X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;ds=sidebyside;f=encoder%2Fratecontrol.c;h=e0e6761148693490d719f3ccba6fe67c217851fe;hb=0dbc490af39525adbefc1757151e5801c79eac3b;hp=c0a4bd769d6a21d5aa4d3e9568c35aac621f477d;hpb=d4ca70f8398bdba2391fbcea4886ee0577494b08;p=x264 diff --git a/encoder/ratecontrol.c b/encoder/ratecontrol.c index c0a4bd76..e0e67611 100644 --- a/encoder/ratecontrol.c +++ b/encoder/ratecontrol.c @@ -1,7 +1,7 @@ -/***************************************************-*- coding: iso-8859-1 -*- - * ratecontrol.c: h264 encoder library (Rate Control) +/***************************************************************************** + * ratecontrol.c: ratecontrol ***************************************************************************** - * Copyright (C) 2005-2008 x264 project + * Copyright (C) 2005-2010 x264 project * * Authors: Loren Merritt * Michael Niedermayer @@ -22,35 +22,43 @@ * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. + * + * This program is also available under a commercial proprietary license. + * For more information, contact us at licensing@x264.com. *****************************************************************************/ #define _ISOC99_SOURCE #undef NDEBUG // always check asserts, the speed effect is far too small to disable them #include -#include -#include #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; + int i_duration; + int i_cpb_duration; } ratecontrol_entry_t; typedef struct @@ -58,6 +66,7 @@ typedef struct double coeff; double count; double decay; + double offset; } predictor_t; struct x264_ratecontrol_t @@ -70,30 +79,35 @@ 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]; + int qp_constant[3]; /* current frame */ ratecontrol_entry_t *rce; int qp; /* qp for current frame */ - int qpm; /* qp for current macroblock */ - float f_qpm; /* qp for current macroblock: precise float for AQ */ + float 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_size; - double buffer_fill_final; /* real buffer as of the last finished frame */ + int64_t buffer_fill_final; 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 */ + double vbv_max_rate; /* # of bits added to buffer_fill per second */ predictor_t *pred; /* predict frame size from satd */ + int single_frame_vbv; + double rate_factor_max_increment; /* Don't allow RF above (CRF + this value). */ /* 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, only includes finished frames */ + int64_t filler_bits_sum; /* sum in bits of finished frames' filler data */ double wanted_bits_window; /* target bitrate * window */ double cbr_decay; double short_term_cplxsum; @@ -105,24 +119,34 @@ struct x264_ratecontrol_t /* 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 last_qscale; - double last_qscale_for[5]; /* last qscale for a specific pict type, used for max_diff & ipb factor stuff */ + double last_qscale_for[3]; /* last qscale for a specific pict type, used for max_diff & ipb factor stuff */ int last_non_b_pict_type; double accum_p_qp; /* for determining I-frame quant */ double accum_p_norm; double last_accum_p_norm; - double lmin[5]; /* min qscale by frame type */ - double lmax[5]; + double lmin[3]; /* min qscale by frame type */ + double lmax[3]; double lstep; /* max change (multiply) in qscale per frame */ + 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; + float frame_size_estimated; /* Access to this variable must be atomic: double is + * not atomic on all arches we care about */ + double frame_size_maximum; /* Maximum frame size due to MinCR */ 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[3][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 */ @@ -130,159 +154,431 @@ struct x264_ratecontrol_t int i_zones; x264_zone_t *zones; x264_zone_t *prev_zone; + + /* hrd stuff */ + int initial_cpb_removal_delay; + int initial_cpb_removal_delay_offset; + double nrt_first_access_unit; /* nominal removal time */ + double previous_cpb_final_arrival_time; + uint64_t hrd_multiply_denom; }; static int parse_zones( x264_t *h ); 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 int 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 */ -static inline double qp2qscale(double qp) +static inline double qp2qscale( double qp ) { - return 0.85 * pow(2.0, ( qp - 12.0 ) / 6.0); + return 0.85 * pow( 2.0, ( qp - 12.0 ) / 6.0 ); } -static inline double qscale2qp(double qscale) +static inline double qscale2qp( double qscale ) { - return 12.0 + 6.0 * log(qscale/0.85) / log(2.0); + return 12.0 + 6.0 * log2( qscale/0.85 ); } /* Texture bitrate is not quite inversely proportional to qscale, * probably due the the changing number of SKIP blocks. * MV bits level off at about qp<=12, because the lambda used * for motion estimation is constant there. */ -static inline double qscale2bits(ratecontrol_entry_t *rce, double qscale) +static inline double qscale2bits( ratecontrol_entry_t *rce, double qscale ) { - if(qscale<0.1) + if( qscale<0.1 ) qscale = 0.1; return (rce->tex_bits + .1) * pow( rce->qscale / qscale, 1.1 ) + rce->mv_bits * pow( X264_MAX(rce->qscale, 1) / X264_MAX(qscale, 1), 0.5 ) + rce->misc_bits; } +static ALWAYS_INLINE uint32_t ac_energy_var( uint64_t sum_ssd, int shift, x264_frame_t *frame, int i ) +{ + uint32_t sum = sum_ssd; + uint32_t ssd = sum_ssd >> 32; + frame->i_pixel_sum[i] += sum; + frame->i_pixel_ssd[i] += ssd; + return ssd - ((uint64_t)sum * sum >> shift); +} + +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 stride = frame->i_stride[i]; + int offset = h->mb.b_interlaced + ? 16 * mb_x + w * (mb_y&~1) * stride + (mb_y&1) * stride + : 16 * mb_x + w * mb_y * stride; + stride <<= h->mb.b_interlaced; + if( i ) + { + ALIGNED_ARRAY_16( pixel, pix,[FENC_STRIDE*8] ); + h->mc.load_deinterleave_8x8x2_fenc( pix, frame->plane[1] + offset, stride ); + return ac_energy_var( h->pixf.var[PIXEL_8x8]( pix, FENC_STRIDE ), 6, frame, i ) + + ac_energy_var( h->pixf.var[PIXEL_8x8]( pix+FENC_STRIDE/2, FENC_STRIDE ), 6, frame, i ); + } + else + return ac_energy_var( h->pixf.var[PIXEL_16x16]( frame->plane[0] + offset, stride ), 8, frame, i ); +} + // 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 x264_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, 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 ); - } - var = X264_MAX(var,1); + uint32_t var = ac_energy_plane( h, mb_x, mb_y, frame, 0 ); + var += ac_energy_plane( h, mb_x, mb_y, frame, 1 ); x264_emms(); return var; } -static const float log2_lut[128] = { - 0.00000, 0.01123, 0.02237, 0.03342, 0.04439, 0.05528, 0.06609, 0.07682, - 0.08746, 0.09803, 0.10852, 0.11894, 0.12928, 0.13955, 0.14975, 0.15987, - 0.16993, 0.17991, 0.18982, 0.19967, 0.20945, 0.21917, 0.22882, 0.23840, - 0.24793, 0.25739, 0.26679, 0.27612, 0.28540, 0.29462, 0.30378, 0.31288, - 0.32193, 0.33092, 0.33985, 0.34873, 0.35755, 0.36632, 0.37504, 0.38370, - 0.39232, 0.40088, 0.40939, 0.41785, 0.42626, 0.43463, 0.44294, 0.45121, - 0.45943, 0.46761, 0.47573, 0.48382, 0.49185, 0.49985, 0.50779, 0.51570, - 0.52356, 0.53138, 0.53916, 0.54689, 0.55459, 0.56224, 0.56986, 0.57743, - 0.58496, 0.59246, 0.59991, 0.60733, 0.61471, 0.62205, 0.62936, 0.63662, - 0.64386, 0.65105, 0.65821, 0.66534, 0.67243, 0.67948, 0.68650, 0.69349, - 0.70044, 0.70736, 0.71425, 0.72110, 0.72792, 0.73471, 0.74147, 0.74819, - 0.75489, 0.76155, 0.76818, 0.77479, 0.78136, 0.78790, 0.79442, 0.80090, - 0.80735, 0.81378, 0.82018, 0.82655, 0.83289, 0.83920, 0.84549, 0.85175, - 0.85798, 0.86419, 0.87036, 0.87652, 0.88264, 0.88874, 0.89482, 0.90087, - 0.90689, 0.91289, 0.91886, 0.92481, 0.93074, 0.93664, 0.94251, 0.94837, - 0.95420, 0.96000, 0.96578, 0.97154, 0.97728, 0.98299, 0.98868, 0.99435, -}; +void x264_adaptive_quant_frame( x264_t *h, x264_frame_t *frame, float *quant_offsets ) +{ + /* 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. */ + float strength; + float avg_adj = 0.f; + /* Initialize frame stats */ + for( int i = 0; i < 3; i++ ) + { + frame->i_pixel_sum[i] = 0; + frame->i_pixel_ssd[i] = 0; + } -static const uint8_t exp2_lut[64] = { - 1, 4, 7, 10, 13, 16, 19, 22, 25, 28, 31, 34, 37, 40, 44, 47, - 50, 53, 57, 60, 64, 67, 71, 74, 78, 81, 85, 89, 93, 96, 100, 104, - 108, 112, 116, 120, 124, 128, 132, 137, 141, 145, 150, 154, 159, 163, 168, 172, - 177, 182, 186, 191, 196, 201, 206, 211, 216, 221, 226, 232, 237, 242, 248, 253, -}; + /* Degenerate cases */ + if( h->param.rc.i_aq_mode == X264_AQ_NONE || h->param.rc.f_aq_strength == 0 ) + { + /* Need to init it anyways for MB tree */ + if( h->param.rc.i_aq_mode && h->param.rc.f_aq_strength == 0 ) + { + if( quant_offsets ) + { + for( int mb_xy = 0; mb_xy < h->mb.i_mb_count; mb_xy++ ) + frame->f_qp_offset[mb_xy] = frame->f_qp_offset_aq[mb_xy] = quant_offsets[mb_xy]; + if( h->frames.b_have_lowres ) + for( int mb_xy = 0; mb_xy < h->mb.i_mb_count; mb_xy++ ) + frame->i_inv_qscale_factor[mb_xy] = x264_exp2fix8( frame->f_qp_offset[mb_xy] ); + } + else + { + 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( int mb_xy = 0; mb_xy < h->mb.i_mb_count; mb_xy++ ) + frame->i_inv_qscale_factor[mb_xy] = 256; + } + } + /* Need variance data for weighted prediction */ + if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART ) + { + for( int mb_y = 0; mb_y < h->mb.i_mb_height; mb_y++ ) + for( int mb_x = 0; mb_x < h->mb.i_mb_width; mb_x++ ) + x264_ac_energy_mb( h, mb_x, mb_y, frame ); + } + else + return; + } + /* Actual adaptive quantization */ + else + { + if( h->param.rc.i_aq_mode == X264_AQ_AUTOVARIANCE ) + { + float bit_depth_correction = powf(1 << (BIT_DEPTH-8), 0.5f); + float avg_adj_pow2 = 0.f; + for( int mb_y = 0; mb_y < h->mb.i_mb_height; mb_y++ ) + for( int mb_x = 0; mb_x < h->mb.i_mb_width; mb_x++ ) + { + uint32_t energy = x264_ac_energy_mb( h, mb_x, mb_y, frame ); + float qp_adj = powf( energy + 1, 0.125f ); + frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride] = qp_adj; + avg_adj += qp_adj; + avg_adj_pow2 += qp_adj * qp_adj; + } + avg_adj /= h->mb.i_mb_count; + avg_adj_pow2 /= h->mb.i_mb_count; + strength = h->param.rc.f_aq_strength * avg_adj / bit_depth_correction; + avg_adj = avg_adj - 0.5f * (avg_adj_pow2 - (14.f * bit_depth_correction)) / avg_adj; + } + else + strength = h->param.rc.f_aq_strength * 1.0397f; -static int x264_exp2fix8( float x ) -{ - int i, f; - x += 8; - if( x <= 0 ) return 0; - if( x >= 16 ) return 0xffff; - i = x; - f = (x-i)*64; - return (exp2_lut[f]+256) << i >> 8; + for( int mb_y = 0; mb_y < h->mb.i_mb_height; mb_y++ ) + for( int mb_x = 0; mb_x < h->mb.i_mb_width; mb_x++ ) + { + float qp_adj; + int mb_xy = mb_x + mb_y*h->mb.i_mb_stride; + if( h->param.rc.i_aq_mode == X264_AQ_AUTOVARIANCE ) + { + qp_adj = frame->f_qp_offset[mb_xy]; + qp_adj = strength * (qp_adj - avg_adj); + } + else + { + uint32_t energy = x264_ac_energy_mb( h, mb_x, mb_y, frame ); + qp_adj = strength * (x264_log2( X264_MAX(energy, 1) ) - (14.427f + 2*(BIT_DEPTH-8))); + } + if( quant_offsets ) + qp_adj += quant_offsets[mb_xy]; + frame->f_qp_offset[mb_xy] = + frame->f_qp_offset_aq[mb_xy] = qp_adj; + if( h->frames.b_have_lowres ) + frame->i_inv_qscale_factor[mb_xy] = x264_exp2fix8(qp_adj); + } + } + + /* Remove mean from SSD calculation */ + for( int i = 0; i < 3; i++ ) + { + uint64_t ssd = frame->i_pixel_ssd[i]; + uint64_t sum = frame->i_pixel_sum[i]; + int width = h->mb.i_mb_width*16>>!!i; + int height = h->mb.i_mb_height*16>>!!i; + frame->i_pixel_ssd[i] = ssd - (sum * sum + width * height / 2) / (width * height); + } } -void x264_adaptive_quant_frame( x264_t *h, x264_frame_t *frame ) +int x264_macroblock_tree_read( x264_t *h, x264_frame_t *frame, float *quant_offsets ) { - /* 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. */ - float strength = h->param.rc.f_aq_strength * 1.0397; - 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++ ) - { - uint32_t energy = ac_energy_mb( h, mb_x, mb_y, frame ); - int lz = x264_clz( energy ); - float qp_adj = strength * (log2_lut[(energy<>24)&0x7f] - lz + 16.573f); - frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride] = qp_adj; + x264_ratecontrol_t *rc = h->rc; + uint8_t i_type_actual = rc->entry[frame->i_frame].pict_type; + + 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( 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 && 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 ); + } + + for( int i = 0; i < h->mb.i_mb_count; i++ ) + { + 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[mb_x + mb_y*h->mb.i_mb_stride] = x264_exp2fix8(qp_adj*(-1.f/6.f)); + frame->i_inv_qscale_factor[i] = x264_exp2fix8(frame->f_qp_offset[i]); } + rc->qpbuf_pos--; + } + else + x264_stack_align( x264_adaptive_quant_frame, h, frame, quant_offsets ); + 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]; + x264_weight_t weights[16][3]; + int refcount[16]; -/***************************************************************************** -* x264_adaptive_quant: - * adjust macroblock QP based on variance (AC energy) of the MB. - * high variance = higher QP - * low variance = lower QP - * This generally increases SSIM and lowers PSNR. -*****************************************************************************/ -void x264_adaptive_quant( x264_t *h ) + 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( int ref = 1; ref < h->i_ref0; ref++ ) + { + int max = -1; + int bestref = 1; + + for( int i = 1; i < h->i_ref0; i++ ) + /* 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 ) { - 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 ); - /* 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; + char *output = x264_malloc( strlen( input ) + strlen( suffix ) + 1 ); + if( !output ) + return NULL; + strcpy( output, input ); + strcat( output, suffix ); + 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_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 ); + } + + 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; + + int vbv_buffer_size = h->param.rc.i_vbv_buffer_size * 1000; + int vbv_max_bitrate = h->param.rc.i_vbv_max_bitrate * 1000; + + /* Init HRD */ + h->sps->vui.hrd.i_bit_rate_unscaled = vbv_max_bitrate; + h->sps->vui.hrd.i_cpb_size_unscaled = vbv_buffer_size; + if( h->param.i_nal_hrd && b_init ) + { + h->sps->vui.hrd.i_cpb_cnt = 1; + h->sps->vui.hrd.b_cbr_hrd = h->param.i_nal_hrd == X264_NAL_HRD_CBR; + h->sps->vui.hrd.i_time_offset_length = 0; + + #define BR_SHIFT 6 + #define CPB_SHIFT 4 + + int bitrate = 1000*h->param.rc.i_vbv_max_bitrate; + int bufsize = 1000*h->param.rc.i_vbv_buffer_size; + + // normalize HRD size and rate to the value / scale notation + h->sps->vui.hrd.i_bit_rate_scale = x264_clip3( x264_ctz( bitrate ) - BR_SHIFT, 0, 15 ); + h->sps->vui.hrd.i_bit_rate_value = bitrate >> ( h->sps->vui.hrd.i_bit_rate_scale + BR_SHIFT ); + h->sps->vui.hrd.i_bit_rate_unscaled = h->sps->vui.hrd.i_bit_rate_value << ( h->sps->vui.hrd.i_bit_rate_scale + BR_SHIFT ); + h->sps->vui.hrd.i_cpb_size_scale = x264_clip3( x264_ctz( bufsize ) - CPB_SHIFT, 0, 15 ); + h->sps->vui.hrd.i_cpb_size_value = bufsize >> ( h->sps->vui.hrd.i_cpb_size_scale + CPB_SHIFT ); + h->sps->vui.hrd.i_cpb_size_unscaled = h->sps->vui.hrd.i_cpb_size_value << ( h->sps->vui.hrd.i_cpb_size_scale + CPB_SHIFT ); + + #undef CPB_SHIFT + #undef BR_SHIFT + + // arbitrary + #define MAX_DURATION 0.5 + + int max_cpb_output_delay = X264_MIN( h->param.i_keyint_max * MAX_DURATION * h->sps->vui.i_time_scale / h->sps->vui.i_num_units_in_tick, INT_MAX ); + int max_dpb_output_delay = h->sps->vui.i_max_dec_frame_buffering * MAX_DURATION * h->sps->vui.i_time_scale / h->sps->vui.i_num_units_in_tick; + int max_delay = (int)(90000.0 * (double)h->sps->vui.hrd.i_cpb_size_unscaled / h->sps->vui.hrd.i_bit_rate_unscaled + 0.5); + + h->sps->vui.hrd.i_initial_cpb_removal_delay_length = 2 + x264_clip3( 32 - x264_clz( max_delay ), 4, 22 ); + h->sps->vui.hrd.i_cpb_removal_delay_length = x264_clip3( 32 - x264_clz( max_cpb_output_delay ), 4, 31 ); + h->sps->vui.hrd.i_dpb_output_delay_length = x264_clip3( 32 - x264_clz( max_dpb_output_delay ), 4, 31 ); + + #undef MAX_DURATION + + vbv_buffer_size = h->sps->vui.hrd.i_cpb_size_unscaled; + vbv_max_bitrate = h->sps->vui.hrd.i_bit_rate_unscaled; + } + else if( h->param.i_nal_hrd && !b_init ) + { + x264_log( h, X264_LOG_WARNING, "VBV parameters cannot be changed when NAL HRD is in use\n" ); + return; + } + + rc->buffer_rate = vbv_max_bitrate / rc->fps; + rc->vbv_max_rate = vbv_max_bitrate; + rc->buffer_size = vbv_buffer_size; + 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( h->param.rc.i_rc_method == X264_RC_CRF && h->param.rc.f_rf_constant_max ) + { + rc->rate_factor_max_increment = h->param.rc.f_rf_constant_max - h->param.rc.f_rf_constant; + if( rc->rate_factor_max_increment <= 0 ) + { + x264_log( h, X264_LOG_WARNING, "CRF max must be greater than CRF\n" ); + rc->rate_factor_max_increment = 0; + } + } + 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_clip3f( X264_MAX( h->param.rc.f_vbv_buffer_init, rc->buffer_rate / rc->buffer_size ), 0, 1); + rc->buffer_fill_final = rc->buffer_size * h->param.rc.f_vbv_buffer_init * h->sps->vui.i_time_scale; + 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; + } + } } int x264_ratecontrol_new( x264_t *h ) { x264_ratecontrol_t *rc; - int i; 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; /* FIXME: use integers */ - if(h->param.i_fps_num > 0 && h->param.i_fps_den > 0) + if( h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ) rc->fps = (float) h->param.i_fps_num / h->param.i_fps_den; 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; @@ -294,50 +590,27 @@ int x264_ratecontrol_new( x264_t *h ) 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 ) + + x264_ratecontrol_init_reconfigurable( h, 1 ); + + if( h->param.i_nal_hrd ) { - if( h->param.rc.i_vbv_buffer_size < 3 * h->param.rc.i_vbv_max_bitrate / rc->fps ) + uint64_t denom = (uint64_t)h->sps->vui.hrd.i_bit_rate_unscaled * h->sps->vui.i_time_scale; + uint64_t num = 180000; + x264_reduce_fraction64( &num, &denom ); + rc->hrd_multiply_denom = 180000 / num; + + double bits_required = log2( 180000 / rc->hrd_multiply_denom ) + + log2( h->sps->vui.i_time_scale ) + + log2( h->sps->vui.hrd.i_cpb_size_unscaled ); + if( bits_required >= 63 ) { - 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 ); + x264_log( h, X264_LOG_ERROR, "HRD with very large timescale and bufsize not supported\n" ); + return -1; } - 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) + + if( rc->rate_tolerance < 0.01 ) { x264_log(h, X264_LOG_WARNING, "bitrate tolerance too small, using .01\n"); rc->rate_tolerance = 0.01; @@ -352,40 +625,42 @@ int x264_ratecontrol_new( x264_t *h ) 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 */ - 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 ); - } - - 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->ip_offset = 6.0 * log2f( h->param.rc.f_ip_factor ); + rc->pb_offset = 6.0 * log2f( h->param.rc.f_pb_factor ); 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 ); + rc->qp_constant[SLICE_TYPE_I] = x264_clip3( h->param.rc.i_qp_constant - rc->ip_offset + 0.5, 0, QP_MAX ); + rc->qp_constant[SLICE_TYPE_B] = x264_clip3( h->param.rc.i_qp_constant + rc->pb_offset + 0.5, 0, QP_MAX ); + 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 = x264_malloc( 5*sizeof(predictor_t) ); - rc->pred_b_from_p = x264_malloc( sizeof(predictor_t) ); - for( i = 0; i < 5; i++ ) + rc->last_qscale = qp2qscale( 26 ); + int num_preds = h->param.b_sliced_threads * h->param.i_threads + 1; + CHECKED_MALLOC( rc->pred, 5 * sizeof(predictor_t) * num_preds ); + CHECKED_MALLOC( rc->pred_b_from_p, sizeof(predictor_t) ); + for( int i = 0; i < 3; i++ ) { rc->last_qscale_for[i] = qp2qscale( ABR_INIT_QP ); rc->lmin[i] = qp2qscale( h->param.rc.i_qp_min ); rc->lmax[i] = qp2qscale( h->param.rc.i_qp_max ); - 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; + for( int j = 0; j < num_preds; j++ ) + { + rc->pred[i+j*5].coeff= 2.0; + rc->pred[i+j*5].count= 1.0; + rc->pred[i+j*5].decay= 0.5; + rc->pred[i+j*5].offset= 0.0; + } + for( int 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]; @@ -408,35 +683,74 @@ int x264_ratecontrol_new( x264_t *h ) 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 ) ) { - int i; + int i, j; + uint32_t k, l; char *opts = stats_buf; stats_in = strchr( stats_buf, '\n' ); if( !stats_in ) return -1; *stats_in = '\0'; stats_in++; + if( sscanf( opts, "#options: %dx%d", &i, &j ) != 2 ) + { + 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; + } - if( ( p = strstr( opts, "bframes=" ) ) && sscanf( p, "bframes=%d", &i ) - && h->param.i_bframe != i ) + if( ( p = strstr( opts, "timebase=" ) ) && sscanf( p, "timebase=%u/%u", &k, &l ) != 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, "timebase specified in stats file not valid\n" ); + return -1; + } + if( k != h->param.i_timebase_num || l != h->param.i_timebase_den ) + { + x264_log( h, X264_LOG_ERROR, "timebase mismatch with 1st pass (%u/%u vs %u/%u)\n", + h->param.i_timebase_num, h->param.i_timebase_den, k, l ); 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" ); + CMP_OPT_FIRST_PASS( "weightp", 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( "open_gop", h->param.i_open_gop ); - 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 ); + if( (p = strstr( opts, "keyint=" )) ) + { + p += 7; + char buf[13] = "infinite "; + if( h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE ) + sprintf( buf, "%d ", h->param.i_keyint_max ); + if( strncmp( p, buf, strlen(buf) ) ) + { + x264_log( h, X264_LOG_ERROR, "different keyint setting than first pass (%.*s vs %.*s)\n", + strlen(buf)-1, buf, strcspn(p, " "), p ); + return -1; + } + } 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" ); @@ -455,28 +769,21 @@ int x264_ratecontrol_new( x264_t *h ) 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 || h->param.rc.i_vbv_buffer_size) && ( p = strstr( opts, "rc_lookahead=" ) ) && sscanf( p, "rc_lookahead=%d", &i ) ) + h->param.rc.i_lookahead = i; } /* find number of pics */ p = stats_in; - for(i=-1; p; i++) - p = strchr(p+1, ';'); - if(i==0) + int num_entries; + for( num_entries = -1; p; num_entries++ ) + p = strchr( p + 1, ';' ); + if( !num_entries ) { x264_log(h, X264_LOG_ERROR, "empty stats file\n"); return -1; } - rc->num_entries = i; + rc->num_entries = num_entries; if( h->param.i_frame_total < rc->num_entries && h->param.i_frame_total > 0 ) { @@ -490,22 +797,21 @@ int x264_ratecontrol_new( x264_t *h ) return -1; } - 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; inum_entries; i++) + for( int i = 0; i < rc->num_entries; i++ ) { ratecontrol_entry_t *rce = &rc->entry[i]; rce->pict_type = SLICE_TYPE_P; - rce->qscale = rce->new_qscale = qp2qscale(20); + rce->qscale = rce->new_qscale = qp2qscale( 20 ); rce->misc_bits = rc->nmb + 10; rce->new_qp = 0; } /* read stats */ p = stats_in; - for(i=0; i < rc->num_entries; i++) + for( int i = 0; i < rc->num_entries; i++ ) { ratecontrol_entry_t *rce; int frame_number; @@ -513,51 +819,89 @@ int x264_ratecontrol_new( x264_t *h ) int e; char *next; float qp; + int ref; next= strchr(p, ';'); - if(next) - { - (*next)=0; //sscanf is unbelievably slow on long strings - next++; - } - e = sscanf(p, " in:%d ", &frame_number); + if( next ) + *next++ = 0; //sscanf is unbelievably slow on long strings + e = sscanf( p, " in:%d ", &frame_number ); - if(frame_number < 0 || frame_number >= rc->num_entries) + if( frame_number < 0 || frame_number >= rc->num_entries ) { - x264_log(h, X264_LOG_ERROR, "bad frame number (%d) at stats line %d\n", frame_number, i); + x264_log( h, X264_LOG_ERROR, "bad frame number (%d) at stats line %d\n", frame_number, i ); return -1; } rce = &rc->entry[frame_number]; rce->direct_mode = 0; - e += sscanf(p, " in:%*d out:%*d type:%c q:%f tex:%d mv:%d misc:%d imb:%d pmb:%d smb:%d d:%c", - &pict_type, &qp, &rce->tex_bits, + e += sscanf( p, " in:%*d out:%*d type:%c dur:%d cpbdur:%d q:%f tex:%d mv:%d misc:%d imb:%d pmb:%d smb:%d d:%c", + &pict_type, &rce->i_duration, &rce->i_cpb_duration, &qp, &rce->tex_bits, &rce->mv_bits, &rce->misc_bits, &rce->i_count, &rce->p_count, - &rce->s_count, &rce->direct_mode); + &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) + if( e < 12 ) { - x264_log(h, X264_LOG_ERROR, "statistics are damaged at line %d, parser out=%d\n", i, e); +parse_error: + x264_log( h, X264_LOG_ERROR, "statistics are damaged at line %d, parser out=%d\n", i, e ); return -1; } - rce->qscale = qp2qscale(qp); + rce->qscale = qp2qscale( qp ); p = next; } - x264_free(stats_buf); + x264_free( stats_buf ); - if(h->param.rc.i_rc_method == X264_RC_ABR) + if( h->param.rc.i_rc_method == X264_RC_ABR ) { - if(init_pass2(h) < 0) return -1; + if( init_pass2( h ) < 0 ) + return -1; } /* else we're using constant quant, so no need to run the bitrate allocation */ } @@ -567,10 +911,9 @@ int x264_ratecontrol_new( x264_t *h ) 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 ) @@ -580,28 +923,53 @@ int x264_ratecontrol_new( x264_t *h ) } 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[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; iparam.i_threads; i++ ) + for( int i = 0; iparam.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 ) ); + h->thread[i]->param = h->param; 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, 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) ) @@ -618,8 +986,9 @@ static int parse_zone( x264_t *h, x264_zone_t *z, char *p ) p += len; if( !*p ) return 0; - z->param = x264_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, '=' ); @@ -636,35 +1005,37 @@ static int parse_zone( x264_t *h, x264_zone_t *z, char *p ) p = NULL; } return 0; +fail: + return -1; } static int parse_zones( x264_t *h ) { x264_ratecontrol_t *rc = h->rc; - int i; if( h->param.rc.psz_zones && !h->param.rc.i_zones ) { - char *p, *tok, UNUSED *saveptr; - char *psz_zones = x264_malloc( strlen(h->param.rc.psz_zones)+1 ); + 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; 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++ ) + for( int 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( h->param.rc.i_zones > 0 ) { - for( i = 0; i < h->param.rc.i_zones; i++ ) + for( int i = 0; i < h->param.rc.i_zones; i++ ) { x264_zone_t z = h->param.rc.zones[i]; if( z.i_start < 0 || z.i_start > z.i_end ) @@ -682,7 +1053,7 @@ static int parse_zones( x264_t *h ) } 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 @@ -690,9 +1061,9 @@ static int parse_zones( x264_t *h ) 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++ ) + for( int i = 1; i < rc->i_zones; i++ ) { if( !rc->zones[i].param ) rc->zones[i].param = rc->zones[0].param; @@ -700,12 +1071,13 @@ static int parse_zones( x264_t *h ) } return 0; +fail: + return -1; } static x264_zone_t *get_zone( x264_t *h, int frame_num ) { - int i; - for( i = h->rc->i_zones-1; i >= 0; i-- ) + for( int i = h->rc->i_zones - 1; i >= 0; i-- ) { x264_zone_t *z = &h->rc->zones[i]; if( frame_num >= z->i_start && frame_num <= z->i_end ) @@ -720,21 +1092,23 @@ void x264_ratecontrol_summary( x264_t *h ) 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 ); } } void x264_ratecontrol_delete( x264_t *h ) { 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", @@ -742,37 +1116,37 @@ void x264_ratecontrol_delete( x264_t *h ) } x264_free( rc->psz_stat_file_tmpname ); } + 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 && 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", + 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[0] ); + x264_free( rc->qp_buffer[1] ); if( rc->zones ) { x264_free( rc->zones[0].param ); - if( h->param.rc.psz_zones ) - for( i=1; ii_zones; i++ ) - if( rc->zones[i].param != rc->zones[0].param ) - x264_free( rc->zones[i].param ); + for( int 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 ); } -void x264_ratecontrol_set_estimated_size( x264_t *h, int bits ) -{ - x264_pthread_mutex_lock( &h->fenc->mutex ); - h->rc->frame_size_estimated = bits; - x264_pthread_mutex_unlock( &h->fenc->mutex ); -} - -int x264_ratecontrol_get_estimated_size( x264_t const *h) -{ - int size; - x264_pthread_mutex_lock( &h->fenc->mutex ); - size = h->rc->frame_size_estimated; - x264_pthread_mutex_unlock( &h->fenc->mutex ); - return size; -} - static void accum_p_qp_update( x264_t *h, float qp ) { x264_ratecontrol_t *rc = h->rc; @@ -786,7 +1160,7 @@ static void accum_p_qp_update( x264_t *h, float qp ) } /* 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; @@ -817,17 +1191,44 @@ void x264_ratecontrol_start( x264_t *h, int i_force_qp ) if( rc->b_vbv ) { - memset( h->fdec->i_row_bits, 0, h->sps->i_mb_height * sizeof(int) ); + memset( h->fdec->i_row_bits, 0, h->mb.i_mb_height * sizeof(int) ); rc->row_pred = &rc->row_preds[h->sh.i_type]; - update_vbv_plan( h ); + rc->buffer_rate = h->fenc->i_cpb_duration * rc->vbv_max_rate * h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale; + update_vbv_plan( h, overhead ); + + const x264_level_t *l = x264_levels; + while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc ) + l++; + + int mincr = l->mincr; + + /* Blu-ray requires this */ + if( l->level_idc == 41 && h->param.i_nal_hrd ) + mincr = 4; + + /* High 10 doesn't require minCR, so just set the maximum to a large value. */ + if( h->sps->i_profile_idc == PROFILE_HIGH10 ) + rc->frame_size_maximum = 1e9; + else + { + /* The spec has a bizarre special case for the first frame. */ + if( h->i_frame == 0 ) + { + //384 * ( Max( PicSizeInMbs, fR * MaxMBPS ) + MaxMBPS * ( tr( 0 ) - tr,n( 0 ) ) ) / MinCR + double fr = 1. / 172; + int pic_size_in_mbs = h->mb.i_mb_width * h->mb.i_mb_height; + rc->frame_size_maximum = 384 * BIT_DEPTH * X264_MAX( pic_size_in_mbs, fr*l->mbps ) / mincr; + } + else + { + //384 * MaxMBPS * ( tr( n ) - tr( n - 1 ) ) / MinCR + rc->frame_size_maximum = 384 * BIT_DEPTH * ((double)h->fenc->i_cpb_duration * h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale) * l->mbps / mincr; + } + } } 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 ) { @@ -854,66 +1255,69 @@ void x264_ratecontrol_start( x264_t *h, int i_force_qp ) if( zone->b_force_qp ) q += zone->i_qp - rc->qp_constant[SLICE_TYPE_P]; else - q -= 6*log(zone->f_bitrate_factor)/log(2); + q -= 6*log2f( zone->f_bitrate_factor ); } } + q = x264_clip3f( q, h->param.rc.i_qp_min, h->param.rc.i_qp_max ); + rc->qpa_rc = rc->qpa_aq = 0; + rc->qp = x264_clip3( (int)(q + 0.5), 0, QP_MAX ); h->fdec->f_qp_avg_rc = h->fdec->f_qp_avg_aq = - rc->qpm = - rc->qp = x264_clip3( (int)(q + 0.5), 0, 51 ); - rc->f_qpm = q; + rc->qpm = q; 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->qpm ); if( h->sh.i_type != SLICE_TYPE_B ) rc->last_non_b_pict_type = h->sh.i_type; } -static double predict_row_size( x264_t *h, int y, int qp ) +static double predict_row_size( x264_t *h, int y, double qp ) { /* 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]->f_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]->f_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( int i = h->i_threadslice_start; i <= y; i++ ) bits += h->fdec->i_row_bits[i]; return bits; } -static double predict_row_size_sum( x264_t *h, int y, int qp ) +static double predict_row_size_sum( x264_t *h, int y, double qp ) { - int i; double bits = row_bits_so_far(h, y); - for( i = y+1; i < h->sps->i_mb_height; i++ ) + for( int i = y+1; i < h->i_threadslice_end; i++ ) bits += predict_row_size( h, i, qp ); return bits; } @@ -927,93 +1331,113 @@ void x264_ratecontrol_mb( x264_t *h, int bits ) x264_emms(); h->fdec->i_row_bits[y] += bits; - rc->qpa_rc += rc->f_qpm; + rc->qpa_rc += rc->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->mb.i_mb_width - 1 || !rc->b_vbv ) return; - h->fdec->i_row_qp[y] = rc->qpm; + h->fdec->f_row_qp[y] = rc->qpm; + + 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]->f_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] ); - if( h->sh.i_type == SLICE_TYPE_B ) + /* 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 ) + float prev_row_qp = h->fdec->f_row_qp[y]; + float qp_min = X264_MAX( prev_row_qp - h->param.rc.i_qp_step, h->param.rc.i_qp_min ); + float qp_absolute_max = h->param.rc.i_qp_max; + if( rc->rate_factor_max_increment ) + qp_absolute_max = X264_MIN( qp_absolute_max, rc->qp_novbv + rc->rate_factor_max_increment ); + float qp_max = X264_MIN( prev_row_qp + h->param.rc.i_qp_step, qp_absolute_max ); + float step_size = 0.5; + + /* 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_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); + qp_min = X264_MAX( qp_min, X264_MAX( h->fref0[0]->f_row_qp[y+1], h->fref1[0]->f_row_qp[y+1] ) ); + rc->qpm = X264_MAX( rc->qpm, 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_slice_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 max_frame_error = X264_MAX( 0.05, 1.0 / (h->mb.i_mb_height) ); + float size_of_other_slices = 0; + if( h->param.b_sliced_threads ) { - 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; + float size_of_other_slices_planned = 0; + for( int i = 0; i < h->param.i_threads; i++ ) + if( h != h->thread[i] ) + { + size_of_other_slices += h->thread[i]->rc->frame_size_estimated; + size_of_other_slices_planned += h->thread[i]->rc->slice_size_planned; + } + float weight = rc->slice_size_planned / rc->frame_size_planned; + size_of_other_slices = (size_of_other_slices - size_of_other_slices_planned) * weight + size_of_other_slices_planned; + } - /* 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; + /* 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; + int b1 = predict_row_size_sum( h, y, rc->qpm ) + size_of_other_slices; - headroom = buffer_left_planned/rc->buffer_size; - if(h->sh.i_type != SLICE_TYPE_I) - headroom /= 2; - rc_tol += headroom; + /* 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 * slice_size_planned ) + return; - if( !rc->b_vbv_min_rate ) - i_qp_min = X264_MAX( i_qp_min, h->sh.i_qp ); + if( h->sh.i_type != SLICE_TYPE_I ) + rc_tol /= 2; - 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 ); - } + if( !rc->b_vbv_min_rate ) + qp_min = X264_MAX( qp_min, rc->qp_novbv ); - /* 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 < 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 += step_size; + b1 = predict_row_size_sum( h, y, rc->qpm ) + 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); + while( rc->qpm > qp_min + && (rc->qpm > h->fdec->f_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 -= step_size; + b1 = predict_row_size_sum( h, y, rc->qpm ) + size_of_other_slices; } + + /* avoid VBV underflow or MinCR violation */ + while( (rc->qpm < qp_absolute_max) + && ((rc->buffer_fill - b1 < rc->buffer_rate * max_frame_error) || + (rc->frame_size_maximum - b1 < rc->frame_size_maximum * max_frame_error))) + { + rc->qpm += step_size; + b1 = predict_row_size_sum( h, y, rc->qpm ) + size_of_other_slices; + } + + h->rc->frame_size_estimated = predict_row_size_sum( h, y, rc->qpm ); } - /* loses the fractional part of the frame-wise qp */ - rc->f_qpm = rc->qpm; } int x264_ratecontrol_qp( x264_t *h ) { - return h->rc->qpm; + x264_emms(); + return x264_clip3( h->rc->qpm + .5, h->param.rc.i_qp_min, h->param.rc.i_qp_max ); +} + +int x264_ratecontrol_mb_qp( x264_t *h ) +{ + x264_emms(); + float qp = h->rc->qpm; + if( h->param.rc.i_aq_mode ) + /* MB-tree currently doesn't adjust quantizers in unreferenced frames. */ + qp += 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]; + return x264_clip3( qp + .5, h->param.rc.i_qp_min, h->param.rc.i_qp_max ); } /* In 2pass, force the same frame types as in the 1st pass */ @@ -1027,65 +1451,58 @@ int x264_ratecontrol_slice_type( x264_t *h, int frame_num ) /* 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]; - 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 ); + 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, QP_MAX ); + 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, QP_MAX ); + 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, QP_MAX ); x264_log(h, X264_LOG_ERROR, "2nd pass has more frames than 1st pass (%d)\n", rc->num_entries); x264_log(h, X264_LOG_ERROR, "continuing anyway, at constant QP=%d\n", h->param.rc.i_qp_constant); if( h->param.i_bframe_adaptive ) x264_log(h, X264_LOG_ERROR, "disabling adaptive B-frames\n"); - for( i = 0; i < h->param.i_threads; i++ ) + for( int 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.b_pre_scenecut = 0; - h->thread[i]->param.i_scenecut_threshold = -1; + h->thread[i]->param.i_scenecut_threshold = 0; + h->thread[i]->param.rc.b_mb_tree = 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 ) - { - 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 */ -void x264_ratecontrol_end( x264_t *h, int bits ) +int x264_ratecontrol_end( x264_t *h, int bits, int *filler ) { x264_ratecontrol_t *rc = h->rc; const int *mbs = h->stat.frame.i_mb_count; - int i; x264_emms(); h->stat.frame.i_mb_count_skip = mbs[P_SKIP] + mbs[B_SKIP]; h->stat.frame.i_mb_count_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4]; h->stat.frame.i_mb_count_p = mbs[P_L0] + mbs[P_8x8]; - for( i = B_DIRECT; i < B_8x8; i++ ) + for( int i = B_DIRECT; i < B_8x8; i++ ) h->stat.frame.i_mb_count_p += mbs[i]; h->fdec->f_qp_avg_rc = rc->qpa_rc /= h->mb.i_mb_count; @@ -1102,61 +1519,135 @@ void x264_ratecontrol_end( x264_t *h, int bits ) ( 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 dur:%d cpbdur:%d 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, + c_type, h->fenc->i_duration, + h->fenc->i_cpb_duration, rc->qpa_rc, h->stat.frame.i_tex_bits, h->stat.frame.i_mv_bits, h->stat.frame.i_misc_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; + + /* Only write information for reference reordering once. */ + int use_old_stats = h->param.rc.b_stat_read && rc->rce->refs > 1; + for( int 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 ) + { + uint8_t i_type = h->sh.i_type; + /* Values are stored as big-endian FIX8.8 */ + for( int i = 0; i < h->mb.i_mb_count; i++ ) + 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[0], 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 ) { if( h->sh.i_type != SLICE_TYPE_B ) - rc->cplxr_sum += bits * qp2qscale(rc->qpa_rc) / rc->last_rceq; + rc->cplxr_sum += bits * qp2qscale( rc->qpa_rc ) / rc->last_rceq; else { /* Depends on the fact that B-frame's QP is an offset from the following P-frame's. * Not perfectly accurate with B-refs, but good enough. */ - rc->cplxr_sum += bits * qp2qscale(rc->qpa_rc) / (rc->last_rceq * fabs(h->param.rc.f_pb_factor)); + rc->cplxr_sum += bits * qp2qscale( rc->qpa_rc ) / (rc->last_rceq * fabs( h->param.rc.f_pb_factor )); } rc->cplxr_sum *= rc->cbr_decay; - rc->wanted_bits_window += rc->bitrate / rc->fps; - rc->wanted_bits_window *= rc->cbr_decay; + double frame_duration = (double)h->fenc->i_duration * h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale; - if( h->param.i_threads == 1 ) - accum_p_qp_update( h, rc->qpa_rc ); + rc->wanted_bits_window += frame_duration * rc->bitrate; + rc->wanted_bits_window *= rc->cbr_decay; } if( rc->b_2pass ) - { - rc->expected_bits_sum += qscale2bits( rc->rce, qp2qscale(rc->rce->new_qp) ); - } + rc->expected_bits_sum += qscale2bits( rc->rce, qp2qscale( rc->rce->new_qp ) ); if( h->mb.b_variable_qp ) { 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), + update_predictor( rc->pred_b_from_p, qp2qscale( rc->qpa_rc ), h->fref1[h->i_ref1-1]->i_satd, rc->bframe_bits / rc->bframes ); - /* In some cases, such as completely blank scenes, pred_b_from_p can go nuts */ - /* Hackily cap the predictor coeff in case this happens. */ - /* FIXME FIXME FIXME */ - rc->pred_b_from_p->coeff = X264_MIN( rc->pred_b_from_p->coeff, 10. ); rc->bframe_bits = 0; } } } - update_vbv( h, bits ); + *filler = update_vbv( h, bits ); + rc->filler_bits_sum += *filler * 8; + + if( h->sps->vui.b_nal_hrd_parameters_present ) + { + if( h->fenc->i_frame == 0 ) + { + // access unit initialises the HRD + h->fenc->hrd_timing.cpb_initial_arrival_time = 0; + rc->initial_cpb_removal_delay = h->initial_cpb_removal_delay; + rc->initial_cpb_removal_delay_offset = h->initial_cpb_removal_delay_offset; + h->fenc->hrd_timing.cpb_removal_time = rc->nrt_first_access_unit = (double)rc->initial_cpb_removal_delay / 90000; + } + else + { + h->fenc->hrd_timing.cpb_removal_time = rc->nrt_first_access_unit + (double)h->fenc->i_cpb_delay * + h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale; + + double cpb_earliest_arrival_time = h->fenc->hrd_timing.cpb_removal_time - (double)rc->initial_cpb_removal_delay / 90000; + if( h->fenc->b_keyframe ) + { + rc->nrt_first_access_unit = h->fenc->hrd_timing.cpb_removal_time; + rc->initial_cpb_removal_delay = h->initial_cpb_removal_delay; + rc->initial_cpb_removal_delay_offset = h->initial_cpb_removal_delay_offset; + } + else + cpb_earliest_arrival_time -= (double)rc->initial_cpb_removal_delay_offset / 90000; + + if( h->sps->vui.hrd.b_cbr_hrd ) + h->fenc->hrd_timing.cpb_initial_arrival_time = rc->previous_cpb_final_arrival_time; + else + h->fenc->hrd_timing.cpb_initial_arrival_time = X264_MAX( rc->previous_cpb_final_arrival_time, cpb_earliest_arrival_time ); + } + int filler_bits = *filler ? X264_MAX( (FILLER_OVERHEAD - h->param.b_annexb), *filler )*8 : 0; + // Equation C-6 + h->fenc->hrd_timing.cpb_final_arrival_time = rc->previous_cpb_final_arrival_time = h->fenc->hrd_timing.cpb_initial_arrival_time + + (double)(bits + filler_bits) / h->sps->vui.hrd.i_bit_rate_unscaled; + + h->fenc->hrd_timing.dpb_output_time = (double)h->fenc->i_dpb_output_delay * h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale + + h->fenc->hrd_timing.cpb_removal_time; + } + + return 0; +fail: + x264_log(h, X264_LOG_ERROR, "ratecontrol_end: stats file could not be written to\n"); + return -1; } /**************************************************************************** @@ -1169,14 +1660,12 @@ void x264_ratecontrol_end( x264_t *h, int bits ) static double get_qscale(x264_t *h, ratecontrol_entry_t *rce, double rate_factor, int frame_num) { x264_ratecontrol_t *rcc= h->rc; - double q; x264_zone_t *zone = get_zone( h, frame_num ); - - q = pow( rce->blurred_complexity, 1 - h->param.rc.f_qcompress ); + double 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) - q = rcc->last_qscale; + if( !isfinite(q) || rce->tex_bits + rce->mv_bits == 0 ) + q = rcc->last_qscale_for[rce->pict_type]; else { rcc->last_rceq = q; @@ -1187,7 +1676,7 @@ static double get_qscale(x264_t *h, ratecontrol_entry_t *rce, double rate_factor if( zone ) { if( zone->b_force_qp ) - q = qp2qscale(zone->i_qp); + q = qp2qscale( zone->i_qp ); else q /= zone->f_bitrate_factor; } @@ -1233,30 +1722,30 @@ static double get_diff_limited_q(x264_t *h, ratecontrol_entry_t *rce, double q) } /* last qscale / qdiff stuff */ - if(rcc->last_non_b_pict_type==pict_type - && (pict_type!=SLICE_TYPE_I || rcc->last_accum_p_norm < 1)) + if( rcc->last_non_b_pict_type == pict_type && + (pict_type!=SLICE_TYPE_I || rcc->last_accum_p_norm < 1) ) { double last_q = rcc->last_qscale_for[pict_type]; double max_qscale = last_q * rcc->lstep; double min_qscale = last_q / rcc->lstep; - if (q > max_qscale) q = max_qscale; - else if(q < min_qscale) q = min_qscale; + if ( q > max_qscale ) q = max_qscale; + else if( q < min_qscale ) q = min_qscale; } rcc->last_qscale_for[pict_type] = q; - if(pict_type!=SLICE_TYPE_B) + if( pict_type != SLICE_TYPE_B ) rcc->last_non_b_pict_type = pict_type; - if(pict_type==SLICE_TYPE_I) + if( pict_type == SLICE_TYPE_I ) { rcc->last_accum_p_norm = rcc->accum_p_norm; rcc->accum_p_norm = 0; rcc->accum_p_qp = 0; } - if(pict_type==SLICE_TYPE_P) + if( pict_type == SLICE_TYPE_P ) { float mask = 1 - pow( (float)rce->i_count / rcc->nmb, 2 ); - rcc->accum_p_qp = mask * (qscale2qp(q) + rcc->accum_p_qp); + rcc->accum_p_qp = mask * (qscale2qp( q ) + rcc->accum_p_qp); rcc->accum_p_norm = mask * (1 + rcc->accum_p_norm); } return q; @@ -1264,57 +1753,105 @@ static double get_diff_limited_q(x264_t *h, ratecontrol_entry_t *rce, double q) 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 -static void update_vbv( x264_t *h, int bits ) +static int update_vbv( x264_t *h, int bits ) { + int filler = 0; + int bitrate = h->sps->vui.hrd.i_bit_rate_unscaled; x264_ratecontrol_t *rcc = h->rc; x264_ratecontrol_t *rct = h->thread[0]->rc; + uint64_t buffer_size = (uint64_t)h->sps->vui.hrd.i_cpb_size_unscaled * h->sps->vui.i_time_scale; if( rcc->last_satd >= h->mb.i_mb_count ) - update_predictor( &rct->pred[h->sh.i_type], qp2qscale(rcc->qpa_rc), rcc->last_satd, bits ); + update_predictor( &rct->pred[h->sh.i_type], qp2qscale( rcc->qpa_rc ), rcc->last_satd, bits ); if( !rcc->b_vbv ) - return; + return filler; + + rct->buffer_fill_final -= (uint64_t)bits * h->sps->vui.i_time_scale; - rct->buffer_fill_final += rct->buffer_rate - 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, (double)rct->buffer_fill_final / h->sps->vui.i_time_scale ); + rct->buffer_fill_final = X264_MAX( rct->buffer_fill_final, 0 ); + rct->buffer_fill_final += (uint64_t)bitrate * h->sps->vui.i_num_units_in_tick * h->fenc->i_cpb_duration; + + if( h->sps->vui.hrd.b_cbr_hrd && rct->buffer_fill_final > buffer_size ) + { + filler = ceil( (rct->buffer_fill_final - buffer_size) / (8. * h->sps->vui.i_time_scale) ); + bits = X264_MAX( (FILLER_OVERHEAD - h->param.b_annexb), filler ) * 8; + rct->buffer_fill_final -= (uint64_t)bits * h->sps->vui.i_time_scale; + } + else + rct->buffer_fill_final = X264_MIN( rct->buffer_fill_final, buffer_size ); + + return filler; +} + +void x264_hrd_fullness( x264_t *h ) +{ + x264_ratecontrol_t *rct = h->thread[0]->rc; + uint64_t denom = (uint64_t)h->sps->vui.hrd.i_bit_rate_unscaled * h->sps->vui.i_time_scale / rct->hrd_multiply_denom; + uint64_t cpb_state = rct->buffer_fill_final; + uint64_t cpb_size = (uint64_t)h->sps->vui.hrd.i_cpb_size_unscaled * h->sps->vui.i_time_scale; + uint64_t multiply_factor = 180000 / rct->hrd_multiply_denom; + + if( rct->buffer_fill_final < 0 || rct->buffer_fill_final > cpb_size ) + { + x264_log( h, X264_LOG_WARNING, "CPB %s: %.0lf bits in a %.0lf-bit buffer\n", + rct->buffer_fill_final < 0 ? "underflow" : "overflow", (float)rct->buffer_fill_final/denom, (float)cpb_size/denom ); + } + + h->initial_cpb_removal_delay = (multiply_factor * cpb_state + denom) / (2*denom); + h->initial_cpb_removal_delay_offset = (multiply_factor * cpb_size + denom) / (2*denom) - h->initial_cpb_removal_delay; } // 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; - if( h->param.i_threads > 1 ) + rcc->buffer_fill = h->thread[0]->rc->buffer_fill_final / h->sps->vui.i_time_scale; + if( h->i_thread_frames > 1 ) { int j = h->rc - h->thread[0]->rc; - int i; - for( i=1; iparam.i_threads; i++ ) + for( int 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 ); + bits = X264_MAX(bits, t->rc->frame_size_estimated); + rcc->buffer_fill -= bits; + rcc->buffer_fill = X264_MAX( rcc->buffer_fill, 0 ); + rcc->buffer_fill += t->rc->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 @@ -1323,52 +1860,122 @@ static double clip_qscale( x264_t *h, int pict_type, double q ) x264_ratecontrol_t *rcc = h->rc; double lmin = rcc->lmin[pict_type]; double lmax = rcc->lmax[pict_type]; + if( rcc->rate_factor_max_increment ) + lmax = X264_MIN( lmax, qp2qscale( rcc->qp_novbv + rcc->rate_factor_max_increment ) ); 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. */ + /* 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 terminate = 0; + + /* Avoid an infinite loop. */ + for( int 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; + double target_fill; + double total_duration = 0; + 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( int j = 0; buffer_fill_cur >= 0 && buffer_fill_cur <= rcc->buffer_size; j++ ) + { + total_duration += h->fenc->f_planned_cpb_duration[j]; + buffer_fill_cur += rcc->vbv_max_rate * h->fenc->f_planned_cpb_duration[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 -= 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 + total_duration * rcc->vbv_max_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 - total_duration * rcc->vbv_max_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 ); + } + + /* Apply MinCR restrictions */ 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 ); + if( bits > rcc->frame_size_maximum ) + q *= bits / rcc->frame_size_maximum; + bits = predict_size( &rcc->pred[h->sh.i_type], q, rcc->last_satd ); /* 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 pbbits = bits; double bbits = predict_size( rcc->pred_b_from_p, q * h->param.rc.f_pb_factor, rcc->last_satd ); double space; + double bframe_cpb_duration = 0; + double minigop_cpb_duration; + for( int i = 0; i < nb; i++ ) + bframe_cpb_duration += h->fenc->f_planned_cpb_duration[1+i]; - if( bbits > rcc->buffer_rate ) + if( bbits * nb > bframe_cpb_duration * rcc->vbv_max_rate ) nb = 0; pbbits += nb * bbits; - space = rcc->buffer_fill + (1+nb)*rcc->buffer_rate - rcc->buffer_size; + minigop_cpb_duration = bframe_cpb_duration + h->fenc->f_planned_cpb_duration[0]; + space = rcc->buffer_fill + minigop_cpb_duration*rcc->vbv_max_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 ); } @@ -1377,19 +1984,19 @@ static double clip_qscale( x264_t *h, int pict_type, double q ) q = X264_MAX( q0, q ); } - if(lmin==lmax) + if( lmin==lmax ) return lmin; - else if(rcc->b_2pass) + else if( rcc->b_2pass ) { - double min2 = log(lmin); - double max2 = log(lmax); + double min2 = log( lmin ); + double max2 = log( lmax ); q = (log(q) - min2)/(max2-min2) - 0.5; - q = 1.0/(1.0 + exp(-4*q)); + q = 1.0/(1.0 + exp( -4*q )); q = q*(max2-min2) + min2; - return exp(q); + return exp( q ); } else - return x264_clip3f(q, lmin, lmax); + return x264_clip3f( q, lmin, lmax ); } // update qscale for 1 frame based on actual bits used so far @@ -1399,19 +2006,18 @@ static float rate_estimate_qscale( x264_t *h ) x264_ratecontrol_t *rcc = h->rc; ratecontrol_entry_t rce; 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]) + - rcc->filler_bits_sum; if( rcc->b_2pass ) { rce = *rcc->rce; - if(pict_type != rce.pict_type) + if( pict_type != rce.pict_type ) { - x264_log(h, X264_LOG_ERROR, "slice=%c but 2pass stats say %c\n", - slice_type_to_char[pict_type], slice_type_to_char[rce.pict_type]); + x264_log( h, X264_LOG_ERROR, "slice=%c but 2pass stats say %c\n", + slice_type_to_char[pict_type], slice_type_to_char[rce.pict_type] ); } } @@ -1432,24 +2038,31 @@ static float rate_estimate_qscale( x264_t *h ) if( h->fref1[0]->i_type == X264_TYPE_BREF ) q1 -= rcc->pb_offset/2; - if(i0 && i1) + if( i0 && i1 ) q = (q0 + q1) / 2 + rcc->ip_offset; - else if(i0) + else if( i0 ) q = q1; - else if(i1) + else if( i1 ) q = q0; else q = (q0*dt1 + q1*dt0) / (dt0 + dt1); - if(h->fenc->b_kept_as_ref) + if( h->fenc->b_kept_as_ref ) q += rcc->pb_offset/2; 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 ); - x264_ratecontrol_set_estimated_size(h, rcc->frame_size_planned); - rcc->last_satd = 0; - return qp2qscale(q); + 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 ); + h->rc->frame_size_estimated = rcc->frame_size_planned; + + /* For row SATDs */ + if( rcc->b_vbv ) + rcc->last_satd = x264_rc_analyse_slice( h ); + rcc->qp_novbv = q; + return qp2qscale( q ); } else { @@ -1457,57 +2070,66 @@ static float rate_estimate_qscale( x264_t *h ) if( rcc->b_2pass ) { - //FIXME adjust abr_buffer based on distance to the end of the video + double lmin = rcc->lmin[pict_type]; + double lmax = rcc->lmax[pict_type]; int64_t diff; int64_t predicted_bits = total_bits; 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; iparam.i_threads; i++ ) + for( int 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)); + bits = X264_MAX(bits, t->rc->frame_size_estimated); 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; + if( h->i_frame < h->i_thread_frames ) + predicted_bits += (int64_t)h->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; + } + + /* Adjust ABR buffer based on distance to the end of the video. */ + if( rcc->num_entries > h->i_frame ) + { + double final_bits = rcc->entry[rcc->num_entries-1].expected_bits; + double video_pos = rce.expected_bits / final_bits; + double scale_factor = sqrt( (1 - video_pos) * rcc->num_entries ); + abr_buffer *= 0.5 * X264_MAX( scale_factor, 0.5 ); } 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->i_frame + 1 - h->i_thread_frames) >= rcc->fps) && (rcc->expected_bits_sum > 0)) { /* Adjust quant based on the difference between * achieved and expected bitrate so far */ - double time = (double)h->fenc->i_frame / rcc->num_entries; - double w = x264_clip3f( time*100, 0.0, 1.0 ); + double cur_time = (double)h->i_frame / rcc->num_entries; + double w = x264_clip3f( cur_time*100, 0.0, 1.0 ); q *= pow( (double)total_bits / rcc->expected_bits_sum, w ); } if( rcc->b_vbv ) { /* Do not overflow vbv */ - double expected_size = qscale2bits(&rce, q); + 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 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 = 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)) || @@ -1517,7 +2139,7 @@ static float rate_estimate_qscale( x264_t *h ) 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 ); } @@ -1533,9 +2155,9 @@ static float rate_estimate_qscale( x264_t *h ) * tradeoff between quality and bitrate precision. But at large * tolerances, the bit distribution approaches that of 2pass. */ - double wanted_bits, overflow=1, lmin, lmax; + double wanted_bits, overflow = 1; - 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; @@ -1556,17 +2178,24 @@ static float rate_estimate_qscale( x264_t *h ) } else { - int i_frame_done = h->fenc->i_frame + 1 - h->param.i_threads; - q = get_qscale( h, &rce, rcc->wanted_bits_window / rcc->cplxr_sum, h->fenc->i_frame ); - // FIXME is it simpler to keep track of wanted_bits in ratecontrol_end? - wanted_bits = i_frame_done * rcc->bitrate / rcc->fps; - if( wanted_bits > 0 ) + /* ABR code can potentially be counterproductive in CBR, so just don't bother. + * Don't run it if the frame complexity is zero either. */ + if( !rcc->b_vbv_min_rate && rcc->last_satd ) { - abr_buffer *= X264_MAX( 1, sqrt(i_frame_done/25) ); - overflow = x264_clip3f( 1.0 + (total_bits - wanted_bits) / abr_buffer, .5, 2 ); - q *= overflow; + // FIXME is it simpler to keep track of wanted_bits in ratecontrol_end? + int i_frame_done = h->i_frame + 1 - h->i_thread_frames; + double time_done = i_frame_done / rcc->fps; + if( h->param.b_vfr_input && i_frame_done > 0 ) + time_done = ((double)(h->fenc->i_reordered_pts - h->i_reordered_pts_delay)) * h->param.i_timebase_num / h->param.i_timebase_den; + wanted_bits = time_done * rcc->bitrate; + if( wanted_bits > 0 ) + { + abr_buffer *= X264_MAX( 1, sqrt( time_done ) ); + overflow = x264_clip3f( 1.0 + (total_bits - wanted_bits) / abr_buffer, .5, 2 ); + q *= overflow; + } } } @@ -1581,8 +2210,8 @@ static float rate_estimate_qscale( x264_t *h ) { /* Asymmetric clipping, because symmetric would prevent * overflow control in areas of rapidly oscillating complexity */ - lmin = rcc->last_qscale_for[pict_type] / rcc->lstep; - lmax = rcc->last_qscale_for[pict_type] * rcc->lstep; + double lmin = rcc->last_qscale_for[pict_type] / rcc->lstep; + double lmax = rcc->last_qscale_for[pict_type] * rcc->lstep; if( overflow > 1.1 && h->i_frame > 3 ) lmax *= rcc->lstep; else if( overflow < 0.9 ) @@ -1590,10 +2219,11 @@ static float rate_estimate_qscale( x264_t *h ) 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->qp_novbv = qscale2qp( q ); //FIXME use get_diff_limited_q() ? q = clip_qscale( h, pict_type, q ); @@ -1603,17 +2233,106 @@ static float rate_estimate_qscale( x264_t *h ) 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) + 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 ); - x264_ratecontrol_set_estimated_size(h, rcc->frame_size_planned); + + /* Always use up the whole VBV in this case. */ + if( rcc->single_frame_vbv ) + rcc->frame_size_planned = rcc->buffer_rate; + h->rc->frame_size_estimated = rcc->frame_size_planned; return q; } } +void x264_threads_normalize_predictors( x264_t *h ) +{ + double totalsize = 0; + for( int i = 0; i < h->param.i_threads; i++ ) + totalsize += h->thread[i]->rc->slice_size_planned; + double factor = h->rc->frame_size_planned / totalsize; + for( int i = 0; i < h->param.i_threads; i++ ) + h->thread[i]->rc->slice_size_planned *= factor; +} + +void x264_threads_distribute_ratecontrol( x264_t *h ) +{ + int row; + x264_ratecontrol_t *rc = h->rc; + + /* Initialize row predictors */ + if( h->i_frame == 0 ) + for( int i = 0; i < h->param.i_threads; i++ ) + { + x264_ratecontrol_t *t = h->thread[i]->rc; + memcpy( t->row_preds, rc->row_preds, sizeof(rc->row_preds) ); + } + + for( int i = 0; i < h->param.i_threads; i++ ) + { + x264_t *t = h->thread[i]; + memcpy( t->rc, rc, offsetof(x264_ratecontrol_t, row_pred) ); + t->rc->row_pred = &t->rc->row_preds[h->sh.i_type]; + /* Calculate the planned slice size. */ + if( 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 = predict_size( &rc->pred[h->sh.i_type + (i+1)*5], rc->qpm, size ); + } + else + t->rc->slice_size_planned = 0; + } + if( rc->b_vbv && rc->frame_size_planned ) + { + x264_threads_normalize_predictors( h ); + + if( rc->single_frame_vbv ) + { + /* Compensate for our max frame error threshold: give more bits (proportionally) to smaller slices. */ + for( int i = 0; i < h->param.i_threads; i++ ) + { + x264_t *t = h->thread[i]; + float max_frame_error = X264_MAX( 0.05, 1.0 / (t->i_threadslice_end - t->i_threadslice_start) ); + t->rc->slice_size_planned += 2 * max_frame_error * rc->frame_size_planned; + } + x264_threads_normalize_predictors( h ); + } + + for( int i = 0; i < h->param.i_threads; i++ ) + h->thread[i]->rc->frame_size_estimated = h->thread[i]->rc->slice_size_planned; + } +} + +void x264_threads_merge_ratecontrol( x264_t *h ) +{ + x264_ratecontrol_t *rc = h->rc; + x264_emms(); + + for( int i = 0; i < h->param.i_threads; i++ ) + { + x264_t *t = h->thread[i]; + x264_ratecontrol_t *rct = h->thread[i]->rc; + if( h->param.rc.i_vbv_buffer_size ) + { + int size = 0; + for( int row = t->i_threadslice_start; row < t->i_threadslice_end; row++ ) + size += h->fdec->i_row_satd[row]; + int bits = t->stat.frame.i_mv_bits + t->stat.frame.i_tex_bits + t->stat.frame.i_misc_bits; + int mb_count = (t->i_threadslice_end - t->i_threadslice_start) * h->mb.i_mb_width; + update_predictor( &rc->pred[h->sh.i_type+(i+1)*5], qp2qscale( rct->qpa_rc/mb_count ), size, bits ); + } + if( !i ) + continue; + rc->qpa_rc += rct->qpa_rc; + rc->qpa_aq += rct->qpa_aq; + } +} + void x264_thread_sync_ratecontrol( x264_t *cur, x264_t *prev, x264_t *next ) { if( cur != prev ) @@ -1621,8 +2340,7 @@ void x264_thread_sync_ratecontrol( x264_t *cur, x264_t *prev, x264_t *next ) #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); @@ -1633,6 +2351,15 @@ void x264_thread_sync_ratecontrol( x264_t *cur, x264_t *prev, x264_t *next ) 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 ) @@ -1640,12 +2367,16 @@ void x264_thread_sync_ratecontrol( x264_t *cur, x264_t *prev, x264_t *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(filler_bits_sum); COPY(wanted_bits_window); COPY(bframe_bits); + COPY(initial_cpb_removal_delay); + COPY(initial_cpb_removal_delay_offset); + COPY(nrt_first_access_unit); + COPY(previous_cpb_final_arrival_time); #undef COPY } //FIXME row_preds[] (not strictly necessary, but would improve prediction) @@ -1662,40 +2393,40 @@ static int find_underflow( x264_t *h, double *fills, int *t0, int *t1, int over const double buffer_max = .9 * rcc->buffer_size; double fill = fills[*t0-1]; double parity = over ? 1. : -1.; - int i, start=-1, end=-1; - for(i = *t0; i < rcc->num_entries; i++) + int start = -1, end = -1; + for( int i = *t0; i < rcc->num_entries; i++ ) { - fill += (rcc->buffer_rate - qscale2bits(&rcc->entry[i], rcc->entry[i].new_qscale)) * parity; + fill += (rcc->entry[i].i_cpb_duration * rcc->vbv_max_rate * h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale - + qscale2bits( &rcc->entry[i], rcc->entry[i].new_qscale )) * parity; fill = x264_clip3f(fill, 0, rcc->buffer_size); fills[i] = fill; - if(fill <= buffer_min || i == 0) + if( fill <= buffer_min || i == 0 ) { - if(end >= 0) + if( end >= 0 ) break; start = i; } - else if(fill >= buffer_max && start >= 0) + else if( fill >= buffer_max && start >= 0 ) end = i; } *t0 = start; *t1 = end; - return start>=0 && end>=0; + return start >= 0 && end >= 0; } static int fix_underflow( x264_t *h, int t0, int t1, double adjustment, double qscale_min, double qscale_max) { x264_ratecontrol_t *rcc = h->rc; double qscale_orig, qscale_new; - int i; int adjusted = 0; - if(t0 > 0) + if( t0 > 0 ) t0++; - for(i = t0; i <= t1; i++) + for( int i = t0; i <= t1; i++ ) { qscale_orig = rcc->entry[i].new_qscale; - qscale_orig = x264_clip3f(qscale_orig, qscale_min, qscale_max); + qscale_orig = x264_clip3f( qscale_orig, qscale_min, qscale_max ); qscale_new = qscale_orig * adjustment; - qscale_new = x264_clip3f(qscale_new, qscale_min, qscale_max); + qscale_new = x264_clip3f( qscale_new, qscale_min, qscale_max ); rcc->entry[i].new_qscale = qscale_new; adjusted = adjusted || (qscale_new != qscale_orig); } @@ -1706,17 +2437,16 @@ static double count_expected_bits( x264_t *h ) { x264_ratecontrol_t *rcc = h->rc; double expected_bits = 0; - int i; - for(i = 0; i < rcc->num_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) { ratecontrol_entry_t *rce = &rcc->entry[i]; rce->expected_bits = expected_bits; - expected_bits += qscale2bits(rce, rce->new_qscale); + expected_bits += qscale2bits( rce, rce->new_qscale ); } return expected_bits; } -static void vbv_pass2( x264_t *h ) +static int vbv_pass2( x264_t *h, double all_available_bits ) { /* 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 @@ -1724,16 +2454,16 @@ static void vbv_pass2( x264_t *h ) * 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 all_available_bits = h->param.rc.i_bitrate * 1000. * rcc->num_entries / rcc->fps; + double *fills; double expected_bits = 0; double adjustment; double prev_bits = 0; - int i, t0, t1; - double qscale_min = qp2qscale(h->param.rc.i_qp_min); - double qscale_max = qp2qscale(h->param.rc.i_qp_max); + int t0, t1; + double qscale_min = qp2qscale( h->param.rc.i_qp_min ); + 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++; @@ -1743,16 +2473,16 @@ static void vbv_pass2( x264_t *h ) iterations++; prev_bits = expected_bits; - if(expected_bits != 0) + if( expected_bits ) { /* not first iteration */ adjustment = X264_MAX(X264_MIN(expected_bits / all_available_bits, 0.999), 0.9); fills[-1] = rcc->buffer_size * h->param.rc.f_vbv_buffer_init; t0 = 0; /* fix overflows */ adj_min = 1; - while(adj_min && find_underflow(h, fills, &t0, &t1, 1)) + while(adj_min && find_underflow( h, fills, &t0, &t1, 1 )) { - adj_min = fix_underflow(h, t0, t1, adjustment, qscale_min, qscale_max); + adj_min = fix_underflow( h, t0, t1, adjustment, qscale_min, qscale_max ); t0 = t1; } } @@ -1761,37 +2491,44 @@ static void vbv_pass2( x264_t *h ) t0 = 0; /* fix underflows -- should be done after overflow, as we'd better undersize target than underflowing VBV */ adj_max = 1; - while(adj_max && find_underflow(h, fills, &t0, &t1, 0)) - adj_max = fix_underflow(h, t0, t1, 1.001, qscale_min, qscale_max); + while( adj_max && find_underflow( h, fills, &t0, &t1, 0 ) ) + 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)) ); + expected_bits = count_expected_bits( h ); + } while( (expected_bits < .995*all_available_bits) && ((int64_t)(expected_bits+.5) > (int64_t)(prev_bits+.5)) ); - if (!adj_max) + if( !adj_max ) x264_log( h, X264_LOG_WARNING, "vbv-maxrate issue, qpmax or vbv-maxrate too low\n"); /* store expected vbv filling values for tracking when encoding */ - for(i = 0; i < rcc->num_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) rcc->entry[i].expected_vbv = rcc->buffer_size - fills[i]; - x264_free(fills-1); + x264_free( fills-1 ); + return 0; +fail: + return -1; } static int init_pass2( x264_t *h ) { x264_ratecontrol_t *rcc = h->rc; uint64_t all_const_bits = 0; - uint64_t all_available_bits = (uint64_t)(h->param.rc.i_bitrate * 1000. * rcc->num_entries / rcc->fps); - double rate_factor, step, step_mult; + double duration = 0; + for( int i = 0; i < rcc->num_entries; i++ ) + duration += rcc->entry[i].i_duration; + duration *= (double)h->sps->vui.i_num_units_in_tick / h->sps->vui.i_time_scale; + uint64_t all_available_bits = h->param.rc.i_bitrate * 1000. * duration; + double rate_factor, step_mult; double qblur = h->param.rc.f_qblur; double cplxblur = h->param.rc.f_complexity_blur; const int filter_size = (int)(qblur*4) | 1; double expected_bits; double *qscale, *blurred_qscale; - int i; + double base_cplx = h->mb.i_mb_count * (h->param.i_bframe ? 120 : 80); /* find total/average complexity & const_bits */ - for(i=0; inum_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) { ratecontrol_entry_t *rce = &rcc->entry[i]; all_const_bits += rce->misc_bits; @@ -1799,8 +2536,8 @@ static int init_pass2( x264_t *h ) if( all_available_bits < all_const_bits) { - x264_log(h, X264_LOG_ERROR, "requested bitrate is too low. estimated minimum is %d kbps\n", - (int)(all_const_bits * rcc->fps / (rcc->num_entries * 1000.))); + x264_log( h, X264_LOG_ERROR, "requested bitrate is too low. estimated minimum is %d kbps\n", + (int)(all_const_bits * rcc->fps / (rcc->num_entries * 1000.)) ); return -1; } @@ -1808,43 +2545,42 @@ static int init_pass2( x264_t *h ) * We don't blur the QPs directly, because then one very simple frame * could drag down the QP of a nearby complex frame and give it more * bits than intended. */ - for(i=0; inum_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) { ratecontrol_entry_t *rce = &rcc->entry[i]; double weight_sum = 0; double cplx_sum = 0; double weight = 1.0; double gaussian_weight; - int j; /* weighted average of cplx of future frames */ - for(j=1; jnum_entries-i; j++) + for( int j = 1; j < cplxblur*2 && j < rcc->num_entries-i; j++ ) { ratecontrol_entry_t *rcj = &rcc->entry[i+j]; weight *= 1 - pow( (float)rcj->i_count / rcc->nmb, 2 ); - if(weight < .0001) + if( weight < .0001 ) break; - gaussian_weight = weight * exp(-j*j/200.0); + gaussian_weight = weight * exp( -j*j/200.0 ); weight_sum += gaussian_weight; cplx_sum += gaussian_weight * (qscale2bits(rcj, 1) - rcj->misc_bits); } /* weighted average of cplx of past frames */ weight = 1.0; - for(j=0; j<=cplxblur*2 && j<=i; j++) + for( int j = 0; j <= cplxblur*2 && j <= i; j++ ) { ratecontrol_entry_t *rcj = &rcc->entry[i-j]; - gaussian_weight = weight * exp(-j*j/200.0); + gaussian_weight = weight * exp( -j*j/200.0 ); weight_sum += gaussian_weight; - cplx_sum += gaussian_weight * (qscale2bits(rcj, 1) - rcj->misc_bits); + cplx_sum += gaussian_weight * (qscale2bits( rcj, 1 ) - rcj->misc_bits); weight *= 1 - pow( (float)rcj->i_count / rcc->nmb, 2 ); - if(weight < .0001) + if( weight < .0001 ) break; } 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; @@ -1856,12 +2592,16 @@ static int init_pass2( x264_t *h ) * The search range is probably overkill, but speed doesn't matter here. */ expected_bits = 1; - for(i=0; inum_entries; i++) - expected_bits += qscale2bits(&rcc->entry[i], get_qscale(h, &rcc->entry[i], 1.0, i)); + for( int i = 0; i < rcc->num_entries; 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(step = 1E4 * step_mult; step > 1E-7 * step_mult; step *= 0.5) + for( double step = 1E4 * step_mult; step > 1E-7 * step_mult; step *= 0.5) { expected_bits = 0; rate_factor += step; @@ -1870,39 +2610,43 @@ static int init_pass2( x264_t *h ) rcc->last_accum_p_norm = 1; rcc->accum_p_norm = 0; + rcc->last_qscale_for[0] = + rcc->last_qscale_for[1] = + rcc->last_qscale_for[2] = pow( base_cplx, 1 - rcc->qcompress ) / rate_factor; + /* find qscale */ - for(i=0; inum_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) { - qscale[i] = get_qscale(h, &rcc->entry[i], rate_factor, 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 */ - for(i=rcc->num_entries-1; i>=0; i--) + for( int i = rcc->num_entries-1; i >= 0; i-- ) { - qscale[i] = get_diff_limited_q(h, &rcc->entry[i], qscale[i]); + qscale[i] = get_diff_limited_q( h, &rcc->entry[i], qscale[i] ); assert(qscale[i] >= 0); } /* smooth curve */ - if(filter_size > 1) + if( filter_size > 1 ) { - assert(filter_size%2==1); - for(i=0; inum_entries; i++) + assert( filter_size%2 == 1 ); + for( int i = 0; i < rcc->num_entries; i++ ) { ratecontrol_entry_t *rce = &rcc->entry[i]; - int j; - double q=0.0, sum=0.0; + double q = 0.0, sum = 0.0; - for(j=0; j= rcc->num_entries) + int idx = i+j-filter_size/2; + double d = idx-i; + double coeff = qblur==0 ? 1.0 : exp( -d*d/(qblur*qblur) ); + if( idx < 0 || idx >= rcc->num_entries ) continue; - if(rce->pict_type != rcc->entry[index].pict_type) + if( rce->pict_type != rcc->entry[idx].pict_type ) continue; - q += qscale[index] * coeff; + q += qscale[idx] * coeff; sum += coeff; } blurred_qscale[i] = q/sum; @@ -1910,57 +2654,59 @@ static int init_pass2( x264_t *h ) } /* find expected bits */ - for(i=0; inum_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) { ratecontrol_entry_t *rce = &rcc->entry[i]; - rce->new_qscale = clip_qscale(h, rce->pict_type, blurred_qscale[i]); + rce->new_qscale = clip_qscale( h, rce->pict_type, blurred_qscale[i] ); assert(rce->new_qscale >= 0); - expected_bits += qscale2bits(rce, rce->new_qscale); + expected_bits += qscale2bits( rce, rce->new_qscale ); } - if(expected_bits > all_available_bits) rate_factor -= step; + if( expected_bits > all_available_bits ) + rate_factor -= step; } - x264_free(qscale); - if(filter_size > 1) - x264_free(blurred_qscale); + x264_free( qscale ); + if( filter_size > 1 ) + x264_free( blurred_qscale ); - if(rcc->b_vbv) - vbv_pass2(h); - expected_bits = count_expected_bits(h); + if( rcc->b_vbv ) + if( vbv_pass2( h, all_available_bits ) ) + return -1; + expected_bits = count_expected_bits( h ); - if(fabs(expected_bits/all_available_bits - 1.0) > 0.01) + if( fabs( expected_bits/all_available_bits - 1.0 ) > 0.01 ) { double avgq = 0; - for(i=0; inum_entries; i++) + for( int i = 0; i < rcc->num_entries; i++ ) avgq += rcc->entry[i].new_qscale; - avgq = qscale2qp(avgq / rcc->num_entries); - - if ((expected_bits > all_available_bits) || (!rcc->b_vbv)) - x264_log(h, X264_LOG_WARNING, "Error: 2pass curve failed to converge\n"); - x264_log(h, X264_LOG_WARNING, "target: %.2f kbit/s, expected: %.2f kbit/s, avg QP: %.4f\n", - (float)h->param.rc.i_bitrate, - expected_bits * rcc->fps / (rcc->num_entries * 1000.), - avgq); - if(expected_bits < all_available_bits && avgq < h->param.rc.i_qp_min + 2) - { - if(h->param.rc.i_qp_min > 0) - x264_log(h, X264_LOG_WARNING, "try reducing target bitrate or reducing qp_min (currently %d)\n", h->param.rc.i_qp_min); + avgq = qscale2qp( avgq / rcc->num_entries ); + + if( expected_bits > all_available_bits || !rcc->b_vbv ) + x264_log( h, X264_LOG_WARNING, "Error: 2pass curve failed to converge\n" ); + x264_log( h, X264_LOG_WARNING, "target: %.2f kbit/s, expected: %.2f kbit/s, avg QP: %.4f\n", + (float)h->param.rc.i_bitrate, + expected_bits * rcc->fps / (rcc->num_entries * 1000.), + avgq ); + if( expected_bits < all_available_bits && avgq < h->param.rc.i_qp_min + 2 ) + { + if( h->param.rc.i_qp_min > 0 ) + x264_log( h, X264_LOG_WARNING, "try reducing target bitrate or reducing qp_min (currently %d)\n", h->param.rc.i_qp_min ); else - x264_log(h, X264_LOG_WARNING, "try reducing target bitrate\n"); + x264_log( h, X264_LOG_WARNING, "try reducing target bitrate\n" ); } - else if(expected_bits > all_available_bits && avgq > h->param.rc.i_qp_max - 2) + else if( expected_bits > all_available_bits && avgq > h->param.rc.i_qp_max - 2 ) { - if(h->param.rc.i_qp_max < 51) - x264_log(h, X264_LOG_WARNING, "try increasing target bitrate or increasing qp_max (currently %d)\n", h->param.rc.i_qp_max); + if( h->param.rc.i_qp_max < QP_MAX ) + x264_log( h, X264_LOG_WARNING, "try increasing target bitrate or increasing qp_max (currently %d)\n", h->param.rc.i_qp_max ); else - x264_log(h, X264_LOG_WARNING, "try increasing target bitrate\n"); + x264_log( h, X264_LOG_WARNING, "try increasing target bitrate\n"); } - else if(!(rcc->b_2pass && rcc->b_vbv)) - x264_log(h, X264_LOG_WARNING, "internal error\n"); + else if( !(rcc->b_2pass && rcc->b_vbv) ) + x264_log( h, X264_LOG_WARNING, "internal error\n" ); } return 0; +fail: + return -1; } - -