git.sesse.net Git - x264/blob - encoder/encoder.c

   1 /*****************************************************************************
   2  * x264: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *****************************************************************************/
  24
  25 #include <math.h>
  26
  27 #include "common/common.h"
  28 #include "common/cpu.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34 #include "me.h"
  35
  36 #ifdef HAVE_VISUALIZE
  37 #include "common/visualize.h"
  38 #endif
  39
  40 //#define DEBUG_MB_TYPE
  41
  42 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
  43
  44 #define bs_write_ue bs_write_ue_big
  45
  46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  47                                    x264_nal_t **pp_nal, int *pi_nal,
  48                                    x264_picture_t *pic_out );
  49
  50 /****************************************************************************
  51  *
  52  ******************************* x264 libs **********************************
  53  *
  54  ****************************************************************************/
  55 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  56 {
  57     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
  58     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  59         return 100;
  60
  61     return (float)(-10.0 * log( f_mse ) / log( 10.0 ));
  62 }
  63
  64 static void x264_frame_dump( x264_t *h )
  65 {
  66     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  67     int i, y;
  68     if( !f )
  69         return;
  70     /* Write the frame in display order */
  71     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
  72     for( i = 0; i < h->fdec->i_plane; i++ )
  73         for( y = 0; y < h->param.i_height >> !!i; y++ )
  74             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
  75     fclose( f );
  76 }
  77
  78
  79 /* Fill "default" values */
  80 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  81                                     x264_sps_t *sps, x264_pps_t *pps,
  82                                     int i_idr_pic_id, int i_frame, int i_qp )
  83 {
  84     x264_param_t *param = &h->param;
  85     int i;
  86
  87     /* First we fill all field */
  88     sh->sps = sps;
  89     sh->pps = pps;
  90
  91     sh->i_first_mb  = 0;
  92     sh->i_last_mb   = h->mb.i_mb_count - 1;
  93     sh->i_pps_id    = pps->i_id;
  94
  95     sh->i_frame_num = i_frame;
  96
  97     sh->b_mbaff = h->param.b_interlaced;
  98     sh->b_field_pic = 0;    /* no field support for now */
  99     sh->b_bottom_field = 0; /* not yet used */
 100
 101     sh->i_idr_pic_id = i_idr_pic_id;
 102
 103     /* poc stuff, fixed later */
 104     sh->i_poc_lsb = 0;
 105     sh->i_delta_poc_bottom = 0;
 106     sh->i_delta_poc[0] = 0;
 107     sh->i_delta_poc[1] = 0;
 108
 109     sh->i_redundant_pic_cnt = 0;
 110
 111     if( !h->mb.b_direct_auto_read )
 112     {
 113         if( h->mb.b_direct_auto_write )
 114             sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 115         else
 116             sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 117     }
 118     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 119
 120     sh->b_num_ref_idx_override = 0;
 121     sh->i_num_ref_idx_l0_active = 1;
 122     sh->i_num_ref_idx_l1_active = 1;
 123
 124     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 125     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 126
 127     /* If the ref list isn't in the default order, construct reordering header */
 128     /* List1 reordering isn't needed yet */
 129     if( sh->b_ref_pic_list_reordering_l0 )
 130     {
 131         int pred_frame_num = i_frame;
 132         for( i = 0; i < h->i_ref0; i++ )
 133         {
 134             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 135             if( diff == 0 )
 136                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
 137             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 138             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
 139             pred_frame_num = h->fref0[i]->i_frame_num;
 140         }
 141     }
 142
 143     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 144
 145     sh->i_qp = i_qp;
 146     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 147     sh->b_sp_for_swidth = 0;
 148     sh->i_qs_delta = 0;
 149
 150     /* If effective qp <= 15, deblocking would have no effect anyway */
 151     if( param->b_deblocking_filter
 152         && ( h->mb.b_variable_qp
 153         || 15 < i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
 154     {
 155         sh->i_disable_deblocking_filter_idc = 0;
 156     }
 157     else
 158     {
 159         sh->i_disable_deblocking_filter_idc = 1;
 160     }
 161     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 162     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 163 }
 164
 165 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 166 {
 167     int i;
 168
 169     if( sh->b_mbaff )
 170     {
 171         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 172         bs_write_ue( s, sh->i_first_mb >> 1 );
 173     }
 174     else
 175         bs_write_ue( s, sh->i_first_mb );
 176
 177     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 178     bs_write_ue( s, sh->i_pps_id );
 179     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 180
 181     if( !sh->sps->b_frame_mbs_only )
 182     {
 183         bs_write1( s, sh->b_field_pic );
 184         if( sh->b_field_pic )
 185             bs_write1( s, sh->b_bottom_field );
 186     }
 187
 188     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 189     {
 190         bs_write_ue( s, sh->i_idr_pic_id );
 191     }
 192
 193     if( sh->sps->i_poc_type == 0 )
 194     {
 195         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc_lsb & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 196         if( sh->pps->b_pic_order && !sh->b_field_pic )
 197         {
 198             bs_write_se( s, sh->i_delta_poc_bottom );
 199         }
 200     }
 201     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 202     {
 203         bs_write_se( s, sh->i_delta_poc[0] );
 204         if( sh->pps->b_pic_order && !sh->b_field_pic )
 205         {
 206             bs_write_se( s, sh->i_delta_poc[1] );
 207         }
 208     }
 209
 210     if( sh->pps->b_redundant_pic_cnt )
 211     {
 212         bs_write_ue( s, sh->i_redundant_pic_cnt );
 213     }
 214
 215     if( sh->i_type == SLICE_TYPE_B )
 216     {
 217         bs_write1( s, sh->b_direct_spatial_mv_pred );
 218     }
 219     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
 220     {
 221         bs_write1( s, sh->b_num_ref_idx_override );
 222         if( sh->b_num_ref_idx_override )
 223         {
 224             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 225             if( sh->i_type == SLICE_TYPE_B )
 226             {
 227                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 228             }
 229         }
 230     }
 231
 232     /* ref pic list reordering */
 233     if( sh->i_type != SLICE_TYPE_I )
 234     {
 235         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 236         if( sh->b_ref_pic_list_reordering_l0 )
 237         {
 238             for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 239             {
 240                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 241                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 242
 243             }
 244             bs_write_ue( s, 3 );
 245         }
 246     }
 247     if( sh->i_type == SLICE_TYPE_B )
 248     {
 249         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 250         if( sh->b_ref_pic_list_reordering_l1 )
 251         {
 252             for( i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 253             {
 254                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 255                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 256             }
 257             bs_write_ue( s, 3 );
 258         }
 259     }
 260
 261     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
 262     {
 263         /* pred_weight_table() */
 264         bs_write_ue( s, sh->weight[0][0].i_denom );
 265         bs_write_ue( s, sh->weight[0][1].i_denom );
 266         for( i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 267         {
 268             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 269             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 270             bs_write1( s, luma_weight_l0_flag );
 271             if( luma_weight_l0_flag )
 272             {
 273                 bs_write_se( s, sh->weight[i][0].i_scale );
 274                 bs_write_se( s, sh->weight[i][0].i_offset );
 275             }
 276             bs_write1( s, chroma_weight_l0_flag );
 277             if( chroma_weight_l0_flag )
 278             {
 279                 int j;
 280                 for( j = 1; j < 3; j++ )
 281                 {
 282                     bs_write_se( s, sh->weight[i][j].i_scale );
 283                     bs_write_se( s, sh->weight[i][j].i_offset );
 284                 }
 285             }
 286         }
 287     }
 288     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 289     {
 290       /* TODO */
 291     }
 292
 293     if( i_nal_ref_idc != 0 )
 294     {
 295         if( sh->i_idr_pic_id >= 0 )
 296         {
 297             bs_write1( s, 0 );  /* no output of prior pics flag */
 298             bs_write1( s, 0 );  /* long term reference flag */
 299         }
 300         else
 301         {
 302             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 303             if( sh->i_mmco_command_count > 0 )
 304             {
 305                 int i;
 306                 for( i = 0; i < sh->i_mmco_command_count; i++ )
 307                 {
 308                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 309                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 310                 }
 311                 bs_write_ue( s, 0 ); /* end command list */
 312             }
 313         }
 314     }
 315
 316     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 317     {
 318         bs_write_ue( s, sh->i_cabac_init_idc );
 319     }
 320     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 321
 322     if( sh->pps->b_deblocking_filter_control )
 323     {
 324         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 325         if( sh->i_disable_deblocking_filter_idc != 1 )
 326         {
 327             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 328             bs_write_se( s, sh->i_beta_offset >> 1 );
 329         }
 330     }
 331 }
 332
 333 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 334 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 335 static int x264_bitstream_check_buffer( x264_t *h )
 336 {
 337     uint8_t *bs_bak = h->out.p_bitstream;
 338     if( ( h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500) )
 339      || ( h->out.bs.p_end - h->out.bs.p < 2500 ) )
 340     {
 341         intptr_t delta;
 342         int i;
 343
 344         h->out.i_bitstream += 100000;
 345         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 346         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 347         delta = h->out.p_bitstream - bs_bak;
 348
 349         h->out.bs.p_start += delta;
 350         h->out.bs.p += delta;
 351         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 352
 353         h->cabac.p_start += delta;
 354         h->cabac.p += delta;
 355         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 356
 357         for( i = 0; i <= h->out.i_nal; i++ )
 358             h->out.nal[i].p_payload += delta;
 359         x264_free( bs_bak );
 360     }
 361     return 0;
 362 fail:
 363     x264_free( bs_bak );
 364     return -1;
 365 }
 366
 367 /****************************************************************************
 368  *
 369  ****************************************************************************
 370  ****************************** External API*********************************
 371  ****************************************************************************
 372  *
 373  ****************************************************************************/
 374
 375 static int x264_validate_parameters( x264_t *h )
 376 {
 377 #ifdef HAVE_MMX
 378     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 379     {
 380         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 381         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 382         return -1;
 383     }
 384 #endif
 385     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 386     {
 387         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 388                   h->param.i_width, h->param.i_height );
 389         return -1;
 390     }
 391
 392     if( h->param.i_width % 2 || h->param.i_height % 2 )
 393     {
 394         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 395                   h->param.i_width, h->param.i_height );
 396         return -1;
 397     }
 398     int i_csp = h->param.i_csp & X264_CSP_MASK;
 399     if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
 400     {
 401         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
 402         return -1;
 403     }
 404
 405     if( h->param.i_threads == X264_THREADS_AUTO )
 406         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 407     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 408     if( h->param.i_threads > 1 )
 409     {
 410 #ifndef HAVE_PTHREAD
 411         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 412         h->param.i_threads = 1;
 413 #endif
 414         /* Avoid absurdly small thread slices as they can reduce performance
 415          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 416         if( h->param.b_sliced_threads )
 417         {
 418             int max_threads = (h->param.i_height+15)/16 / 4;
 419             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 420         }
 421     }
 422     else
 423         h->param.b_sliced_threads = 0;
 424     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 425
 426     if( h->param.b_interlaced )
 427     {
 428         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 429         {
 430             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 431             h->param.analyse.i_me_method = X264_ME_UMH;
 432         }
 433         if( h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 434         {
 435             x264_log( h, X264_LOG_WARNING, "interlace + direct=temporal is not implemented\n" );
 436             h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 437         }
 438         if( h->param.analyse.i_weighted_pred > 0 )
 439         {
 440             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 441             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 442         }
 443     }
 444
 445     /* Detect default ffmpeg settings and terminate with an error. */
 446     {
 447         int score = 0;
 448         score += h->param.analyse.i_me_range == 0;
 449         score += h->param.rc.i_qp_step == 3;
 450         score += h->param.i_keyint_max == 12;
 451         score += h->param.rc.i_qp_min == 2;
 452         score += h->param.rc.i_qp_max == 31;
 453         score += h->param.rc.f_qcompress == 0.5;
 454         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 455         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 456         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 457         if( score >= 5 )
 458         {
 459             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 460             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 461             return -1;
 462         }
 463     }
 464
 465     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 466     {
 467         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 468         return -1;
 469     }
 470     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 471     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 472     if( h->param.rc.i_rc_method == X264_RC_CRF )
 473     {
 474         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 475         h->param.rc.i_bitrate = 0;
 476     }
 477     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 478         && h->param.rc.i_qp_constant == 0 )
 479     {
 480         h->mb.b_lossless = 1;
 481         h->param.i_cqm_preset = X264_CQM_FLAT;
 482         h->param.psz_cqm_file = NULL;
 483         h->param.rc.i_rc_method = X264_RC_CQP;
 484         h->param.rc.f_ip_factor = 1;
 485         h->param.rc.f_pb_factor = 1;
 486         h->param.analyse.b_psnr = 0;
 487         h->param.analyse.b_ssim = 0;
 488         h->param.analyse.i_chroma_qp_offset = 0;
 489         h->param.analyse.i_trellis = 0;
 490         h->param.analyse.b_fast_pskip = 0;
 491         h->param.analyse.i_noise_reduction = 0;
 492         h->param.analyse.f_psy_rd = 0;
 493         h->param.i_bframe = 0;
 494         /* 8x8dct is not useful at all in CAVLC lossless */
 495         if( !h->param.b_cabac )
 496             h->param.analyse.b_transform_8x8 = 0;
 497     }
 498     if( h->param.rc.i_rc_method == X264_RC_CQP )
 499     {
 500         float qp_p = h->param.rc.i_qp_constant;
 501         float qp_i = qp_p - 6*log(h->param.rc.f_ip_factor)/log(2);
 502         float qp_b = qp_p + 6*log(h->param.rc.f_pb_factor)/log(2);
 503         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 504         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 505         h->param.rc.i_aq_mode = 0;
 506         h->param.rc.b_mb_tree = 0;
 507     }
 508     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 509     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 510     if( h->param.rc.i_vbv_buffer_size )
 511     {
 512         if( h->param.rc.i_rc_method == X264_RC_CQP )
 513         {
 514             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 515             h->param.rc.i_vbv_max_bitrate = 0;
 516             h->param.rc.i_vbv_buffer_size = 0;
 517         }
 518         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 519         {
 520             if( h->param.rc.i_rc_method == X264_RC_ABR )
 521             {
 522                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 523                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 524             }
 525             else
 526             {
 527                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 528                 h->param.rc.i_vbv_buffer_size = 0;
 529             }
 530         }
 531         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 532                  h->param.rc.i_rc_method == X264_RC_ABR )
 533         {
 534             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 535             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 536         }
 537     }
 538     else if( h->param.rc.i_vbv_max_bitrate )
 539     {
 540         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 541         h->param.rc.i_vbv_max_bitrate = 0;
 542     }
 543
 544     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 545     if( h->param.b_sliced_threads )
 546         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 547     else
 548     {
 549         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 550         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 551         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 552         if( h->param.b_interlaced && h->param.i_slice_max_size )
 553         {
 554             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 555             h->param.i_slice_max_size = 0;
 556         }
 557         if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 558         {
 559             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 560             h->param.i_slice_max_mbs = 0;
 561         }
 562         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 563             h->param.i_slice_count = 0;
 564     }
 565
 566     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 567     if( h->param.i_keyint_max <= 0 )
 568         h->param.i_keyint_max = 1;
 569     if( h->param.i_scenecut_threshold < 0 )
 570         h->param.i_scenecut_threshold = 0;
 571     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 572     {
 573         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 574         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 575     }
 576     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 577     if( h->param.i_keyint_max == 1 )
 578     {
 579         h->param.i_bframe = 0;
 580         h->param.b_intra_refresh = 0;
 581     }
 582     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 583     if( h->param.i_bframe <= 1 )
 584         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 585     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 586     if( !h->param.i_bframe )
 587     {
 588         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 589         h->param.analyse.i_direct_mv_pred = 0;
 590         h->param.analyse.b_weighted_bipred = 0;
 591     }
 592     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 593     {
 594         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 595         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 596     }
 597     if( h->param.b_intra_refresh && h->param.i_frame_reference > 1 )
 598     {
 599         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 600         h->param.i_frame_reference = 1;
 601     }
 602     if( h->param.b_intra_refresh )
 603         h->param.i_keyint_max = X264_MIN( h->param.i_keyint_max, (h->param.i_width+15)/16 - 1 );
 604     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 605     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 606     {
 607         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 608         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 609         float fps = h->param.i_fps_num > 0 && h->param.i_fps_den > 0 ? (float) h->param.i_fps_num / h->param.i_fps_den : 25.0;
 610         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 611     }
 612
 613     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
 614     {
 615         h->param.i_timebase_num = h->param.i_fps_den;
 616         h->param.i_timebase_den = h->param.i_fps_num;
 617     }
 618
 619     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 620     if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 621         h->param.rc.b_mb_tree = 0;
 622     if( h->param.rc.b_stat_read )
 623         h->param.rc.i_lookahead = 0;
 624 #ifdef HAVE_PTHREAD
 625     if( h->param.i_sync_lookahead )
 626         h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->i_thread_frames + h->param.i_bframe, X264_LOOKAHEAD_MAX );
 627     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 628         h->param.i_sync_lookahead = 0;
 629 #else
 630     h->param.i_sync_lookahead = 0;
 631 #endif
 632
 633     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 634                                 && h->param.i_bframe
 635                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 636
 637     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 638     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 639     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 640     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 641
 642     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 643
 644     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 645         h->param.i_cqm_preset = X264_CQM_FLAT;
 646
 647     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 648         h->param.analyse.i_me_method > X264_ME_TESA )
 649         h->param.analyse.i_me_method = X264_ME_HEX;
 650     if( h->param.analyse.i_me_range < 4 )
 651         h->param.analyse.i_me_range = 4;
 652     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 653         h->param.analyse.i_me_range = 16;
 654     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 655         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 656         h->param.analyse.i_me_method = X264_ME_ESA;
 657     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 658     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 659     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 660                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 661     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 662     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 663         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 664     if( !h->param.analyse.b_transform_8x8 )
 665     {
 666         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 667         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 668     }
 669     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 670     if( !h->param.b_cabac )
 671         h->param.analyse.i_trellis = 0;
 672     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 673     if( !h->param.analyse.b_psy )
 674     {
 675         h->param.analyse.f_psy_rd = 0;
 676         h->param.analyse.f_psy_trellis = 0;
 677     }
 678     if( !h->param.analyse.i_trellis )
 679         h->param.analyse.f_psy_trellis = 0;
 680     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 681     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 682     if( h->param.analyse.i_subpel_refine < 6 )
 683         h->param.analyse.f_psy_rd = 0;
 684     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 685     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 686     /* so we lower the chroma QP offset to compensate */
 687     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 688      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 689     if( h->mb.i_psy_rd )
 690         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 691     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 692     /* Psy trellis has a similar effect. */
 693     if( h->mb.i_psy_trellis )
 694         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 695     else
 696         h->mb.i_psy_trellis = 0;
 697     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 698     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 699     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 700     if( h->param.rc.f_aq_strength == 0 )
 701         h->param.rc.i_aq_mode = 0;
 702     /* MB-tree requires AQ to be on, even if the strength is zero. */
 703     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 704     {
 705         h->param.rc.i_aq_mode = 1;
 706         h->param.rc.f_aq_strength = 0;
 707     }
 708     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 709     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 710         h->param.analyse.i_subpel_refine = 9;
 711
 712     {
 713         const x264_level_t *l = x264_levels;
 714         if( h->param.i_level_idc < 0 )
 715         {
 716             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 717             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 718                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 719             h->sps = h->sps_array;
 720             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 721             do h->param.i_level_idc = l->level_idc;
 722                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 723             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 724         }
 725         else
 726         {
 727             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 728                 l++;
 729             if( l->level_idc == 0 )
 730             {
 731                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 732                 return -1;
 733             }
 734         }
 735         if( h->param.analyse.i_mv_range <= 0 )
 736             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 737         else
 738             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 739     }
 740
 741     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 742     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 743         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 744
 745     if( h->i_thread_frames > 1 )
 746     {
 747         int r = h->param.analyse.i_mv_range_thread;
 748         int r2;
 749         if( r <= 0 )
 750         {
 751             // half of the available space is reserved and divided evenly among the threads,
 752             // the rest is allocated to whichever thread is far enough ahead to use it.
 753             // reserving more space increases quality for some videos, but costs more time
 754             // in thread synchronization.
 755             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 756             r = max_range / 2;
 757         }
 758         r = X264_MAX( r, h->param.analyse.i_me_range );
 759         r = X264_MIN( r, h->param.analyse.i_mv_range );
 760         // round up to use the whole mb row
 761         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 762         if( r2 < r )
 763             r2 += 16;
 764         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 765         h->param.analyse.i_mv_range_thread = r2;
 766     }
 767
 768     if( h->param.rc.f_qblur < 0 )
 769         h->param.rc.f_qblur = 0;
 770     if( h->param.rc.f_complexity_blur < 0 )
 771         h->param.rc.f_complexity_blur = 0;
 772
 773     h->param.i_sps_id &= 31;
 774
 775     if( h->param.i_log_level < X264_LOG_INFO )
 776     {
 777         h->param.analyse.b_psnr = 0;
 778         h->param.analyse.b_ssim = 0;
 779     }
 780
 781     /* ensure the booleans are 0 or 1 so they can be used in math */
 782 #define BOOLIFY(x) h->param.x = !!h->param.x
 783     BOOLIFY( b_cabac );
 784     BOOLIFY( b_constrained_intra );
 785     BOOLIFY( b_deblocking_filter );
 786     BOOLIFY( b_deterministic );
 787     BOOLIFY( b_sliced_threads );
 788     BOOLIFY( b_interlaced );
 789     BOOLIFY( b_intra_refresh );
 790     BOOLIFY( b_visualize );
 791     BOOLIFY( b_aud );
 792     BOOLIFY( b_repeat_headers );
 793     BOOLIFY( b_annexb );
 794     BOOLIFY( analyse.b_transform_8x8 );
 795     BOOLIFY( analyse.b_weighted_bipred );
 796     BOOLIFY( analyse.b_chroma_me );
 797     BOOLIFY( analyse.b_mixed_references );
 798     BOOLIFY( analyse.b_fast_pskip );
 799     BOOLIFY( analyse.b_dct_decimate );
 800     BOOLIFY( analyse.b_psy );
 801     BOOLIFY( analyse.b_psnr );
 802     BOOLIFY( analyse.b_ssim );
 803     BOOLIFY( rc.b_stat_write );
 804     BOOLIFY( rc.b_stat_read );
 805     BOOLIFY( rc.b_mb_tree );
 806 #undef BOOLIFY
 807
 808     return 0;
 809 }
 810
 811 static void mbcmp_init( x264_t *h )
 812 {
 813     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 814     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 815     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 816     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 817     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 818     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 819     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 820     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 821     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 822     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 823 }
 824
 825 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 826 {
 827     /* VUI */
 828     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 829     {
 830         int i_w = param->vui.i_sar_width;
 831         int i_h = param->vui.i_sar_height;
 832         int old_w = h->param.vui.i_sar_width;
 833         int old_h = h->param.vui.i_sar_height;
 834
 835         x264_reduce_fraction( &i_w, &i_h );
 836
 837         while( i_w > 65535 || i_h > 65535 )
 838         {
 839             i_w /= 2;
 840             i_h /= 2;
 841         }
 842
 843         x264_reduce_fraction( &i_w, &i_h );
 844
 845         if( i_w != old_w || i_h != old_h || initial )
 846         {
 847             h->param.vui.i_sar_width = 0;
 848             h->param.vui.i_sar_height = 0;
 849             if( i_w == 0 || i_h == 0 )
 850                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 851             else
 852             {
 853                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 854                 h->param.vui.i_sar_width = i_w;
 855                 h->param.vui.i_sar_height = i_h;
 856             }
 857         }
 858     }
 859 }
 860
 861 /****************************************************************************
 862  * x264_encoder_open:
 863  ****************************************************************************/
 864 x264_t *x264_encoder_open( x264_param_t *param )
 865 {
 866     x264_t *h;
 867     char buf[1000], *p;
 868     int i, qp, i_slicetype_length;
 869
 870     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 871
 872     /* Create a copy of param */
 873     memcpy( &h->param, param, sizeof(x264_param_t) );
 874
 875     if( param->param_free )
 876         param->param_free( param );
 877
 878     if( x264_validate_parameters( h ) < 0 )
 879         goto fail;
 880
 881     if( h->param.psz_cqm_file )
 882         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 883             goto fail;
 884
 885     if( h->param.rc.psz_stat_out )
 886         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 887     if( h->param.rc.psz_stat_in )
 888         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 889
 890     x264_set_aspect_ratio( h, &h->param, 1 );
 891
 892     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 893     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 894
 895     /* Init x264_t */
 896     h->i_frame = -1;
 897     h->i_frame_num = 0;
 898     h->i_idr_pic_id = 0;
 899     if( h->param.b_dts_compress )
 900     {
 901         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
 902         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
 903         if( h->i_dts_compress_multiplier != 1 )
 904             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %d/%d -> %d/%d\n",
 905                       h->param.i_timebase_num, h->param.i_timebase_den,
 906                       h->param.i_timebase_num, h->param.i_timebase_den * h->i_dts_compress_multiplier );
 907         h->param.i_timebase_den *= h->i_dts_compress_multiplier;
 908     }
 909     else
 910         h->i_dts_compress_multiplier = 1;
 911
 912     h->sps = &h->sps_array[0];
 913     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 914
 915     h->pps = &h->pps_array[0];
 916     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 917
 918     x264_validate_levels( h, 1 );
 919
 920     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 921
 922     if( x264_cqm_init( h ) < 0 )
 923         goto fail;
 924
 925     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 926
 927     /* Init frames. */
 928     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 929         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 930     else
 931         h->frames.i_delay = h->param.i_bframe;
 932     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 933         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 934     i_slicetype_length = h->frames.i_delay;
 935     h->frames.i_delay += h->i_thread_frames - 1;
 936     h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
 937     h->frames.i_delay += h->param.i_sync_lookahead;
 938     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 939
 940     h->frames.i_max_ref0 = h->param.i_frame_reference;
 941     h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
 942     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 943     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 944         && ( h->param.rc.i_rc_method == X264_RC_ABR
 945           || h->param.rc.i_rc_method == X264_RC_CRF
 946           || h->param.i_bframe_adaptive
 947           || h->param.i_scenecut_threshold
 948           || h->param.rc.b_mb_tree
 949           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 950     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 951     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 952
 953     h->frames.i_last_keyframe = - h->param.i_keyint_max;
 954     h->frames.i_input    = 0;
 955
 956     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
 957     /* Allocate room for max refs plus a few extra just in case. */
 958     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
 959     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
 960                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
 961     if( h->param.analyse.i_weighted_pred > 0 )
 962         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
 963     h->i_ref0 = 0;
 964     h->i_ref1 = 0;
 965
 966     x264_rdo_init();
 967
 968     /* init CPU functions */
 969     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 970     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 971     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 972     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 973     if( !h->param.b_cabac )
 974         x264_init_vlc_tables();
 975     x264_pixel_init( h->param.cpu, &h->pixf );
 976     x264_dct_init( h->param.cpu, &h->dctf );
 977     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 978     x264_mc_init( h->param.cpu, &h->mc );
 979     x264_quant_init( h, h->param.cpu, &h->quantf );
 980     x264_deblock_init( h->param.cpu, &h->loopf );
 981     x264_dct_init_weights();
 982
 983     mbcmp_init( h );
 984
 985     p = buf + sprintf( buf, "using cpu capabilities:" );
 986     for( i=0; x264_cpu_names[i].flags; i++ )
 987     {
 988         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 989             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 990             continue;
 991         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 992             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
 993             continue;
 994         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 995             && (h->param.cpu & X264_CPU_SSE42) )
 996             continue;
 997         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 998             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 999             p += sprintf( p, " %s", x264_cpu_names[i].name );
1000     }
1001     if( !h->param.cpu )
1002         p += sprintf( p, " none!" );
1003     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1004
1005     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1006         if( x264_analyse_init_costs( h, qp ) )
1007             goto fail;
1008     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1009         goto fail;
1010     if( h->cost_mv[1][2013] != 24 )
1011     {
1012         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1013         goto fail;
1014     }
1015
1016     h->out.i_nal = 0;
1017     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1018         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1019           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1020
1021     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1022     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1023
1024     h->thread[0] = h;
1025     for( i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1026         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1027
1028     if( x264_lookahead_init( h, i_slicetype_length ) )
1029         goto fail;
1030
1031     for( i = 0; i < h->param.i_threads; i++ )
1032     {
1033         int init_nal_count = h->param.i_slice_count + 3;
1034         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1035         if( i > 0 )
1036             *h->thread[i] = *h;
1037
1038         if( allocate_threadlocal_data )
1039         {
1040             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1041             if( !h->thread[i]->fdec )
1042                 goto fail;
1043         }
1044         else
1045             h->thread[i]->fdec = h->thread[0]->fdec;
1046
1047         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1048         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1049         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1050         h->thread[i]->out.i_nals_allocated = init_nal_count;
1051
1052         if( allocate_threadlocal_data && x264_macroblock_cache_init( h->thread[i] ) < 0 )
1053             goto fail;
1054     }
1055
1056     /* Allocate scratch buffer */
1057     for( i = 0; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1058     {
1059         int buf_hpel = (h->param.i_width+48) * sizeof(int16_t);
1060         int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
1061         int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
1062         int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
1063             ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
1064         int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
1065         int buf_nnz = !h->param.b_cabac * h->pps->b_transform_8x8_mode * (h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t));
1066         int scratch_size = X264_MAX4( buf_hpel, buf_ssim, buf_tesa, X264_MAX( buf_mbtree, buf_nnz ) );
1067         CHECKED_MALLOC( h->thread[i]->scratch_buffer, scratch_size );
1068     }
1069
1070     if( x264_ratecontrol_new( h ) < 0 )
1071         goto fail;
1072
1073     if( h->param.psz_dump_yuv )
1074     {
1075         /* create or truncate the reconstructed video file */
1076         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1077         if( !f )
1078         {
1079             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1080             goto fail;
1081         }
1082         else if( !x264_is_regular_file( f ) )
1083         {
1084             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1085             goto fail;
1086         }
1087         fclose( f );
1088     }
1089
1090     x264_log( h, X264_LOG_INFO, "profile %s, level %d.%d\n",
1091         h->sps->i_profile_idc == PROFILE_BASELINE ? "Baseline" :
1092         h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1093         h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1094         "High 4:4:4 Predictive", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1095
1096     return h;
1097 fail:
1098     x264_free( h );
1099     return NULL;
1100 }
1101
1102 /****************************************************************************
1103  * x264_encoder_reconfig:
1104  ****************************************************************************/
1105 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1106 {
1107     h = h->thread[h->thread[0]->i_thread_phase];
1108     x264_set_aspect_ratio( h, param, 0 );
1109 #define COPY(var) h->param.var = param->var
1110     COPY( i_frame_reference ); // but never uses more refs than initially specified
1111     COPY( i_bframe_bias );
1112     if( h->param.i_scenecut_threshold )
1113         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1114     COPY( b_deblocking_filter );
1115     COPY( i_deblocking_filter_alphac0 );
1116     COPY( i_deblocking_filter_beta );
1117     COPY( analyse.inter );
1118     COPY( analyse.intra );
1119     COPY( analyse.i_direct_mv_pred );
1120     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1121     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1122         COPY( analyse.i_me_range );
1123     COPY( analyse.i_noise_reduction );
1124     /* We can't switch out of subme=0 during encoding. */
1125     if( h->param.analyse.i_subpel_refine )
1126         COPY( analyse.i_subpel_refine );
1127     COPY( analyse.i_trellis );
1128     COPY( analyse.b_chroma_me );
1129     COPY( analyse.b_dct_decimate );
1130     COPY( analyse.b_fast_pskip );
1131     COPY( analyse.b_mixed_references );
1132     COPY( analyse.f_psy_rd );
1133     COPY( analyse.f_psy_trellis );
1134     // can only twiddle these if they were enabled to begin with:
1135     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1136         COPY( analyse.i_me_method );
1137     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1138         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1139     if( h->pps->b_transform_8x8_mode )
1140         COPY( analyse.b_transform_8x8 );
1141     if( h->frames.i_max_ref1 > 1 )
1142         COPY( i_bframe_pyramid );
1143     COPY( i_slice_max_size );
1144     COPY( i_slice_max_mbs );
1145     COPY( i_slice_count );
1146     /* VBV can't be turned on if it wasn't on to begin with */
1147     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1148           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1149     {
1150         COPY( rc.i_vbv_max_bitrate );
1151         COPY( rc.i_vbv_buffer_size );
1152         COPY( rc.i_bitrate );
1153     }
1154     COPY( rc.f_rf_constant );
1155 #undef COPY
1156
1157     mbcmp_init( h );
1158
1159     int ret = x264_validate_parameters( h );
1160
1161     /* Supported reconfiguration options (1-pass only):
1162      * vbv-maxrate
1163      * vbv-bufsize
1164      * crf
1165      * bitrate (CBR only) */
1166     if( !ret )
1167         x264_ratecontrol_init_reconfigurable( h, 0 );
1168
1169     return ret;
1170 }
1171
1172 /****************************************************************************
1173  * x264_encoder_parameters:
1174  ****************************************************************************/
1175 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1176 {
1177     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1178 }
1179
1180 /* internal usage */
1181 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1182 {
1183     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1184
1185     nal->i_ref_idc = i_ref_idc;
1186     nal->i_type    = i_type;
1187
1188     nal->i_payload= 0;
1189     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1190 }
1191 /* if number of allocated nals is not enough, re-allocate a larger one. */
1192 static int x264_nal_check_buffer( x264_t *h )
1193 {
1194     if( h->out.i_nal >= h->out.i_nals_allocated )
1195     {
1196         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1197         if( !new_out )
1198             return -1;
1199         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1200         x264_free( h->out.nal );
1201         h->out.nal = new_out;
1202         h->out.i_nals_allocated *= 2;
1203     }
1204     return 0;
1205 }
1206 static int x264_nal_end( x264_t *h )
1207 {
1208     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1209     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1210     h->out.i_nal++;
1211
1212     return x264_nal_check_buffer( h );
1213 }
1214
1215 static int x264_encoder_encapsulate_nals( x264_t *h )
1216 {
1217     int nal_size = 0, i;
1218     for( i = 0; i < h->out.i_nal; i++ )
1219         nal_size += h->out.nal[i].i_payload;
1220
1221     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1222     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1223     {
1224         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1225         if( !buf )
1226             return -1;
1227         x264_free( h->nal_buffer );
1228         h->nal_buffer = buf;
1229     }
1230
1231     uint8_t *nal_buffer = h->nal_buffer;
1232
1233     for( i = 0; i < h->out.i_nal; i++ )
1234     {
1235         int size = x264_nal_encode( nal_buffer, h->param.b_annexb, &h->out.nal[i] );
1236         h->out.nal[i].i_payload = size;
1237         h->out.nal[i].p_payload = nal_buffer;
1238         nal_buffer += size;
1239     }
1240
1241     return nal_buffer - h->nal_buffer;
1242 }
1243
1244 /****************************************************************************
1245  * x264_encoder_headers:
1246  ****************************************************************************/
1247 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1248 {
1249     int frame_size = 0;
1250     /* init bitstream context */
1251     h->out.i_nal = 0;
1252     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1253
1254     /* Write SEI, SPS and PPS. */
1255     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1256     if( x264_sei_version_write( h, &h->out.bs ) )
1257         return -1;
1258     if( x264_nal_end( h ) )
1259         return -1;
1260
1261     /* generate sequence parameters */
1262     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1263     x264_sps_write( &h->out.bs, h->sps );
1264     if( x264_nal_end( h ) )
1265         return -1;
1266
1267     /* generate picture parameters */
1268     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1269     x264_pps_write( &h->out.bs, h->pps );
1270     if( x264_nal_end( h ) )
1271         return -1;
1272
1273     frame_size = x264_encoder_encapsulate_nals( h );
1274
1275     /* now set output*/
1276     *pi_nal = h->out.i_nal;
1277     *pp_nal = &h->out.nal[0];
1278     h->out.i_nal = 0;
1279
1280     return frame_size;
1281 }
1282
1283 /* Check to see whether we have chosen a reference list ordering different
1284  * from the standard's default. */
1285 static inline void x264_reference_check_reorder( x264_t *h )
1286 {
1287     int i;
1288     for( i = 0; i < h->i_ref0 - 1; i++ )
1289         /* P and B-frames use different default orders. */
1290         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1291                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1292         {
1293             h->b_ref_reorder[0] = 1;
1294             break;
1295         }
1296 }
1297
1298 /* return -1 on failure, else return the index of the new reference frame */
1299 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1300 {
1301     int i = h->i_ref0;
1302     int j;
1303     x264_frame_t *newframe;
1304     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1305         return -1;
1306
1307     /* Find a place to insert the duplicate in the reference list. */
1308     for( j = 0; j < i; j++ )
1309         if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1310         {
1311             /* found a place, after j, make sure there is not already a duplicate there */
1312             if( j == i-1 || ( h->fref0[j+1] && h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1313                 break;
1314         }
1315
1316     if( j == i ) /* No room in the reference list for the duplicate. */
1317         return -1;
1318     j++;
1319
1320     newframe = x264_frame_pop_blank_unused( h );
1321
1322     //FIXME: probably don't need to copy everything
1323     *newframe = *h->fref0[i_ref];
1324     newframe->i_reference_count = 1;
1325     newframe->orig = h->fref0[i_ref];
1326     newframe->b_duplicate = 1;
1327     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1328
1329     /* shift the frames to make space for the dupe. */
1330     h->b_ref_reorder[0] = 1;
1331     if( h->i_ref0 < 16 )
1332         ++h->i_ref0;
1333     h->fref0[15] = NULL;
1334     x264_frame_unshift( &h->fref0[j], newframe );
1335
1336     return j;
1337 }
1338
1339 static void x264_weighted_pred_init( x264_t *h )
1340 {
1341     int i_ref;
1342     int i;
1343
1344     /* for now no analysis and set all weights to nothing */
1345     for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1346         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1347
1348     // FIXME: This only supports weighting of one reference frame
1349     // and duplicates of that frame.
1350     h->fenc->i_lines_weighted = 0;
1351
1352     for( i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1353         for( i = 0; i < 3; i++ )
1354             h->sh.weight[i_ref][i].weightfn = NULL;
1355
1356
1357     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1358         return;
1359
1360     int i_padv = PADV << h->param.b_interlaced;
1361     int denom = -1;
1362     int weightluma = 0;
1363     int buffer_next = 0;
1364     int j;
1365     //FIXME: when chroma support is added, move this into loop
1366     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1367     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1368     for( j = 0; j < h->i_ref0; j++ )
1369     {
1370         if( h->fenc->weight[j][0].weightfn )
1371         {
1372             h->sh.weight[j][0] = h->fenc->weight[j][0];
1373             // if weight is useless, don't write it to stream
1374             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1375                 h->sh.weight[j][0].weightfn = NULL;
1376             else
1377             {
1378                 if( !weightluma )
1379                 {
1380                     weightluma = 1;
1381                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1382                     assert( x264_clip3( denom, 0, 7 ) == denom );
1383                 }
1384                 assert( h->sh.weight[j][0].i_denom == denom );
1385                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1386                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1387                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1388                     h->fenc->i_stride[0] * i_padv + PADH;
1389             }
1390         }
1391
1392         //scale full resolution frame
1393         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1394         {
1395             uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1396             uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1397             int stride = h->fenc->i_stride[0];
1398             int width = h->fenc->i_width[0] + PADH*2;
1399             int height = h->fenc->i_lines[0] + i_padv*2;
1400             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1401             h->fenc->i_lines_weighted = height;
1402         }
1403     }
1404     if( !weightluma )
1405         h->sh.weight[0][0].i_denom = 0;
1406 }
1407
1408 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1409 {
1410     int i, b_ok;
1411
1412     /* build ref list 0/1 */
1413     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1414     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1415     if( h->sh.i_type == SLICE_TYPE_I )
1416         return;
1417
1418     for( i = 0; h->frames.reference[i]; i++ )
1419     {
1420         if( h->frames.reference[i]->i_poc < i_poc )
1421         {
1422             h->fref0[h->i_ref0++] = h->frames.reference[i];
1423         }
1424         else if( h->frames.reference[i]->i_poc > i_poc )
1425         {
1426             h->fref1[h->i_ref1++] = h->frames.reference[i];
1427         }
1428     }
1429
1430     /* Order ref0 from higher to lower poc */
1431     do
1432     {
1433         b_ok = 1;
1434         for( i = 0; i < h->i_ref0 - 1; i++ )
1435         {
1436             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1437             {
1438                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1439                 b_ok = 0;
1440                 break;
1441             }
1442         }
1443     } while( !b_ok );
1444
1445     if( h->sh.i_mmco_remove_from_end )
1446         for( i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1447         {
1448             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1449             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1450             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1451         }
1452
1453     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1454     do
1455     {
1456         b_ok = 1;
1457         for( i = 0; i < h->i_ref1 - 1; i++ )
1458         {
1459             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1460             {
1461                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1462                 b_ok = 0;
1463                 break;
1464             }
1465         }
1466     } while( !b_ok );
1467
1468     x264_reference_check_reorder( h );
1469
1470     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1471     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1472     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1473
1474     /* add duplicates */
1475     if( h->fenc->i_type == X264_TYPE_P )
1476     {
1477         int idx = -1;
1478         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1479         {
1480             x264_weight_t w[3];
1481             w[1].weightfn = w[2].weightfn = NULL;
1482             if( h->param.rc.b_stat_read )
1483                 x264_ratecontrol_set_weights( h, h->fenc );
1484
1485             if( !h->fenc->weight[0][0].weightfn )
1486             {
1487                 h->fenc->weight[0][0].i_denom = 0;
1488                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1489                 idx = x264_weighted_reference_duplicate( h, 0, w );
1490             }
1491             else
1492             {
1493                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1494                 {
1495                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1496                 }
1497                 x264_weighted_reference_duplicate( h, 0, weight_none );
1498                 if( h->fenc->weight[0][0].i_offset > -128 )
1499                 {
1500                     w[0] = h->fenc->weight[0][0];
1501                     w[0].i_offset--;
1502                     h->mc.weight_cache( h, &w[0] );
1503                     idx = x264_weighted_reference_duplicate( h, 0, w );
1504                 }
1505             }
1506         }
1507         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1508         {
1509             //weighted offset=-1
1510             x264_weight_t w[3];
1511             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1512             h->fenc->weight[0][0].i_denom = 0;
1513             w[1].weightfn = w[2].weightfn = NULL;
1514             idx = x264_weighted_reference_duplicate( h, 0, w );
1515         }
1516         h->mb.ref_blind_dupe = idx;
1517     }
1518
1519     assert( h->i_ref0 + h->i_ref1 <= 16 );
1520     h->mb.pic.i_fref[0] = h->i_ref0;
1521     h->mb.pic.i_fref[1] = h->i_ref1;
1522 }
1523
1524 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1525 {
1526     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1527     int b_hpel = h->fdec->b_kept_as_ref;
1528     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1529     int b_end = mb_y == h->sps->i_mb_height;
1530     int min_y = mb_y - (1 << h->sh.b_mbaff);
1531     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1532     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1533     if( mb_y & h->sh.b_mbaff )
1534         return;
1535     if( min_y < 0 )
1536         return;
1537
1538     if( !b_end && !h->param.b_sliced_threads )
1539     {
1540         int i, j;
1541         for( j=0; j<=h->sh.b_mbaff; j++ )
1542             for( i=0; i<3; i++ )
1543             {
1544                 memcpy( h->mb.intra_border_backup[j][i],
1545                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1546                         h->sps->i_mb_width*16 >> !!i );
1547             }
1548     }
1549
1550     if( b_deblock )
1551     {
1552         int y;
1553         for( y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1554             x264_frame_deblock_row( h, y );
1555     }
1556
1557     if( b_hpel )
1558     {
1559         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1560         if( h->param.analyse.i_subpel_refine )
1561         {
1562             x264_frame_filter( h, h->fdec, min_y, b_end );
1563             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1564         }
1565     }
1566
1567     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1568         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1569
1570     min_y = X264_MAX( min_y*16-8, 0 );
1571     max_y = b_end ? h->param.i_height : mb_y*16-8;
1572
1573     if( h->param.analyse.b_psnr )
1574     {
1575         int i;
1576         for( i=0; i<3; i++ )
1577             h->stat.frame.i_ssd[i] +=
1578                 x264_pixel_ssd_wxh( &h->pixf,
1579                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1580                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1581                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1582     }
1583
1584     if( h->param.analyse.b_ssim )
1585     {
1586         x264_emms();
1587         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1588          * and overlap by 4 */
1589         min_y += min_y == 0 ? 2 : -6;
1590         h->stat.frame.f_ssim +=
1591             x264_pixel_ssim_wxh( &h->pixf,
1592                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1593                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1594                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1595     }
1596 }
1597
1598 static inline int x264_reference_update( x264_t *h )
1599 {
1600     int i, j;
1601     if( !h->fdec->b_kept_as_ref )
1602     {
1603         if( h->i_thread_frames > 1 )
1604         {
1605             x264_frame_push_unused( h, h->fdec );
1606             h->fdec = x264_frame_pop_unused( h, 1 );
1607             if( !h->fdec )
1608                 return -1;
1609         }
1610         return 0;
1611     }
1612
1613     /* apply mmco from previous frame. */
1614     for( i = 0; i < h->sh.i_mmco_command_count; i++ )
1615         for( j = 0; h->frames.reference[j]; j++ )
1616             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1617                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1618
1619     /* move frame in the buffer */
1620     x264_frame_push( h->frames.reference, h->fdec );
1621     if( h->frames.reference[h->sps->i_num_ref_frames] )
1622         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1623     h->fdec = x264_frame_pop_unused( h, 1 );
1624     if( !h->fdec )
1625         return -1;
1626     return 0;
1627 }
1628
1629 static inline void x264_reference_reset( x264_t *h )
1630 {
1631     while( h->frames.reference[0] )
1632         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1633     h->fdec->i_poc =
1634     h->fenc->i_poc = 0;
1635 }
1636
1637 static inline void x264_reference_hierarchy_reset( x264_t *h )
1638 {
1639     int i, ref;
1640     int b_hasdelayframe = 0;
1641     if( !h->param.i_bframe_pyramid )
1642         return;
1643
1644     /* look for delay frames -- chain must only contain frames that are disposable */
1645     for( i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1646         b_hasdelayframe |= h->frames.current[i]->i_coded
1647                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1648
1649     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1650         return;
1651
1652     /* Remove last BREF. There will never be old BREFs in the
1653      * dpb during a BREF decode when pyramid == STRICT */
1654     for( ref = 0; h->frames.reference[ref]; ref++ )
1655     {
1656         if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1657             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1658         {
1659             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1660             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1661             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1662             x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1663             h->b_ref_reorder[0] = 1;
1664             break;
1665         }
1666     }
1667
1668     /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1669     h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1670 }
1671
1672 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1673 {
1674     /* ------------------------ Create slice header  ----------------------- */
1675     if( i_nal_type == NAL_SLICE_IDR )
1676     {
1677         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1678
1679         /* increment id */
1680         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1681     }
1682     else
1683     {
1684         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1685
1686         /* always set the real higher num of ref frame used */
1687         h->sh.b_num_ref_idx_override = 1;
1688         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1689         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1690     }
1691
1692     h->fdec->i_frame_num = h->sh.i_frame_num;
1693
1694     if( h->sps->i_poc_type == 0 )
1695     {
1696         h->sh.i_poc_lsb = h->fdec->i_poc & ( (1 << h->sps->i_log2_max_poc_lsb) - 1 );
1697         h->sh.i_delta_poc_bottom = 0;
1698     }
1699     else if( h->sps->i_poc_type == 1 )
1700     {
1701         /* FIXME TODO FIXME */
1702     }
1703     else
1704     {
1705         /* Nothing to do ? */
1706     }
1707
1708     x264_macroblock_slice_init( h );
1709 }
1710
1711 static int x264_slice_write( x264_t *h )
1712 {
1713     int i_skip;
1714     int mb_xy, i_mb_x, i_mb_y;
1715     int i, i_list, i_ref, i_skip_bak = 0; /* Shut up GCC. */
1716     bs_t bs_bak;
1717     x264_cabac_t cabac_bak;
1718     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1719     /* Assume no more than 3 bytes of NALU escaping. */
1720     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-3-NALU_OVERHEAD)*8 : INT_MAX;
1721     int starting_bits = bs_pos(&h->out.bs);
1722     bs_realign( &h->out.bs );
1723
1724     /* Slice */
1725     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1726
1727     /* Slice header */
1728     x264_macroblock_thread_init( h );
1729     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1730     if( h->param.b_cabac )
1731     {
1732         /* alignment needed */
1733         bs_align_1( &h->out.bs );
1734
1735         /* init cabac */
1736         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1737         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1738     }
1739     h->mb.i_last_qp = h->sh.i_qp;
1740     h->mb.i_last_dqp = 0;
1741
1742     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1743     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1744     i_skip = 0;
1745
1746     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1747     {
1748         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1749         if( h->param.i_slice_max_size > 0 )
1750         {
1751             /* We don't need the contexts because flushing the CABAC encoder has no context
1752              * dependency and macroblocks are only re-encoded in the case where a slice is
1753              * ended (and thus the content of all contexts are thrown away). */
1754             if( h->param.b_cabac )
1755             {
1756                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1757                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1758                  * backed up. */
1759                 cabac_prevbyte_bak = h->cabac.p[-1];
1760             }
1761             else
1762             {
1763                 bs_bak = h->out.bs;
1764                 i_skip_bak = i_skip;
1765             }
1766         }
1767
1768         if( i_mb_x == 0 && !h->mb.b_reencode_mb && !h->param.b_sliced_threads )
1769             x264_fdec_filter_row( h, i_mb_y );
1770
1771         /* load cache */
1772         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1773
1774         x264_macroblock_analyse( h );
1775
1776         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1777         x264_macroblock_encode( h );
1778
1779         if( x264_bitstream_check_buffer( h ) )
1780             return -1;
1781
1782         if( h->param.b_cabac )
1783         {
1784             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1785                 x264_cabac_encode_terminal( &h->cabac );
1786
1787             if( IS_SKIP( h->mb.i_type ) )
1788                 x264_cabac_mb_skip( h, 1 );
1789             else
1790             {
1791                 if( h->sh.i_type != SLICE_TYPE_I )
1792                     x264_cabac_mb_skip( h, 0 );
1793                 x264_macroblock_write_cabac( h, &h->cabac );
1794             }
1795         }
1796         else
1797         {
1798             if( IS_SKIP( h->mb.i_type ) )
1799                 i_skip++;
1800             else
1801             {
1802                 if( h->sh.i_type != SLICE_TYPE_I )
1803                 {
1804                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1805                     i_skip = 0;
1806                 }
1807                 x264_macroblock_write_cavlc( h );
1808             }
1809         }
1810
1811         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1812         int mb_size = total_bits - mb_spos;
1813
1814         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1815         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1816         {
1817             if( mb_xy != h->sh.i_first_mb )
1818             {
1819                 if( h->param.b_cabac )
1820                 {
1821                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1822                     h->cabac.p[-1] = cabac_prevbyte_bak;
1823                 }
1824                 else
1825                 {
1826                     h->out.bs = bs_bak;
1827                     i_skip = i_skip_bak;
1828                 }
1829                 h->mb.b_reencode_mb = 1;
1830                 h->sh.i_last_mb = mb_xy-1;
1831                 break;
1832             }
1833             else
1834             {
1835                 h->sh.i_last_mb = mb_xy;
1836                 h->mb.b_reencode_mb = 0;
1837             }
1838         }
1839         else
1840             h->mb.b_reencode_mb = 0;
1841
1842 #ifdef HAVE_VISUALIZE
1843         if( h->param.b_visualize )
1844             x264_visualize_mb( h );
1845 #endif
1846
1847         /* save cache */
1848         x264_macroblock_cache_save( h );
1849
1850         /* accumulate mb stats */
1851         h->stat.frame.i_mb_count[h->mb.i_type]++;
1852
1853         if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1854         {
1855             if( h->mb.i_partition != D_8x8 )
1856                     h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1857                 else
1858                     for( i = 0; i < 4; i++ )
1859                         h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1860             if( h->param.i_frame_reference > 1 )
1861                 for( i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1862                     for( i = 0; i < 4; i++ )
1863                     {
1864                         i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1865                         if( i_ref >= 0 )
1866                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1867                     }
1868         }
1869
1870         if( h->param.i_log_level >= X264_LOG_INFO )
1871         {
1872             if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1873             {
1874                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1875                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1876                 int b_intra = IS_INTRA(h->mb.i_type);
1877                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1878                 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1879                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1880             }
1881             if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1882             {
1883                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1884                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1885             }
1886             if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1887             {
1888                 if( h->mb.i_type == I_16x16 )
1889                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1890                 else if( h->mb.i_type == I_8x8 )
1891                     for( i = 0; i < 16; i += 4 )
1892                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1893                 else //if( h->mb.i_type == I_4x4 )
1894                     for( i = 0; i < 16; i++ )
1895                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1896             }
1897         }
1898
1899         x264_ratecontrol_mb( h, mb_size );
1900
1901         if( h->sh.b_mbaff )
1902         {
1903             i_mb_x += i_mb_y & 1;
1904             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1905         }
1906         else
1907             i_mb_x++;
1908         if( i_mb_x == h->sps->i_mb_width )
1909         {
1910             i_mb_y++;
1911             i_mb_x = 0;
1912         }
1913     }
1914
1915     if( h->param.b_cabac )
1916     {
1917         x264_cabac_encode_flush( h, &h->cabac );
1918         h->out.bs.p = h->cabac.p;
1919     }
1920     else
1921     {
1922         if( i_skip > 0 )
1923             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1924         /* rbsp_slice_trailing_bits */
1925         bs_rbsp_trailing( &h->out.bs );
1926         bs_flush( &h->out.bs );
1927     }
1928     if( x264_nal_end( h ) )
1929         return -1;
1930
1931     if( h->sh.i_last_mb == h->mb.i_mb_count-1 )
1932     {
1933         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1934                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
1935                                   - h->stat.frame.i_tex_bits
1936                                   - h->stat.frame.i_mv_bits;
1937         if( !h->param.b_sliced_threads )
1938             x264_fdec_filter_row( h, h->sps->i_mb_height );
1939     }
1940
1941     return 0;
1942 }
1943
1944 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1945 {
1946     if( dst == src )
1947         return;
1948
1949     // reference counting
1950     x264_frame_t **f;
1951     for( f = src->frames.reference; *f; f++ )
1952         (*f)->i_reference_count++;
1953     for( f = dst->frames.reference; *f; f++ )
1954         x264_frame_push_unused( src, *f );
1955     src->fdec->i_reference_count++;
1956     x264_frame_push_unused( src, dst->fdec );
1957
1958     // copy everything except the per-thread pointers and the constants.
1959     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1960     dst->param = src->param;
1961     dst->stat = src->stat;
1962 }
1963
1964 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1965 {
1966     if( dst == src )
1967         return;
1968     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1969 }
1970
1971 static void *x264_slices_write( x264_t *h )
1972 {
1973     int i_slice_num = 0;
1974     int last_thread_mb = h->sh.i_last_mb;
1975     if( h->param.i_sync_lookahead )
1976         x264_lower_thread_priority( 10 );
1977
1978 #ifdef HAVE_MMX
1979     /* Misalign mask has to be set separately for each thread. */
1980     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1981         x264_cpu_mask_misalign_sse();
1982 #endif
1983
1984 #ifdef HAVE_VISUALIZE
1985     if( h->param.b_visualize )
1986         if( x264_visualize_init( h ) )
1987             return (void *)-1;
1988 #endif
1989
1990     /* init stats */
1991     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
1992     h->mb.b_reencode_mb = 0;
1993     while( h->sh.i_first_mb <= last_thread_mb )
1994     {
1995         h->sh.i_last_mb = last_thread_mb;
1996         if( h->param.i_slice_max_mbs )
1997             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
1998         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
1999         {
2000             int height = h->sps->i_mb_height >> h->param.b_interlaced;
2001             int width = h->sps->i_mb_width << h->param.b_interlaced;
2002             i_slice_num++;
2003             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2004         }
2005         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2006         if( x264_stack_align( x264_slice_write, h ) )
2007             return (void *)-1;
2008         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2009     }
2010
2011 #ifdef HAVE_VISUALIZE
2012     if( h->param.b_visualize )
2013     {
2014         x264_visualize_show( h );
2015         x264_visualize_close( h );
2016     }
2017 #endif
2018
2019     return (void *)0;
2020 }
2021
2022 static int x264_threaded_slices_write( x264_t *h )
2023 {
2024     int i, j;
2025     void *ret = NULL;
2026     /* set first/last mb and sync contexts */
2027     for( i = 0; i < h->param.i_threads; i++ )
2028     {
2029         x264_t *t = h->thread[i];
2030         if( i )
2031         {
2032             t->param = h->param;
2033             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2034         }
2035         int height = h->sps->i_mb_height >> h->param.b_interlaced;
2036         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2037         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2038         t->sh.i_first_mb = t->i_threadslice_start * h->sps->i_mb_width;
2039         t->sh.i_last_mb  =   t->i_threadslice_end * h->sps->i_mb_width - 1;
2040     }
2041
2042     x264_analyse_weight_frame( h, h->sps->i_mb_height*16 + 16 );
2043
2044     x264_threads_distribute_ratecontrol( h );
2045
2046     /* dispatch */
2047     for( i = 0; i < h->param.i_threads; i++ )
2048     {
2049         if( x264_pthread_create( &h->thread[i]->thread_handle, NULL, (void*)x264_slices_write, (void*)h->thread[i] ) )
2050             return -1;
2051         h->thread[i]->b_thread_active = 1;
2052     }
2053     for( i = 0; i < h->param.i_threads; i++ )
2054     {
2055         x264_pthread_join( h->thread[i]->thread_handle, &ret );
2056         h->thread[i]->b_thread_active = 0;
2057         if( (intptr_t)ret )
2058             return (intptr_t)ret;
2059     }
2060
2061     /* deblocking and hpel filtering */
2062     for( i = 0; i <= h->sps->i_mb_height; i++ )
2063         x264_fdec_filter_row( h, i );
2064
2065     for( i = 1; i < h->param.i_threads; i++ )
2066     {
2067         x264_t *t = h->thread[i];
2068         for( j = 0; j < t->out.i_nal; j++ )
2069         {
2070             h->out.nal[h->out.i_nal] = t->out.nal[j];
2071             h->out.i_nal++;
2072             x264_nal_check_buffer( h );
2073         }
2074         /* All entries in stat.frame are ints except for ssd/ssim,
2075          * which are only calculated in the main thread. */
2076         for( j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2077             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2078     }
2079
2080     x264_threads_merge_ratecontrol( h );
2081
2082     return 0;
2083 }
2084
2085 /****************************************************************************
2086  * x264_encoder_encode:
2087  *  XXX: i_poc   : is the poc of the current given picture
2088  *       i_frame : is the number of the frame being coded
2089  *  ex:  type frame poc
2090  *       I      0   2*0
2091  *       P      1   2*3
2092  *       B      2   2*1
2093  *       B      3   2*2
2094  *       P      4   2*6
2095  *       B      5   2*4
2096  *       B      6   2*5
2097  ****************************************************************************/
2098 int     x264_encoder_encode( x264_t *h,
2099                              x264_nal_t **pp_nal, int *pi_nal,
2100                              x264_picture_t *pic_in,
2101                              x264_picture_t *pic_out )
2102 {
2103     x264_t *thread_current, *thread_prev, *thread_oldest;
2104     int i_nal_type, i_nal_ref_idc, i_global_qp, i;
2105
2106     if( h->i_thread_frames > 1 )
2107     {
2108         thread_prev    = h->thread[ h->i_thread_phase ];
2109         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2110         thread_current = h->thread[ h->i_thread_phase ];
2111         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2112         x264_thread_sync_context( thread_current, thread_prev );
2113         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2114         h = thread_current;
2115 //      fprintf(stderr, "current: %p  prev: %p  oldest: %p \n", thread_current, thread_prev, thread_oldest);
2116     }
2117     else
2118     {
2119         thread_current =
2120         thread_oldest  = h;
2121     }
2122
2123     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2124     if( x264_reference_update( h ) )
2125         return -1;
2126     h->fdec->i_lines_completed = -1;
2127
2128     /* no data out */
2129     *pi_nal = 0;
2130     *pp_nal = NULL;
2131
2132     /* ------------------- Setup new frame from picture -------------------- */
2133     if( pic_in != NULL )
2134     {
2135         /* 1: Copy the picture to a frame and move it to a buffer */
2136         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2137         if( !fenc )
2138             return -1;
2139
2140         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2141             return -1;
2142
2143         if( h->param.i_width != 16 * h->sps->i_mb_width ||
2144             h->param.i_height != 16 * h->sps->i_mb_height )
2145             x264_frame_expand_border_mod16( h, fenc );
2146
2147         fenc->i_frame = h->frames.i_input++;
2148
2149         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2150             h->frames.i_bframe_delay_time = fenc->i_pts;
2151
2152         if( h->frames.b_have_lowres )
2153         {
2154             if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2155                 x264_weight_plane_analyse( h, fenc );
2156             x264_frame_init_lowres( h, fenc );
2157         }
2158
2159         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2160         {
2161             if( x264_macroblock_tree_read( h, fenc ) )
2162                 return -1;
2163         }
2164         else if( h->param.rc.i_aq_mode )
2165             x264_adaptive_quant_frame( h, fenc );
2166
2167         /* 2: Place the frame into the queue for its slice type decision */
2168         x264_lookahead_put_frame( h, fenc );
2169
2170         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2171         {
2172             /* Nothing yet to encode, waiting for filling of buffers */
2173             pic_out->i_type = X264_TYPE_AUTO;
2174             return 0;
2175         }
2176     }
2177     else
2178     {
2179         /* signal kills for lookahead thread */
2180         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2181         h->lookahead->b_exit_thread = 1;
2182         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2183         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2184     }
2185
2186     h->i_frame++;
2187     /* 3: The picture is analyzed in the lookahead */
2188     if( !h->frames.current[0] )
2189         x264_lookahead_get_frames( h );
2190
2191     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2192         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2193
2194     /* ------------------- Get frame to be encoded ------------------------- */
2195     /* 4: get picture to encode */
2196     h->fenc = x264_frame_shift( h->frames.current );
2197     if( h->fenc->param )
2198     {
2199         x264_encoder_reconfig( h, h->fenc->param );
2200         if( h->fenc->param->param_free )
2201             h->fenc->param->param_free( h->fenc->param );
2202     }
2203
2204     if( h->fenc->b_keyframe )
2205     {
2206         h->frames.i_last_keyframe = h->fenc->i_frame;
2207         if( h->fenc->i_type == X264_TYPE_IDR )
2208             h->i_frame_num = 0;
2209     }
2210     h->sh.i_mmco_command_count =
2211     h->sh.i_mmco_remove_from_end = 0;
2212     h->b_ref_reorder[0] =
2213     h->b_ref_reorder[1] = 0;
2214
2215     /* ------------------- Setup frame context ----------------------------- */
2216     /* 5: Init data dependent of frame type */
2217     if( h->fenc->i_type == X264_TYPE_IDR )
2218     {
2219         /* reset ref pictures */
2220         i_nal_type    = NAL_SLICE_IDR;
2221         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2222         h->sh.i_type = SLICE_TYPE_I;
2223         x264_reference_reset( h );
2224     }
2225     else if( h->fenc->i_type == X264_TYPE_I )
2226     {
2227         i_nal_type    = NAL_SLICE;
2228         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2229         h->sh.i_type = SLICE_TYPE_I;
2230         x264_reference_hierarchy_reset( h );
2231     }
2232     else if( h->fenc->i_type == X264_TYPE_P )
2233     {
2234         i_nal_type    = NAL_SLICE;
2235         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2236         h->sh.i_type = SLICE_TYPE_P;
2237         x264_reference_hierarchy_reset( h );
2238     }
2239     else if( h->fenc->i_type == X264_TYPE_BREF )
2240     {
2241         i_nal_type    = NAL_SLICE;
2242         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2243         h->sh.i_type = SLICE_TYPE_B;
2244         x264_reference_hierarchy_reset( h );
2245     }
2246     else    /* B frame */
2247     {
2248         i_nal_type    = NAL_SLICE;
2249         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2250         h->sh.i_type = SLICE_TYPE_B;
2251     }
2252
2253     h->fdec->i_poc =
2254     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_keyframe);
2255     h->fdec->i_type = h->fenc->i_type;
2256     h->fdec->i_frame = h->fenc->i_frame;
2257     h->fenc->b_kept_as_ref =
2258     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2259
2260
2261
2262     /* ------------------- Init                ----------------------------- */
2263     /* build ref list 0/1 */
2264     x264_reference_build_list( h, h->fdec->i_poc );
2265
2266     /* ---------------------- Write the bitstream -------------------------- */
2267     /* Init bitstream context */
2268     if( h->param.b_sliced_threads )
2269     {
2270         for( i = 0; i < h->param.i_threads; i++ )
2271         {
2272             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2273             h->thread[i]->out.i_nal = 0;
2274         }
2275     }
2276     else
2277     {
2278         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2279         h->out.i_nal = 0;
2280     }
2281
2282     if( h->param.b_aud )
2283     {
2284         int pic_type;
2285
2286         if( h->sh.i_type == SLICE_TYPE_I )
2287             pic_type = 0;
2288         else if( h->sh.i_type == SLICE_TYPE_P )
2289             pic_type = 1;
2290         else if( h->sh.i_type == SLICE_TYPE_B )
2291             pic_type = 2;
2292         else
2293             pic_type = 7;
2294
2295         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2296         bs_write( &h->out.bs, 3, pic_type );
2297         bs_rbsp_trailing( &h->out.bs );
2298         if( x264_nal_end( h ) )
2299             return -1;
2300     }
2301
2302     h->i_nal_type = i_nal_type;
2303     h->i_nal_ref_idc = i_nal_ref_idc;
2304
2305     int overhead = NALU_OVERHEAD;
2306
2307     if( h->param.b_intra_refresh && h->fenc->i_type == X264_TYPE_P )
2308     {
2309         int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2310         float increment = ((float)h->sps->i_mb_width-1) / h->param.i_keyint_max;
2311         if( IS_X264_TYPE_I( h->fref0[0]->i_type ) )
2312             h->fdec->f_pir_position = 0;
2313         else
2314         {
2315             if( h->fref0[0]->i_pir_end_col == h->sps->i_mb_width - 1 )
2316             {
2317                 h->fdec->f_pir_position = 0;
2318                 h->fenc->b_keyframe = 1;
2319             }
2320             else
2321                 h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2322         }
2323         h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2324         h->fdec->f_pir_position += increment * pocdiff;
2325         h->fdec->i_pir_end_col = X264_MIN( h->fdec->f_pir_position+0.5, h->sps->i_mb_width-1 );
2326     }
2327
2328     /* Write SPS and PPS */
2329     if( h->fenc->b_keyframe )
2330     {
2331         if( h->param.b_repeat_headers )
2332         {
2333             if( h->fenc->i_frame == 0 )
2334             {
2335                 /* identify ourself */
2336                 x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2337                 if( x264_sei_version_write( h, &h->out.bs ) )
2338                     return -1;
2339                 if( x264_nal_end( h ) )
2340                     return -1;
2341                 overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2342             }
2343
2344             /* generate sequence parameters */
2345             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2346             x264_sps_write( &h->out.bs, h->sps );
2347             if( x264_nal_end( h ) )
2348                 return -1;
2349             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2350
2351             /* generate picture parameters */
2352             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2353             x264_pps_write( &h->out.bs, h->pps );
2354             if( x264_nal_end( h ) )
2355                 return -1;
2356             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2357         }
2358
2359         if( h->fenc->i_type != X264_TYPE_IDR )
2360         {
2361             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2362             x264_sei_recovery_point_write( h, &h->out.bs, h->param.i_keyint_max );
2363             x264_nal_end( h );
2364             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2365         }
2366     }
2367
2368     /* Init the rate control */
2369     /* FIXME: Include slice header bit cost. */
2370     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2371     i_global_qp = x264_ratecontrol_qp( h );
2372
2373     pic_out->i_qpplus1 =
2374     h->fdec->i_qpplus1 = i_global_qp + 1;
2375
2376     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2377     {
2378         x264_reference_build_list_optimal( h );
2379         x264_reference_check_reorder( h );
2380     }
2381
2382     if( h->sh.i_type == SLICE_TYPE_B )
2383         x264_macroblock_bipred_init( h );
2384
2385     /*------------------------- Weights -------------------------------------*/
2386     x264_weighted_pred_init( h );
2387
2388     /* ------------------------ Create slice header  ----------------------- */
2389     x264_slice_init( h, i_nal_type, i_global_qp );
2390
2391     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2392         h->i_frame_num++;
2393
2394     /* Write frame */
2395     h->i_threadslice_start = 0;
2396     h->i_threadslice_end = h->sps->i_mb_height;
2397     if( h->i_thread_frames > 1 )
2398     {
2399         if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2400             return -1;
2401         h->b_thread_active = 1;
2402     }
2403     else if( h->param.b_sliced_threads )
2404     {
2405         if( x264_threaded_slices_write( h ) )
2406             return -1;
2407     }
2408     else
2409         if( (intptr_t)x264_slices_write( h ) )
2410             return -1;
2411
2412     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2413 }
2414
2415 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2416                                    x264_nal_t **pp_nal, int *pi_nal,
2417                                    x264_picture_t *pic_out )
2418 {
2419     int i, j, i_list, frame_size;
2420     char psz_message[80];
2421
2422     if( h->b_thread_active )
2423     {
2424         void *ret = NULL;
2425         x264_pthread_join( h->thread_handle, &ret );
2426         h->b_thread_active = 0;
2427         if( (intptr_t)ret )
2428             return (intptr_t)ret;
2429     }
2430     if( !h->out.i_nal )
2431     {
2432         pic_out->i_type = X264_TYPE_AUTO;
2433         return 0;
2434     }
2435
2436     x264_frame_push_unused( thread_current, h->fenc );
2437
2438     /* End bitstream, set output  */
2439     *pi_nal = h->out.i_nal;
2440     *pp_nal = h->out.nal;
2441
2442     frame_size = x264_encoder_encapsulate_nals( h );
2443
2444     h->out.i_nal = 0;
2445
2446     /* Set output picture properties */
2447     if( h->sh.i_type == SLICE_TYPE_I )
2448         pic_out->i_type = h->i_nal_type == NAL_SLICE_IDR ? X264_TYPE_IDR : X264_TYPE_I;
2449     else if( h->sh.i_type == SLICE_TYPE_P )
2450         pic_out->i_type = X264_TYPE_P;
2451     else
2452         pic_out->i_type = X264_TYPE_B;
2453
2454     pic_out->b_keyframe = h->fenc->b_keyframe;
2455
2456     pic_out->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2457     if( h->frames.i_bframe_delay )
2458     {
2459         int64_t *i_prev_dts = thread_current->frames.i_prev_dts;
2460         if( h->i_frame <= h->frames.i_bframe_delay )
2461         {
2462             if( h->i_dts_compress_multiplier == 1 )
2463                 pic_out->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2464             else
2465             {
2466                 /* DTS compression */
2467                 if( h->i_frame == 1 )
2468                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2469                 pic_out->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2470             }
2471         }
2472         else
2473             pic_out->i_dts = i_prev_dts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2474         i_prev_dts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2475     }
2476     else
2477         pic_out->i_dts = h->fenc->i_reordered_pts;
2478     assert( pic_out->i_pts >= pic_out->i_dts );
2479
2480     pic_out->img.i_plane = h->fdec->i_plane;
2481     for(i = 0; i < 3; i++)
2482     {
2483         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2484         pic_out->img.plane[i] = h->fdec->plane[i];
2485     }
2486
2487     /* ---------------------- Update encoder state ------------------------- */
2488
2489     /* update rc */
2490     x264_emms();
2491     if( x264_ratecontrol_end( h, frame_size * 8 ) < 0 )
2492         return -1;
2493
2494     x264_noise_reduction_update( thread_current );
2495
2496     /* ---------------------- Compute/Print statistics --------------------- */
2497     x264_thread_sync_stat( h, h->thread[0] );
2498
2499     /* Slice stat */
2500     h->stat.i_frame_count[h->sh.i_type]++;
2501     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2502     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2503
2504     for( i = 0; i < X264_MBTYPE_MAX; i++ )
2505         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2506     for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2507         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2508     for( i = 0; i < 2; i++ )
2509         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2510     for( i = 0; i < 6; i++ )
2511         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2512     for( i = 0; i < 3; i++ )
2513         for( j = 0; j < 13; j++ )
2514             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2515     if( h->sh.i_type != SLICE_TYPE_I )
2516         for( i_list = 0; i_list < 2; i_list++ )
2517             for( i = 0; i < 32; i++ )
2518                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2519     if( h->sh.i_type == SLICE_TYPE_P )
2520     {
2521         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2522         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2523         {
2524             for( i = 0; i < 3; i++ )
2525                 for( j = 0; j < h->i_ref0; j++ )
2526                     if( h->sh.weight[0][i].i_denom != 0 )
2527                     {
2528                         h->stat.i_wpred[i]++;
2529                         break;
2530                     }
2531         }
2532     }
2533     if( h->sh.i_type == SLICE_TYPE_B )
2534     {
2535         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2536         if( h->mb.b_direct_auto_write )
2537         {
2538             //FIXME somewhat arbitrary time constants
2539             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2540             {
2541                 for( i = 0; i < 2; i++ )
2542                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2543             }
2544             for( i = 0; i < 2; i++ )
2545                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2546         }
2547     }
2548
2549     psz_message[0] = '\0';
2550     if( h->param.analyse.b_psnr )
2551     {
2552         int64_t ssd[3] = {
2553             h->stat.frame.i_ssd[0],
2554             h->stat.frame.i_ssd[1],
2555             h->stat.frame.i_ssd[2],
2556         };
2557
2558         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2559         h->stat.f_psnr_average[h->sh.i_type] += x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2560         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2561         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2562         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2563
2564         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2565                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2566                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2567                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2568     }
2569
2570     if( h->param.analyse.b_ssim )
2571     {
2572         double ssim_y = h->stat.frame.f_ssim
2573                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2574         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2575         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2576                   " SSIM Y:%.5f", ssim_y );
2577     }
2578     psz_message[79] = '\0';
2579
2580     x264_log( h, X264_LOG_DEBUG,
2581                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2582               h->i_frame,
2583               h->fdec->f_qp_avg_aq,
2584               h->i_nal_ref_idc,
2585               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2586               h->fdec->i_poc,
2587               h->stat.frame.i_mb_count_i,
2588               h->stat.frame.i_mb_count_p,
2589               h->stat.frame.i_mb_count_skip,
2590               frame_size,
2591               psz_message );
2592
2593     // keep stats all in one place
2594     x264_thread_sync_stat( h->thread[0], h );
2595     // for the use of the next frame
2596     x264_thread_sync_stat( thread_current, h );
2597
2598 #ifdef DEBUG_MB_TYPE
2599 {
2600     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2601         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2602     int mb_xy;
2603     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2604     {
2605         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2606             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2607         else
2608             fprintf( stderr, "? " );
2609
2610         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2611             fprintf( stderr, "\n" );
2612     }
2613 }
2614 #endif
2615
2616     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2617      * by freeing some of its pointers. */
2618      for( i = 0; i < h->i_ref0; i++ )
2619          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2620          {
2621              x264_frame_push_blank_unused( h, h->fref0[i] );
2622              h->fref0[i] = 0;
2623          }
2624
2625     if( h->param.psz_dump_yuv )
2626         x264_frame_dump( h );
2627
2628     return frame_size;
2629 }
2630
2631 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2632 {
2633     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2634         b_print_pcm ? "..PCM" : "",
2635         i_mb_count[I_16x16]/ i_count,
2636         i_mb_count[I_8x8]  / i_count,
2637         i_mb_count[I_4x4]  / i_count );
2638     if( b_print_pcm )
2639         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2640 }
2641
2642 /****************************************************************************
2643  * x264_encoder_close:
2644  ****************************************************************************/
2645 void    x264_encoder_close  ( x264_t *h )
2646 {
2647     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2648     int64_t i_mb_count_size[2][7] = {{0}};
2649     char buf[200];
2650     int i, j, i_list, i_type;
2651     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2652                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2653                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2654
2655     x264_lookahead_delete( h );
2656
2657     if( h->param.i_threads > 1 )
2658     {
2659         // don't strictly have to wait for the other threads, but it's simpler than canceling them
2660         for( i = 0; i < h->param.i_threads; i++ )
2661             if( h->thread[i]->b_thread_active )
2662                 x264_pthread_join( h->thread[i]->thread_handle, NULL );
2663         if( h->i_thread_frames > 1 )
2664         {
2665             for( i = 0; i < h->i_thread_frames; i++ )
2666             {
2667                 if( h->thread[i]->b_thread_active )
2668                 {
2669                     assert( h->thread[i]->fenc->i_reference_count == 1 );
2670                     x264_frame_delete( h->thread[i]->fenc );
2671                 }
2672             }
2673
2674             x264_t *thread_prev = h->thread[h->i_thread_phase];
2675             x264_thread_sync_ratecontrol( h, thread_prev, h );
2676             x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2677             h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2678         }
2679     }
2680     h->i_frame++;
2681
2682     /* Slices used and PSNR */
2683     for( i=0; i<5; i++ )
2684     {
2685         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2686         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2687         int i_slice = slice_order[i];
2688
2689         if( h->stat.i_frame_count[i_slice] > 0 )
2690         {
2691             const int i_count = h->stat.i_frame_count[i_slice];
2692             if( h->param.analyse.b_psnr )
2693             {
2694                 x264_log( h, X264_LOG_INFO,
2695                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
2696                           slice_name[i_slice],
2697                           i_count,
2698                           h->stat.f_frame_qp[i_slice] / i_count,
2699                           (double)h->stat.i_frame_size[i_slice] / i_count,
2700                           h->stat.f_psnr_mean_y[i_slice] / i_count, h->stat.f_psnr_mean_u[i_slice] / i_count, h->stat.f_psnr_mean_v[i_slice] / i_count,
2701                           h->stat.f_psnr_average[i_slice] / i_count,
2702                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2703             }
2704             else
2705             {
2706                 x264_log( h, X264_LOG_INFO,
2707                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2708                           slice_name[i_slice],
2709                           i_count,
2710                           h->stat.f_frame_qp[i_slice] / i_count,
2711                           (double)h->stat.i_frame_size[i_slice] / i_count );
2712             }
2713         }
2714     }
2715     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2716     {
2717         char *p = buf;
2718         int den = 0;
2719         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2720         for( i=0; i<=h->param.i_bframe; i++ )
2721             den += (i+1) * h->stat.i_consecutive_bframes[i];
2722         for( i=0; i<=h->param.i_bframe; i++ )
2723             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2724         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2725     }
2726
2727     for( i_type = 0; i_type < 2; i_type++ )
2728         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2729         {
2730             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2731             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2732         }
2733
2734     /* MB types used */
2735     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2736     {
2737         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2738         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2739         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2740         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2741     }
2742     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2743     {
2744         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2745         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2746         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2747         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2748         x264_log( h, X264_LOG_INFO,
2749                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2750                   buf,
2751                   i_mb_size[PIXEL_16x16] / (i_count*4),
2752                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2753                   i_mb_size[PIXEL_8x8] / (i_count*4),
2754                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2755                   i_mb_size[PIXEL_4x4] / (i_count*4),
2756                   i_mb_count[P_SKIP] / i_count );
2757     }
2758     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2759     {
2760         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2761         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2762         double i_mb_list_count;
2763         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2764         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2765         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2766         for( i = 0; i < X264_PARTTYPE_MAX; i++ )
2767             for( j = 0; j < 2; j++ )
2768             {
2769                 int l0 = x264_mb_type_list_table[i][0][j];
2770                 int l1 = x264_mb_type_list_table[i][1][j];
2771                 if( l0 || l1 )
2772                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2773             }
2774         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2775         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2776         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2777         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2778         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2779         x264_log( h, X264_LOG_INFO,
2780                   "mb B  %s  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%\n",
2781                   buf,
2782                   i_mb_size[PIXEL_16x16] / (i_count*4),
2783                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2784                   i_mb_size[PIXEL_8x8] / (i_count*4),
2785                   i_mb_count[B_DIRECT] / i_count,
2786                   i_mb_count[B_SKIP]   / i_count,
2787                   list_count[0] / i_mb_list_count,
2788                   list_count[1] / i_mb_list_count,
2789                   list_count[2] / i_mb_list_count );
2790     }
2791
2792     x264_ratecontrol_summary( h );
2793
2794     if( h->stat.i_frame_count[SLICE_TYPE_I] + h->stat.i_frame_count[SLICE_TYPE_P] + h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2795     {
2796 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2797 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2798         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2799         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2800                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
2801         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2802         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2803                             h->stat.i_frame_count[SLICE_TYPE_P] +
2804                             h->stat.i_frame_count[SLICE_TYPE_B];
2805         int64_t i_mb_count = i_count * h->mb.i_mb_count;
2806         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2807         float f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2808
2809         if( h->pps->b_transform_8x8_mode )
2810         {
2811             buf[0] = 0;
2812             if( h->stat.i_mb_count_8x8dct[0] )
2813                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2814             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2815         }
2816
2817         if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
2818             && h->stat.i_frame_count[SLICE_TYPE_B] )
2819         {
2820             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
2821                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2822                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2823         }
2824
2825         buf[0] = 0;
2826         if( i_mb_count != i_all_intra )
2827             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2828                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2829                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
2830                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2831         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2832                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2833                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
2834                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
2835
2836         int64_t fixed_pred_modes[3][9] = {{0}};
2837         int64_t sum_pred_modes[3] = {0};
2838         for( i = 0; i <= I_PRED_16x16_DC_128; i++ )
2839         {
2840             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2841             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2842         }
2843         if( sum_pred_modes[0] )
2844             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2845                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2846                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2847                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2848                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2849         for( i = 1; i <= 2; i++ )
2850         {
2851             for( j = 0; j <= I_PRED_8x8_DC_128; j++ )
2852             {
2853                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2854                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2855             }
2856             if( sum_pred_modes[i] )
2857                 x264_log( h, X264_LOG_INFO, "i%d v,h,dc,ddl,ddr,vr,hd,vl,hu: %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n", (3-i)*4,
2858                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2859                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2860                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2861                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2862                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2863                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2864                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2865                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2866                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2867         }
2868
2869         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2870             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
2871                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
2872
2873         for( i_list = 0; i_list < 2; i_list++ )
2874         {
2875             int i_slice;
2876             for( i_slice = 0; i_slice < 2; i_slice++ )
2877             {
2878                 char *p = buf;
2879                 int64_t i_den = 0;
2880                 int i_max = 0;
2881                 for( i = 0; i < 32; i++ )
2882                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2883                     {
2884                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2885                         i_max = i;
2886                     }
2887                 if( i_max == 0 )
2888                     continue;
2889                 for( i = 0; i <= i_max; i++ )
2890                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2891                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
2892             }
2893         }
2894
2895         if( h->param.analyse.b_ssim )
2896         {
2897             x264_log( h, X264_LOG_INFO,
2898                       "SSIM Mean Y:%.7f\n",
2899                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2900         }
2901         if( h->param.analyse.b_psnr )
2902         {
2903             x264_log( h, X264_LOG_INFO,
2904                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2905                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2906                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2907                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2908                       SUM3( h->stat.f_psnr_average ) / i_count,
2909                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
2910                       f_bitrate );
2911         }
2912         else
2913             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
2914     }
2915
2916     /* rc */
2917     x264_ratecontrol_delete( h );
2918
2919     /* param */
2920     if( h->param.rc.psz_stat_out )
2921         free( h->param.rc.psz_stat_out );
2922     if( h->param.rc.psz_stat_in )
2923         free( h->param.rc.psz_stat_in );
2924
2925     x264_cqm_delete( h );
2926     x264_free( h->nal_buffer );
2927     x264_analyse_free_costs( h );
2928
2929     if( h->i_thread_frames > 1)
2930         h = h->thread[h->i_thread_phase];
2931
2932     /* frames */
2933     x264_frame_delete_list( h->frames.unused[0] );
2934     x264_frame_delete_list( h->frames.unused[1] );
2935     x264_frame_delete_list( h->frames.current );
2936     x264_frame_delete_list( h->frames.blank_unused );
2937
2938     h = h->thread[0];
2939
2940     for( i = h->param.i_threads - 1; i >= 0; i-- )
2941     {
2942         x264_frame_t **frame;
2943
2944         if( !h->param.b_sliced_threads || i == 0 )
2945         {
2946             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
2947             {
2948                 assert( (*frame)->i_reference_count > 0 );
2949                 (*frame)->i_reference_count--;
2950                 if( (*frame)->i_reference_count == 0 )
2951                     x264_frame_delete( *frame );
2952             }
2953             frame = &h->thread[i]->fdec;
2954             assert( (*frame)->i_reference_count > 0 );
2955             (*frame)->i_reference_count--;
2956             if( (*frame)->i_reference_count == 0 )
2957                 x264_frame_delete( *frame );
2958             x264_macroblock_cache_end( h->thread[i] );
2959         }
2960         x264_free( h->thread[i]->scratch_buffer );
2961         x264_free( h->thread[i]->out.p_bitstream );
2962         x264_free( h->thread[i]->out.nal);
2963         x264_free( h->thread[i] );
2964     }
2965 }
2966
2967 /****************************************************************************
2968  * x264_encoder_delayed_frames:
2969  ****************************************************************************/
2970 int x264_encoder_delayed_frames( x264_t *h )
2971 {
2972     int delayed_frames = 0;
2973     int i;
2974     if( h->i_thread_frames > 1 )
2975     {
2976         for( i=0; i<h->i_thread_frames; i++ )
2977             delayed_frames += h->thread[i]->b_thread_active;
2978         h = h->thread[h->i_thread_phase];
2979     }
2980     for( i=0; h->frames.current[i]; i++ )
2981         delayed_frames++;
2982     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
2983     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2984     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
2985     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
2986     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
2987     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2988     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
2989     return delayed_frames;
2990 }