#include "common/common.h"
#include "common/cpu.h"
#include "ratecontrol.h"
+#include "me.h"
typedef struct
{
int pict_type;
+ int frame_type;
int kept_as_ref;
- float qscale;
+ double qscale;
int mv_bits;
int tex_bits;
int misc_bits;
uint64_t expected_bits; /*total expected bits up to the current frame (current one excluded)*/
double expected_vbv;
- float new_qscale;
+ double new_qscale;
int new_qp;
int i_count;
int p_count;
int s_count;
float blurred_complexity;
char direct_mode;
+ int16_t weight[2];
+ int16_t i_weight_denom;
+ int refcount[16];
+ int refs;
} ratecontrol_entry_t;
typedef struct
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 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. */
+ uint16_t *qp_buffer[2]; /* Global buffers for converting MB-tree quantizer data. */
+ int qpbuf_pos; /* In order to handle pyramid reordering, QP buffer acts as a stack.
+ * This value is the current position (0 or 1). */
/* MBRC stuff */
double frame_size_estimated;
double frame_size_planned;
- predictor_t *row_pred;
- predictor_t row_preds[5];
+ double slice_size_planned;
+ 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 parse_zones( x264_t *h );
static int init_pass2(x264_t *);
-static float rate_estimate_qscale( x264_t *h, int overhead );
+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, 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 );
+#define CMP_OPT_FIRST_PASS( opt, param_val )\
+{\
+ if( ( p = strstr( opts, opt "=" ) ) && sscanf( p, opt "=%d" , &i ) && param_val != i )\
+ {\
+ x264_log( h, X264_LOG_ERROR, "different " opt " setting than first pass (%d vs %d)\n", param_val, i );\
+ return -1;\
+ }\
+}
+
/* Terminology:
* qp = h.264's quantizer
* qscale = linearized quantizer = Lagrange multiplier
+ rce->misc_bits;
}
+static ALWAYS_INLINE uint32_t ac_energy_plane( x264_t *h, int mb_x, int mb_y, x264_frame_t *frame, int i )
+{
+ int w = i ? 8 : 16;
+ int shift = i ? 6 : 8;
+ int stride = frame->i_stride[i];
+ int offset = h->mb.b_interlaced
+ ? w * (mb_x + (mb_y&~1) * stride) + (mb_y&1) * stride
+ : w * (mb_x + mb_y * stride);
+ int pix = i ? PIXEL_8x8 : PIXEL_16x16;
+ stride <<= h->mb.b_interlaced;
+ uint64_t res = h->pixf.var[pix]( frame->plane[i] + offset, stride );
+ uint32_t sum = (uint32_t)res;
+ uint32_t sqr = res >> 32;
+ return sqr - (sum * sum >> shift);
+}
+
// Find the total AC energy of the block in all planes.
static NOINLINE uint32_t ac_energy_mb( x264_t *h, int mb_x, int mb_y, x264_frame_t *frame )
{
* 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. */
- uint32_t var = 0, i;
- for( i = 0; i < 3; i++ )
- {
- int w = i ? 8 : 16;
- int stride = frame->i_stride[i];
- int offset = h->mb.b_interlaced
- ? w * (mb_x + (mb_y&~1) * stride) + (mb_y&1) * stride
- : 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 );
- }
+ uint32_t var = ac_energy_plane( h, mb_x, mb_y, frame, 0 );
+ var += ac_energy_plane( h, mb_x, mb_y, frame, 1 );
+ var += ac_energy_plane( h, mb_x, mb_y, frame, 2 );
x264_emms();
return var;
}
void x264_adaptive_quant( x264_t *h )
{
x264_emms();
- 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 );
+ /* MB-tree currently doesn't adjust quantizers in unreferenced frames. */
+ float qp_offset = h->fdec->b_kept_as_ref ? h->fenc->f_qp_offset[h->mb.i_mb_xy] : h->fenc->f_qp_offset_aq[h->mb.i_mb_xy];
+ h->mb.i_qp = x264_clip3( h->rc->f_qpm + qp_offset + .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 )
uint8_t i_type_actual = rc->entry[frame->i_frame].pict_type;
int i;
- if( i_type_actual != SLICE_TYPE_B )
+ if( rc->entry[frame->i_frame].kept_as_ref )
{
uint8_t i_type;
+ if( rc->qpbuf_pos < 0 )
+ {
+ do
+ {
+ rc->qpbuf_pos++;
- if( !fread( &i_type, 1, 1, rc->p_mbtree_stat_file_in ) )
- goto fail;
+ if( !fread( &i_type, 1, 1, rc->p_mbtree_stat_file_in ) )
+ goto fail;
+ if( fread( rc->qp_buffer[rc->qpbuf_pos], sizeof(uint16_t), h->mb.i_mb_count, rc->p_mbtree_stat_file_in ) != h->mb.i_mb_count )
+ 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( i_type != i_type_actual && rc->qpbuf_pos == 1 )
+ {
+ x264_log(h, X264_LOG_ERROR, "MB-tree frametype %d doesn't match actual frametype %d.\n", i_type, i_type_actual);
+ return -1;
+ }
+ } while( i_type != i_type_actual );
}
- 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);
+ {
+ frame->f_qp_offset[i] = ((float)(int16_t)endian_fix16( rc->qp_buffer[rc->qpbuf_pos][i] )) * (1/256.0);
+ if( h->frames.b_have_lowres )
+ frame->i_inv_qscale_factor[i] = x264_exp2fix8(frame->f_qp_offset[i]);
+ }
+ rc->qpbuf_pos--;
}
else
x264_adaptive_quant_frame( h, frame );
return -1;
}
+int x264_reference_build_list_optimal( x264_t *h )
+{
+ ratecontrol_entry_t *rce = h->rc->rce;
+ x264_frame_t *frames[16];
+ x264_weight_t weights[16][3];
+ int refcount[16];
+ int ref, i;
+
+ if( rce->refs != h->i_ref0 )
+ return -1;
+
+ memcpy( frames, h->fref0, sizeof(frames) );
+ memcpy( refcount, rce->refcount, sizeof(refcount) );
+ memcpy( weights, h->fenc->weight, sizeof(weights) );
+ memset( &h->fenc->weight[1][0], 0, sizeof(x264_weight_t[15][3]) );
+
+ /* 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++ )
+ if( !frames[i]->b_duplicate || frames[i]->i_frame != h->fref0[ref-1]->i_frame )
+ /* Favor lower POC as a tiebreaker. */
+ COPY2_IF_GT( max, refcount[i], bestref, i );
+
+ /* FIXME: If there are duplicates from frames other than ref0 then it is possible
+ * that the optimal ordering doesnt place every duplicate. */
+
+ refcount[bestref] = -1;
+ h->fref0[ref] = frames[bestref];
+ memcpy( h->fenc->weight[ref], weights[bestref], sizeof(weights[bestref]) );
+ }
+
+ return 0;
+}
+
static char *x264_strcat_filename( char *input, char *suffix )
{
char *output = x264_malloc( strlen( input ) + strlen( suffix ) + 1 );
return output;
}
+void x264_ratecontrol_init_reconfigurable( x264_t *h, int b_init )
+{
+ x264_ratecontrol_t *rc = h->rc;
+ if( !b_init && rc->b_2pass )
+ return;
+
+ 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 < (int)(h->param.rc.i_vbv_max_bitrate / rc->fps) )
+ {
+ 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 );
+ }
+
+ /* We don't support changing the ABR bitrate right now,
+ so if the stream starts as CBR, keep it CBR. */
+ if( rc->b_vbv_min_rate )
+ h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
+ 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;
+ 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);
+ if( b_init )
+ {
+ 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 );
+ 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->b_vbv = 1;
+ rc->b_vbv_min_rate = !rc->b_2pass
+ && h->param.rc.i_rc_method == X264_RC_ABR
+ && h->param.rc.i_vbv_max_bitrate <= h->param.rc.i_bitrate;
+ }
+ }
+ if( h->param.rc.i_rc_method == X264_RC_CRF )
+ {
+ /* 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);
+ 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 );
+ }
+}
+
int x264_ratecontrol_new( x264_t *h )
{
x264_ratecontrol_t *rc;
- int i;
+ int i, j;
x264_emms();
x264_log(h, X264_LOG_ERROR, "constant rate-factor is incompatible with 2pass.\n");
return -1;
}
- 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" );
- h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
- }
- }
- if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
- h->param.rc.i_vbv_max_bitrate > 0)
- x264_log(h, X264_LOG_WARNING, "max bitrate less than average bitrate, ignored.\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 )
- {
- 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 );
- }
- 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->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->b_vbv = 1;
- rc->b_vbv_min_rate = !rc->b_2pass
- && h->param.rc.i_rc_method == X264_RC_ABR
- && h->param.rc.i_vbv_max_bitrate <= h->param.rc.i_bitrate;
- }
- else if( h->param.rc.i_vbv_max_bitrate )
- {
- x264_log(h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize.\n");
- h->param.rc.i_vbv_max_bitrate = 0;
- }
- if(rc->rate_tolerance < 0.01)
+
+ x264_ratecontrol_init_reconfigurable( h, 1 );
+
+ if( rc->rate_tolerance < 0.01 )
{
x264_log(h, X264_LOG_WARNING, "bitrate tolerance too small, using .01\n");
rc->rate_tolerance = 0.01;
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.
- * Try to compensate for MB-tree's effects as well. */
- 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;
- 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->pb_offset = 6.0 * log(h->param.rc.f_pb_factor) / log(2.0);
rc->qp_constant[SLICE_TYPE_P] = h->param.rc.i_qp_constant;
rc->qp_constant[SLICE_TYPE_I] = x264_clip3( h->param.rc.i_qp_constant - rc->ip_offset + 0.5, 0, 51 );
rc->qp_constant[SLICE_TYPE_B] = x264_clip3( h->param.rc.i_qp_constant + rc->pb_offset + 0.5, 0, 51 );
+ h->mb.ip_offset = rc->ip_offset + 0.5;
rc->lstep = pow( 2, h->param.rc.i_qp_step / 6.0 );
rc->last_qscale = qp2qscale(26);
rc->pred[i].count= 1.0;
rc->pred[i].decay= 0.5;
rc->pred[i].offset= 0.0;
- rc->row_preds[i].coeff= .25;
- rc->row_preds[i].count= 1.0;
- rc->row_preds[i].decay= 0.5;
- rc->row_preds[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];
/* check whether 1st pass options were compatible with current options */
if( !strncmp( stats_buf, "#options:", 9 ) )
{
- int i;
+ int i, j;
char *opts = stats_buf;
stats_in = strchr( stats_buf, '\n' );
if( !stats_in )
return -1;
*stats_in = '\0';
stats_in++;
-
- if( ( p = strstr( opts, "bframes=" ) ) && sscanf( p, "bframes=%d", &i )
- && h->param.i_bframe != i )
+ if( sscanf( opts, "#options: %dx%d", &i, &j ) != 2 )
{
- x264_log( h, X264_LOG_ERROR, "different number of B-frames than 1st pass (%d vs %d)\n",
- h->param.i_bframe, i );
+ x264_log( h, X264_LOG_ERROR, "resolution specified in stats file not valid\n" );
+ return -1;
+ }
+ else if( h->param.rc.b_mb_tree && (i != h->param.i_width || j != h->param.i_height) )
+ {
+ x264_log( h, X264_LOG_ERROR, "MB-tree doesn't support different resolution than 1st pass (%dx%d vs %dx%d)\n",
+ h->param.i_width, h->param.i_height, i, j );
return -1;
}
- /* since B-adapt doesn't (yet) take into account B-pyramid,
- * the converse is not a problem */
- if( strstr( opts, "b_pyramid=1" ) && !h->param.b_bframe_pyramid )
- x264_log( h, X264_LOG_WARNING, "1st pass used B-pyramid, 2nd doesn't\n" );
-
- if( ( p = strstr( opts, "keyint=" ) ) && sscanf( p, "keyint=%d", &i )
- && h->param.i_keyint_max != i )
- x264_log( h, X264_LOG_WARNING, "different keyint than 1st pass (%d vs %d)\n",
- h->param.i_keyint_max, i );
+ CMP_OPT_FIRST_PASS( "wpredp", X264_MAX( 0, h->param.analyse.i_weighted_pred ) );
+ CMP_OPT_FIRST_PASS( "bframes", h->param.i_bframe );
+ CMP_OPT_FIRST_PASS( "b_pyramid", h->param.i_bframe_pyramid );
+ CMP_OPT_FIRST_PASS( "intra_refresh", h->param.b_intra_refresh );
+ CMP_OPT_FIRST_PASS( "keyint", h->param.i_keyint_max );
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" );
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);
- switch(pict_type)
+ 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;
+
+ /* find weights */
+ rce->i_weight_denom = -1;
+ char *w = strchr( p, 'w' );
+ if( w )
+ if( sscanf( w, "w:%hd,%hd,%hd", &rce->i_weight_denom, &rce->weight[0], &rce->weight[1] ) != 3 )
+ rce->i_weight_denom = -1;
+
+ if( pict_type != 'b' )
+ rce->kept_as_ref = 1;
+ switch( pict_type )
{
- case 'I': rce->kept_as_ref = 1;
- case 'i': rce->pict_type = SLICE_TYPE_I; break;
- case 'P': rce->pict_type = SLICE_TYPE_P; break;
- case 'B': rce->kept_as_ref = 1;
- case 'b': rce->pict_type = SLICE_TYPE_B; break;
+ case 'I':
+ rce->frame_type = X264_TYPE_IDR;
+ rce->pict_type = SLICE_TYPE_I;
+ break;
+ case 'i':
+ rce->frame_type = X264_TYPE_I;
+ rce->pict_type = SLICE_TYPE_I;
+ break;
+ case 'P':
+ rce->frame_type = X264_TYPE_P;
+ rce->pict_type = SLICE_TYPE_P;
+ break;
+ case 'B':
+ rce->frame_type = X264_TYPE_BREF;
+ rce->pict_type = SLICE_TYPE_B;
+ break;
+ case 'b':
+ rce->frame_type = X264_TYPE_B;
+ rce->pict_type = SLICE_TYPE_B;
+ break;
default: e = -1; break;
}
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_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) );
+ {
+ CHECKED_MALLOC( rc->qp_buffer[0], h->mb.i_mb_count * sizeof(uint16_t) );
+ if( h->param.i_bframe_pyramid && h->param.rc.b_stat_read )
+ CHECKED_MALLOC( rc->qp_buffer[1], h->mb.i_mb_count * sizeof(uint16_t) );
+ rc->qpbuf_pos = -1;
+ }
for( i=0; i<h->param.i_threads; i++ )
{
if( i )
{
rc[i] = rc[0];
- memcpy( &h->thread[i]->param, &h->param, sizeof(x264_param_t) );
+ h->thread[i]->param = h->param;
h->thread[i]->mb.b_variable_qp = h->mb.b_variable_qp;
}
}
static int parse_zone( x264_t *h, x264_zone_t *z, char *p )
{
int len = 0;
- char *tok, UNUSED *saveptr;
+ char *tok, UNUSED *saveptr=NULL;
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) )
int i;
if( h->param.rc.psz_zones && !h->param.rc.i_zones )
{
- char *psz_zones, *p, *tok, UNUSED *saveptr;
+ char *psz_zones, *p;
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;
p = psz_zones;
for( i = 0; i < h->param.rc.i_zones; i++ )
{
- tok = strtok_r( p, "/", &saveptr );
- if( !tok || parse_zone( h, &h->param.rc.zones[i], tok ) )
+ int i_tok = strcspn( p, "/" );
+ p[i_tok] = 0;
+ if( parse_zone( h, &h->param.rc.zones[i], p ) )
return -1;
- p = NULL;
+ p += i_tok + 1;
}
x264_free( psz_zones );
}
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)*12.5 : 0;
+ 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 - rc->qcompress )
* rc->cplxr_sum / rc->wanted_bits_window ) - mbtree_offset );
{
x264_ratecontrol_t *rc = h->rc;
int i;
+ int b_regular_file;
if( rc->p_stat_file_out )
{
+ b_regular_file = x264_is_regular_file( rc->p_stat_file_out );
fclose( rc->p_stat_file_out );
- if( h->i_frame >= rc->num_entries )
+ if( h->i_frame >= rc->num_entries && b_regular_file )
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",
}
if( rc->p_mbtree_stat_file_out )
{
+ b_regular_file = x264_is_regular_file( rc->p_mbtree_stat_file_out );
fclose( rc->p_mbtree_stat_file_out );
- if( h->i_frame >= rc->num_entries )
+ if( h->i_frame >= rc->num_entries && b_regular_file )
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",
x264_free( rc->pred );
x264_free( rc->pred_b_from_p );
x264_free( rc->entry );
- x264_free( rc->qp_buffer );
+ x264_free( rc->qp_buffer[0] );
+ x264_free( rc->qp_buffer[1] );
if( rc->zones )
{
x264_free( rc->zones[0].param );
}
else if( rc->b_abr )
{
- q = qscale2qp( rate_estimate_qscale( h, overhead ) );
+ q = qscale2qp( rate_estimate_qscale( h ) );
}
else if( rc->b_2pass )
{
- rce->new_qscale = rate_estimate_qscale( h, overhead );
+ rce->new_qscale = rate_estimate_qscale( h );
q = qscale2qp( rce->new_qscale );
}
else /* CQP */
/* 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] )
{
- 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( 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
+ {
+ double pred_intra = predict_size( rc->row_pred[1], qp2qscale(qp), 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 )
{
int i;
double bits = 0;
- for( i = 0; i <= y; i++ )
+ for( i = h->i_threadslice_start; i <= y; i++ )
bits += h->fdec->i_row_bits[i];
return bits;
}
{
int i;
double bits = row_bits_so_far(h, y);
- for( i = y+1; i < h->sps->i_mb_height; i++ )
+ for( i = y+1; i < h->i_threadslice_end; i++ )
bits += predict_row_size( h, i, qp );
return bits;
}
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] )
+ update_predictor( rc->row_pred[1], qp2qscale(rc->qpm), h->fdec->i_row_satds[0][0][y], h->fdec->i_row_bits[y] );
+
+ /* tweak quality based on difference from predicted size */
+ if( y < h->i_threadslice_end-1 )
{
- /* 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 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 );
+
+ /* B-frames shouldn't use lower QP than their reference frames. */
+ if( h->sh.i_type == SLICE_TYPE_B )
{
- int i_estimated;
- int avg_qp = X264_MIN(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);
+ 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 );
}
- }
- else
- {
- update_predictor( rc->row_pred, qp2qscale(rc->qpm), h->fdec->i_row_satd[y], h->fdec->i_row_bits[y] );
- /* tweak quality based on difference from predicted size */
- if( y < h->sps->i_mb_height-1 && h->stat.i_frame_count[h->sh.i_type] > 0 )
+ float buffer_left_planned = rc->buffer_fill - rc->frame_size_planned;
+ float slice_size_planned = h->param.b_sliced_threads ? rc->slice_size_planned : rc->frame_size_planned;
+ float size_of_other_slices = rc->frame_size_planned - slice_size_planned;
+ /* 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;
+ float max_frame_error = X264_MAX( 0.05, 1.0 / h->sps->i_mb_height );
+ int b1 = predict_row_size_sum( h, y, rc->qpm );
+
+ /* Assume that if this slice has become larger than expected,
+ * the other slices will have gotten equally larger. */
+ b1 += X264_MAX( size_of_other_slices * b1 / slice_size_planned, size_of_other_slices );
+
+ /* 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-size_of_other_slices) )
+ 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)) )
{
- 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;
- /* More threads means we have to be more cautious in letting ratecontrol use up extra bits.
- * In 2-pass mode we can be more trusting of the planned frame sizes, since they were decided
- * by actual encoding instead of SATD prediction. */
- float rc_tol = h->param.rc.b_stat_read ? (buffer_left_planned / rc->buffer_size) * rc->frame_size_planned
- : (buffer_left_planned / h->param.i_threads);
-
- /* 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 );
- }
+ rc->qpm ++;
+ b1 = predict_row_size_sum( h, y, rc->qpm );
+ b1 += X264_MAX( size_of_other_slices * b1 / slice_size_planned, size_of_other_slices );
+ }
- /* 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 );
- }
+ 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 );
+ b1 += X264_MAX( size_of_other_slices * b1 / slice_size_planned, size_of_other_slices );
+ }
- 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);
+ /* avoid VBV underflow */
+ while( (rc->qpm < h->param.rc.i_qp_max)
+ && (rc->buffer_fill - b1 < rc->buffer_rate * max_frame_error) )
+ {
+ rc->qpm ++;
+ b1 = predict_row_size_sum( h, y, rc->qpm );
+ b1 += X264_MAX( size_of_other_slices * b1 / slice_size_planned, size_of_other_slices );
}
+
+ x264_ratecontrol_set_estimated_size(h, b1);
}
+
/* loses the fractional part of the frame-wise qp */
rc->f_qpm = rc->qpm;
}
}
return X264_TYPE_AUTO;
}
- switch( rc->entry[frame_num].pict_type )
- {
- case SLICE_TYPE_I:
- return rc->entry[frame_num].kept_as_ref ? X264_TYPE_IDR : X264_TYPE_I;
-
- case SLICE_TYPE_B:
- return rc->entry[frame_num].kept_as_ref ? X264_TYPE_BREF : X264_TYPE_B;
-
- case SLICE_TYPE_P:
- default:
- return X264_TYPE_P;
- }
+ return rc->entry[frame_num].frame_type;
}
else
- {
return X264_TYPE_AUTO;
- }
+}
+
+void x264_ratecontrol_set_weights( x264_t *h, x264_frame_t *frm )
+{
+ ratecontrol_entry_t *rce = &h->rc->entry[frm->i_frame];
+ if( h->param.analyse.i_weighted_pred <= 0 )
+ return;
+ if( rce->i_weight_denom >= 0 )
+ SET_WEIGHT( frm->weight[0][0], 1, rce->weight[0], rce->i_weight_denom, rce->weight[1] );
}
/* After encoding one frame, save stats and update ratecontrol state */
dir_avg>0 ? 's' : dir_avg<0 ? 't' : '-' )
: '-';
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;\n",
+ "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_p,
h->stat.frame.i_mb_count_skip,
c_direct) < 0 )
- goto fail;
+ goto fail;
+
+ /* Only write information for reference reordering once. */
+ int use_old_stats = h->param.rc.b_stat_read && rc->rce->refs > 1;
+ for( i = 0; i < (use_old_stats ? rc->rce->refs : h->i_ref0); i++ )
+ {
+ int refcount = use_old_stats ? rc->rce->refcount[i]
+ : 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( h->sh.weight[0][0].weightfn )
+ {
+ if( fprintf( rc->p_stat_file_out, "w:%"PRId32",%"PRId32",%"PRId32, h->sh.weight[0][0].i_denom, h->sh.weight[0][0].i_scale, h->sh.weight[0][0].i_offset ) < 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 )
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 );
+ rc->qp_buffer[0][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 )
+ if( fwrite( rc->qp_buffer[0], sizeof(uint16_t), h->mb.i_mb_count, rc->p_mbtree_stat_file_out ) < h->mb.i_mb_count )
goto fail;
}
}
// avoid NaN's in the rc_eq
if(!isfinite(q) || rce->tex_bits + rce->mv_bits == 0)
- q = rcc->last_qscale;
+ q = rcc->last_qscale_for[rce->pict_type];
else
{
rcc->last_rceq = q;
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 += rcc->buffer_rate;
+ rct->buffer_fill_final = X264_MIN( rct->buffer_fill_final, rcc->buffer_size );
}
// provisionally update VBV according to the planned size of all frames currently in progress
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;// - overhead;
- if( h->param.i_threads > 1 )
+ rcc->buffer_fill = h->thread[0]->rc->buffer_fill_final;
+ if( h->i_thread_frames > 1 )
{
int j = h->rc - h->thread[0]->rc;
int i;
- for( i=1; i<h->param.i_threads; i++ )
+ for( i=1; i<h->i_thread_frames; i++ )
{
- x264_t *t = h->thread[ (j+i)%h->param.i_threads ];
+ x264_t *t = h->thread[ (j+i)%h->i_thread_frames ];
double bits = t->rc->frame_size_planned;
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 );
}
}
+ rcc->buffer_fill = X264_MIN( rcc->buffer_fill, rcc->buffer_size );
+ rcc->buffer_fill -= overhead;
}
// apply VBV constraints and clip qscale to between lmin and lmax
{
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 buffer_fill_cur = rcc->buffer_fill - cur_bits;
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;
/* Loop over the planned future frames. */
for( j = 0; buffer_fill_cur >= 0 && buffer_fill_cur <= rcc->buffer_size; j++ )
{
+ buffer_fill_cur += rcc->buffer_rate;
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;
+ buffer_fill_cur -= cur_bits;
}
/* 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 );
* 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 );
+ /* 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/2 )
- q *= bits*2/rcc->buffer_rate;
+ 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 );
}
}
// update qscale for 1 frame based on actual bits used so far
-static float rate_estimate_qscale( x264_t *h, int overhead )
+static float rate_estimate_qscale( x264_t *h )
{
float q;
x264_ratecontrol_t *rcc = h->rc;
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 ) + overhead;
+ 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;
+ double abr_buffer = 2 * rcc->rate_tolerance * rcc->bitrate * h->i_thread_frames;
if( rcc->b_2pass )
{
if( rcc->b_vbv )
{
- if( h->param.i_threads > 1 )
+ if( h->i_thread_frames > 1 )
{
int j = h->rc - h->thread[0]->rc;
int i;
- for( i=1; i<h->param.i_threads; i++ )
+ for( i=1; i<h->i_thread_frames; i++ )
{
- x264_t *t = h->thread[ (j+i)%h->param.i_threads ];
+ x264_t *t = h->thread[ (j+i)%h->i_thread_frames ];
double bits = t->rc->frame_size_planned;
if( !t->b_thread_active )
continue;
}
else
{
- if( h->fenc->i_frame < h->param.i_threads )
+ if( h->fenc->i_frame < h->i_thread_frames )
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;
+ predicted_bits += (int64_t)(h->i_thread_frames - 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 + 1 - h->param.i_threads) >= rcc->fps) &&
+ if( ((h->fenc->i_frame + 1 - h->i_thread_frames) >= rcc->fps) &&
(rcc->expected_bits_sum > 0))
{
/* Adjust quant based on the difference between
expected_size = qscale2bits(&rce, q);
expected_vbv = rcc->buffer_fill + rcc->buffer_rate - expected_size;
}
- rcc->last_satd = x264_stack_align( x264_rc_analyse_slice, h );
+ rcc->last_satd = x264_rc_analyse_slice( h );
}
q = x264_clip3f( q, lmin, lmax );
}
double wanted_bits, overflow=1, lmin, lmax;
- rcc->last_satd = x264_stack_align( x264_rc_analyse_slice, h );
+ rcc->last_satd = x264_rc_analyse_slice( h );
rcc->short_term_cplxsum *= 0.5;
rcc->short_term_cplxcount *= 0.5;
rcc->short_term_cplxsum += rcc->last_satd;
}
else
{
- int i_frame_done = h->fenc->i_frame + 1 - h->param.i_threads;
+ int i_frame_done = h->fenc->i_frame + 1 - h->i_thread_frames;
q = get_qscale( h, &rce, rcc->wanted_bits_window / rcc->cplxr_sum, h->fenc->i_frame );
q = x264_clip3f(q, lmin, lmax);
}
- else if( h->param.rc.i_rc_method == X264_RC_CRF )
+ else if( h->param.rc.i_rc_method == X264_RC_CRF && rcc->qcompress != 1 )
{
q = qp2qscale( ABR_INIT_QP ) / fabs( h->param.rc.f_ip_factor );
}
rcc->last_qscale = q;
if( !(rcc->b_2pass && !rcc->b_vbv) && h->fenc->i_frame == 0 )
- rcc->last_qscale_for[SLICE_TYPE_P] = q;
+ rcc->last_qscale_for[SLICE_TYPE_P] = q * fabs( h->param.rc.f_ip_factor );
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 ) + overhead;
+ rcc->frame_size_planned = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd );
+
+ /* Always use up the whole VBV in this case. */
+ if( rcc->single_frame_vbv )
+ rcc->frame_size_planned = rcc->buffer_rate;
x264_ratecontrol_set_estimated_size(h, rcc->frame_size_planned);
return q;
}
}
+void x264_threads_distribute_ratecontrol( x264_t *h )
+{
+ int i, row, totalsize = 0;
+ if( h->rc->b_vbv )
+ for( row = 0; row < h->sps->i_mb_height; row++ )
+ totalsize += h->fdec->i_row_satd[row];
+ for( i = 0; i < h->param.i_threads; i++ )
+ {
+ x264_t *t = h->thread[i];
+ x264_ratecontrol_t *rc = h->rc;
+ memcpy( t->rc, rc, sizeof(x264_ratecontrol_t) );
+ /* Calculate the planned slice size. */
+ if( h->rc->b_vbv && rc->frame_size_planned )
+ {
+ int size = 0;
+ for( row = t->i_threadslice_start; row < t->i_threadslice_end; row++ )
+ size += h->fdec->i_row_satd[row];
+ t->rc->slice_size_planned = size * rc->frame_size_planned / totalsize;
+ }
+ else
+ t->rc->slice_size_planned = 0;
+ }
+}
+
+void x264_threads_merge_ratecontrol( x264_t *h )
+{
+ int i, j, k;
+ x264_ratecontrol_t *rc = h->rc;
+ x264_emms();
+
+ for( i = 1; i < h->param.i_threads; i++ )
+ {
+ x264_ratecontrol_t *t = h->thread[i]->rc;
+ rc->qpa_rc += t->qpa_rc;
+ rc->qpa_aq += t->qpa_aq;
+ for( j = 0; j < 5; j++ )
+ for( k = 0; k < 2; k++ )
+ {
+ rc->row_preds[j][k].coeff += t->row_preds[j][k].coeff;
+ rc->row_preds[j][k].offset += t->row_preds[j][k].offset;
+ rc->row_preds[j][k].count += t->row_preds[j][k].count;
+ }
+ }
+ for( j = 0; j < 5; j++ )
+ for( k = 0; k < 2; k++ )
+ {
+ rc->row_preds[j][k].coeff /= h->param.i_threads;
+ rc->row_preds[j][k].offset /= h->param.i_threads;
+ rc->row_preds[j][k].count /= h->param.i_threads;
+ }
+}
+
void x264_thread_sync_ratecontrol( x264_t *cur, x264_t *prev, x264_t *next )
{
if( cur != prev )
#define COPY(var) memcpy(&cur->rc->var, &prev->rc->var, sizeof(cur->rc->var))
/* these vars are updated in x264_ratecontrol_start()
* so copy them from the context that most recently started (prev)
- * to the context that's about to start (cur).
- */
+ * to the context that's about to start (cur). */
COPY(accum_p_qp);
COPY(accum_p_norm);
COPY(last_satd);
COPY(short_term_cplxcount);
COPY(bframes);
COPY(prev_zone);
+ COPY(qpbuf_pos);
+ /* these vars can be updated by x264_ratecontrol_init_reconfigurable */
+ COPY(buffer_rate);
+ COPY(buffer_size);
+ COPY(single_frame_vbv);
+ COPY(cbr_decay);
+ COPY(b_vbv_min_rate);
+ COPY(rate_factor_constant);
+ COPY(bitrate);
#undef COPY
}
if( cur != next )
#define COPY(var) next->rc->var = cur->rc->var
/* these vars are updated in x264_ratecontrol_end()
* so copy them from the context that most recently ended (cur)
- * to the context that's about to end (next)
- */
+ * to the context that's about to end (next) */
COPY(cplxr_sum);
COPY(expected_bits_sum);
COPY(wanted_bits_window);
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) && ((int)(expected_bits+.5) > (int)(prev_bits+.5)) );
+ } 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");
expected_bits = 1;
for(i=0; i<rcc->num_entries; i++)
- expected_bits += qscale2bits(&rcc->entry[i], get_qscale(h, &rcc->entry[i], 1.0, i));
+ {
+ double q = get_qscale(h, &rcc->entry[i], 1.0, i);
+ expected_bits += qscale2bits(&rcc->entry[i], q);
+ rcc->last_qscale_for[rcc->entry[i].pict_type] = q;
+ }
step_mult = all_available_bits / expected_bits;
rate_factor = 0;
for(i=0; i<rcc->num_entries; i++)
{
qscale[i] = get_qscale(h, &rcc->entry[i], rate_factor, i);
+ rcc->last_qscale_for[rcc->entry[i].pict_type] = qscale[i];
}
/* fixed I/B qscale relative to P */