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_I422 )
 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_I444 )
 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/NV21/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
 496         return -1;
 497     }
 498
 499     int w_mod = i_csp < X264_CSP_I444 ? 2 : 1;
 500     int h_mod = (i_csp < X264_CSP_I422 ? 2 : 1) << PARAM_INTERLACED;
 501     if( h->param.i_width % w_mod )
 502     {
 503         x264_log( h, X264_LOG_ERROR, "width not divisible by %d (%dx%d)\n",
 504                   w_mod, h->param.i_width, h->param.i_height );
 505         return -1;
 506     }
 507     if( h->param.i_height % h_mod )
 508     {
 509         x264_log( h, X264_LOG_ERROR, "height not divisible by %d (%dx%d)\n",
 510                   h_mod, h->param.i_width, h->param.i_height );
 511         return -1;
 512     }
 513
 514     if( h->param.crop_rect.i_left   >= h->param.i_width ||
 515         h->param.crop_rect.i_right  >= h->param.i_width ||
 516         h->param.crop_rect.i_top    >= h->param.i_height ||
 517         h->param.crop_rect.i_bottom >= h->param.i_height ||
 518         h->param.crop_rect.i_left + h->param.crop_rect.i_right  >= h->param.i_width ||
 519         h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom >= h->param.i_height )
 520     {
 521         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 522                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 523         return -1;
 524     }
 525     if( h->param.crop_rect.i_left % w_mod || h->param.crop_rect.i_right  % w_mod ||
 526         h->param.crop_rect.i_top  % h_mod || h->param.crop_rect.i_bottom % h_mod )
 527     {
 528         x264_log( h, X264_LOG_ERROR, "crop-rect %u,%u,%u,%u not divisible by %dx%d\n", h->param.crop_rect.i_left,
 529                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom, w_mod, h_mod );
 530         return -1;
 531     }
 532
 533     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
 534     {
 535         h->param.vui.i_sar_width = 0;
 536         h->param.vui.i_sar_height = 0;
 537     }
 538
 539     if( h->param.i_threads == X264_THREADS_AUTO )
 540         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 541     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
 542     if( h->param.i_threads > 1 )
 543     {
 544 #if !HAVE_THREAD
 545         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 546         h->param.i_threads = 1;
 547 #endif
 548         /* Avoid absurdly small thread slices as they can reduce performance
 549          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 550         if( h->param.b_sliced_threads )
 551             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
 552     }
 553     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 554     if( h->param.i_threads == 1 )
 555     {
 556         h->param.b_sliced_threads = 0;
 557         h->param.i_lookahead_threads = 1;
 558     }
 559     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 560     if( h->i_thread_frames > 1 )
 561         h->param.nalu_process = NULL;
 562
 563     if( h->param.b_opencl )
 564     {
 565 #if !HAVE_OPENCL
 566         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
 567         h->param.b_opencl = 0;
 568 #elif BIT_DEPTH > 8
 569         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
 570         h->param.b_opencl = 0;
 571 #else
 572         if( h->param.i_width < 32 || h->param.i_height < 32 )
 573         {
 574             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
 575             h->param.b_opencl = 0;
 576         }
 577 #endif
 578         if( h->param.opencl_device_id && h->param.i_opencl_device )
 579         {
 580             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
 581             h->param.i_opencl_device = 0;
 582         }
 583     }
 584
 585     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 586     if( h->param.i_keyint_max == 1 )
 587     {
 588         h->param.b_intra_refresh = 0;
 589         h->param.analyse.i_weighted_pred = 0;
 590         h->param.i_frame_reference = 1;
 591         h->param.i_dpb_size = 1;
 592     }
 593
 594     if( h->param.i_frame_packing < -1 || h->param.i_frame_packing > 7 )
 595     {
 596         x264_log( h, X264_LOG_WARNING, "ignoring unknown frame packing value\n" );
 597         h->param.i_frame_packing = -1;
 598     }
 599     if( h->param.i_frame_packing == 7 &&
 600         ((h->param.i_width - h->param.crop_rect.i_left - h->param.crop_rect.i_right)  % 3 ||
 601          (h->param.i_height - h->param.crop_rect.i_top - h->param.crop_rect.i_bottom) % 3) )
 602     {
 603         x264_log( h, X264_LOG_ERROR, "cropped resolution %dx%d not compatible with tile format frame packing\n",
 604                   h->param.i_width - h->param.crop_rect.i_left - h->param.crop_rect.i_right,
 605                   h->param.i_height - h->param.crop_rect.i_top - h->param.crop_rect.i_bottom );
 606         return -1;
 607     }
 608
 609     /* Detect default ffmpeg settings and terminate with an error. */
 610     if( b_open )
 611     {
 612         int score = 0;
 613         score += h->param.analyse.i_me_range == 0;
 614         score += h->param.rc.i_qp_step == 3;
 615         score += h->param.i_keyint_max == 12;
 616         score += h->param.rc.i_qp_min == 2;
 617         score += h->param.rc.i_qp_max == 31;
 618         score += h->param.rc.f_qcompress == 0.5;
 619         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 620         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 621         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 622         if( score >= 5 )
 623         {
 624             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 625             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 626             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 627             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 628             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 629             return -1;
 630         }
 631     }
 632
 633     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 634     {
 635         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 636         return -1;
 637     }
 638
 639     if( PARAM_INTERLACED )
 640         h->param.b_pic_struct = 1;
 641
 642     if( h->param.i_avcintra_class )
 643     {
 644         if( BIT_DEPTH != 10 )
 645         {
 646             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
 647             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
 648             return -1;
 649         }
 650
 651         int type = h->param.i_avcintra_class == 200 ? 2 :
 652                    h->param.i_avcintra_class == 100 ? 1 :
 653                    h->param.i_avcintra_class == 50 ? 0 : -1;
 654         if( type < 0 )
 655         {
 656             x264_log( h, X264_LOG_ERROR, "Invalid AVC-Intra class\n" );
 657             return -1;
 658         }
 659
 660         /* [50/100/200][res][fps] */
 661         static const struct
 662         {
 663             uint16_t fps_num;
 664             uint16_t fps_den;
 665             uint8_t interlaced;
 666             uint16_t frame_size;
 667             const uint8_t *cqm_4ic;
 668             const uint8_t *cqm_8iy;
 669         } avcintra_lut[3][2][7] =
 670         {
 671             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 672               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 673               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 674               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 675               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
 676              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 677               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 678               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 679               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 680               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 681               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 682               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
 683             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 684               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 685               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 686               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 687               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 688              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 689               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 690               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 691               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 692               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 693               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 694               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}},
 695             {{{ 60000, 1001, 0, 3724, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 696               {    50,    1, 0, 4472, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 697              {{ 30000, 1001, 1, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 698               {    25,    1, 1, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 699               { 60000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 700               { 30000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 701               {    50,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 702               {    25,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 703               { 24000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
 704         };
 705
 706         int res = -1;
 707         if( i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 && !type )
 708         {
 709             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
 710             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
 711         }
 712         else if( i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 && type )
 713         {
 714             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
 715             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
 716         }
 717         else
 718         {
 719             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra %d\n", h->param.i_avcintra_class );
 720             return -1;
 721         }
 722
 723         if( res < 0 )
 724         {
 725             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %d\n",
 726                       h->param.i_width, h->param.i_height, h->param.i_avcintra_class );
 727             return -1;
 728         }
 729
 730         if( h->param.nalu_process )
 731         {
 732             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
 733             return -1;
 734         }
 735
 736         if( !h->param.b_repeat_headers )
 737         {
 738             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
 739             return -1;
 740         }
 741
 742         int i;
 743         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
 744         x264_reduce_fraction( &fps_num, &fps_den );
 745         for( i = 0; i < 7; i++ )
 746         {
 747             if( avcintra_lut[type][res][i].fps_num == fps_num &&
 748                 avcintra_lut[type][res][i].fps_den == fps_den &&
 749                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
 750             {
 751                 break;
 752             }
 753         }
 754         if( i == 7 )
 755         {
 756             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
 757                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
 758             return -1;
 759         }
 760
 761         h->param.i_keyint_max = 1;
 762         h->param.b_intra_refresh = 0;
 763         h->param.analyse.i_weighted_pred = 0;
 764         h->param.i_frame_reference = 1;
 765         h->param.i_dpb_size = 1;
 766
 767         h->param.b_bluray_compat = 0;
 768         h->param.b_vfr_input = 0;
 769         h->param.b_aud = 1;
 770         h->param.vui.i_chroma_loc = 0;
 771         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 772         h->param.b_deblocking_filter = 0;
 773         h->param.b_stitchable = 1;
 774         h->param.b_pic_struct = 0;
 775         h->param.analyse.b_transform_8x8 = 1;
 776         h->param.analyse.intra = X264_ANALYSE_I8x8;
 777         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
 778         h->param.b_cabac = !type;
 779         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
 780         h->param.rc.i_vbv_max_bitrate =
 781         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
 782         h->param.rc.i_rc_method = X264_RC_ABR;
 783         h->param.rc.f_vbv_buffer_init = 1.0;
 784         h->param.rc.b_filler = 1;
 785         h->param.i_cqm_preset = X264_CQM_CUSTOM;
 786         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
 787         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
 788         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
 789
 790         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
 791         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
 792         h->param.i_slice_max_size = 0;
 793         /* The slice structure only allows a maximum of 2 threads for 1080i/p
 794          * and 1 or 5 threads for 720p */
 795         if( h->param.b_sliced_threads )
 796         {
 797             if( res )
 798                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
 799             else
 800             {
 801                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
 802                 if( h->param.i_threads < 5 )
 803                     h->param.i_threads = 1;
 804             }
 805         }
 806
 807         if( type )
 808             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
 809         else
 810         {
 811             h->param.vui.i_sar_width  = 4;
 812             h->param.vui.i_sar_height = 3;
 813         }
 814
 815         /* Official encoder doesn't appear to go under 13
 816          * and Avid cannot handle negative QPs */
 817         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET + 1 );
 818     }
 819
 820     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 821     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
 822     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 823     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
 824     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
 825     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
 826     if( h->param.rc.i_rc_method == X264_RC_CRF )
 827     {
 828         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 829         h->param.rc.i_bitrate = 0;
 830     }
 831     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 832         && h->param.rc.i_qp_constant == 0 )
 833     {
 834         h->mb.b_lossless = 1;
 835         h->param.i_cqm_preset = X264_CQM_FLAT;
 836         h->param.psz_cqm_file = NULL;
 837         h->param.rc.i_rc_method = X264_RC_CQP;
 838         h->param.rc.f_ip_factor = 1;
 839         h->param.rc.f_pb_factor = 1;
 840         h->param.analyse.b_psnr = 0;
 841         h->param.analyse.b_ssim = 0;
 842         h->param.analyse.i_chroma_qp_offset = 0;
 843         h->param.analyse.i_trellis = 0;
 844         h->param.analyse.b_fast_pskip = 0;
 845         h->param.analyse.i_noise_reduction = 0;
 846         h->param.analyse.b_psy = 0;
 847         h->param.i_bframe = 0;
 848         /* 8x8dct is not useful without RD in CAVLC lossless */
 849         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 850             h->param.analyse.b_transform_8x8 = 0;
 851         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 852         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 853     }
 854     if( h->param.rc.i_rc_method == X264_RC_CQP )
 855     {
 856         float qp_p = h->param.rc.i_qp_constant;
 857         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 858         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 859         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 860         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 861         h->param.rc.i_aq_mode = 0;
 862         h->param.rc.b_mb_tree = 0;
 863         h->param.rc.i_bitrate = 0;
 864     }
 865     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 866     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 867     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
 868     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
 869     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
 870     {
 871         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
 872         return -1;
 873     }
 874     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
 875     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
 876     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
 877     if( h->param.rc.i_vbv_buffer_size )
 878     {
 879         if( h->param.rc.i_rc_method == X264_RC_CQP )
 880         {
 881             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 882             h->param.rc.i_vbv_max_bitrate = 0;
 883             h->param.rc.i_vbv_buffer_size = 0;
 884         }
 885         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 886         {
 887             if( h->param.rc.i_rc_method == X264_RC_ABR )
 888             {
 889                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 890                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 891             }
 892             else
 893             {
 894                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 895                 h->param.rc.i_vbv_buffer_size = 0;
 896             }
 897         }
 898         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 899                  h->param.rc.i_rc_method == X264_RC_ABR )
 900         {
 901             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 902             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
 903         }
 904     }
 905     else if( h->param.rc.i_vbv_max_bitrate )
 906     {
 907         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 908         h->param.rc.i_vbv_max_bitrate = 0;
 909     }
 910
 911     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 912     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 913     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
 914     if( h->param.i_slice_max_mbs )
 915         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
 916     else if( !h->param.i_slice_max_size )
 917         h->param.i_slice_min_mbs = 0;
 918     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
 919     {
 920         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
 921         h->param.i_slice_min_mbs = 0;
 922     }
 923     int mb_width = (h->param.i_width+15)/16;
 924     if( h->param.i_slice_min_mbs > mb_width )
 925     {
 926         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
 927         h->param.i_slice_min_mbs = mb_width;
 928     }
 929
 930     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
 931     if( h->param.b_sliced_threads )
 932         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 933     else
 934     {
 935         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 936         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 937             h->param.i_slice_count = 0;
 938     }
 939     if( h->param.i_slice_count_max > 0 )
 940         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
 941
 942     if( h->param.b_bluray_compat )
 943     {
 944         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
 945         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
 946         h->param.b_aud = 1;
 947         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
 948         h->param.i_slice_max_size = 0;
 949         h->param.i_slice_max_mbs = 0;
 950         h->param.b_intra_refresh = 0;
 951         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
 952         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
 953         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
 954         h->param.i_keyint_min = 1;
 955         /* Due to the proliferation of broken players that don't handle dupes properly. */
 956         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
 957         if( h->param.b_fake_interlaced )
 958             h->param.b_pic_struct = 1;
 959     }
 960
 961     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 962     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 963     if( h->param.i_scenecut_threshold < 0 )
 964         h->param.i_scenecut_threshold = 0;
 965     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
 966     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 967     {
 968         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 969         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 970     }
 971     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 972     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 973     if( h->param.i_bframe <= 1 )
 974         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 975     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 976     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
 977     if( !h->param.i_bframe )
 978     {
 979         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 980         h->param.analyse.i_direct_mv_pred = 0;
 981         h->param.analyse.b_weighted_bipred = 0;
 982         h->param.b_open_gop = 0;
 983     }
 984     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 985     {
 986         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 987         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 988     }
 989     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 990     {
 991         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 992         h->param.i_frame_reference = 1;
 993         h->param.i_dpb_size = 1;
 994     }
 995     if( h->param.b_intra_refresh && h->param.b_open_gop )
 996     {
 997         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 998         h->param.b_open_gop = 0;
 999     }
1000     if( !h->param.i_fps_num || !h->param.i_fps_den )
1001     {
1002         h->param.i_fps_num = 25;
1003         h->param.i_fps_den = 1;
1004     }
1005     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
1006     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
1007         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
1008     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
1009     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
1010     {
1011         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
1012         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
1013         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
1014     }
1015
1016     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
1017     {
1018         h->param.i_timebase_num = h->param.i_fps_den;
1019         h->param.i_timebase_den = h->param.i_fps_num;
1020     }
1021
1022     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
1023     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
1024         h->param.rc.b_mb_tree = 0;
1025     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
1026         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
1027     {
1028         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
1029         h->param.rc.b_mb_tree = 0;
1030     }
1031     if( b_open && h->param.rc.b_stat_read )
1032         h->param.rc.i_lookahead = 0;
1033 #if HAVE_THREAD
1034     if( h->param.i_sync_lookahead < 0 )
1035         h->param.i_sync_lookahead = h->param.i_bframe + 1;
1036     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
1037     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
1038         h->param.i_sync_lookahead = 0;
1039 #else
1040     h->param.i_sync_lookahead = 0;
1041 #endif
1042
1043     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
1044     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
1045     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1046     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1047
1048     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1049
1050     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1051         h->param.i_cqm_preset = X264_CQM_FLAT;
1052
1053     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1054         h->param.analyse.i_me_method > X264_ME_TESA )
1055         h->param.analyse.i_me_method = X264_ME_HEX;
1056     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1057     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1058         h->param.analyse.i_me_range = 16;
1059     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1060         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1061         h->param.analyse.i_me_method = X264_ME_ESA;
1062     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1063     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1064                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1065     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1066     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1067         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1068     if( !h->param.analyse.b_transform_8x8 )
1069     {
1070         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1071         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1072     }
1073     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1074     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 3 );
1075     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1076     if( h->param.rc.f_aq_strength == 0 )
1077         h->param.rc.i_aq_mode = 0;
1078
1079     if( h->param.i_log_level < X264_LOG_INFO )
1080     {
1081         h->param.analyse.b_psnr = 0;
1082         h->param.analyse.b_ssim = 0;
1083     }
1084     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1085     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1086     {
1087         char *s = NULL;
1088
1089         if( h->param.analyse.b_psy )
1090         {
1091             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1092             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1093         }
1094         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1095         {
1096             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1097             s = "ssim";
1098         }
1099         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1100         {
1101             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1102             s = "psnr";
1103         }
1104         if( s )
1105             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1106     }
1107
1108     if( !h->param.analyse.b_psy )
1109     {
1110         h->param.analyse.f_psy_rd = 0;
1111         h->param.analyse.f_psy_trellis = 0;
1112     }
1113     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1114     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1115     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1116     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1117     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1118     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1119     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1120         h->param.analyse.i_chroma_qp_offset += 6;
1121     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1122     /* so we lower the chroma QP offset to compensate */
1123     if( b_open && h->mb.i_psy_rd && !h->param.i_avcintra_class )
1124         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1125     /* Psy trellis has a similar effect. */
1126     if( b_open && h->mb.i_psy_trellis && !h->param.i_avcintra_class )
1127         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1128     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1129     /* MB-tree requires AQ to be on, even if the strength is zero. */
1130     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1131     {
1132         h->param.rc.i_aq_mode = 1;
1133         h->param.rc.f_aq_strength = 0;
1134     }
1135     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1136     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1137         h->param.analyse.i_subpel_refine = 9;
1138
1139     {
1140         const x264_level_t *l = x264_levels;
1141         if( h->param.i_level_idc < 0 )
1142         {
1143             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1144             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1145                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1146             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1147             do h->param.i_level_idc = l->level_idc;
1148                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1149             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1150         }
1151         else
1152         {
1153             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1154                 l++;
1155             if( l->level_idc == 0 )
1156             {
1157                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1158                 return -1;
1159             }
1160         }
1161         if( h->param.analyse.i_mv_range <= 0 )
1162             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1163         else
1164             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1165     }
1166
1167     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1168
1169     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1170     {
1171         if( h->param.b_sliced_threads )
1172             h->param.i_lookahead_threads = h->param.i_threads;
1173         else
1174         {
1175             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1176              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1177              * try to guess at this sort of case.
1178              *
1179              * Tuned by a little bit of real encoding with the various presets. */
1180             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1181             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1182             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1183
1184             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1185             static const uint8_t lookahead_thread_div[2][5][4] =
1186             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1187              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1188
1189             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1190             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1191              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1192              * pretty much arbitrarily. */
1193             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1194         }
1195     }
1196     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1197
1198     if( PARAM_INTERLACED )
1199     {
1200         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1201         {
1202             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1203             h->param.analyse.i_me_method = X264_ME_UMH;
1204         }
1205         if( h->param.analyse.i_weighted_pred > 0 )
1206         {
1207             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1208             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1209         }
1210     }
1211
1212     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1213         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1214
1215     if( h->i_thread_frames > 1 )
1216     {
1217         int r = h->param.analyse.i_mv_range_thread;
1218         int r2;
1219         if( r <= 0 )
1220         {
1221             // half of the available space is reserved and divided evenly among the threads,
1222             // the rest is allocated to whichever thread is far enough ahead to use it.
1223             // reserving more space increases quality for some videos, but costs more time
1224             // in thread synchronization.
1225             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1226             r = max_range / 2;
1227         }
1228         r = X264_MAX( r, h->param.analyse.i_me_range );
1229         r = X264_MIN( r, h->param.analyse.i_mv_range );
1230         // round up to use the whole mb row
1231         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1232         if( r2 < r )
1233             r2 += 16;
1234         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1235         h->param.analyse.i_mv_range_thread = r2;
1236     }
1237
1238     if( h->param.rc.f_rate_tolerance < 0 )
1239         h->param.rc.f_rate_tolerance = 0;
1240     if( h->param.rc.f_qblur < 0 )
1241         h->param.rc.f_qblur = 0;
1242     if( h->param.rc.f_complexity_blur < 0 )
1243         h->param.rc.f_complexity_blur = 0;
1244
1245     h->param.i_sps_id &= 31;
1246
1247     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1248
1249     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1250     {
1251         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1252         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1253     }
1254
1255     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1256        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1257     {
1258         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1259         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1260     }
1261
1262     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1263         h->param.rc.b_filler = 1;
1264
1265     /* ensure the booleans are 0 or 1 so they can be used in math */
1266 #define BOOLIFY(x) h->param.x = !!h->param.x
1267     BOOLIFY( b_cabac );
1268     BOOLIFY( b_constrained_intra );
1269     BOOLIFY( b_deblocking_filter );
1270     BOOLIFY( b_deterministic );
1271     BOOLIFY( b_sliced_threads );
1272     BOOLIFY( b_interlaced );
1273     BOOLIFY( b_intra_refresh );
1274     BOOLIFY( b_aud );
1275     BOOLIFY( b_repeat_headers );
1276     BOOLIFY( b_annexb );
1277     BOOLIFY( b_vfr_input );
1278     BOOLIFY( b_pulldown );
1279     BOOLIFY( b_tff );
1280     BOOLIFY( b_pic_struct );
1281     BOOLIFY( b_fake_interlaced );
1282     BOOLIFY( b_open_gop );
1283     BOOLIFY( b_bluray_compat );
1284     BOOLIFY( b_stitchable );
1285     BOOLIFY( b_full_recon );
1286     BOOLIFY( b_opencl );
1287     BOOLIFY( analyse.b_transform_8x8 );
1288     BOOLIFY( analyse.b_weighted_bipred );
1289     BOOLIFY( analyse.b_chroma_me );
1290     BOOLIFY( analyse.b_mixed_references );
1291     BOOLIFY( analyse.b_fast_pskip );
1292     BOOLIFY( analyse.b_dct_decimate );
1293     BOOLIFY( analyse.b_psy );
1294     BOOLIFY( analyse.b_psnr );
1295     BOOLIFY( analyse.b_ssim );
1296     BOOLIFY( rc.b_stat_write );
1297     BOOLIFY( rc.b_stat_read );
1298     BOOLIFY( rc.b_mb_tree );
1299     BOOLIFY( rc.b_filler );
1300 #undef BOOLIFY
1301
1302     return 0;
1303 }
1304
1305 static void mbcmp_init( x264_t *h )
1306 {
1307     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1308     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1309     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1310     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1311     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1312     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1313     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1314     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1315     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1316                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1317     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1318                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1319     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1320     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1321     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1322     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1323 }
1324
1325 static void chroma_dsp_init( x264_t *h )
1326 {
1327     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1328
1329     switch( CHROMA_FORMAT )
1330     {
1331         case CHROMA_420:
1332             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1333             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1334             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1335             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1336             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1337             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1338             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1339             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1340             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1341             break;
1342         case CHROMA_422:
1343             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1344             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1345             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1346             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1347             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1348             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1349             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1350             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1351             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1352             break;
1353         case CHROMA_444:
1354             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1355             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1356             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1357             break;
1358     }
1359 }
1360
1361 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1362 {
1363     /* VUI */
1364     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1365     {
1366         uint32_t i_w = param->vui.i_sar_width;
1367         uint32_t i_h = param->vui.i_sar_height;
1368         uint32_t old_w = h->param.vui.i_sar_width;
1369         uint32_t old_h = h->param.vui.i_sar_height;
1370
1371         x264_reduce_fraction( &i_w, &i_h );
1372
1373         while( i_w > 65535 || i_h > 65535 )
1374         {
1375             i_w /= 2;
1376             i_h /= 2;
1377         }
1378
1379         x264_reduce_fraction( &i_w, &i_h );
1380
1381         if( i_w != old_w || i_h != old_h || initial )
1382         {
1383             h->param.vui.i_sar_width = 0;
1384             h->param.vui.i_sar_height = 0;
1385             if( i_w == 0 || i_h == 0 )
1386                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1387             else
1388             {
1389                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1390                 h->param.vui.i_sar_width = i_w;
1391                 h->param.vui.i_sar_height = i_h;
1392             }
1393         }
1394     }
1395 }
1396
1397 /****************************************************************************
1398  * x264_encoder_open:
1399  ****************************************************************************/
1400 x264_t *x264_encoder_open( x264_param_t *param )
1401 {
1402     x264_t *h;
1403     char buf[1000], *p;
1404     int qp, i_slicetype_length;
1405
1406     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1407
1408     /* Create a copy of param */
1409     memcpy( &h->param, param, sizeof(x264_param_t) );
1410
1411     if( param->param_free )
1412         param->param_free( param );
1413
1414 #if HAVE_INTEL_DISPATCHER
1415     x264_intel_dispatcher_override();
1416 #endif
1417
1418     if( x264_threading_init() )
1419     {
1420         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1421         goto fail;
1422     }
1423
1424     if( x264_validate_parameters( h, 1 ) < 0 )
1425         goto fail;
1426
1427     if( h->param.psz_cqm_file )
1428         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1429             goto fail;
1430
1431     if( h->param.rc.psz_stat_out )
1432         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1433     if( h->param.rc.psz_stat_in )
1434         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1435
1436     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1437     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1438
1439     /* Init x264_t */
1440     h->i_frame = -1;
1441     h->i_frame_num = 0;
1442
1443     if( h->param.i_avcintra_class )
1444         h->i_idr_pic_id = 5;
1445     else
1446         h->i_idr_pic_id = 0;
1447
1448     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1449     {
1450         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1451         goto fail;
1452     }
1453
1454     x264_set_aspect_ratio( h, &h->param, 1 );
1455
1456     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1457     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1458
1459     x264_validate_levels( h, 1 );
1460
1461     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1462
1463     if( x264_cqm_init( h ) < 0 )
1464         goto fail;
1465
1466     h->mb.i_mb_width = h->sps->i_mb_width;
1467     h->mb.i_mb_height = h->sps->i_mb_height;
1468     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1469
1470     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1471     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1472
1473     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1474      * vectors during prediction, resulting in hpel mvs.
1475      * The chosen solution is to make MBAFF non-adaptive in this case. */
1476     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1477
1478     /* Init frames. */
1479     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1480         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1481     else
1482         h->frames.i_delay = h->param.i_bframe;
1483     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1484         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1485     i_slicetype_length = h->frames.i_delay;
1486     h->frames.i_delay += h->i_thread_frames - 1;
1487     h->frames.i_delay += h->param.i_sync_lookahead;
1488     h->frames.i_delay += h->param.b_vfr_input;
1489     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1490
1491     h->frames.i_max_ref0 = h->param.i_frame_reference;
1492     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1493     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1494     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1495         && ( h->param.rc.i_rc_method == X264_RC_ABR
1496           || h->param.rc.i_rc_method == X264_RC_CRF
1497           || h->param.i_bframe_adaptive
1498           || h->param.i_scenecut_threshold
1499           || h->param.rc.b_mb_tree
1500           || h->param.analyse.i_weighted_pred );
1501     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1502     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1503
1504     h->frames.i_last_idr =
1505     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1506     h->frames.i_input    = 0;
1507     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1508     h->frames.i_poc_last_open_gop = -1;
1509
1510     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1511     /* Allocate room for max refs plus a few extra just in case. */
1512     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1513     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1514                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1515     if( h->param.analyse.i_weighted_pred > 0 )
1516         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1517     h->i_ref[0] = h->i_ref[1] = 0;
1518     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1519     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);
1520     h->i_disp_fields_last_frame = -1;
1521     x264_rdo_init();
1522
1523     /* init CPU functions */
1524     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1525     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1526     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1527     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1528     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1529     x264_pixel_init( h->param.cpu, &h->pixf );
1530     x264_dct_init( h->param.cpu, &h->dctf );
1531     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1532     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1533     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1534     x264_quant_init( h, h->param.cpu, &h->quantf );
1535     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1536     x264_bitstream_init( h->param.cpu, &h->bsf );
1537     if( h->param.b_cabac )
1538         x264_cabac_init( h );
1539     else
1540         x264_stack_align( x264_cavlc_init, h );
1541
1542     mbcmp_init( h );
1543     chroma_dsp_init( h );
1544
1545     p = buf + sprintf( buf, "using cpu capabilities:" );
1546     for( int i = 0; x264_cpu_names[i].flags; i++ )
1547     {
1548         if( !strcmp(x264_cpu_names[i].name, "SSE")
1549             && h->param.cpu & (X264_CPU_SSE2) )
1550             continue;
1551         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1552             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1553             continue;
1554         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1555             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1556             continue;
1557         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1558             && (h->param.cpu & X264_CPU_SSE42) )
1559             continue;
1560         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1561             && (h->param.cpu & X264_CPU_BMI2) )
1562             continue;
1563         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1564             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1565             p += sprintf( p, " %s", x264_cpu_names[i].name );
1566     }
1567     if( !h->param.cpu )
1568         p += sprintf( p, " none!" );
1569     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1570
1571     float *logs = x264_analyse_prepare_costs( h );
1572     if( !logs )
1573         goto fail;
1574     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1575         if( x264_analyse_init_costs( h, logs, qp ) )
1576             goto fail;
1577     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1578         goto fail;
1579     x264_free( logs );
1580
1581     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1582     /* Checks for known miscompilation issues. */
1583     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1584     {
1585         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1586         goto fail;
1587     }
1588
1589     /* Must be volatile or else GCC will optimize it out. */
1590     volatile int temp = 392;
1591     if( x264_clz( temp ) != 23 )
1592     {
1593         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1594 #if ARCH_X86 || ARCH_X86_64
1595         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1596         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1597 #endif
1598         goto fail;
1599     }
1600
1601     h->out.i_nal = 0;
1602     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1603         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1604           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1605
1606     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1607     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1608
1609     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1610
1611     if( h->param.i_threads > 1 &&
1612         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1613         goto fail;
1614     if( h->param.i_lookahead_threads > 1 &&
1615         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1616         goto fail;
1617
1618 #if HAVE_OPENCL
1619     if( h->param.b_opencl )
1620     {
1621         h->opencl.ocl = x264_opencl_load_library();
1622         if( !h->opencl.ocl )
1623         {
1624             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1625             h->param.b_opencl = 0;
1626         }
1627     }
1628 #endif
1629
1630     h->thread[0] = h;
1631     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1632         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1633     if( h->param.i_lookahead_threads > 1 )
1634         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1635         {
1636             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1637             *h->lookahead_thread[i] = *h;
1638         }
1639     *h->reconfig_h = *h;
1640
1641     for( int i = 0; i < h->param.i_threads; i++ )
1642     {
1643         int init_nal_count = h->param.i_slice_count + 3;
1644         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1645         if( i > 0 )
1646             *h->thread[i] = *h;
1647
1648         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1649             goto fail;
1650         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1651             goto fail;
1652
1653         if( allocate_threadlocal_data )
1654         {
1655             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1656             if( !h->thread[i]->fdec )
1657                 goto fail;
1658         }
1659         else
1660             h->thread[i]->fdec = h->thread[0]->fdec;
1661
1662         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1663         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1664         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1665         h->thread[i]->out.i_nals_allocated = init_nal_count;
1666
1667         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1668             goto fail;
1669     }
1670
1671 #if HAVE_OPENCL
1672     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1673         h->param.b_opencl = 0;
1674 #endif
1675
1676     if( x264_lookahead_init( h, i_slicetype_length ) )
1677         goto fail;
1678
1679     for( int i = 0; i < h->param.i_threads; i++ )
1680         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1681             goto fail;
1682
1683     if( x264_ratecontrol_new( h ) < 0 )
1684         goto fail;
1685
1686     if( h->param.i_nal_hrd )
1687     {
1688         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1689         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1690     }
1691
1692     if( h->param.psz_dump_yuv )
1693     {
1694         /* create or truncate the reconstructed video file */
1695         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1696         if( !f )
1697         {
1698             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1699             goto fail;
1700         }
1701         else if( !x264_is_regular_file( f ) )
1702         {
1703             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1704             fclose( f );
1705             goto fail;
1706         }
1707         fclose( f );
1708     }
1709
1710     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1711                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1712                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1713                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1714                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1715                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1716     char level[4];
1717     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1718     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1719         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1720         strcpy( level, "1b" );
1721
1722     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1723     {
1724         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1725             profile, level );
1726     }
1727     else
1728     {
1729         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1730         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1731             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1732     }
1733
1734     return h;
1735 fail:
1736     x264_free( h );
1737     return NULL;
1738 }
1739
1740 /****************************************************************************/
1741 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1742 {
1743     *rc_reconfig = 0;
1744     x264_set_aspect_ratio( h, param, 0 );
1745 #define COPY(var) h->param.var = param->var
1746     COPY( i_frame_reference ); // but never uses more refs than initially specified
1747     COPY( i_bframe_bias );
1748     if( h->param.i_scenecut_threshold )
1749         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1750     COPY( b_deblocking_filter );
1751     COPY( i_deblocking_filter_alphac0 );
1752     COPY( i_deblocking_filter_beta );
1753     COPY( i_frame_packing );
1754     COPY( analyse.inter );
1755     COPY( analyse.intra );
1756     COPY( analyse.i_direct_mv_pred );
1757     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1758     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1759         COPY( analyse.i_me_range );
1760     COPY( analyse.i_noise_reduction );
1761     /* We can't switch out of subme=0 during encoding. */
1762     if( h->param.analyse.i_subpel_refine )
1763         COPY( analyse.i_subpel_refine );
1764     COPY( analyse.i_trellis );
1765     COPY( analyse.b_chroma_me );
1766     COPY( analyse.b_dct_decimate );
1767     COPY( analyse.b_fast_pskip );
1768     COPY( analyse.b_mixed_references );
1769     COPY( analyse.f_psy_rd );
1770     COPY( analyse.f_psy_trellis );
1771     COPY( crop_rect );
1772     // can only twiddle these if they were enabled to begin with:
1773     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1774         COPY( analyse.i_me_method );
1775     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1776         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1777     if( h->pps->b_transform_8x8_mode )
1778         COPY( analyse.b_transform_8x8 );
1779     if( h->frames.i_max_ref1 > 1 )
1780         COPY( i_bframe_pyramid );
1781     COPY( i_slice_max_size );
1782     COPY( i_slice_max_mbs );
1783     COPY( i_slice_min_mbs );
1784     COPY( i_slice_count );
1785     COPY( i_slice_count_max );
1786     COPY( b_tff );
1787
1788     /* VBV can't be turned on if it wasn't on to begin with */
1789     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1790           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1791     {
1792         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1793         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1794         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1795         COPY( rc.i_vbv_max_bitrate );
1796         COPY( rc.i_vbv_buffer_size );
1797         COPY( rc.i_bitrate );
1798     }
1799     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1800     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1801     COPY( rc.f_rf_constant );
1802     COPY( rc.f_rf_constant_max );
1803 #undef COPY
1804
1805     return x264_validate_parameters( h, 0 );
1806 }
1807
1808 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1809 {
1810     int rc_reconfig;
1811     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1812
1813     mbcmp_init( h );
1814     if( !ret )
1815         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1816
1817     /* Supported reconfiguration options (1-pass only):
1818      * vbv-maxrate
1819      * vbv-bufsize
1820      * crf
1821      * bitrate (CBR only) */
1822     if( !ret && rc_reconfig )
1823         x264_ratecontrol_init_reconfigurable( h, 0 );
1824
1825     return ret;
1826 }
1827
1828 /****************************************************************************
1829  * x264_encoder_reconfig:
1830  ****************************************************************************/
1831 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1832 {
1833     h = h->thread[h->thread[0]->i_thread_phase];
1834     x264_param_t param_save = h->reconfig_h->param;
1835     h->reconfig_h->param = h->param;
1836
1837     int rc_reconfig;
1838     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1839     if( !ret )
1840         h->reconfig = 1;
1841     else
1842         h->reconfig_h->param = param_save;
1843
1844     return ret;
1845 }
1846
1847 /****************************************************************************
1848  * x264_encoder_parameters:
1849  ****************************************************************************/
1850 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1851 {
1852     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1853 }
1854
1855 /* internal usage */
1856 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1857 {
1858     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1859
1860     nal->i_ref_idc        = i_ref_idc;
1861     nal->i_type           = i_type;
1862     nal->b_long_startcode = 1;
1863
1864     nal->i_payload= 0;
1865     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1866     nal->i_padding= 0;
1867 }
1868
1869 /* if number of allocated nals is not enough, re-allocate a larger one. */
1870 static int x264_nal_check_buffer( x264_t *h )
1871 {
1872     if( h->out.i_nal >= h->out.i_nals_allocated )
1873     {
1874         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1875         if( !new_out )
1876             return -1;
1877         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1878         x264_free( h->out.nal );
1879         h->out.nal = new_out;
1880         h->out.i_nals_allocated *= 2;
1881     }
1882     return 0;
1883 }
1884
1885 static int x264_nal_end( x264_t *h )
1886 {
1887     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1888     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1889     nal->i_payload = end - nal->p_payload;
1890     /* Assembly implementation of nal_escape reads past the end of the input.
1891      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1892     memset( end, 0xff, 64 );
1893     if( h->param.nalu_process )
1894         h->param.nalu_process( h, nal, h->fenc->opaque );
1895     h->out.i_nal++;
1896
1897     return x264_nal_check_buffer( h );
1898 }
1899
1900 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1901                                            int previous_nal_size, int necessary_size )
1902 {
1903     if( h0->nal_buffer_size < necessary_size )
1904     {
1905         necessary_size *= 2;
1906         uint8_t *buf = x264_malloc( necessary_size );
1907         if( !buf )
1908             return -1;
1909         if( previous_nal_size )
1910             memcpy( buf, h0->nal_buffer, previous_nal_size );
1911
1912         intptr_t delta = buf - h0->nal_buffer;
1913         for( int i = 0; i < start; i++ )
1914             h->out.nal[i].p_payload += delta;
1915
1916         x264_free( h0->nal_buffer );
1917         h0->nal_buffer = buf;
1918         h0->nal_buffer_size = necessary_size;
1919     }
1920
1921     return 0;
1922 }
1923
1924 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1925 {
1926     x264_t *h0 = h->thread[0];
1927     int nal_size = 0, previous_nal_size = 0;
1928
1929     if( h->param.nalu_process )
1930     {
1931         for( int i = start; i < h->out.i_nal; i++ )
1932             nal_size += h->out.nal[i].i_payload;
1933         return nal_size;
1934     }
1935
1936     for( int i = 0; i < start; i++ )
1937         previous_nal_size += h->out.nal[i].i_payload;
1938
1939     for( int i = start; i < h->out.i_nal; i++ )
1940         nal_size += h->out.nal[i].i_payload;
1941
1942     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1943     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1944     for( int i = start; i < h->out.i_nal; i++ )
1945         necessary_size += h->out.nal[i].i_padding;
1946     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1947         return -1;
1948
1949     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1950
1951     for( int i = start; i < h->out.i_nal; i++ )
1952     {
1953         int old_payload_len = h->out.nal[i].i_payload;
1954         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1955                                          h->param.i_avcintra_class;
1956         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1957         nal_buffer += h->out.nal[i].i_payload;
1958         if( h->param.i_avcintra_class )
1959         {
1960             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1961             if( h->out.nal[i].i_padding > 0 )
1962             {
1963                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1964                 nal_buffer += h->out.nal[i].i_padding;
1965                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1966             }
1967             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1968         }
1969     }
1970
1971     x264_emms();
1972
1973     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1974 }
1975
1976 /****************************************************************************
1977  * x264_encoder_headers:
1978  ****************************************************************************/
1979 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1980 {
1981     int frame_size = 0;
1982     /* init bitstream context */
1983     h->out.i_nal = 0;
1984     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1985
1986     /* Write SEI, SPS and PPS. */
1987
1988     /* generate sequence parameters */
1989     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1990     x264_sps_write( &h->out.bs, h->sps );
1991     if( x264_nal_end( h ) )
1992         return -1;
1993
1994     /* generate picture parameters */
1995     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1996     x264_pps_write( &h->out.bs, h->sps, h->pps );
1997     if( x264_nal_end( h ) )
1998         return -1;
1999
2000     /* identify ourselves */
2001     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
2002     if( x264_sei_version_write( h, &h->out.bs ) )
2003         return -1;
2004     if( x264_nal_end( h ) )
2005         return -1;
2006
2007     frame_size = x264_encoder_encapsulate_nals( h, 0 );
2008     if( frame_size < 0 )
2009         return -1;
2010
2011     /* now set output*/
2012     *pi_nal = h->out.i_nal;
2013     *pp_nal = &h->out.nal[0];
2014     h->out.i_nal = 0;
2015
2016     return frame_size;
2017 }
2018
2019 /* Check to see whether we have chosen a reference list ordering different
2020  * from the standard's default. */
2021 static inline void x264_reference_check_reorder( x264_t *h )
2022 {
2023     /* The reorder check doesn't check for missing frames, so just
2024      * force a reorder if one of the reference list is corrupt. */
2025     for( int i = 0; h->frames.reference[i]; i++ )
2026         if( h->frames.reference[i]->b_corrupt )
2027         {
2028             h->b_ref_reorder[0] = 1;
2029             return;
2030         }
2031     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
2032         for( int i = 0; i < h->i_ref[list] - 1; i++ )
2033         {
2034             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
2035             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
2036             /* P and B-frames use different default orders. */
2037             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
2038             {
2039                 h->b_ref_reorder[list] = 1;
2040                 return;
2041             }
2042         }
2043 }
2044
2045 /* return -1 on failure, else return the index of the new reference frame */
2046 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2047 {
2048     int i = h->i_ref[0];
2049     int j = 1;
2050     x264_frame_t *newframe;
2051     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2052         return -1;
2053
2054     //Duplication is only used in X264_WEIGHTP_SMART
2055     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2056         return -1;
2057
2058     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2059      * With high bit depth, it's not worth doing, so turn it off except in the case of
2060      * unweighted dupes. */
2061     if( BIT_DEPTH > 8 && w != x264_weight_none )
2062         return -1;
2063
2064     newframe = x264_frame_pop_blank_unused( h );
2065     if( !newframe )
2066         return -1;
2067
2068     //FIXME: probably don't need to copy everything
2069     *newframe = *h->fref[0][i_ref];
2070     newframe->i_reference_count = 1;
2071     newframe->orig = h->fref[0][i_ref];
2072     newframe->b_duplicate = 1;
2073     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2074
2075     /* shift the frames to make space for the dupe. */
2076     h->b_ref_reorder[0] = 1;
2077     if( h->i_ref[0] < X264_REF_MAX )
2078         ++h->i_ref[0];
2079     h->fref[0][X264_REF_MAX-1] = NULL;
2080     x264_frame_unshift( &h->fref[0][j], newframe );
2081
2082     return j;
2083 }
2084
2085 static void x264_weighted_pred_init( x264_t *h )
2086 {
2087     /* for now no analysis and set all weights to nothing */
2088     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2089         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2090
2091     // FIXME: This only supports weighting of one reference frame
2092     // and duplicates of that frame.
2093     h->fenc->i_lines_weighted = 0;
2094
2095     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2096         for( int i = 0; i < 3; i++ )
2097             h->sh.weight[i_ref][i].weightfn = NULL;
2098
2099
2100     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2101         return;
2102
2103     int i_padv = PADV << PARAM_INTERLACED;
2104     int denom = -1;
2105     int weightplane[2] = { 0, 0 };
2106     int buffer_next = 0;
2107     for( int i = 0; i < 3; i++ )
2108     {
2109         for( int j = 0; j < h->i_ref[0]; j++ )
2110         {
2111             if( h->fenc->weight[j][i].weightfn )
2112             {
2113                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2114                 // if weight is useless, don't write it to stream
2115                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2116                     h->sh.weight[j][i].weightfn = NULL;
2117                 else
2118                 {
2119                     if( !weightplane[!!i] )
2120                     {
2121                         weightplane[!!i] = 1;
2122                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2123                         assert( x264_clip3( denom, 0, 7 ) == denom );
2124                     }
2125
2126                     assert( h->sh.weight[j][i].i_denom == denom );
2127                     if( !i )
2128                     {
2129                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2130                         //scale full resolution frame
2131                         if( h->param.i_threads == 1 )
2132                         {
2133                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2134                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2135                             int stride = h->fenc->i_stride[0];
2136                             int width = h->fenc->i_width[0] + PADH*2;
2137                             int height = h->fenc->i_lines[0] + i_padv*2;
2138                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2139                             h->fenc->i_lines_weighted = height;
2140                         }
2141                     }
2142                 }
2143             }
2144         }
2145     }
2146
2147     if( weightplane[1] )
2148         for( int i = 0; i < h->i_ref[0]; i++ )
2149         {
2150             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2151             {
2152                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2153                 h->sh.weight[i][2].i_offset = 0;
2154             }
2155             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2156             {
2157                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2158                 h->sh.weight[i][1].i_offset = 0;
2159             }
2160         }
2161
2162     if( !weightplane[0] )
2163         h->sh.weight[0][0].i_denom = 0;
2164     if( !weightplane[1] )
2165         h->sh.weight[0][1].i_denom = 0;
2166     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2167 }
2168
2169 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2170 {
2171     if( h->param.i_frame_packing == 5 )
2172         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2173                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2174     else
2175         return abs(h->fenc->i_frame - frame->i_frame);
2176 }
2177
2178 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2179 {
2180     int b_ok;
2181
2182     /* build ref list 0/1 */
2183     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2184     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2185     if( h->sh.i_type == SLICE_TYPE_I )
2186         return;
2187
2188     for( int i = 0; h->frames.reference[i]; i++ )
2189     {
2190         if( h->frames.reference[i]->b_corrupt )
2191             continue;
2192         if( h->frames.reference[i]->i_poc < i_poc )
2193             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2194         else if( h->frames.reference[i]->i_poc > i_poc )
2195             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2196     }
2197
2198     if( h->sh.i_mmco_remove_from_end )
2199     {
2200         /* Order ref0 for MMCO remove */
2201         do
2202         {
2203             b_ok = 1;
2204             for( int i = 0; i < h->i_ref[0] - 1; i++ )
2205             {
2206                 if( h->fref[0][i]->i_frame < h->fref[0][i+1]->i_frame )
2207                 {
2208                     XCHG( x264_frame_t*, h->fref[0][i], h->fref[0][i+1] );
2209                     b_ok = 0;
2210                     break;
2211                 }
2212             }
2213         } while( !b_ok );
2214
2215         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2216         {
2217             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2218             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2219             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2220         }
2221     }
2222
2223     /* Order reference lists by distance from the current frame. */
2224     for( int list = 0; list < 2; list++ )
2225     {
2226         h->fref_nearest[list] = h->fref[list][0];
2227         do
2228         {
2229             b_ok = 1;
2230             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2231             {
2232                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2233                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2234                     h->fref_nearest[list] = h->fref[list][i+1];
2235                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2236                 {
2237                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2238                     b_ok = 0;
2239                     break;
2240                 }
2241             }
2242         } while( !b_ok );
2243     }
2244
2245     x264_reference_check_reorder( h );
2246
2247     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2248     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2249     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2250
2251     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2252     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2253         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2254
2255     /* add duplicates */
2256     if( h->fenc->i_type == X264_TYPE_P )
2257     {
2258         int idx = -1;
2259         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2260         {
2261             x264_weight_t w[3];
2262             w[1].weightfn = w[2].weightfn = NULL;
2263             if( h->param.rc.b_stat_read )
2264                 x264_ratecontrol_set_weights( h, h->fenc );
2265
2266             if( !h->fenc->weight[0][0].weightfn )
2267             {
2268                 h->fenc->weight[0][0].i_denom = 0;
2269                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2270                 idx = x264_weighted_reference_duplicate( h, 0, w );
2271             }
2272             else
2273             {
2274                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2275                 {
2276                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2277                 }
2278                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2279                 if( h->fenc->weight[0][0].i_offset > -128 )
2280                 {
2281                     w[0] = h->fenc->weight[0][0];
2282                     w[0].i_offset--;
2283                     h->mc.weight_cache( h, &w[0] );
2284                     idx = x264_weighted_reference_duplicate( h, 0, w );
2285                 }
2286             }
2287         }
2288         h->mb.ref_blind_dupe = idx;
2289     }
2290
2291     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2292     h->mb.pic.i_fref[0] = h->i_ref[0];
2293     h->mb.pic.i_fref[1] = h->i_ref[1];
2294 }
2295
2296 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2297 {
2298     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2299     int b_hpel = h->fdec->b_kept_as_ref;
2300     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2301     int b_end = mb_y == h->i_threadslice_end;
2302     int b_measure_quality = 1;
2303     int min_y = mb_y - (1 << SLICE_MBAFF);
2304     int b_start = min_y == h->i_threadslice_start;
2305     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2306      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2307     int minpix_y = min_y*16 - 4 * !b_start;
2308     int maxpix_y = mb_y*16 - 4 * !b_end;
2309     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2310     if( h->param.b_sliced_threads )
2311     {
2312         switch( pass )
2313         {
2314             /* During encode: only do deblock if asked for */
2315             default:
2316             case 0:
2317                 b_deblock &= h->param.b_full_recon;
2318                 b_hpel = 0;
2319                 break;
2320             /* During post-encode pass: do deblock if not done yet, do hpel for all
2321              * rows except those between slices. */
2322             case 1:
2323                 b_deblock &= !h->param.b_full_recon;
2324                 b_hpel &= !(b_start && min_y > 0);
2325                 b_measure_quality = 0;
2326                 break;
2327             /* Final pass: do the rows between slices in sequence. */
2328             case 2:
2329                 b_deblock = 0;
2330                 b_measure_quality = 0;
2331                 break;
2332         }
2333     }
2334     if( mb_y & SLICE_MBAFF )
2335         return;
2336     if( min_y < h->i_threadslice_start )
2337         return;
2338
2339     if( b_deblock )
2340         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2341             x264_frame_deblock_row( h, y );
2342
2343     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2344      * but the actual image data is equivalent. For now, maintain this
2345      * consistency by copying deblocked pixels between planes. */
2346     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2347         for( int p = 0; p < h->fdec->i_plane; p++ )
2348             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2349                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2350                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2351                         h->mb.i_mb_width*16*sizeof(pixel) );
2352
2353     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2354         x264_frame_expand_border( h, h->fdec, min_y );
2355     if( b_hpel )
2356     {
2357         int end = mb_y == h->mb.i_mb_height;
2358         /* Can't do hpel until the previous slice is done encoding. */
2359         if( h->param.analyse.i_subpel_refine )
2360         {
2361             x264_frame_filter( h, h->fdec, min_y, end );
2362             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2363         }
2364     }
2365
2366     if( SLICE_MBAFF && pass == 0 )
2367         for( int i = 0; i < 3; i++ )
2368         {
2369             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2370             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2371         }
2372
2373     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2374         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2375
2376     if( b_measure_quality )
2377     {
2378         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2379         if( h->param.analyse.b_psnr )
2380         {
2381             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2382                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2383                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2384                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2385                     h->param.i_width, maxpix_y-minpix_y );
2386             if( !CHROMA444 )
2387             {
2388                 uint64_t ssd_u, ssd_v;
2389                 int v_shift = CHROMA_V_SHIFT;
2390                 x264_pixel_ssd_nv12( &h->pixf,
2391                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2392                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2393                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2394                 h->stat.frame.i_ssd[1] += ssd_u;
2395                 h->stat.frame.i_ssd[2] += ssd_v;
2396             }
2397         }
2398
2399         if( h->param.analyse.b_ssim )
2400         {
2401             int ssim_cnt;
2402             x264_emms();
2403             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2404              * and overlap by 4 */
2405             minpix_y += b_start ? 2 : -6;
2406             h->stat.frame.f_ssim +=
2407                 x264_pixel_ssim_wxh( &h->pixf,
2408                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2409                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2410                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2411             h->stat.frame.i_ssim_cnt += ssim_cnt;
2412         }
2413     }
2414 }
2415
2416 static inline int x264_reference_update( x264_t *h )
2417 {
2418     if( !h->fdec->b_kept_as_ref )
2419     {
2420         if( h->i_thread_frames > 1 )
2421         {
2422             x264_frame_push_unused( h, h->fdec );
2423             h->fdec = x264_frame_pop_unused( h, 1 );
2424             if( !h->fdec )
2425                 return -1;
2426         }
2427         return 0;
2428     }
2429
2430     /* apply mmco from previous frame. */
2431     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2432         for( int j = 0; h->frames.reference[j]; j++ )
2433             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2434                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2435
2436     /* move frame in the buffer */
2437     x264_frame_push( h->frames.reference, h->fdec );
2438     if( h->frames.reference[h->sps->i_num_ref_frames] )
2439         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2440     h->fdec = x264_frame_pop_unused( h, 1 );
2441     if( !h->fdec )
2442         return -1;
2443     return 0;
2444 }
2445
2446 static inline void x264_reference_reset( x264_t *h )
2447 {
2448     while( h->frames.reference[0] )
2449         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2450     h->fdec->i_poc =
2451     h->fenc->i_poc = 0;
2452 }
2453
2454 static inline void x264_reference_hierarchy_reset( x264_t *h )
2455 {
2456     int ref;
2457     int b_hasdelayframe = 0;
2458
2459     /* look for delay frames -- chain must only contain frames that are disposable */
2460     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2461         b_hasdelayframe |= h->frames.current[i]->i_coded
2462                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2463
2464     /* This function must handle b-pyramid and clear frames for open-gop */
2465     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2466         return;
2467
2468     /* Remove last BREF. There will never be old BREFs in the
2469      * dpb during a BREF decode when pyramid == STRICT */
2470     for( ref = 0; h->frames.reference[ref]; ref++ )
2471     {
2472         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2473             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2474             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2475             && h->sh.i_type != SLICE_TYPE_B ) )
2476         {
2477             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2478             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2479             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2480             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2481             h->b_ref_reorder[0] = 1;
2482             ref--;
2483         }
2484     }
2485
2486     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2487     if( h->param.i_bframe_pyramid )
2488         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2489 }
2490
2491 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2492 {
2493     /* ------------------------ Create slice header  ----------------------- */
2494     if( i_nal_type == NAL_SLICE_IDR )
2495     {
2496         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2497
2498         /* alternate id */
2499         if( h->param.i_avcintra_class )
2500         {
2501             switch( h->i_idr_pic_id )
2502             {
2503                 case 5:
2504                     h->i_idr_pic_id = 3;
2505                     break;
2506                 case 3:
2507                     h->i_idr_pic_id = 4;
2508                     break;
2509                 case 4:
2510                 default:
2511                     h->i_idr_pic_id = 5;
2512                     break;
2513             }
2514         }
2515         else
2516             h->i_idr_pic_id ^= 1;
2517     }
2518     else
2519     {
2520         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2521
2522         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2523         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2524         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2525             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2526         {
2527             h->sh.b_num_ref_idx_override = 1;
2528         }
2529     }
2530
2531     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2532     {
2533         h->b_sh_backup = 1;
2534         h->sh_backup = h->sh;
2535     }
2536
2537     h->fdec->i_frame_num = h->sh.i_frame_num;
2538
2539     if( h->sps->i_poc_type == 0 )
2540     {
2541         h->sh.i_poc = h->fdec->i_poc;
2542         if( PARAM_INTERLACED )
2543         {
2544             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2545             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2546         }
2547         else
2548             h->sh.i_delta_poc_bottom = 0;
2549         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2550         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2551     }
2552     else
2553     {
2554         /* Nothing to do ? */
2555     }
2556
2557     x264_macroblock_slice_init( h );
2558 }
2559
2560 typedef struct
2561 {
2562     int skip;
2563     uint8_t cabac_prevbyte;
2564     bs_t bs;
2565     x264_cabac_t cabac;
2566     x264_frame_stat_t stat;
2567     int last_qp;
2568     int last_dqp;
2569     int field_decoding_flag;
2570 } x264_bs_bak_t;
2571
2572 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2573 {
2574     if( full )
2575     {
2576         bak->stat = h->stat.frame;
2577         bak->last_qp = h->mb.i_last_qp;
2578         bak->last_dqp = h->mb.i_last_dqp;
2579         bak->field_decoding_flag = h->mb.field_decoding_flag;
2580     }
2581     else
2582     {
2583         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2584         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2585     }
2586     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2587      * encoder has no context dependency and in this case, a slice is ended (and
2588      * thus the content of all contexts are thrown away). */
2589     if( h->param.b_cabac )
2590     {
2591         if( full )
2592             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2593         else
2594             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2595         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2596          * backed up. */
2597         bak->cabac_prevbyte = h->cabac.p[-1];
2598     }
2599     else
2600     {
2601         bak->bs = h->out.bs;
2602         bak->skip = i_skip;
2603     }
2604 }
2605
2606 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2607 {
2608     if( full )
2609     {
2610         h->stat.frame = bak->stat;
2611         h->mb.i_last_qp = bak->last_qp;
2612         h->mb.i_last_dqp = bak->last_dqp;
2613         h->mb.field_decoding_flag = bak->field_decoding_flag;
2614     }
2615     else
2616     {
2617         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2618         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2619     }
2620     if( h->param.b_cabac )
2621     {
2622         if( full )
2623             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2624         else
2625             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2626         h->cabac.p[-1] = bak->cabac_prevbyte;
2627     }
2628     else
2629     {
2630         h->out.bs = bak->bs;
2631         *skip = bak->skip;
2632     }
2633 }
2634
2635 static intptr_t x264_slice_write( x264_t *h )
2636 {
2637     int i_skip;
2638     int mb_xy, i_mb_x, i_mb_y;
2639     /* NALUs other than the first use a 3-byte startcode.
2640      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2641      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2642      * other inaccuracies. */
2643     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2644     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2645     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2646     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2647     int starting_bits = bs_pos(&h->out.bs);
2648     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2649     int b_hpel = h->fdec->b_kept_as_ref;
2650     int orig_last_mb = h->sh.i_last_mb;
2651     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2652     uint8_t *last_emu_check;
2653 #define BS_BAK_SLICE_MAX_SIZE 0
2654 #define BS_BAK_CAVLC_OVERFLOW 1
2655 #define BS_BAK_SLICE_MIN_MBS  2
2656 #define BS_BAK_ROW_VBV        3
2657     x264_bs_bak_t bs_bak[4];
2658     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2659     bs_realign( &h->out.bs );
2660
2661     /* Slice */
2662     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2663     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2664
2665     /* Slice header */
2666     x264_macroblock_thread_init( h );
2667
2668     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2669     h->mb.i_mb_xy = h->sh.i_first_mb;
2670     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2671     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2672     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2673
2674     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2675     if( h->param.b_cabac )
2676     {
2677         /* alignment needed */
2678         bs_align_1( &h->out.bs );
2679
2680         /* init cabac */
2681         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 );
2682         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2683         last_emu_check = h->cabac.p;
2684     }
2685     else
2686         last_emu_check = h->out.bs.p;
2687     h->mb.i_last_qp = h->sh.i_qp;
2688     h->mb.i_last_dqp = 0;
2689     h->mb.field_decoding_flag = 0;
2690
2691     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2692     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2693     i_skip = 0;
2694
2695     while( 1 )
2696     {
2697         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2698         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2699
2700         if( i_mb_x == 0 )
2701         {
2702             if( x264_bitstream_check_buffer( h ) )
2703                 return -1;
2704             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2705                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2706             if( !h->mb.b_reencode_mb )
2707                 x264_fdec_filter_row( h, i_mb_y, 0 );
2708         }
2709
2710         if( back_up_bitstream )
2711         {
2712             if( back_up_bitstream_cavlc )
2713                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2714             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2715             {
2716                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2717                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2718                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2719             }
2720         }
2721
2722         if( PARAM_INTERLACED )
2723         {
2724             if( h->mb.b_adaptive_mbaff )
2725             {
2726                 if( !(i_mb_y&1) )
2727                 {
2728                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2729                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2730                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2731                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2732                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2733                 }
2734             }
2735             h->mb.field[mb_xy] = MB_INTERLACED;
2736         }
2737
2738         /* load cache */
2739         if( SLICE_MBAFF )
2740             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2741         else
2742             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2743
2744         x264_macroblock_analyse( h );
2745
2746         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2747 reencode:
2748         x264_macroblock_encode( h );
2749
2750         if( h->param.b_cabac )
2751         {
2752             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2753                 x264_cabac_encode_terminal( &h->cabac );
2754
2755             if( IS_SKIP( h->mb.i_type ) )
2756                 x264_cabac_mb_skip( h, 1 );
2757             else
2758             {
2759                 if( h->sh.i_type != SLICE_TYPE_I )
2760                     x264_cabac_mb_skip( h, 0 );
2761                 x264_macroblock_write_cabac( h, &h->cabac );
2762             }
2763         }
2764         else
2765         {
2766             if( IS_SKIP( h->mb.i_type ) )
2767                 i_skip++;
2768             else
2769             {
2770                 if( h->sh.i_type != SLICE_TYPE_I )
2771                 {
2772                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2773                     i_skip = 0;
2774                 }
2775                 x264_macroblock_write_cavlc( h );
2776                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2777                 if( h->mb.b_overflow )
2778                 {
2779                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2780                     h->mb.i_skip_intra = 0;
2781                     h->mb.b_skip_mc = 0;
2782                     h->mb.b_overflow = 0;
2783                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2784                     goto reencode;
2785                 }
2786             }
2787         }
2788
2789         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2790         int mb_size = total_bits - mb_spos;
2791
2792         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2793         {
2794             /* Count the skip run, just in case. */
2795             if( !h->param.b_cabac )
2796                 total_bits += bs_size_ue_big( i_skip );
2797             /* Check for escape bytes. */
2798             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2799             for( ; last_emu_check < end - 2; last_emu_check++ )
2800                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2801                 {
2802                     slice_max_size -= 8;
2803                     last_emu_check++;
2804                 }
2805             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2806             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2807             {
2808                 if( !x264_frame_new_slice( h, h->fdec ) )
2809                 {
2810                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2811                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2812                      * If possible, roll back to the last checkpoint and try again.
2813                      * We could try raising QP, but that would break in the case where a slice spans multiple
2814                      * rows, which the re-encoding infrastructure can't currently handle. */
2815                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2816                     {
2817                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2818                         {
2819                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2820                             slice_max_size = 0;
2821                             goto cont;
2822                         }
2823                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2824                         h->mb.b_reencode_mb = 1;
2825                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2826                         break;
2827                     }
2828                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2829                     {
2830                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2831                         h->mb.b_reencode_mb = 1;
2832                         if( SLICE_MBAFF )
2833                         {
2834                             // set to bottom of previous mbpair
2835                             if( i_mb_x )
2836                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2837                             else
2838                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2839                         }
2840                         else
2841                             h->sh.i_last_mb = mb_xy-1;
2842                         break;
2843                     }
2844                     else
2845                         h->sh.i_last_mb = mb_xy;
2846                 }
2847                 else
2848                     slice_max_size = 0;
2849             }
2850         }
2851 cont:
2852         h->mb.b_reencode_mb = 0;
2853
2854         /* save cache */
2855         x264_macroblock_cache_save( h );
2856
2857         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2858         {
2859             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2860             h->mb.b_reencode_mb = 1;
2861             i_mb_x = 0;
2862             i_mb_y = i_mb_y - SLICE_MBAFF;
2863             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2864             h->sh.i_last_mb = orig_last_mb;
2865             continue;
2866         }
2867
2868         /* accumulate mb stats */
2869         h->stat.frame.i_mb_count[h->mb.i_type]++;
2870
2871         int b_intra = IS_INTRA( h->mb.i_type );
2872         int b_skip = IS_SKIP( h->mb.i_type );
2873         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2874         {
2875             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2876             {
2877                 if( h->mb.i_partition != D_8x8 )
2878                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2879                     else
2880                         for( int i = 0; i < 4; i++ )
2881                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2882                 if( h->param.i_frame_reference > 1 )
2883                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2884                         for( int i = 0; i < 4; i++ )
2885                         {
2886                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2887                             if( i_ref >= 0 )
2888                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2889                         }
2890             }
2891         }
2892
2893         if( h->param.i_log_level >= X264_LOG_INFO )
2894         {
2895             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2896             {
2897                 if( CHROMA444 )
2898                 {
2899                     for( int i = 0; i < 4; i++ )
2900                         if( h->mb.i_cbp_luma & (1 << i) )
2901                             for( int p = 0; p < 3; p++ )
2902                             {
2903                                 int s8 = i*4+p*16;
2904                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2905                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2906                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2907                             }
2908                 }
2909                 else
2910                 {
2911                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2912                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2913                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2914                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2915                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2916                 }
2917             }
2918             if( h->mb.i_cbp_luma && !b_intra )
2919             {
2920                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2921                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2922             }
2923             if( b_intra && h->mb.i_type != I_PCM )
2924             {
2925                 if( h->mb.i_type == I_16x16 )
2926                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2927                 else if( h->mb.i_type == I_8x8 )
2928                     for( int i = 0; i < 16; i += 4 )
2929                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2930                 else //if( h->mb.i_type == I_4x4 )
2931                     for( int i = 0; i < 16; i++ )
2932                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2933                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2934             }
2935             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2936         }
2937
2938         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2939         if( b_deblock )
2940             x264_macroblock_deblock_strength( h );
2941
2942         if( mb_xy == h->sh.i_last_mb )
2943             break;
2944
2945         if( SLICE_MBAFF )
2946         {
2947             i_mb_x += i_mb_y & 1;
2948             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2949         }
2950         else
2951             i_mb_x++;
2952         if( i_mb_x == h->mb.i_mb_width )
2953         {
2954             i_mb_y++;
2955             i_mb_x = 0;
2956         }
2957     }
2958     if( h->sh.i_last_mb < h->sh.i_first_mb )
2959         return 0;
2960
2961     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2962
2963     if( h->param.b_cabac )
2964     {
2965         x264_cabac_encode_flush( h, &h->cabac );
2966         h->out.bs.p = h->cabac.p;
2967     }
2968     else
2969     {
2970         if( i_skip > 0 )
2971             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2972         /* rbsp_slice_trailing_bits */
2973         bs_rbsp_trailing( &h->out.bs );
2974         bs_flush( &h->out.bs );
2975     }
2976     if( x264_nal_end( h ) )
2977         return -1;
2978
2979     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2980     {
2981         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2982                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2983                                   - h->stat.frame.i_tex_bits
2984                                   - h->stat.frame.i_mv_bits;
2985         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2986
2987         if( h->param.b_sliced_threads )
2988         {
2989             /* Tell the main thread we're done. */
2990             x264_threadslice_cond_broadcast( h, 1 );
2991             /* Do hpel now */
2992             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2993                 x264_fdec_filter_row( h, mb_y, 1 );
2994             x264_threadslice_cond_broadcast( h, 2 );
2995             /* Do the first row of hpel, now that the previous slice is done */
2996             if( h->i_thread_idx > 0 )
2997             {
2998                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2999                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
3000             }
3001         }
3002
3003         /* Free mb info after the last thread's done using it */
3004         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
3005         {
3006             h->fdec->mb_info_free( h->fdec->mb_info );
3007             h->fdec->mb_info = NULL;
3008             h->fdec->mb_info_free = NULL;
3009         }
3010     }
3011
3012     return 0;
3013 }
3014
3015 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
3016 {
3017     if( dst == src )
3018         return;
3019
3020     // reference counting
3021     for( x264_frame_t **f = src->frames.reference; *f; f++ )
3022         (*f)->i_reference_count++;
3023     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
3024         x264_frame_push_unused( src, *f );
3025     src->fdec->i_reference_count++;
3026     x264_frame_push_unused( src, dst->fdec );
3027
3028     // copy everything except the per-thread pointers and the constants.
3029     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
3030     dst->param = src->param;
3031     dst->stat = src->stat;
3032     dst->pixf = src->pixf;
3033     dst->reconfig = src->reconfig;
3034 }
3035
3036 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
3037 {
3038     if( dst == src )
3039         return;
3040     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
3041 }
3042
3043 static void *x264_slices_write( x264_t *h )
3044 {
3045     int i_slice_num = 0;
3046     int last_thread_mb = h->sh.i_last_mb;
3047
3048     /* init stats */
3049     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
3050     h->mb.b_reencode_mb = 0;
3051     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
3052     {
3053         h->sh.i_last_mb = last_thread_mb;
3054         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
3055         {
3056             if( h->param.i_slice_max_mbs )
3057             {
3058                 if( SLICE_MBAFF )
3059                 {
3060                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
3061                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
3062                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
3063                         + h->param.i_slice_max_mbs - 1;
3064                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
3065                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
3066                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
3067                 }
3068                 else
3069                 {
3070                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
3071                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
3072                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
3073                 }
3074                 i_slice_num++;
3075             }
3076             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3077             {
3078                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3079                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3080                 i_slice_num++;
3081                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3082             }
3083         }
3084         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3085         if( x264_stack_align( x264_slice_write, h ) )
3086             goto fail;
3087         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3088         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3089         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3090             h->sh.i_first_mb -= h->mb.i_mb_stride;
3091     }
3092
3093     return (void *)0;
3094
3095 fail:
3096     /* Tell other threads we're done, so they wouldn't wait for it */
3097     if( h->param.b_sliced_threads )
3098         x264_threadslice_cond_broadcast( h, 2 );
3099     return (void *)-1;
3100 }
3101
3102 static int x264_threaded_slices_write( x264_t *h )
3103 {
3104     /* set first/last mb and sync contexts */
3105     for( int i = 0; i < h->param.i_threads; i++ )
3106     {
3107         x264_t *t = h->thread[i];
3108         if( i )
3109         {
3110             t->param = h->param;
3111             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3112         }
3113         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3114         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3115         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3116         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3117         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3118     }
3119
3120     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3121
3122     x264_threads_distribute_ratecontrol( h );
3123
3124     /* setup */
3125     for( int i = 0; i < h->param.i_threads; i++ )
3126     {
3127         h->thread[i]->i_thread_idx = i;
3128         h->thread[i]->b_thread_active = 1;
3129         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3130     }
3131     /* dispatch */
3132     for( int i = 0; i < h->param.i_threads; i++ )
3133         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3134     /* wait */
3135     for( int i = 0; i < h->param.i_threads; i++ )
3136         x264_threadslice_cond_wait( h->thread[i], 1 );
3137
3138     x264_threads_merge_ratecontrol( h );
3139
3140     for( int i = 1; i < h->param.i_threads; i++ )
3141     {
3142         x264_t *t = h->thread[i];
3143         for( int j = 0; j < t->out.i_nal; j++ )
3144         {
3145             h->out.nal[h->out.i_nal] = t->out.nal[j];
3146             h->out.i_nal++;
3147             x264_nal_check_buffer( h );
3148         }
3149         /* All entries in stat.frame are ints except for ssd/ssim. */
3150         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3151             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3152         for( int j = 0; j < 3; j++ )
3153             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3154         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3155         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3156     }
3157
3158     return 0;
3159 }
3160
3161 void x264_encoder_intra_refresh( x264_t *h )
3162 {
3163     h = h->thread[h->i_thread_phase];
3164     h->b_queued_intra_refresh = 1;
3165 }
3166
3167 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3168 {
3169     if( h->param.i_bframe )
3170     {
3171         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3172         return -1;
3173     }
3174     if( h->param.b_intra_refresh )
3175     {
3176         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3177         return -1;
3178     }
3179     h = h->thread[h->i_thread_phase];
3180     if( pts >= h->i_last_idr_pts )
3181     {
3182         for( int i = 0; h->frames.reference[i]; i++ )
3183             if( pts <= h->frames.reference[i]->i_pts )
3184                 h->frames.reference[i]->b_corrupt = 1;
3185         if( pts <= h->fdec->i_pts )
3186             h->fdec->b_corrupt = 1;
3187     }
3188     return 0;
3189 }
3190
3191 /****************************************************************************
3192  * x264_encoder_encode:
3193  *  XXX: i_poc   : is the poc of the current given picture
3194  *       i_frame : is the number of the frame being coded
3195  *  ex:  type frame poc
3196  *       I      0   2*0
3197  *       P      1   2*3
3198  *       B      2   2*1
3199  *       B      3   2*2
3200  *       P      4   2*6
3201  *       B      5   2*4
3202  *       B      6   2*5
3203  ****************************************************************************/
3204 int     x264_encoder_encode( x264_t *h,
3205                              x264_nal_t **pp_nal, int *pi_nal,
3206                              x264_picture_t *pic_in,
3207                              x264_picture_t *pic_out )
3208 {
3209     x264_t *thread_current, *thread_prev, *thread_oldest;
3210     int i_nal_type, i_nal_ref_idc, i_global_qp;
3211     int overhead = NALU_OVERHEAD;
3212
3213 #if HAVE_OPENCL
3214     if( h->opencl.b_fatal_error )
3215         return -1;
3216 #endif
3217
3218     if( h->i_thread_frames > 1 )
3219     {
3220         thread_prev    = h->thread[ h->i_thread_phase ];
3221         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3222         thread_current = h->thread[ h->i_thread_phase ];
3223         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3224         x264_thread_sync_context( thread_current, thread_prev );
3225         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3226         h = thread_current;
3227     }
3228     else
3229     {
3230         thread_current =
3231         thread_oldest  = h;
3232     }
3233     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3234
3235     /* no data out */
3236     *pi_nal = 0;
3237     *pp_nal = NULL;
3238
3239     /* ------------------- Setup new frame from picture -------------------- */
3240     if( pic_in != NULL )
3241     {
3242         if( h->lookahead->b_exit_thread )
3243         {
3244             x264_log( h, X264_LOG_ERROR, "lookahead thread is already stopped\n" );
3245             return -1;
3246         }
3247
3248         /* 1: Copy the picture to a frame and move it to a buffer */
3249         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3250         if( !fenc )
3251             return -1;
3252
3253         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3254             return -1;
3255
3256         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3257             h->param.i_height != 16 * h->mb.i_mb_height )
3258             x264_frame_expand_border_mod16( h, fenc );
3259
3260         fenc->i_frame = h->frames.i_input++;
3261
3262         if( fenc->i_frame == 0 )
3263             h->frames.i_first_pts = fenc->i_pts;
3264         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3265             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3266
3267         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3268             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3269
3270         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3271         h->frames.i_largest_pts = fenc->i_pts;
3272
3273         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3274             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3275
3276         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3277         {
3278 #if HAVE_INTERLACED
3279             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3280 #else
3281             int b_interlaced = 0;
3282 #endif
3283             if( b_interlaced )
3284             {
3285                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3286                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3287             }
3288             else
3289                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3290         }
3291
3292         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3293         {
3294             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3295                 return -1;
3296         }
3297         else
3298             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3299
3300         if( pic_in->prop.quant_offsets_free )
3301             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3302
3303         if( h->frames.b_have_lowres )
3304             x264_frame_init_lowres( h, fenc );
3305
3306         /* 2: Place the frame into the queue for its slice type decision */
3307         x264_lookahead_put_frame( h, fenc );
3308
3309         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3310         {
3311             /* Nothing yet to encode, waiting for filling of buffers */
3312             pic_out->i_type = X264_TYPE_AUTO;
3313             return 0;
3314         }
3315     }
3316     else
3317     {
3318         /* signal kills for lookahead thread */
3319         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3320         h->lookahead->b_exit_thread = 1;
3321         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3322         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3323     }
3324
3325     h->i_frame++;
3326     /* 3: The picture is analyzed in the lookahead */
3327     if( !h->frames.current[0] )
3328         x264_lookahead_get_frames( h );
3329
3330     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3331         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3332
3333     /* ------------------- Get frame to be encoded ------------------------- */
3334     /* 4: get picture to encode */
3335     h->fenc = x264_frame_shift( h->frames.current );
3336
3337     /* If applicable, wait for previous frame reconstruction to finish */
3338     if( h->param.b_sliced_threads )
3339         if( x264_threadpool_wait_all( h ) < 0 )
3340             return -1;
3341
3342     if( h->i_frame == h->i_thread_frames - 1 )
3343         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3344     if( h->reconfig )
3345     {
3346         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3347         h->reconfig = 0;
3348     }
3349     if( h->fenc->param )
3350     {
3351         x264_encoder_reconfig_apply( h, h->fenc->param );
3352         if( h->fenc->param->param_free )
3353         {
3354             h->fenc->param->param_free( h->fenc->param );
3355             h->fenc->param = NULL;
3356         }
3357     }
3358
3359     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3360     if( x264_reference_update( h ) )
3361         return -1;
3362     h->fdec->i_lines_completed = -1;
3363
3364     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3365     {
3366         int valid_refs_left = 0;
3367         for( int i = 0; h->frames.reference[i]; i++ )
3368             if( !h->frames.reference[i]->b_corrupt )
3369                 valid_refs_left++;
3370         /* No valid reference frames left: force an IDR. */
3371         if( !valid_refs_left )
3372         {
3373             h->fenc->b_keyframe = 1;
3374             h->fenc->i_type = X264_TYPE_IDR;
3375         }
3376     }
3377
3378     if( h->fenc->b_keyframe )
3379     {
3380         h->frames.i_last_keyframe = h->fenc->i_frame;
3381         if( h->fenc->i_type == X264_TYPE_IDR )
3382         {
3383             h->i_frame_num = 0;
3384             h->frames.i_last_idr = h->fenc->i_frame;
3385         }
3386     }
3387     h->sh.i_mmco_command_count =
3388     h->sh.i_mmco_remove_from_end = 0;
3389     h->b_ref_reorder[0] =
3390     h->b_ref_reorder[1] = 0;
3391     h->fdec->i_poc =
3392     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3393
3394     /* ------------------- Setup frame context ----------------------------- */
3395     /* 5: Init data dependent of frame type */
3396     if( h->fenc->i_type == X264_TYPE_IDR )
3397     {
3398         /* reset ref pictures */
3399         i_nal_type    = NAL_SLICE_IDR;
3400         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3401         h->sh.i_type = SLICE_TYPE_I;
3402         x264_reference_reset( h );
3403         h->frames.i_poc_last_open_gop = -1;
3404     }
3405     else if( h->fenc->i_type == X264_TYPE_I )
3406     {
3407         i_nal_type    = NAL_SLICE;
3408         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3409         h->sh.i_type = SLICE_TYPE_I;
3410         x264_reference_hierarchy_reset( h );
3411         if( h->param.b_open_gop )
3412             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3413     }
3414     else if( h->fenc->i_type == X264_TYPE_P )
3415     {
3416         i_nal_type    = NAL_SLICE;
3417         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3418         h->sh.i_type = SLICE_TYPE_P;
3419         x264_reference_hierarchy_reset( h );
3420         h->frames.i_poc_last_open_gop = -1;
3421     }
3422     else if( h->fenc->i_type == X264_TYPE_BREF )
3423     {
3424         i_nal_type    = NAL_SLICE;
3425         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3426         h->sh.i_type = SLICE_TYPE_B;
3427         x264_reference_hierarchy_reset( h );
3428     }
3429     else    /* B frame */
3430     {
3431         i_nal_type    = NAL_SLICE;
3432         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3433         h->sh.i_type = SLICE_TYPE_B;
3434     }
3435
3436     h->fdec->i_type = h->fenc->i_type;
3437     h->fdec->i_frame = h->fenc->i_frame;
3438     h->fenc->b_kept_as_ref =
3439     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3440
3441     h->fdec->mb_info = h->fenc->mb_info;
3442     h->fdec->mb_info_free = h->fenc->mb_info_free;
3443     h->fenc->mb_info = NULL;
3444     h->fenc->mb_info_free = NULL;
3445
3446     h->fdec->i_pts = h->fenc->i_pts;
3447     if( h->frames.i_bframe_delay )
3448     {
3449         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3450         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3451                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3452                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3453         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3454     }
3455     else
3456         h->fdec->i_dts = h->fenc->i_reordered_pts;
3457     if( h->fenc->i_type == X264_TYPE_IDR )
3458         h->i_last_idr_pts = h->fdec->i_pts;
3459
3460     /* ------------------- Init                ----------------------------- */
3461     /* build ref list 0/1 */
3462     x264_reference_build_list( h, h->fdec->i_poc );
3463
3464     /* ---------------------- Write the bitstream -------------------------- */
3465     /* Init bitstream context */
3466     if( h->param.b_sliced_threads )
3467     {
3468         for( int i = 0; i < h->param.i_threads; i++ )
3469         {
3470             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3471             h->thread[i]->out.i_nal = 0;
3472         }
3473     }
3474     else
3475     {
3476         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3477         h->out.i_nal = 0;
3478     }
3479
3480     if( h->param.b_aud )
3481     {
3482         int pic_type;
3483
3484         if( h->sh.i_type == SLICE_TYPE_I )
3485             pic_type = 0;
3486         else if( h->sh.i_type == SLICE_TYPE_P )
3487             pic_type = 1;
3488         else if( h->sh.i_type == SLICE_TYPE_B )
3489             pic_type = 2;
3490         else
3491             pic_type = 7;
3492
3493         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3494         bs_write( &h->out.bs, 3, pic_type );
3495         bs_rbsp_trailing( &h->out.bs );
3496         if( x264_nal_end( h ) )
3497             return -1;
3498         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3499     }
3500
3501     h->i_nal_type = i_nal_type;
3502     h->i_nal_ref_idc = i_nal_ref_idc;
3503
3504     if( h->param.b_intra_refresh )
3505     {
3506         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3507         {
3508             h->fdec->i_frames_since_pir = 0;
3509             h->b_queued_intra_refresh = 0;
3510             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3511              * the whole frame and counts as an intra refresh. */
3512             h->fdec->f_pir_position = h->mb.i_mb_width;
3513         }
3514         else if( h->fenc->i_type == X264_TYPE_P )
3515         {
3516             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3517             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3518             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3519             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3520             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3521                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3522             {
3523                 h->fdec->f_pir_position = 0;
3524                 h->fdec->i_frames_since_pir = 0;
3525                 h->b_queued_intra_refresh = 0;
3526                 h->fenc->b_keyframe = 1;
3527             }
3528             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3529             h->fdec->f_pir_position += increment * pocdiff;
3530             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3531             /* If our intra refresh has reached the right side of the frame, we're done. */
3532             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3533             {
3534                 h->fdec->f_pir_position = h->mb.i_mb_width;
3535                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3536             }
3537         }
3538     }
3539
3540     if( h->fenc->b_keyframe )
3541     {
3542         /* Write SPS and PPS */
3543         if( h->param.b_repeat_headers )
3544         {
3545             /* generate sequence parameters */
3546             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3547             x264_sps_write( &h->out.bs, h->sps );
3548             if( x264_nal_end( h ) )
3549                 return -1;
3550             /* Pad AUD/SPS to 256 bytes like Panasonic */
3551             if( h->param.i_avcintra_class )
3552                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3553             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3554
3555             /* generate picture parameters */
3556             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3557             x264_pps_write( &h->out.bs, h->sps, h->pps );
3558             if( x264_nal_end( h ) )
3559                 return -1;
3560             if( h->param.i_avcintra_class )
3561                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3562             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3563         }
3564
3565         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3566         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3567         {
3568             x264_hrd_fullness( h );
3569             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3570             x264_sei_buffering_period_write( h, &h->out.bs );
3571             if( x264_nal_end( h ) )
3572                return -1;
3573             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3574         }
3575     }
3576
3577     /* write extra sei */
3578     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3579     {
3580         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3581         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3582                         h->fenc->extra_sei.payloads[i].payload_type );
3583         if( x264_nal_end( h ) )
3584             return -1;
3585         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3586         if( h->fenc->extra_sei.sei_free )
3587         {
3588             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3589             h->fenc->extra_sei.payloads[i].payload = NULL;
3590         }
3591     }
3592
3593     if( h->fenc->extra_sei.sei_free )
3594     {
3595         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3596         h->fenc->extra_sei.payloads = NULL;
3597         h->fenc->extra_sei.sei_free = NULL;
3598     }
3599
3600     if( h->fenc->b_keyframe )
3601     {
3602         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3603         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3604         {
3605             /* identify ourself */
3606             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3607             if( x264_sei_version_write( h, &h->out.bs ) )
3608                 return -1;
3609             if( x264_nal_end( h ) )
3610                 return -1;
3611             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3612         }
3613
3614         if( h->fenc->i_type != X264_TYPE_IDR )
3615         {
3616             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;
3617             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3618             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3619             if( x264_nal_end( h ) )
3620                 return -1;
3621             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3622         }
3623     }
3624
3625     if( h->param.i_frame_packing >= 0 && (h->fenc->b_keyframe || h->param.i_frame_packing == 5) )
3626     {
3627         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3628         x264_sei_frame_packing_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     /* generate sei pic timing */
3635     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3636     {
3637         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3638         x264_sei_pic_timing_write( h, &h->out.bs );
3639         if( x264_nal_end( h ) )
3640             return -1;
3641         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3642     }
3643
3644     /* As required by Blu-ray. */
3645     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3646     {
3647         h->b_sh_backup = 0;
3648         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3649         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3650         if( x264_nal_end( h ) )
3651             return -1;
3652         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3653     }
3654
3655     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3656         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3657
3658     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3659     if( h->param.i_avcintra_class )
3660     {
3661         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3662         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3663         x264_filler_write( h, &h->out.bs, 0 );
3664         if( x264_nal_end( h ) )
3665             return -1;
3666         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3667
3668         /* All lengths are magic lengths that decoders expect to see */
3669         /* "UMID" SEI */
3670         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3671         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3672             return -1;
3673         if( x264_nal_end( h ) )
3674             return -1;
3675         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3676
3677         int unpadded_len;
3678         int total_len;
3679         if( h->param.i_height == 1080 )
3680         {
3681             unpadded_len = 5780;
3682             total_len = 17*512;
3683         }
3684         else
3685         {
3686             unpadded_len = 2900;
3687             total_len = 9*512;
3688         }
3689         /* "VANC" SEI */
3690         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3691         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3692             return -1;
3693         if( x264_nal_end( h ) )
3694             return -1;
3695
3696         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3697         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3698     }
3699
3700     /* Init the rate control */
3701     /* FIXME: Include slice header bit cost. */
3702     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3703     i_global_qp = x264_ratecontrol_qp( h );
3704
3705     pic_out->i_qpplus1 =
3706     h->fdec->i_qpplus1 = i_global_qp + 1;
3707
3708     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3709     {
3710         x264_reference_build_list_optimal( h );
3711         x264_reference_check_reorder( h );
3712     }
3713
3714     if( h->i_ref[0] )
3715         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3716
3717     /* ------------------------ Create slice header  ----------------------- */
3718     x264_slice_init( h, i_nal_type, i_global_qp );
3719
3720     /*------------------------- Weights -------------------------------------*/
3721     if( h->sh.i_type == SLICE_TYPE_B )
3722         x264_macroblock_bipred_init( h );
3723
3724     x264_weighted_pred_init( h );
3725
3726     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3727         h->i_frame_num++;
3728
3729     /* Write frame */
3730     h->i_threadslice_start = 0;
3731     h->i_threadslice_end = h->mb.i_mb_height;
3732     if( h->i_thread_frames > 1 )
3733     {
3734         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3735         h->b_thread_active = 1;
3736     }
3737     else if( h->param.b_sliced_threads )
3738     {
3739         if( x264_threaded_slices_write( h ) )
3740             return -1;
3741     }
3742     else
3743         if( (intptr_t)x264_slices_write( h ) )
3744             return -1;
3745
3746     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3747 }
3748
3749 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3750                                    x264_nal_t **pp_nal, int *pi_nal,
3751                                    x264_picture_t *pic_out )
3752 {
3753     char psz_message[80];
3754
3755     if( !h->param.b_sliced_threads && h->b_thread_active )
3756     {
3757         h->b_thread_active = 0;
3758         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3759             return -1;
3760     }
3761     if( !h->out.i_nal )
3762     {
3763         pic_out->i_type = X264_TYPE_AUTO;
3764         return 0;
3765     }
3766
3767     x264_emms();
3768
3769     /* generate buffering period sei and insert it into place */
3770     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3771     {
3772         x264_hrd_fullness( h );
3773         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3774         x264_sei_buffering_period_write( h, &h->out.bs );
3775         if( x264_nal_end( h ) )
3776            return -1;
3777         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3778         int idx = 0;
3779         while( h->out.nal[idx].i_type == NAL_AUD ||
3780                h->out.nal[idx].i_type == NAL_SPS ||
3781                h->out.nal[idx].i_type == NAL_PPS )
3782             idx++;
3783         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3784         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3785         h->out.nal[idx] = nal_tmp;
3786     }
3787
3788     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3789     if( frame_size < 0 )
3790         return -1;
3791
3792     /* Set output picture properties */
3793     pic_out->i_type = h->fenc->i_type;
3794
3795     pic_out->b_keyframe = h->fenc->b_keyframe;
3796     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3797
3798     pic_out->i_pts = h->fdec->i_pts;
3799     pic_out->i_dts = h->fdec->i_dts;
3800
3801     if( pic_out->i_pts < pic_out->i_dts )
3802         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3803
3804     pic_out->opaque = h->fenc->opaque;
3805
3806     pic_out->img.i_csp = h->fdec->i_csp;
3807 #if HIGH_BIT_DEPTH
3808     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3809 #endif
3810     pic_out->img.i_plane = h->fdec->i_plane;
3811     for( int i = 0; i < pic_out->img.i_plane; i++ )
3812     {
3813         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3814         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3815     }
3816
3817     x264_frame_push_unused( thread_current, h->fenc );
3818
3819     /* ---------------------- Update encoder state ------------------------- */
3820
3821     /* update rc */
3822     int filler = 0;
3823     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3824         return -1;
3825
3826     pic_out->hrd_timing = h->fenc->hrd_timing;
3827     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3828
3829     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3830      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3831     if( h->param.i_avcintra_class )
3832     {
3833         x264_t *h0 = h->thread[0];
3834         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3835         if( ret < 0 )
3836             return -1;
3837         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3838         h->out.nal[h->out.i_nal-1].i_payload += filler;
3839         h->out.nal[h->out.i_nal-1].i_padding = filler;
3840         frame_size += filler;
3841     }
3842     else
3843     {
3844         while( filler > 0 )
3845         {
3846             int f, overhead;
3847             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3848             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3849             {
3850                 int next_size = filler - h->param.i_slice_max_size;
3851                 int overflow = X264_MAX( overhead - next_size, 0 );
3852                 f = h->param.i_slice_max_size - overhead - overflow;
3853             }
3854             else
3855                 f = X264_MAX( 0, filler - overhead );
3856
3857             if( x264_bitstream_check_buffer_filler( h, f ) )
3858                 return -1;
3859             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3860             x264_filler_write( h, &h->out.bs, f );
3861             if( x264_nal_end( h ) )
3862                 return -1;
3863             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3864             if( total_size < 0 )
3865                 return -1;
3866             frame_size += total_size;
3867             filler -= total_size;
3868         }
3869     }
3870
3871     /* End bitstream, set output  */
3872     *pi_nal = h->out.i_nal;
3873     *pp_nal = h->out.nal;
3874
3875     h->out.i_nal = 0;
3876
3877     x264_noise_reduction_update( h );
3878
3879     /* ---------------------- Compute/Print statistics --------------------- */
3880     x264_thread_sync_stat( h, h->thread[0] );
3881
3882     /* Slice stat */
3883     h->stat.i_frame_count[h->sh.i_type]++;
3884     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3885     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3886
3887     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3888         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3889     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3890         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3891     for( int i = 0; i < 2; i++ )
3892         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3893     for( int i = 0; i < 6; i++ )
3894         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3895     for( int i = 0; i < 4; i++ )
3896         for( int j = 0; j < 13; j++ )
3897             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3898     if( h->sh.i_type != SLICE_TYPE_I )
3899         for( int i_list = 0; i_list < 2; i_list++ )
3900             for( int i = 0; i < X264_REF_MAX*2; i++ )
3901                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3902     for( int i = 0; i < 3; i++ )
3903         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3904     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3905     {
3906         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3907         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3908     }
3909     if( h->sh.i_type == SLICE_TYPE_B )
3910     {
3911         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3912         if( h->mb.b_direct_auto_write )
3913         {
3914             //FIXME somewhat arbitrary time constants
3915             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3916                 for( int i = 0; i < 2; i++ )
3917                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3918             for( int i = 0; i < 2; i++ )
3919                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3920         }
3921     }
3922     else
3923         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3924
3925     psz_message[0] = '\0';
3926     double dur = h->fenc->f_duration;
3927     h->stat.f_frame_duration[h->sh.i_type] += dur;
3928     if( h->param.analyse.b_psnr )
3929     {
3930         int64_t ssd[3] =
3931         {
3932             h->stat.frame.i_ssd[0],
3933             h->stat.frame.i_ssd[1],
3934             h->stat.frame.i_ssd[2],
3935         };
3936         int luma_size = h->param.i_width * h->param.i_height;
3937         int chroma_size = CHROMA_SIZE( luma_size );
3938         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3939         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3940         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3941         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3942
3943         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3944         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3945         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3946         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3947         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3948
3949         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3950                                                                     pic_out->prop.f_psnr[1],
3951                                                                     pic_out->prop.f_psnr[2] );
3952     }
3953
3954     if( h->param.analyse.b_ssim )
3955     {
3956         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3957         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3958         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3959                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3960     }
3961     psz_message[79] = '\0';
3962
3963     x264_log( h, X264_LOG_DEBUG,
3964                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3965               h->i_frame,
3966               h->fdec->f_qp_avg_aq,
3967               h->i_nal_ref_idc,
3968               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3969               h->fdec->i_poc,
3970               h->stat.frame.i_mb_count_i,
3971               h->stat.frame.i_mb_count_p,
3972               h->stat.frame.i_mb_count_skip,
3973               frame_size,
3974               psz_message );
3975
3976     // keep stats all in one place
3977     x264_thread_sync_stat( h->thread[0], h );
3978     // for the use of the next frame
3979     x264_thread_sync_stat( thread_current, h );
3980
3981 #ifdef DEBUG_MB_TYPE
3982 {
3983     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3984         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3985     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3986     {
3987         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3988             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3989         else
3990             fprintf( stderr, "? " );
3991
3992         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3993             fprintf( stderr, "\n" );
3994     }
3995 }
3996 #endif
3997
3998     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3999      * by freeing some of its pointers. */
4000     for( int i = 0; i < h->i_ref[0]; i++ )
4001         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
4002         {
4003             x264_frame_push_blank_unused( h, h->fref[0][i] );
4004             h->fref[0][i] = 0;
4005         }
4006
4007     if( h->param.psz_dump_yuv )
4008         x264_frame_dump( h );
4009     x264_emms();
4010
4011     return frame_size;
4012 }
4013
4014 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
4015 {
4016     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
4017         b_print_pcm ? "..PCM" : "",
4018         i_mb_count[I_16x16]/ i_count,
4019         i_mb_count[I_8x8]  / i_count,
4020         i_mb_count[I_4x4]  / i_count );
4021     if( b_print_pcm )
4022         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
4023 }
4024
4025 /****************************************************************************
4026  * x264_encoder_close:
4027  ****************************************************************************/
4028 void    x264_encoder_close  ( x264_t *h )
4029 {
4030     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
4031     int64_t i_mb_count_size[2][7] = {{0}};
4032     char buf[200];
4033     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
4034                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
4035                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
4036
4037     x264_lookahead_delete( h );
4038
4039 #if HAVE_OPENCL
4040     x264_opencl_lookahead_delete( h );
4041     x264_opencl_function_t *ocl = h->opencl.ocl;
4042 #endif
4043
4044     if( h->param.b_sliced_threads )
4045         x264_threadpool_wait_all( h );
4046     if( h->param.i_threads > 1 )
4047         x264_threadpool_delete( h->threadpool );
4048     if( h->param.i_lookahead_threads > 1 )
4049         x264_threadpool_delete( h->lookaheadpool );
4050     if( h->i_thread_frames > 1 )
4051     {
4052         for( int i = 0; i < h->i_thread_frames; i++ )
4053             if( h->thread[i]->b_thread_active )
4054             {
4055                 assert( h->thread[i]->fenc->i_reference_count == 1 );
4056                 x264_frame_delete( h->thread[i]->fenc );
4057             }
4058
4059         x264_t *thread_prev = h->thread[h->i_thread_phase];
4060         x264_thread_sync_ratecontrol( h, thread_prev, h );
4061         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
4062         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
4063     }
4064     h->i_frame++;
4065
4066     /* Slices used and PSNR */
4067     for( int i = 0; i < 3; i++ )
4068     {
4069         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
4070         int i_slice = slice_order[i];
4071
4072         if( h->stat.i_frame_count[i_slice] > 0 )
4073         {
4074             int i_count = h->stat.i_frame_count[i_slice];
4075             double dur =  h->stat.f_frame_duration[i_slice];
4076             if( h->param.analyse.b_psnr )
4077             {
4078                 x264_log( h, X264_LOG_INFO,
4079                           "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",
4080                           slice_type_to_char[i_slice],
4081                           i_count,
4082                           h->stat.f_frame_qp[i_slice] / i_count,
4083                           (double)h->stat.i_frame_size[i_slice] / i_count,
4084                           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,
4085                           h->stat.f_psnr_average[i_slice] / dur,
4086                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4087             }
4088             else
4089             {
4090                 x264_log( h, X264_LOG_INFO,
4091                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4092                           slice_type_to_char[i_slice],
4093                           i_count,
4094                           h->stat.f_frame_qp[i_slice] / i_count,
4095                           (double)h->stat.i_frame_size[i_slice] / i_count );
4096             }
4097         }
4098     }
4099     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4100     {
4101         char *p = buf;
4102         int den = 0;
4103         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4104         for( int i = 0; i <= h->param.i_bframe; i++ )
4105             den += (i+1) * h->stat.i_consecutive_bframes[i];
4106         for( int i = 0; i <= h->param.i_bframe; i++ )
4107             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4108         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4109     }
4110
4111     for( int i_type = 0; i_type < 2; i_type++ )
4112         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4113         {
4114             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4115             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4116         }
4117
4118     /* MB types used */
4119     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4120     {
4121         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4122         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4123         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4124         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4125     }
4126     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4127     {
4128         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4129         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4130         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4131         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4132         x264_log( h, X264_LOG_INFO,
4133                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4134                   buf,
4135                   i_mb_size[PIXEL_16x16] / (i_count*4),
4136                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4137                   i_mb_size[PIXEL_8x8] / (i_count*4),
4138                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4139                   i_mb_size[PIXEL_4x4] / (i_count*4),
4140                   i_mb_count[P_SKIP] / i_count );
4141     }
4142     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4143     {
4144         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4145         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4146         double i_mb_list_count;
4147         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4148         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4149         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4150         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4151             for( int j = 0; j < 2; j++ )
4152             {
4153                 int l0 = x264_mb_type_list_table[i][0][j];
4154                 int l1 = x264_mb_type_list_table[i][1][j];
4155                 if( l0 || l1 )
4156                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4157             }
4158         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4159         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4160         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4161         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4162         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4163         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4164                  i_mb_size[PIXEL_16x16] / (i_count*4),
4165                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4166                  i_mb_size[PIXEL_8x8] / (i_count*4),
4167                  i_mb_count[B_DIRECT] / i_count,
4168                  i_mb_count[B_SKIP]   / i_count );
4169         if( i_mb_list_count != 0 )
4170             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4171                      list_count[0] / i_mb_list_count,
4172                      list_count[1] / i_mb_list_count,
4173                      list_count[2] / i_mb_list_count );
4174         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4175     }
4176
4177     x264_ratecontrol_summary( h );
4178
4179     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 )
4180     {
4181 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4182 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4183         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4184         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4185                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4186         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4187         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4188                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4189         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4190                             h->stat.i_frame_count[SLICE_TYPE_P] +
4191                             h->stat.i_frame_count[SLICE_TYPE_B];
4192         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4193         int64_t i_inter = i_mb_count - i_skip - i_intra;
4194         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4195                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4196                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4197         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4198
4199         if( PARAM_INTERLACED )
4200         {
4201             char *fieldstats = buf;
4202             fieldstats[0] = 0;
4203             if( i_inter )
4204                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4205             if( i_skip )
4206                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4207             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4208                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4209         }
4210
4211         if( h->pps->b_transform_8x8_mode )
4212         {
4213             buf[0] = 0;
4214             if( h->stat.i_mb_count_8x8dct[0] )
4215                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4216             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4217         }
4218
4219         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4220             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4221             && h->stat.i_frame_count[SLICE_TYPE_B] )
4222         {
4223             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4224                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4225                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4226         }
4227
4228         buf[0] = 0;
4229         int csize = CHROMA444 ? 4 : 1;
4230         if( i_mb_count != i_all_intra )
4231             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4232                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4233                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4234                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4235         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4236                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4237                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4238                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4239                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4240
4241         int64_t fixed_pred_modes[4][9] = {{0}};
4242         int64_t sum_pred_modes[4] = {0};
4243         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4244         {
4245             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4246             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4247         }
4248         if( sum_pred_modes[0] )
4249             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4250                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4251                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4252                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4253                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4254         for( int i = 1; i <= 2; i++ )
4255         {
4256             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
4257             {
4258                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
4259                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
4260             }
4261             if( sum_pred_modes[i] )
4262                 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,
4263                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
4264                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
4265                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
4266                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
4267                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
4268                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
4269                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
4270                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
4271                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
4272         }
4273         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
4274         {
4275             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
4276             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
4277         }
4278         if( sum_pred_modes[3] && !CHROMA444 )
4279             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4280                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
4281                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
4282                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
4283                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
4284
4285         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4286             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
4287                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
4288                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
4289
4290         for( int i_list = 0; i_list < 2; i_list++ )
4291             for( int i_slice = 0; i_slice < 2; i_slice++ )
4292             {
4293                 char *p = buf;
4294                 int64_t i_den = 0;
4295                 int i_max = 0;
4296                 for( int i = 0; i < X264_REF_MAX*2; i++ )
4297                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
4298                     {
4299                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
4300                         i_max = i;
4301                     }
4302                 if( i_max == 0 )
4303                     continue;
4304                 for( int i = 0; i <= i_max; i++ )
4305                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
4306                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
4307             }
4308
4309         if( h->param.analyse.b_ssim )
4310         {
4311             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
4312             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
4313         }
4314         if( h->param.analyse.b_psnr )
4315         {
4316             x264_log( h, X264_LOG_INFO,
4317                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
4318                       SUM3( h->stat.f_psnr_mean_y ) / duration,
4319                       SUM3( h->stat.f_psnr_mean_u ) / duration,
4320                       SUM3( h->stat.f_psnr_mean_v ) / duration,
4321                       SUM3( h->stat.f_psnr_average ) / duration,
4322                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
4323                       f_bitrate );
4324         }
4325         else
4326             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
4327     }
4328
4329     /* rc */
4330     x264_ratecontrol_delete( h );
4331
4332     /* param */
4333     if( h->param.rc.psz_stat_out )
4334         free( h->param.rc.psz_stat_out );
4335     if( h->param.rc.psz_stat_in )
4336         free( h->param.rc.psz_stat_in );
4337
4338     x264_cqm_delete( h );
4339     x264_free( h->nal_buffer );
4340     x264_free( h->reconfig_h );
4341     x264_analyse_free_costs( h );
4342
4343     if( h->i_thread_frames > 1 )
4344         h = h->thread[h->i_thread_phase];
4345
4346     /* frames */
4347     x264_frame_delete_list( h->frames.unused[0] );
4348     x264_frame_delete_list( h->frames.unused[1] );
4349     x264_frame_delete_list( h->frames.current );
4350     x264_frame_delete_list( h->frames.blank_unused );
4351
4352     h = h->thread[0];
4353
4354     for( int i = 0; i < h->i_thread_frames; i++ )
4355         if( h->thread[i]->b_thread_active )
4356             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4357                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4358                     x264_frame_delete( h->thread[i]->fref[0][j] );
4359
4360     if( h->param.i_lookahead_threads > 1 )
4361         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4362             x264_free( h->lookahead_thread[i] );
4363
4364     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4365     {
4366         x264_frame_t **frame;
4367
4368         if( !h->param.b_sliced_threads || i == 0 )
4369         {
4370             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4371             {
4372                 assert( (*frame)->i_reference_count > 0 );
4373                 (*frame)->i_reference_count--;
4374                 if( (*frame)->i_reference_count == 0 )
4375                     x264_frame_delete( *frame );
4376             }
4377             frame = &h->thread[i]->fdec;
4378             if( *frame )
4379             {
4380                 assert( (*frame)->i_reference_count > 0 );
4381                 (*frame)->i_reference_count--;
4382                 if( (*frame)->i_reference_count == 0 )
4383                     x264_frame_delete( *frame );
4384             }
4385             x264_macroblock_cache_free( h->thread[i] );
4386         }
4387         x264_macroblock_thread_free( h->thread[i], 0 );
4388         x264_free( h->thread[i]->out.p_bitstream );
4389         x264_free( h->thread[i]->out.nal );
4390         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4391         x264_pthread_cond_destroy( &h->thread[i]->cv );
4392         x264_free( h->thread[i] );
4393     }
4394 #if HAVE_OPENCL
4395     x264_opencl_close_library( ocl );
4396 #endif
4397 }
4398
4399 int x264_encoder_delayed_frames( x264_t *h )
4400 {
4401     int delayed_frames = 0;
4402     if( h->i_thread_frames > 1 )
4403     {
4404         for( int i = 0; i < h->i_thread_frames; i++ )
4405             delayed_frames += h->thread[i]->b_thread_active;
4406         h = h->thread[h->i_thread_phase];
4407     }
4408     for( int i = 0; h->frames.current[i]; i++ )
4409         delayed_frames++;
4410     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4411     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4412     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4413     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4414     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4415     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4416     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4417     return delayed_frames;
4418 }
4419
4420 int x264_encoder_maximum_delayed_frames( x264_t *h )
4421 {
4422     return h->frames.i_delay;
4423 }