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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2010 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  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include <math.h>
  29
  30 #include "common/common.h"
  31
  32 #include "set.h"
  33 #include "analyse.h"
  34 #include "ratecontrol.h"
  35 #include "macroblock.h"
  36 #include "me.h"
  37
  38 #if HAVE_VISUALIZE
  39 #include "common/visualize.h"
  40 #endif
  41
  42 //#define DEBUG_MB_TYPE
  43
  44 #define bs_write_ue bs_write_ue_big
  45
  46 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  47                                    x264_nal_t **pp_nal, int *pi_nal,
  48                                    x264_picture_t *pic_out );
  49
  50 /****************************************************************************
  51  *
  52  ******************************* x264 libs **********************************
  53  *
  54  ****************************************************************************/
  55 static double x264_psnr( double sqe, double size )
  56 {
  57     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  58     if( mse <= 0.0000000001 ) /* Max 100dB */
  59         return 100;
  60
  61     return -10.0 * log10( mse );
  62 }
  63
  64 static double x264_ssim( double ssim )
  65 {
  66     return -10.0 * log10( 1 - ssim );
  67 }
  68
  69 static void x264_frame_dump( x264_t *h )
  70 {
  71     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  72     if( !f )
  73         return;
  74     /* Write the frame in display order */
  75     fseek( f, (uint64_t)h->fdec->i_frame * h->param.i_height * h->param.i_width * 3/2 * sizeof(pixel), SEEK_SET );
  76     for( int y = 0; y < h->param.i_height; y++ )
  77         fwrite( &h->fdec->plane[0][y*h->fdec->i_stride[0]], sizeof(pixel), h->param.i_width, f );
  78     int cw = h->param.i_width>>1;
  79     int ch = h->param.i_height>>1;
  80     pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
  81     pixel *planev = planeu + cw*ch + 16;
  82     h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
  83     fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
  84     fwrite( planev, 1, cw*ch*sizeof(pixel), f );
  85     x264_free( planeu );
  86     fclose( f );
  87 }
  88
  89
  90 /* Fill "default" values */
  91 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  92                                     x264_sps_t *sps, x264_pps_t *pps,
  93                                     int i_idr_pic_id, int i_frame, int i_qp )
  94 {
  95     x264_param_t *param = &h->param;
  96
  97     /* First we fill all fields */
  98     sh->sps = sps;
  99     sh->pps = pps;
 100
 101     sh->i_first_mb  = 0;
 102     sh->i_last_mb   = h->mb.i_mb_count - 1;
 103     sh->i_pps_id    = pps->i_id;
 104
 105     sh->i_frame_num = i_frame;
 106
 107     sh->b_mbaff = h->param.b_interlaced;
 108     sh->b_field_pic = 0;    /* no field support for now */
 109     sh->b_bottom_field = 0; /* not yet used */
 110
 111     sh->i_idr_pic_id = i_idr_pic_id;
 112
 113     /* poc stuff, fixed later */
 114     sh->i_poc = 0;
 115     sh->i_delta_poc_bottom = 0;
 116     sh->i_delta_poc[0] = 0;
 117     sh->i_delta_poc[1] = 0;
 118
 119     sh->i_redundant_pic_cnt = 0;
 120
 121     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 122                                 && h->param.i_bframe
 123                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 124
 125     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 126     {
 127         if( h->fref1[0]->i_poc_l0ref0 == h->fref0[0]->i_poc )
 128         {
 129             if( h->mb.b_direct_auto_write )
 130                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 131             else
 132                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 133         }
 134         else
 135         {
 136             h->mb.b_direct_auto_write = 0;
 137             sh->b_direct_spatial_mv_pred = 1;
 138         }
 139     }
 140     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 141
 142     sh->b_num_ref_idx_override = 0;
 143     sh->i_num_ref_idx_l0_active = 1;
 144     sh->i_num_ref_idx_l1_active = 1;
 145
 146     sh->b_ref_pic_list_reordering_l0 = h->b_ref_reorder[0];
 147     sh->b_ref_pic_list_reordering_l1 = h->b_ref_reorder[1];
 148
 149     /* If the ref list isn't in the default order, construct reordering header */
 150     /* List1 reordering isn't needed yet */
 151     if( sh->b_ref_pic_list_reordering_l0 )
 152     {
 153         int pred_frame_num = i_frame;
 154         for( int i = 0; i < h->i_ref0; i++ )
 155         {
 156             int diff = h->fref0[i]->i_frame_num - pred_frame_num;
 157             sh->ref_pic_list_order[0][i].idc = ( diff > 0 );
 158             sh->ref_pic_list_order[0][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 159             pred_frame_num = h->fref0[i]->i_frame_num;
 160         }
 161     }
 162
 163     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 164
 165     sh->i_qp = i_qp;
 166     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
 167     sh->b_sp_for_swidth = 0;
 168     sh->i_qs_delta = 0;
 169
 170     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 171     /* If effective qp <= 15, deblocking would have no effect anyway */
 172     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 173         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 174     else
 175         sh->i_disable_deblocking_filter_idc = 1;
 176     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 177     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 178 }
 179
 180 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 181 {
 182     if( sh->b_mbaff )
 183     {
 184         assert( sh->i_first_mb % (2*sh->sps->i_mb_width) == 0 );
 185         bs_write_ue( s, sh->i_first_mb >> 1 );
 186     }
 187     else
 188         bs_write_ue( s, sh->i_first_mb );
 189
 190     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 191     bs_write_ue( s, sh->i_pps_id );
 192     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 193
 194     if( !sh->sps->b_frame_mbs_only )
 195     {
 196         bs_write1( s, sh->b_field_pic );
 197         if( sh->b_field_pic )
 198             bs_write1( s, sh->b_bottom_field );
 199     }
 200
 201     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 202         bs_write_ue( s, sh->i_idr_pic_id );
 203
 204     if( sh->sps->i_poc_type == 0 )
 205     {
 206         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 207         if( sh->pps->b_pic_order && !sh->b_field_pic )
 208             bs_write_se( s, sh->i_delta_poc_bottom );
 209     }
 210     else if( sh->sps->i_poc_type == 1 && !sh->sps->b_delta_pic_order_always_zero )
 211     {
 212         bs_write_se( s, sh->i_delta_poc[0] );
 213         if( sh->pps->b_pic_order && !sh->b_field_pic )
 214             bs_write_se( s, sh->i_delta_poc[1] );
 215     }
 216
 217     if( sh->pps->b_redundant_pic_cnt )
 218         bs_write_ue( s, sh->i_redundant_pic_cnt );
 219
 220     if( sh->i_type == SLICE_TYPE_B )
 221         bs_write1( s, sh->b_direct_spatial_mv_pred );
 222
 223     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 224     {
 225         bs_write1( s, sh->b_num_ref_idx_override );
 226         if( sh->b_num_ref_idx_override )
 227         {
 228             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 229             if( sh->i_type == SLICE_TYPE_B )
 230                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 231         }
 232     }
 233
 234     /* ref pic list reordering */
 235     if( sh->i_type != SLICE_TYPE_I )
 236     {
 237         bs_write1( s, sh->b_ref_pic_list_reordering_l0 );
 238         if( sh->b_ref_pic_list_reordering_l0 )
 239         {
 240             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 241             {
 242                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 243                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 244             }
 245             bs_write_ue( s, 3 );
 246         }
 247     }
 248     if( sh->i_type == SLICE_TYPE_B )
 249     {
 250         bs_write1( s, sh->b_ref_pic_list_reordering_l1 );
 251         if( sh->b_ref_pic_list_reordering_l1 )
 252         {
 253             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 254             {
 255                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 256                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 257             }
 258             bs_write_ue( s, 3 );
 259         }
 260     }
 261
 262     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 263     {
 264         /* pred_weight_table() */
 265         bs_write_ue( s, sh->weight[0][0].i_denom );
 266         bs_write_ue( s, sh->weight[0][1].i_denom );
 267         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 268         {
 269             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 270             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 271             bs_write1( s, luma_weight_l0_flag );
 272             if( luma_weight_l0_flag )
 273             {
 274                 bs_write_se( s, sh->weight[i][0].i_scale );
 275                 bs_write_se( s, sh->weight[i][0].i_offset );
 276             }
 277             bs_write1( s, chroma_weight_l0_flag );
 278             if( chroma_weight_l0_flag )
 279             {
 280                 for( int j = 1; j < 3; j++ )
 281                 {
 282                     bs_write_se( s, sh->weight[i][j].i_scale );
 283                     bs_write_se( s, sh->weight[i][j].i_offset );
 284                 }
 285             }
 286         }
 287     }
 288     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 289     {
 290       /* TODO */
 291     }
 292
 293     if( i_nal_ref_idc != 0 )
 294     {
 295         if( sh->i_idr_pic_id >= 0 )
 296         {
 297             bs_write1( s, 0 );  /* no output of prior pics flag */
 298             bs_write1( s, 0 );  /* long term reference flag */
 299         }
 300         else
 301         {
 302             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 303             if( sh->i_mmco_command_count > 0 )
 304             {
 305                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 306                 {
 307                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 308                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 309                 }
 310                 bs_write_ue( s, 0 ); /* end command list */
 311             }
 312         }
 313     }
 314
 315     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 316         bs_write_ue( s, sh->i_cabac_init_idc );
 317
 318     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 319
 320     if( sh->pps->b_deblocking_filter_control )
 321     {
 322         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 323         if( sh->i_disable_deblocking_filter_idc != 1 )
 324         {
 325             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 326             bs_write_se( s, sh->i_beta_offset >> 1 );
 327         }
 328     }
 329 }
 330
 331 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 332 /* reallocate, adding an arbitrary amount of space (100 kilobytes). */
 333 static int x264_bitstream_check_buffer( x264_t *h )
 334 {
 335     uint8_t *bs_bak = h->out.p_bitstream;
 336     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < 2500)) ||
 337         (h->out.bs.p_end - h->out.bs.p < 2500) )
 338     {
 339         h->out.i_bitstream += 100000;
 340         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 341         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - 100000) & ~15 );
 342         intptr_t delta = h->out.p_bitstream - bs_bak;
 343
 344         h->out.bs.p_start += delta;
 345         h->out.bs.p += delta;
 346         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 347
 348         h->cabac.p_start += delta;
 349         h->cabac.p += delta;
 350         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 351
 352         for( int i = 0; i <= h->out.i_nal; i++ )
 353             h->out.nal[i].p_payload += delta;
 354         x264_free( bs_bak );
 355     }
 356     return 0;
 357 fail:
 358     x264_free( bs_bak );
 359     return -1;
 360 }
 361
 362 #if HAVE_PTHREAD
 363 static void x264_encoder_thread_init( x264_t *h )
 364 {
 365     if( h->param.i_sync_lookahead )
 366         x264_lower_thread_priority( 10 );
 367
 368 #if HAVE_MMX
 369     /* Misalign mask has to be set separately for each thread. */
 370     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 371         x264_cpu_mask_misalign_sse();
 372 #endif
 373 }
 374 #endif
 375
 376 /****************************************************************************
 377  *
 378  ****************************************************************************
 379  ****************************** External API*********************************
 380  ****************************************************************************
 381  *
 382  ****************************************************************************/
 383
 384 static int x264_validate_parameters( x264_t *h )
 385 {
 386 #if HAVE_MMX
 387 #ifdef __SSE__
 388     if( !(x264_cpu_detect() & X264_CPU_SSE) )
 389     {
 390         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 391 #else
 392     if( !(x264_cpu_detect() & X264_CPU_MMXEXT) )
 393     {
 394         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 395 #endif
 396         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 397         return -1;
 398     }
 399 #endif
 400     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 401     {
 402         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 403                   h->param.i_width, h->param.i_height );
 404         return -1;
 405     }
 406
 407     if( h->param.i_width % 2 || h->param.i_height % 2 )
 408     {
 409         x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
 410                   h->param.i_width, h->param.i_height );
 411         return -1;
 412     }
 413     int i_csp = h->param.i_csp & X264_CSP_MASK;
 414     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 415     {
 416         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12 supported)\n" );
 417         return -1;
 418     }
 419
 420     if( h->param.i_threads == X264_THREADS_AUTO )
 421         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 422     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 423     if( h->param.i_threads > 1 )
 424     {
 425 #if !HAVE_PTHREAD
 426         x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
 427         h->param.i_threads = 1;
 428 #endif
 429         /* Avoid absurdly small thread slices as they can reduce performance
 430          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 431         if( h->param.b_sliced_threads )
 432         {
 433             int max_threads = (h->param.i_height+15)/16 / 4;
 434             h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 435         }
 436     }
 437     else
 438         h->param.b_sliced_threads = 0;
 439     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 440     if( h->i_thread_frames > 1 )
 441         h->param.nalu_process = NULL;
 442
 443     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 444     if( h->param.i_keyint_max == 1 )
 445     {
 446         h->param.b_intra_refresh = 0;
 447         h->param.analyse.i_weighted_pred = 0;
 448     }
 449
 450     if( h->param.b_interlaced )
 451     {
 452         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 453         {
 454             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 455             h->param.analyse.i_me_method = X264_ME_UMH;
 456         }
 457         if( h->param.analyse.i_weighted_pred > 0 )
 458         {
 459             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 460             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 461         }
 462     }
 463
 464     /* Detect default ffmpeg settings and terminate with an error. */
 465     {
 466         int score = 0;
 467         score += h->param.analyse.i_me_range == 0;
 468         score += h->param.rc.i_qp_step == 3;
 469         score += h->param.i_keyint_max == 12;
 470         score += h->param.rc.i_qp_min == 2;
 471         score += h->param.rc.i_qp_max == 31;
 472         score += h->param.rc.f_qcompress == 0.5;
 473         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 474         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 475         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 476         if( score >= 5 )
 477         {
 478             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 479             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 480             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 481             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 482             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 483             return -1;
 484         }
 485     }
 486
 487     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 488     {
 489         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 490         return -1;
 491     }
 492     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 493     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 494     if( h->param.rc.i_rc_method == X264_RC_CRF )
 495     {
 496         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 497         h->param.rc.i_bitrate = 0;
 498     }
 499     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 500         && h->param.rc.i_qp_constant == 0 )
 501     {
 502         h->mb.b_lossless = 1;
 503         h->param.i_cqm_preset = X264_CQM_FLAT;
 504         h->param.psz_cqm_file = NULL;
 505         h->param.rc.i_rc_method = X264_RC_CQP;
 506         h->param.rc.f_ip_factor = 1;
 507         h->param.rc.f_pb_factor = 1;
 508         h->param.analyse.b_psnr = 0;
 509         h->param.analyse.b_ssim = 0;
 510         h->param.analyse.i_chroma_qp_offset = 0;
 511         h->param.analyse.i_trellis = 0;
 512         h->param.analyse.b_fast_pskip = 0;
 513         h->param.analyse.i_noise_reduction = 0;
 514         h->param.analyse.b_psy = 0;
 515         h->param.i_bframe = 0;
 516         /* 8x8dct is not useful at all in CAVLC lossless */
 517         if( !h->param.b_cabac )
 518             h->param.analyse.b_transform_8x8 = 0;
 519     }
 520     if( h->param.rc.i_rc_method == X264_RC_CQP )
 521     {
 522         float qp_p = h->param.rc.i_qp_constant;
 523         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 524         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 525         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 526         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 527         h->param.rc.i_aq_mode = 0;
 528         h->param.rc.b_mb_tree = 0;
 529     }
 530     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 531     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 532     if( h->param.rc.i_vbv_buffer_size )
 533     {
 534         if( h->param.rc.i_rc_method == X264_RC_CQP )
 535         {
 536             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 537             h->param.rc.i_vbv_max_bitrate = 0;
 538             h->param.rc.i_vbv_buffer_size = 0;
 539         }
 540         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 541         {
 542             if( h->param.rc.i_rc_method == X264_RC_ABR )
 543             {
 544                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 545                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 546             }
 547             else
 548             {
 549                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 550                 h->param.rc.i_vbv_buffer_size = 0;
 551             }
 552         }
 553         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 554                  h->param.rc.i_rc_method == X264_RC_ABR )
 555         {
 556             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 557             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 558         }
 559     }
 560     else if( h->param.rc.i_vbv_max_bitrate )
 561     {
 562         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 563         h->param.rc.i_vbv_max_bitrate = 0;
 564     }
 565
 566     if( h->param.b_interlaced && h->param.i_slice_max_size )
 567     {
 568         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-size is not implemented\n" );
 569         h->param.i_slice_max_size = 0;
 570     }
 571     if( h->param.b_interlaced && h->param.i_slice_max_mbs )
 572     {
 573         x264_log( h, X264_LOG_WARNING, "interlaced + slice-max-mbs is not implemented\n" );
 574         h->param.i_slice_max_mbs = 0;
 575     }
 576     int max_slices = (h->param.i_height+((16<<h->param.b_interlaced)-1))/(16<<h->param.b_interlaced);
 577     if( h->param.b_sliced_threads )
 578         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 579     else
 580     {
 581         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 582         h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 583         h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 584         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 585             h->param.i_slice_count = 0;
 586     }
 587
 588     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 589     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 590     if( h->param.i_scenecut_threshold < 0 )
 591         h->param.i_scenecut_threshold = 0;
 592     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 593     {
 594         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 595         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 596     }
 597     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 598     h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
 599     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 600     if( h->param.i_bframe <= 1 )
 601         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 602     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 603     if( !h->param.i_bframe )
 604     {
 605         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 606         h->param.analyse.i_direct_mv_pred = 0;
 607         h->param.analyse.b_weighted_bipred = 0;
 608         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 609     }
 610     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 611     {
 612         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 613         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 614     }
 615     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 616     {
 617         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 618         h->param.i_frame_reference = 1;
 619         h->param.i_dpb_size = 1;
 620     }
 621     if( h->param.b_intra_refresh && h->param.i_open_gop )
 622     {
 623         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 624         h->param.i_open_gop = X264_OPEN_GOP_NONE;
 625     }
 626     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;
 627     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 628         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 629     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 630     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 631     {
 632         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 633         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 634         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 635     }
 636
 637     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
 638     {
 639         h->param.i_timebase_num = h->param.i_fps_den;
 640         h->param.i_timebase_den = h->param.i_fps_num;
 641     }
 642
 643     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 644     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 645         h->param.rc.b_mb_tree = 0;
 646     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 647         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 648     {
 649         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 650         h->param.rc.b_mb_tree = 0;
 651     }
 652     if( h->param.rc.b_stat_read )
 653         h->param.rc.i_lookahead = 0;
 654 #if HAVE_PTHREAD
 655     if( h->param.i_sync_lookahead < 0 )
 656         h->param.i_sync_lookahead = h->param.i_bframe + 1;
 657     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
 658     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 659         h->param.i_sync_lookahead = 0;
 660 #else
 661     h->param.i_sync_lookahead = 0;
 662 #endif
 663
 664     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 665     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 666     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 667     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 668
 669     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 670
 671     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 672         h->param.i_cqm_preset = X264_CQM_FLAT;
 673
 674     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 675         h->param.analyse.i_me_method > X264_ME_TESA )
 676         h->param.analyse.i_me_method = X264_ME_HEX;
 677     if( h->param.analyse.i_me_range < 4 )
 678         h->param.analyse.i_me_range = 4;
 679     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 680         h->param.analyse.i_me_range = 16;
 681     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 682         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 683         h->param.analyse.i_me_method = X264_ME_ESA;
 684     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 10 );
 685     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 686     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 687                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 688     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 689     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 690         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 691     if( !h->param.analyse.b_transform_8x8 )
 692     {
 693         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 694         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 695     }
 696     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 697     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 698     if( !h->param.analyse.b_psy )
 699     {
 700         h->param.analyse.f_psy_rd = 0;
 701         h->param.analyse.f_psy_trellis = 0;
 702     }
 703     if( !h->param.analyse.i_trellis )
 704         h->param.analyse.f_psy_trellis = 0;
 705     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 706     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 707     if( h->param.analyse.i_subpel_refine < 6 )
 708         h->param.analyse.f_psy_rd = 0;
 709     h->mb.i_psy_rd = FIX8( h->param.analyse.f_psy_rd );
 710     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 711     /* so we lower the chroma QP offset to compensate */
 712     /* This can be triggered repeatedly on multiple calls to parameter_validate, but since encoding
 713      * uses the pps chroma qp offset not the param chroma qp offset, this is not a problem. */
 714     if( h->mb.i_psy_rd )
 715         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 716     h->mb.i_psy_trellis = FIX8( h->param.analyse.f_psy_trellis / 4 );
 717     /* Psy trellis has a similar effect. */
 718     if( h->mb.i_psy_trellis )
 719         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 720     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 721     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 722     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 723     if( h->param.rc.f_aq_strength == 0 )
 724         h->param.rc.i_aq_mode = 0;
 725     /* MB-tree requires AQ to be on, even if the strength is zero. */
 726     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 727     {
 728         h->param.rc.i_aq_mode = 1;
 729         h->param.rc.f_aq_strength = 0;
 730     }
 731     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 732     if( h->param.analyse.i_subpel_refine == 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 733         h->param.analyse.i_subpel_refine = 9;
 734
 735     {
 736         const x264_level_t *l = x264_levels;
 737         if( h->param.i_level_idc < 0 )
 738         {
 739             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 740             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 741                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 742             h->sps = h->sps_array;
 743             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 744             do h->param.i_level_idc = l->level_idc;
 745                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 746             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 747         }
 748         else
 749         {
 750             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 751                 l++;
 752             if( l->level_idc == 0 )
 753             {
 754                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 755                 return -1;
 756             }
 757         }
 758         if( h->param.analyse.i_mv_range <= 0 )
 759             h->param.analyse.i_mv_range = l->mv_range >> h->param.b_interlaced;
 760         else
 761             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> h->param.b_interlaced);
 762     }
 763
 764     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
 765     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy && !h->param.b_interlaced )
 766         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 767     if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND && BIT_DEPTH > 8 )
 768         h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 769
 770     if( h->i_thread_frames > 1 )
 771     {
 772         int r = h->param.analyse.i_mv_range_thread;
 773         int r2;
 774         if( r <= 0 )
 775         {
 776             // half of the available space is reserved and divided evenly among the threads,
 777             // the rest is allocated to whichever thread is far enough ahead to use it.
 778             // reserving more space increases quality for some videos, but costs more time
 779             // in thread synchronization.
 780             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 781             r = max_range / 2;
 782         }
 783         r = X264_MAX( r, h->param.analyse.i_me_range );
 784         r = X264_MIN( r, h->param.analyse.i_mv_range );
 785         // round up to use the whole mb row
 786         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 787         if( r2 < r )
 788             r2 += 16;
 789         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 790         h->param.analyse.i_mv_range_thread = r2;
 791     }
 792
 793     if( h->param.rc.f_qblur < 0 )
 794         h->param.rc.f_qblur = 0;
 795     if( h->param.rc.f_complexity_blur < 0 )
 796         h->param.rc.f_complexity_blur = 0;
 797
 798     h->param.i_sps_id &= 31;
 799
 800     if( h->param.i_log_level < X264_LOG_INFO )
 801     {
 802         h->param.analyse.b_psnr = 0;
 803         h->param.analyse.b_ssim = 0;
 804     }
 805
 806     if( h->param.b_interlaced )
 807         h->param.b_pic_struct = 1;
 808
 809     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 810     {
 811         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 812         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 813     }
 814
 815     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 816        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 817     {
 818         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 819         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 820     }
 821
 822     /* ensure the booleans are 0 or 1 so they can be used in math */
 823 #define BOOLIFY(x) h->param.x = !!h->param.x
 824     BOOLIFY( b_cabac );
 825     BOOLIFY( b_constrained_intra );
 826     BOOLIFY( b_deblocking_filter );
 827     BOOLIFY( b_deterministic );
 828     BOOLIFY( b_sliced_threads );
 829     BOOLIFY( b_interlaced );
 830     BOOLIFY( b_intra_refresh );
 831     BOOLIFY( b_visualize );
 832     BOOLIFY( b_aud );
 833     BOOLIFY( b_repeat_headers );
 834     BOOLIFY( b_annexb );
 835     BOOLIFY( b_vfr_input );
 836     BOOLIFY( b_pic_struct );
 837     BOOLIFY( b_fake_interlaced );
 838     BOOLIFY( analyse.b_transform_8x8 );
 839     BOOLIFY( analyse.b_weighted_bipred );
 840     BOOLIFY( analyse.b_chroma_me );
 841     BOOLIFY( analyse.b_mixed_references );
 842     BOOLIFY( analyse.b_fast_pskip );
 843     BOOLIFY( analyse.b_dct_decimate );
 844     BOOLIFY( analyse.b_psy );
 845     BOOLIFY( analyse.b_psnr );
 846     BOOLIFY( analyse.b_ssim );
 847     BOOLIFY( rc.b_stat_write );
 848     BOOLIFY( rc.b_stat_read );
 849     BOOLIFY( rc.b_mb_tree );
 850 #undef BOOLIFY
 851
 852     return 0;
 853 }
 854
 855 static void mbcmp_init( x264_t *h )
 856 {
 857     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
 858     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
 859     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
 860     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
 861     h->pixf.intra_mbcmp_x3_8x8c = satd ? h->pixf.intra_satd_x3_8x8c : h->pixf.intra_sad_x3_8x8c;
 862     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
 863     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
 864     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
 865     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
 866     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
 867     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
 868 }
 869
 870 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
 871 {
 872     /* VUI */
 873     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
 874     {
 875         uint32_t i_w = param->vui.i_sar_width;
 876         uint32_t i_h = param->vui.i_sar_height;
 877         uint32_t old_w = h->param.vui.i_sar_width;
 878         uint32_t old_h = h->param.vui.i_sar_height;
 879
 880         x264_reduce_fraction( &i_w, &i_h );
 881
 882         while( i_w > 65535 || i_h > 65535 )
 883         {
 884             i_w /= 2;
 885             i_h /= 2;
 886         }
 887
 888         x264_reduce_fraction( &i_w, &i_h );
 889
 890         if( i_w != old_w || i_h != old_h || initial )
 891         {
 892             h->param.vui.i_sar_width = 0;
 893             h->param.vui.i_sar_height = 0;
 894             if( i_w == 0 || i_h == 0 )
 895                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
 896             else
 897             {
 898                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
 899                 h->param.vui.i_sar_width = i_w;
 900                 h->param.vui.i_sar_height = i_h;
 901             }
 902         }
 903     }
 904 }
 905
 906 /****************************************************************************
 907  * x264_encoder_open:
 908  ****************************************************************************/
 909 x264_t *x264_encoder_open( x264_param_t *param )
 910 {
 911     x264_t *h;
 912     char buf[1000], *p;
 913     int qp, i_slicetype_length;
 914
 915     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
 916
 917     /* Create a copy of param */
 918     memcpy( &h->param, param, sizeof(x264_param_t) );
 919
 920     if( param->param_free )
 921         param->param_free( param );
 922
 923     if( x264_validate_parameters( h ) < 0 )
 924         goto fail;
 925
 926     if( h->param.psz_cqm_file )
 927         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
 928             goto fail;
 929
 930     if( h->param.rc.psz_stat_out )
 931         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
 932     if( h->param.rc.psz_stat_in )
 933         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
 934
 935     x264_set_aspect_ratio( h, &h->param, 1 );
 936
 937     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
 938     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
 939
 940     /* Init x264_t */
 941     h->i_frame = -1;
 942     h->i_frame_num = 0;
 943     h->i_idr_pic_id = 0;
 944
 945     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
 946     {
 947         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
 948         goto fail;
 949     }
 950
 951     h->sps = &h->sps_array[0];
 952     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 953
 954     h->pps = &h->pps_array[0];
 955     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
 956
 957     x264_validate_levels( h, 1 );
 958
 959     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
 960
 961     if( x264_cqm_init( h ) < 0 )
 962         goto fail;
 963
 964     h->mb.i_mb_width = h->sps->i_mb_width;
 965     h->mb.i_mb_height = h->sps->i_mb_height;
 966     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
 967
 968     /* Init frames. */
 969     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
 970         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
 971     else
 972         h->frames.i_delay = h->param.i_bframe;
 973     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
 974         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
 975     i_slicetype_length = h->frames.i_delay;
 976     h->frames.i_delay += h->i_thread_frames - 1;
 977     h->frames.i_delay += h->param.i_sync_lookahead;
 978     h->frames.i_delay += h->param.b_vfr_input && (h->param.rc.i_rc_method == X264_RC_ABR || h->param.rc.b_stat_write
 979                                                  || h->param.rc.i_vbv_buffer_size);
 980     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
 981
 982     h->frames.i_max_ref0 = h->param.i_frame_reference;
 983     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
 984     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
 985     h->frames.b_have_lowres = !h->param.rc.b_stat_read
 986         && ( h->param.rc.i_rc_method == X264_RC_ABR
 987           || h->param.rc.i_rc_method == X264_RC_CRF
 988           || h->param.i_bframe_adaptive
 989           || h->param.i_scenecut_threshold
 990           || h->param.rc.b_mb_tree
 991           || h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART );
 992     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
 993     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
 994
 995     h->frames.i_last_idr =
 996     h->frames.i_last_keyframe = - h->param.i_keyint_max;
 997     h->frames.i_input    = 0;
 998     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
 999     h->frames.i_poc_last_open_gop = -1;
1000
1001     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1002     /* Allocate room for max refs plus a few extra just in case. */
1003     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1004     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1005                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1006     if( h->param.analyse.i_weighted_pred > 0 )
1007         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1008     h->i_ref0 = 0;
1009     h->i_ref1 = 0;
1010     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1011     h->i_prev_duration = ((uint64_t)h->param.i_fps_den * h->sps->vui.i_time_scale) / ((uint64_t)h->param.i_fps_num * h->sps->vui.i_num_units_in_tick);
1012     h->i_disp_fields_last_frame = -1;
1013     x264_rdo_init();
1014
1015     /* init CPU functions */
1016     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1017     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1018     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1019     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1020     if( !h->param.b_cabac )
1021         x264_init_vlc_tables();
1022     x264_pixel_init( h->param.cpu, &h->pixf );
1023     x264_dct_init( h->param.cpu, &h->dctf );
1024     x264_zigzag_init( h->param.cpu, &h->zigzagf, h->param.b_interlaced );
1025     x264_mc_init( h->param.cpu, &h->mc );
1026     x264_quant_init( h, h->param.cpu, &h->quantf );
1027     x264_deblock_init( h->param.cpu, &h->loopf );
1028     x264_bitstream_init( h->param.cpu, &h->bsf );
1029     x264_dct_init_weights();
1030
1031     mbcmp_init( h );
1032
1033     p = buf + sprintf( buf, "using cpu capabilities:" );
1034     for( int i = 0; x264_cpu_names[i].flags; i++ )
1035     {
1036         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1037             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1038             continue;
1039         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1040             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1041             continue;
1042         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1043             && (h->param.cpu & X264_CPU_SSE42) )
1044             continue;
1045         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1046             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1047             p += sprintf( p, " %s", x264_cpu_names[i].name );
1048     }
1049     if( !h->param.cpu )
1050         p += sprintf( p, " none!" );
1051     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1052
1053     for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
1054         if( x264_analyse_init_costs( h, qp ) )
1055             goto fail;
1056     if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
1057         goto fail;
1058
1059     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1060     /* Checks for known miscompilation issues. */
1061     if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
1062     {
1063         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1064         goto fail;
1065     }
1066
1067     /* Must be volatile or else GCC will optimize it out. */
1068     volatile int temp = 392;
1069     if( x264_clz( temp ) != 23 )
1070     {
1071         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1072 #if ARCH_X86 || ARCH_X86_64
1073         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1074         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1075 #endif
1076         goto fail;
1077     }
1078
1079     h->out.i_nal = 0;
1080     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1081         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1082           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1083
1084     CHECKED_MALLOC( h->nal_buffer, h->out.i_bitstream * 3/2 + 4 );
1085     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1086
1087     if( h->param.i_threads > 1 &&
1088         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1089         goto fail;
1090
1091     h->thread[0] = h;
1092     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1093         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1094
1095     for( int i = 0; i < h->param.i_threads; i++ )
1096     {
1097         int init_nal_count = h->param.i_slice_count + 3;
1098         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1099         if( i > 0 )
1100             *h->thread[i] = *h;
1101
1102         if( allocate_threadlocal_data )
1103         {
1104             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1105             if( !h->thread[i]->fdec )
1106                 goto fail;
1107         }
1108         else
1109             h->thread[i]->fdec = h->thread[0]->fdec;
1110
1111         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1112         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1113         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1114         h->thread[i]->out.i_nals_allocated = init_nal_count;
1115
1116         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1117             goto fail;
1118     }
1119
1120     if( x264_lookahead_init( h, i_slicetype_length ) )
1121         goto fail;
1122
1123     for( int i = 0; i < h->param.i_threads; i++ )
1124         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1125             goto fail;
1126
1127     if( x264_ratecontrol_new( h ) < 0 )
1128         goto fail;
1129
1130     if( h->param.i_nal_hrd )
1131     {
1132         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1133         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1134     }
1135
1136     if( h->param.psz_dump_yuv )
1137     {
1138         /* create or truncate the reconstructed video file */
1139         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1140         if( !f )
1141         {
1142             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1143             goto fail;
1144         }
1145         else if( !x264_is_regular_file( f ) )
1146         {
1147             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1148             goto fail;
1149         }
1150         fclose( f );
1151     }
1152
1153     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1154                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1155                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1156                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1157                           "High 4:4:4 Predictive";
1158     char level[4];
1159     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1160     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 ) )
1161         strcpy( level, "1b" );
1162
1163     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1164     {
1165         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1166             profile, level );
1167     }
1168     else
1169     {
1170         x264_log( h, X264_LOG_INFO, "profile %s, level %s, bit depth %d\n",
1171             profile, level, BIT_DEPTH );
1172     }
1173
1174     return h;
1175 fail:
1176     x264_free( h );
1177     return NULL;
1178 }
1179
1180 /****************************************************************************
1181  * x264_encoder_reconfig:
1182  ****************************************************************************/
1183 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1184 {
1185     int rc_reconfig = 0;
1186     h = h->thread[h->thread[0]->i_thread_phase];
1187     x264_set_aspect_ratio( h, param, 0 );
1188 #define COPY(var) h->param.var = param->var
1189     COPY( i_frame_reference ); // but never uses more refs than initially specified
1190     COPY( i_bframe_bias );
1191     if( h->param.i_scenecut_threshold )
1192         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1193     COPY( b_deblocking_filter );
1194     COPY( i_deblocking_filter_alphac0 );
1195     COPY( i_deblocking_filter_beta );
1196     COPY( analyse.inter );
1197     COPY( analyse.intra );
1198     COPY( analyse.i_direct_mv_pred );
1199     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1200     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1201         COPY( analyse.i_me_range );
1202     COPY( analyse.i_noise_reduction );
1203     /* We can't switch out of subme=0 during encoding. */
1204     if( h->param.analyse.i_subpel_refine )
1205         COPY( analyse.i_subpel_refine );
1206     COPY( analyse.i_trellis );
1207     COPY( analyse.b_chroma_me );
1208     COPY( analyse.b_dct_decimate );
1209     COPY( analyse.b_fast_pskip );
1210     COPY( analyse.b_mixed_references );
1211     COPY( analyse.f_psy_rd );
1212     COPY( analyse.f_psy_trellis );
1213     // can only twiddle these if they were enabled to begin with:
1214     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1215         COPY( analyse.i_me_method );
1216     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1217         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1218     if( h->pps->b_transform_8x8_mode )
1219         COPY( analyse.b_transform_8x8 );
1220     if( h->frames.i_max_ref1 > 1 )
1221         COPY( i_bframe_pyramid );
1222     COPY( i_slice_max_size );
1223     COPY( i_slice_max_mbs );
1224     COPY( i_slice_count );
1225
1226     /* VBV can't be turned on if it wasn't on to begin with */
1227     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1228           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1229     {
1230         COPY( rc.i_vbv_max_bitrate );
1231         COPY( rc.i_vbv_buffer_size );
1232         COPY( rc.i_bitrate );
1233         rc_reconfig = 1;
1234     }
1235     if( h->param.rc.f_rf_constant != param->rc.f_rf_constant )
1236     {
1237         COPY( rc.f_rf_constant );
1238         rc_reconfig = 1;
1239     }
1240     if( h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max )
1241     {
1242         COPY( rc.f_rf_constant_max );
1243         rc_reconfig = 1;
1244     }
1245
1246 #undef COPY
1247
1248     mbcmp_init( h );
1249
1250     int ret = x264_validate_parameters( h );
1251
1252     /* Supported reconfiguration options (1-pass only):
1253      * vbv-maxrate
1254      * vbv-bufsize
1255      * crf
1256      * bitrate (CBR only) */
1257     if( !ret && rc_reconfig )
1258         x264_ratecontrol_init_reconfigurable( h, 0 );
1259
1260     return ret;
1261 }
1262
1263 /****************************************************************************
1264  * x264_encoder_parameters:
1265  ****************************************************************************/
1266 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1267 {
1268     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1269 }
1270
1271 /* internal usage */
1272 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1273 {
1274     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1275
1276     nal->i_ref_idc        = i_ref_idc;
1277     nal->i_type           = i_type;
1278     nal->b_long_startcode = 1;
1279
1280     nal->i_payload= 0;
1281     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1282 }
1283
1284 /* if number of allocated nals is not enough, re-allocate a larger one. */
1285 static int x264_nal_check_buffer( x264_t *h )
1286 {
1287     if( h->out.i_nal >= h->out.i_nals_allocated )
1288     {
1289         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1290         if( !new_out )
1291             return -1;
1292         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1293         x264_free( h->out.nal );
1294         h->out.nal = new_out;
1295         h->out.i_nals_allocated *= 2;
1296     }
1297     return 0;
1298 }
1299
1300 static int x264_nal_end( x264_t *h )
1301 {
1302     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1303     nal->i_payload = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8] - nal->p_payload;
1304     if( h->param.nalu_process )
1305         h->param.nalu_process( h, nal );
1306     h->out.i_nal++;
1307
1308     return x264_nal_check_buffer( h );
1309 }
1310
1311 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1312 {
1313     int nal_size = 0, previous_nal_size = 0;
1314
1315     if( h->param.nalu_process )
1316     {
1317         for( int i = start; i < h->out.i_nal; i++ )
1318             nal_size += h->out.nal[i].i_payload;
1319         return nal_size;
1320     }
1321
1322     for( int i = 0; i < start; i++ )
1323         previous_nal_size += h->out.nal[i].i_payload;
1324
1325     for( int i = start; i < h->out.i_nal; i++ )
1326         nal_size += h->out.nal[i].i_payload;
1327
1328     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1329     if( h->nal_buffer_size < nal_size * 3/2 + h->out.i_nal * 4 )
1330     {
1331         uint8_t *buf = x264_malloc( nal_size * 2 + h->out.i_nal * 4 );
1332         if( !buf )
1333             return -1;
1334         if( previous_nal_size )
1335             memcpy( buf, h->nal_buffer, previous_nal_size );
1336         x264_free( h->nal_buffer );
1337         h->nal_buffer = buf;
1338     }
1339
1340     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1341
1342     for( int i = start; i < h->out.i_nal; i++ )
1343     {
1344         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1345         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1346         nal_buffer += h->out.nal[i].i_payload;
1347     }
1348
1349     x264_emms();
1350
1351     return nal_buffer - (h->nal_buffer + previous_nal_size);
1352 }
1353
1354 /****************************************************************************
1355  * x264_encoder_headers:
1356  ****************************************************************************/
1357 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1358 {
1359     int frame_size = 0;
1360     /* init bitstream context */
1361     h->out.i_nal = 0;
1362     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1363
1364     /* Write SEI, SPS and PPS. */
1365
1366     /* generate sequence parameters */
1367     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1368     x264_sps_write( &h->out.bs, h->sps );
1369     if( x264_nal_end( h ) )
1370         return -1;
1371
1372     /* generate picture parameters */
1373     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1374     x264_pps_write( &h->out.bs, h->pps );
1375     if( x264_nal_end( h ) )
1376         return -1;
1377
1378     /* identify ourselves */
1379     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1380     if( x264_sei_version_write( h, &h->out.bs ) )
1381         return -1;
1382     if( x264_nal_end( h ) )
1383         return -1;
1384
1385     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1386
1387     /* now set output*/
1388     *pi_nal = h->out.i_nal;
1389     *pp_nal = &h->out.nal[0];
1390     h->out.i_nal = 0;
1391
1392     return frame_size;
1393 }
1394
1395 /* Check to see whether we have chosen a reference list ordering different
1396  * from the standard's default. */
1397 static inline void x264_reference_check_reorder( x264_t *h )
1398 {
1399     /* The reorder check doesn't check for missing frames, so just
1400      * force a reorder if one of the reference list is corrupt. */
1401     for( int i = 0; h->frames.reference[i]; i++ )
1402         if( h->frames.reference[i]->b_corrupt )
1403         {
1404             h->b_ref_reorder[0] = 1;
1405             return;
1406         }
1407     for( int i = 0; i < h->i_ref0 - 1; i++ )
1408         /* P and B-frames use different default orders. */
1409         if( h->sh.i_type == SLICE_TYPE_P ? h->fref0[i]->i_frame_num < h->fref0[i+1]->i_frame_num
1410                                          : h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1411         {
1412             h->b_ref_reorder[0] = 1;
1413             return;
1414         }
1415 }
1416
1417 /* return -1 on failure, else return the index of the new reference frame */
1418 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1419 {
1420     int i = h->i_ref0;
1421     int j = 1;
1422     x264_frame_t *newframe;
1423     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1424         return -1;
1425
1426     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1427      * With high bit depth, it's not worth doing, so turn it off except in the case of
1428      * unweighted dupes. */
1429     if( BIT_DEPTH > 8 && w != weight_none )
1430         return -1;
1431
1432     newframe = x264_frame_pop_blank_unused( h );
1433
1434     //FIXME: probably don't need to copy everything
1435     *newframe = *h->fref0[i_ref];
1436     newframe->i_reference_count = 1;
1437     newframe->orig = h->fref0[i_ref];
1438     newframe->b_duplicate = 1;
1439     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1440
1441     /* shift the frames to make space for the dupe. */
1442     h->b_ref_reorder[0] = 1;
1443     if( h->i_ref0 < X264_REF_MAX )
1444         ++h->i_ref0;
1445     h->fref0[X264_REF_MAX-1] = NULL;
1446     x264_frame_unshift( &h->fref0[j], newframe );
1447
1448     return j;
1449 }
1450
1451 static void x264_weighted_pred_init( x264_t *h )
1452 {
1453     /* for now no analysis and set all weights to nothing */
1454     for( int i_ref = 0; i_ref < h->i_ref0; i_ref++ )
1455         h->fenc->weighted[i_ref] = h->fref0[i_ref]->filtered[0];
1456
1457     // FIXME: This only supports weighting of one reference frame
1458     // and duplicates of that frame.
1459     h->fenc->i_lines_weighted = 0;
1460
1461     for( int i_ref = 0; i_ref < (h->i_ref0 << h->sh.b_mbaff); i_ref++ )
1462         for( int i = 0; i < 3; i++ )
1463             h->sh.weight[i_ref][i].weightfn = NULL;
1464
1465
1466     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1467         return;
1468
1469     int i_padv = PADV << h->param.b_interlaced;
1470     int denom = -1;
1471     int weightluma = 0;
1472     int buffer_next = 0;
1473     //FIXME: when chroma support is added, move this into loop
1474     h->sh.weight[0][1].weightfn = h->sh.weight[0][2].weightfn = NULL;
1475     h->sh.weight[0][1].i_denom = h->sh.weight[0][2].i_denom = 0;
1476     for( int j = 0; j < h->i_ref0; j++ )
1477     {
1478         if( h->fenc->weight[j][0].weightfn )
1479         {
1480             h->sh.weight[j][0] = h->fenc->weight[j][0];
1481             // if weight is useless, don't write it to stream
1482             if( h->sh.weight[j][0].i_scale == 1<<h->sh.weight[j][0].i_denom && h->sh.weight[j][0].i_offset == 0 )
1483                 h->sh.weight[j][0].weightfn = NULL;
1484             else
1485             {
1486                 if( !weightluma )
1487                 {
1488                     weightluma = 1;
1489                     h->sh.weight[0][0].i_denom = denom = h->sh.weight[j][0].i_denom;
1490                     assert( x264_clip3( denom, 0, 7 ) == denom );
1491                 }
1492                 assert( h->sh.weight[j][0].i_denom == denom );
1493                 assert( x264_clip3( h->sh.weight[j][0].i_scale, 0, 127 ) == h->sh.weight[j][0].i_scale );
1494                 assert( x264_clip3( h->sh.weight[j][0].i_offset, -128, 127 ) == h->sh.weight[j][0].i_offset );
1495                 h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] +
1496                     h->fenc->i_stride[0] * i_padv + PADH;
1497             }
1498         }
1499
1500         //scale full resolution frame
1501         if( h->sh.weight[j][0].weightfn && h->param.i_threads == 1 )
1502         {
1503             pixel *src = h->fref0[j]->filtered[0] - h->fref0[j]->i_stride[0]*i_padv - PADH;
1504             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1505             int stride = h->fenc->i_stride[0];
1506             int width = h->fenc->i_width[0] + PADH*2;
1507             int height = h->fenc->i_lines[0] + i_padv*2;
1508             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1509             h->fenc->i_lines_weighted = height;
1510         }
1511     }
1512     if( !weightluma )
1513         h->sh.weight[0][0].i_denom = 0;
1514 }
1515
1516 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1517 {
1518     int b_ok;
1519
1520     /* build ref list 0/1 */
1521     h->mb.pic.i_fref[0] = h->i_ref0 = 0;
1522     h->mb.pic.i_fref[1] = h->i_ref1 = 0;
1523     if( h->sh.i_type == SLICE_TYPE_I )
1524         return;
1525
1526     for( int i = 0; h->frames.reference[i]; i++ )
1527     {
1528         if( h->frames.reference[i]->b_corrupt )
1529             continue;
1530         if( h->frames.reference[i]->i_poc < i_poc )
1531             h->fref0[h->i_ref0++] = h->frames.reference[i];
1532         else if( h->frames.reference[i]->i_poc > i_poc )
1533             h->fref1[h->i_ref1++] = h->frames.reference[i];
1534     }
1535
1536     /* Order ref0 from higher to lower poc */
1537     do
1538     {
1539         b_ok = 1;
1540         for( int i = 0; i < h->i_ref0 - 1; i++ )
1541         {
1542             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
1543             {
1544                 XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
1545                 b_ok = 0;
1546                 break;
1547             }
1548         }
1549     } while( !b_ok );
1550
1551     if( h->sh.i_mmco_remove_from_end )
1552         for( int i = h->i_ref0-1; i >= h->i_ref0 - h->sh.i_mmco_remove_from_end; i-- )
1553         {
1554             int diff = h->i_frame_num - h->fref0[i]->i_frame_num;
1555             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref0[i]->i_poc;
1556             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1557         }
1558
1559     /* Order ref1 from lower to higher poc (bubble sort) for B-frame */
1560     do
1561     {
1562         b_ok = 1;
1563         for( int i = 0; i < h->i_ref1 - 1; i++ )
1564         {
1565             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
1566             {
1567                 XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
1568                 b_ok = 0;
1569                 break;
1570             }
1571         }
1572     } while( !b_ok );
1573
1574     x264_reference_check_reorder( h );
1575
1576     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
1577     h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
1578     h->i_ref0 = X264_MIN( h->i_ref0, h->param.i_frame_reference ); // if reconfig() has lowered the limit
1579
1580     /* add duplicates */
1581     if( h->fenc->i_type == X264_TYPE_P )
1582     {
1583         int idx = -1;
1584         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
1585         {
1586             x264_weight_t w[3];
1587             w[1].weightfn = w[2].weightfn = NULL;
1588             if( h->param.rc.b_stat_read )
1589                 x264_ratecontrol_set_weights( h, h->fenc );
1590
1591             if( !h->fenc->weight[0][0].weightfn )
1592             {
1593                 h->fenc->weight[0][0].i_denom = 0;
1594                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1595                 idx = x264_weighted_reference_duplicate( h, 0, w );
1596             }
1597             else
1598             {
1599                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1600                 {
1601                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1602                 }
1603                 x264_weighted_reference_duplicate( h, 0, weight_none );
1604                 if( h->fenc->weight[0][0].i_offset > -128 )
1605                 {
1606                     w[0] = h->fenc->weight[0][0];
1607                     w[0].i_offset--;
1608                     h->mc.weight_cache( h, &w[0] );
1609                     idx = x264_weighted_reference_duplicate( h, 0, w );
1610                 }
1611             }
1612         }
1613         else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
1614         {
1615             //weighted offset=-1
1616             x264_weight_t w[3];
1617             SET_WEIGHT( w[0], 1, 1, 0, -1 );
1618             h->fenc->weight[0][0].i_denom = 0;
1619             w[1].weightfn = w[2].weightfn = NULL;
1620             idx = x264_weighted_reference_duplicate( h, 0, w );
1621         }
1622         h->mb.ref_blind_dupe = idx;
1623     }
1624
1625     assert( h->i_ref0 + h->i_ref1 <= X264_REF_MAX );
1626     h->mb.pic.i_fref[0] = h->i_ref0;
1627     h->mb.pic.i_fref[1] = h->i_ref1;
1628 }
1629
1630 static void x264_fdec_filter_row( x264_t *h, int mb_y, int b_inloop )
1631 {
1632     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1633     int b_hpel = h->fdec->b_kept_as_ref;
1634     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1635     int b_end = mb_y == h->i_threadslice_end;
1636     int b_measure_quality = 1;
1637     int min_y = mb_y - (1 << h->sh.b_mbaff);
1638     int b_start = min_y == h->i_threadslice_start;
1639     int max_y = b_end ? h->i_threadslice_end : mb_y;
1640     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1641     if( h->param.b_sliced_threads && b_start && min_y && !b_inloop )
1642     {
1643         b_deblock = 0;         /* We already deblocked on the inloop pass. */
1644         b_measure_quality = 0; /* We already measured quality on the inloop pass. */
1645     }
1646     if( mb_y & h->sh.b_mbaff )
1647         return;
1648     if( min_y < h->i_threadslice_start )
1649         return;
1650
1651     if( b_deblock )
1652         for( int y = min_y; y < max_y; y += (1 << h->sh.b_mbaff) )
1653             x264_frame_deblock_row( h, y );
1654
1655     if( b_hpel )
1656     {
1657         int end = mb_y == h->mb.i_mb_height;
1658         x264_frame_expand_border( h, h->fdec, min_y, end );
1659         if( h->param.analyse.i_subpel_refine )
1660         {
1661             x264_frame_filter( h, h->fdec, min_y, end );
1662             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1663         }
1664     }
1665
1666     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1667         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << h->sh.b_mbaff)) );
1668
1669     min_y = min_y*16 - 8 * !b_start;
1670     max_y = b_end ? X264_MIN( h->i_threadslice_end*16 , h->param.i_height ) : mb_y*16 - 8;
1671
1672     if( b_measure_quality )
1673     {
1674         if( h->param.analyse.b_psnr )
1675         {
1676             uint64_t ssd_y = x264_pixel_ssd_wxh( &h->pixf,
1677                 h->fdec->plane[0] + min_y * h->fdec->i_stride[0], h->fdec->i_stride[0],
1678                 h->fenc->plane[0] + min_y * h->fenc->i_stride[0], h->fenc->i_stride[0],
1679                 h->param.i_width, max_y-min_y );
1680             uint64_t ssd_u, ssd_v;
1681             x264_pixel_ssd_nv12( &h->pixf,
1682                 h->fdec->plane[1] + (min_y>>1) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1683                 h->fenc->plane[1] + (min_y>>1) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1684                 h->param.i_width>>1, (max_y-min_y)>>1, &ssd_u, &ssd_v );
1685             h->stat.frame.i_ssd[0] += ssd_y;
1686             h->stat.frame.i_ssd[1] += ssd_u;
1687             h->stat.frame.i_ssd[2] += ssd_v;
1688         }
1689
1690         if( h->param.analyse.b_ssim )
1691         {
1692             x264_emms();
1693             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1694              * and overlap by 4 */
1695             min_y += b_start ? 2 : -6;
1696             h->stat.frame.f_ssim +=
1697                 x264_pixel_ssim_wxh( &h->pixf,
1698                     h->fdec->plane[0] + 2+min_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
1699                     h->fenc->plane[0] + 2+min_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
1700                     h->param.i_width-2, max_y-min_y, h->scratch_buffer );
1701         }
1702     }
1703 }
1704
1705 static inline int x264_reference_update( x264_t *h )
1706 {
1707     if( !h->fdec->b_kept_as_ref )
1708     {
1709         if( h->i_thread_frames > 1 )
1710         {
1711             x264_frame_push_unused( h, h->fdec );
1712             h->fdec = x264_frame_pop_unused( h, 1 );
1713             if( !h->fdec )
1714                 return -1;
1715         }
1716         return 0;
1717     }
1718
1719     /* apply mmco from previous frame. */
1720     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
1721         for( int j = 0; h->frames.reference[j]; j++ )
1722             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
1723                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
1724
1725     /* move frame in the buffer */
1726     x264_frame_push( h->frames.reference, h->fdec );
1727     if( h->frames.reference[h->sps->i_num_ref_frames] )
1728         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
1729     h->fdec = x264_frame_pop_unused( h, 1 );
1730     if( !h->fdec )
1731         return -1;
1732     return 0;
1733 }
1734
1735 static inline void x264_reference_reset( x264_t *h )
1736 {
1737     while( h->frames.reference[0] )
1738         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
1739     h->fdec->i_poc =
1740     h->fenc->i_poc = 0;
1741 }
1742
1743 static inline void x264_reference_hierarchy_reset( x264_t *h )
1744 {
1745     int ref;
1746     int b_hasdelayframe = 0;
1747
1748     /* look for delay frames -- chain must only contain frames that are disposable */
1749     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
1750         b_hasdelayframe |= h->frames.current[i]->i_coded
1751                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
1752
1753     /* This function must handle b-pyramid and clear frames for open-gop */
1754     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
1755         return;
1756
1757     /* Remove last BREF. There will never be old BREFs in the
1758      * dpb during a BREF decode when pyramid == STRICT */
1759     for( ref = 0; h->frames.reference[ref]; ref++ )
1760     {
1761         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
1762             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
1763             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
1764             && h->sh.i_type != SLICE_TYPE_B ) )
1765         {
1766             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
1767             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
1768             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
1769             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
1770             h->b_ref_reorder[0] = 1;
1771             ref--;
1772         }
1773     }
1774
1775     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
1776     if( h->param.i_bframe_pyramid )
1777         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
1778 }
1779
1780 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
1781 {
1782     /* ------------------------ Create slice header  ----------------------- */
1783     if( i_nal_type == NAL_SLICE_IDR )
1784     {
1785         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
1786
1787         /* alternate id */
1788         h->i_idr_pic_id ^= 1;
1789     }
1790     else
1791     {
1792         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
1793
1794         h->sh.i_num_ref_idx_l0_active = h->i_ref0 <= 0 ? 1 : h->i_ref0;
1795         h->sh.i_num_ref_idx_l1_active = h->i_ref1 <= 0 ? 1 : h->i_ref1;
1796         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
1797             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
1798         {
1799             h->sh.b_num_ref_idx_override = 1;
1800         }
1801     }
1802
1803     h->fdec->i_frame_num = h->sh.i_frame_num;
1804
1805     if( h->sps->i_poc_type == 0 )
1806     {
1807         h->sh.i_poc = h->fdec->i_poc;
1808         if( h->param.b_interlaced )
1809         {
1810             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
1811             if( h->sh.i_delta_poc_bottom == -1 )
1812                 h->sh.i_poc = h->fdec->i_poc + 1;
1813         }
1814         else
1815             h->sh.i_delta_poc_bottom = 0;
1816     }
1817     else if( h->sps->i_poc_type == 1 )
1818     {
1819         /* FIXME TODO FIXME */
1820     }
1821     else
1822     {
1823         /* Nothing to do ? */
1824     }
1825
1826     x264_macroblock_slice_init( h );
1827 }
1828
1829 static int x264_slice_write( x264_t *h )
1830 {
1831     int i_skip;
1832     int mb_xy, i_mb_x, i_mb_y;
1833     int i_skip_bak = 0; /* Shut up GCC. */
1834     bs_t bs_bak;
1835     x264_cabac_t cabac_bak;
1836     uint8_t cabac_prevbyte_bak = 0; /* Shut up GCC. */
1837     int mv_bits_bak = 0;
1838     int tex_bits_bak = 0;
1839     /* NALUs other than the first use a 3-byte startcode.
1840      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
1841      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
1842      * other inaccuracies. */
1843     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
1844     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
1845     int starting_bits = bs_pos(&h->out.bs);
1846     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1847     int b_hpel = h->fdec->b_kept_as_ref;
1848     b_deblock &= b_hpel || h->param.psz_dump_yuv;
1849     bs_realign( &h->out.bs );
1850
1851     /* Slice */
1852     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
1853     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
1854
1855     /* Slice header */
1856     x264_macroblock_thread_init( h );
1857
1858     /* If this isn't the first slice in the threadslice, set the slice QP
1859      * equal to the last QP in the previous slice for more accurate
1860      * CABAC initialization. */
1861     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
1862     {
1863         h->sh.i_qp = h->mb.i_last_qp;
1864         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
1865     }
1866
1867     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
1868     if( h->param.b_cabac )
1869     {
1870         /* alignment needed */
1871         bs_align_1( &h->out.bs );
1872
1873         /* init cabac */
1874         x264_cabac_context_init( &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
1875         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
1876     }
1877     h->mb.i_last_qp = h->sh.i_qp;
1878     h->mb.i_last_dqp = 0;
1879
1880     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
1881     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
1882     i_skip = 0;
1883
1884     while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
1885     {
1886         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1887
1888         if( x264_bitstream_check_buffer( h ) )
1889             return -1;
1890
1891         if( slice_max_size )
1892         {
1893             mv_bits_bak = h->stat.frame.i_mv_bits;
1894             tex_bits_bak = h->stat.frame.i_tex_bits;
1895             /* We don't need the contexts because flushing the CABAC encoder has no context
1896              * dependency and macroblocks are only re-encoded in the case where a slice is
1897              * ended (and thus the content of all contexts are thrown away). */
1898             if( h->param.b_cabac )
1899             {
1900                 memcpy( &cabac_bak, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
1901                 /* x264's CABAC writer modifies the previous byte during carry, so it has to be
1902                  * backed up. */
1903                 cabac_prevbyte_bak = h->cabac.p[-1];
1904             }
1905             else
1906             {
1907                 bs_bak = h->out.bs;
1908                 i_skip_bak = i_skip;
1909             }
1910         }
1911
1912         if( i_mb_x == 0 && !h->mb.b_reencode_mb )
1913             x264_fdec_filter_row( h, i_mb_y, 1 );
1914
1915         /* load cache */
1916         x264_macroblock_cache_load( h, i_mb_x, i_mb_y );
1917
1918         x264_macroblock_analyse( h );
1919
1920         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
1921         x264_macroblock_encode( h );
1922
1923         if( h->param.b_cabac )
1924         {
1925             if( mb_xy > h->sh.i_first_mb && !(h->sh.b_mbaff && (i_mb_y&1)) )
1926                 x264_cabac_encode_terminal( &h->cabac );
1927
1928             if( IS_SKIP( h->mb.i_type ) )
1929                 x264_cabac_mb_skip( h, 1 );
1930             else
1931             {
1932                 if( h->sh.i_type != SLICE_TYPE_I )
1933                     x264_cabac_mb_skip( h, 0 );
1934                 x264_macroblock_write_cabac( h, &h->cabac );
1935             }
1936         }
1937         else
1938         {
1939             if( IS_SKIP( h->mb.i_type ) )
1940                 i_skip++;
1941             else
1942             {
1943                 if( h->sh.i_type != SLICE_TYPE_I )
1944                 {
1945                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
1946                     i_skip = 0;
1947                 }
1948                 x264_macroblock_write_cavlc( h );
1949             }
1950         }
1951
1952         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
1953         int mb_size = total_bits - mb_spos;
1954
1955         if( slice_max_size )
1956         {
1957             /* Count the skip run, just in case. */
1958             if( !h->param.b_cabac )
1959                 total_bits += bs_size_ue_big( i_skip );
1960             /* HACK: we assume no more than 3 bytes of NALU escaping, but
1961              * this can fail in CABAC streams with an extremely large number of identical
1962              * blocks in sequence (e.g. all-black intra blocks).
1963              * Thus, every 64 blocks, pretend we've used a byte.
1964              * For reference, a seqeuence of identical empty-CBP i16x16 blocks will use
1965              * one byte after 26 macroblocks, assuming a perfectly adapted CABAC.
1966              * That's 78 macroblocks to generate the 3-byte sequence to trigger an escape. */
1967             else if( ((mb_xy - h->sh.i_first_mb) & 63) == 63 )
1968                 slice_max_size -= 8;
1969             /* We'll just re-encode this last macroblock if we go over the max slice size. */
1970             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
1971             {
1972                 if( mb_xy != h->sh.i_first_mb )
1973                 {
1974                     h->stat.frame.i_mv_bits = mv_bits_bak;
1975                     h->stat.frame.i_tex_bits = tex_bits_bak;
1976                     if( h->param.b_cabac )
1977                     {
1978                         memcpy( &h->cabac, &cabac_bak, offsetof(x264_cabac_t, f8_bits_encoded) );
1979                         h->cabac.p[-1] = cabac_prevbyte_bak;
1980                     }
1981                     else
1982                     {
1983                         h->out.bs = bs_bak;
1984                         i_skip = i_skip_bak;
1985                     }
1986                     h->mb.b_reencode_mb = 1;
1987                     h->sh.i_last_mb = mb_xy-1;
1988                     break;
1989                 }
1990                 else
1991                 {
1992                     h->sh.i_last_mb = mb_xy;
1993                     h->mb.b_reencode_mb = 0;
1994                 }
1995             }
1996             else
1997                 h->mb.b_reencode_mb = 0;
1998         }
1999
2000 #if HAVE_VISUALIZE
2001         if( h->param.b_visualize )
2002             x264_visualize_mb( h );
2003 #endif
2004
2005         /* save cache */
2006         x264_macroblock_cache_save( h );
2007
2008         /* accumulate mb stats */
2009         h->stat.frame.i_mb_count[h->mb.i_type]++;
2010
2011         int b_intra = IS_INTRA( h->mb.i_type );
2012         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2013         {
2014             if( !b_intra && !IS_SKIP( h->mb.i_type ) && !IS_DIRECT( h->mb.i_type ) )
2015             {
2016                 if( h->mb.i_partition != D_8x8 )
2017                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2018                     else
2019                         for( int i = 0; i < 4; i++ )
2020                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2021                 if( h->param.i_frame_reference > 1 )
2022                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2023                         for( int i = 0; i < 4; i++ )
2024                         {
2025                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2026                             if( i_ref >= 0 )
2027                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2028                         }
2029             }
2030         }
2031
2032         if( h->param.i_log_level >= X264_LOG_INFO )
2033         {
2034             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2035             {
2036                 int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2037                            + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2038                 h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2039                 h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2040                 h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2041             }
2042             if( h->mb.i_cbp_luma && !b_intra )
2043             {
2044                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2045                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2046             }
2047             if( b_intra && h->mb.i_type != I_PCM )
2048             {
2049                 if( h->mb.i_type == I_16x16 )
2050                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2051                 else if( h->mb.i_type == I_8x8 )
2052                     for( int i = 0; i < 16; i += 4 )
2053                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2054                 else //if( h->mb.i_type == I_4x4 )
2055                     for( int i = 0; i < 16; i++ )
2056                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2057                 h->stat.frame.i_mb_pred_mode[3][x264_mb_pred_mode8x8c_fix[h->mb.i_chroma_pred_mode]]++;
2058             }
2059         }
2060
2061         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2062         if( b_deblock )
2063         {
2064             int mvy_limit = 4 >> h->sh.b_mbaff;
2065             uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
2066             x264_macroblock_cache_load_deblock( h );
2067             if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
2068                 memset( bs, 3, 2*4*4*sizeof(uint8_t) );
2069             else
2070                 h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
2071                                            bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
2072         }
2073
2074         x264_ratecontrol_mb( h, mb_size );
2075
2076         if( h->sh.b_mbaff )
2077         {
2078             i_mb_x += i_mb_y & 1;
2079             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2080         }
2081         else
2082             i_mb_x++;
2083         if( i_mb_x == h->mb.i_mb_width )
2084         {
2085             i_mb_y++;
2086             i_mb_x = 0;
2087         }
2088     }
2089     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2090
2091     if( h->param.b_cabac )
2092     {
2093         x264_cabac_encode_flush( h, &h->cabac );
2094         h->out.bs.p = h->cabac.p;
2095     }
2096     else
2097     {
2098         if( i_skip > 0 )
2099             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2100         /* rbsp_slice_trailing_bits */
2101         bs_rbsp_trailing( &h->out.bs );
2102         bs_flush( &h->out.bs );
2103     }
2104     if( x264_nal_end( h ) )
2105         return -1;
2106
2107     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2108     {
2109         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2110                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2111                                   - h->stat.frame.i_tex_bits
2112                                   - h->stat.frame.i_mv_bits;
2113         x264_fdec_filter_row( h, h->i_threadslice_end, 1 );
2114     }
2115
2116     return 0;
2117 }
2118
2119 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2120 {
2121     if( dst == src )
2122         return;
2123
2124     // reference counting
2125     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2126         (*f)->i_reference_count++;
2127     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2128         x264_frame_push_unused( src, *f );
2129     src->fdec->i_reference_count++;
2130     x264_frame_push_unused( src, dst->fdec );
2131
2132     // copy everything except the per-thread pointers and the constants.
2133     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2134     dst->param = src->param;
2135     dst->stat = src->stat;
2136 }
2137
2138 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2139 {
2140     if( dst == src )
2141         return;
2142     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2143 }
2144
2145 static void *x264_slices_write( x264_t *h )
2146 {
2147     int i_slice_num = 0;
2148     int last_thread_mb = h->sh.i_last_mb;
2149
2150 #if HAVE_VISUALIZE
2151     if( h->param.b_visualize )
2152         if( x264_visualize_init( h ) )
2153             return (void *)-1;
2154 #endif
2155
2156     /* init stats */
2157     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2158     h->mb.b_reencode_mb = 0;
2159     while( h->sh.i_first_mb <= last_thread_mb )
2160     {
2161         h->sh.i_last_mb = last_thread_mb;
2162         if( h->param.i_slice_max_mbs )
2163             h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2164         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2165         {
2166             int height = h->mb.i_mb_height >> h->param.b_interlaced;
2167             int width = h->mb.i_mb_width << h->param.b_interlaced;
2168             i_slice_num++;
2169             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2170         }
2171         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2172         if( x264_stack_align( x264_slice_write, h ) )
2173             return (void *)-1;
2174         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2175     }
2176
2177 #if HAVE_VISUALIZE
2178     if( h->param.b_visualize )
2179     {
2180         x264_visualize_show( h );
2181         x264_visualize_close( h );
2182     }
2183 #endif
2184
2185     return (void *)0;
2186 }
2187
2188 static int x264_threaded_slices_write( x264_t *h )
2189 {
2190     /* set first/last mb and sync contexts */
2191     for( int i = 0; i < h->param.i_threads; i++ )
2192     {
2193         x264_t *t = h->thread[i];
2194         if( i )
2195         {
2196             t->param = h->param;
2197             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2198         }
2199         int height = h->mb.i_mb_height >> h->param.b_interlaced;
2200         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2201         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << h->param.b_interlaced;
2202         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2203         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2204     }
2205
2206     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2207
2208     x264_threads_distribute_ratecontrol( h );
2209
2210     /* dispatch */
2211     for( int i = 0; i < h->param.i_threads; i++ )
2212     {
2213         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2214         h->thread[i]->b_thread_active = 1;
2215     }
2216     for( int i = 0; i < h->param.i_threads; i++ )
2217     {
2218         h->thread[i]->b_thread_active = 0;
2219         if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) )
2220             return -1;
2221     }
2222
2223     /* Go back and fix up the hpel on the borders between slices. */
2224     for( int i = 1; i < h->param.i_threads; i++ )
2225     {
2226         x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 1, 0 );
2227         if( h->sh.b_mbaff )
2228             x264_fdec_filter_row( h->thread[i], h->thread[i]->i_threadslice_start + 2, 0 );
2229     }
2230
2231     x264_threads_merge_ratecontrol( h );
2232
2233     for( int i = 1; i < h->param.i_threads; i++ )
2234     {
2235         x264_t *t = h->thread[i];
2236         for( int j = 0; j < t->out.i_nal; j++ )
2237         {
2238             h->out.nal[h->out.i_nal] = t->out.nal[j];
2239             h->out.i_nal++;
2240             x264_nal_check_buffer( h );
2241         }
2242         /* All entries in stat.frame are ints except for ssd/ssim. */
2243         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2244             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2245         for( int j = 0; j < 3; j++ )
2246             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2247         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2248     }
2249
2250     return 0;
2251 }
2252
2253 void x264_encoder_intra_refresh( x264_t *h )
2254 {
2255     h = h->thread[h->i_thread_phase];
2256     h->b_queued_intra_refresh = 1;
2257 }
2258
2259 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2260 {
2261     if( h->param.i_bframe )
2262     {
2263         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2264         return -1;
2265     }
2266     if( h->param.b_intra_refresh )
2267     {
2268         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2269         return -1;
2270     }
2271     h = h->thread[h->i_thread_phase];
2272     if( pts >= h->i_last_idr_pts )
2273     {
2274         for( int i = 0; h->frames.reference[i]; i++ )
2275             if( pts <= h->frames.reference[i]->i_pts )
2276                 h->frames.reference[i]->b_corrupt = 1;
2277         if( pts <= h->fdec->i_pts )
2278             h->fdec->b_corrupt = 1;
2279     }
2280     return 0;
2281 }
2282
2283 /****************************************************************************
2284  * x264_encoder_encode:
2285  *  XXX: i_poc   : is the poc of the current given picture
2286  *       i_frame : is the number of the frame being coded
2287  *  ex:  type frame poc
2288  *       I      0   2*0
2289  *       P      1   2*3
2290  *       B      2   2*1
2291  *       B      3   2*2
2292  *       P      4   2*6
2293  *       B      5   2*4
2294  *       B      6   2*5
2295  ****************************************************************************/
2296 int     x264_encoder_encode( x264_t *h,
2297                              x264_nal_t **pp_nal, int *pi_nal,
2298                              x264_picture_t *pic_in,
2299                              x264_picture_t *pic_out )
2300 {
2301     x264_t *thread_current, *thread_prev, *thread_oldest;
2302     int i_nal_type, i_nal_ref_idc, i_global_qp;
2303     int overhead = NALU_OVERHEAD;
2304
2305     if( h->i_thread_frames > 1 )
2306     {
2307         thread_prev    = h->thread[ h->i_thread_phase ];
2308         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2309         thread_current = h->thread[ h->i_thread_phase ];
2310         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2311         x264_thread_sync_context( thread_current, thread_prev );
2312         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2313         h = thread_current;
2314     }
2315     else
2316     {
2317         thread_current =
2318         thread_oldest  = h;
2319     }
2320 #if HAVE_MMX
2321     if( h->i_thread_frames == 1 && h->param.cpu&X264_CPU_SSE_MISALIGN )
2322         x264_cpu_mask_misalign_sse();
2323 #endif
2324
2325     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2326     if( x264_reference_update( h ) )
2327         return -1;
2328     h->fdec->i_lines_completed = -1;
2329
2330     /* no data out */
2331     *pi_nal = 0;
2332     *pp_nal = NULL;
2333
2334     /* ------------------- Setup new frame from picture -------------------- */
2335     if( pic_in != NULL )
2336     {
2337         /* 1: Copy the picture to a frame and move it to a buffer */
2338         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2339         if( !fenc )
2340             return -1;
2341
2342         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2343             return -1;
2344
2345         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2346             h->param.i_height != 16 * h->mb.i_mb_height )
2347             x264_frame_expand_border_mod16( h, fenc );
2348
2349         fenc->i_frame = h->frames.i_input++;
2350
2351         if( fenc->i_frame == 0 )
2352             h->frames.i_first_pts = fenc->i_pts;
2353         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2354             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2355
2356         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2357             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2358
2359         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2360         h->frames.i_largest_pts = fenc->i_pts;
2361
2362         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2363             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2364
2365         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2366         {
2367             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2368             if( b_interlaced )
2369             {
2370                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2371                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2372             }
2373             else
2374                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2375         }
2376
2377         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2378         {
2379             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2380                 return -1;
2381         }
2382         else
2383             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2384
2385         if( pic_in->prop.quant_offsets_free )
2386             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2387
2388         if( h->frames.b_have_lowres )
2389             x264_frame_init_lowres( h, fenc );
2390
2391         /* 2: Place the frame into the queue for its slice type decision */
2392         x264_lookahead_put_frame( h, fenc );
2393
2394         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2395         {
2396             /* Nothing yet to encode, waiting for filling of buffers */
2397             pic_out->i_type = X264_TYPE_AUTO;
2398             return 0;
2399         }
2400     }
2401     else
2402     {
2403         /* signal kills for lookahead thread */
2404         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2405         h->lookahead->b_exit_thread = 1;
2406         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2407         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2408     }
2409
2410     h->i_frame++;
2411     /* 3: The picture is analyzed in the lookahead */
2412     if( !h->frames.current[0] )
2413         x264_lookahead_get_frames( h );
2414
2415     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2416         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2417
2418     /* ------------------- Get frame to be encoded ------------------------- */
2419     /* 4: get picture to encode */
2420     h->fenc = x264_frame_shift( h->frames.current );
2421     if( h->i_frame == h->i_thread_frames - 1 )
2422         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2423     if( h->fenc->param )
2424     {
2425         x264_encoder_reconfig( h, h->fenc->param );
2426         if( h->fenc->param->param_free )
2427             h->fenc->param->param_free( h->fenc->param );
2428     }
2429
2430     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2431     {
2432         int valid_refs_left = 0;
2433         for( int i = 0; h->frames.reference[i]; i++ )
2434             if( !h->frames.reference[i]->b_corrupt )
2435                 valid_refs_left++;
2436         /* No valid reference frames left: force an IDR. */
2437         if( !valid_refs_left )
2438         {
2439             h->fenc->b_keyframe = 1;
2440             h->fenc->i_type = X264_TYPE_IDR;
2441         }
2442     }
2443
2444     if( h->fenc->b_keyframe )
2445     {
2446         h->frames.i_last_keyframe = h->fenc->i_frame;
2447         if( h->fenc->i_type == X264_TYPE_IDR )
2448         {
2449             h->i_frame_num = 0;
2450             h->frames.i_last_idr = h->fenc->i_frame;
2451         }
2452     }
2453     h->sh.i_mmco_command_count =
2454     h->sh.i_mmco_remove_from_end = 0;
2455     h->b_ref_reorder[0] =
2456     h->b_ref_reorder[1] = 0;
2457     h->fdec->i_poc =
2458     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2459
2460     /* ------------------- Setup frame context ----------------------------- */
2461     /* 5: Init data dependent of frame type */
2462     if( h->fenc->i_type == X264_TYPE_IDR )
2463     {
2464         /* reset ref pictures */
2465         i_nal_type    = NAL_SLICE_IDR;
2466         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2467         h->sh.i_type = SLICE_TYPE_I;
2468         x264_reference_reset( h );
2469         h->frames.i_poc_last_open_gop = -1;
2470     }
2471     else if( h->fenc->i_type == X264_TYPE_I )
2472     {
2473         i_nal_type    = NAL_SLICE;
2474         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2475         h->sh.i_type = SLICE_TYPE_I;
2476         x264_reference_hierarchy_reset( h );
2477         if( h->param.i_open_gop )
2478             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2479     }
2480     else if( h->fenc->i_type == X264_TYPE_P )
2481     {
2482         i_nal_type    = NAL_SLICE;
2483         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2484         h->sh.i_type = SLICE_TYPE_P;
2485         x264_reference_hierarchy_reset( h );
2486         h->frames.i_poc_last_open_gop = -1;
2487     }
2488     else if( h->fenc->i_type == X264_TYPE_BREF )
2489     {
2490         i_nal_type    = NAL_SLICE;
2491         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2492         h->sh.i_type = SLICE_TYPE_B;
2493         x264_reference_hierarchy_reset( h );
2494     }
2495     else    /* B frame */
2496     {
2497         i_nal_type    = NAL_SLICE;
2498         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2499         h->sh.i_type = SLICE_TYPE_B;
2500     }
2501
2502     h->fdec->i_type = h->fenc->i_type;
2503     h->fdec->i_frame = h->fenc->i_frame;
2504     h->fenc->b_kept_as_ref =
2505     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2506
2507     h->fdec->i_pts = h->fenc->i_pts;
2508     if( h->frames.i_bframe_delay )
2509     {
2510         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2511         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2512                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2513                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2514         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2515     }
2516     else
2517         h->fdec->i_dts = h->fenc->i_reordered_pts;
2518     if( h->fenc->i_type == X264_TYPE_IDR )
2519         h->i_last_idr_pts = h->fdec->i_pts;
2520
2521     /* ------------------- Init                ----------------------------- */
2522     /* build ref list 0/1 */
2523     x264_reference_build_list( h, h->fdec->i_poc );
2524
2525     /* ---------------------- Write the bitstream -------------------------- */
2526     /* Init bitstream context */
2527     if( h->param.b_sliced_threads )
2528     {
2529         for( int i = 0; i < h->param.i_threads; i++ )
2530         {
2531             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
2532             h->thread[i]->out.i_nal = 0;
2533         }
2534     }
2535     else
2536     {
2537         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
2538         h->out.i_nal = 0;
2539     }
2540
2541     if( h->param.b_aud )
2542     {
2543         int pic_type;
2544
2545         if( h->sh.i_type == SLICE_TYPE_I )
2546             pic_type = 0;
2547         else if( h->sh.i_type == SLICE_TYPE_P )
2548             pic_type = 1;
2549         else if( h->sh.i_type == SLICE_TYPE_B )
2550             pic_type = 2;
2551         else
2552             pic_type = 7;
2553
2554         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
2555         bs_write( &h->out.bs, 3, pic_type );
2556         bs_rbsp_trailing( &h->out.bs );
2557         if( x264_nal_end( h ) )
2558             return -1;
2559         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2560     }
2561
2562     h->i_nal_type = i_nal_type;
2563     h->i_nal_ref_idc = i_nal_ref_idc;
2564
2565     if( h->param.b_intra_refresh )
2566     {
2567         if( IS_X264_TYPE_I( h->fenc->i_type ) )
2568         {
2569             h->fdec->i_frames_since_pir = 0;
2570             h->b_queued_intra_refresh = 0;
2571             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
2572              * the whole frame and counts as an intra refresh. */
2573             h->fdec->f_pir_position = h->mb.i_mb_width;
2574         }
2575         else if( h->fenc->i_type == X264_TYPE_P )
2576         {
2577             int pocdiff = (h->fdec->i_poc - h->fref0[0]->i_poc)/2;
2578             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
2579             h->fdec->f_pir_position = h->fref0[0]->f_pir_position;
2580             h->fdec->i_frames_since_pir = h->fref0[0]->i_frames_since_pir + pocdiff;
2581             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
2582                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
2583             {
2584                 h->fdec->f_pir_position = 0;
2585                 h->fdec->i_frames_since_pir = 0;
2586                 h->b_queued_intra_refresh = 0;
2587                 h->fenc->b_keyframe = 1;
2588             }
2589             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
2590             h->fdec->f_pir_position += increment * pocdiff;
2591             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
2592             /* If our intra refresh has reached the right side of the frame, we're done. */
2593             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
2594                 h->fdec->f_pir_position = h->mb.i_mb_width;
2595         }
2596     }
2597
2598     if( h->fenc->b_keyframe )
2599     {
2600         /* Write SPS and PPS */
2601         if( h->param.b_repeat_headers )
2602         {
2603             /* generate sequence parameters */
2604             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
2605             x264_sps_write( &h->out.bs, h->sps );
2606             if( x264_nal_end( h ) )
2607                 return -1;
2608             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2609
2610             /* generate picture parameters */
2611             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
2612             x264_pps_write( &h->out.bs, h->pps );
2613             if( x264_nal_end( h ) )
2614                 return -1;
2615             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
2616         }
2617
2618         /* buffering period sei is written in x264_encoder_frame_end */
2619     }
2620
2621     /* write extra sei */
2622     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
2623     {
2624         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2625         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
2626                         h->fenc->extra_sei.payloads[i].payload_type );
2627         if( x264_nal_end( h ) )
2628             return -1;
2629         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2630         if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads[i].payload )
2631             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
2632     }
2633
2634     if( h->fenc->extra_sei.sei_free && h->fenc->extra_sei.payloads )
2635         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
2636
2637     if( h->fenc->b_keyframe )
2638     {
2639         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
2640         {
2641             /* identify ourself */
2642             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2643             if( x264_sei_version_write( h, &h->out.bs ) )
2644                 return -1;
2645             if( x264_nal_end( h ) )
2646                 return -1;
2647             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2648         }
2649
2650         if( h->fenc->i_type != X264_TYPE_IDR )
2651         {
2652             int time_to_recovery = h->param.i_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
2653             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2654             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
2655             x264_nal_end( h );
2656             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2657         }
2658     }
2659
2660     /* generate sei pic timing */
2661     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
2662     {
2663         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2664         x264_sei_pic_timing_write( h, &h->out.bs );
2665         if( x264_nal_end( h ) )
2666             return -1;
2667         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
2668     }
2669
2670     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
2671         h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
2672
2673     /* Init the rate control */
2674     /* FIXME: Include slice header bit cost. */
2675     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
2676     i_global_qp = x264_ratecontrol_qp( h );
2677
2678     pic_out->i_qpplus1 =
2679     h->fdec->i_qpplus1 = i_global_qp + 1;
2680
2681     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
2682     {
2683         x264_reference_build_list_optimal( h );
2684         x264_reference_check_reorder( h );
2685     }
2686
2687     if( h->i_ref0 )
2688         h->fdec->i_poc_l0ref0 = h->fref0[0]->i_poc;
2689
2690     if( h->sh.i_type == SLICE_TYPE_B )
2691         x264_macroblock_bipred_init( h );
2692
2693     /*------------------------- Weights -------------------------------------*/
2694     x264_weighted_pred_init( h );
2695
2696     /* ------------------------ Create slice header  ----------------------- */
2697     x264_slice_init( h, i_nal_type, i_global_qp );
2698
2699     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
2700         h->i_frame_num++;
2701
2702     /* Write frame */
2703     h->i_threadslice_start = 0;
2704     h->i_threadslice_end = h->mb.i_mb_height;
2705     if( h->i_thread_frames > 1 )
2706     {
2707         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
2708         h->b_thread_active = 1;
2709     }
2710     else if( h->param.b_sliced_threads )
2711     {
2712         if( x264_threaded_slices_write( h ) )
2713             return -1;
2714     }
2715     else
2716         if( (intptr_t)x264_slices_write( h ) )
2717             return -1;
2718
2719     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2720 }
2721
2722 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
2723                                    x264_nal_t **pp_nal, int *pi_nal,
2724                                    x264_picture_t *pic_out )
2725 {
2726     char psz_message[80];
2727
2728     if( h->b_thread_active )
2729     {
2730         h->b_thread_active = 0;
2731         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
2732             return -1;
2733     }
2734     if( !h->out.i_nal )
2735     {
2736         pic_out->i_type = X264_TYPE_AUTO;
2737         return 0;
2738     }
2739
2740     x264_emms();
2741     /* generate sei buffering period and insert it into place */
2742     if( h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
2743     {
2744         x264_hrd_fullness( h );
2745         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2746         x264_sei_buffering_period_write( h, &h->out.bs );
2747         if( x264_nal_end( h ) )
2748            return -1;
2749         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
2750         int idx = 0;
2751         while( h->out.nal[idx].i_type == NAL_AUD ||
2752                h->out.nal[idx].i_type == NAL_SPS ||
2753                h->out.nal[idx].i_type == NAL_PPS )
2754             idx++;
2755         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
2756         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
2757         h->out.nal[idx] = nal_tmp;
2758     }
2759
2760     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
2761
2762     /* Set output picture properties */
2763     pic_out->i_type = h->fenc->i_type;
2764
2765     pic_out->b_keyframe = h->fenc->b_keyframe;
2766
2767     pic_out->i_pts = h->fdec->i_pts;
2768     pic_out->i_dts = h->fdec->i_dts;
2769
2770     if( pic_out->i_pts < pic_out->i_dts )
2771         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
2772
2773     pic_out->img.i_csp = X264_CSP_NV12;
2774 #if X264_HIGH_BIT_DEPTH
2775     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
2776 #endif
2777     pic_out->img.i_plane = h->fdec->i_plane;
2778     for( int i = 0; i < 2; i++ )
2779     {
2780         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
2781         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
2782     }
2783
2784     x264_frame_push_unused( thread_current, h->fenc );
2785
2786     /* ---------------------- Update encoder state ------------------------- */
2787
2788     /* update rc */
2789     int filler = 0;
2790     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
2791         return -1;
2792
2793     pic_out->hrd_timing = h->fenc->hrd_timing;
2794
2795     while( filler > 0 )
2796     {
2797         int f, overhead;
2798         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
2799         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
2800         {
2801             int next_size = filler - h->param.i_slice_max_size;
2802             int overflow = X264_MAX( overhead - next_size, 0 );
2803             f = h->param.i_slice_max_size - overhead - overflow;
2804         }
2805         else
2806             f = X264_MAX( 0, filler - overhead );
2807
2808         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
2809         x264_filler_write( h, &h->out.bs, f );
2810         if( x264_nal_end( h ) )
2811             return -1;
2812         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
2813         frame_size += total_size;
2814         filler -= total_size;
2815     }
2816
2817     /* End bitstream, set output  */
2818     *pi_nal = h->out.i_nal;
2819     *pp_nal = h->out.nal;
2820
2821     h->out.i_nal = 0;
2822
2823     x264_noise_reduction_update( h );
2824
2825     /* ---------------------- Compute/Print statistics --------------------- */
2826     x264_thread_sync_stat( h, h->thread[0] );
2827
2828     /* Slice stat */
2829     h->stat.i_frame_count[h->sh.i_type]++;
2830     h->stat.i_frame_size[h->sh.i_type] += frame_size;
2831     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
2832
2833     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
2834         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
2835     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
2836         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
2837     for( int i = 0; i < 2; i++ )
2838         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
2839     for( int i = 0; i < 6; i++ )
2840         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
2841     for( int i = 0; i < 4; i++ )
2842         for( int j = 0; j < 13; j++ )
2843             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
2844     if( h->sh.i_type != SLICE_TYPE_I )
2845         for( int i_list = 0; i_list < 2; i_list++ )
2846             for( int i = 0; i < X264_REF_MAX*2; i++ )
2847                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
2848     if( h->sh.i_type == SLICE_TYPE_P )
2849     {
2850         h->stat.i_consecutive_bframes[h->fdec->i_frame - h->fref0[0]->i_frame - 1]++;
2851         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
2852             for( int i = 0; i < 3; i++ )
2853                 for( int j = 0; j < h->i_ref0; j++ )
2854                     if( h->sh.weight[0][i].i_denom != 0 )
2855                     {
2856                         h->stat.i_wpred[i]++;
2857                         break;
2858                     }
2859     }
2860     if( h->sh.i_type == SLICE_TYPE_B )
2861     {
2862         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
2863         if( h->mb.b_direct_auto_write )
2864         {
2865             //FIXME somewhat arbitrary time constants
2866             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
2867                 for( int i = 0; i < 2; i++ )
2868                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
2869             for( int i = 0; i < 2; i++ )
2870                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
2871         }
2872     }
2873
2874     psz_message[0] = '\0';
2875     double dur = h->fenc->f_duration;
2876     h->stat.f_frame_duration[h->sh.i_type] += dur;
2877     if( h->param.analyse.b_psnr )
2878     {
2879         int64_t ssd[3] =
2880         {
2881             h->stat.frame.i_ssd[0],
2882             h->stat.frame.i_ssd[1],
2883             h->stat.frame.i_ssd[2],
2884         };
2885
2886         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
2887         h->stat.f_psnr_average[h->sh.i_type] += dur * x264_psnr( ssd[0] + ssd[1] + ssd[2], 3 * h->param.i_width * h->param.i_height / 2 );
2888         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * x264_psnr( ssd[0], h->param.i_width * h->param.i_height );
2889         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4 );
2890         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4 );
2891
2892         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f",
2893                   x264_psnr( ssd[0], h->param.i_width * h->param.i_height ),
2894                   x264_psnr( ssd[1], h->param.i_width * h->param.i_height / 4),
2895                   x264_psnr( ssd[2], h->param.i_width * h->param.i_height / 4) );
2896     }
2897
2898     if( h->param.analyse.b_ssim )
2899     {
2900         double ssim_y = h->stat.frame.f_ssim
2901                       / (((h->param.i_width-6)>>2) * ((h->param.i_height-6)>>2));
2902         h->stat.f_ssim_mean_y[h->sh.i_type] += ssim_y * dur;
2903         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
2904                   " SSIM Y:%.5f", ssim_y );
2905     }
2906     psz_message[79] = '\0';
2907
2908     x264_log( h, X264_LOG_DEBUG,
2909                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
2910               h->i_frame,
2911               h->fdec->f_qp_avg_aq,
2912               h->i_nal_ref_idc,
2913               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
2914               h->fdec->i_poc,
2915               h->stat.frame.i_mb_count_i,
2916               h->stat.frame.i_mb_count_p,
2917               h->stat.frame.i_mb_count_skip,
2918               frame_size,
2919               psz_message );
2920
2921     // keep stats all in one place
2922     x264_thread_sync_stat( h->thread[0], h );
2923     // for the use of the next frame
2924     x264_thread_sync_stat( thread_current, h );
2925
2926 #ifdef DEBUG_MB_TYPE
2927 {
2928     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
2929         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
2930     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
2931     {
2932         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
2933             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
2934         else
2935             fprintf( stderr, "? " );
2936
2937         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
2938             fprintf( stderr, "\n" );
2939     }
2940 }
2941 #endif
2942
2943     /* Remove duplicates, must be done near the end as breaks h->fref0 array
2944      * by freeing some of its pointers. */
2945     for( int i = 0; i < h->i_ref0; i++ )
2946         if( h->fref0[i] && h->fref0[i]->b_duplicate )
2947         {
2948             x264_frame_push_blank_unused( h, h->fref0[i] );
2949             h->fref0[i] = 0;
2950         }
2951
2952     if( h->param.psz_dump_yuv )
2953         x264_frame_dump( h );
2954     x264_emms();
2955
2956     return frame_size;
2957 }
2958
2959 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
2960 {
2961     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
2962         b_print_pcm ? "..PCM" : "",
2963         i_mb_count[I_16x16]/ i_count,
2964         i_mb_count[I_8x8]  / i_count,
2965         i_mb_count[I_4x4]  / i_count );
2966     if( b_print_pcm )
2967         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
2968 }
2969
2970 /****************************************************************************
2971  * x264_encoder_close:
2972  ****************************************************************************/
2973 void    x264_encoder_close  ( x264_t *h )
2974 {
2975     int64_t i_yuv_size = 3 * h->param.i_width * h->param.i_height / 2;
2976     int64_t i_mb_count_size[2][7] = {{0}};
2977     char buf[200];
2978     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
2979                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
2980                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
2981
2982     x264_lookahead_delete( h );
2983
2984     if( h->param.i_threads > 1 )
2985         x264_threadpool_delete( h->threadpool );
2986     if( h->i_thread_frames > 1 )
2987     {
2988         for( int i = 0; i < h->i_thread_frames; i++ )
2989             if( h->thread[i]->b_thread_active )
2990             {
2991                 assert( h->thread[i]->fenc->i_reference_count == 1 );
2992                 x264_frame_delete( h->thread[i]->fenc );
2993             }
2994
2995         x264_t *thread_prev = h->thread[h->i_thread_phase];
2996         x264_thread_sync_ratecontrol( h, thread_prev, h );
2997         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
2998         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
2999     }
3000     h->i_frame++;
3001
3002     /* Slices used and PSNR */
3003     for( int i = 0; i < 3; i++ )
3004     {
3005         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3006         int i_slice = slice_order[i];
3007
3008         if( h->stat.i_frame_count[i_slice] > 0 )
3009         {
3010             int i_count = h->stat.i_frame_count[i_slice];
3011             double dur =  h->stat.f_frame_duration[i_slice];
3012             if( h->param.analyse.b_psnr )
3013             {
3014                 x264_log( h, X264_LOG_INFO,
3015                           "frame %c:%-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",
3016                           slice_type_to_char[i_slice],
3017                           i_count,
3018                           h->stat.f_frame_qp[i_slice] / i_count,
3019                           (double)h->stat.i_frame_size[i_slice] / i_count,
3020                           h->stat.f_psnr_mean_y[i_slice] / dur, h->stat.f_psnr_mean_u[i_slice] / dur, h->stat.f_psnr_mean_v[i_slice] / dur,
3021                           h->stat.f_psnr_average[i_slice] / dur,
3022                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3023             }
3024             else
3025             {
3026                 x264_log( h, X264_LOG_INFO,
3027                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3028                           slice_type_to_char[i_slice],
3029                           i_count,
3030                           h->stat.f_frame_qp[i_slice] / i_count,
3031                           (double)h->stat.i_frame_size[i_slice] / i_count );
3032             }
3033         }
3034     }
3035     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_P] )
3036     {
3037         char *p = buf;
3038         int den = 0;
3039         // weight by number of frames (including the P-frame) that are in a sequence of N B-frames
3040         for( int i = 0; i <= h->param.i_bframe; i++ )
3041             den += (i+1) * h->stat.i_consecutive_bframes[i];
3042         for( int i = 0; i <= h->param.i_bframe; i++ )
3043             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3044         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3045     }
3046
3047     for( int i_type = 0; i_type < 2; i_type++ )
3048         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3049         {
3050             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3051             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3052         }
3053
3054     /* MB types used */
3055     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3056     {
3057         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3058         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3059         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3060         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3061     }
3062     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3063     {
3064         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3065         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3066         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3067         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3068         x264_log( h, X264_LOG_INFO,
3069                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3070                   buf,
3071                   i_mb_size[PIXEL_16x16] / (i_count*4),
3072                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3073                   i_mb_size[PIXEL_8x8] / (i_count*4),
3074                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3075                   i_mb_size[PIXEL_4x4] / (i_count*4),
3076                   i_mb_count[P_SKIP] / i_count );
3077     }
3078     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3079     {
3080         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3081         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3082         double i_mb_list_count;
3083         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3084         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3085         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3086         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3087             for( int j = 0; j < 2; j++ )
3088             {
3089                 int l0 = x264_mb_type_list_table[i][0][j];
3090                 int l1 = x264_mb_type_list_table[i][1][j];
3091                 if( l0 || l1 )
3092                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3093             }
3094         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3095         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3096         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3097         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3098         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3099         x264_log( h, X264_LOG_INFO,
3100                   "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",
3101                   buf,
3102                   i_mb_size[PIXEL_16x16] / (i_count*4),
3103                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3104                   i_mb_size[PIXEL_8x8] / (i_count*4),
3105                   i_mb_count[B_DIRECT] / i_count,
3106                   i_mb_count[B_SKIP]   / i_count,
3107                   list_count[0] / i_mb_list_count,
3108                   list_count[1] / i_mb_list_count,
3109                   list_count[2] / i_mb_list_count );
3110     }
3111
3112     x264_ratecontrol_summary( h );
3113
3114     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 )
3115     {
3116 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3117 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3118         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3119         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3120                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3121         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3122         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3123                             h->stat.i_frame_count[SLICE_TYPE_P] +
3124                             h->stat.i_frame_count[SLICE_TYPE_B];
3125         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3126                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3127                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3128         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3129         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3130
3131         if( h->pps->b_transform_8x8_mode )
3132         {
3133             buf[0] = 0;
3134             if( h->stat.i_mb_count_8x8dct[0] )
3135                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3136             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3137         }
3138
3139         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3140             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3141             && h->stat.i_frame_count[SLICE_TYPE_B] )
3142         {
3143             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3144                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3145                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3146         }
3147
3148         buf[0] = 0;
3149         if( i_mb_count != i_all_intra )
3150             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3151                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3152                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)  ),
3153                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)) );
3154         x264_log( h, X264_LOG_INFO, "coded y,uvDC,uvAC intra: %.1f%% %.1f%% %.1f%%%s\n",
3155                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3156                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra  ),
3157                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra  ), buf );
3158
3159         int64_t fixed_pred_modes[4][9] = {{0}};
3160         int64_t sum_pred_modes[4] = {0};
3161         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3162         {
3163             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3164             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3165         }
3166         if( sum_pred_modes[0] )
3167             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3168                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3169                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3170                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3171                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3172         for( int i = 1; i <= 2; i++ )
3173         {
3174             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3175             {
3176                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3177                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3178             }
3179             if( sum_pred_modes[i] )
3180                 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,
3181                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3182                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3183                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3184                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3185                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3186                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3187                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3188                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3189                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3190         }
3191         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3192         {
3193             fixed_pred_modes[3][x264_mb_pred_mode8x8c_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3194             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3195         }
3196         if( sum_pred_modes[3] )
3197             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3198                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3199                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3200                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3201                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3202
3203         if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3204             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%%\n",
3205                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3206
3207         for( int i_list = 0; i_list < 2; i_list++ )
3208             for( int i_slice = 0; i_slice < 2; i_slice++ )
3209             {
3210                 char *p = buf;
3211                 int64_t i_den = 0;
3212                 int i_max = 0;
3213                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3214                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3215                     {
3216                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3217                         i_max = i;
3218                     }
3219                 if( i_max == 0 )
3220                     continue;
3221                 for( int i = 0; i <= i_max; i++ )
3222                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3223                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3224             }
3225
3226         if( h->param.analyse.b_ssim )
3227         {
3228             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3229             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3230         }
3231         if( h->param.analyse.b_psnr )
3232         {
3233             x264_log( h, X264_LOG_INFO,
3234                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3235                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3236                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3237                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3238                       SUM3( h->stat.f_psnr_average ) / duration,
3239                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3240                       f_bitrate );
3241         }
3242         else
3243             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3244     }
3245
3246     /* rc */
3247     x264_ratecontrol_delete( h );
3248
3249     /* param */
3250     if( h->param.rc.psz_stat_out )
3251         free( h->param.rc.psz_stat_out );
3252     if( h->param.rc.psz_stat_in )
3253         free( h->param.rc.psz_stat_in );
3254
3255     x264_cqm_delete( h );
3256     x264_free( h->nal_buffer );
3257     x264_analyse_free_costs( h );
3258
3259     if( h->i_thread_frames > 1)
3260         h = h->thread[h->i_thread_phase];
3261
3262     /* frames */
3263     x264_frame_delete_list( h->frames.unused[0] );
3264     x264_frame_delete_list( h->frames.unused[1] );
3265     x264_frame_delete_list( h->frames.current );
3266     x264_frame_delete_list( h->frames.blank_unused );
3267
3268     h = h->thread[0];
3269
3270     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3271     {
3272         x264_frame_t **frame;
3273
3274         if( !h->param.b_sliced_threads || i == 0 )
3275         {
3276             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3277             {
3278                 assert( (*frame)->i_reference_count > 0 );
3279                 (*frame)->i_reference_count--;
3280                 if( (*frame)->i_reference_count == 0 )
3281                     x264_frame_delete( *frame );
3282             }
3283             frame = &h->thread[i]->fdec;
3284             assert( (*frame)->i_reference_count > 0 );
3285             (*frame)->i_reference_count--;
3286             if( (*frame)->i_reference_count == 0 )
3287                 x264_frame_delete( *frame );
3288             x264_macroblock_cache_free( h->thread[i] );
3289         }
3290         x264_macroblock_thread_free( h->thread[i], 0 );
3291         x264_free( h->thread[i]->out.p_bitstream );
3292         x264_free( h->thread[i]->out.nal);
3293         x264_free( h->thread[i] );
3294     }
3295 }
3296
3297 /****************************************************************************
3298  * x264_encoder_delayed_frames:
3299  ****************************************************************************/
3300 int x264_encoder_delayed_frames( x264_t *h )
3301 {
3302     int delayed_frames = 0;
3303     if( h->i_thread_frames > 1 )
3304     {
3305         for( int i = 0; i < h->i_thread_frames; i++ )
3306             delayed_frames += h->thread[i]->b_thread_active;
3307         h = h->thread[h->i_thread_phase];
3308     }
3309     for( int i = 0; h->frames.current[i]; i++ )
3310         delayed_frames++;
3311     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3312     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3313     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3314     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3315     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3316     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3317     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3318     return delayed_frames;
3319 }