vpx/vp9/encoder/vp9_ratectrl.c

606 lines
21 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <limits.h>
#include <assert.h>
#include <math.h>
2010-05-18 17:58:33 +02:00
#include "vp9/common/vp9_alloccommon.h"
#include "vp9/common/vp9_modecont.h"
#include "vp9/common/vp9_common.h"
#include "vp9/encoder/vp9_ratectrl.h"
#include "vp9/common/vp9_entropymode.h"
2010-05-18 17:58:33 +02:00
#include "vpx_mem/vpx_mem.h"
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/encoder/vp9_encodemv.h"
#include "vp9/common/vp9_quant_common.h"
#include "vp9/common/vp9_seg_common.h"
2010-05-18 17:58:33 +02:00
#define MIN_BPB_FACTOR 0.005
#define MAX_BPB_FACTOR 50
2010-05-18 17:58:33 +02:00
#ifdef MODE_STATS
extern unsigned int y_modes[VP9_YMODES];
extern unsigned int uv_modes[VP9_UV_MODES];
extern unsigned int b_modes[B_MODE_COUNT];
2010-05-18 17:58:33 +02:00
extern unsigned int inter_y_modes[MB_MODE_COUNT];
extern unsigned int inter_uv_modes[VP9_UV_MODES];
extern unsigned int inter_b_modes[B_MODE_COUNT];
2010-05-18 17:58:33 +02:00
#endif
// Bits Per MB at different Q (Multiplied by 512)
#define BPER_MB_NORMBITS 9
// % adjustment to target kf size based on seperation from previous frame
static const int kf_boost_seperation_adjustment[16] = {
30, 40, 50, 55, 60, 65, 70, 75,
80, 85, 90, 95, 100, 100, 100, 100,
2010-05-18 17:58:33 +02:00
};
static const int gf_adjust_table[101] = {
100,
115, 130, 145, 160, 175, 190, 200, 210, 220, 230,
240, 260, 270, 280, 290, 300, 310, 320, 330, 340,
350, 360, 370, 380, 390, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
400, 400, 400, 400, 400, 400, 400, 400, 400, 400,
2010-05-18 17:58:33 +02:00
};
static const int gf_intra_usage_adjustment[20] = {
125, 120, 115, 110, 105, 100, 95, 85, 80, 75,
70, 65, 60, 55, 50, 50, 50, 50, 50, 50,
2010-05-18 17:58:33 +02:00
};
static const int gf_interval_table[101] = {
7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
2010-05-18 17:58:33 +02:00
};
static const unsigned int prior_key_frame_weight[KEY_FRAME_CONTEXT] = { 1, 2, 3, 4, 5 };
// These functions use formulaic calculations to make playing with the
// quantizer tables easier. If necessary they can be replaced by lookup
// tables if and when things settle down in the experimental bitstream
double vp9_convert_qindex_to_q(int qindex) {
// Convert the index to a real Q value (scaled down to match old Q values)
return vp9_ac_yquant(qindex) / 4.0;
}
int vp9_gfboost_qadjust(int qindex) {
const double q = vp9_convert_qindex_to_q(qindex);
return (int)((0.00000828 * q * q * q) +
(-0.0055 * q * q) +
(1.32 * q) + 79.3);
}
static int kfboost_qadjust(int qindex) {
const double q = vp9_convert_qindex_to_q(qindex);
return (int)((0.00000973 * q * q * q) +
(-0.00613 * q * q) +
(1.316 * q) + 121.2);
}
int vp9_bits_per_mb(FRAME_TYPE frame_type, int qindex,
double correction_factor) {
const double q = vp9_convert_qindex_to_q(qindex);
int enumerator = frame_type == KEY_FRAME ? 4000000 : 2500000;
// q based adjustment to baseline enumerator
enumerator += (int)(enumerator * q) >> 12;
return (int)(0.5 + (enumerator * correction_factor / q));
}
2010-05-18 17:58:33 +02:00
void vp9_save_coding_context(VP9_COMP *cpi) {
CODING_CONTEXT *const cc = &cpi->coding_context;
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
2010-05-18 17:58:33 +02:00
// Stores a snapshot of key state variables which can subsequently be
// restored with a call to vp9_restore_coding_context. These functions are
// intended for use in a re-code loop in vp9_compress_frame where the
// quantizer value is adjusted between loop iterations.
2010-05-18 17:58:33 +02:00
cc->nmvc = cm->fc.nmvc;
vp9_copy(cc->nmvjointcost, cpi->mb.nmvjointcost);
vp9_copy(cc->nmvcosts, cpi->mb.nmvcosts);
vp9_copy(cc->nmvcosts_hp, cpi->mb.nmvcosts_hp);
vp9_copy(cc->vp9_mode_contexts, cm->fc.vp9_mode_contexts);
vp9_copy(cc->ymode_prob, cm->fc.ymode_prob);
vp9_copy(cc->sb_ymode_prob, cm->fc.sb_ymode_prob);
vp9_copy(cc->bmode_prob, cm->fc.bmode_prob);
vp9_copy(cc->uv_mode_prob, cm->fc.uv_mode_prob);
vp9_copy(cc->i8x8_mode_prob, cm->fc.i8x8_mode_prob);
vp9_copy(cc->sub_mv_ref_prob, cm->fc.sub_mv_ref_prob);
vp9_copy(cc->mbsplit_prob, cm->fc.mbsplit_prob);
vp9_copy(cc->partition_prob, cm->fc.partition_prob);
// Stats
2010-05-18 17:58:33 +02:00
#ifdef MODE_STATS
vp9_copy(cc->y_modes, y_modes);
vp9_copy(cc->uv_modes, uv_modes);
vp9_copy(cc->b_modes, b_modes);
vp9_copy(cc->inter_y_modes, inter_y_modes);
vp9_copy(cc->inter_uv_modes, inter_uv_modes);
vp9_copy(cc->inter_b_modes, inter_b_modes);
2010-05-18 17:58:33 +02:00
#endif
vp9_copy(cc->segment_pred_probs, cm->segment_pred_probs);
vp9_copy(cc->ref_pred_probs_update, cpi->ref_pred_probs_update);
vp9_copy(cc->ref_pred_probs, cm->ref_pred_probs);
vp9_copy(cc->prob_comppred, cm->prob_comppred);
vpx_memcpy(cpi->coding_context.last_frame_seg_map_copy,
cm->last_frame_seg_map, (cm->mb_rows * cm->mb_cols));
2010-05-18 17:58:33 +02:00
vp9_copy(cc->last_ref_lf_deltas, xd->last_ref_lf_deltas);
vp9_copy(cc->last_mode_lf_deltas, xd->last_mode_lf_deltas);
vp9_copy(cc->coef_probs_4x4, cm->fc.coef_probs_4x4);
vp9_copy(cc->coef_probs_8x8, cm->fc.coef_probs_8x8);
vp9_copy(cc->coef_probs_16x16, cm->fc.coef_probs_16x16);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
vp9_copy(cc->coef_probs_32x32, cm->fc.coef_probs_32x32);
vp9_copy(cc->switchable_interp_prob, cm->fc.switchable_interp_prob);
#if CONFIG_COMP_INTERINTRA_PRED
cc->interintra_prob = cm->fc.interintra_prob;
#endif
#if CONFIG_CODE_NONZEROCOUNT
vp9_copy(cc->nzc_probs_4x4, cm->fc.nzc_probs_4x4);
vp9_copy(cc->nzc_probs_8x8, cm->fc.nzc_probs_8x8);
vp9_copy(cc->nzc_probs_16x16, cm->fc.nzc_probs_16x16);
vp9_copy(cc->nzc_probs_32x32, cm->fc.nzc_probs_32x32);
vp9_copy(cc->nzc_pcat_probs, cm->fc.nzc_pcat_probs);
#endif
}
2010-05-18 17:58:33 +02:00
void vp9_restore_coding_context(VP9_COMP *cpi) {
CODING_CONTEXT *const cc = &cpi->coding_context;
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
2010-05-18 17:58:33 +02:00
// Restore key state variables to the snapshot state stored in the
// previous call to vp9_save_coding_context.
2010-05-18 17:58:33 +02:00
cm->fc.nmvc = cc->nmvc;
vp9_copy(cpi->mb.nmvjointcost, cc->nmvjointcost);
vp9_copy(cpi->mb.nmvcosts, cc->nmvcosts);
vp9_copy(cpi->mb.nmvcosts_hp, cc->nmvcosts_hp);
vp9_copy(cm->fc.vp9_mode_contexts, cc->vp9_mode_contexts);
vp9_copy(cm->fc.ymode_prob, cc->ymode_prob);
vp9_copy(cm->fc.sb_ymode_prob, cc->sb_ymode_prob);
vp9_copy(cm->fc.bmode_prob, cc->bmode_prob);
vp9_copy(cm->fc.i8x8_mode_prob, cc->i8x8_mode_prob);
vp9_copy(cm->fc.uv_mode_prob, cc->uv_mode_prob);
vp9_copy(cm->fc.sub_mv_ref_prob, cc->sub_mv_ref_prob);
vp9_copy(cm->fc.mbsplit_prob, cc->mbsplit_prob);
vp9_copy(cm->fc.partition_prob, cc->partition_prob);
// Stats
2010-05-18 17:58:33 +02:00
#ifdef MODE_STATS
vp9_copy(y_modes, cc->y_modes);
vp9_copy(uv_modes, cc->uv_modes);
vp9_copy(b_modes, cc->b_modes);
vp9_copy(inter_y_modes, cc->inter_y_modes);
vp9_copy(inter_uv_modes, cc->inter_uv_modes);
vp9_copy(inter_b_modes, cc->inter_b_modes);
2010-05-18 17:58:33 +02:00
#endif
vp9_copy(cm->segment_pred_probs, cc->segment_pred_probs);
vp9_copy(cpi->ref_pred_probs_update, cc->ref_pred_probs_update);
vp9_copy(cm->ref_pred_probs, cc->ref_pred_probs);
vp9_copy(cm->prob_comppred, cc->prob_comppred);
vpx_memcpy(cm->last_frame_seg_map,
cpi->coding_context.last_frame_seg_map_copy,
(cm->mb_rows * cm->mb_cols));
2010-05-18 17:58:33 +02:00
vp9_copy(xd->last_ref_lf_deltas, cc->last_ref_lf_deltas);
vp9_copy(xd->last_mode_lf_deltas, cc->last_mode_lf_deltas);
vp9_copy(cm->fc.coef_probs_4x4, cc->coef_probs_4x4);
vp9_copy(cm->fc.coef_probs_8x8, cc->coef_probs_8x8);
vp9_copy(cm->fc.coef_probs_16x16, cc->coef_probs_16x16);
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
vp9_copy(cm->fc.coef_probs_32x32, cc->coef_probs_32x32);
vp9_copy(cm->fc.switchable_interp_prob, cc->switchable_interp_prob);
#if CONFIG_COMP_INTERINTRA_PRED
cm->fc.interintra_prob = cc->interintra_prob;
#endif
#if CONFIG_CODE_NONZEROCOUNT
vp9_copy(cm->fc.nzc_probs_4x4, cc->nzc_probs_4x4);
vp9_copy(cm->fc.nzc_probs_8x8, cc->nzc_probs_8x8);
vp9_copy(cm->fc.nzc_probs_16x16, cc->nzc_probs_16x16);
vp9_copy(cm->fc.nzc_probs_32x32, cc->nzc_probs_32x32);
vp9_copy(cm->fc.nzc_pcat_probs, cc->nzc_pcat_probs);
#endif
2010-05-18 17:58:33 +02:00
}
void vp9_setup_key_frame(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
vp9_setup_past_independence(cm, xd);
2010-05-18 17:58:33 +02:00
// interval before next GF
cpi->frames_till_gf_update_due = cpi->baseline_gf_interval;
/* All buffers are implicitly updated on key frames. */
cpi->refresh_golden_frame = 1;
cpi->refresh_alt_ref_frame = 1;
2010-05-18 17:58:33 +02:00
}
void vp9_setup_inter_frame(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &cpi->mb.e_mbd;
if (cm->error_resilient_mode)
vp9_setup_past_independence(cm, xd);
assert(cm->frame_context_idx < NUM_FRAME_CONTEXTS);
vpx_memcpy(&cm->fc, &cm->frame_contexts[cm->frame_context_idx],
sizeof(cm->fc));
}
2010-05-18 17:58:33 +02:00
static int estimate_bits_at_q(int frame_kind, int q, int mbs,
double correction_factor) {
const int bpm = (int)(vp9_bits_per_mb(frame_kind, q, correction_factor));
// Attempt to retain reasonable accuracy without overflow. The cutoff is
// chosen such that the maximum product of Bpm and MBs fits 31 bits. The
// largest Bpm takes 20 bits.
return (mbs > (1 << 11)) ? (bpm >> BPER_MB_NORMBITS) * mbs
: (bpm * mbs) >> BPER_MB_NORMBITS;
}
static void calc_iframe_target_size(VP9_COMP *cpi) {
// boost defaults to half second
int target;
2010-05-18 17:58:33 +02:00
// Clear down mmx registers to allow floating point in what follows
vp9_clear_system_state(); // __asm emms;
2010-05-18 17:58:33 +02:00
// New Two pass RC
target = cpi->per_frame_bandwidth;
2010-05-18 17:58:33 +02:00
if (cpi->oxcf.rc_max_intra_bitrate_pct) {
int max_rate = cpi->per_frame_bandwidth
* cpi->oxcf.rc_max_intra_bitrate_pct / 100;
2010-05-18 17:58:33 +02:00
if (target > max_rate)
target = max_rate;
}
2010-05-18 17:58:33 +02:00
cpi->this_frame_target = target;
2010-05-18 17:58:33 +02:00
}
// Do the best we can to define the parameters for the next GF based
// on what information we have available.
//
// In this experimental code only two pass is supported
// so we just use the interval determined in the two pass code.
static void calc_gf_params(VP9_COMP *cpi) {
// Set the gf interval
cpi->frames_till_gf_update_due = cpi->baseline_gf_interval;
2010-05-18 17:58:33 +02:00
}
static void calc_pframe_target_size(VP9_COMP *cpi) {
const int min_frame_target = MAX(cpi->min_frame_bandwidth,
cpi->av_per_frame_bandwidth >> 5);
if (cpi->refresh_alt_ref_frame) {
// Special alt reference frame case
// Per frame bit target for the alt ref frame
cpi->per_frame_bandwidth = cpi->twopass.gf_bits;
cpi->this_frame_target = cpi->per_frame_bandwidth;
} else {
// Normal frames (gf,and inter)
cpi->this_frame_target = cpi->per_frame_bandwidth;
}
2010-05-18 17:58:33 +02:00
// Sanity check that the total sum of adjustments is not above the maximum allowed
// That is that having allowed for KF and GF penalties we have not pushed the
// current interframe target to low. If the adjustment we apply here is not capable of recovering
// all the extra bits we have spent in the KF or GF then the remainder will have to be recovered over
// a longer time span via other buffer / rate control mechanisms.
if (cpi->this_frame_target < min_frame_target)
cpi->this_frame_target = min_frame_target;
2010-05-18 17:58:33 +02:00
if (!cpi->refresh_alt_ref_frame)
// Note the baseline target data rate for this inter frame.
cpi->inter_frame_target = cpi->this_frame_target;
// Adjust target frame size for Golden Frames:
if (cpi->frames_till_gf_update_due == 0) {
const int q = (cpi->oxcf.fixed_q < 0) ? cpi->last_q[INTER_FRAME]
: cpi->oxcf.fixed_q;
cpi->refresh_golden_frame = 1;
calc_gf_params(cpi);
// If we are using alternate ref instead of gf then do not apply the boost
// It will instead be applied to the altref update
// Jims modified boost
if (!cpi->source_alt_ref_active) {
if (cpi->oxcf.fixed_q < 0) {
// The spend on the GF is defined in the two pass code
// for two pass encodes
cpi->this_frame_target = cpi->per_frame_bandwidth;
} else {
cpi->this_frame_target =
(estimate_bits_at_q(1, q, cpi->common.MBs, 1.0)
* cpi->last_boost) / 100;
}
} else {
// If there is an active ARF at this location use the minimum
// bits on this frame even if it is a constructed arf.
// The active maximum quantizer insures that an appropriate
// number of bits will be spent if needed for constructed ARFs.
cpi->this_frame_target = 0;
2010-05-18 17:58:33 +02:00
}
}
2010-05-18 17:58:33 +02:00
}
void vp9_update_rate_correction_factors(VP9_COMP *cpi, int damp_var) {
const int q = cpi->common.base_qindex;
int correction_factor = 100;
double rate_correction_factor;
double adjustment_limit;
2010-05-18 17:58:33 +02:00
int projected_size_based_on_q = 0;
2010-05-18 17:58:33 +02:00
// Clear down mmx registers to allow floating point in what follows
vp9_clear_system_state(); // __asm emms;
2010-05-18 17:58:33 +02:00
if (cpi->common.frame_type == KEY_FRAME) {
rate_correction_factor = cpi->key_frame_rate_correction_factor;
} else {
if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
rate_correction_factor = cpi->gf_rate_correction_factor;
2010-05-18 17:58:33 +02:00
else
rate_correction_factor = cpi->rate_correction_factor;
}
// Work out how big we would have expected the frame to be at this Q given
// the current correction factor.
// Stay in double to avoid int overflow when values are large
projected_size_based_on_q = estimate_bits_at_q(cpi->common.frame_type, q,
cpi->common.MBs,
rate_correction_factor);
// Work out a size correction factor.
// if ( cpi->this_frame_target > 0 )
// correction_factor = (100 * cpi->projected_frame_size) / cpi->this_frame_target;
if (projected_size_based_on_q > 0)
correction_factor = (100 * cpi->projected_frame_size) / projected_size_based_on_q;
2010-05-18 17:58:33 +02:00
// More heavily damped adjustment used if we have been oscillating either side of target
switch (damp_var) {
2010-05-18 17:58:33 +02:00
case 0:
adjustment_limit = 0.75;
break;
2010-05-18 17:58:33 +02:00
case 1:
adjustment_limit = 0.375;
break;
2010-05-18 17:58:33 +02:00
case 2:
default:
adjustment_limit = 0.25;
break;
}
// if ( (correction_factor > 102) && (Q < cpi->active_worst_quality) )
if (correction_factor > 102) {
// We are not already at the worst allowable quality
correction_factor = (int)(100.5 + ((correction_factor - 100) * adjustment_limit));
rate_correction_factor = ((rate_correction_factor * correction_factor) / 100);
// Keep rate_correction_factor within limits
if (rate_correction_factor > MAX_BPB_FACTOR)
rate_correction_factor = MAX_BPB_FACTOR;
}
// else if ( (correction_factor < 99) && (Q > cpi->active_best_quality) )
else if (correction_factor < 99) {
// We are not already at the best allowable quality
correction_factor = (int)(100.5 - ((100 - correction_factor) * adjustment_limit));
rate_correction_factor = ((rate_correction_factor * correction_factor) / 100);
// Keep rate_correction_factor within limits
if (rate_correction_factor < MIN_BPB_FACTOR)
rate_correction_factor = MIN_BPB_FACTOR;
}
if (cpi->common.frame_type == KEY_FRAME)
cpi->key_frame_rate_correction_factor = rate_correction_factor;
else {
if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
cpi->gf_rate_correction_factor = rate_correction_factor;
2010-05-18 17:58:33 +02:00
else
cpi->rate_correction_factor = rate_correction_factor;
}
2010-05-18 17:58:33 +02:00
}
int vp9_regulate_q(VP9_COMP *cpi, int target_bits_per_frame) {
int q = cpi->active_worst_quality;
2010-05-18 17:58:33 +02:00
int i;
int last_error = INT_MAX;
int target_bits_per_mb;
int bits_per_mb_at_this_q;
double correction_factor;
// Select the appropriate correction factor based upon type of frame.
if (cpi->common.frame_type == KEY_FRAME)
correction_factor = cpi->key_frame_rate_correction_factor;
else {
if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame)
correction_factor = cpi->gf_rate_correction_factor;
2010-05-18 17:58:33 +02:00
else
correction_factor = cpi->rate_correction_factor;
}
2010-05-18 17:58:33 +02:00
// Calculate required scaling factor based on target frame size and size of frame produced using previous Q
if (target_bits_per_frame >= (INT_MAX >> BPER_MB_NORMBITS))
target_bits_per_mb = (target_bits_per_frame / cpi->common.MBs) << BPER_MB_NORMBITS; // Case where we would overflow int
else
target_bits_per_mb = (target_bits_per_frame << BPER_MB_NORMBITS) / cpi->common.MBs;
2010-05-18 17:58:33 +02:00
i = cpi->active_best_quality;
2010-05-18 17:58:33 +02:00
do {
bits_per_mb_at_this_q = (int)vp9_bits_per_mb(cpi->common.frame_type, i,
correction_factor);
2010-05-18 17:58:33 +02:00
if (bits_per_mb_at_this_q <= target_bits_per_mb) {
if ((target_bits_per_mb - bits_per_mb_at_this_q) <= last_error)
q = i;
else
q = i - 1;
2010-05-18 17:58:33 +02:00
break;
} else {
last_error = bits_per_mb_at_this_q - target_bits_per_mb;
}
} while (++i <= cpi->active_worst_quality);
2010-05-18 17:58:33 +02:00
return q;
2010-05-18 17:58:33 +02:00
}
static int estimate_keyframe_frequency(VP9_COMP *cpi) {
int i;
2010-05-18 17:58:33 +02:00
// Average key frame frequency
int av_key_frame_frequency = 0;
2010-05-18 17:58:33 +02:00
/* First key frame at start of sequence is a special case. We have no
* frequency data.
*/
if (cpi->key_frame_count == 1) {
/* Assume a default of 1 kf every 2 seconds, or the max kf interval,
* whichever is smaller.
*/
int key_freq = cpi->oxcf.key_freq > 0 ? cpi->oxcf.key_freq : 1;
av_key_frame_frequency = (int)cpi->output_frame_rate * 2;
2010-05-18 17:58:33 +02:00
if (cpi->oxcf.auto_key && av_key_frame_frequency > key_freq)
av_key_frame_frequency = cpi->oxcf.key_freq;
2010-05-18 17:58:33 +02:00
cpi->prior_key_frame_distance[KEY_FRAME_CONTEXT - 1]
= av_key_frame_frequency;
} else {
unsigned int total_weight = 0;
int last_kf_interval =
(cpi->frames_since_key > 0) ? cpi->frames_since_key : 1;
/* reset keyframe context and calculate weighted average of last
* KEY_FRAME_CONTEXT keyframes
*/
for (i = 0; i < KEY_FRAME_CONTEXT; i++) {
if (i < KEY_FRAME_CONTEXT - 1)
cpi->prior_key_frame_distance[i]
= cpi->prior_key_frame_distance[i + 1];
else
cpi->prior_key_frame_distance[i] = last_kf_interval;
av_key_frame_frequency += prior_key_frame_weight[i]
* cpi->prior_key_frame_distance[i];
total_weight += prior_key_frame_weight[i];
2010-05-18 17:58:33 +02:00
}
av_key_frame_frequency /= total_weight;
}
return av_key_frame_frequency;
}
void vp9_adjust_key_frame_context(VP9_COMP *cpi) {
// Clear down mmx registers to allow floating point in what follows
vp9_clear_system_state();
2010-05-18 17:58:33 +02:00
cpi->frames_since_key = 0;
cpi->key_frame_count++;
2010-05-18 17:58:33 +02:00
}
void vp9_compute_frame_size_bounds(VP9_COMP *cpi, int *frame_under_shoot_limit,
int *frame_over_shoot_limit) {
// Set-up bounds on acceptable frame size:
if (cpi->oxcf.fixed_q >= 0) {
// Fixed Q scenario: frame size never outranges target (there is no target!)
*frame_under_shoot_limit = 0;
*frame_over_shoot_limit = INT_MAX;
} else {
if (cpi->common.frame_type == KEY_FRAME) {
*frame_over_shoot_limit = cpi->this_frame_target * 9 / 8;
*frame_under_shoot_limit = cpi->this_frame_target * 7 / 8;
} else {
if (cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame) {
*frame_over_shoot_limit = cpi->this_frame_target * 9 / 8;
*frame_under_shoot_limit = cpi->this_frame_target * 7 / 8;
} else {
// Stron overshoot limit for constrained quality
if (cpi->oxcf.end_usage == USAGE_CONSTRAINED_QUALITY) {
*frame_over_shoot_limit = cpi->this_frame_target * 11 / 8;
*frame_under_shoot_limit = cpi->this_frame_target * 2 / 8;
} else {
*frame_over_shoot_limit = cpi->this_frame_target * 11 / 8;
*frame_under_shoot_limit = cpi->this_frame_target * 5 / 8;
2010-05-18 17:58:33 +02:00
}
}
2010-05-18 17:58:33 +02:00
}
// For very small rate targets where the fractional adjustment
// (eg * 7/8) may be tiny make sure there is at least a minimum
// range.
*frame_over_shoot_limit += 200;
*frame_under_shoot_limit -= 200;
if (*frame_under_shoot_limit < 0)
*frame_under_shoot_limit = 0;
}
2010-05-18 17:58:33 +02:00
}
// return of 0 means drop frame
int vp9_pick_frame_size(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
if (cm->frame_type == KEY_FRAME)
calc_iframe_target_size(cpi);
else
calc_pframe_target_size(cpi);
return 1;
}