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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2012 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
  36 #if HAVE_VISUALIZE
  37 #include "common/visualize.h"
  38 #endif
  39
  40 //#define DEBUG_MB_TYPE
  41
  42 #define bs_write_ue bs_write_ue_big
  43
  44 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  45                                    x264_nal_t **pp_nal, int *pi_nal,
  46                                    x264_picture_t *pic_out );
  47
  48 /****************************************************************************
  49  *
  50  ******************************* x264 libs **********************************
  51  *
  52  ****************************************************************************/
  53 static double x264_psnr( double sqe, double size )
  54 {
  55     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  56     if( mse <= 0.0000000001 ) /* Max 100dB */
  57         return 100;
  58
  59     return -10.0 * log10( mse );
  60 }
  61
  62 static double x264_ssim( double ssim )
  63 {
  64     double inv_ssim = 1 - ssim;
  65     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
  66         return 100;
  67
  68     return -10.0 * log10( inv_ssim );
  69 }
  70
  71 static int x264_threadpool_wait_all( x264_t *h )
  72 {
  73     for( int i = 0; i < h->param.i_threads; i++ )
  74         if( h->thread[i]->b_thread_active )
  75         {
  76             h->thread[i]->b_thread_active = 0;
  77             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
  78                 return -1;
  79         }
  80     return 0;
  81 }
  82
  83 static void x264_frame_dump( x264_t *h )
  84 {
  85     FILE *f = fopen( h->param.psz_dump_yuv, "r+b" );
  86     if( !f )
  87         return;
  88
  89     /* Wait for the threads to finish deblocking */
  90     if( h->param.b_sliced_threads )
  91         x264_threadpool_wait_all( h );
  92
  93     /* Write the frame in display order */
  94     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
  95     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
  96     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
  97         for( int y = 0; y < h->param.i_height; y++ )
  98             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
  99     if( !CHROMA444 )
 100     {
 101         int cw = h->param.i_width>>1;
 102         int ch = h->param.i_height>>CHROMA_V_SHIFT;
 103         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
 104         pixel *planev = planeu + cw*ch + 16;
 105         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
 106         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
 107         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
 108         x264_free( planeu );
 109     }
 110     fclose( f );
 111 }
 112
 113 /* Fill "default" values */
 114 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 115                                     x264_sps_t *sps, x264_pps_t *pps,
 116                                     int i_idr_pic_id, int i_frame, int i_qp )
 117 {
 118     x264_param_t *param = &h->param;
 119
 120     /* First we fill all fields */
 121     sh->sps = sps;
 122     sh->pps = pps;
 123
 124     sh->i_first_mb  = 0;
 125     sh->i_last_mb   = h->mb.i_mb_count - 1;
 126     sh->i_pps_id    = pps->i_id;
 127
 128     sh->i_frame_num = i_frame;
 129
 130     sh->b_mbaff = PARAM_INTERLACED;
 131     sh->b_field_pic = 0;    /* no field support for now */
 132     sh->b_bottom_field = 0; /* not yet used */
 133
 134     sh->i_idr_pic_id = i_idr_pic_id;
 135
 136     /* poc stuff, fixed later */
 137     sh->i_poc = 0;
 138     sh->i_delta_poc_bottom = 0;
 139     sh->i_delta_poc[0] = 0;
 140     sh->i_delta_poc[1] = 0;
 141
 142     sh->i_redundant_pic_cnt = 0;
 143
 144     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 145                                 && h->param.i_bframe
 146                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 147
 148     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 149     {
 150         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
 151         {
 152             if( h->mb.b_direct_auto_write )
 153                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 154             else
 155                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 156         }
 157         else
 158         {
 159             h->mb.b_direct_auto_write = 0;
 160             sh->b_direct_spatial_mv_pred = 1;
 161         }
 162     }
 163     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 164
 165     sh->b_num_ref_idx_override = 0;
 166     sh->i_num_ref_idx_l0_active = 1;
 167     sh->i_num_ref_idx_l1_active = 1;
 168
 169     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
 170     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
 171
 172     /* If the ref list isn't in the default order, construct reordering header */
 173     for( int list = 0; list < 2; list++ )
 174     {
 175         if( sh->b_ref_pic_list_reordering[list] )
 176         {
 177             int pred_frame_num = i_frame;
 178             for( int i = 0; i < h->i_ref[list]; i++ )
 179             {
 180                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
 181                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
 182                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 183                 pred_frame_num = h->fref[list][i]->i_frame_num;
 184             }
 185         }
 186     }
 187
 188     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 189
 190     sh->i_qp = SPEC_QP(i_qp);
 191     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
 192     sh->b_sp_for_swidth = 0;
 193     sh->i_qs_delta = 0;
 194
 195     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 196     /* If effective qp <= 15, deblocking would have no effect anyway */
 197     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 198         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 199     else
 200         sh->i_disable_deblocking_filter_idc = 1;
 201     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 202     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 203 }
 204
 205 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 206 {
 207     if( sh->b_mbaff )
 208     {
 209         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
 210         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
 211         assert( (first_y&1) == 0 );
 212         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
 213     }
 214     else
 215         bs_write_ue( s, sh->i_first_mb );
 216
 217     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 218     bs_write_ue( s, sh->i_pps_id );
 219     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 220
 221     if( !sh->sps->b_frame_mbs_only )
 222     {
 223         bs_write1( s, sh->b_field_pic );
 224         if( sh->b_field_pic )
 225             bs_write1( s, sh->b_bottom_field );
 226     }
 227
 228     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 229         bs_write_ue( s, sh->i_idr_pic_id );
 230
 231     if( sh->sps->i_poc_type == 0 )
 232     {
 233         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 234         if( sh->pps->b_pic_order && !sh->b_field_pic )
 235             bs_write_se( s, sh->i_delta_poc_bottom );
 236     }
 237
 238     if( sh->pps->b_redundant_pic_cnt )
 239         bs_write_ue( s, sh->i_redundant_pic_cnt );
 240
 241     if( sh->i_type == SLICE_TYPE_B )
 242         bs_write1( s, sh->b_direct_spatial_mv_pred );
 243
 244     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 245     {
 246         bs_write1( s, sh->b_num_ref_idx_override );
 247         if( sh->b_num_ref_idx_override )
 248         {
 249             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 250             if( sh->i_type == SLICE_TYPE_B )
 251                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 252         }
 253     }
 254
 255     /* ref pic list reordering */
 256     if( sh->i_type != SLICE_TYPE_I )
 257     {
 258         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
 259         if( sh->b_ref_pic_list_reordering[0] )
 260         {
 261             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 262             {
 263                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 264                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 265             }
 266             bs_write_ue( s, 3 );
 267         }
 268     }
 269     if( sh->i_type == SLICE_TYPE_B )
 270     {
 271         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
 272         if( sh->b_ref_pic_list_reordering[1] )
 273         {
 274             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 275             {
 276                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 277                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 278             }
 279             bs_write_ue( s, 3 );
 280         }
 281     }
 282
 283     sh->b_weighted_pred = 0;
 284     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 285     {
 286         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
 287         /* pred_weight_table() */
 288         bs_write_ue( s, sh->weight[0][0].i_denom );
 289         bs_write_ue( s, sh->weight[0][1].i_denom );
 290         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 291         {
 292             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 293             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 294             bs_write1( s, luma_weight_l0_flag );
 295             if( luma_weight_l0_flag )
 296             {
 297                 bs_write_se( s, sh->weight[i][0].i_scale );
 298                 bs_write_se( s, sh->weight[i][0].i_offset );
 299             }
 300             bs_write1( s, chroma_weight_l0_flag );
 301             if( chroma_weight_l0_flag )
 302             {
 303                 for( int j = 1; j < 3; j++ )
 304                 {
 305                     bs_write_se( s, sh->weight[i][j].i_scale );
 306                     bs_write_se( s, sh->weight[i][j].i_offset );
 307                 }
 308             }
 309         }
 310     }
 311     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 312     {
 313       /* TODO */
 314     }
 315
 316     if( i_nal_ref_idc != 0 )
 317     {
 318         if( sh->i_idr_pic_id >= 0 )
 319         {
 320             bs_write1( s, 0 );  /* no output of prior pics flag */
 321             bs_write1( s, 0 );  /* long term reference flag */
 322         }
 323         else
 324         {
 325             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 326             if( sh->i_mmco_command_count > 0 )
 327             {
 328                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 329                 {
 330                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 331                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 332                 }
 333                 bs_write_ue( s, 0 ); /* end command list */
 334             }
 335         }
 336     }
 337
 338     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 339         bs_write_ue( s, sh->i_cabac_init_idc );
 340
 341     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 342
 343     if( sh->pps->b_deblocking_filter_control )
 344     {
 345         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 346         if( sh->i_disable_deblocking_filter_idc != 1 )
 347         {
 348             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 349             bs_write_se( s, sh->i_beta_offset >> 1 );
 350         }
 351     }
 352 }
 353
 354 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 355 /* reallocate, adding an arbitrary amount of space. */
 356 static int x264_bitstream_check_buffer( x264_t *h )
 357 {
 358     uint8_t *bs_bak = h->out.p_bitstream;
 359     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
 360     if( (h->param.b_cabac && (h->cabac.p_end - h->cabac.p < max_row_size)) ||
 361         (h->out.bs.p_end - h->out.bs.p < max_row_size) )
 362     {
 363         h->out.i_bitstream += max_row_size;
 364         CHECKED_MALLOC( h->out.p_bitstream, h->out.i_bitstream );
 365         h->mc.memcpy_aligned( h->out.p_bitstream, bs_bak, (h->out.i_bitstream - max_row_size) & ~15 );
 366         intptr_t delta = h->out.p_bitstream - bs_bak;
 367
 368         h->out.bs.p_start += delta;
 369         h->out.bs.p += delta;
 370         h->out.bs.p_end = h->out.p_bitstream + h->out.i_bitstream;
 371
 372         h->cabac.p_start += delta;
 373         h->cabac.p += delta;
 374         h->cabac.p_end = h->out.p_bitstream + h->out.i_bitstream;
 375
 376         for( int i = 0; i <= h->out.i_nal; i++ )
 377             h->out.nal[i].p_payload += delta;
 378         x264_free( bs_bak );
 379     }
 380     return 0;
 381 fail:
 382     x264_free( bs_bak );
 383     return -1;
 384 }
 385
 386 #if HAVE_THREAD
 387 static void x264_encoder_thread_init( x264_t *h )
 388 {
 389     if( h->param.i_sync_lookahead )
 390         x264_lower_thread_priority( 10 );
 391
 392 #if HAVE_MMX
 393     /* Misalign mask has to be set separately for each thread. */
 394     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 395         x264_cpu_mask_misalign_sse();
 396 #endif
 397 }
 398
 399 static void x264_lookahead_thread_init( x264_t *h )
 400 {
 401 #if HAVE_MMX
 402     /* Misalign mask has to be set separately for each thread. */
 403     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
 404         x264_cpu_mask_misalign_sse();
 405 #endif
 406 }
 407 #endif
 408
 409 /****************************************************************************
 410  *
 411  ****************************************************************************
 412  ****************************** External API*********************************
 413  ****************************************************************************
 414  *
 415  ****************************************************************************/
 416
 417 static int x264_validate_parameters( x264_t *h, int b_open )
 418 {
 419 #if HAVE_MMX
 420 #ifdef __SSE__
 421     if( b_open && !(x264_cpu_detect() & X264_CPU_SSE) )
 422     {
 423         x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm support\n");
 424 #else
 425     if( b_open && !(x264_cpu_detect() & X264_CPU_MMX2) )
 426     {
 427         x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm support\n");
 428 #endif
 429         x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm support (configure --disable-asm)\n");
 430         return -1;
 431     }
 432 #endif
 433
 434 #if HAVE_INTERLACED
 435     h->param.b_interlaced = !!PARAM_INTERLACED;
 436 #else
 437     if( h->param.b_interlaced )
 438     {
 439         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
 440         return -1;
 441     }
 442 #endif
 443
 444     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 445     {
 446         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 447                   h->param.i_width, h->param.i_height );
 448         return -1;
 449     }
 450
 451     int i_csp = h->param.i_csp & X264_CSP_MASK;
 452 #if X264_CHROMA_FORMAT
 453     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
 454     {
 455         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
 456         return -1;
 457     }
 458     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_NV16 )
 459     {
 460         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
 461         return -1;
 462     }
 463     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
 464     {
 465         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
 466         return -1;
 467     }
 468 #endif
 469     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 470     {
 471         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
 472         return -1;
 473     }
 474
 475     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
 476     {
 477         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
 478                   h->param.i_width, h->param.i_height );
 479         return -1;
 480     }
 481
 482     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
 483     {
 484         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
 485                   h->param.i_width, h->param.i_height );
 486         return -1;
 487     }
 488
 489     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
 490     {
 491         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
 492                   h->param.i_width, h->param.i_height );
 493         return -1;
 494     }
 495
 496     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
 497         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
 498     {
 499         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 500                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 501         return -1;
 502     }
 503
 504     if( h->param.i_threads == X264_THREADS_AUTO )
 505         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 506     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
 507         h->param.i_lookahead_threads = h->param.i_threads / (h->param.b_sliced_threads?1:6);
 508     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
 509     if( h->param.i_threads > 1 )
 510     {
 511 #if !HAVE_THREAD
 512         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 513         h->param.i_threads = 1;
 514 #endif
 515         /* Avoid absurdly small thread slices as they can reduce performance
 516          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 517         if( h->param.b_sliced_threads )
 518             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
 519     }
 520     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 521     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
 522     if( h->param.i_threads == 1 )
 523     {
 524         h->param.b_sliced_threads = 0;
 525         h->param.i_lookahead_threads = 1;
 526     }
 527     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 528     if( h->i_thread_frames > 1 )
 529         h->param.nalu_process = NULL;
 530
 531     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 532     if( h->param.i_keyint_max == 1 )
 533     {
 534         h->param.b_intra_refresh = 0;
 535         h->param.analyse.i_weighted_pred = 0;
 536     }
 537
 538     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
 539
 540     /* Detect default ffmpeg settings and terminate with an error. */
 541     if( b_open )
 542     {
 543         int score = 0;
 544         score += h->param.analyse.i_me_range == 0;
 545         score += h->param.rc.i_qp_step == 3;
 546         score += h->param.i_keyint_max == 12;
 547         score += h->param.rc.i_qp_min == 2;
 548         score += h->param.rc.i_qp_max == 31;
 549         score += h->param.rc.f_qcompress == 0.5;
 550         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 551         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 552         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 553         if( score >= 5 )
 554         {
 555             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 556             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 557             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 558             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 559             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 560             return -1;
 561         }
 562     }
 563
 564     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 565     {
 566         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 567         return -1;
 568     }
 569     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 570     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
 571     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 572     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
 573     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
 574     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
 575     if( h->param.rc.i_rc_method == X264_RC_CRF )
 576     {
 577         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 578         h->param.rc.i_bitrate = 0;
 579     }
 580     if( (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 581         && h->param.rc.i_qp_constant == 0 )
 582     {
 583         h->mb.b_lossless = 1;
 584         h->param.i_cqm_preset = X264_CQM_FLAT;
 585         h->param.psz_cqm_file = NULL;
 586         h->param.rc.i_rc_method = X264_RC_CQP;
 587         h->param.rc.f_ip_factor = 1;
 588         h->param.rc.f_pb_factor = 1;
 589         h->param.analyse.b_psnr = 0;
 590         h->param.analyse.b_ssim = 0;
 591         h->param.analyse.i_chroma_qp_offset = 0;
 592         h->param.analyse.i_trellis = 0;
 593         h->param.analyse.b_fast_pskip = 0;
 594         h->param.analyse.i_noise_reduction = 0;
 595         h->param.analyse.b_psy = 0;
 596         h->param.i_bframe = 0;
 597         /* 8x8dct is not useful without RD in CAVLC lossless */
 598         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 599             h->param.analyse.b_transform_8x8 = 0;
 600     }
 601     if( h->param.rc.i_rc_method == X264_RC_CQP )
 602     {
 603         float qp_p = h->param.rc.i_qp_constant;
 604         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 605         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 606         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 607         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 608         h->param.rc.i_aq_mode = 0;
 609         h->param.rc.b_mb_tree = 0;
 610         h->param.rc.i_bitrate = 0;
 611     }
 612     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 613     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 614     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
 615     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
 616     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
 617     {
 618         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
 619         return -1;
 620     }
 621     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
 622     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
 623     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
 624     if( h->param.rc.i_vbv_buffer_size )
 625     {
 626         if( h->param.rc.i_rc_method == X264_RC_CQP )
 627         {
 628             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 629             h->param.rc.i_vbv_max_bitrate = 0;
 630             h->param.rc.i_vbv_buffer_size = 0;
 631         }
 632         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 633         {
 634             if( h->param.rc.i_rc_method == X264_RC_ABR )
 635             {
 636                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 637                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 638             }
 639             else
 640             {
 641                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 642                 h->param.rc.i_vbv_buffer_size = 0;
 643             }
 644         }
 645         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 646                  h->param.rc.i_rc_method == X264_RC_ABR )
 647         {
 648             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 649             h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 650         }
 651     }
 652     else if( h->param.rc.i_vbv_max_bitrate )
 653     {
 654         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 655         h->param.rc.i_vbv_max_bitrate = 0;
 656     }
 657
 658     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 659     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 660
 661     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
 662     if( h->param.b_sliced_threads )
 663         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 664     else
 665     {
 666         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 667         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 668             h->param.i_slice_count = 0;
 669     }
 670
 671     if( h->param.b_bluray_compat )
 672     {
 673         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
 674         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
 675         h->param.b_aud = 1;
 676         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
 677         h->param.i_slice_max_size = 0;
 678         h->param.i_slice_max_mbs = 0;
 679         h->param.b_intra_refresh = 0;
 680         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
 681         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
 682         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
 683         h->param.i_keyint_min = 1;
 684         /* Due to the proliferation of broken players that don't handle dupes properly. */
 685         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
 686         if( h->param.b_fake_interlaced )
 687             h->param.b_pic_struct = 1;
 688     }
 689
 690     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 691     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 692     if( h->param.i_scenecut_threshold < 0 )
 693         h->param.i_scenecut_threshold = 0;
 694     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
 695     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 696     {
 697         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 698         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 699     }
 700     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 701     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 702     if( h->param.i_bframe <= 1 )
 703         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 704     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 705     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
 706     if( !h->param.i_bframe )
 707     {
 708         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 709         h->param.analyse.i_direct_mv_pred = 0;
 710         h->param.analyse.b_weighted_bipred = 0;
 711         h->param.b_open_gop = 0;
 712     }
 713     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 714     {
 715         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 716         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 717     }
 718     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 719     {
 720         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 721         h->param.i_frame_reference = 1;
 722         h->param.i_dpb_size = 1;
 723     }
 724     if( h->param.b_intra_refresh && h->param.b_open_gop )
 725     {
 726         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 727         h->param.b_open_gop = 0;
 728     }
 729     if( !h->param.i_fps_num || !h->param.i_fps_den )
 730     {
 731         h->param.i_fps_num = 25;
 732         h->param.i_fps_den = 1;
 733     }
 734     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
 735     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 736         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 737     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 738     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 739     {
 740         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 741         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 742         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 743     }
 744
 745     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
 746     {
 747         h->param.i_timebase_num = h->param.i_fps_den;
 748         h->param.i_timebase_den = h->param.i_fps_num;
 749     }
 750
 751     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 752     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 753         h->param.rc.b_mb_tree = 0;
 754     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 755         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 756     {
 757         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 758         h->param.rc.b_mb_tree = 0;
 759     }
 760     if( b_open && h->param.rc.b_stat_read )
 761         h->param.rc.i_lookahead = 0;
 762 #if HAVE_THREAD
 763     if( h->param.i_sync_lookahead < 0 )
 764         h->param.i_sync_lookahead = h->param.i_bframe + 1;
 765     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
 766     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
 767         h->param.i_sync_lookahead = 0;
 768 #else
 769     h->param.i_sync_lookahead = 0;
 770 #endif
 771
 772     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
 773     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 774     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
 775     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
 776
 777     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
 778
 779     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
 780         h->param.i_cqm_preset = X264_CQM_FLAT;
 781
 782     if( h->param.analyse.i_me_method < X264_ME_DIA ||
 783         h->param.analyse.i_me_method > X264_ME_TESA )
 784         h->param.analyse.i_me_method = X264_ME_HEX;
 785     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
 786     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
 787         h->param.analyse.i_me_range = 16;
 788     if( h->param.analyse.i_me_method == X264_ME_TESA &&
 789         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
 790         h->param.analyse.i_me_method = X264_ME_ESA;
 791     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
 792     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
 793                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 794     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
 795     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
 796         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
 797     if( !h->param.analyse.b_transform_8x8 )
 798     {
 799         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 800         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 801     }
 802     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
 803     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
 804     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
 805     if( h->param.rc.f_aq_strength == 0 )
 806         h->param.rc.i_aq_mode = 0;
 807
 808     if( h->param.i_log_level < X264_LOG_INFO )
 809     {
 810         h->param.analyse.b_psnr = 0;
 811         h->param.analyse.b_ssim = 0;
 812     }
 813     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
 814     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
 815     {
 816         char *s = NULL;
 817
 818         if( h->param.analyse.b_psy )
 819         {
 820             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
 821             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
 822         }
 823         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
 824         {
 825             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
 826             s = "ssim";
 827         }
 828         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
 829         {
 830             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
 831             s = "psnr";
 832         }
 833         if( s )
 834             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
 835     }
 836
 837     if( !h->param.analyse.b_psy )
 838     {
 839         h->param.analyse.f_psy_rd = 0;
 840         h->param.analyse.f_psy_trellis = 0;
 841     }
 842     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
 843     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
 844     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
 845     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
 846     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
 847     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
 848     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
 849         h->param.analyse.i_chroma_qp_offset += 6;
 850     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
 851     /* so we lower the chroma QP offset to compensate */
 852     if( b_open && h->mb.i_psy_rd )
 853         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
 854     /* Psy trellis has a similar effect. */
 855     if( b_open && h->mb.i_psy_trellis )
 856         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
 857     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
 858     /* MB-tree requires AQ to be on, even if the strength is zero. */
 859     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
 860     {
 861         h->param.rc.i_aq_mode = 1;
 862         h->param.rc.f_aq_strength = 0;
 863     }
 864     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 865     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
 866         h->param.analyse.i_subpel_refine = 9;
 867
 868     {
 869         const x264_level_t *l = x264_levels;
 870         if( h->param.i_level_idc < 0 )
 871         {
 872             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
 873             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
 874                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
 875             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
 876             do h->param.i_level_idc = l->level_idc;
 877                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
 878             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
 879         }
 880         else
 881         {
 882             while( l->level_idc && l->level_idc != h->param.i_level_idc )
 883                 l++;
 884             if( l->level_idc == 0 )
 885             {
 886                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
 887                 return -1;
 888             }
 889         }
 890         if( h->param.analyse.i_mv_range <= 0 )
 891             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
 892         else
 893             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
 894     }
 895
 896     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
 897
 898     if( PARAM_INTERLACED )
 899     {
 900         if( h->param.analyse.i_me_method >= X264_ME_ESA )
 901         {
 902             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
 903             h->param.analyse.i_me_method = X264_ME_UMH;
 904         }
 905         if( h->param.analyse.i_weighted_pred > 0 )
 906         {
 907             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
 908             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
 909         }
 910     }
 911
 912     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
 913         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
 914
 915     if( h->i_thread_frames > 1 )
 916     {
 917         int r = h->param.analyse.i_mv_range_thread;
 918         int r2;
 919         if( r <= 0 )
 920         {
 921             // half of the available space is reserved and divided evenly among the threads,
 922             // the rest is allocated to whichever thread is far enough ahead to use it.
 923             // reserving more space increases quality for some videos, but costs more time
 924             // in thread synchronization.
 925             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
 926             r = max_range / 2;
 927         }
 928         r = X264_MAX( r, h->param.analyse.i_me_range );
 929         r = X264_MIN( r, h->param.analyse.i_mv_range );
 930         // round up to use the whole mb row
 931         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
 932         if( r2 < r )
 933             r2 += 16;
 934         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
 935         h->param.analyse.i_mv_range_thread = r2;
 936     }
 937
 938     if( h->param.rc.f_rate_tolerance < 0 )
 939         h->param.rc.f_rate_tolerance = 0;
 940     if( h->param.rc.f_qblur < 0 )
 941         h->param.rc.f_qblur = 0;
 942     if( h->param.rc.f_complexity_blur < 0 )
 943         h->param.rc.f_complexity_blur = 0;
 944
 945     h->param.i_sps_id &= 31;
 946
 947     if( PARAM_INTERLACED )
 948         h->param.b_pic_struct = 1;
 949
 950     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
 951
 952     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
 953     {
 954         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
 955         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 956     }
 957
 958     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
 959        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
 960     {
 961         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
 962         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
 963     }
 964
 965     /* ensure the booleans are 0 or 1 so they can be used in math */
 966 #define BOOLIFY(x) h->param.x = !!h->param.x
 967     BOOLIFY( b_cabac );
 968     BOOLIFY( b_constrained_intra );
 969     BOOLIFY( b_deblocking_filter );
 970     BOOLIFY( b_deterministic );
 971     BOOLIFY( b_sliced_threads );
 972     BOOLIFY( b_interlaced );
 973     BOOLIFY( b_intra_refresh );
 974     BOOLIFY( b_visualize );
 975     BOOLIFY( b_aud );
 976     BOOLIFY( b_repeat_headers );
 977     BOOLIFY( b_annexb );
 978     BOOLIFY( b_vfr_input );
 979     BOOLIFY( b_pulldown );
 980     BOOLIFY( b_tff );
 981     BOOLIFY( b_pic_struct );
 982     BOOLIFY( b_fake_interlaced );
 983     BOOLIFY( b_open_gop );
 984     BOOLIFY( b_bluray_compat );
 985     BOOLIFY( b_full_recon );
 986     BOOLIFY( analyse.b_transform_8x8 );
 987     BOOLIFY( analyse.b_weighted_bipred );
 988     BOOLIFY( analyse.b_chroma_me );
 989     BOOLIFY( analyse.b_mixed_references );
 990     BOOLIFY( analyse.b_fast_pskip );
 991     BOOLIFY( analyse.b_dct_decimate );
 992     BOOLIFY( analyse.b_psy );
 993     BOOLIFY( analyse.b_psnr );
 994     BOOLIFY( analyse.b_ssim );
 995     BOOLIFY( rc.b_stat_write );
 996     BOOLIFY( rc.b_stat_read );
 997     BOOLIFY( rc.b_mb_tree );
 998 #undef BOOLIFY
 999
1000     return 0;
1001 }
1002
1003 static void mbcmp_init( x264_t *h )
1004 {
1005     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1006     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1007     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1008     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1009     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1010     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1011     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1012     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1013     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1014                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1015     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1016                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1017     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1018     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1019     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1020     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1021 }
1022
1023 static void chroma_dsp_init( x264_t *h )
1024 {
1025     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1026
1027     switch( CHROMA_FORMAT )
1028     {
1029         case CHROMA_420:
1030             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1031             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1032             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1033             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1034             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1035             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1036             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1037             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1038             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1039             break;
1040         case CHROMA_422:
1041             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1042             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1043             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1044             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1045             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1046             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1047             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1048             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1049             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1050             break;
1051         case CHROMA_444:
1052             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1053             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1054             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1055             break;
1056     }
1057 }
1058
1059 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1060 {
1061     /* VUI */
1062     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1063     {
1064         uint32_t i_w = param->vui.i_sar_width;
1065         uint32_t i_h = param->vui.i_sar_height;
1066         uint32_t old_w = h->param.vui.i_sar_width;
1067         uint32_t old_h = h->param.vui.i_sar_height;
1068
1069         x264_reduce_fraction( &i_w, &i_h );
1070
1071         while( i_w > 65535 || i_h > 65535 )
1072         {
1073             i_w /= 2;
1074             i_h /= 2;
1075         }
1076
1077         x264_reduce_fraction( &i_w, &i_h );
1078
1079         if( i_w != old_w || i_h != old_h || initial )
1080         {
1081             h->param.vui.i_sar_width = 0;
1082             h->param.vui.i_sar_height = 0;
1083             if( i_w == 0 || i_h == 0 )
1084                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1085             else
1086             {
1087                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1088                 h->param.vui.i_sar_width = i_w;
1089                 h->param.vui.i_sar_height = i_h;
1090             }
1091             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1092         }
1093     }
1094 }
1095
1096 /****************************************************************************
1097  * x264_encoder_open:
1098  ****************************************************************************/
1099 x264_t *x264_encoder_open( x264_param_t *param )
1100 {
1101     x264_t *h;
1102     char buf[1000], *p;
1103     int qp, i_slicetype_length;
1104
1105     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1106
1107     /* Create a copy of param */
1108     memcpy( &h->param, param, sizeof(x264_param_t) );
1109
1110     if( param->param_free )
1111         param->param_free( param );
1112
1113     if( x264_threading_init() )
1114     {
1115         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1116         goto fail;
1117     }
1118
1119     if( x264_validate_parameters( h, 1 ) < 0 )
1120         goto fail;
1121
1122     if( h->param.psz_cqm_file )
1123         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1124             goto fail;
1125
1126     if( h->param.rc.psz_stat_out )
1127         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1128     if( h->param.rc.psz_stat_in )
1129         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1130
1131     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1132     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1133
1134     /* Init x264_t */
1135     h->i_frame = -1;
1136     h->i_frame_num = 0;
1137     h->i_idr_pic_id = 0;
1138
1139     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1140     {
1141         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1142         goto fail;
1143     }
1144
1145     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1146     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1147
1148     x264_set_aspect_ratio( h, &h->param, 1 );
1149
1150     x264_validate_levels( h, 1 );
1151
1152     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1153
1154     if( x264_cqm_init( h ) < 0 )
1155         goto fail;
1156
1157     h->mb.i_mb_width = h->sps->i_mb_width;
1158     h->mb.i_mb_height = h->sps->i_mb_height;
1159     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1160
1161     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1162     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1163
1164     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1165      * vectors during prediction, resulting in hpel mvs.
1166      * The chosen solution is to make MBAFF non-adaptive in this case. */
1167     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1168
1169     /* Init frames. */
1170     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1171         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1172     else
1173         h->frames.i_delay = h->param.i_bframe;
1174     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1175         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1176     i_slicetype_length = h->frames.i_delay;
1177     h->frames.i_delay += h->i_thread_frames - 1;
1178     h->frames.i_delay += h->param.i_sync_lookahead;
1179     h->frames.i_delay += h->param.b_vfr_input;
1180     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1181
1182     h->frames.i_max_ref0 = h->param.i_frame_reference;
1183     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1184     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1185     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1186         && ( h->param.rc.i_rc_method == X264_RC_ABR
1187           || h->param.rc.i_rc_method == X264_RC_CRF
1188           || h->param.i_bframe_adaptive
1189           || h->param.i_scenecut_threshold
1190           || h->param.rc.b_mb_tree
1191           || h->param.analyse.i_weighted_pred );
1192     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1193     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1194
1195     h->frames.i_last_idr =
1196     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1197     h->frames.i_input    = 0;
1198     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1199     h->frames.i_poc_last_open_gop = -1;
1200
1201     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1202     /* Allocate room for max refs plus a few extra just in case. */
1203     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1204     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1205                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1206     if( h->param.analyse.i_weighted_pred > 0 )
1207         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1208     h->i_ref[0] = h->i_ref[1] = 0;
1209     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1210     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);
1211     h->i_disp_fields_last_frame = -1;
1212     x264_rdo_init();
1213
1214     /* init CPU functions */
1215     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1216     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1217     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1218     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1219     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1220     x264_pixel_init( h->param.cpu, &h->pixf );
1221     x264_dct_init( h->param.cpu, &h->dctf );
1222     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1223     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1224     x264_mc_init( h->param.cpu, &h->mc );
1225     x264_quant_init( h, h->param.cpu, &h->quantf );
1226     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1227     x264_bitstream_init( h->param.cpu, &h->bsf );
1228     if( h->param.b_cabac )
1229         x264_cabac_init( h );
1230     else
1231         x264_stack_align( x264_cavlc_init, h );
1232
1233     mbcmp_init( h );
1234     chroma_dsp_init( h );
1235
1236     p = buf + sprintf( buf, "using cpu capabilities:" );
1237     for( int i = 0; x264_cpu_names[i].flags; i++ )
1238     {
1239         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1240             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1241             continue;
1242         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1243             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1244             continue;
1245         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1246             && (h->param.cpu & X264_CPU_SSE42) )
1247             continue;
1248         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1249             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1250             p += sprintf( p, " %s", x264_cpu_names[i].name );
1251     }
1252     if( !h->param.cpu )
1253         p += sprintf( p, " none!" );
1254     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1255
1256     float *logs = x264_analyse_prepare_costs( h );
1257     if( !logs )
1258         goto fail;
1259     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1260         if( x264_analyse_init_costs( h, logs, qp ) )
1261             goto fail;
1262     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1263         goto fail;
1264     x264_free( logs );
1265
1266     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1267     /* Checks for known miscompilation issues. */
1268     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1269     {
1270         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1271         goto fail;
1272     }
1273
1274     /* Must be volatile or else GCC will optimize it out. */
1275     volatile int temp = 392;
1276     if( x264_clz( temp ) != 23 )
1277     {
1278         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1279 #if ARCH_X86 || ARCH_X86_64
1280         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a-targeted build on a CPU that\n" );
1281         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1282 #endif
1283         goto fail;
1284     }
1285
1286     h->out.i_nal = 0;
1287     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1288         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1289           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1290
1291     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4;
1292     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1293
1294     if( h->param.i_threads > 1 &&
1295         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1296         goto fail;
1297     if( h->param.i_lookahead_threads > 1 &&
1298         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, (void*)x264_lookahead_thread_init, h ) )
1299         goto fail;
1300
1301     h->thread[0] = h;
1302     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1303         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1304     if( h->param.i_lookahead_threads > 1 )
1305         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1306         {
1307             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1308             *h->lookahead_thread[i] = *h;
1309         }
1310
1311     for( int i = 0; i < h->param.i_threads; i++ )
1312     {
1313         int init_nal_count = h->param.i_slice_count + 3;
1314         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1315         if( i > 0 )
1316             *h->thread[i] = *h;
1317
1318         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1319             goto fail;
1320         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1321             goto fail;
1322
1323         if( allocate_threadlocal_data )
1324         {
1325             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1326             if( !h->thread[i]->fdec )
1327                 goto fail;
1328         }
1329         else
1330             h->thread[i]->fdec = h->thread[0]->fdec;
1331
1332         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1333         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1334         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1335         h->thread[i]->out.i_nals_allocated = init_nal_count;
1336
1337         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1338             goto fail;
1339     }
1340
1341     if( x264_lookahead_init( h, i_slicetype_length ) )
1342         goto fail;
1343
1344     for( int i = 0; i < h->param.i_threads; i++ )
1345         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1346             goto fail;
1347
1348     if( x264_ratecontrol_new( h ) < 0 )
1349         goto fail;
1350
1351     if( h->param.i_nal_hrd )
1352     {
1353         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1354         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1355     }
1356
1357     if( h->param.psz_dump_yuv )
1358     {
1359         /* create or truncate the reconstructed video file */
1360         FILE *f = fopen( h->param.psz_dump_yuv, "w" );
1361         if( !f )
1362         {
1363             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1364             goto fail;
1365         }
1366         else if( !x264_is_regular_file( f ) )
1367         {
1368             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1369             goto fail;
1370         }
1371         fclose( f );
1372     }
1373
1374     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1375                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1376                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1377                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1378                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1379                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1380     char level[4];
1381     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1382     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1383         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1384         strcpy( level, "1b" );
1385
1386     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1387     {
1388         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1389             profile, level );
1390     }
1391     else
1392     {
1393         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1394         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1395             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1396     }
1397
1398     return h;
1399 fail:
1400     x264_free( h );
1401     return NULL;
1402 }
1403
1404 /****************************************************************************
1405  * x264_encoder_reconfig:
1406  ****************************************************************************/
1407 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1408 {
1409     /* If the previous frame isn't done encoding, reconfiguring is probably dangerous. */
1410     if( h->param.b_sliced_threads )
1411         if( x264_threadpool_wait_all( h ) < 0 )
1412             return -1;
1413
1414     int rc_reconfig = 0;
1415     h = h->thread[h->thread[0]->i_thread_phase];
1416     x264_set_aspect_ratio( h, param, 0 );
1417 #define COPY(var) h->param.var = param->var
1418     COPY( i_frame_reference ); // but never uses more refs than initially specified
1419     COPY( i_bframe_bias );
1420     if( h->param.i_scenecut_threshold )
1421         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1422     COPY( b_deblocking_filter );
1423     COPY( i_deblocking_filter_alphac0 );
1424     COPY( i_deblocking_filter_beta );
1425     COPY( i_frame_packing );
1426     COPY( analyse.inter );
1427     COPY( analyse.intra );
1428     COPY( analyse.i_direct_mv_pred );
1429     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1430     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1431         COPY( analyse.i_me_range );
1432     COPY( analyse.i_noise_reduction );
1433     /* We can't switch out of subme=0 during encoding. */
1434     if( h->param.analyse.i_subpel_refine )
1435         COPY( analyse.i_subpel_refine );
1436     COPY( analyse.i_trellis );
1437     COPY( analyse.b_chroma_me );
1438     COPY( analyse.b_dct_decimate );
1439     COPY( analyse.b_fast_pskip );
1440     COPY( analyse.b_mixed_references );
1441     COPY( analyse.f_psy_rd );
1442     COPY( analyse.f_psy_trellis );
1443     COPY( crop_rect );
1444     // can only twiddle these if they were enabled to begin with:
1445     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1446         COPY( analyse.i_me_method );
1447     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1448         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1449     if( h->pps->b_transform_8x8_mode )
1450         COPY( analyse.b_transform_8x8 );
1451     if( h->frames.i_max_ref1 > 1 )
1452         COPY( i_bframe_pyramid );
1453     COPY( i_slice_max_size );
1454     COPY( i_slice_max_mbs );
1455     COPY( i_slice_count );
1456     COPY( b_tff );
1457
1458     /* VBV can't be turned on if it wasn't on to begin with */
1459     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1460           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1461     {
1462         rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1463         rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1464         rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1465         COPY( rc.i_vbv_max_bitrate );
1466         COPY( rc.i_vbv_buffer_size );
1467         COPY( rc.i_bitrate );
1468     }
1469     rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1470     rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1471     COPY( rc.f_rf_constant );
1472     COPY( rc.f_rf_constant_max );
1473 #undef COPY
1474
1475     mbcmp_init( h );
1476
1477     int ret = x264_validate_parameters( h, 0 );
1478
1479     /* Supported reconfiguration options (1-pass only):
1480      * vbv-maxrate
1481      * vbv-bufsize
1482      * crf
1483      * bitrate (CBR only) */
1484     if( !ret && rc_reconfig )
1485         x264_ratecontrol_init_reconfigurable( h, 0 );
1486
1487     return ret;
1488 }
1489
1490 /****************************************************************************
1491  * x264_encoder_parameters:
1492  ****************************************************************************/
1493 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1494 {
1495     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1496 }
1497
1498 /* internal usage */
1499 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1500 {
1501     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1502
1503     nal->i_ref_idc        = i_ref_idc;
1504     nal->i_type           = i_type;
1505     nal->b_long_startcode = 1;
1506
1507     nal->i_payload= 0;
1508     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1509 }
1510
1511 /* if number of allocated nals is not enough, re-allocate a larger one. */
1512 static int x264_nal_check_buffer( x264_t *h )
1513 {
1514     if( h->out.i_nal >= h->out.i_nals_allocated )
1515     {
1516         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1517         if( !new_out )
1518             return -1;
1519         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1520         x264_free( h->out.nal );
1521         h->out.nal = new_out;
1522         h->out.i_nals_allocated *= 2;
1523     }
1524     return 0;
1525 }
1526
1527 static int x264_nal_end( x264_t *h )
1528 {
1529     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1530     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1531     nal->i_payload = end - nal->p_payload;
1532     /* nal_escape_mmx reads past the end of the input.
1533      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1534     memset( end, 0xff, 32 );
1535     if( h->param.nalu_process )
1536         h->param.nalu_process( h, nal, h->fenc->opaque );
1537     h->out.i_nal++;
1538
1539     return x264_nal_check_buffer( h );
1540 }
1541
1542 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1543 {
1544     int nal_size = 0, previous_nal_size = 0;
1545
1546     if( h->param.nalu_process )
1547     {
1548         for( int i = start; i < h->out.i_nal; i++ )
1549             nal_size += h->out.nal[i].i_payload;
1550         return nal_size;
1551     }
1552
1553     for( int i = 0; i < start; i++ )
1554         previous_nal_size += h->out.nal[i].i_payload;
1555
1556     for( int i = start; i < h->out.i_nal; i++ )
1557         nal_size += h->out.nal[i].i_payload;
1558
1559     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1560     int necessary_size = nal_size * 3/2 + h->out.i_nal * 4;
1561     if( h->nal_buffer_size < necessary_size )
1562     {
1563         h->nal_buffer_size = necessary_size * 2;
1564         uint8_t *buf = x264_malloc( h->nal_buffer_size );
1565         if( !buf )
1566             return -1;
1567         if( previous_nal_size )
1568             memcpy( buf, h->nal_buffer, previous_nal_size );
1569         x264_free( h->nal_buffer );
1570         h->nal_buffer = buf;
1571     }
1572
1573     uint8_t *nal_buffer = h->nal_buffer + previous_nal_size;
1574
1575     for( int i = start; i < h->out.i_nal; i++ )
1576     {
1577         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS;
1578         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1579         nal_buffer += h->out.nal[i].i_payload;
1580     }
1581
1582     x264_emms();
1583
1584     return nal_buffer - (h->nal_buffer + previous_nal_size);
1585 }
1586
1587 /****************************************************************************
1588  * x264_encoder_headers:
1589  ****************************************************************************/
1590 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1591 {
1592     int frame_size = 0;
1593     /* init bitstream context */
1594     h->out.i_nal = 0;
1595     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1596
1597     /* Write SEI, SPS and PPS. */
1598
1599     /* generate sequence parameters */
1600     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1601     x264_sps_write( &h->out.bs, h->sps );
1602     if( x264_nal_end( h ) )
1603         return -1;
1604
1605     /* generate picture parameters */
1606     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1607     x264_pps_write( &h->out.bs, h->sps, h->pps );
1608     if( x264_nal_end( h ) )
1609         return -1;
1610
1611     /* identify ourselves */
1612     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1613     if( x264_sei_version_write( h, &h->out.bs ) )
1614         return -1;
1615     if( x264_nal_end( h ) )
1616         return -1;
1617
1618     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1619     if( frame_size < 0 )
1620         return -1;
1621
1622     /* now set output*/
1623     *pi_nal = h->out.i_nal;
1624     *pp_nal = &h->out.nal[0];
1625     h->out.i_nal = 0;
1626
1627     return frame_size;
1628 }
1629
1630 /* Check to see whether we have chosen a reference list ordering different
1631  * from the standard's default. */
1632 static inline void x264_reference_check_reorder( x264_t *h )
1633 {
1634     /* The reorder check doesn't check for missing frames, so just
1635      * force a reorder if one of the reference list is corrupt. */
1636     for( int i = 0; h->frames.reference[i]; i++ )
1637         if( h->frames.reference[i]->b_corrupt )
1638         {
1639             h->b_ref_reorder[0] = 1;
1640             return;
1641         }
1642     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1643         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1644         {
1645             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1646             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1647             /* P and B-frames use different default orders. */
1648             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1649             {
1650                 h->b_ref_reorder[list] = 1;
1651                 return;
1652             }
1653         }
1654 }
1655
1656 /* return -1 on failure, else return the index of the new reference frame */
1657 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
1658 {
1659     int i = h->i_ref[0];
1660     int j = 1;
1661     x264_frame_t *newframe;
1662     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
1663         return -1;
1664
1665     //Duplication is only used in X264_WEIGHTP_SMART
1666     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
1667         return -1;
1668
1669     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
1670      * With high bit depth, it's not worth doing, so turn it off except in the case of
1671      * unweighted dupes. */
1672     if( BIT_DEPTH > 8 && w != x264_weight_none )
1673         return -1;
1674
1675     newframe = x264_frame_pop_blank_unused( h );
1676     if( !newframe )
1677         return -1;
1678
1679     //FIXME: probably don't need to copy everything
1680     *newframe = *h->fref[0][i_ref];
1681     newframe->i_reference_count = 1;
1682     newframe->orig = h->fref[0][i_ref];
1683     newframe->b_duplicate = 1;
1684     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
1685
1686     /* shift the frames to make space for the dupe. */
1687     h->b_ref_reorder[0] = 1;
1688     if( h->i_ref[0] < X264_REF_MAX )
1689         ++h->i_ref[0];
1690     h->fref[0][X264_REF_MAX-1] = NULL;
1691     x264_frame_unshift( &h->fref[0][j], newframe );
1692
1693     return j;
1694 }
1695
1696 static void x264_weighted_pred_init( x264_t *h )
1697 {
1698     /* for now no analysis and set all weights to nothing */
1699     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
1700         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
1701
1702     // FIXME: This only supports weighting of one reference frame
1703     // and duplicates of that frame.
1704     h->fenc->i_lines_weighted = 0;
1705
1706     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
1707         for( int i = 0; i < 3; i++ )
1708             h->sh.weight[i_ref][i].weightfn = NULL;
1709
1710
1711     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
1712         return;
1713
1714     int i_padv = PADV << PARAM_INTERLACED;
1715     int denom = -1;
1716     int weightplane[2] = { 0, 0 };
1717     int buffer_next = 0;
1718     for( int i = 0; i < 3; i++ )
1719     {
1720         for( int j = 0; j < h->i_ref[0]; j++ )
1721         {
1722             if( h->fenc->weight[j][i].weightfn )
1723             {
1724                 h->sh.weight[j][i] = h->fenc->weight[j][i];
1725                 // if weight is useless, don't write it to stream
1726                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
1727                     h->sh.weight[j][i].weightfn = NULL;
1728                 else
1729                 {
1730                     if( !weightplane[!!i] )
1731                     {
1732                         weightplane[!!i] = 1;
1733                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
1734                         assert( x264_clip3( denom, 0, 7 ) == denom );
1735                     }
1736
1737                     assert( h->sh.weight[j][i].i_denom == denom );
1738                     if( !i )
1739                     {
1740                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
1741                         //scale full resolution frame
1742                         if( h->param.i_threads == 1 )
1743                         {
1744                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
1745                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
1746                             int stride = h->fenc->i_stride[0];
1747                             int width = h->fenc->i_width[0] + PADH*2;
1748                             int height = h->fenc->i_lines[0] + i_padv*2;
1749                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
1750                             h->fenc->i_lines_weighted = height;
1751                         }
1752                     }
1753                 }
1754             }
1755         }
1756     }
1757
1758     if( weightplane[1] )
1759         for( int i = 0; i < h->i_ref[0]; i++ )
1760         {
1761             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
1762             {
1763                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
1764                 h->sh.weight[i][2].i_offset = 0;
1765             }
1766             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
1767             {
1768                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
1769                 h->sh.weight[i][1].i_offset = 0;
1770             }
1771         }
1772
1773     if( !weightplane[0] )
1774         h->sh.weight[0][0].i_denom = 0;
1775     if( !weightplane[1] )
1776         h->sh.weight[0][1].i_denom = 0;
1777     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
1778 }
1779
1780 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
1781 {
1782     if( h->param.i_frame_packing == 5 )
1783         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
1784                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
1785     else
1786         return abs(h->fenc->i_frame - frame->i_frame);
1787 }
1788
1789 static inline void x264_reference_build_list( x264_t *h, int i_poc )
1790 {
1791     int b_ok;
1792
1793     /* build ref list 0/1 */
1794     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
1795     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
1796     if( h->sh.i_type == SLICE_TYPE_I )
1797         return;
1798
1799     for( int i = 0; h->frames.reference[i]; i++ )
1800     {
1801         if( h->frames.reference[i]->b_corrupt )
1802             continue;
1803         if( h->frames.reference[i]->i_poc < i_poc )
1804             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
1805         else if( h->frames.reference[i]->i_poc > i_poc )
1806             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
1807     }
1808
1809     /* Order reference lists by distance from the current frame. */
1810     for( int list = 0; list < 2; list++ )
1811     {
1812         h->fref_nearest[list] = h->fref[list][0];
1813         do
1814         {
1815             b_ok = 1;
1816             for( int i = 0; i < h->i_ref[list] - 1; i++ )
1817             {
1818                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
1819                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
1820                     h->fref_nearest[list] = h->fref[list][i+1];
1821                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
1822                 {
1823                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
1824                     b_ok = 0;
1825                     break;
1826                 }
1827             }
1828         } while( !b_ok );
1829     }
1830
1831     if( h->sh.i_mmco_remove_from_end )
1832         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
1833         {
1834             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
1835             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
1836             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
1837         }
1838
1839     x264_reference_check_reorder( h );
1840
1841     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
1842     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
1843     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
1844
1845     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
1846     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
1847         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
1848
1849     /* add duplicates */
1850     if( h->fenc->i_type == X264_TYPE_P )
1851     {
1852         int idx = -1;
1853         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
1854         {
1855             x264_weight_t w[3];
1856             w[1].weightfn = w[2].weightfn = NULL;
1857             if( h->param.rc.b_stat_read )
1858                 x264_ratecontrol_set_weights( h, h->fenc );
1859
1860             if( !h->fenc->weight[0][0].weightfn )
1861             {
1862                 h->fenc->weight[0][0].i_denom = 0;
1863                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
1864                 idx = x264_weighted_reference_duplicate( h, 0, w );
1865             }
1866             else
1867             {
1868                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
1869                 {
1870                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
1871                 }
1872                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
1873                 if( h->fenc->weight[0][0].i_offset > -128 )
1874                 {
1875                     w[0] = h->fenc->weight[0][0];
1876                     w[0].i_offset--;
1877                     h->mc.weight_cache( h, &w[0] );
1878                     idx = x264_weighted_reference_duplicate( h, 0, w );
1879                 }
1880             }
1881         }
1882         h->mb.ref_blind_dupe = idx;
1883     }
1884
1885     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
1886     h->mb.pic.i_fref[0] = h->i_ref[0];
1887     h->mb.pic.i_fref[1] = h->i_ref[1];
1888 }
1889
1890 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
1891 {
1892     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
1893     int b_hpel = h->fdec->b_kept_as_ref;
1894     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
1895     int b_end = mb_y == h->i_threadslice_end;
1896     int b_measure_quality = 1;
1897     int min_y = mb_y - (1 << SLICE_MBAFF);
1898     int b_start = min_y == h->i_threadslice_start;
1899     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
1900      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
1901     int minpix_y = min_y*16 - 4 * !b_start;
1902     int maxpix_y = mb_y*16 - 4 * !b_end;
1903     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
1904     if( h->param.b_sliced_threads )
1905     {
1906         switch( pass )
1907         {
1908             /* During encode: only do deblock if asked for */
1909             default:
1910             case 0:
1911                 b_deblock &= h->param.b_full_recon;
1912                 b_hpel = 0;
1913                 break;
1914             /* During post-encode pass: do deblock if not done yet, do hpel for all
1915              * rows except those between slices. */
1916             case 1:
1917                 b_deblock &= !h->param.b_full_recon;
1918                 b_hpel &= !(b_start && min_y > 0);
1919                 b_measure_quality = 0;
1920                 break;
1921             /* Final pass: do the rows between slices in sequence. */
1922             case 2:
1923                 b_deblock = 0;
1924                 b_measure_quality = 0;
1925                 break;
1926         }
1927     }
1928     if( mb_y & SLICE_MBAFF )
1929         return;
1930     if( min_y < h->i_threadslice_start )
1931         return;
1932
1933     if( b_deblock )
1934         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
1935             x264_frame_deblock_row( h, y );
1936
1937     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
1938      * but the actual image data is equivalent. For now, maintain this
1939      * consistency by copying deblocked pixels between planes. */
1940     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
1941         for( int p = 0; p < h->fdec->i_plane; p++ )
1942             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
1943                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
1944                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
1945                         h->mb.i_mb_width*16*sizeof(pixel) );
1946
1947     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
1948         x264_frame_expand_border( h, h->fdec, min_y );
1949     if( b_hpel )
1950     {
1951         int end = mb_y == h->mb.i_mb_height;
1952         /* Can't do hpel until the previous slice is done encoding. */
1953         if( h->param.analyse.i_subpel_refine )
1954         {
1955             x264_frame_filter( h, h->fdec, min_y, end );
1956             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
1957         }
1958     }
1959
1960     if( SLICE_MBAFF && pass == 0 )
1961         for( int i = 0; i < 3; i++ )
1962         {
1963             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
1964             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
1965         }
1966
1967     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
1968         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
1969
1970     if( b_measure_quality )
1971     {
1972         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
1973         if( h->param.analyse.b_psnr )
1974         {
1975             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
1976                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
1977                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
1978                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
1979                     h->param.i_width, maxpix_y-minpix_y );
1980             if( !CHROMA444 )
1981             {
1982                 uint64_t ssd_u, ssd_v;
1983                 int v_shift = CHROMA_V_SHIFT;
1984                 x264_pixel_ssd_nv12( &h->pixf,
1985                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
1986                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
1987                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
1988                 h->stat.frame.i_ssd[1] += ssd_u;
1989                 h->stat.frame.i_ssd[2] += ssd_v;
1990             }
1991         }
1992
1993         if( h->param.analyse.b_ssim )
1994         {
1995             int ssim_cnt;
1996             x264_emms();
1997             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
1998              * and overlap by 4 */
1999             minpix_y += b_start ? 2 : -6;
2000             h->stat.frame.f_ssim +=
2001                 x264_pixel_ssim_wxh( &h->pixf,
2002                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2003                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2004                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2005             h->stat.frame.i_ssim_cnt += ssim_cnt;
2006         }
2007     }
2008 }
2009
2010 static inline int x264_reference_update( x264_t *h )
2011 {
2012     if( !h->fdec->b_kept_as_ref )
2013     {
2014         if( h->i_thread_frames > 1 )
2015         {
2016             x264_frame_push_unused( h, h->fdec );
2017             h->fdec = x264_frame_pop_unused( h, 1 );
2018             if( !h->fdec )
2019                 return -1;
2020         }
2021         return 0;
2022     }
2023
2024     /* apply mmco from previous frame. */
2025     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2026         for( int j = 0; h->frames.reference[j]; j++ )
2027             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2028                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2029
2030     /* move frame in the buffer */
2031     x264_frame_push( h->frames.reference, h->fdec );
2032     if( h->frames.reference[h->sps->i_num_ref_frames] )
2033         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2034     h->fdec = x264_frame_pop_unused( h, 1 );
2035     if( !h->fdec )
2036         return -1;
2037     return 0;
2038 }
2039
2040 static inline void x264_reference_reset( x264_t *h )
2041 {
2042     while( h->frames.reference[0] )
2043         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2044     h->fdec->i_poc =
2045     h->fenc->i_poc = 0;
2046 }
2047
2048 static inline void x264_reference_hierarchy_reset( x264_t *h )
2049 {
2050     int ref;
2051     int b_hasdelayframe = 0;
2052
2053     /* look for delay frames -- chain must only contain frames that are disposable */
2054     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2055         b_hasdelayframe |= h->frames.current[i]->i_coded
2056                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2057
2058     /* This function must handle b-pyramid and clear frames for open-gop */
2059     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2060         return;
2061
2062     /* Remove last BREF. There will never be old BREFs in the
2063      * dpb during a BREF decode when pyramid == STRICT */
2064     for( ref = 0; h->frames.reference[ref]; ref++ )
2065     {
2066         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2067             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2068             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2069             && h->sh.i_type != SLICE_TYPE_B ) )
2070         {
2071             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2072             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2073             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2074             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2075             h->b_ref_reorder[0] = 1;
2076             ref--;
2077         }
2078     }
2079
2080     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2081     if( h->param.i_bframe_pyramid )
2082         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2083 }
2084
2085 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2086 {
2087     /* ------------------------ Create slice header  ----------------------- */
2088     if( i_nal_type == NAL_SLICE_IDR )
2089     {
2090         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2091
2092         /* alternate id */
2093         h->i_idr_pic_id ^= 1;
2094     }
2095     else
2096     {
2097         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2098
2099         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2100         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2101         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2102             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2103         {
2104             h->sh.b_num_ref_idx_override = 1;
2105         }
2106     }
2107
2108     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2109     {
2110         h->b_sh_backup = 1;
2111         h->sh_backup = h->sh;
2112     }
2113
2114     h->fdec->i_frame_num = h->sh.i_frame_num;
2115
2116     if( h->sps->i_poc_type == 0 )
2117     {
2118         h->sh.i_poc = h->fdec->i_poc;
2119         if( PARAM_INTERLACED )
2120         {
2121             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2122             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2123         }
2124         else
2125             h->sh.i_delta_poc_bottom = 0;
2126         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2127         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2128     }
2129     else
2130     {
2131         /* Nothing to do ? */
2132     }
2133
2134     x264_macroblock_slice_init( h );
2135 }
2136
2137 typedef struct
2138 {
2139     int skip;
2140     uint8_t cabac_prevbyte;
2141     bs_t bs;
2142     x264_cabac_t cabac;
2143     x264_frame_stat_t stat;
2144     int last_qp;
2145     int last_dqp;
2146     int field_decoding_flag;
2147 } x264_bs_bak_t;
2148
2149 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2150 {
2151     if( full )
2152     {
2153         bak->stat = h->stat.frame;
2154         bak->last_qp = h->mb.i_last_qp;
2155         bak->last_dqp = h->mb.i_last_dqp;
2156         bak->field_decoding_flag = h->mb.field_decoding_flag;
2157     }
2158     else
2159     {
2160         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2161         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2162     }
2163     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2164      * encoder has no context dependency and in this case, a slice is ended (and
2165      * thus the content of all contexts are thrown away). */
2166     if( h->param.b_cabac )
2167     {
2168         if( full )
2169             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2170         else
2171             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2172         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2173          * backed up. */
2174         bak->cabac_prevbyte = h->cabac.p[-1];
2175     }
2176     else
2177     {
2178         bak->bs = h->out.bs;
2179         bak->skip = i_skip;
2180     }
2181 }
2182
2183 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2184 {
2185     if( full )
2186     {
2187         h->stat.frame = bak->stat;
2188         h->mb.i_last_qp = bak->last_qp;
2189         h->mb.i_last_dqp = bak->last_dqp;
2190         h->mb.field_decoding_flag = bak->field_decoding_flag;
2191     }
2192     else
2193     {
2194         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2195         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2196     }
2197     if( h->param.b_cabac )
2198     {
2199         if( full )
2200             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2201         else
2202             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2203         h->cabac.p[-1] = bak->cabac_prevbyte;
2204     }
2205     else
2206     {
2207         h->out.bs = bak->bs;
2208         *skip = bak->skip;
2209     }
2210 }
2211
2212 static int x264_slice_write( x264_t *h )
2213 {
2214     int i_skip;
2215     int mb_xy, i_mb_x, i_mb_y;
2216     /* NALUs other than the first use a 3-byte startcode.
2217      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2218      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2219      * other inaccuracies. */
2220     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2221     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2222     int back_up_bitstream = slice_max_size || (!h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH);
2223     int starting_bits = bs_pos(&h->out.bs);
2224     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2225     int b_hpel = h->fdec->b_kept_as_ref;
2226     int orig_last_mb = h->sh.i_last_mb;
2227     uint8_t *last_emu_check;
2228     x264_bs_bak_t bs_bak[2];
2229     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2230     bs_realign( &h->out.bs );
2231
2232     /* Slice */
2233     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2234     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2235
2236     /* Slice header */
2237     x264_macroblock_thread_init( h );
2238
2239     /* If this isn't the first slice in the threadslice, set the slice QP
2240      * equal to the last QP in the previous slice for more accurate
2241      * CABAC initialization. */
2242     if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
2243     {
2244         h->sh.i_qp = h->mb.i_last_qp;
2245         h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2246     }
2247
2248     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2249     if( h->param.b_cabac )
2250     {
2251         /* alignment needed */
2252         bs_align_1( &h->out.bs );
2253
2254         /* init cabac */
2255         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 );
2256         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2257         last_emu_check = h->cabac.p;
2258     }
2259     else
2260         last_emu_check = h->out.bs.p;
2261     h->mb.i_last_qp = h->sh.i_qp;
2262     h->mb.i_last_dqp = 0;
2263     h->mb.field_decoding_flag = 0;
2264
2265     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2266     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2267     i_skip = 0;
2268
2269     while( 1 )
2270     {
2271         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2272         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2273
2274         if( i_mb_x == 0 )
2275         {
2276             if( x264_bitstream_check_buffer( h ) )
2277                 return -1;
2278             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2279                 x264_bitstream_backup( h, &bs_bak[1], i_skip, 1 );
2280             if( !h->mb.b_reencode_mb )
2281                 x264_fdec_filter_row( h, i_mb_y, 0 );
2282         }
2283
2284         if( !(i_mb_y & SLICE_MBAFF) && back_up_bitstream )
2285             x264_bitstream_backup( h, &bs_bak[0], i_skip, 0 );
2286
2287         if( PARAM_INTERLACED )
2288         {
2289             if( h->mb.b_adaptive_mbaff )
2290             {
2291                 if( !(i_mb_y&1) )
2292                 {
2293                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2294                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2295                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2296                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2297                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2298                 }
2299             }
2300             h->mb.field[mb_xy] = MB_INTERLACED;
2301         }
2302
2303         /* load cache */
2304         if( SLICE_MBAFF )
2305             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2306         else
2307             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2308
2309         x264_macroblock_analyse( h );
2310
2311         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2312 reencode:
2313         x264_macroblock_encode( h );
2314
2315         if( h->param.b_cabac )
2316         {
2317             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2318                 x264_cabac_encode_terminal( &h->cabac );
2319
2320             if( IS_SKIP( h->mb.i_type ) )
2321                 x264_cabac_mb_skip( h, 1 );
2322             else
2323             {
2324                 if( h->sh.i_type != SLICE_TYPE_I )
2325                     x264_cabac_mb_skip( h, 0 );
2326                 x264_macroblock_write_cabac( h, &h->cabac );
2327             }
2328         }
2329         else
2330         {
2331             if( IS_SKIP( h->mb.i_type ) )
2332                 i_skip++;
2333             else
2334             {
2335                 if( h->sh.i_type != SLICE_TYPE_I )
2336                 {
2337                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2338                     i_skip = 0;
2339                 }
2340                 x264_macroblock_write_cavlc( h );
2341                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2342                 if( h->mb.b_overflow )
2343                 {
2344                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2345                     h->mb.i_skip_intra = 0;
2346                     h->mb.b_skip_mc = 0;
2347                     h->mb.b_overflow = 0;
2348                     x264_bitstream_restore( h, &bs_bak[0], &i_skip, 0 );
2349                     goto reencode;
2350                 }
2351             }
2352         }
2353
2354         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2355         int mb_size = total_bits - mb_spos;
2356
2357         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2358         {
2359             /* Count the skip run, just in case. */
2360             if( !h->param.b_cabac )
2361                 total_bits += bs_size_ue_big( i_skip );
2362             /* Check for escape bytes. */
2363             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2364             for( ; last_emu_check < end - 2; last_emu_check++ )
2365                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2366                 {
2367                     slice_max_size -= 8;
2368                     last_emu_check++;
2369                 }
2370             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2371             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2372             {
2373                 if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2374                 {
2375                     x264_bitstream_restore( h, &bs_bak[0], &i_skip, 0 );
2376                     h->mb.b_reencode_mb = 1;
2377                     if( SLICE_MBAFF )
2378                     {
2379                         // set to bottom of previous mbpair
2380                         if( i_mb_x )
2381                             h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2382                         else
2383                             h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2384                     }
2385                     else
2386                         h->sh.i_last_mb = mb_xy-1;
2387                     break;
2388                 }
2389                 else
2390                     h->sh.i_last_mb = mb_xy;
2391             }
2392         }
2393         h->mb.b_reencode_mb = 0;
2394
2395 #if HAVE_VISUALIZE
2396         if( h->param.b_visualize )
2397             x264_visualize_mb( h );
2398 #endif
2399
2400         /* save cache */
2401         x264_macroblock_cache_save( h );
2402
2403         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2404         {
2405             x264_bitstream_restore( h, &bs_bak[1], &i_skip, 1 );
2406             h->mb.b_reencode_mb = 1;
2407             i_mb_x = 0;
2408             i_mb_y = i_mb_y - SLICE_MBAFF;
2409             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2410             h->sh.i_last_mb = orig_last_mb;
2411             continue;
2412         }
2413
2414         /* accumulate mb stats */
2415         h->stat.frame.i_mb_count[h->mb.i_type]++;
2416
2417         int b_intra = IS_INTRA( h->mb.i_type );
2418         int b_skip = IS_SKIP( h->mb.i_type );
2419         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2420         {
2421             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2422             {
2423                 if( h->mb.i_partition != D_8x8 )
2424                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2425                     else
2426                         for( int i = 0; i < 4; i++ )
2427                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2428                 if( h->param.i_frame_reference > 1 )
2429                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2430                         for( int i = 0; i < 4; i++ )
2431                         {
2432                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2433                             if( i_ref >= 0 )
2434                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2435                         }
2436             }
2437         }
2438
2439         if( h->param.i_log_level >= X264_LOG_INFO )
2440         {
2441             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2442             {
2443                 if( CHROMA444 )
2444                 {
2445                     for( int i = 0; i < 4; i++ )
2446                         if( h->mb.i_cbp_luma & (1 << i) )
2447                             for( int p = 0; p < 3; p++ )
2448                             {
2449                                 int s8 = i*4+p*16;
2450                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2451                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2452                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2453                             }
2454                 }
2455                 else
2456                 {
2457                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2458                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2459                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2460                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2461                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2462                 }
2463             }
2464             if( h->mb.i_cbp_luma && !b_intra )
2465             {
2466                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2467                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2468             }
2469             if( b_intra && h->mb.i_type != I_PCM )
2470             {
2471                 if( h->mb.i_type == I_16x16 )
2472                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2473                 else if( h->mb.i_type == I_8x8 )
2474                     for( int i = 0; i < 16; i += 4 )
2475                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2476                 else //if( h->mb.i_type == I_4x4 )
2477                     for( int i = 0; i < 16; i++ )
2478                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2479                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2480             }
2481             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2482         }
2483
2484         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2485         if( b_deblock )
2486             x264_macroblock_deblock_strength( h );
2487
2488         if( mb_xy == h->sh.i_last_mb )
2489             break;
2490
2491         if( SLICE_MBAFF )
2492         {
2493             i_mb_x += i_mb_y & 1;
2494             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2495         }
2496         else
2497             i_mb_x++;
2498         if( i_mb_x == h->mb.i_mb_width )
2499         {
2500             i_mb_y++;
2501             i_mb_x = 0;
2502         }
2503     }
2504     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2505
2506     if( h->param.b_cabac )
2507     {
2508         x264_cabac_encode_flush( h, &h->cabac );
2509         h->out.bs.p = h->cabac.p;
2510     }
2511     else
2512     {
2513         if( i_skip > 0 )
2514             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2515         /* rbsp_slice_trailing_bits */
2516         bs_rbsp_trailing( &h->out.bs );
2517         bs_flush( &h->out.bs );
2518     }
2519     if( x264_nal_end( h ) )
2520         return -1;
2521
2522     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2523     {
2524         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2525                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2526                                   - h->stat.frame.i_tex_bits
2527                                   - h->stat.frame.i_mv_bits;
2528         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2529
2530         if( h->param.b_sliced_threads )
2531         {
2532             /* Tell the main thread we're done. */
2533             x264_threadslice_cond_broadcast( h, 1 );
2534             /* Do hpel now */
2535             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2536                 x264_fdec_filter_row( h, mb_y, 1 );
2537             x264_threadslice_cond_broadcast( h, 2 );
2538             /* Do the first row of hpel, now that the previous slice is done */
2539             if( h->i_thread_idx > 0 )
2540             {
2541                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2542                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2543             }
2544         }
2545
2546         /* Free mb info after the last thread's done using it */
2547         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2548         {
2549             h->fdec->mb_info_free( h->fdec->mb_info );
2550             h->fdec->mb_info = NULL;
2551             h->fdec->mb_info_free = NULL;
2552         }
2553     }
2554
2555     return 0;
2556 }
2557
2558 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2559 {
2560     if( dst == src )
2561         return;
2562
2563     // reference counting
2564     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2565         (*f)->i_reference_count++;
2566     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2567         x264_frame_push_unused( src, *f );
2568     src->fdec->i_reference_count++;
2569     x264_frame_push_unused( src, dst->fdec );
2570
2571     // copy everything except the per-thread pointers and the constants.
2572     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.type) - offsetof(x264_t, i_frame) );
2573     dst->param = src->param;
2574     dst->stat = src->stat;
2575     dst->pixf = src->pixf;
2576 }
2577
2578 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2579 {
2580     if( dst == src )
2581         return;
2582     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2583 }
2584
2585 static void *x264_slices_write( x264_t *h )
2586 {
2587     int i_slice_num = 0;
2588     int last_thread_mb = h->sh.i_last_mb;
2589
2590 #if HAVE_VISUALIZE
2591     if( h->param.b_visualize )
2592         if( x264_visualize_init( h ) )
2593             goto fail;
2594 #endif
2595
2596     /* init stats */
2597     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2598     h->mb.b_reencode_mb = 0;
2599     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2600     {
2601         h->sh.i_last_mb = last_thread_mb;
2602         if( h->param.i_slice_max_mbs )
2603         {
2604             if( SLICE_MBAFF )
2605             {
2606                 // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2607                 int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2608                     + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2609                     + h->param.i_slice_max_mbs - 1;
2610                 int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2611                 int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2612                 h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2613             }
2614             else
2615                 h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2616         }
2617         else if( h->param.i_slice_count && !h->param.b_sliced_threads )
2618         {
2619             int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2620             int width = h->mb.i_mb_width << PARAM_INTERLACED;
2621             i_slice_num++;
2622             h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
2623         }
2624         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
2625         if( x264_stack_align( x264_slice_write, h ) )
2626             goto fail;
2627         h->sh.i_first_mb = h->sh.i_last_mb + 1;
2628         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
2629         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
2630             h->sh.i_first_mb -= h->mb.i_mb_stride;
2631     }
2632
2633 #if HAVE_VISUALIZE
2634     if( h->param.b_visualize )
2635     {
2636         x264_visualize_show( h );
2637         x264_visualize_close( h );
2638     }
2639 #endif
2640
2641     return (void *)0;
2642
2643 fail:
2644     /* Tell other threads we're done, so they wouldn't wait for it */
2645     if( h->param.b_sliced_threads )
2646         x264_threadslice_cond_broadcast( h, 2 );
2647     return (void *)-1;
2648 }
2649
2650 static int x264_threaded_slices_write( x264_t *h )
2651 {
2652     /* set first/last mb and sync contexts */
2653     for( int i = 0; i < h->param.i_threads; i++ )
2654     {
2655         x264_t *t = h->thread[i];
2656         if( i )
2657         {
2658             t->param = h->param;
2659             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
2660         }
2661         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
2662         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2663         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
2664         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
2665         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
2666     }
2667
2668     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
2669
2670     x264_threads_distribute_ratecontrol( h );
2671
2672     /* setup */
2673     for( int i = 0; i < h->param.i_threads; i++ )
2674     {
2675         h->thread[i]->i_thread_idx = i;
2676         h->thread[i]->b_thread_active = 1;
2677         x264_threadslice_cond_broadcast( h->thread[i], 0 );
2678     }
2679     /* dispatch */
2680     for( int i = 0; i < h->param.i_threads; i++ )
2681         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
2682     /* wait */
2683     for( int i = 0; i < h->param.i_threads; i++ )
2684         x264_threadslice_cond_wait( h->thread[i], 1 );
2685
2686     x264_threads_merge_ratecontrol( h );
2687
2688     for( int i = 1; i < h->param.i_threads; i++ )
2689     {
2690         x264_t *t = h->thread[i];
2691         for( int j = 0; j < t->out.i_nal; j++ )
2692         {
2693             h->out.nal[h->out.i_nal] = t->out.nal[j];
2694             h->out.i_nal++;
2695             x264_nal_check_buffer( h );
2696         }
2697         /* All entries in stat.frame are ints except for ssd/ssim. */
2698         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
2699             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
2700         for( int j = 0; j < 3; j++ )
2701             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
2702         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
2703         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
2704     }
2705
2706     return 0;
2707 }
2708
2709 void x264_encoder_intra_refresh( x264_t *h )
2710 {
2711     h = h->thread[h->i_thread_phase];
2712     h->b_queued_intra_refresh = 1;
2713 }
2714
2715 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
2716 {
2717     if( h->param.i_bframe )
2718     {
2719         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
2720         return -1;
2721     }
2722     if( h->param.b_intra_refresh )
2723     {
2724         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
2725         return -1;
2726     }
2727     h = h->thread[h->i_thread_phase];
2728     if( pts >= h->i_last_idr_pts )
2729     {
2730         for( int i = 0; h->frames.reference[i]; i++ )
2731             if( pts <= h->frames.reference[i]->i_pts )
2732                 h->frames.reference[i]->b_corrupt = 1;
2733         if( pts <= h->fdec->i_pts )
2734             h->fdec->b_corrupt = 1;
2735     }
2736     return 0;
2737 }
2738
2739 /****************************************************************************
2740  * x264_encoder_encode:
2741  *  XXX: i_poc   : is the poc of the current given picture
2742  *       i_frame : is the number of the frame being coded
2743  *  ex:  type frame poc
2744  *       I      0   2*0
2745  *       P      1   2*3
2746  *       B      2   2*1
2747  *       B      3   2*2
2748  *       P      4   2*6
2749  *       B      5   2*4
2750  *       B      6   2*5
2751  ****************************************************************************/
2752 int     x264_encoder_encode( x264_t *h,
2753                              x264_nal_t **pp_nal, int *pi_nal,
2754                              x264_picture_t *pic_in,
2755                              x264_picture_t *pic_out )
2756 {
2757     x264_t *thread_current, *thread_prev, *thread_oldest;
2758     int i_nal_type, i_nal_ref_idc, i_global_qp;
2759     int overhead = NALU_OVERHEAD;
2760
2761     if( h->i_thread_frames > 1 )
2762     {
2763         thread_prev    = h->thread[ h->i_thread_phase ];
2764         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
2765         thread_current = h->thread[ h->i_thread_phase ];
2766         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
2767         x264_thread_sync_context( thread_current, thread_prev );
2768         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
2769         h = thread_current;
2770     }
2771     else
2772     {
2773         thread_current =
2774         thread_oldest  = h;
2775     }
2776 #if HAVE_MMX
2777     if( h->param.cpu&X264_CPU_SSE_MISALIGN )
2778         x264_cpu_mask_misalign_sse();
2779 #endif
2780     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
2781
2782     /* no data out */
2783     *pi_nal = 0;
2784     *pp_nal = NULL;
2785
2786     /* ------------------- Setup new frame from picture -------------------- */
2787     if( pic_in != NULL )
2788     {
2789         /* 1: Copy the picture to a frame and move it to a buffer */
2790         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
2791         if( !fenc )
2792             return -1;
2793
2794         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
2795             return -1;
2796
2797         if( h->param.i_width != 16 * h->mb.i_mb_width ||
2798             h->param.i_height != 16 * h->mb.i_mb_height )
2799             x264_frame_expand_border_mod16( h, fenc );
2800
2801         fenc->i_frame = h->frames.i_input++;
2802
2803         if( fenc->i_frame == 0 )
2804             h->frames.i_first_pts = fenc->i_pts;
2805         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
2806             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
2807
2808         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
2809             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
2810
2811         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
2812         h->frames.i_largest_pts = fenc->i_pts;
2813
2814         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
2815             fenc->i_pic_struct = PIC_STRUCT_AUTO;
2816
2817         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
2818         {
2819 #if HAVE_INTERLACED
2820             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
2821 #else
2822             int b_interlaced = 0;
2823 #endif
2824             if( b_interlaced )
2825             {
2826                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
2827                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
2828             }
2829             else
2830                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
2831         }
2832
2833         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
2834         {
2835             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
2836                 return -1;
2837         }
2838         else
2839             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
2840
2841         if( pic_in->prop.quant_offsets_free )
2842             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
2843
2844         if( h->frames.b_have_lowres )
2845             x264_frame_init_lowres( h, fenc );
2846
2847         /* 2: Place the frame into the queue for its slice type decision */
2848         x264_lookahead_put_frame( h, fenc );
2849
2850         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
2851         {
2852             /* Nothing yet to encode, waiting for filling of buffers */
2853             pic_out->i_type = X264_TYPE_AUTO;
2854             return 0;
2855         }
2856     }
2857     else
2858     {
2859         /* signal kills for lookahead thread */
2860         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
2861         h->lookahead->b_exit_thread = 1;
2862         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
2863         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
2864     }
2865
2866     h->i_frame++;
2867     /* 3: The picture is analyzed in the lookahead */
2868     if( !h->frames.current[0] )
2869         x264_lookahead_get_frames( h );
2870
2871     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
2872         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
2873
2874     /* ------------------- Get frame to be encoded ------------------------- */
2875     /* 4: get picture to encode */
2876     h->fenc = x264_frame_shift( h->frames.current );
2877
2878     /* If applicable, wait for previous frame reconstruction to finish */
2879     if( h->param.b_sliced_threads )
2880         if( x264_threadpool_wait_all( h ) < 0 )
2881             return -1;
2882
2883     if( h->i_frame == h->i_thread_frames - 1 )
2884         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
2885     if( h->fenc->param )
2886     {
2887         x264_encoder_reconfig( h, h->fenc->param );
2888         if( h->fenc->param->param_free )
2889         {
2890             h->fenc->param->param_free( h->fenc->param );
2891             h->fenc->param = NULL;
2892         }
2893     }
2894
2895     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
2896     if( x264_reference_update( h ) )
2897         return -1;
2898     h->fdec->i_lines_completed = -1;
2899
2900     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
2901     {
2902         int valid_refs_left = 0;
2903         for( int i = 0; h->frames.reference[i]; i++ )
2904             if( !h->frames.reference[i]->b_corrupt )
2905                 valid_refs_left++;
2906         /* No valid reference frames left: force an IDR. */
2907         if( !valid_refs_left )
2908         {
2909             h->fenc->b_keyframe = 1;
2910             h->fenc->i_type = X264_TYPE_IDR;
2911         }
2912     }
2913
2914     if( h->fenc->b_keyframe )
2915     {
2916         h->frames.i_last_keyframe = h->fenc->i_frame;
2917         if( h->fenc->i_type == X264_TYPE_IDR )
2918         {
2919             h->i_frame_num = 0;
2920             h->frames.i_last_idr = h->fenc->i_frame;
2921         }
2922     }
2923     h->sh.i_mmco_command_count =
2924     h->sh.i_mmco_remove_from_end = 0;
2925     h->b_ref_reorder[0] =
2926     h->b_ref_reorder[1] = 0;
2927     h->fdec->i_poc =
2928     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
2929
2930     /* ------------------- Setup frame context ----------------------------- */
2931     /* 5: Init data dependent of frame type */
2932     if( h->fenc->i_type == X264_TYPE_IDR )
2933     {
2934         /* reset ref pictures */
2935         i_nal_type    = NAL_SLICE_IDR;
2936         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
2937         h->sh.i_type = SLICE_TYPE_I;
2938         x264_reference_reset( h );
2939         h->frames.i_poc_last_open_gop = -1;
2940     }
2941     else if( h->fenc->i_type == X264_TYPE_I )
2942     {
2943         i_nal_type    = NAL_SLICE;
2944         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2945         h->sh.i_type = SLICE_TYPE_I;
2946         x264_reference_hierarchy_reset( h );
2947         if( h->param.b_open_gop )
2948             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
2949     }
2950     else if( h->fenc->i_type == X264_TYPE_P )
2951     {
2952         i_nal_type    = NAL_SLICE;
2953         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
2954         h->sh.i_type = SLICE_TYPE_P;
2955         x264_reference_hierarchy_reset( h );
2956         h->frames.i_poc_last_open_gop = -1;
2957     }
2958     else if( h->fenc->i_type == X264_TYPE_BREF )
2959     {
2960         i_nal_type    = NAL_SLICE;
2961         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
2962         h->sh.i_type = SLICE_TYPE_B;
2963         x264_reference_hierarchy_reset( h );
2964     }
2965     else    /* B frame */
2966     {
2967         i_nal_type    = NAL_SLICE;
2968         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
2969         h->sh.i_type = SLICE_TYPE_B;
2970     }
2971
2972     h->fdec->i_type = h->fenc->i_type;
2973     h->fdec->i_frame = h->fenc->i_frame;
2974     h->fenc->b_kept_as_ref =
2975     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
2976
2977     h->fdec->mb_info = h->fenc->mb_info;
2978     h->fdec->mb_info_free = h->fenc->mb_info_free;
2979     h->fenc->mb_info = NULL;
2980     h->fenc->mb_info_free = NULL;
2981
2982     h->fdec->i_pts = h->fenc->i_pts;
2983     if( h->frames.i_bframe_delay )
2984     {
2985         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
2986         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
2987                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
2988                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
2989         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
2990     }
2991     else
2992         h->fdec->i_dts = h->fenc->i_reordered_pts;
2993     if( h->fenc->i_type == X264_TYPE_IDR )
2994         h->i_last_idr_pts = h->fdec->i_pts;
2995
2996     /* ------------------- Init                ----------------------------- */
2997     /* build ref list 0/1 */
2998     x264_reference_build_list( h, h->fdec->i_poc );
2999
3000     /* ---------------------- Write the bitstream -------------------------- */
3001     /* Init bitstream context */
3002     if( h->param.b_sliced_threads )
3003     {
3004         for( int i = 0; i < h->param.i_threads; i++ )
3005         {
3006             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3007             h->thread[i]->out.i_nal = 0;
3008         }
3009     }
3010     else
3011     {
3012         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3013         h->out.i_nal = 0;
3014     }
3015
3016     if( h->param.b_aud )
3017     {
3018         int pic_type;
3019
3020         if( h->sh.i_type == SLICE_TYPE_I )
3021             pic_type = 0;
3022         else if( h->sh.i_type == SLICE_TYPE_P )
3023             pic_type = 1;
3024         else if( h->sh.i_type == SLICE_TYPE_B )
3025             pic_type = 2;
3026         else
3027             pic_type = 7;
3028
3029         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3030         bs_write( &h->out.bs, 3, pic_type );
3031         bs_rbsp_trailing( &h->out.bs );
3032         if( x264_nal_end( h ) )
3033             return -1;
3034         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3035     }
3036
3037     h->i_nal_type = i_nal_type;
3038     h->i_nal_ref_idc = i_nal_ref_idc;
3039
3040     if( h->param.b_intra_refresh )
3041     {
3042         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3043         {
3044             h->fdec->i_frames_since_pir = 0;
3045             h->b_queued_intra_refresh = 0;
3046             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3047              * the whole frame and counts as an intra refresh. */
3048             h->fdec->f_pir_position = h->mb.i_mb_width;
3049         }
3050         else if( h->fenc->i_type == X264_TYPE_P )
3051         {
3052             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3053             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3054             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3055             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3056             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3057                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3058             {
3059                 h->fdec->f_pir_position = 0;
3060                 h->fdec->i_frames_since_pir = 0;
3061                 h->b_queued_intra_refresh = 0;
3062                 h->fenc->b_keyframe = 1;
3063             }
3064             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3065             h->fdec->f_pir_position += increment * pocdiff;
3066             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3067             /* If our intra refresh has reached the right side of the frame, we're done. */
3068             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3069             {
3070                 h->fdec->f_pir_position = h->mb.i_mb_width;
3071                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3072             }
3073         }
3074     }
3075
3076     if( h->fenc->b_keyframe )
3077     {
3078         /* Write SPS and PPS */
3079         if( h->param.b_repeat_headers )
3080         {
3081             /* generate sequence parameters */
3082             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3083             x264_sps_write( &h->out.bs, h->sps );
3084             if( x264_nal_end( h ) )
3085                 return -1;
3086             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3087
3088             /* generate picture parameters */
3089             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3090             x264_pps_write( &h->out.bs, h->sps, h->pps );
3091             if( x264_nal_end( h ) )
3092                 return -1;
3093             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3094         }
3095
3096         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3097         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3098         {
3099             x264_hrd_fullness( h );
3100             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3101             x264_sei_buffering_period_write( h, &h->out.bs );
3102             if( x264_nal_end( h ) )
3103                return -1;
3104             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3105         }
3106     }
3107
3108     /* write extra sei */
3109     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3110     {
3111         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3112         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3113                         h->fenc->extra_sei.payloads[i].payload_type );
3114         if( x264_nal_end( h ) )
3115             return -1;
3116         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3117         if( h->fenc->extra_sei.sei_free )
3118         {
3119             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3120             h->fenc->extra_sei.payloads[i].payload = NULL;
3121         }
3122     }
3123
3124     if( h->fenc->extra_sei.sei_free )
3125     {
3126         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3127         h->fenc->extra_sei.payloads = NULL;
3128         h->fenc->extra_sei.sei_free = NULL;
3129     }
3130
3131     if( h->fenc->b_keyframe )
3132     {
3133         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 )
3134         {
3135             /* identify ourself */
3136             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3137             if( x264_sei_version_write( h, &h->out.bs ) )
3138                 return -1;
3139             if( x264_nal_end( h ) )
3140                 return -1;
3141             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3142         }
3143
3144         if( h->fenc->i_type != X264_TYPE_IDR )
3145         {
3146             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;
3147             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3148             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3149             if( x264_nal_end( h ) )
3150                 return -1;
3151             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3152         }
3153
3154         if ( h->param.i_frame_packing >= 0 )
3155         {
3156             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3157             x264_sei_frame_packing_write( h, &h->out.bs );
3158             if( x264_nal_end( h ) )
3159                 return -1;
3160             overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3161         }
3162     }
3163
3164     /* generate sei pic timing */
3165     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3166     {
3167         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3168         x264_sei_pic_timing_write( h, &h->out.bs );
3169         if( x264_nal_end( h ) )
3170             return -1;
3171         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3172     }
3173
3174     /* As required by Blu-ray. */
3175     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3176     {
3177         h->b_sh_backup = 0;
3178         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3179         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3180         if( x264_nal_end( h ) )
3181             return -1;
3182         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal-1);
3183     }
3184
3185     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3186         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3187
3188     /* Init the rate control */
3189     /* FIXME: Include slice header bit cost. */
3190     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3191     i_global_qp = x264_ratecontrol_qp( h );
3192
3193     pic_out->i_qpplus1 =
3194     h->fdec->i_qpplus1 = i_global_qp + 1;
3195
3196     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3197     {
3198         x264_reference_build_list_optimal( h );
3199         x264_reference_check_reorder( h );
3200     }
3201
3202     if( h->i_ref[0] )
3203         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3204
3205     /* ------------------------ Create slice header  ----------------------- */
3206     x264_slice_init( h, i_nal_type, i_global_qp );
3207
3208     /*------------------------- Weights -------------------------------------*/
3209     if( h->sh.i_type == SLICE_TYPE_B )
3210         x264_macroblock_bipred_init( h );
3211
3212     x264_weighted_pred_init( h );
3213
3214     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3215         h->i_frame_num++;
3216
3217     /* Write frame */
3218     h->i_threadslice_start = 0;
3219     h->i_threadslice_end = h->mb.i_mb_height;
3220     if( h->i_thread_frames > 1 )
3221     {
3222         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3223         h->b_thread_active = 1;
3224     }
3225     else if( h->param.b_sliced_threads )
3226     {
3227         if( x264_threaded_slices_write( h ) )
3228             return -1;
3229     }
3230     else
3231         if( (intptr_t)x264_slices_write( h ) )
3232             return -1;
3233
3234     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3235 }
3236
3237 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3238                                    x264_nal_t **pp_nal, int *pi_nal,
3239                                    x264_picture_t *pic_out )
3240 {
3241     char psz_message[80];
3242
3243     if( !h->param.b_sliced_threads && h->b_thread_active )
3244     {
3245         h->b_thread_active = 0;
3246         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3247             return -1;
3248     }
3249     if( !h->out.i_nal )
3250     {
3251         pic_out->i_type = X264_TYPE_AUTO;
3252         return 0;
3253     }
3254
3255     x264_emms();
3256
3257     /* generate buffering period sei and insert it into place */
3258     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3259     {
3260         x264_hrd_fullness( h );
3261         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3262         x264_sei_buffering_period_write( h, &h->out.bs );
3263         if( x264_nal_end( h ) )
3264            return -1;
3265         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3266         int idx = 0;
3267         while( h->out.nal[idx].i_type == NAL_AUD ||
3268                h->out.nal[idx].i_type == NAL_SPS ||
3269                h->out.nal[idx].i_type == NAL_PPS )
3270             idx++;
3271         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3272         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3273         h->out.nal[idx] = nal_tmp;
3274     }
3275
3276     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3277     if( frame_size < 0 )
3278         return -1;
3279
3280     /* Set output picture properties */
3281     pic_out->i_type = h->fenc->i_type;
3282
3283     pic_out->b_keyframe = h->fenc->b_keyframe;
3284     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3285
3286     pic_out->i_pts = h->fdec->i_pts;
3287     pic_out->i_dts = h->fdec->i_dts;
3288
3289     if( pic_out->i_pts < pic_out->i_dts )
3290         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3291
3292     pic_out->opaque = h->fenc->opaque;
3293
3294     pic_out->img.i_csp = h->fdec->i_csp;
3295 #if HIGH_BIT_DEPTH
3296     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3297 #endif
3298     pic_out->img.i_plane = h->fdec->i_plane;
3299     for( int i = 0; i < pic_out->img.i_plane; i++ )
3300     {
3301         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3302         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3303     }
3304
3305     x264_frame_push_unused( thread_current, h->fenc );
3306
3307     /* ---------------------- Update encoder state ------------------------- */
3308
3309     /* update rc */
3310     int filler = 0;
3311     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3312         return -1;
3313
3314     pic_out->hrd_timing = h->fenc->hrd_timing;
3315     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3316
3317     while( filler > 0 )
3318     {
3319         int f, overhead;
3320         overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3321         if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3322         {
3323             int next_size = filler - h->param.i_slice_max_size;
3324             int overflow = X264_MAX( overhead - next_size, 0 );
3325             f = h->param.i_slice_max_size - overhead - overflow;
3326         }
3327         else
3328             f = X264_MAX( 0, filler - overhead );
3329
3330         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3331         x264_filler_write( h, &h->out.bs, f );
3332         if( x264_nal_end( h ) )
3333             return -1;
3334         int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3335         if( total_size < 0 )
3336             return -1;
3337         frame_size += total_size;
3338         filler -= total_size;
3339     }
3340
3341     /* End bitstream, set output  */
3342     *pi_nal = h->out.i_nal;
3343     *pp_nal = h->out.nal;
3344
3345     h->out.i_nal = 0;
3346
3347     x264_noise_reduction_update( h );
3348
3349     /* ---------------------- Compute/Print statistics --------------------- */
3350     x264_thread_sync_stat( h, h->thread[0] );
3351
3352     /* Slice stat */
3353     h->stat.i_frame_count[h->sh.i_type]++;
3354     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3355     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3356
3357     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3358         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3359     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3360         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3361     for( int i = 0; i < 2; i++ )
3362         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3363     for( int i = 0; i < 6; i++ )
3364         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3365     for( int i = 0; i < 4; i++ )
3366         for( int j = 0; j < 13; j++ )
3367             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3368     if( h->sh.i_type != SLICE_TYPE_I )
3369         for( int i_list = 0; i_list < 2; i_list++ )
3370             for( int i = 0; i < X264_REF_MAX*2; i++ )
3371                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3372     for( int i = 0; i < 3; i++ )
3373         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3374     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3375     {
3376         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3377         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3378     }
3379     if( h->sh.i_type == SLICE_TYPE_B )
3380     {
3381         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3382         if( h->mb.b_direct_auto_write )
3383         {
3384             //FIXME somewhat arbitrary time constants
3385             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3386                 for( int i = 0; i < 2; i++ )
3387                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3388             for( int i = 0; i < 2; i++ )
3389                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3390         }
3391     }
3392     else
3393         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3394
3395     psz_message[0] = '\0';
3396     double dur = h->fenc->f_duration;
3397     h->stat.f_frame_duration[h->sh.i_type] += dur;
3398     if( h->param.analyse.b_psnr )
3399     {
3400         int64_t ssd[3] =
3401         {
3402             h->stat.frame.i_ssd[0],
3403             h->stat.frame.i_ssd[1],
3404             h->stat.frame.i_ssd[2],
3405         };
3406         int luma_size = h->param.i_width * h->param.i_height;
3407         int chroma_size = CHROMA_SIZE( luma_size );
3408         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3409         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3410         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3411         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3412
3413         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3414         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3415         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3416         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3417         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3418
3419         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3420                                                                     pic_out->prop.f_psnr[1],
3421                                                                     pic_out->prop.f_psnr[2] );
3422     }
3423
3424     if( h->param.analyse.b_ssim )
3425     {
3426         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3427         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3428         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3429                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3430     }
3431     psz_message[79] = '\0';
3432
3433     x264_log( h, X264_LOG_DEBUG,
3434                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3435               h->i_frame,
3436               h->fdec->f_qp_avg_aq,
3437               h->i_nal_ref_idc,
3438               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3439               h->fdec->i_poc,
3440               h->stat.frame.i_mb_count_i,
3441               h->stat.frame.i_mb_count_p,
3442               h->stat.frame.i_mb_count_skip,
3443               frame_size,
3444               psz_message );
3445
3446     // keep stats all in one place
3447     x264_thread_sync_stat( h->thread[0], h );
3448     // for the use of the next frame
3449     x264_thread_sync_stat( thread_current, h );
3450
3451 #ifdef DEBUG_MB_TYPE
3452 {
3453     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3454         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3455     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3456     {
3457         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3458             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3459         else
3460             fprintf( stderr, "? " );
3461
3462         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3463             fprintf( stderr, "\n" );
3464     }
3465 }
3466 #endif
3467
3468     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3469      * by freeing some of its pointers. */
3470     for( int i = 0; i < h->i_ref[0]; i++ )
3471         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3472         {
3473             x264_frame_push_blank_unused( h, h->fref[0][i] );
3474             h->fref[0][i] = 0;
3475         }
3476
3477     if( h->param.psz_dump_yuv )
3478         x264_frame_dump( h );
3479     x264_emms();
3480
3481     return frame_size;
3482 }
3483
3484 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3485 {
3486     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3487         b_print_pcm ? "..PCM" : "",
3488         i_mb_count[I_16x16]/ i_count,
3489         i_mb_count[I_8x8]  / i_count,
3490         i_mb_count[I_4x4]  / i_count );
3491     if( b_print_pcm )
3492         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3493 }
3494
3495 /****************************************************************************
3496  * x264_encoder_close:
3497  ****************************************************************************/
3498 void    x264_encoder_close  ( x264_t *h )
3499 {
3500     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3501     int64_t i_mb_count_size[2][7] = {{0}};
3502     char buf[200];
3503     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3504                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3505                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3506
3507     x264_lookahead_delete( h );
3508
3509     if( h->param.b_sliced_threads )
3510         x264_threadpool_wait_all( h );
3511     if( h->param.i_threads > 1 )
3512         x264_threadpool_delete( h->threadpool );
3513     if( h->param.i_lookahead_threads > 1 )
3514         x264_threadpool_delete( h->lookaheadpool );
3515     if( h->i_thread_frames > 1 )
3516     {
3517         for( int i = 0; i < h->i_thread_frames; i++ )
3518             if( h->thread[i]->b_thread_active )
3519             {
3520                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3521                 x264_frame_delete( h->thread[i]->fenc );
3522             }
3523
3524         x264_t *thread_prev = h->thread[h->i_thread_phase];
3525         x264_thread_sync_ratecontrol( h, thread_prev, h );
3526         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3527         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3528     }
3529     h->i_frame++;
3530
3531     /* Slices used and PSNR */
3532     for( int i = 0; i < 3; i++ )
3533     {
3534         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3535         int i_slice = slice_order[i];
3536
3537         if( h->stat.i_frame_count[i_slice] > 0 )
3538         {
3539             int i_count = h->stat.i_frame_count[i_slice];
3540             double dur =  h->stat.f_frame_duration[i_slice];
3541             if( h->param.analyse.b_psnr )
3542             {
3543                 x264_log( h, X264_LOG_INFO,
3544                           "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",
3545                           slice_type_to_char[i_slice],
3546                           i_count,
3547                           h->stat.f_frame_qp[i_slice] / i_count,
3548                           (double)h->stat.i_frame_size[i_slice] / i_count,
3549                           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,
3550                           h->stat.f_psnr_average[i_slice] / dur,
3551                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
3552             }
3553             else
3554             {
3555                 x264_log( h, X264_LOG_INFO,
3556                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
3557                           slice_type_to_char[i_slice],
3558                           i_count,
3559                           h->stat.f_frame_qp[i_slice] / i_count,
3560                           (double)h->stat.i_frame_size[i_slice] / i_count );
3561             }
3562         }
3563     }
3564     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
3565     {
3566         char *p = buf;
3567         int den = 0;
3568         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
3569         for( int i = 0; i <= h->param.i_bframe; i++ )
3570             den += (i+1) * h->stat.i_consecutive_bframes[i];
3571         for( int i = 0; i <= h->param.i_bframe; i++ )
3572             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
3573         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
3574     }
3575
3576     for( int i_type = 0; i_type < 2; i_type++ )
3577         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3578         {
3579             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
3580             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
3581         }
3582
3583     /* MB types used */
3584     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
3585     {
3586         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
3587         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
3588         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3589         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
3590     }
3591     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3592     {
3593         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
3594         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
3595         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
3596         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3597         x264_log( h, X264_LOG_INFO,
3598                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
3599                   buf,
3600                   i_mb_size[PIXEL_16x16] / (i_count*4),
3601                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3602                   i_mb_size[PIXEL_8x8] / (i_count*4),
3603                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
3604                   i_mb_size[PIXEL_4x4] / (i_count*4),
3605                   i_mb_count[P_SKIP] / i_count );
3606     }
3607     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
3608     {
3609         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
3610         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
3611         double i_mb_list_count;
3612         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
3613         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
3614         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
3615         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3616             for( int j = 0; j < 2; j++ )
3617             {
3618                 int l0 = x264_mb_type_list_table[i][0][j];
3619                 int l1 = x264_mb_type_list_table[i][1][j];
3620                 if( l0 || l1 )
3621                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
3622             }
3623         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
3624         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
3625         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
3626         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
3627         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
3628         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
3629                  i_mb_size[PIXEL_16x16] / (i_count*4),
3630                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
3631                  i_mb_size[PIXEL_8x8] / (i_count*4),
3632                  i_mb_count[B_DIRECT] / i_count,
3633                  i_mb_count[B_SKIP]   / i_count );
3634         if( i_mb_list_count != 0 )
3635             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
3636                      list_count[0] / i_mb_list_count,
3637                      list_count[1] / i_mb_list_count,
3638                      list_count[2] / i_mb_list_count );
3639         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
3640     }
3641
3642     x264_ratecontrol_summary( h );
3643
3644     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 )
3645     {
3646 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
3647 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
3648         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
3649         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
3650                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
3651         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
3652         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
3653                        + SUM3b( h->stat.i_mb_count, B_SKIP );
3654         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
3655                             h->stat.i_frame_count[SLICE_TYPE_P] +
3656                             h->stat.i_frame_count[SLICE_TYPE_B];
3657         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
3658         int64_t i_inter = i_mb_count - i_skip - i_intra;
3659         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
3660                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
3661                                 h->stat.f_frame_duration[SLICE_TYPE_B];
3662         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
3663
3664         if( PARAM_INTERLACED )
3665         {
3666             char *fieldstats = buf;
3667             fieldstats[0] = 0;
3668             if( i_inter )
3669                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
3670             if( i_skip )
3671                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
3672             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
3673                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
3674         }
3675
3676         if( h->pps->b_transform_8x8_mode )
3677         {
3678             buf[0] = 0;
3679             if( h->stat.i_mb_count_8x8dct[0] )
3680                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
3681             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
3682         }
3683
3684         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
3685             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
3686             && h->stat.i_frame_count[SLICE_TYPE_B] )
3687         {
3688             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
3689                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
3690                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
3691         }
3692
3693         buf[0] = 0;
3694         int csize = CHROMA444 ? 4 : 1;
3695         if( i_mb_count != i_all_intra )
3696             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
3697                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
3698                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
3699                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
3700         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
3701                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
3702                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
3703                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
3704                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
3705
3706         int64_t fixed_pred_modes[4][9] = {{0}};
3707         int64_t sum_pred_modes[4] = {0};
3708         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
3709         {
3710             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
3711             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
3712         }
3713         if( sum_pred_modes[0] )
3714             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3715                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
3716                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
3717                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
3718                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
3719         for( int i = 1; i <= 2; i++ )
3720         {
3721             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
3722             {
3723                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
3724                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
3725             }
3726             if( sum_pred_modes[i] )
3727                 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,
3728                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
3729                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
3730                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
3731                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
3732                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
3733                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
3734                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
3735                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
3736                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
3737         }
3738         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
3739         {
3740             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
3741             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
3742         }
3743         if( sum_pred_modes[3] && !CHROMA444 )
3744             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
3745                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
3746                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
3747                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
3748                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
3749
3750         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
3751             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
3752                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
3753                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
3754
3755         for( int i_list = 0; i_list < 2; i_list++ )
3756             for( int i_slice = 0; i_slice < 2; i_slice++ )
3757             {
3758                 char *p = buf;
3759                 int64_t i_den = 0;
3760                 int i_max = 0;
3761                 for( int i = 0; i < X264_REF_MAX*2; i++ )
3762                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
3763                     {
3764                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
3765                         i_max = i;
3766                     }
3767                 if( i_max == 0 )
3768                     continue;
3769                 for( int i = 0; i <= i_max; i++ )
3770                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
3771                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
3772             }
3773
3774         if( h->param.analyse.b_ssim )
3775         {
3776             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
3777             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
3778         }
3779         if( h->param.analyse.b_psnr )
3780         {
3781             x264_log( h, X264_LOG_INFO,
3782                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
3783                       SUM3( h->stat.f_psnr_mean_y ) / duration,
3784                       SUM3( h->stat.f_psnr_mean_u ) / duration,
3785                       SUM3( h->stat.f_psnr_mean_v ) / duration,
3786                       SUM3( h->stat.f_psnr_average ) / duration,
3787                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
3788                       f_bitrate );
3789         }
3790         else
3791             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
3792     }
3793
3794     /* rc */
3795     x264_ratecontrol_delete( h );
3796
3797     /* param */
3798     if( h->param.rc.psz_stat_out )
3799         free( h->param.rc.psz_stat_out );
3800     if( h->param.rc.psz_stat_in )
3801         free( h->param.rc.psz_stat_in );
3802
3803     x264_cqm_delete( h );
3804     x264_free( h->nal_buffer );
3805     x264_analyse_free_costs( h );
3806
3807     if( h->i_thread_frames > 1 )
3808         h = h->thread[h->i_thread_phase];
3809
3810     /* frames */
3811     x264_frame_delete_list( h->frames.unused[0] );
3812     x264_frame_delete_list( h->frames.unused[1] );
3813     x264_frame_delete_list( h->frames.current );
3814     x264_frame_delete_list( h->frames.blank_unused );
3815
3816     h = h->thread[0];
3817
3818     for( int i = 0; i < h->i_thread_frames; i++ )
3819         if( h->thread[i]->b_thread_active )
3820             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
3821                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
3822                     x264_frame_delete( h->thread[i]->fref[0][j] );
3823
3824     if( h->param.i_lookahead_threads > 1 )
3825         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
3826             x264_free( h->lookahead_thread[i] );
3827
3828     for( int i = h->param.i_threads - 1; i >= 0; i-- )
3829     {
3830         x264_frame_t **frame;
3831
3832         if( !h->param.b_sliced_threads || i == 0 )
3833         {
3834             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
3835             {
3836                 assert( (*frame)->i_reference_count > 0 );
3837                 (*frame)->i_reference_count--;
3838                 if( (*frame)->i_reference_count == 0 )
3839                     x264_frame_delete( *frame );
3840             }
3841             frame = &h->thread[i]->fdec;
3842             if( *frame )
3843             {
3844                 assert( (*frame)->i_reference_count > 0 );
3845                 (*frame)->i_reference_count--;
3846                 if( (*frame)->i_reference_count == 0 )
3847                     x264_frame_delete( *frame );
3848             }
3849             x264_macroblock_cache_free( h->thread[i] );
3850         }
3851         x264_macroblock_thread_free( h->thread[i], 0 );
3852         x264_free( h->thread[i]->out.p_bitstream );
3853         x264_free( h->thread[i]->out.nal );
3854         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
3855         x264_pthread_cond_destroy( &h->thread[i]->cv );
3856         x264_free( h->thread[i] );
3857     }
3858 }
3859
3860 int x264_encoder_delayed_frames( x264_t *h )
3861 {
3862     int delayed_frames = 0;
3863     if( h->i_thread_frames > 1 )
3864     {
3865         for( int i = 0; i < h->i_thread_frames; i++ )
3866             delayed_frames += h->thread[i]->b_thread_active;
3867         h = h->thread[h->i_thread_phase];
3868     }
3869     for( int i = 0; h->frames.current[i]; i++ )
3870         delayed_frames++;
3871     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
3872     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3873     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
3874     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
3875     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
3876     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3877     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
3878     return delayed_frames;
3879 }
3880
3881 int x264_encoder_maximum_delayed_frames( x264_t *h )
3882 {
3883     return h->frames.i_delay;
3884 }