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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2015 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 "common/common.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34 #include "me.h"
  35 #if HAVE_INTEL_DISPATCHER
  36 #include "extras/intel_dispatcher.h"
  37 #endif
  38
  39 //#define DEBUG_MB_TYPE
  40
  41 #define bs_write_ue bs_write_ue_big
  42
  43 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  44                                    x264_nal_t **pp_nal, int *pi_nal,
  45                                    x264_picture_t *pic_out );
  46
  47 /****************************************************************************
  48  *
  49  ******************************* x264 libs **********************************
  50  *
  51  ****************************************************************************/
  52 static double x264_psnr( double sqe, double size )
  53 {
  54     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  55     if( mse <= 0.0000000001 ) /* Max 100dB */
  56         return 100;
  57
  58     return -10.0 * log10( mse );
  59 }
  60
  61 static double x264_ssim( double ssim )
  62 {
  63     double inv_ssim = 1 - ssim;
  64     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
  65         return 100;
  66
  67     return -10.0 * log10( inv_ssim );
  68 }
  69
  70 static int x264_threadpool_wait_all( x264_t *h )
  71 {
  72     for( int i = 0; i < h->param.i_threads; i++ )
  73         if( h->thread[i]->b_thread_active )
  74         {
  75             h->thread[i]->b_thread_active = 0;
  76             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
  77                 return -1;
  78         }
  79     return 0;
  80 }
  81
  82 static void x264_frame_dump( x264_t *h )
  83 {
  84     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
  85     if( !f )
  86         return;
  87
  88     /* Wait for the threads to finish deblocking */
  89     if( h->param.b_sliced_threads )
  90         x264_threadpool_wait_all( h );
  91
  92     /* Write the frame in display order */
  93     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
  94     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
  95     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
  96         for( int y = 0; y < h->param.i_height; y++ )
  97             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
  98     if( !CHROMA444 )
  99     {
 100         int cw = h->param.i_width>>1;
 101         int ch = h->param.i_height>>CHROMA_V_SHIFT;
 102         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
 103         if( planeu )
 104         {
 105             pixel *planev = planeu + cw*ch + 16;
 106             h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
 107             fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
 108             fwrite( planev, 1, cw*ch*sizeof(pixel), f );
 109             x264_free( planeu );
 110         }
 111     }
 112     fclose( f );
 113 }
 114
 115 /* Fill "default" values */
 116 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 117                                     x264_sps_t *sps, x264_pps_t *pps,
 118                                     int i_idr_pic_id, int i_frame, int i_qp )
 119 {
 120     x264_param_t *param = &h->param;
 121
 122     /* First we fill all fields */
 123     sh->sps = sps;
 124     sh->pps = pps;
 125
 126     sh->i_first_mb  = 0;
 127     sh->i_last_mb   = h->mb.i_mb_count - 1;
 128     sh->i_pps_id    = pps->i_id;
 129
 130     sh->i_frame_num = i_frame;
 131
 132     sh->b_mbaff = PARAM_INTERLACED;
 133     sh->b_field_pic = 0;    /* no field support for now */
 134     sh->b_bottom_field = 0; /* not yet used */
 135
 136     sh->i_idr_pic_id = i_idr_pic_id;
 137
 138     /* poc stuff, fixed later */
 139     sh->i_poc = 0;
 140     sh->i_delta_poc_bottom = 0;
 141     sh->i_delta_poc[0] = 0;
 142     sh->i_delta_poc[1] = 0;
 143
 144     sh->i_redundant_pic_cnt = 0;
 145
 146     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 147                                 && h->param.i_bframe
 148                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 149
 150     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 151     {
 152         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
 153         {
 154             if( h->mb.b_direct_auto_write )
 155                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 156             else
 157                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 158         }
 159         else
 160         {
 161             h->mb.b_direct_auto_write = 0;
 162             sh->b_direct_spatial_mv_pred = 1;
 163         }
 164     }
 165     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 166
 167     sh->b_num_ref_idx_override = 0;
 168     sh->i_num_ref_idx_l0_active = 1;
 169     sh->i_num_ref_idx_l1_active = 1;
 170
 171     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
 172     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
 173
 174     /* If the ref list isn't in the default order, construct reordering header */
 175     for( int list = 0; list < 2; list++ )
 176     {
 177         if( sh->b_ref_pic_list_reordering[list] )
 178         {
 179             int pred_frame_num = i_frame;
 180             for( int i = 0; i < h->i_ref[list]; i++ )
 181             {
 182                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
 183                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
 184                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 185                 pred_frame_num = h->fref[list][i]->i_frame_num;
 186             }
 187         }
 188     }
 189
 190     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 191
 192     sh->i_qp = SPEC_QP(i_qp);
 193     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
 194     sh->b_sp_for_swidth = 0;
 195     sh->i_qs_delta = 0;
 196
 197     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 198     /* If effective qp <= 15, deblocking would have no effect anyway */
 199     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 200         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 201     else
 202         sh->i_disable_deblocking_filter_idc = 1;
 203     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 204     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 205 }
 206
 207 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 208 {
 209     if( sh->b_mbaff )
 210     {
 211         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
 212         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
 213         assert( (first_y&1) == 0 );
 214         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
 215     }
 216     else
 217         bs_write_ue( s, sh->i_first_mb );
 218
 219     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 220     bs_write_ue( s, sh->i_pps_id );
 221     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 222
 223     if( !sh->sps->b_frame_mbs_only )
 224     {
 225         bs_write1( s, sh->b_field_pic );
 226         if( sh->b_field_pic )
 227             bs_write1( s, sh->b_bottom_field );
 228     }
 229
 230     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 231         bs_write_ue( s, sh->i_idr_pic_id );
 232
 233     if( sh->sps->i_poc_type == 0 )
 234     {
 235         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 236         if( sh->pps->b_pic_order && !sh->b_field_pic )
 237             bs_write_se( s, sh->i_delta_poc_bottom );
 238     }
 239
 240     if( sh->pps->b_redundant_pic_cnt )
 241         bs_write_ue( s, sh->i_redundant_pic_cnt );
 242
 243     if( sh->i_type == SLICE_TYPE_B )
 244         bs_write1( s, sh->b_direct_spatial_mv_pred );
 245
 246     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 247     {
 248         bs_write1( s, sh->b_num_ref_idx_override );
 249         if( sh->b_num_ref_idx_override )
 250         {
 251             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 252             if( sh->i_type == SLICE_TYPE_B )
 253                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 254         }
 255     }
 256
 257     /* ref pic list reordering */
 258     if( sh->i_type != SLICE_TYPE_I )
 259     {
 260         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
 261         if( sh->b_ref_pic_list_reordering[0] )
 262         {
 263             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 264             {
 265                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 266                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 267             }
 268             bs_write_ue( s, 3 );
 269         }
 270     }
 271     if( sh->i_type == SLICE_TYPE_B )
 272     {
 273         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
 274         if( sh->b_ref_pic_list_reordering[1] )
 275         {
 276             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 277             {
 278                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 279                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 280             }
 281             bs_write_ue( s, 3 );
 282         }
 283     }
 284
 285     sh->b_weighted_pred = 0;
 286     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 287     {
 288         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
 289         /* pred_weight_table() */
 290         bs_write_ue( s, sh->weight[0][0].i_denom );
 291         bs_write_ue( s, sh->weight[0][1].i_denom );
 292         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 293         {
 294             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 295             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 296             bs_write1( s, luma_weight_l0_flag );
 297             if( luma_weight_l0_flag )
 298             {
 299                 bs_write_se( s, sh->weight[i][0].i_scale );
 300                 bs_write_se( s, sh->weight[i][0].i_offset );
 301             }
 302             bs_write1( s, chroma_weight_l0_flag );
 303             if( chroma_weight_l0_flag )
 304             {
 305                 for( int j = 1; j < 3; j++ )
 306                 {
 307                     bs_write_se( s, sh->weight[i][j].i_scale );
 308                     bs_write_se( s, sh->weight[i][j].i_offset );
 309                 }
 310             }
 311         }
 312     }
 313     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 314     {
 315       /* TODO */
 316     }
 317
 318     if( i_nal_ref_idc != 0 )
 319     {
 320         if( sh->i_idr_pic_id >= 0 )
 321         {
 322             bs_write1( s, 0 );  /* no output of prior pics flag */
 323             bs_write1( s, 0 );  /* long term reference flag */
 324         }
 325         else
 326         {
 327             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 328             if( sh->i_mmco_command_count > 0 )
 329             {
 330                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 331                 {
 332                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 333                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 334                 }
 335                 bs_write_ue( s, 0 ); /* end command list */
 336             }
 337         }
 338     }
 339
 340     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 341         bs_write_ue( s, sh->i_cabac_init_idc );
 342
 343     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 344
 345     if( sh->pps->b_deblocking_filter_control )
 346     {
 347         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 348         if( sh->i_disable_deblocking_filter_idc != 1 )
 349         {
 350             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 351             bs_write_se( s, sh->i_beta_offset >> 1 );
 352         }
 353     }
 354 }
 355
 356 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 357 /* reallocate, adding an arbitrary amount of space. */
 358 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
 359 {
 360     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
 361         (h->out.bs.p_end - h->out.bs.p < size) )
 362     {
 363         int buf_size = h->out.i_bitstream + size;
 364         uint8_t *buf = x264_malloc( buf_size );
 365         if( !buf )
 366             return -1;
 367         int aligned_size = h->out.i_bitstream & ~15;
 368         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
 369         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
 370
 371         intptr_t delta = buf - h->out.p_bitstream;
 372
 373         h->out.bs.p_start += delta;
 374         h->out.bs.p += delta;
 375         h->out.bs.p_end = buf + buf_size;
 376
 377         h->cabac.p_start += delta;
 378         h->cabac.p += delta;
 379         h->cabac.p_end = buf + buf_size;
 380
 381         for( int i = 0; i <= i_nal; i++ )
 382             h->out.nal[i].p_payload += delta;
 383
 384         x264_free( h->out.p_bitstream );
 385         h->out.p_bitstream = buf;
 386         h->out.i_bitstream = buf_size;
 387     }
 388     return 0;
 389 }
 390
 391 static int x264_bitstream_check_buffer( x264_t *h )
 392 {
 393     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
 394     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
 395 }
 396
 397 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
 398 {
 399     filler += 32; // add padding for safety
 400     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
 401 }
 402
 403 #if HAVE_THREAD
 404 static void x264_encoder_thread_init( x264_t *h )
 405 {
 406     if( h->param.i_sync_lookahead )
 407         x264_lower_thread_priority( 10 );
 408 }
 409 #endif
 410
 411 /****************************************************************************
 412  *
 413  ****************************************************************************
 414  ****************************** External API*********************************
 415  ****************************************************************************
 416  *
 417  ****************************************************************************/
 418
 419 static int x264_validate_parameters( x264_t *h, int b_open )
 420 {
 421     if( !h->param.pf_log )
 422     {
 423         x264_log( NULL, X264_LOG_ERROR, "pf_log not set! did you forget to call x264_param_default?\n" );
 424         return -1;
 425     }
 426
 427 #if HAVE_MMX
 428     if( b_open )
 429     {
 430         int cpuflags = x264_cpu_detect();
 431         int fail = 0;
 432 #ifdef __SSE__
 433         if( !(cpuflags & X264_CPU_SSE) )
 434         {
 435             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
 436             fail = 1;
 437         }
 438 #else
 439         if( !(cpuflags & X264_CPU_MMX2) )
 440         {
 441             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
 442             fail = 1;
 443         }
 444 #endif
 445         if( !fail && !(cpuflags & X264_CPU_CMOV) )
 446         {
 447             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
 448             fail = 1;
 449         }
 450         if( fail )
 451         {
 452             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
 453             return -1;
 454         }
 455     }
 456 #endif
 457
 458 #if HAVE_INTERLACED
 459     h->param.b_interlaced = !!PARAM_INTERLACED;
 460 #else
 461     if( h->param.b_interlaced )
 462     {
 463         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
 464         return -1;
 465     }
 466 #endif
 467
 468     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 469     {
 470         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 471                   h->param.i_width, h->param.i_height );
 472         return -1;
 473     }
 474
 475     int i_csp = h->param.i_csp & X264_CSP_MASK;
 476 #if X264_CHROMA_FORMAT
 477     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
 478     {
 479         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
 480         return -1;
 481     }
 482     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_V210 )
 483     {
 484         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
 485         return -1;
 486     }
 487     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
 488     {
 489         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
 490         return -1;
 491     }
 492 #endif
 493     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 494     {
 495         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
 496         return -1;
 497     }
 498
 499     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
 500     {
 501         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
 502                   h->param.i_width, h->param.i_height );
 503         return -1;
 504     }
 505
 506     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
 507     {
 508         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
 509                   h->param.i_width, h->param.i_height );
 510         return -1;
 511     }
 512
 513     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
 514     {
 515         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
 516                   h->param.i_width, h->param.i_height );
 517         return -1;
 518     }
 519
 520     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
 521         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
 522     {
 523         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 524                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 525         return -1;
 526     }
 527
 528     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
 529     {
 530         h->param.vui.i_sar_width = 0;
 531         h->param.vui.i_sar_height = 0;
 532     }
 533
 534     if( h->param.i_threads == X264_THREADS_AUTO )
 535         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 536     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
 537     if( h->param.i_threads > 1 )
 538     {
 539 #if !HAVE_THREAD
 540         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 541         h->param.i_threads = 1;
 542 #endif
 543         /* Avoid absurdly small thread slices as they can reduce performance
 544          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 545         if( h->param.b_sliced_threads )
 546             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
 547     }
 548     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 549     if( h->param.i_threads == 1 )
 550     {
 551         h->param.b_sliced_threads = 0;
 552         h->param.i_lookahead_threads = 1;
 553     }
 554     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 555     if( h->i_thread_frames > 1 )
 556         h->param.nalu_process = NULL;
 557
 558     if( h->param.b_opencl )
 559     {
 560 #if !HAVE_OPENCL
 561         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
 562         h->param.b_opencl = 0;
 563 #elif BIT_DEPTH > 8
 564         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
 565         h->param.b_opencl = 0;
 566 #else
 567         if( h->param.i_width < 32 || h->param.i_height < 32 )
 568         {
 569             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
 570             h->param.b_opencl = 0;
 571         }
 572 #endif
 573         if( h->param.opencl_device_id && h->param.i_opencl_device )
 574         {
 575             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
 576             h->param.i_opencl_device = 0;
 577         }
 578     }
 579
 580     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 581     if( h->param.i_keyint_max == 1 )
 582     {
 583         h->param.b_intra_refresh = 0;
 584         h->param.analyse.i_weighted_pred = 0;
 585         h->param.i_frame_reference = 1;
 586         h->param.i_dpb_size = 1;
 587     }
 588
 589     if( h->param.i_frame_packing < -1 || h->param.i_frame_packing > 6 )
 590     {
 591         x264_log( h, X264_LOG_WARNING, "ignoring unknown frame packing value\n" );
 592         h->param.i_frame_packing = -1;
 593     }
 594
 595     /* Detect default ffmpeg settings and terminate with an error. */
 596     if( b_open )
 597     {
 598         int score = 0;
 599         score += h->param.analyse.i_me_range == 0;
 600         score += h->param.rc.i_qp_step == 3;
 601         score += h->param.i_keyint_max == 12;
 602         score += h->param.rc.i_qp_min == 2;
 603         score += h->param.rc.i_qp_max == 31;
 604         score += h->param.rc.f_qcompress == 0.5;
 605         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 606         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 607         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 608         if( score >= 5 )
 609         {
 610             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 611             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 612             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 613             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 614             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 615             return -1;
 616         }
 617     }
 618
 619     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 620     {
 621         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 622         return -1;
 623     }
 624
 625     if( PARAM_INTERLACED )
 626         h->param.b_pic_struct = 1;
 627
 628     if( h->param.i_avcintra_class )
 629     {
 630         if( BIT_DEPTH != 10 )
 631         {
 632             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
 633             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
 634             return -1;
 635         }
 636
 637         int type = h->param.i_avcintra_class == 200 ? 2 :
 638                    h->param.i_avcintra_class == 100 ? 1 :
 639                    h->param.i_avcintra_class == 50 ? 0 : -1;
 640         if( type < 0 )
 641         {
 642             x264_log( h, X264_LOG_ERROR, "Invalid AVC-Intra class\n" );
 643             return -1;
 644         }
 645
 646         /* [50/100/200][res][fps] */
 647         static const struct
 648         {
 649             uint16_t fps_num;
 650             uint16_t fps_den;
 651             uint8_t interlaced;
 652             uint16_t frame_size;
 653             const uint8_t *cqm_4ic;
 654             const uint8_t *cqm_8iy;
 655         } avcintra_lut[3][2][7] =
 656         {
 657             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 658               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 659               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 660               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 661               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
 662              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 663               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 664               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 665               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 666               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 667               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 668               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
 669             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 670               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 671               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 672               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 673               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 674              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 675               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 676               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 677               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 678               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 679               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 680               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}},
 681             {{{ 60000, 1001, 0, 3724, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 682               {    50,    1, 0, 4472, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 683              {{ 30000, 1001, 1, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 684               {    25,    1, 1, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 685               { 60000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 686               { 30000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 687               {    50,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 688               {    25,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 689               { 24000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
 690         };
 691
 692         int res = -1;
 693         if( i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 && !type )
 694         {
 695             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
 696             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
 697         }
 698         else if( i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 && type )
 699         {
 700             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
 701             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
 702         }
 703         else
 704         {
 705             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra %d\n", h->param.i_avcintra_class );
 706             return -1;
 707         }
 708
 709         if( res < 0 )
 710         {
 711             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %d\n",
 712                       h->param.i_width, h->param.i_height, h->param.i_avcintra_class );
 713             return -1;
 714         }
 715
 716         if( h->param.nalu_process )
 717         {
 718             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
 719             return -1;
 720         }
 721
 722         if( !h->param.b_repeat_headers )
 723         {
 724             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
 725             return -1;
 726         }
 727
 728         int i;
 729         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
 730         x264_reduce_fraction( &fps_num, &fps_den );
 731         for( i = 0; i < 7; i++ )
 732         {
 733             if( avcintra_lut[type][res][i].fps_num == fps_num &&
 734                 avcintra_lut[type][res][i].fps_den == fps_den &&
 735                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
 736             {
 737                 break;
 738             }
 739         }
 740         if( i == 7 )
 741         {
 742             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
 743                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
 744             return -1;
 745         }
 746
 747         h->param.i_keyint_max = 1;
 748         h->param.b_intra_refresh = 0;
 749         h->param.analyse.i_weighted_pred = 0;
 750         h->param.i_frame_reference = 1;
 751         h->param.i_dpb_size = 1;
 752
 753         h->param.b_bluray_compat = 0;
 754         h->param.b_vfr_input = 0;
 755         h->param.b_aud = 1;
 756         h->param.vui.i_chroma_loc = 0;
 757         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 758         h->param.b_deblocking_filter = 0;
 759         h->param.b_stitchable = 1;
 760         h->param.b_pic_struct = 0;
 761         h->param.analyse.b_transform_8x8 = 1;
 762         h->param.analyse.intra = X264_ANALYSE_I8x8;
 763         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
 764         h->param.b_cabac = !type;
 765         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
 766         h->param.rc.i_vbv_max_bitrate =
 767         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
 768         h->param.rc.i_rc_method = X264_RC_ABR;
 769         h->param.rc.f_vbv_buffer_init = 1.0;
 770         h->param.rc.b_filler = 1;
 771         h->param.i_cqm_preset = X264_CQM_CUSTOM;
 772         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
 773         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
 774         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
 775
 776         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
 777         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
 778         h->param.i_slice_max_size = 0;
 779         /* The slice structure only allows a maximum of 2 threads for 1080i/p
 780          * and 1 or 5 threads for 720p */
 781         if( h->param.b_sliced_threads )
 782         {
 783             if( res )
 784                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
 785             else
 786             {
 787                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
 788                 if( h->param.i_threads < 5 )
 789                     h->param.i_threads = 1;
 790             }
 791         }
 792
 793         if( type )
 794             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
 795         else
 796         {
 797             h->param.vui.i_sar_width  = 4;
 798             h->param.vui.i_sar_height = 3;
 799         }
 800
 801         /* Official encoder doesn't appear to go under 13
 802          * and Avid cannot handle negative QPs */
 803         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET + 1 );
 804     }
 805
 806     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 807     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
 808     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 809     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
 810     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
 811     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
 812     if( h->param.rc.i_rc_method == X264_RC_CRF )
 813     {
 814         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 815         h->param.rc.i_bitrate = 0;
 816     }
 817     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 818         && h->param.rc.i_qp_constant == 0 )
 819     {
 820         h->mb.b_lossless = 1;
 821         h->param.i_cqm_preset = X264_CQM_FLAT;
 822         h->param.psz_cqm_file = NULL;
 823         h->param.rc.i_rc_method = X264_RC_CQP;
 824         h->param.rc.f_ip_factor = 1;
 825         h->param.rc.f_pb_factor = 1;
 826         h->param.analyse.b_psnr = 0;
 827         h->param.analyse.b_ssim = 0;
 828         h->param.analyse.i_chroma_qp_offset = 0;
 829         h->param.analyse.i_trellis = 0;
 830         h->param.analyse.b_fast_pskip = 0;
 831         h->param.analyse.i_noise_reduction = 0;
 832         h->param.analyse.b_psy = 0;
 833         h->param.i_bframe = 0;
 834         /* 8x8dct is not useful without RD in CAVLC lossless */
 835         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 836             h->param.analyse.b_transform_8x8 = 0;
 837         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 838         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 839     }
 840     if( h->param.rc.i_rc_method == X264_RC_CQP )
 841     {
 842         float qp_p = h->param.rc.i_qp_constant;
 843         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 844         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 845         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 846         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 847         h->param.rc.i_aq_mode = 0;
 848         h->param.rc.b_mb_tree = 0;
 849         h->param.rc.i_bitrate = 0;
 850     }
 851     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 852     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 853     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
 854     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
 855     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
 856     {
 857         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
 858         return -1;
 859     }
 860     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
 861     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
 862     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
 863     if( h->param.rc.i_vbv_buffer_size )
 864     {
 865         if( h->param.rc.i_rc_method == X264_RC_CQP )
 866         {
 867             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 868             h->param.rc.i_vbv_max_bitrate = 0;
 869             h->param.rc.i_vbv_buffer_size = 0;
 870         }
 871         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 872         {
 873             if( h->param.rc.i_rc_method == X264_RC_ABR )
 874             {
 875                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 876                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 877             }
 878             else
 879             {
 880                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 881                 h->param.rc.i_vbv_buffer_size = 0;
 882             }
 883         }
 884         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 885                  h->param.rc.i_rc_method == X264_RC_ABR )
 886         {
 887             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 888             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
 889         }
 890     }
 891     else if( h->param.rc.i_vbv_max_bitrate )
 892     {
 893         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 894         h->param.rc.i_vbv_max_bitrate = 0;
 895     }
 896
 897     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 898     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 899     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
 900     if( h->param.i_slice_max_mbs )
 901         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
 902     else if( !h->param.i_slice_max_size )
 903         h->param.i_slice_min_mbs = 0;
 904     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
 905     {
 906         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
 907         h->param.i_slice_min_mbs = 0;
 908     }
 909     int mb_width = (h->param.i_width+15)/16;
 910     if( h->param.i_slice_min_mbs > mb_width )
 911     {
 912         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
 913         h->param.i_slice_min_mbs = mb_width;
 914     }
 915
 916     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
 917     if( h->param.b_sliced_threads )
 918         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 919     else
 920     {
 921         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 922         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 923             h->param.i_slice_count = 0;
 924     }
 925     if( h->param.i_slice_count_max > 0 )
 926         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
 927
 928     if( h->param.b_bluray_compat )
 929     {
 930         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
 931         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
 932         h->param.b_aud = 1;
 933         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
 934         h->param.i_slice_max_size = 0;
 935         h->param.i_slice_max_mbs = 0;
 936         h->param.b_intra_refresh = 0;
 937         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
 938         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
 939         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
 940         h->param.i_keyint_min = 1;
 941         /* Due to the proliferation of broken players that don't handle dupes properly. */
 942         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
 943         if( h->param.b_fake_interlaced )
 944             h->param.b_pic_struct = 1;
 945     }
 946
 947     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 948     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 949     if( h->param.i_scenecut_threshold < 0 )
 950         h->param.i_scenecut_threshold = 0;
 951     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
 952     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 953     {
 954         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 955         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 956     }
 957     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 958     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 959     if( h->param.i_bframe <= 1 )
 960         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 961     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 962     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
 963     if( !h->param.i_bframe )
 964     {
 965         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 966         h->param.analyse.i_direct_mv_pred = 0;
 967         h->param.analyse.b_weighted_bipred = 0;
 968         h->param.b_open_gop = 0;
 969     }
 970     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 971     {
 972         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 973         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 974     }
 975     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 976     {
 977         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 978         h->param.i_frame_reference = 1;
 979         h->param.i_dpb_size = 1;
 980     }
 981     if( h->param.b_intra_refresh && h->param.b_open_gop )
 982     {
 983         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 984         h->param.b_open_gop = 0;
 985     }
 986     if( !h->param.i_fps_num || !h->param.i_fps_den )
 987     {
 988         h->param.i_fps_num = 25;
 989         h->param.i_fps_den = 1;
 990     }
 991     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
 992     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 993         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 994     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 995     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 996     {
 997         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 998         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 999         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
1000     }
1001
1002     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
1003     {
1004         h->param.i_timebase_num = h->param.i_fps_den;
1005         h->param.i_timebase_den = h->param.i_fps_num;
1006     }
1007
1008     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
1009     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
1010         h->param.rc.b_mb_tree = 0;
1011     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
1012         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
1013     {
1014         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
1015         h->param.rc.b_mb_tree = 0;
1016     }
1017     if( b_open && h->param.rc.b_stat_read )
1018         h->param.rc.i_lookahead = 0;
1019 #if HAVE_THREAD
1020     if( h->param.i_sync_lookahead < 0 )
1021         h->param.i_sync_lookahead = h->param.i_bframe + 1;
1022     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
1023     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
1024         h->param.i_sync_lookahead = 0;
1025 #else
1026     h->param.i_sync_lookahead = 0;
1027 #endif
1028
1029     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
1030     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
1031     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1032     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1033
1034     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1035
1036     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1037         h->param.i_cqm_preset = X264_CQM_FLAT;
1038
1039     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1040         h->param.analyse.i_me_method > X264_ME_TESA )
1041         h->param.analyse.i_me_method = X264_ME_HEX;
1042     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1043     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1044         h->param.analyse.i_me_range = 16;
1045     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1046         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1047         h->param.analyse.i_me_method = X264_ME_ESA;
1048     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1049     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1050                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1051     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1052     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1053         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1054     if( !h->param.analyse.b_transform_8x8 )
1055     {
1056         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1057         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1058     }
1059     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1060     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 3 );
1061     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1062     if( h->param.rc.f_aq_strength == 0 )
1063         h->param.rc.i_aq_mode = 0;
1064
1065     if( h->param.i_log_level < X264_LOG_INFO )
1066     {
1067         h->param.analyse.b_psnr = 0;
1068         h->param.analyse.b_ssim = 0;
1069     }
1070     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1071     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1072     {
1073         char *s = NULL;
1074
1075         if( h->param.analyse.b_psy )
1076         {
1077             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1078             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1079         }
1080         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1081         {
1082             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1083             s = "ssim";
1084         }
1085         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1086         {
1087             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1088             s = "psnr";
1089         }
1090         if( s )
1091             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1092     }
1093
1094     if( !h->param.analyse.b_psy )
1095     {
1096         h->param.analyse.f_psy_rd = 0;
1097         h->param.analyse.f_psy_trellis = 0;
1098     }
1099     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1100     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1101     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1102     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1103     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1104     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1105     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1106         h->param.analyse.i_chroma_qp_offset += 6;
1107     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1108     /* so we lower the chroma QP offset to compensate */
1109     if( b_open && h->mb.i_psy_rd && !h->param.i_avcintra_class )
1110         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1111     /* Psy trellis has a similar effect. */
1112     if( b_open && h->mb.i_psy_trellis && !h->param.i_avcintra_class )
1113         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1114     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1115     /* MB-tree requires AQ to be on, even if the strength is zero. */
1116     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1117     {
1118         h->param.rc.i_aq_mode = 1;
1119         h->param.rc.f_aq_strength = 0;
1120     }
1121     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1122     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1123         h->param.analyse.i_subpel_refine = 9;
1124
1125     {
1126         const x264_level_t *l = x264_levels;
1127         if( h->param.i_level_idc < 0 )
1128         {
1129             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1130             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1131                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1132             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1133             do h->param.i_level_idc = l->level_idc;
1134                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1135             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1136         }
1137         else
1138         {
1139             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1140                 l++;
1141             if( l->level_idc == 0 )
1142             {
1143                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1144                 return -1;
1145             }
1146         }
1147         if( h->param.analyse.i_mv_range <= 0 )
1148             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1149         else
1150             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1151     }
1152
1153     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1154
1155     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1156     {
1157         if( h->param.b_sliced_threads )
1158             h->param.i_lookahead_threads = h->param.i_threads;
1159         else
1160         {
1161             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1162              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1163              * try to guess at this sort of case.
1164              *
1165              * Tuned by a little bit of real encoding with the various presets. */
1166             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1167             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1168             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1169
1170             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1171             static const uint8_t lookahead_thread_div[2][5][4] =
1172             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1173              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1174
1175             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1176             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1177              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1178              * pretty much arbitrarily. */
1179             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1180         }
1181     }
1182     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1183
1184     if( PARAM_INTERLACED )
1185     {
1186         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1187         {
1188             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1189             h->param.analyse.i_me_method = X264_ME_UMH;
1190         }
1191         if( h->param.analyse.i_weighted_pred > 0 )
1192         {
1193             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1194             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1195         }
1196     }
1197
1198     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1199         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1200
1201     if( h->i_thread_frames > 1 )
1202     {
1203         int r = h->param.analyse.i_mv_range_thread;
1204         int r2;
1205         if( r <= 0 )
1206         {
1207             // half of the available space is reserved and divided evenly among the threads,
1208             // the rest is allocated to whichever thread is far enough ahead to use it.
1209             // reserving more space increases quality for some videos, but costs more time
1210             // in thread synchronization.
1211             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1212             r = max_range / 2;
1213         }
1214         r = X264_MAX( r, h->param.analyse.i_me_range );
1215         r = X264_MIN( r, h->param.analyse.i_mv_range );
1216         // round up to use the whole mb row
1217         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1218         if( r2 < r )
1219             r2 += 16;
1220         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1221         h->param.analyse.i_mv_range_thread = r2;
1222     }
1223
1224     if( h->param.rc.f_rate_tolerance < 0 )
1225         h->param.rc.f_rate_tolerance = 0;
1226     if( h->param.rc.f_qblur < 0 )
1227         h->param.rc.f_qblur = 0;
1228     if( h->param.rc.f_complexity_blur < 0 )
1229         h->param.rc.f_complexity_blur = 0;
1230
1231     h->param.i_sps_id &= 31;
1232
1233     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1234
1235     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1236     {
1237         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1238         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1239     }
1240
1241     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1242        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1243     {
1244         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1245         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1246     }
1247
1248     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1249         h->param.rc.b_filler = 1;
1250
1251     /* ensure the booleans are 0 or 1 so they can be used in math */
1252 #define BOOLIFY(x) h->param.x = !!h->param.x
1253     BOOLIFY( b_cabac );
1254     BOOLIFY( b_constrained_intra );
1255     BOOLIFY( b_deblocking_filter );
1256     BOOLIFY( b_deterministic );
1257     BOOLIFY( b_sliced_threads );
1258     BOOLIFY( b_interlaced );
1259     BOOLIFY( b_intra_refresh );
1260     BOOLIFY( b_aud );
1261     BOOLIFY( b_repeat_headers );
1262     BOOLIFY( b_annexb );
1263     BOOLIFY( b_vfr_input );
1264     BOOLIFY( b_pulldown );
1265     BOOLIFY( b_tff );
1266     BOOLIFY( b_pic_struct );
1267     BOOLIFY( b_fake_interlaced );
1268     BOOLIFY( b_open_gop );
1269     BOOLIFY( b_bluray_compat );
1270     BOOLIFY( b_stitchable );
1271     BOOLIFY( b_full_recon );
1272     BOOLIFY( b_opencl );
1273     BOOLIFY( analyse.b_transform_8x8 );
1274     BOOLIFY( analyse.b_weighted_bipred );
1275     BOOLIFY( analyse.b_chroma_me );
1276     BOOLIFY( analyse.b_mixed_references );
1277     BOOLIFY( analyse.b_fast_pskip );
1278     BOOLIFY( analyse.b_dct_decimate );
1279     BOOLIFY( analyse.b_psy );
1280     BOOLIFY( analyse.b_psnr );
1281     BOOLIFY( analyse.b_ssim );
1282     BOOLIFY( rc.b_stat_write );
1283     BOOLIFY( rc.b_stat_read );
1284     BOOLIFY( rc.b_mb_tree );
1285     BOOLIFY( rc.b_filler );
1286 #undef BOOLIFY
1287
1288     return 0;
1289 }
1290
1291 static void mbcmp_init( x264_t *h )
1292 {
1293     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1294     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1295     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1296     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1297     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1298     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1299     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1300     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1301     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1302                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1303     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1304                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1305     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1306     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1307     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1308     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1309 }
1310
1311 static void chroma_dsp_init( x264_t *h )
1312 {
1313     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1314
1315     switch( CHROMA_FORMAT )
1316     {
1317         case CHROMA_420:
1318             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1319             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1320             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1321             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1322             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1323             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1324             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1325             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1326             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1327             break;
1328         case CHROMA_422:
1329             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1330             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1331             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1332             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1333             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1334             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1335             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1336             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1337             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1338             break;
1339         case CHROMA_444:
1340             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1341             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1342             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1343             break;
1344     }
1345 }
1346
1347 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1348 {
1349     /* VUI */
1350     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1351     {
1352         uint32_t i_w = param->vui.i_sar_width;
1353         uint32_t i_h = param->vui.i_sar_height;
1354         uint32_t old_w = h->param.vui.i_sar_width;
1355         uint32_t old_h = h->param.vui.i_sar_height;
1356
1357         x264_reduce_fraction( &i_w, &i_h );
1358
1359         while( i_w > 65535 || i_h > 65535 )
1360         {
1361             i_w /= 2;
1362             i_h /= 2;
1363         }
1364
1365         x264_reduce_fraction( &i_w, &i_h );
1366
1367         if( i_w != old_w || i_h != old_h || initial )
1368         {
1369             h->param.vui.i_sar_width = 0;
1370             h->param.vui.i_sar_height = 0;
1371             if( i_w == 0 || i_h == 0 )
1372                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1373             else
1374             {
1375                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1376                 h->param.vui.i_sar_width = i_w;
1377                 h->param.vui.i_sar_height = i_h;
1378             }
1379         }
1380     }
1381 }
1382
1383 /****************************************************************************
1384  * x264_encoder_open:
1385  ****************************************************************************/
1386 x264_t *x264_encoder_open( x264_param_t *param )
1387 {
1388     x264_t *h;
1389     char buf[1000], *p;
1390     int qp, i_slicetype_length;
1391
1392     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1393
1394     /* Create a copy of param */
1395     memcpy( &h->param, param, sizeof(x264_param_t) );
1396
1397     if( param->param_free )
1398         param->param_free( param );
1399
1400 #if HAVE_INTEL_DISPATCHER
1401     x264_intel_dispatcher_override();
1402 #endif
1403
1404     if( x264_threading_init() )
1405     {
1406         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1407         goto fail;
1408     }
1409
1410     if( x264_validate_parameters( h, 1 ) < 0 )
1411         goto fail;
1412
1413     if( h->param.psz_cqm_file )
1414         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1415             goto fail;
1416
1417     if( h->param.rc.psz_stat_out )
1418         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1419     if( h->param.rc.psz_stat_in )
1420         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1421
1422     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1423     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1424
1425     /* Init x264_t */
1426     h->i_frame = -1;
1427     h->i_frame_num = 0;
1428
1429     if( h->param.i_avcintra_class )
1430         h->i_idr_pic_id = 5;
1431     else
1432         h->i_idr_pic_id = 0;
1433
1434     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1435     {
1436         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1437         goto fail;
1438     }
1439
1440     x264_set_aspect_ratio( h, &h->param, 1 );
1441
1442     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1443     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1444
1445     x264_validate_levels( h, 1 );
1446
1447     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1448
1449     if( x264_cqm_init( h ) < 0 )
1450         goto fail;
1451
1452     h->mb.i_mb_width = h->sps->i_mb_width;
1453     h->mb.i_mb_height = h->sps->i_mb_height;
1454     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1455
1456     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1457     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1458
1459     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1460      * vectors during prediction, resulting in hpel mvs.
1461      * The chosen solution is to make MBAFF non-adaptive in this case. */
1462     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1463
1464     /* Init frames. */
1465     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1466         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1467     else
1468         h->frames.i_delay = h->param.i_bframe;
1469     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1470         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1471     i_slicetype_length = h->frames.i_delay;
1472     h->frames.i_delay += h->i_thread_frames - 1;
1473     h->frames.i_delay += h->param.i_sync_lookahead;
1474     h->frames.i_delay += h->param.b_vfr_input;
1475     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1476
1477     h->frames.i_max_ref0 = h->param.i_frame_reference;
1478     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1479     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1480     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1481         && ( h->param.rc.i_rc_method == X264_RC_ABR
1482           || h->param.rc.i_rc_method == X264_RC_CRF
1483           || h->param.i_bframe_adaptive
1484           || h->param.i_scenecut_threshold
1485           || h->param.rc.b_mb_tree
1486           || h->param.analyse.i_weighted_pred );
1487     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1488     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1489
1490     h->frames.i_last_idr =
1491     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1492     h->frames.i_input    = 0;
1493     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1494     h->frames.i_poc_last_open_gop = -1;
1495
1496     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1497     /* Allocate room for max refs plus a few extra just in case. */
1498     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1499     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1500                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1501     if( h->param.analyse.i_weighted_pred > 0 )
1502         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1503     h->i_ref[0] = h->i_ref[1] = 0;
1504     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1505     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);
1506     h->i_disp_fields_last_frame = -1;
1507     x264_rdo_init();
1508
1509     /* init CPU functions */
1510     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1511     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1512     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1513     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1514     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1515     x264_pixel_init( h->param.cpu, &h->pixf );
1516     x264_dct_init( h->param.cpu, &h->dctf );
1517     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1518     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1519     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1520     x264_quant_init( h, h->param.cpu, &h->quantf );
1521     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1522     x264_bitstream_init( h->param.cpu, &h->bsf );
1523     if( h->param.b_cabac )
1524         x264_cabac_init( h );
1525     else
1526         x264_stack_align( x264_cavlc_init, h );
1527
1528     mbcmp_init( h );
1529     chroma_dsp_init( h );
1530
1531     p = buf + sprintf( buf, "using cpu capabilities:" );
1532     for( int i = 0; x264_cpu_names[i].flags; i++ )
1533     {
1534         if( !strcmp(x264_cpu_names[i].name, "SSE")
1535             && h->param.cpu & (X264_CPU_SSE2) )
1536             continue;
1537         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1538             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1539             continue;
1540         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1541             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1542             continue;
1543         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1544             && (h->param.cpu & X264_CPU_SSE42) )
1545             continue;
1546         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1547             && (h->param.cpu & X264_CPU_BMI2) )
1548             continue;
1549         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1550             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1551             p += sprintf( p, " %s", x264_cpu_names[i].name );
1552     }
1553     if( !h->param.cpu )
1554         p += sprintf( p, " none!" );
1555     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1556
1557     float *logs = x264_analyse_prepare_costs( h );
1558     if( !logs )
1559         goto fail;
1560     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1561         if( x264_analyse_init_costs( h, logs, qp ) )
1562             goto fail;
1563     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1564         goto fail;
1565     x264_free( logs );
1566
1567     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1568     /* Checks for known miscompilation issues. */
1569     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1570     {
1571         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1572         goto fail;
1573     }
1574
1575     /* Must be volatile or else GCC will optimize it out. */
1576     volatile int temp = 392;
1577     if( x264_clz( temp ) != 23 )
1578     {
1579         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1580 #if ARCH_X86 || ARCH_X86_64
1581         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1582         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1583 #endif
1584         goto fail;
1585     }
1586
1587     h->out.i_nal = 0;
1588     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1589         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1590           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1591
1592     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1593     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1594
1595     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1596
1597     if( h->param.i_threads > 1 &&
1598         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1599         goto fail;
1600     if( h->param.i_lookahead_threads > 1 &&
1601         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1602         goto fail;
1603
1604 #if HAVE_OPENCL
1605     if( h->param.b_opencl )
1606     {
1607         h->opencl.ocl = x264_opencl_load_library();
1608         if( !h->opencl.ocl )
1609         {
1610             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1611             h->param.b_opencl = 0;
1612         }
1613     }
1614 #endif
1615
1616     h->thread[0] = h;
1617     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1618         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1619     if( h->param.i_lookahead_threads > 1 )
1620         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1621         {
1622             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1623             *h->lookahead_thread[i] = *h;
1624         }
1625     *h->reconfig_h = *h;
1626
1627     for( int i = 0; i < h->param.i_threads; i++ )
1628     {
1629         int init_nal_count = h->param.i_slice_count + 3;
1630         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1631         if( i > 0 )
1632             *h->thread[i] = *h;
1633
1634         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1635             goto fail;
1636         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1637             goto fail;
1638
1639         if( allocate_threadlocal_data )
1640         {
1641             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1642             if( !h->thread[i]->fdec )
1643                 goto fail;
1644         }
1645         else
1646             h->thread[i]->fdec = h->thread[0]->fdec;
1647
1648         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1649         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1650         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1651         h->thread[i]->out.i_nals_allocated = init_nal_count;
1652
1653         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1654             goto fail;
1655     }
1656
1657 #if HAVE_OPENCL
1658     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1659         h->param.b_opencl = 0;
1660 #endif
1661
1662     if( x264_lookahead_init( h, i_slicetype_length ) )
1663         goto fail;
1664
1665     for( int i = 0; i < h->param.i_threads; i++ )
1666         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1667             goto fail;
1668
1669     if( x264_ratecontrol_new( h ) < 0 )
1670         goto fail;
1671
1672     if( h->param.i_nal_hrd )
1673     {
1674         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1675         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1676     }
1677
1678     if( h->param.psz_dump_yuv )
1679     {
1680         /* create or truncate the reconstructed video file */
1681         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1682         if( !f )
1683         {
1684             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1685             goto fail;
1686         }
1687         else if( !x264_is_regular_file( f ) )
1688         {
1689             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1690             goto fail;
1691         }
1692         fclose( f );
1693     }
1694
1695     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1696                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1697                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1698                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1699                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1700                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1701     char level[4];
1702     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1703     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1704         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1705         strcpy( level, "1b" );
1706
1707     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1708     {
1709         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1710             profile, level );
1711     }
1712     else
1713     {
1714         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1715         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1716             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1717     }
1718
1719     return h;
1720 fail:
1721     x264_free( h );
1722     return NULL;
1723 }
1724
1725 /****************************************************************************/
1726 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1727 {
1728     *rc_reconfig = 0;
1729     x264_set_aspect_ratio( h, param, 0 );
1730 #define COPY(var) h->param.var = param->var
1731     COPY( i_frame_reference ); // but never uses more refs than initially specified
1732     COPY( i_bframe_bias );
1733     if( h->param.i_scenecut_threshold )
1734         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1735     COPY( b_deblocking_filter );
1736     COPY( i_deblocking_filter_alphac0 );
1737     COPY( i_deblocking_filter_beta );
1738     COPY( i_frame_packing );
1739     COPY( analyse.inter );
1740     COPY( analyse.intra );
1741     COPY( analyse.i_direct_mv_pred );
1742     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1743     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1744         COPY( analyse.i_me_range );
1745     COPY( analyse.i_noise_reduction );
1746     /* We can't switch out of subme=0 during encoding. */
1747     if( h->param.analyse.i_subpel_refine )
1748         COPY( analyse.i_subpel_refine );
1749     COPY( analyse.i_trellis );
1750     COPY( analyse.b_chroma_me );
1751     COPY( analyse.b_dct_decimate );
1752     COPY( analyse.b_fast_pskip );
1753     COPY( analyse.b_mixed_references );
1754     COPY( analyse.f_psy_rd );
1755     COPY( analyse.f_psy_trellis );
1756     COPY( crop_rect );
1757     // can only twiddle these if they were enabled to begin with:
1758     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1759         COPY( analyse.i_me_method );
1760     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1761         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1762     if( h->pps->b_transform_8x8_mode )
1763         COPY( analyse.b_transform_8x8 );
1764     if( h->frames.i_max_ref1 > 1 )
1765         COPY( i_bframe_pyramid );
1766     COPY( i_slice_max_size );
1767     COPY( i_slice_max_mbs );
1768     COPY( i_slice_min_mbs );
1769     COPY( i_slice_count );
1770     COPY( i_slice_count_max );
1771     COPY( b_tff );
1772
1773     /* VBV can't be turned on if it wasn't on to begin with */
1774     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1775           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1776     {
1777         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1778         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1779         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1780         COPY( rc.i_vbv_max_bitrate );
1781         COPY( rc.i_vbv_buffer_size );
1782         COPY( rc.i_bitrate );
1783     }
1784     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1785     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1786     COPY( rc.f_rf_constant );
1787     COPY( rc.f_rf_constant_max );
1788 #undef COPY
1789
1790     return x264_validate_parameters( h, 0 );
1791 }
1792
1793 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1794 {
1795     int rc_reconfig;
1796     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1797
1798     mbcmp_init( h );
1799     if( !ret )
1800         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1801
1802     /* Supported reconfiguration options (1-pass only):
1803      * vbv-maxrate
1804      * vbv-bufsize
1805      * crf
1806      * bitrate (CBR only) */
1807     if( !ret && rc_reconfig )
1808         x264_ratecontrol_init_reconfigurable( h, 0 );
1809
1810     return ret;
1811 }
1812
1813 /****************************************************************************
1814  * x264_encoder_reconfig:
1815  ****************************************************************************/
1816 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1817 {
1818     h = h->thread[h->thread[0]->i_thread_phase];
1819     x264_param_t param_save = h->reconfig_h->param;
1820     h->reconfig_h->param = h->param;
1821
1822     int rc_reconfig;
1823     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1824     if( !ret )
1825         h->reconfig = 1;
1826     else
1827         h->reconfig_h->param = param_save;
1828
1829     return ret;
1830 }
1831
1832 /****************************************************************************
1833  * x264_encoder_parameters:
1834  ****************************************************************************/
1835 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1836 {
1837     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1838 }
1839
1840 /* internal usage */
1841 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1842 {
1843     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1844
1845     nal->i_ref_idc        = i_ref_idc;
1846     nal->i_type           = i_type;
1847     nal->b_long_startcode = 1;
1848
1849     nal->i_payload= 0;
1850     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1851     nal->i_padding= 0;
1852 }
1853
1854 /* if number of allocated nals is not enough, re-allocate a larger one. */
1855 static int x264_nal_check_buffer( x264_t *h )
1856 {
1857     if( h->out.i_nal >= h->out.i_nals_allocated )
1858     {
1859         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1860         if( !new_out )
1861             return -1;
1862         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1863         x264_free( h->out.nal );
1864         h->out.nal = new_out;
1865         h->out.i_nals_allocated *= 2;
1866     }
1867     return 0;
1868 }
1869
1870 static int x264_nal_end( x264_t *h )
1871 {
1872     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1873     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1874     nal->i_payload = end - nal->p_payload;
1875     /* Assembly implementation of nal_escape reads past the end of the input.
1876      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1877     memset( end, 0xff, 64 );
1878     if( h->param.nalu_process )
1879         h->param.nalu_process( h, nal, h->fenc->opaque );
1880     h->out.i_nal++;
1881
1882     return x264_nal_check_buffer( h );
1883 }
1884
1885 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1886                                            int previous_nal_size, int necessary_size )
1887 {
1888     if( h0->nal_buffer_size < necessary_size )
1889     {
1890         necessary_size *= 2;
1891         uint8_t *buf = x264_malloc( necessary_size );
1892         if( !buf )
1893             return -1;
1894         if( previous_nal_size )
1895             memcpy( buf, h0->nal_buffer, previous_nal_size );
1896
1897         intptr_t delta = buf - h0->nal_buffer;
1898         for( int i = 0; i < start; i++ )
1899             h->out.nal[i].p_payload += delta;
1900
1901         x264_free( h0->nal_buffer );
1902         h0->nal_buffer = buf;
1903         h0->nal_buffer_size = necessary_size;
1904     }
1905
1906     return 0;
1907 }
1908
1909 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1910 {
1911     x264_t *h0 = h->thread[0];
1912     int nal_size = 0, previous_nal_size = 0;
1913
1914     if( h->param.nalu_process )
1915     {
1916         for( int i = start; i < h->out.i_nal; i++ )
1917             nal_size += h->out.nal[i].i_payload;
1918         return nal_size;
1919     }
1920
1921     for( int i = 0; i < start; i++ )
1922         previous_nal_size += h->out.nal[i].i_payload;
1923
1924     for( int i = start; i < h->out.i_nal; i++ )
1925         nal_size += h->out.nal[i].i_payload;
1926
1927     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1928     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1929     for( int i = start; i < h->out.i_nal; i++ )
1930         necessary_size += h->out.nal[i].i_padding;
1931     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1932         return -1;
1933
1934     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1935
1936     for( int i = start; i < h->out.i_nal; i++ )
1937     {
1938         int old_payload_len = h->out.nal[i].i_payload;
1939         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1940                                          h->param.i_avcintra_class;
1941         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1942         nal_buffer += h->out.nal[i].i_payload;
1943         if( h->param.i_avcintra_class )
1944         {
1945             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1946             if( h->out.nal[i].i_padding > 0 )
1947             {
1948                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1949                 nal_buffer += h->out.nal[i].i_padding;
1950                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1951             }
1952             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1953         }
1954     }
1955
1956     x264_emms();
1957
1958     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1959 }
1960
1961 /****************************************************************************
1962  * x264_encoder_headers:
1963  ****************************************************************************/
1964 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1965 {
1966     int frame_size = 0;
1967     /* init bitstream context */
1968     h->out.i_nal = 0;
1969     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1970
1971     /* Write SEI, SPS and PPS. */
1972
1973     /* generate sequence parameters */
1974     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1975     x264_sps_write( &h->out.bs, h->sps );
1976     if( x264_nal_end( h ) )
1977         return -1;
1978
1979     /* generate picture parameters */
1980     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1981     x264_pps_write( &h->out.bs, h->sps, h->pps );
1982     if( x264_nal_end( h ) )
1983         return -1;
1984
1985     /* identify ourselves */
1986     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1987     if( x264_sei_version_write( h, &h->out.bs ) )
1988         return -1;
1989     if( x264_nal_end( h ) )
1990         return -1;
1991
1992     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1993     if( frame_size < 0 )
1994         return -1;
1995
1996     /* now set output*/
1997     *pi_nal = h->out.i_nal;
1998     *pp_nal = &h->out.nal[0];
1999     h->out.i_nal = 0;
2000
2001     return frame_size;
2002 }
2003
2004 /* Check to see whether we have chosen a reference list ordering different
2005  * from the standard's default. */
2006 static inline void x264_reference_check_reorder( x264_t *h )
2007 {
2008     /* The reorder check doesn't check for missing frames, so just
2009      * force a reorder if one of the reference list is corrupt. */
2010     for( int i = 0; h->frames.reference[i]; i++ )
2011         if( h->frames.reference[i]->b_corrupt )
2012         {
2013             h->b_ref_reorder[0] = 1;
2014             return;
2015         }
2016     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
2017         for( int i = 0; i < h->i_ref[list] - 1; i++ )
2018         {
2019             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
2020             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
2021             /* P and B-frames use different default orders. */
2022             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
2023             {
2024                 h->b_ref_reorder[list] = 1;
2025                 return;
2026             }
2027         }
2028 }
2029
2030 /* return -1 on failure, else return the index of the new reference frame */
2031 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2032 {
2033     int i = h->i_ref[0];
2034     int j = 1;
2035     x264_frame_t *newframe;
2036     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2037         return -1;
2038
2039     //Duplication is only used in X264_WEIGHTP_SMART
2040     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2041         return -1;
2042
2043     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2044      * With high bit depth, it's not worth doing, so turn it off except in the case of
2045      * unweighted dupes. */
2046     if( BIT_DEPTH > 8 && w != x264_weight_none )
2047         return -1;
2048
2049     newframe = x264_frame_pop_blank_unused( h );
2050     if( !newframe )
2051         return -1;
2052
2053     //FIXME: probably don't need to copy everything
2054     *newframe = *h->fref[0][i_ref];
2055     newframe->i_reference_count = 1;
2056     newframe->orig = h->fref[0][i_ref];
2057     newframe->b_duplicate = 1;
2058     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2059
2060     /* shift the frames to make space for the dupe. */
2061     h->b_ref_reorder[0] = 1;
2062     if( h->i_ref[0] < X264_REF_MAX )
2063         ++h->i_ref[0];
2064     h->fref[0][X264_REF_MAX-1] = NULL;
2065     x264_frame_unshift( &h->fref[0][j], newframe );
2066
2067     return j;
2068 }
2069
2070 static void x264_weighted_pred_init( x264_t *h )
2071 {
2072     /* for now no analysis and set all weights to nothing */
2073     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2074         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2075
2076     // FIXME: This only supports weighting of one reference frame
2077     // and duplicates of that frame.
2078     h->fenc->i_lines_weighted = 0;
2079
2080     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2081         for( int i = 0; i < 3; i++ )
2082             h->sh.weight[i_ref][i].weightfn = NULL;
2083
2084
2085     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2086         return;
2087
2088     int i_padv = PADV << PARAM_INTERLACED;
2089     int denom = -1;
2090     int weightplane[2] = { 0, 0 };
2091     int buffer_next = 0;
2092     for( int i = 0; i < 3; i++ )
2093     {
2094         for( int j = 0; j < h->i_ref[0]; j++ )
2095         {
2096             if( h->fenc->weight[j][i].weightfn )
2097             {
2098                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2099                 // if weight is useless, don't write it to stream
2100                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2101                     h->sh.weight[j][i].weightfn = NULL;
2102                 else
2103                 {
2104                     if( !weightplane[!!i] )
2105                     {
2106                         weightplane[!!i] = 1;
2107                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2108                         assert( x264_clip3( denom, 0, 7 ) == denom );
2109                     }
2110
2111                     assert( h->sh.weight[j][i].i_denom == denom );
2112                     if( !i )
2113                     {
2114                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2115                         //scale full resolution frame
2116                         if( h->param.i_threads == 1 )
2117                         {
2118                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2119                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2120                             int stride = h->fenc->i_stride[0];
2121                             int width = h->fenc->i_width[0] + PADH*2;
2122                             int height = h->fenc->i_lines[0] + i_padv*2;
2123                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2124                             h->fenc->i_lines_weighted = height;
2125                         }
2126                     }
2127                 }
2128             }
2129         }
2130     }
2131
2132     if( weightplane[1] )
2133         for( int i = 0; i < h->i_ref[0]; i++ )
2134         {
2135             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2136             {
2137                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2138                 h->sh.weight[i][2].i_offset = 0;
2139             }
2140             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2141             {
2142                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2143                 h->sh.weight[i][1].i_offset = 0;
2144             }
2145         }
2146
2147     if( !weightplane[0] )
2148         h->sh.weight[0][0].i_denom = 0;
2149     if( !weightplane[1] )
2150         h->sh.weight[0][1].i_denom = 0;
2151     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2152 }
2153
2154 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2155 {
2156     if( h->param.i_frame_packing == 5 )
2157         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2158                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2159     else
2160         return abs(h->fenc->i_frame - frame->i_frame);
2161 }
2162
2163 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2164 {
2165     int b_ok;
2166
2167     /* build ref list 0/1 */
2168     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2169     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2170     if( h->sh.i_type == SLICE_TYPE_I )
2171         return;
2172
2173     for( int i = 0; h->frames.reference[i]; i++ )
2174     {
2175         if( h->frames.reference[i]->b_corrupt )
2176             continue;
2177         if( h->frames.reference[i]->i_poc < i_poc )
2178             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2179         else if( h->frames.reference[i]->i_poc > i_poc )
2180             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2181     }
2182
2183     if( h->sh.i_mmco_remove_from_end )
2184     {
2185         /* Order ref0 for MMCO remove */
2186         do
2187         {
2188             b_ok = 1;
2189             for( int i = 0; i < h->i_ref[0] - 1; i++ )
2190             {
2191                 if( h->fref[0][i]->i_frame < h->fref[0][i+1]->i_frame )
2192                 {
2193                     XCHG( x264_frame_t*, h->fref[0][i], h->fref[0][i+1] );
2194                     b_ok = 0;
2195                     break;
2196                 }
2197             }
2198         } while( !b_ok );
2199
2200         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2201         {
2202             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2203             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2204             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2205         }
2206     }
2207
2208     /* Order reference lists by distance from the current frame. */
2209     for( int list = 0; list < 2; list++ )
2210     {
2211         h->fref_nearest[list] = h->fref[list][0];
2212         do
2213         {
2214             b_ok = 1;
2215             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2216             {
2217                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2218                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2219                     h->fref_nearest[list] = h->fref[list][i+1];
2220                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2221                 {
2222                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2223                     b_ok = 0;
2224                     break;
2225                 }
2226             }
2227         } while( !b_ok );
2228     }
2229
2230     x264_reference_check_reorder( h );
2231
2232     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2233     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2234     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2235
2236     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2237     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2238         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2239
2240     /* add duplicates */
2241     if( h->fenc->i_type == X264_TYPE_P )
2242     {
2243         int idx = -1;
2244         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2245         {
2246             x264_weight_t w[3];
2247             w[1].weightfn = w[2].weightfn = NULL;
2248             if( h->param.rc.b_stat_read )
2249                 x264_ratecontrol_set_weights( h, h->fenc );
2250
2251             if( !h->fenc->weight[0][0].weightfn )
2252             {
2253                 h->fenc->weight[0][0].i_denom = 0;
2254                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2255                 idx = x264_weighted_reference_duplicate( h, 0, w );
2256             }
2257             else
2258             {
2259                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2260                 {
2261                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2262                 }
2263                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2264                 if( h->fenc->weight[0][0].i_offset > -128 )
2265                 {
2266                     w[0] = h->fenc->weight[0][0];
2267                     w[0].i_offset--;
2268                     h->mc.weight_cache( h, &w[0] );
2269                     idx = x264_weighted_reference_duplicate( h, 0, w );
2270                 }
2271             }
2272         }
2273         h->mb.ref_blind_dupe = idx;
2274     }
2275
2276     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2277     h->mb.pic.i_fref[0] = h->i_ref[0];
2278     h->mb.pic.i_fref[1] = h->i_ref[1];
2279 }
2280
2281 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2282 {
2283     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2284     int b_hpel = h->fdec->b_kept_as_ref;
2285     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2286     int b_end = mb_y == h->i_threadslice_end;
2287     int b_measure_quality = 1;
2288     int min_y = mb_y - (1 << SLICE_MBAFF);
2289     int b_start = min_y == h->i_threadslice_start;
2290     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2291      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2292     int minpix_y = min_y*16 - 4 * !b_start;
2293     int maxpix_y = mb_y*16 - 4 * !b_end;
2294     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2295     if( h->param.b_sliced_threads )
2296     {
2297         switch( pass )
2298         {
2299             /* During encode: only do deblock if asked for */
2300             default:
2301             case 0:
2302                 b_deblock &= h->param.b_full_recon;
2303                 b_hpel = 0;
2304                 break;
2305             /* During post-encode pass: do deblock if not done yet, do hpel for all
2306              * rows except those between slices. */
2307             case 1:
2308                 b_deblock &= !h->param.b_full_recon;
2309                 b_hpel &= !(b_start && min_y > 0);
2310                 b_measure_quality = 0;
2311                 break;
2312             /* Final pass: do the rows between slices in sequence. */
2313             case 2:
2314                 b_deblock = 0;
2315                 b_measure_quality = 0;
2316                 break;
2317         }
2318     }
2319     if( mb_y & SLICE_MBAFF )
2320         return;
2321     if( min_y < h->i_threadslice_start )
2322         return;
2323
2324     if( b_deblock )
2325         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2326             x264_frame_deblock_row( h, y );
2327
2328     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2329      * but the actual image data is equivalent. For now, maintain this
2330      * consistency by copying deblocked pixels between planes. */
2331     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2332         for( int p = 0; p < h->fdec->i_plane; p++ )
2333             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2334                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2335                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2336                         h->mb.i_mb_width*16*sizeof(pixel) );
2337
2338     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2339         x264_frame_expand_border( h, h->fdec, min_y );
2340     if( b_hpel )
2341     {
2342         int end = mb_y == h->mb.i_mb_height;
2343         /* Can't do hpel until the previous slice is done encoding. */
2344         if( h->param.analyse.i_subpel_refine )
2345         {
2346             x264_frame_filter( h, h->fdec, min_y, end );
2347             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2348         }
2349     }
2350
2351     if( SLICE_MBAFF && pass == 0 )
2352         for( int i = 0; i < 3; i++ )
2353         {
2354             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2355             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2356         }
2357
2358     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2359         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2360
2361     if( b_measure_quality )
2362     {
2363         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2364         if( h->param.analyse.b_psnr )
2365         {
2366             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2367                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2368                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2369                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2370                     h->param.i_width, maxpix_y-minpix_y );
2371             if( !CHROMA444 )
2372             {
2373                 uint64_t ssd_u, ssd_v;
2374                 int v_shift = CHROMA_V_SHIFT;
2375                 x264_pixel_ssd_nv12( &h->pixf,
2376                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2377                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2378                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2379                 h->stat.frame.i_ssd[1] += ssd_u;
2380                 h->stat.frame.i_ssd[2] += ssd_v;
2381             }
2382         }
2383
2384         if( h->param.analyse.b_ssim )
2385         {
2386             int ssim_cnt;
2387             x264_emms();
2388             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2389              * and overlap by 4 */
2390             minpix_y += b_start ? 2 : -6;
2391             h->stat.frame.f_ssim +=
2392                 x264_pixel_ssim_wxh( &h->pixf,
2393                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2394                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2395                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2396             h->stat.frame.i_ssim_cnt += ssim_cnt;
2397         }
2398     }
2399 }
2400
2401 static inline int x264_reference_update( x264_t *h )
2402 {
2403     if( !h->fdec->b_kept_as_ref )
2404     {
2405         if( h->i_thread_frames > 1 )
2406         {
2407             x264_frame_push_unused( h, h->fdec );
2408             h->fdec = x264_frame_pop_unused( h, 1 );
2409             if( !h->fdec )
2410                 return -1;
2411         }
2412         return 0;
2413     }
2414
2415     /* apply mmco from previous frame. */
2416     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2417         for( int j = 0; h->frames.reference[j]; j++ )
2418             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2419                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2420
2421     /* move frame in the buffer */
2422     x264_frame_push( h->frames.reference, h->fdec );
2423     if( h->frames.reference[h->sps->i_num_ref_frames] )
2424         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2425     h->fdec = x264_frame_pop_unused( h, 1 );
2426     if( !h->fdec )
2427         return -1;
2428     return 0;
2429 }
2430
2431 static inline void x264_reference_reset( x264_t *h )
2432 {
2433     while( h->frames.reference[0] )
2434         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2435     h->fdec->i_poc =
2436     h->fenc->i_poc = 0;
2437 }
2438
2439 static inline void x264_reference_hierarchy_reset( x264_t *h )
2440 {
2441     int ref;
2442     int b_hasdelayframe = 0;
2443
2444     /* look for delay frames -- chain must only contain frames that are disposable */
2445     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2446         b_hasdelayframe |= h->frames.current[i]->i_coded
2447                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2448
2449     /* This function must handle b-pyramid and clear frames for open-gop */
2450     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2451         return;
2452
2453     /* Remove last BREF. There will never be old BREFs in the
2454      * dpb during a BREF decode when pyramid == STRICT */
2455     for( ref = 0; h->frames.reference[ref]; ref++ )
2456     {
2457         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2458             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2459             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2460             && h->sh.i_type != SLICE_TYPE_B ) )
2461         {
2462             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2463             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2464             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2465             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2466             h->b_ref_reorder[0] = 1;
2467             ref--;
2468         }
2469     }
2470
2471     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2472     if( h->param.i_bframe_pyramid )
2473         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2474 }
2475
2476 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2477 {
2478     /* ------------------------ Create slice header  ----------------------- */
2479     if( i_nal_type == NAL_SLICE_IDR )
2480     {
2481         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2482
2483         /* alternate id */
2484         if( h->param.i_avcintra_class )
2485         {
2486             switch( h->i_idr_pic_id )
2487             {
2488                 case 5:
2489                     h->i_idr_pic_id = 3;
2490                     break;
2491                 case 3:
2492                     h->i_idr_pic_id = 4;
2493                     break;
2494                 case 4:
2495                 default:
2496                     h->i_idr_pic_id = 5;
2497                     break;
2498             }
2499         }
2500         else
2501             h->i_idr_pic_id ^= 1;
2502     }
2503     else
2504     {
2505         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2506
2507         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2508         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2509         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2510             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2511         {
2512             h->sh.b_num_ref_idx_override = 1;
2513         }
2514     }
2515
2516     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2517     {
2518         h->b_sh_backup = 1;
2519         h->sh_backup = h->sh;
2520     }
2521
2522     h->fdec->i_frame_num = h->sh.i_frame_num;
2523
2524     if( h->sps->i_poc_type == 0 )
2525     {
2526         h->sh.i_poc = h->fdec->i_poc;
2527         if( PARAM_INTERLACED )
2528         {
2529             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2530             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2531         }
2532         else
2533             h->sh.i_delta_poc_bottom = 0;
2534         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2535         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2536     }
2537     else
2538     {
2539         /* Nothing to do ? */
2540     }
2541
2542     x264_macroblock_slice_init( h );
2543 }
2544
2545 typedef struct
2546 {
2547     int skip;
2548     uint8_t cabac_prevbyte;
2549     bs_t bs;
2550     x264_cabac_t cabac;
2551     x264_frame_stat_t stat;
2552     int last_qp;
2553     int last_dqp;
2554     int field_decoding_flag;
2555 } x264_bs_bak_t;
2556
2557 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2558 {
2559     if( full )
2560     {
2561         bak->stat = h->stat.frame;
2562         bak->last_qp = h->mb.i_last_qp;
2563         bak->last_dqp = h->mb.i_last_dqp;
2564         bak->field_decoding_flag = h->mb.field_decoding_flag;
2565     }
2566     else
2567     {
2568         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2569         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2570     }
2571     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2572      * encoder has no context dependency and in this case, a slice is ended (and
2573      * thus the content of all contexts are thrown away). */
2574     if( h->param.b_cabac )
2575     {
2576         if( full )
2577             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2578         else
2579             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2580         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2581          * backed up. */
2582         bak->cabac_prevbyte = h->cabac.p[-1];
2583     }
2584     else
2585     {
2586         bak->bs = h->out.bs;
2587         bak->skip = i_skip;
2588     }
2589 }
2590
2591 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2592 {
2593     if( full )
2594     {
2595         h->stat.frame = bak->stat;
2596         h->mb.i_last_qp = bak->last_qp;
2597         h->mb.i_last_dqp = bak->last_dqp;
2598         h->mb.field_decoding_flag = bak->field_decoding_flag;
2599     }
2600     else
2601     {
2602         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2603         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2604     }
2605     if( h->param.b_cabac )
2606     {
2607         if( full )
2608             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2609         else
2610             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2611         h->cabac.p[-1] = bak->cabac_prevbyte;
2612     }
2613     else
2614     {
2615         h->out.bs = bak->bs;
2616         *skip = bak->skip;
2617     }
2618 }
2619
2620 static intptr_t x264_slice_write( x264_t *h )
2621 {
2622     int i_skip;
2623     int mb_xy, i_mb_x, i_mb_y;
2624     /* NALUs other than the first use a 3-byte startcode.
2625      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2626      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2627      * other inaccuracies. */
2628     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2629     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2630     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2631     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2632     int starting_bits = bs_pos(&h->out.bs);
2633     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2634     int b_hpel = h->fdec->b_kept_as_ref;
2635     int orig_last_mb = h->sh.i_last_mb;
2636     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2637     uint8_t *last_emu_check;
2638 #define BS_BAK_SLICE_MAX_SIZE 0
2639 #define BS_BAK_CAVLC_OVERFLOW 1
2640 #define BS_BAK_SLICE_MIN_MBS  2
2641 #define BS_BAK_ROW_VBV        3
2642     x264_bs_bak_t bs_bak[4];
2643     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2644     bs_realign( &h->out.bs );
2645
2646     /* Slice */
2647     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2648     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2649
2650     /* Slice header */
2651     x264_macroblock_thread_init( h );
2652
2653     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2654     h->mb.i_mb_xy = h->sh.i_first_mb;
2655     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2656     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2657     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2658
2659     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2660     if( h->param.b_cabac )
2661     {
2662         /* alignment needed */
2663         bs_align_1( &h->out.bs );
2664
2665         /* init cabac */
2666         x264_cabac_context_init( h, &h->cabac, h->sh.i_type, x264_clip3( h->sh.i_qp-QP_BD_OFFSET, 0, 51 ), h->sh.i_cabac_init_idc );
2667         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2668         last_emu_check = h->cabac.p;
2669     }
2670     else
2671         last_emu_check = h->out.bs.p;
2672     h->mb.i_last_qp = h->sh.i_qp;
2673     h->mb.i_last_dqp = 0;
2674     h->mb.field_decoding_flag = 0;
2675
2676     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2677     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2678     i_skip = 0;
2679
2680     while( 1 )
2681     {
2682         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2683         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2684
2685         if( i_mb_x == 0 )
2686         {
2687             if( x264_bitstream_check_buffer( h ) )
2688                 return -1;
2689             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2690                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2691             if( !h->mb.b_reencode_mb )
2692                 x264_fdec_filter_row( h, i_mb_y, 0 );
2693         }
2694
2695         if( back_up_bitstream )
2696         {
2697             if( back_up_bitstream_cavlc )
2698                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2699             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2700             {
2701                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2702                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2703                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2704             }
2705         }
2706
2707         if( PARAM_INTERLACED )
2708         {
2709             if( h->mb.b_adaptive_mbaff )
2710             {
2711                 if( !(i_mb_y&1) )
2712                 {
2713                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2714                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2715                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2716                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2717                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2718                 }
2719             }
2720             h->mb.field[mb_xy] = MB_INTERLACED;
2721         }
2722
2723         /* load cache */
2724         if( SLICE_MBAFF )
2725             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2726         else
2727             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2728
2729         x264_macroblock_analyse( h );
2730
2731         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2732 reencode:
2733         x264_macroblock_encode( h );
2734
2735         if( h->param.b_cabac )
2736         {
2737             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2738                 x264_cabac_encode_terminal( &h->cabac );
2739
2740             if( IS_SKIP( h->mb.i_type ) )
2741                 x264_cabac_mb_skip( h, 1 );
2742             else
2743             {
2744                 if( h->sh.i_type != SLICE_TYPE_I )
2745                     x264_cabac_mb_skip( h, 0 );
2746                 x264_macroblock_write_cabac( h, &h->cabac );
2747             }
2748         }
2749         else
2750         {
2751             if( IS_SKIP( h->mb.i_type ) )
2752                 i_skip++;
2753             else
2754             {
2755                 if( h->sh.i_type != SLICE_TYPE_I )
2756                 {
2757                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2758                     i_skip = 0;
2759                 }
2760                 x264_macroblock_write_cavlc( h );
2761                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2762                 if( h->mb.b_overflow )
2763                 {
2764                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2765                     h->mb.i_skip_intra = 0;
2766                     h->mb.b_skip_mc = 0;
2767                     h->mb.b_overflow = 0;
2768                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2769                     goto reencode;
2770                 }
2771             }
2772         }
2773
2774         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2775         int mb_size = total_bits - mb_spos;
2776
2777         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2778         {
2779             /* Count the skip run, just in case. */
2780             if( !h->param.b_cabac )
2781                 total_bits += bs_size_ue_big( i_skip );
2782             /* Check for escape bytes. */
2783             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2784             for( ; last_emu_check < end - 2; last_emu_check++ )
2785                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2786                 {
2787                     slice_max_size -= 8;
2788                     last_emu_check++;
2789                 }
2790             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2791             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2792             {
2793                 if( !x264_frame_new_slice( h, h->fdec ) )
2794                 {
2795                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2796                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2797                      * If possible, roll back to the last checkpoint and try again.
2798                      * We could try raising QP, but that would break in the case where a slice spans multiple
2799                      * rows, which the re-encoding infrastructure can't currently handle. */
2800                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2801                     {
2802                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2803                         {
2804                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2805                             slice_max_size = 0;
2806                             goto cont;
2807                         }
2808                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2809                         h->mb.b_reencode_mb = 1;
2810                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2811                         break;
2812                     }
2813                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2814                     {
2815                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2816                         h->mb.b_reencode_mb = 1;
2817                         if( SLICE_MBAFF )
2818                         {
2819                             // set to bottom of previous mbpair
2820                             if( i_mb_x )
2821                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2822                             else
2823                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2824                         }
2825                         else
2826                             h->sh.i_last_mb = mb_xy-1;
2827                         break;
2828                     }
2829                     else
2830                         h->sh.i_last_mb = mb_xy;
2831                 }
2832                 else
2833                     slice_max_size = 0;
2834             }
2835         }
2836 cont:
2837         h->mb.b_reencode_mb = 0;
2838
2839         /* save cache */
2840         x264_macroblock_cache_save( h );
2841
2842         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2843         {
2844             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2845             h->mb.b_reencode_mb = 1;
2846             i_mb_x = 0;
2847             i_mb_y = i_mb_y - SLICE_MBAFF;
2848             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2849             h->sh.i_last_mb = orig_last_mb;
2850             continue;
2851         }
2852
2853         /* accumulate mb stats */
2854         h->stat.frame.i_mb_count[h->mb.i_type]++;
2855
2856         int b_intra = IS_INTRA( h->mb.i_type );
2857         int b_skip = IS_SKIP( h->mb.i_type );
2858         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2859         {
2860             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2861             {
2862                 if( h->mb.i_partition != D_8x8 )
2863                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2864                     else
2865                         for( int i = 0; i < 4; i++ )
2866                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2867                 if( h->param.i_frame_reference > 1 )
2868                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2869                         for( int i = 0; i < 4; i++ )
2870                         {
2871                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2872                             if( i_ref >= 0 )
2873                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2874                         }
2875             }
2876         }
2877
2878         if( h->param.i_log_level >= X264_LOG_INFO )
2879         {
2880             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2881             {
2882                 if( CHROMA444 )
2883                 {
2884                     for( int i = 0; i < 4; i++ )
2885                         if( h->mb.i_cbp_luma & (1 << i) )
2886                             for( int p = 0; p < 3; p++ )
2887                             {
2888                                 int s8 = i*4+p*16;
2889                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2890                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2891                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2892                             }
2893                 }
2894                 else
2895                 {
2896                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2897                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2898                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2899                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2900                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2901                 }
2902             }
2903             if( h->mb.i_cbp_luma && !b_intra )
2904             {
2905                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2906                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2907             }
2908             if( b_intra && h->mb.i_type != I_PCM )
2909             {
2910                 if( h->mb.i_type == I_16x16 )
2911                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2912                 else if( h->mb.i_type == I_8x8 )
2913                     for( int i = 0; i < 16; i += 4 )
2914                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2915                 else //if( h->mb.i_type == I_4x4 )
2916                     for( int i = 0; i < 16; i++ )
2917                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2918                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2919             }
2920             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2921         }
2922
2923         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2924         if( b_deblock )
2925             x264_macroblock_deblock_strength( h );
2926
2927         if( mb_xy == h->sh.i_last_mb )
2928             break;
2929
2930         if( SLICE_MBAFF )
2931         {
2932             i_mb_x += i_mb_y & 1;
2933             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2934         }
2935         else
2936             i_mb_x++;
2937         if( i_mb_x == h->mb.i_mb_width )
2938         {
2939             i_mb_y++;
2940             i_mb_x = 0;
2941         }
2942     }
2943     if( h->sh.i_last_mb < h->sh.i_first_mb )
2944         return 0;
2945
2946     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2947
2948     if( h->param.b_cabac )
2949     {
2950         x264_cabac_encode_flush( h, &h->cabac );
2951         h->out.bs.p = h->cabac.p;
2952     }
2953     else
2954     {
2955         if( i_skip > 0 )
2956             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2957         /* rbsp_slice_trailing_bits */
2958         bs_rbsp_trailing( &h->out.bs );
2959         bs_flush( &h->out.bs );
2960     }
2961     if( x264_nal_end( h ) )
2962         return -1;
2963
2964     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2965     {
2966         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2967                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2968                                   - h->stat.frame.i_tex_bits
2969                                   - h->stat.frame.i_mv_bits;
2970         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2971
2972         if( h->param.b_sliced_threads )
2973         {
2974             /* Tell the main thread we're done. */
2975             x264_threadslice_cond_broadcast( h, 1 );
2976             /* Do hpel now */
2977             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2978                 x264_fdec_filter_row( h, mb_y, 1 );
2979             x264_threadslice_cond_broadcast( h, 2 );
2980             /* Do the first row of hpel, now that the previous slice is done */
2981             if( h->i_thread_idx > 0 )
2982             {
2983                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2984                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2985             }
2986         }
2987
2988         /* Free mb info after the last thread's done using it */
2989         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2990         {
2991             h->fdec->mb_info_free( h->fdec->mb_info );
2992             h->fdec->mb_info = NULL;
2993             h->fdec->mb_info_free = NULL;
2994         }
2995     }
2996
2997     return 0;
2998 }
2999
3000 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
3001 {
3002     if( dst == src )
3003         return;
3004
3005     // reference counting
3006     for( x264_frame_t **f = src->frames.reference; *f; f++ )
3007         (*f)->i_reference_count++;
3008     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
3009         x264_frame_push_unused( src, *f );
3010     src->fdec->i_reference_count++;
3011     x264_frame_push_unused( src, dst->fdec );
3012
3013     // copy everything except the per-thread pointers and the constants.
3014     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
3015     dst->param = src->param;
3016     dst->stat = src->stat;
3017     dst->pixf = src->pixf;
3018     dst->reconfig = src->reconfig;
3019 }
3020
3021 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
3022 {
3023     if( dst == src )
3024         return;
3025     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
3026 }
3027
3028 static void *x264_slices_write( x264_t *h )
3029 {
3030     int i_slice_num = 0;
3031     int last_thread_mb = h->sh.i_last_mb;
3032
3033     /* init stats */
3034     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
3035     h->mb.b_reencode_mb = 0;
3036     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
3037     {
3038         h->sh.i_last_mb = last_thread_mb;
3039         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
3040         {
3041             if( h->param.i_slice_max_mbs )
3042             {
3043                 if( SLICE_MBAFF )
3044                 {
3045                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
3046                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
3047                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
3048                         + h->param.i_slice_max_mbs - 1;
3049                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
3050                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
3051                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
3052                 }
3053                 else
3054                 {
3055                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
3056                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
3057                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
3058                 }
3059                 i_slice_num++;
3060             }
3061             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3062             {
3063                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3064                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3065                 i_slice_num++;
3066                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3067             }
3068         }
3069         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3070         if( x264_stack_align( x264_slice_write, h ) )
3071             goto fail;
3072         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3073         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3074         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3075             h->sh.i_first_mb -= h->mb.i_mb_stride;
3076     }
3077
3078     return (void *)0;
3079
3080 fail:
3081     /* Tell other threads we're done, so they wouldn't wait for it */
3082     if( h->param.b_sliced_threads )
3083         x264_threadslice_cond_broadcast( h, 2 );
3084     return (void *)-1;
3085 }
3086
3087 static int x264_threaded_slices_write( x264_t *h )
3088 {
3089     /* set first/last mb and sync contexts */
3090     for( int i = 0; i < h->param.i_threads; i++ )
3091     {
3092         x264_t *t = h->thread[i];
3093         if( i )
3094         {
3095             t->param = h->param;
3096             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3097         }
3098         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3099         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3100         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3101         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3102         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3103     }
3104
3105     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3106
3107     x264_threads_distribute_ratecontrol( h );
3108
3109     /* setup */
3110     for( int i = 0; i < h->param.i_threads; i++ )
3111     {
3112         h->thread[i]->i_thread_idx = i;
3113         h->thread[i]->b_thread_active = 1;
3114         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3115     }
3116     /* dispatch */
3117     for( int i = 0; i < h->param.i_threads; i++ )
3118         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3119     /* wait */
3120     for( int i = 0; i < h->param.i_threads; i++ )
3121         x264_threadslice_cond_wait( h->thread[i], 1 );
3122
3123     x264_threads_merge_ratecontrol( h );
3124
3125     for( int i = 1; i < h->param.i_threads; i++ )
3126     {
3127         x264_t *t = h->thread[i];
3128         for( int j = 0; j < t->out.i_nal; j++ )
3129         {
3130             h->out.nal[h->out.i_nal] = t->out.nal[j];
3131             h->out.i_nal++;
3132             x264_nal_check_buffer( h );
3133         }
3134         /* All entries in stat.frame are ints except for ssd/ssim. */
3135         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3136             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3137         for( int j = 0; j < 3; j++ )
3138             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3139         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3140         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3141     }
3142
3143     return 0;
3144 }
3145
3146 void x264_encoder_intra_refresh( x264_t *h )
3147 {
3148     h = h->thread[h->i_thread_phase];
3149     h->b_queued_intra_refresh = 1;
3150 }
3151
3152 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3153 {
3154     if( h->param.i_bframe )
3155     {
3156         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3157         return -1;
3158     }
3159     if( h->param.b_intra_refresh )
3160     {
3161         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3162         return -1;
3163     }
3164     h = h->thread[h->i_thread_phase];
3165     if( pts >= h->i_last_idr_pts )
3166     {
3167         for( int i = 0; h->frames.reference[i]; i++ )
3168             if( pts <= h->frames.reference[i]->i_pts )
3169                 h->frames.reference[i]->b_corrupt = 1;
3170         if( pts <= h->fdec->i_pts )
3171             h->fdec->b_corrupt = 1;
3172     }
3173     return 0;
3174 }
3175
3176 /****************************************************************************
3177  * x264_encoder_encode:
3178  *  XXX: i_poc   : is the poc of the current given picture
3179  *       i_frame : is the number of the frame being coded
3180  *  ex:  type frame poc
3181  *       I      0   2*0
3182  *       P      1   2*3
3183  *       B      2   2*1
3184  *       B      3   2*2
3185  *       P      4   2*6
3186  *       B      5   2*4
3187  *       B      6   2*5
3188  ****************************************************************************/
3189 int     x264_encoder_encode( x264_t *h,
3190                              x264_nal_t **pp_nal, int *pi_nal,
3191                              x264_picture_t *pic_in,
3192                              x264_picture_t *pic_out )
3193 {
3194     x264_t *thread_current, *thread_prev, *thread_oldest;
3195     int i_nal_type, i_nal_ref_idc, i_global_qp;
3196     int overhead = NALU_OVERHEAD;
3197
3198 #if HAVE_OPENCL
3199     if( h->opencl.b_fatal_error )
3200         return -1;
3201 #endif
3202
3203     if( h->i_thread_frames > 1 )
3204     {
3205         thread_prev    = h->thread[ h->i_thread_phase ];
3206         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3207         thread_current = h->thread[ h->i_thread_phase ];
3208         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3209         x264_thread_sync_context( thread_current, thread_prev );
3210         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3211         h = thread_current;
3212     }
3213     else
3214     {
3215         thread_current =
3216         thread_oldest  = h;
3217     }
3218     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3219
3220     /* no data out */
3221     *pi_nal = 0;
3222     *pp_nal = NULL;
3223
3224     /* ------------------- Setup new frame from picture -------------------- */
3225     if( pic_in != NULL )
3226     {
3227         /* 1: Copy the picture to a frame and move it to a buffer */
3228         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3229         if( !fenc )
3230             return -1;
3231
3232         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3233             return -1;
3234
3235         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3236             h->param.i_height != 16 * h->mb.i_mb_height )
3237             x264_frame_expand_border_mod16( h, fenc );
3238
3239         fenc->i_frame = h->frames.i_input++;
3240
3241         if( fenc->i_frame == 0 )
3242             h->frames.i_first_pts = fenc->i_pts;
3243         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3244             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3245
3246         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3247             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3248
3249         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3250         h->frames.i_largest_pts = fenc->i_pts;
3251
3252         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3253             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3254
3255         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3256         {
3257 #if HAVE_INTERLACED
3258             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3259 #else
3260             int b_interlaced = 0;
3261 #endif
3262             if( b_interlaced )
3263             {
3264                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3265                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3266             }
3267             else
3268                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3269         }
3270
3271         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3272         {
3273             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3274                 return -1;
3275         }
3276         else
3277             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3278
3279         if( pic_in->prop.quant_offsets_free )
3280             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3281
3282         if( h->frames.b_have_lowres )
3283             x264_frame_init_lowres( h, fenc );
3284
3285         /* 2: Place the frame into the queue for its slice type decision */
3286         x264_lookahead_put_frame( h, fenc );
3287
3288         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3289         {
3290             /* Nothing yet to encode, waiting for filling of buffers */
3291             pic_out->i_type = X264_TYPE_AUTO;
3292             return 0;
3293         }
3294     }
3295     else
3296     {
3297         /* signal kills for lookahead thread */
3298         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3299         h->lookahead->b_exit_thread = 1;
3300         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3301         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3302     }
3303
3304     h->i_frame++;
3305     /* 3: The picture is analyzed in the lookahead */
3306     if( !h->frames.current[0] )
3307         x264_lookahead_get_frames( h );
3308
3309     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3310         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3311
3312     /* ------------------- Get frame to be encoded ------------------------- */
3313     /* 4: get picture to encode */
3314     h->fenc = x264_frame_shift( h->frames.current );
3315
3316     /* If applicable, wait for previous frame reconstruction to finish */
3317     if( h->param.b_sliced_threads )
3318         if( x264_threadpool_wait_all( h ) < 0 )
3319             return -1;
3320
3321     if( h->i_frame == h->i_thread_frames - 1 )
3322         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3323     if( h->reconfig )
3324     {
3325         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3326         h->reconfig = 0;
3327     }
3328     if( h->fenc->param )
3329     {
3330         x264_encoder_reconfig_apply( h, h->fenc->param );
3331         if( h->fenc->param->param_free )
3332         {
3333             h->fenc->param->param_free( h->fenc->param );
3334             h->fenc->param = NULL;
3335         }
3336     }
3337
3338     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3339     if( x264_reference_update( h ) )
3340         return -1;
3341     h->fdec->i_lines_completed = -1;
3342
3343     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3344     {
3345         int valid_refs_left = 0;
3346         for( int i = 0; h->frames.reference[i]; i++ )
3347             if( !h->frames.reference[i]->b_corrupt )
3348                 valid_refs_left++;
3349         /* No valid reference frames left: force an IDR. */
3350         if( !valid_refs_left )
3351         {
3352             h->fenc->b_keyframe = 1;
3353             h->fenc->i_type = X264_TYPE_IDR;
3354         }
3355     }
3356
3357     if( h->fenc->b_keyframe )
3358     {
3359         h->frames.i_last_keyframe = h->fenc->i_frame;
3360         if( h->fenc->i_type == X264_TYPE_IDR )
3361         {
3362             h->i_frame_num = 0;
3363             h->frames.i_last_idr = h->fenc->i_frame;
3364         }
3365     }
3366     h->sh.i_mmco_command_count =
3367     h->sh.i_mmco_remove_from_end = 0;
3368     h->b_ref_reorder[0] =
3369     h->b_ref_reorder[1] = 0;
3370     h->fdec->i_poc =
3371     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3372
3373     /* ------------------- Setup frame context ----------------------------- */
3374     /* 5: Init data dependent of frame type */
3375     if( h->fenc->i_type == X264_TYPE_IDR )
3376     {
3377         /* reset ref pictures */
3378         i_nal_type    = NAL_SLICE_IDR;
3379         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3380         h->sh.i_type = SLICE_TYPE_I;
3381         x264_reference_reset( h );
3382         h->frames.i_poc_last_open_gop = -1;
3383     }
3384     else if( h->fenc->i_type == X264_TYPE_I )
3385     {
3386         i_nal_type    = NAL_SLICE;
3387         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3388         h->sh.i_type = SLICE_TYPE_I;
3389         x264_reference_hierarchy_reset( h );
3390         if( h->param.b_open_gop )
3391             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3392     }
3393     else if( h->fenc->i_type == X264_TYPE_P )
3394     {
3395         i_nal_type    = NAL_SLICE;
3396         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3397         h->sh.i_type = SLICE_TYPE_P;
3398         x264_reference_hierarchy_reset( h );
3399         h->frames.i_poc_last_open_gop = -1;
3400     }
3401     else if( h->fenc->i_type == X264_TYPE_BREF )
3402     {
3403         i_nal_type    = NAL_SLICE;
3404         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3405         h->sh.i_type = SLICE_TYPE_B;
3406         x264_reference_hierarchy_reset( h );
3407     }
3408     else    /* B frame */
3409     {
3410         i_nal_type    = NAL_SLICE;
3411         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3412         h->sh.i_type = SLICE_TYPE_B;
3413     }
3414
3415     h->fdec->i_type = h->fenc->i_type;
3416     h->fdec->i_frame = h->fenc->i_frame;
3417     h->fenc->b_kept_as_ref =
3418     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3419
3420     h->fdec->mb_info = h->fenc->mb_info;
3421     h->fdec->mb_info_free = h->fenc->mb_info_free;
3422     h->fenc->mb_info = NULL;
3423     h->fenc->mb_info_free = NULL;
3424
3425     h->fdec->i_pts = h->fenc->i_pts;
3426     if( h->frames.i_bframe_delay )
3427     {
3428         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3429         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3430                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3431                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3432         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3433     }
3434     else
3435         h->fdec->i_dts = h->fenc->i_reordered_pts;
3436     if( h->fenc->i_type == X264_TYPE_IDR )
3437         h->i_last_idr_pts = h->fdec->i_pts;
3438
3439     /* ------------------- Init                ----------------------------- */
3440     /* build ref list 0/1 */
3441     x264_reference_build_list( h, h->fdec->i_poc );
3442
3443     /* ---------------------- Write the bitstream -------------------------- */
3444     /* Init bitstream context */
3445     if( h->param.b_sliced_threads )
3446     {
3447         for( int i = 0; i < h->param.i_threads; i++ )
3448         {
3449             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3450             h->thread[i]->out.i_nal = 0;
3451         }
3452     }
3453     else
3454     {
3455         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3456         h->out.i_nal = 0;
3457     }
3458
3459     if( h->param.b_aud )
3460     {
3461         int pic_type;
3462
3463         if( h->sh.i_type == SLICE_TYPE_I )
3464             pic_type = 0;
3465         else if( h->sh.i_type == SLICE_TYPE_P )
3466             pic_type = 1;
3467         else if( h->sh.i_type == SLICE_TYPE_B )
3468             pic_type = 2;
3469         else
3470             pic_type = 7;
3471
3472         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3473         bs_write( &h->out.bs, 3, pic_type );
3474         bs_rbsp_trailing( &h->out.bs );
3475         if( x264_nal_end( h ) )
3476             return -1;
3477         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3478     }
3479
3480     h->i_nal_type = i_nal_type;
3481     h->i_nal_ref_idc = i_nal_ref_idc;
3482
3483     if( h->param.b_intra_refresh )
3484     {
3485         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3486         {
3487             h->fdec->i_frames_since_pir = 0;
3488             h->b_queued_intra_refresh = 0;
3489             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3490              * the whole frame and counts as an intra refresh. */
3491             h->fdec->f_pir_position = h->mb.i_mb_width;
3492         }
3493         else if( h->fenc->i_type == X264_TYPE_P )
3494         {
3495             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3496             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3497             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3498             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3499             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3500                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3501             {
3502                 h->fdec->f_pir_position = 0;
3503                 h->fdec->i_frames_since_pir = 0;
3504                 h->b_queued_intra_refresh = 0;
3505                 h->fenc->b_keyframe = 1;
3506             }
3507             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3508             h->fdec->f_pir_position += increment * pocdiff;
3509             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3510             /* If our intra refresh has reached the right side of the frame, we're done. */
3511             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3512             {
3513                 h->fdec->f_pir_position = h->mb.i_mb_width;
3514                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3515             }
3516         }
3517     }
3518
3519     if( h->fenc->b_keyframe )
3520     {
3521         /* Write SPS and PPS */
3522         if( h->param.b_repeat_headers )
3523         {
3524             /* generate sequence parameters */
3525             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3526             x264_sps_write( &h->out.bs, h->sps );
3527             if( x264_nal_end( h ) )
3528                 return -1;
3529             /* Pad AUD/SPS to 256 bytes like Panasonic */
3530             if( h->param.i_avcintra_class )
3531                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3532             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3533
3534             /* generate picture parameters */
3535             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3536             x264_pps_write( &h->out.bs, h->sps, h->pps );
3537             if( x264_nal_end( h ) )
3538                 return -1;
3539             if( h->param.i_avcintra_class )
3540                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3541             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3542         }
3543
3544         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3545         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3546         {
3547             x264_hrd_fullness( h );
3548             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3549             x264_sei_buffering_period_write( h, &h->out.bs );
3550             if( x264_nal_end( h ) )
3551                return -1;
3552             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3553         }
3554     }
3555
3556     /* write extra sei */
3557     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3558     {
3559         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3560         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3561                         h->fenc->extra_sei.payloads[i].payload_type );
3562         if( x264_nal_end( h ) )
3563             return -1;
3564         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3565         if( h->fenc->extra_sei.sei_free )
3566         {
3567             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3568             h->fenc->extra_sei.payloads[i].payload = NULL;
3569         }
3570     }
3571
3572     if( h->fenc->extra_sei.sei_free )
3573     {
3574         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3575         h->fenc->extra_sei.payloads = NULL;
3576         h->fenc->extra_sei.sei_free = NULL;
3577     }
3578
3579     if( h->fenc->b_keyframe )
3580     {
3581         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3582         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3583         {
3584             /* identify ourself */
3585             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3586             if( x264_sei_version_write( h, &h->out.bs ) )
3587                 return -1;
3588             if( x264_nal_end( h ) )
3589                 return -1;
3590             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3591         }
3592
3593         if( h->fenc->i_type != X264_TYPE_IDR )
3594         {
3595             int time_to_recovery = h->param.b_open_gop ? 0 : X264_MIN( h->mb.i_mb_width - 1, h->param.i_keyint_max ) + h->param.i_bframe - 1;
3596             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3597             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3598             if( x264_nal_end( h ) )
3599                 return -1;
3600             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3601         }
3602     }
3603
3604     if( h->param.i_frame_packing >= 0 && (h->fenc->b_keyframe || h->param.i_frame_packing == 5) )
3605     {
3606         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3607         x264_sei_frame_packing_write( h, &h->out.bs );
3608         if( x264_nal_end( h ) )
3609             return -1;
3610         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3611     }
3612
3613     /* generate sei pic timing */
3614     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3615     {
3616         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3617         x264_sei_pic_timing_write( h, &h->out.bs );
3618         if( x264_nal_end( h ) )
3619             return -1;
3620         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3621     }
3622
3623     /* As required by Blu-ray. */
3624     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3625     {
3626         h->b_sh_backup = 0;
3627         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3628         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3629         if( x264_nal_end( h ) )
3630             return -1;
3631         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3632     }
3633
3634     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3635         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3636
3637     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3638     if( h->param.i_avcintra_class )
3639     {
3640         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3641         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3642         x264_filler_write( h, &h->out.bs, 0 );
3643         if( x264_nal_end( h ) )
3644             return -1;
3645         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3646
3647         /* All lengths are magic lengths that decoders expect to see */
3648         /* "UMID" SEI */
3649         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3650         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3651             return -1;
3652         if( x264_nal_end( h ) )
3653             return -1;
3654         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3655
3656         int unpadded_len;
3657         int total_len;
3658         if( h->param.i_height == 1080 )
3659         {
3660             unpadded_len = 5780;
3661             total_len = 17*512;
3662         }
3663         else
3664         {
3665             unpadded_len = 2900;
3666             total_len = 9*512;
3667         }
3668         /* "VANC" SEI */
3669         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3670         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3671             return -1;
3672         if( x264_nal_end( h ) )
3673             return -1;
3674
3675         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3676         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3677     }
3678
3679     /* Init the rate control */
3680     /* FIXME: Include slice header bit cost. */
3681     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3682     i_global_qp = x264_ratecontrol_qp( h );
3683
3684     pic_out->i_qpplus1 =
3685     h->fdec->i_qpplus1 = i_global_qp + 1;
3686
3687     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3688     {
3689         x264_reference_build_list_optimal( h );
3690         x264_reference_check_reorder( h );
3691     }
3692
3693     if( h->i_ref[0] )
3694         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3695
3696     /* ------------------------ Create slice header  ----------------------- */
3697     x264_slice_init( h, i_nal_type, i_global_qp );
3698
3699     /*------------------------- Weights -------------------------------------*/
3700     if( h->sh.i_type == SLICE_TYPE_B )
3701         x264_macroblock_bipred_init( h );
3702
3703     x264_weighted_pred_init( h );
3704
3705     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3706         h->i_frame_num++;
3707
3708     /* Write frame */
3709     h->i_threadslice_start = 0;
3710     h->i_threadslice_end = h->mb.i_mb_height;
3711     if( h->i_thread_frames > 1 )
3712     {
3713         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3714         h->b_thread_active = 1;
3715     }
3716     else if( h->param.b_sliced_threads )
3717     {
3718         if( x264_threaded_slices_write( h ) )
3719             return -1;
3720     }
3721     else
3722         if( (intptr_t)x264_slices_write( h ) )
3723             return -1;
3724
3725     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3726 }
3727
3728 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3729                                    x264_nal_t **pp_nal, int *pi_nal,
3730                                    x264_picture_t *pic_out )
3731 {
3732     char psz_message[80];
3733
3734     if( !h->param.b_sliced_threads && h->b_thread_active )
3735     {
3736         h->b_thread_active = 0;
3737         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3738             return -1;
3739     }
3740     if( !h->out.i_nal )
3741     {
3742         pic_out->i_type = X264_TYPE_AUTO;
3743         return 0;
3744     }
3745
3746     x264_emms();
3747
3748     /* generate buffering period sei and insert it into place */
3749     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3750     {
3751         x264_hrd_fullness( h );
3752         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3753         x264_sei_buffering_period_write( h, &h->out.bs );
3754         if( x264_nal_end( h ) )
3755            return -1;
3756         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3757         int idx = 0;
3758         while( h->out.nal[idx].i_type == NAL_AUD ||
3759                h->out.nal[idx].i_type == NAL_SPS ||
3760                h->out.nal[idx].i_type == NAL_PPS )
3761             idx++;
3762         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3763         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3764         h->out.nal[idx] = nal_tmp;
3765     }
3766
3767     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3768     if( frame_size < 0 )
3769         return -1;
3770
3771     /* Set output picture properties */
3772     pic_out->i_type = h->fenc->i_type;
3773
3774     pic_out->b_keyframe = h->fenc->b_keyframe;
3775     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3776
3777     pic_out->i_pts = h->fdec->i_pts;
3778     pic_out->i_dts = h->fdec->i_dts;
3779
3780     if( pic_out->i_pts < pic_out->i_dts )
3781         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3782
3783     pic_out->opaque = h->fenc->opaque;
3784
3785     pic_out->img.i_csp = h->fdec->i_csp;
3786 #if HIGH_BIT_DEPTH
3787     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3788 #endif
3789     pic_out->img.i_plane = h->fdec->i_plane;
3790     for( int i = 0; i < pic_out->img.i_plane; i++ )
3791     {
3792         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3793         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3794     }
3795
3796     x264_frame_push_unused( thread_current, h->fenc );
3797
3798     /* ---------------------- Update encoder state ------------------------- */
3799
3800     /* update rc */
3801     int filler = 0;
3802     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3803         return -1;
3804
3805     pic_out->hrd_timing = h->fenc->hrd_timing;
3806     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3807
3808     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3809      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3810     if( h->param.i_avcintra_class )
3811     {
3812         x264_t *h0 = h->thread[0];
3813         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3814         if( ret < 0 )
3815             return -1;
3816         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3817         h->out.nal[h->out.i_nal-1].i_payload += filler;
3818         h->out.nal[h->out.i_nal-1].i_padding = filler;
3819         frame_size += filler;
3820     }
3821     else
3822     {
3823         while( filler > 0 )
3824         {
3825             int f, overhead;
3826             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3827             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3828             {
3829                 int next_size = filler - h->param.i_slice_max_size;
3830                 int overflow = X264_MAX( overhead - next_size, 0 );
3831                 f = h->param.i_slice_max_size - overhead - overflow;
3832             }
3833             else
3834                 f = X264_MAX( 0, filler - overhead );
3835
3836             if( x264_bitstream_check_buffer_filler( h, f ) )
3837                 return -1;
3838             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3839             x264_filler_write( h, &h->out.bs, f );
3840             if( x264_nal_end( h ) )
3841                 return -1;
3842             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3843             if( total_size < 0 )
3844                 return -1;
3845             frame_size += total_size;
3846             filler -= total_size;
3847         }
3848     }
3849
3850     /* End bitstream, set output  */
3851     *pi_nal = h->out.i_nal;
3852     *pp_nal = h->out.nal;
3853
3854     h->out.i_nal = 0;
3855
3856     x264_noise_reduction_update( h );
3857
3858     /* ---------------------- Compute/Print statistics --------------------- */
3859     x264_thread_sync_stat( h, h->thread[0] );
3860
3861     /* Slice stat */
3862     h->stat.i_frame_count[h->sh.i_type]++;
3863     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3864     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3865
3866     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3867         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3868     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3869         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3870     for( int i = 0; i < 2; i++ )
3871         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3872     for( int i = 0; i < 6; i++ )
3873         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3874     for( int i = 0; i < 4; i++ )
3875         for( int j = 0; j < 13; j++ )
3876             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3877     if( h->sh.i_type != SLICE_TYPE_I )
3878         for( int i_list = 0; i_list < 2; i_list++ )
3879             for( int i = 0; i < X264_REF_MAX*2; i++ )
3880                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3881     for( int i = 0; i < 3; i++ )
3882         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3883     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3884     {
3885         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3886         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3887     }
3888     if( h->sh.i_type == SLICE_TYPE_B )
3889     {
3890         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3891         if( h->mb.b_direct_auto_write )
3892         {
3893             //FIXME somewhat arbitrary time constants
3894             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3895                 for( int i = 0; i < 2; i++ )
3896                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3897             for( int i = 0; i < 2; i++ )
3898                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3899         }
3900     }
3901     else
3902         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3903
3904     psz_message[0] = '\0';
3905     double dur = h->fenc->f_duration;
3906     h->stat.f_frame_duration[h->sh.i_type] += dur;
3907     if( h->param.analyse.b_psnr )
3908     {
3909         int64_t ssd[3] =
3910         {
3911             h->stat.frame.i_ssd[0],
3912             h->stat.frame.i_ssd[1],
3913             h->stat.frame.i_ssd[2],
3914         };
3915         int luma_size = h->param.i_width * h->param.i_height;
3916         int chroma_size = CHROMA_SIZE( luma_size );
3917         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3918         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3919         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3920         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3921
3922         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3923         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3924         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3925         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3926         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3927
3928         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3929                                                                     pic_out->prop.f_psnr[1],
3930                                                                     pic_out->prop.f_psnr[2] );
3931     }
3932
3933     if( h->param.analyse.b_ssim )
3934     {
3935         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3936         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3937         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3938                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3939     }
3940     psz_message[79] = '\0';
3941
3942     x264_log( h, X264_LOG_DEBUG,
3943                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3944               h->i_frame,
3945               h->fdec->f_qp_avg_aq,
3946               h->i_nal_ref_idc,
3947               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3948               h->fdec->i_poc,
3949               h->stat.frame.i_mb_count_i,
3950               h->stat.frame.i_mb_count_p,
3951               h->stat.frame.i_mb_count_skip,
3952               frame_size,
3953               psz_message );
3954
3955     // keep stats all in one place
3956     x264_thread_sync_stat( h->thread[0], h );
3957     // for the use of the next frame
3958     x264_thread_sync_stat( thread_current, h );
3959
3960 #ifdef DEBUG_MB_TYPE
3961 {
3962     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3963         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3964     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3965     {
3966         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3967             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3968         else
3969             fprintf( stderr, "? " );
3970
3971         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3972             fprintf( stderr, "\n" );
3973     }
3974 }
3975 #endif
3976
3977     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3978      * by freeing some of its pointers. */
3979     for( int i = 0; i < h->i_ref[0]; i++ )
3980         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3981         {
3982             x264_frame_push_blank_unused( h, h->fref[0][i] );
3983             h->fref[0][i] = 0;
3984         }
3985
3986     if( h->param.psz_dump_yuv )
3987         x264_frame_dump( h );
3988     x264_emms();
3989
3990     return frame_size;
3991 }
3992
3993 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3994 {
3995     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3996         b_print_pcm ? "..PCM" : "",
3997         i_mb_count[I_16x16]/ i_count,
3998         i_mb_count[I_8x8]  / i_count,
3999         i_mb_count[I_4x4]  / i_count );
4000     if( b_print_pcm )
4001         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
4002 }
4003
4004 /****************************************************************************
4005  * x264_encoder_close:
4006  ****************************************************************************/
4007 void    x264_encoder_close  ( x264_t *h )
4008 {
4009     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
4010     int64_t i_mb_count_size[2][7] = {{0}};
4011     char buf[200];
4012     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
4013                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
4014                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
4015
4016     x264_lookahead_delete( h );
4017
4018 #if HAVE_OPENCL
4019     x264_opencl_lookahead_delete( h );
4020     x264_opencl_function_t *ocl = h->opencl.ocl;
4021 #endif
4022
4023     if( h->param.b_sliced_threads )
4024         x264_threadpool_wait_all( h );
4025     if( h->param.i_threads > 1 )
4026         x264_threadpool_delete( h->threadpool );
4027     if( h->param.i_lookahead_threads > 1 )
4028         x264_threadpool_delete( h->lookaheadpool );
4029     if( h->i_thread_frames > 1 )
4030     {
4031         for( int i = 0; i < h->i_thread_frames; i++ )
4032             if( h->thread[i]->b_thread_active )
4033             {
4034                 assert( h->thread[i]->fenc->i_reference_count == 1 );
4035                 x264_frame_delete( h->thread[i]->fenc );
4036             }
4037
4038         x264_t *thread_prev = h->thread[h->i_thread_phase];
4039         x264_thread_sync_ratecontrol( h, thread_prev, h );
4040         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
4041         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
4042     }
4043     h->i_frame++;
4044
4045     /* Slices used and PSNR */
4046     for( int i = 0; i < 3; i++ )
4047     {
4048         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
4049         int i_slice = slice_order[i];
4050
4051         if( h->stat.i_frame_count[i_slice] > 0 )
4052         {
4053             int i_count = h->stat.i_frame_count[i_slice];
4054             double dur =  h->stat.f_frame_duration[i_slice];
4055             if( h->param.analyse.b_psnr )
4056             {
4057                 x264_log( h, X264_LOG_INFO,
4058                           "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",
4059                           slice_type_to_char[i_slice],
4060                           i_count,
4061                           h->stat.f_frame_qp[i_slice] / i_count,
4062                           (double)h->stat.i_frame_size[i_slice] / i_count,
4063                           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,
4064                           h->stat.f_psnr_average[i_slice] / dur,
4065                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4066             }
4067             else
4068             {
4069                 x264_log( h, X264_LOG_INFO,
4070                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4071                           slice_type_to_char[i_slice],
4072                           i_count,
4073                           h->stat.f_frame_qp[i_slice] / i_count,
4074                           (double)h->stat.i_frame_size[i_slice] / i_count );
4075             }
4076         }
4077     }
4078     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4079     {
4080         char *p = buf;
4081         int den = 0;
4082         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4083         for( int i = 0; i <= h->param.i_bframe; i++ )
4084             den += (i+1) * h->stat.i_consecutive_bframes[i];
4085         for( int i = 0; i <= h->param.i_bframe; i++ )
4086             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4087         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4088     }
4089
4090     for( int i_type = 0; i_type < 2; i_type++ )
4091         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4092         {
4093             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4094             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4095         }
4096
4097     /* MB types used */
4098     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4099     {
4100         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4101         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4102         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4103         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4104     }
4105     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4106     {
4107         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4108         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4109         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4110         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4111         x264_log( h, X264_LOG_INFO,
4112                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4113                   buf,
4114                   i_mb_size[PIXEL_16x16] / (i_count*4),
4115                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4116                   i_mb_size[PIXEL_8x8] / (i_count*4),
4117                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4118                   i_mb_size[PIXEL_4x4] / (i_count*4),
4119                   i_mb_count[P_SKIP] / i_count );
4120     }
4121     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4122     {
4123         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4124         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4125         double i_mb_list_count;
4126         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4127         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4128         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4129         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4130             for( int j = 0; j < 2; j++ )
4131             {
4132                 int l0 = x264_mb_type_list_table[i][0][j];
4133                 int l1 = x264_mb_type_list_table[i][1][j];
4134                 if( l0 || l1 )
4135                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4136             }
4137         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4138         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4139         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4140         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4141         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4142         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4143                  i_mb_size[PIXEL_16x16] / (i_count*4),
4144                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4145                  i_mb_size[PIXEL_8x8] / (i_count*4),
4146                  i_mb_count[B_DIRECT] / i_count,
4147                  i_mb_count[B_SKIP]   / i_count );
4148         if( i_mb_list_count != 0 )
4149             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4150                      list_count[0] / i_mb_list_count,
4151                      list_count[1] / i_mb_list_count,
4152                      list_count[2] / i_mb_list_count );
4153         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4154     }
4155
4156     x264_ratecontrol_summary( h );
4157
4158     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 )
4159     {
4160 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4161 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4162         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4163         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4164                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4165         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4166         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4167                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4168         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4169                             h->stat.i_frame_count[SLICE_TYPE_P] +
4170                             h->stat.i_frame_count[SLICE_TYPE_B];
4171         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4172         int64_t i_inter = i_mb_count - i_skip - i_intra;
4173         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4174                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4175                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4176         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4177
4178         if( PARAM_INTERLACED )
4179         {
4180             char *fieldstats = buf;
4181             fieldstats[0] = 0;
4182             if( i_inter )
4183                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4184             if( i_skip )
4185                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4186             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4187                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4188         }
4189
4190         if( h->pps->b_transform_8x8_mode )
4191         {
4192             buf[0] = 0;
4193             if( h->stat.i_mb_count_8x8dct[0] )
4194                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4195             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4196         }
4197
4198         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4199             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4200             && h->stat.i_frame_count[SLICE_TYPE_B] )
4201         {
4202             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4203                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4204                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4205         }
4206
4207         buf[0] = 0;
4208         int csize = CHROMA444 ? 4 : 1;
4209         if( i_mb_count != i_all_intra )
4210             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4211                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4212                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4213                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4214         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4215                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4216                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4217                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4218                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4219
4220         int64_t fixed_pred_modes[4][9] = {{0}};
4221         int64_t sum_pred_modes[4] = {0};
4222         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4223         {
4224             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4225             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4226         }
4227         if( sum_pred_modes[0] )
4228             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4229                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4230                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4231                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4232                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4233         for( int i = 1; i <= 2; i++ )
4234         {
4235             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
4236             {
4237                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
4238                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
4239             }
4240             if( sum_pred_modes[i] )
4241                 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,
4242                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
4243                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
4244                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
4245                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
4246                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
4247                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
4248                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
4249                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
4250                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
4251         }
4252         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
4253         {
4254             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
4255             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
4256         }
4257         if( sum_pred_modes[3] && !CHROMA444 )
4258             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4259                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
4260                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
4261                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
4262                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
4263
4264         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4265             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
4266                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
4267                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
4268
4269         for( int i_list = 0; i_list < 2; i_list++ )
4270             for( int i_slice = 0; i_slice < 2; i_slice++ )
4271             {
4272                 char *p = buf;
4273                 int64_t i_den = 0;
4274                 int i_max = 0;
4275                 for( int i = 0; i < X264_REF_MAX*2; i++ )
4276                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
4277                     {
4278                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
4279                         i_max = i;
4280                     }
4281                 if( i_max == 0 )
4282                     continue;
4283                 for( int i = 0; i <= i_max; i++ )
4284                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
4285                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
4286             }
4287
4288         if( h->param.analyse.b_ssim )
4289         {
4290             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
4291             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
4292         }
4293         if( h->param.analyse.b_psnr )
4294         {
4295             x264_log( h, X264_LOG_INFO,
4296                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
4297                       SUM3( h->stat.f_psnr_mean_y ) / duration,
4298                       SUM3( h->stat.f_psnr_mean_u ) / duration,
4299                       SUM3( h->stat.f_psnr_mean_v ) / duration,
4300                       SUM3( h->stat.f_psnr_average ) / duration,
4301                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
4302                       f_bitrate );
4303         }
4304         else
4305             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
4306     }
4307
4308     /* rc */
4309     x264_ratecontrol_delete( h );
4310
4311     /* param */
4312     if( h->param.rc.psz_stat_out )
4313         free( h->param.rc.psz_stat_out );
4314     if( h->param.rc.psz_stat_in )
4315         free( h->param.rc.psz_stat_in );
4316
4317     x264_cqm_delete( h );
4318     x264_free( h->nal_buffer );
4319     x264_free( h->reconfig_h );
4320     x264_analyse_free_costs( h );
4321
4322     if( h->i_thread_frames > 1 )
4323         h = h->thread[h->i_thread_phase];
4324
4325     /* frames */
4326     x264_frame_delete_list( h->frames.unused[0] );
4327     x264_frame_delete_list( h->frames.unused[1] );
4328     x264_frame_delete_list( h->frames.current );
4329     x264_frame_delete_list( h->frames.blank_unused );
4330
4331     h = h->thread[0];
4332
4333     for( int i = 0; i < h->i_thread_frames; i++ )
4334         if( h->thread[i]->b_thread_active )
4335             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4336                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4337                     x264_frame_delete( h->thread[i]->fref[0][j] );
4338
4339     if( h->param.i_lookahead_threads > 1 )
4340         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4341             x264_free( h->lookahead_thread[i] );
4342
4343     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4344     {
4345         x264_frame_t **frame;
4346
4347         if( !h->param.b_sliced_threads || i == 0 )
4348         {
4349             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4350             {
4351                 assert( (*frame)->i_reference_count > 0 );
4352                 (*frame)->i_reference_count--;
4353                 if( (*frame)->i_reference_count == 0 )
4354                     x264_frame_delete( *frame );
4355             }
4356             frame = &h->thread[i]->fdec;
4357             if( *frame )
4358             {
4359                 assert( (*frame)->i_reference_count > 0 );
4360                 (*frame)->i_reference_count--;
4361                 if( (*frame)->i_reference_count == 0 )
4362                     x264_frame_delete( *frame );
4363             }
4364             x264_macroblock_cache_free( h->thread[i] );
4365         }
4366         x264_macroblock_thread_free( h->thread[i], 0 );
4367         x264_free( h->thread[i]->out.p_bitstream );
4368         x264_free( h->thread[i]->out.nal );
4369         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4370         x264_pthread_cond_destroy( &h->thread[i]->cv );
4371         x264_free( h->thread[i] );
4372     }
4373 #if HAVE_OPENCL
4374     x264_opencl_close_library( ocl );
4375 #endif
4376 }
4377
4378 int x264_encoder_delayed_frames( x264_t *h )
4379 {
4380     int delayed_frames = 0;
4381     if( h->i_thread_frames > 1 )
4382     {
4383         for( int i = 0; i < h->i_thread_frames; i++ )
4384             delayed_frames += h->thread[i]->b_thread_active;
4385         h = h->thread[h->i_thread_phase];
4386     }
4387     for( int i = 0; h->frames.current[i]; i++ )
4388         delayed_frames++;
4389     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4390     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4391     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4392     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4393     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4394     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4395     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4396     return delayed_frames;
4397 }
4398
4399 int x264_encoder_maximum_delayed_frames( x264_t *h )
4400 {
4401     return h->frames.i_delay;
4402 }