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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2016 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *
  24  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include "common/common.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34 #include "me.h"
  35 #if HAVE_INTEL_DISPATCHER
  36 #include "extras/intel_dispatcher.h"
  37 #endif
  38
  39 //#define DEBUG_MB_TYPE
  40
  41 #define bs_write_ue bs_write_ue_big
  42
  43 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  44                                    x264_nal_t **pp_nal, int *pi_nal,
  45                                    x264_picture_t *pic_out );
  46
  47 /****************************************************************************
  48  *
  49  ******************************* x264 libs **********************************
  50  *
  51  ****************************************************************************/
  52 static double x264_psnr( double sqe, double size )
  53 {
  54     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  55     if( mse <= 0.0000000001 ) /* Max 100dB */
  56         return 100;
  57
  58     return -10.0 * log10( mse );
  59 }
  60
  61 static double x264_ssim( double ssim )
  62 {
  63     double inv_ssim = 1 - ssim;
  64     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
  65         return 100;
  66
  67     return -10.0 * log10( inv_ssim );
  68 }
  69
  70 static int x264_threadpool_wait_all( x264_t *h )
  71 {
  72     for( int i = 0; i < h->param.i_threads; i++ )
  73         if( h->thread[i]->b_thread_active )
  74         {
  75             h->thread[i]->b_thread_active = 0;
  76             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
  77                 return -1;
  78         }
  79     return 0;
  80 }
  81
  82 static void x264_frame_dump( x264_t *h )
  83 {
  84     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
  85     if( !f )
  86         return;
  87
  88     /* Wait for the threads to finish deblocking */
  89     if( h->param.b_sliced_threads )
  90         x264_threadpool_wait_all( h );
  91
  92     /* Write the frame in display order */
  93     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
  94     if( !fseek( f, (int64_t)h->fdec->i_frame * frame_size, SEEK_SET ) )
  95     {
  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             if( planeu )
 105             {
 106                 pixel *planev = planeu + cw*ch + 16;
 107                 h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
 108                 fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
 109                 fwrite( planev, 1, cw*ch*sizeof(pixel), f );
 110                 x264_free( planeu );
 111             }
 112         }
 113     }
 114     fclose( f );
 115 }
 116
 117 /* Fill "default" values */
 118 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 119                                     x264_sps_t *sps, x264_pps_t *pps,
 120                                     int i_idr_pic_id, int i_frame, int i_qp )
 121 {
 122     x264_param_t *param = &h->param;
 123
 124     /* First we fill all fields */
 125     sh->sps = sps;
 126     sh->pps = pps;
 127
 128     sh->i_first_mb  = 0;
 129     sh->i_last_mb   = h->mb.i_mb_count - 1;
 130     sh->i_pps_id    = pps->i_id;
 131
 132     sh->i_frame_num = i_frame;
 133
 134     sh->b_mbaff = PARAM_INTERLACED;
 135     sh->b_field_pic = 0;    /* no field support for now */
 136     sh->b_bottom_field = 0; /* not yet used */
 137
 138     sh->i_idr_pic_id = i_idr_pic_id;
 139
 140     /* poc stuff, fixed later */
 141     sh->i_poc = 0;
 142     sh->i_delta_poc_bottom = 0;
 143     sh->i_delta_poc[0] = 0;
 144     sh->i_delta_poc[1] = 0;
 145
 146     sh->i_redundant_pic_cnt = 0;
 147
 148     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 149                                 && h->param.i_bframe
 150                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 151
 152     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 153     {
 154         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
 155         {
 156             if( h->mb.b_direct_auto_write )
 157                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 158             else
 159                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 160         }
 161         else
 162         {
 163             h->mb.b_direct_auto_write = 0;
 164             sh->b_direct_spatial_mv_pred = 1;
 165         }
 166     }
 167     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 168
 169     sh->b_num_ref_idx_override = 0;
 170     sh->i_num_ref_idx_l0_active = 1;
 171     sh->i_num_ref_idx_l1_active = 1;
 172
 173     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
 174     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
 175
 176     /* If the ref list isn't in the default order, construct reordering header */
 177     for( int list = 0; list < 2; list++ )
 178     {
 179         if( sh->b_ref_pic_list_reordering[list] )
 180         {
 181             int pred_frame_num = i_frame;
 182             for( int i = 0; i < h->i_ref[list]; i++ )
 183             {
 184                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
 185                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
 186                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 187                 pred_frame_num = h->fref[list][i]->i_frame_num;
 188             }
 189         }
 190     }
 191
 192     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 193
 194     sh->i_qp = SPEC_QP(i_qp);
 195     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
 196     sh->b_sp_for_swidth = 0;
 197     sh->i_qs_delta = 0;
 198
 199     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 200     /* If effective qp <= 15, deblocking would have no effect anyway */
 201     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 202         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 203     else
 204         sh->i_disable_deblocking_filter_idc = 1;
 205     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 206     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 207 }
 208
 209 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 210 {
 211     if( sh->b_mbaff )
 212     {
 213         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
 214         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
 215         assert( (first_y&1) == 0 );
 216         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
 217     }
 218     else
 219         bs_write_ue( s, sh->i_first_mb );
 220
 221     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 222     bs_write_ue( s, sh->i_pps_id );
 223     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 224
 225     if( !sh->sps->b_frame_mbs_only )
 226     {
 227         bs_write1( s, sh->b_field_pic );
 228         if( sh->b_field_pic )
 229             bs_write1( s, sh->b_bottom_field );
 230     }
 231
 232     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 233         bs_write_ue( s, sh->i_idr_pic_id );
 234
 235     if( sh->sps->i_poc_type == 0 )
 236     {
 237         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 238         if( sh->pps->b_pic_order && !sh->b_field_pic )
 239             bs_write_se( s, sh->i_delta_poc_bottom );
 240     }
 241
 242     if( sh->pps->b_redundant_pic_cnt )
 243         bs_write_ue( s, sh->i_redundant_pic_cnt );
 244
 245     if( sh->i_type == SLICE_TYPE_B )
 246         bs_write1( s, sh->b_direct_spatial_mv_pred );
 247
 248     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 249     {
 250         bs_write1( s, sh->b_num_ref_idx_override );
 251         if( sh->b_num_ref_idx_override )
 252         {
 253             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 254             if( sh->i_type == SLICE_TYPE_B )
 255                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 256         }
 257     }
 258
 259     /* ref pic list reordering */
 260     if( sh->i_type != SLICE_TYPE_I )
 261     {
 262         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
 263         if( sh->b_ref_pic_list_reordering[0] )
 264         {
 265             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 266             {
 267                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 268                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 269             }
 270             bs_write_ue( s, 3 );
 271         }
 272     }
 273     if( sh->i_type == SLICE_TYPE_B )
 274     {
 275         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
 276         if( sh->b_ref_pic_list_reordering[1] )
 277         {
 278             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 279             {
 280                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 281                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 282             }
 283             bs_write_ue( s, 3 );
 284         }
 285     }
 286
 287     sh->b_weighted_pred = 0;
 288     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 289     {
 290         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
 291         /* pred_weight_table() */
 292         bs_write_ue( s, sh->weight[0][0].i_denom );
 293         bs_write_ue( s, sh->weight[0][1].i_denom );
 294         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 295         {
 296             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 297             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 298             bs_write1( s, luma_weight_l0_flag );
 299             if( luma_weight_l0_flag )
 300             {
 301                 bs_write_se( s, sh->weight[i][0].i_scale );
 302                 bs_write_se( s, sh->weight[i][0].i_offset );
 303             }
 304             bs_write1( s, chroma_weight_l0_flag );
 305             if( chroma_weight_l0_flag )
 306             {
 307                 for( int j = 1; j < 3; j++ )
 308                 {
 309                     bs_write_se( s, sh->weight[i][j].i_scale );
 310                     bs_write_se( s, sh->weight[i][j].i_offset );
 311                 }
 312             }
 313         }
 314     }
 315     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 316     {
 317       /* TODO */
 318     }
 319
 320     if( i_nal_ref_idc != 0 )
 321     {
 322         if( sh->i_idr_pic_id >= 0 )
 323         {
 324             bs_write1( s, 0 );  /* no output of prior pics flag */
 325             bs_write1( s, 0 );  /* long term reference flag */
 326         }
 327         else
 328         {
 329             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 330             if( sh->i_mmco_command_count > 0 )
 331             {
 332                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 333                 {
 334                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 335                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 336                 }
 337                 bs_write_ue( s, 0 ); /* end command list */
 338             }
 339         }
 340     }
 341
 342     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 343         bs_write_ue( s, sh->i_cabac_init_idc );
 344
 345     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 346
 347     if( sh->pps->b_deblocking_filter_control )
 348     {
 349         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 350         if( sh->i_disable_deblocking_filter_idc != 1 )
 351         {
 352             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 353             bs_write_se( s, sh->i_beta_offset >> 1 );
 354         }
 355     }
 356 }
 357
 358 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 359 /* reallocate, adding an arbitrary amount of space. */
 360 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
 361 {
 362     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
 363         (h->out.bs.p_end - h->out.bs.p < size) )
 364     {
 365         int buf_size = h->out.i_bitstream + size;
 366         uint8_t *buf = x264_malloc( buf_size );
 367         if( !buf )
 368             return -1;
 369         int aligned_size = h->out.i_bitstream & ~15;
 370         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
 371         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
 372
 373         intptr_t delta = buf - h->out.p_bitstream;
 374
 375         h->out.bs.p_start += delta;
 376         h->out.bs.p += delta;
 377         h->out.bs.p_end = buf + buf_size;
 378
 379         h->cabac.p_start += delta;
 380         h->cabac.p += delta;
 381         h->cabac.p_end = buf + buf_size;
 382
 383         for( int i = 0; i <= i_nal; i++ )
 384             h->out.nal[i].p_payload += delta;
 385
 386         x264_free( h->out.p_bitstream );
 387         h->out.p_bitstream = buf;
 388         h->out.i_bitstream = buf_size;
 389     }
 390     return 0;
 391 }
 392
 393 static int x264_bitstream_check_buffer( x264_t *h )
 394 {
 395     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
 396     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
 397 }
 398
 399 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
 400 {
 401     filler += 32; // add padding for safety
 402     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
 403 }
 404
 405 #if HAVE_THREAD
 406 static void x264_encoder_thread_init( x264_t *h )
 407 {
 408     if( h->param.i_sync_lookahead )
 409         x264_lower_thread_priority( 10 );
 410 }
 411 #endif
 412
 413 /****************************************************************************
 414  *
 415  ****************************************************************************
 416  ****************************** External API*********************************
 417  ****************************************************************************
 418  *
 419  ****************************************************************************/
 420
 421 static int x264_validate_parameters( x264_t *h, int b_open )
 422 {
 423     if( !h->param.pf_log )
 424     {
 425         x264_log( NULL, X264_LOG_ERROR, "pf_log not set! did you forget to call x264_param_default?\n" );
 426         return -1;
 427     }
 428
 429 #if HAVE_MMX
 430     if( b_open )
 431     {
 432         int cpuflags = x264_cpu_detect();
 433         int fail = 0;
 434 #ifdef __SSE__
 435         if( !(cpuflags & X264_CPU_SSE) )
 436         {
 437             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
 438             fail = 1;
 439         }
 440 #else
 441         if( !(cpuflags & X264_CPU_MMX2) )
 442         {
 443             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
 444             fail = 1;
 445         }
 446 #endif
 447         if( !fail && !(cpuflags & X264_CPU_CMOV) )
 448         {
 449             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
 450             fail = 1;
 451         }
 452         if( fail )
 453         {
 454             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
 455             return -1;
 456         }
 457     }
 458 #endif
 459
 460 #if HAVE_INTERLACED
 461     h->param.b_interlaced = !!PARAM_INTERLACED;
 462 #else
 463     if( h->param.b_interlaced )
 464     {
 465         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
 466         return -1;
 467     }
 468 #endif
 469
 470     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 471     {
 472         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 473                   h->param.i_width, h->param.i_height );
 474         return -1;
 475     }
 476
 477     int i_csp = h->param.i_csp & X264_CSP_MASK;
 478 #if X264_CHROMA_FORMAT
 479     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 )
 480     {
 481         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
 482         return -1;
 483     }
 484     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 )
 485     {
 486         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
 487         return -1;
 488     }
 489     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
 490     {
 491         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
 492         return -1;
 493     }
 494 #endif
 495     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 496     {
 497         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/NV21/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
 498         return -1;
 499     }
 500
 501     int w_mod = i_csp < X264_CSP_I444 ? 2 : 1;
 502     int h_mod = (i_csp < X264_CSP_I422 ? 2 : 1) << PARAM_INTERLACED;
 503     if( h->param.i_width % w_mod )
 504     {
 505         x264_log( h, X264_LOG_ERROR, "width not divisible by %d (%dx%d)\n",
 506                   w_mod, h->param.i_width, h->param.i_height );
 507         return -1;
 508     }
 509     if( h->param.i_height % h_mod )
 510     {
 511         x264_log( h, X264_LOG_ERROR, "height not divisible by %d (%dx%d)\n",
 512                   h_mod, h->param.i_width, h->param.i_height );
 513         return -1;
 514     }
 515
 516     if( h->param.crop_rect.i_left   >= h->param.i_width ||
 517         h->param.crop_rect.i_right  >= h->param.i_width ||
 518         h->param.crop_rect.i_top    >= h->param.i_height ||
 519         h->param.crop_rect.i_bottom >= h->param.i_height ||
 520         h->param.crop_rect.i_left + h->param.crop_rect.i_right  >= h->param.i_width ||
 521         h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom >= h->param.i_height )
 522     {
 523         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 524                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 525         return -1;
 526     }
 527     if( h->param.crop_rect.i_left % w_mod || h->param.crop_rect.i_right  % w_mod ||
 528         h->param.crop_rect.i_top  % h_mod || h->param.crop_rect.i_bottom % h_mod )
 529     {
 530         x264_log( h, X264_LOG_ERROR, "crop-rect %u,%u,%u,%u not divisible by %dx%d\n", h->param.crop_rect.i_left,
 531                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom, w_mod, h_mod );
 532         return -1;
 533     }
 534
 535     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
 536     {
 537         h->param.vui.i_sar_width = 0;
 538         h->param.vui.i_sar_height = 0;
 539     }
 540
 541     if( h->param.i_threads == X264_THREADS_AUTO )
 542     {
 543         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 544         /* Avoid too many threads as they don't improve performance and
 545          * complicate VBV. Capped at an arbitrary 2 rows per thread. */
 546         int max_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 2 );
 547         h->param.i_threads = X264_MIN( h->param.i_threads, max_threads );
 548     }
 549     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
 550     if( h->param.i_threads > 1 )
 551     {
 552 #if !HAVE_THREAD
 553         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 554         h->param.i_threads = 1;
 555 #endif
 556         /* Avoid absurdly small thread slices as they can reduce performance
 557          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 558         if( h->param.b_sliced_threads )
 559             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
 560     }
 561     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 562     if( h->param.i_threads == 1 )
 563     {
 564         h->param.b_sliced_threads = 0;
 565         h->param.i_lookahead_threads = 1;
 566     }
 567     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 568     if( h->i_thread_frames > 1 )
 569         h->param.nalu_process = NULL;
 570
 571     if( h->param.b_opencl )
 572     {
 573 #if !HAVE_OPENCL
 574         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
 575         h->param.b_opencl = 0;
 576 #elif BIT_DEPTH > 8
 577         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
 578         h->param.b_opencl = 0;
 579 #else
 580         if( h->param.i_width < 32 || h->param.i_height < 32 )
 581         {
 582             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
 583             h->param.b_opencl = 0;
 584         }
 585 #endif
 586         if( h->param.opencl_device_id && h->param.i_opencl_device )
 587         {
 588             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
 589             h->param.i_opencl_device = 0;
 590         }
 591     }
 592
 593     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 594     if( h->param.i_keyint_max == 1 )
 595     {
 596         h->param.b_intra_refresh = 0;
 597         h->param.analyse.i_weighted_pred = 0;
 598         h->param.i_frame_reference = 1;
 599         h->param.i_dpb_size = 1;
 600     }
 601
 602     if( h->param.i_frame_packing < -1 || h->param.i_frame_packing > 7 )
 603     {
 604         x264_log( h, X264_LOG_WARNING, "ignoring unknown frame packing value\n" );
 605         h->param.i_frame_packing = -1;
 606     }
 607     if( h->param.i_frame_packing == 7 &&
 608         ((h->param.i_width - h->param.crop_rect.i_left - h->param.crop_rect.i_right)  % 3 ||
 609          (h->param.i_height - h->param.crop_rect.i_top - h->param.crop_rect.i_bottom) % 3) )
 610     {
 611         x264_log( h, X264_LOG_ERROR, "cropped resolution %dx%d not compatible with tile format frame packing\n",
 612                   h->param.i_width - h->param.crop_rect.i_left - h->param.crop_rect.i_right,
 613                   h->param.i_height - h->param.crop_rect.i_top - h->param.crop_rect.i_bottom );
 614         return -1;
 615     }
 616
 617     /* Detect default ffmpeg settings and terminate with an error. */
 618     if( b_open )
 619     {
 620         int score = 0;
 621         score += h->param.analyse.i_me_range == 0;
 622         score += h->param.rc.i_qp_step == 3;
 623         score += h->param.i_keyint_max == 12;
 624         score += h->param.rc.i_qp_min == 2;
 625         score += h->param.rc.i_qp_max == 31;
 626         score += h->param.rc.f_qcompress == 0.5;
 627         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 628         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 629         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 630         if( score >= 5 )
 631         {
 632             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 633             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 634             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 635             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 636             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 637             return -1;
 638         }
 639     }
 640
 641     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 642     {
 643         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 644         return -1;
 645     }
 646
 647     if( PARAM_INTERLACED )
 648         h->param.b_pic_struct = 1;
 649
 650     if( h->param.i_avcintra_class )
 651     {
 652         if( BIT_DEPTH != 10 )
 653         {
 654             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
 655             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
 656             return -1;
 657         }
 658
 659         int type = h->param.i_avcintra_class == 200 ? 2 :
 660                    h->param.i_avcintra_class == 100 ? 1 :
 661                    h->param.i_avcintra_class == 50 ? 0 : -1;
 662         if( type < 0 )
 663         {
 664             x264_log( h, X264_LOG_ERROR, "Invalid AVC-Intra class\n" );
 665             return -1;
 666         }
 667
 668         /* [50/100/200][res][fps] */
 669         static const struct
 670         {
 671             uint16_t fps_num;
 672             uint16_t fps_den;
 673             uint8_t interlaced;
 674             uint16_t frame_size;
 675             const uint8_t *cqm_4ic;
 676             const uint8_t *cqm_8iy;
 677         } avcintra_lut[3][2][7] =
 678         {
 679             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 680               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 681               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 682               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 683               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
 684              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 685               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 686               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 687               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 688               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 689               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 690               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
 691             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 692               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 693               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 694               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 695               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 696              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 697               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 698               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 699               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 700               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 701               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 702               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}},
 703             {{{ 60000, 1001, 0, 3724, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 704               {    50,    1, 0, 4472, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 705              {{ 30000, 1001, 1, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 706               {    25,    1, 1, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 707               { 60000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 708               { 30000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 709               {    50,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 710               {    25,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 711               { 24000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
 712         };
 713
 714         int res = -1;
 715         if( i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 && !type )
 716         {
 717             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
 718             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
 719         }
 720         else if( i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 && type )
 721         {
 722             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
 723             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
 724         }
 725         else
 726         {
 727             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra %d\n", h->param.i_avcintra_class );
 728             return -1;
 729         }
 730
 731         if( res < 0 )
 732         {
 733             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %d\n",
 734                       h->param.i_width, h->param.i_height, h->param.i_avcintra_class );
 735             return -1;
 736         }
 737
 738         if( h->param.nalu_process )
 739         {
 740             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
 741             return -1;
 742         }
 743
 744         if( !h->param.b_repeat_headers )
 745         {
 746             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
 747             return -1;
 748         }
 749
 750         int i;
 751         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
 752         x264_reduce_fraction( &fps_num, &fps_den );
 753         for( i = 0; i < 7; i++ )
 754         {
 755             if( avcintra_lut[type][res][i].fps_num == fps_num &&
 756                 avcintra_lut[type][res][i].fps_den == fps_den &&
 757                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
 758             {
 759                 break;
 760             }
 761         }
 762         if( i == 7 )
 763         {
 764             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
 765                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
 766             return -1;
 767         }
 768
 769         h->param.i_keyint_max = 1;
 770         h->param.b_intra_refresh = 0;
 771         h->param.analyse.i_weighted_pred = 0;
 772         h->param.i_frame_reference = 1;
 773         h->param.i_dpb_size = 1;
 774
 775         h->param.b_bluray_compat = 0;
 776         h->param.b_vfr_input = 0;
 777         h->param.b_aud = 1;
 778         h->param.vui.i_chroma_loc = 0;
 779         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 780         h->param.b_deblocking_filter = 0;
 781         h->param.b_stitchable = 1;
 782         h->param.b_pic_struct = 0;
 783         h->param.analyse.b_transform_8x8 = 1;
 784         h->param.analyse.intra = X264_ANALYSE_I8x8;
 785         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
 786         h->param.b_cabac = !type;
 787         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
 788         h->param.rc.i_vbv_max_bitrate =
 789         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
 790         h->param.rc.i_rc_method = X264_RC_ABR;
 791         h->param.rc.f_vbv_buffer_init = 1.0;
 792         h->param.rc.b_filler = 1;
 793         h->param.i_cqm_preset = X264_CQM_CUSTOM;
 794         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
 795         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
 796         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
 797
 798         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
 799         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
 800         h->param.i_slice_max_size = 0;
 801         /* The slice structure only allows a maximum of 2 threads for 1080i/p
 802          * and 1 or 5 threads for 720p */
 803         if( h->param.b_sliced_threads )
 804         {
 805             if( res )
 806                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
 807             else
 808             {
 809                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
 810                 if( h->param.i_threads < 5 )
 811                     h->param.i_threads = 1;
 812             }
 813         }
 814
 815         if( type )
 816             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
 817         else
 818         {
 819             h->param.vui.i_sar_width  = 4;
 820             h->param.vui.i_sar_height = 3;
 821         }
 822
 823         /* Official encoder doesn't appear to go under 13
 824          * and Avid cannot handle negative QPs */
 825         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET + 1 );
 826     }
 827
 828     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 829     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
 830     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 831     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
 832     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
 833     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
 834     if( h->param.rc.i_rc_method == X264_RC_CRF )
 835     {
 836         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 837         h->param.rc.i_bitrate = 0;
 838     }
 839     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 840         && h->param.rc.i_qp_constant == 0 )
 841     {
 842         h->mb.b_lossless = 1;
 843         h->param.i_cqm_preset = X264_CQM_FLAT;
 844         h->param.psz_cqm_file = NULL;
 845         h->param.rc.i_rc_method = X264_RC_CQP;
 846         h->param.rc.f_ip_factor = 1;
 847         h->param.rc.f_pb_factor = 1;
 848         h->param.analyse.b_psnr = 0;
 849         h->param.analyse.b_ssim = 0;
 850         h->param.analyse.i_chroma_qp_offset = 0;
 851         h->param.analyse.i_trellis = 0;
 852         h->param.analyse.b_fast_pskip = 0;
 853         h->param.analyse.i_noise_reduction = 0;
 854         h->param.analyse.b_psy = 0;
 855         h->param.i_bframe = 0;
 856         /* 8x8dct is not useful without RD in CAVLC lossless */
 857         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 858             h->param.analyse.b_transform_8x8 = 0;
 859         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
 860         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
 861     }
 862     if( h->param.rc.i_rc_method == X264_RC_CQP )
 863     {
 864         float qp_p = h->param.rc.i_qp_constant;
 865         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 866         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 867         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 868         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 869         h->param.rc.i_aq_mode = 0;
 870         h->param.rc.b_mb_tree = 0;
 871         h->param.rc.i_bitrate = 0;
 872     }
 873     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 874     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 875     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
 876     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
 877     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
 878     {
 879         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
 880         return -1;
 881     }
 882     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
 883     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
 884     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
 885     if( h->param.rc.i_vbv_buffer_size )
 886     {
 887         if( h->param.rc.i_rc_method == X264_RC_CQP )
 888         {
 889             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 890             h->param.rc.i_vbv_max_bitrate = 0;
 891             h->param.rc.i_vbv_buffer_size = 0;
 892         }
 893         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 894         {
 895             if( h->param.rc.i_rc_method == X264_RC_ABR )
 896             {
 897                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 898                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 899             }
 900             else
 901             {
 902                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 903                 h->param.rc.i_vbv_buffer_size = 0;
 904             }
 905         }
 906         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 907                  h->param.rc.i_rc_method == X264_RC_ABR )
 908         {
 909             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 910             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
 911         }
 912     }
 913     else if( h->param.rc.i_vbv_max_bitrate )
 914     {
 915         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 916         h->param.rc.i_vbv_max_bitrate = 0;
 917     }
 918
 919     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 920     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 921     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
 922     if( h->param.i_slice_max_mbs )
 923         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
 924     else if( !h->param.i_slice_max_size )
 925         h->param.i_slice_min_mbs = 0;
 926     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
 927     {
 928         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
 929         h->param.i_slice_min_mbs = 0;
 930     }
 931     int mb_width = (h->param.i_width+15)/16;
 932     if( h->param.i_slice_min_mbs > mb_width )
 933     {
 934         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
 935         h->param.i_slice_min_mbs = mb_width;
 936     }
 937
 938     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
 939     if( h->param.b_sliced_threads )
 940         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 941     else
 942     {
 943         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 944         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 945             h->param.i_slice_count = 0;
 946     }
 947     if( h->param.i_slice_count_max > 0 )
 948         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
 949
 950     if( h->param.b_bluray_compat )
 951     {
 952         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
 953         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
 954         h->param.b_aud = 1;
 955         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
 956         h->param.i_slice_max_size = 0;
 957         h->param.i_slice_max_mbs = 0;
 958         h->param.b_intra_refresh = 0;
 959         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
 960         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
 961         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
 962         h->param.i_keyint_min = 1;
 963         /* Due to the proliferation of broken players that don't handle dupes properly. */
 964         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
 965         if( h->param.b_fake_interlaced )
 966             h->param.b_pic_struct = 1;
 967     }
 968
 969     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 970     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 971     if( h->param.i_scenecut_threshold < 0 )
 972         h->param.i_scenecut_threshold = 0;
 973     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
 974     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 975     {
 976         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 977         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 978     }
 979     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 980     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 981     if( h->param.i_bframe <= 1 )
 982         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 983     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 984     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
 985     if( !h->param.i_bframe )
 986     {
 987         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 988         h->param.analyse.i_direct_mv_pred = 0;
 989         h->param.analyse.b_weighted_bipred = 0;
 990         h->param.b_open_gop = 0;
 991     }
 992     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 993     {
 994         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 995         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 996     }
 997     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 998     {
 999         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
1000         h->param.i_frame_reference = 1;
1001         h->param.i_dpb_size = 1;
1002     }
1003     if( h->param.b_intra_refresh && h->param.b_open_gop )
1004     {
1005         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
1006         h->param.b_open_gop = 0;
1007     }
1008     if( !h->param.i_fps_num || !h->param.i_fps_den )
1009     {
1010         h->param.i_fps_num = 25;
1011         h->param.i_fps_den = 1;
1012     }
1013     float fps = (float)h->param.i_fps_num / h->param.i_fps_den;
1014     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
1015         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, (int)fps );
1016     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
1017     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
1018     {
1019         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
1020         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
1021         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
1022     }
1023
1024     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
1025     {
1026         h->param.i_timebase_num = h->param.i_fps_den;
1027         h->param.i_timebase_den = h->param.i_fps_num;
1028     }
1029
1030     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
1031     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
1032         h->param.rc.b_mb_tree = 0;
1033     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
1034         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
1035     {
1036         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
1037         h->param.rc.b_mb_tree = 0;
1038     }
1039     if( b_open && h->param.rc.b_stat_read )
1040         h->param.rc.i_lookahead = 0;
1041 #if HAVE_THREAD
1042     if( h->param.i_sync_lookahead < 0 )
1043         h->param.i_sync_lookahead = h->param.i_bframe + 1;
1044     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
1045     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
1046         h->param.i_sync_lookahead = 0;
1047 #else
1048     h->param.i_sync_lookahead = 0;
1049 #endif
1050
1051     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
1052     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
1053     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1054     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1055
1056     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1057
1058     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1059         h->param.i_cqm_preset = X264_CQM_FLAT;
1060
1061     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1062         h->param.analyse.i_me_method > X264_ME_TESA )
1063         h->param.analyse.i_me_method = X264_ME_HEX;
1064     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1065     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1066         h->param.analyse.i_me_range = 16;
1067     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1068         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1069         h->param.analyse.i_me_method = X264_ME_ESA;
1070     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1071     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1072                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1073     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1074     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1075         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1076     if( !h->param.analyse.b_transform_8x8 )
1077     {
1078         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1079         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1080     }
1081     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1082     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 3 );
1083     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1084     if( h->param.rc.f_aq_strength == 0 )
1085         h->param.rc.i_aq_mode = 0;
1086
1087     if( h->param.i_log_level < X264_LOG_INFO )
1088     {
1089         h->param.analyse.b_psnr = 0;
1090         h->param.analyse.b_ssim = 0;
1091     }
1092     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1093     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1094     {
1095         char *s = NULL;
1096
1097         if( h->param.analyse.b_psy )
1098         {
1099             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1100             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1101         }
1102         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1103         {
1104             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1105             s = "ssim";
1106         }
1107         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1108         {
1109             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1110             s = "psnr";
1111         }
1112         if( s )
1113             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1114     }
1115
1116     if( !h->param.analyse.b_psy )
1117     {
1118         h->param.analyse.f_psy_rd = 0;
1119         h->param.analyse.f_psy_trellis = 0;
1120     }
1121     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1122     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1123     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1124     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1125     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1126     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1127     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1128         h->param.analyse.i_chroma_qp_offset += 6;
1129     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1130     /* so we lower the chroma QP offset to compensate */
1131     if( b_open && h->mb.i_psy_rd && !h->param.i_avcintra_class )
1132         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1133     /* Psy trellis has a similar effect. */
1134     if( b_open && h->mb.i_psy_trellis && !h->param.i_avcintra_class )
1135         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1136     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1137     /* MB-tree requires AQ to be on, even if the strength is zero. */
1138     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1139     {
1140         h->param.rc.i_aq_mode = 1;
1141         h->param.rc.f_aq_strength = 0;
1142     }
1143     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1144     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1145         h->param.analyse.i_subpel_refine = 9;
1146
1147     {
1148         const x264_level_t *l = x264_levels;
1149         if( h->param.i_level_idc < 0 )
1150         {
1151             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1152             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1153                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1154             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1155             do h->param.i_level_idc = l->level_idc;
1156                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1157             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1158         }
1159         else
1160         {
1161             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1162                 l++;
1163             if( l->level_idc == 0 )
1164             {
1165                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1166                 return -1;
1167             }
1168         }
1169         if( h->param.analyse.i_mv_range <= 0 )
1170             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1171         else
1172             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1173     }
1174
1175     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1176
1177     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1178     {
1179         if( h->param.b_sliced_threads )
1180             h->param.i_lookahead_threads = h->param.i_threads;
1181         else
1182         {
1183             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1184              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1185              * try to guess at this sort of case.
1186              *
1187              * Tuned by a little bit of real encoding with the various presets. */
1188             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1189             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1190             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1191
1192             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1193             static const uint8_t lookahead_thread_div[2][5][4] =
1194             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1195              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1196
1197             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1198             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1199              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1200              * pretty much arbitrarily. */
1201             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1202         }
1203     }
1204     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1205
1206     if( PARAM_INTERLACED )
1207     {
1208         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1209         {
1210             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1211             h->param.analyse.i_me_method = X264_ME_UMH;
1212         }
1213         if( h->param.analyse.i_weighted_pred > 0 )
1214         {
1215             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1216             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1217         }
1218     }
1219
1220     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1221         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1222
1223     if( h->i_thread_frames > 1 )
1224     {
1225         int r = h->param.analyse.i_mv_range_thread;
1226         int r2;
1227         if( r <= 0 )
1228         {
1229             // half of the available space is reserved and divided evenly among the threads,
1230             // the rest is allocated to whichever thread is far enough ahead to use it.
1231             // reserving more space increases quality for some videos, but costs more time
1232             // in thread synchronization.
1233             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1234             r = max_range / 2;
1235         }
1236         r = X264_MAX( r, h->param.analyse.i_me_range );
1237         r = X264_MIN( r, h->param.analyse.i_mv_range );
1238         // round up to use the whole mb row
1239         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1240         if( r2 < r )
1241             r2 += 16;
1242         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1243         h->param.analyse.i_mv_range_thread = r2;
1244     }
1245
1246     if( h->param.rc.f_rate_tolerance < 0 )
1247         h->param.rc.f_rate_tolerance = 0;
1248     if( h->param.rc.f_qblur < 0 )
1249         h->param.rc.f_qblur = 0;
1250     if( h->param.rc.f_complexity_blur < 0 )
1251         h->param.rc.f_complexity_blur = 0;
1252     if( h->param.sc.i_buffer_size < 0 || h->param.sc.f_speed <= 0 )
1253         h->param.sc.i_buffer_size = 0;
1254
1255     h->param.i_sps_id &= 31;
1256
1257     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1258
1259     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1260     {
1261         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1262         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1263     }
1264
1265     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1266        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1267     {
1268         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1269         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1270     }
1271
1272     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1273         h->param.rc.b_filler = 1;
1274
1275     h->param.sc.max_preset = x264_clip3( h->param.sc.max_preset, 1, SC_PRESETS );
1276
1277     /* ensure the booleans are 0 or 1 so they can be used in math */
1278 #define BOOLIFY(x) h->param.x = !!h->param.x
1279     BOOLIFY( b_cabac );
1280     BOOLIFY( b_constrained_intra );
1281     BOOLIFY( b_deblocking_filter );
1282     BOOLIFY( b_deterministic );
1283     BOOLIFY( b_sliced_threads );
1284     BOOLIFY( b_interlaced );
1285     BOOLIFY( b_intra_refresh );
1286     BOOLIFY( b_aud );
1287     BOOLIFY( b_repeat_headers );
1288     BOOLIFY( b_annexb );
1289     BOOLIFY( b_vfr_input );
1290     BOOLIFY( b_pulldown );
1291     BOOLIFY( b_tff );
1292     BOOLIFY( b_pic_struct );
1293     BOOLIFY( b_fake_interlaced );
1294     BOOLIFY( b_open_gop );
1295     BOOLIFY( b_bluray_compat );
1296     BOOLIFY( b_stitchable );
1297     BOOLIFY( b_full_recon );
1298     BOOLIFY( b_opencl );
1299     BOOLIFY( analyse.b_transform_8x8 );
1300     BOOLIFY( analyse.b_weighted_bipred );
1301     BOOLIFY( analyse.b_chroma_me );
1302     BOOLIFY( analyse.b_mixed_references );
1303     BOOLIFY( analyse.b_fast_pskip );
1304     BOOLIFY( analyse.b_dct_decimate );
1305     BOOLIFY( analyse.b_psy );
1306     BOOLIFY( analyse.b_psnr );
1307     BOOLIFY( analyse.b_ssim );
1308     BOOLIFY( rc.b_stat_write );
1309     BOOLIFY( rc.b_stat_read );
1310     BOOLIFY( rc.b_mb_tree );
1311     BOOLIFY( rc.b_filler );
1312 #undef BOOLIFY
1313
1314     return 0;
1315 }
1316
1317 static void mbcmp_init( x264_t *h )
1318 {
1319     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1320     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1321     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1322     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1323     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1324     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1325     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1326     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1327     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1328                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1329     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1330                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1331     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1332     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1333     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1334     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1335 }
1336
1337 static void chroma_dsp_init( x264_t *h )
1338 {
1339     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1340
1341     switch( CHROMA_FORMAT )
1342     {
1343         case CHROMA_420:
1344             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1345             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1346             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1347             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1348             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1349             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1350             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1351             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1352             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1353             break;
1354         case CHROMA_422:
1355             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1356             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1357             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1358             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1359             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1360             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1361             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1362             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1363             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1364             break;
1365         case CHROMA_444:
1366             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1367             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1368             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1369             break;
1370     }
1371 }
1372
1373 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1374 {
1375     /* VUI */
1376     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1377     {
1378         uint32_t i_w = param->vui.i_sar_width;
1379         uint32_t i_h = param->vui.i_sar_height;
1380         uint32_t old_w = h->param.vui.i_sar_width;
1381         uint32_t old_h = h->param.vui.i_sar_height;
1382
1383         x264_reduce_fraction( &i_w, &i_h );
1384
1385         while( i_w > 65535 || i_h > 65535 )
1386         {
1387             i_w /= 2;
1388             i_h /= 2;
1389         }
1390
1391         x264_reduce_fraction( &i_w, &i_h );
1392
1393         if( i_w != old_w || i_h != old_h || initial )
1394         {
1395             h->param.vui.i_sar_width = 0;
1396             h->param.vui.i_sar_height = 0;
1397             if( i_w == 0 || i_h == 0 )
1398                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1399             else
1400             {
1401                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1402                 h->param.vui.i_sar_width = i_w;
1403                 h->param.vui.i_sar_height = i_h;
1404             }
1405         }
1406     }
1407 }
1408
1409 /****************************************************************************
1410  * x264_encoder_open:
1411  ****************************************************************************/
1412 x264_t *x264_encoder_open( x264_param_t *param )
1413 {
1414     x264_t *h;
1415     char buf[1000], *p;
1416     int i_slicetype_length;
1417
1418     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1419
1420     /* Create a copy of param */
1421     memcpy( &h->param, param, sizeof(x264_param_t) );
1422
1423     if( param->param_free )
1424         param->param_free( param );
1425
1426 #if HAVE_INTEL_DISPATCHER
1427     x264_intel_dispatcher_override();
1428 #endif
1429
1430     if( x264_threading_init() )
1431     {
1432         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1433         goto fail;
1434     }
1435
1436     if( x264_validate_parameters( h, 1 ) < 0 )
1437         goto fail;
1438
1439     if( h->param.psz_cqm_file )
1440         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1441             goto fail;
1442
1443     if( h->param.rc.psz_stat_out )
1444         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1445     if( h->param.rc.psz_stat_in )
1446         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1447
1448     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1449     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1450
1451     /* Init x264_t */
1452     h->i_frame = -1;
1453     h->i_frame_num = 0;
1454
1455     if( h->param.i_avcintra_class )
1456         h->i_idr_pic_id = 5;
1457     else
1458         h->i_idr_pic_id = 0;
1459
1460     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1461     {
1462         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1463         goto fail;
1464     }
1465
1466     x264_set_aspect_ratio( h, &h->param, 1 );
1467
1468     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1469     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1470
1471     x264_validate_levels( h, 1 );
1472
1473     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1474
1475     if( x264_cqm_init( h ) < 0 )
1476         goto fail;
1477
1478     h->mb.i_mb_width = h->sps->i_mb_width;
1479     h->mb.i_mb_height = h->sps->i_mb_height;
1480     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1481
1482     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1483     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1484
1485     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1486      * vectors during prediction, resulting in hpel mvs.
1487      * The chosen solution is to make MBAFF non-adaptive in this case. */
1488     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1489
1490     /* Init frames. */
1491     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1492         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1493     else
1494         h->frames.i_delay = h->param.i_bframe;
1495     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1496         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1497     i_slicetype_length = h->frames.i_delay;
1498     h->frames.i_delay += h->i_thread_frames - 1;
1499     h->frames.i_delay += h->param.i_sync_lookahead;
1500     h->frames.i_delay += h->param.b_vfr_input;
1501     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1502
1503     h->frames.i_max_ref0 = h->param.i_frame_reference;
1504     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1505     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1506     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1507         && ( h->param.rc.i_rc_method == X264_RC_ABR
1508           || h->param.rc.i_rc_method == X264_RC_CRF
1509           || h->param.i_bframe_adaptive
1510           || h->param.i_scenecut_threshold
1511           || h->param.rc.b_mb_tree
1512           || h->param.analyse.i_weighted_pred );
1513     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1514     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1515
1516     h->frames.i_last_idr =
1517     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1518     h->frames.i_input    = 0;
1519     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1520     h->frames.i_poc_last_open_gop = -1;
1521
1522     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1523     /* Allocate room for max refs plus a few extra just in case. */
1524     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1525     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1526                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1527     if( h->param.analyse.i_weighted_pred > 0 )
1528         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1529     h->i_ref[0] = h->i_ref[1] = 0;
1530     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1531     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);
1532     h->i_disp_fields_last_frame = -1;
1533     x264_rdo_init();
1534
1535     /* init CPU functions */
1536     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1537     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1538     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1539     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1540     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1541     x264_pixel_init( h->param.cpu, &h->pixf );
1542     x264_dct_init( h->param.cpu, &h->dctf );
1543     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1544     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1545     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1546     x264_quant_init( h, h->param.cpu, &h->quantf );
1547     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1548     x264_bitstream_init( h->param.cpu, &h->bsf );
1549     if( h->param.b_cabac )
1550         x264_cabac_init( h );
1551     else
1552         x264_stack_align( x264_cavlc_init, h );
1553
1554     mbcmp_init( h );
1555     chroma_dsp_init( h );
1556
1557     if( h->param.sc.i_buffer_size )
1558         x264_speedcontrol_new( h );
1559
1560
1561     p = buf + sprintf( buf, "using cpu capabilities:" );
1562     for( int i = 0; x264_cpu_names[i].flags; i++ )
1563     {
1564         if( !strcmp(x264_cpu_names[i].name, "SSE")
1565             && h->param.cpu & (X264_CPU_SSE2) )
1566             continue;
1567         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1568             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1569             continue;
1570         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1571             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1572             continue;
1573         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1574             && (h->param.cpu & X264_CPU_SSE42) )
1575             continue;
1576         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1577             && (h->param.cpu & X264_CPU_BMI2) )
1578             continue;
1579         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1580             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1581             p += sprintf( p, " %s", x264_cpu_names[i].name );
1582     }
1583     if( !h->param.cpu )
1584         p += sprintf( p, " none!" );
1585     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1586
1587     if( x264_analyse_init_costs( h ) )
1588         goto fail;
1589
1590     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1591     /* Checks for known miscompilation issues. */
1592     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1593     {
1594         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1595         goto fail;
1596     }
1597
1598     /* Must be volatile or else GCC will optimize it out. */
1599     volatile int temp = 392;
1600     if( x264_clz( temp ) != 23 )
1601     {
1602         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1603 #if ARCH_X86 || ARCH_X86_64
1604         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1605         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1606 #endif
1607         goto fail;
1608     }
1609
1610     h->out.i_nal = 0;
1611     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1612         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1613           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1614
1615     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1616     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1617
1618     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1619
1620     if( h->param.i_threads > 1 &&
1621         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1622         goto fail;
1623     if( h->param.i_lookahead_threads > 1 &&
1624         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1625         goto fail;
1626
1627 #if HAVE_OPENCL
1628     if( h->param.b_opencl )
1629     {
1630         h->opencl.ocl = x264_opencl_load_library();
1631         if( !h->opencl.ocl )
1632         {
1633             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1634             h->param.b_opencl = 0;
1635         }
1636     }
1637 #endif
1638
1639     h->thread[0] = h;
1640     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1641         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1642     if( h->param.i_lookahead_threads > 1 )
1643         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1644         {
1645             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1646             *h->lookahead_thread[i] = *h;
1647         }
1648     *h->reconfig_h = *h;
1649
1650     for( int i = 0; i < h->param.i_threads; i++ )
1651     {
1652         int init_nal_count = h->param.i_slice_count + 3;
1653         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1654         if( i > 0 )
1655             *h->thread[i] = *h;
1656
1657         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1658             goto fail;
1659         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1660             goto fail;
1661
1662         if( allocate_threadlocal_data )
1663         {
1664             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1665             if( !h->thread[i]->fdec )
1666                 goto fail;
1667         }
1668         else
1669             h->thread[i]->fdec = h->thread[0]->fdec;
1670
1671         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1672         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1673         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1674         h->thread[i]->out.i_nals_allocated = init_nal_count;
1675
1676         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1677             goto fail;
1678     }
1679
1680 #if HAVE_OPENCL
1681     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1682         h->param.b_opencl = 0;
1683 #endif
1684
1685     if( x264_lookahead_init( h, i_slicetype_length ) )
1686         goto fail;
1687
1688     for( int i = 0; i < h->param.i_threads; i++ )
1689         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1690             goto fail;
1691
1692     if( x264_ratecontrol_new( h ) < 0 )
1693         goto fail;
1694
1695     if( h->param.i_nal_hrd )
1696     {
1697         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1698         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1699     }
1700
1701     if( h->param.psz_dump_yuv )
1702     {
1703         /* create or truncate the reconstructed video file */
1704         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1705         if( !f )
1706         {
1707             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1708             goto fail;
1709         }
1710         else if( !x264_is_regular_file( f ) )
1711         {
1712             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1713             fclose( f );
1714             goto fail;
1715         }
1716         fclose( f );
1717     }
1718
1719     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1720                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1721                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1722                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1723                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1724                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1725     char level[4];
1726     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1727     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1728         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1729         strcpy( level, "1b" );
1730
1731     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1732     {
1733         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1734             profile, level );
1735     }
1736     else
1737     {
1738         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1739         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1740             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1741     }
1742
1743     return h;
1744 fail:
1745     x264_free( h );
1746     return NULL;
1747 }
1748
1749 /****************************************************************************/
1750 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1751 {
1752     *rc_reconfig = 0;
1753     x264_set_aspect_ratio( h, param, 0 );
1754 #define COPY(var) h->param.var = param->var
1755     COPY( i_frame_reference ); // but never uses more refs than initially specified
1756     COPY( i_bframe_bias );
1757     if( h->param.i_scenecut_threshold )
1758         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1759     COPY( b_deblocking_filter );
1760     COPY( i_deblocking_filter_alphac0 );
1761     COPY( i_deblocking_filter_beta );
1762     COPY( i_frame_packing );
1763     COPY( analyse.inter );
1764     COPY( analyse.intra );
1765     COPY( analyse.i_direct_mv_pred );
1766     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1767     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1768         COPY( analyse.i_me_range );
1769     COPY( analyse.i_noise_reduction );
1770     /* We can't switch out of subme=0 during encoding. */
1771     if( h->param.analyse.i_subpel_refine )
1772         COPY( analyse.i_subpel_refine );
1773     COPY( analyse.i_trellis );
1774     COPY( analyse.b_chroma_me );
1775     COPY( analyse.b_dct_decimate );
1776     COPY( analyse.b_fast_pskip );
1777     COPY( analyse.b_mixed_references );
1778     COPY( analyse.f_psy_rd );
1779     COPY( analyse.f_psy_trellis );
1780     COPY( crop_rect );
1781     // can only twiddle these if they were enabled to begin with:
1782     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1783         COPY( analyse.i_me_method );
1784     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1785         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1786     if( h->pps->b_transform_8x8_mode )
1787         COPY( analyse.b_transform_8x8 );
1788     if( h->frames.i_max_ref1 > 1 )
1789         COPY( i_bframe_pyramid );
1790     COPY( i_slice_max_size );
1791     COPY( i_slice_max_mbs );
1792     COPY( i_slice_min_mbs );
1793     COPY( i_slice_count );
1794     COPY( i_slice_count_max );
1795     COPY( b_tff );
1796
1797     /* VBV can't be turned on if it wasn't on to begin with */
1798     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1799           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1800     {
1801         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1802         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1803         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1804         COPY( rc.i_vbv_max_bitrate );
1805         COPY( rc.i_vbv_buffer_size );
1806         COPY( rc.i_bitrate );
1807     }
1808     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1809     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1810     COPY( rc.f_rf_constant );
1811     COPY( rc.f_rf_constant_max );
1812 #undef COPY
1813
1814     return x264_validate_parameters( h, 0 );
1815 }
1816
1817 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1818 {
1819     int rc_reconfig;
1820     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1821
1822     mbcmp_init( h );
1823     if( !ret )
1824         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1825
1826     /* Supported reconfiguration options (1-pass only):
1827      * vbv-maxrate
1828      * vbv-bufsize
1829      * crf
1830      * bitrate (CBR only) */
1831     if( !ret && rc_reconfig )
1832         x264_ratecontrol_init_reconfigurable( h, 0 );
1833
1834     return ret;
1835 }
1836
1837 /****************************************************************************
1838  * x264_encoder_reconfig:
1839  ****************************************************************************/
1840 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1841 {
1842     h = h->thread[h->thread[0]->i_thread_phase];
1843     x264_param_t param_save = h->reconfig_h->param;
1844     h->reconfig_h->param = h->param;
1845
1846     int rc_reconfig;
1847     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1848     if( !ret )
1849         h->reconfig = 1;
1850     else
1851         h->reconfig_h->param = param_save;
1852
1853     return ret;
1854 }
1855
1856 /****************************************************************************
1857  * x264_encoder_parameters:
1858  ****************************************************************************/
1859 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1860 {
1861     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1862 }
1863
1864 /* internal usage */
1865 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1866 {
1867     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1868
1869     nal->i_ref_idc        = i_ref_idc;
1870     nal->i_type           = i_type;
1871     nal->b_long_startcode = 1;
1872
1873     nal->i_payload= 0;
1874     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1875     nal->i_padding= 0;
1876 }
1877
1878 /* if number of allocated nals is not enough, re-allocate a larger one. */
1879 static int x264_nal_check_buffer( x264_t *h )
1880 {
1881     if( h->out.i_nal >= h->out.i_nals_allocated )
1882     {
1883         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1884         if( !new_out )
1885             return -1;
1886         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1887         x264_free( h->out.nal );
1888         h->out.nal = new_out;
1889         h->out.i_nals_allocated *= 2;
1890     }
1891     return 0;
1892 }
1893
1894 static int x264_nal_end( x264_t *h )
1895 {
1896     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1897     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1898     nal->i_payload = end - nal->p_payload;
1899     /* Assembly implementation of nal_escape reads past the end of the input.
1900      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1901     memset( end, 0xff, 64 );
1902     if( h->param.nalu_process )
1903         h->param.nalu_process( h, nal, h->fenc->opaque );
1904     h->out.i_nal++;
1905
1906     return x264_nal_check_buffer( h );
1907 }
1908
1909 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1910                                            int previous_nal_size, int necessary_size )
1911 {
1912     if( h0->nal_buffer_size < necessary_size )
1913     {
1914         necessary_size *= 2;
1915         uint8_t *buf = x264_malloc( necessary_size );
1916         if( !buf )
1917             return -1;
1918         if( previous_nal_size )
1919             memcpy( buf, h0->nal_buffer, previous_nal_size );
1920
1921         intptr_t delta = buf - h0->nal_buffer;
1922         for( int i = 0; i < start; i++ )
1923             h->out.nal[i].p_payload += delta;
1924
1925         x264_free( h0->nal_buffer );
1926         h0->nal_buffer = buf;
1927         h0->nal_buffer_size = necessary_size;
1928     }
1929
1930     return 0;
1931 }
1932
1933 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1934 {
1935     x264_t *h0 = h->thread[0];
1936     int nal_size = 0, previous_nal_size = 0;
1937
1938     if( h->param.nalu_process )
1939     {
1940         for( int i = start; i < h->out.i_nal; i++ )
1941             nal_size += h->out.nal[i].i_payload;
1942         return nal_size;
1943     }
1944
1945     for( int i = 0; i < start; i++ )
1946         previous_nal_size += h->out.nal[i].i_payload;
1947
1948     for( int i = start; i < h->out.i_nal; i++ )
1949         nal_size += h->out.nal[i].i_payload;
1950
1951     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1952     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1953     for( int i = start; i < h->out.i_nal; i++ )
1954         necessary_size += h->out.nal[i].i_padding;
1955     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1956         return -1;
1957
1958     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1959
1960     for( int i = start; i < h->out.i_nal; i++ )
1961     {
1962         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1963                                          h->param.i_avcintra_class;
1964         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1965         nal_buffer += h->out.nal[i].i_payload;
1966     }
1967
1968     x264_emms();
1969
1970     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1971 }
1972
1973 /****************************************************************************
1974  * x264_encoder_headers:
1975  ****************************************************************************/
1976 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1977 {
1978     int frame_size = 0;
1979     /* init bitstream context */
1980     h->out.i_nal = 0;
1981     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1982
1983     /* Write SEI, SPS and PPS. */
1984
1985     /* generate sequence parameters */
1986     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1987     x264_sps_write( &h->out.bs, h->sps );
1988     if( x264_nal_end( h ) )
1989         return -1;
1990
1991     /* generate picture parameters */
1992     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1993     x264_pps_write( &h->out.bs, h->sps, h->pps );
1994     if( x264_nal_end( h ) )
1995         return -1;
1996
1997     /* identify ourselves */
1998     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1999     if( x264_sei_version_write( h, &h->out.bs ) )
2000         return -1;
2001     if( x264_nal_end( h ) )
2002         return -1;
2003
2004     frame_size = x264_encoder_encapsulate_nals( h, 0 );
2005     if( frame_size < 0 )
2006         return -1;
2007
2008     /* now set output*/
2009     *pi_nal = h->out.i_nal;
2010     *pp_nal = &h->out.nal[0];
2011     h->out.i_nal = 0;
2012
2013     return frame_size;
2014 }
2015
2016 /* Check to see whether we have chosen a reference list ordering different
2017  * from the standard's default. */
2018 static inline void x264_reference_check_reorder( x264_t *h )
2019 {
2020     /* The reorder check doesn't check for missing frames, so just
2021      * force a reorder if one of the reference list is corrupt. */
2022     for( int i = 0; h->frames.reference[i]; i++ )
2023         if( h->frames.reference[i]->b_corrupt )
2024         {
2025             h->b_ref_reorder[0] = 1;
2026             return;
2027         }
2028     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
2029         for( int i = 0; i < h->i_ref[list] - 1; i++ )
2030         {
2031             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
2032             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
2033             /* P and B-frames use different default orders. */
2034             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
2035             {
2036                 h->b_ref_reorder[list] = 1;
2037                 return;
2038             }
2039         }
2040 }
2041
2042 /* return -1 on failure, else return the index of the new reference frame */
2043 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2044 {
2045     int i = h->i_ref[0];
2046     int j = 1;
2047     x264_frame_t *newframe;
2048     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2049         return -1;
2050
2051     //Duplication is only used in X264_WEIGHTP_SMART
2052     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2053         return -1;
2054
2055     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2056      * With high bit depth, it's not worth doing, so turn it off except in the case of
2057      * unweighted dupes. */
2058     if( BIT_DEPTH > 8 && w != x264_weight_none )
2059         return -1;
2060
2061     newframe = x264_frame_pop_blank_unused( h );
2062     if( !newframe )
2063         return -1;
2064
2065     //FIXME: probably don't need to copy everything
2066     *newframe = *h->fref[0][i_ref];
2067     newframe->i_reference_count = 1;
2068     newframe->orig = h->fref[0][i_ref];
2069     newframe->b_duplicate = 1;
2070     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2071
2072     /* shift the frames to make space for the dupe. */
2073     h->b_ref_reorder[0] = 1;
2074     if( h->i_ref[0] < X264_REF_MAX )
2075         ++h->i_ref[0];
2076     h->fref[0][X264_REF_MAX-1] = NULL;
2077     x264_frame_unshift( &h->fref[0][j], newframe );
2078
2079     return j;
2080 }
2081
2082 static void x264_weighted_pred_init( x264_t *h )
2083 {
2084     /* for now no analysis and set all weights to nothing */
2085     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2086         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2087
2088     // FIXME: This only supports weighting of one reference frame
2089     // and duplicates of that frame.
2090     h->fenc->i_lines_weighted = 0;
2091
2092     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2093         for( int i = 0; i < 3; i++ )
2094             h->sh.weight[i_ref][i].weightfn = NULL;
2095
2096
2097     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2098         return;
2099
2100     int i_padv = PADV << PARAM_INTERLACED;
2101     int denom = -1;
2102     int weightplane[2] = { 0, 0 };
2103     int buffer_next = 0;
2104     for( int i = 0; i < 3; i++ )
2105     {
2106         for( int j = 0; j < h->i_ref[0]; j++ )
2107         {
2108             if( h->fenc->weight[j][i].weightfn )
2109             {
2110                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2111                 // if weight is useless, don't write it to stream
2112                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2113                     h->sh.weight[j][i].weightfn = NULL;
2114                 else
2115                 {
2116                     if( !weightplane[!!i] )
2117                     {
2118                         weightplane[!!i] = 1;
2119                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2120                         assert( x264_clip3( denom, 0, 7 ) == denom );
2121                     }
2122
2123                     assert( h->sh.weight[j][i].i_denom == denom );
2124                     if( !i )
2125                     {
2126                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2127                         //scale full resolution frame
2128                         if( h->param.i_threads == 1 )
2129                         {
2130                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2131                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2132                             int stride = h->fenc->i_stride[0];
2133                             int width = h->fenc->i_width[0] + PADH*2;
2134                             int height = h->fenc->i_lines[0] + i_padv*2;
2135                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2136                             h->fenc->i_lines_weighted = height;
2137                         }
2138                     }
2139                 }
2140             }
2141         }
2142     }
2143
2144     if( weightplane[1] )
2145         for( int i = 0; i < h->i_ref[0]; i++ )
2146         {
2147             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2148             {
2149                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2150                 h->sh.weight[i][2].i_offset = 0;
2151             }
2152             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2153             {
2154                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2155                 h->sh.weight[i][1].i_offset = 0;
2156             }
2157         }
2158
2159     if( !weightplane[0] )
2160         h->sh.weight[0][0].i_denom = 0;
2161     if( !weightplane[1] )
2162         h->sh.weight[0][1].i_denom = 0;
2163     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2164 }
2165
2166 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2167 {
2168     if( h->param.i_frame_packing == 5 )
2169         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2170                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2171     else
2172         return abs(h->fenc->i_frame - frame->i_frame);
2173 }
2174
2175 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2176 {
2177     int b_ok;
2178
2179     /* build ref list 0/1 */
2180     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2181     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2182     if( h->sh.i_type == SLICE_TYPE_I )
2183         return;
2184
2185     for( int i = 0; h->frames.reference[i]; i++ )
2186     {
2187         if( h->frames.reference[i]->b_corrupt )
2188             continue;
2189         if( h->frames.reference[i]->i_poc < i_poc )
2190             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2191         else if( h->frames.reference[i]->i_poc > i_poc )
2192             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2193     }
2194
2195     if( h->sh.i_mmco_remove_from_end )
2196     {
2197         /* Order ref0 for MMCO remove */
2198         do
2199         {
2200             b_ok = 1;
2201             for( int i = 0; i < h->i_ref[0] - 1; i++ )
2202             {
2203                 if( h->fref[0][i]->i_frame < h->fref[0][i+1]->i_frame )
2204                 {
2205                     XCHG( x264_frame_t*, h->fref[0][i], h->fref[0][i+1] );
2206                     b_ok = 0;
2207                     break;
2208                 }
2209             }
2210         } while( !b_ok );
2211
2212         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2213         {
2214             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2215             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2216             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2217         }
2218     }
2219
2220     /* Order reference lists by distance from the current frame. */
2221     for( int list = 0; list < 2; list++ )
2222     {
2223         h->fref_nearest[list] = h->fref[list][0];
2224         do
2225         {
2226             b_ok = 1;
2227             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2228             {
2229                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2230                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2231                     h->fref_nearest[list] = h->fref[list][i+1];
2232                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2233                 {
2234                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2235                     b_ok = 0;
2236                     break;
2237                 }
2238             }
2239         } while( !b_ok );
2240     }
2241
2242     x264_reference_check_reorder( h );
2243
2244     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2245     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2246     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2247
2248     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2249     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2250         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2251
2252     /* add duplicates */
2253     if( h->fenc->i_type == X264_TYPE_P )
2254     {
2255         int idx = -1;
2256         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2257         {
2258             x264_weight_t w[3];
2259             w[1].weightfn = w[2].weightfn = NULL;
2260             if( h->param.rc.b_stat_read )
2261                 x264_ratecontrol_set_weights( h, h->fenc );
2262
2263             if( !h->fenc->weight[0][0].weightfn )
2264             {
2265                 h->fenc->weight[0][0].i_denom = 0;
2266                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2267                 idx = x264_weighted_reference_duplicate( h, 0, w );
2268             }
2269             else
2270             {
2271                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2272                 {
2273                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2274                 }
2275                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2276                 if( h->fenc->weight[0][0].i_offset > -128 )
2277                 {
2278                     w[0] = h->fenc->weight[0][0];
2279                     w[0].i_offset--;
2280                     h->mc.weight_cache( h, &w[0] );
2281                     idx = x264_weighted_reference_duplicate( h, 0, w );
2282                 }
2283             }
2284         }
2285         h->mb.ref_blind_dupe = idx;
2286     }
2287
2288     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2289     h->mb.pic.i_fref[0] = h->i_ref[0];
2290     h->mb.pic.i_fref[1] = h->i_ref[1];
2291 }
2292
2293 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2294 {
2295     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2296     int b_hpel = h->fdec->b_kept_as_ref;
2297     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2298     int b_end = mb_y == h->i_threadslice_end;
2299     int b_measure_quality = 1;
2300     int min_y = mb_y - (1 << SLICE_MBAFF);
2301     int b_start = min_y == h->i_threadslice_start;
2302     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2303      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2304     int minpix_y = min_y*16 - 4 * !b_start;
2305     int maxpix_y = mb_y*16 - 4 * !b_end;
2306     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2307     if( h->param.b_sliced_threads )
2308     {
2309         switch( pass )
2310         {
2311             /* During encode: only do deblock if asked for */
2312             default:
2313             case 0:
2314                 b_deblock &= h->param.b_full_recon;
2315                 b_hpel = 0;
2316                 break;
2317             /* During post-encode pass: do deblock if not done yet, do hpel for all
2318              * rows except those between slices. */
2319             case 1:
2320                 b_deblock &= !h->param.b_full_recon;
2321                 b_hpel &= !(b_start && min_y > 0);
2322                 b_measure_quality = 0;
2323                 break;
2324             /* Final pass: do the rows between slices in sequence. */
2325             case 2:
2326                 b_deblock = 0;
2327                 b_measure_quality = 0;
2328                 break;
2329         }
2330     }
2331     if( mb_y & SLICE_MBAFF )
2332         return;
2333     if( min_y < h->i_threadslice_start )
2334         return;
2335
2336     if( b_deblock )
2337         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2338             x264_frame_deblock_row( h, y );
2339
2340     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2341      * but the actual image data is equivalent. For now, maintain this
2342      * consistency by copying deblocked pixels between planes. */
2343     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2344         for( int p = 0; p < h->fdec->i_plane; p++ )
2345             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2346                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2347                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2348                         h->mb.i_mb_width*16*sizeof(pixel) );
2349
2350     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2351         x264_frame_expand_border( h, h->fdec, min_y );
2352     if( b_hpel )
2353     {
2354         int end = mb_y == h->mb.i_mb_height;
2355         /* Can't do hpel until the previous slice is done encoding. */
2356         if( h->param.analyse.i_subpel_refine )
2357         {
2358             x264_frame_filter( h, h->fdec, min_y, end );
2359             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2360         }
2361     }
2362
2363     if( SLICE_MBAFF && pass == 0 )
2364         for( int i = 0; i < 3; i++ )
2365         {
2366             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2367             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2368         }
2369
2370     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2371         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2372
2373     if( b_measure_quality )
2374     {
2375         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2376         if( h->param.analyse.b_psnr )
2377         {
2378             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2379                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2380                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2381                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2382                     h->param.i_width, maxpix_y-minpix_y );
2383             if( !CHROMA444 )
2384             {
2385                 uint64_t ssd_u, ssd_v;
2386                 int v_shift = CHROMA_V_SHIFT;
2387                 x264_pixel_ssd_nv12( &h->pixf,
2388                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2389                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2390                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2391                 h->stat.frame.i_ssd[1] += ssd_u;
2392                 h->stat.frame.i_ssd[2] += ssd_v;
2393             }
2394         }
2395
2396         if( h->param.analyse.b_ssim )
2397         {
2398             int ssim_cnt;
2399             x264_emms();
2400             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2401              * and overlap by 4 */
2402             minpix_y += b_start ? 2 : -6;
2403             h->stat.frame.f_ssim +=
2404                 x264_pixel_ssim_wxh( &h->pixf,
2405                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2406                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2407                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2408             h->stat.frame.i_ssim_cnt += ssim_cnt;
2409         }
2410     }
2411 }
2412
2413 static inline int x264_reference_update( x264_t *h )
2414 {
2415     if( !h->fdec->b_kept_as_ref )
2416     {
2417         if( h->i_thread_frames > 1 )
2418         {
2419             x264_frame_push_unused( h, h->fdec );
2420             h->fdec = x264_frame_pop_unused( h, 1 );
2421             if( !h->fdec )
2422                 return -1;
2423         }
2424         return 0;
2425     }
2426
2427     /* apply mmco from previous frame. */
2428     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2429         for( int j = 0; h->frames.reference[j]; j++ )
2430             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2431                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2432
2433     /* move frame in the buffer */
2434     x264_frame_push( h->frames.reference, h->fdec );
2435     if( h->frames.reference[h->sps->i_num_ref_frames] )
2436         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2437     h->fdec = x264_frame_pop_unused( h, 1 );
2438     if( !h->fdec )
2439         return -1;
2440     return 0;
2441 }
2442
2443 static inline void x264_reference_reset( x264_t *h )
2444 {
2445     while( h->frames.reference[0] )
2446         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2447     h->fdec->i_poc =
2448     h->fenc->i_poc = 0;
2449 }
2450
2451 static inline void x264_reference_hierarchy_reset( x264_t *h )
2452 {
2453     int ref;
2454     int b_hasdelayframe = 0;
2455
2456     /* look for delay frames -- chain must only contain frames that are disposable */
2457     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2458         b_hasdelayframe |= h->frames.current[i]->i_coded
2459                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2460
2461     /* This function must handle b-pyramid and clear frames for open-gop */
2462     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2463         return;
2464
2465     /* Remove last BREF. There will never be old BREFs in the
2466      * dpb during a BREF decode when pyramid == STRICT */
2467     for( ref = 0; h->frames.reference[ref]; ref++ )
2468     {
2469         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2470             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2471             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2472             && h->sh.i_type != SLICE_TYPE_B ) )
2473         {
2474             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2475             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2476             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2477             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2478             h->b_ref_reorder[0] = 1;
2479             ref--;
2480         }
2481     }
2482
2483     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2484     if( h->param.i_bframe_pyramid )
2485         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2486 }
2487
2488 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2489 {
2490     /* ------------------------ Create slice header  ----------------------- */
2491     if( i_nal_type == NAL_SLICE_IDR )
2492     {
2493         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2494
2495         /* alternate id */
2496         if( h->param.i_avcintra_class )
2497         {
2498             switch( h->i_idr_pic_id )
2499             {
2500                 case 5:
2501                     h->i_idr_pic_id = 3;
2502                     break;
2503                 case 3:
2504                     h->i_idr_pic_id = 4;
2505                     break;
2506                 case 4:
2507                 default:
2508                     h->i_idr_pic_id = 5;
2509                     break;
2510             }
2511         }
2512         else
2513             h->i_idr_pic_id ^= 1;
2514     }
2515     else
2516     {
2517         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2518
2519         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2520         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2521         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2522             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2523         {
2524             h->sh.b_num_ref_idx_override = 1;
2525         }
2526     }
2527
2528     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2529     {
2530         h->b_sh_backup = 1;
2531         h->sh_backup = h->sh;
2532     }
2533
2534     h->fdec->i_frame_num = h->sh.i_frame_num;
2535
2536     if( h->sps->i_poc_type == 0 )
2537     {
2538         h->sh.i_poc = h->fdec->i_poc;
2539         if( PARAM_INTERLACED )
2540         {
2541             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2542             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2543         }
2544         else
2545             h->sh.i_delta_poc_bottom = 0;
2546         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2547         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2548     }
2549     else
2550     {
2551         /* Nothing to do ? */
2552     }
2553
2554     x264_macroblock_slice_init( h );
2555 }
2556
2557 typedef struct
2558 {
2559     int skip;
2560     uint8_t cabac_prevbyte;
2561     bs_t bs;
2562     x264_cabac_t cabac;
2563     x264_frame_stat_t stat;
2564     int last_qp;
2565     int last_dqp;
2566     int field_decoding_flag;
2567 } x264_bs_bak_t;
2568
2569 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2570 {
2571     if( full )
2572     {
2573         bak->stat = h->stat.frame;
2574         bak->last_qp = h->mb.i_last_qp;
2575         bak->last_dqp = h->mb.i_last_dqp;
2576         bak->field_decoding_flag = h->mb.field_decoding_flag;
2577     }
2578     else
2579     {
2580         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2581         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2582     }
2583     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2584      * encoder has no context dependency and in this case, a slice is ended (and
2585      * thus the content of all contexts are thrown away). */
2586     if( h->param.b_cabac )
2587     {
2588         if( full )
2589             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2590         else
2591             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2592         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2593          * backed up. */
2594         bak->cabac_prevbyte = h->cabac.p[-1];
2595     }
2596     else
2597     {
2598         bak->bs = h->out.bs;
2599         bak->skip = i_skip;
2600     }
2601 }
2602
2603 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2604 {
2605     if( full )
2606     {
2607         h->stat.frame = bak->stat;
2608         h->mb.i_last_qp = bak->last_qp;
2609         h->mb.i_last_dqp = bak->last_dqp;
2610         h->mb.field_decoding_flag = bak->field_decoding_flag;
2611     }
2612     else
2613     {
2614         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2615         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2616     }
2617     if( h->param.b_cabac )
2618     {
2619         if( full )
2620             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2621         else
2622             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2623         h->cabac.p[-1] = bak->cabac_prevbyte;
2624     }
2625     else
2626     {
2627         h->out.bs = bak->bs;
2628         *skip = bak->skip;
2629     }
2630 }
2631
2632 static intptr_t x264_slice_write( x264_t *h )
2633 {
2634     int i_skip;
2635     int mb_xy, i_mb_x, i_mb_y;
2636     /* NALUs other than the first use a 3-byte startcode.
2637      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2638      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2639      * other inaccuracies. */
2640     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2641     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2642     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2643     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2644     int starting_bits = bs_pos(&h->out.bs);
2645     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2646     int b_hpel = h->fdec->b_kept_as_ref;
2647     int orig_last_mb = h->sh.i_last_mb;
2648     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2649     uint8_t *last_emu_check;
2650 #define BS_BAK_SLICE_MAX_SIZE 0
2651 #define BS_BAK_CAVLC_OVERFLOW 1
2652 #define BS_BAK_SLICE_MIN_MBS  2
2653 #define BS_BAK_ROW_VBV        3
2654     x264_bs_bak_t bs_bak[4];
2655     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2656     bs_realign( &h->out.bs );
2657
2658     /* Slice */
2659     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2660     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2661
2662     /* Slice header */
2663     x264_macroblock_thread_init( h );
2664
2665     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2666     h->mb.i_mb_xy = h->sh.i_first_mb;
2667     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2668     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2669     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2670
2671     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2672     if( h->param.b_cabac )
2673     {
2674         /* alignment needed */
2675         bs_align_1( &h->out.bs );
2676
2677         /* init cabac */
2678         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 );
2679         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2680         last_emu_check = h->cabac.p;
2681     }
2682     else
2683         last_emu_check = h->out.bs.p;
2684     h->mb.i_last_qp = h->sh.i_qp;
2685     h->mb.i_last_dqp = 0;
2686     h->mb.field_decoding_flag = 0;
2687
2688     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2689     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2690     i_skip = 0;
2691
2692     while( 1 )
2693     {
2694         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2695         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2696
2697         if( i_mb_x == 0 )
2698         {
2699             if( x264_bitstream_check_buffer( h ) )
2700                 return -1;
2701             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2702                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2703             if( !h->mb.b_reencode_mb )
2704                 x264_fdec_filter_row( h, i_mb_y, 0 );
2705         }
2706
2707         if( back_up_bitstream )
2708         {
2709             if( back_up_bitstream_cavlc )
2710                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2711             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2712             {
2713                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2714                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2715                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2716             }
2717         }
2718
2719         if( PARAM_INTERLACED )
2720         {
2721             if( h->mb.b_adaptive_mbaff )
2722             {
2723                 if( !(i_mb_y&1) )
2724                 {
2725                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2726                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2727                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2728                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2729                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2730                 }
2731             }
2732             h->mb.field[mb_xy] = MB_INTERLACED;
2733         }
2734
2735         /* load cache */
2736         if( SLICE_MBAFF )
2737             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2738         else
2739             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2740
2741         x264_macroblock_analyse( h );
2742
2743         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2744 reencode:
2745         x264_macroblock_encode( h );
2746
2747         if( h->param.b_cabac )
2748         {
2749             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2750                 x264_cabac_encode_terminal( &h->cabac );
2751
2752             if( IS_SKIP( h->mb.i_type ) )
2753                 x264_cabac_mb_skip( h, 1 );
2754             else
2755             {
2756                 if( h->sh.i_type != SLICE_TYPE_I )
2757                     x264_cabac_mb_skip( h, 0 );
2758                 x264_macroblock_write_cabac( h, &h->cabac );
2759             }
2760         }
2761         else
2762         {
2763             if( IS_SKIP( h->mb.i_type ) )
2764                 i_skip++;
2765             else
2766             {
2767                 if( h->sh.i_type != SLICE_TYPE_I )
2768                 {
2769                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2770                     i_skip = 0;
2771                 }
2772                 x264_macroblock_write_cavlc( h );
2773                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2774                 if( h->mb.b_overflow )
2775                 {
2776                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2777                     h->mb.i_skip_intra = 0;
2778                     h->mb.b_skip_mc = 0;
2779                     h->mb.b_overflow = 0;
2780                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2781                     goto reencode;
2782                 }
2783             }
2784         }
2785
2786         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2787         int mb_size = total_bits - mb_spos;
2788
2789         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2790         {
2791             /* Count the skip run, just in case. */
2792             if( !h->param.b_cabac )
2793                 total_bits += bs_size_ue_big( i_skip );
2794             /* Check for escape bytes. */
2795             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2796             for( ; last_emu_check < end - 2; last_emu_check++ )
2797                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2798                 {
2799                     slice_max_size -= 8;
2800                     last_emu_check++;
2801                 }
2802             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2803             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2804             {
2805                 if( !x264_frame_new_slice( h, h->fdec ) )
2806                 {
2807                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2808                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2809                      * If possible, roll back to the last checkpoint and try again.
2810                      * We could try raising QP, but that would break in the case where a slice spans multiple
2811                      * rows, which the re-encoding infrastructure can't currently handle. */
2812                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2813                     {
2814                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2815                         {
2816                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2817                             slice_max_size = 0;
2818                             goto cont;
2819                         }
2820                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2821                         h->mb.b_reencode_mb = 1;
2822                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2823                         break;
2824                     }
2825                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2826                     {
2827                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2828                         h->mb.b_reencode_mb = 1;
2829                         if( SLICE_MBAFF )
2830                         {
2831                             // set to bottom of previous mbpair
2832                             if( i_mb_x )
2833                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2834                             else
2835                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2836                         }
2837                         else
2838                             h->sh.i_last_mb = mb_xy-1;
2839                         break;
2840                     }
2841                     else
2842                         h->sh.i_last_mb = mb_xy;
2843                 }
2844                 else
2845                     slice_max_size = 0;
2846             }
2847         }
2848 cont:
2849         h->mb.b_reencode_mb = 0;
2850
2851         /* save cache */
2852         x264_macroblock_cache_save( h );
2853
2854         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2855         {
2856             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2857             h->mb.b_reencode_mb = 1;
2858             i_mb_x = 0;
2859             i_mb_y = i_mb_y - SLICE_MBAFF;
2860             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2861             h->sh.i_last_mb = orig_last_mb;
2862             continue;
2863         }
2864
2865         /* accumulate mb stats */
2866         h->stat.frame.i_mb_count[h->mb.i_type]++;
2867
2868         int b_intra = IS_INTRA( h->mb.i_type );
2869         int b_skip = IS_SKIP( h->mb.i_type );
2870         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2871         {
2872             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2873             {
2874                 if( h->mb.i_partition != D_8x8 )
2875                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2876                     else
2877                         for( int i = 0; i < 4; i++ )
2878                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2879                 if( h->param.i_frame_reference > 1 )
2880                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2881                         for( int i = 0; i < 4; i++ )
2882                         {
2883                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2884                             if( i_ref >= 0 )
2885                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2886                         }
2887             }
2888         }
2889
2890         if( h->param.i_log_level >= X264_LOG_INFO )
2891         {
2892             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2893             {
2894                 if( CHROMA444 )
2895                 {
2896                     for( int i = 0; i < 4; i++ )
2897                         if( h->mb.i_cbp_luma & (1 << i) )
2898                             for( int p = 0; p < 3; p++ )
2899                             {
2900                                 int s8 = i*4+p*16;
2901                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2902                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2903                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2904                             }
2905                 }
2906                 else
2907                 {
2908                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2909                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2910                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2911                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2912                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2913                 }
2914             }
2915             if( h->mb.i_cbp_luma && !b_intra )
2916             {
2917                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2918                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2919             }
2920             if( b_intra && h->mb.i_type != I_PCM )
2921             {
2922                 if( h->mb.i_type == I_16x16 )
2923                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2924                 else if( h->mb.i_type == I_8x8 )
2925                     for( int i = 0; i < 16; i += 4 )
2926                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2927                 else //if( h->mb.i_type == I_4x4 )
2928                     for( int i = 0; i < 16; i++ )
2929                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2930                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2931             }
2932             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2933         }
2934
2935         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2936         if( b_deblock )
2937             x264_macroblock_deblock_strength( h );
2938
2939         if( mb_xy == h->sh.i_last_mb )
2940             break;
2941
2942         if( SLICE_MBAFF )
2943         {
2944             i_mb_x += i_mb_y & 1;
2945             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2946         }
2947         else
2948             i_mb_x++;
2949         if( i_mb_x == h->mb.i_mb_width )
2950         {
2951             i_mb_y++;
2952             i_mb_x = 0;
2953         }
2954     }
2955     if( h->sh.i_last_mb < h->sh.i_first_mb )
2956         return 0;
2957
2958     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2959
2960     if( h->param.b_cabac )
2961     {
2962         x264_cabac_encode_flush( h, &h->cabac );
2963         h->out.bs.p = h->cabac.p;
2964     }
2965     else
2966     {
2967         if( i_skip > 0 )
2968             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2969         /* rbsp_slice_trailing_bits */
2970         bs_rbsp_trailing( &h->out.bs );
2971         bs_flush( &h->out.bs );
2972     }
2973     if( x264_nal_end( h ) )
2974         return -1;
2975
2976     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2977     {
2978         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2979                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2980                                   - h->stat.frame.i_tex_bits
2981                                   - h->stat.frame.i_mv_bits;
2982         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2983
2984         if( h->param.b_sliced_threads )
2985         {
2986             /* Tell the main thread we're done. */
2987             x264_threadslice_cond_broadcast( h, 1 );
2988             /* Do hpel now */
2989             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2990                 x264_fdec_filter_row( h, mb_y, 1 );
2991             x264_threadslice_cond_broadcast( h, 2 );
2992             /* Do the first row of hpel, now that the previous slice is done */
2993             if( h->i_thread_idx > 0 )
2994             {
2995                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2996                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2997             }
2998         }
2999
3000         /* Free mb info after the last thread's done using it */
3001         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
3002         {
3003             h->fdec->mb_info_free( h->fdec->mb_info );
3004             h->fdec->mb_info = NULL;
3005             h->fdec->mb_info_free = NULL;
3006         }
3007     }
3008
3009     return 0;
3010 }
3011
3012 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
3013 {
3014     if( dst == src )
3015         return;
3016
3017     // reference counting
3018     for( x264_frame_t **f = src->frames.reference; *f; f++ )
3019         (*f)->i_reference_count++;
3020     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
3021         x264_frame_push_unused( src, *f );
3022     src->fdec->i_reference_count++;
3023     x264_frame_push_unused( src, dst->fdec );
3024
3025     // copy everything except the per-thread pointers and the constants.
3026     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
3027     dst->param = src->param;
3028     dst->stat = src->stat;
3029     dst->pixf = src->pixf;
3030     dst->reconfig = src->reconfig;
3031 }
3032
3033 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
3034 {
3035     if( dst != src )
3036         memcpy( &dst->stat, &src->stat, offsetof(x264_t, stat.frame) - offsetof(x264_t, stat) );
3037 }
3038
3039 static void *x264_slices_write( x264_t *h )
3040 {
3041     int i_slice_num = 0;
3042     int last_thread_mb = h->sh.i_last_mb;
3043
3044     /* init stats */
3045     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
3046     h->mb.b_reencode_mb = 0;
3047     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
3048     {
3049         h->sh.i_last_mb = last_thread_mb;
3050         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
3051         {
3052             if( h->param.i_slice_max_mbs )
3053             {
3054                 if( SLICE_MBAFF )
3055                 {
3056                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
3057                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
3058                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
3059                         + h->param.i_slice_max_mbs - 1;
3060                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
3061                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
3062                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
3063                 }
3064                 else
3065                 {
3066                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
3067                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
3068                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
3069                 }
3070                 i_slice_num++;
3071             }
3072             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3073             {
3074                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3075                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3076                 i_slice_num++;
3077                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3078             }
3079         }
3080         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3081         if( x264_stack_align( x264_slice_write, h ) )
3082             goto fail;
3083         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3084         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3085         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3086             h->sh.i_first_mb -= h->mb.i_mb_stride;
3087     }
3088
3089     return (void *)0;
3090
3091 fail:
3092     /* Tell other threads we're done, so they wouldn't wait for it */
3093     if( h->param.b_sliced_threads )
3094         x264_threadslice_cond_broadcast( h, 2 );
3095     return (void *)-1;
3096 }
3097
3098 static int x264_threaded_slices_write( x264_t *h )
3099 {
3100     /* set first/last mb and sync contexts */
3101     for( int i = 0; i < h->param.i_threads; i++ )
3102     {
3103         x264_t *t = h->thread[i];
3104         if( i )
3105         {
3106             t->param = h->param;
3107             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3108         }
3109         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3110         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3111         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3112         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3113         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3114     }
3115
3116     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3117
3118     x264_threads_distribute_ratecontrol( h );
3119
3120     /* setup */
3121     for( int i = 0; i < h->param.i_threads; i++ )
3122     {
3123         h->thread[i]->i_thread_idx = i;
3124         h->thread[i]->b_thread_active = 1;
3125         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3126     }
3127     /* dispatch */
3128     for( int i = 0; i < h->param.i_threads; i++ )
3129         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3130     /* wait */
3131     for( int i = 0; i < h->param.i_threads; i++ )
3132         x264_threadslice_cond_wait( h->thread[i], 1 );
3133
3134     x264_threads_merge_ratecontrol( h );
3135
3136     for( int i = 1; i < h->param.i_threads; i++ )
3137     {
3138         x264_t *t = h->thread[i];
3139         for( int j = 0; j < t->out.i_nal; j++ )
3140         {
3141             h->out.nal[h->out.i_nal] = t->out.nal[j];
3142             h->out.i_nal++;
3143             x264_nal_check_buffer( h );
3144         }
3145         /* All entries in stat.frame are ints except for ssd/ssim. */
3146         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3147             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3148         for( int j = 0; j < 3; j++ )
3149             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3150         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3151         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3152     }
3153
3154     return 0;
3155 }
3156
3157 void x264_encoder_intra_refresh( x264_t *h )
3158 {
3159     h = h->thread[h->i_thread_phase];
3160     h->b_queued_intra_refresh = 1;
3161 }
3162
3163 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3164 {
3165     if( h->param.i_bframe )
3166     {
3167         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3168         return -1;
3169     }
3170     if( h->param.b_intra_refresh )
3171     {
3172         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3173         return -1;
3174     }
3175     h = h->thread[h->i_thread_phase];
3176     if( pts >= h->i_last_idr_pts )
3177     {
3178         for( int i = 0; h->frames.reference[i]; i++ )
3179             if( pts <= h->frames.reference[i]->i_pts )
3180                 h->frames.reference[i]->b_corrupt = 1;
3181         if( pts <= h->fdec->i_pts )
3182             h->fdec->b_corrupt = 1;
3183     }
3184     return 0;
3185 }
3186
3187 /****************************************************************************
3188  * x264_encoder_encode:
3189  *  XXX: i_poc   : is the poc of the current given picture
3190  *       i_frame : is the number of the frame being coded
3191  *  ex:  type frame poc
3192  *       I      0   2*0
3193  *       P      1   2*3
3194  *       B      2   2*1
3195  *       B      3   2*2
3196  *       P      4   2*6
3197  *       B      5   2*4
3198  *       B      6   2*5
3199  ****************************************************************************/
3200 int     x264_encoder_encode( x264_t *h,
3201                              x264_nal_t **pp_nal, int *pi_nal,
3202                              x264_picture_t *pic_in,
3203                              x264_picture_t *pic_out )
3204 {
3205     x264_t *thread_current, *thread_prev, *thread_oldest;
3206     int i_nal_type, i_nal_ref_idc, i_global_qp;
3207     int overhead = NALU_OVERHEAD;
3208
3209 #if HAVE_OPENCL
3210     if( h->opencl.b_fatal_error )
3211         return -1;
3212 #endif
3213
3214     if( h->i_thread_frames > 1 )
3215     {
3216         thread_prev    = h->thread[ h->i_thread_phase ];
3217         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3218         thread_current = h->thread[ h->i_thread_phase ];
3219         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3220         x264_thread_sync_context( thread_current, thread_prev );
3221         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3222         h = thread_current;
3223     }
3224     else
3225     {
3226         thread_current =
3227         thread_oldest  = h;
3228     }
3229     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3230
3231     /* no data out */
3232     *pi_nal = 0;
3233     *pp_nal = NULL;
3234
3235     /* ------------------- Setup new frame from picture -------------------- */
3236     if( pic_in != NULL )
3237     {
3238         if( h->lookahead->b_exit_thread )
3239         {
3240             x264_log( h, X264_LOG_ERROR, "lookahead thread is already stopped\n" );
3241             return -1;
3242         }
3243
3244         /* 1: Copy the picture to a frame and move it to a buffer */
3245         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3246         if( !fenc )
3247             return -1;
3248
3249         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3250             return -1;
3251
3252         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3253             h->param.i_height != 16 * h->mb.i_mb_height )
3254             x264_frame_expand_border_mod16( h, fenc );
3255
3256         fenc->i_frame = h->frames.i_input++;
3257
3258         if( fenc->i_frame == 0 )
3259             h->frames.i_first_pts = fenc->i_pts;
3260         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3261             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3262
3263         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3264             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3265
3266         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3267         h->frames.i_largest_pts = fenc->i_pts;
3268
3269         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3270             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3271
3272         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3273         {
3274 #if HAVE_INTERLACED
3275             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3276 #else
3277             int b_interlaced = 0;
3278 #endif
3279             if( b_interlaced )
3280             {
3281                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3282                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3283             }
3284             else
3285                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3286         }
3287
3288         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3289         {
3290             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3291                 return -1;
3292         }
3293         else
3294             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3295
3296         if( pic_in->prop.quant_offsets_free )
3297             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3298
3299         if( h->frames.b_have_lowres )
3300             x264_frame_init_lowres( h, fenc );
3301
3302         /* 2: Place the frame into the queue for its slice type decision */
3303         x264_lookahead_put_frame( h, fenc );
3304
3305         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3306         {
3307             /* Nothing yet to encode, waiting for filling of buffers */
3308             pic_out->i_type = X264_TYPE_AUTO;
3309             return 0;
3310         }
3311     }
3312     else
3313     {
3314         /* signal kills for lookahead thread */
3315         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3316         h->lookahead->b_exit_thread = 1;
3317         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3318         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3319     }
3320
3321     h->i_frame++;
3322     /* 3: The picture is analyzed in the lookahead */
3323     if( !h->frames.current[0] )
3324         x264_lookahead_get_frames( h );
3325
3326     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3327         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3328
3329     /* ------------------- Get frame to be encoded ------------------------- */
3330     /* 4: get picture to encode */
3331     h->fenc = x264_frame_shift( h->frames.current );
3332
3333     /* If applicable, wait for previous frame reconstruction to finish */
3334     if( h->param.b_sliced_threads )
3335         if( x264_threadpool_wait_all( h ) < 0 )
3336             return -1;
3337
3338     if( h->i_frame == 0 )
3339         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3340     if( h->reconfig )
3341     {
3342         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3343         h->reconfig = 0;
3344     }
3345     if( h->fenc->param )
3346     {
3347         x264_encoder_reconfig_apply( h, h->fenc->param );
3348         if( h->fenc->param->param_free )
3349         {
3350             h->fenc->param->param_free( h->fenc->param );
3351             h->fenc->param = NULL;
3352         }
3353     }
3354
3355     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3356     if( x264_reference_update( h ) )
3357         return -1;
3358     h->fdec->i_lines_completed = -1;
3359
3360     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3361     {
3362         int valid_refs_left = 0;
3363         for( int i = 0; h->frames.reference[i]; i++ )
3364             if( !h->frames.reference[i]->b_corrupt )
3365                 valid_refs_left++;
3366         /* No valid reference frames left: force an IDR. */
3367         if( !valid_refs_left )
3368         {
3369             h->fenc->b_keyframe = 1;
3370             h->fenc->i_type = X264_TYPE_IDR;
3371         }
3372     }
3373
3374     if( h->fenc->b_keyframe )
3375     {
3376         h->frames.i_last_keyframe = h->fenc->i_frame;
3377         if( h->fenc->i_type == X264_TYPE_IDR )
3378         {
3379             h->i_frame_num = 0;
3380             h->frames.i_last_idr = h->fenc->i_frame;
3381         }
3382     }
3383     h->sh.i_mmco_command_count =
3384     h->sh.i_mmco_remove_from_end = 0;
3385     h->b_ref_reorder[0] =
3386     h->b_ref_reorder[1] = 0;
3387     h->fdec->i_poc =
3388     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3389
3390     /* ------------------- Setup frame context ----------------------------- */
3391     /* 5: Init data dependent of frame type */
3392     if( h->fenc->i_type == X264_TYPE_IDR )
3393     {
3394         /* reset ref pictures */
3395         i_nal_type    = NAL_SLICE_IDR;
3396         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3397         h->sh.i_type = SLICE_TYPE_I;
3398         x264_reference_reset( h );
3399         h->frames.i_poc_last_open_gop = -1;
3400     }
3401     else if( h->fenc->i_type == X264_TYPE_I )
3402     {
3403         i_nal_type    = NAL_SLICE;
3404         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3405         h->sh.i_type = SLICE_TYPE_I;
3406         x264_reference_hierarchy_reset( h );
3407         if( h->param.b_open_gop )
3408             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3409     }
3410     else if( h->fenc->i_type == X264_TYPE_P )
3411     {
3412         i_nal_type    = NAL_SLICE;
3413         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3414         h->sh.i_type = SLICE_TYPE_P;
3415         x264_reference_hierarchy_reset( h );
3416         h->frames.i_poc_last_open_gop = -1;
3417     }
3418     else if( h->fenc->i_type == X264_TYPE_BREF )
3419     {
3420         i_nal_type    = NAL_SLICE;
3421         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3422         h->sh.i_type = SLICE_TYPE_B;
3423         x264_reference_hierarchy_reset( h );
3424     }
3425     else    /* B frame */
3426     {
3427         i_nal_type    = NAL_SLICE;
3428         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3429         h->sh.i_type = SLICE_TYPE_B;
3430     }
3431
3432     h->fdec->i_type = h->fenc->i_type;
3433     h->fdec->i_frame = h->fenc->i_frame;
3434     h->fenc->b_kept_as_ref =
3435     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3436
3437     h->fdec->mb_info = h->fenc->mb_info;
3438     h->fdec->mb_info_free = h->fenc->mb_info_free;
3439     h->fenc->mb_info = NULL;
3440     h->fenc->mb_info_free = NULL;
3441
3442     h->fdec->i_pts = h->fenc->i_pts;
3443     if( h->frames.i_bframe_delay )
3444     {
3445         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3446         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3447                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3448                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3449         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3450     }
3451     else
3452         h->fdec->i_dts = h->fenc->i_reordered_pts;
3453     if( h->fenc->i_type == X264_TYPE_IDR )
3454         h->i_last_idr_pts = h->fdec->i_pts;
3455
3456     /* ------------------- Init                ----------------------------- */
3457     /* build ref list 0/1 */
3458     x264_reference_build_list( h, h->fdec->i_poc );
3459
3460     /* ---------------------- Write the bitstream -------------------------- */
3461     /* Init bitstream context */
3462     if( h->param.b_sliced_threads )
3463     {
3464         for( int i = 0; i < h->param.i_threads; i++ )
3465         {
3466             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3467             h->thread[i]->out.i_nal = 0;
3468         }
3469     }
3470     else
3471     {
3472         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3473         h->out.i_nal = 0;
3474     }
3475
3476     if( h->param.b_aud )
3477     {
3478         int pic_type;
3479
3480         if( h->sh.i_type == SLICE_TYPE_I )
3481             pic_type = 0;
3482         else if( h->sh.i_type == SLICE_TYPE_P )
3483             pic_type = 1;
3484         else if( h->sh.i_type == SLICE_TYPE_B )
3485             pic_type = 2;
3486         else
3487             pic_type = 7;
3488
3489         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3490         bs_write( &h->out.bs, 3, pic_type );
3491         bs_rbsp_trailing( &h->out.bs );
3492         if( x264_nal_end( h ) )
3493             return -1;
3494         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3495     }
3496
3497     h->i_nal_type = i_nal_type;
3498     h->i_nal_ref_idc = i_nal_ref_idc;
3499
3500     if( h->param.b_intra_refresh )
3501     {
3502         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3503         {
3504             h->fdec->i_frames_since_pir = 0;
3505             h->b_queued_intra_refresh = 0;
3506             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3507              * the whole frame and counts as an intra refresh. */
3508             h->fdec->f_pir_position = h->mb.i_mb_width;
3509         }
3510         else if( h->fenc->i_type == X264_TYPE_P )
3511         {
3512             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3513             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3514             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3515             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3516             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3517                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3518             {
3519                 h->fdec->f_pir_position = 0;
3520                 h->fdec->i_frames_since_pir = 0;
3521                 h->b_queued_intra_refresh = 0;
3522                 h->fenc->b_keyframe = 1;
3523             }
3524             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3525             h->fdec->f_pir_position += increment * pocdiff;
3526             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3527             /* If our intra refresh has reached the right side of the frame, we're done. */
3528             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3529             {
3530                 h->fdec->f_pir_position = h->mb.i_mb_width;
3531                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3532             }
3533         }
3534     }
3535
3536     if( h->fenc->b_keyframe )
3537     {
3538         /* Write SPS and PPS */
3539         if( h->param.b_repeat_headers )
3540         {
3541             /* generate sequence parameters */
3542             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3543             x264_sps_write( &h->out.bs, h->sps );
3544             if( x264_nal_end( h ) )
3545                 return -1;
3546             /* Pad AUD/SPS to 256 bytes like Panasonic */
3547             if( h->param.i_avcintra_class )
3548                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3549             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3550
3551             /* generate picture parameters */
3552             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3553             x264_pps_write( &h->out.bs, h->sps, h->pps );
3554             if( x264_nal_end( h ) )
3555                 return -1;
3556             if( h->param.i_avcintra_class )
3557                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3558             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3559         }
3560
3561         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3562         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3563         {
3564             x264_hrd_fullness( h );
3565             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3566             x264_sei_buffering_period_write( h, &h->out.bs );
3567             if( x264_nal_end( h ) )
3568                return -1;
3569             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3570         }
3571     }
3572
3573     /* write extra sei */
3574     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3575     {
3576         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3577         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3578                         h->fenc->extra_sei.payloads[i].payload_type );
3579         if( x264_nal_end( h ) )
3580             return -1;
3581         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3582         if( h->fenc->extra_sei.sei_free )
3583         {
3584             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3585             h->fenc->extra_sei.payloads[i].payload = NULL;
3586         }
3587     }
3588
3589     if( h->fenc->extra_sei.sei_free )
3590     {
3591         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3592         h->fenc->extra_sei.payloads = NULL;
3593         h->fenc->extra_sei.sei_free = NULL;
3594     }
3595
3596     if( h->fenc->b_keyframe )
3597     {
3598         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3599         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3600         {
3601             /* identify ourself */
3602             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3603             if( x264_sei_version_write( h, &h->out.bs ) )
3604                 return -1;
3605             if( x264_nal_end( h ) )
3606                 return -1;
3607             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3608         }
3609
3610         if( h->fenc->i_type != X264_TYPE_IDR )
3611         {
3612             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;
3613             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3614             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3615             if( x264_nal_end( h ) )
3616                 return -1;
3617             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3618         }
3619     }
3620
3621     if( h->param.i_frame_packing >= 0 && (h->fenc->b_keyframe || h->param.i_frame_packing == 5) )
3622     {
3623         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3624         x264_sei_frame_packing_write( h, &h->out.bs );
3625         if( x264_nal_end( h ) )
3626             return -1;
3627         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3628     }
3629
3630     /* generate sei pic timing */
3631     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3632     {
3633         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3634         x264_sei_pic_timing_write( h, &h->out.bs );
3635         if( x264_nal_end( h ) )
3636             return -1;
3637         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3638     }
3639
3640     /* As required by Blu-ray. */
3641     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3642     {
3643         h->b_sh_backup = 0;
3644         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3645         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3646         if( x264_nal_end( h ) )
3647             return -1;
3648         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3649     }
3650
3651     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3652         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3653
3654     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3655     if( h->param.i_avcintra_class )
3656     {
3657         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3658         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3659         x264_filler_write( h, &h->out.bs, 0 );
3660         if( x264_nal_end( h ) )
3661             return -1;
3662         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3663
3664         /* All lengths are magic lengths that decoders expect to see */
3665         /* "UMID" SEI */
3666         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3667         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3668             return -1;
3669         if( x264_nal_end( h ) )
3670             return -1;
3671         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3672
3673         int unpadded_len;
3674         int total_len;
3675         if( h->param.i_height == 1080 )
3676         {
3677             unpadded_len = 5780;
3678             total_len = 17*512;
3679         }
3680         else
3681         {
3682             unpadded_len = 2900;
3683             total_len = 9*512;
3684         }
3685         /* "VANC" SEI */
3686         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3687         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3688             return -1;
3689         if( x264_nal_end( h ) )
3690             return -1;
3691
3692         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3693         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3694     }
3695
3696     /* Init the speed control */
3697     if( h->param.sc.i_buffer_size )
3698         x264_speedcontrol_frame( h );
3699
3700     /* Init the rate control */
3701     /* FIXME: Include slice header bit cost. */
3702     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3703     i_global_qp = x264_ratecontrol_qp( h );
3704
3705     pic_out->i_qpplus1 =
3706     h->fdec->i_qpplus1 = i_global_qp + 1;
3707
3708     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3709     {
3710         x264_reference_build_list_optimal( h );
3711         x264_reference_check_reorder( h );
3712     }
3713
3714     if( h->i_ref[0] )
3715         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3716
3717     /* ------------------------ Create slice header  ----------------------- */
3718     x264_slice_init( h, i_nal_type, i_global_qp );
3719
3720     /*------------------------- Weights -------------------------------------*/
3721     if( h->sh.i_type == SLICE_TYPE_B )
3722         x264_macroblock_bipred_init( h );
3723
3724     x264_weighted_pred_init( h );
3725
3726     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3727         h->i_frame_num++;
3728
3729     /* Write frame */
3730     h->i_threadslice_start = 0;
3731     h->i_threadslice_end = h->mb.i_mb_height;
3732     if( h->i_thread_frames > 1 )
3733     {
3734         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3735         h->b_thread_active = 1;
3736     }
3737     else if( h->param.b_sliced_threads )
3738     {
3739         if( x264_threaded_slices_write( h ) )
3740             return -1;
3741     }
3742     else
3743         if( (intptr_t)x264_slices_write( h ) )
3744             return -1;
3745
3746     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3747 }
3748
3749 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3750                                    x264_nal_t **pp_nal, int *pi_nal,
3751                                    x264_picture_t *pic_out )
3752 {
3753     char psz_message[80];
3754
3755     if( !h->param.b_sliced_threads && h->b_thread_active )
3756     {
3757         h->b_thread_active = 0;
3758         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3759             return -1;
3760     }
3761     if( !h->out.i_nal )
3762     {
3763         pic_out->i_type = X264_TYPE_AUTO;
3764         return 0;
3765     }
3766
3767     x264_emms();
3768
3769     /* generate buffering period sei and insert it into place */
3770     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3771     {
3772         x264_hrd_fullness( h );
3773         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3774         x264_sei_buffering_period_write( h, &h->out.bs );
3775         if( x264_nal_end( h ) )
3776            return -1;
3777         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3778         int idx = 0;
3779         while( h->out.nal[idx].i_type == NAL_AUD ||
3780                h->out.nal[idx].i_type == NAL_SPS ||
3781                h->out.nal[idx].i_type == NAL_PPS )
3782             idx++;
3783         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3784         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3785         h->out.nal[idx] = nal_tmp;
3786     }
3787
3788     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3789     if( frame_size < 0 )
3790         return -1;
3791
3792     /* Set output picture properties */
3793     pic_out->i_type = h->fenc->i_type;
3794
3795     pic_out->b_keyframe = h->fenc->b_keyframe;
3796     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3797
3798     pic_out->i_pts = h->fdec->i_pts;
3799     pic_out->i_dts = h->fdec->i_dts;
3800
3801     if( pic_out->i_pts < pic_out->i_dts )
3802         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3803
3804     pic_out->opaque = h->fenc->opaque;
3805
3806     pic_out->img.i_csp = h->fdec->i_csp;
3807 #if HIGH_BIT_DEPTH
3808     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3809 #endif
3810     pic_out->img.i_plane = h->fdec->i_plane;
3811     for( int i = 0; i < pic_out->img.i_plane; i++ )
3812     {
3813         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3814         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3815     }
3816
3817     x264_frame_push_unused( thread_current, h->fenc );
3818
3819     /* ---------------------- Update encoder state ------------------------- */
3820
3821     /* update rc */
3822     int filler = 0;
3823     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3824         return -1;
3825
3826     pic_out->hrd_timing = h->fenc->hrd_timing;
3827     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3828
3829     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3830      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3831     if( h->param.i_avcintra_class )
3832     {
3833         if( x264_check_encapsulated_buffer( h, h->thread[0], h->out.i_nal, frame_size, frame_size + filler ) < 0 )
3834             return -1;
3835
3836         x264_nal_t *nal = &h->out.nal[h->out.i_nal-1];
3837         memset( nal->p_payload + nal->i_payload, 0, filler );
3838         nal->i_payload += filler;
3839         nal->i_padding = filler;
3840         frame_size += filler;
3841
3842         /* Fix up the size header for mp4/etc */
3843         if( !h->param.b_annexb )
3844         {
3845             /* Size doesn't include the size of the header we're writing now. */
3846             uint8_t *nal_data = nal->p_payload;
3847             int chunk_size = nal->i_payload - 4;
3848             nal_data[0] = chunk_size >> 24;
3849             nal_data[1] = chunk_size >> 16;
3850             nal_data[2] = chunk_size >> 8;
3851             nal_data[3] = chunk_size >> 0;
3852         }
3853     }
3854     else
3855     {
3856         while( filler > 0 )
3857         {
3858             int f, overhead = FILLER_OVERHEAD - h->param.b_annexb;
3859             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3860             {
3861                 int next_size = filler - h->param.i_slice_max_size;
3862                 int overflow = X264_MAX( overhead - next_size, 0 );
3863                 f = h->param.i_slice_max_size - overhead - overflow;
3864             }
3865             else
3866                 f = X264_MAX( 0, filler - overhead );
3867
3868             if( x264_bitstream_check_buffer_filler( h, f ) )
3869                 return -1;
3870             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3871             x264_filler_write( h, &h->out.bs, f );
3872             if( x264_nal_end( h ) )
3873                 return -1;
3874             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3875             if( total_size < 0 )
3876                 return -1;
3877             frame_size += total_size;
3878             filler -= total_size;
3879         }
3880     }
3881
3882     /* End bitstream, set output  */
3883     *pi_nal = h->out.i_nal;
3884     *pp_nal = h->out.nal;
3885
3886     h->out.i_nal = 0;
3887
3888     x264_noise_reduction_update( h );
3889
3890     if( h->param.sc.i_buffer_size )
3891         x264_speedcontrol_frame_end( h );
3892
3893     /* ---------------------- Compute/Print statistics --------------------- */
3894     x264_thread_sync_stat( h, h->thread[0] );
3895
3896     /* Slice stat */
3897     h->stat.i_frame_count[h->sh.i_type]++;
3898     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3899     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3900
3901     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3902         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3903     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3904         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3905     for( int i = 0; i < 2; i++ )
3906         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3907     for( int i = 0; i < 6; i++ )
3908         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3909     for( int i = 0; i < 4; i++ )
3910         for( int j = 0; j < 13; j++ )
3911             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3912     if( h->sh.i_type != SLICE_TYPE_I )
3913         for( int i_list = 0; i_list < 2; i_list++ )
3914             for( int i = 0; i < X264_REF_MAX*2; i++ )
3915                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3916     for( int i = 0; i < 3; i++ )
3917         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3918     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3919     {
3920         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3921         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3922     }
3923     if( h->sh.i_type == SLICE_TYPE_B )
3924     {
3925         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3926         if( h->mb.b_direct_auto_write )
3927         {
3928             //FIXME somewhat arbitrary time constants
3929             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3930                 for( int i = 0; i < 2; i++ )
3931                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3932             for( int i = 0; i < 2; i++ )
3933                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3934         }
3935     }
3936     else
3937         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3938
3939     psz_message[0] = '\0';
3940     double dur = h->fenc->f_duration;
3941     h->stat.f_frame_duration[h->sh.i_type] += dur;
3942     if( h->param.analyse.b_psnr )
3943     {
3944         int64_t ssd[3] =
3945         {
3946             h->stat.frame.i_ssd[0],
3947             h->stat.frame.i_ssd[1],
3948             h->stat.frame.i_ssd[2],
3949         };
3950         int luma_size = h->param.i_width * h->param.i_height;
3951         int chroma_size = CHROMA_SIZE( luma_size );
3952         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3953         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3954         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3955         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3956
3957         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3958         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3959         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3960         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3961         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3962
3963         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3964                                                                     pic_out->prop.f_psnr[1],
3965                                                                     pic_out->prop.f_psnr[2] );
3966     }
3967
3968     if( h->param.analyse.b_ssim )
3969     {
3970         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3971         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3972         int msg_len = strlen(psz_message);
3973         snprintf( psz_message + msg_len, 80 - msg_len, " SSIM Y:%.5f", pic_out->prop.f_ssim );
3974     }
3975     psz_message[79] = '\0';
3976
3977     x264_log( h, X264_LOG_DEBUG,
3978               "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3979               h->i_frame,
3980               h->fdec->f_qp_avg_aq,
3981               h->i_nal_ref_idc,
3982               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3983               h->fdec->i_poc,
3984               h->stat.frame.i_mb_count_i,
3985               h->stat.frame.i_mb_count_p,
3986               h->stat.frame.i_mb_count_skip,
3987               frame_size,
3988               psz_message );
3989
3990     // keep stats all in one place
3991     x264_thread_sync_stat( h->thread[0], h );
3992     // for the use of the next frame
3993     x264_thread_sync_stat( thread_current, h );
3994
3995 #ifdef DEBUG_MB_TYPE
3996 {
3997     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3998         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3999     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
4000     {
4001         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
4002             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
4003         else
4004             fprintf( stderr, "? " );
4005
4006         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
4007             fprintf( stderr, "\n" );
4008     }
4009 }
4010 #endif
4011
4012     /* Remove duplicates, must be done near the end as breaks h->fref0 array
4013      * by freeing some of its pointers. */
4014     for( int i = 0; i < h->i_ref[0]; i++ )
4015         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
4016         {
4017             x264_frame_push_blank_unused( h, h->fref[0][i] );
4018             h->fref[0][i] = 0;
4019         }
4020
4021     if( h->param.psz_dump_yuv )
4022         x264_frame_dump( h );
4023     x264_emms();
4024
4025     return frame_size;
4026 }
4027
4028 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
4029 {
4030     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
4031         b_print_pcm ? "..PCM" : "",
4032         i_mb_count[I_16x16]/ i_count,
4033         i_mb_count[I_8x8]  / i_count,
4034         i_mb_count[I_4x4]  / i_count );
4035     if( b_print_pcm )
4036         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
4037 }
4038
4039 /****************************************************************************
4040  * x264_encoder_close:
4041  ****************************************************************************/
4042 void    x264_encoder_close  ( x264_t *h )
4043 {
4044     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
4045     int64_t i_mb_count_size[2][7] = {{0}};
4046     char buf[200];
4047     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
4048                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
4049                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
4050
4051     x264_lookahead_delete( h );
4052
4053 #if HAVE_OPENCL
4054     x264_opencl_lookahead_delete( h );
4055     x264_opencl_function_t *ocl = h->opencl.ocl;
4056 #endif
4057
4058     if( h->param.b_sliced_threads )
4059         x264_threadpool_wait_all( h );
4060     if( h->param.i_threads > 1 )
4061         x264_threadpool_delete( h->threadpool );
4062     if( h->param.i_lookahead_threads > 1 )
4063         x264_threadpool_delete( h->lookaheadpool );
4064     if( h->i_thread_frames > 1 )
4065     {
4066         for( int i = 0; i < h->i_thread_frames; i++ )
4067             if( h->thread[i]->b_thread_active )
4068             {
4069                 assert( h->thread[i]->fenc->i_reference_count == 1 );
4070                 x264_frame_delete( h->thread[i]->fenc );
4071             }
4072
4073         x264_t *thread_prev = h->thread[h->i_thread_phase];
4074         x264_thread_sync_ratecontrol( h, thread_prev, h );
4075         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
4076         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
4077     }
4078     h->i_frame++;
4079
4080     /* Slices used and PSNR */
4081     for( int i = 0; i < 3; i++ )
4082     {
4083         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
4084         int i_slice = slice_order[i];
4085
4086         if( h->stat.i_frame_count[i_slice] > 0 )
4087         {
4088             int i_count = h->stat.i_frame_count[i_slice];
4089             double dur =  h->stat.f_frame_duration[i_slice];
4090             if( h->param.analyse.b_psnr )
4091             {
4092                 x264_log( h, X264_LOG_INFO,
4093                           "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",
4094                           slice_type_to_char[i_slice],
4095                           i_count,
4096                           h->stat.f_frame_qp[i_slice] / i_count,
4097                           (double)h->stat.i_frame_size[i_slice] / i_count,
4098                           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,
4099                           h->stat.f_psnr_average[i_slice] / dur,
4100                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4101             }
4102             else
4103             {
4104                 x264_log( h, X264_LOG_INFO,
4105                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4106                           slice_type_to_char[i_slice],
4107                           i_count,
4108                           h->stat.f_frame_qp[i_slice] / i_count,
4109                           (double)h->stat.i_frame_size[i_slice] / i_count );
4110             }
4111         }
4112     }
4113     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4114     {
4115         char *p = buf;
4116         int den = 0;
4117         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4118         for( int i = 0; i <= h->param.i_bframe; i++ )
4119             den += (i+1) * h->stat.i_consecutive_bframes[i];
4120         for( int i = 0; i <= h->param.i_bframe; i++ )
4121             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4122         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4123     }
4124
4125     for( int i_type = 0; i_type < 2; i_type++ )
4126         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4127         {
4128             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4129             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4130         }
4131
4132     /* MB types used */
4133     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4134     {
4135         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4136         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4137         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4138         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4139     }
4140     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4141     {
4142         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4143         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4144         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4145         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4146         x264_log( h, X264_LOG_INFO,
4147                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4148                   buf,
4149                   i_mb_size[PIXEL_16x16] / (i_count*4),
4150                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4151                   i_mb_size[PIXEL_8x8] / (i_count*4),
4152                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4153                   i_mb_size[PIXEL_4x4] / (i_count*4),
4154                   i_mb_count[P_SKIP] / i_count );
4155     }
4156     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4157     {
4158         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4159         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4160         double i_mb_list_count;
4161         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4162         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4163         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4164         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4165             for( int j = 0; j < 2; j++ )
4166             {
4167                 int l0 = x264_mb_type_list_table[i][0][j];
4168                 int l1 = x264_mb_type_list_table[i][1][j];
4169                 if( l0 || l1 )
4170                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4171             }
4172         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4173         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4174         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4175         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4176         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4177         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4178                  i_mb_size[PIXEL_16x16] / (i_count*4),
4179                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4180                  i_mb_size[PIXEL_8x8] / (i_count*4),
4181                  i_mb_count[B_DIRECT] / i_count,
4182                  i_mb_count[B_SKIP]   / i_count );
4183         if( i_mb_list_count != 0 )
4184             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4185                      list_count[0] / i_mb_list_count,
4186                      list_count[1] / i_mb_list_count,
4187                      list_count[2] / i_mb_list_count );
4188         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4189     }
4190
4191     x264_ratecontrol_summary( h );
4192
4193     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 )
4194     {
4195 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4196 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4197         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4198         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4199                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4200         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4201         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4202                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4203         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4204                             h->stat.i_frame_count[SLICE_TYPE_P] +
4205                             h->stat.i_frame_count[SLICE_TYPE_B];
4206         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4207         int64_t i_inter = i_mb_count - i_skip - i_intra;
4208         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4209                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4210                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4211         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4212
4213         if( PARAM_INTERLACED )
4214         {
4215             char *fieldstats = buf;
4216             fieldstats[0] = 0;
4217             if( i_inter )
4218                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4219             if( i_skip )
4220                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4221             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4222                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4223         }
4224
4225         if( h->pps->b_transform_8x8_mode )
4226         {
4227             buf[0] = 0;
4228             if( h->stat.i_mb_count_8x8dct[0] )
4229                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4230             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4231         }
4232
4233         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4234             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4235             && h->stat.i_frame_count[SLICE_TYPE_B] )
4236         {
4237             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4238                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4239                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4240         }
4241
4242         buf[0] = 0;
4243         int csize = CHROMA444 ? 4 : 1;
4244         if( i_mb_count != i_all_intra )
4245             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4246                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4247                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4248                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4249         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4250                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4251                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4252                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4253                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4254
4255         int64_t fixed_pred_modes[4][9] = {{0}};
4256         int64_t sum_pred_modes[4] = {0};
4257         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4258         {
4259             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4260             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4261         }
4262         if( sum_pred_modes[0] )
4263             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4264                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4265                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4266                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4267                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4268         for( int i = 1; i <= 2; i++ )
4269         {
4270             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
4271             {
4272                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
4273                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
4274             }
4275             if( sum_pred_modes[i] )
4276                 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,
4277                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
4278                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
4279                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
4280                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
4281                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
4282                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
4283                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
4284                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
4285                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
4286         }
4287         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
4288         {
4289             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
4290             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
4291         }
4292         if( sum_pred_modes[3] && !CHROMA444 )
4293             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4294                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
4295                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
4296                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
4297                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
4298
4299         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4300             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
4301                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
4302                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
4303
4304         for( int i_list = 0; i_list < 2; i_list++ )
4305             for( int i_slice = 0; i_slice < 2; i_slice++ )
4306             {
4307                 char *p = buf;
4308                 int64_t i_den = 0;
4309                 int i_max = 0;
4310                 for( int i = 0; i < X264_REF_MAX*2; i++ )
4311                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
4312                     {
4313                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
4314                         i_max = i;
4315                     }
4316                 if( i_max == 0 )
4317                     continue;
4318                 for( int i = 0; i <= i_max; i++ )
4319                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
4320                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
4321             }
4322
4323         if( h->param.analyse.b_ssim )
4324         {
4325             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
4326             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
4327         }
4328         if( h->param.analyse.b_psnr )
4329         {
4330             x264_log( h, X264_LOG_INFO,
4331                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
4332                       SUM3( h->stat.f_psnr_mean_y ) / duration,
4333                       SUM3( h->stat.f_psnr_mean_u ) / duration,
4334                       SUM3( h->stat.f_psnr_mean_v ) / duration,
4335                       SUM3( h->stat.f_psnr_average ) / duration,
4336                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
4337                       f_bitrate );
4338         }
4339         else
4340             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
4341     }
4342
4343     /* rc */
4344     x264_ratecontrol_delete( h );
4345     x264_speedcontrol_delete( h );
4346
4347     /* param */
4348     if( h->param.rc.psz_stat_out )
4349         free( h->param.rc.psz_stat_out );
4350     if( h->param.rc.psz_stat_in )
4351         free( h->param.rc.psz_stat_in );
4352
4353     x264_cqm_delete( h );
4354     x264_free( h->nal_buffer );
4355     x264_free( h->reconfig_h );
4356     x264_analyse_free_costs( h );
4357
4358     if( h->i_thread_frames > 1 )
4359         h = h->thread[h->i_thread_phase];
4360
4361     /* frames */
4362     x264_frame_delete_list( h->frames.unused[0] );
4363     x264_frame_delete_list( h->frames.unused[1] );
4364     x264_frame_delete_list( h->frames.current );
4365     x264_frame_delete_list( h->frames.blank_unused );
4366
4367     h = h->thread[0];
4368
4369     for( int i = 0; i < h->i_thread_frames; i++ )
4370         if( h->thread[i]->b_thread_active )
4371             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4372                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4373                     x264_frame_delete( h->thread[i]->fref[0][j] );
4374
4375     if( h->param.i_lookahead_threads > 1 )
4376         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4377             x264_free( h->lookahead_thread[i] );
4378
4379     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4380     {
4381         x264_frame_t **frame;
4382
4383         if( !h->param.b_sliced_threads || i == 0 )
4384         {
4385             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4386             {
4387                 assert( (*frame)->i_reference_count > 0 );
4388                 (*frame)->i_reference_count--;
4389                 if( (*frame)->i_reference_count == 0 )
4390                     x264_frame_delete( *frame );
4391             }
4392             frame = &h->thread[i]->fdec;
4393             if( *frame )
4394             {
4395                 assert( (*frame)->i_reference_count > 0 );
4396                 (*frame)->i_reference_count--;
4397                 if( (*frame)->i_reference_count == 0 )
4398                     x264_frame_delete( *frame );
4399             }
4400             x264_macroblock_cache_free( h->thread[i] );
4401         }
4402         x264_macroblock_thread_free( h->thread[i], 0 );
4403         x264_free( h->thread[i]->out.p_bitstream );
4404         x264_free( h->thread[i]->out.nal );
4405         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4406         x264_pthread_cond_destroy( &h->thread[i]->cv );
4407         x264_free( h->thread[i] );
4408     }
4409 #if HAVE_OPENCL
4410     x264_opencl_close_library( ocl );
4411 #endif
4412 }
4413
4414 int x264_encoder_delayed_frames( x264_t *h )
4415 {
4416     int delayed_frames = 0;
4417     if( h->i_thread_frames > 1 )
4418     {
4419         for( int i = 0; i < h->i_thread_frames; i++ )
4420             delayed_frames += h->thread[i]->b_thread_active;
4421         h = h->thread[h->i_thread_phase];
4422     }
4423     for( int i = 0; h->frames.current[i]; i++ )
4424         delayed_frames++;
4425     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4426     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4427     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4428     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4429     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4430     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4431     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4432     return delayed_frames;
4433 }
4434
4435 int x264_encoder_maximum_delayed_frames( x264_t *h )
4436 {
4437     return h->frames.i_delay;
4438 }