#define _ISOC99_SOURCE
#undef NDEBUG // always check asserts, the speed effect is far too small to disable them
#include <math.h>
-#include <limits.h>
-#include <assert.h>
#include "common/common.h"
#include "common/cpu.h"
#include "ratecontrol.h"
+#include "me.h"
typedef struct
{
int mv_bits;
int tex_bits;
int misc_bits;
- uint64_t expected_bits;
+ uint64_t expected_bits; /*total expected bits up to the current frame (current one excluded)*/
double expected_vbv;
float new_qscale;
int new_qp;
int s_count;
float blurred_complexity;
char direct_mode;
+ int refcount[16];
+ int refs;
} ratecontrol_entry_t;
typedef struct
double coeff;
double count;
double decay;
+ double offset;
} predictor_t;
struct x264_ratecontrol_t
double fps;
double bitrate;
double rate_tolerance;
+ double qcompress;
int nmb; /* number of macroblocks in a frame */
int qp_constant[5];
float f_qpm; /* qp for current macroblock: precise float for AQ */
float qpa_rc; /* average of macroblocks' qp before aq */
float qpa_aq; /* average of macroblocks' qp after aq */
+ float qp_novbv; /* QP for the current frame if 1-pass VBV was disabled. */
int qp_force;
/* VBV stuff */
double buffer_fill; /* planned buffer, if all in-progress frames hit their bit budget */
double buffer_rate; /* # of bits added to buffer_fill after each frame */
predictor_t *pred; /* predict frame size from satd */
+ int single_frame_vbv;
/* ABR stuff */
int last_satd;
double last_rceq;
double cplxr_sum; /* sum of bits*qscale/rceq */
- double expected_bits_sum; /* sum of qscale2bits after rceq, ratefactor, and overflow */
+ double expected_bits_sum; /* sum of qscale2bits after rceq, ratefactor, and overflow, only includes finished frames */
double wanted_bits_window; /* target bitrate * window */
double cbr_decay;
double short_term_cplxsum;
/* 2pass stuff */
FILE *p_stat_file_out;
char *psz_stat_file_tmpname;
+ FILE *p_mbtree_stat_file_out;
+ char *psz_mbtree_stat_file_tmpname;
+ char *psz_mbtree_stat_file_name;
+ FILE *p_mbtree_stat_file_in;
int num_entries; /* number of ratecontrol_entry_ts */
ratecontrol_entry_t *entry; /* FIXME: copy needed data and free this once init is done */
double lmin[5]; /* min qscale by frame type */
double lmax[5];
double lstep; /* max change (multiply) in qscale per frame */
+ uint16_t *qp_buffer; /* Global buffer for converting MB-tree quantizer data. */
/* MBRC stuff */
double frame_size_estimated;
double frame_size_planned;
- predictor_t *row_pred;
- predictor_t row_preds[5];
+ predictor_t (*row_pred)[2];
+ predictor_t row_preds[5][2];
predictor_t *pred_b_from_p; /* predict B-frame size from P-frame satd */
int bframes; /* # consecutive B-frames before this P-frame */
int bframe_bits; /* total cost of those frames */
static int init_pass2(x264_t *);
static float rate_estimate_qscale( x264_t *h );
static void update_vbv( x264_t *h, int bits );
-static void update_vbv_plan( x264_t *h );
+static void update_vbv_plan( x264_t *h, int overhead );
static double predict_size( predictor_t *p, double q, double var );
static void update_predictor( predictor_t *p, double q, double var, double bits );
}
// Find the total AC energy of the block in all planes.
-static NOINLINE int ac_energy_mb( x264_t *h, int mb_x, int mb_y, x264_frame_t *frame )
+static NOINLINE uint32_t ac_energy_mb( x264_t *h, int mb_x, int mb_y, x264_frame_t *frame )
{
/* This function contains annoying hacks because GCC has a habit of reordering emms
* and putting it after floating point ops. As a result, we put the emms at the end of the
* function and make sure that its always called before the float math. Noinline makes
* sure no reordering goes on. */
- unsigned int var=0, sad, i;
- for( i=0; i<3; i++ )
+ uint32_t var = 0, i;
+ for( i = 0; i < 3; i++ )
{
int w = i ? 8 : 16;
int stride = frame->i_stride[i];
: w * (mb_x + mb_y * stride);
int pix = i ? PIXEL_8x8 : PIXEL_16x16;
stride <<= h->mb.b_interlaced;
- var += h->pixf.var[pix]( frame->plane[i]+offset, stride, &sad );
+ var += h->pixf.var[pix]( frame->plane[i]+offset, stride );
}
- var = X264_MAX(var,1);
x264_emms();
return var;
}
void x264_adaptive_quant_frame( x264_t *h, x264_frame_t *frame )
{
+ /* constants chosen to result in approximately the same overall bitrate as without AQ.
+ * FIXME: while they're written in 5 significant digits, they're only tuned to 2. */
int mb_x, mb_y;
- for( mb_y=0; mb_y<h->sps->i_mb_height; mb_y++ )
- for( mb_x=0; mb_x<h->sps->i_mb_width; mb_x++ )
+ float strength;
+ float avg_adj = 0.f;
+ /* Need to init it anyways for MB tree. */
+ if( h->param.rc.f_aq_strength == 0 )
+ {
+ int mb_xy;
+ memset( frame->f_qp_offset, 0, h->mb.i_mb_count * sizeof(float) );
+ memset( frame->f_qp_offset_aq, 0, h->mb.i_mb_count * sizeof(float) );
+ if( h->frames.b_have_lowres )
+ for( mb_xy = 0; mb_xy < h->mb.i_mb_count; mb_xy++ )
+ frame->i_inv_qscale_factor[mb_xy] = 256;
+ return;
+ }
+
+ if( h->param.rc.i_aq_mode == X264_AQ_AUTOVARIANCE )
+ {
+ for( mb_y = 0; mb_y < h->sps->i_mb_height; mb_y++ )
+ for( mb_x = 0; mb_x < h->sps->i_mb_width; mb_x++ )
+ {
+ uint32_t energy = ac_energy_mb( h, mb_x, mb_y, frame );
+ float qp_adj = x264_log2( energy + 2 );
+ qp_adj *= qp_adj;
+ frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride] = qp_adj;
+ avg_adj += qp_adj;
+ }
+ avg_adj /= h->mb.i_mb_count;
+ strength = h->param.rc.f_aq_strength * avg_adj * (1.f / 6000.f);
+ }
+ else
+ strength = h->param.rc.f_aq_strength * 1.0397f;
+
+ for( mb_y = 0; mb_y < h->sps->i_mb_height; mb_y++ )
+ for( mb_x = 0; mb_x < h->sps->i_mb_width; mb_x++ )
{
- int energy = ac_energy_mb( h, mb_x, mb_y, frame );
- /* 10 constant chosen to result in approximately the same overall bitrate as without AQ. */
- float qp_adj = h->param.rc.f_aq_strength * 1.5 * (logf(energy) - 10.0);
- frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride] = qp_adj;
+ float qp_adj;
+ if( h->param.rc.i_aq_mode == X264_AQ_AUTOVARIANCE )
+ {
+ qp_adj = frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride];
+ qp_adj = strength * (qp_adj - avg_adj);
+ }
+ else
+ {
+ uint32_t energy = ac_energy_mb( h, mb_x, mb_y, frame );
+ qp_adj = strength * (x264_log2( X264_MAX(energy, 1) ) - 14.427f);
+ }
+ frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride] =
+ frame->f_qp_offset_aq[mb_x + mb_y*h->mb.i_mb_stride] = qp_adj;
+ if( h->frames.b_have_lowres )
+ frame->i_inv_qscale_factor[mb_x + mb_y*h->mb.i_mb_stride] = x264_exp2fix8(qp_adj);
}
}
+
/*****************************************************************************
* x264_adaptive_quant:
* adjust macroblock QP based on variance (AC energy) of the MB.
*****************************************************************************/
void x264_adaptive_quant( x264_t *h )
{
- float qp, qp_adj;
x264_emms();
- qp = h->rc->f_qpm;
- qp_adj = h->fenc->f_qp_offset[h->mb.i_mb_x + h->mb.i_mb_y*h->mb.i_mb_stride];
- h->mb.i_qp = x264_clip3( qp + qp_adj + .5, h->param.rc.i_qp_min, h->param.rc.i_qp_max );
- /* If the QP of this MB is within 1 of the previous MB, code the same QP as the previous MB,
- * to lower the bit cost of the qp_delta. */
- if( abs(h->mb.i_qp - h->mb.i_last_qp) == 1 )
- h->mb.i_qp = h->mb.i_last_qp;
- h->mb.i_chroma_qp = h->chroma_qp_table[h->mb.i_qp];
+ h->mb.i_qp = x264_clip3( h->rc->f_qpm + h->fenc->f_qp_offset[h->mb.i_mb_xy] + .5, h->param.rc.i_qp_min, h->param.rc.i_qp_max );
+}
+
+int x264_macroblock_tree_read( x264_t *h, x264_frame_t *frame )
+{
+ x264_ratecontrol_t *rc = h->rc;
+ uint8_t i_type_actual = rc->entry[frame->i_frame].pict_type;
+ int i;
+
+ if( i_type_actual != SLICE_TYPE_B )
+ {
+ uint8_t i_type;
+
+ if( !fread( &i_type, 1, 1, rc->p_mbtree_stat_file_in ) )
+ goto fail;
+
+ if( i_type != i_type_actual )
+ {
+ x264_log(h, X264_LOG_ERROR, "MB-tree frametype %d doesn't match actual frametype %d.\n", i_type,i_type_actual);
+ return -1;
+ }
+
+ if( fread( rc->qp_buffer, sizeof(uint16_t), h->mb.i_mb_count, rc->p_mbtree_stat_file_in ) != h->mb.i_mb_count )
+ goto fail;
+
+ for( i = 0; i < h->mb.i_mb_count; i++ )
+ {
+ frame->f_qp_offset[i] = ((float)(int16_t)endian_fix16( rc->qp_buffer[i] )) * (1/256.0);
+ if( h->frames.b_have_lowres )
+ frame->i_inv_qscale_factor[i] = x264_exp2fix8(frame->f_qp_offset[i]);
+ }
+ }
+ else
+ x264_adaptive_quant_frame( h, frame );
+ return 0;
+fail:
+ x264_log(h, X264_LOG_ERROR, "Incomplete MB-tree stats file.\n");
+ return -1;
+}
+
+int x264_reference_build_list_optimal( x264_t *h )
+{
+ ratecontrol_entry_t *rce = h->rc->rce;
+ x264_frame_t *frames[16];
+ int ref, i;
+
+ if( rce->refs != h->i_ref0 )
+ return -1;
+
+ memcpy( frames, h->fref0, sizeof(frames) );
+
+ /* For now don't reorder ref 0; it seems to lower quality
+ in most cases due to skips. */
+ for( ref = 1; ref < h->i_ref0; ref++ )
+ {
+ int max = -1;
+ int bestref = 1;
+ for( i = 1; i < h->i_ref0; i++ )
+ /* Favor lower POC as a tiebreaker. */
+ COPY2_IF_GT( max, rce->refcount[i], bestref, i );
+ rce->refcount[bestref] = -1;
+ h->fref0[ref] = frames[bestref];
+ }
+
+ return 0;
+}
+
+static char *x264_strcat_filename( char *input, char *suffix )
+{
+ char *output = x264_malloc( strlen( input ) + strlen( suffix ) + 1 );
+ if( !output )
+ return NULL;
+ strcpy( output, input );
+ strcat( output, suffix );
+ return output;
}
int x264_ratecontrol_new( x264_t *h )
{
x264_ratecontrol_t *rc;
- int i;
+ int i, j;
x264_emms();
- rc = h->rc = x264_malloc( h->param.i_threads * sizeof(x264_ratecontrol_t) );
- memset( rc, 0, h->param.i_threads * sizeof(x264_ratecontrol_t) );
+ CHECKED_MALLOCZERO( h->rc, h->param.i_threads * sizeof(x264_ratecontrol_t) );
+ rc = h->rc;
rc->b_abr = h->param.rc.i_rc_method != X264_RC_CQP && !h->param.rc.b_stat_read;
rc->b_2pass = h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.b_stat_read;
else
rc->fps = 25.0;
+ if( h->param.rc.b_mb_tree )
+ {
+ h->param.rc.f_pb_factor = 1;
+ rc->qcompress = 1;
+ }
+ else
+ rc->qcompress = h->param.rc.f_qcompress;
+
rc->bitrate = h->param.rc.i_bitrate * 1000.;
rc->rate_tolerance = h->param.rc.f_rate_tolerance;
rc->nmb = h->mb.i_mb_count;
if( h->param.rc.i_vbv_buffer_size )
{
if( h->param.rc.i_rc_method == X264_RC_CQP )
+ {
x264_log(h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n");
+ h->param.rc.i_vbv_max_bitrate = 0;
+ h->param.rc.i_vbv_buffer_size = 0;
+ }
else if( h->param.rc.i_vbv_max_bitrate == 0 )
{
x264_log( h, X264_LOG_DEBUG, "VBV maxrate unspecified, assuming CBR\n" );
else if( h->param.rc.i_vbv_max_bitrate > 0 &&
h->param.rc.i_vbv_buffer_size > 0 )
{
- if( h->param.rc.i_vbv_buffer_size < 3 * h->param.rc.i_vbv_max_bitrate / rc->fps )
+ if( h->param.rc.i_vbv_buffer_size < (int)(h->param.rc.i_vbv_max_bitrate / rc->fps) )
{
- h->param.rc.i_vbv_buffer_size = 3 * h->param.rc.i_vbv_max_bitrate / rc->fps;
- x264_log( h, X264_LOG_WARNING, "VBV buffer size too small, using %d kbit\n",
+ h->param.rc.i_vbv_buffer_size = h->param.rc.i_vbv_max_bitrate / rc->fps;
+ x264_log( h, X264_LOG_WARNING, "VBV buffer size cannot be smaller than one frame, using %d kbit\n",
h->param.rc.i_vbv_buffer_size );
}
if( h->param.rc.f_vbv_buffer_init > 1. )
h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init / h->param.rc.i_vbv_buffer_size, 0, 1 );
rc->buffer_rate = h->param.rc.i_vbv_max_bitrate * 1000. / rc->fps;
rc->buffer_size = h->param.rc.i_vbv_buffer_size * 1000.;
+ rc->single_frame_vbv = rc->buffer_rate * 1.1 > rc->buffer_size;
+ h->param.rc.f_vbv_buffer_init = X264_MAX( h->param.rc.f_vbv_buffer_init, rc->buffer_rate / rc->buffer_size );
rc->buffer_fill_final = rc->buffer_size * h->param.rc.f_vbv_buffer_init;
rc->cbr_decay = 1.0 - rc->buffer_rate / rc->buffer_size
* 0.5 * X264_MAX(0, 1.5 - rc->buffer_rate * rc->fps / rc->bitrate);
rc->accum_p_norm = .01;
rc->accum_p_qp = ABR_INIT_QP * rc->accum_p_norm;
/* estimated ratio that produces a reasonable QP for the first I-frame */
- rc->cplxr_sum = .01 * pow( 7.0e5, h->param.rc.f_qcompress ) * pow( h->mb.i_mb_count, 0.5 );
+ rc->cplxr_sum = .01 * pow( 7.0e5, rc->qcompress ) * pow( h->mb.i_mb_count, 0.5 );
rc->wanted_bits_window = 1.0 * rc->bitrate / rc->fps;
rc->last_non_b_pict_type = SLICE_TYPE_I;
}
if( h->param.rc.i_rc_method == X264_RC_CRF )
{
- /* arbitrary rescaling to make CRF somewhat similar to QP */
+ /* Arbitrary rescaling to make CRF somewhat similar to QP.
+ * Try to compensate for MB-tree's effects as well. */
double base_cplx = h->mb.i_mb_count * (h->param.i_bframe ? 120 : 80);
- rc->rate_factor_constant = pow( base_cplx, 1 - h->param.rc.f_qcompress )
- / qp2qscale( h->param.rc.f_rf_constant );
+ double mbtree_offset = h->param.rc.b_mb_tree ? (1.0-h->param.rc.f_qcompress)*13.5 : 0;
+ rc->rate_factor_constant = pow( base_cplx, 1 - rc->qcompress )
+ / qp2qscale( h->param.rc.f_rf_constant + mbtree_offset );
}
rc->ip_offset = 6.0 * log(h->param.rc.f_ip_factor) / log(2.0);
rc->lstep = pow( 2, h->param.rc.i_qp_step / 6.0 );
rc->last_qscale = qp2qscale(26);
- rc->pred = x264_malloc( 5*sizeof(predictor_t) );
- rc->pred_b_from_p = x264_malloc( sizeof(predictor_t) );
+ CHECKED_MALLOC( rc->pred, 5*sizeof(predictor_t) );
+ CHECKED_MALLOC( rc->pred_b_from_p, sizeof(predictor_t) );
for( i = 0; i < 5; i++ )
{
rc->last_qscale_for[i] = qp2qscale( ABR_INIT_QP );
rc->pred[i].coeff= 2.0;
rc->pred[i].count= 1.0;
rc->pred[i].decay= 0.5;
- rc->row_preds[i].coeff= .25;
- rc->row_preds[i].count= 1.0;
- rc->row_preds[i].decay= 0.5;
+ rc->pred[i].offset= 0.0;
+ for( j = 0; j < 2; j++ )
+ {
+ rc->row_preds[i][j].coeff= .25;
+ rc->row_preds[i][j].count= 1.0;
+ rc->row_preds[i][j].decay= 0.5;
+ rc->row_preds[i][j].offset= 0.0;
+ }
}
*rc->pred_b_from_p = rc->pred[0];
x264_log(h, X264_LOG_ERROR, "ratecontrol_init: can't open stats file\n");
return -1;
}
+ if( h->param.rc.b_mb_tree )
+ {
+ char *mbtree_stats_in = x264_strcat_filename( h->param.rc.psz_stat_in, ".mbtree" );
+ if( !mbtree_stats_in )
+ return -1;
+ rc->p_mbtree_stat_file_in = fopen( mbtree_stats_in, "rb" );
+ x264_free( mbtree_stats_in );
+ if( !rc->p_mbtree_stat_file_in )
+ {
+ x264_log(h, X264_LOG_ERROR, "ratecontrol_init: can't open mbtree stats file\n");
+ return -1;
+ }
+ }
/* check whether 1st pass options were compatible with current options */
if( !strncmp( stats_buf, "#options:", 9 ) )
if( strstr( opts, "qp=0" ) && h->param.rc.i_rc_method == X264_RC_ABR )
x264_log( h, X264_LOG_WARNING, "1st pass was lossless, bitrate prediction will be inaccurate\n" );
+ if( !strstr( opts, "direct=3" ) && h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO )
+ {
+ x264_log( h, X264_LOG_WARNING, "direct=auto not used on the first pass\n" );
+ h->mb.b_direct_auto_write = 1;
+ }
+
if( ( p = strstr( opts, "b_adapt=" ) ) && sscanf( p, "b_adapt=%d", &i ) && i >= X264_B_ADAPT_NONE && i <= X264_B_ADAPT_TRELLIS )
h->param.i_bframe_adaptive = i;
- else
+ else if( h->param.i_bframe )
{
x264_log( h, X264_LOG_ERROR, "b_adapt method specified in stats file not valid\n" );
return -1;
}
- if( ( p = strstr( opts, "scenecut=" ) ) && sscanf( p, "scenecut=%d", &i ) && i >= -1 && i <= 100 )
- {
- h->param.i_scenecut_threshold = i;
- h->param.b_pre_scenecut = !!strstr( p, "(pre)" );
- }
- else
- {
- x264_log( h, X264_LOG_ERROR, "scenecut method specified in stats file not valid\n" );
- return -1;
- }
+ if( h->param.rc.b_mb_tree && ( p = strstr( opts, "rc_lookahead=" ) ) && sscanf( p, "rc_lookahead=%d", &i ) )
+ h->param.rc.i_lookahead = i;
}
/* find number of pics */
x264_log( h, X264_LOG_WARNING, "2nd pass has fewer frames than 1st pass (%d vs %d)\n",
h->param.i_frame_total, rc->num_entries );
}
- if( h->param.i_frame_total > rc->num_entries + h->param.i_bframe )
+ if( h->param.i_frame_total > rc->num_entries )
{
x264_log( h, X264_LOG_ERROR, "2nd pass has more frames than 1st pass (%d vs %d)\n",
h->param.i_frame_total, rc->num_entries );
return -1;
}
- /* FIXME: ugly padding because VfW drops delayed B-frames */
- rc->num_entries += h->param.i_bframe;
-
- rc->entry = (ratecontrol_entry_t*) x264_malloc(rc->num_entries * sizeof(ratecontrol_entry_t));
- memset(rc->entry, 0, rc->num_entries * sizeof(ratecontrol_entry_t));
+ CHECKED_MALLOCZERO( rc->entry, rc->num_entries * sizeof(ratecontrol_entry_t) );
/* init all to skipped p frames */
for(i=0; i<rc->num_entries; i++)
/* read stats */
p = stats_in;
- for(i=0; i < rc->num_entries - h->param.i_bframe; i++)
+ for(i=0; i < rc->num_entries; i++)
{
ratecontrol_entry_t *rce;
int frame_number;
int e;
char *next;
float qp;
+ int ref;
next= strchr(p, ';');
if(next)
&rce->mv_bits, &rce->misc_bits, &rce->i_count, &rce->p_count,
&rce->s_count, &rce->direct_mode);
+ p = strstr( p, "ref:" );
+ if( !p )
+ goto parse_error;
+ p += 4;
+ for( ref = 0; ref < 16; ref++ )
+ {
+ if( sscanf( p, " %d", &rce->refcount[ref] ) != 1 )
+ break;
+ p = strchr( p+1, ' ' );
+ if( !p )
+ goto parse_error;
+ }
+ rce->refs = ref;
+
switch(pict_type)
{
case 'I': rce->kept_as_ref = 1;
}
if(e < 10)
{
+parse_error:
x264_log(h, X264_LOG_ERROR, "statistics are damaged at line %d, parser out=%d\n", i, e);
return -1;
}
if( h->param.rc.b_stat_write )
{
char *p;
-
- rc->psz_stat_file_tmpname = x264_malloc( strlen(h->param.rc.psz_stat_out) + 6 );
- strcpy( rc->psz_stat_file_tmpname, h->param.rc.psz_stat_out );
- strcat( rc->psz_stat_file_tmpname, ".temp" );
+ rc->psz_stat_file_tmpname = x264_strcat_filename( h->param.rc.psz_stat_out, ".temp" );
+ if( !rc->psz_stat_file_tmpname )
+ return -1;
rc->p_stat_file_out = fopen( rc->psz_stat_file_tmpname, "wb" );
if( rc->p_stat_file_out == NULL )
}
p = x264_param2string( &h->param, 1 );
- fprintf( rc->p_stat_file_out, "#options: %s\n", p );
+ if( p )
+ fprintf( rc->p_stat_file_out, "#options: %s\n", p );
x264_free( p );
+ if( h->param.rc.b_mb_tree && !h->param.rc.b_stat_read )
+ {
+ rc->psz_mbtree_stat_file_tmpname = x264_strcat_filename( h->param.rc.psz_stat_out, ".mbtree.temp" );
+ rc->psz_mbtree_stat_file_name = x264_strcat_filename( h->param.rc.psz_stat_out, ".mbtree" );
+ if( !rc->psz_mbtree_stat_file_tmpname || !rc->psz_mbtree_stat_file_name )
+ return -1;
+
+ rc->p_mbtree_stat_file_out = fopen( rc->psz_mbtree_stat_file_tmpname, "wb" );
+ if( rc->p_mbtree_stat_file_out == NULL )
+ {
+ x264_log(h, X264_LOG_ERROR, "ratecontrol_init: can't open mbtree stats file\n");
+ return -1;
+ }
+ }
}
+ if( h->param.rc.b_mb_tree && (h->param.rc.b_stat_read || h->param.rc.b_stat_write) )
+ CHECKED_MALLOC( rc->qp_buffer, h->mb.i_mb_count * sizeof(uint16_t) );
+
for( i=0; i<h->param.i_threads; i++ )
{
h->thread[i]->rc = rc+i;
if( i )
{
rc[i] = rc[0];
- memcpy( &h->thread[i]->param, &h->param, sizeof( x264_param_t ) );
+ memcpy( &h->thread[i]->param, &h->param, sizeof(x264_param_t) );
+ h->thread[i]->mb.b_variable_qp = h->mb.b_variable_qp;
}
}
return 0;
+fail:
+ return -1;
}
static int parse_zone( x264_t *h, x264_zone_t *z, char *p )
{
int len = 0;
- char *tok, *saveptr;
+ char *tok, UNUSED *saveptr;
z->param = NULL;
z->f_bitrate_factor = 1;
if( 3 <= sscanf(p, "%u,%u,q=%u%n", &z->i_start, &z->i_end, &z->i_qp, &len) )
p += len;
if( !*p )
return 0;
- z->param = malloc( sizeof(x264_param_t) );
+ CHECKED_MALLOC( z->param, sizeof(x264_param_t) );
memcpy( z->param, &h->param, sizeof(x264_param_t) );
+ z->param->param_free = x264_free;
while( (tok = strtok_r( p, ",", &saveptr )) )
{
char *val = strchr( tok, '=' );
p = NULL;
}
return 0;
+fail:
+ return -1;
}
static int parse_zones( x264_t *h )
int i;
if( h->param.rc.psz_zones && !h->param.rc.i_zones )
{
- char *p, *tok, *saveptr;
- char *psz_zones = x264_malloc( strlen(h->param.rc.psz_zones)+1 );
+ char *psz_zones, *p, *tok, UNUSED *saveptr;
+ CHECKED_MALLOC( psz_zones, strlen( h->param.rc.psz_zones )+1 );
strcpy( psz_zones, h->param.rc.psz_zones );
h->param.rc.i_zones = 1;
for( p = psz_zones; *p; p++ )
h->param.rc.i_zones += (*p == '/');
- h->param.rc.zones = x264_malloc( h->param.rc.i_zones * sizeof(x264_zone_t) );
+ CHECKED_MALLOC( h->param.rc.zones, h->param.rc.i_zones * sizeof(x264_zone_t) );
p = psz_zones;
for( i = 0; i < h->param.rc.i_zones; i++ )
{
}
rc->i_zones = h->param.rc.i_zones + 1;
- rc->zones = x264_malloc( rc->i_zones * sizeof(x264_zone_t) );
+ CHECKED_MALLOC( rc->zones, rc->i_zones * sizeof(x264_zone_t) );
memcpy( rc->zones+1, h->param.rc.zones, (rc->i_zones-1) * sizeof(x264_zone_t) );
// default zone to fall back to if none of the others match
rc->zones[0].i_end = INT_MAX;
rc->zones[0].b_force_qp = 0;
rc->zones[0].f_bitrate_factor = 1;
- rc->zones[0].param = x264_malloc( sizeof(x264_param_t) );
+ CHECKED_MALLOC( rc->zones[0].param, sizeof(x264_param_t) );
memcpy( rc->zones[0].param, &h->param, sizeof(x264_param_t) );
for( i = 1; i < rc->i_zones; i++ )
{
}
return 0;
+fail:
+ return -1;
}
static x264_zone_t *get_zone( x264_t *h, int frame_num )
if( rc->b_abr && h->param.rc.i_rc_method == X264_RC_ABR && rc->cbr_decay > .9999 )
{
double base_cplx = h->mb.i_mb_count * (h->param.i_bframe ? 120 : 80);
+ double mbtree_offset = h->param.rc.b_mb_tree ? (1.0-h->param.rc.f_qcompress)*13.5 : 0;
x264_log( h, X264_LOG_INFO, "final ratefactor: %.2f\n",
- qscale2qp( pow( base_cplx, 1 - h->param.rc.f_qcompress )
- * rc->cplxr_sum / rc->wanted_bits_window ) );
+ qscale2qp( pow( base_cplx, 1 - rc->qcompress )
+ * rc->cplxr_sum / rc->wanted_bits_window ) - mbtree_offset );
}
}
if( rc->p_stat_file_out )
{
fclose( rc->p_stat_file_out );
- if( h->i_frame >= rc->num_entries - h->param.i_bframe )
+ if( h->i_frame >= rc->num_entries )
if( rename( rc->psz_stat_file_tmpname, h->param.rc.psz_stat_out ) != 0 )
{
x264_log( h, X264_LOG_ERROR, "failed to rename \"%s\" to \"%s\"\n",
}
x264_free( rc->psz_stat_file_tmpname );
}
+ if( rc->p_mbtree_stat_file_out )
+ {
+ fclose( rc->p_mbtree_stat_file_out );
+ if( h->i_frame >= rc->num_entries )
+ if( rename( rc->psz_mbtree_stat_file_tmpname, rc->psz_mbtree_stat_file_name ) != 0 )
+ {
+ x264_log( h, X264_LOG_ERROR, "failed to rename \"%s\" to \"%s\"\n",
+ rc->psz_mbtree_stat_file_tmpname, rc->psz_mbtree_stat_file_name );
+ }
+ x264_free( rc->psz_mbtree_stat_file_tmpname );
+ x264_free( rc->psz_mbtree_stat_file_name );
+ }
+ if( rc->p_mbtree_stat_file_in )
+ fclose( rc->p_mbtree_stat_file_in );
x264_free( rc->pred );
x264_free( rc->pred_b_from_p );
x264_free( rc->entry );
+ x264_free( rc->qp_buffer );
if( rc->zones )
{
x264_free( rc->zones[0].param );
- if( h->param.rc.psz_zones )
- for( i=1; i<rc->i_zones; i++ )
- if( rc->zones[i].param != rc->zones[0].param )
- x264_free( rc->zones[i].param );
+ for( i=1; i<rc->i_zones; i++ )
+ if( rc->zones[i].param != rc->zones[0].param && rc->zones[i].param->param_free )
+ rc->zones[i].param->param_free( rc->zones[i].param );
x264_free( rc->zones );
}
x264_free( rc );
}
/* Before encoding a frame, choose a QP for it */
-void x264_ratecontrol_start( x264_t *h, int i_force_qp )
+void x264_ratecontrol_start( x264_t *h, int i_force_qp, int overhead )
{
x264_ratecontrol_t *rc = h->rc;
ratecontrol_entry_t *rce = NULL;
{
memset( h->fdec->i_row_bits, 0, h->sps->i_mb_height * sizeof(int) );
rc->row_pred = &rc->row_preds[h->sh.i_type];
- update_vbv_plan( h );
+ update_vbv_plan( h, overhead );
}
if( h->sh.i_type != SLICE_TYPE_B )
- {
- rc->bframes = 0;
- while( h->frames.current[rc->bframes] && IS_X264_TYPE_B(h->frames.current[rc->bframes]->i_type) )
- rc->bframes++;
- }
+ rc->bframes = h->fenc->i_bframes;
if( i_force_qp )
{
}
}
+ q = x264_clip3f( q, h->param.rc.i_qp_min, h->param.rc.i_qp_max );
+
rc->qpa_rc =
rc->qpa_aq = 0;
h->fdec->f_qp_avg_rc =
if( rce )
rce->new_qp = rc->qp;
- /* accum_p_qp needs to be here so that future frames can benefit from the
- * data before this frame is done. but this only works because threading
- * guarantees to not re-encode any frames. so the non-threaded case does
- * accum_p_qp later. */
- if( h->param.i_threads > 1 )
- accum_p_qp_update( h, rc->qp );
+ accum_p_qp_update( h, rc->qp );
if( h->sh.i_type != SLICE_TYPE_B )
rc->last_non_b_pict_type = h->sh.i_type;
/* average between two predictors:
* absolute SATD, and scaled bit cost of the colocated row in the previous frame */
x264_ratecontrol_t *rc = h->rc;
- double pred_s = predict_size( rc->row_pred, qp2qscale(qp), h->fdec->i_row_satd[y] );
+ double pred_s = predict_size( rc->row_pred[0], qp2qscale(qp), h->fdec->i_row_satd[y] );
double pred_t = 0;
- if( h->sh.i_type != SLICE_TYPE_I
- && h->fref0[0]->i_type == h->fdec->i_type
- && h->fref0[0]->i_row_satd[y] > 0
- && (abs(h->fref0[0]->i_row_satd[y] - h->fdec->i_row_satd[y]) < h->fdec->i_row_satd[y]/2))
+ if( h->sh.i_type == SLICE_TYPE_I || qp >= h->fref0[0]->i_row_qp[y] )
+ {
+ if( h->sh.i_type == SLICE_TYPE_P
+ && h->fref0[0]->i_type == h->fdec->i_type
+ && h->fref0[0]->i_row_satd[y] > 0
+ && (abs(h->fref0[0]->i_row_satd[y] - h->fdec->i_row_satd[y]) < h->fdec->i_row_satd[y]/2))
+ {
+ pred_t = h->fref0[0]->i_row_bits[y] * h->fdec->i_row_satd[y] / h->fref0[0]->i_row_satd[y]
+ * qp2qscale(h->fref0[0]->i_row_qp[y]) / qp2qscale(qp);
+ }
+ if( pred_t == 0 )
+ pred_t = pred_s;
+ return (pred_s + pred_t) / 2;
+ }
+ /* Our QP is lower than the reference! */
+ else
{
- pred_t = h->fref0[0]->i_row_bits[y] * h->fdec->i_row_satd[y] / h->fref0[0]->i_row_satd[y]
- * qp2qscale(h->fref0[0]->i_row_qp[y]) / qp2qscale(qp);
+ double newq = qp2qscale(qp);
+ double oldq = qp2qscale(h->fref0[0]->i_row_qp[y]);
+ double pred_intra = predict_size( rc->row_pred[1], (1 - newq / oldq) * newq, h->fdec->i_row_satds[0][0][y] );
+ /* Sum: better to overestimate than underestimate by using only one of the two predictors. */
+ return pred_intra + pred_s;
}
- if( pred_t == 0 )
- pred_t = pred_s;
-
- return (pred_s + pred_t) / 2;
}
static double row_bits_so_far( x264_t *h, int y )
rc->qpa_rc += rc->f_qpm;
rc->qpa_aq += h->mb.i_qp;
- if( h->mb.i_mb_x != h->sps->i_mb_width - 1 || !rc->b_vbv)
+ if( h->mb.i_mb_x != h->sps->i_mb_width - 1 || !rc->b_vbv )
return;
h->fdec->i_row_qp[y] = rc->qpm;
- if( h->sh.i_type == SLICE_TYPE_B )
+ update_predictor( rc->row_pred[0], qp2qscale(rc->qpm), h->fdec->i_row_satd[y], h->fdec->i_row_bits[y] );
+ if( h->sh.i_type == SLICE_TYPE_P && rc->qpm < h->fref0[0]->i_row_qp[y] )
{
- /* B-frames shouldn't use lower QP than their reference frames.
- * This code is a bit overzealous in limiting B-frame quantizers, but it helps avoid
- * underflows due to the fact that B-frames are not explicitly covered by VBV. */
- if( y < h->sps->i_mb_height-1 )
- {
- int i_estimated;
- int avg_qp = X264_MAX(h->fref0[0]->i_row_qp[y+1], h->fref1[0]->i_row_qp[y+1])
- + rc->pb_offset * ((h->fenc->i_type == X264_TYPE_BREF) ? 0.5 : 1);
- rc->qpm = X264_MIN(X264_MAX( rc->qp, avg_qp), 51); //avg_qp could go higher than 51 due to pb_offset
- i_estimated = row_bits_so_far(h, y); //FIXME: compute full estimated size
- if (i_estimated > h->rc->frame_size_planned)
- x264_ratecontrol_set_estimated_size(h, i_estimated);
- }
+ double newq = qp2qscale(rc->qpm);
+ double oldq = qp2qscale(h->fref0[0]->i_row_qp[y]);
+ update_predictor( rc->row_pred[1], (1 - newq / oldq) * newq, h->fdec->i_row_satds[0][0][y], h->fdec->i_row_bits[y] );
}
- else
+
+ /* tweak quality based on difference from predicted size */
+ if( y < h->sps->i_mb_height-1 )
{
- update_predictor( rc->row_pred, qp2qscale(rc->qpm), h->fdec->i_row_satd[y], h->fdec->i_row_bits[y] );
+ int prev_row_qp = h->fdec->i_row_qp[y];
+ int i_qp_max = X264_MIN( prev_row_qp + h->param.rc.i_qp_step, h->param.rc.i_qp_max );
+ int i_qp_min = X264_MAX( prev_row_qp - h->param.rc.i_qp_step, h->param.rc.i_qp_min );
- /* tweak quality based on difference from predicted size */
- if( y < h->sps->i_mb_height-1 && h->stat.i_slice_count[h->sh.i_type] > 0 )
+ /* B-frames shouldn't use lower QP than their reference frames. */
+ if( h->sh.i_type == SLICE_TYPE_B )
{
- int prev_row_qp = h->fdec->i_row_qp[y];
- int b0 = predict_row_size_sum( h, y, rc->qpm );
- int b1 = b0;
- int i_qp_max = X264_MIN( prev_row_qp + h->param.rc.i_qp_step, h->param.rc.i_qp_max );
- int i_qp_min = X264_MAX( prev_row_qp - h->param.rc.i_qp_step, h->param.rc.i_qp_min );
- float buffer_left_planned = rc->buffer_fill - rc->frame_size_planned;
- float rc_tol = 1;
- float headroom = 0;
-
- /* Don't modify the row QPs until a sufficent amount of the bits of the frame have been processed, in case a flat */
- /* area at the top of the frame was measured inaccurately. */
- if(row_bits_so_far(h,y) < 0.05 * rc->frame_size_planned)
- return;
-
- headroom = buffer_left_planned/rc->buffer_size;
- if(h->sh.i_type != SLICE_TYPE_I)
- headroom /= 2;
- rc_tol += headroom;
-
- if( !rc->b_vbv_min_rate )
- i_qp_min = X264_MAX( i_qp_min, h->sh.i_qp );
-
- while( rc->qpm < i_qp_max
- && (b1 > rc->frame_size_planned * rc_tol
- || (rc->buffer_fill - b1 < buffer_left_planned * 0.5)))
- {
- rc->qpm ++;
- b1 = predict_row_size_sum( h, y, rc->qpm );
- }
+ i_qp_min = X264_MAX( i_qp_min, X264_MAX( h->fref0[0]->i_row_qp[y+1], h->fref1[0]->i_row_qp[y+1] ) );
+ rc->qpm = X264_MAX( rc->qpm, i_qp_min );
+ }
- /* avoid VBV underflow */
- while( (rc->qpm < h->param.rc.i_qp_max)
- && (rc->buffer_fill - b1 < rc->buffer_size * 0.005))
- {
- rc->qpm ++;
- b1 = predict_row_size_sum( h, y, rc->qpm );
- }
+ int b0 = predict_row_size_sum( h, y, rc->qpm );
+ int b1 = b0;
+ float buffer_left_planned = rc->buffer_fill - rc->frame_size_planned;
- while( rc->qpm > i_qp_min
- && rc->qpm > h->fdec->i_row_qp[0]
- && ((b1 < rc->frame_size_planned * 0.8 && rc->qpm <= prev_row_qp)
- || b1 < (rc->buffer_fill - rc->buffer_size + rc->buffer_rate) * 1.1) )
- {
- rc->qpm --;
- b1 = predict_row_size_sum( h, y, rc->qpm );
- }
- x264_ratecontrol_set_estimated_size(h, b1);
+ /* More threads means we have to be more cautious in letting ratecontrol use up extra bits. */
+ float rc_tol = buffer_left_planned / h->param.i_threads * rc->rate_tolerance;
+
+ /* Don't modify the row QPs until a sufficent amount of the bits of the frame have been processed, in case a flat */
+ /* area at the top of the frame was measured inaccurately. */
+ if( row_bits_so_far(h,y) < 0.05 * rc->frame_size_planned )
+ return;
+
+ if( h->sh.i_type != SLICE_TYPE_I )
+ rc_tol /= 2;
+
+ if( !rc->b_vbv_min_rate )
+ i_qp_min = X264_MAX( i_qp_min, h->sh.i_qp );
+
+ while( rc->qpm < i_qp_max
+ && ((b1 > rc->frame_size_planned + rc_tol) ||
+ (rc->buffer_fill - b1 < buffer_left_planned * 0.5) ||
+ (b1 > rc->frame_size_planned && rc->qpm < rc->qp_novbv)) )
+ {
+ rc->qpm ++;
+ b1 = predict_row_size_sum( h, y, rc->qpm );
+ }
+
+ while( rc->qpm > i_qp_min
+ && (rc->qpm > h->fdec->i_row_qp[0] || rc->single_frame_vbv)
+ && ((b1 < rc->frame_size_planned * 0.8 && rc->qpm <= prev_row_qp)
+ || b1 < (rc->buffer_fill - rc->buffer_size + rc->buffer_rate) * 1.1) )
+ {
+ rc->qpm --;
+ b1 = predict_row_size_sum( h, y, rc->qpm );
+ }
+
+ /* avoid VBV underflow */
+ while( (rc->qpm < h->param.rc.i_qp_max)
+ && (rc->buffer_fill - b1 < rc->buffer_rate * 0.05 ) )
+ {
+ rc->qpm ++;
+ b1 = predict_row_size_sum( h, y, rc->qpm );
}
+
+ x264_ratecontrol_set_estimated_size(h, b1);
}
+
/* loses the fractional part of the frame-wise qp */
rc->f_qpm = rc->qpm;
}
/* We could try to initialize everything required for ABR and
* adaptive B-frames, but that would be complicated.
* So just calculate the average QP used so far. */
+ int i;
- h->param.rc.i_qp_constant = (h->stat.i_slice_count[SLICE_TYPE_P] == 0) ? 24
- : 1 + h->stat.f_slice_qp[SLICE_TYPE_P] / h->stat.i_slice_count[SLICE_TYPE_P];
+ h->param.rc.i_qp_constant = (h->stat.i_frame_count[SLICE_TYPE_P] == 0) ? 24
+ : 1 + h->stat.f_frame_qp[SLICE_TYPE_P] / h->stat.i_frame_count[SLICE_TYPE_P];
rc->qp_constant[SLICE_TYPE_P] = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
rc->qp_constant[SLICE_TYPE_I] = x264_clip3( (int)( qscale2qp( qp2qscale( h->param.rc.i_qp_constant ) / fabs( h->param.rc.f_ip_factor )) + 0.5 ), 0, 51 );
rc->qp_constant[SLICE_TYPE_B] = x264_clip3( (int)( qscale2qp( qp2qscale( h->param.rc.i_qp_constant ) * fabs( h->param.rc.f_pb_factor )) + 0.5 ), 0, 51 );
if( h->param.i_bframe_adaptive )
x264_log(h, X264_LOG_ERROR, "disabling adaptive B-frames\n");
- rc->b_abr = 0;
- rc->b_2pass = 0;
- h->param.rc.i_rc_method = X264_RC_CQP;
- h->param.rc.b_stat_read = 0;
- h->param.i_bframe_adaptive = 0;
- if( h->param.i_bframe > 1 )
- h->param.i_bframe = 1;
- return X264_TYPE_P;
+ for( i = 0; i < h->param.i_threads; i++ )
+ {
+ h->thread[i]->rc->b_abr = 0;
+ h->thread[i]->rc->b_2pass = 0;
+ h->thread[i]->param.rc.i_rc_method = X264_RC_CQP;
+ h->thread[i]->param.rc.b_stat_read = 0;
+ h->thread[i]->param.i_bframe_adaptive = 0;
+ h->thread[i]->param.i_scenecut_threshold = 0;
+ if( h->thread[i]->param.i_bframe > 1 )
+ h->thread[i]->param.i_bframe = 1;
+ }
+ return X264_TYPE_AUTO;
}
switch( rc->entry[frame_num].pict_type )
{
}
/* After encoding one frame, save stats and update ratecontrol state */
-void x264_ratecontrol_end( x264_t *h, int bits )
+int x264_ratecontrol_end( x264_t *h, int bits )
{
x264_ratecontrol_t *rc = h->rc;
const int *mbs = h->stat.frame.i_mb_count;
( dir_frame>0 ? 's' : dir_frame<0 ? 't' :
dir_avg>0 ? 's' : dir_avg<0 ? 't' : '-' )
: '-';
- fprintf( rc->p_stat_file_out,
- "in:%d out:%d type:%c q:%.2f tex:%d mv:%d misc:%d imb:%d pmb:%d smb:%d d:%c;\n",
+ if( fprintf( rc->p_stat_file_out,
+ "in:%d out:%d type:%c q:%.2f tex:%d mv:%d misc:%d imb:%d pmb:%d smb:%d d:%c ref:",
h->fenc->i_frame, h->i_frame,
c_type, rc->qpa_rc,
h->stat.frame.i_tex_bits,
h->stat.frame.i_mb_count_i,
h->stat.frame.i_mb_count_p,
h->stat.frame.i_mb_count_skip,
- c_direct);
+ c_direct) < 0 )
+ goto fail;
+
+ for( i = 0; i < h->i_ref0; i++ )
+ {
+ int refcount = h->param.b_interlaced ? h->stat.frame.i_mb_count_ref[0][i*2]
+ + h->stat.frame.i_mb_count_ref[0][i*2+1] :
+ h->stat.frame.i_mb_count_ref[0][i];
+ if( fprintf( rc->p_stat_file_out, "%d ", refcount ) < 0 )
+ goto fail;
+ }
+
+ if( fprintf( rc->p_stat_file_out, ";\n" ) < 0 )
+ goto fail;
+
+ /* Don't re-write the data in multi-pass mode. */
+ if( h->param.rc.b_mb_tree && h->fenc->b_kept_as_ref && !h->param.rc.b_stat_read )
+ {
+ uint8_t i_type = h->sh.i_type;
+ int i;
+ /* Values are stored as big-endian FIX8.8 */
+ for( i = 0; i < h->mb.i_mb_count; i++ )
+ rc->qp_buffer[i] = endian_fix16( h->fenc->f_qp_offset[i]*256.0 );
+ if( fwrite( &i_type, 1, 1, rc->p_mbtree_stat_file_out ) < 1 )
+ goto fail;
+ if( fwrite( rc->qp_buffer, sizeof(uint16_t), h->mb.i_mb_count, rc->p_mbtree_stat_file_out ) < h->mb.i_mb_count )
+ goto fail;
+ }
}
if( rc->b_abr )
rc->cplxr_sum *= rc->cbr_decay;
rc->wanted_bits_window += rc->bitrate / rc->fps;
rc->wanted_bits_window *= rc->cbr_decay;
-
- if( h->param.i_threads == 1 )
- accum_p_qp_update( h, rc->qpa_rc );
}
if( rc->b_2pass )
if( h->sh.i_type == SLICE_TYPE_B )
{
rc->bframe_bits += bits;
- if( !h->frames.current[0] || !IS_X264_TYPE_B(h->frames.current[0]->i_type) )
+ if( h->fenc->b_last_minigop_bframe )
{
update_predictor( rc->pred_b_from_p, qp2qscale(rc->qpa_rc),
h->fref1[h->i_ref1-1]->i_satd, rc->bframe_bits / rc->bframes );
}
update_vbv( h, bits );
+ return 0;
+fail:
+ x264_log(h, X264_LOG_ERROR, "ratecontrol_end: stats file could not be written to\n");
+ return -1;
}
/****************************************************************************
double q;
x264_zone_t *zone = get_zone( h, frame_num );
- q = pow( rce->blurred_complexity, 1 - h->param.rc.f_qcompress );
+ q = pow( rce->blurred_complexity, 1 - rcc->qcompress );
// avoid NaN's in the rc_eq
if(!isfinite(q) || rce->tex_bits + rce->mv_bits == 0)
static double predict_size( predictor_t *p, double q, double var )
{
- return p->coeff*var / (q*p->count);
+ return (p->coeff*var + p->offset) / (q*p->count);
}
static void update_predictor( predictor_t *p, double q, double var, double bits )
{
+ const double range = 1.5;
if( var < 10 )
return;
- p->count *= p->decay;
- p->coeff *= p->decay;
- p->count ++;
- p->coeff += bits*q / var;
+ double old_coeff = p->coeff / p->count;
+ double new_coeff = bits*q / var;
+ double new_coeff_clipped = x264_clip3f( new_coeff, old_coeff/range, old_coeff*range );
+ double new_offset = bits*q - new_coeff_clipped * var;
+ if( new_offset >= 0 )
+ new_coeff = new_coeff_clipped;
+ else
+ new_offset = 0;
+ p->count *= p->decay;
+ p->coeff *= p->decay;
+ p->offset *= p->decay;
+ p->count ++;
+ p->coeff += new_coeff;
+ p->offset += new_offset;
}
// update VBV after encoding a frame
if( !rcc->b_vbv )
return;
- rct->buffer_fill_final += rct->buffer_rate - bits;
+ rct->buffer_fill_final -= bits;
if( rct->buffer_fill_final < 0 )
- x264_log( h, X264_LOG_WARNING, "VBV underflow (%.0f bits)\n", rct->buffer_fill_final );
- rct->buffer_fill_final = x264_clip3f( rct->buffer_fill_final, 0, rct->buffer_size );
+ x264_log( h, X264_LOG_WARNING, "VBV underflow (frame %d, %.0f bits)\n", h->i_frame, rct->buffer_fill_final );
+ rct->buffer_fill_final = X264_MAX( rct->buffer_fill_final, 0 );
+ rct->buffer_fill_final += rct->buffer_rate;
+ rct->buffer_fill_final = X264_MIN( rct->buffer_fill_final, rct->buffer_size );
}
// provisionally update VBV according to the planned size of all frames currently in progress
-static void update_vbv_plan( x264_t *h )
+static void update_vbv_plan( x264_t *h, int overhead )
{
x264_ratecontrol_t *rcc = h->rc;
- rcc->buffer_fill = h->thread[0]->rc->buffer_fill_final;
+ rcc->buffer_fill = h->thread[0]->rc->buffer_fill_final - overhead;
if( h->param.i_threads > 1 )
{
int j = h->rc - h->thread[0]->rc;
if( !t->b_thread_active )
continue;
bits = X264_MAX(bits, x264_ratecontrol_get_estimated_size(t));
- rcc->buffer_fill += rcc->buffer_rate - bits;
- rcc->buffer_fill = x264_clip3( rcc->buffer_fill, 0, rcc->buffer_size );
+ rcc->buffer_fill -= bits;
+ rcc->buffer_fill = X264_MAX( rcc->buffer_fill, 0 );
+ rcc->buffer_fill += rcc->buffer_rate;
+ rcc->buffer_fill = X264_MIN( rcc->buffer_fill, rcc->buffer_size );
}
}
}
double q0 = q;
/* B-frames are not directly subject to VBV,
- * since they are controlled by the P-frames' QPs.
- * FIXME: in 2pass we could modify previous frames' QP too,
- * instead of waiting for the buffer to fill */
- if( rcc->b_vbv &&
- ( pict_type == SLICE_TYPE_P ||
- ( pict_type == SLICE_TYPE_I && rcc->last_non_b_pict_type == SLICE_TYPE_I ) ) )
- {
- if( rcc->buffer_fill/rcc->buffer_size < 0.5 )
- q /= x264_clip3f( 2.0*rcc->buffer_fill/rcc->buffer_size, 0.5, 1.0 );
- }
+ * since they are controlled by the P-frames' QPs. */
if( rcc->b_vbv && rcc->last_satd > 0 )
{
- /* Now a hard threshold to make sure the frame fits in VBV.
- * This one is mostly for I-frames. */
- double bits = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd );
- double qf = 1.0;
- if( bits > rcc->buffer_fill/2 )
- qf = x264_clip3f( rcc->buffer_fill/(2*bits), 0.2, 1.0 );
- q /= qf;
- bits *= qf;
- if( bits < rcc->buffer_rate/2 )
- q *= bits*2/rcc->buffer_rate;
- q = X264_MAX( q0, q );
+ /* Lookahead VBV: raise the quantizer as necessary such that no frames in
+ * the lookahead overflow and such that the buffer is in a reasonable state
+ * by the end of the lookahead. */
+ if( h->param.rc.i_lookahead )
+ {
+ int j, iterations, terminate = 0;
+
+ /* Avoid an infinite loop. */
+ for( iterations = 0; iterations < 1000 && terminate != 3; iterations++ )
+ {
+ double frame_q[3];
+ double cur_bits = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd );
+ double buffer_fill_cur = rcc->buffer_fill - cur_bits + rcc->buffer_rate;
+ double target_fill;
+ frame_q[0] = h->sh.i_type == SLICE_TYPE_I ? q * h->param.rc.f_ip_factor : q;
+ frame_q[1] = frame_q[0] * h->param.rc.f_pb_factor;
+ frame_q[2] = frame_q[0] / h->param.rc.f_ip_factor;
+
+ /* Loop over the planned future frames. */
+ for( j = 0; buffer_fill_cur >= 0 && buffer_fill_cur <= rcc->buffer_size; j++ )
+ {
+ int i_type = h->fenc->i_planned_type[j];
+ int i_satd = h->fenc->i_planned_satd[j];
+ if( i_type == X264_TYPE_AUTO )
+ break;
+ i_type = IS_X264_TYPE_I( i_type ) ? SLICE_TYPE_I : IS_X264_TYPE_B( i_type ) ? SLICE_TYPE_B : SLICE_TYPE_P;
+ cur_bits = predict_size( &rcc->pred[i_type], frame_q[i_type], i_satd );
+ buffer_fill_cur = buffer_fill_cur - cur_bits + rcc->buffer_rate;
+ }
+ /* Try to get to get the buffer at least 50% filled, but don't set an impossible goal. */
+ target_fill = X264_MIN( rcc->buffer_fill + j * rcc->buffer_rate * 0.5, rcc->buffer_size * 0.5 );
+ if( buffer_fill_cur < target_fill )
+ {
+ q *= 1.01;
+ terminate |= 1;
+ continue;
+ }
+ /* Try to get the buffer no more than 80% filled, but don't set an impossible goal. */
+ target_fill = x264_clip3f( rcc->buffer_fill - j * rcc->buffer_rate * 0.5, rcc->buffer_size * 0.8, rcc->buffer_size );
+ if( rcc->b_vbv_min_rate && buffer_fill_cur > target_fill )
+ {
+ q /= 1.01;
+ terminate |= 2;
+ continue;
+ }
+ break;
+ }
+ }
+ /* Fallback to old purely-reactive algorithm: no lookahead. */
+ else
+ {
+ if( ( pict_type == SLICE_TYPE_P ||
+ ( pict_type == SLICE_TYPE_I && rcc->last_non_b_pict_type == SLICE_TYPE_I ) ) &&
+ rcc->buffer_fill/rcc->buffer_size < 0.5 )
+ {
+ q /= x264_clip3f( 2.0*rcc->buffer_fill/rcc->buffer_size, 0.5, 1.0 );
+ }
+
+ /* Now a hard threshold to make sure the frame fits in VBV.
+ * This one is mostly for I-frames. */
+ double bits = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd );
+ double qf = 1.0;
+ /* For small VBVs, allow the frame to use up the entire VBV. */
+ double max_fill_factor = h->param.rc.i_vbv_buffer_size >= 5*h->param.rc.i_vbv_max_bitrate / rcc->fps ? 2 : 1;
+ /* For single-frame VBVs, request that the frame use up the entire VBV. */
+ double min_fill_factor = rcc->single_frame_vbv ? 1 : 2;
+
+ if( bits > rcc->buffer_fill/max_fill_factor )
+ qf = x264_clip3f( rcc->buffer_fill/(max_fill_factor*bits), 0.2, 1.0 );
+ q /= qf;
+ bits *= qf;
+ if( bits < rcc->buffer_rate/min_fill_factor )
+ q *= bits*min_fill_factor/rcc->buffer_rate;
+ q = X264_MAX( q0, q );
+ }
/* Check B-frame complexity, and use up any bits that would
* overflow before the next P-frame. */
- if( h->sh.i_type == SLICE_TYPE_P )
+ if( h->sh.i_type == SLICE_TYPE_P && !rcc->single_frame_vbv )
{
int nb = rcc->bframes;
+ double bits = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd );
double pbbits = bits;
double bbits = predict_size( rcc->pred_b_from_p, q * h->param.rc.f_pb_factor, rcc->last_satd );
double space;
-
- if( bbits > rcc->buffer_rate )
+ if( bbits > rcc->buffer_rate )
nb = 0;
pbbits += nb * bbits;
space = rcc->buffer_fill + (1+nb)*rcc->buffer_rate - rcc->buffer_size;
if( pbbits < space )
{
- q *= X264_MAX( pbbits / space,
- bits / (0.5 * rcc->buffer_size) );
+ q *= X264_MAX( pbbits / space, bits / (0.5 * rcc->buffer_size) );
}
q = X264_MAX( q0-5, q );
}
int pict_type = h->sh.i_type;
double lmin = rcc->lmin[pict_type];
double lmax = rcc->lmax[pict_type];
- int64_t total_bits = 8*(h->stat.i_slice_size[SLICE_TYPE_I]
- + h->stat.i_slice_size[SLICE_TYPE_P]
- + h->stat.i_slice_size[SLICE_TYPE_B]);
+ int64_t total_bits = 8*(h->stat.i_frame_size[SLICE_TYPE_I]
+ + h->stat.i_frame_size[SLICE_TYPE_P]
+ + h->stat.i_frame_size[SLICE_TYPE_B]);
if( rcc->b_2pass )
{
else
q += rcc->pb_offset;
- rcc->frame_size_planned = predict_size( rcc->pred_b_from_p, q, h->fref1[h->i_ref1-1]->i_satd );
+ if( rcc->b_2pass && rcc->b_vbv )
+ rcc->frame_size_planned = qscale2bits( &rce, q );
+ else
+ rcc->frame_size_planned = predict_size( rcc->pred_b_from_p, q, h->fref1[h->i_ref1-1]->i_satd );
x264_ratecontrol_set_estimated_size(h, rcc->frame_size_planned);
- rcc->last_satd = 0;
+
+ /* For row SATDs */
+ if( rcc->b_vbv )
+ rcc->last_satd = x264_rc_analyse_slice( h );
return qp2qscale(q);
}
else
{
double abr_buffer = 2 * rcc->rate_tolerance * rcc->bitrate;
+
if( rcc->b_2pass )
{
//FIXME adjust abr_buffer based on distance to the end of the video
- int64_t diff = total_bits - (int64_t)rce.expected_bits;
+ int64_t diff;
+ int64_t predicted_bits = total_bits;
+
+ if( rcc->b_vbv )
+ {
+ if( h->param.i_threads > 1 )
+ {
+ int j = h->rc - h->thread[0]->rc;
+ int i;
+ for( i=1; i<h->param.i_threads; i++ )
+ {
+ x264_t *t = h->thread[ (j+i)%h->param.i_threads ];
+ double bits = t->rc->frame_size_planned;
+ if( !t->b_thread_active )
+ continue;
+ bits = X264_MAX(bits, x264_ratecontrol_get_estimated_size(t));
+ predicted_bits += (int64_t)bits;
+ }
+ }
+ }
+ else
+ {
+ if( h->fenc->i_frame < h->param.i_threads )
+ predicted_bits += (int64_t)h->fenc->i_frame * rcc->bitrate / rcc->fps;
+ else
+ predicted_bits += (int64_t)(h->param.i_threads - 1) * rcc->bitrate / rcc->fps;
+ }
+
+ diff = predicted_bits - (int64_t)rce.expected_bits;
q = rce.new_qscale;
q /= x264_clip3f((double)(abr_buffer - diff) / abr_buffer, .5, 2);
- if( h->fenc->i_frame > 30 )
+ if( ((h->fenc->i_frame + 1 - h->param.i_threads) >= rcc->fps) &&
+ (rcc->expected_bits_sum > 0))
{
/* Adjust quant based on the difference between
* achieved and expected bitrate so far */
}
if( rcc->b_vbv )
{
+ /* Do not overflow vbv */
double expected_size = qscale2bits(&rce, q);
double expected_vbv = rcc->buffer_fill + rcc->buffer_rate - expected_size;
double expected_fullness = rce.expected_vbv / rcc->buffer_size;
double qmax = q*(2 - expected_fullness);
double size_constraint = 1 + expected_fullness;
+ qmax = X264_MAX(qmax, rce.new_qscale);
if (expected_fullness < .05)
qmax = lmax;
qmax = X264_MIN(qmax, lmax);
- while( (expected_vbv < rce.expected_vbv/size_constraint) && (q < qmax) )
+ while( ((expected_vbv < rce.expected_vbv/size_constraint) && (q < qmax)) ||
+ ((expected_vbv < 0) && (q < lmax)))
{
q *= 1.05;
expected_size = qscale2bits(&rce, q);
{
q = qp2qscale( ABR_INIT_QP ) / fabs( h->param.rc.f_ip_factor );
}
+ rcc->qp_novbv = qscale2qp(q);
//FIXME use get_diff_limited_q() ?
q = clip_qscale( h, pict_type, q );
if( !(rcc->b_2pass && !rcc->b_vbv) && h->fenc->i_frame == 0 )
rcc->last_qscale_for[SLICE_TYPE_P] = q;
- if( rcc->b_2pass && rcc->b_vbv)
+ if( rcc->b_2pass && rcc->b_vbv )
rcc->frame_size_planned = qscale2bits(&rce, q);
else
rcc->frame_size_planned = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd );
return expected_bits;
}
-static void vbv_pass2( x264_t *h )
+static int vbv_pass2( x264_t *h )
{
/* for each interval of buffer_full .. underflow, uniformly increase the qp of all
* frames in the interval until either buffer is full at some intermediate frame or the
* Then do the converse to put bits back into overflow areas until target size is met */
x264_ratecontrol_t *rcc = h->rc;
- double *fills = x264_malloc((rcc->num_entries+1)*sizeof(double));
+ double *fills;
double all_available_bits = h->param.rc.i_bitrate * 1000. * rcc->num_entries / rcc->fps;
double expected_bits = 0;
double adjustment;
double qscale_max = qp2qscale(h->param.rc.i_qp_max);
int iterations = 0;
int adj_min, adj_max;
+ CHECKED_MALLOC( fills, (rcc->num_entries+1)*sizeof(double) );
fills++;
adj_max = fix_underflow(h, t0, t1, 1.001, qscale_min, qscale_max);
expected_bits = count_expected_bits(h);
- } while(expected_bits < .995 * all_available_bits && expected_bits > prev_bits);
+ } while((expected_bits < .995*all_available_bits) && ((int64_t)(expected_bits+.5) > (int64_t)(prev_bits+.5)) );
if (!adj_max)
x264_log( h, X264_LOG_WARNING, "vbv-maxrate issue, qpmax or vbv-maxrate too low\n");
rcc->entry[i].expected_vbv = rcc->buffer_size - fills[i];
x264_free(fills-1);
+ return 0;
+fail:
+ return -1;
}
static int init_pass2( x264_t *h )
rce->blurred_complexity = cplx_sum / weight_sum;
}
- qscale = x264_malloc(sizeof(double)*rcc->num_entries);
- if(filter_size > 1)
- blurred_qscale = x264_malloc(sizeof(double)*rcc->num_entries);
+ CHECKED_MALLOC( qscale, sizeof(double)*rcc->num_entries );
+ if( filter_size > 1 )
+ CHECKED_MALLOC( blurred_qscale, sizeof(double)*rcc->num_entries );
else
blurred_qscale = qscale;
x264_free(blurred_qscale);
if(rcc->b_vbv)
- vbv_pass2(h);
+ if( vbv_pass2( h ) )
+ return -1;
expected_bits = count_expected_bits(h);
if(fabs(expected_bits/all_available_bits - 1.0) > 0.01)
}
return 0;
+fail:
+ return -1;
}
-
-
projects / x264 / blobdiff