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 bs_write_ue bs_write_ue_big
  43
  44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  45                                    x264_nal_t **pp_nal, int *pi_nal,
  46                                    x264_picture_t *pic_out );
  47
  48 /****************************************************************************
  49  *
  50  ******************************* x264 libs **********************************
  51  *
  52  ****************************************************************************/
  53 static float x264_psnr( int64_t i_sqe, int64_t i_size )
  54 {
  55     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
  56     if( f_mse <= 0.0000000001 ) /* Max 100dB */
  57         return 100;
  58
  59     return -10.0 * log10( f_mse );
  60 }
  61
  62 static void x264_frame_dump( x264_t *h )
  63 {
  64     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  65     if( !f )
  66         return;
  67     /* Write the frame in display order */
  68     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2, SEEK_SET );
  69     for( int i = 0; i < h->fdec->i_plane; i++ )
  70         for( int y = 0; y < h->param.i_height >> !!i; y++ )
  71             fwrite( &h->fdec->plane[i][y*h->fdec->i_stride[i]], 1, h->param.i_width >> !!i, f );
  72     fclose( f );
  73 }
  74
  75
  76 /* Fill "default" values */
  77 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  78                                     x264_sps_t *sps, x264_pps_t *pps,
  79                                     int i_idr_pic_id, int i_frame, int i_qp )
  80 {
  81     x264_param_t *param = &h->param;
  82
  83     /* First we fill all fields */
  84     sh->sps = sps;
  85     sh->pps = pps;
  86
  87     sh->i_first_mb  = 0;
  88     sh->i_last_mb   = h->mb.i_mb_count - 1;
  89     sh->i_pps_id    = pps->i_id;
  90
  91     sh->i_frame_num = i_frame;
  92
  93     sh->b_mbaff = h->param.b_interlaced;
  94     sh->b_field_pic = 0;    /* no field support for now */
  95     sh->b_bottom_field = 0; /* not yet used */
  96
  97     sh->i_idr_pic_id = i_idr_pic_id;
  98
  99     /* poc stuff, fixed later */
 100     sh->i_poc = 0;
 101     sh->i_delta_poc_bottom = 0;
 102     sh->i_delta_poc[0] = 0;
 103     sh->i_delta_poc[1] = 0;
 104
 105     sh->i_redundant_pic_cnt = 0;
 106
 107     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 108                                 && h->param.i_bframe
 109                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 110
 111     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 112     {
 113         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
 114         {
 115             if( h->mb.b_direct_auto_write )
 116                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 117             else
 118                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 119         }
 120         else
 121         {
 122             h->mb.b_direct_auto_write = 0;
 123             sh->b_direct_spatial_mv_pred = 1;
 124         }
 125     }
 126     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 127
 128     sh->b_num_ref_idx_override = 0;
 129     sh->i_num_ref_idx_l0_active = 1;
 130     sh->i_num_ref_idx_l1_active = 1;
 131
 132     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 133     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 134
 135     /* If the ref list isn't in the default order, construct reordering header */
 136     /* List1 reordering isn't needed yet */
 137     if( sh->b_ref_pic_list_reordering_l0 )
 138     {
 139         int pred_frame_num = i_frame;
 140         for( int i = 0; i < h->i_ref0; i++ )
 141         {
 142             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 143             if( diff == 0 )
 144                 x264_log( h, X264_LOG_ERROR, "diff frame num == 0\n" );
 145             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 146             sh->ref_pic_list_order[0][i].arg = abs( diff ) - 1;
 147             pred_frame_num = h->fref0[i]->i_frame_num;
 148         }
 149     }
 150
 151     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 152
 153     sh->i_qp = i_qp;
 154     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 155     sh->b_sp_for_swidth = 0;
 156     sh->i_qs_delta = 0;
 157
 158     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 159     /* If effective qp <= 15, deblocking would have no effect anyway */
 160     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 161         sh->i_disable_deblocking_filter_idc = 0;
 162     else
 163         sh->i_disable_deblocking_filter_idc = 1;
 164     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 165     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 166 }
 167
 168 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 169 {
 170     if( sh->b_mbaff )
 171     {
 172         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 173         bs_write_ue( s, sh->i_first_mb >> 1 );
 174     }
 175     else
 176         bs_write_ue( s, sh->i_first_mb );
 177
 178     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 179     bs_write_ue( s, sh->i_pps_id );
 180     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 181
 182     if( !sh->sps->b_frame_mbs_only )
 183     {
 184         bs_write1( s, sh->b_field_pic );
 185         if( sh->b_field_pic )
 186             bs_write1( s, sh->b_bottom_field );
 187     }
 188
 189     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 190         bs_write_ue( s, sh->i_idr_pic_id );
 191
 192     if( sh->sps->i_poc_type == 0 )
 193     {
 194         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 195         if( sh->pps->b_pic_order && !sh->b_field_pic )
 196             bs_write_se( s, sh->i_delta_poc_bottom );
 197     }
 198     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 199     {
 200         bs_write_se( s, sh->i_delta_poc[0] );
 201         if( sh->pps->b_pic_order && !sh->b_field_pic )
 202             bs_write_se( s, sh->i_delta_poc[1] );
 203     }
 204
 205     if( sh->pps->b_redundant_pic_cnt )
 206         bs_write_ue( s, sh->i_redundant_pic_cnt );
 207
 208     if( sh->i_type == SLICE_TYPE_B )
 209         bs_write1( s, sh->b_direct_spatial_mv_pred );
 210
 211     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_B )
 212     {
 213         bs_write1( s, sh->b_num_ref_idx_override );
 214         if( sh->b_num_ref_idx_override )
 215         {
 216             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 217             if( sh->i_type == SLICE_TYPE_B )
 218                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 219         }
 220     }
 221
 222     /* ref pic list reordering */
 223     if( sh->i_type != SLICE_TYPE_I )
 224     {
 225         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 226         if( sh->b_ref_pic_list_reordering_l0 )
 227         {
 228             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 229             {
 230                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 231                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 232             }
 233             bs_write_ue( s, 3 );
 234         }
 235     }
 236     if( sh->i_type == SLICE_TYPE_B )
 237     {
 238         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 239         if( sh->b_ref_pic_list_reordering_l1 )
 240         {
 241             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 242             {
 243                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 244                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 245             }
 246             bs_write_ue( s, 3 );
 247         }
 248     }
 249
 250     if( sh->pps->b_weighted_pred && ( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_SP ) )
 251     {
 252         /* pred_weight_table() */
 253         bs_write_ue( s, sh->weight[0][0].i_denom );
 254         bs_write_ue( s, sh->weight[0][1].i_denom );
 255         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 256         {
 257             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 258             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 259             bs_write1( s, luma_weight_l0_flag );
 260             if( luma_weight_l0_flag )
 261             {
 262                 bs_write_se( s, sh->weight[i][0].i_scale );
 263                 bs_write_se( s, sh->weight[i][0].i_offset );
 264             }
 265             bs_write1( s, chroma_weight_l0_flag );
 266             if( chroma_weight_l0_flag )
 267             {
 268                 for( int j = 1; j < 3; j++ )
 269                 {
 270                     bs_write_se( s, sh->weight[i][j].i_scale );
 271                     bs_write_se( s, sh->weight[i][j].i_offset );
 272                 }
 273             }
 274         }
 275     }
 276     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 277     {
 278       /* TODO */
 279     }
 280
 281     if( i_nal_ref_idc != 0 )
 282     {
 283         if( sh->i_idr_pic_id >= 0 )
 284         {
 285             bs_write1( s, 0 );  /* no output of prior pics flag */
 286             bs_write1( s, 0 );  /* long term reference flag */
 287         }
 288         else
 289         {
 290             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 291             if( sh->i_mmco_command_count > 0 )
 292             {
 293                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 294                 {
 295                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 296                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 297                 }
 298                 bs_write_ue( s, 0 ); /* end command list */
 299             }
 300         }
 301     }
 302
 303     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 304         bs_write_ue( s, sh->i_cabac_init_idc );
 305
 306     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 307
 308     if( sh->pps->b_deblocking_filter_control )
 309     {
 310         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 311         if( sh->i_disable_deblocking_filter_idc != 1 )
 312         {
 313             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 314             bs_write_se( s, sh->i_beta_offset >> 1 );
 315         }
 316     }
 317 }
 318
 319 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 320 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 321 static int x264_bitstream_check_buffer( x264_t *h )
 322 {
 323     uint8_t *bs_bak = h->out.p_bitstream;
 324     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
 325         (h->out.bs.p_end - h->out.bs.p < 2500) )
 326     {
 327         h->out.i_bitstream += 100000;
 328         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 329         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 330         intptr_t delta = h->out.p_bitstream - bs_bak;
 331
 332         h->out.bs.p_start += delta;
 333         h->out.bs.p += delta;
 334         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 335
 336         h->cabac.p_start += delta;
 337         h->cabac.p += delta;
 338         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 339
 340         for( int i = 0; i <= h->out.i_nal; i++ )
 341             h->out.nal[i].p_payload += delta;
 342         x264_free( bs_bak );
 343     }
 344     return 0;
 345 fail:
 346     x264_free( bs_bak );
 347     return -1;
 348 }
 349
 350 /****************************************************************************
 351  *
 352  ****************************************************************************
 353  ****************************** External API*********************************
 354  ****************************************************************************
 355  *
 356  ****************************************************************************/
 357
 358 static int x264_validate_parameters( x264_t *h )
 359 {
 360 #ifdef HAVE_MMX
 361     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 362     {
 363         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 364         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 365         return -1;
 366     }
 367 #endif
 368     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 369     {
 370         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 371                   h->param.i_width, h->param.i_height );
 372         return -1;
 373     }
 374
 375     if( h->param.i_width % 2 || h->param.i_height % 2 )
 376     {
 377         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 378                   h->param.i_width, h->param.i_height );
 379         return -1;
 380     }
 381     int i_csp = h->param.i_csp & X264_CSP_MASK;
 382     if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
 383     {
 384         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12 supported)\n" );
 385         return -1;
 386     }
 387
 388     if( h->param.i_threads == X264_THREADS_AUTO )
 389         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 390     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 391     if( h->param.i_threads > 1 )
 392     {
 393 #ifndef HAVE_PTHREAD
 394         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 395         h->param.i_threads = 1;
 396 #endif
 397         /* Avoid absurdly small thread slices as they can reduce performance
 398          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 399         if( h->param.b_sliced_threads )
 400         {
 401             int max_threads = (h->param.i_height+15)/16 / 4;
 402             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 403         }
 404     }
 405     else
 406         h->param.b_sliced_threads = 0;
 407     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 408
 409     if( h->param.b_interlaced )
 410     {
 411         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 412         {
 413             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 414             h->param.analyse.i_me_method = X264_ME_UMH;
 415         }
 416         if( h->param.analyse.i_weighted_pred > 0 )
 417         {
 418             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 419             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 420         }
 421     }
 422
 423     /* Detect default ffmpeg settings and terminate with an error. */
 424     {
 425         int score = 0;
 426         score += h->param.analyse.i_me_range == 0;
 427         score += h->param.rc.i_qp_step == 3;
 428         score += h->param.i_keyint_max == 12;
 429         score += h->param.rc.i_qp_min == 2;
 430         score += h->param.rc.i_qp_max == 31;
 431         score += h->param.rc.f_qcompress == 0.5;
 432         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 433         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 434         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 435         if( score >= 5 )
 436         {
 437             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 438             x264_log( h, X264_LOG_ERROR, "use an encoding preset (vpre)\n" );
 439             return -1;
 440         }
 441     }
 442
 443     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 444     {
 445         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 446         return -1;
 447     }
 448     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, 0, 51 );
 449     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
 450     if( h->param.rc.i_rc_method == X264_RC_CRF )
 451     {
 452         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant;
 453         h->param.rc.i_bitrate = 0;
 454     }
 455     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 456         && h->param.rc.i_qp_constant == 0 )
 457     {
 458         h->mb.b_lossless = 1;
 459         h->param.i_cqm_preset = X264_CQM_FLAT;
 460         h->param.psz_cqm_file = NULL;
 461         h->param.rc.i_rc_method = X264_RC_CQP;
 462         h->param.rc.f_ip_factor = 1;
 463         h->param.rc.f_pb_factor = 1;
 464         h->param.analyse.b_psnr = 0;
 465         h->param.analyse.b_ssim = 0;
 466         h->param.analyse.i_chroma_qp_offset = 0;
 467         h->param.analyse.i_trellis = 0;
 468         h->param.analyse.b_fast_pskip = 0;
 469         h->param.analyse.i_noise_reduction = 0;
 470         h->param.analyse.f_psy_rd = 0;
 471         h->param.i_bframe = 0;
 472         /* 8x8dct is not useful at all in CAVLC lossless */
 473         if( !h->param.b_cabac )
 474             h->param.analyse.b_transform_8x8 = 0;
 475     }
 476     if( h->param.rc.i_rc_method == X264_RC_CQP )
 477     {
 478         float qp_p = h->param.rc.i_qp_constant;
 479         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 480         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 481         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, 51 );
 482         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, 51 );
 483         h->param.rc.i_aq_mode = 0;
 484         h->param.rc.b_mb_tree = 0;
 485     }
 486     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, 51 );
 487     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 488     if( h->param.rc.i_vbv_buffer_size )
 489     {
 490         if( h->param.rc.i_rc_method == X264_RC_CQP )
 491         {
 492             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 493             h->param.rc.i_vbv_max_bitrate = 0;
 494             h->param.rc.i_vbv_buffer_size = 0;
 495         }
 496         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 497         {
 498             if( h->param.rc.i_rc_method == X264_RC_ABR )
 499             {
 500                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 501                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 502             }
 503             else
 504             {
 505                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 506                 h->param.rc.i_vbv_buffer_size = 0;
 507             }
 508         }
 509         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 510                  h->param.rc.i_rc_method == X264_RC_ABR )
 511         {
 512             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 513             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 514         }
 515     }
 516     else if( h->param.rc.i_vbv_max_bitrate )
 517     {
 518         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 519         h->param.rc.i_vbv_max_bitrate = 0;
 520     }
 521
 522     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 523     if( h->param.b_sliced_threads )
 524         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 525     else
 526     {
 527         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 528         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 529         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 530         if( h->param.b_interlaced && h->param.i_slice_max_size )
 531         {
 532             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 533             h->param.i_slice_max_size = 0;
 534         }
 535         if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 536         {
 537             x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 538             h->param.i_slice_max_mbs = 0;
 539         }
 540         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 541             h->param.i_slice_count = 0;
 542     }
 543
 544     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
 545     if( h->param.i_keyint_max <= 0 )
 546         h->param.i_keyint_max = 1;
 547     if( h->param.i_scenecut_threshold < 0 )
 548         h->param.i_scenecut_threshold = 0;
 549     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 550     {
 551         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 552         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 553     }
 554     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
 555     if( h->param.i_keyint_max == 1 )
 556     {
 557         h->param.i_bframe = 0;
 558         h->param.b_intra_refresh = 0;
 559     }
 560     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 561     if( h->param.i_bframe <= 1 )
 562         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 563     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 564     if( !h->param.i_bframe )
 565     {
 566         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 567         h->param.analyse.i_direct_mv_pred = 0;
 568         h->param.analyse.b_weighted_bipred = 0;
 569     }
 570     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 571     {
 572         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 573         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 574     }
 575     if( h->param.b_intra_refresh && h->param.i_frame_reference > 1 )
 576     {
 577         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 578         h->param.i_frame_reference = 1;
 579     }
 580     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 581     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 582     {
 583         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 584         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 585         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;
 586         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 587     }
 588
 589     if( !h->param.i_timebase_num || !h->param.i_timebase_den )
 590     {
 591         h->param.i_timebase_num = h->param.i_fps_den;
 592         h->param.i_timebase_den = h->param.i_fps_num;
 593     }
 594
 595     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 596     if( !h->param.rc.i_lookahead || h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 597         h->param.rc.b_mb_tree = 0;
 598     if( h->param.rc.b_stat_read )
 599         h->param.rc.i_lookahead = 0;
 600 #ifdef HAVE_PTHREAD
 601     if( h->param.i_sync_lookahead )
 602         h->param.i_sync_lookahead = x264_clip3( h->param.i_sync_lookahead, h->i_thread_frames + h->param.i_bframe, X264_LOOKAHEAD_MAX );
 603     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 604         h->param.i_sync_lookahead = 0;
 605 #else
 606     h->param.i_sync_lookahead = 0;
 607 #endif
 608
 609     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 610     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 611     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 612     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 613
 614     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 615
 616     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 617         h->param.i_cqm_preset = X264_CQM_FLAT;
 618
 619     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 620         h->param.analyse.i_me_method > X264_ME_TESA )
 621         h->param.analyse.i_me_method = X264_ME_HEX;
 622     if( h->param.analyse.i_me_range < 4 )
 623         h->param.analyse.i_me_range = 4;
 624     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 625         h->param.analyse.i_me_range = 16;
 626     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 627         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 628         h->param.analyse.i_me_method = X264_ME_ESA;
 629     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 630     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 631     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 632                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 633     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 634     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 635         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 636     if( !h->param.analyse.b_transform_8x8 )
 637     {
 638         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 639         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 640     }
 641     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 642     if( !h->param.b_cabac )
 643         h->param.analyse.i_trellis = 0;
 644     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 645     if( !h->param.analyse.b_psy )
 646     {
 647         h->param.analyse.f_psy_rd = 0;
 648         h->param.analyse.f_psy_trellis = 0;
 649     }
 650     if( !h->param.analyse.i_trellis )
 651         h->param.analyse.f_psy_trellis = 0;
 652     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 653     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 654     if( h->param.analyse.i_subpel_refine < 6 )
 655         h->param.analyse.f_psy_rd = 0;
 656     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 657     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 658     /* so we lower the chroma QP offset to compensate */
 659     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 660      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 661     if( h->mb.i_psy_rd )
 662         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 663     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 664     /* Psy trellis has a similar effect. */
 665     if( h->mb.i_psy_trellis )
 666         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 667     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 668     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 669     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 670     if( h->param.rc.f_aq_strength == 0 )
 671         h->param.rc.i_aq_mode = 0;
 672     /* MB-tree requires AQ to be on, even if the strength is zero. */
 673     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 674     {
 675         h->param.rc.i_aq_mode = 1;
 676         h->param.rc.f_aq_strength = 0;
 677     }
 678     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 679     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 680         h->param.analyse.i_subpel_refine = 9;
 681
 682     {
 683         const x264_level_t *l = x264_levels;
 684         if( h->param.i_level_idc < 0 )
 685         {
 686             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 687             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 688                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 689             h->sps = h->sps_array;
 690             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 691             do h->param.i_level_idc = l->level_idc;
 692                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 693             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 694         }
 695         else
 696         {
 697             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 698                 l++;
 699             if( l->level_idc == 0 )
 700             {
 701                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 702                 return -1;
 703             }
 704         }
 705         if( h->param.analyse.i_mv_range <= 0 )
 706             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 707         else
 708             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 709     }
 710
 711     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, 0, X264_WEIGHTP_SMART );
 712     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 713         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 714
 715     if( h->i_thread_frames > 1 )
 716     {
 717         int r = h->param.analyse.i_mv_range_thread;
 718         int r2;
 719         if( r <= 0 )
 720         {
 721             // half of the available space is reserved and divided evenly among the threads,
 722             // the rest is allocated to whichever thread is far enough ahead to use it.
 723             // reserving more space increases quality for some videos, but costs more time
 724             // in thread synchronization.
 725             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 726             r = max_range / 2;
 727         }
 728         r = X264_MAX( r, h->param.analyse.i_me_range );
 729         r = X264_MIN( r, h->param.analyse.i_mv_range );
 730         // round up to use the whole mb row
 731         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 732         if( r2 < r )
 733             r2 += 16;
 734         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 735         h->param.analyse.i_mv_range_thread = r2;
 736     }
 737
 738     if( h->param.rc.f_qblur < 0 )
 739         h->param.rc.f_qblur = 0;
 740     if( h->param.rc.f_complexity_blur < 0 )
 741         h->param.rc.f_complexity_blur = 0;
 742
 743     h->param.i_sps_id &= 31;
 744
 745     if( h->param.i_log_level < X264_LOG_INFO )
 746     {
 747         h->param.analyse.b_psnr = 0;
 748         h->param.analyse.b_ssim = 0;
 749     }
 750
 751     if( h->param.b_interlaced )
 752         h->param.b_pic_struct = 1;
 753
 754     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 755     {
 756         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 757         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 758     }
 759
 760     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 761        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 762     {
 763         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 764         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 765     }
 766
 767     /* ensure the booleans are 0 or 1 so they can be used in math */
 768 #define BOOLIFY(x) h->param.x = !!h->param.x
 769     BOOLIFY( b_cabac );
 770     BOOLIFY( b_constrained_intra );
 771     BOOLIFY( b_deblocking_filter );
 772     BOOLIFY( b_deterministic );
 773     BOOLIFY( b_sliced_threads );
 774     BOOLIFY( b_interlaced );
 775     BOOLIFY( b_intra_refresh );
 776     BOOLIFY( b_visualize );
 777     BOOLIFY( b_aud );
 778     BOOLIFY( b_repeat_headers );
 779     BOOLIFY( b_annexb );
 780     BOOLIFY( b_vfr_input );
 781     BOOLIFY( b_pic_struct );
 782     BOOLIFY( analyse.b_transform_8x8 );
 783     BOOLIFY( analyse.b_weighted_bipred );
 784     BOOLIFY( analyse.b_chroma_me );
 785     BOOLIFY( analyse.b_mixed_references );
 786     BOOLIFY( analyse.b_fast_pskip );
 787     BOOLIFY( analyse.b_dct_decimate );
 788     BOOLIFY( analyse.b_psy );
 789     BOOLIFY( analyse.b_psnr );
 790     BOOLIFY( analyse.b_ssim );
 791     BOOLIFY( rc.b_stat_write );
 792     BOOLIFY( rc.b_stat_read );
 793     BOOLIFY( rc.b_mb_tree );
 794 #undef BOOLIFY
 795
 796     return 0;
 797 }
 798
 799 static void mbcmp_init( x264_t *h )
 800 {
 801     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 802     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 803     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 804     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 805     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 806     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 807     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 808     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 809     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 810     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 811 }
 812
 813 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 814 {
 815     /* VUI */
 816     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 817     {
 818         int i_w = param->vui.i_sar_width;
 819         int i_h = param->vui.i_sar_height;
 820         int old_w = h->param.vui.i_sar_width;
 821         int old_h = h->param.vui.i_sar_height;
 822
 823         x264_reduce_fraction( &i_w, &i_h );
 824
 825         while( i_w > 65535 || i_h > 65535 )
 826         {
 827             i_w /= 2;
 828             i_h /= 2;
 829         }
 830
 831         x264_reduce_fraction( &i_w, &i_h );
 832
 833         if( i_w != old_w || i_h != old_h || initial )
 834         {
 835             h->param.vui.i_sar_width = 0;
 836             h->param.vui.i_sar_height = 0;
 837             if( i_w == 0 || i_h == 0 )
 838                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 839             else
 840             {
 841                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 842                 h->param.vui.i_sar_width = i_w;
 843                 h->param.vui.i_sar_height = i_h;
 844             }
 845         }
 846     }
 847 }
 848
 849 /****************************************************************************
 850  * x264_encoder_open:
 851  ****************************************************************************/
 852 x264_t *x264_encoder_open( x264_param_t *param )
 853 {
 854     x264_t *h;
 855     char buf[1000], *p;
 856     int qp, i_slicetype_length;
 857
 858     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 859
 860     /* Create a copy of param */
 861     memcpy( &h->param, param, sizeof(x264_param_t) );
 862
 863     if( param->param_free )
 864         param->param_free( param );
 865
 866     if( x264_validate_parameters( h ) < 0 )
 867         goto fail;
 868
 869     if( h->param.psz_cqm_file )
 870         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 871             goto fail;
 872
 873     if( h->param.rc.psz_stat_out )
 874         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 875     if( h->param.rc.psz_stat_in )
 876         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 877
 878     x264_set_aspect_ratio( h, &h->param, 1 );
 879
 880     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 881     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 882
 883     /* Init x264_t */
 884     h->i_frame = -1;
 885     h->i_frame_num = 0;
 886     h->i_idr_pic_id = 0;
 887     if( h->param.b_dts_compress )
 888     {
 889         /* h->i_dts_compress_multiplier == h->frames.i_bframe_delay + 1 */
 890         h->i_dts_compress_multiplier = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 3 : 2) : 1;
 891         if( h->i_dts_compress_multiplier != 1 )
 892         {
 893             x264_log( h, X264_LOG_DEBUG, "DTS compresion changed timebase: %d/%d -> %d/%d\n",
 894                       h->param.i_timebase_num, h->param.i_timebase_den,
 895                       h->param.i_timebase_num, h->param.i_timebase_den * h->i_dts_compress_multiplier );
 896             h->param.i_timebase_den *= h->i_dts_compress_multiplier;
 897         }
 898     }
 899     else
 900         h->i_dts_compress_multiplier = 1;
 901
 902     h->sps = &h->sps_array[0];
 903     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 904
 905     h->pps = &h->pps_array[0];
 906     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 907
 908     x264_validate_levels( h, 1 );
 909
 910     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 911
 912     if( x264_cqm_init( h ) < 0 )
 913         goto fail;
 914
 915     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 916
 917     /* Init frames. */
 918     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS )
 919         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 920     else
 921         h->frames.i_delay = h->param.i_bframe;
 922     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 923         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 924     i_slicetype_length = h->frames.i_delay;
 925     h->frames.i_delay += h->i_thread_frames - 1;
 926     h->frames.i_delay = X264_MIN( h->frames.i_delay, X264_LOOKAHEAD_MAX );
 927     h->frames.i_delay += h->param.i_sync_lookahead;
 928     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 929                                                  || h->param.rc.i_vbv_buffer_size);
 930     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 931
 932     h->frames.i_max_ref0 = h->param.i_frame_reference;
 933     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
 934     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 935     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 936         && ( h->param.rc.i_rc_method == X264_RC_ABR
 937           || h->param.rc.i_rc_method == X264_RC_CRF
 938           || h->param.i_bframe_adaptive
 939           || h->param.i_scenecut_threshold
 940           || h->param.rc.b_mb_tree
 941           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 942     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 943     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 944
 945     h->frames.i_last_keyframe = - h->param.i_keyint_max;
 946     h->frames.i_input    = 0;
 947     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
 948
 949     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
 950     /* Allocate room for max refs plus a few extra just in case. */
 951     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + 20) * sizeof(x264_frame_t *) );
 952     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
 953                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
 954     if( h->param.analyse.i_weighted_pred > 0 )
 955         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
 956     h->i_ref0 = 0;
 957     h->i_ref1 = 0;
 958     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = h->i_prev_duration = 0;
 959     h->i_disp_fields_last_frame = -1;
 960     x264_rdo_init();
 961
 962     /* init CPU functions */
 963     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
 964     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
 965     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
 966     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 967     if( !h->param.b_cabac )
 968         x264_init_vlc_tables();
 969     x264_pixel_init( h->param.cpu, &h->pixf );
 970     x264_dct_init( h->param.cpu, &h->dctf );
 971     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
 972     x264_mc_init( h->param.cpu, &h->mc );
 973     x264_quant_init( h, h->param.cpu, &h->quantf );
 974     x264_deblock_init( h->param.cpu, &h->loopf );
 975     x264_dct_init_weights();
 976
 977     mbcmp_init( h );
 978
 979     p = buf + sprintf( buf, "using cpu capabilities:" );
 980     for( int i = 0; x264_cpu_names[i].flags; i++ )
 981     {
 982         if( !strcmp(x264_cpu_names[i].name, "SSE2")
 983             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
 984             continue;
 985         if( !strcmp(x264_cpu_names[i].name, "SSE3")
 986             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
 987             continue;
 988         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
 989             && (h->param.cpu & X264_CPU_SSE42) )
 990             continue;
 991         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
 992             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
 993             p += sprintf( p, " %s", x264_cpu_names[i].name );
 994     }
 995     if( !h->param.cpu )
 996         p += sprintf( p, " none!" );
 997     x264_log( h, X264_LOG_INFO, "%s\n", buf );
 998
 999     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1000         if( x264_analyse_init_costs( h, qp ) )
1001             goto fail;
1002     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1003         goto fail;
1004     if( h->cost_mv[1][2013] != 24 )
1005     {
1006         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1007         goto fail;
1008     }
1009
1010     h->out.i_nal = 0;
1011     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1012         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1013           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1014
1015     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1016     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1017
1018     h->thread[0] = h;
1019     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1020         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1021
1022     if( x264_lookahead_init( h, i_slicetype_length ) )
1023         goto fail;
1024
1025     for( int i = 0; i < h->param.i_threads; i++ )
1026     {
1027         int init_nal_count = h->param.i_slice_count + 3;
1028         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1029         if( i > 0 )
1030             *h->thread[i] = *h;
1031
1032         if( allocate_threadlocal_data )
1033         {
1034             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1035             if( !h->thread[i]->fdec )
1036                 goto fail;
1037         }
1038         else
1039             h->thread[i]->fdec = h->thread[0]->fdec;
1040
1041         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1042         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1043         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1044         h->thread[i]->out.i_nals_allocated = init_nal_count;
1045
1046         if( allocate_threadlocal_data && x264_macroblock_cache_init( h->thread[i] ) < 0 )
1047             goto fail;
1048     }
1049
1050     /* Allocate scratch buffer */
1051     for( int i = 0; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1052     {
1053         int buf_hpel = (h->fdec->i_width[0]+48) * sizeof(int16_t);
1054         int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
1055         int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
1056         int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
1057             ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
1058         int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
1059         int buf_nnz = !h->param.b_cabac * h->pps->b_transform_8x8_mode * (h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t));
1060         int scratch_size = X264_MAX4( buf_hpel, buf_ssim, buf_tesa, X264_MAX( buf_mbtree, buf_nnz ) );
1061         CHECKED_MALLOC( h->thread[i]->scratch_buffer, scratch_size );
1062     }
1063
1064     if( x264_ratecontrol_new( h ) < 0 )
1065         goto fail;
1066
1067     if( h->param.i_nal_hrd )
1068     {
1069         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1070         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1071     }
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     int rc_reconfig = 0;
1108     h = h->thread[h->thread[0]->i_thread_phase];
1109     x264_set_aspect_ratio( h, param, 0 );
1110 #define COPY(var) h->param.var = param->var
1111     COPY( i_frame_reference ); // but never uses more refs than initially specified
1112     COPY( i_bframe_bias );
1113     if( h->param.i_scenecut_threshold )
1114         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1115     COPY( b_deblocking_filter );
1116     COPY( i_deblocking_filter_alphac0 );
1117     COPY( i_deblocking_filter_beta );
1118     COPY( analyse.inter );
1119     COPY( analyse.intra );
1120     COPY( analyse.i_direct_mv_pred );
1121     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1122     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1123         COPY( analyse.i_me_range );
1124     COPY( analyse.i_noise_reduction );
1125     /* We can't switch out of subme=0 during encoding. */
1126     if( h->param.analyse.i_subpel_refine )
1127         COPY( analyse.i_subpel_refine );
1128     COPY( analyse.i_trellis );
1129     COPY( analyse.b_chroma_me );
1130     COPY( analyse.b_dct_decimate );
1131     COPY( analyse.b_fast_pskip );
1132     COPY( analyse.b_mixed_references );
1133     COPY( analyse.f_psy_rd );
1134     COPY( analyse.f_psy_trellis );
1135     // can only twiddle these if they were enabled to begin with:
1136     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1137         COPY( analyse.i_me_method );
1138     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1139         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1140     if( h->pps->b_transform_8x8_mode )
1141         COPY( analyse.b_transform_8x8 );
1142     if( h->frames.i_max_ref1 > 1 )
1143         COPY( i_bframe_pyramid );
1144     COPY( i_slice_max_size );
1145     COPY( i_slice_max_mbs );
1146     COPY( i_slice_count );
1147
1148     /* VBV can't be turned on if it wasn't on to begin with */
1149     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1150           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1151     {
1152         COPY( rc.i_vbv_max_bitrate );
1153         COPY( rc.i_vbv_buffer_size );
1154         COPY( rc.i_bitrate );
1155         rc_reconfig = 1;
1156     }
1157     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1158     {
1159         COPY( rc.f_rf_constant );
1160         rc_reconfig = 1;
1161     }
1162     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1163     {
1164         COPY( rc.f_rf_constant_max );
1165         rc_reconfig = 1;
1166     }
1167
1168 #undef COPY
1169
1170     mbcmp_init( h );
1171
1172     int ret = x264_validate_parameters( h );
1173
1174     /* Supported reconfiguration options (1-pass only):
1175      * vbv-maxrate
1176      * vbv-bufsize
1177      * crf
1178      * bitrate (CBR only) */
1179     if( !ret && rc_reconfig )
1180         x264_ratecontrol_init_reconfigurable( h, 0 );
1181
1182     return ret;
1183 }
1184
1185 /****************************************************************************
1186  * x264_encoder_parameters:
1187  ****************************************************************************/
1188 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1189 {
1190     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1191 }
1192
1193 /* internal usage */
1194 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1195 {
1196     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1197
1198     nal->i_ref_idc = i_ref_idc;
1199     nal->i_type    = i_type;
1200
1201     nal->i_payload= 0;
1202     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1203 }
1204
1205 /* if number of allocated nals is not enough, re-allocate a larger one. */
1206 static int x264_nal_check_buffer( x264_t *h )
1207 {
1208     if( h->out.i_nal >= h->out.i_nals_allocated )
1209     {
1210         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1211         if( !new_out )
1212             return -1;
1213         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1214         x264_free( h->out.nal );
1215         h->out.nal = new_out;
1216         h->out.i_nals_allocated *= 2;
1217     }
1218     return 0;
1219 }
1220
1221 static int x264_nal_end( x264_t *h )
1222 {
1223     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1224     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1225     h->out.i_nal++;
1226
1227     return x264_nal_check_buffer( h );
1228 }
1229
1230 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1231 {
1232     int nal_size = 0, previous_nal_size = 0;
1233
1234     for( int i = 0; i < start; i++ )
1235         previous_nal_size += h->out.nal[i].i_payload;
1236
1237     for( int i = start; i < h->out.i_nal; i++ )
1238         nal_size += h->out.nal[i].i_payload;
1239
1240     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1241     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1242     {
1243         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1244         if( !buf )
1245             return -1;
1246         if( previous_nal_size )
1247             memcpy( buf, h->nal_buffer, previous_nal_size );
1248         x264_free( h->nal_buffer );
1249         h->nal_buffer = buf;
1250     }
1251
1252     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1253
1254     for( int i = start; i < h->out.i_nal; i++ )
1255     {
1256         int long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1257         int size = x264_nal_encode( nal_buffer, &h->out.nal[i], h->param.b_annexb, long_startcode );
1258         h->out.nal[i].i_payload = size;
1259         h->out.nal[i].p_payload = nal_buffer;
1260         nal_buffer += size;
1261     }
1262
1263     return nal_buffer - (h->nal_buffer + previous_nal_size);
1264 }
1265
1266 /****************************************************************************
1267  * x264_encoder_headers:
1268  ****************************************************************************/
1269 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1270 {
1271     int frame_size = 0;
1272     /* init bitstream context */
1273     h->out.i_nal = 0;
1274     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1275
1276     /* Write SEI, SPS and PPS. */
1277
1278     /* generate sequence parameters */
1279     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1280     x264_sps_write( &h->out.bs, h->sps );
1281     if( x264_nal_end( h ) )
1282         return -1;
1283
1284     /* generate picture parameters */
1285     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1286     x264_pps_write( &h->out.bs, h->pps );
1287     if( x264_nal_end( h ) )
1288         return -1;
1289
1290     /* identify ourselves */
1291     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1292     if( x264_sei_version_write( h, &h->out.bs ) )
1293         return -1;
1294     if( x264_nal_end( h ) )
1295         return -1;
1296
1297     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1298
1299     /* now set output*/
1300     *pi_nal = h->out.i_nal;
1301     *pp_nal = &h->out.nal[0];
1302     h->out.i_nal = 0;
1303
1304     return frame_size;
1305 }
1306
1307 /* Check to see whether we have chosen a reference list ordering different
1308  * from the standard's default. */
1309 static inline void x264_reference_check_reorder( x264_t *h )
1310 {
1311     for( int i = 0; i < h->i_ref0 - 1; i++ )
1312         /* P and B-frames use different default orders. */
1313         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1314                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1315         {
1316             h->b_ref_reorder[0] = 1;
1317             break;
1318         }
1319 }
1320
1321 /* return -1 on failure, else return the index of the new reference frame */
1322 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1323 {
1324     int i = h->i_ref0;
1325     int j;
1326     x264_frame_t *newframe;
1327     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1328         return -1;
1329
1330     /* Find a place to insert the duplicate in the reference list. */
1331     for( j = 0; j < i; j++ )
1332         if( h->fref0[i_ref]->i_frame != h->fref0[j]->i_frame )
1333         {
1334             /* found a place, after j, make sure there is not already a duplicate there */
1335             if( j == i-1 || ( h->fref0[j+1] && h->fref0[i_ref]->i_frame != h->fref0[j+1]->i_frame ) )
1336                 break;
1337         }
1338
1339     if( j == i ) /* No room in the reference list for the duplicate. */
1340         return -1;
1341     j++;
1342
1343     newframe = x264_frame_pop_blank_unused( h );
1344
1345     //FIXME: probably don't need to copy everything
1346     *newframe = *h->fref0[i_ref];
1347     newframe->i_reference_count = 1;
1348     newframe->orig = h->fref0[i_ref];
1349     newframe->b_duplicate = 1;
1350     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1351
1352     /* shift the frames to make space for the dupe. */
1353     h->b_ref_reorder[0] = 1;
1354     if( h->i_ref0 < 16 )
1355         ++h->i_ref0;
1356     h->fref0[15] = NULL;
1357     x264_frame_unshift( &h->fref0[j], newframe );
1358
1359     return j;
1360 }
1361
1362 static void x264_weighted_pred_init( x264_t *h )
1363 {
1364     /* for now no analysis and set all weights to nothing */
1365     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1366         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1367
1368     // FIXME: This only supports weighting of one reference frame
1369     // and duplicates of that frame.
1370     h->fenc->i_lines_weighted = 0;
1371
1372     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1373         for( int i = 0; i < 3; i++ )
1374             h->sh.weight[i_ref][i].weightfn = NULL;
1375
1376
1377     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1378         return;
1379
1380     int i_padv = PADV << h->param.b_interlaced;
1381     int denom = -1;
1382     int weightluma = 0;
1383     int buffer_next = 0;
1384     //FIXME: when chroma support is added, move this into loop
1385     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1386     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1387     for( int j = 0; j < h->i_ref0; j++ )
1388     {
1389         if( h->fenc->weight[j][0].weightfn )
1390         {
1391             h->sh.weight[j][0] = h->fenc->weight[j][0];
1392             // if weight is useless, don't write it to stream
1393             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1394                 h->sh.weight[j][0].weightfn = NULL;
1395             else
1396             {
1397                 if( !weightluma )
1398                 {
1399                     weightluma = 1;
1400                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1401                     assert( x264_clip3( denom, 0, 7 ) == denom );
1402                 }
1403                 assert( h->sh.weight[j][0].i_denom == denom );
1404                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1405                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1406                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1407                     h->fenc->i_stride[0] * i_padv + PADH;
1408             }
1409         }
1410
1411         //scale full resolution frame
1412         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1413         {
1414             uint8_t *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1415             uint8_t *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1416             int stride = h->fenc->i_stride[0];
1417             int width = h->fenc->i_width[0] + PADH*2;
1418             int height = h->fenc->i_lines[0] + i_padv*2;
1419             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1420             h->fenc->i_lines_weighted = height;
1421         }
1422     }
1423     if( !weightluma )
1424         h->sh.weight[0][0].i_denom = 0;
1425 }
1426
1427 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1428 {
1429     int b_ok;
1430
1431     /* build ref list 0/1 */
1432     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1433     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1434     if( h->sh.i_type == SLICE_TYPE_I )
1435         return;
1436
1437     for( int i = 0; h->frames.reference[i]; i++ )
1438     {
1439         if( h->frames.reference[i]->i_poc < i_poc )
1440             h->fref0[h->i_ref0++] = h->frames.reference[i];
1441         else if( h->frames.reference[i]->i_poc > i_poc )
1442             h->fref1[h->i_ref1++] = h->frames.reference[i];
1443     }
1444
1445     /* Order ref0 from higher to lower poc */
1446     do
1447     {
1448         b_ok = 1;
1449         for( int i = 0; i < h->i_ref0 - 1; i++ )
1450         {
1451             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1452             {
1453                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1454                 b_ok = 0;
1455                 break;
1456             }
1457         }
1458     } while( !b_ok );
1459
1460     if( h->sh.i_mmco_remove_from_end )
1461         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1462         {
1463             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1464             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1465             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1466         }
1467
1468     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1469     do
1470     {
1471         b_ok = 1;
1472         for( int i = 0; i < h->i_ref1 - 1; i++ )
1473         {
1474             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1475             {
1476                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1477                 b_ok = 0;
1478                 break;
1479             }
1480         }
1481     } while( !b_ok );
1482
1483     x264_reference_check_reorder( h );
1484
1485     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1486     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1487     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1488
1489     /* add duplicates */
1490     if( h->fenc->i_type == X264_TYPE_P )
1491     {
1492         int idx = -1;
1493         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1494         {
1495             x264_weight_t w[3];
1496             w[1].weightfn = w[2].weightfn = NULL;
1497             if( h->param.rc.b_stat_read )
1498                 x264_ratecontrol_set_weights( h, h->fenc );
1499
1500             if( !h->fenc->weight[0][0].weightfn )
1501             {
1502                 h->fenc->weight[0][0].i_denom = 0;
1503                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1504                 idx = x264_weighted_reference_duplicate( h, 0, w );
1505             }
1506             else
1507             {
1508                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1509                 {
1510                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1511                 }
1512                 x264_weighted_reference_duplicate( h, 0, weight_none );
1513                 if( h->fenc->weight[0][0].i_offset > -128 )
1514                 {
1515                     w[0] = h->fenc->weight[0][0];
1516                     w[0].i_offset--;
1517                     h->mc.weight_cache( h, &w[0] );
1518                     idx = x264_weighted_reference_duplicate( h, 0, w );
1519                 }
1520             }
1521         }
1522         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1523         {
1524             //weighted offset=-1
1525             x264_weight_t w[3];
1526             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1527             h->fenc->weight[0][0].i_denom = 0;
1528             w[1].weightfn = w[2].weightfn = NULL;
1529             idx = x264_weighted_reference_duplicate( h, 0, w );
1530         }
1531         h->mb.ref_blind_dupe = idx;
1532     }
1533
1534     assert( h->i_ref0 + h->i_ref1 <= 16 );
1535     h->mb.pic.i_fref[0] = h->i_ref0;
1536     h->mb.pic.i_fref[1] = h->i_ref1;
1537 }
1538
1539 static void x264_fdec_filter_row( x264_t *h, int mb_y )
1540 {
1541     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1542     int b_hpel = h->fdec->b_kept_as_ref;
1543     int b_deblock = !h->sh.i_disable_deblocking_filter_idc;
1544     int b_end = mb_y == h->sps->i_mb_height;
1545     int min_y = mb_y - (1 << h->sh.b_mbaff);
1546     int max_y = b_end ? h->sps->i_mb_height : mb_y;
1547     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1548     if( mb_y & h->sh.b_mbaff )
1549         return;
1550     if( min_y < 0 )
1551         return;
1552
1553     if( !b_end && !h->param.b_sliced_threads )
1554         for( int j = 0; j <= h->sh.b_mbaff; j++ )
1555             for( int i = 0; i < 3; i++ )
1556             {
1557                 memcpy( h->mb.intra_border_backup[j][i],
1558                         h->fdec->plane[i] + ((mb_y*16 >> !!i) + j - 1 - h->sh.b_mbaff) * h->fdec->i_stride[i],
1559                         h->sps->i_mb_width*16 >> !!i );
1560             }
1561
1562     if( b_deblock )
1563         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1564             x264_frame_deblock_row( h, y );
1565
1566     if( b_hpel )
1567     {
1568         x264_frame_expand_border( h, h->fdec, min_y, b_end );
1569         if( h->param.analyse.i_subpel_refine )
1570         {
1571             x264_frame_filter( h, h->fdec, min_y, b_end );
1572             x264_frame_expand_border_filtered( h, h->fdec, min_y, b_end );
1573         }
1574     }
1575
1576     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1577         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1578
1579     min_y = X264_MAX( min_y*16-8, 0 );
1580     max_y = b_end ? h->param.i_height : mb_y*16-8;
1581
1582     if( h->param.analyse.b_psnr )
1583         for( int i = 0; i < 3; i++ )
1584             h->stat.frame.i_ssd[i] +=
1585                 x264_pixel_ssd_wxh( &h->pixf,
1586                     h->fdec->plane[i] + (min_y>>!!i) * h->fdec->i_stride[i], h->fdec->i_stride[i],
1587                     h->fenc->plane[i] + (min_y>>!!i) * h->fenc->i_stride[i], h->fenc->i_stride[i],
1588                     h->param.i_width >> !!i, (max_y-min_y) >> !!i );
1589
1590     if( h->param.analyse.b_ssim )
1591     {
1592         x264_emms();
1593         /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1594          * and overlap by 4 */
1595         min_y += min_y == 0 ? 2 : -6;
1596         h->stat.frame.f_ssim +=
1597             x264_pixel_ssim_wxh( &h->pixf,
1598                 h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1599                 h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1600                 h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1601     }
1602 }
1603
1604 static inline int x264_reference_update( x264_t *h )
1605 {
1606     if( !h->fdec->b_kept_as_ref )
1607     {
1608         if( h->i_thread_frames > 1 )
1609         {
1610             x264_frame_push_unused( h, h->fdec );
1611             h->fdec = x264_frame_pop_unused( h, 1 );
1612             if( !h->fdec )
1613                 return -1;
1614         }
1615         return 0;
1616     }
1617
1618     /* apply mmco from previous frame. */
1619     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1620         for( int j = 0; h->frames.reference[j]; j++ )
1621             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1622                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1623
1624     /* move frame in the buffer */
1625     x264_frame_push( h->frames.reference, h->fdec );
1626     if( h->frames.reference[h->sps->i_num_ref_frames] )
1627         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1628     h->fdec = x264_frame_pop_unused( h, 1 );
1629     if( !h->fdec )
1630         return -1;
1631     return 0;
1632 }
1633
1634 static inline void x264_reference_reset( x264_t *h )
1635 {
1636     while( h->frames.reference[0] )
1637         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1638     h->fdec->i_poc =
1639     h->fenc->i_poc = 0;
1640 }
1641
1642 static inline void x264_reference_hierarchy_reset( x264_t *h )
1643 {
1644     int ref;
1645     int b_hasdelayframe = 0;
1646     if( !h->param.i_bframe_pyramid )
1647         return;
1648
1649     /* look for delay frames -- chain must only contain frames that are disposable */
1650     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1651         b_hasdelayframe |= h->frames.current[i]->i_coded
1652                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1653
1654     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe )
1655         return;
1656
1657     /* Remove last BREF. There will never be old BREFs in the
1658      * dpb during a BREF decode when pyramid == STRICT */
1659     for( ref = 0; h->frames.reference[ref]; ref++ )
1660     {
1661         if( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1662             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1663         {
1664             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1665             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1666             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1667             x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1668             h->b_ref_reorder[0] = 1;
1669             break;
1670         }
1671     }
1672
1673     /* Prepare to room in the dpb for the delayed display time of the later b-frame's */
1674     h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1675 }
1676
1677 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1678 {
1679     /* ------------------------ Create slice header  ----------------------- */
1680     if( i_nal_type == NAL_SLICE_IDR )
1681     {
1682         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1683
1684         /* increment id */
1685         h->i_idr_pic_id = ( h->i_idr_pic_id + 1 ) % 65536;
1686     }
1687     else
1688     {
1689         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1690
1691         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1692         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1693         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1694             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1695         {
1696             h->sh.b_num_ref_idx_override = 1;
1697         }
1698     }
1699
1700     h->fdec->i_frame_num = h->sh.i_frame_num;
1701
1702     if( h->sps->i_poc_type == 0 )
1703     {
1704         h->sh.i_poc = h->fdec->i_poc;
1705         if( h->param.b_interlaced )
1706         {
1707             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1708             if( h->sh.i_delta_poc_bottom == -1 )
1709                 h->sh.i_poc = h->fdec->i_poc + 1;
1710         }
1711         else
1712             h->sh.i_delta_poc_bottom = 0;
1713     }
1714     else if( h->sps->i_poc_type == 1 )
1715     {
1716         /* FIXME TODO FIXME */
1717     }
1718     else
1719     {
1720         /* Nothing to do ? */
1721     }
1722
1723     x264_macroblock_slice_init( h );
1724 }
1725
1726 static int x264_slice_write( x264_t *h )
1727 {
1728     int i_skip;
1729     int mb_xy, i_mb_x, i_mb_y;
1730     int i_skip_bak = 0; /* Shut up GCC. */
1731     bs_t bs_bak;
1732     x264_cabac_t cabac_bak;
1733     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1734     /* Assume no more than 3 bytes of NALU escaping.
1735      * NALUs other than the first use a 3-byte startcode. */
1736     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 3;
1737     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : INT_MAX;
1738     int starting_bits = bs_pos(&h->out.bs);
1739     bs_realign( &h->out.bs );
1740
1741     /* Slice */
1742     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1743
1744     /* Slice header */
1745     x264_macroblock_thread_init( h );
1746     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1747     if( h->param.b_cabac )
1748     {
1749         /* alignment needed */
1750         bs_align_1( &h->out.bs );
1751
1752         /* init cabac */
1753         x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
1754         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1755     }
1756     h->mb.i_last_qp = h->sh.i_qp;
1757     h->mb.i_last_dqp = 0;
1758
1759     i_mb_y = h->sh.i_first_mb / h->sps->i_mb_width;
1760     i_mb_x = h->sh.i_first_mb % h->sps->i_mb_width;
1761     i_skip = 0;
1762
1763     while( (mb_xy = i_mb_x + i_mb_y * h->sps->i_mb_width) <= h->sh.i_last_mb )
1764     {
1765         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1766         if( h->param.i_slice_max_size > 0 )
1767         {
1768             /* We don't need the contexts because flushing the CABAC encoder has no context
1769              * dependency and macroblocks are only re-encoded in the case where a slice is
1770              * ended (and thus the content of all contexts are thrown away). */
1771             if( h->param.b_cabac )
1772             {
1773                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1774                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1775                  * backed up. */
1776                 cabac_prevbyte_bak = h->cabac.p[-1];
1777             }
1778             else
1779             {
1780                 bs_bak = h->out.bs;
1781                 i_skip_bak = i_skip;
1782             }
1783         }
1784
1785         if( i_mb_x == 0 && !h->mb.b_reencode_mb && !h->param.b_sliced_threads )
1786             x264_fdec_filter_row( h, i_mb_y );
1787
1788         /* load cache */
1789         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1790
1791         x264_macroblock_analyse( h );
1792
1793         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1794         x264_macroblock_encode( h );
1795
1796         if( x264_bitstream_check_buffer( h ) )
1797             return -1;
1798
1799         if( h->param.b_cabac )
1800         {
1801             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1802                 x264_cabac_encode_terminal( &h->cabac );
1803
1804             if( IS_SKIP( h->mb.i_type ) )
1805                 x264_cabac_mb_skip( h, 1 );
1806             else
1807             {
1808                 if( h->sh.i_type != SLICE_TYPE_I )
1809                     x264_cabac_mb_skip( h, 0 );
1810                 x264_macroblock_write_cabac( h, &h->cabac );
1811             }
1812         }
1813         else
1814         {
1815             if( IS_SKIP( h->mb.i_type ) )
1816                 i_skip++;
1817             else
1818             {
1819                 if( h->sh.i_type != SLICE_TYPE_I )
1820                 {
1821                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1822                     i_skip = 0;
1823                 }
1824                 x264_macroblock_write_cavlc( h );
1825             }
1826         }
1827
1828         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1829         int mb_size = total_bits - mb_spos;
1830
1831         /* We'll just re-encode this last macroblock if we go over the max slice size. */
1832         if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1833         {
1834             if( mb_xy != h->sh.i_first_mb )
1835             {
1836                 if( h->param.b_cabac )
1837                 {
1838                     memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1839                     h->cabac.p[-1] = cabac_prevbyte_bak;
1840                 }
1841                 else
1842                 {
1843                     h->out.bs = bs_bak;
1844                     i_skip = i_skip_bak;
1845                 }
1846                 h->mb.b_reencode_mb = 1;
1847                 h->sh.i_last_mb = mb_xy-1;
1848                 break;
1849             }
1850             else
1851             {
1852                 h->sh.i_last_mb = mb_xy;
1853                 h->mb.b_reencode_mb = 0;
1854             }
1855         }
1856         else
1857             h->mb.b_reencode_mb = 0;
1858
1859 #ifdef HAVE_VISUALIZE
1860         if( h->param.b_visualize )
1861             x264_visualize_mb( h );
1862 #endif
1863
1864         /* save cache */
1865         x264_macroblock_cache_save( h );
1866
1867         /* accumulate mb stats */
1868         h->stat.frame.i_mb_count[h->mb.i_type]++;
1869
1870         if( !IS_INTRA(h->mb.i_type) && !IS_SKIP(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
1871         {
1872             if( h->mb.i_partition != D_8x8 )
1873                     h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
1874                 else
1875                     for( int i = 0; i < 4; i++ )
1876                         h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
1877             if( h->param.i_frame_reference > 1 )
1878                 for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
1879                     for( int i = 0; i < 4; i++ )
1880                     {
1881                         int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
1882                         if( i_ref >= 0 )
1883                             h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
1884                     }
1885         }
1886
1887         if( h->param.i_log_level >= X264_LOG_INFO )
1888         {
1889             if( h->mb.i_cbp_luma || h->mb.i_cbp_chroma )
1890             {
1891                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
1892                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
1893                 int b_intra = IS_INTRA(h->mb.i_type);
1894                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
1895                 h->stat.frame.i_mb_cbp[!b_intra + 2] += h->mb.i_cbp_chroma >= 1;
1896                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma == 2;
1897             }
1898             if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
1899             {
1900                 h->stat.frame.i_mb_count_8x8dct[0] ++;
1901                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
1902             }
1903             if( IS_INTRA(h->mb.i_type) && h->mb.i_type != I_PCM )
1904             {
1905                 if( h->mb.i_type == I_16x16 )
1906                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
1907                 else if( h->mb.i_type == I_8x8 )
1908                     for( int i = 0; i < 16; i += 4 )
1909                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1910                 else //if( h->mb.i_type == I_4x4 )
1911                     for( int i = 0; i < 16; i++ )
1912                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
1913             }
1914         }
1915
1916         x264_ratecontrol_mb( h, mb_size );
1917
1918         if( h->sh.b_mbaff )
1919         {
1920             i_mb_x += i_mb_y & 1;
1921             i_mb_y ^= i_mb_x < h->sps->i_mb_width;
1922         }
1923         else
1924             i_mb_x++;
1925         if( i_mb_x == h->sps->i_mb_width )
1926         {
1927             i_mb_y++;
1928             i_mb_x = 0;
1929         }
1930     }
1931
1932     if( h->param.b_cabac )
1933     {
1934         x264_cabac_encode_flush( h, &h->cabac );
1935         h->out.bs.p = h->cabac.p;
1936     }
1937     else
1938     {
1939         if( i_skip > 0 )
1940             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
1941         /* rbsp_slice_trailing_bits */
1942         bs_rbsp_trailing( &h->out.bs );
1943         bs_flush( &h->out.bs );
1944     }
1945     if( x264_nal_end( h ) )
1946         return -1;
1947
1948     if( h->sh.i_last_mb == h->mb.i_mb_count-1 )
1949     {
1950         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
1951                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
1952                                   - h->stat.frame.i_tex_bits
1953                                   - h->stat.frame.i_mv_bits;
1954         if( !h->param.b_sliced_threads )
1955             x264_fdec_filter_row( h, h->sps->i_mb_height );
1956     }
1957
1958     return 0;
1959 }
1960
1961 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
1962 {
1963     if( dst == src )
1964         return;
1965
1966     // reference counting
1967     for( x264_frame_t **f = src->frames.reference; *f; f++ )
1968         (*f)->i_reference_count++;
1969     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
1970         x264_frame_push_unused( src, *f );
1971     src->fdec->i_reference_count++;
1972     x264_frame_push_unused( src, dst->fdec );
1973
1974     // copy everything except the per-thread pointers and the constants.
1975     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
1976     dst->param = src->param;
1977     dst->stat = src->stat;
1978 }
1979
1980 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
1981 {
1982     if( dst == src )
1983         return;
1984     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
1985 }
1986
1987 static void *x264_slices_write( x264_t *h )
1988 {
1989     int i_slice_num = 0;
1990     int last_thread_mb = h->sh.i_last_mb;
1991     if( h->param.i_sync_lookahead )
1992         x264_lower_thread_priority( 10 );
1993
1994 #ifdef HAVE_MMX
1995     /* Misalign mask has to be set separately for each thread. */
1996     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
1997         x264_cpu_mask_misalign_sse();
1998 #endif
1999
2000 #ifdef HAVE_VISUALIZE
2001     if( h->param.b_visualize )
2002         if( x264_visualize_init( h ) )
2003             return (void *)-1;
2004 #endif
2005
2006     /* init stats */
2007     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2008     h->mb.b_reencode_mb = 0;
2009     while( h->sh.i_first_mb <= last_thread_mb )
2010     {
2011         h->sh.i_last_mb = last_thread_mb;
2012         if( h->param.i_slice_max_mbs )
2013             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2014         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2015         {
2016             int height = h->sps->i_mb_height >> h->param.b_interlaced;
2017             int width = h->sps->i_mb_width << h->param.b_interlaced;
2018             i_slice_num++;
2019             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2020         }
2021         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2022         if( x264_stack_align( x264_slice_write, h ) )
2023             return (void *)-1;
2024         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2025     }
2026
2027 #ifdef HAVE_VISUALIZE
2028     if( h->param.b_visualize )
2029     {
2030         x264_visualize_show( h );
2031         x264_visualize_close( h );
2032     }
2033 #endif
2034
2035     return (void *)0;
2036 }
2037
2038 static int x264_threaded_slices_write( x264_t *h )
2039 {
2040     void *ret = NULL;
2041     /* set first/last mb and sync contexts */
2042     for( int i = 0; i < h->param.i_threads; i++ )
2043     {
2044         x264_t *t = h->thread[i];
2045         if( i )
2046         {
2047             t->param = h->param;
2048             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2049         }
2050         int height = h->sps->i_mb_height >> h->param.b_interlaced;
2051         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2052         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2053         t->sh.i_first_mb = t->i_threadslice_start * h->sps->i_mb_width;
2054         t->sh.i_last_mb  =   t->i_threadslice_end * h->sps->i_mb_width - 1;
2055     }
2056
2057     x264_stack_align( x264_analyse_weight_frame, h, h->sps->i_mb_height*16 + 16 );
2058
2059     x264_threads_distribute_ratecontrol( h );
2060
2061     /* dispatch */
2062     for( int i = 0; i < h->param.i_threads; i++ )
2063     {
2064         if( x264_pthread_create( &h->thread[i]->thread_handle, NULL, (void*)x264_slices_write, (void*)h->thread[i] ) )
2065             return -1;
2066         h->thread[i]->b_thread_active = 1;
2067     }
2068     for( int i = 0; i < h->param.i_threads; i++ )
2069     {
2070         x264_pthread_join( h->thread[i]->thread_handle, &ret );
2071         h->thread[i]->b_thread_active = 0;
2072         if( (intptr_t)ret )
2073             return (intptr_t)ret;
2074     }
2075
2076     /* deblocking and hpel filtering */
2077     for( int i = 0; i <= h->sps->i_mb_height; i++ )
2078         x264_stack_align( x264_fdec_filter_row, h, i );
2079
2080     x264_threads_merge_ratecontrol( h );
2081
2082     for( int i = 1; i < h->param.i_threads; i++ )
2083     {
2084         x264_t *t = h->thread[i];
2085         for( int j = 0; j < t->out.i_nal; j++ )
2086         {
2087             h->out.nal[h->out.i_nal] = t->out.nal[j];
2088             h->out.i_nal++;
2089             x264_nal_check_buffer( h );
2090         }
2091         /* All entries in stat.frame are ints except for ssd/ssim,
2092          * which are only calculated in the main thread. */
2093         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2094             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2095     }
2096
2097     return 0;
2098 }
2099
2100 /****************************************************************************
2101  * x264_encoder_encode:
2102  *  XXX: i_poc   : is the poc of the current given picture
2103  *       i_frame : is the number of the frame being coded
2104  *  ex:  type frame poc
2105  *       I      0   2*0
2106  *       P      1   2*3
2107  *       B      2   2*1
2108  *       B      3   2*2
2109  *       P      4   2*6
2110  *       B      5   2*4
2111  *       B      6   2*5
2112  ****************************************************************************/
2113 int     x264_encoder_encode( x264_t *h,
2114                              x264_nal_t **pp_nal, int *pi_nal,
2115                              x264_picture_t *pic_in,
2116                              x264_picture_t *pic_out )
2117 {
2118     x264_t *thread_current, *thread_prev, *thread_oldest;
2119     int i_nal_type, i_nal_ref_idc, i_global_qp;
2120     int overhead = NALU_OVERHEAD;
2121
2122     if( h->i_thread_frames > 1 )
2123     {
2124         thread_prev    = h->thread[ h->i_thread_phase ];
2125         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2126         thread_current = h->thread[ h->i_thread_phase ];
2127         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2128         x264_thread_sync_context( thread_current, thread_prev );
2129         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2130         h = thread_current;
2131     }
2132     else
2133     {
2134         thread_current =
2135         thread_oldest  = h;
2136     }
2137
2138     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2139     if( x264_reference_update( h ) )
2140         return -1;
2141     h->fdec->i_lines_completed = -1;
2142
2143     /* no data out */
2144     *pi_nal = 0;
2145     *pp_nal = NULL;
2146
2147     /* ------------------- Setup new frame from picture -------------------- */
2148     if( pic_in != NULL )
2149     {
2150         /* 1: Copy the picture to a frame and move it to a buffer */
2151         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2152         if( !fenc )
2153             return -1;
2154
2155         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2156             return -1;
2157
2158         if( h->param.i_width != 16 * h->sps->i_mb_width ||
2159             h->param.i_height != 16 * h->sps->i_mb_height )
2160             x264_frame_expand_border_mod16( h, fenc );
2161
2162         fenc->i_frame = h->frames.i_input++;
2163
2164         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2165             h->frames.i_bframe_delay_time = fenc->i_pts;
2166
2167         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2168             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2169
2170         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2171         h->frames.i_largest_pts = fenc->i_pts;
2172
2173         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2174             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2175
2176         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2177         {
2178             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2179             if( b_interlaced )
2180             {
2181                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2182                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2183             }
2184             else
2185                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2186         }
2187
2188         if( h->frames.b_have_lowres )
2189         {
2190             if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2191                 x264_weight_plane_analyse( h, fenc );
2192             x264_frame_init_lowres( h, fenc );
2193         }
2194
2195         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2196         {
2197             if( x264_macroblock_tree_read( h, fenc ) )
2198                 return -1;
2199         }
2200         else if( h->param.rc.i_aq_mode )
2201             x264_adaptive_quant_frame( h, fenc );
2202
2203         /* 2: Place the frame into the queue for its slice type decision */
2204         x264_lookahead_put_frame( h, fenc );
2205
2206         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2207         {
2208             /* Nothing yet to encode, waiting for filling of buffers */
2209             pic_out->i_type = X264_TYPE_AUTO;
2210             return 0;
2211         }
2212     }
2213     else
2214     {
2215         /* signal kills for lookahead thread */
2216         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2217         h->lookahead->b_exit_thread = 1;
2218         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2219         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2220     }
2221
2222     h->i_frame++;
2223     /* 3: The picture is analyzed in the lookahead */
2224     if( !h->frames.current[0] )
2225         x264_lookahead_get_frames( h );
2226
2227     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2228         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2229
2230     /* ------------------- Get frame to be encoded ------------------------- */
2231     /* 4: get picture to encode */
2232     h->fenc = x264_frame_shift( h->frames.current );
2233     if( h->i_frame == 0 )
2234         h->first_pts = h->fenc->i_reordered_pts;
2235     if( h->fenc->param )
2236     {
2237         x264_encoder_reconfig( h, h->fenc->param );
2238         if( h->fenc->param->param_free )
2239             h->fenc->param->param_free( h->fenc->param );
2240     }
2241
2242     if( h->fenc->b_keyframe )
2243     {
2244         h->frames.i_last_keyframe = h->fenc->i_frame;
2245         if( h->fenc->i_type == X264_TYPE_IDR )
2246             h->i_frame_num = 0;
2247     }
2248     h->sh.i_mmco_command_count =
2249     h->sh.i_mmco_remove_from_end = 0;
2250     h->b_ref_reorder[0] =
2251     h->b_ref_reorder[1] = 0;
2252
2253     /* ------------------- Setup frame context ----------------------------- */
2254     /* 5: Init data dependent of frame type */
2255     if( h->fenc->i_type == X264_TYPE_IDR )
2256     {
2257         /* reset ref pictures */
2258         i_nal_type    = NAL_SLICE_IDR;
2259         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2260         h->sh.i_type = SLICE_TYPE_I;
2261         x264_reference_reset( h );
2262     }
2263     else if( h->fenc->i_type == X264_TYPE_I )
2264     {
2265         i_nal_type    = NAL_SLICE;
2266         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2267         h->sh.i_type = SLICE_TYPE_I;
2268         x264_reference_hierarchy_reset( h );
2269     }
2270     else if( h->fenc->i_type == X264_TYPE_P )
2271     {
2272         i_nal_type    = NAL_SLICE;
2273         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2274         h->sh.i_type = SLICE_TYPE_P;
2275         x264_reference_hierarchy_reset( h );
2276     }
2277     else if( h->fenc->i_type == X264_TYPE_BREF )
2278     {
2279         i_nal_type    = NAL_SLICE;
2280         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2281         h->sh.i_type = SLICE_TYPE_B;
2282         x264_reference_hierarchy_reset( h );
2283     }
2284     else    /* B frame */
2285     {
2286         i_nal_type    = NAL_SLICE;
2287         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2288         h->sh.i_type = SLICE_TYPE_B;
2289     }
2290
2291     h->fdec->i_poc =
2292     h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_keyframe);
2293     h->fdec->i_type = h->fenc->i_type;
2294     h->fdec->i_frame = h->fenc->i_frame;
2295     h->fenc->b_kept_as_ref =
2296     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2297
2298
2299
2300     /* ------------------- Init                ----------------------------- */
2301     /* build ref list 0/1 */
2302     x264_reference_build_list( h, h->fdec->i_poc );
2303
2304     /* ---------------------- Write the bitstream -------------------------- */
2305     /* Init bitstream context */
2306     if( h->param.b_sliced_threads )
2307     {
2308         for( int i = 0; i < h->param.i_threads; i++ )
2309         {
2310             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2311             h->thread[i]->out.i_nal = 0;
2312         }
2313     }
2314     else
2315     {
2316         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2317         h->out.i_nal = 0;
2318     }
2319
2320     if( h->param.b_aud )
2321     {
2322         int pic_type;
2323
2324         if( h->sh.i_type == SLICE_TYPE_I )
2325             pic_type = 0;
2326         else if( h->sh.i_type == SLICE_TYPE_P )
2327             pic_type = 1;
2328         else if( h->sh.i_type == SLICE_TYPE_B )
2329             pic_type = 2;
2330         else
2331             pic_type = 7;
2332
2333         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2334         bs_write( &h->out.bs, 3, pic_type );
2335         bs_rbsp_trailing( &h->out.bs );
2336         if( x264_nal_end( h ) )
2337             return -1;
2338         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2339     }
2340
2341     h->i_nal_type = i_nal_type;
2342     h->i_nal_ref_idc = i_nal_ref_idc;
2343
2344     if( h->param.b_intra_refresh && h->fenc->i_type == X264_TYPE_P )
2345     {
2346         int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2347         float increment = X264_MAX( ((float)h->sps->i_mb_width-1) / h->param.i_keyint_max, 1 );
2348         int max_position = (int)(increment * h->param.i_keyint_max);
2349         if( IS_X264_TYPE_I( h->fref0[0]->i_type ) )
2350             h->fdec->f_pir_position = 0;
2351         else
2352         {
2353             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2354             if( h->fdec->f_pir_position+0.5 >= max_position )
2355             {
2356                 h->fdec->f_pir_position = 0;
2357                 h->fenc->b_keyframe = 1;
2358             }
2359         }
2360         h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2361         h->fdec->f_pir_position += increment * pocdiff;
2362         h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2363     }
2364
2365     if( h->fenc->b_keyframe )
2366     {
2367         /* Write SPS and PPS */
2368         if( h->param.b_repeat_headers )
2369         {
2370             /* generate sequence parameters */
2371             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2372             x264_sps_write( &h->out.bs, h->sps );
2373             if( x264_nal_end( h ) )
2374                 return -1;
2375             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2376
2377             /* generate picture parameters */
2378             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2379             x264_pps_write( &h->out.bs, h->pps );
2380             if( x264_nal_end( h ) )
2381                 return -1;
2382             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2383         }
2384
2385         /* buffering period sei is written in x264_encoder_frame_end */
2386
2387         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2388         {
2389             /* identify ourself */
2390             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2391             if( x264_sei_version_write( h, &h->out.bs ) )
2392                 return -1;
2393             if( x264_nal_end( h ) )
2394                 return -1;
2395             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2396         }
2397
2398         if( h->fenc->i_type != X264_TYPE_IDR )
2399         {
2400             int time_to_recovery = X264_MIN( h->sps->i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe;
2401             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2402             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2403             x264_nal_end( h );
2404             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2405         }
2406     }
2407
2408     /* generate sei pic timing */
2409     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2410     {
2411         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2412         x264_sei_pic_timing_write( h, &h->out.bs );
2413         if( x264_nal_end( h ) )
2414             return -1;
2415         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2416     }
2417
2418     /* Init the rate control */
2419     /* FIXME: Include slice header bit cost. */
2420     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2421     i_global_qp = x264_ratecontrol_qp( h );
2422
2423     pic_out->i_qpplus1 =
2424     h->fdec->i_qpplus1 = i_global_qp + 1;
2425
2426     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2427     {
2428         x264_reference_build_list_optimal( h );
2429         x264_reference_check_reorder( h );
2430     }
2431
2432     if( h->i_ref0 )
2433         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2434
2435     if( h->sh.i_type == SLICE_TYPE_B )
2436         x264_macroblock_bipred_init( h );
2437
2438     /*------------------------- Weights -------------------------------------*/
2439     x264_weighted_pred_init( h );
2440
2441     /* ------------------------ Create slice header  ----------------------- */
2442     x264_slice_init( h, i_nal_type, i_global_qp );
2443
2444     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2445         h->i_frame_num++;
2446
2447     /* Write frame */
2448     h->i_threadslice_start = 0;
2449     h->i_threadslice_end = h->sps->i_mb_height;
2450     if( h->i_thread_frames > 1 )
2451     {
2452         if( x264_pthread_create( &h->thread_handle, NULL, (void*)x264_slices_write, h ) )
2453             return -1;
2454         h->b_thread_active = 1;
2455     }
2456     else if( h->param.b_sliced_threads )
2457     {
2458         if( x264_threaded_slices_write( h ) )
2459             return -1;
2460     }
2461     else
2462         if( (intptr_t)x264_slices_write( h ) )
2463             return -1;
2464
2465     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2466 }
2467
2468 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2469                                    x264_nal_t **pp_nal, int *pi_nal,
2470                                    x264_picture_t *pic_out )
2471 {
2472     char psz_message[80];
2473
2474     if( h->b_thread_active )
2475     {
2476         void *ret = NULL;
2477         x264_pthread_join( h->thread_handle, &ret );
2478         h->b_thread_active = 0;
2479         if( (intptr_t)ret )
2480             return (intptr_t)ret;
2481     }
2482     if( !h->out.i_nal )
2483     {
2484         pic_out->i_type = X264_TYPE_AUTO;
2485         return 0;
2486     }
2487
2488     x264_emms();
2489     /* generate sei buffering period and insert it into place */
2490     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2491     {
2492         h->initial_cpb_removal_delay = x264_hrd_fullness( h );
2493
2494         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2495         x264_sei_buffering_period_write( h, &h->out.bs );
2496         if( x264_nal_end( h ) )
2497            return -1;
2498         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2499         int idx = 0;
2500         while( h->out.nal[idx].i_type == NAL_AUD ||
2501                h->out.nal[idx].i_type == NAL_SPS ||
2502                h->out.nal[idx].i_type == NAL_PPS )
2503             idx++;
2504         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2505         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2506         h->out.nal[idx] = nal_tmp;
2507     }
2508
2509     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2510
2511     /* Set output picture properties */
2512     pic_out->i_type = h->fenc->i_type;
2513
2514     pic_out->b_keyframe = h->fenc->b_keyframe;
2515
2516     pic_out->i_pts = h->fenc->i_pts *= h->i_dts_compress_multiplier;
2517     if( h->frames.i_bframe_delay )
2518     {
2519         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2520         if( h->i_frame <= h->frames.i_bframe_delay )
2521         {
2522             if( h->i_dts_compress_multiplier == 1 )
2523                 pic_out->i_dts = h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2524             else
2525             {
2526                 /* DTS compression */
2527                 if( h->i_frame == 1 )
2528                     thread_current->frames.i_init_delta = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2529                 pic_out->i_dts = h->i_frame * thread_current->frames.i_init_delta / h->i_dts_compress_multiplier;
2530             }
2531         }
2532         else
2533             pic_out->i_dts = prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ];
2534         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts * h->i_dts_compress_multiplier;
2535     }
2536     else
2537         pic_out->i_dts = h->fenc->i_reordered_pts;
2538     if( pic_out->i_pts < pic_out->i_dts )
2539         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2540
2541     pic_out->img.i_plane = h->fdec->i_plane;
2542     for( int i = 0; i < 3; i++ )
2543     {
2544         pic_out->img.i_stride[i] = h->fdec->i_stride[i];
2545         pic_out->img.plane[i] = h->fdec->plane[i];
2546     }
2547
2548     x264_frame_push_unused( thread_current, h->fenc );
2549
2550     /* ---------------------- Update encoder state ------------------------- */
2551
2552     /* update rc */
2553     int filler = 0;
2554     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2555         return -1;
2556
2557     pic_out->hrd_timing = h->fenc->hrd_timing;
2558
2559     while( filler > 0 )
2560     {
2561         int f, overhead;
2562         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2563         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2564         {
2565             int next_size = filler - h->param.i_slice_max_size;
2566             int overflow = X264_MAX( overhead - next_size, 0 );
2567             f = h->param.i_slice_max_size - overhead - overflow;
2568         }
2569         else
2570             f = X264_MAX( 0, filler - overhead );
2571
2572         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2573         x264_filler_write( h, &h->out.bs, f );
2574         if( x264_nal_end( h ) )
2575             return -1;
2576         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2577         frame_size += total_size;
2578         filler -= total_size;
2579     }
2580
2581     /* End bitstream, set output  */
2582     *pi_nal = h->out.i_nal;
2583     *pp_nal = h->out.nal;
2584
2585     h->out.i_nal = 0;
2586
2587     x264_noise_reduction_update( h );
2588
2589     /* ---------------------- Compute/Print statistics --------------------- */
2590     x264_thread_sync_stat( h, h->thread[0] );
2591
2592     /* Slice stat */
2593     h->stat.i_frame_count[h->sh.i_type]++;
2594     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2595     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2596
2597     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2598         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2599     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2600         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2601     for( int i = 0; i < 2; i++ )
2602         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2603     for( int i = 0; i < 6; i++ )
2604         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2605     for( int i = 0; i < 3; i++ )
2606         for( int j = 0; j < 13; j++ )
2607             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2608     if( h->sh.i_type != SLICE_TYPE_I )
2609         for( int i_list = 0; i_list < 2; i_list++ )
2610             for( int i = 0; i < 32; i++ )
2611                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2612     if( h->sh.i_type == SLICE_TYPE_P )
2613     {
2614         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2615         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2616             for( int i = 0; i < 3; i++ )
2617                 for( int j = 0; j < h->i_ref0; j++ )
2618                     if( h->sh.weight[0][i].i_denom != 0 )
2619                     {
2620                         h->stat.i_wpred[i]++;
2621                         break;
2622                     }
2623     }
2624     if( h->sh.i_type == SLICE_TYPE_B )
2625     {
2626         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2627         if( h->mb.b_direct_auto_write )
2628         {
2629             //FIXME somewhat arbitrary time constants
2630             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2631                 for( int i = 0; i < 2; i++ )
2632                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2633             for( int i = 0; i < 2; i++ )
2634                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2635         }
2636     }
2637
2638     psz_message[0] = '\0';
2639     if( h->param.analyse.b_psnr )
2640     {
2641         int64_t ssd[3] = {
2642             h->stat.frame.i_ssd[0],
2643             h->stat.frame.i_ssd[1],
2644             h->stat.frame.i_ssd[2],
2645         };
2646
2647         h->stat.i_ssd_global[h->sh.i_type] += ssd[0] + ssd[1] + ssd[2];
2648         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 );
2649         h->stat.f_psnr_mean_y[h->sh.i_type] += x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2650         h->stat.f_psnr_mean_u[h->sh.i_type] += x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2651         h->stat.f_psnr_mean_v[h->sh.i_type] += x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2652
2653         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2654                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2655                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2656                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2657     }
2658
2659     if( h->param.analyse.b_ssim )
2660     {
2661         double ssim_y = h->stat.frame.f_ssim
2662                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2663         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y;
2664         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2665                   " SSIM Y:%.5f", ssim_y );
2666     }
2667     psz_message[79] = '\0';
2668
2669     x264_log( h, X264_LOG_DEBUG,
2670                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2671               h->i_frame,
2672               h->fdec->f_qp_avg_aq,
2673               h->i_nal_ref_idc,
2674               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2675               h->fdec->i_poc,
2676               h->stat.frame.i_mb_count_i,
2677               h->stat.frame.i_mb_count_p,
2678               h->stat.frame.i_mb_count_skip,
2679               frame_size,
2680               psz_message );
2681
2682     // keep stats all in one place
2683     x264_thread_sync_stat( h->thread[0], h );
2684     // for the use of the next frame
2685     x264_thread_sync_stat( thread_current, h );
2686
2687 #ifdef DEBUG_MB_TYPE
2688 {
2689     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2690         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2691     for( int mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
2692     {
2693         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2694             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2695         else
2696             fprintf( stderr, "? " );
2697
2698         if( (mb_xy+1) % h->sps->i_mb_width == 0 )
2699             fprintf( stderr, "\n" );
2700     }
2701 }
2702 #endif
2703
2704     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2705      * by freeing some of its pointers. */
2706      for( int i = 0; i < h->i_ref0; i++ )
2707          if( h->fref0[i] && h->fref0[i]->b_duplicate )
2708          {
2709              x264_frame_push_blank_unused( h, h->fref0[i] );
2710              h->fref0[i] = 0;
2711          }
2712
2713     if( h->param.psz_dump_yuv )
2714         x264_frame_dump( h );
2715
2716     return frame_size;
2717 }
2718
2719 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2720 {
2721     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2722         b_print_pcm ? "..PCM" : "",
2723         i_mb_count[I_16x16]/ i_count,
2724         i_mb_count[I_8x8]  / i_count,
2725         i_mb_count[I_4x4]  / i_count );
2726     if( b_print_pcm )
2727         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2728 }
2729
2730 /****************************************************************************
2731  * x264_encoder_close:
2732  ****************************************************************************/
2733 void    x264_encoder_close  ( x264_t *h )
2734 {
2735     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2736     int64_t i_mb_count_size[2][7] = {{0}};
2737     char buf[200];
2738     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2739                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2740                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2741
2742     x264_lookahead_delete( h );
2743
2744     if( h->param.i_threads > 1 )
2745     {
2746         // don't strictly have to wait for the other threads, but it's simpler than canceling them
2747         for( int i = 0; i < h->param.i_threads; i++ )
2748             if( h->thread[i]->b_thread_active )
2749                 x264_pthread_join( h->thread[i]->thread_handle, NULL );
2750         if( h->i_thread_frames > 1 )
2751         {
2752             for( int i = 0; i < h->i_thread_frames; i++ )
2753                 if( h->thread[i]->b_thread_active )
2754                 {
2755                     assert( h->thread[i]->fenc->i_reference_count == 1 );
2756                     x264_frame_delete( h->thread[i]->fenc );
2757                 }
2758
2759             x264_t *thread_prev = h->thread[h->i_thread_phase];
2760             x264_thread_sync_ratecontrol( h, thread_prev, h );
2761             x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2762             h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2763         }
2764     }
2765     h->i_frame++;
2766
2767     /* Slices used and PSNR */
2768     for( int i = 0; i < 5; i++ )
2769     {
2770         static const int slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_SI, SLICE_TYPE_P, SLICE_TYPE_SP, SLICE_TYPE_B };
2771         static const char *slice_name[] = { "P", "B", "I", "SP", "SI" };
2772         int i_slice = slice_order[i];
2773
2774         if( h->stat.i_frame_count[i_slice] > 0 )
2775         {
2776             const int i_count = h->stat.i_frame_count[i_slice];
2777             if( h->param.analyse.b_psnr )
2778             {
2779                 x264_log( h, X264_LOG_INFO,
2780                           "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",
2781                           slice_name[i_slice],
2782                           i_count,
2783                           h->stat.f_frame_qp[i_slice] / i_count,
2784                           (double)h->stat.i_frame_size[i_slice] / i_count,
2785                           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,
2786                           h->stat.f_psnr_average[i_slice] / i_count,
2787                           x264_psnr( h->stat.i_ssd_global[i_slice], i_count * i_yuv_size ) );
2788             }
2789             else
2790             {
2791                 x264_log( h, X264_LOG_INFO,
2792                           "frame %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
2793                           slice_name[i_slice],
2794                           i_count,
2795                           h->stat.f_frame_qp[i_slice] / i_count,
2796                           (double)h->stat.i_frame_size[i_slice] / i_count );
2797             }
2798         }
2799     }
2800     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
2801     {
2802         char *p = buf;
2803         int den = 0;
2804         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
2805         for( int i = 0; i <= h->param.i_bframe; i++ )
2806             den += (i+1) * h->stat.i_consecutive_bframes[i];
2807         for( int i = 0; i <= h->param.i_bframe; i++ )
2808             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
2809         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
2810     }
2811
2812     for( int i_type = 0; i_type < 2; i_type++ )
2813         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2814         {
2815             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
2816             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
2817         }
2818
2819     /* MB types used */
2820     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
2821     {
2822         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
2823         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
2824         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2825         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
2826     }
2827     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2828     {
2829         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
2830         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
2831         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
2832         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2833         x264_log( h, X264_LOG_INFO,
2834                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
2835                   buf,
2836                   i_mb_size[PIXEL_16x16] / (i_count*4),
2837                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2838                   i_mb_size[PIXEL_8x8] / (i_count*4),
2839                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
2840                   i_mb_size[PIXEL_4x4] / (i_count*4),
2841                   i_mb_count[P_SKIP] / i_count );
2842     }
2843     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
2844     {
2845         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
2846         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
2847         double i_mb_list_count;
2848         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
2849         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
2850         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
2851         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2852             for( int j = 0; j < 2; j++ )
2853             {
2854                 int l0 = x264_mb_type_list_table[i][0][j];
2855                 int l1 = x264_mb_type_list_table[i][1][j];
2856                 if( l0 || l1 )
2857                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
2858             }
2859         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
2860         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
2861         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
2862         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
2863         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
2864         x264_log( h, X264_LOG_INFO,
2865                   "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",
2866                   buf,
2867                   i_mb_size[PIXEL_16x16] / (i_count*4),
2868                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
2869                   i_mb_size[PIXEL_8x8] / (i_count*4),
2870                   i_mb_count[B_DIRECT] / i_count,
2871                   i_mb_count[B_SKIP]   / i_count,
2872                   list_count[0] / i_mb_list_count,
2873                   list_count[1] / i_mb_list_count,
2874                   list_count[2] / i_mb_list_count );
2875     }
2876
2877     x264_ratecontrol_summary( h );
2878
2879     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 )
2880     {
2881 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
2882 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
2883         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
2884         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
2885                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
2886         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
2887         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
2888                             h->stat.i_frame_count[SLICE_TYPE_P] +
2889                             h->stat.i_frame_count[SLICE_TYPE_B];
2890         int64_t i_mb_count = i_count * h->mb.i_mb_count;
2891         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
2892         float f_bitrate;
2893         /* duration algorithm fails with one frame */
2894         if( i_count == 1 )
2895             f_bitrate = fps * SUM3(h->stat.i_frame_size) / i_count / 125;
2896         else
2897         {
2898             float duration = (float)(2 * h->frames.i_largest_pts - h->frames.i_second_largest_pts) * h->param.i_timebase_num / h->param.i_timebase_den;
2899             f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
2900         }
2901
2902         if( h->pps->b_transform_8x8_mode )
2903         {
2904             buf[0] = 0;
2905             if( h->stat.i_mb_count_8x8dct[0] )
2906                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
2907             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
2908         }
2909
2910         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
2911             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
2912             && h->stat.i_frame_count[SLICE_TYPE_B] )
2913         {
2914             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
2915                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
2916                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
2917         }
2918
2919         buf[0] = 0;
2920         if( i_mb_count != i_all_intra )
2921             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
2922                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
2923                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
2924                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
2925         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
2926                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
2927                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
2928                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
2929
2930         int64_t fixed_pred_modes[3][9] = {{0}};
2931         int64_t sum_pred_modes[3] = {0};
2932         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
2933         {
2934             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
2935             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
2936         }
2937         if( sum_pred_modes[0] )
2938             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
2939                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
2940                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
2941                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
2942                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
2943         for( int i = 1; i <= 2; i++ )
2944         {
2945             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
2946             {
2947                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
2948                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
2949             }
2950             if( sum_pred_modes[i] )
2951                 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,
2952                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
2953                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
2954                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
2955                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
2956                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
2957                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
2958                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
2959                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
2960                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
2961         }
2962
2963         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
2964             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
2965                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
2966
2967         for( int i_list = 0; i_list < 2; i_list++ )
2968             for( int i_slice = 0; i_slice < 2; i_slice++ )
2969             {
2970                 char *p = buf;
2971                 int64_t i_den = 0;
2972                 int i_max = 0;
2973                 for( int i = 0; i < 32; i++ )
2974                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
2975                     {
2976                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
2977                         i_max = i;
2978                     }
2979                 if( i_max == 0 )
2980                     continue;
2981                 for( int i = 0; i <= i_max; i++ )
2982                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
2983                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
2984             }
2985
2986         if( h->param.analyse.b_ssim )
2987         {
2988             x264_log( h, X264_LOG_INFO,
2989                       "SSIM Mean Y:%.7f\n",
2990                       SUM3( h->stat.f_ssim_mean_y ) / i_count );
2991         }
2992         if( h->param.analyse.b_psnr )
2993         {
2994             x264_log( h, X264_LOG_INFO,
2995                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
2996                       SUM3( h->stat.f_psnr_mean_y ) / i_count,
2997                       SUM3( h->stat.f_psnr_mean_u ) / i_count,
2998                       SUM3( h->stat.f_psnr_mean_v ) / i_count,
2999                       SUM3( h->stat.f_psnr_average ) / i_count,
3000                       x264_psnr( SUM3( h->stat.i_ssd_global ), i_count * i_yuv_size ),
3001                       f_bitrate );
3002         }
3003         else
3004             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3005     }
3006
3007     /* rc */
3008     x264_ratecontrol_delete( h );
3009
3010     /* param */
3011     if( h->param.rc.psz_stat_out )
3012         free( h->param.rc.psz_stat_out );
3013     if( h->param.rc.psz_stat_in )
3014         free( h->param.rc.psz_stat_in );
3015
3016     x264_cqm_delete( h );
3017     x264_free( h->nal_buffer );
3018     x264_analyse_free_costs( h );
3019
3020     if( h->i_thread_frames > 1)
3021         h = h->thread[h->i_thread_phase];
3022
3023     /* frames */
3024     x264_frame_delete_list( h->frames.unused[0] );
3025     x264_frame_delete_list( h->frames.unused[1] );
3026     x264_frame_delete_list( h->frames.current );
3027     x264_frame_delete_list( h->frames.blank_unused );
3028
3029     h = h->thread[0];
3030
3031     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3032     {
3033         x264_frame_t **frame;
3034
3035         if( !h->param.b_sliced_threads || i == 0 )
3036         {
3037             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3038             {
3039                 assert( (*frame)->i_reference_count > 0 );
3040                 (*frame)->i_reference_count--;
3041                 if( (*frame)->i_reference_count == 0 )
3042                     x264_frame_delete( *frame );
3043             }
3044             frame = &h->thread[i]->fdec;
3045             assert( (*frame)->i_reference_count > 0 );
3046             (*frame)->i_reference_count--;
3047             if( (*frame)->i_reference_count == 0 )
3048                 x264_frame_delete( *frame );
3049             x264_macroblock_cache_end( h->thread[i] );
3050         }
3051         x264_free( h->thread[i]->scratch_buffer );
3052         x264_free( h->thread[i]->out.p_bitstream );
3053         x264_free( h->thread[i]->out.nal);
3054         x264_free( h->thread[i] );
3055     }
3056 }
3057
3058 /****************************************************************************
3059  * x264_encoder_delayed_frames:
3060  ****************************************************************************/
3061 int x264_encoder_delayed_frames( x264_t *h )
3062 {
3063     int delayed_frames = 0;
3064     if( h->i_thread_frames > 1 )
3065     {
3066         for( int i = 0; i < h->i_thread_frames; i++ )
3067             delayed_frames += h->thread[i]->b_thread_active;
3068         h = h->thread[h->i_thread_phase];
3069     }
3070     for( int i = 0; h->frames.current[i]; i++ )
3071         delayed_frames++;
3072     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3073     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3074     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3075     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3076     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3077     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3078     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3079     return delayed_frames;
3080 }