blob: 009bab7c50336eebac64043ea2aecbf47c53c88a [file] [log] [blame]
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* 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.
*/
#include <limits.h>
#include <math.h>
#include <stdio.h>
#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
#include "vpx_ports/system_state.h"
#include "vpx_scale/vpx_scale.h"
#include "vpx_scale/yv12config.h"
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_quant_common.h"
#include "vp9/common/vp9_reconinter.h" // vp9_setup_dst_planes()
#include "vp9/encoder/vp9_aq_variance.h"
#include "vp9/encoder/vp9_block.h"
#include "vp9/encoder/vp9_encodeframe.h"
#include "vp9/encoder/vp9_encodemb.h"
#include "vp9/encoder/vp9_encodemv.h"
#include "vp9/encoder/vp9_encoder.h"
#include "vp9/encoder/vp9_ethread.h"
#include "vp9/encoder/vp9_extend.h"
#include "vp9/encoder/vp9_firstpass.h"
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/encoder/vp9_rd.h"
#include "vpx_dsp/variance.h"
#define OUTPUT_FPF 0
#define ARF_STATS_OUTPUT 0
#define COMPLEXITY_STATS_OUTPUT 0
#define FIRST_PASS_Q 10.0
#define NORMAL_BOOST 100
#define MIN_ARF_GF_BOOST 250
#define MIN_DECAY_FACTOR 0.01
#define NEW_MV_MODE_PENALTY 32
#define DARK_THRESH 64
#define LOW_I_THRESH 24000
#define NCOUNT_INTRA_THRESH 8192
#define NCOUNT_INTRA_FACTOR 3
#define DOUBLE_DIVIDE_CHECK(x) ((x) < 0 ? (x)-0.000001 : (x) + 0.000001)
#if ARF_STATS_OUTPUT
unsigned int arf_count = 0;
#endif
// Resets the first pass file to the given position using a relative seek from
// the current position.
static void reset_fpf_position(TWO_PASS *p, const FIRSTPASS_STATS *position) {
p->stats_in = position;
}
// Read frame stats at an offset from the current position.
static const FIRSTPASS_STATS *read_frame_stats(const TWO_PASS *p, int offset) {
if ((offset >= 0 && p->stats_in + offset >= p->stats_in_end) ||
(offset < 0 && p->stats_in + offset < p->stats_in_start)) {
return NULL;
}
return &p->stats_in[offset];
}
static int input_stats(TWO_PASS *p, FIRSTPASS_STATS *fps) {
if (p->stats_in >= p->stats_in_end) return EOF;
*fps = *p->stats_in;
++p->stats_in;
return 1;
}
static void output_stats(FIRSTPASS_STATS *stats) {
(void)stats;
// TEMP debug code
#if OUTPUT_FPF
{
FILE *fpfile;
fpfile = fopen("firstpass.stt", "a");
fprintf(fpfile,
"%12.0lf %12.4lf %12.2lf %12.2lf %12.2lf %12.0lf %12.4lf %12.4lf"
"%12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf"
"%12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.0lf %12.4lf %12.0lf"
"%12.4lf"
"\n",
stats->frame, stats->weight, stats->intra_error, stats->coded_error,
stats->sr_coded_error, stats->frame_noise_energy, stats->pcnt_inter,
stats->pcnt_motion, stats->pcnt_second_ref, stats->pcnt_neutral,
stats->pcnt_intra_low, stats->pcnt_intra_high,
stats->intra_skip_pct, stats->intra_smooth_pct,
stats->inactive_zone_rows, stats->inactive_zone_cols, stats->MVr,
stats->mvr_abs, stats->MVc, stats->mvc_abs, stats->MVrv,
stats->MVcv, stats->mv_in_out_count, stats->count, stats->duration);
fclose(fpfile);
}
#endif
}
#if CONFIG_FP_MB_STATS
static void output_fpmb_stats(uint8_t *this_frame_mb_stats, VP9_COMMON *cm,
struct vpx_codec_pkt_list *pktlist) {
struct vpx_codec_cx_pkt pkt;
pkt.kind = VPX_CODEC_FPMB_STATS_PKT;
pkt.data.firstpass_mb_stats.buf = this_frame_mb_stats;
pkt.data.firstpass_mb_stats.sz = cm->initial_mbs * sizeof(uint8_t);
vpx_codec_pkt_list_add(pktlist, &pkt);
}
#endif
static void zero_stats(FIRSTPASS_STATS *section) {
section->frame = 0.0;
section->weight = 0.0;
section->intra_error = 0.0;
section->coded_error = 0.0;
section->sr_coded_error = 0.0;
section->frame_noise_energy = 0.0;
section->pcnt_inter = 0.0;
section->pcnt_motion = 0.0;
section->pcnt_second_ref = 0.0;
section->pcnt_neutral = 0.0;
section->intra_skip_pct = 0.0;
section->intra_smooth_pct = 0.0;
section->pcnt_intra_low = 0.0;
section->pcnt_intra_high = 0.0;
section->inactive_zone_rows = 0.0;
section->inactive_zone_cols = 0.0;
section->MVr = 0.0;
section->mvr_abs = 0.0;
section->MVc = 0.0;
section->mvc_abs = 0.0;
section->MVrv = 0.0;
section->MVcv = 0.0;
section->mv_in_out_count = 0.0;
section->count = 0.0;
section->duration = 1.0;
section->spatial_layer_id = 0;
}
static void accumulate_stats(FIRSTPASS_STATS *section,
const FIRSTPASS_STATS *frame) {
section->frame += frame->frame;
section->weight += frame->weight;
section->spatial_layer_id = frame->spatial_layer_id;
section->intra_error += frame->intra_error;
section->coded_error += frame->coded_error;
section->sr_coded_error += frame->sr_coded_error;
section->frame_noise_energy += frame->frame_noise_energy;
section->pcnt_inter += frame->pcnt_inter;
section->pcnt_motion += frame->pcnt_motion;
section->pcnt_second_ref += frame->pcnt_second_ref;
section->pcnt_neutral += frame->pcnt_neutral;
section->intra_skip_pct += frame->intra_skip_pct;
section->intra_smooth_pct += frame->intra_smooth_pct;
section->pcnt_intra_low += frame->pcnt_intra_low;
section->pcnt_intra_high += frame->pcnt_intra_high;
section->inactive_zone_rows += frame->inactive_zone_rows;
section->inactive_zone_cols += frame->inactive_zone_cols;
section->MVr += frame->MVr;
section->mvr_abs += frame->mvr_abs;
section->MVc += frame->MVc;
section->mvc_abs += frame->mvc_abs;
section->MVrv += frame->MVrv;
section->MVcv += frame->MVcv;
section->mv_in_out_count += frame->mv_in_out_count;
section->count += frame->count;
section->duration += frame->duration;
}
static void subtract_stats(FIRSTPASS_STATS *section,
const FIRSTPASS_STATS *frame) {
section->frame -= frame->frame;
section->weight -= frame->weight;
section->intra_error -= frame->intra_error;
section->coded_error -= frame->coded_error;
section->sr_coded_error -= frame->sr_coded_error;
section->frame_noise_energy -= frame->frame_noise_energy;
section->pcnt_inter -= frame->pcnt_inter;
section->pcnt_motion -= frame->pcnt_motion;
section->pcnt_second_ref -= frame->pcnt_second_ref;
section->pcnt_neutral -= frame->pcnt_neutral;
section->intra_skip_pct -= frame->intra_skip_pct;
section->intra_smooth_pct -= frame->intra_smooth_pct;
section->pcnt_intra_low -= frame->pcnt_intra_low;
section->pcnt_intra_high -= frame->pcnt_intra_high;
section->inactive_zone_rows -= frame->inactive_zone_rows;
section->inactive_zone_cols -= frame->inactive_zone_cols;
section->MVr -= frame->MVr;
section->mvr_abs -= frame->mvr_abs;
section->MVc -= frame->MVc;
section->mvc_abs -= frame->mvc_abs;
section->MVrv -= frame->MVrv;
section->MVcv -= frame->MVcv;
section->mv_in_out_count -= frame->mv_in_out_count;
section->count -= frame->count;
section->duration -= frame->duration;
}
// Calculate an active area of the image that discounts formatting
// bars and partially discounts other 0 energy areas.
#define MIN_ACTIVE_AREA 0.5
#define MAX_ACTIVE_AREA 1.0
static double calculate_active_area(const FRAME_INFO *frame_info,
const FIRSTPASS_STATS *this_frame) {
double active_pct;
active_pct =
1.0 -
((this_frame->intra_skip_pct / 2) +
((this_frame->inactive_zone_rows * 2) / (double)frame_info->mb_rows));
return fclamp(active_pct, MIN_ACTIVE_AREA, MAX_ACTIVE_AREA);
}
// Get the average weighted error for the clip (or corpus)
static double get_distribution_av_err(VP9_COMP *cpi, TWO_PASS *const twopass) {
const double av_weight =
twopass->total_stats.weight / twopass->total_stats.count;
if (cpi->oxcf.vbr_corpus_complexity)
return av_weight * twopass->mean_mod_score;
else
return (twopass->total_stats.coded_error * av_weight) /
twopass->total_stats.count;
}
#define ACT_AREA_CORRECTION 0.5
// Calculate a modified Error used in distributing bits between easier and
// harder frames.
static double calculate_mod_frame_score(const VP9_COMP *cpi,
const VP9EncoderConfig *oxcf,
const FIRSTPASS_STATS *this_frame,
const double av_err) {
double modified_score =
av_err * pow(this_frame->coded_error * this_frame->weight /
DOUBLE_DIVIDE_CHECK(av_err),
oxcf->two_pass_vbrbias / 100.0);
// Correction for active area. Frames with a reduced active area
// (eg due to formatting bars) have a higher error per mb for the
// remaining active MBs. The correction here assumes that coding
// 0.5N blocks of complexity 2X is a little easier than coding N
// blocks of complexity X.
modified_score *= pow(calculate_active_area(&cpi->frame_info, this_frame),
ACT_AREA_CORRECTION);
return modified_score;
}
static double calc_norm_frame_score(const VP9EncoderConfig *oxcf,
const FRAME_INFO *frame_info,
const FIRSTPASS_STATS *this_frame,
double mean_mod_score, double av_err) {
double modified_score =
av_err * pow(this_frame->coded_error * this_frame->weight /
DOUBLE_DIVIDE_CHECK(av_err),
oxcf->two_pass_vbrbias / 100.0);
const double min_score = (double)(oxcf->two_pass_vbrmin_section) / 100.0;
const double max_score = (double)(oxcf->two_pass_vbrmax_section) / 100.0;
// Correction for active area. Frames with a reduced active area
// (eg due to formatting bars) have a higher error per mb for the
// remaining active MBs. The correction here assumes that coding
// 0.5N blocks of complexity 2X is a little easier than coding N
// blocks of complexity X.
modified_score *=
pow(calculate_active_area(frame_info, this_frame), ACT_AREA_CORRECTION);
// Normalize to a midpoint score.
modified_score /= DOUBLE_DIVIDE_CHECK(mean_mod_score);
return fclamp(modified_score, min_score, max_score);
}
static double calculate_norm_frame_score(const VP9_COMP *cpi,
const TWO_PASS *twopass,
const VP9EncoderConfig *oxcf,
const FIRSTPASS_STATS *this_frame,
const double av_err) {
return calc_norm_frame_score(oxcf, &cpi->frame_info, this_frame,
twopass->mean_mod_score, av_err);
}
// This function returns the maximum target rate per frame.
static int frame_max_bits(const RATE_CONTROL *rc,
const VP9EncoderConfig *oxcf) {
int64_t max_bits = ((int64_t)rc->avg_frame_bandwidth *
(int64_t)oxcf->two_pass_vbrmax_section) /
100;
if (max_bits < 0)
max_bits = 0;
else if (max_bits > rc->max_frame_bandwidth)
max_bits = rc->max_frame_bandwidth;
return (int)max_bits;
}
void vp9_init_first_pass(VP9_COMP *cpi) {
zero_stats(&cpi->twopass.total_stats);
}
void vp9_end_first_pass(VP9_COMP *cpi) {
output_stats(&cpi->twopass.total_stats);
cpi->twopass.first_pass_done = 1;
vpx_free(cpi->twopass.fp_mb_float_stats);
cpi->twopass.fp_mb_float_stats = NULL;
}
static vpx_variance_fn_t get_block_variance_fn(BLOCK_SIZE bsize) {
switch (bsize) {
case BLOCK_8X8: return vpx_mse8x8;
case BLOCK_16X8: return vpx_mse16x8;
case BLOCK_8X16: return vpx_mse8x16;
default: return vpx_mse16x16;
}
}
static unsigned int get_prediction_error(BLOCK_SIZE bsize,
const struct buf_2d *src,
const struct buf_2d *ref) {
unsigned int sse;
const vpx_variance_fn_t fn = get_block_variance_fn(bsize);
fn(src->buf, src->stride, ref->buf, ref->stride, &sse);
return sse;
}
#if CONFIG_VP9_HIGHBITDEPTH
static vpx_variance_fn_t highbd_get_block_variance_fn(BLOCK_SIZE bsize,
int bd) {
switch (bd) {
default:
switch (bsize) {
case BLOCK_8X8: return vpx_highbd_8_mse8x8;
case BLOCK_16X8: return vpx_highbd_8_mse16x8;
case BLOCK_8X16: return vpx_highbd_8_mse8x16;
default: return vpx_highbd_8_mse16x16;
}
break;
case 10:
switch (bsize) {
case BLOCK_8X8: return vpx_highbd_10_mse8x8;
case BLOCK_16X8: return vpx_highbd_10_mse16x8;
case BLOCK_8X16: return vpx_highbd_10_mse8x16;
default: return vpx_highbd_10_mse16x16;
}
break;
case 12:
switch (bsize) {
case BLOCK_8X8: return vpx_highbd_12_mse8x8;
case BLOCK_16X8: return vpx_highbd_12_mse16x8;
case BLOCK_8X16: return vpx_highbd_12_mse8x16;
default: return vpx_highbd_12_mse16x16;
}
break;
}
}
static unsigned int highbd_get_prediction_error(BLOCK_SIZE bsize,
const struct buf_2d *src,
const struct buf_2d *ref,
int bd) {
unsigned int sse;
const vpx_variance_fn_t fn = highbd_get_block_variance_fn(bsize, bd);
fn(src->buf, src->stride, ref->buf, ref->stride, &sse);
return sse;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
// Refine the motion search range according to the frame dimension
// for first pass test.
static int get_search_range(const VP9_COMP *cpi) {
int sr = 0;
const int dim = VPXMIN(cpi->initial_width, cpi->initial_height);
while ((dim << sr) < MAX_FULL_PEL_VAL) ++sr;
return sr;
}
// Reduce limits to keep the motion search within MV_MAX of ref_mv. Not doing
// this can be problematic for big videos (8K) and may cause assert failure
// (or memory violation) in mv_cost. Limits are only modified if they would
// be non-empty. Returns 1 if limits are non-empty.
static int intersect_limits_with_mv_max(MvLimits *mv_limits, const MV *ref_mv) {
const int row_min =
VPXMAX(mv_limits->row_min, (ref_mv->row + 7 - MV_MAX) >> 3);
const int row_max =
VPXMIN(mv_limits->row_max, (ref_mv->row - 1 + MV_MAX) >> 3);
const int col_min =
VPXMAX(mv_limits->col_min, (ref_mv->col + 7 - MV_MAX) >> 3);
const int col_max =
VPXMIN(mv_limits->col_max, (ref_mv->col - 1 + MV_MAX) >> 3);
if (row_min > row_max || col_min > col_max) {
return 0;
}
mv_limits->row_min = row_min;
mv_limits->row_max = row_max;
mv_limits->col_min = col_min;
mv_limits->col_max = col_max;
return 1;
}
static void first_pass_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
const MV *ref_mv, MV *best_mv,
int *best_motion_err) {
MACROBLOCKD *const xd = &x->e_mbd;
MV tmp_mv = { 0, 0 };
MV ref_mv_full = { ref_mv->row >> 3, ref_mv->col >> 3 };
int num00, tmp_err, n;
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
vp9_variance_fn_ptr_t v_fn_ptr = cpi->fn_ptr[bsize];
const int new_mv_mode_penalty = NEW_MV_MODE_PENALTY;
int step_param = 3;
int further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
const int sr = get_search_range(cpi);
const MvLimits tmp_mv_limits = x->mv_limits;
step_param += sr;
further_steps -= sr;
if (!intersect_limits_with_mv_max(&x->mv_limits, ref_mv)) {
return;
}
// Override the default variance function to use MSE.
v_fn_ptr.vf = get_block_variance_fn(bsize);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
v_fn_ptr.vf = highbd_get_block_variance_fn(bsize, xd->bd);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
// Center the initial step/diamond search on best mv.
tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg, &ref_mv_full, &tmp_mv,
step_param, x->sadperbit16, &num00,
&v_fn_ptr, ref_mv);
if (tmp_err < INT_MAX)
tmp_err = vp9_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty) tmp_err += new_mv_mode_penalty;
if (tmp_err < *best_motion_err) {
*best_motion_err = tmp_err;
*best_mv = tmp_mv;
}
// Carry out further step/diamond searches as necessary.
n = num00;
num00 = 0;
while (n < further_steps) {
++n;
if (num00) {
--num00;
} else {
tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg, &ref_mv_full, &tmp_mv,
step_param + n, x->sadperbit16, &num00,
&v_fn_ptr, ref_mv);
if (tmp_err < INT_MAX)
tmp_err = vp9_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty)
tmp_err += new_mv_mode_penalty;
if (tmp_err < *best_motion_err) {
*best_motion_err = tmp_err;
*best_mv = tmp_mv;
}
}
}
x->mv_limits = tmp_mv_limits;
}
static BLOCK_SIZE get_bsize(const VP9_COMMON *cm, int mb_row, int mb_col) {
if (2 * mb_col + 1 < cm->mi_cols) {
return 2 * mb_row + 1 < cm->mi_rows ? BLOCK_16X16 : BLOCK_16X8;
} else {
return 2 * mb_row + 1 < cm->mi_rows ? BLOCK_8X16 : BLOCK_8X8;
}
}
static int find_fp_qindex(vpx_bit_depth_t bit_depth) {
int i;
for (i = 0; i < QINDEX_RANGE; ++i)
if (vp9_convert_qindex_to_q(i, bit_depth) >= FIRST_PASS_Q) break;
if (i == QINDEX_RANGE) i--;
return i;
}
static void set_first_pass_params(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
if (!cpi->refresh_alt_ref_frame &&
(cm->current_video_frame == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY))) {
cm->frame_type = KEY_FRAME;
} else {
cm->frame_type = INTER_FRAME;
}
// Do not use periodic key frames.
cpi->rc.frames_to_key = INT_MAX;
}
// Scale an sse threshold to account for 8/10/12 bit.
static int scale_sse_threshold(VP9_COMMON *cm, int thresh) {
int ret_val = thresh;
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth) {
switch (cm->bit_depth) {
case VPX_BITS_8: ret_val = thresh; break;
case VPX_BITS_10: ret_val = thresh << 4; break;
default:
assert(cm->bit_depth == VPX_BITS_12);
ret_val = thresh << 8;
break;
}
}
#else
(void)cm;
#endif // CONFIG_VP9_HIGHBITDEPTH
return ret_val;
}
// This threshold is used to track blocks where to all intents and purposes
// the intra prediction error 0. Though the metric we test against
// is technically a sse we are mainly interested in blocks where all the pixels
// in the 8 bit domain have an error of <= 1 (where error = sse) so a
// linear scaling for 10 and 12 bit gives similar results.
#define UL_INTRA_THRESH 50
static int get_ul_intra_threshold(VP9_COMMON *cm) {
int ret_val = UL_INTRA_THRESH;
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth) {
switch (cm->bit_depth) {
case VPX_BITS_8: ret_val = UL_INTRA_THRESH; break;
case VPX_BITS_10: ret_val = UL_INTRA_THRESH << 2; break;
default:
assert(cm->bit_depth == VPX_BITS_12);
ret_val = UL_INTRA_THRESH << 4;
break;
}
}
#else
(void)cm;
#endif // CONFIG_VP9_HIGHBITDEPTH
return ret_val;
}
#define SMOOTH_INTRA_THRESH 4000
static int get_smooth_intra_threshold(VP9_COMMON *cm) {
int ret_val = SMOOTH_INTRA_THRESH;
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth) {
switch (cm->bit_depth) {
case VPX_BITS_8: ret_val = SMOOTH_INTRA_THRESH; break;
case VPX_BITS_10: ret_val = SMOOTH_INTRA_THRESH << 4; break;
default:
assert(cm->bit_depth == VPX_BITS_12);
ret_val = SMOOTH_INTRA_THRESH << 8;
break;
}
}
#else
(void)cm;
#endif // CONFIG_VP9_HIGHBITDEPTH
return ret_val;
}
#define FP_DN_THRESH 8
#define FP_MAX_DN_THRESH 24
#define KERNEL_SIZE 3
// Baseline Kernal weights for first pass noise metric
static uint8_t fp_dn_kernal_3[KERNEL_SIZE * KERNEL_SIZE] = { 1, 2, 1, 2, 4,
2, 1, 2, 1 };
// Estimate noise at a single point based on the impace of a spatial kernal
// on the point value
static int fp_estimate_point_noise(uint8_t *src_ptr, const int stride) {
int sum_weight = 0;
int sum_val = 0;
int i, j;
int max_diff = 0;
int diff;
int dn_diff;
uint8_t *tmp_ptr;
uint8_t *kernal_ptr;
uint8_t dn_val;
uint8_t centre_val = *src_ptr;
kernal_ptr = fp_dn_kernal_3;
// Apply the kernal
tmp_ptr = src_ptr - stride - 1;
for (i = 0; i < KERNEL_SIZE; ++i) {
for (j = 0; j < KERNEL_SIZE; ++j) {
diff = abs((int)centre_val - (int)tmp_ptr[j]);
max_diff = VPXMAX(max_diff, diff);
if (diff <= FP_DN_THRESH) {
sum_weight += *kernal_ptr;
sum_val += (int)tmp_ptr[j] * (int)*kernal_ptr;
}
++kernal_ptr;
}
tmp_ptr += stride;
}
if (max_diff < FP_MAX_DN_THRESH)
// Update the source value with the new filtered value
dn_val = (sum_val + (sum_weight >> 1)) / sum_weight;
else
dn_val = *src_ptr;
// return the noise energy as the square of the difference between the
// denoised and raw value.
dn_diff = (int)*src_ptr - (int)dn_val;
return dn_diff * dn_diff;
}
#if CONFIG_VP9_HIGHBITDEPTH
static int fp_highbd_estimate_point_noise(uint8_t *src_ptr, const int stride) {
int sum_weight = 0;
int sum_val = 0;
int i, j;
int max_diff = 0;
int diff;
int dn_diff;
uint8_t *tmp_ptr;
uint16_t *tmp_ptr16;
uint8_t *kernal_ptr;
uint16_t dn_val;
uint16_t centre_val = *CONVERT_TO_SHORTPTR(src_ptr);
kernal_ptr = fp_dn_kernal_3;
// Apply the kernal
tmp_ptr = src_ptr - stride - 1;
for (i = 0; i < KERNEL_SIZE; ++i) {
tmp_ptr16 = CONVERT_TO_SHORTPTR(tmp_ptr);
for (j = 0; j < KERNEL_SIZE; ++j) {
diff = abs((int)centre_val - (int)tmp_ptr16[j]);
max_diff = VPXMAX(max_diff, diff);
if (diff <= FP_DN_THRESH) {
sum_weight += *kernal_ptr;
sum_val += (int)tmp_ptr16[j] * (int)*kernal_ptr;
}
++kernal_ptr;
}
tmp_ptr += stride;
}
if (max_diff < FP_MAX_DN_THRESH)
// Update the source value with the new filtered value
dn_val = (sum_val + (sum_weight >> 1)) / sum_weight;
else
dn_val = *CONVERT_TO_SHORTPTR(src_ptr);
// return the noise energy as the square of the difference between the
// denoised and raw value.
dn_diff = (int)(*CONVERT_TO_SHORTPTR(src_ptr)) - (int)dn_val;
return dn_diff * dn_diff;
}
#endif
// Estimate noise for a block.
static int fp_estimate_block_noise(MACROBLOCK *x, BLOCK_SIZE bsize) {
#if CONFIG_VP9_HIGHBITDEPTH
MACROBLOCKD *xd = &x->e_mbd;
#endif
uint8_t *src_ptr = &x->plane[0].src.buf[0];
const int width = num_4x4_blocks_wide_lookup[bsize] * 4;
const int height = num_4x4_blocks_high_lookup[bsize] * 4;
int w, h;
int stride = x->plane[0].src.stride;
int block_noise = 0;
// Sampled points to reduce cost overhead.
for (h = 0; h < height; h += 2) {
for (w = 0; w < width; w += 2) {
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
block_noise += fp_highbd_estimate_point_noise(src_ptr, stride);
else
block_noise += fp_estimate_point_noise(src_ptr, stride);
#else
block_noise += fp_estimate_point_noise(src_ptr, stride);
#endif
++src_ptr;
}
src_ptr += (stride - width);
}
return block_noise << 2; // Scale << 2 to account for sampling.
}
// This function is called to test the functionality of row based
// multi-threading in unit tests for bit-exactness
static void accumulate_floating_point_stats(VP9_COMP *cpi,
TileDataEnc *first_tile_col) {
VP9_COMMON *const cm = &cpi->common;
int mb_row, mb_col;
first_tile_col->fp_data.intra_factor = 0;
first_tile_col->fp_data.brightness_factor = 0;
first_tile_col->fp_data.neutral_count = 0;
for (mb_row = 0; mb_row < cm->mb_rows; ++mb_row) {
for (mb_col = 0; mb_col < cm->mb_cols; ++mb_col) {
const int mb_index = mb_row * cm->mb_cols + mb_col;
first_tile_col->fp_data.intra_factor +=
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_intra_factor;
first_tile_col->fp_data.brightness_factor +=
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_brightness_factor;
first_tile_col->fp_data.neutral_count +=
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_neutral_count;
}
}
}
static void first_pass_stat_calc(VP9_COMP *cpi, FIRSTPASS_STATS *fps,
FIRSTPASS_DATA *fp_acc_data) {
VP9_COMMON *const cm = &cpi->common;
// The minimum error here insures some bit allocation to frames even
// in static regions. The allocation per MB declines for larger formats
// where the typical "real" energy per MB also falls.
// Initial estimate here uses sqrt(mbs) to define the min_err, where the
// number of mbs is proportional to the image area.
const int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE) ? cpi->initial_mbs
: cpi->common.MBs;
const double min_err = 200 * sqrt(num_mbs);
// Clamp the image start to rows/2. This number of rows is discarded top
// and bottom as dead data so rows / 2 means the frame is blank.
if ((fp_acc_data->image_data_start_row > cm->mb_rows / 2) ||
(fp_acc_data->image_data_start_row == INVALID_ROW)) {
fp_acc_data->image_data_start_row = cm->mb_rows / 2;
}
// Exclude any image dead zone
if (fp_acc_data->image_data_start_row > 0) {
fp_acc_data->intra_skip_count =
VPXMAX(0, fp_acc_data->intra_skip_count -
(fp_acc_data->image_data_start_row * cm->mb_cols * 2));
}
fp_acc_data->intra_factor = fp_acc_data->intra_factor / (double)num_mbs;
fp_acc_data->brightness_factor =
fp_acc_data->brightness_factor / (double)num_mbs;
fps->weight = fp_acc_data->intra_factor * fp_acc_data->brightness_factor;
fps->frame = cm->current_video_frame;
fps->spatial_layer_id = cpi->svc.spatial_layer_id;
fps->coded_error =
((double)(fp_acc_data->coded_error >> 8) + min_err) / num_mbs;
fps->sr_coded_error =
((double)(fp_acc_data->sr_coded_error >> 8) + min_err) / num_mbs;
fps->intra_error =
((double)(fp_acc_data->intra_error >> 8) + min_err) / num_mbs;
fps->frame_noise_energy =
(double)(fp_acc_data->frame_noise_energy) / (double)num_mbs;
fps->count = 1.0;
fps->pcnt_inter = (double)(fp_acc_data->intercount) / num_mbs;
fps->pcnt_second_ref = (double)(fp_acc_data->second_ref_count) / num_mbs;
fps->pcnt_neutral = (double)(fp_acc_data->neutral_count) / num_mbs;
fps->pcnt_intra_low = (double)(fp_acc_data->intra_count_low) / num_mbs;
fps->pcnt_intra_high = (double)(fp_acc_data->intra_count_high) / num_mbs;
fps->intra_skip_pct = (double)(fp_acc_data->intra_skip_count) / num_mbs;
fps->intra_smooth_pct = (double)(fp_acc_data->intra_smooth_count) / num_mbs;
fps->inactive_zone_rows = (double)(fp_acc_data->image_data_start_row);
// Currently set to 0 as most issues relate to letter boxing.
fps->inactive_zone_cols = (double)0;
if (fp_acc_data->mvcount > 0) {
fps->MVr = (double)(fp_acc_data->sum_mvr) / fp_acc_data->mvcount;
fps->mvr_abs = (double)(fp_acc_data->sum_mvr_abs) / fp_acc_data->mvcount;
fps->MVc = (double)(fp_acc_data->sum_mvc) / fp_acc_data->mvcount;
fps->mvc_abs = (double)(fp_acc_data->sum_mvc_abs) / fp_acc_data->mvcount;
fps->MVrv = ((double)(fp_acc_data->sum_mvrs) -
((double)(fp_acc_data->sum_mvr) * (fp_acc_data->sum_mvr) /
fp_acc_data->mvcount)) /
fp_acc_data->mvcount;
fps->MVcv = ((double)(fp_acc_data->sum_mvcs) -
((double)(fp_acc_data->sum_mvc) * (fp_acc_data->sum_mvc) /
fp_acc_data->mvcount)) /
fp_acc_data->mvcount;
fps->mv_in_out_count =
(double)(fp_acc_data->sum_in_vectors) / (fp_acc_data->mvcount * 2);
fps->pcnt_motion = (double)(fp_acc_data->mvcount) / num_mbs;
} else {
fps->MVr = 0.0;
fps->mvr_abs = 0.0;
fps->MVc = 0.0;
fps->mvc_abs = 0.0;
fps->MVrv = 0.0;
fps->MVcv = 0.0;
fps->mv_in_out_count = 0.0;
fps->pcnt_motion = 0.0;
}
}
static void accumulate_fp_mb_row_stat(TileDataEnc *this_tile,
FIRSTPASS_DATA *fp_acc_data) {
this_tile->fp_data.intra_factor += fp_acc_data->intra_factor;
this_tile->fp_data.brightness_factor += fp_acc_data->brightness_factor;
this_tile->fp_data.coded_error += fp_acc_data->coded_error;
this_tile->fp_data.sr_coded_error += fp_acc_data->sr_coded_error;
this_tile->fp_data.frame_noise_energy += fp_acc_data->frame_noise_energy;
this_tile->fp_data.intra_error += fp_acc_data->intra_error;
this_tile->fp_data.intercount += fp_acc_data->intercount;
this_tile->fp_data.second_ref_count += fp_acc_data->second_ref_count;
this_tile->fp_data.neutral_count += fp_acc_data->neutral_count;
this_tile->fp_data.intra_count_low += fp_acc_data->intra_count_low;
this_tile->fp_data.intra_count_high += fp_acc_data->intra_count_high;
this_tile->fp_data.intra_skip_count += fp_acc_data->intra_skip_count;
this_tile->fp_data.mvcount += fp_acc_data->mvcount;
this_tile->fp_data.sum_mvr += fp_acc_data->sum_mvr;
this_tile->fp_data.sum_mvr_abs += fp_acc_data->sum_mvr_abs;
this_tile->fp_data.sum_mvc += fp_acc_data->sum_mvc;
this_tile->fp_data.sum_mvc_abs += fp_acc_data->sum_mvc_abs;
this_tile->fp_data.sum_mvrs += fp_acc_data->sum_mvrs;
this_tile->fp_data.sum_mvcs += fp_acc_data->sum_mvcs;
this_tile->fp_data.sum_in_vectors += fp_acc_data->sum_in_vectors;
this_tile->fp_data.intra_smooth_count += fp_acc_data->intra_smooth_count;
this_tile->fp_data.image_data_start_row =
VPXMIN(this_tile->fp_data.image_data_start_row,
fp_acc_data->image_data_start_row) == INVALID_ROW
? VPXMAX(this_tile->fp_data.image_data_start_row,
fp_acc_data->image_data_start_row)
: VPXMIN(this_tile->fp_data.image_data_start_row,
fp_acc_data->image_data_start_row);
}
#if CONFIG_RATE_CTRL
static void store_fp_motion_vector(VP9_COMP *cpi, const MV *mv,
const int mb_row, const int mb_col,
const int is_second_mv) {
VP9_COMMON *const cm = &cpi->common;
const int mb_index = mb_row * cm->mb_cols + mb_col;
MOTION_VECTOR_INFO *this_motion_vector_info =
&cpi->fp_motion_vector_info[mb_index];
if (!is_second_mv) {
this_motion_vector_info->ref_frame[0] = LAST_FRAME;
this_motion_vector_info->mv[0].as_mv.row = mv->row;
this_motion_vector_info->mv[0].as_mv.col = mv->col;
return;
}
this_motion_vector_info->ref_frame[1] = GOLDEN_FRAME;
this_motion_vector_info->mv[1].as_mv.row = mv->row;
this_motion_vector_info->mv[1].as_mv.col = mv->col;
}
#endif // CONFIG_RATE_CTRL
#define NZ_MOTION_PENALTY 128
#define INTRA_MODE_PENALTY 1024
void vp9_first_pass_encode_tile_mb_row(VP9_COMP *cpi, ThreadData *td,
FIRSTPASS_DATA *fp_acc_data,
TileDataEnc *tile_data, MV *best_ref_mv,
int mb_row) {
int mb_col;
MACROBLOCK *const x = &td->mb;
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &x->e_mbd;
TileInfo tile = tile_data->tile_info;
const int mb_col_start = ROUND_POWER_OF_TWO(tile.mi_col_start, 1);
const int mb_col_end = ROUND_POWER_OF_TWO(tile.mi_col_end, 1);
struct macroblock_plane *const p = x->plane;
struct macroblockd_plane *const pd = xd->plane;
const PICK_MODE_CONTEXT *ctx = &td->pc_root->none;
int i, c;
int num_mb_cols = get_num_cols(tile_data->tile_info, 1);
int recon_yoffset, recon_uvoffset;
const int intrapenalty = INTRA_MODE_PENALTY;
const MV zero_mv = { 0, 0 };
int recon_y_stride, recon_uv_stride, uv_mb_height;
YV12_BUFFER_CONFIG *const lst_yv12 = get_ref_frame_buffer(cpi, LAST_FRAME);
YV12_BUFFER_CONFIG *gld_yv12 = get_ref_frame_buffer(cpi, GOLDEN_FRAME);
YV12_BUFFER_CONFIG *const new_yv12 = get_frame_new_buffer(cm);
const YV12_BUFFER_CONFIG *first_ref_buf = lst_yv12;
MODE_INFO mi_above, mi_left;
double mb_intra_factor;
double mb_brightness_factor;
double mb_neutral_count;
int scaled_low_intra_thresh = scale_sse_threshold(cm, LOW_I_THRESH);
// First pass code requires valid last and new frame buffers.
assert(new_yv12 != NULL);
assert(frame_is_intra_only(cm) || (lst_yv12 != NULL));
xd->mi = cm->mi_grid_visible + xd->mi_stride * (mb_row << 1) + mb_col_start;
xd->mi[0] = cm->mi + xd->mi_stride * (mb_row << 1) + mb_col_start;
for (i = 0; i < MAX_MB_PLANE; ++i) {
p[i].coeff = ctx->coeff_pbuf[i][1];
p[i].qcoeff = ctx->qcoeff_pbuf[i][1];
pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
p[i].eobs = ctx->eobs_pbuf[i][1];
}
recon_y_stride = new_yv12->y_stride;
recon_uv_stride = new_yv12->uv_stride;
uv_mb_height = 16 >> (new_yv12->y_height > new_yv12->uv_height);
// Reset above block coeffs.
recon_yoffset = (mb_row * recon_y_stride * 16) + mb_col_start * 16;
recon_uvoffset =
(mb_row * recon_uv_stride * uv_mb_height) + mb_col_start * uv_mb_height;
// Set up limit values for motion vectors to prevent them extending
// outside the UMV borders.
x->mv_limits.row_min = -((mb_row * 16) + BORDER_MV_PIXELS_B16);
x->mv_limits.row_max =
((cm->mb_rows - 1 - mb_row) * 16) + BORDER_MV_PIXELS_B16;
for (mb_col = mb_col_start, c = 0; mb_col < mb_col_end; ++mb_col, c++) {
int this_error;
int this_intra_error;
const int use_dc_pred = (mb_col || mb_row) && (!mb_col || !mb_row);
const BLOCK_SIZE bsize = get_bsize(cm, mb_row, mb_col);
double log_intra;
int level_sample;
const int mb_index = mb_row * cm->mb_cols + mb_col;
#if CONFIG_FP_MB_STATS
const int mb_index = mb_row * cm->mb_cols + mb_col;
#endif
(*(cpi->row_mt_sync_read_ptr))(&tile_data->row_mt_sync, mb_row, c);
// Adjust to the next column of MBs.
x->plane[0].src.buf = cpi->Source->y_buffer +
mb_row * 16 * x->plane[0].src.stride + mb_col * 16;
x->plane[1].src.buf = cpi->Source->u_buffer +
mb_row * uv_mb_height * x->plane[1].src.stride +
mb_col * uv_mb_height;
x->plane[2].src.buf = cpi->Source->v_buffer +
mb_row * uv_mb_height * x->plane[1].src.stride +
mb_col * uv_mb_height;
vpx_clear_system_state();
xd->plane[0].dst.buf = new_yv12->y_buffer + recon_yoffset;
xd->plane[1].dst.buf = new_yv12->u_buffer + recon_uvoffset;
xd->plane[2].dst.buf = new_yv12->v_buffer + recon_uvoffset;
xd->mi[0]->sb_type = bsize;
xd->mi[0]->ref_frame[0] = INTRA_FRAME;
set_mi_row_col(xd, &tile, mb_row << 1, num_8x8_blocks_high_lookup[bsize],
mb_col << 1, num_8x8_blocks_wide_lookup[bsize], cm->mi_rows,
cm->mi_cols);
// Are edges available for intra prediction?
// Since the firstpass does not populate the mi_grid_visible,
// above_mi/left_mi must be overwritten with a nonzero value when edges
// are available. Required by vp9_predict_intra_block().
xd->above_mi = (mb_row != 0) ? &mi_above : NULL;
xd->left_mi = ((mb_col << 1) > tile.mi_col_start) ? &mi_left : NULL;
// Do intra 16x16 prediction.
x->skip_encode = 0;
x->fp_src_pred = 0;
// Do intra prediction based on source pixels for tile boundaries
if (mb_col == mb_col_start && mb_col != 0) {
xd->left_mi = &mi_left;
x->fp_src_pred = 1;
}
xd->mi[0]->mode = DC_PRED;
xd->mi[0]->tx_size =
use_dc_pred ? (bsize >= BLOCK_16X16 ? TX_16X16 : TX_8X8) : TX_4X4;
// Fix - zero the 16x16 block first. This ensures correct this_error for
// block sizes smaller than 16x16.
vp9_zero_array(x->plane[0].src_diff, 256);
vp9_encode_intra_block_plane(x, bsize, 0, 0);
this_error = vpx_get_mb_ss(x->plane[0].src_diff);
this_intra_error = this_error;
// Keep a record of blocks that have very low intra error residual
// (i.e. are in effect completely flat and untextured in the intra
// domain). In natural videos this is uncommon, but it is much more
// common in animations, graphics and screen content, so may be used
// as a signal to detect these types of content.
if (this_error < get_ul_intra_threshold(cm)) {
++(fp_acc_data->intra_skip_count);
} else if ((mb_col > 0) &&
(fp_acc_data->image_data_start_row == INVALID_ROW)) {
fp_acc_data->image_data_start_row = mb_row;
}
// Blocks that are mainly smooth in the intra domain.
// Some special accounting for CQ but also these are better for testing
// noise levels.
if (this_error < get_smooth_intra_threshold(cm)) {
++(fp_acc_data->intra_smooth_count);
}
// Special case noise measurement for first frame.
if (cm->current_video_frame == 0) {
if (this_intra_error < scale_sse_threshold(cm, LOW_I_THRESH)) {
fp_acc_data->frame_noise_energy += fp_estimate_block_noise(x, bsize);
} else {
fp_acc_data->frame_noise_energy += (int64_t)SECTION_NOISE_DEF;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth) {
switch (cm->bit_depth) {
case VPX_BITS_8: break;
case VPX_BITS_10: this_error >>= 4; break;
default:
assert(cm->bit_depth == VPX_BITS_12);
this_error >>= 8;
break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
vpx_clear_system_state();
log_intra = log(this_error + 1.0);
if (log_intra < 10.0) {
mb_intra_factor = 1.0 + ((10.0 - log_intra) * 0.05);
fp_acc_data->intra_factor += mb_intra_factor;
if (cpi->row_mt_bit_exact)
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_intra_factor =
mb_intra_factor;
} else {
fp_acc_data->intra_factor += 1.0;
if (cpi->row_mt_bit_exact)
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_intra_factor = 1.0;
}
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth)
level_sample = CONVERT_TO_SHORTPTR(x->plane[0].src.buf)[0];
else
level_sample = x->plane[0].src.buf[0];
#else
level_sample = x->plane[0].src.buf[0];
#endif
if ((level_sample < DARK_THRESH) && (log_intra < 9.0)) {
mb_brightness_factor = 1.0 + (0.01 * (DARK_THRESH - level_sample));
fp_acc_data->brightness_factor += mb_brightness_factor;
if (cpi->row_mt_bit_exact)
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_brightness_factor =
mb_brightness_factor;
} else {
fp_acc_data->brightness_factor += 1.0;
if (cpi->row_mt_bit_exact)
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_brightness_factor =
1.0;
}
// Intrapenalty below deals with situations where the intra and inter
// error scores are very low (e.g. a plain black frame).
// We do not have special cases in first pass for 0,0 and nearest etc so
// all inter modes carry an overhead cost estimate for the mv.
// When the error score is very low this causes us to pick all or lots of
// INTRA modes and throw lots of key frames.
// This penalty adds a cost matching that of a 0,0 mv to the intra case.
this_error += intrapenalty;
// Accumulate the intra error.
fp_acc_data->intra_error += (int64_t)this_error;
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
// initialization
cpi->twopass.frame_mb_stats_buf[mb_index] = 0;
}
#endif
// Set up limit values for motion vectors to prevent them extending
// outside the UMV borders.
x->mv_limits.col_min = -((mb_col * 16) + BORDER_MV_PIXELS_B16);
x->mv_limits.col_max =
((cm->mb_cols - 1 - mb_col) * 16) + BORDER_MV_PIXELS_B16;
// Other than for the first frame do a motion search.
if (cm->current_video_frame > 0) {
int tmp_err, motion_error, this_motion_error, raw_motion_error;
// Assume 0,0 motion with no mv overhead.
MV mv = { 0, 0 }, tmp_mv = { 0, 0 };
struct buf_2d unscaled_last_source_buf_2d;
vp9_variance_fn_ptr_t v_fn_ptr = cpi->fn_ptr[bsize];
xd->plane[0].pre[0].buf = first_ref_buf->y_buffer + recon_yoffset;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &xd->plane[0].pre[0], xd->bd);
this_motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &xd->plane[0].pre[0], 8);
} else {
motion_error =
get_prediction_error(bsize, &x->plane[0].src, &xd->plane[0].pre[0]);
this_motion_error = motion_error;
}
#else
motion_error =
get_prediction_error(bsize, &x->plane[0].src, &xd->plane[0].pre[0]);
this_motion_error = motion_error;
#endif // CONFIG_VP9_HIGHBITDEPTH
// Compute the motion error of the 0,0 motion using the last source
// frame as the reference. Skip the further motion search on
// reconstructed frame if this error is very small.
unscaled_last_source_buf_2d.buf =
cpi->unscaled_last_source->y_buffer + recon_yoffset;
unscaled_last_source_buf_2d.stride = cpi->unscaled_last_source->y_stride;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
raw_motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &unscaled_last_source_buf_2d, xd->bd);
} else {
raw_motion_error = get_prediction_error(bsize, &x->plane[0].src,
&unscaled_last_source_buf_2d);
}
#else
raw_motion_error = get_prediction_error(bsize, &x->plane[0].src,
&unscaled_last_source_buf_2d);
#endif // CONFIG_VP9_HIGHBITDEPTH
if (raw_motion_error > NZ_MOTION_PENALTY) {
// Test last reference frame using the previous best mv as the
// starting point (best reference) for the search.
first_pass_motion_search(cpi, x, best_ref_mv, &mv, &motion_error);
v_fn_ptr.vf = get_block_variance_fn(bsize);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
v_fn_ptr.vf = highbd_get_block_variance_fn(bsize, 8);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
this_motion_error =
vp9_get_mvpred_var(x, &mv, best_ref_mv, &v_fn_ptr, 0);
// If the current best reference mv is not centered on 0,0 then do a
// 0,0 based search as well.
if (!is_zero_mv(best_ref_mv)) {
tmp_err = INT_MAX;
first_pass_motion_search(cpi, x, &zero_mv, &tmp_mv, &tmp_err);
if (tmp_err < motion_error) {
motion_error = tmp_err;
mv = tmp_mv;
this_motion_error =
vp9_get_mvpred_var(x, &tmp_mv, &zero_mv, &v_fn_ptr, 0);
}
}
#if CONFIG_RATE_CTRL
store_fp_motion_vector(cpi, &mv, mb_row, mb_col, /*is_second_mv=*/0);
#endif // CONFIG_RAGE_CTRL
// Search in an older reference frame.
if ((cm->current_video_frame > 1) && gld_yv12 != NULL) {
// Assume 0,0 motion with no mv overhead.
int gf_motion_error;
xd->plane[0].pre[0].buf = gld_yv12->y_buffer + recon_yoffset;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
gf_motion_error = highbd_get_prediction_error(
bsize, &x->plane[0].src, &xd->plane[0].pre[0], xd->bd);
} else {
gf_motion_error = get_prediction_error(bsize, &x->plane[0].src,
&xd->plane[0].pre[0]);
}
#else
gf_motion_error = get_prediction_error(bsize, &x->plane[0].src,
&xd->plane[0].pre[0]);
#endif // CONFIG_VP9_HIGHBITDEPTH
first_pass_motion_search(cpi, x, &zero_mv, &tmp_mv, &gf_motion_error);
#if CONFIG_RATE_CTRL
store_fp_motion_vector(cpi, &tmp_mv, mb_row, mb_col,
/*is_second_mv=*/1);
#endif // CONFIG_RAGE_CTRL
if (gf_motion_error < motion_error && gf_motion_error < this_error)
++(fp_acc_data->second_ref_count);
// Reset to last frame as reference buffer.
xd->plane[0].pre[0].buf = first_ref_buf->y_buffer + recon_yoffset;
xd->plane[1].pre[0].buf = first_ref_buf->u_buffer + recon_uvoffset;
xd->plane[2].pre[0].buf = first_ref_buf->v_buffer + recon_uvoffset;
// In accumulating a score for the older reference frame take the
// best of the motion predicted score and the intra coded error
// (just as will be done for) accumulation of "coded_error" for
// the last frame.
if (gf_motion_error < this_error)
fp_acc_data->sr_coded_error += gf_motion_error;
else
fp_acc_data->sr_coded_error += this_error;
} else {
fp_acc_data->sr_coded_error += motion_error;
}
} else {
fp_acc_data->sr_coded_error += motion_error;
}
// Start by assuming that intra mode is best.
best_ref_mv->row = 0;
best_ref_mv->col = 0;
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
// intra prediction statistics
cpi->twopass.frame_mb_stats_buf[mb_index] = 0;
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_DCINTRA_MASK;
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_MOTION_ZERO_MASK;
if (this_error > FPMB_ERROR_LARGE_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_ERROR_LARGE_MASK;
} else if (this_error < FPMB_ERROR_SMALL_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_ERROR_SMALL_MASK;
}
}
#endif
if (motion_error <= this_error) {
vpx_clear_system_state();
// Keep a count of cases where the inter and intra were very close
// and very low. This helps with scene cut detection for example in
// cropped clips with black bars at the sides or top and bottom.
if (((this_error - intrapenalty) * 9 <= motion_error * 10) &&
(this_error < (2 * intrapenalty))) {
fp_acc_data->neutral_count += 1.0;
if (cpi->row_mt_bit_exact)
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_neutral_count =
1.0;
// Also track cases where the intra is not much worse than the inter
// and use this in limiting the GF/arf group length.
} else if ((this_error > NCOUNT_INTRA_THRESH) &&
(this_error < (NCOUNT_INTRA_FACTOR * motion_error))) {
mb_neutral_count =
(double)motion_error / DOUBLE_DIVIDE_CHECK((double)this_error);
fp_acc_data->neutral_count += mb_neutral_count;
if (cpi->row_mt_bit_exact)
cpi->twopass.fp_mb_float_stats[mb_index].frame_mb_neutral_count =
mb_neutral_count;
}
mv.row *= 8;
mv.col *= 8;
this_error = motion_error;
xd->mi[0]->mode = NEWMV;
xd->mi[0]->mv[0].as_mv = mv;
xd->mi[0]->tx_size = TX_4X4;
xd->mi[0]->ref_frame[0] = LAST_FRAME;
xd->mi[0]->ref_frame[1] = NONE;
vp9_build_inter_predictors_sby(xd, mb_row << 1, mb_col << 1, bsize);
vp9_encode_sby_pass1(x, bsize);
fp_acc_data->sum_mvr += mv.row;
fp_acc_data->sum_mvr_abs += abs(mv.row);
fp_acc_data->sum_mvc += mv.col;
fp_acc_data->sum_mvc_abs += abs(mv.col);
fp_acc_data->sum_mvrs += mv.row * mv.row;
fp_acc_data->sum_mvcs += mv.col * mv.col;
++(fp_acc_data->intercount);
*best_ref_mv = mv;
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
// inter prediction statistics
cpi->twopass.frame_mb_stats_buf[mb_index] = 0;
cpi->twopass.frame_mb_stats_buf[mb_index] &= ~FPMB_DCINTRA_MASK;
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_MOTION_ZERO_MASK;
if (this_error > FPMB_ERROR_LARGE_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_ERROR_LARGE_MASK;
} else if (this_error < FPMB_ERROR_SMALL_TH) {
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_ERROR_SMALL_MASK;
}
}
#endif
if (!is_zero_mv(&mv)) {
++(fp_acc_data->mvcount);
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
cpi->twopass.frame_mb_stats_buf[mb_index] &= ~FPMB_MOTION_ZERO_MASK;
// check estimated motion direction
if (mv.as_mv.col > 0 && mv.as_mv.col >= abs(mv.as_mv.row)) {
// right direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_RIGHT_MASK;
} else if (mv.as_mv.row < 0 &&
abs(mv.as_mv.row) >= abs(mv.as_mv.col)) {
// up direction
cpi->twopass.frame_mb_stats_buf[mb_index] |= FPMB_MOTION_UP_MASK;
} else if (mv.as_mv.col < 0 &&
abs(mv.as_mv.col) >= abs(mv.as_mv.row)) {
// left direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_LEFT_MASK;
} else {
// down direction
cpi->twopass.frame_mb_stats_buf[mb_index] |=
FPMB_MOTION_DOWN_MASK;
}
}
#endif
// Does the row vector point inwards or outwards?
if (mb_row < cm->mb_rows / 2) {
if (mv.row > 0)
--(fp_acc_data->sum_in_vectors);
else if (mv.row < 0)
++(fp_acc_data->sum_in_vectors);
} else if (mb_row > cm->mb_rows / 2) {
if (mv.row > 0)
++(fp_acc_data->sum_in_vectors);
else if (mv.row < 0)
--(fp_acc_data->sum_in_vectors);
}
// Does the col vector point inwards or outwards?
if (mb_col < cm->mb_cols / 2) {
if (mv.col > 0)
--(fp_acc_data->sum_in_vectors);
else if (mv.col < 0)
++(fp_acc_data->sum_in_vectors);
} else if (mb_col > cm->mb_cols / 2) {
if (mv.col > 0)
++(fp_acc_data->sum_in_vectors);
else if (mv.col < 0)
--(fp_acc_data->sum_in_vectors);
}
}
if (this_intra_error < scaled_low_intra_thresh) {
fp_acc_data->frame_noise_energy += fp_estimate_block_noise(x, bsize);
} else {
fp_acc_data->frame_noise_energy += (int64_t)SECTION_NOISE_DEF;
}
} else { // Intra < inter error
if (this_intra_error < scaled_low_intra_thresh) {
fp_acc_data->frame_noise_energy += fp_estimate_block_noise(x, bsize);
if (this_motion_error < scaled_low_intra_thresh) {
fp_acc_data->intra_count_low += 1.0;
} else {
fp_acc_data->intra_count_high += 1.0;
}
} else {
fp_acc_data->frame_noise_energy += (int64_t)SECTION_NOISE_DEF;
fp_acc_data->intra_count_high += 1.0;
}
}
} else {
fp_acc_data->sr_coded_error += (int64_t)this_error;
}
fp_acc_data->coded_error += (int64_t)this_error;
recon_yoffset += 16;
recon_uvoffset += uv_mb_height;
// Accumulate row level stats to the corresponding tile stats
if (cpi->row_mt && mb_col == mb_col_end - 1)
accumulate_fp_mb_row_stat(tile_data, fp_acc_data);
(*(cpi->row_mt_sync_write_ptr))(&tile_data->row_mt_sync, mb_row, c,
num_mb_cols);
}
vpx_clear_system_state();
}
static void first_pass_encode(VP9_COMP *cpi, FIRSTPASS_DATA *fp_acc_data) {
VP9_COMMON *const cm = &cpi->common;
int mb_row;
TileDataEnc tile_data;
TileInfo *tile = &tile_data.tile_info;
MV zero_mv = { 0, 0 };
MV best_ref_mv;
// Tiling is ignored in the first pass.
vp9_tile_init(tile, cm, 0, 0);
for (mb_row = 0; mb_row < cm->mb_rows; ++mb_row) {
best_ref_mv = zero_mv;
vp9_first_pass_encode_tile_mb_row(cpi, &cpi->td, fp_acc_data, &tile_data,
&best_ref_mv, mb_row);
}
}
void vp9_first_pass(VP9_COMP *cpi, const struct lookahead_entry *source) {
MACROBLOCK *const x = &cpi->td.mb;
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &x->e_mbd;
TWO_PASS *twopass = &cpi->twopass;
YV12_BUFFER_CONFIG *const lst_yv12 = get_ref_frame_buffer(cpi, LAST_FRAME);
YV12_BUFFER_CONFIG *gld_yv12 = get_ref_frame_buffer(cpi, GOLDEN_FRAME);
YV12_BUFFER_CONFIG *const new_yv12 = get_frame_new_buffer(cm);
const YV12_BUFFER_CONFIG *first_ref_buf = lst_yv12;
BufferPool *const pool = cm->buffer_pool;
FIRSTPASS_DATA fp_temp_data;
FIRSTPASS_DATA *fp_acc_data = &fp_temp_data;
vpx_clear_system_state();
vp9_zero(fp_temp_data);
fp_acc_data->image_data_start_row = INVALID_ROW;
// First pass code requires valid last and new frame buffers.
assert(new_yv12 != NULL);
assert(frame_is_intra_only(cm) || (lst_yv12 != NULL));
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
vp9_zero_array(cpi->twopass.frame_mb_stats_buf, cm->initial_mbs);
}
#endif
set_first_pass_params(cpi);
vp9_set_quantizer(cpi, find_fp_qindex(cm->bit_depth));
vp9_setup_block_planes(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
vp9_setup_src_planes(x, cpi->Source, 0, 0);
vp9_setup_dst_planes(xd->plane, new_yv12, 0, 0);
if (!frame_is_intra_only(cm)) {
vp9_setup_pre_planes(xd, 0, first_ref_buf, 0, 0, NULL);
}
xd->mi = cm->mi_grid_visible;
xd->mi[0] = cm->mi;
vp9_frame_init_quantizer(cpi);
x->skip_recode = 0;
vp9_init_mv_probs(cm);
vp9_initialize_rd_consts(cpi);
cm->log2_tile_rows = 0;
if (cpi->row_mt_bit_exact && cpi->twopass.fp_mb_float_stats == NULL)
CHECK_MEM_ERROR(
cm, cpi->twopass.fp_mb_float_stats,
vpx_calloc(cm->MBs * sizeof(*cpi->twopass.fp_mb_float_stats), 1));
{
FIRSTPASS_STATS fps;
TileDataEnc *first_tile_col;
if (!cpi->row_mt) {
cm->log2_tile_cols = 0;
cpi->row_mt_sync_read_ptr = vp9_row_mt_sync_read_dummy;
cpi->row_mt_sync_write_ptr = vp9_row_mt_sync_write_dummy;
first_pass_encode(cpi, fp_acc_data);
first_pass_stat_calc(cpi, &fps, fp_acc_data);
} else {
cpi->row_mt_sync_read_ptr = vp9_row_mt_sync_read;
cpi->row_mt_sync_write_ptr = vp9_row_mt_sync_write;
if (cpi->row_mt_bit_exact) {
cm->log2_tile_cols = 0;
vp9_zero_array(cpi->twopass.fp_mb_float_stats, cm->MBs);
}
vp9_encode_fp_row_mt(cpi);
first_tile_col = &cpi->tile_data[0];
if (cpi->row_mt_bit_exact)
accumulate_floating_point_stats(cpi, first_tile_col);
first_pass_stat_calc(cpi, &fps, &(first_tile_col->fp_data));
}
// Dont allow a value of 0 for duration.
// (Section duration is also defaulted to minimum of 1.0).
fps.duration = VPXMAX(1.0, (double)(source->ts_end - source->ts_start));
// Don't want to do output stats with a stack variable!
twopass->this_frame_stats = fps;
output_stats(&twopass->this_frame_stats);
accumulate_stats(&twopass->total_stats, &fps);
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
output_fpmb_stats(twopass->frame_mb_stats_buf, cm, cpi->output_pkt_list);
}
#endif
}
// Copy the previous Last Frame back into gf and and arf buffers if
// the prediction is good enough... but also don't allow it to lag too far.
if ((twopass->sr_update_lag > 3) ||
((cm->current_video_frame > 0) &&
(twopass->this_frame_stats.pcnt_inter > 0.20) &&
((twopass->this_frame_stats.intra_error /
DOUBLE_DIVIDE_CHECK(twopass->this_frame_stats.coded_error)) > 2.0))) {
if (gld_yv12 != NULL) {
ref_cnt_fb(pool->frame_bufs, &cm->ref_frame_map[cpi->gld_fb_idx],
cm->ref_frame_map[cpi->lst_fb_idx]);
}
twopass->sr_update_lag = 1;
} else {
++twopass->sr_update_lag;
}
vpx_extend_frame_borders(new_yv12);
// The frame we just compressed now becomes the last frame.
ref_cnt_fb(pool->frame_bufs, &cm->ref_frame_map[cpi->lst_fb_idx],
cm->new_fb_idx);
// Special case for the first frame. Copy into the GF buffer as a second
// reference.
if (cm->current_video_frame == 0 && cpi->gld_fb_idx != INVALID_IDX) {
ref_cnt_fb(pool->frame_bufs, &cm->ref_frame_map[cpi->gld_fb_idx],
cm->ref_frame_map[cpi->lst_fb_idx]);
}
// Use this to see what the first pass reconstruction looks like.
if (0) {
char filename[512];
FILE *recon_file;
snprintf(filename, sizeof(filename), "enc%04d.yuv",
(int)cm->current_video_frame);
if (cm->current_video_frame == 0)
recon_file = fopen(filename, "wb");
else
recon_file = fopen(filename, "ab");
(void)fwrite(lst_yv12->buffer_alloc, lst_yv12->frame_size, 1, recon_file);
fclose(recon_file);
}
// In the first pass, every frame is considered as a show frame.
update_frame_indexes(cm, /*show_frame=*/1);
if (cpi->use_svc) vp9_inc_frame_in_layer(cpi);
}
static const double q_pow_term[(QINDEX_RANGE >> 5) + 1] = { 0.65, 0.70, 0.75,
0.85, 0.90, 0.90,
0.90, 1.00, 1.25 };
static double calc_correction_factor(double err_per_mb, double err_divisor,
int q) {
const double error_term = err_per_mb / DOUBLE_DIVIDE_CHECK(err_divisor);
const int index = q >> 5;
double power_term;
assert((index >= 0) && (index < (QINDEX_RANGE >> 5)));
// Adjustment based on quantizer to the power term.
power_term =
q_pow_term[index] +
(((q_pow_term[index + 1] - q_pow_term[index]) * (q % 32)) / 32.0);
// Calculate correction factor.
if (power_term < 1.0) assert(error_term >= 0.0);
return fclamp(pow(error_term, power_term), 0.05, 5.0);
}
static double wq_err_divisor(VP9_COMP *cpi) {
const VP9_COMMON *const cm = &cpi->common;
unsigned int screen_area = (cm->width * cm->height);
// Use a different error per mb factor for calculating boost for
// different formats.
if (screen_area <= 640 * 360) {
return 115.0;
} else if (screen_area < 1280 * 720) {
return 125.0;
} else if (screen_area <= 1920 * 1080) {
return 130.0;
} else if (screen_area < 3840 * 2160) {
return 150.0;
}
// Fall through to here only for 4K and above.
return 200.0;
}
#define NOISE_FACTOR_MIN 0.9
#define NOISE_FACTOR_MAX 1.1
static int get_twopass_worst_quality(VP9_COMP *cpi, const double section_err,
double inactive_zone, double section_noise,
int section_target_bandwidth) {
const RATE_CONTROL *const rc = &cpi->rc;
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
TWO_PASS *const twopass = &cpi->twopass;
double last_group_rate_err;
// Clamp the target rate to VBR min / max limts.
const int target_rate =
vp9_rc_clamp_pframe_target_size(cpi, section_target_bandwidth);
double noise_factor = pow((section_noise / SECTION_NOISE_DEF), 0.5);
noise_factor = fclamp(noise_factor, NOISE_FACTOR_MIN, NOISE_FACTOR_MAX);
inactive_zone = fclamp(inactive_zone, 0.0, 1.0);
// TODO(jimbankoski): remove #if here or below when this has been
// well tested.
#if CONFIG_ALWAYS_ADJUST_BPM
// based on recent history adjust expectations of bits per macroblock.
last_group_rate_err =
(double)twopass->rolling_arf_group_actual_bits /
DOUBLE_DIVIDE_CHECK((double)twopass->rolling_arf_group_target_bits);
last_group_rate_err = VPXMAX(0.25, VPXMIN(4.0, last_group_rate_err));
twopass->bpm_factor *= (3.0 + last_group_rate_err) / 4.0;
twopass->bpm_factor = VPXMAX(0.25, VPXMIN(4.0, twopass->bpm_factor));
#endif
if (target_rate <= 0) {
return rc->worst_quality; // Highest value allowed
} else {
const int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE)
? cpi->initial_mbs
: cpi->common.MBs;
const double active_pct = VPXMAX(0.01, 1.0 - inactive_zone);
const int active_mbs = (int)VPXMAX(1, (double)num_mbs * active_pct);
const double av_err_per_mb = section_err / active_pct;
const double speed_term = 1.0 + 0.04 * oxcf->speed;
const int target_norm_bits_per_mb =
(int)(((uint64_t)target_rate << BPER_MB_NORMBITS) / active_mbs);
int q;
// TODO(jimbankoski): remove #if here or above when this has been
// well tested.
#if !CONFIG_ALWAYS_ADJUST_BPM
// based on recent history adjust expectations of bits per macroblock.
last_group_rate_err =
(double)twopass->rolling_arf_group_actual_bits /
DOUBLE_DIVIDE_CHECK((double)twopass->rolling_arf_group_target_bits);
last_group_rate_err = VPXMAX(0.25, VPXMIN(4.0, last_group_rate_err));
twopass->bpm_factor *= (3.0 + last_group_rate_err) / 4.0;
twopass->bpm_factor = VPXMAX(0.25, VPXMIN(4.0, twopass->bpm_factor));
#endif
// Try and pick a max Q that will be high enough to encode the
// content at the given rate.
for (q = rc->best_quality; q < rc->worst_quality; ++q) {
const double factor =
calc_correction_factor(av_err_per_mb, wq_err_divisor(cpi), q);
const int bits_per_mb = vp9_rc_bits_per_mb(
INTER_FRAME, q,
factor * speed_term * cpi->twopass.bpm_factor * noise_factor,
cpi->common.bit_depth);
if (bits_per_mb <= target_norm_bits_per_mb) break;
}
// Restriction on active max q for constrained quality mode.
if (cpi->oxcf.rc_mode == VPX_CQ) q = VPXMAX(q, oxcf->cq_level);
return q;
}
}
static void setup_rf_level_maxq(VP9_COMP *cpi) {
int i;
RATE_CONTROL *const rc = &cpi->rc;
for (i = INTER_NORMAL; i < RATE_FACTOR_LEVELS; ++i) {
int qdelta = vp9_frame_type_qdelta(cpi, i, rc->worst_quality);
rc->rf_level_maxq[i] = VPXMAX(rc->worst_quality + qdelta, rc->best_quality);
}
}
static void init_subsampling(VP9_COMP *cpi) {
const VP9_COMMON *const cm = &cpi->common;
RATE_CONTROL *const rc = &cpi->rc;
const int w = cm->width;
const int h = cm->height;
int i;
for (i = 0; i < FRAME_SCALE_STEPS; ++i) {
// Note: Frames with odd-sized dimensions may result from this scaling.
rc->frame_width[i] = (w * 16) / frame_scale_factor[i];
rc->frame_height[i] = (h * 16) / frame_scale_factor[i];
}
setup_rf_level_maxq(cpi);
}
void calculate_coded_size(VP9_COMP *cpi, int *scaled_frame_width,
int *scaled_frame_height) {
RATE_CONTROL *const rc = &cpi->rc;
*scaled_frame_width = rc->frame_width[rc->frame_size_selector];
*scaled_frame_height = rc->frame_height[rc->frame_size_selector];
}
void vp9_init_second_pass(VP9_COMP *cpi) {
VP9EncoderConfig *const oxcf = &cpi->oxcf;
RATE_CONTROL *const rc = &cpi->rc;
TWO_PASS *const twopass = &cpi->twopass;
double frame_rate;
FIRSTPASS_STATS *stats;
zero_stats(&twopass->total_stats);
zero_stats(&twopass->total_left_stats);
if (!twopass->stats_in_end) return;
stats = &twopass->total_stats;
*stats = *twopass->stats_in_end;
twopass->total_left_stats = *stats;
// Scan the first pass file and calculate a modified score for each
// frame that is used to distribute bits. The modified score is assumed
// to provide a linear basis for bit allocation. I.e a frame A with a score
// that is double that of frame B will be allocated 2x as many bits.
{
double modified_score_total = 0.0;
const FIRSTPASS_STATS *s = twopass->stats_in;
double av_err;
if (oxcf->vbr_corpus_complexity) {
twopass->mean_mod_score = (double)oxcf->vbr_corpus_complexity / 10.0;
av_err = get_distribution_av_err(cpi, twopass);
} else {
av_err = get_distribution_av_err(cpi, twopass);
// The first scan is unclamped and gives a raw average.
while (s < twopass->stats_in_end) {
modified_score_total += calculate_mod_frame_score(cpi, oxcf, s, av_err);
++s;
}
// The average error from this first scan is used to define the midpoint
// error for the rate distribution function.
twopass->mean_mod_score =
modified_score_total / DOUBLE_DIVIDE_CHECK(stats->count);
}
// Second scan using clamps based on the previous cycle average.
// This may modify the total and average somewhat but we dont bother with
// further itterations.
modified_score_total = 0.0;
s = twopass->stats_in;
while (s < twopass->stats_in_end) {
modified_score_total +=
calculate_norm_frame_score(cpi, twopass, oxcf, s, av_err);
++s;
}
twopass->normalized_score_left = modified_score_total;
// If using Corpus wide VBR mode then update the clip target bandwidth to
// reflect how the clip compares to the rest of the corpus.
if (oxcf->vbr_corpus_complexity) {
oxcf->target_bandwidth =
(int64_t)((double)oxcf->target_bandwidth *
(twopass->normalized_score_left / stats->count));
}
#if COMPLEXITY_STATS_OUTPUT
{
FILE *compstats;
compstats = fopen("complexity_stats.stt", "a");
fprintf(compstats, "%10.3lf\n",
twopass->normalized_score_left / stats->count);
fclose(compstats);
}
#endif
}
frame_rate = 10000000.0 * stats->count / stats->duration;
// Each frame can have a different duration, as the frame rate in the source
// isn't guaranteed to be constant. The frame rate prior to the first frame
// encoded in the second pass is a guess. However, the sum duration is not.
// It is calculated based on the actual durations of all frames from the
// first pass.
vp9_new_framerate(cpi, frame_rate);
twopass->bits_left =
(int64_t)(stats->duration * oxcf->target_bandwidth / 10000000.0);
// This variable monitors how far behind the second ref update is lagging.
twopass->sr_update_lag = 1;
// Reset the vbr bits off target counters
rc->vbr_bits_off_target = 0;
rc->vbr_bits_off_target_fast = 0;
rc->rate_error_estimate = 0;
// Static sequence monitor variables.
twopass->kf_zeromotion_pct = 100;
twopass->last_kfgroup_zeromotion_pct = 100;
// Initialize bits per macro_block estimate correction factor.
twopass->bpm_factor = 1.0;
// Initialize actual and target bits counters for ARF groups so that
// at the start we have a neutral bpm adjustment.
twopass->rolling_arf_group_target_bits = 1;
twopass->rolling_arf_group_actual_bits = 1;
if (oxcf->resize_mode != RESIZE_NONE) {
init_subsampling(cpi);
}
// Initialize the arnr strangth adjustment to 0
twopass->arnr_strength_adjustment = 0;
}
#define SR_DIFF_PART 0.0015
#define INTRA_PART 0.005
#define DEFAULT_DECAY_LIMIT 0.75
#define LOW_SR_DIFF_TRHESH 0.1
#define SR_DIFF_MAX 128.0
#define LOW_CODED_ERR_PER_MB 10.0
#define NCOUNT_FRAME_II_THRESH 6.0
static double get_sr_decay_rate(const FRAME_INFO *frame_info,
const FIRSTPASS_STATS *frame) {
double sr_diff = (frame->sr_coded_error - frame->coded_error);
double sr_decay = 1.0;
double modified_pct_inter;
double modified_pcnt_intra;
const double motion_amplitude_part =
frame->pcnt_motion *
((frame->mvc_abs + frame->mvr_abs) /
(frame_info->frame_height + frame_info->frame_width));
modified_pct_inter = frame->pcnt_inter;
if ((frame->coded_error > LOW_CODED_ERR_PER_MB) &&
((frame->intra_error / DOUBLE_DIVIDE_CHECK(frame->coded_error)) <
(double)NCOUNT_FRAME_II_THRESH)) {
modified_pct_inter =
frame->pcnt_inter + frame->pcnt_intra_low - frame->pcnt_neutral;
}
modified_pcnt_intra = 100 * (1.0 - modified_pct_inter);
if ((sr_diff > LOW_SR_DIFF_TRHESH)) {
sr_diff = VPXMIN(sr_diff, SR_DIFF_MAX);
sr_decay = 1.0 - (SR_DIFF_PART * sr_diff) - motion_amplitude_part -
(INTRA_PART * modified_pcnt_intra);
}
return VPXMAX(sr_decay, DEFAULT_DECAY_LIMIT);
}
// This function gives an estimate of how badly we believe the prediction
// quality is decaying from frame to frame.
static double get_zero_motion_factor(const FRAME_INFO *frame_info,
const FIRSTPASS_STATS *frame_stats) {
const double zero_motion_pct =
frame_stats->pcnt_inter - frame_stats->pcnt_motion;
double sr_decay = get_sr_decay_rate(frame_info, frame_stats);
return VPXMIN(sr_decay, zero_motion_pct);
}
#define ZM_POWER_FACTOR 0.75
static double get_prediction_decay_rate(const FRAME_INFO *frame_info,
const FIRSTPASS_STATS *frame_stats) {
const double sr_decay_rate = get_sr_decay_rate(frame_info, frame_stats);
const double zero_motion_factor =
(0.95 * pow((frame_stats->pcnt_inter - frame_stats->pcnt_motion),
ZM_POWER_FACTOR));
return VPXMAX(zero_motion_factor,
(sr_decay_rate + ((1.0 - sr_decay_rate) * zero_motion_factor)));
}
static int get_show_idx(const TWO_PASS *twopass) {
return (int)(twopass->stats_in - twopass->stats_in_start);
}
// Function to test for a condition where a complex transition is followed
// by a static section. For example in slide shows where there is a fade
// between slides. This is to help with more optimal kf and gf positioning.
static int check_transition_to_still(const FIRST_PASS_INFO *first_pass_info,
int show_idx, int still_interval) {
int j;
int num_frames = fps_get_num_frames(first_pass_info);
if (show_idx + still_interval > num_frames) {
return 0;
}
// Look ahead a few frames to see if static condition persists...
for (j = 0; j < still_interval; ++j) {
const FIRSTPASS_STATS *stats =
fps_get_frame_stats(first_pass_info, show_idx + j);
if (stats->pcnt_inter - stats->pcnt_motion < 0.999) break;
}
// Only if it does do we signal a transition to still.
return j == still_interval;
}
// This function detects a flash through the high relative pcnt_second_ref
// score in the frame following a flash frame. The offset passed in should
// reflect this.
static int detect_flash_from_frame_stats(const FIRSTPASS_STATS *frame_stats) {
// What we are looking for here is a situation where there is a
// brief break in prediction (such as a flash) but subsequent frames
// are reasonably well predicted by an earlier (pre flash) frame.
// The recovery after a flash is indicated by a high pcnt_second_ref
// useage or a second ref coded error notabley lower than the last
// frame coded error.
if (frame_stats == NULL) {
return 0;
}
return (frame_stats->sr_coded_error < frame_stats->coded_error) ||
((frame_stats->pcnt_second_ref > frame_stats->pcnt_inter) &&
(frame_stats->pcnt_second_ref >= 0.5));
}
static int detect_flash(const TWO_PASS *twopass, int offset) {
const FIRSTPASS_STATS *const next_frame = read_frame_stats(twopass, offset);
return detect_flash_from_frame_stats(next_frame);
}
// Update the motion related elements to the GF arf boost calculation.
static void accumulate_frame_motion_stats(const FIRSTPASS_STATS *stats,
double *mv_in_out,
double *mv_in_out_accumulator,
double *abs_mv_in_out_accumulator,
double *mv_ratio_accumulator) {
const double pct = stats->pcnt_motion;
// Accumulate Motion In/Out of frame stats.
*mv_in_out = stats->mv_in_out_count * pct;
*mv_in_out_accumulator += *mv_in_out;
*abs_mv_in_out_accumulator += fabs(*mv_in_out);
// Accumulate a measure of how uniform (or conversely how random) the motion
// field is (a ratio of abs(mv) / mv).
if (pct > 0.05) {
const double mvr_ratio =
fabs(stats->mvr_abs) / DOUBLE_DIVIDE_CHECK(fabs(stats->MVr));
const double mvc_ratio =
fabs(stats->mvc_abs) / DOUBLE_DIVIDE_CHECK(fabs(stats->MVc));
*mv_ratio_accumulator +=
pct * (mvr_ratio < stats->mvr_abs ? mvr_ratio : stats->mvr_abs);
*mv_ratio_accumulator +=
pct * (mvc_ratio < stats->mvc_abs ? mvc_ratio : stats->mvc_abs);
}
}
#define BASELINE_ERR_PER_MB 12500.0
#define GF_MAX_BOOST 96.0
static double calc_frame_boost(const FRAME_INFO *frame_info,
const FIRSTPASS_STATS *this_frame,
int avg_frame_qindex,
double this_frame_mv_in_out) {
double frame_boost;
const double lq =
vp9_convert_qindex_to_q(avg_frame_qindex, frame_info->bit_depth);
const double boost_q_correction = VPXMIN((0.5 + (lq * 0.015)), 1.5);
const double active_area = calculate_active_area(frame_info, this_frame);
// Underlying boost factor is based on inter error ratio.
frame_boost = (BASELINE_ERR_PER_MB * active_area) /
DOUBLE_DIVIDE_CHECK(this_frame->coded_error);
// Small adjustment for cases where there is a zoom out
if (this_frame_mv_in_out > 0.0)
frame_boost += frame_boost * (this_frame_mv_in_out * 2.0);
// Q correction and scalling
frame_boost = frame_boost * boost_q_correction;
return VPXMIN(frame_boost, GF_MAX_BOOST * boost_q_correction);
}
static double kf_err_per_mb(VP9_COMP *cpi) {
const VP9_COMMON *const cm = &cpi->common;
unsigned int screen_area = (cm->width * cm->height);
// Use a different error per mb factor for calculating boost for
// different formats.
if (screen_area < 1280 * 720) {
return 2000.0;
} else if (screen_area < 1920 * 1080) {
return 500.0;
}
return 250.0;
}
static double calc_kf_frame_boost(VP9_COMP *cpi,
const FIRSTPASS_STATS *this_frame,
double *sr_accumulator,
double this_frame_mv_in_out,
double max_boost) {
double frame_boost;
const double lq = vp9_convert_qindex_to_q(
cpi->rc.avg_frame_qindex[INTER_FRAME], cpi->common.bit_depth);
const double boost_q_correction = VPXMIN((0.50 + (lq * 0.015)), 2.00);
const double active_area =
calculate_active_area(&cpi->frame_info, this_frame);
// Underlying boost factor is based on inter error ratio.
frame_boost = (kf_err_per_mb(cpi) * active_area) /
DOUBLE_DIVIDE_CHECK(this_frame->coded_error + *sr_accumulator);
// Update the accumulator for second ref error difference.
// This is intended to give an indication of how much the coded error is
// increasing over time.
*sr_accumulator += (this_frame->sr_coded_error - this_frame->coded_error);
*sr_accumulator = VPXMAX(0.0, *sr_accumulator);
// Small adjustment for cases where there is a zoom out
if (this_frame_mv_in_out > 0.0)
frame_boost += frame_boost * (this_frame_mv_in_out * 2.0);
// Q correction and scaling
// The 40.0 value here is an experimentally derived baseline minimum.
// This value is in line with the minimum per frame boost in the alt_ref
// boost calculation.
frame_boost = ((frame_boost + 40.0) * boost_q_correction);
return VPXMIN(frame_boost, max_boost * boost_q_correction);
}
static int compute_arf_boost(const FRAME_INFO *frame_info,
const FIRST_PASS_INFO *first_pass_info,
int arf_show_idx, int f_frames, int b_frames,
int avg_frame_qindex) {
int i;
double boost_score = 0.0;
double mv_ratio_accumulator = 0.0;
double decay_accumulator = 1.0;
double this_frame_mv_in_out = 0.0;
double mv_in_out_accumulator = 0.0;
double abs_mv_in_out_accumulator = 0.0;
int arf_boost;
int flash_detected = 0;
// Search forward from the proposed arf/next gf position.
for (i = 0; i < f_frames; ++i) {
const FIRSTPASS_STATS *this_frame =
fps_get_frame_stats(first_pass_info, arf_show_idx + i);
const FIRSTPASS_STATS *next_frame =
fps_get_frame_stats(first_pass_info, arf_show_idx + i + 1);
if (this_frame == NULL) break;
// Update the motion related elements to the boost calculation.
accumulate_frame_motion_stats(
this_frame, &this_frame_mv_in_out, &mv_in_out_accumulator,
&abs_mv_in_out_accumulator, &mv_ratio_accumulator);
// We want to discount the flash frame itself and the recovery
// frame that follows as both will have poor scores.
flash_detected = detect_flash_from_frame_stats(this_frame) ||
detect_flash_from_frame_stats(next_frame);
// Accumulate the effect of prediction quality decay.
if (!flash_detected) {
decay_accumulator *= get_prediction_decay_rate(frame_info, this_frame);
decay_accumulator = decay_accumulator < MIN_DECAY_FACTOR
? MIN_DECAY_FACTOR
: decay_accumulator;
}
boost_score += decay_accumulator * calc_frame_boost(frame_info, this_frame,
avg_frame_qindex,
this_frame_mv_in_out);
}
arf_boost = (int)boost_score;
// Reset for backward looking loop.
boost_score = 0.0;
mv_ratio_accumulator = 0.0;
decay_accumulator = 1.0;
this_frame_mv_in_out = 0.0;
mv_in_out_accumulator = 0.0;
abs_mv_in_out_accumulator = 0.0;
// Search backward towards last gf position.
for (i = -1; i >= -b_frames; --i) {
const FIRSTPASS_STATS *this_frame =
fps_get_frame_stats(first_pass_info, arf_show_idx + i);
const FIRSTPASS_STATS *next_frame =
fps_get_frame_stats(first_pass_info, arf_show_idx + i + 1);
if (this_frame == NULL) break;
// Update the motion related elements to the boost calculation.
accumulate_frame_motion_stats(
this_frame, &this_frame_mv_in_out, &mv_in_out_accumulator,
&abs_mv_in_out_accumulator, &mv_ratio_accumulator);
// We want to discount the the flash frame itself and the recovery
// frame that follows as both will have poor scores.
flash_detected = detect_flash_from_frame_stats(this_frame) ||
detect_flash_from_frame_stats(next_frame);
// Cumulative effect of prediction quality decay.
if (!flash_detected) {
decay_accumulator *= get_prediction_decay_rate(frame_info, this_frame);
decay_accumulator = decay_accumulator < MIN_DECAY_FACTOR
? MIN_DECAY_FACTOR
: decay_accumulator;
}
boost_score += decay_accumulator * calc_frame_boost(frame_info, this_frame,
avg_frame_qindex,
this_frame_mv_in_out);
}
arf_boost += (int)boost_score;
if (arf_boost < ((b_frames + f_frames) * 40))
arf_boost = ((b_frames + f_frames) * 40);
arf_boost = VPXMAX(arf_boost, MIN_ARF_GF_BOOST);
return arf_boost;
}
static int calc_arf_boost(VP9_COMP *cpi, int f_frames, int b_frames) {
const FRAME_INFO *frame_info = &cpi->frame_info;
TWO_PASS *const twopass = &cpi->twopass;
const int avg_inter_frame_qindex = cpi->rc.avg_frame_qindex[INTER_FRAME];
int arf_show_idx = get_show_idx(twopass);
return compute_arf_boost(frame_info, &twopass->first_pass_info, arf_show_idx,
f_frames, b_frames, avg_inter_frame_qindex);
}
// Calculate a section intra ratio used in setting max loop filter.
static int calculate_section_intra_ratio(const FIRSTPASS_STATS *begin,
const FIRSTPASS_STATS *end,
int section_length) {
const FIRSTPASS_STATS *s = begin;
double intra_error = 0.0;
double coded_error = 0.0;
int i = 0;
while (s < end && i < section_length) {
intra_error += s->intra_error;
coded_error += s->coded_error;
++s;
++i;
}
return (int)(intra_error / DOUBLE_DIVIDE_CHECK(coded_error));
}
// Calculate the total bits to allocate in this GF/ARF group.
static int64_t calculate_total_gf_group_bits(VP9_COMP *cpi,
double gf_group_err) {
VP9_COMMON *const cm = &cpi->common;
const RATE_CONTROL *const rc = &cpi->rc;
const TWO_PASS *const twopass = &cpi->twopass;
const int max_bits = frame_max_bits(rc, &cpi->oxcf);
int64_t total_group_bits;
const int is_key_frame = frame_is_intra_only(cm);
const int arf_active_or_kf = is_key_frame || rc->source_alt_ref_active;
int gop_frames =
rc->baseline_gf_interval + rc->source_alt_ref_pending - arf_active_or_kf;
// Calculate the bits to be allocated to the group as a whole.
if ((twopass->kf_group_bits > 0) && (twopass->kf_group_error_left > 0.0)) {
int key_frame_interval = rc->frames_since_key + rc->frames_to_key;
int distance_from_next_key_frame =
rc->frames_to_key -
(rc->baseline_gf_interval + rc->source_alt_ref_pending);
int max_gf_bits_bias = rc->avg_frame_bandwidth;
double gf_interval_bias_bits_normalize_factor =
(double)rc->baseline_gf_interval / 16;
total_group_bits = (int64_t)(twopass->kf_group_bits *
(gf_group_err / twopass->kf_group_error_left));
// TODO(ravi): Experiment with different values of max_gf_bits_bias
total_group_bits +=
(int64_t)((double)distance_from_next_key_frame / key_frame_interval *
max_gf_bits_bias * gf_interval_bias_bits_normalize_factor);
} else {
total_group_bits = 0;
}
// Clamp odd edge cases.
total_group_bits = (total_group_bits < 0)
? 0
: (total_group_bits > twopass->kf_group_bits)
? twopass->kf_group_bits
: total_group_bits;
// Clip based on user supplied data rate variability limit.
if (total_group_bits > (int64_t)max_bits * gop_frames)
total_group_bits = (int64_t)max_bits * gop_frames;
return total_group_bits;
}
// Calculate the number bits extra to assign to boosted frames in a group.
static int calculate_boost_bits(int frame_count, int boost,
int64_t total_group_bits) {
int allocation_chunks;
// return 0 for invalid inputs (could arise e.g. through rounding errors)
if (!boost || (total_group_bits <= 0) || (frame_count < 0)) return 0;
allocation_chunks = (frame_count * NORMAL_BOOST) + boost;
// Prevent overflow.
if (boost > 1023) {
int divisor = boost >> 10;
boost /= divisor;
allocation_chunks /= divisor;
}
// Calculate the number of extra bits for use in the boosted frame or frames.
return VPXMAX((int)(((int64_t)boost * total_group_bits) / allocation_chunks),
0);
}
// Used in corpus vbr: Calculates the total normalized group complexity score
// for a given number of frames starting at the current position in the stats
// file.
static double calculate_group_score(VP9_COMP *cpi, double av_score,
int frame_count) {
VP9EncoderConfig *const oxcf = &cpi->oxcf;
TWO_PASS *const twopass = &cpi->twopass;
const FIRSTPASS_STATS *s = twopass->stats_in;
double score_total = 0.0;
int i = 0;
// We dont ever want to return a 0 score here.
if (frame_count == 0) return 1.0;
while ((i < frame_count) && (s < twopass->stats_in_end)) {
score_total += calculate_norm_frame_score(cpi, twopass, oxcf, s, av_score);
++s;
++i;
}
return score_total;
}
static void find_arf_order(VP9_COMP *cpi, GF_GROUP *gf_group,
int *index_counter, int depth, int start, int end) {
TWO_PASS *twopass = &cpi->twopass;
const FIRSTPASS_STATS *const start_pos = twopass->stats_in;
FIRSTPASS_STATS fpf_frame;
const int mid = (start + end + 1) >> 1;
const int min_frame_interval = 2;
int idx;
// Process regular P frames
if ((end - start < min_frame_interval) ||
(depth > gf_group->allowed_max_layer_depth)) {
for (idx = start; idx <= end; ++idx) {
gf_group->update_type[*index_counter] = LF_UPDATE;
gf_group->arf_src_offset[*index_counter] = 0;
gf_group->frame_gop_index[*index_counter] = idx;
gf_group->rf_level[*index_counter] = INTER_NORMAL;
gf_group->layer_depth[*index_counter] = depth;
gf_group->gfu_boost[*index_counter] = NORMAL_BOOST;
++(*index_counter);
}
gf_group->max_layer_depth = VPXMAX(gf_group->max_layer_depth, depth);
return;
}
assert(abs(mid - start) >= 1 && abs(mid - end) >= 1);
// Process ARF frame
gf_group->layer_depth[*index_counter] = depth;
gf_group->update_type[*index_counter] = ARF_UPDATE;
gf_group->arf_src_offset[*index_counter] = mid - start;
gf_group->frame_gop_index[*index_counter] = mid;
gf_group->rf_level[*index_counter] = GF_ARF_LOW;
for (idx = 0; idx <= mid; ++idx)
if (EOF == input_stats(twopass, &fpf_frame)) break;
gf_group->gfu_boost[*index_counter] =
VPXMAX(MIN_ARF_GF_BOOST,
calc_arf_boost(cpi, end - mid + 1, mid - start) >> depth);
reset_fpf_position(twopass, start_pos);
++(*index_counter);
find_arf_order(cpi, gf_group, index_counter, depth + 1, start, mid - 1);
gf_group->update_type[*index_counter] = USE_BUF_FRAME;
gf_group->arf_src_offset[*index_counter] = 0;
gf_group->frame_gop_index[*index_counter] = mid;
gf_group->rf_level[*index_counter] = INTER_NORMAL;
gf_group->layer_depth[*index_counter] = depth;
++(*index_counter);
find_arf_order(cpi, gf_group, index_counter, depth + 1, mid + 1, end);
}
static INLINE void set_gf_overlay_frame_type(GF_GROUP *gf_group,
int frame_index,
int source_alt_ref_active) {
if (source_alt_ref_active) {
gf_group->update_type[frame_index] = OVERLAY_UPDATE;
gf_group->rf_level[frame_index] = INTER_NORMAL;
gf_group->layer_depth[frame_index] = MAX_ARF_LAYERS - 1;
gf_group->gfu_boost[frame_index] = NORMAL_BOOST;
} else {
gf_group->update_type[frame_index] = GF_UPDATE;
gf_group->rf_level[frame_index] = GF_ARF_STD;
gf_group->layer_depth[frame_index] = 0;
}
}
static void define_gf_group_structure(VP9_COMP *cpi) {
RATE_CONTROL *const rc = &cpi->rc;
TWO_PASS *const twopass = &cpi->twopass;
GF_GROUP *const gf_group = &twopass->gf_group;
int frame_index = 0;
int key_frame = cpi->common.frame_type == KEY_FRAME;
int layer_depth = 1;
int gop_frames =
rc->baseline_gf_interval - (key_frame || rc->source_alt_ref_pending);
gf_group->frame_start = cpi->common.current_video_frame;
gf_group->frame_end = gf_group->frame_start + rc->baseline_gf_interval;
gf_group->max_layer_depth = 0;
gf_group->allowed_max_layer_depth = 0;
// For key frames the frame target rate is already set and it
// is also the golden frame.
// === [frame_index == 0] ===
if (!key_frame)
set_gf_overlay_frame_type(gf_group, frame_index, rc->source_alt_ref_active);
++frame_index;
// === [frame_index == 1] ===
if (rc->source_alt_ref_pending) {
gf_group->update_type[frame_index] = ARF_UPDATE;
gf_group->rf_level[frame_index] = GF_ARF_STD;
gf_group->layer_depth[frame_index] = layer_depth;
gf_group->arf_src_offset[frame_index] =
(unsigned char)(rc->baseline_gf_interval - 1);
gf_group->frame_gop_index[frame_index] = rc->baseline_gf_interval;
gf_group->max_layer_depth = 1;
++frame_index;
++layer_depth;
gf_group->allowed_max_layer_depth = cpi->oxcf.enable_auto_arf;
}
find_arf_order(cpi, gf_group, &frame_index, layer_depth, 1, gop_frames);
set_gf_overlay_frame_type(gf_group, frame_index, rc->source_alt_ref_pending);
gf_group->arf_src_offset[frame_index] = 0;
gf_group->frame_gop_index[frame_index] = rc->baseline_gf_interval;
// Set the frame ops number.
gf_group->gf_group_size = frame_index;
}
static void allocate_gf_group_bits(VP9_COMP *cpi, int64_t gf_group_bits,
int gf_arf_bits) {
VP9EncoderConfig *const oxcf = &cpi->oxcf;
RATE_CONTROL *const rc = &cpi->rc;
TWO_PASS *const twopass = &cpi->twopass;
GF_GROUP *const gf_group = &twopass->gf_group;
FIRSTPASS_STATS frame_stats;
int i;
int frame_index = 0;
int target_frame_size;
int key_frame;
const int max_bits = frame_max_bits(&cpi->rc, oxcf);
int64_t total_group_bits = gf_group_bits;
int mid_frame_idx;
int normal_frames;
int normal_frame_bits;
int last_frame_reduction = 0;
double av_score = 1.0;
double tot_norm_frame_score = 1.0;
double this_frame_score = 1.0;
// Define the GF structure and specify
int gop_frames = gf_group->gf_group_size;
key_frame = cpi->common.frame_type == KEY_FRAME;
// For key frames the frame target rate is already set and it
// is also the golden frame.
// === [frame_index == 0] ===
if (!key_frame) {
gf_group->bit_allocation[frame_index] =
rc->source_alt_ref_active ? 0 : gf_arf_bits;
}
// Deduct the boost bits for arf (or gf if it is not a key frame)
// from the group total.
if (rc->source_alt_ref_pending || !key_frame) total_group_bits -= gf_arf_bits;
++frame_index;
// === [frame_index == 1] ===
// Store the bits to spend on the ARF if there is one.
if (rc->source_alt_ref_pending) {
gf_group->bit_allocation[frame_index] = gf_arf_bits;
++frame_index;
}
// Define middle frame
mid_frame_idx = frame_index + (rc->baseline_gf_interval >> 1) - 1;
normal_frames = (rc->baseline_gf_interval - 1);
if (normal_frames > 1)
normal_frame_bits = (int)(total_group_bits / normal_frames);
else
normal_frame_bits = (int)total_group_bits;
gf_group->gfu_boost[1] = rc->gfu_boost;
if (cpi->multi_layer_arf) {
int idx;
int arf_depth_bits[MAX_ARF_LAYERS] = { 0 };
int arf_depth_count[MAX_ARF_LAYERS] = { 0 };
int arf_depth_boost[MAX_ARF_LAYERS] = { 0 };
int total_arfs = 1; // Account for the base layer ARF.
for (idx = 0; idx < gop_frames; ++idx) {
if (gf_group->update_type[idx] == ARF_UPDATE) {
arf_depth_boost[gf_group->layer_depth[idx]] += gf_group->gfu_boost[idx];
++arf_depth_count[gf_group->layer_depth[idx]];
}
}
for (idx = 2; idx < MAX_ARF_LAYERS; ++idx) {
if (arf_depth_boost[idx] == 0) break;
arf_depth_bits[idx] = calculate_boost_bits(
rc->baseline_gf_interval - total_arfs - arf_depth_count[idx],
arf_depth_boost[idx], total_group_bits);
total_group_bits -= arf_depth_bits[idx];
total_arfs += arf_depth_count[idx];
}
// offset the base layer arf
normal_frames -= (total_arfs - 1);
if (normal_frames > 1)
normal_frame_bits = (int)(total_group_bits / normal_frames);
else
normal_frame_bits = (int)total_group_bits;
target_frame_size = normal_frame_bits;
target_frame_size =
clamp(target_frame_size, 0, VPXMIN(max_bits, (int)total_group_bits));
// The first layer ARF has its bit allocation assigned.
for (idx = frame_index; idx < gop_frames; ++idx) {
switch (gf_group->update_type[idx]) {
case ARF_UPDATE:
gf_group->bit_allocation[idx] =
(int)(((int64_t)arf_depth_bits[gf_group->layer_depth[idx]] *
gf_group->gfu_boost[idx]) /
arf_depth_boost[gf_group->layer_depth[idx]]);
break;
case USE_BUF_FRAME: gf_group->bit_allocation[idx] = 0; break;
default: gf_group->bit_allocation[idx] = target_frame_size; break;
}
}
gf_group->bit_allocation[idx] = 0;
return;
}
if (oxcf->vbr_corpus_complexity) {
av_score = get_distribution_av_err(cpi, twopass);
tot_norm_frame_score = calculate_group_score(cpi, av_score, normal_frames);
}
// Allocate bits to the other frames in the group.
for (i = 0; i < normal_frames; ++i) {
if (EOF == input_stats(twopass, &frame_stats)) break;
if (oxcf->vbr_corpus_complexity) {
this_frame_score = calculate_norm_frame_score(cpi, twopass, oxcf,
&frame_stats, av_score);
normal_frame_bits = (int)((double)total_group_bits *
(this_frame_score / tot_norm_frame_score));
}
target_frame_size = normal_frame_bits;
if ((i == (normal_frames - 1)) && (i >= 1)) {
last_frame_reduction = normal_frame_bits / 16;
target_frame_size -= last_frame_reduction;
}
target_frame_size =
clamp(target_frame_size, 0, VPXMIN(max_bits, (int)total_group_bits));
gf_group->bit_allocation[frame_index] = target_frame_size;
++frame_index;
}
// Add in some extra bits for the middle frame in the group.
gf_group->bit_allocation[mid_frame_idx] += last_frame_reduction;
// Note:
// We need to configure the frame at the end of the sequence + 1 that will be
// the start frame for the next group. Otherwise prior to the call to
// vp9_rc_get_second_pass_params() the data will be undefined.
}
// Adjusts the ARNF filter for a GF group.
static void adjust_group_arnr_filter(VP9_COMP *cpi, double section_noise,
double section_inter,
double section_motion) {
TWO_PASS *const twopass = &cpi->twopass;
double section_zeromv = section_inter - section_motion;
twopass->arnr_strength_adjustment = 0;
if (section_noise < 150) {
twopass->arnr_strength_adjustment -= 1;
if (section_noise < 75) twopass->arnr_strength_adjustment -= 1;
} else if (section_noise > 250)
twopass->arnr_strength_adjustment += 1;
if (section_zeromv > 0.50) twopass->arnr_strength_adjustment += 1;
}
// Analyse and define a gf/arf group.
#define ARF_ABS_ZOOM_THRESH 4.0
#define MAX_GF_BOOST 5400
typedef struct RANGE {
int min;
int max;
} RANGE;
/* get_gop_coding_frame_num() depends on several fields in RATE_CONTROL *rc as
* follows.
* Static fields:
* (The following fields will remain unchanged after initialization of encoder.)
* rc->static_scene_max_gf_interval
* rc->min_gf_interval
*
* Dynamic fields:
* (The following fields will be updated before or after coding each frame.)
* rc->frames_to_key
* rc->frames_since_key
* rc->source_alt_ref_active
*
* Special case: if CONFIG_RATE_CTRL is true, the external arf indexes will
* determine the arf position.
*
* TODO(angiebird): Separate the dynamic fields and static fields into two
* structs.
*/
static int get_gop_coding_frame_num(
int *use_alt_ref, const FRAME_INFO *frame_info,
const FIRST_PASS_INFO *first_pass_info, const RATE_CONTROL *rc,
int gf_start_show_idx, const RANGE *active_gf_interval,
double gop_intra_factor, int lag_in_frames) {
double loop_decay_rate = 1.00;
double mv_ratio_accumulator = 0.0;
double this_frame_mv_in_out = 0.0;
double mv_in_out_accumulator = 0.0;
double abs_mv_in_out_accumulator = 0.0;
double sr_accumulator = 0.0;
// Motion breakout threshold for loop below depends on image size.
double mv_ratio_accumulator_thresh =
(frame_info->frame_height + frame_info->frame_width) / 4.0;
double zero_motion_accumulator = 1.0;
int gop_coding_frames;
*use_alt_ref = 1;
gop_coding_frames = 0;
while (gop_coding_frames < rc->static_scene_max_gf_interval &&
gop_coding_frames < rc->frames_to_key) {
const FIRSTPASS_STATS *next_next_frame;
const FIRSTPASS_STATS *next_frame;
int flash_detected;
++gop_coding_frames;
next_frame = fps_get_frame_stats(first_pass_info,
gf_start_show_idx + gop_coding_frames);
if (next_frame == NULL) {
break;
}
// Test for the case where there is a brief flash but the prediction
// quality back to an earlier frame is then restored.
next_next_frame = fps_get_frame_stats(
first_pass_info, gf_start_show_idx + gop_coding_frames + 1);
flash_detected = detect_flash_from_frame_stats(next_next_frame);
// Update the motion related elements to the boost calculation.
accumulate_frame_motion_stats(
next_frame, &this_frame_mv_in_out, &mv_in_out_accumulator,
&abs_mv_in_out_accumulator, &mv_ratio_accumulator);
// Monitor for static sections.
if ((rc->frames_since_key + gop_coding_frames - 1) > 1) {
zero_motion_accumulator =
VPXMIN(zero_motion_accumulator,
get_zero_motion_factor(frame_info, next_frame));
}
// Accumulate the effect of prediction quality decay.
if (!flash_detected) {
double last_loop_decay_rate = loop_decay_rate;
loop_decay_rate = get_prediction_decay_rate(frame_info, next_frame);
// Break clause to detect very still sections after motion. For example,
// a static image after a fade or other transition.
if (gop_coding_frames > rc->min_gf_interval && loop_decay_rate >= 0.999 &&
last_loop_decay_rate < 0.9) {
int still_interval = 5;
if (check_transition_to_still(first_pass_info,
gf_start_show_idx + gop_coding_frames,
still_interval)) {
*use_alt_ref = 0;
break;
}
}
// Update the accumulator for second ref error difference.
// This is intended to give an indication of how much the coded error is
// increasing over time.
if (gop_coding_frames == 1) {
sr_accumulator += next_frame->coded_error;
} else {
sr_accumulator +=
(next_frame->sr_coded_error - next_frame->coded_error);
}
}
// Break out conditions.
// Break at maximum of active_gf_interval->max unless almost totally
// static.
//
// Note that the addition of a test of rc->source_alt_ref_active is
// deliberate. The effect of this is that after a normal altref group even
// if the material is static there will be one normal length GF group
// before allowing longer GF groups. The reason for this is that in cases
// such as slide shows where slides are separated by a complex transition
// such as a fade, the arf group spanning the transition may not be coded
// at a very high quality and hence this frame (with its overlay) is a
// poor golden frame to use for an extended group.
if ((gop_coding_frames >= active_gf_interval->max) &&
((zero_motion_accumulator < 0.995) || (rc->source_alt_ref_active))) {
break;
}
if (
// Don't break out with a very short interval.
(gop_coding_frames >= active_gf_interval->min) &&
// If possible dont break very close to a kf
((rc->frames_to_key - gop_coding_frames) >= rc->min_gf_interval) &&
(gop_coding_frames & 0x01) && (!flash_detected) &&
((mv_ratio_accumulator > mv_ratio_accumulator_thresh) ||
(abs_mv_in_out_accumulator > ARF_ABS_ZOOM_THRESH) ||
(sr_accumulator > gop_intra_factor * next_frame->intra_error))) {
break;
}
}
*use_alt_ref &= zero_motion_accumulator < 0.995;
*use_alt_ref &= gop_coding_frames < lag_in_frames;
*use_alt_ref &= gop_coding_frames >= rc->min_gf_interval;
return gop_coding_frames;
}
static RANGE get_active_gf_inverval_range(
const FRAME_INFO *frame_info, const RATE_CONTROL *rc, int arf_active_or_kf,
int gf_start_show_idx, int active_worst_quality, int last_boosted_qindex) {
RANGE active_gf_interval;
#if CONFIG_RATE_CTRL
(void)frame_info;
(void)gf_start_show_idx;
(void)active_worst_quality;
(void)last_boosted_qindex;
active_gf_interval.min = rc->min_gf_interval + arf_active_or_kf + 2;
active_gf_interval.max = 16 + arf_active_or_kf;
if ((active_gf_interval.max <= rc->frames_to_key) &&
(active_gf_interval.max >= (rc->frames_to_key - rc->min_gf_interval))) {
active_gf_interval.min = rc->frames_to_key / 2;
active_gf_interval.max = rc->frames_to_key / 2;
}
#else
int int_max_q = (int)(vp9_convert_qindex_to_q(active_worst_quality,
frame_info->bit_depth));
int q_term = (gf_start_show_idx == 0)
? int_max_q / 32
: (int)(vp9_convert_qindex_to_q(last_boosted_qindex,
frame_info->bit_depth) /
6);
active_gf_interval.min =
rc->min_gf_interval + arf_active_or_kf + VPXMIN(2, int_max_q / 200);
active_gf_interval.min =
VPXMIN(active_gf_interval.min, rc->max_gf_interval + arf_active_or_kf);
// The value chosen depends on the active Q range. At low Q we have
// bits to spare and are better with a smaller interval and smaller boost.
// At high Q when there are few bits to spare we are better with a longer
// interval to spread the cost of the GF.
active_gf_interval.max = 11 + arf_active_or_kf + VPXMIN(5, q_term);
// Force max GF interval to be odd.
active_gf_interval.max = active_gf_interval.max | 0x01;
// We have: active_gf_interval.min <=
// rc->max_gf_interval + arf_active_or_kf.
if (active_gf_interval.max < active_gf_interval.min) {
active_gf_interval.max = active_gf_interval.min;
} else {
active_gf_interval.max =
VPXMIN(active_gf_interval.max, rc->max_gf_interval + arf_active_or_kf);
}
// Would the active max drop us out just before the near the next kf?
if ((active_gf_interval.max <= rc->frames_to_key) &&
(active_gf_interval.max