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

   1 /*****************************************************************************
   2  * encoder.c: top-level encoder functions
   3  *****************************************************************************
   4  * Copyright (C) 2003-2014 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *
  24  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include "common/common.h"
  29
  30 #include "set.h"
  31 #include "analyse.h"
  32 #include "ratecontrol.h"
  33 #include "macroblock.h"
  34 #include "me.h"
  35
  36 //#define DEBUG_MB_TYPE
  37
  38 #define bs_write_ue bs_write_ue_big
  39
  40 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
  41                                    x264_nal_t **pp_nal, int *pi_nal,
  42                                    x264_picture_t *pic_out );
  43
  44 /****************************************************************************
  45  *
  46  ******************************* x264 libs **********************************
  47  *
  48  ****************************************************************************/
  49 static double x264_psnr( double sqe, double size )
  50 {
  51     double mse = sqe / (PIXEL_MAX*PIXEL_MAX * size);
  52     if( mse <= 0.0000000001 ) /* Max 100dB */
  53         return 100;
  54
  55     return -10.0 * log10( mse );
  56 }
  57
  58 static double x264_ssim( double ssim )
  59 {
  60     double inv_ssim = 1 - ssim;
  61     if( inv_ssim <= 0.0000000001 ) /* Max 100dB */
  62         return 100;
  63
  64     return -10.0 * log10( inv_ssim );
  65 }
  66
  67 static int x264_threadpool_wait_all( x264_t *h )
  68 {
  69     for( int i = 0; i < h->param.i_threads; i++ )
  70         if( h->thread[i]->b_thread_active )
  71         {
  72             h->thread[i]->b_thread_active = 0;
  73             if( (intptr_t)x264_threadpool_wait( h->threadpool, h->thread[i] ) < 0 )
  74                 return -1;
  75         }
  76     return 0;
  77 }
  78
  79 static void x264_frame_dump( x264_t *h )
  80 {
  81     FILE *f = x264_fopen( h->param.psz_dump_yuv, "r+b" );
  82     if( !f )
  83         return;
  84
  85     /* Wait for the threads to finish deblocking */
  86     if( h->param.b_sliced_threads )
  87         x264_threadpool_wait_all( h );
  88
  89     /* Write the frame in display order */
  90     int frame_size = FRAME_SIZE( h->param.i_height * h->param.i_width * sizeof(pixel) );
  91     fseek( f, (uint64_t)h->fdec->i_frame * frame_size, SEEK_SET );
  92     for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
  93         for( int y = 0; y < h->param.i_height; y++ )
  94             fwrite( &h->fdec->plane[p][y*h->fdec->i_stride[p]], sizeof(pixel), h->param.i_width, f );
  95     if( !CHROMA444 )
  96     {
  97         int cw = h->param.i_width>>1;
  98         int ch = h->param.i_height>>CHROMA_V_SHIFT;
  99         pixel *planeu = x264_malloc( (cw*ch*2+32)*sizeof(pixel) );
 100         pixel *planev = planeu + cw*ch + 16;
 101         h->mc.plane_copy_deinterleave( planeu, cw, planev, cw, h->fdec->plane[1], h->fdec->i_stride[1], cw, ch );
 102         fwrite( planeu, 1, cw*ch*sizeof(pixel), f );
 103         fwrite( planev, 1, cw*ch*sizeof(pixel), f );
 104         x264_free( planeu );
 105     }
 106     fclose( f );
 107 }
 108
 109 /* Fill "default" values */
 110 static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 111                                     x264_sps_t *sps, x264_pps_t *pps,
 112                                     int i_idr_pic_id, int i_frame, int i_qp )
 113 {
 114     x264_param_t *param = &h->param;
 115
 116     /* First we fill all fields */
 117     sh->sps = sps;
 118     sh->pps = pps;
 119
 120     sh->i_first_mb  = 0;
 121     sh->i_last_mb   = h->mb.i_mb_count - 1;
 122     sh->i_pps_id    = pps->i_id;
 123
 124     sh->i_frame_num = i_frame;
 125
 126     sh->b_mbaff = PARAM_INTERLACED;
 127     sh->b_field_pic = 0;    /* no field support for now */
 128     sh->b_bottom_field = 0; /* not yet used */
 129
 130     sh->i_idr_pic_id = i_idr_pic_id;
 131
 132     /* poc stuff, fixed later */
 133     sh->i_poc = 0;
 134     sh->i_delta_poc_bottom = 0;
 135     sh->i_delta_poc[0] = 0;
 136     sh->i_delta_poc[1] = 0;
 137
 138     sh->i_redundant_pic_cnt = 0;
 139
 140     h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
 141                                 && h->param.i_bframe
 142                                 && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 143
 144     if( !h->mb.b_direct_auto_read && sh->i_type == SLICE_TYPE_B )
 145     {
 146         if( h->fref[1][0]->i_poc_l0ref0 == h->fref[0][0]->i_poc )
 147         {
 148             if( h->mb.b_direct_auto_write )
 149                 sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
 150             else
 151                 sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
 152         }
 153         else
 154         {
 155             h->mb.b_direct_auto_write = 0;
 156             sh->b_direct_spatial_mv_pred = 1;
 157         }
 158     }
 159     /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 160
 161     sh->b_num_ref_idx_override = 0;
 162     sh->i_num_ref_idx_l0_active = 1;
 163     sh->i_num_ref_idx_l1_active = 1;
 164
 165     sh->b_ref_pic_list_reordering[0] = h->b_ref_reorder[0];
 166     sh->b_ref_pic_list_reordering[1] = h->b_ref_reorder[1];
 167
 168     /* If the ref list isn't in the default order, construct reordering header */
 169     for( int list = 0; list < 2; list++ )
 170     {
 171         if( sh->b_ref_pic_list_reordering[list] )
 172         {
 173             int pred_frame_num = i_frame;
 174             for( int i = 0; i < h->i_ref[list]; i++ )
 175             {
 176                 int diff = h->fref[list][i]->i_frame_num - pred_frame_num;
 177                 sh->ref_pic_list_order[list][i].idc = ( diff > 0 );
 178                 sh->ref_pic_list_order[list][i].arg = (abs(diff) - 1) & ((1 << sps->i_log2_max_frame_num) - 1);
 179                 pred_frame_num = h->fref[list][i]->i_frame_num;
 180             }
 181         }
 182     }
 183
 184     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 185
 186     sh->i_qp = SPEC_QP(i_qp);
 187     sh->i_qp_delta = sh->i_qp - pps->i_pic_init_qp;
 188     sh->b_sp_for_swidth = 0;
 189     sh->i_qs_delta = 0;
 190
 191     int deblock_thresh = i_qp + 2 * X264_MIN(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta);
 192     /* If effective qp <= 15, deblocking would have no effect anyway */
 193     if( param->b_deblocking_filter && (h->mb.b_variable_qp || 15 < deblock_thresh ) )
 194         sh->i_disable_deblocking_filter_idc = param->b_sliced_threads ? 2 : 0;
 195     else
 196         sh->i_disable_deblocking_filter_idc = 1;
 197     sh->i_alpha_c0_offset = param->i_deblocking_filter_alphac0 << 1;
 198     sh->i_beta_offset = param->i_deblocking_filter_beta << 1;
 199 }
 200
 201 static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal_ref_idc )
 202 {
 203     if( sh->b_mbaff )
 204     {
 205         int first_x = sh->i_first_mb % sh->sps->i_mb_width;
 206         int first_y = sh->i_first_mb / sh->sps->i_mb_width;
 207         assert( (first_y&1) == 0 );
 208         bs_write_ue( s, (2*first_x + sh->sps->i_mb_width*(first_y&~1) + (first_y&1)) >> 1 );
 209     }
 210     else
 211         bs_write_ue( s, sh->i_first_mb );
 212
 213     bs_write_ue( s, sh->i_type + 5 );   /* same type things */
 214     bs_write_ue( s, sh->i_pps_id );
 215     bs_write( s, sh->sps->i_log2_max_frame_num, sh->i_frame_num & ((1<<sh->sps->i_log2_max_frame_num)-1) );
 216
 217     if( !sh->sps->b_frame_mbs_only )
 218     {
 219         bs_write1( s, sh->b_field_pic );
 220         if( sh->b_field_pic )
 221             bs_write1( s, sh->b_bottom_field );
 222     }
 223
 224     if( sh->i_idr_pic_id >= 0 ) /* NAL IDR */
 225         bs_write_ue( s, sh->i_idr_pic_id );
 226
 227     if( sh->sps->i_poc_type == 0 )
 228     {
 229         bs_write( s, sh->sps->i_log2_max_poc_lsb, sh->i_poc & ((1<<sh->sps->i_log2_max_poc_lsb)-1) );
 230         if( sh->pps->b_pic_order && !sh->b_field_pic )
 231             bs_write_se( s, sh->i_delta_poc_bottom );
 232     }
 233
 234     if( sh->pps->b_redundant_pic_cnt )
 235         bs_write_ue( s, sh->i_redundant_pic_cnt );
 236
 237     if( sh->i_type == SLICE_TYPE_B )
 238         bs_write1( s, sh->b_direct_spatial_mv_pred );
 239
 240     if( sh->i_type == SLICE_TYPE_P || sh->i_type == SLICE_TYPE_B )
 241     {
 242         bs_write1( s, sh->b_num_ref_idx_override );
 243         if( sh->b_num_ref_idx_override )
 244         {
 245             bs_write_ue( s, sh->i_num_ref_idx_l0_active - 1 );
 246             if( sh->i_type == SLICE_TYPE_B )
 247                 bs_write_ue( s, sh->i_num_ref_idx_l1_active - 1 );
 248         }
 249     }
 250
 251     /* ref pic list reordering */
 252     if( sh->i_type != SLICE_TYPE_I )
 253     {
 254         bs_write1( s, sh->b_ref_pic_list_reordering[0] );
 255         if( sh->b_ref_pic_list_reordering[0] )
 256         {
 257             for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 258             {
 259                 bs_write_ue( s, sh->ref_pic_list_order[0][i].idc );
 260                 bs_write_ue( s, sh->ref_pic_list_order[0][i].arg );
 261             }
 262             bs_write_ue( s, 3 );
 263         }
 264     }
 265     if( sh->i_type == SLICE_TYPE_B )
 266     {
 267         bs_write1( s, sh->b_ref_pic_list_reordering[1] );
 268         if( sh->b_ref_pic_list_reordering[1] )
 269         {
 270             for( int i = 0; i < sh->i_num_ref_idx_l1_active; i++ )
 271             {
 272                 bs_write_ue( s, sh->ref_pic_list_order[1][i].idc );
 273                 bs_write_ue( s, sh->ref_pic_list_order[1][i].arg );
 274             }
 275             bs_write_ue( s, 3 );
 276         }
 277     }
 278
 279     sh->b_weighted_pred = 0;
 280     if( sh->pps->b_weighted_pred && sh->i_type == SLICE_TYPE_P )
 281     {
 282         sh->b_weighted_pred = sh->weight[0][0].weightfn || sh->weight[0][1].weightfn || sh->weight[0][2].weightfn;
 283         /* pred_weight_table() */
 284         bs_write_ue( s, sh->weight[0][0].i_denom );
 285         bs_write_ue( s, sh->weight[0][1].i_denom );
 286         for( int i = 0; i < sh->i_num_ref_idx_l0_active; i++ )
 287         {
 288             int luma_weight_l0_flag = !!sh->weight[i][0].weightfn;
 289             int chroma_weight_l0_flag = !!sh->weight[i][1].weightfn || !!sh->weight[i][2].weightfn;
 290             bs_write1( s, luma_weight_l0_flag );
 291             if( luma_weight_l0_flag )
 292             {
 293                 bs_write_se( s, sh->weight[i][0].i_scale );
 294                 bs_write_se( s, sh->weight[i][0].i_offset );
 295             }
 296             bs_write1( s, chroma_weight_l0_flag );
 297             if( chroma_weight_l0_flag )
 298             {
 299                 for( int j = 1; j < 3; j++ )
 300                 {
 301                     bs_write_se( s, sh->weight[i][j].i_scale );
 302                     bs_write_se( s, sh->weight[i][j].i_offset );
 303                 }
 304             }
 305         }
 306     }
 307     else if( sh->pps->b_weighted_bipred == 1 && sh->i_type == SLICE_TYPE_B )
 308     {
 309       /* TODO */
 310     }
 311
 312     if( i_nal_ref_idc != 0 )
 313     {
 314         if( sh->i_idr_pic_id >= 0 )
 315         {
 316             bs_write1( s, 0 );  /* no output of prior pics flag */
 317             bs_write1( s, 0 );  /* long term reference flag */
 318         }
 319         else
 320         {
 321             bs_write1( s, sh->i_mmco_command_count > 0 ); /* adaptive_ref_pic_marking_mode_flag */
 322             if( sh->i_mmco_command_count > 0 )
 323             {
 324                 for( int i = 0; i < sh->i_mmco_command_count; i++ )
 325                 {
 326                     bs_write_ue( s, 1 ); /* mark short term ref as unused */
 327                     bs_write_ue( s, sh->mmco[i].i_difference_of_pic_nums - 1 );
 328                 }
 329                 bs_write_ue( s, 0 ); /* end command list */
 330             }
 331         }
 332     }
 333
 334     if( sh->pps->b_cabac && sh->i_type != SLICE_TYPE_I )
 335         bs_write_ue( s, sh->i_cabac_init_idc );
 336
 337     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
 338
 339     if( sh->pps->b_deblocking_filter_control )
 340     {
 341         bs_write_ue( s, sh->i_disable_deblocking_filter_idc );
 342         if( sh->i_disable_deblocking_filter_idc != 1 )
 343         {
 344             bs_write_se( s, sh->i_alpha_c0_offset >> 1 );
 345             bs_write_se( s, sh->i_beta_offset >> 1 );
 346         }
 347     }
 348 }
 349
 350 /* If we are within a reasonable distance of the end of the memory allocated for the bitstream, */
 351 /* reallocate, adding an arbitrary amount of space. */
 352 static int x264_bitstream_check_buffer_internal( x264_t *h, int size, int b_cabac, int i_nal )
 353 {
 354     if( (b_cabac && (h->cabac.p_end - h->cabac.p < size)) ||
 355         (h->out.bs.p_end - h->out.bs.p < size) )
 356     {
 357         int buf_size = h->out.i_bitstream + size;
 358         uint8_t *buf = x264_malloc( buf_size );
 359         if( !buf )
 360             return -1;
 361         int aligned_size = h->out.i_bitstream & ~15;
 362         h->mc.memcpy_aligned( buf, h->out.p_bitstream, aligned_size );
 363         memcpy( buf + aligned_size, h->out.p_bitstream + aligned_size, h->out.i_bitstream - aligned_size );
 364
 365         intptr_t delta = buf - h->out.p_bitstream;
 366
 367         h->out.bs.p_start += delta;
 368         h->out.bs.p += delta;
 369         h->out.bs.p_end = buf + buf_size;
 370
 371         h->cabac.p_start += delta;
 372         h->cabac.p += delta;
 373         h->cabac.p_end = buf + buf_size;
 374
 375         for( int i = 0; i <= i_nal; i++ )
 376             h->out.nal[i].p_payload += delta;
 377
 378         x264_free( h->out.p_bitstream );
 379         h->out.p_bitstream = buf;
 380         h->out.i_bitstream = buf_size;
 381     }
 382     return 0;
 383 }
 384
 385 static int x264_bitstream_check_buffer( x264_t *h )
 386 {
 387     int max_row_size = (2500 << SLICE_MBAFF) * h->mb.i_mb_width;
 388     return x264_bitstream_check_buffer_internal( h, max_row_size, h->param.b_cabac, h->out.i_nal );
 389 }
 390
 391 static int x264_bitstream_check_buffer_filler( x264_t *h, int filler )
 392 {
 393     filler += 32; // add padding for safety
 394     return x264_bitstream_check_buffer_internal( h, filler, 0, -1 );
 395 }
 396
 397 #if HAVE_THREAD
 398 static void x264_encoder_thread_init( x264_t *h )
 399 {
 400     if( h->param.i_sync_lookahead )
 401         x264_lower_thread_priority( 10 );
 402 }
 403 #endif
 404
 405 /****************************************************************************
 406  *
 407  ****************************************************************************
 408  ****************************** External API*********************************
 409  ****************************************************************************
 410  *
 411  ****************************************************************************/
 412
 413 static int x264_validate_parameters( x264_t *h, int b_open )
 414 {
 415 #if HAVE_MMX
 416     if( b_open )
 417     {
 418         int cpuflags = x264_cpu_detect();
 419         int fail = 0;
 420 #ifdef __SSE__
 421         if( !(cpuflags & X264_CPU_SSE) )
 422         {
 423             x264_log( h, X264_LOG_ERROR, "your cpu does not support SSE1, but x264 was compiled with asm\n");
 424             fail = 1;
 425         }
 426 #else
 427         if( !(cpuflags & X264_CPU_MMX2) )
 428         {
 429             x264_log( h, X264_LOG_ERROR, "your cpu does not support MMXEXT, but x264 was compiled with asm\n");
 430             fail = 1;
 431         }
 432 #endif
 433         if( !fail && !(cpuflags & X264_CPU_CMOV) )
 434         {
 435             x264_log( h, X264_LOG_ERROR, "your cpu does not support CMOV, but x264 was compiled with asm\n");
 436             fail = 1;
 437         }
 438         if( fail )
 439         {
 440             x264_log( h, X264_LOG_ERROR, "to run x264, recompile without asm (configure --disable-asm)\n");
 441             return -1;
 442         }
 443     }
 444 #endif
 445
 446 #if HAVE_INTERLACED
 447     h->param.b_interlaced = !!PARAM_INTERLACED;
 448 #else
 449     if( h->param.b_interlaced )
 450     {
 451         x264_log( h, X264_LOG_ERROR, "not compiled with interlaced support\n" );
 452         return -1;
 453     }
 454 #endif
 455
 456     if( h->param.i_width <= 0 || h->param.i_height <= 0 )
 457     {
 458         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
 459                   h->param.i_width, h->param.i_height );
 460         return -1;
 461     }
 462
 463     int i_csp = h->param.i_csp & X264_CSP_MASK;
 464 #if X264_CHROMA_FORMAT
 465     if( CHROMA_FORMAT != CHROMA_420 && i_csp >= X264_CSP_I420 && i_csp <= X264_CSP_NV12 )
 466     {
 467         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:0 support\n" );
 468         return -1;
 469     }
 470     else if( CHROMA_FORMAT != CHROMA_422 && i_csp >= X264_CSP_I422 && i_csp <= X264_CSP_V210 )
 471     {
 472         x264_log( h, X264_LOG_ERROR, "not compiled with 4:2:2 support\n" );
 473         return -1;
 474     }
 475     else if( CHROMA_FORMAT != CHROMA_444 && i_csp >= X264_CSP_I444 && i_csp <= X264_CSP_RGB )
 476     {
 477         x264_log( h, X264_LOG_ERROR, "not compiled with 4:4:4 support\n" );
 478         return -1;
 479     }
 480 #endif
 481     if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
 482     {
 483         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420/YV12/NV12/I422/YV16/NV16/I444/YV24/BGR/BGRA/RGB supported)\n" );
 484         return -1;
 485     }
 486
 487     if( i_csp < X264_CSP_I444 && h->param.i_width % 2 )
 488     {
 489         x264_log( h, X264_LOG_ERROR, "width not divisible by 2 (%dx%d)\n",
 490                   h->param.i_width, h->param.i_height );
 491         return -1;
 492     }
 493
 494     if( i_csp < X264_CSP_I422 && PARAM_INTERLACED && h->param.i_height % 4 )
 495     {
 496         x264_log( h, X264_LOG_ERROR, "height not divisible by 4 (%dx%d)\n",
 497                   h->param.i_width, h->param.i_height );
 498         return -1;
 499     }
 500
 501     if( (i_csp < X264_CSP_I422 || PARAM_INTERLACED) && h->param.i_height % 2 )
 502     {
 503         x264_log( h, X264_LOG_ERROR, "height not divisible by 2 (%dx%d)\n",
 504                   h->param.i_width, h->param.i_height );
 505         return -1;
 506     }
 507
 508     if( (h->param.crop_rect.i_left + h->param.crop_rect.i_right ) >= h->param.i_width ||
 509         (h->param.crop_rect.i_top  + h->param.crop_rect.i_bottom) >= h->param.i_height )
 510     {
 511         x264_log( h, X264_LOG_ERROR, "invalid crop-rect %u,%u,%u,%u\n", h->param.crop_rect.i_left,
 512                   h->param.crop_rect.i_top, h->param.crop_rect.i_right,  h->param.crop_rect.i_bottom );
 513         return -1;
 514     }
 515
 516     if( h->param.vui.i_sar_width <= 0 || h->param.vui.i_sar_height <= 0 )
 517     {
 518         h->param.vui.i_sar_width = 0;
 519         h->param.vui.i_sar_height = 0;
 520     }
 521
 522     if( h->param.i_threads == X264_THREADS_AUTO )
 523         h->param.i_threads = x264_cpu_num_processors() * (h->param.b_sliced_threads?2:3)/2;
 524     int max_sliced_threads = X264_MAX( 1, (h->param.i_height+15)/16 / 4 );
 525     if( h->param.i_threads > 1 )
 526     {
 527 #if !HAVE_THREAD
 528         x264_log( h, X264_LOG_WARNING, "not compiled with thread support!\n");
 529         h->param.i_threads = 1;
 530 #endif
 531         /* Avoid absurdly small thread slices as they can reduce performance
 532          * and VBV compliance.  Capped at an arbitrary 4 rows per thread. */
 533         if( h->param.b_sliced_threads )
 534             h->param.i_threads = X264_MIN( h->param.i_threads, max_sliced_threads );
 535     }
 536     h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_THREAD_MAX );
 537     if( h->param.i_threads == 1 )
 538     {
 539         h->param.b_sliced_threads = 0;
 540         h->param.i_lookahead_threads = 1;
 541     }
 542     h->i_thread_frames = h->param.b_sliced_threads ? 1 : h->param.i_threads;
 543     if( h->i_thread_frames > 1 )
 544         h->param.nalu_process = NULL;
 545
 546     if( h->param.b_opencl )
 547     {
 548 #if !HAVE_OPENCL
 549         x264_log( h, X264_LOG_WARNING, "OpenCL: not compiled with OpenCL support, disabling\n" );
 550         h->param.b_opencl = 0;
 551 #elif BIT_DEPTH > 8
 552         x264_log( h, X264_LOG_WARNING, "OpenCL lookahead does not support high bit depth, disabling opencl\n" );
 553         h->param.b_opencl = 0;
 554 #else
 555         if( h->param.i_width < 32 || h->param.i_height < 32 )
 556         {
 557             x264_log( h, X264_LOG_WARNING, "OpenCL: frame size is too small, disabling opencl\n" );
 558             h->param.b_opencl = 0;
 559         }
 560 #endif
 561         if( h->param.opencl_device_id && h->param.i_opencl_device )
 562         {
 563             x264_log( h, X264_LOG_WARNING, "OpenCL: device id and device skip count configured; dropping skip\n" );
 564             h->param.i_opencl_device = 0;
 565         }
 566     }
 567
 568     h->param.i_keyint_max = x264_clip3( h->param.i_keyint_max, 1, X264_KEYINT_MAX_INFINITE );
 569     if( h->param.i_keyint_max == 1 )
 570     {
 571         h->param.b_intra_refresh = 0;
 572         h->param.analyse.i_weighted_pred = 0;
 573         h->param.i_frame_reference = 1;
 574         h->param.i_dpb_size = 1;
 575     }
 576
 577     h->param.i_frame_packing = x264_clip3( h->param.i_frame_packing, -1, 5 );
 578
 579     /* Detect default ffmpeg settings and terminate with an error. */
 580     if( b_open )
 581     {
 582         int score = 0;
 583         score += h->param.analyse.i_me_range == 0;
 584         score += h->param.rc.i_qp_step == 3;
 585         score += h->param.i_keyint_max == 12;
 586         score += h->param.rc.i_qp_min == 2;
 587         score += h->param.rc.i_qp_max == 31;
 588         score += h->param.rc.f_qcompress == 0.5;
 589         score += fabs(h->param.rc.f_ip_factor - 1.25) < 0.01;
 590         score += fabs(h->param.rc.f_pb_factor - 1.25) < 0.01;
 591         score += h->param.analyse.inter == 0 && h->param.analyse.i_subpel_refine == 8;
 592         if( score >= 5 )
 593         {
 594             x264_log( h, X264_LOG_ERROR, "broken ffmpeg default settings detected\n" );
 595             x264_log( h, X264_LOG_ERROR, "use an encoding preset (e.g. -vpre medium)\n" );
 596             x264_log( h, X264_LOG_ERROR, "preset usage: -vpre <speed> -vpre <profile>\n" );
 597             x264_log( h, X264_LOG_ERROR, "speed presets are listed in x264 --help\n" );
 598             x264_log( h, X264_LOG_ERROR, "profile is optional; x264 defaults to high\n" );
 599             return -1;
 600         }
 601     }
 602
 603     if( h->param.rc.i_rc_method < 0 || h->param.rc.i_rc_method > 2 )
 604     {
 605         x264_log( h, X264_LOG_ERROR, "no ratecontrol method specified\n" );
 606         return -1;
 607     }
 608
 609     if( PARAM_INTERLACED )
 610         h->param.b_pic_struct = 1;
 611
 612     if( h->param.i_avcintra_class )
 613     {
 614         if( BIT_DEPTH != 10 )
 615         {
 616             x264_log( h, X264_LOG_ERROR, "%2d-bit AVC-Intra is not widely compatible\n", BIT_DEPTH );
 617             x264_log( h, X264_LOG_ERROR, "10-bit x264 is required to encode AVC-Intra\n" );
 618             return -1;
 619         }
 620
 621         int type = h->param.i_avcintra_class == 200 ? 2 :
 622                    h->param.i_avcintra_class == 100 ? 1 :
 623                    h->param.i_avcintra_class == 50 ? 0 : -1;
 624         if( type < 0 )
 625         {
 626             x264_log( h, X264_LOG_ERROR, "Invalid AVC-Intra class\n" );
 627             return -1;
 628         }
 629
 630         /* [50/100/200][res][fps] */
 631         static const struct
 632         {
 633             uint16_t fps_num;
 634             uint16_t fps_den;
 635             uint8_t interlaced;
 636             uint16_t frame_size;
 637             const uint8_t *cqm_4ic;
 638             const uint8_t *cqm_8iy;
 639         } avcintra_lut[3][2][7] =
 640         {
 641             {{{ 60000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 642               {    50,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 643               { 30000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 644               {    25,    1, 0, 1100, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 645               { 24000, 1001, 0,  912, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }},
 646              {{ 30000, 1001, 1, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 647               {    25,    1, 1, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_1080i_8iy },
 648               { 60000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 649               { 30000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 650               {    50,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 651               {    25,    1, 0, 2196, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy },
 652               { 24000, 1001, 0, 1820, x264_cqm_avci50_4ic, x264_cqm_avci50_p_8iy }}},
 653             {{{ 60000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 654               {    50,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 655               { 30000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 656               {    25,    1, 0, 2224, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 657               { 24000, 1001, 0, 1848, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 658              {{ 30000, 1001, 1, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 659               {    25,    1, 1, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 660               { 60000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 661               { 30000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 662               {    50,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 663               {    25,    1, 0, 4444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 664               { 24000, 1001, 0, 3692, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}},
 665             {{{ 60000, 1001, 0, 3724, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy },
 666               {    50,    1, 0, 4472, x264_cqm_avci100_720p_4ic, x264_cqm_avci100_720p_8iy }},
 667              {{ 30000, 1001, 1, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 668               {    25,    1, 1, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080i_8iy },
 669               { 60000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 670               { 30000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 671               {    50,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 672               {    25,    1, 0, 8940, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy },
 673               { 24000, 1001, 0, 7444, x264_cqm_avci100_1080_4ic, x264_cqm_avci100_1080p_8iy }}}
 674         };
 675
 676         int res = -1;
 677         if( i_csp >= X264_CSP_I420 && i_csp < X264_CSP_I422 && !type )
 678         {
 679             if(      h->param.i_width == 1440 && h->param.i_height == 1080 ) res =  1;
 680             else if( h->param.i_width ==  960 && h->param.i_height ==  720 ) res =  0;
 681         }
 682         else if( i_csp >= X264_CSP_I422 && i_csp < X264_CSP_I444 && type )
 683         {
 684             if(      h->param.i_width == 1920 && h->param.i_height == 1080 ) res =  1;
 685             else if( h->param.i_width == 1280 && h->param.i_height ==  720 ) res =  0;
 686         }
 687         else
 688         {
 689             x264_log( h, X264_LOG_ERROR, "Invalid colorspace for AVC-Intra %d\n", h->param.i_avcintra_class );
 690             return -1;
 691         }
 692
 693         if( res < 0 )
 694         {
 695             x264_log( h, X264_LOG_ERROR, "Resolution %dx%d invalid for AVC-Intra %d\n",
 696                       h->param.i_width, h->param.i_height, h->param.i_avcintra_class );
 697             return -1;
 698         }
 699
 700         if( h->param.nalu_process )
 701         {
 702             x264_log( h, X264_LOG_ERROR, "nalu_process is not supported in AVC-Intra mode\n" );
 703             return -1;
 704         }
 705
 706         if( !h->param.b_repeat_headers )
 707         {
 708             x264_log( h, X264_LOG_ERROR, "Separate headers not supported in AVC-Intra mode\n" );
 709             return -1;
 710         }
 711
 712         int i;
 713         uint32_t fps_num = h->param.i_fps_num, fps_den = h->param.i_fps_den;
 714         x264_reduce_fraction( &fps_num, &fps_den );
 715         for( i = 0; i < 7; i++ )
 716         {
 717             if( avcintra_lut[type][res][i].fps_num == fps_num &&
 718                 avcintra_lut[type][res][i].fps_den == fps_den &&
 719                 avcintra_lut[type][res][i].interlaced == PARAM_INTERLACED )
 720             {
 721                 break;
 722             }
 723         }
 724         if( i == 7 )
 725         {
 726             x264_log( h, X264_LOG_ERROR, "FPS %d/%d%c not compatible with AVC-Intra\n",
 727                       h->param.i_fps_num, h->param.i_fps_den, PARAM_INTERLACED ? 'i' : 'p' );
 728             return -1;
 729         }
 730
 731         h->param.i_keyint_max = 1;
 732         h->param.b_intra_refresh = 0;
 733         h->param.analyse.i_weighted_pred = 0;
 734         h->param.i_frame_reference = 1;
 735         h->param.i_dpb_size = 1;
 736
 737         h->param.b_bluray_compat = 0;
 738         h->param.b_vfr_input = 0;
 739         h->param.b_aud = 1;
 740         h->param.vui.i_chroma_loc = 0;
 741         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
 742         h->param.b_deblocking_filter = 0;
 743         h->param.b_stitchable = 1;
 744         h->param.b_pic_struct = 0;
 745         h->param.analyse.b_transform_8x8 = 1;
 746         h->param.analyse.intra = X264_ANALYSE_I8x8;
 747         h->param.analyse.i_chroma_qp_offset = res && type ? 3 : 4;
 748         h->param.b_cabac = !type;
 749         h->param.rc.i_vbv_buffer_size = avcintra_lut[type][res][i].frame_size;
 750         h->param.rc.i_vbv_max_bitrate =
 751         h->param.rc.i_bitrate = h->param.rc.i_vbv_buffer_size * fps_num / fps_den;
 752         h->param.rc.i_rc_method = X264_RC_ABR;
 753         h->param.rc.f_vbv_buffer_init = 1.0;
 754         h->param.rc.b_filler = 1;
 755         h->param.i_cqm_preset = X264_CQM_CUSTOM;
 756         memcpy( h->param.cqm_4iy, x264_cqm_jvt4i, sizeof(h->param.cqm_4iy) );
 757         memcpy( h->param.cqm_4ic, avcintra_lut[type][res][i].cqm_4ic, sizeof(h->param.cqm_4ic) );
 758         memcpy( h->param.cqm_8iy, avcintra_lut[type][res][i].cqm_8iy, sizeof(h->param.cqm_8iy) );
 759
 760         /* Need exactly 10 slices of equal MB count... why?  $deity knows... */
 761         h->param.i_slice_max_mbs = ((h->param.i_width + 15) / 16) * ((h->param.i_height + 15) / 16) / 10;
 762         h->param.i_slice_max_size = 0;
 763         /* The slice structure only allows a maximum of 2 threads for 1080i/p
 764          * and 1 or 5 threads for 720p */
 765         if( h->param.b_sliced_threads )
 766         {
 767             if( res )
 768                 h->param.i_threads = X264_MIN( 2, h->param.i_threads );
 769             else
 770             {
 771                 h->param.i_threads = X264_MIN( 5, h->param.i_threads );
 772                 if( h->param.i_threads < 5 )
 773                     h->param.i_threads = 1;
 774             }
 775         }
 776
 777         if( type )
 778             h->param.vui.i_sar_width = h->param.vui.i_sar_height = 1;
 779         else
 780         {
 781             h->param.vui.i_sar_width  = 4;
 782             h->param.vui.i_sar_height = 3;
 783         }
 784
 785         /* Official encoder doesn't appear to go under 13
 786          * and Avid cannot handle negative QPs */
 787         h->param.rc.i_qp_min = X264_MAX( h->param.rc.i_qp_min, QP_BD_OFFSET + 1 );
 788     }
 789
 790     h->param.rc.f_rf_constant = x264_clip3f( h->param.rc.f_rf_constant, -QP_BD_OFFSET, 51 );
 791     h->param.rc.f_rf_constant_max = x264_clip3f( h->param.rc.f_rf_constant_max, -QP_BD_OFFSET, 51 );
 792     h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, QP_MAX );
 793     h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 0, 11 );
 794     h->param.rc.f_ip_factor = X264_MAX( h->param.rc.f_ip_factor, 0.01f );
 795     h->param.rc.f_pb_factor = X264_MAX( h->param.rc.f_pb_factor, 0.01f );
 796     if( h->param.rc.i_rc_method == X264_RC_CRF )
 797     {
 798         h->param.rc.i_qp_constant = h->param.rc.f_rf_constant + QP_BD_OFFSET;
 799         h->param.rc.i_bitrate = 0;
 800     }
 801     if( b_open && (h->param.rc.i_rc_method == X264_RC_CQP || h->param.rc.i_rc_method == X264_RC_CRF)
 802         && h->param.rc.i_qp_constant == 0 )
 803     {
 804         h->mb.b_lossless = 1;
 805         h->param.i_cqm_preset = X264_CQM_FLAT;
 806         h->param.psz_cqm_file = NULL;
 807         h->param.rc.i_rc_method = X264_RC_CQP;
 808         h->param.rc.f_ip_factor = 1;
 809         h->param.rc.f_pb_factor = 1;
 810         h->param.analyse.b_psnr = 0;
 811         h->param.analyse.b_ssim = 0;
 812         h->param.analyse.i_chroma_qp_offset = 0;
 813         h->param.analyse.i_trellis = 0;
 814         h->param.analyse.b_fast_pskip = 0;
 815         h->param.analyse.i_noise_reduction = 0;
 816         h->param.analyse.b_psy = 0;
 817         h->param.i_bframe = 0;
 818         /* 8x8dct is not useful without RD in CAVLC lossless */
 819         if( !h->param.b_cabac && h->param.analyse.i_subpel_refine < 6 )
 820             h->param.analyse.b_transform_8x8 = 0;
 821     }
 822     if( h->param.rc.i_rc_method == X264_RC_CQP )
 823     {
 824         float qp_p = h->param.rc.i_qp_constant;
 825         float qp_i = qp_p - 6*log2f( h->param.rc.f_ip_factor );
 826         float qp_b = qp_p + 6*log2f( h->param.rc.f_pb_factor );
 827         h->param.rc.i_qp_min = x264_clip3( (int)(X264_MIN3( qp_p, qp_i, qp_b )), 0, QP_MAX );
 828         h->param.rc.i_qp_max = x264_clip3( (int)(X264_MAX3( qp_p, qp_i, qp_b ) + .999), 0, QP_MAX );
 829         h->param.rc.i_aq_mode = 0;
 830         h->param.rc.b_mb_tree = 0;
 831         h->param.rc.i_bitrate = 0;
 832     }
 833     h->param.rc.i_qp_max = x264_clip3( h->param.rc.i_qp_max, 0, QP_MAX );
 834     h->param.rc.i_qp_min = x264_clip3( h->param.rc.i_qp_min, 0, h->param.rc.i_qp_max );
 835     h->param.rc.i_qp_step = x264_clip3( h->param.rc.i_qp_step, 2, QP_MAX );
 836     h->param.rc.i_bitrate = x264_clip3( h->param.rc.i_bitrate, 0, 2000000 );
 837     if( h->param.rc.i_rc_method == X264_RC_ABR && !h->param.rc.i_bitrate )
 838     {
 839         x264_log( h, X264_LOG_ERROR, "bitrate not specified\n" );
 840         return -1;
 841     }
 842     h->param.rc.i_vbv_buffer_size = x264_clip3( h->param.rc.i_vbv_buffer_size, 0, 2000000 );
 843     h->param.rc.i_vbv_max_bitrate = x264_clip3( h->param.rc.i_vbv_max_bitrate, 0, 2000000 );
 844     h->param.rc.f_vbv_buffer_init = x264_clip3f( h->param.rc.f_vbv_buffer_init, 0, 2000000 );
 845     if( h->param.rc.i_vbv_buffer_size )
 846     {
 847         if( h->param.rc.i_rc_method == X264_RC_CQP )
 848         {
 849             x264_log( h, X264_LOG_WARNING, "VBV is incompatible with constant QP, ignored.\n" );
 850             h->param.rc.i_vbv_max_bitrate = 0;
 851             h->param.rc.i_vbv_buffer_size = 0;
 852         }
 853         else if( h->param.rc.i_vbv_max_bitrate == 0 )
 854         {
 855             if( h->param.rc.i_rc_method == X264_RC_ABR )
 856             {
 857                 x264_log( h, X264_LOG_WARNING, "VBV maxrate unspecified, assuming CBR\n" );
 858                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate;
 859             }
 860             else
 861             {
 862                 x264_log( h, X264_LOG_WARNING, "VBV bufsize set but maxrate unspecified, ignored\n" );
 863                 h->param.rc.i_vbv_buffer_size = 0;
 864             }
 865         }
 866         else if( h->param.rc.i_vbv_max_bitrate < h->param.rc.i_bitrate &&
 867                  h->param.rc.i_rc_method == X264_RC_ABR )
 868         {
 869             x264_log( h, X264_LOG_WARNING, "max bitrate less than average bitrate, assuming CBR\n" );
 870             h->param.rc.i_bitrate = h->param.rc.i_vbv_max_bitrate;
 871         }
 872     }
 873     else if( h->param.rc.i_vbv_max_bitrate )
 874     {
 875         x264_log( h, X264_LOG_WARNING, "VBV maxrate specified, but no bufsize, ignored\n" );
 876         h->param.rc.i_vbv_max_bitrate = 0;
 877     }
 878
 879     h->param.i_slice_max_size = X264_MAX( h->param.i_slice_max_size, 0 );
 880     h->param.i_slice_max_mbs = X264_MAX( h->param.i_slice_max_mbs, 0 );
 881     h->param.i_slice_min_mbs = X264_MAX( h->param.i_slice_min_mbs, 0 );
 882     if( h->param.i_slice_max_mbs )
 883         h->param.i_slice_min_mbs = X264_MIN( h->param.i_slice_min_mbs, h->param.i_slice_max_mbs/2 );
 884     else if( !h->param.i_slice_max_size )
 885         h->param.i_slice_min_mbs = 0;
 886     if( PARAM_INTERLACED && h->param.i_slice_min_mbs )
 887     {
 888         x264_log( h, X264_LOG_WARNING, "interlace + slice-min-mbs is not implemented\n" );
 889         h->param.i_slice_min_mbs = 0;
 890     }
 891     int mb_width = (h->param.i_width+15)/16;
 892     if( h->param.i_slice_min_mbs > mb_width )
 893     {
 894         x264_log( h, X264_LOG_WARNING, "slice-min-mbs > row mb size (%d) not implemented\n", mb_width );
 895         h->param.i_slice_min_mbs = mb_width;
 896     }
 897
 898     int max_slices = (h->param.i_height+((16<<PARAM_INTERLACED)-1))/(16<<PARAM_INTERLACED);
 899     if( h->param.b_sliced_threads )
 900         h->param.i_slice_count = x264_clip3( h->param.i_threads, 0, max_slices );
 901     else
 902     {
 903         h->param.i_slice_count = x264_clip3( h->param.i_slice_count, 0, max_slices );
 904         if( h->param.i_slice_max_mbs || h->param.i_slice_max_size )
 905             h->param.i_slice_count = 0;
 906     }
 907     if( h->param.i_slice_count_max > 0 )
 908         h->param.i_slice_count_max = X264_MAX( h->param.i_slice_count, h->param.i_slice_count_max );
 909
 910     if( h->param.b_bluray_compat )
 911     {
 912         h->param.i_bframe_pyramid = X264_MIN( X264_B_PYRAMID_STRICT, h->param.i_bframe_pyramid );
 913         h->param.i_bframe = X264_MIN( h->param.i_bframe, 3 );
 914         h->param.b_aud = 1;
 915         h->param.i_nal_hrd = X264_MAX( h->param.i_nal_hrd, X264_NAL_HRD_VBR );
 916         h->param.i_slice_max_size = 0;
 917         h->param.i_slice_max_mbs = 0;
 918         h->param.b_intra_refresh = 0;
 919         h->param.i_frame_reference = X264_MIN( h->param.i_frame_reference, 6 );
 920         h->param.i_dpb_size = X264_MIN( h->param.i_dpb_size, 6 );
 921         /* Don't use I-frames, because Blu-ray treats them the same as IDR. */
 922         h->param.i_keyint_min = 1;
 923         /* Due to the proliferation of broken players that don't handle dupes properly. */
 924         h->param.analyse.i_weighted_pred = X264_MIN( h->param.analyse.i_weighted_pred, X264_WEIGHTP_SIMPLE );
 925         if( h->param.b_fake_interlaced )
 926             h->param.b_pic_struct = 1;
 927     }
 928
 929     h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, X264_REF_MAX );
 930     h->param.i_dpb_size = x264_clip3( h->param.i_dpb_size, 1, X264_REF_MAX );
 931     if( h->param.i_scenecut_threshold < 0 )
 932         h->param.i_scenecut_threshold = 0;
 933     h->param.analyse.i_direct_mv_pred = x264_clip3( h->param.analyse.i_direct_mv_pred, X264_DIRECT_PRED_NONE, X264_DIRECT_PRED_AUTO );
 934     if( !h->param.analyse.i_subpel_refine && h->param.analyse.i_direct_mv_pred > X264_DIRECT_PRED_SPATIAL )
 935     {
 936         x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
 937         h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
 938     }
 939     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
 940     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
 941     if( h->param.i_bframe <= 1 )
 942         h->param.i_bframe_pyramid = X264_B_PYRAMID_NONE;
 943     h->param.i_bframe_pyramid = x264_clip3( h->param.i_bframe_pyramid, X264_B_PYRAMID_NONE, X264_B_PYRAMID_NORMAL );
 944     h->param.i_bframe_adaptive = x264_clip3( h->param.i_bframe_adaptive, X264_B_ADAPT_NONE, X264_B_ADAPT_TRELLIS );
 945     if( !h->param.i_bframe )
 946     {
 947         h->param.i_bframe_adaptive = X264_B_ADAPT_NONE;
 948         h->param.analyse.i_direct_mv_pred = 0;
 949         h->param.analyse.b_weighted_bipred = 0;
 950         h->param.b_open_gop = 0;
 951     }
 952     if( h->param.b_intra_refresh && h->param.i_bframe_pyramid == X264_B_PYRAMID_NORMAL )
 953     {
 954         x264_log( h, X264_LOG_WARNING, "b-pyramid normal + intra-refresh is not supported\n" );
 955         h->param.i_bframe_pyramid = X264_B_PYRAMID_STRICT;
 956     }
 957     if( h->param.b_intra_refresh && (h->param.i_frame_reference > 1 || h->param.i_dpb_size > 1) )
 958     {
 959         x264_log( h, X264_LOG_WARNING, "ref > 1 + intra-refresh is not supported\n" );
 960         h->param.i_frame_reference = 1;
 961         h->param.i_dpb_size = 1;
 962     }
 963     if( h->param.b_intra_refresh && h->param.b_open_gop )
 964     {
 965         x264_log( h, X264_LOG_WARNING, "intra-refresh is not compatible with open-gop\n" );
 966         h->param.b_open_gop = 0;
 967     }
 968     if( !h->param.i_fps_num || !h->param.i_fps_den )
 969     {
 970         h->param.i_fps_num = 25;
 971         h->param.i_fps_den = 1;
 972     }
 973     float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
 974     if( h->param.i_keyint_min == X264_KEYINT_MIN_AUTO )
 975         h->param.i_keyint_min = X264_MIN( h->param.i_keyint_max / 10, fps );
 976     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
 977     h->param.rc.i_lookahead = x264_clip3( h->param.rc.i_lookahead, 0, X264_LOOKAHEAD_MAX );
 978     {
 979         int maxrate = X264_MAX( h->param.rc.i_vbv_max_bitrate, h->param.rc.i_bitrate );
 980         float bufsize = maxrate ? (float)h->param.rc.i_vbv_buffer_size / maxrate : 0;
 981         h->param.rc.i_lookahead = X264_MIN( h->param.rc.i_lookahead, X264_MAX( h->param.i_keyint_max, bufsize*fps ) );
 982     }
 983
 984     if( !h->param.i_timebase_num || !h->param.i_timebase_den || !(h->param.b_vfr_input || h->param.b_pulldown) )
 985     {
 986         h->param.i_timebase_num = h->param.i_fps_den;
 987         h->param.i_timebase_den = h->param.i_fps_num;
 988     }
 989
 990     h->param.rc.f_qcompress = x264_clip3f( h->param.rc.f_qcompress, 0.0, 1.0 );
 991     if( h->param.i_keyint_max == 1 || h->param.rc.f_qcompress == 1 )
 992         h->param.rc.b_mb_tree = 0;
 993     if( (!h->param.b_intra_refresh && h->param.i_keyint_max != X264_KEYINT_MAX_INFINITE) &&
 994         !h->param.rc.i_lookahead && h->param.rc.b_mb_tree )
 995     {
 996         x264_log( h, X264_LOG_WARNING, "lookaheadless mb-tree requires intra refresh or infinite keyint\n" );
 997         h->param.rc.b_mb_tree = 0;
 998     }
 999     if( b_open && h->param.rc.b_stat_read )
1000         h->param.rc.i_lookahead = 0;
1001 #if HAVE_THREAD
1002     if( h->param.i_sync_lookahead < 0 )
1003         h->param.i_sync_lookahead = h->param.i_bframe + 1;
1004     h->param.i_sync_lookahead = X264_MIN( h->param.i_sync_lookahead, X264_LOOKAHEAD_MAX );
1005     if( h->param.rc.b_stat_read || h->i_thread_frames == 1 )
1006         h->param.i_sync_lookahead = 0;
1007 #else
1008     h->param.i_sync_lookahead = 0;
1009 #endif
1010
1011     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
1012     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
1013     h->param.analyse.i_luma_deadzone[0] = x264_clip3( h->param.analyse.i_luma_deadzone[0], 0, 32 );
1014     h->param.analyse.i_luma_deadzone[1] = x264_clip3( h->param.analyse.i_luma_deadzone[1], 0, 32 );
1015
1016     h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
1017
1018     if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
1019         h->param.i_cqm_preset = X264_CQM_FLAT;
1020
1021     if( h->param.analyse.i_me_method < X264_ME_DIA ||
1022         h->param.analyse.i_me_method > X264_ME_TESA )
1023         h->param.analyse.i_me_method = X264_ME_HEX;
1024     h->param.analyse.i_me_range = x264_clip3( h->param.analyse.i_me_range, 4, 1024 );
1025     if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
1026         h->param.analyse.i_me_range = 16;
1027     if( h->param.analyse.i_me_method == X264_ME_TESA &&
1028         (h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1) )
1029         h->param.analyse.i_me_method = X264_ME_ESA;
1030     h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
1031     h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
1032                               X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1033     h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
1034     if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
1035         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1036     if( !h->param.analyse.b_transform_8x8 )
1037     {
1038         h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
1039         h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
1040     }
1041     h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
1042     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 2 );
1043     h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
1044     if( h->param.rc.f_aq_strength == 0 )
1045         h->param.rc.i_aq_mode = 0;
1046
1047     if( h->param.i_log_level < X264_LOG_INFO )
1048     {
1049         h->param.analyse.b_psnr = 0;
1050         h->param.analyse.b_ssim = 0;
1051     }
1052     /* Warn users trying to measure PSNR/SSIM with psy opts on. */
1053     if( b_open && (h->param.analyse.b_psnr || h->param.analyse.b_ssim) )
1054     {
1055         char *s = NULL;
1056
1057         if( h->param.analyse.b_psy )
1058         {
1059             s = h->param.analyse.b_psnr ? "psnr" : "ssim";
1060             x264_log( h, X264_LOG_WARNING, "--%s used with psy on: results will be invalid!\n", s );
1061         }
1062         else if( !h->param.rc.i_aq_mode && h->param.analyse.b_ssim )
1063         {
1064             x264_log( h, X264_LOG_WARNING, "--ssim used with AQ off: results will be invalid!\n" );
1065             s = "ssim";
1066         }
1067         else if(  h->param.rc.i_aq_mode && h->param.analyse.b_psnr )
1068         {
1069             x264_log( h, X264_LOG_WARNING, "--psnr used with AQ on: results will be invalid!\n" );
1070             s = "psnr";
1071         }
1072         if( s )
1073             x264_log( h, X264_LOG_WARNING, "--tune %s should be used if attempting to benchmark %s!\n", s, s );
1074     }
1075
1076     if( !h->param.analyse.b_psy )
1077     {
1078         h->param.analyse.f_psy_rd = 0;
1079         h->param.analyse.f_psy_trellis = 0;
1080     }
1081     h->param.analyse.f_psy_rd = x264_clip3f( h->param.analyse.f_psy_rd, 0, 10 );
1082     h->param.analyse.f_psy_trellis = x264_clip3f( h->param.analyse.f_psy_trellis, 0, 10 );
1083     h->mb.i_psy_rd = h->param.analyse.i_subpel_refine >= 6 ? FIX8( h->param.analyse.f_psy_rd ) : 0;
1084     h->mb.i_psy_trellis = h->param.analyse.i_trellis ? FIX8( h->param.analyse.f_psy_trellis / 4 ) : 0;
1085     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -32, 32);
1086     /* In 4:4:4 mode, chroma gets twice as much resolution, so we can halve its quality. */
1087     if( b_open && i_csp >= X264_CSP_I444 && i_csp < X264_CSP_BGR && h->param.analyse.b_psy )
1088         h->param.analyse.i_chroma_qp_offset += 6;
1089     /* Psy RDO increases overall quantizers to improve the quality of luma--this indirectly hurts chroma quality */
1090     /* so we lower the chroma QP offset to compensate */
1091     if( b_open && h->mb.i_psy_rd && !h->param.i_avcintra_class )
1092         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_rd < 0.25 ? 1 : 2;
1093     /* Psy trellis has a similar effect. */
1094     if( b_open && h->mb.i_psy_trellis && !h->param.i_avcintra_class )
1095         h->param.analyse.i_chroma_qp_offset -= h->param.analyse.f_psy_trellis < 0.25 ? 1 : 2;
1096     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
1097     /* MB-tree requires AQ to be on, even if the strength is zero. */
1098     if( !h->param.rc.i_aq_mode && h->param.rc.b_mb_tree )
1099     {
1100         h->param.rc.i_aq_mode = 1;
1101         h->param.rc.f_aq_strength = 0;
1102     }
1103     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
1104     if( h->param.analyse.i_subpel_refine >= 10 && (h->param.analyse.i_trellis != 2 || !h->param.rc.i_aq_mode) )
1105         h->param.analyse.i_subpel_refine = 9;
1106
1107     {
1108         const x264_level_t *l = x264_levels;
1109         if( h->param.i_level_idc < 0 )
1110         {
1111             int maxrate_bak = h->param.rc.i_vbv_max_bitrate;
1112             if( h->param.rc.i_rc_method == X264_RC_ABR && h->param.rc.i_vbv_buffer_size <= 0 )
1113                 h->param.rc.i_vbv_max_bitrate = h->param.rc.i_bitrate * 2;
1114             x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1115             do h->param.i_level_idc = l->level_idc;
1116                 while( l[1].level_idc && x264_validate_levels( h, 0 ) && l++ );
1117             h->param.rc.i_vbv_max_bitrate = maxrate_bak;
1118         }
1119         else
1120         {
1121             while( l->level_idc && l->level_idc != h->param.i_level_idc )
1122                 l++;
1123             if( l->level_idc == 0 )
1124             {
1125                 x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
1126                 return -1;
1127             }
1128         }
1129         if( h->param.analyse.i_mv_range <= 0 )
1130             h->param.analyse.i_mv_range = l->mv_range >> PARAM_INTERLACED;
1131         else
1132             h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 512 >> PARAM_INTERLACED);
1133     }
1134
1135     h->param.analyse.i_weighted_pred = x264_clip3( h->param.analyse.i_weighted_pred, X264_WEIGHTP_NONE, X264_WEIGHTP_SMART );
1136
1137     if( h->param.i_lookahead_threads == X264_THREADS_AUTO )
1138     {
1139         if( h->param.b_sliced_threads )
1140             h->param.i_lookahead_threads = h->param.i_threads;
1141         else
1142         {
1143             /* If we're using much slower lookahead settings than encoding settings, it helps a lot to use
1144              * more lookahead threads.  This typically happens in the first pass of a two-pass encode, so
1145              * try to guess at this sort of case.
1146              *
1147              * Tuned by a little bit of real encoding with the various presets. */
1148             int badapt = h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS;
1149             int subme = X264_MIN( h->param.analyse.i_subpel_refine / 3, 3 ) + (h->param.analyse.i_subpel_refine > 1);
1150             int bframes = X264_MIN( (h->param.i_bframe - 1) / 3, 3 );
1151
1152             /* [b-adapt 0/1 vs 2][quantized subme][quantized bframes] */
1153             static const uint8_t lookahead_thread_div[2][5][4] =
1154             {{{6,6,6,6}, {3,3,3,3}, {4,4,4,4}, {6,6,6,6}, {12,12,12,12}},
1155              {{3,2,1,1}, {2,1,1,1}, {4,3,2,1}, {6,4,3,2}, {12, 9, 6, 4}}};
1156
1157             h->param.i_lookahead_threads = h->param.i_threads / lookahead_thread_div[badapt][subme][bframes];
1158             /* Since too many lookahead threads significantly degrades lookahead accuracy, limit auto
1159              * lookahead threads to about 8 macroblock rows high each at worst.  This number is chosen
1160              * pretty much arbitrarily. */
1161             h->param.i_lookahead_threads = X264_MIN( h->param.i_lookahead_threads, h->param.i_height / 128 );
1162         }
1163     }
1164     h->param.i_lookahead_threads = x264_clip3( h->param.i_lookahead_threads, 1, X264_MIN( max_sliced_threads, X264_LOOKAHEAD_THREAD_MAX ) );
1165
1166     if( PARAM_INTERLACED )
1167     {
1168         if( h->param.analyse.i_me_method >= X264_ME_ESA )
1169         {
1170             x264_log( h, X264_LOG_WARNING, "interlace + me=esa is not implemented\n" );
1171             h->param.analyse.i_me_method = X264_ME_UMH;
1172         }
1173         if( h->param.analyse.i_weighted_pred > 0 )
1174         {
1175             x264_log( h, X264_LOG_WARNING, "interlace + weightp is not implemented\n" );
1176             h->param.analyse.i_weighted_pred = X264_WEIGHTP_NONE;
1177         }
1178     }
1179
1180     if( !h->param.analyse.i_weighted_pred && h->param.rc.b_mb_tree && h->param.analyse.b_psy )
1181         h->param.analyse.i_weighted_pred = X264_WEIGHTP_FAKE;
1182
1183     if( h->i_thread_frames > 1 )
1184     {
1185         int r = h->param.analyse.i_mv_range_thread;
1186         int r2;
1187         if( r <= 0 )
1188         {
1189             // half of the available space is reserved and divided evenly among the threads,
1190             // the rest is allocated to whichever thread is far enough ahead to use it.
1191             // reserving more space increases quality for some videos, but costs more time
1192             // in thread synchronization.
1193             int max_range = (h->param.i_height + X264_THREAD_HEIGHT) / h->i_thread_frames - X264_THREAD_HEIGHT;
1194             r = max_range / 2;
1195         }
1196         r = X264_MAX( r, h->param.analyse.i_me_range );
1197         r = X264_MIN( r, h->param.analyse.i_mv_range );
1198         // round up to use the whole mb row
1199         r2 = (r & ~15) + ((-X264_THREAD_HEIGHT) & 15);
1200         if( r2 < r )
1201             r2 += 16;
1202         x264_log( h, X264_LOG_DEBUG, "using mv_range_thread = %d\n", r2 );
1203         h->param.analyse.i_mv_range_thread = r2;
1204     }
1205
1206     if( h->param.rc.f_rate_tolerance < 0 )
1207         h->param.rc.f_rate_tolerance = 0;
1208     if( h->param.rc.f_qblur < 0 )
1209         h->param.rc.f_qblur = 0;
1210     if( h->param.rc.f_complexity_blur < 0 )
1211         h->param.rc.f_complexity_blur = 0;
1212
1213     h->param.i_sps_id &= 31;
1214
1215     h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
1216
1217     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
1218     {
1219         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
1220         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
1221     }
1222
1223     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
1224        (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
1225     {
1226         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
1227         h->param.i_nal_hrd = X264_NAL_HRD_VBR;
1228     }
1229
1230     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
1231         h->param.rc.b_filler = 1;
1232
1233     /* ensure the booleans are 0 or 1 so they can be used in math */
1234 #define BOOLIFY(x) h->param.x = !!h->param.x
1235     BOOLIFY( b_cabac );
1236     BOOLIFY( b_constrained_intra );
1237     BOOLIFY( b_deblocking_filter );
1238     BOOLIFY( b_deterministic );
1239     BOOLIFY( b_sliced_threads );
1240     BOOLIFY( b_interlaced );
1241     BOOLIFY( b_intra_refresh );
1242     BOOLIFY( b_aud );
1243     BOOLIFY( b_repeat_headers );
1244     BOOLIFY( b_annexb );
1245     BOOLIFY( b_vfr_input );
1246     BOOLIFY( b_pulldown );
1247     BOOLIFY( b_tff );
1248     BOOLIFY( b_pic_struct );
1249     BOOLIFY( b_fake_interlaced );
1250     BOOLIFY( b_open_gop );
1251     BOOLIFY( b_bluray_compat );
1252     BOOLIFY( b_stitchable );
1253     BOOLIFY( b_full_recon );
1254     BOOLIFY( b_opencl );
1255     BOOLIFY( analyse.b_transform_8x8 );
1256     BOOLIFY( analyse.b_weighted_bipred );
1257     BOOLIFY( analyse.b_chroma_me );
1258     BOOLIFY( analyse.b_mixed_references );
1259     BOOLIFY( analyse.b_fast_pskip );
1260     BOOLIFY( analyse.b_dct_decimate );
1261     BOOLIFY( analyse.b_psy );
1262     BOOLIFY( analyse.b_psnr );
1263     BOOLIFY( analyse.b_ssim );
1264     BOOLIFY( rc.b_stat_write );
1265     BOOLIFY( rc.b_stat_read );
1266     BOOLIFY( rc.b_mb_tree );
1267     BOOLIFY( rc.b_filler );
1268 #undef BOOLIFY
1269
1270     return 0;
1271 }
1272
1273 static void mbcmp_init( x264_t *h )
1274 {
1275     int satd = !h->mb.b_lossless && h->param.analyse.i_subpel_refine > 1;
1276     memcpy( h->pixf.mbcmp, satd ? h->pixf.satd : h->pixf.sad_aligned, sizeof(h->pixf.mbcmp) );
1277     memcpy( h->pixf.mbcmp_unaligned, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.mbcmp_unaligned) );
1278     h->pixf.intra_mbcmp_x3_16x16 = satd ? h->pixf.intra_satd_x3_16x16 : h->pixf.intra_sad_x3_16x16;
1279     h->pixf.intra_mbcmp_x3_8x16c = satd ? h->pixf.intra_satd_x3_8x16c : h->pixf.intra_sad_x3_8x16c;
1280     h->pixf.intra_mbcmp_x3_8x8c  = satd ? h->pixf.intra_satd_x3_8x8c  : h->pixf.intra_sad_x3_8x8c;
1281     h->pixf.intra_mbcmp_x3_8x8 = satd ? h->pixf.intra_sa8d_x3_8x8 : h->pixf.intra_sad_x3_8x8;
1282     h->pixf.intra_mbcmp_x3_4x4 = satd ? h->pixf.intra_satd_x3_4x4 : h->pixf.intra_sad_x3_4x4;
1283     h->pixf.intra_mbcmp_x9_4x4 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1284                                : satd ? h->pixf.intra_satd_x9_4x4 : h->pixf.intra_sad_x9_4x4;
1285     h->pixf.intra_mbcmp_x9_8x8 = h->param.b_cpu_independent || h->mb.b_lossless ? NULL
1286                                : satd ? h->pixf.intra_sa8d_x9_8x8 : h->pixf.intra_sad_x9_8x8;
1287     satd &= h->param.analyse.i_me_method == X264_ME_TESA;
1288     memcpy( h->pixf.fpelcmp, satd ? h->pixf.satd : h->pixf.sad, sizeof(h->pixf.fpelcmp) );
1289     memcpy( h->pixf.fpelcmp_x3, satd ? h->pixf.satd_x3 : h->pixf.sad_x3, sizeof(h->pixf.fpelcmp_x3) );
1290     memcpy( h->pixf.fpelcmp_x4, satd ? h->pixf.satd_x4 : h->pixf.sad_x4, sizeof(h->pixf.fpelcmp_x4) );
1291 }
1292
1293 static void chroma_dsp_init( x264_t *h )
1294 {
1295     memcpy( h->luma2chroma_pixel, x264_luma2chroma_pixel[CHROMA_FORMAT], sizeof(h->luma2chroma_pixel) );
1296
1297     switch( CHROMA_FORMAT )
1298     {
1299         case CHROMA_420:
1300             memcpy( h->predict_chroma, h->predict_8x8c, sizeof(h->predict_chroma) );
1301             h->mc.prefetch_fenc = h->mc.prefetch_fenc_420;
1302             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_420;
1303             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_420_intra;
1304             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_420_mbaff;
1305             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_420_intra_mbaff;
1306             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x8c;
1307             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last4;
1308             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run4;
1309             break;
1310         case CHROMA_422:
1311             memcpy( h->predict_chroma, h->predict_8x16c, sizeof(h->predict_chroma) );
1312             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422;
1313             h->loopf.deblock_chroma[0] = h->loopf.deblock_h_chroma_422;
1314             h->loopf.deblock_chroma_intra[0] = h->loopf.deblock_h_chroma_422_intra;
1315             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_chroma_422_mbaff;
1316             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_chroma_422_intra_mbaff;
1317             h->pixf.intra_mbcmp_x3_chroma = h->pixf.intra_mbcmp_x3_8x16c;
1318             h->quantf.coeff_last[DCT_CHROMA_DC] = h->quantf.coeff_last8;
1319             h->quantf.coeff_level_run[DCT_CHROMA_DC] = h->quantf.coeff_level_run8;
1320             break;
1321         case CHROMA_444:
1322             h->mc.prefetch_fenc = h->mc.prefetch_fenc_422; /* FIXME: doesn't cover V plane */
1323             h->loopf.deblock_chroma_mbaff = h->loopf.deblock_luma_mbaff;
1324             h->loopf.deblock_chroma_intra_mbaff = h->loopf.deblock_luma_intra_mbaff;
1325             break;
1326     }
1327 }
1328
1329 static void x264_set_aspect_ratio( x264_t *h, x264_param_t *param, int initial )
1330 {
1331     /* VUI */
1332     if( param->vui.i_sar_width > 0 && param->vui.i_sar_height > 0 )
1333     {
1334         uint32_t i_w = param->vui.i_sar_width;
1335         uint32_t i_h = param->vui.i_sar_height;
1336         uint32_t old_w = h->param.vui.i_sar_width;
1337         uint32_t old_h = h->param.vui.i_sar_height;
1338
1339         x264_reduce_fraction( &i_w, &i_h );
1340
1341         while( i_w > 65535 || i_h > 65535 )
1342         {
1343             i_w /= 2;
1344             i_h /= 2;
1345         }
1346
1347         x264_reduce_fraction( &i_w, &i_h );
1348
1349         if( i_w != old_w || i_h != old_h || initial )
1350         {
1351             h->param.vui.i_sar_width = 0;
1352             h->param.vui.i_sar_height = 0;
1353             if( i_w == 0 || i_h == 0 )
1354                 x264_log( h, X264_LOG_WARNING, "cannot create valid sample aspect ratio\n" );
1355             else
1356             {
1357                 x264_log( h, initial?X264_LOG_INFO:X264_LOG_DEBUG, "using SAR=%d/%d\n", i_w, i_h );
1358                 h->param.vui.i_sar_width = i_w;
1359                 h->param.vui.i_sar_height = i_h;
1360             }
1361         }
1362     }
1363 }
1364
1365 /****************************************************************************
1366  * x264_encoder_open:
1367  ****************************************************************************/
1368 x264_t *x264_encoder_open( x264_param_t *param )
1369 {
1370     x264_t *h;
1371     char buf[1000], *p;
1372     int qp, i_slicetype_length;
1373
1374     CHECKED_MALLOCZERO( h, sizeof(x264_t) );
1375
1376     /* Create a copy of param */
1377     memcpy( &h->param, param, sizeof(x264_param_t) );
1378
1379     if( param->param_free )
1380         param->param_free( param );
1381
1382     if( x264_threading_init() )
1383     {
1384         x264_log( h, X264_LOG_ERROR, "unable to initialize threading\n" );
1385         goto fail;
1386     }
1387
1388     if( x264_validate_parameters( h, 1 ) < 0 )
1389         goto fail;
1390
1391     if( h->param.psz_cqm_file )
1392         if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
1393             goto fail;
1394
1395     if( h->param.rc.psz_stat_out )
1396         h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
1397     if( h->param.rc.psz_stat_in )
1398         h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
1399
1400     x264_reduce_fraction( &h->param.i_fps_num, &h->param.i_fps_den );
1401     x264_reduce_fraction( &h->param.i_timebase_num, &h->param.i_timebase_den );
1402
1403     /* Init x264_t */
1404     h->i_frame = -1;
1405     h->i_frame_num = 0;
1406     h->i_idr_pic_id = 0;
1407
1408     if( (uint64_t)h->param.i_timebase_den * 2 > UINT32_MAX )
1409     {
1410         x264_log( h, X264_LOG_ERROR, "Effective timebase denominator %u exceeds H.264 maximum\n", h->param.i_timebase_den );
1411         goto fail;
1412     }
1413
1414     x264_set_aspect_ratio( h, &h->param, 1 );
1415
1416     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1417     x264_pps_init( h->pps, h->param.i_sps_id, &h->param, h->sps );
1418
1419     x264_validate_levels( h, 1 );
1420
1421     h->chroma_qp_table = i_chroma_qp_table + 12 + h->pps->i_chroma_qp_index_offset;
1422
1423     if( x264_cqm_init( h ) < 0 )
1424         goto fail;
1425
1426     h->mb.i_mb_width = h->sps->i_mb_width;
1427     h->mb.i_mb_height = h->sps->i_mb_height;
1428     h->mb.i_mb_count = h->mb.i_mb_width * h->mb.i_mb_height;
1429
1430     h->mb.chroma_h_shift = CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422;
1431     h->mb.chroma_v_shift = CHROMA_FORMAT == CHROMA_420;
1432
1433     /* Adaptive MBAFF and subme 0 are not supported as we require halving motion
1434      * vectors during prediction, resulting in hpel mvs.
1435      * The chosen solution is to make MBAFF non-adaptive in this case. */
1436     h->mb.b_adaptive_mbaff = PARAM_INTERLACED && h->param.analyse.i_subpel_refine;
1437
1438     /* Init frames. */
1439     if( h->param.i_bframe_adaptive == X264_B_ADAPT_TRELLIS && !h->param.rc.b_stat_read )
1440         h->frames.i_delay = X264_MAX(h->param.i_bframe,3)*4;
1441     else
1442         h->frames.i_delay = h->param.i_bframe;
1443     if( h->param.rc.b_mb_tree || h->param.rc.i_vbv_buffer_size )
1444         h->frames.i_delay = X264_MAX( h->frames.i_delay, h->param.rc.i_lookahead );
1445     i_slicetype_length = h->frames.i_delay;
1446     h->frames.i_delay += h->i_thread_frames - 1;
1447     h->frames.i_delay += h->param.i_sync_lookahead;
1448     h->frames.i_delay += h->param.b_vfr_input;
1449     h->frames.i_bframe_delay = h->param.i_bframe ? (h->param.i_bframe_pyramid ? 2 : 1) : 0;
1450
1451     h->frames.i_max_ref0 = h->param.i_frame_reference;
1452     h->frames.i_max_ref1 = X264_MIN( h->sps->vui.i_num_reorder_frames, h->param.i_frame_reference );
1453     h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering;
1454     h->frames.b_have_lowres = !h->param.rc.b_stat_read
1455         && ( h->param.rc.i_rc_method == X264_RC_ABR
1456           || h->param.rc.i_rc_method == X264_RC_CRF
1457           || h->param.i_bframe_adaptive
1458           || h->param.i_scenecut_threshold
1459           || h->param.rc.b_mb_tree
1460           || h->param.analyse.i_weighted_pred );
1461     h->frames.b_have_lowres |= h->param.rc.b_stat_read && h->param.rc.i_vbv_buffer_size > 0;
1462     h->frames.b_have_sub8x8_esa = !!(h->param.analyse.inter & X264_ANALYSE_PSUB8x8);
1463
1464     h->frames.i_last_idr =
1465     h->frames.i_last_keyframe = - h->param.i_keyint_max;
1466     h->frames.i_input    = 0;
1467     h->frames.i_largest_pts = h->frames.i_second_largest_pts = -1;
1468     h->frames.i_poc_last_open_gop = -1;
1469
1470     CHECKED_MALLOCZERO( h->frames.unused[0], (h->frames.i_delay + 3) * sizeof(x264_frame_t *) );
1471     /* Allocate room for max refs plus a few extra just in case. */
1472     CHECKED_MALLOCZERO( h->frames.unused[1], (h->i_thread_frames + X264_REF_MAX + 4) * sizeof(x264_frame_t *) );
1473     CHECKED_MALLOCZERO( h->frames.current, (h->param.i_sync_lookahead + h->param.i_bframe
1474                         + h->i_thread_frames + 3) * sizeof(x264_frame_t *) );
1475     if( h->param.analyse.i_weighted_pred > 0 )
1476         CHECKED_MALLOCZERO( h->frames.blank_unused, h->i_thread_frames * 4 * sizeof(x264_frame_t *) );
1477     h->i_ref[0] = h->i_ref[1] = 0;
1478     h->i_cpb_delay = h->i_coded_fields = h->i_disp_fields = 0;
1479     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);
1480     h->i_disp_fields_last_frame = -1;
1481     x264_rdo_init();
1482
1483     /* init CPU functions */
1484     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
1485     x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
1486     x264_predict_8x16c_init( h->param.cpu, h->predict_8x16c );
1487     x264_predict_8x8_init( h->param.cpu, h->predict_8x8, &h->predict_8x8_filter );
1488     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
1489     x264_pixel_init( h->param.cpu, &h->pixf );
1490     x264_dct_init( h->param.cpu, &h->dctf );
1491     x264_zigzag_init( h->param.cpu, &h->zigzagf_progressive, &h->zigzagf_interlaced );
1492     memcpy( &h->zigzagf, PARAM_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
1493     x264_mc_init( h->param.cpu, &h->mc, h->param.b_cpu_independent );
1494     x264_quant_init( h, h->param.cpu, &h->quantf );
1495     x264_deblock_init( h->param.cpu, &h->loopf, PARAM_INTERLACED );
1496     x264_bitstream_init( h->param.cpu, &h->bsf );
1497     if( h->param.b_cabac )
1498         x264_cabac_init( h );
1499     else
1500         x264_stack_align( x264_cavlc_init, h );
1501
1502     mbcmp_init( h );
1503     chroma_dsp_init( h );
1504
1505     p = buf + sprintf( buf, "using cpu capabilities:" );
1506     for( int i = 0; x264_cpu_names[i].flags; i++ )
1507     {
1508         if( !strcmp(x264_cpu_names[i].name, "SSE")
1509             && h->param.cpu & (X264_CPU_SSE2) )
1510             continue;
1511         if( !strcmp(x264_cpu_names[i].name, "SSE2")
1512             && h->param.cpu & (X264_CPU_SSE2_IS_FAST|X264_CPU_SSE2_IS_SLOW) )
1513             continue;
1514         if( !strcmp(x264_cpu_names[i].name, "SSE3")
1515             && (h->param.cpu & X264_CPU_SSSE3 || !(h->param.cpu & X264_CPU_CACHELINE_64)) )
1516             continue;
1517         if( !strcmp(x264_cpu_names[i].name, "SSE4.1")
1518             && (h->param.cpu & X264_CPU_SSE42) )
1519             continue;
1520         if( !strcmp(x264_cpu_names[i].name, "BMI1")
1521             && (h->param.cpu & X264_CPU_BMI2) )
1522             continue;
1523         if( (h->param.cpu & x264_cpu_names[i].flags) == x264_cpu_names[i].flags
1524             && (!i || x264_cpu_names[i].flags != x264_cpu_names[i-1].flags) )
1525             p += sprintf( p, " %s", x264_cpu_names[i].name );
1526     }
1527     if( !h->param.cpu )
1528         p += sprintf( p, " none!" );
1529     x264_log( h, X264_LOG_INFO, "%s\n", buf );
1530
1531     float *logs = x264_analyse_prepare_costs( h );
1532     if( !logs )
1533         goto fail;
1534     for( qp = X264_MIN( h->param.rc.i_qp_min, QP_MAX_SPEC ); qp <= h->param.rc.i_qp_max; qp++ )
1535         if( x264_analyse_init_costs( h, logs, qp ) )
1536             goto fail;
1537     if( x264_analyse_init_costs( h, logs, X264_LOOKAHEAD_QP ) )
1538         goto fail;
1539     x264_free( logs );
1540
1541     static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
1542     /* Checks for known miscompilation issues. */
1543     if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
1544     {
1545         x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
1546         goto fail;
1547     }
1548
1549     /* Must be volatile or else GCC will optimize it out. */
1550     volatile int temp = 392;
1551     if( x264_clz( temp ) != 23 )
1552     {
1553         x264_log( h, X264_LOG_ERROR, "CLZ test failed: x264 has been miscompiled!\n" );
1554 #if ARCH_X86 || ARCH_X86_64
1555         x264_log( h, X264_LOG_ERROR, "Are you attempting to run an SSE4a/LZCNT-targeted build on a CPU that\n" );
1556         x264_log( h, X264_LOG_ERROR, "doesn't support it?\n" );
1557 #endif
1558         goto fail;
1559     }
1560
1561     h->out.i_nal = 0;
1562     h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 4
1563         * ( h->param.rc.i_rc_method == X264_RC_ABR ? pow( 0.95, h->param.rc.i_qp_min )
1564           : pow( 0.95, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
1565
1566     h->nal_buffer_size = h->out.i_bitstream * 3/2 + 4 + 64; /* +4 for startcode, +64 for nal_escape assembly padding */
1567     CHECKED_MALLOC( h->nal_buffer, h->nal_buffer_size );
1568
1569     CHECKED_MALLOC( h->reconfig_h, sizeof(x264_t) );
1570
1571     if( h->param.i_threads > 1 &&
1572         x264_threadpool_init( &h->threadpool, h->param.i_threads, (void*)x264_encoder_thread_init, h ) )
1573         goto fail;
1574     if( h->param.i_lookahead_threads > 1 &&
1575         x264_threadpool_init( &h->lookaheadpool, h->param.i_lookahead_threads, NULL, NULL ) )
1576         goto fail;
1577
1578 #if HAVE_OPENCL
1579     if( h->param.b_opencl )
1580     {
1581         h->opencl.ocl = x264_opencl_load_library();
1582         if( !h->opencl.ocl )
1583         {
1584             x264_log( h, X264_LOG_WARNING, "failed to load OpenCL\n" );
1585             h->param.b_opencl = 0;
1586         }
1587     }
1588 #endif
1589
1590     h->thread[0] = h;
1591     for( int i = 1; i < h->param.i_threads + !!h->param.i_sync_lookahead; i++ )
1592         CHECKED_MALLOC( h->thread[i], sizeof(x264_t) );
1593     if( h->param.i_lookahead_threads > 1 )
1594         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
1595         {
1596             CHECKED_MALLOC( h->lookahead_thread[i], sizeof(x264_t) );
1597             *h->lookahead_thread[i] = *h;
1598         }
1599     *h->reconfig_h = *h;
1600
1601     for( int i = 0; i < h->param.i_threads; i++ )
1602     {
1603         int init_nal_count = h->param.i_slice_count + 3;
1604         int allocate_threadlocal_data = !h->param.b_sliced_threads || !i;
1605         if( i > 0 )
1606             *h->thread[i] = *h;
1607
1608         if( x264_pthread_mutex_init( &h->thread[i]->mutex, NULL ) )
1609             goto fail;
1610         if( x264_pthread_cond_init( &h->thread[i]->cv, NULL ) )
1611             goto fail;
1612
1613         if( allocate_threadlocal_data )
1614         {
1615             h->thread[i]->fdec = x264_frame_pop_unused( h, 1 );
1616             if( !h->thread[i]->fdec )
1617                 goto fail;
1618         }
1619         else
1620             h->thread[i]->fdec = h->thread[0]->fdec;
1621
1622         CHECKED_MALLOC( h->thread[i]->out.p_bitstream, h->out.i_bitstream );
1623         /* Start each thread with room for init_nal_count NAL units; it'll realloc later if needed. */
1624         CHECKED_MALLOC( h->thread[i]->out.nal, init_nal_count*sizeof(x264_nal_t) );
1625         h->thread[i]->out.i_nals_allocated = init_nal_count;
1626
1627         if( allocate_threadlocal_data && x264_macroblock_cache_allocate( h->thread[i] ) < 0 )
1628             goto fail;
1629     }
1630
1631 #if HAVE_OPENCL
1632     if( h->param.b_opencl && x264_opencl_lookahead_init( h ) < 0 )
1633         h->param.b_opencl = 0;
1634 #endif
1635
1636     if( x264_lookahead_init( h, i_slicetype_length ) )
1637         goto fail;
1638
1639     for( int i = 0; i < h->param.i_threads; i++ )
1640         if( x264_macroblock_thread_allocate( h->thread[i], 0 ) < 0 )
1641             goto fail;
1642
1643     if( x264_ratecontrol_new( h ) < 0 )
1644         goto fail;
1645
1646     if( h->param.i_nal_hrd )
1647     {
1648         x264_log( h, X264_LOG_DEBUG, "HRD bitrate: %i bits/sec\n", h->sps->vui.hrd.i_bit_rate_unscaled );
1649         x264_log( h, X264_LOG_DEBUG, "CPB size: %i bits\n", h->sps->vui.hrd.i_cpb_size_unscaled );
1650     }
1651
1652     if( h->param.psz_dump_yuv )
1653     {
1654         /* create or truncate the reconstructed video file */
1655         FILE *f = x264_fopen( h->param.psz_dump_yuv, "w" );
1656         if( !f )
1657         {
1658             x264_log( h, X264_LOG_ERROR, "dump_yuv: can't write to %s\n", h->param.psz_dump_yuv );
1659             goto fail;
1660         }
1661         else if( !x264_is_regular_file( f ) )
1662         {
1663             x264_log( h, X264_LOG_ERROR, "dump_yuv: incompatible with non-regular file %s\n", h->param.psz_dump_yuv );
1664             goto fail;
1665         }
1666         fclose( f );
1667     }
1668
1669     const char *profile = h->sps->i_profile_idc == PROFILE_BASELINE ? "Constrained Baseline" :
1670                           h->sps->i_profile_idc == PROFILE_MAIN ? "Main" :
1671                           h->sps->i_profile_idc == PROFILE_HIGH ? "High" :
1672                           h->sps->i_profile_idc == PROFILE_HIGH10 ? (h->sps->b_constraint_set3 == 1 ? "High 10 Intra" : "High 10") :
1673                           h->sps->i_profile_idc == PROFILE_HIGH422 ? (h->sps->b_constraint_set3 == 1 ? "High 4:2:2 Intra" : "High 4:2:2") :
1674                           h->sps->b_constraint_set3 == 1 ? "High 4:4:4 Intra" : "High 4:4:4 Predictive";
1675     char level[4];
1676     snprintf( level, sizeof(level), "%d.%d", h->sps->i_level_idc/10, h->sps->i_level_idc%10 );
1677     if( h->sps->i_level_idc == 9 || ( h->sps->i_level_idc == 11 && h->sps->b_constraint_set3 &&
1678         (h->sps->i_profile_idc == PROFILE_BASELINE || h->sps->i_profile_idc == PROFILE_MAIN) ) )
1679         strcpy( level, "1b" );
1680
1681     if( h->sps->i_profile_idc < PROFILE_HIGH10 )
1682     {
1683         x264_log( h, X264_LOG_INFO, "profile %s, level %s\n",
1684             profile, level );
1685     }
1686     else
1687     {
1688         static const char * const subsampling[4] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
1689         x264_log( h, X264_LOG_INFO, "profile %s, level %s, %s %d-bit\n",
1690             profile, level, subsampling[CHROMA_FORMAT], BIT_DEPTH );
1691     }
1692
1693     return h;
1694 fail:
1695     x264_free( h );
1696     return NULL;
1697 }
1698
1699 /****************************************************************************/
1700 static int x264_encoder_try_reconfig( x264_t *h, x264_param_t *param, int *rc_reconfig )
1701 {
1702     *rc_reconfig = 0;
1703     x264_set_aspect_ratio( h, param, 0 );
1704 #define COPY(var) h->param.var = param->var
1705     COPY( i_frame_reference ); // but never uses more refs than initially specified
1706     COPY( i_bframe_bias );
1707     if( h->param.i_scenecut_threshold )
1708         COPY( i_scenecut_threshold ); // can't turn it on or off, only vary the threshold
1709     COPY( b_deblocking_filter );
1710     COPY( i_deblocking_filter_alphac0 );
1711     COPY( i_deblocking_filter_beta );
1712     COPY( i_frame_packing );
1713     COPY( analyse.inter );
1714     COPY( analyse.intra );
1715     COPY( analyse.i_direct_mv_pred );
1716     /* Scratch buffer prevents me_range from being increased for esa/tesa */
1717     if( h->param.analyse.i_me_method < X264_ME_ESA || param->analyse.i_me_range < h->param.analyse.i_me_range )
1718         COPY( analyse.i_me_range );
1719     COPY( analyse.i_noise_reduction );
1720     /* We can't switch out of subme=0 during encoding. */
1721     if( h->param.analyse.i_subpel_refine )
1722         COPY( analyse.i_subpel_refine );
1723     COPY( analyse.i_trellis );
1724     COPY( analyse.b_chroma_me );
1725     COPY( analyse.b_dct_decimate );
1726     COPY( analyse.b_fast_pskip );
1727     COPY( analyse.b_mixed_references );
1728     COPY( analyse.f_psy_rd );
1729     COPY( analyse.f_psy_trellis );
1730     COPY( crop_rect );
1731     // can only twiddle these if they were enabled to begin with:
1732     if( h->param.analyse.i_me_method >= X264_ME_ESA || param->analyse.i_me_method < X264_ME_ESA )
1733         COPY( analyse.i_me_method );
1734     if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->frames.b_have_sub8x8_esa )
1735         h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
1736     if( h->pps->b_transform_8x8_mode )
1737         COPY( analyse.b_transform_8x8 );
1738     if( h->frames.i_max_ref1 > 1 )
1739         COPY( i_bframe_pyramid );
1740     COPY( i_slice_max_size );
1741     COPY( i_slice_max_mbs );
1742     COPY( i_slice_min_mbs );
1743     COPY( i_slice_count );
1744     COPY( i_slice_count_max );
1745     COPY( b_tff );
1746
1747     /* VBV can't be turned on if it wasn't on to begin with */
1748     if( h->param.rc.i_vbv_max_bitrate > 0 && h->param.rc.i_vbv_buffer_size > 0 &&
1749           param->rc.i_vbv_max_bitrate > 0 &&   param->rc.i_vbv_buffer_size > 0 )
1750     {
1751         *rc_reconfig |= h->param.rc.i_vbv_max_bitrate != param->rc.i_vbv_max_bitrate;
1752         *rc_reconfig |= h->param.rc.i_vbv_buffer_size != param->rc.i_vbv_buffer_size;
1753         *rc_reconfig |= h->param.rc.i_bitrate != param->rc.i_bitrate;
1754         COPY( rc.i_vbv_max_bitrate );
1755         COPY( rc.i_vbv_buffer_size );
1756         COPY( rc.i_bitrate );
1757     }
1758     *rc_reconfig |= h->param.rc.f_rf_constant != param->rc.f_rf_constant;
1759     *rc_reconfig |= h->param.rc.f_rf_constant_max != param->rc.f_rf_constant_max;
1760     COPY( rc.f_rf_constant );
1761     COPY( rc.f_rf_constant_max );
1762 #undef COPY
1763
1764     return x264_validate_parameters( h, 0 );
1765 }
1766
1767 int x264_encoder_reconfig_apply( x264_t *h, x264_param_t *param )
1768 {
1769     int rc_reconfig;
1770     int ret = x264_encoder_try_reconfig( h, param, &rc_reconfig );
1771
1772     mbcmp_init( h );
1773     if( !ret )
1774         x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
1775
1776     /* Supported reconfiguration options (1-pass only):
1777      * vbv-maxrate
1778      * vbv-bufsize
1779      * crf
1780      * bitrate (CBR only) */
1781     if( !ret && rc_reconfig )
1782         x264_ratecontrol_init_reconfigurable( h, 0 );
1783
1784     return ret;
1785 }
1786
1787 /****************************************************************************
1788  * x264_encoder_reconfig:
1789  ****************************************************************************/
1790 int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
1791 {
1792     h = h->thread[h->thread[0]->i_thread_phase];
1793     x264_param_t param_save = h->reconfig_h->param;
1794     h->reconfig_h->param = h->param;
1795
1796     int rc_reconfig;
1797     int ret = x264_encoder_try_reconfig( h->reconfig_h, param, &rc_reconfig );
1798     if( !ret )
1799         h->reconfig = 1;
1800     else
1801         h->reconfig_h->param = param_save;
1802
1803     return ret;
1804 }
1805
1806 /****************************************************************************
1807  * x264_encoder_parameters:
1808  ****************************************************************************/
1809 void x264_encoder_parameters( x264_t *h, x264_param_t *param )
1810 {
1811     memcpy( param, &h->thread[h->i_thread_phase]->param, sizeof(x264_param_t) );
1812 }
1813
1814 /* internal usage */
1815 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
1816 {
1817     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1818
1819     nal->i_ref_idc        = i_ref_idc;
1820     nal->i_type           = i_type;
1821     nal->b_long_startcode = 1;
1822
1823     nal->i_payload= 0;
1824     nal->p_payload= &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1825     nal->i_padding= 0;
1826 }
1827
1828 /* if number of allocated nals is not enough, re-allocate a larger one. */
1829 static int x264_nal_check_buffer( x264_t *h )
1830 {
1831     if( h->out.i_nal >= h->out.i_nals_allocated )
1832     {
1833         x264_nal_t *new_out = x264_malloc( sizeof(x264_nal_t) * (h->out.i_nals_allocated*2) );
1834         if( !new_out )
1835             return -1;
1836         memcpy( new_out, h->out.nal, sizeof(x264_nal_t) * (h->out.i_nals_allocated) );
1837         x264_free( h->out.nal );
1838         h->out.nal = new_out;
1839         h->out.i_nals_allocated *= 2;
1840     }
1841     return 0;
1842 }
1843
1844 static int x264_nal_end( x264_t *h )
1845 {
1846     x264_nal_t *nal = &h->out.nal[h->out.i_nal];
1847     uint8_t *end = &h->out.p_bitstream[bs_pos( &h->out.bs ) / 8];
1848     nal->i_payload = end - nal->p_payload;
1849     /* Assembly implementation of nal_escape reads past the end of the input.
1850      * While undefined padding wouldn't actually affect the output, it makes valgrind unhappy. */
1851     memset( end, 0xff, 64 );
1852     if( h->param.nalu_process )
1853         h->param.nalu_process( h, nal, h->fenc->opaque );
1854     h->out.i_nal++;
1855
1856     return x264_nal_check_buffer( h );
1857 }
1858
1859 static int x264_check_encapsulated_buffer( x264_t *h, x264_t *h0, int start,
1860                                            int previous_nal_size, int necessary_size )
1861 {
1862     if( h0->nal_buffer_size < necessary_size )
1863     {
1864         necessary_size *= 2;
1865         uint8_t *buf = x264_malloc( necessary_size );
1866         if( !buf )
1867             return -1;
1868         if( previous_nal_size )
1869             memcpy( buf, h0->nal_buffer, previous_nal_size );
1870
1871         intptr_t delta = buf - h0->nal_buffer;
1872         for( int i = 0; i < start; i++ )
1873             h->out.nal[i].p_payload += delta;
1874
1875         x264_free( h0->nal_buffer );
1876         h0->nal_buffer = buf;
1877         h0->nal_buffer_size = necessary_size;
1878     }
1879
1880     return 0;
1881 }
1882
1883 static int x264_encoder_encapsulate_nals( x264_t *h, int start )
1884 {
1885     x264_t *h0 = h->thread[0];
1886     int nal_size = 0, previous_nal_size = 0;
1887
1888     if( h->param.nalu_process )
1889     {
1890         for( int i = start; i < h->out.i_nal; i++ )
1891             nal_size += h->out.nal[i].i_payload;
1892         return nal_size;
1893     }
1894
1895     for( int i = 0; i < start; i++ )
1896         previous_nal_size += h->out.nal[i].i_payload;
1897
1898     for( int i = start; i < h->out.i_nal; i++ )
1899         nal_size += h->out.nal[i].i_payload;
1900
1901     /* Worst-case NAL unit escaping: reallocate the buffer if it's too small. */
1902     int necessary_size = previous_nal_size + nal_size * 3/2 + h->out.i_nal * 4 + 4 + 64;
1903     for( int i = start; i < h->out.i_nal; i++ )
1904         necessary_size += h->out.nal[i].i_padding;
1905     if( x264_check_encapsulated_buffer( h, h0, start, previous_nal_size, necessary_size ) )
1906         return -1;
1907
1908     uint8_t *nal_buffer = h0->nal_buffer + previous_nal_size;
1909
1910     for( int i = start; i < h->out.i_nal; i++ )
1911     {
1912         int old_payload_len = h->out.nal[i].i_payload;
1913         h->out.nal[i].b_long_startcode = !i || h->out.nal[i].i_type == NAL_SPS || h->out.nal[i].i_type == NAL_PPS ||
1914                                          h->param.i_avcintra_class;
1915         x264_nal_encode( h, nal_buffer, &h->out.nal[i] );
1916         nal_buffer += h->out.nal[i].i_payload;
1917         if( h->param.i_avcintra_class )
1918         {
1919             h->out.nal[i].i_padding -= h->out.nal[i].i_payload - (old_payload_len + NALU_OVERHEAD);
1920             if( h->out.nal[i].i_padding > 0 )
1921             {
1922                 memset( nal_buffer, 0, h->out.nal[i].i_padding );
1923                 nal_buffer += h->out.nal[i].i_padding;
1924                 h->out.nal[i].i_payload += h->out.nal[i].i_padding;
1925             }
1926             h->out.nal[i].i_padding = X264_MAX( h->out.nal[i].i_padding, 0 );
1927         }
1928     }
1929
1930     x264_emms();
1931
1932     return nal_buffer - (h0->nal_buffer + previous_nal_size);
1933 }
1934
1935 /****************************************************************************
1936  * x264_encoder_headers:
1937  ****************************************************************************/
1938 int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
1939 {
1940     int frame_size = 0;
1941     /* init bitstream context */
1942     h->out.i_nal = 0;
1943     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
1944
1945     /* Write SEI, SPS and PPS. */
1946
1947     /* generate sequence parameters */
1948     x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
1949     x264_sps_write( &h->out.bs, h->sps );
1950     if( x264_nal_end( h ) )
1951         return -1;
1952
1953     /* generate picture parameters */
1954     x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
1955     x264_pps_write( &h->out.bs, h->sps, h->pps );
1956     if( x264_nal_end( h ) )
1957         return -1;
1958
1959     /* identify ourselves */
1960     x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
1961     if( x264_sei_version_write( h, &h->out.bs ) )
1962         return -1;
1963     if( x264_nal_end( h ) )
1964         return -1;
1965
1966     frame_size = x264_encoder_encapsulate_nals( h, 0 );
1967     if( frame_size < 0 )
1968         return -1;
1969
1970     /* now set output*/
1971     *pi_nal = h->out.i_nal;
1972     *pp_nal = &h->out.nal[0];
1973     h->out.i_nal = 0;
1974
1975     return frame_size;
1976 }
1977
1978 /* Check to see whether we have chosen a reference list ordering different
1979  * from the standard's default. */
1980 static inline void x264_reference_check_reorder( x264_t *h )
1981 {
1982     /* The reorder check doesn't check for missing frames, so just
1983      * force a reorder if one of the reference list is corrupt. */
1984     for( int i = 0; h->frames.reference[i]; i++ )
1985         if( h->frames.reference[i]->b_corrupt )
1986         {
1987             h->b_ref_reorder[0] = 1;
1988             return;
1989         }
1990     for( int list = 0; list <= (h->sh.i_type == SLICE_TYPE_B); list++ )
1991         for( int i = 0; i < h->i_ref[list] - 1; i++ )
1992         {
1993             int framenum_diff = h->fref[list][i+1]->i_frame_num - h->fref[list][i]->i_frame_num;
1994             int poc_diff = h->fref[list][i+1]->i_poc - h->fref[list][i]->i_poc;
1995             /* P and B-frames use different default orders. */
1996             if( h->sh.i_type == SLICE_TYPE_P ? framenum_diff > 0 : list == 1 ? poc_diff < 0 : poc_diff > 0 )
1997             {
1998                 h->b_ref_reorder[list] = 1;
1999                 return;
2000             }
2001         }
2002 }
2003
2004 /* return -1 on failure, else return the index of the new reference frame */
2005 int x264_weighted_reference_duplicate( x264_t *h, int i_ref, const x264_weight_t *w )
2006 {
2007     int i = h->i_ref[0];
2008     int j = 1;
2009     x264_frame_t *newframe;
2010     if( i <= 1 ) /* empty list, definitely can't duplicate frame */
2011         return -1;
2012
2013     //Duplication is only used in X264_WEIGHTP_SMART
2014     if( h->param.analyse.i_weighted_pred != X264_WEIGHTP_SMART )
2015         return -1;
2016
2017     /* Duplication is a hack to compensate for crappy rounding in motion compensation.
2018      * With high bit depth, it's not worth doing, so turn it off except in the case of
2019      * unweighted dupes. */
2020     if( BIT_DEPTH > 8 && w != x264_weight_none )
2021         return -1;
2022
2023     newframe = x264_frame_pop_blank_unused( h );
2024     if( !newframe )
2025         return -1;
2026
2027     //FIXME: probably don't need to copy everything
2028     *newframe = *h->fref[0][i_ref];
2029     newframe->i_reference_count = 1;
2030     newframe->orig = h->fref[0][i_ref];
2031     newframe->b_duplicate = 1;
2032     memcpy( h->fenc->weight[j], w, sizeof(h->fenc->weight[i]) );
2033
2034     /* shift the frames to make space for the dupe. */
2035     h->b_ref_reorder[0] = 1;
2036     if( h->i_ref[0] < X264_REF_MAX )
2037         ++h->i_ref[0];
2038     h->fref[0][X264_REF_MAX-1] = NULL;
2039     x264_frame_unshift( &h->fref[0][j], newframe );
2040
2041     return j;
2042 }
2043
2044 static void x264_weighted_pred_init( x264_t *h )
2045 {
2046     /* for now no analysis and set all weights to nothing */
2047     for( int i_ref = 0; i_ref < h->i_ref[0]; i_ref++ )
2048         h->fenc->weighted[i_ref] = h->fref[0][i_ref]->filtered[0][0];
2049
2050     // FIXME: This only supports weighting of one reference frame
2051     // and duplicates of that frame.
2052     h->fenc->i_lines_weighted = 0;
2053
2054     for( int i_ref = 0; i_ref < (h->i_ref[0] << SLICE_MBAFF); i_ref++ )
2055         for( int i = 0; i < 3; i++ )
2056             h->sh.weight[i_ref][i].weightfn = NULL;
2057
2058
2059     if( h->sh.i_type != SLICE_TYPE_P || h->param.analyse.i_weighted_pred <= 0 )
2060         return;
2061
2062     int i_padv = PADV << PARAM_INTERLACED;
2063     int denom = -1;
2064     int weightplane[2] = { 0, 0 };
2065     int buffer_next = 0;
2066     for( int i = 0; i < 3; i++ )
2067     {
2068         for( int j = 0; j < h->i_ref[0]; j++ )
2069         {
2070             if( h->fenc->weight[j][i].weightfn )
2071             {
2072                 h->sh.weight[j][i] = h->fenc->weight[j][i];
2073                 // if weight is useless, don't write it to stream
2074                 if( h->sh.weight[j][i].i_scale == 1<<h->sh.weight[j][i].i_denom && h->sh.weight[j][i].i_offset == 0 )
2075                     h->sh.weight[j][i].weightfn = NULL;
2076                 else
2077                 {
2078                     if( !weightplane[!!i] )
2079                     {
2080                         weightplane[!!i] = 1;
2081                         h->sh.weight[0][!!i].i_denom = denom = h->sh.weight[j][i].i_denom;
2082                         assert( x264_clip3( denom, 0, 7 ) == denom );
2083                     }
2084
2085                     assert( h->sh.weight[j][i].i_denom == denom );
2086                     if( !i )
2087                     {
2088                         h->fenc->weighted[j] = h->mb.p_weight_buf[buffer_next++] + h->fenc->i_stride[0] * i_padv + PADH;
2089                         //scale full resolution frame
2090                         if( h->param.i_threads == 1 )
2091                         {
2092                             pixel *src = h->fref[0][j]->filtered[0][0] - h->fref[0][j]->i_stride[0]*i_padv - PADH;
2093                             pixel *dst = h->fenc->weighted[j] - h->fenc->i_stride[0]*i_padv - PADH;
2094                             int stride = h->fenc->i_stride[0];
2095                             int width = h->fenc->i_width[0] + PADH*2;
2096                             int height = h->fenc->i_lines[0] + i_padv*2;
2097                             x264_weight_scale_plane( h, dst, stride, src, stride, width, height, &h->sh.weight[j][0] );
2098                             h->fenc->i_lines_weighted = height;
2099                         }
2100                     }
2101                 }
2102             }
2103         }
2104     }
2105
2106     if( weightplane[1] )
2107         for( int i = 0; i < h->i_ref[0]; i++ )
2108         {
2109             if( h->sh.weight[i][1].weightfn && !h->sh.weight[i][2].weightfn )
2110             {
2111                 h->sh.weight[i][2].i_scale = 1 << h->sh.weight[0][1].i_denom;
2112                 h->sh.weight[i][2].i_offset = 0;
2113             }
2114             else if( h->sh.weight[i][2].weightfn && !h->sh.weight[i][1].weightfn )
2115             {
2116                 h->sh.weight[i][1].i_scale = 1 << h->sh.weight[0][1].i_denom;
2117                 h->sh.weight[i][1].i_offset = 0;
2118             }
2119         }
2120
2121     if( !weightplane[0] )
2122         h->sh.weight[0][0].i_denom = 0;
2123     if( !weightplane[1] )
2124         h->sh.weight[0][1].i_denom = 0;
2125     h->sh.weight[0][2].i_denom = h->sh.weight[0][1].i_denom;
2126 }
2127
2128 static inline int x264_reference_distance( x264_t *h, x264_frame_t *frame )
2129 {
2130     if( h->param.i_frame_packing == 5 )
2131         return abs((h->fenc->i_frame&~1) - (frame->i_frame&~1)) +
2132                   ((h->fenc->i_frame&1) != (frame->i_frame&1));
2133     else
2134         return abs(h->fenc->i_frame - frame->i_frame);
2135 }
2136
2137 static inline void x264_reference_build_list( x264_t *h, int i_poc )
2138 {
2139     int b_ok;
2140
2141     /* build ref list 0/1 */
2142     h->mb.pic.i_fref[0] = h->i_ref[0] = 0;
2143     h->mb.pic.i_fref[1] = h->i_ref[1] = 0;
2144     if( h->sh.i_type == SLICE_TYPE_I )
2145         return;
2146
2147     for( int i = 0; h->frames.reference[i]; i++ )
2148     {
2149         if( h->frames.reference[i]->b_corrupt )
2150             continue;
2151         if( h->frames.reference[i]->i_poc < i_poc )
2152             h->fref[0][h->i_ref[0]++] = h->frames.reference[i];
2153         else if( h->frames.reference[i]->i_poc > i_poc )
2154             h->fref[1][h->i_ref[1]++] = h->frames.reference[i];
2155     }
2156
2157     /* Order reference lists by distance from the current frame. */
2158     for( int list = 0; list < 2; list++ )
2159     {
2160         h->fref_nearest[list] = h->fref[list][0];
2161         do
2162         {
2163             b_ok = 1;
2164             for( int i = 0; i < h->i_ref[list] - 1; i++ )
2165             {
2166                 if( list ? h->fref[list][i+1]->i_poc < h->fref_nearest[list]->i_poc
2167                          : h->fref[list][i+1]->i_poc > h->fref_nearest[list]->i_poc )
2168                     h->fref_nearest[list] = h->fref[list][i+1];
2169                 if( x264_reference_distance( h, h->fref[list][i] ) > x264_reference_distance( h, h->fref[list][i+1] ) )
2170                 {
2171                     XCHG( x264_frame_t*, h->fref[list][i], h->fref[list][i+1] );
2172                     b_ok = 0;
2173                     break;
2174                 }
2175             }
2176         } while( !b_ok );
2177     }
2178
2179     if( h->sh.i_mmco_remove_from_end )
2180         for( int i = h->i_ref[0]-1; i >= h->i_ref[0] - h->sh.i_mmco_remove_from_end; i-- )
2181         {
2182             int diff = h->i_frame_num - h->fref[0][i]->i_frame_num;
2183             h->sh.mmco[h->sh.i_mmco_command_count].i_poc = h->fref[0][i]->i_poc;
2184             h->sh.mmco[h->sh.i_mmco_command_count++].i_difference_of_pic_nums = diff;
2185         }
2186
2187     x264_reference_check_reorder( h );
2188
2189     h->i_ref[1] = X264_MIN( h->i_ref[1], h->frames.i_max_ref1 );
2190     h->i_ref[0] = X264_MIN( h->i_ref[0], h->frames.i_max_ref0 );
2191     h->i_ref[0] = X264_MIN( h->i_ref[0], h->param.i_frame_reference ); // if reconfig() has lowered the limit
2192
2193     /* For Blu-ray compliance, don't reference frames outside of the minigop. */
2194     if( IS_X264_TYPE_B( h->fenc->i_type ) && h->param.b_bluray_compat )
2195         h->i_ref[0] = X264_MIN( h->i_ref[0], IS_X264_TYPE_B( h->fref[0][0]->i_type ) + 1 );
2196
2197     /* add duplicates */
2198     if( h->fenc->i_type == X264_TYPE_P )
2199     {
2200         int idx = -1;
2201         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
2202         {
2203             x264_weight_t w[3];
2204             w[1].weightfn = w[2].weightfn = NULL;
2205             if( h->param.rc.b_stat_read )
2206                 x264_ratecontrol_set_weights( h, h->fenc );
2207
2208             if( !h->fenc->weight[0][0].weightfn )
2209             {
2210                 h->fenc->weight[0][0].i_denom = 0;
2211                 SET_WEIGHT( w[0], 1, 1, 0, -1 );
2212                 idx = x264_weighted_reference_duplicate( h, 0, w );
2213             }
2214             else
2215             {
2216                 if( h->fenc->weight[0][0].i_scale == 1<<h->fenc->weight[0][0].i_denom )
2217                 {
2218                     SET_WEIGHT( h->fenc->weight[0][0], 1, 1, 0, h->fenc->weight[0][0].i_offset );
2219                 }
2220                 x264_weighted_reference_duplicate( h, 0, x264_weight_none );
2221                 if( h->fenc->weight[0][0].i_offset > -128 )
2222                 {
2223                     w[0] = h->fenc->weight[0][0];
2224                     w[0].i_offset--;
2225                     h->mc.weight_cache( h, &w[0] );
2226                     idx = x264_weighted_reference_duplicate( h, 0, w );
2227                 }
2228             }
2229         }
2230         h->mb.ref_blind_dupe = idx;
2231     }
2232
2233     assert( h->i_ref[0] + h->i_ref[1] <= X264_REF_MAX );
2234     h->mb.pic.i_fref[0] = h->i_ref[0];
2235     h->mb.pic.i_fref[1] = h->i_ref[1];
2236 }
2237
2238 static void x264_fdec_filter_row( x264_t *h, int mb_y, int pass )
2239 {
2240     /* mb_y is the mb to be encoded next, not the mb to be filtered here */
2241     int b_hpel = h->fdec->b_kept_as_ref;
2242     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2243     int b_end = mb_y == h->i_threadslice_end;
2244     int b_measure_quality = 1;
2245     int min_y = mb_y - (1 << SLICE_MBAFF);
2246     int b_start = min_y == h->i_threadslice_start;
2247     /* Even in interlaced mode, deblocking never modifies more than 4 pixels
2248      * above each MB, as bS=4 doesn't happen for the top of interlaced mbpairs. */
2249     int minpix_y = min_y*16 - 4 * !b_start;
2250     int maxpix_y = mb_y*16 - 4 * !b_end;
2251     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2252     if( h->param.b_sliced_threads )
2253     {
2254         switch( pass )
2255         {
2256             /* During encode: only do deblock if asked for */
2257             default:
2258             case 0:
2259                 b_deblock &= h->param.b_full_recon;
2260                 b_hpel = 0;
2261                 break;
2262             /* During post-encode pass: do deblock if not done yet, do hpel for all
2263              * rows except those between slices. */
2264             case 1:
2265                 b_deblock &= !h->param.b_full_recon;
2266                 b_hpel &= !(b_start && min_y > 0);
2267                 b_measure_quality = 0;
2268                 break;
2269             /* Final pass: do the rows between slices in sequence. */
2270             case 2:
2271                 b_deblock = 0;
2272                 b_measure_quality = 0;
2273                 break;
2274         }
2275     }
2276     if( mb_y & SLICE_MBAFF )
2277         return;
2278     if( min_y < h->i_threadslice_start )
2279         return;
2280
2281     if( b_deblock )
2282         for( int y = min_y; y < mb_y; y += (1 << SLICE_MBAFF) )
2283             x264_frame_deblock_row( h, y );
2284
2285     /* FIXME: Prediction requires different borders for interlaced/progressive mc,
2286      * but the actual image data is equivalent. For now, maintain this
2287      * consistency by copying deblocked pixels between planes. */
2288     if( PARAM_INTERLACED && (!h->param.b_sliced_threads || pass == 1) )
2289         for( int p = 0; p < h->fdec->i_plane; p++ )
2290             for( int i = minpix_y>>(CHROMA_V_SHIFT && p); i < maxpix_y>>(CHROMA_V_SHIFT && p); i++ )
2291                 memcpy( h->fdec->plane_fld[p] + i*h->fdec->i_stride[p],
2292                         h->fdec->plane[p]     + i*h->fdec->i_stride[p],
2293                         h->mb.i_mb_width*16*sizeof(pixel) );
2294
2295     if( h->fdec->b_kept_as_ref && (!h->param.b_sliced_threads || pass == 1) )
2296         x264_frame_expand_border( h, h->fdec, min_y );
2297     if( b_hpel )
2298     {
2299         int end = mb_y == h->mb.i_mb_height;
2300         /* Can't do hpel until the previous slice is done encoding. */
2301         if( h->param.analyse.i_subpel_refine )
2302         {
2303             x264_frame_filter( h, h->fdec, min_y, end );
2304             x264_frame_expand_border_filtered( h, h->fdec, min_y, end );
2305         }
2306     }
2307
2308     if( SLICE_MBAFF && pass == 0 )
2309         for( int i = 0; i < 3; i++ )
2310         {
2311             XCHG( pixel *, h->intra_border_backup[0][i], h->intra_border_backup[3][i] );
2312             XCHG( pixel *, h->intra_border_backup[1][i], h->intra_border_backup[4][i] );
2313         }
2314
2315     if( h->i_thread_frames > 1 && h->fdec->b_kept_as_ref )
2316         x264_frame_cond_broadcast( h->fdec, mb_y*16 + (b_end ? 10000 : -(X264_THREAD_HEIGHT << SLICE_MBAFF)) );
2317
2318     if( b_measure_quality )
2319     {
2320         maxpix_y = X264_MIN( maxpix_y, h->param.i_height );
2321         if( h->param.analyse.b_psnr )
2322         {
2323             for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
2324                 h->stat.frame.i_ssd[p] += x264_pixel_ssd_wxh( &h->pixf,
2325                     h->fdec->plane[p] + minpix_y * h->fdec->i_stride[p], h->fdec->i_stride[p],
2326                     h->fenc->plane[p] + minpix_y * h->fenc->i_stride[p], h->fenc->i_stride[p],
2327                     h->param.i_width, maxpix_y-minpix_y );
2328             if( !CHROMA444 )
2329             {
2330                 uint64_t ssd_u, ssd_v;
2331                 int v_shift = CHROMA_V_SHIFT;
2332                 x264_pixel_ssd_nv12( &h->pixf,
2333                     h->fdec->plane[1] + (minpix_y>>v_shift) * h->fdec->i_stride[1], h->fdec->i_stride[1],
2334                     h->fenc->plane[1] + (minpix_y>>v_shift) * h->fenc->i_stride[1], h->fenc->i_stride[1],
2335                     h->param.i_width>>1, (maxpix_y-minpix_y)>>v_shift, &ssd_u, &ssd_v );
2336                 h->stat.frame.i_ssd[1] += ssd_u;
2337                 h->stat.frame.i_ssd[2] += ssd_v;
2338             }
2339         }
2340
2341         if( h->param.analyse.b_ssim )
2342         {
2343             int ssim_cnt;
2344             x264_emms();
2345             /* offset by 2 pixels to avoid alignment of ssim blocks with dct blocks,
2346              * and overlap by 4 */
2347             minpix_y += b_start ? 2 : -6;
2348             h->stat.frame.f_ssim +=
2349                 x264_pixel_ssim_wxh( &h->pixf,
2350                     h->fdec->plane[0] + 2+minpix_y*h->fdec->i_stride[0], h->fdec->i_stride[0],
2351                     h->fenc->plane[0] + 2+minpix_y*h->fenc->i_stride[0], h->fenc->i_stride[0],
2352                     h->param.i_width-2, maxpix_y-minpix_y, h->scratch_buffer, &ssim_cnt );
2353             h->stat.frame.i_ssim_cnt += ssim_cnt;
2354         }
2355     }
2356 }
2357
2358 static inline int x264_reference_update( x264_t *h )
2359 {
2360     if( !h->fdec->b_kept_as_ref )
2361     {
2362         if( h->i_thread_frames > 1 )
2363         {
2364             x264_frame_push_unused( h, h->fdec );
2365             h->fdec = x264_frame_pop_unused( h, 1 );
2366             if( !h->fdec )
2367                 return -1;
2368         }
2369         return 0;
2370     }
2371
2372     /* apply mmco from previous frame. */
2373     for( int i = 0; i < h->sh.i_mmco_command_count; i++ )
2374         for( int j = 0; h->frames.reference[j]; j++ )
2375             if( h->frames.reference[j]->i_poc == h->sh.mmco[i].i_poc )
2376                 x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[j] ) );
2377
2378     /* move frame in the buffer */
2379     x264_frame_push( h->frames.reference, h->fdec );
2380     if( h->frames.reference[h->sps->i_num_ref_frames] )
2381         x264_frame_push_unused( h, x264_frame_shift( h->frames.reference ) );
2382     h->fdec = x264_frame_pop_unused( h, 1 );
2383     if( !h->fdec )
2384         return -1;
2385     return 0;
2386 }
2387
2388 static inline void x264_reference_reset( x264_t *h )
2389 {
2390     while( h->frames.reference[0] )
2391         x264_frame_push_unused( h, x264_frame_pop( h->frames.reference ) );
2392     h->fdec->i_poc =
2393     h->fenc->i_poc = 0;
2394 }
2395
2396 static inline void x264_reference_hierarchy_reset( x264_t *h )
2397 {
2398     int ref;
2399     int b_hasdelayframe = 0;
2400
2401     /* look for delay frames -- chain must only contain frames that are disposable */
2402     for( int i = 0; h->frames.current[i] && IS_DISPOSABLE( h->frames.current[i]->i_type ); i++ )
2403         b_hasdelayframe |= h->frames.current[i]->i_coded
2404                         != h->frames.current[i]->i_frame + h->sps->vui.i_num_reorder_frames;
2405
2406     /* This function must handle b-pyramid and clear frames for open-gop */
2407     if( h->param.i_bframe_pyramid != X264_B_PYRAMID_STRICT && !b_hasdelayframe && h->frames.i_poc_last_open_gop == -1 )
2408         return;
2409
2410     /* Remove last BREF. There will never be old BREFs in the
2411      * dpb during a BREF decode when pyramid == STRICT */
2412     for( ref = 0; h->frames.reference[ref]; ref++ )
2413     {
2414         if( ( h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT
2415             && h->frames.reference[ref]->i_type == X264_TYPE_BREF )
2416             || ( h->frames.reference[ref]->i_poc < h->frames.i_poc_last_open_gop
2417             && h->sh.i_type != SLICE_TYPE_B ) )
2418         {
2419             int diff = h->i_frame_num - h->frames.reference[ref]->i_frame_num;
2420             h->sh.mmco[h->sh.i_mmco_command_count].i_difference_of_pic_nums = diff;
2421             h->sh.mmco[h->sh.i_mmco_command_count++].i_poc = h->frames.reference[ref]->i_poc;
2422             x264_frame_push_unused( h, x264_frame_shift( &h->frames.reference[ref] ) );
2423             h->b_ref_reorder[0] = 1;
2424             ref--;
2425         }
2426     }
2427
2428     /* Prepare room in the dpb for the delayed display time of the later b-frame's */
2429     if( h->param.i_bframe_pyramid )
2430         h->sh.i_mmco_remove_from_end = X264_MAX( ref + 2 - h->frames.i_max_dpb, 0 );
2431 }
2432
2433 static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
2434 {
2435     /* ------------------------ Create slice header  ----------------------- */
2436     if( i_nal_type == NAL_SLICE_IDR )
2437     {
2438         x264_slice_header_init( h, &h->sh, h->sps, h->pps, h->i_idr_pic_id, h->i_frame_num, i_global_qp );
2439
2440         /* alternate id */
2441         h->i_idr_pic_id ^= 1;
2442     }
2443     else
2444     {
2445         x264_slice_header_init( h, &h->sh, h->sps, h->pps, -1, h->i_frame_num, i_global_qp );
2446
2447         h->sh.i_num_ref_idx_l0_active = h->i_ref[0] <= 0 ? 1 : h->i_ref[0];
2448         h->sh.i_num_ref_idx_l1_active = h->i_ref[1] <= 0 ? 1 : h->i_ref[1];
2449         if( h->sh.i_num_ref_idx_l0_active != h->pps->i_num_ref_idx_l0_default_active ||
2450             (h->sh.i_type == SLICE_TYPE_B && h->sh.i_num_ref_idx_l1_active != h->pps->i_num_ref_idx_l1_default_active) )
2451         {
2452             h->sh.b_num_ref_idx_override = 1;
2453         }
2454     }
2455
2456     if( h->fenc->i_type == X264_TYPE_BREF && h->param.b_bluray_compat && h->sh.i_mmco_command_count )
2457     {
2458         h->b_sh_backup = 1;
2459         h->sh_backup = h->sh;
2460     }
2461
2462     h->fdec->i_frame_num = h->sh.i_frame_num;
2463
2464     if( h->sps->i_poc_type == 0 )
2465     {
2466         h->sh.i_poc = h->fdec->i_poc;
2467         if( PARAM_INTERLACED )
2468         {
2469             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
2470             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
2471         }
2472         else
2473             h->sh.i_delta_poc_bottom = 0;
2474         h->fdec->i_delta_poc[0] = h->sh.i_delta_poc_bottom == -1;
2475         h->fdec->i_delta_poc[1] = h->sh.i_delta_poc_bottom ==  1;
2476     }
2477     else
2478     {
2479         /* Nothing to do ? */
2480     }
2481
2482     x264_macroblock_slice_init( h );
2483 }
2484
2485 typedef struct
2486 {
2487     int skip;
2488     uint8_t cabac_prevbyte;
2489     bs_t bs;
2490     x264_cabac_t cabac;
2491     x264_frame_stat_t stat;
2492     int last_qp;
2493     int last_dqp;
2494     int field_decoding_flag;
2495 } x264_bs_bak_t;
2496
2497 static ALWAYS_INLINE void x264_bitstream_backup( x264_t *h, x264_bs_bak_t *bak, int i_skip, int full )
2498 {
2499     if( full )
2500     {
2501         bak->stat = h->stat.frame;
2502         bak->last_qp = h->mb.i_last_qp;
2503         bak->last_dqp = h->mb.i_last_dqp;
2504         bak->field_decoding_flag = h->mb.field_decoding_flag;
2505     }
2506     else
2507     {
2508         bak->stat.i_mv_bits = h->stat.frame.i_mv_bits;
2509         bak->stat.i_tex_bits = h->stat.frame.i_tex_bits;
2510     }
2511     /* In the per-MB backup, we don't need the contexts because flushing the CABAC
2512      * encoder has no context dependency and in this case, a slice is ended (and
2513      * thus the content of all contexts are thrown away). */
2514     if( h->param.b_cabac )
2515     {
2516         if( full )
2517             memcpy( &bak->cabac, &h->cabac, sizeof(x264_cabac_t) );
2518         else
2519             memcpy( &bak->cabac, &h->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2520         /* x264's CABAC writer modifies the previous byte during carry, so it has to be
2521          * backed up. */
2522         bak->cabac_prevbyte = h->cabac.p[-1];
2523     }
2524     else
2525     {
2526         bak->bs = h->out.bs;
2527         bak->skip = i_skip;
2528     }
2529 }
2530
2531 static ALWAYS_INLINE void x264_bitstream_restore( x264_t *h, x264_bs_bak_t *bak, int *skip, int full )
2532 {
2533     if( full )
2534     {
2535         h->stat.frame = bak->stat;
2536         h->mb.i_last_qp = bak->last_qp;
2537         h->mb.i_last_dqp = bak->last_dqp;
2538         h->mb.field_decoding_flag = bak->field_decoding_flag;
2539     }
2540     else
2541     {
2542         h->stat.frame.i_mv_bits = bak->stat.i_mv_bits;
2543         h->stat.frame.i_tex_bits = bak->stat.i_tex_bits;
2544     }
2545     if( h->param.b_cabac )
2546     {
2547         if( full )
2548             memcpy( &h->cabac, &bak->cabac, sizeof(x264_cabac_t) );
2549         else
2550             memcpy( &h->cabac, &bak->cabac, offsetof(x264_cabac_t, f8_bits_encoded) );
2551         h->cabac.p[-1] = bak->cabac_prevbyte;
2552     }
2553     else
2554     {
2555         h->out.bs = bak->bs;
2556         *skip = bak->skip;
2557     }
2558 }
2559
2560 static intptr_t x264_slice_write( x264_t *h )
2561 {
2562     int i_skip;
2563     int mb_xy, i_mb_x, i_mb_y;
2564     /* NALUs other than the first use a 3-byte startcode.
2565      * Add one extra byte for the rbsp, and one more for the final CABAC putbyte.
2566      * Then add an extra 5 bytes just in case, to account for random NAL escapes and
2567      * other inaccuracies. */
2568     int overhead_guess = (NALU_OVERHEAD - (h->param.b_annexb && h->out.i_nal)) + 1 + h->param.b_cabac + 5;
2569     int slice_max_size = h->param.i_slice_max_size > 0 ? (h->param.i_slice_max_size-overhead_guess)*8 : 0;
2570     int back_up_bitstream_cavlc = !h->param.b_cabac && h->sps->i_profile_idc < PROFILE_HIGH;
2571     int back_up_bitstream = slice_max_size || back_up_bitstream_cavlc;
2572     int starting_bits = bs_pos(&h->out.bs);
2573     int b_deblock = h->sh.i_disable_deblocking_filter_idc != 1;
2574     int b_hpel = h->fdec->b_kept_as_ref;
2575     int orig_last_mb = h->sh.i_last_mb;
2576     int thread_last_mb = h->i_threadslice_end * h->mb.i_mb_width - 1;
2577     uint8_t *last_emu_check;
2578 #define BS_BAK_SLICE_MAX_SIZE 0
2579 #define BS_BAK_CAVLC_OVERFLOW 1
2580 #define BS_BAK_SLICE_MIN_MBS  2
2581 #define BS_BAK_ROW_VBV        3
2582     x264_bs_bak_t bs_bak[4];
2583     b_deblock &= b_hpel || h->param.b_full_recon || h->param.psz_dump_yuv;
2584     bs_realign( &h->out.bs );
2585
2586     /* Slice */
2587     x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
2588     h->out.nal[h->out.i_nal].i_first_mb = h->sh.i_first_mb;
2589
2590     /* Slice header */
2591     x264_macroblock_thread_init( h );
2592
2593     /* Set the QP equal to the first QP in the slice for more accurate CABAC initialization. */
2594     h->mb.i_mb_xy = h->sh.i_first_mb;
2595     h->sh.i_qp = x264_ratecontrol_mb_qp( h );
2596     h->sh.i_qp = SPEC_QP( h->sh.i_qp );
2597     h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
2598
2599     x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
2600     if( h->param.b_cabac )
2601     {
2602         /* alignment needed */
2603         bs_align_1( &h->out.bs );
2604
2605         /* init cabac */
2606         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 );
2607         x264_cabac_encode_init ( &h->cabac, h->out.bs.p, h->out.bs.p_end );
2608         last_emu_check = h->cabac.p;
2609     }
2610     else
2611         last_emu_check = h->out.bs.p;
2612     h->mb.i_last_qp = h->sh.i_qp;
2613     h->mb.i_last_dqp = 0;
2614     h->mb.field_decoding_flag = 0;
2615
2616     i_mb_y = h->sh.i_first_mb / h->mb.i_mb_width;
2617     i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
2618     i_skip = 0;
2619
2620     while( 1 )
2621     {
2622         mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width;
2623         int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2624
2625         if( i_mb_x == 0 )
2626         {
2627             if( x264_bitstream_check_buffer( h ) )
2628                 return -1;
2629             if( !(i_mb_y & SLICE_MBAFF) && h->param.rc.i_vbv_buffer_size )
2630                 x264_bitstream_backup( h, &bs_bak[BS_BAK_ROW_VBV], i_skip, 1 );
2631             if( !h->mb.b_reencode_mb )
2632                 x264_fdec_filter_row( h, i_mb_y, 0 );
2633         }
2634
2635         if( back_up_bitstream )
2636         {
2637             if( back_up_bitstream_cavlc )
2638                 x264_bitstream_backup( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], i_skip, 0 );
2639             if( slice_max_size && !(i_mb_y & SLICE_MBAFF) )
2640             {
2641                 x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], i_skip, 0 );
2642                 if( (thread_last_mb+1-mb_xy) == h->param.i_slice_min_mbs )
2643                     x264_bitstream_backup( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], i_skip, 0 );
2644             }
2645         }
2646
2647         if( PARAM_INTERLACED )
2648         {
2649             if( h->mb.b_adaptive_mbaff )
2650             {
2651                 if( !(i_mb_y&1) )
2652                 {
2653                     /* FIXME: VSAD is fast but fairly poor at choosing the best interlace type. */
2654                     h->mb.b_interlaced = x264_field_vsad( h, i_mb_x, i_mb_y );
2655                     memcpy( &h->zigzagf, MB_INTERLACED ? &h->zigzagf_interlaced : &h->zigzagf_progressive, sizeof(h->zigzagf) );
2656                     if( !MB_INTERLACED && (i_mb_y+2) == h->mb.i_mb_height )
2657                         x264_expand_border_mbpair( h, i_mb_x, i_mb_y );
2658                 }
2659             }
2660             h->mb.field[mb_xy] = MB_INTERLACED;
2661         }
2662
2663         /* load cache */
2664         if( SLICE_MBAFF )
2665             x264_macroblock_cache_load_interlaced( h, i_mb_x, i_mb_y );
2666         else
2667             x264_macroblock_cache_load_progressive( h, i_mb_x, i_mb_y );
2668
2669         x264_macroblock_analyse( h );
2670
2671         /* encode this macroblock -> be careful it can change the mb type to P_SKIP if needed */
2672 reencode:
2673         x264_macroblock_encode( h );
2674
2675         if( h->param.b_cabac )
2676         {
2677             if( mb_xy > h->sh.i_first_mb && !(SLICE_MBAFF && (i_mb_y&1)) )
2678                 x264_cabac_encode_terminal( &h->cabac );
2679
2680             if( IS_SKIP( h->mb.i_type ) )
2681                 x264_cabac_mb_skip( h, 1 );
2682             else
2683             {
2684                 if( h->sh.i_type != SLICE_TYPE_I )
2685                     x264_cabac_mb_skip( h, 0 );
2686                 x264_macroblock_write_cabac( h, &h->cabac );
2687             }
2688         }
2689         else
2690         {
2691             if( IS_SKIP( h->mb.i_type ) )
2692                 i_skip++;
2693             else
2694             {
2695                 if( h->sh.i_type != SLICE_TYPE_I )
2696                 {
2697                     bs_write_ue( &h->out.bs, i_skip );  /* skip run */
2698                     i_skip = 0;
2699                 }
2700                 x264_macroblock_write_cavlc( h );
2701                 /* If there was a CAVLC level code overflow, try again at a higher QP. */
2702                 if( h->mb.b_overflow )
2703                 {
2704                     h->mb.i_chroma_qp = h->chroma_qp_table[++h->mb.i_qp];
2705                     h->mb.i_skip_intra = 0;
2706                     h->mb.b_skip_mc = 0;
2707                     h->mb.b_overflow = 0;
2708                     x264_bitstream_restore( h, &bs_bak[BS_BAK_CAVLC_OVERFLOW], &i_skip, 0 );
2709                     goto reencode;
2710                 }
2711             }
2712         }
2713
2714         int total_bits = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
2715         int mb_size = total_bits - mb_spos;
2716
2717         if( slice_max_size && (!SLICE_MBAFF || (i_mb_y&1)) )
2718         {
2719             /* Count the skip run, just in case. */
2720             if( !h->param.b_cabac )
2721                 total_bits += bs_size_ue_big( i_skip );
2722             /* Check for escape bytes. */
2723             uint8_t *end = h->param.b_cabac ? h->cabac.p : h->out.bs.p;
2724             for( ; last_emu_check < end - 2; last_emu_check++ )
2725                 if( last_emu_check[0] == 0 && last_emu_check[1] == 0 && last_emu_check[2] <= 3 )
2726                 {
2727                     slice_max_size -= 8;
2728                     last_emu_check++;
2729                 }
2730             /* We'll just re-encode this last macroblock if we go over the max slice size. */
2731             if( total_bits - starting_bits > slice_max_size && !h->mb.b_reencode_mb )
2732             {
2733                 if( !x264_frame_new_slice( h, h->fdec ) )
2734                 {
2735                     /* Handle the most obnoxious slice-min-mbs edge case: we need to end the slice
2736                      * because it's gone over the maximum size, but doing so would violate slice-min-mbs.
2737                      * If possible, roll back to the last checkpoint and try again.
2738                      * We could try raising QP, but that would break in the case where a slice spans multiple
2739                      * rows, which the re-encoding infrastructure can't currently handle. */
2740                     if( mb_xy <= thread_last_mb && (thread_last_mb+1-mb_xy) < h->param.i_slice_min_mbs )
2741                     {
2742                         if( thread_last_mb-h->param.i_slice_min_mbs < h->sh.i_first_mb+h->param.i_slice_min_mbs )
2743                         {
2744                             x264_log( h, X264_LOG_WARNING, "slice-max-size violated (frame %d, cause: slice-min-mbs)\n", h->i_frame );
2745                             slice_max_size = 0;
2746                             goto cont;
2747                         }
2748                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MIN_MBS], &i_skip, 0 );
2749                         h->mb.b_reencode_mb = 1;
2750                         h->sh.i_last_mb = thread_last_mb-h->param.i_slice_min_mbs;
2751                         break;
2752                     }
2753                     if( mb_xy-SLICE_MBAFF*h->mb.i_mb_stride != h->sh.i_first_mb )
2754                     {
2755                         x264_bitstream_restore( h, &bs_bak[BS_BAK_SLICE_MAX_SIZE], &i_skip, 0 );
2756                         h->mb.b_reencode_mb = 1;
2757                         if( SLICE_MBAFF )
2758                         {
2759                             // set to bottom of previous mbpair
2760                             if( i_mb_x )
2761                                 h->sh.i_last_mb = mb_xy-1+h->mb.i_mb_stride*(!(i_mb_y&1));
2762                             else
2763                                 h->sh.i_last_mb = (i_mb_y-2+!(i_mb_y&1))*h->mb.i_mb_stride + h->mb.i_mb_width - 1;
2764                         }
2765                         else
2766                             h->sh.i_last_mb = mb_xy-1;
2767                         break;
2768                     }
2769                     else
2770                         h->sh.i_last_mb = mb_xy;
2771                 }
2772                 else
2773                     slice_max_size = 0;
2774             }
2775         }
2776 cont:
2777         h->mb.b_reencode_mb = 0;
2778
2779         /* save cache */
2780         x264_macroblock_cache_save( h );
2781
2782         if( x264_ratecontrol_mb( h, mb_size ) < 0 )
2783         {
2784             x264_bitstream_restore( h, &bs_bak[BS_BAK_ROW_VBV], &i_skip, 1 );
2785             h->mb.b_reencode_mb = 1;
2786             i_mb_x = 0;
2787             i_mb_y = i_mb_y - SLICE_MBAFF;
2788             h->mb.i_mb_prev_xy = i_mb_y * h->mb.i_mb_stride - 1;
2789             h->sh.i_last_mb = orig_last_mb;
2790             continue;
2791         }
2792
2793         /* accumulate mb stats */
2794         h->stat.frame.i_mb_count[h->mb.i_type]++;
2795
2796         int b_intra = IS_INTRA( h->mb.i_type );
2797         int b_skip = IS_SKIP( h->mb.i_type );
2798         if( h->param.i_log_level >= X264_LOG_INFO || h->param.rc.b_stat_write )
2799         {
2800             if( !b_intra && !b_skip && !IS_DIRECT( h->mb.i_type ) )
2801             {
2802                 if( h->mb.i_partition != D_8x8 )
2803                         h->stat.frame.i_mb_partition[h->mb.i_partition] += 4;
2804                     else
2805                         for( int i = 0; i < 4; i++ )
2806                             h->stat.frame.i_mb_partition[h->mb.i_sub_partition[i]] ++;
2807                 if( h->param.i_frame_reference > 1 )
2808                     for( int i_list = 0; i_list <= (h->sh.i_type == SLICE_TYPE_B); i_list++ )
2809                         for( int i = 0; i < 4; i++ )
2810                         {
2811                             int i_ref = h->mb.cache.ref[i_list][ x264_scan8[4*i] ];
2812                             if( i_ref >= 0 )
2813                                 h->stat.frame.i_mb_count_ref[i_list][i_ref] ++;
2814                         }
2815             }
2816         }
2817
2818         if( h->param.i_log_level >= X264_LOG_INFO )
2819         {
2820             if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
2821             {
2822                 if( CHROMA444 )
2823                 {
2824                     for( int i = 0; i < 4; i++ )
2825                         if( h->mb.i_cbp_luma & (1 << i) )
2826                             for( int p = 0; p < 3; p++ )
2827                             {
2828                                 int s8 = i*4+p*16;
2829                                 int nnz8x8 = M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+0] )
2830                                            | M16( &h->mb.cache.non_zero_count[x264_scan8[s8]+8] );
2831                                 h->stat.frame.i_mb_cbp[!b_intra + p*2] += !!nnz8x8;
2832                             }
2833                 }
2834                 else
2835                 {
2836                     int cbpsum = (h->mb.i_cbp_luma&1) + ((h->mb.i_cbp_luma>>1)&1)
2837                                + ((h->mb.i_cbp_luma>>2)&1) + (h->mb.i_cbp_luma>>3);
2838                     h->stat.frame.i_mb_cbp[!b_intra + 0] += cbpsum;
2839                     h->stat.frame.i_mb_cbp[!b_intra + 2] += !!h->mb.i_cbp_chroma;
2840                     h->stat.frame.i_mb_cbp[!b_intra + 4] += h->mb.i_cbp_chroma >> 1;
2841                 }
2842             }
2843             if( h->mb.i_cbp_luma && !b_intra )
2844             {
2845                 h->stat.frame.i_mb_count_8x8dct[0] ++;
2846                 h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
2847             }
2848             if( b_intra && h->mb.i_type != I_PCM )
2849             {
2850                 if( h->mb.i_type == I_16x16 )
2851                     h->stat.frame.i_mb_pred_mode[0][h->mb.i_intra16x16_pred_mode]++;
2852                 else if( h->mb.i_type == I_8x8 )
2853                     for( int i = 0; i < 16; i += 4 )
2854                         h->stat.frame.i_mb_pred_mode[1][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2855                 else //if( h->mb.i_type == I_4x4 )
2856                     for( int i = 0; i < 16; i++ )
2857                         h->stat.frame.i_mb_pred_mode[2][h->mb.cache.intra4x4_pred_mode[x264_scan8[i]]]++;
2858                 h->stat.frame.i_mb_pred_mode[3][x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode]]++;
2859             }
2860             h->stat.frame.i_mb_field[b_intra?0:b_skip?2:1] += MB_INTERLACED;
2861         }
2862
2863         /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
2864         if( b_deblock )
2865             x264_macroblock_deblock_strength( h );
2866
2867         if( mb_xy == h->sh.i_last_mb )
2868             break;
2869
2870         if( SLICE_MBAFF )
2871         {
2872             i_mb_x += i_mb_y & 1;
2873             i_mb_y ^= i_mb_x < h->mb.i_mb_width;
2874         }
2875         else
2876             i_mb_x++;
2877         if( i_mb_x == h->mb.i_mb_width )
2878         {
2879             i_mb_y++;
2880             i_mb_x = 0;
2881         }
2882     }
2883     if( h->sh.i_last_mb < h->sh.i_first_mb )
2884         return 0;
2885
2886     h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
2887
2888     if( h->param.b_cabac )
2889     {
2890         x264_cabac_encode_flush( h, &h->cabac );
2891         h->out.bs.p = h->cabac.p;
2892     }
2893     else
2894     {
2895         if( i_skip > 0 )
2896             bs_write_ue( &h->out.bs, i_skip );  /* last skip run */
2897         /* rbsp_slice_trailing_bits */
2898         bs_rbsp_trailing( &h->out.bs );
2899         bs_flush( &h->out.bs );
2900     }
2901     if( x264_nal_end( h ) )
2902         return -1;
2903
2904     if( h->sh.i_last_mb == (h->i_threadslice_end * h->mb.i_mb_width - 1) )
2905     {
2906         h->stat.frame.i_misc_bits = bs_pos( &h->out.bs )
2907                                   + (h->out.i_nal*NALU_OVERHEAD * 8)
2908                                   - h->stat.frame.i_tex_bits
2909                                   - h->stat.frame.i_mv_bits;
2910         x264_fdec_filter_row( h, h->i_threadslice_end, 0 );
2911
2912         if( h->param.b_sliced_threads )
2913         {
2914             /* Tell the main thread we're done. */
2915             x264_threadslice_cond_broadcast( h, 1 );
2916             /* Do hpel now */
2917             for( int mb_y = h->i_threadslice_start; mb_y <= h->i_threadslice_end; mb_y++ )
2918                 x264_fdec_filter_row( h, mb_y, 1 );
2919             x264_threadslice_cond_broadcast( h, 2 );
2920             /* Do the first row of hpel, now that the previous slice is done */
2921             if( h->i_thread_idx > 0 )
2922             {
2923                 x264_threadslice_cond_wait( h->thread[h->i_thread_idx-1], 2 );
2924                 x264_fdec_filter_row( h, h->i_threadslice_start + (1 << SLICE_MBAFF), 2 );
2925             }
2926         }
2927
2928         /* Free mb info after the last thread's done using it */
2929         if( h->fdec->mb_info_free && (!h->param.b_sliced_threads || h->i_thread_idx == (h->param.i_threads-1)) )
2930         {
2931             h->fdec->mb_info_free( h->fdec->mb_info );
2932             h->fdec->mb_info = NULL;
2933             h->fdec->mb_info_free = NULL;
2934         }
2935     }
2936
2937     return 0;
2938 }
2939
2940 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
2941 {
2942     if( dst == src )
2943         return;
2944
2945     // reference counting
2946     for( x264_frame_t **f = src->frames.reference; *f; f++ )
2947         (*f)->i_reference_count++;
2948     for( x264_frame_t **f = dst->frames.reference; *f; f++ )
2949         x264_frame_push_unused( src, *f );
2950     src->fdec->i_reference_count++;
2951     x264_frame_push_unused( src, dst->fdec );
2952
2953     // copy everything except the per-thread pointers and the constants.
2954     memcpy( &dst->i_frame, &src->i_frame, offsetof(x264_t, mb.base) - offsetof(x264_t, i_frame) );
2955     dst->param = src->param;
2956     dst->stat = src->stat;
2957     dst->pixf = src->pixf;
2958     dst->reconfig = src->reconfig;
2959 }
2960
2961 static void x264_thread_sync_stat( x264_t *dst, x264_t *src )
2962 {
2963     if( dst == src )
2964         return;
2965     memcpy( &dst->stat.i_frame_count, &src->stat.i_frame_count, sizeof(dst->stat) - sizeof(dst->stat.frame) );
2966 }
2967
2968 static void *x264_slices_write( x264_t *h )
2969 {
2970     int i_slice_num = 0;
2971     int last_thread_mb = h->sh.i_last_mb;
2972
2973     /* init stats */
2974     memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
2975     h->mb.b_reencode_mb = 0;
2976     while( h->sh.i_first_mb + SLICE_MBAFF*h->mb.i_mb_stride <= last_thread_mb )
2977     {
2978         h->sh.i_last_mb = last_thread_mb;
2979         if( !i_slice_num || !x264_frame_new_slice( h, h->fdec ) )
2980         {
2981             if( h->param.i_slice_max_mbs )
2982             {
2983                 if( SLICE_MBAFF )
2984                 {
2985                     // convert first to mbaff form, add slice-max-mbs, then convert back to normal form
2986                     int last_mbaff = 2*(h->sh.i_first_mb % h->mb.i_mb_width)
2987                         + h->mb.i_mb_width*(h->sh.i_first_mb / h->mb.i_mb_width)
2988                         + h->param.i_slice_max_mbs - 1;
2989                     int last_x = (last_mbaff % (2*h->mb.i_mb_width))/2;
2990                     int last_y = (last_mbaff / (2*h->mb.i_mb_width))*2 + 1;
2991                     h->sh.i_last_mb = last_x + h->mb.i_mb_stride*last_y;
2992                 }
2993                 else
2994                 {
2995                     h->sh.i_last_mb = h->sh.i_first_mb + h->param.i_slice_max_mbs - 1;
2996                     if( h->sh.i_last_mb < last_thread_mb && last_thread_mb - h->sh.i_last_mb < h->param.i_slice_min_mbs )
2997                         h->sh.i_last_mb = last_thread_mb - h->param.i_slice_min_mbs;
2998                 }
2999                 i_slice_num++;
3000             }
3001             else if( h->param.i_slice_count && !h->param.b_sliced_threads )
3002             {
3003                 int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3004                 int width = h->mb.i_mb_width << PARAM_INTERLACED;
3005                 i_slice_num++;
3006                 h->sh.i_last_mb = (height * i_slice_num + h->param.i_slice_count/2) / h->param.i_slice_count * width - 1;
3007             }
3008         }
3009         h->sh.i_last_mb = X264_MIN( h->sh.i_last_mb, last_thread_mb );
3010         if( x264_stack_align( x264_slice_write, h ) )
3011             goto fail;
3012         h->sh.i_first_mb = h->sh.i_last_mb + 1;
3013         // if i_first_mb is not the last mb in a row then go to the next mb in MBAFF order
3014         if( SLICE_MBAFF && h->sh.i_first_mb % h->mb.i_mb_width )
3015             h->sh.i_first_mb -= h->mb.i_mb_stride;
3016     }
3017
3018     return (void *)0;
3019
3020 fail:
3021     /* Tell other threads we're done, so they wouldn't wait for it */
3022     if( h->param.b_sliced_threads )
3023         x264_threadslice_cond_broadcast( h, 2 );
3024     return (void *)-1;
3025 }
3026
3027 static int x264_threaded_slices_write( x264_t *h )
3028 {
3029     /* set first/last mb and sync contexts */
3030     for( int i = 0; i < h->param.i_threads; i++ )
3031     {
3032         x264_t *t = h->thread[i];
3033         if( i )
3034         {
3035             t->param = h->param;
3036             memcpy( &t->i_frame, &h->i_frame, offsetof(x264_t, rc) - offsetof(x264_t, i_frame) );
3037         }
3038         int height = h->mb.i_mb_height >> PARAM_INTERLACED;
3039         t->i_threadslice_start = ((height *  i    + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3040         t->i_threadslice_end   = ((height * (i+1) + h->param.i_slice_count/2) / h->param.i_threads) << PARAM_INTERLACED;
3041         t->sh.i_first_mb = t->i_threadslice_start * h->mb.i_mb_width;
3042         t->sh.i_last_mb  =   t->i_threadslice_end * h->mb.i_mb_width - 1;
3043     }
3044
3045     x264_stack_align( x264_analyse_weight_frame, h, h->mb.i_mb_height*16 + 16 );
3046
3047     x264_threads_distribute_ratecontrol( h );
3048
3049     /* setup */
3050     for( int i = 0; i < h->param.i_threads; i++ )
3051     {
3052         h->thread[i]->i_thread_idx = i;
3053         h->thread[i]->b_thread_active = 1;
3054         x264_threadslice_cond_broadcast( h->thread[i], 0 );
3055     }
3056     /* dispatch */
3057     for( int i = 0; i < h->param.i_threads; i++ )
3058         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h->thread[i] );
3059     /* wait */
3060     for( int i = 0; i < h->param.i_threads; i++ )
3061         x264_threadslice_cond_wait( h->thread[i], 1 );
3062
3063     x264_threads_merge_ratecontrol( h );
3064
3065     for( int i = 1; i < h->param.i_threads; i++ )
3066     {
3067         x264_t *t = h->thread[i];
3068         for( int j = 0; j < t->out.i_nal; j++ )
3069         {
3070             h->out.nal[h->out.i_nal] = t->out.nal[j];
3071             h->out.i_nal++;
3072             x264_nal_check_buffer( h );
3073         }
3074         /* All entries in stat.frame are ints except for ssd/ssim. */
3075         for( int j = 0; j < (offsetof(x264_t,stat.frame.i_ssd) - offsetof(x264_t,stat.frame.i_mv_bits)) / sizeof(int); j++ )
3076             ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
3077         for( int j = 0; j < 3; j++ )
3078             h->stat.frame.i_ssd[j] += t->stat.frame.i_ssd[j];
3079         h->stat.frame.f_ssim += t->stat.frame.f_ssim;
3080         h->stat.frame.i_ssim_cnt += t->stat.frame.i_ssim_cnt;
3081     }
3082
3083     return 0;
3084 }
3085
3086 void x264_encoder_intra_refresh( x264_t *h )
3087 {
3088     h = h->thread[h->i_thread_phase];
3089     h->b_queued_intra_refresh = 1;
3090 }
3091
3092 int x264_encoder_invalidate_reference( x264_t *h, int64_t pts )
3093 {
3094     if( h->param.i_bframe )
3095     {
3096         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with B-frames enabled\n" );
3097         return -1;
3098     }
3099     if( h->param.b_intra_refresh )
3100     {
3101         x264_log( h, X264_LOG_ERROR, "x264_encoder_invalidate_reference is not supported with intra refresh enabled\n" );
3102         return -1;
3103     }
3104     h = h->thread[h->i_thread_phase];
3105     if( pts >= h->i_last_idr_pts )
3106     {
3107         for( int i = 0; h->frames.reference[i]; i++ )
3108             if( pts <= h->frames.reference[i]->i_pts )
3109                 h->frames.reference[i]->b_corrupt = 1;
3110         if( pts <= h->fdec->i_pts )
3111             h->fdec->b_corrupt = 1;
3112     }
3113     return 0;
3114 }
3115
3116 /****************************************************************************
3117  * x264_encoder_encode:
3118  *  XXX: i_poc   : is the poc of the current given picture
3119  *       i_frame : is the number of the frame being coded
3120  *  ex:  type frame poc
3121  *       I      0   2*0
3122  *       P      1   2*3
3123  *       B      2   2*1
3124  *       B      3   2*2
3125  *       P      4   2*6
3126  *       B      5   2*4
3127  *       B      6   2*5
3128  ****************************************************************************/
3129 int     x264_encoder_encode( x264_t *h,
3130                              x264_nal_t **pp_nal, int *pi_nal,
3131                              x264_picture_t *pic_in,
3132                              x264_picture_t *pic_out )
3133 {
3134     x264_t *thread_current, *thread_prev, *thread_oldest;
3135     int i_nal_type, i_nal_ref_idc, i_global_qp;
3136     int overhead = NALU_OVERHEAD;
3137
3138 #if HAVE_OPENCL
3139     if( h->opencl.b_fatal_error )
3140         return -1;
3141 #endif
3142
3143     if( h->i_thread_frames > 1 )
3144     {
3145         thread_prev    = h->thread[ h->i_thread_phase ];
3146         h->i_thread_phase = (h->i_thread_phase + 1) % h->i_thread_frames;
3147         thread_current = h->thread[ h->i_thread_phase ];
3148         thread_oldest  = h->thread[ (h->i_thread_phase + 1) % h->i_thread_frames ];
3149         x264_thread_sync_context( thread_current, thread_prev );
3150         x264_thread_sync_ratecontrol( thread_current, thread_prev, thread_oldest );
3151         h = thread_current;
3152     }
3153     else
3154     {
3155         thread_current =
3156         thread_oldest  = h;
3157     }
3158     h->i_cpb_delay_pir_offset = h->i_cpb_delay_pir_offset_next;
3159
3160     /* no data out */
3161     *pi_nal = 0;
3162     *pp_nal = NULL;
3163
3164     /* ------------------- Setup new frame from picture -------------------- */
3165     if( pic_in != NULL )
3166     {
3167         /* 1: Copy the picture to a frame and move it to a buffer */
3168         x264_frame_t *fenc = x264_frame_pop_unused( h, 0 );
3169         if( !fenc )
3170             return -1;
3171
3172         if( x264_frame_copy_picture( h, fenc, pic_in ) < 0 )
3173             return -1;
3174
3175         if( h->param.i_width != 16 * h->mb.i_mb_width ||
3176             h->param.i_height != 16 * h->mb.i_mb_height )
3177             x264_frame_expand_border_mod16( h, fenc );
3178
3179         fenc->i_frame = h->frames.i_input++;
3180
3181         if( fenc->i_frame == 0 )
3182             h->frames.i_first_pts = fenc->i_pts;
3183         if( h->frames.i_bframe_delay && fenc->i_frame == h->frames.i_bframe_delay )
3184             h->frames.i_bframe_delay_time = fenc->i_pts - h->frames.i_first_pts;
3185
3186         if( h->param.b_vfr_input && fenc->i_pts <= h->frames.i_largest_pts )
3187             x264_log( h, X264_LOG_WARNING, "non-strictly-monotonic PTS\n" );
3188
3189         h->frames.i_second_largest_pts = h->frames.i_largest_pts;
3190         h->frames.i_largest_pts = fenc->i_pts;
3191
3192         if( (fenc->i_pic_struct < PIC_STRUCT_AUTO) || (fenc->i_pic_struct > PIC_STRUCT_TRIPLE) )
3193             fenc->i_pic_struct = PIC_STRUCT_AUTO;
3194
3195         if( fenc->i_pic_struct == PIC_STRUCT_AUTO )
3196         {
3197 #if HAVE_INTERLACED
3198             int b_interlaced = fenc->param ? fenc->param->b_interlaced : h->param.b_interlaced;
3199 #else
3200             int b_interlaced = 0;
3201 #endif
3202             if( b_interlaced )
3203             {
3204                 int b_tff = fenc->param ? fenc->param->b_tff : h->param.b_tff;
3205                 fenc->i_pic_struct = b_tff ? PIC_STRUCT_TOP_BOTTOM : PIC_STRUCT_BOTTOM_TOP;
3206             }
3207             else
3208                 fenc->i_pic_struct = PIC_STRUCT_PROGRESSIVE;
3209         }
3210
3211         if( h->param.rc.b_mb_tree && h->param.rc.b_stat_read )
3212         {
3213             if( x264_macroblock_tree_read( h, fenc, pic_in->prop.quant_offsets ) )
3214                 return -1;
3215         }
3216         else
3217             x264_stack_align( x264_adaptive_quant_frame, h, fenc, pic_in->prop.quant_offsets );
3218
3219         if( pic_in->prop.quant_offsets_free )
3220             pic_in->prop.quant_offsets_free( pic_in->prop.quant_offsets );
3221
3222         if( h->frames.b_have_lowres )
3223             x264_frame_init_lowres( h, fenc );
3224
3225         /* 2: Place the frame into the queue for its slice type decision */
3226         x264_lookahead_put_frame( h, fenc );
3227
3228         if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
3229         {
3230             /* Nothing yet to encode, waiting for filling of buffers */
3231             pic_out->i_type = X264_TYPE_AUTO;
3232             return 0;
3233         }
3234     }
3235     else
3236     {
3237         /* signal kills for lookahead thread */
3238         x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
3239         h->lookahead->b_exit_thread = 1;
3240         x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
3241         x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
3242     }
3243
3244     h->i_frame++;
3245     /* 3: The picture is analyzed in the lookahead */
3246     if( !h->frames.current[0] )
3247         x264_lookahead_get_frames( h );
3248
3249     if( !h->frames.current[0] && x264_lookahead_is_empty( h ) )
3250         return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3251
3252     /* ------------------- Get frame to be encoded ------------------------- */
3253     /* 4: get picture to encode */
3254     h->fenc = x264_frame_shift( h->frames.current );
3255
3256     /* If applicable, wait for previous frame reconstruction to finish */
3257     if( h->param.b_sliced_threads )
3258         if( x264_threadpool_wait_all( h ) < 0 )
3259             return -1;
3260
3261     if( h->i_frame == h->i_thread_frames - 1 )
3262         h->i_reordered_pts_delay = h->fenc->i_reordered_pts;
3263     if( h->reconfig )
3264     {
3265         x264_encoder_reconfig_apply( h, &h->reconfig_h->param );
3266         h->reconfig = 0;
3267     }
3268     if( h->fenc->param )
3269     {
3270         x264_encoder_reconfig_apply( h, h->fenc->param );
3271         if( h->fenc->param->param_free )
3272         {
3273             h->fenc->param->param_free( h->fenc->param );
3274             h->fenc->param = NULL;
3275         }
3276     }
3277
3278     // ok to call this before encoding any frames, since the initial values of fdec have b_kept_as_ref=0
3279     if( x264_reference_update( h ) )
3280         return -1;
3281     h->fdec->i_lines_completed = -1;
3282
3283     if( !IS_X264_TYPE_I( h->fenc->i_type ) )
3284     {
3285         int valid_refs_left = 0;
3286         for( int i = 0; h->frames.reference[i]; i++ )
3287             if( !h->frames.reference[i]->b_corrupt )
3288                 valid_refs_left++;
3289         /* No valid reference frames left: force an IDR. */
3290         if( !valid_refs_left )
3291         {
3292             h->fenc->b_keyframe = 1;
3293             h->fenc->i_type = X264_TYPE_IDR;
3294         }
3295     }
3296
3297     if( h->fenc->b_keyframe )
3298     {
3299         h->frames.i_last_keyframe = h->fenc->i_frame;
3300         if( h->fenc->i_type == X264_TYPE_IDR )
3301         {
3302             h->i_frame_num = 0;
3303             h->frames.i_last_idr = h->fenc->i_frame;
3304         }
3305     }
3306     h->sh.i_mmco_command_count =
3307     h->sh.i_mmco_remove_from_end = 0;
3308     h->b_ref_reorder[0] =
3309     h->b_ref_reorder[1] = 0;
3310     h->fdec->i_poc =
3311     h->fenc->i_poc = 2 * ( h->fenc->i_frame - X264_MAX( h->frames.i_last_idr, 0 ) );
3312
3313     /* ------------------- Setup frame context ----------------------------- */
3314     /* 5: Init data dependent of frame type */
3315     if( h->fenc->i_type == X264_TYPE_IDR )
3316     {
3317         /* reset ref pictures */
3318         i_nal_type    = NAL_SLICE_IDR;
3319         i_nal_ref_idc = NAL_PRIORITY_HIGHEST;
3320         h->sh.i_type = SLICE_TYPE_I;
3321         x264_reference_reset( h );
3322         h->frames.i_poc_last_open_gop = -1;
3323     }
3324     else if( h->fenc->i_type == X264_TYPE_I )
3325     {
3326         i_nal_type    = NAL_SLICE;
3327         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3328         h->sh.i_type = SLICE_TYPE_I;
3329         x264_reference_hierarchy_reset( h );
3330         if( h->param.b_open_gop )
3331             h->frames.i_poc_last_open_gop = h->fenc->b_keyframe ? h->fenc->i_poc : -1;
3332     }
3333     else if( h->fenc->i_type == X264_TYPE_P )
3334     {
3335         i_nal_type    = NAL_SLICE;
3336         i_nal_ref_idc = NAL_PRIORITY_HIGH; /* Not completely true but for now it is (as all I/P are kept as ref)*/
3337         h->sh.i_type = SLICE_TYPE_P;
3338         x264_reference_hierarchy_reset( h );
3339         h->frames.i_poc_last_open_gop = -1;
3340     }
3341     else if( h->fenc->i_type == X264_TYPE_BREF )
3342     {
3343         i_nal_type    = NAL_SLICE;
3344         i_nal_ref_idc = h->param.i_bframe_pyramid == X264_B_PYRAMID_STRICT ? NAL_PRIORITY_LOW : NAL_PRIORITY_HIGH;
3345         h->sh.i_type = SLICE_TYPE_B;
3346         x264_reference_hierarchy_reset( h );
3347     }
3348     else    /* B frame */
3349     {
3350         i_nal_type    = NAL_SLICE;
3351         i_nal_ref_idc = NAL_PRIORITY_DISPOSABLE;
3352         h->sh.i_type = SLICE_TYPE_B;
3353     }
3354
3355     h->fdec->i_type = h->fenc->i_type;
3356     h->fdec->i_frame = h->fenc->i_frame;
3357     h->fenc->b_kept_as_ref =
3358     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE && h->param.i_keyint_max > 1;
3359
3360     h->fdec->mb_info = h->fenc->mb_info;
3361     h->fdec->mb_info_free = h->fenc->mb_info_free;
3362     h->fenc->mb_info = NULL;
3363     h->fenc->mb_info_free = NULL;
3364
3365     h->fdec->i_pts = h->fenc->i_pts;
3366     if( h->frames.i_bframe_delay )
3367     {
3368         int64_t *prev_reordered_pts = thread_current->frames.i_prev_reordered_pts;
3369         h->fdec->i_dts = h->i_frame > h->frames.i_bframe_delay
3370                        ? prev_reordered_pts[ (h->i_frame - h->frames.i_bframe_delay) % h->frames.i_bframe_delay ]
3371                        : h->fenc->i_reordered_pts - h->frames.i_bframe_delay_time;
3372         prev_reordered_pts[ h->i_frame % h->frames.i_bframe_delay ] = h->fenc->i_reordered_pts;
3373     }
3374     else
3375         h->fdec->i_dts = h->fenc->i_reordered_pts;
3376     if( h->fenc->i_type == X264_TYPE_IDR )
3377         h->i_last_idr_pts = h->fdec->i_pts;
3378
3379     /* ------------------- Init                ----------------------------- */
3380     /* build ref list 0/1 */
3381     x264_reference_build_list( h, h->fdec->i_poc );
3382
3383     /* ---------------------- Write the bitstream -------------------------- */
3384     /* Init bitstream context */
3385     if( h->param.b_sliced_threads )
3386     {
3387         for( int i = 0; i < h->param.i_threads; i++ )
3388         {
3389             bs_init( &h->thread[i]->out.bs, h->thread[i]->out.p_bitstream, h->thread[i]->out.i_bitstream );
3390             h->thread[i]->out.i_nal = 0;
3391         }
3392     }
3393     else
3394     {
3395         bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
3396         h->out.i_nal = 0;
3397     }
3398
3399     if( h->param.b_aud )
3400     {
3401         int pic_type;
3402
3403         if( h->sh.i_type == SLICE_TYPE_I )
3404             pic_type = 0;
3405         else if( h->sh.i_type == SLICE_TYPE_P )
3406             pic_type = 1;
3407         else if( h->sh.i_type == SLICE_TYPE_B )
3408             pic_type = 2;
3409         else
3410             pic_type = 7;
3411
3412         x264_nal_start( h, NAL_AUD, NAL_PRIORITY_DISPOSABLE );
3413         bs_write( &h->out.bs, 3, pic_type );
3414         bs_rbsp_trailing( &h->out.bs );
3415         if( x264_nal_end( h ) )
3416             return -1;
3417         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3418     }
3419
3420     h->i_nal_type = i_nal_type;
3421     h->i_nal_ref_idc = i_nal_ref_idc;
3422
3423     if( h->param.b_intra_refresh )
3424     {
3425         if( IS_X264_TYPE_I( h->fenc->i_type ) )
3426         {
3427             h->fdec->i_frames_since_pir = 0;
3428             h->b_queued_intra_refresh = 0;
3429             /* PIR is currently only supported with ref == 1, so any intra frame effectively refreshes
3430              * the whole frame and counts as an intra refresh. */
3431             h->fdec->f_pir_position = h->mb.i_mb_width;
3432         }
3433         else if( h->fenc->i_type == X264_TYPE_P )
3434         {
3435             int pocdiff = (h->fdec->i_poc - h->fref[0][0]->i_poc)/2;
3436             float increment = X264_MAX( ((float)h->mb.i_mb_width-1) / h->param.i_keyint_max, 1 );
3437             h->fdec->f_pir_position = h->fref[0][0]->f_pir_position;
3438             h->fdec->i_frames_since_pir = h->fref[0][0]->i_frames_since_pir + pocdiff;
3439             if( h->fdec->i_frames_since_pir >= h->param.i_keyint_max ||
3440                 (h->b_queued_intra_refresh && h->fdec->f_pir_position + 0.5 >= h->mb.i_mb_width) )
3441             {
3442                 h->fdec->f_pir_position = 0;
3443                 h->fdec->i_frames_since_pir = 0;
3444                 h->b_queued_intra_refresh = 0;
3445                 h->fenc->b_keyframe = 1;
3446             }
3447             h->fdec->i_pir_start_col = h->fdec->f_pir_position+0.5;
3448             h->fdec->f_pir_position += increment * pocdiff;
3449             h->fdec->i_pir_end_col = h->fdec->f_pir_position+0.5;
3450             /* If our intra refresh has reached the right side of the frame, we're done. */
3451             if( h->fdec->i_pir_end_col >= h->mb.i_mb_width - 1 )
3452             {
3453                 h->fdec->f_pir_position = h->mb.i_mb_width;
3454                 h->fdec->i_pir_end_col = h->mb.i_mb_width - 1;
3455             }
3456         }
3457     }
3458
3459     if( h->fenc->b_keyframe )
3460     {
3461         /* Write SPS and PPS */
3462         if( h->param.b_repeat_headers )
3463         {
3464             /* generate sequence parameters */
3465             x264_nal_start( h, NAL_SPS, NAL_PRIORITY_HIGHEST );
3466             x264_sps_write( &h->out.bs, h->sps );
3467             if( x264_nal_end( h ) )
3468                 return -1;
3469             /* Pad AUD/SPS to 256 bytes like Panasonic */
3470             if( h->param.i_avcintra_class )
3471                 h->out.nal[h->out.i_nal-1].i_padding = 256 - bs_pos( &h->out.bs ) / 8 - 2*NALU_OVERHEAD;
3472             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3473
3474             /* generate picture parameters */
3475             x264_nal_start( h, NAL_PPS, NAL_PRIORITY_HIGHEST );
3476             x264_pps_write( &h->out.bs, h->sps, h->pps );
3477             if( x264_nal_end( h ) )
3478                 return -1;
3479             if( h->param.i_avcintra_class )
3480                 h->out.nal[h->out.i_nal-1].i_padding = 256 - h->out.nal[h->out.i_nal-1].i_payload - NALU_OVERHEAD;
3481             overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + NALU_OVERHEAD;
3482         }
3483
3484         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
3485         if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
3486         {
3487             x264_hrd_fullness( h );
3488             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3489             x264_sei_buffering_period_write( h, &h->out.bs );
3490             if( x264_nal_end( h ) )
3491                return -1;
3492             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3493         }
3494     }
3495
3496     /* write extra sei */
3497     for( int i = 0; i < h->fenc->extra_sei.num_payloads; i++ )
3498     {
3499         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3500         x264_sei_write( &h->out.bs, h->fenc->extra_sei.payloads[i].payload, h->fenc->extra_sei.payloads[i].payload_size,
3501                         h->fenc->extra_sei.payloads[i].payload_type );
3502         if( x264_nal_end( h ) )
3503             return -1;
3504         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3505         if( h->fenc->extra_sei.sei_free )
3506         {
3507             h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads[i].payload );
3508             h->fenc->extra_sei.payloads[i].payload = NULL;
3509         }
3510     }
3511
3512     if( h->fenc->extra_sei.sei_free )
3513     {
3514         h->fenc->extra_sei.sei_free( h->fenc->extra_sei.payloads );
3515         h->fenc->extra_sei.payloads = NULL;
3516         h->fenc->extra_sei.sei_free = NULL;
3517     }
3518
3519     if( h->fenc->b_keyframe )
3520     {
3521         /* Avid's decoder strictly wants two SEIs for AVC-Intra so we can't insert the x264 SEI */
3522         if( h->param.b_repeat_headers && h->fenc->i_frame == 0 && !h->param.i_avcintra_class )
3523         {
3524             /* identify ourself */
3525             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3526             if( x264_sei_version_write( h, &h->out.bs ) )
3527                 return -1;
3528             if( x264_nal_end( h ) )
3529                 return -1;
3530             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3531         }
3532
3533         if( h->fenc->i_type != X264_TYPE_IDR )
3534         {
3535             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;
3536             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3537             x264_sei_recovery_point_write( h, &h->out.bs, time_to_recovery );
3538             if( x264_nal_end( h ) )
3539                 return -1;
3540             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3541         }
3542
3543         if( h->param.i_frame_packing >= 0 )
3544         {
3545             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3546             x264_sei_frame_packing_write( h, &h->out.bs );
3547             if( x264_nal_end( h ) )
3548                 return -1;
3549             overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3550         }
3551     }
3552
3553     /* generate sei pic timing */
3554     if( h->sps->vui.b_pic_struct_present || h->sps->vui.b_nal_hrd_parameters_present )
3555     {
3556         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3557         x264_sei_pic_timing_write( h, &h->out.bs );
3558         if( x264_nal_end( h ) )
3559             return -1;
3560         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3561     }
3562
3563     /* As required by Blu-ray. */
3564     if( !IS_X264_TYPE_B( h->fenc->i_type ) && h->b_sh_backup )
3565     {
3566         h->b_sh_backup = 0;
3567         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3568         x264_sei_dec_ref_pic_marking_write( h, &h->out.bs );
3569         if( x264_nal_end( h ) )
3570             return -1;
3571         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3572     }
3573
3574     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
3575         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
3576
3577     /* Filler space: 10 or 18 SEIs' worth of space, depending on resolution */
3578     if( h->param.i_avcintra_class )
3579     {
3580         /* Write an empty filler NAL to mimic the AUD in the P2 format*/
3581         x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3582         x264_filler_write( h, &h->out.bs, 0 );
3583         if( x264_nal_end( h ) )
3584             return -1;
3585         overhead += h->out.nal[h->out.i_nal-1].i_payload + NALU_OVERHEAD;
3586
3587         /* All lengths are magic lengths that decoders expect to see */
3588         /* "UMID" SEI */
3589         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3590         if( x264_sei_avcintra_umid_write( h, &h->out.bs ) < 0 )
3591             return -1;
3592         if( x264_nal_end( h ) )
3593             return -1;
3594         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
3595
3596         int unpadded_len;
3597         int total_len;
3598         if( h->param.i_height == 1080 )
3599         {
3600             unpadded_len = 5780;
3601             total_len = 17*512;
3602         }
3603         else
3604         {
3605             unpadded_len = 2900;
3606             total_len = 9*512;
3607         }
3608         /* "VANC" SEI */
3609         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3610         if( x264_sei_avcintra_vanc_write( h, &h->out.bs, unpadded_len ) < 0 )
3611             return -1;
3612         if( x264_nal_end( h ) )
3613             return -1;
3614
3615         h->out.nal[h->out.i_nal-1].i_padding = total_len - h->out.nal[h->out.i_nal-1].i_payload - SEI_OVERHEAD;
3616         overhead += h->out.nal[h->out.i_nal-1].i_payload + h->out.nal[h->out.i_nal-1].i_padding + SEI_OVERHEAD;
3617     }
3618
3619     /* Init the rate control */
3620     /* FIXME: Include slice header bit cost. */
3621     x264_ratecontrol_start( h, h->fenc->i_qpplus1, overhead*8 );
3622     i_global_qp = x264_ratecontrol_qp( h );
3623
3624     pic_out->i_qpplus1 =
3625     h->fdec->i_qpplus1 = i_global_qp + 1;
3626
3627     if( h->param.rc.b_stat_read && h->sh.i_type != SLICE_TYPE_I )
3628     {
3629         x264_reference_build_list_optimal( h );
3630         x264_reference_check_reorder( h );
3631     }
3632
3633     if( h->i_ref[0] )
3634         h->fdec->i_poc_l0ref0 = h->fref[0][0]->i_poc;
3635
3636     /* ------------------------ Create slice header  ----------------------- */
3637     x264_slice_init( h, i_nal_type, i_global_qp );
3638
3639     /*------------------------- Weights -------------------------------------*/
3640     if( h->sh.i_type == SLICE_TYPE_B )
3641         x264_macroblock_bipred_init( h );
3642
3643     x264_weighted_pred_init( h );
3644
3645     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
3646         h->i_frame_num++;
3647
3648     /* Write frame */
3649     h->i_threadslice_start = 0;
3650     h->i_threadslice_end = h->mb.i_mb_height;
3651     if( h->i_thread_frames > 1 )
3652     {
3653         x264_threadpool_run( h->threadpool, (void*)x264_slices_write, h );
3654         h->b_thread_active = 1;
3655     }
3656     else if( h->param.b_sliced_threads )
3657     {
3658         if( x264_threaded_slices_write( h ) )
3659             return -1;
3660     }
3661     else
3662         if( (intptr_t)x264_slices_write( h ) )
3663             return -1;
3664
3665     return x264_encoder_frame_end( thread_oldest, thread_current, pp_nal, pi_nal, pic_out );
3666 }
3667
3668 static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
3669                                    x264_nal_t **pp_nal, int *pi_nal,
3670                                    x264_picture_t *pic_out )
3671 {
3672     char psz_message[80];
3673
3674     if( !h->param.b_sliced_threads && h->b_thread_active )
3675     {
3676         h->b_thread_active = 0;
3677         if( (intptr_t)x264_threadpool_wait( h->threadpool, h ) )
3678             return -1;
3679     }
3680     if( !h->out.i_nal )
3681     {
3682         pic_out->i_type = X264_TYPE_AUTO;
3683         return 0;
3684     }
3685
3686     x264_emms();
3687
3688     /* generate buffering period sei and insert it into place */
3689     if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
3690     {
3691         x264_hrd_fullness( h );
3692         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
3693         x264_sei_buffering_period_write( h, &h->out.bs );
3694         if( x264_nal_end( h ) )
3695            return -1;
3696         /* buffering period sei must follow AUD, SPS and PPS and precede all other SEIs */
3697         int idx = 0;
3698         while( h->out.nal[idx].i_type == NAL_AUD ||
3699                h->out.nal[idx].i_type == NAL_SPS ||
3700                h->out.nal[idx].i_type == NAL_PPS )
3701             idx++;
3702         x264_nal_t nal_tmp = h->out.nal[h->out.i_nal-1];
3703         memmove( &h->out.nal[idx+1], &h->out.nal[idx], (h->out.i_nal-idx-1)*sizeof(x264_nal_t) );
3704         h->out.nal[idx] = nal_tmp;
3705     }
3706
3707     int frame_size = x264_encoder_encapsulate_nals( h, 0 );
3708     if( frame_size < 0 )
3709         return -1;
3710
3711     /* Set output picture properties */
3712     pic_out->i_type = h->fenc->i_type;
3713
3714     pic_out->b_keyframe = h->fenc->b_keyframe;
3715     pic_out->i_pic_struct = h->fenc->i_pic_struct;
3716
3717     pic_out->i_pts = h->fdec->i_pts;
3718     pic_out->i_dts = h->fdec->i_dts;
3719
3720     if( pic_out->i_pts < pic_out->i_dts )
3721         x264_log( h, X264_LOG_WARNING, "invalid DTS: PTS is less than DTS\n" );
3722
3723     pic_out->opaque = h->fenc->opaque;
3724
3725     pic_out->img.i_csp = h->fdec->i_csp;
3726 #if HIGH_BIT_DEPTH
3727     pic_out->img.i_csp |= X264_CSP_HIGH_DEPTH;
3728 #endif
3729     pic_out->img.i_plane = h->fdec->i_plane;
3730     for( int i = 0; i < pic_out->img.i_plane; i++ )
3731     {
3732         pic_out->img.i_stride[i] = h->fdec->i_stride[i] * sizeof(pixel);
3733         pic_out->img.plane[i] = (uint8_t*)h->fdec->plane[i];
3734     }
3735
3736     x264_frame_push_unused( thread_current, h->fenc );
3737
3738     /* ---------------------- Update encoder state ------------------------- */
3739
3740     /* update rc */
3741     int filler = 0;
3742     if( x264_ratecontrol_end( h, frame_size * 8, &filler ) < 0 )
3743         return -1;
3744
3745     pic_out->hrd_timing = h->fenc->hrd_timing;
3746     pic_out->prop.f_crf_avg = h->fdec->f_crf_avg;
3747
3748     /* Filler in AVC-Intra mode is written as zero bytes to the last slice
3749      * We don't know the size of the last slice until encapsulation so we add filler to the encapsulated NAL */
3750     if( h->param.i_avcintra_class )
3751     {
3752         x264_t *h0 = h->thread[0];
3753         int ret = x264_check_encapsulated_buffer( h, h0, h->out.i_nal, frame_size, frame_size + filler );
3754         if( ret < 0 )
3755             return -1;
3756         memset( h->out.nal[0].p_payload + frame_size, 0, filler );
3757         h->out.nal[h->out.i_nal-1].i_payload += filler;
3758         h->out.nal[h->out.i_nal-1].i_padding = filler;
3759         frame_size += filler;
3760     }
3761     else
3762     {
3763         while( filler > 0 )
3764         {
3765             int f, overhead;
3766             overhead = (FILLER_OVERHEAD - h->param.b_annexb);
3767             if( h->param.i_slice_max_size && filler > h->param.i_slice_max_size )
3768             {
3769                 int next_size = filler - h->param.i_slice_max_size;
3770                 int overflow = X264_MAX( overhead - next_size, 0 );
3771                 f = h->param.i_slice_max_size - overhead - overflow;
3772             }
3773             else
3774                 f = X264_MAX( 0, filler - overhead );
3775
3776             if( x264_bitstream_check_buffer_filler( h, f ) )
3777                 return -1;
3778             x264_nal_start( h, NAL_FILLER, NAL_PRIORITY_DISPOSABLE );
3779             x264_filler_write( h, &h->out.bs, f );
3780             if( x264_nal_end( h ) )
3781                 return -1;
3782             int total_size = x264_encoder_encapsulate_nals( h, h->out.i_nal-1 );
3783             if( total_size < 0 )
3784                 return -1;
3785             frame_size += total_size;
3786             filler -= total_size;
3787         }
3788     }
3789
3790     /* End bitstream, set output  */
3791     *pi_nal = h->out.i_nal;
3792     *pp_nal = h->out.nal;
3793
3794     h->out.i_nal = 0;
3795
3796     x264_noise_reduction_update( h );
3797
3798     /* ---------------------- Compute/Print statistics --------------------- */
3799     x264_thread_sync_stat( h, h->thread[0] );
3800
3801     /* Slice stat */
3802     h->stat.i_frame_count[h->sh.i_type]++;
3803     h->stat.i_frame_size[h->sh.i_type] += frame_size;
3804     h->stat.f_frame_qp[h->sh.i_type] += h->fdec->f_qp_avg_aq;
3805
3806     for( int i = 0; i < X264_MBTYPE_MAX; i++ )
3807         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
3808     for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
3809         h->stat.i_mb_partition[h->sh.i_type][i] += h->stat.frame.i_mb_partition[i];
3810     for( int i = 0; i < 2; i++ )
3811         h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
3812     for( int i = 0; i < 6; i++ )
3813         h->stat.i_mb_cbp[i] += h->stat.frame.i_mb_cbp[i];
3814     for( int i = 0; i < 4; i++ )
3815         for( int j = 0; j < 13; j++ )
3816             h->stat.i_mb_pred_mode[i][j] += h->stat.frame.i_mb_pred_mode[i][j];
3817     if( h->sh.i_type != SLICE_TYPE_I )
3818         for( int i_list = 0; i_list < 2; i_list++ )
3819             for( int i = 0; i < X264_REF_MAX*2; i++ )
3820                 h->stat.i_mb_count_ref[h->sh.i_type][i_list][i] += h->stat.frame.i_mb_count_ref[i_list][i];
3821     for( int i = 0; i < 3; i++ )
3822         h->stat.i_mb_field[i] += h->stat.frame.i_mb_field[i];
3823     if( h->sh.i_type == SLICE_TYPE_P && h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE )
3824     {
3825         h->stat.i_wpred[0] += !!h->sh.weight[0][0].weightfn;
3826         h->stat.i_wpred[1] += !!h->sh.weight[0][1].weightfn || !!h->sh.weight[0][2].weightfn;
3827     }
3828     if( h->sh.i_type == SLICE_TYPE_B )
3829     {
3830         h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
3831         if( h->mb.b_direct_auto_write )
3832         {
3833             //FIXME somewhat arbitrary time constants
3834             if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
3835                 for( int i = 0; i < 2; i++ )
3836                     h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
3837             for( int i = 0; i < 2; i++ )
3838                 h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
3839         }
3840     }
3841     else
3842         h->stat.i_consecutive_bframes[h->fenc->i_bframes]++;
3843
3844     psz_message[0] = '\0';
3845     double dur = h->fenc->f_duration;
3846     h->stat.f_frame_duration[h->sh.i_type] += dur;
3847     if( h->param.analyse.b_psnr )
3848     {
3849         int64_t ssd[3] =
3850         {
3851             h->stat.frame.i_ssd[0],
3852             h->stat.frame.i_ssd[1],
3853             h->stat.frame.i_ssd[2],
3854         };
3855         int luma_size = h->param.i_width * h->param.i_height;
3856         int chroma_size = CHROMA_SIZE( luma_size );
3857         pic_out->prop.f_psnr[0] = x264_psnr( ssd[0], luma_size );
3858         pic_out->prop.f_psnr[1] = x264_psnr( ssd[1], chroma_size );
3859         pic_out->prop.f_psnr[2] = x264_psnr( ssd[2], chroma_size );
3860         pic_out->prop.f_psnr_avg = x264_psnr( ssd[0] + ssd[1] + ssd[2], luma_size + chroma_size*2 );
3861
3862         h->stat.f_ssd_global[h->sh.i_type]   += dur * (ssd[0] + ssd[1] + ssd[2]);
3863         h->stat.f_psnr_average[h->sh.i_type] += dur * pic_out->prop.f_psnr_avg;
3864         h->stat.f_psnr_mean_y[h->sh.i_type]  += dur * pic_out->prop.f_psnr[0];
3865         h->stat.f_psnr_mean_u[h->sh.i_type]  += dur * pic_out->prop.f_psnr[1];
3866         h->stat.f_psnr_mean_v[h->sh.i_type]  += dur * pic_out->prop.f_psnr[2];
3867
3868         snprintf( psz_message, 80, " PSNR Y:%5.2f U:%5.2f V:%5.2f", pic_out->prop.f_psnr[0],
3869                                                                     pic_out->prop.f_psnr[1],
3870                                                                     pic_out->prop.f_psnr[2] );
3871     }
3872
3873     if( h->param.analyse.b_ssim )
3874     {
3875         pic_out->prop.f_ssim = h->stat.frame.f_ssim / h->stat.frame.i_ssim_cnt;
3876         h->stat.f_ssim_mean_y[h->sh.i_type] += pic_out->prop.f_ssim * dur;
3877         snprintf( psz_message + strlen(psz_message), 80 - strlen(psz_message),
3878                   " SSIM Y:%.5f", pic_out->prop.f_ssim );
3879     }
3880     psz_message[79] = '\0';
3881
3882     x264_log( h, X264_LOG_DEBUG,
3883                   "frame=%4d QP=%.2f NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
3884               h->i_frame,
3885               h->fdec->f_qp_avg_aq,
3886               h->i_nal_ref_idc,
3887               h->sh.i_type == SLICE_TYPE_I ? 'I' : (h->sh.i_type == SLICE_TYPE_P ? 'P' : 'B' ),
3888               h->fdec->i_poc,
3889               h->stat.frame.i_mb_count_i,
3890               h->stat.frame.i_mb_count_p,
3891               h->stat.frame.i_mb_count_skip,
3892               frame_size,
3893               psz_message );
3894
3895     // keep stats all in one place
3896     x264_thread_sync_stat( h->thread[0], h );
3897     // for the use of the next frame
3898     x264_thread_sync_stat( thread_current, h );
3899
3900 #ifdef DEBUG_MB_TYPE
3901 {
3902     static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
3903         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
3904     for( int mb_xy = 0; mb_xy < h->mb.i_mb_width * h->mb.i_mb_height; mb_xy++ )
3905     {
3906         if( h->mb.type[mb_xy] < X264_MBTYPE_MAX && h->mb.type[mb_xy] >= 0 )
3907             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
3908         else
3909             fprintf( stderr, "? " );
3910
3911         if( (mb_xy+1) % h->mb.i_mb_width == 0 )
3912             fprintf( stderr, "\n" );
3913     }
3914 }
3915 #endif
3916
3917     /* Remove duplicates, must be done near the end as breaks h->fref0 array
3918      * by freeing some of its pointers. */
3919     for( int i = 0; i < h->i_ref[0]; i++ )
3920         if( h->fref[0][i] && h->fref[0][i]->b_duplicate )
3921         {
3922             x264_frame_push_blank_unused( h, h->fref[0][i] );
3923             h->fref[0][i] = 0;
3924         }
3925
3926     if( h->param.psz_dump_yuv )
3927         x264_frame_dump( h );
3928     x264_emms();
3929
3930     return frame_size;
3931 }
3932
3933 static void x264_print_intra( int64_t *i_mb_count, double i_count, int b_print_pcm, char *intra )
3934 {
3935     intra += sprintf( intra, "I16..4%s: %4.1f%% %4.1f%% %4.1f%%",
3936         b_print_pcm ? "..PCM" : "",
3937         i_mb_count[I_16x16]/ i_count,
3938         i_mb_count[I_8x8]  / i_count,
3939         i_mb_count[I_4x4]  / i_count );
3940     if( b_print_pcm )
3941         sprintf( intra, " %4.1f%%", i_mb_count[I_PCM]  / i_count );
3942 }
3943
3944 /****************************************************************************
3945  * x264_encoder_close:
3946  ****************************************************************************/
3947 void    x264_encoder_close  ( x264_t *h )
3948 {
3949     int64_t i_yuv_size = FRAME_SIZE( h->param.i_width * h->param.i_height );
3950     int64_t i_mb_count_size[2][7] = {{0}};
3951     char buf[200];
3952     int b_print_pcm = h->stat.i_mb_count[SLICE_TYPE_I][I_PCM]
3953                    || h->stat.i_mb_count[SLICE_TYPE_P][I_PCM]
3954                    || h->stat.i_mb_count[SLICE_TYPE_B][I_PCM];
3955
3956     x264_lookahead_delete( h );
3957
3958 #if HAVE_OPENCL
3959     x264_opencl_lookahead_delete( h );
3960     x264_opencl_function_t *ocl = h->opencl.ocl;
3961 #endif
3962
3963     if( h->param.b_sliced_threads )
3964         x264_threadpool_wait_all( h );
3965     if( h->param.i_threads > 1 )
3966         x264_threadpool_delete( h->threadpool );
3967     if( h->param.i_lookahead_threads > 1 )
3968         x264_threadpool_delete( h->lookaheadpool );
3969     if( h->i_thread_frames > 1 )
3970     {
3971         for( int i = 0; i < h->i_thread_frames; i++ )
3972             if( h->thread[i]->b_thread_active )
3973             {
3974                 assert( h->thread[i]->fenc->i_reference_count == 1 );
3975                 x264_frame_delete( h->thread[i]->fenc );
3976             }
3977
3978         x264_t *thread_prev = h->thread[h->i_thread_phase];
3979         x264_thread_sync_ratecontrol( h, thread_prev, h );
3980         x264_thread_sync_ratecontrol( thread_prev, thread_prev, h );
3981         h->i_frame = thread_prev->i_frame + 1 - h->i_thread_frames;
3982     }
3983     h->i_frame++;
3984
3985     /* Slices used and PSNR */
3986     for( int i = 0; i < 3; i++ )
3987     {
3988         static const uint8_t slice_order[] = { SLICE_TYPE_I, SLICE_TYPE_P, SLICE_TYPE_B };
3989         int i_slice = slice_order[i];
3990
3991         if( h->stat.i_frame_count[i_slice] > 0 )
3992         {
3993             int i_count = h->stat.i_frame_count[i_slice];
3994             double dur =  h->stat.f_frame_duration[i_slice];
3995             if( h->param.analyse.b_psnr )
3996             {
3997                 x264_log( h, X264_LOG_INFO,
3998                           "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",
3999                           slice_type_to_char[i_slice],
4000                           i_count,
4001                           h->stat.f_frame_qp[i_slice] / i_count,
4002                           (double)h->stat.i_frame_size[i_slice] / i_count,
4003                           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,
4004                           h->stat.f_psnr_average[i_slice] / dur,
4005                           x264_psnr( h->stat.f_ssd_global[i_slice], dur * i_yuv_size ) );
4006             }
4007             else
4008             {
4009                 x264_log( h, X264_LOG_INFO,
4010                           "frame %c:%-5d Avg QP:%5.2f  size:%6.0f\n",
4011                           slice_type_to_char[i_slice],
4012                           i_count,
4013                           h->stat.f_frame_qp[i_slice] / i_count,
4014                           (double)h->stat.i_frame_size[i_slice] / i_count );
4015             }
4016         }
4017     }
4018     if( h->param.i_bframe && h->stat.i_frame_count[SLICE_TYPE_B] )
4019     {
4020         char *p = buf;
4021         int den = 0;
4022         // weight by number of frames (including the I/P-frames) that are in a sequence of N B-frames
4023         for( int i = 0; i <= h->param.i_bframe; i++ )
4024             den += (i+1) * h->stat.i_consecutive_bframes[i];
4025         for( int i = 0; i <= h->param.i_bframe; i++ )
4026             p += sprintf( p, " %4.1f%%", 100. * (i+1) * h->stat.i_consecutive_bframes[i] / den );
4027         x264_log( h, X264_LOG_INFO, "consecutive B-frames:%s\n", buf );
4028     }
4029
4030     for( int i_type = 0; i_type < 2; i_type++ )
4031         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4032         {
4033             if( i == D_DIRECT_8x8 ) continue; /* direct is counted as its own type */
4034             i_mb_count_size[i_type][x264_mb_partition_pixel_table[i]] += h->stat.i_mb_partition[i_type][i];
4035         }
4036
4037     /* MB types used */
4038     if( h->stat.i_frame_count[SLICE_TYPE_I] > 0 )
4039     {
4040         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
4041         double i_count = h->stat.i_frame_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
4042         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4043         x264_log( h, X264_LOG_INFO, "mb I  %s\n", buf );
4044     }
4045     if( h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4046     {
4047         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
4048         double i_count = h->stat.i_frame_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
4049         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_P];
4050         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4051         x264_log( h, X264_LOG_INFO,
4052                   "mb P  %s  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
4053                   buf,
4054                   i_mb_size[PIXEL_16x16] / (i_count*4),
4055                   (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4056                   i_mb_size[PIXEL_8x8] / (i_count*4),
4057                   (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
4058                   i_mb_size[PIXEL_4x4] / (i_count*4),
4059                   i_mb_count[P_SKIP] / i_count );
4060     }
4061     if( h->stat.i_frame_count[SLICE_TYPE_B] > 0 )
4062     {
4063         int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
4064         double i_count = h->stat.i_frame_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
4065         double i_mb_list_count;
4066         int64_t *i_mb_size = i_mb_count_size[SLICE_TYPE_B];
4067         int64_t list_count[3] = {0}; /* 0 == L0, 1 == L1, 2 == BI */
4068         x264_print_intra( i_mb_count, i_count, b_print_pcm, buf );
4069         for( int i = 0; i < X264_PARTTYPE_MAX; i++ )
4070             for( int j = 0; j < 2; j++ )
4071             {
4072                 int l0 = x264_mb_type_list_table[i][0][j];
4073                 int l1 = x264_mb_type_list_table[i][1][j];
4074                 if( l0 || l1 )
4075                     list_count[l1+l0*l1] += h->stat.i_mb_count[SLICE_TYPE_B][i] * 2;
4076             }
4077         list_count[0] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L0_8x8];
4078         list_count[1] += h->stat.i_mb_partition[SLICE_TYPE_B][D_L1_8x8];
4079         list_count[2] += h->stat.i_mb_partition[SLICE_TYPE_B][D_BI_8x8];
4080         i_mb_count[B_DIRECT] += (h->stat.i_mb_partition[SLICE_TYPE_B][D_DIRECT_8x8]+2)/4;
4081         i_mb_list_count = (list_count[0] + list_count[1] + list_count[2]) / 100.0;
4082         sprintf( buf + strlen(buf), "  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%",
4083                  i_mb_size[PIXEL_16x16] / (i_count*4),
4084                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
4085                  i_mb_size[PIXEL_8x8] / (i_count*4),
4086                  i_mb_count[B_DIRECT] / i_count,
4087                  i_mb_count[B_SKIP]   / i_count );
4088         if( i_mb_list_count != 0 )
4089             sprintf( buf + strlen(buf), "  L0:%4.1f%% L1:%4.1f%% BI:%4.1f%%",
4090                      list_count[0] / i_mb_list_count,
4091                      list_count[1] / i_mb_list_count,
4092                      list_count[2] / i_mb_list_count );
4093         x264_log( h, X264_LOG_INFO, "mb B  %s\n", buf );
4094     }
4095
4096     x264_ratecontrol_summary( h );
4097
4098     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 )
4099     {
4100 #define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
4101 #define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
4102         int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
4103         int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
4104                                  + SUM3b( h->stat.i_mb_count, I_16x16 );
4105         int64_t i_all_intra = i_intra + SUM3b( h->stat.i_mb_count, I_PCM);
4106         int64_t i_skip = SUM3b( h->stat.i_mb_count, P_SKIP )
4107                        + SUM3b( h->stat.i_mb_count, B_SKIP );
4108         const int i_count = h->stat.i_frame_count[SLICE_TYPE_I] +
4109                             h->stat.i_frame_count[SLICE_TYPE_P] +
4110                             h->stat.i_frame_count[SLICE_TYPE_B];
4111         int64_t i_mb_count = (int64_t)i_count * h->mb.i_mb_count;
4112         int64_t i_inter = i_mb_count - i_skip - i_intra;
4113         const double duration = h->stat.f_frame_duration[SLICE_TYPE_I] +
4114                                 h->stat.f_frame_duration[SLICE_TYPE_P] +
4115                                 h->stat.f_frame_duration[SLICE_TYPE_B];
4116         float f_bitrate = SUM3(h->stat.i_frame_size) / duration / 125;
4117
4118         if( PARAM_INTERLACED )
4119         {
4120             char *fieldstats = buf;
4121             fieldstats[0] = 0;
4122             if( i_inter )
4123                 fieldstats += sprintf( fieldstats, " inter:%.1f%%", h->stat.i_mb_field[1] * 100.0 / i_inter );
4124             if( i_skip )
4125                 fieldstats += sprintf( fieldstats, " skip:%.1f%%", h->stat.i_mb_field[2] * 100.0 / i_skip );
4126             x264_log( h, X264_LOG_INFO, "field mbs: intra: %.1f%%%s\n",
4127                       h->stat.i_mb_field[0] * 100.0 / i_intra, buf );
4128         }
4129
4130         if( h->pps->b_transform_8x8_mode )
4131         {
4132             buf[0] = 0;
4133             if( h->stat.i_mb_count_8x8dct[0] )
4134                 sprintf( buf, " inter:%.1f%%", 100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
4135             x264_log( h, X264_LOG_INFO, "8x8 transform intra:%.1f%%%s\n", 100. * i_i8x8 / i_intra, buf );
4136         }
4137
4138         if( (h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO ||
4139             (h->stat.i_direct_frames[0] && h->stat.i_direct_frames[1]))
4140             && h->stat.i_frame_count[SLICE_TYPE_B] )
4141         {
4142             x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%% temporal:%.1f%%\n",
4143                       h->stat.i_direct_frames[1] * 100. / h->stat.i_frame_count[SLICE_TYPE_B],
4144                       h->stat.i_direct_frames[0] * 100. / h->stat.i_frame_count[SLICE_TYPE_B] );
4145         }
4146
4147         buf[0] = 0;
4148         int csize = CHROMA444 ? 4 : 1;
4149         if( i_mb_count != i_all_intra )
4150             sprintf( buf, " inter: %.1f%% %.1f%% %.1f%%",
4151                      h->stat.i_mb_cbp[1] * 100.0 / ((i_mb_count - i_all_intra)*4),
4152                      h->stat.i_mb_cbp[3] * 100.0 / ((i_mb_count - i_all_intra)*csize),
4153                      h->stat.i_mb_cbp[5] * 100.0 / ((i_mb_count - i_all_intra)*csize) );
4154         x264_log( h, X264_LOG_INFO, "coded y,%s,%s intra: %.1f%% %.1f%% %.1f%%%s\n",
4155                   CHROMA444?"u":"uvDC", CHROMA444?"v":"uvAC",
4156                   h->stat.i_mb_cbp[0] * 100.0 / (i_all_intra*4),
4157                   h->stat.i_mb_cbp[2] * 100.0 / (i_all_intra*csize),
4158                   h->stat.i_mb_cbp[4] * 100.0 / (i_all_intra*csize), buf );
4159
4160         int64_t fixed_pred_modes[4][9] = {{0}};
4161         int64_t sum_pred_modes[4] = {0};
4162         for( int i = 0; i <= I_PRED_16x16_DC_128; i++ )
4163         {
4164             fixed_pred_modes[0][x264_mb_pred_mode16x16_fix[i]] += h->stat.i_mb_pred_mode[0][i];
4165             sum_pred_modes[0] += h->stat.i_mb_pred_mode[0][i];
4166         }
4167         if( sum_pred_modes[0] )
4168             x264_log( h, X264_LOG_INFO, "i16 v,h,dc,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4169                       fixed_pred_modes[0][0] * 100.0 / sum_pred_modes[0],
4170                       fixed_pred_modes[0][1] * 100.0 / sum_pred_modes[0],
4171                       fixed_pred_modes[0][2] * 100.0 / sum_pred_modes[0],
4172                       fixed_pred_modes[0][3] * 100.0 / sum_pred_modes[0] );
4173         for( int i = 1; i <= 2; i++ )
4174         {
4175             for( int j = 0; j <= I_PRED_8x8_DC_128; j++ )
4176             {
4177                 fixed_pred_modes[i][x264_mb_pred_mode4x4_fix(j)] += h->stat.i_mb_pred_mode[i][j];
4178                 sum_pred_modes[i] += h->stat.i_mb_pred_mode[i][j];
4179             }
4180             if( sum_pred_modes[i] )
4181                 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,
4182                           fixed_pred_modes[i][0] * 100.0 / sum_pred_modes[i],
4183                           fixed_pred_modes[i][1] * 100.0 / sum_pred_modes[i],
4184                           fixed_pred_modes[i][2] * 100.0 / sum_pred_modes[i],
4185                           fixed_pred_modes[i][3] * 100.0 / sum_pred_modes[i],
4186                           fixed_pred_modes[i][4] * 100.0 / sum_pred_modes[i],
4187                           fixed_pred_modes[i][5] * 100.0 / sum_pred_modes[i],
4188                           fixed_pred_modes[i][6] * 100.0 / sum_pred_modes[i],
4189                           fixed_pred_modes[i][7] * 100.0 / sum_pred_modes[i],
4190                           fixed_pred_modes[i][8] * 100.0 / sum_pred_modes[i] );
4191         }
4192         for( int i = 0; i <= I_PRED_CHROMA_DC_128; i++ )
4193         {
4194             fixed_pred_modes[3][x264_mb_chroma_pred_mode_fix[i]] += h->stat.i_mb_pred_mode[3][i];
4195             sum_pred_modes[3] += h->stat.i_mb_pred_mode[3][i];
4196         }
4197         if( sum_pred_modes[3] && !CHROMA444 )
4198             x264_log( h, X264_LOG_INFO, "i8c dc,h,v,p: %2.0f%% %2.0f%% %2.0f%% %2.0f%%\n",
4199                       fixed_pred_modes[3][0] * 100.0 / sum_pred_modes[3],
4200                       fixed_pred_modes[3][1] * 100.0 / sum_pred_modes[3],
4201                       fixed_pred_modes[3][2] * 100.0 / sum_pred_modes[3],
4202                       fixed_pred_modes[3][3] * 100.0 / sum_pred_modes[3] );
4203
4204         if( h->param.analyse.i_weighted_pred >= X264_WEIGHTP_SIMPLE && h->stat.i_frame_count[SLICE_TYPE_P] > 0 )
4205             x264_log( h, X264_LOG_INFO, "Weighted P-Frames: Y:%.1f%% UV:%.1f%%\n",
4206                       h->stat.i_wpred[0] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P],
4207                       h->stat.i_wpred[1] * 100.0 / h->stat.i_frame_count[SLICE_TYPE_P] );
4208
4209         for( int i_list = 0; i_list < 2; i_list++ )
4210             for( int i_slice = 0; i_slice < 2; i_slice++ )
4211             {
4212                 char *p = buf;
4213                 int64_t i_den = 0;
4214                 int i_max = 0;
4215                 for( int i = 0; i < X264_REF_MAX*2; i++ )
4216                     if( h->stat.i_mb_count_ref[i_slice][i_list][i] )
4217                     {
4218                         i_den += h->stat.i_mb_count_ref[i_slice][i_list][i];
4219                         i_max = i;
4220                     }
4221                 if( i_max == 0 )
4222                     continue;
4223                 for( int i = 0; i <= i_max; i++ )
4224                     p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i_list][i] / i_den );
4225                 x264_log( h, X264_LOG_INFO, "ref %c L%d:%s\n", "PB"[i_slice], i_list, buf );
4226             }
4227
4228         if( h->param.analyse.b_ssim )
4229         {
4230             float ssim = SUM3( h->stat.f_ssim_mean_y ) / duration;
4231             x264_log( h, X264_LOG_INFO, "SSIM Mean Y:%.7f (%6.3fdb)\n", ssim, x264_ssim( ssim ) );
4232         }
4233         if( h->param.analyse.b_psnr )
4234         {
4235             x264_log( h, X264_LOG_INFO,
4236                       "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
4237                       SUM3( h->stat.f_psnr_mean_y ) / duration,
4238                       SUM3( h->stat.f_psnr_mean_u ) / duration,
4239                       SUM3( h->stat.f_psnr_mean_v ) / duration,
4240                       SUM3( h->stat.f_psnr_average ) / duration,
4241                       x264_psnr( SUM3( h->stat.f_ssd_global ), duration * i_yuv_size ),
4242                       f_bitrate );
4243         }
4244         else
4245             x264_log( h, X264_LOG_INFO, "kb/s:%.2f\n", f_bitrate );
4246     }
4247
4248     /* rc */
4249     x264_ratecontrol_delete( h );
4250
4251     /* param */
4252     if( h->param.rc.psz_stat_out )
4253         free( h->param.rc.psz_stat_out );
4254     if( h->param.rc.psz_stat_in )
4255         free( h->param.rc.psz_stat_in );
4256
4257     x264_cqm_delete( h );
4258     x264_free( h->nal_buffer );
4259     x264_free( h->reconfig_h );
4260     x264_analyse_free_costs( h );
4261
4262     if( h->i_thread_frames > 1 )
4263         h = h->thread[h->i_thread_phase];
4264
4265     /* frames */
4266     x264_frame_delete_list( h->frames.unused[0] );
4267     x264_frame_delete_list( h->frames.unused[1] );
4268     x264_frame_delete_list( h->frames.current );
4269     x264_frame_delete_list( h->frames.blank_unused );
4270
4271     h = h->thread[0];
4272
4273     for( int i = 0; i < h->i_thread_frames; i++ )
4274         if( h->thread[i]->b_thread_active )
4275             for( int j = 0; j < h->thread[i]->i_ref[0]; j++ )
4276                 if( h->thread[i]->fref[0][j] && h->thread[i]->fref[0][j]->b_duplicate )
4277                     x264_frame_delete( h->thread[i]->fref[0][j] );
4278
4279     if( h->param.i_lookahead_threads > 1 )
4280         for( int i = 0; i < h->param.i_lookahead_threads; i++ )
4281             x264_free( h->lookahead_thread[i] );
4282
4283     for( int i = h->param.i_threads - 1; i >= 0; i-- )
4284     {
4285         x264_frame_t **frame;
4286
4287         if( !h->param.b_sliced_threads || i == 0 )
4288         {
4289             for( frame = h->thread[i]->frames.reference; *frame; frame++ )
4290             {
4291                 assert( (*frame)->i_reference_count > 0 );
4292                 (*frame)->i_reference_count--;
4293                 if( (*frame)->i_reference_count == 0 )
4294                     x264_frame_delete( *frame );
4295             }
4296             frame = &h->thread[i]->fdec;
4297             if( *frame )
4298             {
4299                 assert( (*frame)->i_reference_count > 0 );
4300                 (*frame)->i_reference_count--;
4301                 if( (*frame)->i_reference_count == 0 )
4302                     x264_frame_delete( *frame );
4303             }
4304             x264_macroblock_cache_free( h->thread[i] );
4305         }
4306         x264_macroblock_thread_free( h->thread[i], 0 );
4307         x264_free( h->thread[i]->out.p_bitstream );
4308         x264_free( h->thread[i]->out.nal );
4309         x264_pthread_mutex_destroy( &h->thread[i]->mutex );
4310         x264_pthread_cond_destroy( &h->thread[i]->cv );
4311         x264_free( h->thread[i] );
4312     }
4313 #if HAVE_OPENCL
4314     x264_opencl_close_library( ocl );
4315 #endif
4316 }
4317
4318 int x264_encoder_delayed_frames( x264_t *h )
4319 {
4320     int delayed_frames = 0;
4321     if( h->i_thread_frames > 1 )
4322     {
4323         for( int i = 0; i < h->i_thread_frames; i++ )
4324             delayed_frames += h->thread[i]->b_thread_active;
4325         h = h->thread[h->i_thread_phase];
4326     }
4327     for( int i = 0; h->frames.current[i]; i++ )
4328         delayed_frames++;
4329     x264_pthread_mutex_lock( &h->lookahead->ofbuf.mutex );
4330     x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
4331     x264_pthread_mutex_lock( &h->lookahead->next.mutex );
4332     delayed_frames += h->lookahead->ifbuf.i_size + h->lookahead->next.i_size + h->lookahead->ofbuf.i_size;
4333     x264_pthread_mutex_unlock( &h->lookahead->next.mutex );
4334     x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
4335     x264_pthread_mutex_unlock( &h->lookahead->ofbuf.mutex );
4336     return delayed_frames;
4337 }
4338
4339 int x264_encoder_maximum_delayed_frames( x264_t *h )
4340 {
4341     return h->frames.i_delay;
4342 }