16x16(orig + col, orig_stride, recon + col, recon_stride, &sse);
total_sse += sse;
}
/* Handle odd-sized width */
if (col < cols) {
unsigned int border_row, border_col;
const uint8_t *border_orig = orig;
const uint8_t *border_recon = recon;
for (border_row = 0; border_row < 16; border_row++) {
for (border_col = col; border_col < cols; border_col++) {
diff = border_orig[border_col] - border_recon[border_col];
total_sse += diff * diff;
2010-05-18 19:58:33 +04:00
}
border_orig += orig_stride;
border_recon += recon_stride;
}
2010-05-18 19:58:33 +04:00
}
orig += orig_stride * 16;
recon += recon_stride * 16;
}
2010-05-18 19:58:33 +04:00
/* Handle odd-sized height */
for (; row < rows; row++) {
for (col = 0; col < cols; col++) {
diff = orig[col] - recon[col];
total_sse += diff * diff;
2010-05-18 19:58:33 +04:00
}
orig += orig_stride;
recon += recon_stride;
}
return total_sse;
2010-05-18 19:58:33 +04:00
}
typedef struct {
double psnr[4]; // total/y/u/v
uint64_t sse[4]; // total/y/u/v
uint32_t samples[4]; // total/y/u/v
} PSNR_STATS;
static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
PSNR_STATS *psnr) {
const int widths[3] = {a->y_width, a->uv_width, a->uv_width };
const int heights[3] = {a->y_height, a->uv_height, a->uv_height};
const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer };
const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer };
const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
int i;
uint64_t total_sse = 0;
uint32_t total_samples = 0;
2010-05-18 19:58:33 +04:00
for (i = 0; i < 3; ++i) {
const int w = widths[i];
const int h = heights[i];
const uint32_t samples = w * h;
const uint64_t sse = calc_plane_error(a_planes[i], a_strides[i],
b_planes[i], b_strides[i],
w, h);
psnr->sse[1 + i] = sse;
psnr->samples[1 + i] = samples;
psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, 255.0, (double)sse);
total_sse += sse;
total_samples += samples;
}
psnr->sse[0] = total_sse;
psnr->samples[0] = total_samples;
psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, 255.0,
(double)total_sse);
2010-05-18 19:58:33 +04:00
}
static void generate_psnr_packet(VP9_COMP *cpi) {
struct vpx_codec_cx_pkt pkt;
int i;
PSNR_STATS psnr;
calc_psnr(cpi->Source, cpi->common.frame_to_show, &psnr);
for (i = 0; i < 4; ++i) {
pkt.data.psnr.samples[i] = psnr.samples[i];
pkt.data.psnr.sse[i] = psnr.sse[i];
pkt.data.psnr.psnr[i] = psnr.psnr[i];
}
pkt.kind = VPX_CODEC_PSNR_PKT;
vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
}
2010-05-18 19:58:33 +04:00
int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags) {
if (ref_frame_flags > 7)
return -1;
2010-05-18 19:58:33 +04:00
cpi->ref_frame_flags = ref_frame_flags;
return 0;
2010-05-18 19:58:33 +04:00
}
void vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags) {
cpi->ext_refresh_golden_frame = (ref_frame_flags & VP9_GOLD_FLAG) != 0;
cpi->ext_refresh_alt_ref_frame = (ref_frame_flags & VP9_ALT_FLAG) != 0;
cpi->ext_refresh_last_frame = (ref_frame_flags & VP9_LAST_FLAG) != 0;
cpi->ext_refresh_frame_flags_pending = 1;
2010-05-18 19:58:33 +04:00
}
static YV12_BUFFER_CONFIG *get_vp9_ref_frame_buffer(VP9_COMP *cpi,
VP9_REFFRAME ref_frame_flag) {
MV_REFERENCE_FRAME ref_frame = NONE;
if (ref_frame_flag == VP9_LAST_FLAG)
ref_frame = LAST_FRAME;
else if (ref_frame_flag == VP9_GOLD_FLAG)
ref_frame = GOLDEN_FRAME;
else if (ref_frame_flag == VP9_ALT_FLAG)
ref_frame = ALTREF_FRAME;
return ref_frame == NONE ? NULL : get_ref_frame_buffer(cpi, ref_frame);
}
2010-05-18 19:58:33 +04:00
int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
YV12_BUFFER_CONFIG *cfg = get_vp9_ref_frame_buffer(cpi, ref_frame_flag);
if (cfg) {
vp8_yv12_copy_frame(cfg, sd);
return 0;
} else {
return -1;
}
2010-05-18 19:58:33 +04:00
}
int vp9_get_reference_enc(VP9_COMP *cpi, int index, YV12_BUFFER_CONFIG **fb) {
VP9_COMMON *cm = &cpi->common;
if (index < 0 || index >= REF_FRAMES)
return -1;
*fb = &cm->frame_bufs[cm->ref_frame_map[index]].buf;
return 0;
}
int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
YV12_BUFFER_CONFIG *cfg = get_vp9_ref_frame_buffer(cpi, ref_frame_flag);
if (cfg) {
vp8_yv12_copy_frame(sd, cfg);
return 0;
} else {
return -1;
}
}
int vp9_update_entropy(VP9_COMP * cpi, int update) {
cpi->ext_refresh_frame_context = update;
cpi->ext_refresh_frame_context_pending = 1;
return 0;
}
2010-05-18 19:58:33 +04:00
#ifdef OUTPUT_YUV_SRC
void vp9_write_yuv_frame(YV12_BUFFER_CONFIG *s) {
uint8_t *src = s->y_buffer;
int h = s->y_height;
do {
fwrite(src, s->y_width, 1, yuv_file);
src += s->y_stride;
} while (--h);
src = s->u_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, yuv_file);
src += s->uv_stride;
} while (--h);
src = s->v_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, yuv_file);
src += s->uv_stride;
} while (--h);
}
#endif
2010-05-18 19:58:33 +04:00
#ifdef OUTPUT_YUV_REC
void vp9_write_yuv_rec_frame(VP9_COMMON *cm) {
YV12_BUFFER_CONFIG *s = cm->frame_to_show;
uint8_t *src = s->y_buffer;
int h = cm->height;
do {
fwrite(src, s->y_width, 1, yuv_rec_file);
src += s->y_stride;
} while (--h);
src = s->u_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, yuv_rec_file);
src += s->uv_stride;
} while (--h);
src = s->v_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, yuv_rec_file);
src += s->uv_stride;
} while (--h);
#if CONFIG_ALPHA
if (s->alpha_buffer) {
src = s->alpha_buffer;
h = s->alpha_height;
do {
fwrite(src, s->alpha_width, 1, yuv_rec_file);
src += s->alpha_stride;
} while (--h);
}
#endif
[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 21:35:28 +04:00
fflush(yuv_rec_file);
}
#endif
static void scale_and_extend_frame_nonnormative(YV12_BUFFER_CONFIG *src_fb,
YV12_BUFFER_CONFIG *dst_fb) {
const int in_w = src_fb->y_crop_width;
const int in_h = src_fb->y_crop_height;
const int out_w = dst_fb->y_crop_width;
const int out_h = dst_fb->y_crop_height;
const int in_w_uv = src_fb->uv_crop_width;
const int in_h_uv = src_fb->uv_crop_height;
const int out_w_uv = dst_fb->uv_crop_width;
const int out_h_uv = dst_fb->uv_crop_height;
int i;
uint8_t *srcs[4] = {src_fb->y_buffer, src_fb->u_buffer, src_fb->v_buffer,
src_fb->alpha_buffer};
int src_strides[4] = {src_fb->y_stride, src_fb->uv_stride, src_fb->uv_stride,
src_fb->alpha_stride};
uint8_t *dsts[4] = {dst_fb->y_buffer, dst_fb->u_buffer, dst_fb->v_buffer,
dst_fb->alpha_buffer};
int dst_strides[4] = {dst_fb->y_stride, dst_fb->uv_stride, dst_fb->uv_stride,
dst_fb->alpha_stride};
for (i = 0; i < MAX_MB_PLANE; ++i) {
if (i == 0 || i == 3) {
// Y and alpha planes
vp9_resize_plane(srcs[i], in_h, in_w, src_strides[i],
dsts[i], out_h, out_w, dst_strides[i]);
} else {
// Chroma planes
vp9_resize_plane(srcs[i], in_h_uv, in_w_uv, src_strides[i],
dsts[i], out_h_uv, out_w_uv, dst_strides[i]);
}
}
vp8_yv12_extend_frame_borders(dst_fb);
}
static void scale_and_extend_frame(YV12_BUFFER_CONFIG *src_fb,
YV12_BUFFER_CONFIG *dst_fb) {
const int in_w = src_fb->y_crop_width;
const int in_h = src_fb->y_crop_height;
const int out_w = dst_fb->y_crop_width;
const int out_h = dst_fb->y_crop_height;
int x, y, i;
uint8_t *srcs[4] = {src_fb->y_buffer, src_fb->u_buffer, src_fb->v_buffer,
src_fb->alpha_buffer};
int src_strides[4] = {src_fb->y_stride, src_fb->uv_stride, src_fb->uv_stride,
src_fb->alpha_stride};
uint8_t *dsts[4] = {dst_fb->y_buffer, dst_fb->u_buffer, dst_fb->v_buffer,
dst_fb->alpha_buffer};
int dst_strides[4] = {dst_fb->y_stride, dst_fb->uv_stride, dst_fb->uv_stride,
dst_fb->alpha_stride};
for (y = 0; y < out_h; y += 16) {
for (x = 0; x < out_w; x += 16) {
for (i = 0; i < MAX_MB_PLANE; ++i) {
const int factor = (i == 0 || i == 3 ? 1 : 2);
const int x_q4 = x * (16 / factor) * in_w / out_w;
const int y_q4 = y * (16 / factor) * in_h / out_h;
const int src_stride = src_strides[i];
const int dst_stride = dst_strides[i];
uint8_t *src = srcs[i] + y / factor * in_h / out_h * src_stride +
x / factor * in_w / out_w;
uint8_t *dst = dsts[i] + y / factor * dst_stride + x / factor;
vp9_convolve8(src, src_stride, dst, dst_stride,
vp9_sub_pel_filters_8[x_q4 & 0xf], 16 * in_w / out_w,
vp9_sub_pel_filters_8[y_q4 & 0xf], 16 * in_h / out_h,
16 / factor, 16 / factor);
}
}
}
vp8_yv12_extend_frame_borders(dst_fb);
}
static int find_fp_qindex() {
int i;
for (i = 0; i < QINDEX_RANGE; i++) {
if (vp9_convert_qindex_to_q(i) >= 30.0) {
break;
}
}
experiment extending the quantizer range Prior to this change, VP8 min quantizer is 4, which caps the highest quality around 51DB. This experimental change extends the min quantizer to 1, removes the cap and allows the highest quality to be around ~73DB, consistent with the fdct/idct round trip error. To test this change, at configure time use options: --enable-experimental --enable-extend_qrange The following is a brief log of changes in each of the patch sets patch set 1: In this commit, the quantization/dequantization constants are kept unchanged, instead scaling factor 4 is rolled into fdct/idct. Fixed Q0 encoding tests on mobile: Before: 9560.567kbps Overall PSNR:50.255DB VPXSSIM:98.288 Now: 18035.774kbps Overall PSNR:73.022DB VPXSSIM:99.991 patch set 2: regenerated dc/ac quantizer lookup tables based on the scaling factor rolled in the fdct/idct. Also slightly extended the range towards the high quantizer end. patch set 3: slightly tweaked the quantizer tables and generated bits_per_mb table based on Paul's suggestions. patch set 4: fix a typo in idct, re-calculated tables relating active max Q to active min Q patch set 5: added rdmult lookup table based on Q patch set 6: fix rdmult scale: dct coefficient has scaled up by 4 patch set 7: make transform coefficients to be within 16bits patch set 8: normalize 2nd order quantizers patch set 9: fix mis-spellings patch set 10: change the configure script and macros to allow experimental code to be enabled at configure time with --enable-extend_qrange patch set 11: rebase for merge Change-Id: Ib50641ddd44aba2a52ed890222c309faa31cc59c
2010-12-02 02:50:14 +03:00
if (i == QINDEX_RANGE)
i--;
return i;
}
2010-05-18 19:58:33 +04:00
#define WRITE_RECON_BUFFER 0
#if WRITE_RECON_BUFFER
void write_cx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame) {
FILE *yframe;
int i;
char filename[255];
2010-05-18 19:58:33 +04:00
snprintf(filename, sizeof(filename), "cx\\y%04d.raw", this_frame);
yframe = fopen(filename, "wb");
2010-05-18 19:58:33 +04:00
for (i = 0; i < frame->y_height; i++)
fwrite(frame->y_buffer + i * frame->y_stride,
frame->y_width, 1, yframe);
2010-05-18 19:58:33 +04:00
fclose(yframe);
snprintf(filename, sizeof(filename), "cx\\u%04d.raw", this_frame);
yframe = fopen(filename, "wb");
2010-05-18 19:58:33 +04:00
for (i = 0; i < frame->uv_height; i++)
fwrite(frame->u_buffer + i * frame->uv_stride,
frame->uv_width, 1, yframe);
2010-05-18 19:58:33 +04:00
fclose(yframe);
snprintf(filename, sizeof(filename), "cx\\v%04d.raw", this_frame);
yframe = fopen(filename, "wb");
2010-05-18 19:58:33 +04:00
for (i = 0; i < frame->uv_height; i++)
fwrite(frame->v_buffer + i * frame->uv_stride,
frame->uv_width, 1, yframe);
2010-05-18 19:58:33 +04:00
fclose(yframe);
2010-05-18 19:58:33 +04:00
}
#endif
// Function to test for conditions that indicate we should loop
// back and recode a frame.
static int recode_loop_test(const VP9_COMP *cpi,
int high_limit, int low_limit,
int q, int maxq, int minq) {
const VP9_COMMON *const cm = &cpi->common;
const RATE_CONTROL *const rc = &cpi->rc;
int force_recode = 0;
// Special case trap if maximum allowed frame size exceeded.
if (rc->projected_frame_size > rc->max_frame_bandwidth) {
force_recode = 1;
// Is frame recode allowed.
// Yes if either recode mode 1 is selected or mode 2 is selected
// and the frame is a key frame, golden frame or alt_ref_frame
} else if ((cpi->sf.recode_loop == ALLOW_RECODE) ||
((cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF) &&
(cm->frame_type == KEY_FRAME ||
cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame))) {
// General over and under shoot tests
if ((rc->projected_frame_size > high_limit && q < maxq) ||
(rc->projected_frame_size < low_limit && q > minq)) {
force_recode = 1;
} else if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) {
// Deal with frame undershoot and whether or not we are
// below the automatically set cq level.
if (q > cpi->cq_target_quality &&
rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) {
force_recode = 1;
}
}
}
return force_recode;
}
void vp9_update_reference_frames(VP9_COMP *cpi) {
VP9_COMMON * const cm = &cpi->common;
// At this point the new frame has been encoded.
// If any buffer copy / swapping is signaled it should be done here.
if (cm->frame_type == KEY_FRAME) {
ref_cnt_fb(cm->frame_bufs,
&cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
ref_cnt_fb(cm->frame_bufs,
&cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
}
#if CONFIG_MULTIPLE_ARF
else if (!cpi->multi_arf_enabled && cpi->refresh_golden_frame &&
!cpi->refresh_alt_ref_frame) {
#else
else if (cpi->refresh_golden_frame && !cpi->refresh_alt_ref_frame &&
!cpi->use_svc) {
#endif
/* Preserve the previously existing golden frame and update the frame in
* the alt ref slot instead. This is highly specific to the current use of
* alt-ref as a forward reference, and this needs to be generalized as
* other uses are implemented (like RTC/temporal scaling)
*
* The update to the buffer in the alt ref slot was signaled in
* vp9_pack_bitstream(), now swap the buffer pointers so that it's treated
* as the golden frame next time.
*/
int tmp;
ref_cnt_fb(cm->frame_bufs,
&cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
tmp = cpi->alt_fb_idx;
cpi->alt_fb_idx = cpi->gld_fb_idx;
cpi->gld_fb_idx = tmp;
} else { /* For non key/golden frames */
if (cpi->refresh_alt_ref_frame) {
int arf_idx = cpi->alt_fb_idx;
#if CONFIG_MULTIPLE_ARF
if (cpi->multi_arf_enabled) {
arf_idx = cpi->arf_buffer_idx[cpi->sequence_number + 1];
}
#endif
ref_cnt_fb(cm->frame_bufs,
&cm->ref_frame_map[arf_idx], cm->new_fb_idx);
}
if (cpi->refresh_golden_frame) {
ref_cnt_fb(cm->frame_bufs,
&cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
}
}
if (cpi->refresh_last_frame) {
ref_cnt_fb(cm->frame_bufs,
&cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
}
}
static void loopfilter_frame(VP9_COMP *cpi, VP9_COMMON *cm) {
MACROBLOCKD *xd = &cpi->mb.e_mbd;
struct loopfilter *lf = &cm->lf;
if (xd->lossless) {
lf->filter_level = 0;
} else {
struct vpx_usec_timer timer;
vp9_clear_system_state();
vpx_usec_timer_start(&timer);
vp9_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
vpx_usec_timer_mark(&timer);
cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
}
if (lf->filter_level > 0) {
vp9_loop_filter_frame(cm, xd, lf->filter_level, 0, 0);
}
vp9_extend_frame_inner_borders(cm->frame_to_show);
}
void vp9_scale_references(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
MV_REFERENCE_FRAME ref_frame;
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
const int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, ref_frame)];
YV12_BUFFER_CONFIG *const ref = &cm->frame_bufs[idx].buf;
if (ref->y_crop_width != cm->width ||
ref->y_crop_height != cm->height) {
const int new_fb = get_free_fb(cm);
vp9_realloc_frame_buffer(&cm->frame_bufs[new_fb].buf,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL);
scale_and_extend_frame(ref, &cm->frame_bufs[new_fb].buf);
cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
} else {
cpi->scaled_ref_idx[ref_frame - 1] = idx;
cm->frame_bufs[idx].ref_count++;
}
}
}
static void release_scaled_references(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
int i;
for (i = 0; i < 3; i++)
cm->frame_bufs[cpi->scaled_ref_idx[i]].ref_count--;
}
static void full_to_model_count(unsigned int *model_count,
unsigned int *full_count) {
int n;
model_count[ZERO_TOKEN] = full_count[ZERO_TOKEN];
model_count[ONE_TOKEN] = full_count[ONE_TOKEN];
model_count[TWO_TOKEN] = full_count[TWO_TOKEN];
for (n = THREE_TOKEN; n < EOB_TOKEN; ++n)
model_count[TWO_TOKEN] += full_count[n];
model_count[EOB_MODEL_TOKEN] = full_count[EOB_TOKEN];
}
static void full_to_model_counts(vp9_coeff_count_model *model_count,
vp9_coeff_count *full_count) {
int i, j, k, l;
for (i = 0; i < PLANE_TYPES; ++i)
for (j = 0; j < REF_TYPES; ++j)
for (k = 0; k < COEF_BANDS; ++k)
for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
full_to_model_count(model_count[i][j][k][l], full_count[i][j][k][l]);
}
#if 0 && CONFIG_INTERNAL_STATS
static void output_frame_level_debug_stats(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
FILE *const f = fopen("tmp.stt", cm->current_video_frame ? "a" : "w");
int recon_err;
vp9_clear_system_state();
recon_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm));
if (cpi->twopass.total_left_stats.coded_error != 0.0)
fprintf(f, "%10u %10d %10d %10d %10d %10d "
"%10"PRId64" %10"PRId64" %10d "
"%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
"%6d %6d %5d %5d %5d "
"%10"PRId64" %10.3lf"
"%10lf %8u %10d %10d %10d\n",
cpi->common.current_video_frame, cpi->rc.this_frame_target,
cpi->rc.projected_frame_size,
cpi->rc.projected_frame_size / cpi->common.MBs,
(cpi->rc.projected_frame_size - cpi->rc.this_frame_target),
cpi->rc.total_target_vs_actual,
(cpi->oxcf.starting_buffer_level - cpi->rc.bits_off_target),
cpi->rc.total_actual_bits, cm->base_qindex,
vp9_convert_qindex_to_q(cm->base_qindex),
(double)vp9_dc_quant(cm->base_qindex, 0) / 4.0,
cpi->rc.avg_q,
vp9_convert_qindex_to_q(cpi->rc.ni_av_qi),
vp9_convert_qindex_to_q(cpi->cq_target_quality),
cpi->refresh_last_frame, cpi->refresh_golden_frame,
cpi->refresh_alt_ref_frame, cm->frame_type, cpi->rc.gfu_boost,
cpi->twopass.bits_left,
cpi->twopass.total_left_stats.coded_error,
cpi->twopass.bits_left /
(1 + cpi->twopass.total_left_stats.coded_error),
cpi->tot_recode_hits, recon_err, cpi->rc.kf_boost,
cpi->twopass.kf_zeromotion_pct);
fclose(f);
if (0) {
FILE *const fmodes = fopen("Modes.stt", "a");
int i;
2010-05-18 19:58:33 +04:00
fprintf(fmodes, "%6d:%1d:%1d:%1d ", cpi->common.current_video_frame,
cm->frame_type, cpi->refresh_golden_frame,
cpi->refresh_alt_ref_frame);
2010-05-18 19:58:33 +04:00
for (i = 0; i < MAX_MODES; ++i)
fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
2011-10-05 14:26:00 +04:00
fprintf(fmodes, "\n");
2010-05-18 19:58:33 +04:00
fclose(fmodes);
}
}
#endif
2010-05-18 19:58:33 +04:00
static void encode_without_recode_loop(VP9_COMP *cpi,
size_t *size,
uint8_t *dest,
int q) {
VP9_COMMON *const cm = &cpi->common;
vp9_clear_system_state();
vp9_set_quantizer(cpi, q);
// Set up entropy context depending on frame type. The decoder mandates
// the use of the default context, index 0, for keyframes and inter
// frames where the error_resilient_mode or intra_only flag is set. For
// other inter-frames the encoder currently uses only two contexts;
// context 1 for ALTREF frames and context 0 for the others.
if (cm->frame_type == KEY_FRAME) {
vp9_setup_key_frame(cpi);
} else {
if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc) {
cpi->common.frame_context_idx = cpi->refresh_alt_ref_frame;
}
vp9_setup_inter_frame(cpi);
}
// Variance adaptive and in frame q adjustment experiments are mutually
// exclusive.
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
vp9_vaq_frame_setup(cpi);
} else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
setup_in_frame_q_adj(cpi);
} else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
vp9_setup_cyclic_refresh_aq(cpi);
}
// transform / motion compensation build reconstruction frame
vp9_encode_frame(cpi);
// Update the skip mb flag probabilities based on the distribution
// seen in the last encoder iteration.
// update_base_skip_probs(cpi);
vp9_clear_system_state();
}
static void encode_with_recode_loop(VP9_COMP *cpi,
size_t *size,
uint8_t *dest,
int q,
int bottom_index,
int top_index) {
VP9_COMMON *const cm = &cpi->common;
RATE_CONTROL *const rc = &cpi->rc;
int loop_count = 0;
int loop = 0;
int overshoot_seen = 0;
int undershoot_seen = 0;
int q_low = bottom_index, q_high = top_index;
int frame_over_shoot_limit;
int frame_under_shoot_limit;
// Decide frame size bounds
vp9_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
&frame_under_shoot_limit,
&frame_over_shoot_limit);
do {
vp9_clear_system_state();
2010-05-18 19:58:33 +04:00
vp9_set_quantizer(cpi, q);
if (loop_count == 0) {
// Set up entropy context depending on frame type. The decoder mandates
// the use of the default context, index 0, for keyframes and inter
// frames where the error_resilient_mode or intra_only flag is set. For
// other inter-frames the encoder currently uses only two contexts;
// context 1 for ALTREF frames and context 0 for the others.
if (cm->frame_type == KEY_FRAME) {
vp9_setup_key_frame(cpi);
} else {
if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc) {
cpi->common.frame_context_idx = cpi->refresh_alt_ref_frame;
}
vp9_setup_inter_frame(cpi);
}
}
// Variance adaptive and in frame q adjustment experiments are mutually
// exclusive.
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
vp9_vaq_frame_setup(cpi);
} else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
setup_in_frame_q_adj(cpi);
}
// transform / motion compensation build reconstruction frame
vp9_encode_frame(cpi);
// Update the skip mb flag probabilities based on the distribution
// seen in the last encoder iteration.
// update_base_skip_probs(cpi);
vp9_clear_system_state();
// Dummy pack of the bitstream using up to date stats to get an
// accurate estimate of output frame size to determine if we need
// to recode.
if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) {
vp9_save_coding_context(cpi);
cpi->dummy_packing = 1;
if (!cpi->sf.use_nonrd_pick_mode)
vp9_pack_bitstream(cpi, dest, size);
rc->projected_frame_size = (int)(*size) << 3;
vp9_restore_coding_context(cpi);
if (frame_over_shoot_limit == 0)
frame_over_shoot_limit = 1;
}
if (cpi->oxcf.end_usage == USAGE_CONSTANT_QUALITY) {
loop = 0;
} else {
if ((cm->frame_type == KEY_FRAME) &&
rc->this_key_frame_forced &&
(rc->projected_frame_size < rc->max_frame_bandwidth)) {
int last_q = q;
int kf_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm));
int high_err_target = cpi->ambient_err;
int low_err_target = cpi->ambient_err >> 1;
// Prevent possible divide by zero error below for perfect KF
kf_err += !kf_err;
// The key frame is not good enough or we can afford
// to make it better without undue risk of popping.
if ((kf_err > high_err_target &&
rc->projected_frame_size <= frame_over_shoot_limit) ||
(kf_err > low_err_target &&
rc->projected_frame_size <= frame_under_shoot_limit)) {
// Lower q_high
q_high = q > q_low ? q - 1 : q_low;
// Adjust Q
q = (q * high_err_target) / kf_err;
q = MIN(q, (q_high + q_low) >> 1);
} else if (kf_err < low_err_target &&
rc->projected_frame_size >= frame_under_shoot_limit) {
// The key frame is much better than the previous frame
// Raise q_low
q_low = q < q_high ? q + 1 : q_high;
// Adjust Q
q = (q * low_err_target) / kf_err;
q = MIN(q, (q_high + q_low + 1) >> 1);
}
// Clamp Q to upper and lower limits:
q = clamp(q, q_low, q_high);
loop = q != last_q;
} else if (recode_loop_test(
cpi, frame_over_shoot_limit, frame_under_shoot_limit,
q, MAX(q_high, top_index), bottom_index)) {
// Is the projected frame size out of range and are we allowed
// to attempt to recode.
int last_q = q;
int retries = 0;
// Frame size out of permitted range:
// Update correction factor & compute new Q to try...
// Frame is too large
if (rc->projected_frame_size > rc->this_frame_target) {
// Special case if the projected size is > the max allowed.
if (rc->projected_frame_size >= rc->max_frame_bandwidth)
q_high = rc->worst_quality;
// Raise Qlow as to at least the current value
q_low = q < q_high ? q + 1 : q_high;
if (undershoot_seen || loop_count > 1) {
// Update rate_correction_factor unless
vp9_rc_update_rate_correction_factors(cpi, 1);
q = (q_high + q_low + 1) / 2;
} else {
// Update rate_correction_factor unless
vp9_rc_update_rate_correction_factors(cpi, 0);
q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, MAX(q_high, top_index));
while (q < q_low && retries < 10) {
vp9_rc_update_rate_correction_factors(cpi, 0);
q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, MAX(q_high, top_index));
retries++;
}
}
overshoot_seen = 1;
} else {
// Frame is too small
q_high = q > q_low ? q - 1 : q_low;
if (overshoot_seen || loop_count > 1) {
vp9_rc_update_rate_correction_factors(cpi, 1);
q = (q_high + q_low) / 2;
} else {
vp9_rc_update_rate_correction_factors(cpi, 0);
q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, top_index);
// Special case reset for qlow for constrained quality.
// This should only trigger where there is very substantial
// undershoot on a frame and the auto cq level is above
// the user passsed in value.
if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY &&
q < q_low) {
q_low = q;
}
while (q > q_high && retries < 10) {
vp9_rc_update_rate_correction_factors(cpi, 0);
q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, top_index);
retries++;
}
}
undershoot_seen = 1;
}
// Clamp Q to upper and lower limits:
q = clamp(q, q_low, q_high);
loop = q != last_q;
} else {
loop = 0;
}
}
// Special case for overlay frame.
if (rc->is_src_frame_alt_ref &&
rc->projected_frame_size < rc->max_frame_bandwidth)
loop = 0;
if (loop) {
loop_count++;
#if CONFIG_INTERNAL_STATS
cpi->tot_recode_hits++;
#endif
}
} while (loop);
}
static void get_ref_frame_flags(VP9_COMP *cpi) {
if (cpi->refresh_last_frame & cpi->refresh_golden_frame)
cpi->gold_is_last = 1;
else if (cpi->refresh_last_frame ^ cpi->refresh_golden_frame)
cpi->gold_is_last = 0;
if (cpi->refresh_last_frame & cpi->refresh_alt_ref_frame)
cpi->alt_is_last = 1;
else if (cpi->refresh_last_frame ^ cpi->refresh_alt_ref_frame)
cpi->alt_is_last = 0;
if (cpi->refresh_alt_ref_frame & cpi->refresh_golden_frame)
cpi->gold_is_alt = 1;
else if (cpi->refresh_alt_ref_frame ^ cpi->refresh_golden_frame)
cpi->gold_is_alt = 0;
cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
if (cpi->gold_is_last)
cpi->ref_frame_flags &= ~VP9_GOLD_FLAG;
if (cpi->rc.frames_till_gf_update_due == INT_MAX)
cpi->ref_frame_flags &= ~VP9_GOLD_FLAG;
if (cpi->alt_is_last)
cpi->ref_frame_flags &= ~VP9_ALT_FLAG;
if (cpi->gold_is_alt)
cpi->ref_frame_flags &= ~VP9_ALT_FLAG;
}
static void set_ext_overrides(VP9_COMP *cpi) {
// Overrides the defaults with the externally supplied values with
// vp9_update_reference() and vp9_update_entropy() calls
// Note: The overrides are valid only for the next frame passed
// to encode_frame_to_data_rate() function
if (cpi->ext_refresh_frame_context_pending) {
cpi->common.refresh_frame_context = cpi->ext_refresh_frame_context;
cpi->ext_refresh_frame_context_pending = 0;
}
if (cpi->ext_refresh_frame_flags_pending) {
cpi->refresh_last_frame = cpi->ext_refresh_last_frame;
cpi->refresh_golden_frame = cpi->ext_refresh_golden_frame;
cpi->refresh_alt_ref_frame = cpi->ext_refresh_alt_ref_frame;
cpi->ext_refresh_frame_flags_pending = 0;
}
}
static void encode_frame_to_data_rate(VP9_COMP *cpi,
size_t *size,
uint8_t *dest,
unsigned int *frame_flags) {
VP9_COMMON *const cm = &cpi->common;
TX_SIZE t;
int q;
int top_index;
int bottom_index;
const SPEED_FEATURES *const sf = &cpi->sf;
const unsigned int max_mv_def = MIN(cm->width, cm->height);
struct segmentation *const seg = &cm->seg;
set_ext_overrides(cpi);
/* Scale the source buffer, if required. */
if (cm->mi_cols * MI_SIZE != cpi->un_scaled_source->y_width ||
cm->mi_rows * MI_SIZE != cpi->un_scaled_source->y_height) {
scale_and_extend_frame_nonnormative(cpi->un_scaled_source,
&cpi->scaled_source);
cpi->Source = &cpi->scaled_source;
} else {
cpi->Source = cpi->un_scaled_source;
}
vp9_scale_references(cpi);
vp9_clear_system_state();
// Enable or disable mode based tweaking of the zbin.
// For 2 pass only used where GF/ARF prediction quality
// is above a threshold.
cpi->zbin_mode_boost = 0;
cpi->zbin_mode_boost_enabled = 0;
// Current default encoder behavior for the altref sign bias.
cm->ref_frame_sign_bias[ALTREF_FRAME] = cpi->rc.source_alt_ref_active;
// Set default state for segment based loop filter update flags.
cm->lf.mode_ref_delta_update = 0;
// Initialize cpi->mv_step_param to default based on max resolution.
cpi->mv_step_param = vp9_init_search_range(cpi, max_mv_def);
// Initialize cpi->max_mv_magnitude and cpi->mv_step_param if appropriate.
if (sf->auto_mv_step_size) {
if (frame_is_intra_only(cm)) {
// Initialize max_mv_magnitude for use in the first INTER frame
// after a key/intra-only frame.
cpi->max_mv_magnitude = max_mv_def;
} else {
if (cm->show_frame)
// Allow mv_steps to correspond to twice the max mv magnitude found
// in the previous frame, capped by the default max_mv_magnitude based
// on resolution.
cpi->mv_step_param = vp9_init_search_range(cpi, MIN(max_mv_def, 2 *
cpi->max_mv_magnitude));
cpi->max_mv_magnitude = 0;
}
}
// Set various flags etc to special state if it is a key frame.
if (frame_is_intra_only(cm)) {
vp9_setup_key_frame(cpi);
// Reset the loop filter deltas and segmentation map.
vp9_reset_segment_features(&cm->seg);
// If segmentation is enabled force a map update for key frames.
if (seg->enabled) {
seg->update_map = 1;
seg->update_data = 1;
}
// The alternate reference frame cannot be active for a key frame.
cpi->rc.source_alt_ref_active = 0;
cm->error_resilient_mode = (cpi->oxcf.error_resilient_mode != 0);
cm->frame_parallel_decoding_mode =
(cpi->oxcf.frame_parallel_decoding_mode != 0);
// By default, encoder assumes decoder can use prev_mi.
cm->coding_use_prev_mi = 1;
if (cm->error_resilient_mode) {
cm->coding_use_prev_mi = 0;
cm->frame_parallel_decoding_mode = 1;
cm->reset_frame_context = 0;
cm->refresh_frame_context = 0;
} else if (cm->intra_only) {
// Only reset the current context.
cm->reset_frame_context = 2;
}
}
// Configure experimental use of segmentation for enhanced coding of
// static regions if indicated.
// Only allowed in second pass of two pass (as requires lagged coding)
// and if the relevant speed feature flag is set.
if (cpi->pass == 2 && cpi->sf.static_segmentation)
configure_static_seg_features(cpi);
// For 1 pass CBR, check if we are dropping this frame.
// Never drop on key frame.
if (cpi->pass == 0 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER &&
cm->frame_type != KEY_FRAME) {
if (vp9_rc_drop_frame(cpi)) {
vp9_rc_postencode_update_drop_frame(cpi);
++cm->current_video_frame;
return;
}
}
vp9_clear_system_state();
vp9_zero(cpi->rd_tx_select_threshes);
2010-05-18 19:58:33 +04:00
#if CONFIG_VP9_POSTPROC
if (cpi->oxcf.noise_sensitivity > 0) {
int l = 0;
switch (cpi->oxcf.noise_sensitivity) {
case 1:
l = 20;
break;
case 2:
l = 40;
break;
case 3:
l = 60;
break;
case 4:
case 5:
l = 100;
break;
case 6:
l = 150;
break;
}
vp9_denoise(cpi->Source, cpi->Source, l);
}
2010-05-18 19:58:33 +04:00
#endif
#ifdef OUTPUT_YUV_SRC
vp9_write_yuv_frame(cpi->Source);
2010-05-18 19:58:33 +04:00
#endif
// Decide q and q bounds.
q = vp9_rc_pick_q_and_bounds(cpi, &bottom_index, &top_index);
if (!frame_is_intra_only(cm)) {
cm->interp_filter = DEFAULT_INTERP_FILTER;
/* TODO: Decide this more intelligently */
set_high_precision_mv(cpi, q < HIGH_PRECISION_MV_QTHRESH);
}
vp9_set_speed_features(cpi);
if (cpi->sf.recode_loop == DISALLOW_RECODE) {
encode_without_recode_loop(cpi, size, dest, q);
} else {
encode_with_recode_loop(cpi, size, dest, q, bottom_index, top_index);
}
// Special case code to reduce pulsing when key frames are forced at a
// fixed interval. Note the reconstruction error if it is the frame before
// the force key frame
if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
cpi->ambient_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm));
}
// If the encoder forced a KEY_FRAME decision
if (cm->frame_type == KEY_FRAME)
cpi->refresh_last_frame = 1;
2010-05-18 19:58:33 +04:00
cm->frame_to_show = get_frame_new_buffer(cm);
#if WRITE_RECON_BUFFER
if (cm->show_frame)
write_cx_frame_to_file(cm->frame_to_show,
cm->current_video_frame);
else
write_cx_frame_to_file(cm->frame_to_show,
cm->current_video_frame + 1000);
#endif
// Pick the loop filter level for the frame.
loopfilter_frame(cpi, cm);
#if WRITE_RECON_BUFFER
if (cm->show_frame)
write_cx_frame_to_file(cm->frame_to_show,
cm->current_video_frame + 2000);
else
write_cx_frame_to_file(cm->frame_to_show,
cm->current_video_frame + 3000);
#endif
// build the bitstream
cpi->dummy_packing = 0;
vp9_pack_bitstream(cpi, dest, size);
if (cm->seg.update_map)
update_reference_segmentation_map(cpi);
release_scaled_references(cpi);
vp9_update_reference_frames(cpi);
for (t = TX_4X4; t <= TX_32X32; t++)
full_to_model_counts(cm->counts.coef[t], cpi->coef_counts[t]);
if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode)
vp9_adapt_coef_probs(cm);
if (!frame_is_intra_only(cm)) {
if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) {
vp9_adapt_mode_probs(cm);
vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv);
}
}
#if 0
output_frame_level_debug_stats(cpi);
2010-05-18 19:58:33 +04:00
#endif
if (cpi->refresh_golden_frame == 1)
cm->frame_flags |= FRAMEFLAGS_GOLDEN;
else
cm->frame_flags &= ~FRAMEFLAGS_GOLDEN;
2010-05-18 19:58:33 +04:00
if (cpi->refresh_alt_ref_frame == 1)
cm->frame_flags |= FRAMEFLAGS_ALTREF;
else
cm->frame_flags &= ~FRAMEFLAGS_ALTREF;
2010-05-18 19:58:33 +04:00
get_ref_frame_flags(cpi);
2010-05-18 19:58:33 +04:00
vp9_rc_postencode_update(cpi, *size);
2010-05-18 19:58:33 +04:00
if (cm->frame_type == KEY_FRAME) {
// Tell the caller that the frame was coded as a key frame
*frame_flags = cm->frame_flags | FRAMEFLAGS_KEY;
2010-05-18 19:58:33 +04:00
#if CONFIG_MULTIPLE_ARF
// Reset the sequence number.
if (cpi->multi_arf_enabled) {
cpi->sequence_number = 0;
cpi->frame_coding_order_period = cpi->new_frame_coding_order_period;
cpi->new_frame_coding_order_period = -1;
}
#endif
} else {
*frame_flags = cm->frame_flags&~FRAMEFLAGS_KEY;
#if CONFIG_MULTIPLE_ARF
/* Increment position in the coded frame sequence. */
if (cpi->multi_arf_enabled) {
++cpi->sequence_number;
if (cpi->sequence_number >= cpi->frame_coding_order_period) {
cpi->sequence_number = 0;
cpi->frame_coding_order_period = cpi->new_frame_coding_order_period;
cpi->new_frame_coding_order_period = -1;
}
cpi->this_frame_weight = cpi->arf_weight[cpi->sequence_number];
assert(cpi->this_frame_weight >= 0);
}
#endif
}
2010-05-18 19:58:33 +04:00
// Clear the one shot update flags for segmentation map and mode/ref loop
// filter deltas.
cm->seg.update_map = 0;
cm->seg.update_data = 0;
cm->lf.mode_ref_delta_update = 0;
2010-05-18 19:58:33 +04:00
// keep track of the last coded dimensions
cm->last_width = cm->width;
cm->last_height = cm->height;
2010-05-18 19:58:33 +04:00
// reset to normal state now that we are done.
if (!cm->show_existing_frame)
cm->last_show_frame = cm->show_frame;
if (cm->show_frame) {
vp9_swap_mi_and_prev_mi(cm);
// Don't increment frame counters if this was an altref buffer
// update not a real frame
++cm->current_video_frame;
}
// restore prev_mi
cm->prev_mi = cm->prev_mip + cm->mode_info_stride + 1;
cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mode_info_stride + 1;
2010-05-18 19:58:33 +04:00
}
static void SvcEncode(VP9_COMP *cpi, size_t *size, uint8_t *dest,
unsigned int *frame_flags) {
vp9_rc_get_svc_params(cpi);
encode_frame_to_data_rate(cpi, size, dest, frame_flags);
}
static void Pass0Encode(VP9_COMP *cpi, size_t *size, uint8_t *dest,
unsigned int *frame_flags) {
if (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
vp9_rc_get_one_pass_cbr_params(cpi);
} else {
vp9_rc_get_one_pass_vbr_params(cpi);
}
encode_frame_to_data_rate(cpi, size, dest, frame_flags);
}
static void Pass1Encode(VP9_COMP *cpi, size_t *size, uint8_t *dest,
unsigned int *frame_flags) {
(void) size;
(void) dest;
(void) frame_flags;
vp9_rc_get_first_pass_params(cpi);
vp9_set_quantizer(cpi, find_fp_qindex());
vp9_first_pass(cpi);
}
static void Pass2Encode(VP9_COMP *cpi, size_t *size,
uint8_t *dest, unsigned int *frame_flags) {
cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
vp9_rc_get_second_pass_params(cpi);
encode_frame_to_data_rate(cpi, size, dest, frame_flags);
vp9_twopass_postencode_update(cpi, *size);
2010-05-18 19:58:33 +04:00
}
static void check_initial_width(VP9_COMP *cpi, int subsampling_x,
int subsampling_y) {
VP9_COMMON *const cm = &cpi->common;
if (!cpi->initial_width) {
cm->subsampling_x = subsampling_x;
cm->subsampling_y = subsampling_y;
alloc_raw_frame_buffers(cpi);
cpi->initial_width = cm->width;
cpi->initial_height = cm->height;
}
}
int vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
int64_t end_time) {
VP9_COMMON *cm = &cpi->common;
struct vpx_usec_timer timer;
int res = 0;
const int subsampling_x = sd->uv_width < sd->y_width;
const int subsampling_y = sd->uv_height < sd->y_height;
check_initial_width(cpi, subsampling_x, subsampling_y);
vpx_usec_timer_start(&timer);
if (vp9_lookahead_push(cpi->lookahead,
sd, time_stamp, end_time, frame_flags))
res = -1;
vpx_usec_timer_mark(&timer);
cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
2010-05-18 19:58:33 +04:00
if (cm->version == 0 && (subsampling_x != 1 || subsampling_y != 1)) {
vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
"Non-4:2:0 color space requires profile >= 1");
res = -1;
}
return res;
2010-05-18 19:58:33 +04:00
}
static int frame_is_reference(const VP9_COMP *cpi) {
const VP9_COMMON *cm = &cpi->common;
return cm->frame_type == KEY_FRAME ||
cpi->refresh_last_frame ||
cpi->refresh_golden_frame ||
cpi->refresh_alt_ref_frame ||
cm->refresh_frame_context ||
cm->lf.mode_ref_delta_update ||
cm->seg.update_map ||
cm->seg.update_data;
}
#if CONFIG_MULTIPLE_ARF
int is_next_frame_arf(VP9_COMP *cpi) {
// Negative entry in frame_coding_order indicates an ARF at this position.
return cpi->frame_coding_order[cpi->sequence_number + 1] < 0 ? 1 : 0;
}
#endif
void adjust_frame_rate(VP9_COMP *cpi) {
int64_t this_duration;
int step = 0;
if (cpi->source->ts_start == cpi->first_time_stamp_ever) {
this_duration = cpi->source->ts_end - cpi->source->ts_start;
step = 1;
} else {
int64_t last_duration = cpi->last_end_time_stamp_seen
- cpi->last_time_stamp_seen;
this_duration = cpi->source->ts_end - cpi->last_end_time_stamp_seen;
// do a step update if the duration changes by 10%
if (last_duration)
step = (int)((this_duration - last_duration) * 10 / last_duration);
}
if (this_duration) {
if (step) {
vp9_new_framerate(cpi, 10000000.0 / this_duration);
} else {
// Average this frame's rate into the last second's average
// frame rate. If we haven't seen 1 second yet, then average
// over the whole interval seen.
const double interval = MIN((double)(cpi->source->ts_end
- cpi->first_time_stamp_ever), 10000000.0);
double avg_duration = 10000000.0 / cpi->oxcf.framerate;
avg_duration *= (interval - avg_duration + this_duration);
avg_duration /= interval;
vp9_new_framerate(cpi, 10000000.0 / avg_duration);
}
}
cpi->last_time_stamp_seen = cpi->source->ts_start;
cpi->last_end_time_stamp_seen = cpi->source->ts_end;
}
int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
size_t *size, uint8_t *dest,
int64_t *time_stamp, int64_t *time_end, int flush) {
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
struct vpx_usec_timer cmptimer;
YV12_BUFFER_CONFIG *force_src_buffer = NULL;
MV_REFERENCE_FRAME ref_frame;
2010-05-18 19:58:33 +04:00
if (!cpi)
return -1;
2010-05-18 19:58:33 +04:00
vpx_usec_timer_start(&cmptimer);
2010-05-18 19:58:33 +04:00
cpi->source = NULL;
2010-05-18 19:58:33 +04:00
set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV);
// Normal defaults
cm->reset_frame_context = 0;
cm->refresh_frame_context = 1;
cpi->refresh_last_frame = 1;
cpi->refresh_golden_frame = 0;
cpi->refresh_alt_ref_frame = 0;
// Should we code an alternate reference frame.
if (cpi->oxcf.play_alternate && cpi->rc.source_alt_ref_pending) {
int frames_to_arf;
#if CONFIG_MULTIPLE_ARF
assert(!cpi->multi_arf_enabled ||
cpi->frame_coding_order[cpi->sequence_number] < 0);
if (cpi->multi_arf_enabled && (cpi->pass == 2))
frames_to_arf = (-cpi->frame_coding_order[cpi->sequence_number])
- cpi->next_frame_in_order;
else
#endif
frames_to_arf = cpi->rc.frames_till_gf_update_due;
assert(frames_to_arf <= cpi->rc.frames_to_key);
if ((cpi->source = vp9_lookahead_peek(cpi->lookahead, frames_to_arf))) {
#if CONFIG_MULTIPLE_ARF
cpi->alt_ref_source[cpi->arf_buffered] = cpi->source;
#else
cpi->alt_ref_source = cpi->source;
#endif
if (cpi->oxcf.arnr_max_frames > 0) {
// Produce the filtered ARF frame.
// TODO(agrange) merge these two functions.
vp9_configure_arnr_filter(cpi, frames_to_arf, cpi->rc.gfu_boost);
vp9_temporal_filter_prepare(cpi, frames_to_arf);
vp9_extend_frame_borders(&cpi->alt_ref_buffer);
force_src_buffer = &cpi->alt_ref_buffer;
}
cm->show_frame = 0;
cpi->refresh_alt_ref_frame = 1;
cpi->refresh_golden_frame = 0;
cpi->refresh_last_frame = 0;
cpi->rc.is_src_frame_alt_ref = 0;
#if CONFIG_MULTIPLE_ARF
if (!cpi->multi_arf_enabled)
#endif
cpi->rc.source_alt_ref_pending = 0;
} else {
cpi->rc.source_alt_ref_pending = 0;
}
}
if (!cpi->source) {
#if CONFIG_MULTIPLE_ARF
int i;
#endif
if ((cpi->source = vp9_lookahead_pop(cpi->lookahead, flush))) {
cm->show_frame = 1;
cm->intra_only = 0;
#if CONFIG_MULTIPLE_ARF
// Is this frame the ARF overlay.
cpi->rc.is_src_frame_alt_ref = 0;
for (i = 0; i < cpi->arf_buffered; ++i) {
if (cpi->source == cpi->alt_ref_source[i]) {
cpi->rc.is_src_frame_alt_ref = 1;
cpi->refresh_golden_frame = 1;
break;
}
}
#else
cpi->rc.is_src_frame_alt_ref = cpi->alt_ref_source
&& (cpi->source == cpi->alt_ref_source);
#endif
if (cpi->rc.is_src_frame_alt_ref) {
// Current frame is an ARF overlay frame.
#if CONFIG_MULTIPLE_ARF
cpi->alt_ref_source[i] = NULL;
#else
cpi->alt_ref_source = NULL;
#endif
// Don't refresh the last buffer for an ARF overlay frame. It will
// become the GF so preserve last as an alternative prediction option.
cpi->refresh_last_frame = 0;
}
#if CONFIG_MULTIPLE_ARF
++cpi->next_frame_in_order;
#endif
}
}
if (cpi->source) {
cpi->un_scaled_source = cpi->Source = force_src_buffer ? force_src_buffer
: &cpi->source->img;
*time_stamp = cpi->source->ts_start;
*time_end = cpi->source->ts_end;
*frame_flags = cpi->source->flags;
#if CONFIG_MULTIPLE_ARF
if ((cm->frame_type != KEY_FRAME) && (cpi->pass == 2))
cpi->rc.source_alt_ref_pending = is_next_frame_arf(cpi);
#endif
} else {
*size = 0;
if (flush && cpi->pass == 1 && !cpi->twopass.first_pass_done) {
vp9_end_first_pass(cpi); /* get last stats packet */
cpi->twopass.first_pass_done = 1;
2010-05-18 19:58:33 +04:00
}
return -1;
}
2010-05-18 19:58:33 +04:00
if (cpi->source->ts_start < cpi->first_time_stamp_ever) {
cpi->first_time_stamp_ever = cpi->source->ts_start;
cpi->last_end_time_stamp_seen = cpi->source->ts_start;
}
// adjust frame rates based on timestamps given
if (cm->show_frame) {
adjust_frame_rate(cpi);
}
2010-05-18 19:58:33 +04:00
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
vp9_update_layer_framerate(cpi);
vp9_restore_layer_context(cpi);
}
// start with a 0 size frame
*size = 0;
2010-05-18 19:58:33 +04:00
// Clear down mmx registers
vp9_clear_system_state();
/* find a free buffer for the new frame, releasing the reference previously
* held.
*/
cm->frame_bufs[cm->new_fb_idx].ref_count--;
cm->new_fb_idx = get_free_fb(cm);
#if CONFIG_MULTIPLE_ARF
/* Set up the correct ARF frame. */
if (cpi->refresh_alt_ref_frame) {
++cpi->arf_buffered;
}
if (cpi->multi_arf_enabled && (cm->frame_type != KEY_FRAME) &&
(cpi->pass == 2)) {
cpi->alt_fb_idx = cpi->arf_buffer_idx[cpi->sequence_number];
}
#endif
cm->frame_flags = *frame_flags;
// Reset the frame pointers to the current frame size
vp9_realloc_frame_buffer(get_frame_new_buffer(cm),
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL);
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
const int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, ref_frame)];
YV12_BUFFER_CONFIG *const buf = &cm->frame_bufs[idx].buf;
RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - 1];
ref_buf->buf = buf;
ref_buf->idx = idx;
vp9_setup_scale_factors_for_frame(&ref_buf->sf,
buf->y_crop_width, buf->y_crop_height,
cm->width, cm->height);
if (vp9_is_scaled(&ref_buf->sf))
vp9_extend_frame_borders(buf);
}
set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME);
xd->interp_kernel = vp9_get_interp_kernel(
DEFAULT_INTERP_FILTER == SWITCHABLE ? EIGHTTAP : DEFAULT_INTERP_FILTER);
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
vp9_vaq_init();
}
if (cpi->pass == 1 &&
(!cpi->use_svc || cpi->svc.number_temporal_layers == 1)) {
Pass1Encode(cpi, size, dest, frame_flags);
} else if (cpi->pass == 2 && !cpi->use_svc) {
Pass2Encode(cpi, size, dest, frame_flags);
} else if (cpi->use_svc) {
SvcEncode(cpi, size, dest, frame_flags);
} else {
// One pass encode
Pass0Encode(cpi, size, dest, frame_flags);
}
2010-05-18 19:58:33 +04:00
if (cm->refresh_frame_context)
cm->frame_contexts[cm->frame_context_idx] = cm->fc;
2010-05-18 19:58:33 +04:00
// Frame was dropped, release scaled references.
if (*size == 0) {
release_scaled_references(cpi);
}
if (*size > 0) {
cpi->droppable = !frame_is_reference(cpi);
}
2010-05-18 19:58:33 +04:00
// Save layer specific state.
if (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
vp9_save_layer_context(cpi);
}
vpx_usec_timer_mark(&cmptimer);
cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
if (cpi->b_calculate_psnr && cpi->pass != 1 && cm->show_frame)
generate_psnr_packet(cpi);
2010-05-18 19:58:33 +04:00
#if CONFIG_INTERNAL_STATS
2010-05-18 19:58:33 +04:00
if (cpi->pass != 1) {
cpi->bytes += (int)(*size);
2010-05-18 19:58:33 +04:00
if (cm->show_frame) {
cpi->count++;
2010-05-18 19:58:33 +04:00
if (cpi->b_calculate_psnr) {
YV12_BUFFER_CONFIG *orig = cpi->Source;
YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
PSNR_STATS psnr;
calc_psnr(orig, recon, &psnr);
cpi->total += psnr.psnr[0];
cpi->total_y += psnr.psnr[1];
cpi->total_u += psnr.psnr[2];
cpi->total_v += psnr.psnr[3];
cpi->total_sq_error += psnr.sse[0];
cpi->total_samples += psnr.samples[0];
{
PSNR_STATS psnr2;
double frame_ssim2 = 0, weight = 0;
#if CONFIG_VP9_POSTPROC
vp9_deblock(cm->frame_to_show, &cm->post_proc_buffer,
cm->lf.filter_level * 10 / 6);
#endif
vp9_clear_system_state();
calc_psnr(orig, pp, &psnr2);
cpi->totalp += psnr2.psnr[0];
cpi->totalp_y += psnr2.psnr[1];
cpi->totalp_u += psnr2.psnr[2];
cpi->totalp_v += psnr2.psnr[3];
cpi->totalp_sq_error += psnr2.sse[0];
cpi->totalp_samples += psnr2.samples[0];
2010-05-18 19:58:33 +04:00
frame_ssim2 = vp9_calc_ssim(orig, recon, 1, &weight);
cpi->summed_quality += frame_ssim2 * weight;
cpi->summed_weights += weight;
frame_ssim2 = vp9_calc_ssim(orig, &cm->post_proc_buffer, 1, &weight);
cpi->summedp_quality += frame_ssim2 * weight;
cpi->summedp_weights += weight;
experiment extending the quantizer range Prior to this change, VP8 min quantizer is 4, which caps the highest quality around 51DB. This experimental change extends the min quantizer to 1, removes the cap and allows the highest quality to be around ~73DB, consistent with the fdct/idct round trip error. To test this change, at configure time use options: --enable-experimental --enable-extend_qrange The following is a brief log of changes in each of the patch sets patch set 1: In this commit, the quantization/dequantization constants are kept unchanged, instead scaling factor 4 is rolled into fdct/idct. Fixed Q0 encoding tests on mobile: Before: 9560.567kbps Overall PSNR:50.255DB VPXSSIM:98.288 Now: 18035.774kbps Overall PSNR:73.022DB VPXSSIM:99.991 patch set 2: regenerated dc/ac quantizer lookup tables based on the scaling factor rolled in the fdct/idct. Also slightly extended the range towards the high quantizer end. patch set 3: slightly tweaked the quantizer tables and generated bits_per_mb table based on Paul's suggestions. patch set 4: fix a typo in idct, re-calculated tables relating active max Q to active min Q patch set 5: added rdmult lookup table based on Q patch set 6: fix rdmult scale: dct coefficient has scaled up by 4 patch set 7: make transform coefficients to be within 16bits patch set 8: normalize 2nd order quantizers patch set 9: fix mis-spellings patch set 10: change the configure script and macros to allow experimental code to be enabled at configure time with --enable-extend_qrange patch set 11: rebase for merge Change-Id: Ib50641ddd44aba2a52ed890222c309faa31cc59c
2010-12-02 02:50:14 +03:00
#if 0
{
FILE *f = fopen("q_used.stt", "a");
fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
cpi->common.current_video_frame, y2, u2, v2,
frame_psnr2, frame_ssim2);
fclose(f);
}
experiment extending the quantizer range Prior to this change, VP8 min quantizer is 4, which caps the highest quality around 51DB. This experimental change extends the min quantizer to 1, removes the cap and allows the highest quality to be around ~73DB, consistent with the fdct/idct round trip error. To test this change, at configure time use options: --enable-experimental --enable-extend_qrange The following is a brief log of changes in each of the patch sets patch set 1: In this commit, the quantization/dequantization constants are kept unchanged, instead scaling factor 4 is rolled into fdct/idct. Fixed Q0 encoding tests on mobile: Before: 9560.567kbps Overall PSNR:50.255DB VPXSSIM:98.288 Now: 18035.774kbps Overall PSNR:73.022DB VPXSSIM:99.991 patch set 2: regenerated dc/ac quantizer lookup tables based on the scaling factor rolled in the fdct/idct. Also slightly extended the range towards the high quantizer end. patch set 3: slightly tweaked the quantizer tables and generated bits_per_mb table based on Paul's suggestions. patch set 4: fix a typo in idct, re-calculated tables relating active max Q to active min Q patch set 5: added rdmult lookup table based on Q patch set 6: fix rdmult scale: dct coefficient has scaled up by 4 patch set 7: make transform coefficients to be within 16bits patch set 8: normalize 2nd order quantizers patch set 9: fix mis-spellings patch set 10: change the configure script and macros to allow experimental code to be enabled at configure time with --enable-extend_qrange patch set 11: rebase for merge Change-Id: Ib50641ddd44aba2a52ed890222c309faa31cc59c
2010-12-02 02:50:14 +03:00
#endif
}
}
2010-05-18 19:58:33 +04:00
if (cpi->b_calculate_ssimg) {
double y, u, v, frame_all;
frame_all = vp9_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u, &v);
cpi->total_ssimg_y += y;
cpi->total_ssimg_u += u;
cpi->total_ssimg_v += v;
cpi->total_ssimg_all += frame_all;
}
2010-05-18 19:58:33 +04:00
}
}
2010-05-18 19:58:33 +04:00
#endif
return 0;
2010-05-18 19:58:33 +04:00
}
int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
vp9_ppflags_t *flags) {
VP9_COMMON *cm = &cpi->common;
2010-05-18 19:58:33 +04:00
if (!cm->show_frame) {
return -1;
} else {
int ret;
#if CONFIG_VP9_POSTPROC
ret = vp9_post_proc_frame(cm, dest, flags);
2010-05-18 19:58:33 +04:00
#else
if (cm->frame_to_show) {
*dest = *cm->frame_to_show;
dest->y_width = cm->width;
dest->y_height = cm->height;
dest->uv_width = cm->width >> cm->subsampling_x;
dest->uv_height = cm->height >> cm->subsampling_y;
ret = 0;
} else {
ret = -1;
2010-05-18 19:58:33 +04:00
}
#endif // !CONFIG_VP9_POSTPROC
vp9_clear_system_state();
return ret;
}
}
2010-05-18 19:58:33 +04:00
int vp9_set_roimap(VP9_COMP *cpi, unsigned char *map, unsigned int rows,
unsigned int cols, int delta_q[MAX_SEGMENTS],
int delta_lf[MAX_SEGMENTS],
unsigned int threshold[MAX_SEGMENTS]) {
signed char feature_data[SEG_LVL_MAX][MAX_SEGMENTS];
struct segmentation *seg = &cpi->common.seg;
int i;
2010-05-18 19:58:33 +04:00
if (cpi->common.mb_rows != rows || cpi->common.mb_cols != cols)
return -1;
2010-05-18 19:58:33 +04:00
if (!map) {
vp9_disable_segmentation(seg);
return 0;
}
// Set the segmentation Map
vp9_set_segmentation_map(cpi, map);
// Activate segmentation.
vp9_enable_segmentation(seg);
// Set up the quant, LF and breakout threshold segment data
for (i = 0; i < MAX_SEGMENTS; i++) {
feature_data[SEG_LVL_ALT_Q][i] = delta_q[i];
feature_data[SEG_LVL_ALT_LF][i] = delta_lf[i];
cpi->segment_encode_breakout[i] = threshold[i];
}
// Enable the loop and quant changes in the feature mask
for (i = 0; i < MAX_SEGMENTS; i++) {
if (delta_q[i])
vp9_enable_segfeature(seg, i, SEG_LVL_ALT_Q);
else
vp9_disable_segfeature(seg, i, SEG_LVL_ALT_Q);
2010-05-18 19:58:33 +04:00
if (delta_lf[i])
vp9_enable_segfeature(seg, i, SEG_LVL_ALT_LF);
else
vp9_disable_segfeature(seg, i, SEG_LVL_ALT_LF);
}
2011-10-05 14:26:00 +04:00
// Initialize the feature data structure
// SEGMENT_DELTADATA 0, SEGMENT_ABSDATA 1
vp9_set_segment_data(seg, &feature_data[0][0], SEGMENT_DELTADATA);
2010-05-18 19:58:33 +04:00
return 0;
2010-05-18 19:58:33 +04:00
}
int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map,
unsigned int rows, unsigned int cols) {
if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) {
if (map) {
vpx_memcpy(cpi->active_map, map, rows * cols);
cpi->active_map_enabled = 1;
} else {
cpi->active_map_enabled = 0;
}
2010-05-18 19:58:33 +04:00
return 0;
} else {
// cpi->active_map_enabled = 0;
return -1;
}
2010-05-18 19:58:33 +04:00
}
int vp9_set_internal_size(VP9_COMP *cpi,
VPX_SCALING horiz_mode, VPX_SCALING vert_mode) {
VP9_COMMON *cm = &cpi->common;
int hr = 0, hs = 0, vr = 0, vs = 0;
2010-05-18 19:58:33 +04:00
if (horiz_mode > ONETWO || vert_mode > ONETWO)
return -1;
2010-05-18 19:58:33 +04:00
Scale2Ratio(horiz_mode, &hr, &hs);
Scale2Ratio(vert_mode, &vr, &vs);
// always go to the next whole number
cm->width = (hs - 1 + cpi->oxcf.width * hr) / hs;
cm->height = (vs - 1 + cpi->oxcf.height * vr) / vs;
assert(cm->width <= cpi->initial_width);
assert(cm->height <= cpi->initial_height);
update_frame_size(cpi);
return 0;
2010-05-18 19:58:33 +04:00
}
int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
unsigned int height) {
VP9_COMMON *cm = &cpi->common;
check_initial_width(cpi, 1, 1);
if (width) {
cm->width = width;
if (cm->width * 5 < cpi->initial_width) {
cm->width = cpi->initial_width / 5 + 1;
printf("Warning: Desired width too small, changed to %d\n", cm->width);
}
if (cm->width > cpi->initial_width) {
cm->width = cpi->initial_width;
printf("Warning: Desired width too large, changed to %d\n", cm->width);
}
}
if (height) {
cm->height = height;
if (cm->height * 5 < cpi->initial_height) {
cm->height = cpi->initial_height / 5 + 1;
printf("Warning: Desired height too small, changed to %d\n", cm->height);
}
if (cm->height > cpi->initial_height) {
cm->height = cpi->initial_height;
printf("Warning: Desired height too large, changed to %d\n", cm->height);
}
}
assert(cm->width <= cpi->initial_width);
assert(cm->height <= cpi->initial_height);
update_frame_size(cpi);
return 0;
}
void vp9_set_svc(VP9_COMP *cpi, int use_svc) {
cpi->use_svc = use_svc;
return;
}
2010-05-18 19:58:33 +04:00
int vp9_calc_ss_err(const YV12_BUFFER_CONFIG *source,
const YV12_BUFFER_CONFIG *reference) {
int i, j;
int total = 0;
2010-05-18 19:58:33 +04:00
const uint8_t *src = source->y_buffer;
const uint8_t *ref = reference->y_buffer;
2010-05-18 19:58:33 +04:00
// Loop through the Y plane raw and reconstruction data summing
// (square differences)
for (i = 0; i < source->y_height; i += 16) {
for (j = 0; j < source->y_width; j += 16) {
unsigned int sse;
total += vp9_mse16x16(src + j, source->y_stride,
ref + j, reference->y_stride, &sse);
2010-05-18 19:58:33 +04:00
}
src += 16 * source->y_stride;
ref += 16 * reference->y_stride;
}
return total;
2010-05-18 19:58:33 +04:00
}
int vp9_get_quantizer(VP9_COMP *cpi) {
return cpi->common.base_qindex;
2010-05-18 19:58:33 +04:00
}