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     /* ensure the booleans are 0 or 1 so they can be used in math */
1276 #define BOOLIFY(x) h->param.x = !!h->param.x
1277     BOOLIFY( b_cabac );
1278     BOOLIFY( b_constrained_intra );
1279     BOOLIFY( b_deblocking_filter );
1280     BOOLIFY( b_deterministic );
1281     BOOLIFY( b_sliced_threads );
1282     BOOLIFY( b_interlaced );
1283     BOOLIFY( b_intra_refresh );
1284     BOOLIFY( b_aud );
1285     BOOLIFY( b_repeat_headers );
1286     BOOLIFY( b_annexb );
1287     BOOLIFY( b_vfr_input );
1288     BOOLIFY( b_pulldown );
1289     BOOLIFY( b_tff );
1290     BOOLIFY( b_pic_struct );
1291     BOOLIFY( b_fake_interlaced );
1292     BOOLIFY( b_open_gop );
1293     BOOLIFY( b_bluray_compat );
1294     BOOLIFY( b_stitchable );
1295     BOOLIFY( b_full_recon );
1296     BOOLIFY( b_opencl );
1297     BOOLIFY( analyse.b_transform_8x8 );
1298     BOOLIFY( analyse.b_weighted_bipred );
1299     BOOLIFY( analyse.b_chroma_me );
1300     BOOLIFY( analyse.b_mixed_references );
1301     BOOLIFY( analyse.b_fast_pskip );
1302     BOOLIFY( analyse.b_dct_decimate );
1303     BOOLIFY( analyse.b_psy );
1304     BOOLIFY( analyse.b_psnr );
1305     BOOLIFY( analyse.b_ssim );
1306     BOOLIFY( rc.b_stat_write );
1307     BOOLIFY( rc.b_stat_read );
1308     BOOLIFY( rc.b_mb_tree );
1309     BOOLIFY( rc.b_filler );
1310 #undef BOOLIFY
1311
1312     return 0;
1313 }
1314
1315 static void mbcmp_init( x264_t *h )
1316 {
1317     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1318     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1319     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1320     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1321     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1322     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1323     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1324     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1325     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1326                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1327     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1328                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1329     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1330     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1331     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1332     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1333 }
1334
1335 static void chroma_dsp_init( x264_t *h )
1336 {
1337     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1338
1339     switch( CHROMA_FORMAT )
1340     {
1341         case CHROMA_420:
1342             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1343             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1344             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1345             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1346             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1347             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1348             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1349             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1350             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1351             break;
1352         case CHROMA_422:
1353             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1354             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1355             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1356             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1357             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1358             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1359             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1360             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1361             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1362             break;
1363         case CHROMA_444:
1364             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1365             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1366             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1367             break;
1368     }
1369 }
1370
1371 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1372 {
1373     /* VUI */
1374     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1375     {
1376         uint32_t i_w = param->vui.i_sar_width;
1377         uint32_t i_h = param->vui.i_sar_height;
1378         uint32_t old_w = h->param.vui.i_sar_width;
1379         uint32_t old_h = h->param.vui.i_sar_height;
1380
1381         x264_reduce_fraction( &i_w, &i_h );
1382
1383         while( i_w > 65535 || i_h > 65535 )
1384         {
1385             i_w /= 2;
1386             i_h /= 2;
1387         }
1388
1389         x264_reduce_fraction( &i_w, &i_h );
1390
1391         if( i_w != old_w || i_h != old_h || initial )
1392         {
1393             h->param.vui.i_sar_width = 0;
1394             h->param.vui.i_sar_height = 0;
1395             if( i_w == 0 || i_h == 0 )
1396                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1397             else
1398             {
1399                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1400                 h->param.vui.i_sar_width = i_w;
1401                 h->param.vui.i_sar_height = i_h;
1402             }
1403         }
1404     }
1405 }
1406
1407 /****************************************************************************
1408  * x264_encoder_open:
1409  ****************************************************************************/
1410 x264_t *x264_encoder_open( x264_param_t *param )
1411 {
1412     x264_t *h;
1413     char buf[1000], *p;
1414     int i_slicetype_length;
1415
1416     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1417
1418     /* Create a copy of param */
1419     memcpy( &h->param, param, sizeof(x264_param_t) );
1420
1421     if( param->param_free )
1422         param->param_free( param );
1423
1424 #if HAVE_INTEL_DISPATCHER
1425     x264_intel_dispatcher_override();
1426 #endif
1427
1428     if( x264_threading_init() )
1429     {
1430         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1431         goto fail;
1432     }
1433
1434     if( x264_validate_parameters( h, 1 ) < 0 )
1435         goto fail;
1436
1437     if( h->param.psz_cqm_file )
1438         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1439             goto fail;
1440
1441     if( h->param.rc.psz_stat_out )
1442         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1443     if( h->param.rc.psz_stat_in )
1444         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1445
1446     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1447     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1448
1449     /* Init x264_t */
1450     h->i_frame = -1;
1451     h->i_frame_num = 0;
1452
1453     if( h->param.i_avcintra_class )
1454         h->i_idr_pic_id = 5;
1455     else
1456         h->i_idr_pic_id = 0;
1457
1458     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1459     {
1460         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1461         goto fail;
1462     }
1463
1464     x264_set_aspect_ratio( h, &h->param, 1 );
1465
1466     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1467     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1468
1469     x264_validate_levels( h, 1 );
1470
1471     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1472
1473     if( x264_cqm_init( h ) < 0 )
1474         goto fail;
1475
1476     h->mb.i_mb_width = h->sps->i_mb_width;
1477     h->mb.i_mb_height = h->sps->i_mb_height;
1478     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1479
1480     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1481     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1482
1483     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1484      * vectors during prediction, resulting in hpel mvs.
1485      * The chosen solution is to make MBAFF non-adaptive in this case. */
1486     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1487
1488     /* Init frames. */
1489     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1490         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1491     else
1492         h->frames.i_delay = h->param.i_bframe;
1493     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1494         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1495     i_slicetype_length = h->frames.i_delay;
1496     h->frames.i_delay += h->i_thread_frames - 1;
1497     h->frames.i_delay += h->param.i_sync_lookahead;
1498     h->frames.i_delay += h->param.b_vfr_input;
1499     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1500
1501     h->frames.i_max_ref0 = h->param.i_frame_reference;
1502     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1503     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1504     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1505         && ( h->param.rc.i_rc_method == X264_RC_ABR
1506           || h->param.rc.i_rc_method == X264_RC_CRF
1507           || h->param.i_bframe_adaptive
1508           || h->param.i_scenecut_threshold
1509           || h->param.rc.b_mb_tree
1510           || h->param.analyse.i_weighted_pred );
1511     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1512     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1513
1514     h->frames.i_last_idr =
1515     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1516     h->frames.i_input    = 0;
1517     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1518     h->frames.i_poc_last_open_gop = -1;
1519
1520     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1521     /* Allocate room for max refs plus a few extra just in case. */
1522     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1523     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1524                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1525     if( h->param.analyse.i_weighted_pred > 0 )
1526         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1527     h->i_ref[0] = h->i_ref[1] = 0;
1528     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1529     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);
1530     h->i_disp_fields_last_frame = -1;
1531     x264_rdo_init();
1532
1533     /* init CPU functions */
1534     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1535     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1536     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1537     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1538     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1539     x264_pixel_init( h->param.cpu, &h->pixf );
1540     x264_dct_init( h->param.cpu, &h->dctf );
1541     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1542     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1543     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1544     x264_quant_init( h, h->param.cpu, &h->quantf );
1545     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1546     x264_bitstream_init( h->param.cpu, &h->bsf );
1547     if( h->param.b_cabac )
1548         x264_cabac_init( h );
1549     else
1550         x264_stack_align( x264_cavlc_init, h );
1551
1552     mbcmp_init( h );
1553     chroma_dsp_init( h );
1554
1555     if( h->param.sc.i_buffer_size )
1556         x264_speedcontrol_new( h );
1557
1558
1559     p = buf + sprintf( buf, "using cpu capabilities:" );
1560     for( int i = 0; x264_cpu_names[i].flags; i++ )
1561     {
1562         if( !strcmp(x264_cpu_names[i].name, "SSE")
1563             && h->param.cpu & (X264_CPU_SSE2) )
1564             continue;
1565         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1566             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1567             continue;
1568         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1569             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1570             continue;
1571         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1572             && (h->param.cpu & X264_CPU_SSE42) )
1573             continue;
1574         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1575             && (h->param.cpu & X264_CPU_BMI2) )
1576             continue;
1577         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1578             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1579             p += sprintf( p, " %s", x264_cpu_names[i].name );
1580     }
1581     if( !h->param.cpu )
1582         p += sprintf( p, " none!" );
1583     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1584
1585     if( x264_analyse_init_costs( h ) )
1586         goto fail;
1587
1588     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1589     /* Checks for known miscompilation issues. */
1590     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1591     {
1592         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1593         goto fail;
1594     }
1595
1596     /* Must be volatile or else GCC will optimize it out. */
1597     volatile int temp = 392;
1598     if( x264_clz( temp ) != 23 )
1599     {
1600         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1601 #if ARCH_X86 || ARCH_X86_64
1602         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1603         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1604 #endif
1605         goto fail;
1606     }
1607
1608     h->out.i_nal = 0;
1609     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1610         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1611           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1612
1613     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1614     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1615
1616     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1617
1618     if( h->param.i_threads > 1 &&
1619         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1620         goto fail;
1621     if( h->param.i_lookahead_threads > 1 &&
1622         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1623         goto fail;
1624
1625 #if HAVE_OPENCL
1626     if( h->param.b_opencl )
1627     {
1628         h->opencl.ocl = x264_opencl_load_library();
1629         if( !h->opencl.ocl )
1630         {
1631             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1632             h->param.b_opencl = 0;
1633         }
1634     }
1635 #endif
1636
1637     h->thread[0] = h;
1638     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1639         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1640     if( h->param.i_lookahead_threads > 1 )
1641         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1642         {
1643             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1644             *h->lookahead_thread[i] = *h;
1645         }
1646     *h->reconfig_h = *h;
1647
1648     for( int i = 0; i < h->param.i_threads; i++ )
1649     {
1650         int init_nal_count = h->param.i_slice_count + 3;
1651         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1652         if( i > 0 )
1653             *h->thread[i] = *h;
1654
1655         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1656             goto fail;
1657         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1658             goto fail;
1659
1660         if( allocate_threadlocal_data )
1661         {
1662             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1663             if( !h->thread[i]->fdec )
1664                 goto fail;
1665         }
1666         else
1667             h->thread[i]->fdec = h->thread[0]->fdec;
1668
1669         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1670         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1671         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1672         h->thread[i]->out.i_nals_allocated = init_nal_count;
1673
1674         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1675             goto fail;
1676     }
1677
1678 #if HAVE_OPENCL
1679     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1680         h->param.b_opencl = 0;
1681 #endif
1682
1683     if( x264_lookahead_init( h, i_slicetype_length ) )
1684         goto fail;
1685
1686     for( int i = 0; i < h->param.i_threads; i++ )
1687         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1688             goto fail;
1689
1690     if( x264_ratecontrol_new( h ) < 0 )
1691         goto fail;
1692
1693     if( h->param.i_nal_hrd )
1694     {
1695         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1696         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1697     }
1698
1699     if( h->param.psz_dump_yuv )
1700     {
1701         /* create or truncate the reconstructed video file */
1702         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1703         if( !f )
1704         {
1705             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1706             goto fail;
1707         }
1708         else if( !x264_is_regular_file( f ) )
1709         {
1710             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1711             fclose( f );
1712             goto fail;
1713         }
1714         fclose( f );
1715     }
1716
1717     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1718                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1719                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1720                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1721                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1722                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1723     char level[4];
1724     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1725     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1726         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1727         strcpy( level, "1b" );
1728
1729     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1730     {
1731         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1732             profile, level );
1733     }
1734     else
1735     {
1736         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1737         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1738             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1739     }
1740
1741     return h;
1742 fail:
1743     x264_free( h );
1744     return NULL;
1745 }
1746
1747 /****************************************************************************/
1748 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1749 {
1750     *rc_reconfig = 0;
1751     x264_set_aspect_ratio( h, param, 0 );
1752 #define COPY(var) h->param.var = param->var
1753     COPY( i_frame_reference ); // but never uses more refs than initially specified
1754     COPY( i_bframe_bias );
1755     if( h->param.i_scenecut_threshold )
1756         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1757     COPY( b_deblocking_filter );
1758     COPY( i_deblocking_filter_alphac0 );
1759     COPY( i_deblocking_filter_beta );
1760     COPY( i_frame_packing );
1761     COPY( analyse.inter );
1762     COPY( analyse.intra );
1763     COPY( analyse.i_direct_mv_pred );
1764     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1765     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1766         COPY( analyse.i_me_range );
1767     COPY( analyse.i_noise_reduction );
1768     /* We can't switch out of subme=0 during encoding. */
1769     if( h->param.analyse.i_subpel_refine )
1770         COPY( analyse.i_subpel_refine );
1771     COPY( analyse.i_trellis );
1772     COPY( analyse.b_chroma_me );
1773     COPY( analyse.b_dct_decimate );
1774     COPY( analyse.b_fast_pskip );
1775     COPY( analyse.b_mixed_references );
1776     COPY( analyse.f_psy_rd );
1777     COPY( analyse.f_psy_trellis );
1778     COPY( crop_rect );
1779     // can only twiddle these if they were enabled to begin with:
1780     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1781         COPY( analyse.i_me_method );
1782     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1783         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1784     if( h->pps->b_transform_8x8_mode )
1785         COPY( analyse.b_transform_8x8 );
1786     if( h->frames.i_max_ref1 > 1 )
1787         COPY( i_bframe_pyramid );
1788     COPY( i_slice_max_size );
1789     COPY( i_slice_max_mbs );
1790     COPY( i_slice_min_mbs );
1791     COPY( i_slice_count );
1792     COPY( i_slice_count_max );
1793     COPY( b_tff );
1794
1795     /* VBV can't be turned on if it wasn't on to begin with */
1796     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1797           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1798     {
1799         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1800         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1801         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1802         COPY( rc.i_vbv_max_bitrate );
1803         COPY( rc.i_vbv_buffer_size );
1804         COPY( rc.i_bitrate );
1805     }
1806     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1807     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1808     COPY( rc.f_rf_constant );
1809     COPY( rc.f_rf_constant_max );
1810 #undef COPY
1811
1812     return x264_validate_parameters( h, 0 );
1813 }
1814
1815 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1816 {
1817     int rc_reconfig;
1818     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1819
1820     mbcmp_init( h );
1821     if( !ret )
1822         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1823
1824     /* Supported reconfiguration options (1-pass only):
1825      * vbv-maxrate
1826      * vbv-bufsize
1827      * crf
1828      * bitrate (CBR only) */
1829     if( !ret && rc_reconfig )
1830         x264_ratecontrol_init_reconfigurable( h, 0 );
1831
1832     return ret;
1833 }
1834
1835 /****************************************************************************
1836  * x264_encoder_reconfig:
1837  ****************************************************************************/
1838 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1839 {
1840     h = h->thread[h->thread[0]->i_thread_phase];
1841     x264_param_t param_save = h->reconfig_h->param;
1842     h->reconfig_h->param = h->param;
1843
1844     int rc_reconfig;
1845     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1846     if( !ret )
1847         h->reconfig = 1;
1848     else
1849         h->reconfig_h->param = param_save;
1850
1851     return ret;
1852 }
1853
1854 /****************************************************************************
1855  * x264_encoder_parameters:
1856  ****************************************************************************/
1857 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1858 {
1859     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1860 }
1861
1862 /* internal usage */
1863 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1864 {
1865     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1866
1867     nal->i_ref_idc        = i_ref_idc;
1868     nal->i_type           = i_type;
1869     nal->b_long_startcode = 1;
1870
1871     nal->i_payload= 0;
1872     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1873     nal->i_padding= 0;
1874 }
1875
1876 /* if number of allocated nals is not enough, re-allocate a larger one. */
1877 static int x264_nal_check_buffer( x264_t *h )
1878 {
1879     if( h->out.i_nal >= h->out.i_nals_allocated )
1880     {
1881         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1882         if( !new_out )
1883             return -1;
1884         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1885         x264_free( h->out.nal );
1886         h->out.nal = new_out;
1887         h->out.i_nals_allocated *= 2;
1888     }
1889     return 0;
1890 }
1891
1892 static int x264_nal_end( x264_t *h )
1893 {
1894     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1895     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1896     nal->i_payload = end - nal->p_payload;
1897     /* Assembly implementation of nal_escape reads past the end of the input.
1898      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1899     memset( end, 0xff, 64 );
1900     if( h->param.nalu_process )
1901         h->param.nalu_process( h, nal, h->fenc->opaque );
1902     h->out.i_nal++;
1903
1904     return x264_nal_check_buffer( h );
1905 }
1906
1907 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1908                                            int previous_nal_size, int necessary_size )
1909 {
1910     if( h0->nal_buffer_size < necessary_size )
1911     {
1912         necessary_size *= 2;
1913         uint8_t *buf = x264_malloc( necessary_size );
1914         if( !buf )
1915             return -1;
1916         if( previous_nal_size )
1917             memcpy( buf, h0->nal_buffer, previous_nal_size );
1918
1919         intptr_t delta = buf - h0->nal_buffer;
1920         for( int i = 0; i < start; i++ )
1921             h->out.nal[i].p_payload += delta;
1922
1923         x264_free( h0->nal_buffer );
1924         h0->nal_buffer = buf;
1925         h0->nal_buffer_size = necessary_size;
1926     }
1927
1928     return 0;
1929 }
1930
1931 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1932 {
1933     x264_t *h0 = h->thread[0];
1934     int nal_size = 0, previous_nal_size = 0;
1935
1936     if( h->param.nalu_process )
1937     {
1938         for( int i = start; i < h->out.i_nal; i++ )
1939             nal_size += h->out.nal[i].i_payload;
1940         return nal_size;
1941     }
1942
1943     for( int i = 0; i < start; i++ )
1944         previous_nal_size += h->out.nal[i].i_payload;
1945
1946     for( int i = start; i < h->out.i_nal; i++ )
1947         nal_size += h->out.nal[i].i_payload;
1948
1949     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1950     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1951     for( int i = start; i < h->out.i_nal; i++ )
1952         necessary_size += h->out.nal[i].i_padding;
1953     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1954         return -1;
1955
1956     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1957
1958     for( int i = start; i < h->out.i_nal; i++ )
1959     {
1960         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1961                                          h->param.i_avcintra_class;
1962         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1963         nal_buffer += h->out.nal[i].i_payload;
1964     }
1965
1966     x264_emms();
1967
1968     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1969 }
1970
1971 /****************************************************************************
1972  * x264_encoder_headers:
1973  ****************************************************************************/
1974 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1975 {
1976     int frame_size = 0;
1977     /* init bitstream context */
1978     h->out.i_nal = 0;
1979     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1980
1981     /* Write SEI, SPS and PPS. */
1982
1983     /* generate sequence parameters */
1984     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1985     x264_sps_write( &h->out.bs, h->sps );
1986     if( x264_nal_end( h ) )
1987         return -1;
1988
1989     /* generate picture parameters */
1990     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1991     x264_pps_write( &h->out.bs, h->sps, h->pps );
1992     if( x264_nal_end( h ) )
1993         return -1;
1994
1995     /* identify ourselves */
1996     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1997     if( x264_sei_version_write( h, &h->out.bs ) )
1998         return -1;
1999     if( x264_nal_end( h ) )
2000         return -1;
2001
2002     frame_size = x264_encoder_encapsulate_nals( h, 0 );
2003     if( frame_size < 0 )
2004         return -1;
2005
2006     /* now set output*/
2007     *pi_nal = h->out.i_nal;
2008     *pp_nal = &h->out.nal[0];
2009     h->out.i_nal = 0;
2010
2011     return frame_size;
2012 }
2013
2014 /* Check to see whether we have chosen a reference list ordering different
2015  * from the standard's default. */
2016 static inline void x264_reference_check_reorder( x264_t *h )
2017 {
2018     /* The reorder check doesn't check for missing frames, so just
2019      * force a reorder if one of the reference list is corrupt. */
2020     for( int i = 0; h->frames.reference[i]; i++ )
2021         if( h->frames.reference[i]->b_corrupt )
2022         {
2023             h->b_ref_reorder[0] = 1;
2024             return;
2025         }
2026     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
2027         for( int i = 0; i < h->i_ref[list] - 1; i++ )
2028         {
2029             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
2030             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
2031             /* P and B-frames use different default orders. */
2032             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
2033             {
2034                 h->b_ref_reorder[list] = 1;
2035                 return;
2036             }
2037         }
2038 }
2039
2040 /* return -1 on failure, else return the index of the new reference frame */
2041 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2042 {
2043     int i = h->i_ref[0];
2044     int j = 1;
2045     x264_frame_t *newframe;
2046     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2047         return -1;
2048
2049     //Duplication is only used in X264_WEIGHTP_SMART
2050     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2051         return -1;
2052
2053     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2054      * With high bit depth, it's not worth doing, so turn it off except in the case of
2055      * unweighted dupes. */
2056     if( BIT_DEPTH > 8 && w != x264_weight_none )
2057         return -1;
2058
2059     newframe = x264_frame_pop_blank_unused( h );
2060     if( !newframe )
2061         return -1;
2062
2063     //FIXME: probably don't need to copy everything
2064     *newframe = *h->fref[0][i_ref];
2065     newframe->i_reference_count = 1;
2066     newframe->orig = h->fref[0][i_ref];
2067     newframe->b_duplicate = 1;
2068     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2069
2070     /* shift the frames to make space for the dupe. */
2071     h->b_ref_reorder[0] = 1;
2072     if( h->i_ref[0] < X264_REF_MAX )
2073         ++h->i_ref[0];
2074     h->fref[0][X264_REF_MAX-1] = NULL;
2075     x264_frame_unshift( &h->fref[0][j], newframe );
2076
2077     return j;
2078 }
2079
2080 static void x264_weighted_pred_init( x264_t *h )
2081 {
2082     /* for now no analysis and set all weights to nothing */
2083     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2084         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2085
2086     // FIXME: This only supports weighting of one reference frame
2087     // and duplicates of that frame.
2088     h->fenc->i_lines_weighted = 0;
2089
2090     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2091         for( int i = 0; i < 3; i++ )
2092             h->sh.weight[i_ref][i].weightfn = NULL;
2093
2094
2095     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2096         return;
2097
2098     int i_padv = PADV << PARAM_INTERLACED;
2099     int denom = -1;
2100     int weightplane[2] = { 0, 0 };
2101     int buffer_next = 0;
2102     for( int i = 0; i < 3; i++ )
2103     {
2104         for( int j = 0; j < h->i_ref[0]; j++ )
2105         {
2106             if( h->fenc->weight[j][i].weightfn )
2107             {
2108                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2109                 // if weight is useless, don't write it to stream
2110                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2111                     h->sh.weight[j][i].weightfn = NULL;
2112                 else
2113                 {
2114                     if( !weightplane[!!i] )
2115                     {
2116                         weightplane[!!i] = 1;
2117                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2118                         assert( x264_clip3( denom, 0, 7 ) == denom );
2119                     }
2120
2121                     assert( h->sh.weight[j][i].i_denom == denom );
2122                     if( !i )
2123                     {
2124                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2125                         //scale full resolution frame
2126                         if( h->param.i_threads == 1 )
2127                         {
2128                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2129                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2130                             int stride = h->fenc->i_stride[0];
2131                             int width = h->fenc->i_width[0] + PADH*2;
2132                             int height = h->fenc->i_lines[0] + i_padv*2;
2133                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2134                             h->fenc->i_lines_weighted = height;
2135                         }
2136                     }
2137                 }
2138             }
2139         }
2140     }
2141
2142     if( weightplane[1] )
2143         for( int i = 0; i < h->i_ref[0]; i++ )
2144         {
2145             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2146             {
2147                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2148                 h->sh.weight[i][2].i_offset = 0;
2149             }
2150             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2151             {
2152                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2153                 h->sh.weight[i][1].i_offset = 0;
2154             }
2155         }
2156
2157     if( !weightplane[0] )
2158         h->sh.weight[0][0].i_denom = 0;
2159     if( !weightplane[1] )
2160         h->sh.weight[0][1].i_denom = 0;
2161     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2162 }
2163
2164 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2165 {
2166     if( h->param.i_frame_packing == 5 )
2167         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2168                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2169     else
2170         return abs(h->fenc->i_frame - frame->i_frame);
2171 }
2172
2173 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2174 {
2175     int b_ok;
2176
2177     /* build ref list 0/1 */
2178     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2179     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2180     if( h->sh.i_type == SLICE_TYPE_I )
2181         return;
2182
2183     for( int i = 0; h->frames.reference[i]; i++ )
2184     {
2185         if( h->frames.reference[i]->b_corrupt )
2186             continue;
2187         if( h->frames.reference[i]->i_poc < i_poc )
2188             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2189         else if( h->frames.reference[i]->i_poc > i_poc )
2190             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2191     }
2192
2193     if( h->sh.i_mmco_remove_from_end )
2194     {
2195         /* Order ref0 for MMCO remove */
2196         do
2197         {
2198             b_ok = 1;
2199             for( int i = 0; i < h->i_ref[0] - 1; i++ )
2200             {
2201                 if( h->fref[0][i]->i_frame < h->fref[0][i+1]->i_frame )
2202                 {
2203                     XCHG( x264_frame_t*, h->fref[0][i], h->fref[0][i+1] );
2204                     b_ok = 0;
2205                     break;
2206                 }
2207             }
2208         } while( !b_ok );
2209
2210         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2211         {
2212             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2213             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2214             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2215         }
2216     }
2217
2218     /* Order reference lists by distance from the current frame. */
2219     for( int list = 0; list < 2; list++ )
2220     {
2221         h->fref_nearest[list] = h->fref[list][0];
2222         do
2223         {
2224             b_ok = 1;
2225             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2226             {
2227                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2228                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2229                     h->fref_nearest[list] = h->fref[list][i+1];
2230                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2231                 {
2232                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2233                     b_ok = 0;
2234                     break;
2235                 }
2236             }
2237         } while( !b_ok );
2238     }
2239
2240     x264_reference_check_reorder( h );
2241
2242     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2243     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2244     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2245
2246     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2247     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2248         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2249
2250     /* add duplicates */
2251     if( h->fenc->i_type == X264_TYPE_P )
2252     {
2253         int idx = -1;
2254         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2255         {
2256             x264_weight_t w[3];
2257             w[1].weightfn = w[2].weightfn = NULL;
2258             if( h->param.rc.b_stat_read )
2259                 x264_ratecontrol_set_weights( h, h->fenc );
2260
2261             if( !h->fenc->weight[0][0].weightfn )
2262             {
2263                 h->fenc->weight[0][0].i_denom = 0;
2264                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2265                 idx = x264_weighted_reference_duplicate( h, 0, w );
2266             }
2267             else
2268             {
2269                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2270                 {
2271                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2272                 }
2273                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2274                 if( h->fenc->weight[0][0].i_offset > -128 )
2275                 {
2276                     w[0] = h->fenc->weight[0][0];
2277                     w[0].i_offset--;
2278                     h->mc.weight_cache( h, &w[0] );
2279                     idx = x264_weighted_reference_duplicate( h, 0, w );
2280                 }
2281             }
2282         }
2283         h->mb.ref_blind_dupe = idx;
2284     }
2285
2286     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2287     h->mb.pic.i_fref[0] = h->i_ref[0];
2288     h->mb.pic.i_fref[1] = h->i_ref[1];
2289 }
2290
2291 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2292 {
2293     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2294     int b_hpel = h->fdec->b_kept_as_ref;
2295     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2296     int b_end = mb_y == h->i_threadslice_end;
2297     int b_measure_quality = 1;
2298     int min_y = mb_y - (1 << SLICE_MBAFF);
2299     int b_start = min_y == h->i_threadslice_start;
2300     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2301      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2302     int minpix_y = min_y*16 - 4 * !b_start;
2303     int maxpix_y = mb_y*16 - 4 * !b_end;
2304     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2305     if( h->param.b_sliced_threads )
2306     {
2307         switch( pass )
2308         {
2309             /* During encode: only do deblock if asked for */
2310             default:
2311             case 0:
2312                 b_deblock &= h->param.b_full_recon;
2313                 b_hpel = 0;
2314                 break;
2315             /* During post-encode pass: do deblock if not done yet, do hpel for all
2316              * rows except those between slices. */
2317             case 1:
2318                 b_deblock &= !h->param.b_full_recon;
2319                 b_hpel &= !(b_start && min_y > 0);
2320                 b_measure_quality = 0;
2321                 break;
2322             /* Final pass: do the rows between slices in sequence. */
2323             case 2:
2324                 b_deblock = 0;
2325                 b_measure_quality = 0;
2326                 break;
2327         }
2328     }
2329     if( mb_y & SLICE_MBAFF )
2330         return;
2331     if( min_y < h->i_threadslice_start )
2332         return;
2333
2334     if( b_deblock )
2335         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2336             x264_frame_deblock_row( h, y );
2337
2338     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2339      * but the actual image data is equivalent. For now, maintain this
2340      * consistency by copying deblocked pixels between planes. */
2341     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2342         for( int p = 0; p < h->fdec->i_plane; p++ )
2343             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2344                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2345                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2346                         h->mb.i_mb_width*16*sizeof(pixel) );
2347
2348     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2349         x264_frame_expand_border( h, h->fdec, min_y );
2350     if( b_hpel )
2351     {
2352         int end = mb_y == h->mb.i_mb_height;
2353         /* Can't do hpel until the previous slice is done encoding. */
2354         if( h->param.analyse.i_subpel_refine )
2355         {
2356             x264_frame_filter( h, h->fdec, min_y, end );
2357             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2358         }
2359     }
2360
2361     if( SLICE_MBAFF && pass == 0 )
2362         for( int i = 0; i < 3; i++ )
2363         {
2364             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2365             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2366         }
2367
2368     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2369         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2370
2371     if( b_measure_quality )
2372     {
2373         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2374         if( h->param.analyse.b_psnr )
2375         {
2376             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2377                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2378                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2379                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2380                     h->param.i_width, maxpix_y-minpix_y );
2381             if( !CHROMA444 )
2382             {
2383                 uint64_t ssd_u, ssd_v;
2384                 int v_shift = CHROMA_V_SHIFT;
2385                 x264_pixel_ssd_nv12( &h->pixf,
2386                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2387                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2388                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2389                 h->stat.frame.i_ssd[1] += ssd_u;
2390                 h->stat.frame.i_ssd[2] += ssd_v;
2391             }
2392         }
2393
2394         if( h->param.analyse.b_ssim )
2395         {
2396             int ssim_cnt;
2397             x264_emms();
2398             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2399              * and overlap by 4 */
2400             minpix_y += b_start ? 2 : -6;
2401             h->stat.frame.f_ssim +=
2402                 x264_pixel_ssim_wxh( &h->pixf,
2403                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2404                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2405                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2406             h->stat.frame.i_ssim_cnt += ssim_cnt;
2407         }
2408     }
2409 }
2410
2411 static inline int x264_reference_update( x264_t *h )
2412 {
2413     if( !h->fdec->b_kept_as_ref )
2414     {
2415         if( h->i_thread_frames > 1 )
2416         {
2417             x264_frame_push_unused( h, h->fdec );
2418             h->fdec = x264_frame_pop_unused( h, 1 );
2419             if( !h->fdec )
2420                 return -1;
2421         }
2422         return 0;
2423     }
2424
2425     /* apply mmco from previous frame. */
2426     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2427         for( int j = 0; h->frames.reference[j]; j++ )
2428             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2429                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2430
2431     /* move frame in the buffer */
2432     x264_frame_push( h->frames.reference, h->fdec );
2433     if( h->frames.reference[h->sps->i_num_ref_frames] )
2434         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2435     h->fdec = x264_frame_pop_unused( h, 1 );
2436     if( !h->fdec )
2437         return -1;
2438     return 0;
2439 }
2440
2441 static inline void x264_reference_reset( x264_t *h )
2442 {
2443     while( h->frames.reference[0] )
2444         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2445     h->fdec->i_poc =
2446     h->fenc->i_poc = 0;
2447 }
2448
2449 static inline void x264_reference_hierarchy_reset( x264_t *h )
2450 {
2451     int ref;
2452     int b_hasdelayframe = 0;
2453
2454     /* look for delay frames -- chain must only contain frames that are disposable */
2455     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2456         b_hasdelayframe |= h->frames.current[i]->i_coded
2457                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2458
2459     /* This function must handle b-pyramid and clear frames for open-gop */
2460     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2461         return;
2462
2463     /* Remove last BREF. There will never be old BREFs in the
2464      * dpb during a BREF decode when pyramid == STRICT */
2465     for( ref = 0; h->frames.reference[ref]; ref++ )
2466     {
2467         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2468             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2469             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2470             && h->sh.i_type != SLICE_TYPE_B ) )
2471         {
2472             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2473             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2474             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2475             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2476             h->b_ref_reorder[0] = 1;
2477             ref--;
2478         }
2479     }
2480
2481     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2482     if( h->param.i_bframe_pyramid )
2483         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2484 }
2485
2486 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2487 {
2488     /* ------------------------ Create slice header  ----------------------- */
2489     if( i_nal_type == NAL_SLICE_IDR )
2490     {
2491         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2492
2493         /* alternate id */
2494         if( h->param.i_avcintra_class )
2495         {
2496             switch( h->i_idr_pic_id )
2497             {
2498                 case 5:
2499                     h->i_idr_pic_id = 3;
2500                     break;
2501                 case 3:
2502                     h->i_idr_pic_id = 4;
2503                     break;
2504                 case 4:
2505                 default:
2506                     h->i_idr_pic_id = 5;
2507                     break;
2508             }
2509         }
2510         else
2511             h->i_idr_pic_id ^= 1;
2512     }
2513     else
2514     {
2515         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2516
2517         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2518         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2519         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2520             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2521         {
2522             h->sh.b_num_ref_idx_override = 1;
2523         }
2524     }
2525
2526     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2527     {
2528         h->b_sh_backup = 1;
2529         h->sh_backup = h->sh;
2530     }
2531
2532     h->fdec->i_frame_num = h->sh.i_frame_num;
2533
2534     if( h->sps->i_poc_type == 0 )
2535     {
2536         h->sh.i_poc = h->fdec->i_poc;
2537         if( PARAM_INTERLACED )
2538         {
2539             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2540             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2541         }
2542         else
2543             h->sh.i_delta_poc_bottom = 0;
2544         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2545         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2546     }
2547     else
2548     {
2549         /* Nothing to do ? */
2550     }
2551
2552     x264_macroblock_slice_init( h );
2553 }
2554
2555 typedef struct
2556 {
2557     int skip;
2558     uint8_t cabac_prevbyte;
2559     bs_t bs;
2560     x264_cabac_t cabac;
2561     x264_frame_stat_t stat;
2562     int last_qp;
2563     int last_dqp;
2564     int field_decoding_flag;
2565 } x264_bs_bak_t;
2566
2567 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2568 {
2569     if( full )
2570     {
2571         bak->stat = h->stat.frame;
2572         bak->last_qp = h->mb.i_last_qp;
2573         bak->last_dqp = h->mb.i_last_dqp;
2574         bak->field_decoding_flag = h->mb.field_decoding_flag;
2575     }
2576     else
2577     {
2578         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2579         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2580     }
2581     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2582      * encoder has no context dependency and in this case, a slice is ended (and
2583      * thus the content of all contexts are thrown away). */
2584     if( h->param.b_cabac )
2585     {
2586         if( full )
2587             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2588         else
2589             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2590         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2591          * backed up. */
2592         bak->cabac_prevbyte = h->cabac.p[-1];
2593     }
2594     else
2595     {
2596         bak->bs = h->out.bs;
2597         bak->skip = i_skip;
2598     }
2599 }
2600
2601 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2602 {
2603     if( full )
2604     {
2605         h->stat.frame = bak->stat;
2606         h->mb.i_last_qp = bak->last_qp;
2607         h->mb.i_last_dqp = bak->last_dqp;
2608         h->mb.field_decoding_flag = bak->field_decoding_flag;
2609     }
2610     else
2611     {
2612         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2613         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2614     }
2615     if( h->param.b_cabac )
2616     {
2617         if( full )
2618             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2619         else
2620             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2621         h->cabac.p[-1] = bak->cabac_prevbyte;
2622     }
2623     else
2624     {
2625         h->out.bs = bak->bs;
2626         *skip = bak->skip;
2627     }
2628 }
2629
2630 static intptr_t x264_slice_write( x264_t *h )
2631 {
2632     int i_skip;
2633     int mb_xy, i_mb_x, i_mb_y;
2634     /* NALUs other than the first use a 3-byte startcode.
2635      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2636      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2637      * other inaccuracies. */
2638     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2639     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2640     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2641     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2642     int starting_bits = bs_pos(&h->out.bs);
2643     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2644     int b_hpel = h->fdec->b_kept_as_ref;
2645     int orig_last_mb = h->sh.i_last_mb;
2646     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2647     uint8_t *last_emu_check;
2648 #define BS_BAK_SLICE_MAX_SIZE 0
2649 #define BS_BAK_CAVLC_OVERFLOW 1
2650 #define BS_BAK_SLICE_MIN_MBS  2
2651 #define BS_BAK_ROW_VBV        3
2652     x264_bs_bak_t bs_bak[4];
2653     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2654     bs_realign( &h->out.bs );
2655
2656     /* Slice */
2657     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2658     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2659
2660     /* Slice header */
2661     x264_macroblock_thread_init( h );
2662
2663     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2664     h->mb.i_mb_xy = h->sh.i_first_mb;
2665     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2666     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2667     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2668
2669     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2670     if( h->param.b_cabac )
2671     {
2672         /* alignment needed */
2673         bs_align_1( &h->out.bs );
2674
2675         /* init cabac */
2676         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 );
2677         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2678         last_emu_check = h->cabac.p;
2679     }
2680     else
2681         last_emu_check = h->out.bs.p;
2682     h->mb.i_last_qp = h->sh.i_qp;
2683     h->mb.i_last_dqp = 0;
2684     h->mb.field_decoding_flag = 0;
2685
2686     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2687     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2688     i_skip = 0;
2689
2690     while( 1 )
2691     {
2692         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2693         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2694
2695         if( i_mb_x == 0 )
2696         {
2697             if( x264_bitstream_check_buffer( h ) )
2698                 return -1;
2699             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2700                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2701             if( !h->mb.b_reencode_mb )
2702                 x264_fdec_filter_row( h, i_mb_y, 0 );
2703         }
2704
2705         if( back_up_bitstream )
2706         {
2707             if( back_up_bitstream_cavlc )
2708                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2709             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2710             {
2711                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2712                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2713                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2714             }
2715         }
2716
2717         if( PARAM_INTERLACED )
2718         {
2719             if( h->mb.b_adaptive_mbaff )
2720             {
2721                 if( !(i_mb_y&1) )
2722                 {
2723                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2724                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2725                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2726                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2727                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2728                 }
2729             }
2730             h->mb.field[mb_xy] = MB_INTERLACED;
2731         }
2732
2733         /* load cache */
2734         if( SLICE_MBAFF )
2735             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2736         else
2737             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2738
2739         x264_macroblock_analyse( h );
2740
2741         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2742 reencode:
2743         x264_macroblock_encode( h );
2744
2745         if( h->param.b_cabac )
2746         {
2747             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2748                 x264_cabac_encode_terminal( &h->cabac );
2749
2750             if( IS_SKIP( h->mb.i_type ) )
2751                 x264_cabac_mb_skip( h, 1 );
2752             else
2753             {
2754                 if( h->sh.i_type != SLICE_TYPE_I )
2755                     x264_cabac_mb_skip( h, 0 );
2756                 x264_macroblock_write_cabac( h, &h->cabac );
2757             }
2758         }
2759         else
2760         {
2761             if( IS_SKIP( h->mb.i_type ) )
2762                 i_skip++;
2763             else
2764             {
2765                 if( h->sh.i_type != SLICE_TYPE_I )
2766                 {
2767                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2768                     i_skip = 0;
2769                 }
2770                 x264_macroblock_write_cavlc( h );
2771                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2772                 if( h->mb.b_overflow )
2773                 {
2774                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2775                     h->mb.i_skip_intra = 0;
2776                     h->mb.b_skip_mc = 0;
2777                     h->mb.b_overflow = 0;
2778                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2779                     goto reencode;
2780                 }
2781             }
2782         }
2783
2784         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2785         int mb_size = total_bits - mb_spos;
2786
2787         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2788         {
2789             /* Count the skip run, just in case. */
2790             if( !h->param.b_cabac )
2791                 total_bits += bs_size_ue_big( i_skip );
2792             /* Check for escape bytes. */
2793             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2794             for( ; last_emu_check < end - 2; last_emu_check++ )
2795                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2796                 {
2797                     slice_max_size -= 8;
2798                     last_emu_check++;
2799                 }
2800             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2801             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2802             {
2803                 if( !x264_frame_new_slice( h, h->fdec ) )
2804                 {
2805                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2806                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2807                      * If possible, roll back to the last checkpoint and try again.
2808                      * We could try raising QP, but that would break in the case where a slice spans multiple
2809                      * rows, which the re-encoding infrastructure can't currently handle. */
2810                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2811                     {
2812                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2813                         {
2814                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2815                             slice_max_size = 0;
2816                             goto cont;
2817                         }
2818                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2819                         h->mb.b_reencode_mb = 1;
2820                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2821                         break;
2822                     }
2823                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2824                     {
2825                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2826                         h->mb.b_reencode_mb = 1;
2827                         if( SLICE_MBAFF )
2828                         {
2829                             // set to bottom of previous mbpair
2830                             if( i_mb_x )
2831                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2832                             else
2833                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2834                         }
2835                         else
2836                             h->sh.i_last_mb = mb_xy-1;
2837                         break;
2838                     }
2839                     else
2840                         h->sh.i_last_mb = mb_xy;
2841                 }
2842                 else
2843                     slice_max_size = 0;
2844             }
2845         }
2846 cont:
2847         h->mb.b_reencode_mb = 0;
2848
2849         /* save cache */
2850         x264_macroblock_cache_save( h );
2851
2852         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2853         {
2854             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2855             h->mb.b_reencode_mb = 1;
2856             i_mb_x = 0;
2857             i_mb_y = i_mb_y - SLICE_MBAFF;
2858             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2859             h->sh.i_last_mb = orig_last_mb;
2860             continue;
2861         }
2862
2863         /* accumulate mb stats */
2864         h->stat.frame.i_mb_count[h->mb.i_type]++;
2865
2866         int b_intra = IS_INTRA( h->mb.i_type );
2867         int b_skip = IS_SKIP( h->mb.i_type );
2868         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2869         {
2870             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2871             {
2872                 if( h->mb.i_partition != D_8x8 )
2873                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2874                     else
2875                         for( int i = 0; i < 4; i++ )
2876                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2877                 if( h->param.i_frame_reference > 1 )
2878                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2879                         for( int i = 0; i < 4; i++ )
2880                         {
2881                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2882                             if( i_ref >= 0 )
2883                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2884                         }
2885             }
2886         }
2887
2888         if( h->param.i_log_level >= X264_LOG_INFO )
2889         {
2890             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2891             {
2892                 if( CHROMA444 )
2893                 {
2894                     for( int i = 0; i < 4; i++ )
2895                         if( h->mb.i_cbp_luma & (1 << i) )
2896                             for( int p = 0; p < 3; p++ )
2897                             {
2898                                 int s8 = i*4+p*16;
2899                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2900                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2901                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2902                             }
2903                 }
2904                 else
2905                 {
2906                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2907                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2908                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2909                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2910                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2911                 }
2912             }
2913             if( h->mb.i_cbp_luma && !b_intra )
2914             {
2915                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2916                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2917             }
2918             if( b_intra && h->mb.i_type != I_PCM )
2919             {
2920                 if( h->mb.i_type == I_16x16 )
2921                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2922                 else if( h->mb.i_type == I_8x8 )
2923                     for( int i = 0; i < 16; i += 4 )
2924                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2925                 else //if( h->mb.i_type == I_4x4 )
2926                     for( int i = 0; i < 16; i++ )
2927                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2928                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2929             }
2930             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2931         }
2932
2933         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2934         if( b_deblock )
2935             x264_macroblock_deblock_strength( h );
2936
2937         if( mb_xy == h->sh.i_last_mb )
2938             break;
2939
2940         if( SLICE_MBAFF )
2941         {
2942             i_mb_x += i_mb_y & 1;
2943             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2944         }
2945         else
2946             i_mb_x++;
2947         if( i_mb_x == h->mb.i_mb_width )
2948         {
2949             i_mb_y++;
2950             i_mb_x = 0;
2951         }
2952     }
2953     if( h->sh.i_last_mb < h->sh.i_first_mb )
2954         return 0;
2955
2956     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2957
2958     if( h->param.b_cabac )
2959     {
2960         x264_cabac_encode_flush( h, &h->cabac );
2961         h->out.bs.p = h->cabac.p;
2962     }
2963     else
2964     {
2965         if( i_skip > 0 )
2966             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2967         /* rbsp_slice_trailing_bits */
2968         bs_rbsp_trailing( &h->out.bs );
2969         bs_flush( &h->out.bs );
2970     }
2971     if( x264_nal_end( h ) )
2972         return -1;
2973
2974     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2975     {
2976         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2977                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2978                                   - h->stat.frame.i_tex_bits
2979                                   - h->stat.frame.i_mv_bits;
2980         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2981
2982         if( h->param.b_sliced_threads )
2983         {
2984             /* Tell the main thread we're done. */
2985             x264_threadslice_cond_broadcast( h, 1 );
2986             /* Do hpel now */
2987             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2988                 x264_fdec_filter_row( h, mb_y, 1 );
2989             x264_threadslice_cond_broadcast( h, 2 );
2990             /* Do the first row of hpel, now that the previous slice is done */
2991             if( h->i_thread_idx > 0 )
2992             {
2993                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2994                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2995             }
2996         }
2997
2998         /* Free mb info after the last thread's done using it */
2999         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
3000         {
3001             h->fdec->mb_info_free( h->fdec->mb_info );
3002             h->fdec->mb_info = NULL;
3003             h->fdec->mb_info_free = NULL;
3004         }
3005     }
3006
3007     return 0;
3008 }
3009
3010 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
3011 {
3012     if( dst == src )
3013         return;
3014
3015     // reference counting
3016     for( x264_frame_t **f = src->frames.reference; *f; f++ )
3017         (*f)->i_reference_count++;
3018     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
3019         x264_frame_push_unused( src, *f );
3020     src->fdec->i_reference_count++;
3021     x264_frame_push_unused( src, dst->fdec );
3022
3023     // copy everything except the per-thread pointers and the constants.
3024     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
3025     dst->param = src->param;
3026     dst->stat = src->stat;
3027     dst->pixf = src->pixf;
3028     dst->reconfig = src->reconfig;
3029 }
3030
3031 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
3032 {
3033     if( dst != src )
3034         memcpy( &dst->stat, &src->stat, offsetof(x264_t, stat.frame) - offsetof(x264_t, stat) );
3035 }
3036
3037 static void *x264_slices_write( x264_t *h )
3038 {
3039     int i_slice_num = 0;
3040     int last_thread_mb = h->sh.i_last_mb;
3041
3042     /* init stats */
3043     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
3044     h->mb.b_reencode_mb = 0;
3045     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
3046     {
3047         h->sh.i_last_mb = last_thread_mb;
3048         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
3049         {
3050             if( h->param.i_slice_max_mbs )
3051             {
3052                 if( SLICE_MBAFF )
3053                 {
3054                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
3055                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
3056                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
3057                         + h->param.i_slice_max_mbs - 1;
3058                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
3059                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
3060                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
3061                 }
3062                 else
3063                 {
3064                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
3065                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
3066                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
3067                 }
3068                 i_slice_num++;
3069             }
3070             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3071             {
3072                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3073                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3074                 i_slice_num++;
3075                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3076             }
3077         }
3078         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3079         if( x264_stack_align( x264_slice_write, h ) )
3080             goto fail;
3081         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3082         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3083         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3084             h->sh.i_first_mb -= h->mb.i_mb_stride;
3085     }
3086
3087     return (void *)0;
3088
3089 fail:
3090     /* Tell other threads we're done, so they wouldn't wait for it */
3091     if( h->param.b_sliced_threads )
3092         x264_threadslice_cond_broadcast( h, 2 );
3093     return (void *)-1;
3094 }
3095
3096 static int x264_threaded_slices_write( x264_t *h )
3097 {
3098     /* set first/last mb and sync contexts */
3099     for( int i = 0; i < h->param.i_threads; i++ )
3100     {
3101         x264_t *t = h->thread[i];
3102         if( i )
3103         {
3104             t->param = h->param;
3105             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3106         }
3107         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3108         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3109         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3110         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3111         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3112     }
3113
3114     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3115
3116     x264_threads_distribute_ratecontrol( h );
3117
3118     /* setup */
3119     for( int i = 0; i < h->param.i_threads; i++ )
3120     {
3121         h->thread[i]->i_thread_idx = i;
3122         h->thread[i]->b_thread_active = 1;
3123         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3124     }
3125     /* dispatch */
3126     for( int i = 0; i < h->param.i_threads; i++ )
3127         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3128     /* wait */
3129     for( int i = 0; i < h->param.i_threads; i++ )
3130         x264_threadslice_cond_wait( h->thread[i], 1 );
3131
3132     x264_threads_merge_ratecontrol( h );
3133
3134     for( int i = 1; i < h->param.i_threads; i++ )
3135     {
3136         x264_t *t = h->thread[i];
3137         for( int j = 0; j < t->out.i_nal; j++ )
3138         {
3139             h->out.nal[h->out.i_nal] = t->out.nal[j];
3140             h->out.i_nal++;
3141             x264_nal_check_buffer( h );
3142         }
3143         /* All entries in stat.frame are ints except for ssd/ssim. */
3144         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3145             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3146         for( int j = 0; j < 3; j++ )
3147             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3148         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3149         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3150     }
3151
3152     return 0;
3153 }
3154
3155 void x264_encoder_intra_refresh( x264_t *h )
3156 {
3157     h = h->thread[h->i_thread_phase];
3158     h->b_queued_intra_refresh = 1;
3159 }
3160
3161 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3162 {
3163     if( h->param.i_bframe )
3164     {
3165         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3166         return -1;
3167     }
3168     if( h->param.b_intra_refresh )
3169     {
3170         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3171         return -1;
3172     }
3173     h = h->thread[h->i_thread_phase];
3174     if( pts >= h->i_last_idr_pts )
3175     {
3176         for( int i = 0; h->frames.reference[i]; i++ )
3177             if( pts <= h->frames.reference[i]->i_pts )
3178                 h->frames.reference[i]->b_corrupt = 1;
3179         if( pts <= h->fdec->i_pts )
3180             h->fdec->b_corrupt = 1;
3181     }
3182     return 0;
3183 }
3184
3185 /****************************************************************************
3186  * x264_encoder_encode:
3187  *  XXX: i_poc   : is the poc of the current given picture
3188  *       i_frame : is the number of the frame being coded
3189  *  ex:  type frame poc
3190  *       I      0   2*0
3191  *       P      1   2*3
3192  *       B      2   2*1
3193  *       B      3   2*2
3194  *       P      4   2*6
3195  *       B      5   2*4
3196  *       B      6   2*5
3197  ****************************************************************************/
3198 int     x264_encoder_encode( x264_t *h,
3199                              x264_nal_t **pp_nal, int *pi_nal,
3200                              x264_picture_t *pic_in,
3201                              x264_picture_t *pic_out )
3202 {
3203     x264_t *thread_current, *thread_prev, *thread_oldest;
3204     int i_nal_type, i_nal_ref_idc, i_global_qp;
3205     int overhead = NALU_OVERHEAD;
3206
3207 #if HAVE_OPENCL
3208     if( h->opencl.b_fatal_error )
3209         return -1;
3210 #endif
3211
3212     if( h->i_thread_frames > 1 )
3213     {
3214         thread_prev    = h->thread[ h->i_thread_phase ];
3215         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3216         thread_current = h->thread[ h->i_thread_phase ];
3217         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3218         x264_thread_sync_context( thread_current, thread_prev );
3219         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3220         h = thread_current;
3221     }
3222     else
3223     {
3224         thread_current =
3225         thread_oldest  = h;
3226     }
3227     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3228
3229     /* no data out */
3230     *pi_nal = 0;
3231     *pp_nal = NULL;
3232
3233     /* ------------------- Setup new frame from picture -------------------- */
3234     if( pic_in != NULL )
3235     {
3236         if( h->lookahead->b_exit_thread )
3237         {
3238             x264_log( h, X264_LOG_ERROR, "lookahead thread is already stopped\n" );
3239             return -1;
3240         }
3241
3242         /* 1: Copy the picture to a frame and move it to a buffer */
3243         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3244         if( !fenc )
3245             return -1;
3246
3247         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3248             return -1;
3249
3250         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3251             h->param.i_height != 16 * h->mb.i_mb_height )
3252             x264_frame_expand_border_mod16( h, fenc );
3253
3254         fenc->i_frame = h->frames.i_input++;
3255
3256         if( fenc->i_frame == 0 )
3257             h->frames.i_first_pts = fenc->i_pts;
3258         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3259             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3260
3261         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3262             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3263
3264         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3265         h->frames.i_largest_pts = fenc->i_pts;
3266
3267         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3268             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3269
3270         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3271         {
3272 #if HAVE_INTERLACED
3273             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3274 #else
3275             int b_interlaced = 0;
3276 #endif
3277             if( b_interlaced )
3278             {
3279                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3280                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3281             }
3282             else
3283                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3284         }
3285
3286         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3287         {
3288             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3289                 return -1;
3290         }
3291         else
3292             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3293
3294         if( pic_in->prop.quant_offsets_free )
3295             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3296
3297         if( h->frames.b_have_lowres )
3298             x264_frame_init_lowres( h, fenc );
3299
3300         /* 2: Place the frame into the queue for its slice type decision */
3301         x264_lookahead_put_frame( h, fenc );
3302
3303         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3304         {
3305             /* Nothing yet to encode, waiting for filling of buffers */
3306             pic_out->i_type = X264_TYPE_AUTO;
3307             return 0;
3308         }
3309     }
3310     else
3311     {
3312         /* signal kills for lookahead thread */
3313         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3314         h->lookahead->b_exit_thread = 1;
3315         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3316         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3317     }
3318
3319     h->i_frame++;
3320     /* 3: The picture is analyzed in the lookahead */
3321     if( !h->frames.current[0] )
3322         x264_lookahead_get_frames( h );
3323
3324     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3325         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3326
3327     /* ------------------- Get frame to be encoded ------------------------- */
3328     /* 4: get picture to encode */
3329     h->fenc = x264_frame_shift( h->frames.current );
3330
3331     /* If applicable, wait for previous frame reconstruction to finish */
3332     if( h->param.b_sliced_threads )
3333         if( x264_threadpool_wait_all( h ) < 0 )
3334             return -1;
3335
3336     if( h->i_frame == 0 )
3337         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3338     if( h->reconfig )
3339     {
3340         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3341         h->reconfig = 0;
3342     }
3343     if( h->fenc->param )
3344     {
3345         x264_encoder_reconfig_apply( h, h->fenc->param );
3346         if( h->fenc->param->param_free )
3347         {
3348             h->fenc->param->param_free( h->fenc->param );
3349             h->fenc->param = NULL;
3350         }
3351     }
3352
3353     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3354     if( x264_reference_update( h ) )
3355         return -1;
3356     h->fdec->i_lines_completed = -1;
3357
3358     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3359     {
3360         int valid_refs_left = 0;
3361         for( int i = 0; h->frames.reference[i]; i++ )
3362             if( !h->frames.reference[i]->b_corrupt )
3363                 valid_refs_left++;
3364         /* No valid reference frames left: force an IDR. */
3365         if( !valid_refs_left )
3366         {
3367             h->fenc->b_keyframe = 1;
3368             h->fenc->i_type = X264_TYPE_IDR;
3369         }
3370     }
3371
3372     if( h->fenc->b_keyframe )
3373     {
3374         h->frames.i_last_keyframe = h->fenc->i_frame;
3375         if( h->fenc->i_type == X264_TYPE_IDR )
3376         {
3377             h->i_frame_num = 0;
3378             h->frames.i_last_idr = h->fenc->i_frame;
3379         }
3380     }
3381     h->sh.i_mmco_command_count =
3382     h->sh.i_mmco_remove_from_end = 0;
3383     h->b_ref_reorder[0] =
3384     h->b_ref_reorder[1] = 0;
3385     h->fdec->i_poc =
3386     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3387
3388     /* ------------------- Setup frame context ----------------------------- */
3389     /* 5: Init data dependent of frame type */
3390     if( h->fenc->i_type == X264_TYPE_IDR )
3391     {
3392         /* reset ref pictures */
3393         i_nal_type    = NAL_SLICE_IDR;
3394         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3395         h->sh.i_type = SLICE_TYPE_I;
3396         x264_reference_reset( h );
3397         h->frames.i_poc_last_open_gop = -1;
3398     }
3399     else if( h->fenc->i_type == X264_TYPE_I )
3400     {
3401         i_nal_type    = NAL_SLICE;
3402         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3403         h->sh.i_type = SLICE_TYPE_I;
3404         x264_reference_hierarchy_reset( h );
3405         if( h->param.b_open_gop )
3406             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3407     }
3408     else if( h->fenc->i_type == X264_TYPE_P )
3409     {
3410         i_nal_type    = NAL_SLICE;
3411         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3412         h->sh.i_type = SLICE_TYPE_P;
3413         x264_reference_hierarchy_reset( h );
3414         h->frames.i_poc_last_open_gop = -1;
3415     }
3416     else if( h->fenc->i_type == X264_TYPE_BREF )
3417     {
3418         i_nal_type    = NAL_SLICE;
3419         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3420         h->sh.i_type = SLICE_TYPE_B;
3421         x264_reference_hierarchy_reset( h );
3422     }
3423     else    /* B frame */
3424     {
3425         i_nal_type    = NAL_SLICE;
3426         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3427         h->sh.i_type = SLICE_TYPE_B;
3428     }
3429
3430     h->fdec->i_type = h->fenc->i_type;
3431     h->fdec->i_frame = h->fenc->i_frame;
3432     h->fenc->b_kept_as_ref =
3433     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3434
3435     h->fdec->mb_info = h->fenc->mb_info;
3436     h->fdec->mb_info_free = h->fenc->mb_info_free;
3437     h->fenc->mb_info = NULL;
3438     h->fenc->mb_info_free = NULL;
3439
3440     h->fdec->i_pts = h->fenc->i_pts;
3441     if( h->frames.i_bframe_delay )
3442     {
3443         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3444         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3445                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3446                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3447         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3448     }
3449     else
3450         h->fdec->i_dts = h->fenc->i_reordered_pts;
3451     if( h->fenc->i_type == X264_TYPE_IDR )
3452         h->i_last_idr_pts = h->fdec->i_pts;
3453
3454     /* ------------------- Init                ----------------------------- */
3455     /* build ref list 0/1 */
3456     x264_reference_build_list( h, h->fdec->i_poc );
3457
3458     /* ---------------------- Write the bitstream -------------------------- */
3459     /* Init bitstream context */
3460     if( h->param.b_sliced_threads )
3461     {
3462         for( int i = 0; i < h->param.i_threads; i++ )
3463         {
3464             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3465             h->thread[i]->out.i_nal = 0;
3466         }
3467     }
3468     else
3469     {
3470         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3471         h->out.i_nal = 0;
3472     }
3473
3474     if( h->param.b_aud )
3475     {
3476         int pic_type;
3477
3478         if( h->sh.i_type == SLICE_TYPE_I )
3479             pic_type = 0;
3480         else if( h->sh.i_type == SLICE_TYPE_P )
3481             pic_type = 1;
3482         else if( h->sh.i_type == SLICE_TYPE_B )
3483             pic_type = 2;
3484         else
3485             pic_type = 7;
3486
3487         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3488         bs_write( &h->out.bs, 3, pic_type );
3489         bs_rbsp_trailing( &h->out.bs );
3490         if( x264_nal_end( h ) )
3491             return -1;
3492         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3493     }
3494
3495     h->i_nal_type = i_nal_type;
3496     h->i_nal_ref_idc = i_nal_ref_idc;
3497
3498     if( h->param.b_intra_refresh )
3499     {
3500         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3501         {
3502             h->fdec->i_frames_since_pir = 0;
3503             h->b_queued_intra_refresh = 0;
3504             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3505              * the whole frame and counts as an intra refresh. */
3506             h->fdec->f_pir_position = h->mb.i_mb_width;
3507         }
3508         else if( h->fenc->i_type == X264_TYPE_P )
3509         {
3510             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3511             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3512             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3513             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3514             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3515                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3516             {
3517                 h->fdec->f_pir_position = 0;
3518                 h->fdec->i_frames_since_pir = 0;
3519                 h->b_queued_intra_refresh = 0;
3520                 h->fenc->b_keyframe = 1;
3521             }
3522             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3523             h->fdec->f_pir_position += increment * pocdiff;
3524             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3525             /* If our intra refresh has reached the right side of the frame, we're done. */
3526             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3527             {
3528                 h->fdec->f_pir_position = h->mb.i_mb_width;
3529                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3530             }
3531         }
3532     }
3533
3534     if( h->fenc->b_keyframe )
3535     {
3536         /* Write SPS and PPS */
3537         if( h->param.b_repeat_headers )
3538         {
3539             /* generate sequence parameters */
3540             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3541             x264_sps_write( &h->out.bs, h->sps );
3542             if( x264_nal_end( h ) )
3543                 return -1;
3544             /* Pad AUD/SPS to 256 bytes like Panasonic */
3545             if( h->param.i_avcintra_class )
3546                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3547             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3548
3549             /* generate picture parameters */
3550             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3551             x264_pps_write( &h->out.bs, h->sps, h->pps );
3552             if( x264_nal_end( h ) )
3553                 return -1;
3554             if( h->param.i_avcintra_class )
3555                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3556             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3557         }
3558
3559         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3560         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3561         {
3562             x264_hrd_fullness( h );
3563             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3564             x264_sei_buffering_period_write( h, &h->out.bs );
3565             if( x264_nal_end( h ) )
3566                return -1;
3567             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3568         }
3569     }
3570
3571     /* write extra sei */
3572     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3573     {
3574         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3575         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3576                         h->fenc->extra_sei.payloads[i].payload_type );
3577         if( x264_nal_end( h ) )
3578             return -1;
3579         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3580         if( h->fenc->extra_sei.sei_free )
3581         {
3582             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3583             h->fenc->extra_sei.payloads[i].payload = NULL;
3584         }
3585     }
3586
3587     if( h->fenc->extra_sei.sei_free )
3588     {
3589         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3590         h->fenc->extra_sei.payloads = NULL;
3591         h->fenc->extra_sei.sei_free = NULL;
3592     }
3593
3594     if( h->fenc->b_keyframe )
3595     {
3596         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3597         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3598         {
3599             /* identify ourself */
3600             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3601             if( x264_sei_version_write( h, &h->out.bs ) )
3602                 return -1;
3603             if( x264_nal_end( h ) )
3604                 return -1;
3605             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3606         }
3607
3608         if( h->fenc->i_type != X264_TYPE_IDR )
3609         {
3610             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;
3611             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3612             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3613             if( x264_nal_end( h ) )
3614                 return -1;
3615             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3616         }
3617     }
3618
3619     if( h->param.i_frame_packing >= 0 && (h->fenc->b_keyframe || h->param.i_frame_packing == 5) )
3620     {
3621         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3622         x264_sei_frame_packing_write( h, &h->out.bs );
3623         if( x264_nal_end( h ) )
3624             return -1;
3625         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3626     }
3627
3628     /* generate sei pic timing */
3629     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3630     {
3631         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3632         x264_sei_pic_timing_write( h, &h->out.bs );
3633         if( x264_nal_end( h ) )
3634             return -1;
3635         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3636     }
3637
3638     /* As required by Blu-ray. */
3639     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3640     {
3641         h->b_sh_backup = 0;
3642         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3643         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3644         if( x264_nal_end( h ) )
3645             return -1;
3646         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3647     }
3648
3649     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3650         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3651
3652     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3653     if( h->param.i_avcintra_class )
3654     {
3655         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3656         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3657         x264_filler_write( h, &h->out.bs, 0 );
3658         if( x264_nal_end( h ) )
3659             return -1;
3660         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3661
3662         /* All lengths are magic lengths that decoders expect to see */
3663         /* "UMID" SEI */
3664         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3665         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3666             return -1;
3667         if( x264_nal_end( h ) )
3668             return -1;
3669         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3670
3671         int unpadded_len;
3672         int total_len;
3673         if( h->param.i_height == 1080 )
3674         {
3675             unpadded_len = 5780;
3676             total_len = 17*512;
3677         }
3678         else
3679         {
3680             unpadded_len = 2900;
3681             total_len = 9*512;
3682         }
3683         /* "VANC" SEI */
3684         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3685         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3686             return -1;
3687         if( x264_nal_end( h ) )
3688             return -1;
3689
3690         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3691         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3692     }
3693
3694     /* Init the speed control */
3695     if( h->param.sc.i_buffer_size )
3696         x264_speedcontrol_frame( h );
3697
3698     /* Init the rate control */
3699     /* FIXME: Include slice header bit cost. */
3700     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3701     i_global_qp = x264_ratecontrol_qp( h );
3702
3703     pic_out->i_qpplus1 =
3704     h->fdec->i_qpplus1 = i_global_qp + 1;
3705
3706     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3707     {
3708         x264_reference_build_list_optimal( h );
3709         x264_reference_check_reorder( h );
3710     }
3711
3712     if( h->i_ref[0] )
3713         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3714
3715     /* ------------------------ Create slice header  ----------------------- */
3716     x264_slice_init( h, i_nal_type, i_global_qp );
3717
3718     /*------------------------- Weights -------------------------------------*/
3719     if( h->sh.i_type == SLICE_TYPE_B )
3720         x264_macroblock_bipred_init( h );
3721
3722     x264_weighted_pred_init( h );
3723
3724     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3725         h->i_frame_num++;
3726
3727     /* Write frame */
3728     h->i_threadslice_start = 0;
3729     h->i_threadslice_end = h->mb.i_mb_height;
3730     if( h->i_thread_frames > 1 )
3731     {
3732         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3733         h->b_thread_active = 1;
3734     }
3735     else if( h->param.b_sliced_threads )
3736     {
3737         if( x264_threaded_slices_write( h ) )
3738             return -1;
3739     }
3740     else
3741         if( (intptr_t)x264_slices_write( h ) )
3742             return -1;
3743
3744     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3745 }
3746
3747 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3748                                    x264_nal_t **pp_nal, int *pi_nal,
3749                                    x264_picture_t *pic_out )
3750 {
3751     char psz_message[80];
3752
3753     if( !h->param.b_sliced_threads && h->b_thread_active )
3754     {
3755         h->b_thread_active = 0;
3756         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3757             return -1;
3758     }
3759     if( !h->out.i_nal )
3760     {
3761         pic_out->i_type = X264_TYPE_AUTO;
3762         return 0;
3763     }
3764
3765     x264_emms();
3766
3767     /* generate buffering period sei and insert it into place */
3768     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3769     {
3770         x264_hrd_fullness( h );
3771         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3772         x264_sei_buffering_period_write( h, &h->out.bs );
3773         if( x264_nal_end( h ) )
3774            return -1;
3775         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3776         int idx = 0;
3777         while( h->out.nal[idx].i_type == NAL_AUD ||
3778                h->out.nal[idx].i_type == NAL_SPS ||
3779                h->out.nal[idx].i_type == NAL_PPS )
3780             idx++;
3781         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3782         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3783         h->out.nal[idx] = nal_tmp;
3784     }
3785
3786     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3787     if( frame_size < 0 )
3788         return -1;
3789
3790     /* Set output picture properties */
3791     pic_out->i_type = h->fenc->i_type;
3792
3793     pic_out->b_keyframe = h->fenc->b_keyframe;
3794     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3795
3796     pic_out->i_pts = h->fdec->i_pts;
3797     pic_out->i_dts = h->fdec->i_dts;
3798
3799     if( pic_out->i_pts < pic_out->i_dts )
3800         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3801
3802     pic_out->opaque = h->fenc->opaque;
3803
3804     pic_out->img.i_csp = h->fdec->i_csp;
3805 #if HIGH_BIT_DEPTH
3806     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3807 #endif
3808     pic_out->img.i_plane = h->fdec->i_plane;
3809     for( int i = 0; i < pic_out->img.i_plane; i++ )
3810     {
3811         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3812         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3813     }
3814
3815     x264_frame_push_unused( thread_current, h->fenc );
3816
3817     /* ---------------------- Update encoder state ------------------------- */
3818
3819     /* update rc */
3820     int filler = 0;
3821     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3822         return -1;
3823
3824     pic_out->hrd_timing = h->fenc->hrd_timing;
3825     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3826
3827     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3828      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3829     if( h->param.i_avcintra_class )
3830     {
3831         if( x264_check_encapsulated_buffer( h, h->thread[0], h->out.i_nal, frame_size, frame_size + filler ) < 0 )
3832             return -1;
3833
3834         x264_nal_t *nal = &h->out.nal[h->out.i_nal-1];
3835         memset( nal->p_payload + nal->i_payload, 0, filler );
3836         nal->i_payload += filler;
3837         nal->i_padding = filler;
3838         frame_size += filler;
3839
3840         /* Fix up the size header for mp4/etc */
3841         if( !h->param.b_annexb )
3842         {
3843             /* Size doesn't include the size of the header we're writing now. */
3844             uint8_t *nal_data = nal->p_payload;
3845             int chunk_size = nal->i_payload - 4;
3846             nal_data[0] = chunk_size >> 24;
3847             nal_data[1] = chunk_size >> 16;
3848             nal_data[2] = chunk_size >> 8;
3849             nal_data[3] = chunk_size >> 0;
3850         }
3851     }
3852     else
3853     {
3854         while( filler > 0 )
3855         {
3856             int f, overhead = FILLER_OVERHEAD - h->param.b_annexb;
3857             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3858             {
3859                 int next_size = filler - h->param.i_slice_max_size;
3860                 int overflow = X264_MAX( overhead - next_size, 0 );
3861                 f = h->param.i_slice_max_size - overhead - overflow;
3862             }
3863             else
3864                 f = X264_MAX( 0, filler - overhead );
3865
3866             if( x264_bitstream_check_buffer_filler( h, f ) )
3867                 return -1;
3868             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3869             x264_filler_write( h, &h->out.bs, f );
3870             if( x264_nal_end( h ) )
3871                 return -1;
3872             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3873             if( total_size < 0 )
3874                 return -1;
3875             frame_size += total_size;
3876             filler -= total_size;
3877         }
3878     }
3879
3880     /* End bitstream, set output  */
3881     *pi_nal = h->out.i_nal;
3882     *pp_nal = h->out.nal;
3883
3884     h->out.i_nal = 0;
3885
3886     x264_noise_reduction_update( h );
3887
3888     if( h->param.sc.i_buffer_size )
3889         x264_speedcontrol_frame_end( h );
3890
3891     /* ---------------------- Compute/Print statistics --------------------- */
3892     x264_thread_sync_stat( h, h->thread[0] );
3893
3894     /* Slice stat */
3895     h->stat.i_frame_count[h->sh.i_type]++;
3896     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3897     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3898
3899     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3900         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3901     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3902         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3903     for( int i = 0; i < 2; i++ )
3904         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3905     for( int i = 0; i < 6; i++ )
3906         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3907     for( int i = 0; i < 4; i++ )
3908         for( int j = 0; j < 13; j++ )
3909             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3910     if( h->sh.i_type != SLICE_TYPE_I )
3911         for( int i_list = 0; i_list < 2; i_list++ )
3912             for( int i = 0; i < X264_REF_MAX*2; i++ )
3913                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3914     for( int i = 0; i < 3; i++ )
3915         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3916     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3917     {
3918         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3919         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3920     }
3921     if( h->sh.i_type == SLICE_TYPE_B )
3922     {
3923         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3924         if( h->mb.b_direct_auto_write )
3925         {
3926             //FIXME somewhat arbitrary time constants
3927             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3928                 for( int i = 0; i < 2; i++ )
3929                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3930             for( int i = 0; i < 2; i++ )
3931                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3932         }
3933     }
3934     else
3935         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3936
3937     psz_message[0] = '\0';
3938     double dur = h->fenc->f_duration;
3939     h->stat.f_frame_duration[h->sh.i_type] += dur;
3940     if( h->param.analyse.b_psnr )
3941     {
3942         int64_t ssd[3] =
3943         {
3944             h->stat.frame.i_ssd[0],
3945             h->stat.frame.i_ssd[1],
3946             h->stat.frame.i_ssd[2],
3947         };
3948         int luma_size = h->param.i_width * h->param.i_height;
3949         int chroma_size = CHROMA_SIZE( luma_size );
3950         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3951         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3952         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3953         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3954
3955         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3956         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3957         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3958         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3959         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3960
3961         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3962                                                                     pic_out->prop.f_psnr[1],
3963                                                                     pic_out->prop.f_psnr[2] );
3964     }
3965
3966     if( h->param.analyse.b_ssim )
3967     {
3968         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3969         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3970         int msg_len = strlen(psz_message);
3971         snprintf( psz_message + msg_len, 80 - msg_len, " SSIM Y:%.5f", pic_out->prop.f_ssim );
3972     }
3973     psz_message[79] = '\0';
3974
3975     x264_log( h, X264_LOG_DEBUG,
3976               "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3977               h->i_frame,
3978               h->fdec->f_qp_avg_aq,
3979               h->i_nal_ref_idc,
3980               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3981               h->fdec->i_poc,
3982               h->stat.frame.i_mb_count_i,
3983               h->stat.frame.i_mb_count_p,
3984               h->stat.frame.i_mb_count_skip,
3985               frame_size,
3986               psz_message );
3987
3988     // keep stats all in one place
3989     x264_thread_sync_stat( h->thread[0], h );
3990     // for the use of the next frame
3991     x264_thread_sync_stat( thread_current, h );
3992
3993 #ifdef DEBUG_MB_TYPE
3994 {
3995     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3996         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3997     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3998     {
3999         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
4000             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
4001         else
4002             fprintf( stderr, "? " );
4003
4004         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
4005             fprintf( stderr, "\n" );
4006     }
4007 }
4008 #endif
4009
4010     /* Remove duplicates, must be done near the end as breaks h->fref0 array
4011      * by freeing some of its pointers. */
4012     for( int i = 0; i < h->i_ref[0]; i++ )
4013         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
4014         {
4015             x264_frame_push_blank_unused( h, h->fref[0][i] );
4016             h->fref[0][i] = 0;
4017         }
4018
4019     if( h->param.psz_dump_yuv )
4020         x264_frame_dump( h );
4021     x264_emms();
4022
4023     return frame_size;
4024 }
4025
4026 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
4027 {
4028     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
4029         b_print_pcm ? "..PCM" : "",
4030         i_mb_count[I_16x16]/ i_count,
4031         i_mb_count[I_8x8]  / i_count,
4032         i_mb_count[I_4x4]  / i_count );
4033     if( b_print_pcm )
4034         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
4035 }
4036
4037 /****************************************************************************
4038  * x264_encoder_close:
4039  ****************************************************************************/
4040 void    x264_encoder_close  ( x264_t *h )
4041 {
4042     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
4043     int64_t i_mb_count_size[2][7] = {{0}};
4044     char buf[200];
4045     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
4046                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
4047                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
4048
4049     x264_lookahead_delete( h );
4050
4051 #if HAVE_OPENCL
4052     x264_opencl_lookahead_delete( h );
4053     x264_opencl_function_t *ocl = h->opencl.ocl;
4054 #endif
4055
4056     if( h->param.b_sliced_threads )
4057         x264_threadpool_wait_all( h );
4058     if( h->param.i_threads > 1 )
4059         x264_threadpool_delete( h->threadpool );
4060     if( h->param.i_lookahead_threads > 1 )
4061         x264_threadpool_delete( h->lookaheadpool );
4062     if( h->i_thread_frames > 1 )
4063     {
4064         for( int i = 0; i < h->i_thread_frames; i++ )
4065             if( h->thread[i]->b_thread_active )
4066             {
4067                 assert( h->thread[i]->fenc->i_reference_count == 1 );
4068                 x264_frame_delete( h->thread[i]->fenc );
4069             }
4070
4071         x264_t *thread_prev = h->thread[h->i_thread_phase];
4072         x264_thread_sync_ratecontrol( h, thread_prev, h );
4073         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
4074         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
4075     }
4076     h->i_frame++;
4077
4078     /* Slices used and PSNR */
4079     for( int i = 0; i < 3; i++ )
4080     {
4081         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
4082         int i_slice = slice_order[i];
4083
4084         if( h->stat.i_frame_count[i_slice] > 0 )
4085         {
4086             int i_count = h->stat.i_frame_count[i_slice];
4087             double dur =  h->stat.f_frame_duration[i_slice];
4088             if( h->param.analyse.b_psnr )
4089             {
4090                 x264_log( h, X264_LOG_INFO,
4091                           "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",
4092                           slice_type_to_char[i_slice],
4093                           i_count,
4094                           h->stat.f_frame_qp[i_slice] / i_count,
4095                           (double)h->stat.i_frame_size[i_slice] / i_count,
4096                           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,
4097                           h->stat.f_psnr_average[i_slice] / dur,
4098                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4099             }
4100             else
4101             {
4102                 x264_log( h, X264_LOG_INFO,
4103                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4104                           slice_type_to_char[i_slice],
4105                           i_count,
4106                           h->stat.f_frame_qp[i_slice] / i_count,
4107                           (double)h->stat.i_frame_size[i_slice] / i_count );
4108             }
4109         }
4110     }
4111     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4112     {
4113         char *p = buf;
4114         int den = 0;
4115         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4116         for( int i = 0; i <= h->param.i_bframe; i++ )
4117             den += (i+1) * h->stat.i_consecutive_bframes[i];
4118         for( int i = 0; i <= h->param.i_bframe; i++ )
4119             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4120         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4121     }
4122
4123     for( int i_type = 0; i_type < 2; i_type++ )
4124         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4125         {
4126             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4127             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4128         }
4129
4130     /* MB types used */
4131     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4132     {
4133         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4134         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4135         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4136         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4137     }
4138     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4139     {
4140         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4141         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4142         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4143         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4144         x264_log( h, X264_LOG_INFO,
4145                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4146                   buf,
4147                   i_mb_size[PIXEL_16x16] / (i_count*4),
4148                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4149                   i_mb_size[PIXEL_8x8] / (i_count*4),
4150                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4151                   i_mb_size[PIXEL_4x4] / (i_count*4),
4152                   i_mb_count[P_SKIP] / i_count );
4153     }
4154     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4155     {
4156         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4157         double i_count = (double)h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4158         double i_mb_list_count;
4159         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4160         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4161         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4162         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4163             for( int j = 0; j < 2; j++ )
4164             {
4165                 int l0 = x264_mb_type_list_table[i][0][j];
4166                 int l1 = x264_mb_type_list_table[i][1][j];
4167                 if( l0 || l1 )
4168                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4169             }
4170         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4171         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4172         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4173         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4174         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4175         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4176                  i_mb_size[PIXEL_16x16] / (i_count*4),
4177                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4178                  i_mb_size[PIXEL_8x8] / (i_count*4),
4179                  i_mb_count[B_DIRECT] / i_count,
4180                  i_mb_count[B_SKIP]   / i_count );
4181         if( i_mb_list_count != 0 )
4182             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4183                      list_count[0] / i_mb_list_count,
4184                      list_count[1] / i_mb_list_count,
4185                      list_count[2] / i_mb_list_count );
4186         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4187     }
4188
4189     x264_ratecontrol_summary( h );
4190
4191     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 )
4192     {
4193 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4194 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4195         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4196         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4197                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4198         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4199         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4200                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4201         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4202                             h->stat.i_frame_count[SLICE_TYPE_P] +
4203                             h->stat.i_frame_count[SLICE_TYPE_B];
4204         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4205         int64_t i_inter = i_mb_count - i_skip - i_intra;
4206         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4207                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4208                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4209         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4210
4211         if( PARAM_INTERLACED )
4212         {
4213             char *fieldstats = buf;
4214             fieldstats[0] = 0;
4215             if( i_inter )
4216                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4217             if( i_skip )
4218                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4219             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4220                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4221         }
4222
4223         if( h->pps->b_transform_8x8_mode )
4224         {
4225             buf[0] = 0;
4226             if( h->stat.i_mb_count_8x8dct[0] )
4227                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4228             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4229         }
4230
4231         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4232             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4233             && h->stat.i_frame_count[SLICE_TYPE_B] )
4234         {
4235             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4236                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4237                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4238         }
4239
4240         buf[0] = 0;
4241         int csize = CHROMA444 ? 4 : 1;
4242         if( i_mb_count != i_all_intra )
4243             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4244                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4245                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4246                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4247         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4248                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4249                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4250                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4251                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4252
4253         int64_t fixed_pred_modes[4][9] = {{0}};
4254         int64_t sum_pred_modes[4] = {0};
4255         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4256         {
4257             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4258             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4259         }
4260         if( sum_pred_modes[0] )
4261             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4262                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4263                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4264                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4265                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4266         for( int i = 1; i <= 2; i++ )
4267         {
4268             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
4269             {
4270                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
4271                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
4272             }
4273             if( sum_pred_modes[i] )
4274                 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,
4275                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
4276                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
4277                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
4278                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
4279                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
4280                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
4281                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
4282                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
4283                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
4284         }
4285         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
4286         {
4287             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
4288             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
4289         }
4290         if( sum_pred_modes[3] && !CHROMA444 )
4291             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4292                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
4293                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
4294                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
4295                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
4296
4297         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4298             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
4299                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
4300                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
4301
4302         for( int i_list = 0; i_list < 2; i_list++ )
4303             for( int i_slice = 0; i_slice < 2; i_slice++ )
4304             {
4305                 char *p = buf;
4306                 int64_t i_den = 0;
4307                 int i_max = 0;
4308                 for( int i = 0; i < X264_REF_MAX*2; i++ )
4309                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
4310                     {
4311                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
4312                         i_max = i;
4313                     }
4314                 if( i_max == 0 )
4315                     continue;
4316                 for( int i = 0; i <= i_max; i++ )
4317                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
4318                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
4319             }
4320
4321         if( h->param.analyse.b_ssim )
4322         {
4323             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
4324             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
4325         }
4326         if( h->param.analyse.b_psnr )
4327         {
4328             x264_log( h, X264_LOG_INFO,
4329                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
4330                       SUM3( h->stat.f_psnr_mean_y ) / duration,
4331                       SUM3( h->stat.f_psnr_mean_u ) / duration,
4332                       SUM3( h->stat.f_psnr_mean_v ) / duration,
4333                       SUM3( h->stat.f_psnr_average ) / duration,
4334                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
4335                       f_bitrate );
4336         }
4337         else
4338             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
4339     }
4340
4341     /* rc */
4342     x264_ratecontrol_delete( h );
4343     x264_speedcontrol_delete( h );
4344
4345     /* param */
4346     if( h->param.rc.psz_stat_out )
4347         free( h->param.rc.psz_stat_out );
4348     if( h->param.rc.psz_stat_in )
4349         free( h->param.rc.psz_stat_in );
4350
4351     x264_cqm_delete( h );
4352     x264_free( h->nal_buffer );
4353     x264_free( h->reconfig_h );
4354     x264_analyse_free_costs( h );
4355
4356     if( h->i_thread_frames > 1 )
4357         h = h->thread[h->i_thread_phase];
4358
4359     /* frames */
4360     x264_frame_delete_list( h->frames.unused[0] );
4361     x264_frame_delete_list( h->frames.unused[1] );
4362     x264_frame_delete_list( h->frames.current );
4363     x264_frame_delete_list( h->frames.blank_unused );
4364
4365     h = h->thread[0];
4366
4367     for( int i = 0; i < h->i_thread_frames; i++ )
4368         if( h->thread[i]->b_thread_active )
4369             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4370                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4371                     x264_frame_delete( h->thread[i]->fref[0][j] );
4372
4373     if( h->param.i_lookahead_threads > 1 )
4374         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4375             x264_free( h->lookahead_thread[i] );
4376
4377     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4378     {
4379         x264_frame_t **frame;
4380
4381         if( !h->param.b_sliced_threads || i == 0 )
4382         {
4383             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4384             {
4385                 assert( (*frame)->i_reference_count > 0 );
4386                 (*frame)->i_reference_count--;
4387                 if( (*frame)->i_reference_count == 0 )
4388                     x264_frame_delete( *frame );
4389             }
4390             frame = &h->thread[i]->fdec;
4391             if( *frame )
4392             {
4393                 assert( (*frame)->i_reference_count > 0 );
4394                 (*frame)->i_reference_count--;
4395                 if( (*frame)->i_reference_count == 0 )
4396                     x264_frame_delete( *frame );
4397             }
4398             x264_macroblock_cache_free( h->thread[i] );
4399         }
4400         x264_macroblock_thread_free( h->thread[i], 0 );
4401         x264_free( h->thread[i]->out.p_bitstream );
4402         x264_free( h->thread[i]->out.nal );
4403         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4404         x264_pthread_cond_destroy( &h->thread[i]->cv );
4405         x264_free( h->thread[i] );
4406     }
4407 #if HAVE_OPENCL
4408     x264_opencl_close_library( ocl );
4409 #endif
4410 }
4411
4412 int x264_encoder_delayed_frames( x264_t *h )
4413 {
4414     int delayed_frames = 0;
4415     if( h->i_thread_frames > 1 )
4416     {
4417         for( int i = 0; i < h->i_thread_frames; i++ )
4418             delayed_frames += h->thread[i]->b_thread_active;
4419         h = h->thread[h->i_thread_phase];
4420     }
4421     for( int i = 0; h->frames.current[i]; i++ )
4422         delayed_frames++;
4423     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4424     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4425     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4426     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4427     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4428     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4429     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4430     return delayed_frames;
4431 }
4432
4433 int x264_encoder_maximum_delayed_frames( x264_t *h )
4434 {
4435     return h->frames.i_delay;
4436 }